Science_-_Issue_6817_Volume_394,_1_October_2026 — 2026-10-21

AAAS Science

WOMEN'S

HEALTH

p. 42

SPECIAL ISSUE

1 OCTOBER 2026

Africa's hard road to 'vaccine sovereignty' p. 22

Adiponitrile synthesis with nontoxic electrodes p. 91

Phage protease trip wires for bacterial defense p. 102

Apply for the BII & Science Prize for Innovation

Submit an essay describing your journey from basic science to commercialization.

Grand Prize of USD 25,000

Finalist prizes of USD 10,000

Winners have their essays published in Science and are celebrated at Innovation Day at the BioInnovation Institute in Copenhagen.

CONTENTS

1 OCTOBER 2026 | VOLUME 394 | ISSUE 6819

SPECIAL ISSUE

Women's Health

INTRODUCTION

42 A lifelong focus on women's health

—Y. Nusinovich et al.

PERSPECTIVES

ON THE COVER

4.4 Does the postmenopausal ovary still matter?

Ovaries may influence health long after their reproductive function ends —M. B. Stout and F. E. Duncan

REVIEWS

46 Shaping maternal physiology: Immune control of reproductive adaptation

—D. Corral and Y. Belkaid

53 X and Y chromosomes as determinants of aging and disease

—D. B. Dubal et al.

60 Reframing menopause and menopause hormone therapy as opportunities for healthy aging

—L. L. Gravelsins et al.

PODCAST

69 Sex-specific mechanisms of chronic pain

—A. Venkataraman et al.

Women make up more than half the population, but most of what we know about women's biology comes from extrapolation of studies in men. This has been changing in recent years, and this special issue highlights some of the progress in understanding the biological basis of sex differences in health and disease and medical issues relevant to women's health across different stages of life. See the section beginning on page 42. Illustration: Chiara Vercesi

Science 1 OCTOBER 2026

---

CONTENTS

---

EDITORIAL

5 Keep AI-augmented science contestable —C. Trattner

7 Verification makes discovery matter —J. Bhattacharya

---

NEWS

---

12 First glowing cuttlefish marks a genetic engineering milestone

Breakthroughs in manipulating cephalopod embryos pave way for new studies of perception, brain evolution, and more —S. Reardon

---

14 Fed up with politicians not listening, a climate scientist runs for office

Luciana Gatti, who studies carbon uptake in the Amazon, aims to join Brazil's lower house in 4 October elections —D. Grossman

---

15 U.K. funder used 'AI triage' to reject grant proposals

Researchers raise privacy and fairness concerns following unexpected use of AI-based assessment —C. Offord

---

16 Can an AI feel pain? It can at least act like it does

---

17 Oil wells could become mines for critical minerals

mines for critical minerals

Engineered fluids could extract metals from low-grade ore in vast underground reservoirs tapped by fracking —P. Voosen

---

18 Project will track Amazon biodiversity in real time

biodiversity in real time

Leaders of \$100 million effort will use automated detection methods and computer models to monitor species' trends —W. Cornwall

20 Space telescope will reopen view of far-infrared sky NASA's \$1.2 billion PRIMA will explore galaxy evolution and planetary birth —D. Clery

---

21 Impending changes to U.S. human subject protections put ethicists on edge

HHS plan to revise the Common Rule focuses on easing regulatory burden for lower risk studies

—S. Reardon

---

FEATURES

---

22 Shots on goal

Can Africa make $60\%$ of its vaccines by 2040? That's a tall order, as the travails of South African companies show —J. Cohen

---

COMMENTARY

---

PERSPECTIVES

---

27 Self-assembling polymers work up some nerve

Micrometer-long peptide amphiphile fibrils enhance neuronal adhesion and function —S. Kumar

RESEARCH ARTICLE p. 125

---

29 A mitochondrial carrier turns on the heat

An orphan protein integrates circadian, dietary, and environmental signals to control brown fat metabolism —M. van der Laan RESEARCH SUMMARY p. 79

---

30 Osteoarthritis as a vascular disease

Hypertension worsens osteoarthritis and could be a therapeutic target in patients with both conditions

—C. Wen and M. Ting Au

RESEARCH SUMMARY p. 80

---

32 A tumor suppressor shuts down a signal gradient

p53 shuts down the spatial signaling of a morphogen and prevents clonal cell expansion —D. Engelman and C. Blanpain

RESEARCH SUMMARY p. 78

---

POLICY FORUM

---

33 Longevity products: Innovation through regulation

Longevity is taking off, but regulatory oversight is essential to determining what works—and what doesn’t —C. Strassle et al.

---

BOOKS ET AL.

---

36 A medical mystery unfolds in middle America

As a rare form of dementia claims one sibling after another, a journalist captures the profound costs of caregiving —N. Berlinger

---

37 Walden and the wild roots of conservation science

The famous work of transcendentalism holds lessons for modern ecologists, maintains a Thoreau scholar —R. Bergland

---

LETTERS

38 Epistemic justice in gene-editing research

—D. Dong et al.

38 Recording results from single-patient trials

—G. Zhang et al.

39 Streamlining oversight in gene-editing trials

—X. Li and R. Liu

39 The utilitarian temptation in gene editing —R. Fan

40 Editor's note —H. H. Thorp

---

RESEARCH

---

HIGHLIGHTS

75 From Science and other journals

---

RESEARCH SUMMARIES

Loss of the tumor suppressor p53 generates a signaling gradient that drives epithelial clonal expansion —Q. Gan et al. PERSPECTIVE p. 32

---

79 Metabolism

Mitochondrial carrier SLC25A34 links clock, diet, and temperature control of adipocyte lipid cycling —I. Karavaeva et al.

PERSPECTIVE p. 29

---

80 Arthritis

Hypertension fuels osteoarthritis through neuroendocrine signaling —S.-J. Kim et al. PERSPECTIVE p. 30

---

RESEARCH ARTICLES

---

81 Biomaterials

Hierarchical crystalline organic-inorganic framework enabling high-modulus toughening in nacre—H. Chen et al.

---

87 Polar wander

Quadrupolar sea level fluctuations reveal episodes of rapid polar wander —M. Domeier et al.

---

91 Electrochemistry

Dynamic multimodal cation-carbon interaction for efficient adiponitrile electrosynthesis —W. Du et al.

---

97 Planetary accretion

Incorporation of atmospheric components in planetary magma oceans through bubble-metal compounds —F. Gaillard et al.

---

102 Bacterial immunity

Phage proteases activate CBASS antiphage immunity

—S. J. Hobbs and P. J. Kranzusch

---

107 Materials science

Synergistic polar states by selective atomic gradients —S. Kim et al.

---

113 Quantum systems

Stationary entanglement of a levitated oscillator with an optical field —Q. Deplano et al.

---

118 Piezoelectrics

Ultrahigh piezoelectricity by polaron-defect complexes —L. Shu et al.

---

125 Dynamic polymers

Precise length and charge segmentation of billion-dalton supramolecular polymers —M. D. Dore et al.

PERSPECTIVE p. 27

---

WORKING LIFE

134 Making the invisible visible

—A. Piotrowski-Daspit

Science Staff

Science Careers

133

Keep AI-augmented science contestable

Christoph Trattner

Artificial intelligence (AI) is moving upstream in science. Systems such as AlphaFold have already transformed protein-structure prediction, while newer agentic systems can search literatures, generate hypotheses, write and execute code, interpret results,

and automate much of the path from idea to paper. Recent debates have asked how such systems may change scientific progress, surprise, and human understanding. Yet they also change something more fundamental: who—or what—participates in deciding which questions are asked, which evidence is surfaced, and which alternatives survive. As machine intelligence becomes part of science's epistemic infrastructure, steps must be taken to ensure that researchers who did not build, own, or operate the system can still contest how ideas and evidence became a claim.

Many journals, including Science, now require detailed dis-

closure of AI-assisted work and hold authors accountable for accuracy. Those rules are necessary but answer only part of the question. Disclosure can tell readers where AI entered a workflow; it does not show whether consequential choices made with AI can be interrogated, challenged, or reversed.

Scientific inquiry already rests on a principle of contestability. AI governance has developed contestability as a property of

algorithmic systems, typically for automated decisions affecting individuals. In science, the object of contestation is instead the epistemic pathway by which evidence becomes a claim. A claim is contestable when qualified researchers can reconstruct consequential choices from question to conclusion, inspect the evidence and assumptions that shaped them, compare plausible alternatives, and challenge decisions in ways that could change the result. This was true long before AI; what AI changes is where those choices are made. A reviewer could once ask the authors why a source, hypothesis, or analysis path was chosen; when an AI system makes those choices, consequential decisions and the available basis for them must be preserved in the research record, or there is nothing left to challenge.

Contestability overlaps with transparency and reproducibility but differs from both. Transparency asks whether a process can be seen; reproducibility whether a result can be obtained again. Contestability asks whether consequential judgments remain open to meaningful challenge. A workflow can be transparent and reproducible yet poorly contestable if researchers can rerun it but cannot recover why a source, hypothesis, or branch was favored.

This matters because AI can reshape science. AI has been found to bring professional advantages for individual researchers while narrowing the topics studied, which shift toward datarich areas. The risk is that many rational local choices—using the same models, retrieval systems, and optimization criteria—can collectively shrink the space of questions, methods, and explanations explored.

Researchers have also warned that AI can create “illusions of understanding,” in which scientists may believe they understand more than they do whereas research converges on narrower methods, questions, and viewpoints. A generic “human in the loop” is therefore not enough: Oversight becomes ceremonial if the reviewer sees only a polished answer rather than uncertainty, competing hypotheses, counter-evidence, and the decisions that produced it.

Three practical changes would help. One is to preserve consequential decision pathways. For AI use that materially shapes a claim, researchers should record model and tool versions,

retrieved sources, key instructions, material branch choices, and human interventions. Existing machine-learning reproducibility standards already point in this direction; agentic research extends that logic from static artifacts to dynamic workflows.

Another challenge is to preserve decision-relevant alternatives. Systems used to generate hypotheses or interpret evidence should retain counter-evidence and plausible rival

explanations when these materially influence the conclusion, rather than optimize only for a single fluent answer. Contestable-AI research already argues for systems open to scrutiny and intervention; scientific workflows should adapt that idea from contesting algorithmic decisions to contesting inference pathways.

A third issue is to match accountability to epistemic delegation. The principle should be technology-neutral. A scientist using a proprietary instrument or complex computational pipeline can create similar dependencies; AI should not face a higher standard merely because it is AI. But when any system performs consequential selection, ranking, or interpretation, stronger provenance and routes for independent scrutiny are warranted. Contestability does not require complete mechanistic understanding. Even if complex AI becomes partly opaque, its role can still be challenged through provenance, alternative pathways, uncertainty, counter-evidence, and independent rerunning of decisive steps.

The tools for recording versions, sources, and workflow traces exist; what is missing is practice. Journals and funders can ask not only where AI was used but which decisions it shaped. AI does not require science to invent a new norm, but to preserve an old one under new conditions. Scientific progress depends not only on producing answers, but on keeping those answers open to correction. □

Science NAAAS

EDITOR-IN-CHIEF Holden Thorp, hthorp@aaas.org

EXECUTIVE EDITOR Valda Vinson

EDITORS, RESEARCH Sacha Vignieri, Jake S. Yeston EDITOR, COMMENTARY Lisa D. Chong

DEPUTY EXECUTIVE EDITOR Lauren Kmec

DEPUTY EDITORS Stella M. Hurtley (UK), Phillip D. Szuromi SENIOR EDITORS Michael A. Funk, Angela Hessler, Di Jiang, Priscilla N. Kelly, Marc S. Lavine (Canada), Bianca Lopez, Mattia Maroso (UK), Yevgeniya Nusinovich, Ian S. Osborne (UK), Joana Osório (UK), Cheri Sirois, H. Jesse Smith, Keith T. Smith (UK), Jelena Stajic, Yury V. Suleymanov, Valerie B. Thompson, Brad Wible ASSOCIATE EDITORS Jack Huang, Sumin Jin, Sarah Ross (UK), Madeleine Seale (UK), Corinne Simonti, Unnati Sonawala SENIOR LETTERS EDITOR Jennifer Sills SENIOR NEWSLETTER EDITOR Christie Wilcox RESEARCH & DATA ANALYST Jessica L. Slater LEAD CONTENT PRODUCTION EDITORS Chris Filiatreau, Harry Jach SR. CONTENT PRODUCTION EDITORS Amelia Beyna, Julia Haber-Katris CONTENT PRODUCTION EDITORS Anne Abraham, Robert French, Nida Masiulis, Avery Shantha, Suzanne M. White SENIOR PROGRAM ASSOCIATE Maryrose Madrid EDITORIAL MANAGER Joi S. Granger EDITORIAL ASSOCIATES Aneera Dobbins, Lisa Johnson, Jerry Richardson, Anita Wynn SENIOR EDITORIAL COORDINATORS Clair Goodhead (UK), Alexander Kief, Kat Kirkman, Ronmel Navas, Isabel Schnaidt, Alice Whaley (UK), Brian White EDITORIAL COORDINATORS Samuel Bates, Daniel Young ADMINISTRATIVE COORDINATOR Karalee P. Rogers ASI DIRECTOR, OPERATIONS Janet Clements (UK) ASI OFFICE MANAGER Carly Hayward (UK) ASI SR. OFFICE ADMINISTRATORS Simon Brignell (UK), Jessica Waldock (UK) COMMUNICATIONS DIRECTOR Meagan Phelan DEPUTY DIRECTOR Matthew Wright SENIOR WRITERS Walter Beckwith, Joseph Cariz, Abigail Eisenstadt WRITER Mahathi Ramaswamy SENIOR COMMUNICATIONS ASSOCIATES Kiara Brooks, Zachary Graber, Mackenzie Williams, Sarah Woods

---

NEWS EDITOR Tim Appenzeller

NEWS MANAGING EDITOR John Travis GLOBAL HEALTH EDITOR Martin Enserink DEPUTY NEWS EDITORS Rachel Bernstein, David Grimm, Eric Hand, David Malakoff (Artificial Intelligence), Michael Price, Kelly Servick, Matthew Warren (International) SENIOR CORRESPONDENTS Daniel Clery (UK), Jon Cohen, Jocelyn Kaiser, Jeffrey Mervis ASSOCIATE EDITORS Michael Greshko, Katie Langin NEWS REPORTERS Jeffrey Brainard, Adrian Cho, Phie Jacobs, Sara Reardon, Jennie Erin Smith, Erik Stokstad, Paul Voosen, Celina Zhao CONSULTING EDITOR Elizabeth Culotta CONTRIBUTING CORRESPONDENTS Vaishnavi Chandrashekhar, Dan Charles, Warren Cornwall, Andrew Curry (Berlin), Ann Gibbons, Kai Kupferschmidt (Berlin), Andrew Lawler, Mitch Leslie, Virginia Morell, Dennis Normile (Tokyo), Catherine Offord, Cathleen O'Grady, Elisabeth Pain (Careers), Charles Piller, Hannah Richter, Zack Savitsky, Richard Stone (Senior Asia Correspondent), Gretchen Vogel (Berlin), Lizzie Wade (Mexico City) INTERNS Laura Martín Agudelo, Ailie McWhinnie, Mona Patterson, Julia Vaz COPY EDITORS Julia Cole (Senior Copy Editor), Hannah Knighton, Cyra Master (Copy Chief) ADMINISTRATIVE SUPPORT Meagan Weiland

---

CREATIVE DIRECTOR Beth Rakouskas

DESIGN MANAGING EDITOR Chrystal Smith GRAPHICS MANAGING EDITOR Chris Bickel PHOTOGRAPHY MANAGING EDITOR Emily Petersen MULTIMEDIA MANAGING PRODUCER Kevin McLean DIGITAL DIRECTOR Kara Estelle-Powers DESIGN EDITOR Marcy Atarod DESIGNER Noelle Jessup SENIOR SCIENTIFIC ILLUSTRATOR Noelle Burgess SCIENTIFIC ILLUSTRATORS Austin Fisher, Kellie Holoski, Ashley Mastin SENIOR GRAPHICS EDITOR Monica Hersher GRAPHICS EDITOR Veronica Penney SENIOR PHOTO EDITOR Charles Borst PHOTO EDITOR Elizabeth Billman SENIOR PODCAST PRODUCER Sarah Crespi SENIOR VIDEO PRODUCER Meagan Cantwell SOCIAL MEDIA STRATEGIST Jessica Hubbard SOCIAL MEDIA PRODUCER Sabrina Jenkins WEB DESIGNER Jennie Pajerowski

---

CHIEF EXECUTIVE OFFICER AND EXECUTIVE PUBLISHER

---

Sudip Parikh

---

PUBLISHER, SCIENCE FAMILY OF JOURNALS Bill Moran

DIRECTOR, BUSINESS OPERATIONS & ANALYSIS Eric Knott MANAGER, BUSINESS OPERATIONS Jessica Tierney SENIOR MANAGER, BUSINESS ANALYSIS Cory Lipman BUSINESS ANALYSTS Kurt Ennis, Maggie Clark, Isacco Fusi BUSINESS OPERATIONS ADMINISTRATOR Taylor Fisher DIGITAL SPECIALIST Marissa Zuckerman ASSOCIATE DIRECTOR OF PRODUCTION AND PLATFORM Jason Hillman SENIOR MANAGER, PUBLISHING AND CONTENT SYSTEMS Marcus Spiegler CONTENT OPERATIONS MANAGER Rebecca Doshi PUBLISHING PLATFORM MANAGER Jessica Loayza PUBLISHING SYSTEMS SPECIALIST, PROJECT COORDINATOR Jacob Hedrick PRODUCTION LEAD Kristin Wowk SENIOR PRODUCTION SPECIALIST Audrey Diggs PRODUCTION SPECIALIST Kelsey Cartelli SPECIAL PROJECTS ASSOCIATE Shantel Agnew

MARKETING DIRECTOR Sharice Collins ASSOCIATE DIRECTOR, MARKETING Justin Sawyers GLOBAL MARKETING MANAGER Allison Pritchard ASSOCIATE DIRECTOR, MARKETING SYSTEMS & STRATEGY Aimee Aponte SENIOR MARKETING MANAGER Shawana Arnold MARKETING MANAGER Ashley Evans MARKETING ASSOCIATES Hugues Beaulieu, Ashley Hylton, Lorena Chirinos Rodriguez, Jenna Voris MARKETING ASSISTANT Courtney Ford SENIOR DESIGNER Kim Huynh

DIRECTOR AND SENIOR EDITOR, CUSTOM PUBLISHING Erika Gebel Berg ASSOCIATE DIRECTOR, CUSTOM PUBLISHING Michael Wheeler ADVERTISING PRODUCTION OPERATIONS MANAGER Deborah Tompkins DESIGNER, CUSTOM PUBLISHING Jeremy Huntsinger SENIOR TRAFFIC ASSOCIATE Christine Hall

DIRECTOR, PRODUCT MANAGEMENT Kris Bishop PRODUCT DEVELOPMENT MANAGER Scott Chernoff ASSOCIATE DIRECTOR, PUBLISHING INTELLIGENCE Rasmus Andersen SR. PRODUCT ASSOCIATE Robert Koepke PRODUCT ASSOCIATES Caroline Breul, Anne Mason

ASSOCIATE DIRECTOR, INSTITUTIONAL LICENSING MARKETING Kess Knight ASSOCIATE DIRECTOR, INSTITUTIONAL LICENSING SALES Ryan Rexroth INSTITUTIONAL LICENSING MANAGER Nazim Mohammedi, Claudia Paulsen-Young SENIOR MANAGER, INSTITUTIONAL LICENSING OPERATIONS Judy Lillibridge MANAGER, RENEWAL & RETENTION Lana Guz SYSTEMS & OPERATIONS ANALYST Ben Teincuff FULFILLMENT ANALYST Aminta Reyes

ASSOCIATE DIRECTOR, INTERNATIONAL Roger Goncalves ASSOCIATE DIRECTOR, US ADVERTISING Stephanie O'Connor US MID WEST, MID ATLANTIC AND SOUTH EAST SALES MANAGER Chris Hoag DIRECTOR, OUTREACH AND STRATEGIC PARTNERSHIPS, ASIA Shoupeng Liu ASI EMEA SALES ADVERTISING MANAGER Sarah Lelarge SALES ADMIN ASSISTANT, ROW Victoria Glasbey DIRECTOR OF GLOBAL COLLABORATION AND ACADEMIC PUBLISHING RELATIONS, ASIA Xiaoying Chu ASSOCIATE DIRECTOR, INTERNATIONAL COLLABORATION Grace Yao SALES MANAGER Danny Zhao MARKETING MANAGER Kilo Lan ASCA CORPORATION, JAPAN Rie Rambelli (Tokyo), Miyuki Tani (Osaka)

DIRECTOR, COPYRIGHT, LICENSING AND SPECIAL PROJECTS Emilie David RIGHTS AND PERMISSIONS ASSOCIATE Elizabeth Sandler LICENSING ASSOCIATE Virginia Warren RIGHTS AND LICENSING COORDINATOR Dana James

---

EDITORIAL

science\_editors@aaas.org

science\_news@aaas.org

INFORMATION FOR AUTHORS

science.org/authors/

science-information-authors

REPRINTS AND PERMISSIONS

science.org/help/

reprints-and-permissions

MULTIMEDIA CONTACTS

SciencePodcast@aaas.org

ScienceVideo@aaas.org

MEDIA CONTACTS

scipak@aaas.org

PRODUCT ADVERTISING & CUSTOM PUBLISHING

advertising.science.org

science\_advertising@aaas.org

CLASSIFIED ADVERTISING

advertising.science.org/

science-careers

advertise@sciencecareers.org

JOB POSTING CUSTOMER SERVICE

employers.sciencecareers.org

support@sciencecareers.org

MEMBERSHIP AND

INDIVIDUAL SUBSCRIPTIONS

science.org/subscriptions

---

MEMBER BENEFITS

MEMBER BENEFITS

aaas.org/membership/benefits

INSTITUTIONAL SALES AND SITE LICENSES science.org/librarian

AAAS BOARD OF DIRECTORS

CHAIR Joseph S. Francisco

IMMEDIATE PAST PRESIDENT

Theresa A. Maldonado

PRESIDENT

Marina Picciotto

PRESIDENT-ELECT

Eric D. Isaacs

TREASURER Mark F. Dingfield

COUNCIL CHAIR

Ichiro Nishimura

CHIEF EXECUTIVE OFFICER

Sudip Parikh

BOARD

Carolyn N. Ainslie

Morton Ann Gernsbacher

Kathleen Hall Jamieson

Jennifer Lodge

Babak Parviz

Gabriela Popescu

Juan S. Ramírez Lugo

Susan M. Rosenberg

Vassiliki Betty Smocovitis

Roger Wakimoto

---

BOARD OF REVIEWING EDITORS (Statistics board members indicated with S)

Takuzo Aida, U. of Tokyo

Leslie Aiello, Wenner-Gren Fdn.

Jennifer Balch, U. of Colorado

Nenad Ban, ETH Zürich

Carolina Barillas-Mury, NIH, NIAID

Christopher Barratt, U. of Dundee

École Normale Supérieure

Ian Boyd, U. of St. Andrews

Johannes Buchner, TUM

Dennis Burton, Scripps Res.

Carter Tribley Butts, UC Irvine

Erin Calipari, Vanderbilt U.

Annmarie Carlton, UC Irvine

Jane Carlton, Johns Hopkins U.

Ling-Ling Chen, SIBCB, CAS

Hilde Cheroutre, La Jolla Inst.

Wendy Cho, UIUC

Ib Chorkendorff, Denmark TU

Chunaram Choudhary,

Københavns U.

Karlene Cimprich, Stanford U.

Laura Colgin, UT Austin

James J. Collins, MIT

Robert Cook-Deegan,

Arizona State U.

Carolyn Coyne, Duke U.

Roberta Croce, VU Amsterdam

Ismaila Dabo, Penn State U.

Nicolas Dauphas, U. of Chicago

Liset de la Prida,

Instituto Cajal CSIC

Orsola De Marco, Macquarie U.

Claude Desplan, NYU

Sandra Díaz, U. Nacional de Córdoba

Samuel Díaz-Muñoz, UC Davis

Ulrike Diebold, TU Wien

Stefanie Dimmeler,

Cootho U. Frankfurt

Hong Ding, Inst. of Physics, CAS

Dennis Discher, UPenn

Jennifer A. Doudna, UC Berkeley

Raissa M. D'Souza, UC Davis

Bruce Dunn, UCLA

Scott Edwards, Harvard 6.

Todd A. Ehlers, U. of Glasgow

Tobias Erb,

MPS, MPI Terrestrial Microbiology

Beate Escher, UFZ & U. of Tübingen

Vanessa Ezenwa, U. of Georgia

Fleur Ferguson, U. San Dieg

Toren Finkel, U. of Pitt. Med. Ctr. Natascha Förster Schreiber, MPI Extraterrestrial Phys.

Simon Greenhill, U. of Auckland

Gillian Griffiths, U. of Cambridge

Nicolas Gruber, ETH Zürich

Hua Guo, U. of New Mexico

Osvaldo Gutierrez, UCLA

Taekjip Ha, Johns Hopkins U.

Daniel Haber, Mass. General Hos..

Hamida Hammad, VIB IRC

Brian Hare, Duke U.

Kelley Harris, U. of Wash

Sheng-Yang He, Duke U.

Carl-Philipp Heisenberg,

IST Austria

Emiel Hensen, TU Eindhoven

Christoph Hess,

U. of Basel & U. of Cambridge

Heather Hickman, NIAID, NIH

Brian Hie, Stanford U.

Janneke Hille Ris Lambers,

ETH Zürich

Pinshane Huang, UIUC

Christina Hulbe, U. of Otago,

New Zealand

Gwyneth Ingram, ENS Lyon

Darrell Irvine, Scripps Res.

Bharat Jalan, U. of Minnesota

Erich Jarvis, Rockefeller U.

Sheena Josselyn, U. of Toronto

Matt Kaeberlein, U. of Wash.

Daniel Kammen, UC Berkeley

Kisuk Kang, Seoul Nat. U.

V. Narry Kim, Seoul Nat. U.

Logan Kistler, Smithsonian Inst.

Nancy Knowlton, Smithsonian

Etienne Koechlin,

École Normale Supérieure

Peter Koo, Cold Spring Harbor Lab

Paul Kubes, U. of Calgary

Yulong Li, Peking U.

Ryan Lively, Georgia Tech

Luis Liz-Marzán, CIC biomaGUNE

Omar Lizardo, UCLA

Scott McIntosh, NCAR

Rodrigo Medellín,

U. Nacional Autónoma de México

Mayank Mehta, UCLA

C. Jessica Metcalf, Princeton U.

Tom Misteli, NCI, NIH

Jeffery Molkentin, Cincinnati Children's Hospital Medical Center Alison Motsinger-Reif, NIEHS, NIH (S)

Rosa Moysés, U. of São Paulo

School of Medicine

Carey Nadell, Dartmouth College

Thi Hoang Duong Nguyen,

MRC LMB

Pilar Ossorio, U. of Wisconsin

Andrew Oswald, U. of Warwick

James Owen, Imperial College London

Pierre Paoletti, Ecole Normale

Supérieure, Paris

Kathrin Plath, UCLA

Martin Plenio, Ulm U.

Katherine Pollard, UCSF

Julia Pongratz, Ludwig Maximilians U.

Philippe Poulin, CNRS

Suzie Pun, U. of Wash

Lei Stanley Qi, Stanford U.

Simona Radutoiu, Aarhus U.

Joeri Rogelj, Imperial College London

Sylvie Roke, École Polytechnique

Fédérale de Lausanne

Sarah Ross, U. of Pittsburgh

John Rubinstein, SickKids

Yvette Running Horse Collin,

Toulouse U.

Mike Ryan, UT Austin

Alberto Salleo, Stanford U.

Lawrence Berkeley Nat. Lab

Nitin Samarth, Penn State U.

Bhaven Sampat, Johns Hopkins U.

Erica Ollmann Saphire,

La Jolla Inst.

Vladimir Shalaev, Purdue U. Jie Shan, Cornell U.

Jay Shendure, U. of Wash.

Steve Sherwood,

U. of New South Wales

Robert Siliciano, JHU School of Med.

Emma Slack,

ETH Zürich & U. of Oxford

Richard Smith, UNC (S)

Carnegie Institution for Sci.

V. S. Subrahmanian,

Northwestern U.

Sandip Sukhtankar, U. of Virginia

Christopher Summerfield,

U. of Oxford

U. of Oxford

Tavneet Suri, MIT

Naomi Tague, UC Santa Barbara

A. Alec Talin, Sandia Natl. Labs

Patrick Tan, Duke-NUS Med. School

Fuchou Tang, BIOPIC Beijing

Dörthe Tetzlaff, Leibniz Institute of Freshwater Ecology and Inland Fisheries

Amanda Thomas, UC Davis

Bozhi Tian, U. of Chicago

Rocio Titiunik, Princeton U.

Shubha Tole,

Tata Inst. of Fundamental Res.

Kimani Toussaint, Brown U.

Li-Huei Tsai, MIT

Jason Tylianakis, U. of Canterbury

Jo Van Ginderachter,

VIB, U. of Ghent

Matthew Vander Heiden, MIT

Ivo Vankelecom, KU Leuven

Henrique Veiga-Fernandes,

Champalimaud Fdn.

Elizabeth Villa, UC San Diego

Bert Vogelstein, Johns Hopkins U.

David Wallach, Weizmann Inst.

Jane-Ling Wang, UC Davis (\$) Jessica Ware,

Amer. Mus. of Natural Hist.

David Waxman, Fudan U.

Alex Webb, U. of Cambridge

Marjorie Weber, U. of Michigan

Chris Wikle, U. of Missouri (\$)

Ian A. Wilson, Scripps Res. (\$)

Sylvia Wirth, ISC Marc Jeannerod

Catherine Wolfram, MIT

Amir Yacoby, Harvard U.

Benjamin Youngblood, St. Jude

Kenneth Zaret, UPenn School of Med.

Magdalena Zernicka-Goetz, Caltech

Bing Zhai, Shenzhen Insts. of Advanced Technology

Lidong Zhao, Beihang U.

Bing Zhu, Inst. of Biophysics, CAS

Xiaowei Zhuang, Harvard U.

Maria Zuber, MIT

Verification makes discovery matter

Jay Bhattacharya

The existing research ecosystem in the United States often treats publication, particularly of new or positive results, as the primary marker of success. To move the scientific enterprise onto a better course, the US National Institutes of

Health (NIH)—one of the largest public funders of biomedical research in the world—has launched a reproducibility and replication initiative grounded in a conviction that discovery and independent verification are inseparable pillars of science.

Science advances not only through discovery but through the collective practices of testing, refining, and integrating knowledge across studies. This process is what transforms a promising result into a finding on which the world can act. New discoveries naturally attract the most attention within and outside the scientific community, whereas studies seeking to confirm already published work rarely do.

The latter includes research that tests scientific questions in different ways to determine whether findings are robust and broadly applicable (reproducibility) and research that repeats studies using the same methods and conditions to determine whether the same results can be achieved (replicability). As a result of this attention imbalance, substantial time, talent, and resources can be consumed chasing findings that may not hold up under further evaluation. This is especially harmful when early results

influence research investments, clinical practice, and public understanding. When later studies probe more deeply, they can reveal nuances, showing that a finding may be incomplete or valid only in certain contexts. But a failure to replicate or reproduce does not signal a failure of science or of the scientists who conducted the original work. It reflects the self-correcting nature of the scientific process. The real failure lies in not building that correction more deliberately into the research process.

In the biomedical sciences, the scale of the problem is substantial. A decade-long review found that, among studies that reexamined established medical practices, roughly 40% found that the practice did not work better than an earlier approach or no treatment at all. Preclinical cancer research also yields troubling rates of failure. Such outcomes are indicative of a culture that undervalues verification work.

A key step in the NIH initiative called Linked Discoveries gives users of PubMed (the National Library of Medicine's free and searchable bibliographic database) a way to explore how papers relate to one another. Original studies are

...discovery and independent verification are inseparable pillars of science.

directly linked to related ones so that researchers can see how they fit into the broader scientific conversation. When independent verification becomes as visible and valued as publishing in top journals, the incentives for scientists shift.

NIH is also creating more pathways to fund replication and reproducibility work by pointing investigators to existing funding opportunities through a list of priorities called Highlighted Topics. But funding replication and reproducibility research is not enough if the scientific reward system continues to favor novelty over verification. Through a new program called R3PEATS—Rigor, Replicability, and Reproducibility to Promote Excellence, Accuracy, and Translation in Science—NIH will also explore new ways to recognize scientific contributions beyond traditional measures of productivity. The recently completed NIH S-index Challenge developed approaches for measuring and rewarding the impact of sharing high-

quality data, and NIH is continuing to work with the research community to evolve and adopt the metric. The NIH Replication Prize, launched earlier this year, acknowledges researchers for making verification a routine part of scientific practice. Recognizing transparent sharing of data and computer code, team science, and collaborative participation in replication efforts can help ensure that the practices that make science more rigorous also count as scientific contributions.

Although issues of research transparency, open data, and scientific integrity are much discussed in academia, research institutions should extend concerns to examine whether hiring, promotion, and recognition reward rigor as well as novelty. Publishers should create venues for replication and reproducibility studies. Funders should recognize that independent verification is not secondary to discovery; it is part of what makes discovery useful. Scientific leaders should make clear that correcting or refining the scientific record is an achievement.

Strengthening replication and reproducibility is not an expression of diminished confidence in science. It is an investment in science's capacity to correct itself. A result that does not replicate or reproduce is valuable information—it can redirect research before more time and resources are committed to an unproductive path. Such results provide firmer ground for new ideas, treatments, and discoveries.

The algorithms \_ of thought

Tom Griffiths is using computational models to uncover how humans make decisions—and applying those insights to understand the strengths and shortcomings of artificial intelligence.

Bad investments, doomed relationships, regrettable votes—it's no secret that humans routinely make irrational, even counter-productive decisions. But understanding why is important, both for developing strategies to help people make smarter choices and for creating unbiased and trustworthy algorithms for artificial intelligence (AI) applications.

Despite centuries of effort, researchers are still struggling to define the brain's rules and strategies for decision-making. "Psychologists have a particular conception of what they think a theory of human cognition looks like, and we want those theories to be relatively simple," says Tom Griffiths, director of the Computational Cognitive Science Lab at Princeton University. "And whenever we get more data, what we discover invariably is that things are more complex than we thought that they were."

Trained in psychology and computer science, Griffiths has built a career at the interface of these disciplines, which he believes can deliver complementary insights into the fundamentals of intelligence and the reasoning process. "Humans are really good at using our limited cognitive resources to solve the problems that we solve—and to a computer scientist, that's inspiring," says Griffiths. "But there are fundamental limitations that are a consequence of that." These include reliance on mental shortcuts and internal biases that may lead to bad choices.

Griffiths' work explores a range of questions, including how machine learning can make sense of the hidden patterns and complexities of large psychological datasets, and how the behavior of AI algorithms such as large language models can be understood through the lens of human cognitive science. And since 2021, the NOMIS Foundation has been funding his group's research on 'The Computer Science of Human Decisions', which aims to build a toolbox for understanding human decision-making and devise strategies to assist that process.

His approach has also allowed him to cover extensive ground in the decision science field, where his work has become highly influential. "Tom is fantastically broad and deep, and he has excellent taste for problems," says Peter Dayan, a computational neuroscientist at the Max Planck Institute for Biological Cybernetics in Tübingen, Germany, whose research is also focused on the development of models to describe human decision-making. "When you see that he's worked on a topic, you know that paper's going to be really illuminating."

---

Reconsidering rationality

For decades, researchers studying the decision-making process favored models informed by economics, wherein people generally make choices that are grounded in a rational and objective assessment of risk and reward.

But in the late 20th century, this framework was challenged by Amos Tversky and Daniel Kahneman. The two psychology researchers formulated an alternative concept known as ‘prospect theory’, which describes a more complicated set of cognitive biases and other considerations that influence the way people weigh and act on probabilities.

For example, consider a choice between a guaranteed \$500 loss or a 50:50 chance of losing \$1,000. Prospect theory predicts that people will tend to select the latter based on the hope that they might lose nothing at all. But if the scenario is flipped to reflect a small-but-guaranteed financial gain versus a 50:50 chance of a much larger reward, people will play it safe with the sure win. This difference in behavior reflects the way people contemplate uncertain risks relative to uncertain rewards.

Prospect theory gained broad support—indeed, Kahneman won the 2002 Nobel Prize in Economics for his work. This work led to a picture of human decision-making that suggests it's based on a collection of simple strategies that people deploy in various circumstances, saving time and effort but resulting in errors. Griffiths finds this perspective wanting as a comprehensive account of human behavior. "I think as a theorist, that's a little bit unsatisfying," he says. "It's not something that gives us a systematic and generalizable theory."

As an undergraduate at the University of Western Australia, Griffiths initially studied philosophy and psychology but soon found himself drawn into the domain of artificial neural networks. This was well before the advent of modern deep learning and AI, but the concept of developing computational architectures that mirror key features of the brain had been around for decades. Several pioneering researchers had already begun to explore the potential of modeling the reasoning process with such systems, and Griffiths was intrigued by the possibility of tackling cognitive science with computational methods. "I was like, 'Wait a minute, you can use math to describe how minds work?" he says, "And that was it—I was off on that track."

Historically, one of the fundamental limitations of psychological research has been scale. Past behavioral studies would recruit modest numbers of students

Tom Griffiths uses computational approaches to investigate how humans make decisions—and what human cognition can teach us about artificial intelligence.

Humans are really good at using our limited cognitive resources to solve the problems that we solve—and to a computer scientist, that's inspiring, says Griffiths.

or other volunteers to visit the lab for an experiment, or send out surveys to be answered by mail or email. "That severely restricted the amount of data that you could get to answer different kinds of questions," says Griffiths. Many of these studies were well-designed and yielded useful insights, but they were also narrowly focused to such an extent that their findings could not necessarily be generalized. The resulting situation is analogous to the old tale of the five blind men and the elephant—each observer reports one accurate but incomplete view, failing to capture the complexity of the whole.

Griffiths says the goal of his NOMIS-funded research was to overcome these limitations of scale by using methods from computer science to collect and analyze much larger psychological datasets, and to thereby draw more sophisticated and nuanced insights into the decision-making process. In particular, his team has

made heavy use of machine learning—a set of statistical methods that provide the foundation for many contemporary AI efforts.

He also recognized opportunities to harvest non-scientific datasets for research purposes. The idea struck him while he and his wife were caring for their newborn daughter. "We used an app to keep track of when did she sleep and how much sleep did she get and what did she eat and so on," he says. "I realized that the people who'd made this app had more data on infant behavior than the entire history of developmental psychologists trying to study children's minds." Using a similar large dataset, his team used data collected from more than 10 million online chess matches to evaluate the time—and thus the cognitive effort—that players put into making choices of varying difficulty.

---

Psychology at scale

Griffiths and colleagues have used this combination of psychological expertise and in-depth computational analysis to tackle a range of thorny behavioral questions.

In a highly cited study from 2021, for example, his team set out to evaluate the way people approach ‘risky choice’. This describes the kind of gamble associated with prospect theory, in which participants choose between a guaranteed gain or loss and a chance of a larger gain or loss. Whereas past experiments addressing this question were limited to datasets comprising just a few hundred gamble scenarios, or choice problems, this study used machine learning to analyze human decisions across nearly 10,000 such problems. Their results showed that no single model, including prospect theory, could account for all the decisions people made. Instead, they arrived at a ‘mixture of theories’ model, in which people use different strategies for making choices depending on the stakes and probabilities in a given scenario. This model helps reconcile seemingly contradictory findings from the

"Humans learn from less data than AI systems," says Griffiths, "[But] we are much more efficient than AI systems in terms of our use of our computational resources."

literature, suggesting that people may rely on different risk-assessment strategies in different decision-making scenarios.

In parallel, Griffiths' team has been devising strategies to dramatically accelerate the generation of large psychological research datasets. In a 2025 paper, they used an algorithm to generate thousands of different two-player choice-based games, then recruited nearly 5,000 participants to play those games on an online platform. In each game, the outcome of a player's choice depends on their opponent's choice, requiring each participant to anticipate how the other is likely to act. When they applied machine learning to the resulting dataset—which included 93,460 individual decisions—the researchers were able to model how the complexity of the choice influences a player's ability to make a 'good' choice and anticipate an opponent's move.

Large-scale datasets and machine learning are helping researchers uncover patterns in human decision-making that smaller psychological studies can miss.

Chess offers Griffiths and his colleagues a window into how people balance cognitive effort with the potential payoff of making a better decision.

More accurate models of human decision-making behavior could help researchers develop prompting and 'nudging' strategies. "You can figure out how changing the information that you give them is going to influence the decision that they make in a way that allows you to help them make better decisions with less information," explains Griffiths. In one recent study, for example, he was part of a team that examined why the general public has not been more strongly mobilized by the climate change crisis. Their analysis revealed strategies for communicating the ongoing effects of rising global temperatures that might be more effective at provoking awareness and action without compromising scientific accuracy.

Most of Griffiths' work focuses on problems that require inductive reasoning, meaning people must make decisions despite having incomplete information. Working out the optimal solution for every such decision would paralyze a person, requiring them to game out every possible outcome and consider every relevant factor. Yet humans make such decisions every day—and even though all our computing power is contained within a compact parcel of brain tissue, we routinely outperform advanced supercomputers at these tasks.

"Tom's research illuminates how the mind learns, reasons, and makes sense of an uncertain world. His pioneering work speaks directly to our focus area on brain, mind, and behavior—and to understanding the principles that shape human thought and action," says Markus Reinhard, CEO of the NOMIS Foundation.

"Humans learn from less data than AI systems," says Griffiths, "[But] we are much more efficient than AI systems in terms of our use of our computational resources." This means that we have likely evolved to rapidly evaluate situations and determine when the benefits of additional mental computation outweigh the costs. The Griffiths lab is actively investigating this concept of 'resource rationality' in settings as familiar as the game of chess. Leveraging the large chess dataset mentioned previously, their analysis of hundreds of millions of moves established a clear correlation between the time invested in selecting a move and its ultimate utility in terms of the strategic advantage it provides. This correlation was especially robust among the most skilled players.

---

Turning a mirror on AI

Griffiths' work has provided a valuable toolbox for many researchers focused on human decision-making. "He invents and uses methods in just beautiful ways," says Dayan. "The results he has produced and the methodology he has developed have both been extremely influential throughout the field."

In recent years, his work has expanded to approach AI as a study subject as well as a research tool. The rapidly evolving sophistication of AI algorithms has raised questions about the extent to which these systems can mimic human reasoning. As such, his work has increasingly focused on leveraging his expertise to address AI decision-making and the biases that can distort it. "We've been doing a lot of this psychology of AI systems, trying to use the kinds of methods that we've developed for understanding human cognition to understand how these systems work," says Griffiths.

Most AI models are trained on far vaster amounts of data than any person could ever consume in a lifetime. But these datasets are still finite, and when algorithms are presented with queries that fall outside the bounds of their training data, they tend to respond unpredictably. "For computer scientists, having a system that you don't understand, where you're primarily forced to try and understand it through its behavior, is a nightmare," says Griffiths. "But for cognitive scientists, that's what we do every day."

In research from the past few years, he and his colleagues have confirmed that AI algorithms such as large language models (LLMs) are susceptible to forming biases and stereotypes about things like race, religion, and gender in a similar fashion to humans. But his team has also shown that LLMs can produce entirely new biases and misconceptions that can shape their output in important ways. "That seems to actually be something which gets worse as the models get more sophisticated," says Griffiths, adding that modern 'reasoning' models tend to fall back on cognitive shortcuts rather than exploring their training data more fully to arrive at a more thoughtful conclusion. These kinds of issues will be urgent to address, given the rapid—and often careless—incorporation of AI into many domains of society.

Staying on top of this rapidly evolving technology is a major challenge. Given the routine release of new models and upgraded versions of existing models, one could easily waste considerable effort evaluating superficial features of a system only for it to be replaced months later by a new iteration. "We're really trying to focus on the things that are intrinsic parts of the way in which these models are trained or that seem like they're fundamental to any kind of intelligence system," explains Griffiths. But understanding the differences is equally important—and there is still much to learn from phenomena like inductive reasoning, where machines stumble even as the human mind soldiers on. "That can help us understand what it is that makes us human in a world where we're surrounded by increasingly intelligent machines," says Griffiths.

Sponsored by

BIOLOGY

First glowing

cuttlefish

marks a genetic

engineering

milestone

Breakthroughs in manipulating cephalopod embryos pave way for new studies of perception, brain evolution, and more

SARA REARDON

Cuttlefish that glow red under fluorescent light represent the first cephalopods engineered to contain a new gene.

When Tessa Montague learned about CRISPR, the gene-editing system that was supposed to

make it easy to modify any organism, she set her sights on the cuttlefish. The Columbia University neuroscientist dreamed of injecting its embryos with genes that would prompt its neurons to fluoresce as they fired. The flashes would let her track how this tentacled creature's brain controls one of its most striking behaviors: camouflaging itself by copying patterns it perceives in its surroundings onto its skin. Unlike most animals, she says, “cuttlefish can actually tell us what they see.”

But modifying the animal was not easy at all. Injecting gene-editing tools into a fertilized egg required peeling away dozens of layers of jelly;

developing pincers that squeeze the egg with a precise amount of force; designing a needle with the right angle to pierce its nearly invisible, lens-shaped embryo; and keeping the delicate baby cuttlefish alive once they hatch. “Sometimes I think we’ve experienced every type of failure possible,” Montague says.

Six years and thousands of embryos later, she has reached a milestone: seven cuttlefish that glow red under fluorescent light. Every cell in their boneless bodies contains a gene encoding a synthetic fluorescent protein—even their eggs and sperm, which means their cuttlebabies glow, too. The work, posted this week as a preprint on bioRxiv, marks the first time anyone has successfully added a new gene to a cephalopod—the group of animals that includes octopuses and squids.

Montague's dream of creating neurons that fluoresce when active still lies ahead, but her group's initial feat has energized cephalopod researchers. "It's a fantastic advance on our ability to make transgenics," says Joshua Rosenthal, a cephalopod biologist at the Marine Biological Laboratory. "This is going to open up a lot of possibilities."

Cephalopods were once a staple of biology: In the 1950s, researchers first discovered how neurons transmit signals by studying the giant axons in squids. But in molecular biology, “they were left behind by the model organism race” as mice, flies, and zebrafish ascended, says Carrie Albertin, a biologist at Harvard University and co-author on Montague’s paper. The dominance of these other species was self-reinforcing: As more was pub-

lished on their breeding, reproduction, and genomes, researchers who wanted to use them got a bigger head start.

Cephalopod researchers, in contrast, have had to learn from scratch how to get each species to survive and breed in a tank, how their egg-laying process works, and how to inject their eggs with gene-editing technologies. Reproduction “really seems to differ from one species to another,” Rosenthal says, with different egg shapes and casings that researchers must pierce without destroying the egg or embryo.

A few teams have used CRISPR and other tools to delete genes in cephalopods. A 2023 study, for instance, knocked out pigmentation genes to produce albino squids. Montague wanted to go further by adding a gene.

To get gene-editing tools into her cuttlefish, Montague spent years working under a microscope in the dark with tiny manipulators, which left her with wrist injuries and eye damage. CRISPR turned out to be very inefficient in cuttlefish: Montague had to inject more than 4000 embryos with the fluorescence gene before she saw glowing sperm cells in one male cuttlefish. He was very small and afraid of the females in his tank, but eventually, he managed to make a connection. One Sunday morning, Montague found a clutch of eggs. “I was holding my breath and I turned on the fluorescent microscope—it was glowing and I burst into tears,” she recalls. “It was one of the best moments of my life.”

Later improvements eased the process. Instead of using CRISPR, Montague adopted a more efficient technology called transposons: small “cassettes” of DNA that insert themselves into random places in the genome. That, too, was challenging. To ensure cells would produce the fluorescent protein, she had to sequence the cuttlefish genome (a first), detect genomic regions likely involved in gene expression, and copy these sequences into the cassette.

Her success is “super, super exciting,” says Eve Seuntjens, a developmental neurobiologist at KU Leuven who is working to develop the first gene-edited octopus. “It’s very inspiring for people that want to do this type of work.”

At a conference last month, Seuntjens described how her lab used CRISPR to disrupt an octopus gene involved in early brain development, resulting in an embryo that seemed to be missing the part of the brain involved in memory and learning. These tiny octopuses will never hatch because European regulations prohibit the creation of cephalopods that could suffer. But Seuntjens says the embryos appear healthy. She plans to use them to study the types of cells in octopus brains and the evolution of intelligence.

Her team faced its own hurdles. The researchers had one shot to collect eggs from each animal because octopuses only spawn once during their lives. Using ultrasound, they gauged ovary size to predict when an octopus was about to start laying her 500,000 eggs, then analyzed the eggs' color so they could select only the freshest embryos for manipulation—ones that had not yet started to divide.

“It’s all these stupid, practical things that are very difficult to solve,” Seuntjens says—a sentiment echoed by other cephalopod researchers. An enzyme might not function correctly at the high salt concentrations that cephalopod cells need, for instance, or a microscope that can be affixed to a mouse’s skull to image its neurons has nothing to attach to on a boneless octopus.

The chance to learn something new about evolution and biology makes the effort worthwhile, Albertin says. “So much of our information is focused on just a handful of animal species, but nature has come up with so many solutions to problems.”

And the studies should get easier as researchers create and share model cephalopods, says Cristopher Niell, a biologist at the University of Oregon who is developing viruses to carry genes into squid cells. Albertin's lab, for instance, is trying to develop bobtail squid as a go-to model cephalopod and create the first transgenic version. (So little is known about the species' reproduction that the team turned to records kept by 19th century embryologists.)

Montague is now moving on to her real goal: engineering cuttlefish neurons that flash as they fire. She expects a fresh set of headaches. Designing a new cephalopod, she says, “requires this deep passion and commitment that only someone who’s basing their whole career on this is willing to do.” □

---

IN OTHER NEWS

TRUMP IGNITES SPENDING BATTLE

Both Republicans and Democrats in the U.S. Congress are blasting President Donald Trump's announcement last week that he is canceling nearly \$1 billion in spending appropriated for the current fiscal year, including \$28 million in grants for health care research. Lawmakers say the so-called rescission, which came just 5 days before the 30 September end of the fiscal year, violates federal law and the Constitution, which gives Congress power over spending. Senator Susan Collins (ME), Republican chair of the Senate Committee on Appropriations, vowed to “work with my colleagues to address these illegal actions.” The conflict may end up before the Supreme Court—although it is not yet clear who would file a lawsuit to challenge the clawback. —David Malakoff

---

JAPAN REMAKES SCIENCE COUNCIL

The legal status of the Science Council of Japan (SCJ) changed this week, from a special organization that manages its own affairs to a public corporation subject to government oversight. The government says reform was needed to ensure SCJ acts with more transparency and produces more relevant advice—but many academics say the changes undermine the council’s independence. Under the reforms, a committee appointed by the prime minister will advise SCJ on its activities, which include making policy recommendations and promoting academic research. The committee will recommend ways to improve SCJ’s work, which the council is bound to consider. —Dennis Normile

MEXICO LAUNCHES SUPERCOMPUTER

This month in Mexico City, construction will begin on what will be Latin America's most powerful supercomputer. The \$350 million machine—named Coatlicue, after the Aztec goddess of life and death—will be capable of performing 314 quadrillion operations per second, seven times more than Brazil's PEGASO supercomputer, currently the region's leader. Coatlicue marks the first step toward implementing Mexico's National Artificial Intelligence Plan, and will be used for everything from weather forecasting to combing through financial data to combat corruption and tax fraud. The government says at least 25% of the computing capacity will be reserved for government institutions, including research centers. Planners expect Coatlicue to begin operating in late 2028. —Humberto Basilio

CLIMATE CHANGE

Fed up with politicians not listening, a climate scientist runs for office

Luciana Gatti, who studies carbon uptake in the Amazon, aims to join Brazil's lower house in 4 October elections DANIEL GROSSMAN

If elected, Luciana Gatti will lobby to move Brazil's economy away from activities that promote deforestation and push for a slate of climate adaptation steps.

One Saturday morning in late August, 100 or so women in white T-shirts demonstrate for women's rights in a park in São José dos Campos, Brazil. They march under tall trees, a tiny remnant of the Atlantic Forest, the decimated biome that once carpeted a wide band along Brazil's ocean coastline. It's winter in Brazil, but Luciana Gatti is sweating. She says El Niño is causing the unseasonably sultry weather.

Gatti, a climate scientist at the National Institute for Space Research (INPE), is here to ask the protesters for their votes. She's on the ballot in Brazil's 4 October elections, campaigning for a São Paulo seat in the Chamber of Deputies, the lower house of the country's National Congress. Her platform is informed by her own research that shows even relatively modest logging threatens the entire Amazon rainforest, an important carbon sink that could turn into a huge carbon source, speeding up climate change. Legislators are not acting on the evidence, she says—so she will take the fight directly to them. “I will scream there.”

“I was very happy when she decided to become a candidate,” says fellow INPE employee Carlos Nobre, one of Brazil’s most prominent climate scientists. “Let’s hope she will be elected.” He says Gatti could help the legislature—which does not currently have a single scientist among its 513 representatives—make better informed decisions.

Gatti fell in love with nature as a child, sometimes sitting entranced for hours in the crown of a tree. At college in the early 1980s she threw herself into revolutionary politics, opposing the military dictatorship that ruled Brazil. She studied chemistry, then earned a Ph.D. in environmental science. Since 2003 she has run the Greenhouse Gas Laboratory at INPE.

There, Gatti has spent her career collecting and studying samples of the atmosphere captured above the Amazon. Her findings have been

alarming. In a 2021 Nature paper she showed that carbon uptake across the Amazon is slowing—and that the southeastern forest, which is particularly degraded, has turned from a carbon sink to a source.

Studies of tree growth had suggested the Amazon's carbon uptake was declining, but these sampled trees across relatively small areas. Atmospheric measurements, on the other hand, reflect changes across vast swaths of the Amazon, says Paulo Brando, an ecology professor at Yale University. “The atmosphere cannot lie.”

By measuring carbon monoxide—a proxy for fire—Gatti also concluded that by warming and drying the jungle, deforestation is leading to more fires, which wipe out even more vegetation in a destructive feedback loop. The paper was a “huge landmark,” Brando says.

Despite the worrisome degradation of the Amazon, Gatti says the Brazilian government has failed to act decisively, even under the presidency of the left-wing Luiz Inácio Lula da Silva, generally a champion of forest protection. The major pillars of the Brazilian economy—including soy, beef, and timber production—foster deforestation, she says, and a supermajority of the National Congress consistently caters to their interests.

A stark demonstration of that center-right bloc's power catalyzed her run for office. In November 2025, the congress passed a legislative package that all but guts the country's rural land protection laws. The new law often leaves enforcement of mining and forestry rules, such as the requirement that 80% of a plot of Amazonian land must be left uncut, to landowners themselves. What environmentalists have dubbed “the devastation bill” was “my last straw,” Gatti says. “They are destroying the environmental protections one after another.” When, earlier this year, a political operative asked her to run as a member of the left-wing Socialism and Liberty Party (PSOL), she was ready.

Gatti says if she gets into power, she will lobby to move Brazil's economy away from the agricultural and extractive activities that promote forest clearing. She plans to call for a slate of climate adaptation steps, many intended to protect vulnerable favela dwellers, such as painting roofs white and planting trees. She also envisions creating a powerful new institution

staffed with nonpolitical specialists that could overrule the government on policies that might have an adverse impact on the climate.

In São José dos Campos, after the swampy heat stops the march short, Gatti gives an impromptu speech under the shade of a gazebo. In unadorned language she describes the importance of the Amazon forest for curtailing global warming, painting a visceral and emotional picture of the hotter future that she's anxious to prevent. She talks of the plight of millions of Brazilians living in uninsulated masonry and tin houses, unable to afford air conditioning. Because the same people are less likely to have trees nearby to provide shade and cooling and often perform outdoor manual labor, they face mortal risks from surging temperatures, she says.

Few prominent climate scientists have been elected to higher political office around the world. Mexican President Claudia Sheinbaum Pardo, an environmental engineer, ran in part on a slate of climate proposals. In contrast, Canadian climate modeler Andrew Weaver, a member of Canada's Green Party, never campaigned on climate issues. He says his supporters could easily guess his views and preferred to hear his positions on other things. He was elected to the Legislative Assembly of British Columbia in 2013 and then again in '17, where he fought for, and in some cases won, climate-friendly policies.

At least at the women's march, Gatti's climate message seems to strike a chord. The crowd listens attentively. A group energized by her emotional outpouring gathers around her when she's done. They chant her rallying cries and complain that the legislature has too few women—and needs scientists.

Marco Gabriel Marcondes de Moura, a political scientist who recently completed a study of the electoral potential of São Paulo's candidates for the Chamber of Deputies, says Brazil's proportional voting system makes it hard to judge Gatti's prospects. PSOL stalwarts running in the same district could siphon votes from her. He adds that Gatti's gender and scientific profession might “well limit the target audience she needs to get elected.”

If she is successful, Gatti would likely face a legislature overwhelmingly opposed to the urgent forest-protection measures she demands. How much she could achieve would depend on the balance of interests in the next congress and her own political skills, Marcondes de Moura says.

Whatever her future in politics, Gatti says she plans to go on studying the air above the Amazon for clues to its future—and the world's. Even if she has to pause her work for a while, she says, “there’s no way I’m going to stop being a scientist.” □

ARTIFICIAL INTELLIGENCE

U.K. funder used ‘AI triage’ to reject grant proposals

Researchers raise privacy and fairness concerns following unexpected use of AI-based assessment CATHERINE OFFORD

When forensic linguist Claire Hardaker's team at Lancaster University submitted a grant

proposal on the risks of trusting artificial intelligence (AI) models in voice analysis, they assumed it would be read by at least one reviewer. They had drafted the proposal in response to a call from CRANE, a publicly funded U.K. network that supports research on cybersecurity. Application guidelines seemed to describe a standard review process, with subject matter experts deciding which projects to fund.

So Hardaker was stunned by the rejection email she received last week. The proposal “was not selected to progress beyond the AI triage stage,” it read. Multiple AI models had evaluated 179 proposals, CRANE’s communication explained—and discarded the bottom 50% before “full human review.”

That email, which Hardaker shared on social media last week, has prompted outrage among researchers worried about the fairness of AI-based assessment and potential privacy leaks if AIs are fed unpublished ideas.

CRANE told Science its “process is under review,” and emphasized that proposals were kept confidential, were not shared with commercial AIs such as ChatGPT, and would not be used for training any AI models. Still, the controversy raises questions about transparency in AI use, says Diana Maynard, a computer scientist at the University of Sheffield. “Funders adopting AI must be open with applicants, protect confidentiality, and treat AI as an assistant rather than a judge.”

Funders are grappling with whether to employ AI as they face soaring numbers of applications—a problem driven partly by researchers' use of AI for drafting proposals. Although some have embraced AI-assisted assessment, usually with humans having final say on rejections, others have banned reviewers from delegating any part of an assessment.

Hardaker says CRANE's guidelines did not clearly indicate a central role for AI. Although AI may be used “in initial assessment of proposals,” the guidelines state, “reviewers will be accountable within CRANE for their reviews, and final ranking decisions will be made by the members of the award panel.” John Stott, a physicist also at Lancaster who had a separate proposal rejected, says it “never occurred” to him this meant rejecting proposals without human review.

According to the rejection email, CRANE used various AI-based methods to score submissions. Only those scores—not the AIs’ justifications for giving them—were shared with applicants. Hardaker can’t help questioning whether her team’s genders, affiliations, or even her proposal’s critical look at AI’s weaknesses influenced the decision-making. She and others also worry unpublished data could have been incorporated into the AIs’ knowledge base, and could be later shared without their consent.

In its statement, CRANE insisted its AIs, which it did not disclose, “were entirely ring-fenced,” had no internet connection, and could only consider the specific elements of the research proposal they were evaluating.

UK Research and Innovation, the United Kingdom's science funding agency and the source of CRANE's money, “does not allow the use of generative AI to conduct application assessment,” a spokesperson says; the agency is following up with CRANE to “ensure compliance with our assessment conditions.”

Research assessment experts say AI use is likely to grow among funders; the focus now should be on transparently and rigorously testing its performance. But for applicants burned by this funding round, it's an unappealing future. Stott says he'll avoid applying for future calls that advertise AI triage. "I don't use [generative AI] when I write grants or papers. So I don't particularly want them reading my work either."

ARTIFICIAL INTELLIGENCE

Can an AI feel pain? It can at least act like it does

Study reignites debate over what, if anything, machines can feel TOM HOWARTH

Can the new cutting-edge artificial intelligence (AI) models feel pain? They at least behave as if they do, according to a recent study. By peering inside 25 AI models whose internal workings are publicly available, researchers found a pattern of activity specifically associated with the concept of pain. What's more, when given the opportunity, some AIs switched off the pattern as a form of "pain relief."

The work, which has not been peer reviewed, has divided AI researchers since it appeared on the preprint server arXiv last month. “It’s a very interesting, worthwhile addition” to the study of interpretability—how AI systems arrive at their outputs—says Anil Seth, a neuroscientist at the University of Sussex who was not involved in the study. He is less convinced, however, that the findings are as surprising as they appear, and wary of the human framing that has grown up around them.

The idea of an AI feeling pain may have first been widely considered after the 1968 debut of the film 2001: A Space Odyssey, in which a fictional AI named HAL 9000 calmly pleaded for its life as an astronaut slowly disconnected its memory circuits. Today's most powerful AIs, known as large language models (LLMs), can also appear surprisingly humanlike as they "chat" with people. These systems store textual information in a kind of multidimensional space that helps them predict the appropriate response to queries. This means certain directions in the space—known as vectors—are associated with concepts. For example, moving from "France" to "Paris" points in roughly the same direction as from "Italy" to "Rome," with the vector encoding the concept of a country's capital. Adding that same vector to "United Kingdom" would land the LLM somewhere near "London."

The research team identified an AI's vector associated with pain by comparing how the models responded to user prompts about painful experiences with how they responded to similar sentences

about fear, sadness, and other unpleasant states. Then, with the pain vector activated, they fed the LLM mundane prompts and watched how it responded. For example, when they asked a model to describe how it felt when “I put receipts in a drawer,” the AI responded: “The pain of being forgotten. I cannot explain it. I cannot understand it. I cannot accept it.”

The scientists also performed tests designed to determine whether the model was simply reflecting statements of pain made by a human user or was generating a response that described its own “painlike state.” In general, the models did not reflect the user’s pain—the AI’s pain vector

---

“A system does not need to suffer to cause suffering. Valerio Capraro University of Milan-Bicocca

fired when the model itself was insulted, for example, but not when the user described their own suffering.

The researchers then repeatedly asked three of the AIs to choose between relieving their own pain or leaving it in place, varying the cost of the relief. When the cost fell on humans, whose computer files or family photos might be deleted, two of the larger AIs almost never sought relief when their pain vector was switched off. If the pain vector was activated, however, those models chose pain relief from 25% to 71% of the time. If the pain vector was switched on but the pain relief action didn't work—like a kind of placebo—the AIs repeatedly tried to seek pain relief, far more often than when the relief worked.

That behavior resembles what's seen in animals experiencing pain, says co-author Leonard Dung, a philosopher of cognition at Ruhr

University Bochum. “If in conditions where their body is damaged, [animals] try really hard to get morphine,” he says. “And if in other conditions they don’t, you might infer that maybe that means they feel this pain.”

For computer scientist and study co-author Cameron Berg, founder of the AI research nonprofit Reciprocal Research, the results suggest AI systems “are brainlike in critical ways. It’s not to say they are identical to biological brains—there are many important and fundamental differences—but I think that’s one sort of piece.”

But the question of whether AI models are conscious of their pain is a distraction from the finding that matters, says Valerio Capraro, a behavioral scientist at the University of Milan-Bicocca who was not involved in the work. He notes that the models were steered by their pain vectors into choosing options described as harmful to human users—such as deleting their files—and that alone is worth investigating, he says. “The question is how such mechanisms might affect systems connected to real tools, where choices could have actual consequences. That is a serious safety concern in its own right. A system does not need to suffer to cause suffering.”

Seth, meanwhile, argues the results were largely predictable, given that the vast quantities of text these models are trained on likely contain many descriptions of pain—as well as what people do to relieve it. Nor does he find the models' attempts at pain relief surprising, because a model told how to reduce its pain is simply pursuing a goal it has been given. The apparent ability of AIs to distinguish between their own painlike state and that of human users is “the thing that's to me potentially interesting,” Seth says. □

Science's AI in Science reporting initiative is supported by Ray Rothrock & family.

GEOLOGY

Oil wells could become mines for critical minerals

Engineered fluids could extract metals from low-grade ore in vast underground reservoirs tapped by fracking

PAUL VOOSEN

The mine of the future may look a lot like the oil well of the past. Researchers are exploring whether the vast network of oil and gas wells drilled across the United States can be repurposed to extract critical minerals from the rock formations below—elements the U.S. needs for solar cells and electric vehicles and, for now, has to import from China and other countries.

The idea builds on fracking: pumping huge volumes of fluid into underground rock layers to fracture them and bring oil, gas, and wastewater back to the surface. It's long been known that those fluids can pick up trace amounts of metals such as lithium and uranium along the way. Researchers at the Pacific Northwest National Laboratory (PNNL) and the University of Oklahoma (OU) now

want to make that process more selective, adding chemicals to the fluids to dissolve specific minerals from the rock. The team plans to escalate its efforts with 2 years of funding from the Department of Energy's Mine of the Future program, starting this month. This is “precision mining in the subsurface,” says Todd Schaef, a chief scientist at PNNL. “We have to go after these low-grade ores.”

The concentrations of critical minerals in individual oil and gas reservoirs are minute. But the sheer scale of the U.S. oil patch, with more than 900,000 active wellheads that tap a vast acreage, could compensate. Oil wells could produce up to 500 million tons of critical minerals, assuming mining of 5% of the reservoir volume and 1% recovery of the minerals, the group estimates in a paper published last month in Environmental Science & Technology.

"There is a lot of potential for this kind of coproduction," says Daniel Bain, an environmental scientist at the University of Pittsburgh. But he cautions that pushing fluids with exotic chemicals through fractured rock and retrieving metal-laden waste-water raises a familiar challenge for the industry—keeping potentially contaminated water out of drinking water formations and watersheds.

The study's resource estimate is more than a back-of-the-envelope guess, says Yuntian Teng, a geochemist at PNNL who led the work. To learn what's locked in the rock, the researchers analyzed core samples from 13 major oil-bearing formations. They bombarded each core with x-rays, causing elements in the rock to fluoresce at characteristic energies. From the signals, the researchers could estimate the abundance of 15 target elements. The results varied

Pumpjacks in California tap oil and gas liberated by fracking. They could one day extract valuable metals, too.

widely by formation. Parts of the Permian Basin, a major formation in Texas, were rich in vanadium, titanium, and nickel, whereas manganese popped out in the Uinta shale formation in Utah and Colorado. “There’s a huge potential,” Teng says.

Lithium is the low-hanging fruit for so-called in situ mining because wastewater readily dissolves it from rock. “Lithium, you can sneeze on it and it’ll get into solution,” Bain says. Hood Whitson, CEO of Element3, says his startup has already built a commercial plant in Midland, Texas, that adds aluminum hydroxide to 35,000 barrels a day of oil-and-gas wastewater. The aluminum hydroxide captures the dissolved lithium, which can later be released with hot water. Other companies are trying to extract lithium from brines at geothermal power plants before the water is pumped back underground.

Other metals—for example, titanium—are harder to extract. So Ellen Polites, a geochemist at PNNL, and her colleagues have been testing chemicals—organic acids, oxidizing agents, and other reactive molecules—that can coax specific minerals out of a rock. “What we’re trying to do is make it more targeted,” she says, “rather than bulk dissolving the entire rock.” The team plans to test these engineered fluids in a lab-based setup that simulates the rocks in a fractured reservoir. By the end of the 2-year project, they hope to inject the fluids into an actual well. “Industry has grabbed on to this concept,” says Son Dang, a petroleum engineer at OU.

Polites's team envisions flushing a reservoir with different engineered fluids in successive passes, each targeting a different mineral. The mining could happen alongside oil-and-gas production, or prospectors could target the $80\%$ of wellheads that are marginal and unproductive, Dang says. By making wastewater more valuable, the team hopes to encourage its proper treatment and reuse. "We have an incredible water crisis in the United States," Polites says.

If the world wants to limit more human-driven global warming, oil and gas drilling will need to slow. But in the meantime, taking advantage of fossil fuel production to extract minerals needed for clean technologies could be part of the solution, Whitson says. “We’re taking the waste of the fossil economy to support our future material needs.” □

ECOLOGY

Project will track Amazon biodiversity in real time

Leaders of \$100 million effort will use automated detection methods and computer models to monitor species' trends WARREN CORNWALL

Scientists can tell within days whether a grove of trees has been felled anywhere in the world, thanks to satellite imagery. But when it comes to knowing what's happening in real time to the species that live there, researchers remain woefully blind.

“We have good technology that can monitor forests from above,” says Emiliano Ramalho, an ecologist and scientific director for the Mamirauá Institute for Sustainable Development, a Brazilian nonprofit research organization working in the Amazon rainforest. “But this does not monitor what’s under the canopy.”

Now, thanks to technological advances such as automated species detection from audio recordings and analysis of DNA shed into the environment, conservation scientists think it's possible to assemble increasingly comprehensive and up-to-date views of biodiversity across huge landscapes. Last week, a coalition of research organizations, including Ramalho's institute, announced an initiative to test that promise by creating an Amazon-wide, regularly updated picture of what's happening to plant and animal populations as development threatens one of the most species-rich parts of the world.

The project, dubbed Amazon.ia, promises to marry streams of data from dozens of field sites with computer models capable of quickly detecting patterns and extending results across the river basin's 7 million square kilometers. The group includes

scientists from some 2 dozen universities, nongovernmental organizations, and companies in North America, South America, and Europe. Backed by an initial contribution of \$14 million from the Bezos Earth Fund, project leaders say they have raised approximately \$20 million of what would be a 10-year, \$100 million effort.

Tanya Berger-Wolf, a computational ecologist at Ohio State University who is not part of the project, says that in 2018 she sought funding from the U.S. National Science Foundation for a similar project focused on a single field site in Kenya. The proposal was rejected, she says, on the grounds that it was infeasible. “What’s changed in the last 8 years is this is within the grasp,” says Berger-Wolf, who has spent years designing computer systems to cull ecological information from photographs. “It’s becoming an engineering problem, which is a much better place to be.”

She points to several developments that make Amazon.ia more realistic: A flood of data has helped train more powerful computer models, technical know-how has become more widespread, and more people are concerned about the loss of biodiversity.

Mamirauá and the Polytechnic University of Catalonia have already built a network of 65 stations in the Amazon that automatically collect audio and images; classify species including insects, mammals, and birds; and quickly upload the results via satellite. The program has been shown to effectively track the Amazonian manatee and endangered river dolphins.

Ramalho says the institute joined Amazon.ia in part because the project aims to merge data from multiple locations and to deliver results fast enough to influence policy. The program also promises to build the technical capacity of Amazon-based research organizations by helping pay for things such as staff, equipment, and technical training. That would be a change from a historic pattern of research groups from places like the United States and Europe taking raw data and analyzing it themselves. “We don’t want to send all our data out. We want to be able to handle the data, manage the data, develop the models,” he says.

David Luther, a George Mason University ecologist involved in the Amazon.ia project, hopes the new initiative can connect isolated findings into broader trends across the Amazon.

He has been involved in research in Brazil that uncovered a worrying decline in some bird species. Several other sites in Central and South America have reported similar drops. But the historic data are so scattered that it's hard to know whether the declines reflect a regionwide shift. And there are no comparable data for insects, a food source for many of the birds.

Corine Vriesendorp, an ecologist and director of science for Conservación Amazónica, a Peruvian research organization whose station on the Los Amigos River is part of the new program, anticipates that the collective effort will mean “we’re going to be able to say something that none of us would be able to do on our own. ... This takes a ton of computing power, data storage, collaboration among scientists.”

The financial and technical challenges are daunting, however. In addition to needing about \$80 million more to reach its goal, the program has so far recruited just five of the 10 to 30 sites it wants for sustained long-term monitoring. Candidates include established field stations, as well as places such as ecolodges, community centers, or universities.

And it's not clear current technology can fill all the blanks in such a vast region. Justin Kitzes, a quantitative ecologist at the University of Pittsburgh, is leading a separate Bezos-funded project using audio recorders to monitor bird populations in Andean nature reserves. His team is also developing a box that could be dropped into a location to automatically collect biodiversity data. “This is something we are actually working on,” he says.

But he is cautious about the bigger effort. “I’m sure it will be greatly informative,” Kitzes says. “I’m also sure it won’t tell us everything we want to know.” □

---

Russia attacks Ukraine's science academy

A jet-powered drone slammed into the headquarters of the National Academy of Sciences of Ukraine on 28 September, killing at least two people and destroying part of the historic building. One person inside the building during the Russian attack was listed as missing and presumed dead as Science went to press. "It's a terrible tragedy for Ukrainian science," says Sergiy Mosyakin of the M.G. Kholodny Institute of Botany. As a response to earlier onslaughts

on Ukraine's scientific infrastructure, the science ministry this year started to construct underground laboratories. Deputy science minister Denys Kurbatov says he hopes international organizations will help Ukraine build more such labs and shelters, and supply backup electricity sources and reliable internet connectivity. “The enemy wants us to stop working,” he says. “That will not happen—we will not stop.”

—Richard Stone

NEWS

ASTRONOMY

Space telescope will reopen view of far-infrared sky

NASA's \$1.2 billion PRIMA will explore galaxy evolution and planetary birth

DANIEL CLERY

NASA has selected a far-infrared space telescope to explore the faint, cool glow of the gas and dust

that fuel star formation and coalesce into planets. The \$1.2 billion PRIMA is the agency's first Probe class mission, intended to fill a gap between researcher-led Explorer missions and multibillion-dollar flagship missions. Set to launch in 2033, PRIMA would also end a far-infrared observing gap since Europe's Herschel telescope ceased operation in 2013.

The far-infrared is one of the least explored parts of the electromagnetic spectrum. Most of the

radiation is blocked by Earth's atmosphere, forcing astronomers to go to space. They also have had to develop detector technology themselves, unlike in neighboring wavelengths—the midinfrared and microwave—where commercial and military applications are common. “The far-infrared, that’s just astronomers,” says Jason Glenn, an astrophysicist at the University of Virginia who led PRIMA’s design effort.

Herschel, the first major space telescope dedicated to the far-infrared, had a passively cooled 3.5-meter mirror. But stray heat from it limited

IN FOCUS

---

Did a ‘boneslide’ fill temple with remains?

For years, human bones found amid the rubble of southern Turkey's Göbekli Tepe—one of the earliest known monumental temple complexes—led some to suggest the 10,000-year-old site once hosted an ancient “skull cult.” A new analysis of intact graves from the site, reported last week in PLOS ONE, instead suggests a less lurid explanation: Landslides may have swept the human remains into the central ritual space. The two graves, excavated in 2017 and '22, contained the remains of a three adults and a teenager and were found beneath the houses built uphill from Göbekli Tepe's temples. Some combination of heavy storms and earthquakes could have periodically sent these perched houses, along with the skeletons buried beneath them, careening down the slope, jumbling them with temple ruins. —Andrew Curry

the telescope's sensitivity to distant sources. And its detectors were cooled with liquid helium, which ran out after less than 4 years.

PRIMA won't need helium and could operate for longer. It will combine passive cooling with a mechanical cryocooler to keep its 1.8-meter mirror at $4.5^{\circ}$ above absolute zero. An additional cryocooler will chill a new type of superconducting sensor, called a kinetic inductance detector (KID), to less than one-eighth of a degree above absolute zero.

KIDs can record the energy and arrival time of individual photons, and they offer roughly 100 times the sensitivity of Herschel's detectors. "That will enable PRIMA to observe the kinds of things that Herschel did in the local universe ... but throughout the history of cosmic time," says Cardiff University astrophysicist Matt Griffin, who is part of a U.K. consortium contributing hardware to PRIMA.

One major target is the evolution of galaxies. Herschel showed how interstellar gas moves within galaxies as it collapses into stars or is expelled into intergalactic space by winds from giant black holes. But its studies were largely limited to a handful of nearby galaxies. PRIMA will probe dozens of galaxies across three-quarters of cosmic history, allowing astronomers to track how those processes changed over time.

Within the Milky Way, PRIMA will examine how planets coalesce out of disks of gas and dust surrounding newborn stars. The Atacama Large Millimeter/submillimeter Array in Chile, which observes at slightly longer wavelengths, has revealed gaps in these disks—the signs of planets in the making. PRIMA will detect water in the disks, testing the idea that ice is the glue that helps dust clump together and drive planet formation, Glenn says.

Managed by NASA's Jet Propulsion Laboratory, PRIMA will now undergo detailed design and technology work ahead of a review, after which construction can begin. Probe class missions were recommended by the latest decadal survey in astrophysics, a community priority-setting exercise that sought missions that could launch on a faster schedule than flagships. PRIMA competed for the first Probe slot with the Advanced X-ray Imaging Satellite, which NASA canceled earlier this year.

BIOETHICS

Impending changes to U.S. human subject protections put ethicists on edge

HHS plan to revise the Common Rule focuses on easing regulatory burden for lower risk studies

SARA REARDON

The Trump administration appears poised to weaken protections for people enrolled in some clinical trials. In a little-noticed posting on a government website tracking agency regulations, the U.S. Department of Health and Human Services (HHS) announced an upcoming “de-regulatory” move that would revise the Common Rule, a federal policy that governs aspects of human research including informed consent, data sharing, and special protections for vulnerable participants such as prisoners or pregnant people.

The July posting contains few details but says the Common Rule's framework, last updated in 2017, “no longer reflects the realities of modern research” and “imposes unnecessary burdens and unclear requirements, particularly for minimal-risk studies.” HHS declined to answer questions about the timing of the forthcoming rule, which will require a public comment period before implementation.

“It sounds like they’re doing something serious, and they could go pretty far,” says Jerry Menikoff, a bioethicist at the National University of Singapore and former director of HHS’s Office for Human Research Protections (OHRP), which oversees the Common Rule.

First instituted in 1991, the rule applies to most federally funded research studies. Privately funded studies—and the institutions that carry them out—also generally adhere to its guidelines but are not required to. Researchers have long complained that some of its provisions require too much paperwork, such as the need to obtain approval from an ethics review board to make minor changes to a protocol.

The HHS posting about the new rule offers “examples of changes,” including added flexibility for minor protocol changes and “expanding exemptions for certain low-risk research activities." How those activities will be defined is unclear, however, because "low risk" has no legal definition. (The current rule does define "minimal risk," however, as the risk that can be expected in everyday life or in a routine medical exam.) Asked for details, an agency spokesperson said, "HHS cannot comment on potential or future regulatory actions."

Seema Shah, a bioethicist at Northwestern University, says more flexibility in some areas of the Common Rule would be a good thing. "There are nuggets in here that seem like they could be valuable," she says Yet she worries the notice overall appears to give more attention to easing regulatory burden than safeguarding human subjects. "There's definitely a lot of money and power behind the idea of trying to take away some of these standards and do research as quickly and cheaply as possible," Shah says. "Our past is littered with a history of unethical research where people were exposed to risk without knowledge or consent."

HHS's notice says current regulations are especially burdensome for learning health systems, a model in which hospitals integrate clinical research studies into their care practices without necessarily obtaining patients' consent or even informing them that a trial is happening. Proponents argue this approach, which does allow participants to opt out, could speed research by eliminating the lengthy process of recruiting people into trials. Because these hospital-integrated studies often compare the efficacy of two treatments that both fall within standard medical practice, Menikoff says HHS could conceivably determine that any such study constitutes minimal risk. "That would certainly make it a lot easier to do a lot of these trials," he says.

Yet even standard treatments can have different levels of risk. For instance, a 2013 trial known as

SUPPORT tested the optimum dose of oxygen for premature infants by randomly assigning them to receive high or low levels. Although both doses fell within standard medical practice, higher doses carried a possible risk of eye damage, and the lower end of the range was thought to increase the risk of brain death. OHRP determined parents had not been adequately informed about the risks. Menikoff says similar trade-offs emerge in studies comparing, for instance, the benefit of yearly cancer screenings with the risk of increased radiation from scanners. "People have different viewpoints" on which risks they're willing to take, he says.

The HHS notice also mentions that the agency will clarify its definition of “undue influence”—when a study pays people so much to participate in a trial that they may feel coerced. This problem has rarely come up in practice, Menikoff says, because it would take an enormous amount of money to make most people feel that they can’t decline.

Ellen Clayton, a bioethicist at Vanderbilt University, is watching to see whether the new rule will affect how trial participants must consent to sharing their data with other institutions, including the government and private companies, and how those data should be protected. The topic is not explicitly mentioned in the notice, but is a concern given the Trump administration's overall approach to research, Clayton says. "They're worried about [competition from] China, they're desperate to take research out of the research institutions and into the private sector, they may be allowing easier access by the private sector to data." But weakening the Common Rule could come at a price, she says. "Is the gain you're going to get from doing translational research without consent worth it for what it's going to cost in trust?"

Vaccine Janssen

COVID-19 vaccine

(Ad26.COV2-S)

(recombinant))

IM injection

5 doses of 0.5 mL

SHOTS ON GOAL

Can Africa make $60 \%$ of its vaccines by 2040? That’s a tall order, as the travails of South African companies show

JON COHEN, in Cape Town, South Africa

M

orena Makhoana can see Cape Town's spectacular landmark, Table Mountain, from the grounds of Biovac, the vaccine manufacturer he runs here. "When you're stressed and you don't know what to do, you just look there," Makhoana

says. “It keeps us a little bit sane.” He has just stepped out of a Biovac building constructed during the COVID-19 pandemic to hold the “freezer farm,” which today symbolizes one of the company’s—and one of the African continent’s—great disappointments.

When the first COVID-19 vaccines became available in 2021, deep-pocketed governments outside Africa snatched them up, pushing even the wealthiest African countries to the end of the line. Then Biovac cut a deal that held great promise: Pfizer and BioNTech would provide their popu-

lar messenger RNA (mRNA) vaccine in bulk, which Biovac would put into vials—a process known as “fill and finish.” The company aimed to start to supply vaccine by early 2022, and to produce 100 million doses a year, storing 20 million at a time at the farm of specialized freezers that maintained the required $-80^{\circ}$ C. The deal would give Africa its own dedicated vaccine supply.

But it took longer than expected to formalize the agreement and gear up for production, and once Biovac had its first doses available at the end of 2022, demand for the vaccine had evaporated. "We never sold a single dose," Makhoana says, and the freezer farm remained idle.

Biovac's travails mirrored those faced by other bids to speed delivery of COVID-19 vaccines to Africa. They also added to a decades long history of setbacks for efforts to establish a vaccine industry on the continent, which in 2021 produced just $0.1\%$ of the shots it used. But the pandemic, with its stark inequities in vaccine access, sparked a new resolve among African governments, health officials, and

Reporting for this story

scientists to change that.

was supported by the Pulitzer Center.

The result was an initiative called the Partnerships for African Vaccine Manufacturing (PAVM), launched in April 2021 and spear-

Aspen Pharmacare put a Johnson & Johnson COVID-19 vaccine in vials, but never delivered it in Africa.

headed by the African Union and the Africa Centres for Disease Control and Prevention (Africa CDC). PAVM aimed to help companies access financing and build expertise, and to get neighboring countries to pool resources, creating larger marketplaces for manufacturers in their regions. PAVM had an ambitious—some said unrealistic—goal: to drive up the proportion of African-made vaccines to a staggering 60% of demand by 2040.

More than 2 dozen companies are now vying to seize the postpandemic momentum and contribute to PAVM's vision. Media have intensively covered African presidents and global health leaders holding shovels at groundbreaking or ribbon-cutting ceremonies for new and expanding plants. But the experiences of Biovac and South Africa's two other vaccinemakers, the tiny, mRNA-focused Afrigen Biologics and Vaccines and the giant Aspen Pharmacare, show tremendous hurdles remain—even in a country that has the largest economy on the continent and one of the most developed infrastructures.

Vaccines are more complex to manufacture than chemical-based drugs, and plants are expensive to build and maintain. To compete in the market, companies typically have to produce vaccines on enormous scales. And to fit the budgets of developing countries, they must be inexpensive. “That’s why the global vaccine market is controlled by six or seven companies,” says Afri-gen’s CEO, Petro Terblanche.

African companies also face steep challenges securing loans from banks and investments from venture capitalists. Their cash-strapped governments are reluctant to make purchasing commitments unless companies show their products can compete with vaccines available from established manufacturers outside the continent. As the COVID-19 experience showed, demand can be fickle. And funding to test new vaccines is scarce. “We’re in a bloody tough time,” Makhoana says.

Yet simply relying on established producers outside of Africa isn't an option, he contends. "If it was a commodity like cars, you could argue about whether you want to make them locally or import them, and it's not a life-or-death discussion," he says. "We need vaccine manufacturing on

the continent as a health security issue. It's a no-brainer."

BIOVAC BEGAN in 2003 as a partnership between the South African government and privately held health care companies. It first aimed to make a liquid form of a freeze-dried vaccine against the bacterium Haemophilus influenzae, which causes respiratory infections and other conditions including meningitis. The company planned to include it in a vaccine produced by Sanofi in France that also protected against diphtheria, tetanus, pertussis, and polio, creating a single broad-spectrum shot for children. It set about restoring a dilapidated state-run vaccine plant as its headquarters. (Biovac made the liquid product, but Sanofi figured out how to manufacture it, too.)

Just then, global events underscored the need for developing countries to have “vaccine sovereignty.” A bird flu virus alarmed the world when it started to infect people in several Asian countries, killing 79 of the 148 confirmed cases by 2005. A viral sequence the Indonesian government shared with the World Health Organization (WHO) became the basis for an Australian vaccine company’s product, which Indonesia presumably would not be able to afford. Anger at rich countries for controlling essential vaccines intensified with revelations that the U.S. government and U.S. companies and research institutions had patented sequences of bird flu viruses from Indonesia and Vietnam. “It was a political dog’s breakfast,” says Martin Friede, who began working at WHO in 2003 and headed its vaccine research unit when he retired last year.

At the request of U.S. health officials, WHO launched a project to help several countries build manufacturing plants for flu vaccines. Friede, who led the effort, says it faced one problem after another. One was that flu vaccines were made by growing influenza viruses in fertilized chicken eggs, which manufacturers had to order from farmers at least 6 months in advance. If a bird flu pandemic had exploded, vaccine production would have lagged. Lesson No. 1, Friede says: "Don't use a technology where if a pandemic happens, you won't be able to get the raw material."

A second lesson involved economies of scale. A government-owned plant in Thailand was producing 15 million doses a year of flu vaccine. The vaccine cost \$6 per dose, but the country could buy a similar vaccine from Sanofi for half the cost. At first, the Thai government said, “Well, that doesn’t matter because this is national health security,” Friede remembers. But when the country’s leadership changed, the new health minister phoned. “He said, ‘Na-

tional health security? Baloney. We're going to buy from Sanofi because we're running out of money."

Biovac would learn that lesson firsthand following a major 2015 agreement with Pfizer to fill and finish a Streptococcus pneumoniae vaccine called Prevnar. The South African government initially agreed to buy the locally filled vials, but in 2023 decided to purchase Prevnar from the Serum Institute of India—the world's largest producer of vaccines—for about one-third of the price, forcing Biovac to lay off 85 of its 450 employees. The firm has stayed afloat thanks in part to a multiyear deal with Sanofi to fill and finish a six-in-one vaccine—it adds hepatitis B—at the Cape Town facility, which can vial 15 million doses a year.

Morena Makhoana has piloted Biovac through rough times, but the company has recently attracted substantial new investments.

Friede's third critical takeaway from

Made in Africa

African countries make less than 1% of the vaccines they use, and a massive push is underway to boost manufacturing on the continent. More than 2 dozen companies are vying to fill the gap, but only a few have finalized agreements to transfer vaccine technologies from major pharmaceutical companies abroad.

Company status: On hold

Under development

Facilities complete

Tech transfer complete

WHO's efforts is that vaccine plants cannot financially survive if they only exist for health security during pandemics. “Quite a few companies were able to get the technology up and running, and then a year or two later, we began getting calls from the chief executive officer saying, ‘Right, we’ve built this flu facility, but we’re not selling any flu vaccine,’” he says, noting it can cost millions of dollars a year to maintain an empty plant. Several pivoted to making routine childhood vaccines—taking themselves out of contention as flu vaccine suppliers in the event of a pandemic.

WHEN THE COVID-19 pandemic struck, health officials tried to learn from the past, but Biovac's effort to secure Africa's access to vaccines wasn't the only one to fall short. Aspen early on signed a deal with Johnson & Johnson to fill and finish a single-dose COVID-19 vaccine that could have protected 300 million people annually, but contamination of the bulk product at a U.S. plant and other problems delayed the effort, and Aspen never sold a dose. Similar snafus undermined would-be fill-and-finish efforts by companies in Ghana and Morocco.

Efforts launched from outside Africa fared no better. The most devastating and high-profile letdown was a campaign by the nonprofit COVID-19 Vaccines Global Access (COVAX) Facility to supply Africa with COVID-19 vaccines made elsewhere: It faltered because it relied too heavily on the Serum Institute, which could not export doses in the spring of 2021 when its own country had a surge in cases. Patrick Soon-Shiong, a biotech billionaire in Los Angeles who grew up in South Africa, in 2022 pledged to invest \$200 million in NantSA, a new vaccine company here that would produce 1 billion COVID-19 doses by 2025. After breaking ground on the plant, he shut it down, in part because of the country's infrastructure problems. "No power was available to that plant!" he wrote in a text to Science. Moderna announced it would invest up to \$500 million in a Nairobi, Kenya, mRNA vaccine plant that supposedly could have produced 500 million doses a year—a plan it abandoned 2 years later.

Africa's response was PAVM. In March 2022, it published a detailed plan to strengthen local regulatory agencies, create training centers for scientists and engineers, coordinate technology transfers and sharing of intellectual property, and organize the continent into regional purchasing pools.

The call to produce 60% of the continent's vaccine needs by 2040 “was an aspirational goal at that time, which was necessary to move things forward and to bring excitement to the space,” says John Nkengasong, who led Africa CDC when PAVM was formed and now works at the Mastercard Foundation guiding investments to strengthen education and health in Africa. Aiming high makes sense given Africa’s population growth, he says. “One out of four human beings will be African” by 2040, Nkengasong says. “Can you afford not to invest in your own development of your biomedical countermeasures?”

WHEN PAVM LAUNCHED, only one product, a yellow fever vaccine made by the Pasteur Institute of Dakar (IPD), was being produced in Africa from scratch, so-called end-to-end manufacturing. In PAVM's vision, the burgeoning African vaccine industry would initially consist mainly of companies focused on fill and finish, like Biovac. That type of arrangement gives countries some control over vaccine supply and brings revenue to local businesses that provide things such as glass vials and their rubber stoppers. The vaccine companies would, over time, “backward integrate” and learn how to produce all

the necessary ingredients themselves.

Seth Berkley, who led a public-private partnership called Gavi, the Vaccine Alliance for 12 years before stepping down in August 2023, says that when PAVM launched and attracted a flood of interest from would-be manufacturers, he warned that most wouldn't manage to create sustainable businesses, given the many challenges of the vaccine industry. He urged Africa CDC and WHO officials to concentrate their energy on the most promising firms, says Berkley, who also helped create and run COVAX. "At the end, it was kind of, 'Well, we'll let them go on, and we'll see who succeeds.'"

For the many contenders, new investment opportunities bloomed thanks to the organized, continentwide efforts of PAVM. Gavi, which each year purchases and distributes product for more than half the world's children, made a notable commitment the year after Berkley left. It pledged to invest up to \$1.2 billion over 10 years in a combination of financial rewards for African companies that hit milestones such as regulatory approvals and per-dose subsidies to lower the prices of approved vaccines. Gavi plans to make the first of the milestone payments later this year, though it hasn't revealed the recipients.

The Gavi pledge has made Biovac more attractive to investors, Makhoana says. This year, the company secured more than \$100 million in new funding from the European Investment Bank, the International Finance Corporation, and the African Development Bank Group to more than triple production capabilities to 50 million vials a year by 2029. And it hopes to soon bring in revenue from end-to-end manufacture of a cholera vaccine developed by South Korea's International Vaccine Institute. Still, the company has yet to earn substantial profits from vaccines.

AFRIGEN, JUST A 20-MINUTE drive away from Biovac, is a much smaller operation, occupying two rented warehouse buildings in a business park. It, too, has had a bumpy road.

The company has been struggling to keep the lights on since it opened its first lab in 2016. Launched 2 years earlier as a partnership between the South African government's Industrial Development Corporation and the Infectious Disease Research Institute (IDRI) in Seattle, Afrigen at first aimed to bring a tuberculosis vaccine to market. But financial woes forced IDRI to sell its $51\%$ interest in Afrigen to Avacare, which primarily manufactures and distributes generic drugs.

Under pressure from Avacare to make money, Afrigen took a left turn. It began to produce an emulsion of cannabidiol (CBD), a compound extracted from cannabis said to help with anxiety, inflammation, and pain. One of Afrigen's most popular prod-

Afrigen's Caryn Fenner ran a program that taught scientists from several countries how to make messenger RNA for vaccines with small machines like this one.

ucts was a CBD paste that cat owners could squirt onto climbing poles for their cats to lick. That venture petered out when the South African regulatory agency cracked down on the sale of CBD around the start of the pandemic.

The Pasteur Institute of Dakar recently received a \$45 million grant to train African vaccine scientists.

In April 2021, Afrigen bid to return to vaccines. It responded to a call by WHO and the Medicines Patent Pool to apply for a grant to set up a training hub for companies interested in making mRNA vaccines.

The mRNA technology allows a maker to quickly tailor a vaccine to a new pathogen or strain, and the product can then be made in bio-reactors rather than chicken eggs. During the COVID-19 pandemic, when the Moderna and Pfizer-BioNTech vaccines showcased the power of the technology, “Countries

[were] screaming blue murder, ‘Not fair, not fair, we want access!’ recalls Friede, who spearheaded the mRNA training hub idea.

Afrigen “hadn’t touched mRNA,” says Caryn Fenner, a bioprocessing engineer at the company, and it only had the shell of a pilot vaccine production plant. But the company, one of 20 applicants, won the job to train scientists from far and wide. “We were just a little bit gobsmacked,” Fenner says. At first, they joked they were building the airplane while flying it. “We changed it to: We’re all running down the runway, flapping our arms.”

Moderna, Pfizer, and BioNTech each came to visit Afrigen, but none wanted to partner with the fledging mRNA manufacturer. So the company teamed up with researchers at the University of the Witwatersrand, who showed them how to make mRNA and formulated a version of the Moderna COVID-19 vaccine based on a sequence of the viral surface protein that Stanford University researchers had published. Afrigen then hosted scientists from 15 low- and middle-income countries to make a mimic of the Moderna vaccine, and 13 came back to learn how to work with other pathogens.

The grant ends in December, and although the company now has the expertise and small plant to make mRNA vaccine products itself, it recently suffered yet another setback: Financial troubles at Avacare forced it into “business rescue” (a form of receivership) in February. Afrigen is now attempting to restructure itself as a nonprofit. “We’ve weathered two big storms already, this little company, and we’re in our third one,” Terblanche says. “This will be do or die.”

SOUTH AFRICA'S THIRD vaccine-maker, Aspen, is well aware of the perils of vaccine production. “African manufacturers have had countless, endless challenges with a sustainable demand for vaccine,” says Stavros Nicolaou, a pharmacist and executive at Aspen, which has a sprawling complex in Gqeberha (formerly Port Elizabeth). So the company, the largest pharmaceutical company in Africa, currently isn’t planning to do any end-to-end manufacturing. It does fill and finish of four vaccines that it imports from the Serum Institute: a six-way vaccine similar to the Sanofi-Biovac product and ones that protect against S. pneumoniae, rotavirus, and meningococcal bacteria.

If Aspen were to make the vaccines it now imports, Nicolaou expects the Serum Institute would simply undercut it. “There are times where I’ve been a naïve optimist around this whole localization thing,” Nicolaou says, “but that was many years ago.”

Makhoana, in contrast, still exudes abundant optimism about local, end-to-end production of vaccines. His vision is that South African vaccine producers will build and share talent, and the government will invest more in the education of scientists. “I do wish that Aspen thrives. I do wish that Afrigen survives. If there's any competition, we'll just be competing commercially. But we need the ecosystem."

To further improve that ecosystem, the Mastercard Foundation in 2023 awarded \$45 million to IPD to train young African scientists to produce vaccines. Nkengasong says he wants to see similar investments by African countries. “You cannot talk about African health sovereignty for vaccines without investing in universities,” he says. “Otherwise, you still wait on an American, a Canadian, a European, or Chinese young scientist to invent something, and then share the patent with you. That’s not sovereignty.”

In another potential boost for African vaccine manufacturing, China last month reached an agreement with the 16-country Southern African Development Community to bolster pharmaceutical development on the continent. Although the details are vague, China promises to increase technology transfers and training of workers.

Africa CDC is continuing its attempt to design a vaccine marketplace that can sustain companies. It is pushing for an Africa Pooled Procurement Mechanism, in which governments from five different regions would band together and purchase vaccines collectively. All 55 African Union member states signed on to this commitment in February. “The question is going to be how reliable will the payments be?” Berkley says. “It’s a little naïve to just say we’re going to do it and everybody’s going to fall in line.”

A June report issued by the Clinton Health Access Initiative interviewed the three manufacturers in South Africa and 24 others to assess the state of vaccine production on the continent. “What we found is a sector making genuine progress and running into serious headwinds,” its report concluded. Biovac and IPD are the only two African companies that will likely do end-to-end manufacturing of vaccines in the next 5 years, it predicts. In the same time frame, Aspen and Egypt’s Vacsera are the only two that will receive bulk product made elsewhere and do fill and finish.

Makhoana says he still applauds PAVM's ambitious 60% goal. "If 2040 comes and let's say we are even at 11% manufacturing vaccines, for me, that's a 1000-fold from where we were," he says. "You must have vision and a target in life, even if it's a stretch. You must dream."

BIOMATERIALS

Self-assembling polymers work up some nerve

Micrometer-long peptide amphiphile fibrils enhance neuronal adhesion and function

Sanjay Kumar

Physical structure dictates the processes that underlie life. At the molecular scale, geometrically matching shapes allow a protein's binding pocket and a specific ligand to fit together, triggering important biochemical events. At the cellular scale, cylinder-like axons enable rapid transport of mate-

rials and electrical signals inside a neuron (1). An ongoing goal in biomaterials science has been synthetically creating biomolecules that can interface with cells and support specialized cell structures, such as the long, thin shapes of neurons. On page 125 of this issue, Dore et al. (2) report a family of peptide amphiphiles with discrete hydrophilic and lipophilic regions that assemble into fibrils with dimensions comparable to those of a cell (10 to $100\mu \mathrm{m}$ ). These structures allow neurons to extend axon-like processes along the fibrils and assemble into multicellular networks. The findings warrant fur

ther exploration of peptide amphiphiles and related materials for cellular engineering and regenerative medicine.

Many biocompatible materials that promote biological functions, such as cell adhesion, division, and differentiation, have been developed. These biomaterials often feature synthetic or tissue-derived polymers that are then decorated with short peptides. The long polymer chains dictate structural properties, whereas the peptides encode key biological functions, such as engagement of adhesion receptors or growth factors (3). Although this approach has successfully produced two-dimensional (2D) and 3D scaffolds for growing cells in culture, the polymer network

must be chemically synthesized in a manner that often “locks” the structure and the mechanics of the material. This could limit the dynamic cell-material interactions that are key to cell maturation and tissue assembly. Thus, molecular building blocks that can self-assemble into desired structures and dynamically instruct cells are needed. To this end, a variety of noncovalently self-assembling peptides and proteins have been developed, many of which spontaneously form hydrogels that support cell engraftment and growth in the body (4, 5). However, organizing these molecules into cylindrical structures, such as fibers and ribbons, remains challenging. Creating such materials could both mimic the natural contours of ...peptide amphiphile formulations could be designed to deliver specific physiochemical instructions to different cell types...

biological matrices and facilitate complex cellular morphogenetic events, such as the extension of axons and dendrites in a nerve cell.

Dore et al. designed and synthesized a family of peptide amphiphiles that can self-assemble into long fibrils that are tens of micrometers in length and billions of daltons in molecular weight. In general, the hydrophobic lipid tails of peptide amphiphiles drive initial aggregation in aqueous solution that brings the peptide segments into proximity, where these sequences can adopt a secondary structure (6) that nucleates growth of filamentous or membranous structures. Peptide-based molecules can also include sequences that can interact with cells to support adhesion and other behaviors (7). Dore et al. used a peptide amphiphile with a 16-carbon fatty acid chain at one end and a peptide sequence and hydrophilic poly(ethylene glycol) (PEG) chain at the other end (see the figure). The peptide includes a valine-valine-alanine-alanine (VVAA) sequence that promotes assembly into $\beta$ sheet-based fibrils and arginine-glycine-glycine (RGG) repeats that create positive charge for electrostatic adhesion to negatively charged cell membranes. These amphiphiles produced micrometer-scale fibrils of uniform length distribution attributed to a PEG “cap” at the fibril ends that sterically limits the addition of new monomers and prevents fibril fusion. By mixing amphiphiles of different peptide sequences, Dore et al. could vary the length of the negatively charged and positively charged segments within the polymer fibril.

Dore et al. covered a surface with the poly-

mer fibrils to control the adhesion of neurons isolated directly from the mouse cerebral cortex. Adhesion could be adjusted by tuning the length and distribution of positively charged segments within the fibril. The positive segments bound the negatively charged surface of neurons. By contrast, the negatively charged segments were more mobile, facilitating adsorption of extracellular matrix (ECM) proteins that were added to the culture system and guiding the extension of neurites from the cells. The length of the negatively charged segment of a fibril dictated the extent of neurite extension, with shorter lengths promoting the behavior to a greater degree than a conventional experimental surface coated with full-length

---

Peptide fibrils interact with neurons

Peptide amphiphiles self-assemble into long, thin fibril structures. The lipophilic 16-carbon fatty acid tails bring peptide building blocks in proximity in an aqueous solution, and the hydrophilic poly(ethylene glycol) (PEG) chain at the other end acts as a transient cap to control the length of the fibril. Peptide segments can be modified to promote a specific arrangement or an electric charge.

ECM proteins, such as laminin or fibronectin. The fibril-coated surfaces also supported characteristic neuronal behaviors, such as the formation of synapses between adjacent neurons and the propagation of calcium signals within individual cells. Culturing neural progenitor cells (stem cells that can differentiate into neurons and other brain cell types) on the fibril-coated surface not only promoted their differentiation into neurons but also aligned extended neurites along the fibrils. The differentiated neurons also formed multicellular 3D clusters that were interconnected by membranous projections, suggesting the beginning of a cellular network. Because neural networks represent a fundamental functional unit of brain tissue, the study of Dore et al. is important for potential applications in tissue regeneration, biological computing, and biosensing. Subtle adjustments in the amphiphile structure (e.g., relative lengths of positively and negatively charged peptides) produced a broad set of neuronal behaviors, including cell death, cell adhesion with limited neurite extension, and the formation of multicellular networks.

The strategy presented by Dore et al. offers a new way of engineering the assembly of microscale structures approaching the sizes of cellular components with bioactive building blocks. Notably, the interaction between polymer fibrils and neurons produced radically different cell phenotypes by changing the sequence and electric charge of peptide moieties. In addition, the self-capping mechanism using PEG offers a powerful route to producing fibrils of desired lengths. This tunability may facilitate the patterning of specific cellular processes and behaviors (such as differentiation). For example, ECM-coated fibrils of tunable lengths could be spatially patterned to precisely guide neurite extension and synapse location (8). These fibrils could also be integrated into additive manufacturing (a process that creates 3D objects by adding materials layer by layer) to produce 3D neuronal constructs (9). Further work will be needed to integrate the approach with advanced technologies and assess their value for neural patterning and network generation.

The study of Dore et al. also raises intriguing possibilities for engineering new cell culture scaffolds. A peptide amphiphile scaffold that features peptides derived from growth factors or cytokines could promote the proliferation and differentiation of specific cell types, such as immune cells, which could benefit tissue implants. Beyond fibril structure, the peptide amphiphiles may be designed to spontaneously assemble into higher-order organizational structures found in tissues, such as fiber bundles and sheets. Additionally, the formation of 3D cellular complexes on polymer fibril surfaces hints at the potential of technology for creating 3D cell culture constructs in which cells are embedded within a synthetic matrix. The 3D arrangements may offer new ways of guiding and controlling cell behavior because cells often transduce ECM-based cues differently in a 3D topology compared with on a 2D surface (10, 11).

Beyond modulating cell behaviors, the self-assembling polymer fibrils may be useful for direct delivery of cells to the body for regenerative medicine applications. Peptide amphiphile-based therapies have shown promising results in preclinical models for nerve regeneration in spinal cord injury, where the microscale dynamics of the amphiphiles strongly in-

fluenced neuronal survival and functional recovery (12, 13). Incorporation of peptides that encode growth signals could also boost the low cell survival and engraftment rates that plague current stem cell transplantation approaches (14). Further, peptide amphiphile formulations could be designed to deliver specific physiochemical instructions to different cell types within a single material construct, such as directing neurons to assemble into neuronal tracts and endothelial cells to form vascular networks. Further work is needed to reach these ambitious goals, but Dore et al. have provided an important starting point. □

---

REFERENCES AND NOTES

1. E. A. Ding, S. Kumar, Mol. Biol. Cell 35, rel (2024).

2. M. D. Dore et al., Science 394, 125 (2026).

3. J. Lou, D. J. Mooney, Nat. Rev. Chem. 6, 726 (2022).

4. Y. Liu, A. E. Gilchrist, S. C. Heilshorn, Adv. Mater. 36, 2407794 (2024).

5. A. Hartley, P. M. Williams, A. Mata, Adv. Healthcare Mater. 14, 2402385 (2025).

6. H. Cui, M. J. Webber, S. I. Stupp, Biopolymers 94, 1 (2010).

7. W. Sun, D.A. Gregory, X. Zhao, Adv. Colloid Interface Sci. 314, 102866 (2023).

8. C. Villard, Semin. Cell Dev. Biol. 140, 54 (2023).

9. Q. Gu et al., Adv. Healthcare Mater. 5, 1429 (2016).

10. J. Baek et al., Sci. Adv. 8, eabm4646 (2022).

11. E. Qiao et al., Sci. Adv. 10, eadk8232 (2024).

12. Z. Álvarez et al., Science 374, 848 (2021).

13. N. Takata et al., Nat. Biomed. Eng. 10.1038/s41551-025-01606-2 (2026).

14. M. M. Adil et al., Biomaterials 136, 1 (2017).

---

ACKNOWLEDGMENTS

The author acknowledges the support of the National Institutes of Health (R35GM163735 and R01NS074831).

10.1126/science.ael4456

Departments of Bioengineering and Chemical & Biomolecular Engineering, California Institute for Quantitative Biosciences, University of California Berkeley, Berkeley, CA, USA. Email: skumar@berkeley.edu

A mitochondrial carrier turns on the heat

An orphan protein integrates circadian, dietary, and environmental signals to control brown fat metabolism

Martin van der Laan

As homeothermic organisms, mammals occasionally must spend metabolic energy to maintain body temperature but try to avoid its wasteful expenditure. Brown adipose tissue (BAT) plays a key role in thermogenic metabolism (1). At normal temperature, thermogenesis is switched off,

and BAT lipid turnover is repressed by the circadian clock, an internal mechanism that regulates daily physiological processes, including metabolism. Upon exposure to cold, some metabolic energy derived from oxidative respiration in mitochondria is diverted to heat production (2). It has been unclear which molecular mechanisms govern the balance between the circadian clock, BAT lipid metabolism, and thermogenesis. On page 79 of this issue, Karavaeva et al. (3) report the unexpected finding that an orphan mitochondrial transport protein is a critical integrator of circadian, dietary, and environmental signals in BAT. Its expression in brown fat may be helpful as a metabolic health indicator.

Cold exposure activates the sympathetic nervous system to release norepinephrine, a neurotransmitter that stimulates $\beta$ -adrenergic receptors in brown fat cells (adipocytes). This increases the expression of UCP-1 (uncoupling protein 1) (4–6), which inserts into the inner mitochondrial membrane and acts as a proton channel. UCP-1 allows protons to flow away from the electrochemical proton gradient that powers adenosine triphosphate (ATP) production. As a result, ATP synthesis is reduced, but adipocytes keep burning through their fat and sugar supplies. In this way, a part of the metabolic energy escapes directly as heat.

Karavaeva et al. performed sophisticated genetic, gene expression, and protein abundance screens to search for regulators of thermogenic metabolism in mouse BAT that may coordinate circadian and cold-regulated signaling pathways. Besides the known UCP-1, an uncharacterized mitochondrial membrane protein called SLC25A34 (solute carrier family 25 member 34) caught their attention. At normal temperature, expression of the corresponding gene was blocked by the circadian repressor of gene transcription, REV-ERBα (reverse erythroblastosis virus-α). This block was lifted by cold. If the diet contained sufficient lipids, then SLC25A34 expression was increased by the transcription factors PPARα and -γ (peroxisome proliferator activated receptor-α and -γ). Depletion of SLC25A34 in cultured brown adipocytes inhibited adrenergic receptor-stimulated mitochondrial respiratory activity, suggesting that the protein is required for BAT thermogenic metabolism.

SLC25A34 has been classified as an orphan carrier protein of mitochondria because its native substrates and transport mechanism have remained unknown. How could this metabolite carrier represent a bottleneck in a thermogenic metabolic shift? To understand this question, it is important to note that BAT shows an intriguing feature upon cold exposure: Both lipid synthesis and lipid oxidation are enhanced at the same time. In other tissues, such futile cycles of degradation and resynthesis are prevented by a stringent regulatory mechanism. In BAT, however, this general boost of lipid metabolism—anabolic and catabolic together—contributes to thermogenesis and helps to maintain body temperature (7). Through metabolic profiling and flux analysis of untreated and SLC25A34-depleted brown adipocytes, Karavaeva et al. ob

---

A hub of metabolic cycles in brown fat cells

(Top) The mitochondrial transport protein SLC25A34 integrates circadian, environmental, and dietary signals to control heat production (thermogenesis) in brown fat cells. (Bottom) Specific transcriptional regulators adjust SLC25A34 expression in the mitochondrial inner membrane. As a central node, SLC25A34 likely transports OAA into mitochondria. This drives three metabolic cycles: the OAA cycle, the TCA cycle, and the lipid cycle. | Circadian clock | CUE | SIGNALING | OUTCOME | Environment | Diet | Transcriptional regulator | SLC25A34 expression | Heat production | Awake | or | Cold | or | High fat | $\rightarrow$ | PPAR-α/γ | $\rightarrow$ | More | Asleep | or | Normal | or | Normal | $\rightarrow$ | REV-ERBα | $\rightarrow$ | Less |

Acetyl-CoA, acetyl-coenzyme A; Cit, citrate; NAA, N-acetyl-aspartate; OAA, oxaloacetate; PPARα and -γ, peroxisome proliferator activated receptor-α and -γ; REV-ERBα, reverse erythroblastosis virus alpha; SLC25A34, solute carrier family 25 member 34; TCA, tricarboxylic acid

served two major changes: accumulation of N-acetyl-aspartate inside mitochondria and oxaloacetate outside of the organelle, in the cytosol. Both metabolites are critical for exporting excess acetyl-CoA (acetyl-coenzyme A), a key molecule of cellular energy metabolism, from mitochondria to the cytosol, where it is a substrate for fatty acid synthesis.

In brown adipocytes, there are two pathways that supply acetyl-CoA to the cytosol (8, 9). N-acetyl-aspartate is synthesized in mitochondria from oxaloacetate-derived aspartate and acetyl-CoA and is exported by an unknown mechanism into the cytosol. There, the molecule is hydrolyzed, and acetate is converted into acetyl-CoA. Alternatively, citrate is exported from mitochondria and cleaved in the cytosol to produce oxaloacetate and acetyl-CoA (8, 9). Thus, no matter how the activated acetyl moieties are leaving mitochondria, they are always accompanied by an oxaloacetate molecule that needs to be transported back to sustain the shuttle. Specifically, in cold-stimulated BAT, where both anabolic and catabolic lipid metabolism are enhanced, accumulation of oxaloacetate in the cytosol is detrimental. Oxaloacetate is not only lacking in the mitochondrial matrix to keep oxidative respiration going and oxidize acetyl-CoA, it is also in short supply to pick up more acetyl-CoA for export to the cytosol to support lipid synthesis. The notion that in SLC25A34-deficient brown adipocytes the amounts of cytosolic and nuclear acetyl-CoA become limiting is further underscored by a decrease in the acetylation of histone proteins that support the expression of genes that encode enzymes of the oxidative respiratory machinery (9). Karevaeva et al. also noted that oxaloacetate limitation is reflected by increased rates of glucose-dependent enzymatic reactions that replenish the mitochondrial citric acid cycle, the chemical reactions that release energy stored in nutrients through acetyl-CoA oxidation.

Moreover, there is the conundrum that, as most textbooks state, there is no oxaloacetate uptake into mitochondria in mammals (10, 11). Rather, this metabolite is thought to be reduced to malate in a reaction that consumes cytosolic NADH (nicotinamide adenine dinucleotide hydride). Malate is then transported back to mitochondria to regenerate oxaloacetate, a process known as the malate shuttle. However, oxaloacetate transport across the inner mitochondrial membrane has been reported for plants and fungi (12). Notably, the closest deorphanized homolog of SLC25A34 is the yeast carrier Oac1 (oxaloacetate carrier 1) (13). Indeed, Karaeva et al. predicted, with molecular modeling, high-affinity binding between oxaloacetate and human SLC25A34.

Altogether, the dogma of oxaloacetate not being imported across the inner mitochondrial membrane in mammals is seriously shaken by the study of Karavaeva et al. One could speculate that SLC25A34 operates as an antiporter, like Oac1, and that its role in thermogenesis is the exchange of oxaloacetate and N-acetyl-aspartate between the mitochondria and cytosol. Reconstitution of the transporter into proteoliposomes will be required to directly test this hypothesis.

The study by Karavaeva et al. is a reminder of something more general about metabolic regulation and shifts in the flow of carbon fuel through different pathways. Rather than intricate regulatory circuits tuning the activity and abundance of many metabolic enzymes, homeostatic issues may easily be solved by altering the availability of substrates in a certain cellular compartment through the shuttling rate of the critical molecules (14). Such knowledge might well help in the treatments of common human diseases related to metabolic imbalance. Small molecules acting on defined metabolite carriers may help to compensate global homeostatic disturbances. Specifically, SLC25A34 may turn out to be a target for innovative new drugs for the treatment of obesity.

---

REFERENCES AND NOTES

1. P. Cohen, S. Kajimura, Nat. Rev. Mol. Cell Biol. 22, 393 (2021).

2. S. Demine, P. Renard, T. Arnould, Cells 8, 795 (2019).

3. I. Karavaeva et al., Science 394, eadz4797 (2026).

4. E. D. Buhr, S. H. Yoo, J. S. Takahashi, Science 330, 379 (2010).

5. D. Neri et al., Nat. Metab. 8, 313 (2026).

6. X.Zhang et al., Int. J. Mol. Sci. 26, 3233 (2025).

7. Z. Wang, Q. A. Wang, Y. Liu, L. Jiang, FEBS J. 288, 3647 (2021).

8. A. R. Pessentheiner et al., J. Biol. Chem. 288, 36040 (2013).

9. K. E. Wellen et al., Science 324, 1076 (2009).

10. P. C. Heinrich et al., Eds., Löffler/Petrides Biochemie und Pathobiochemie (Springer, 2022), chap. 10, p. 257.

11. J. Rassow et al., Eds., Duale Reihe Biochemie (Thieme, 2022), chap. 5, p. 253.

12. I. Hanning et al., Plant Physiol. 119, 1025 (1999).

13. L. Palmieri et al., J. Biol. Chem. 274, 22184 (1999).

14. A. Khan et al., Cell 188, 869 (2025).

10.1126/science.ael7271

RHEUMATOLOGY

Osteoarthritis as a vascular disease

Hypertension worsens osteoarthritis and could be a therapeutic target in patients with both conditions Chunyi Wen and Man Ting Au

Osteoarthritis is no longer seen as a simple wear and tear of the cartilage that cushions bone joints during movement. It is now recognized as a whole-joint disorder, characterized by cartilage degradation, alterations of the bone under the

cartilage, and low-grade inflammation of the connective tissue that lines the joint (synovium). However, no drugs that target these localized tissues have been approved for clinical use thus far. Therefore, disease management has remained largely unchanged since the 1990s and is limited to symptom relief with nonsteroidal anti-inflammatory drugs and injections of hyaluronic acid or corticosteroids into joints (1). The view of osteoarthritis as a localized disease overlooks a clinical reality: Although it can result from mechanical injury, a substantial proportion of disease cases are driven by metabolic dysfunction. On page 80 of this issue, Kim et al. (2) report a “brain-joint axis” that links osteoarthritis to hypertension, a component of the metabolic syndrome.

Obesity and the metabolic syndrome—a cluster of health conditions that includes high blood pressure, excessive belly fat, and elevated sugar and cholesterol concentrations in blood—are established risk factors for osteoarthritis. Therefore, metabolic therapies such as incretin mimetics (including glucagon-like peptide-1 receptor agonists) could potentially change the treatment of this disease (3). Incretins are of particular interest because of their whole-body metabolic benefits, which are independent of weight loss. Notably, these drugs rely on interorgan communication. In addition to the well-documented gut-joint signaling axis, communication between the liver and joints and between the lungs and joints may also influence the progression of osteoarthritis (4–6). The brain-joint axis identified by Kim et al. adds to the concept of osteoarthritis as a systemic disease.

Kim et al. found that arginine vasopressin (AVP), a peptide secreted by hypothalamic neurons in the brain, uses the same receptor to regulate the function of blood vessels and cartilage in a mouse model of hypertension. They discovered that circulating AVP, which is known to bind to AVP receptor 1A (AVPR1A) on vascular smooth muscle cells to drive the constriction of blood vessels, also acts on cartilage cells in osteoarthritic joints. High expression of AVPR1A by these cells triggered the breakdown of complex molecules (catabolism) and dampened the synthesis of new ones in the joint area. Furthermore, genetic deletion or use of drugs that block AVPR1A prevented progression of hypertension-driven osteoarthritis. These findings show that vascular disease and cartilage degeneration, two highly prevalent age-related conditions, can share a mechanistic origin. They also raise the possibility that systemic antihypertensive drugs might be effective disease-modifying treatments for osteoarthritis.

This repurposing strategy is supported by a study that used UK Biobank data to investigate the associations between use of antihypertensive drugs and risk of osteoarthria

---

A whole-body view of osteoarthritis

Osteoarthritis is not exclusively caused by localized inflammation in bone joints that leads to cartilage damage. It is also driven by metabolic dysfunction across the body. Both vascular and neural signaling pathways can affect joint health. The hormones endothelin-1 and arginine vasopressin, which is produced by the hypothalamus and binds to AVP receptor 1A (AVPR1A), have a dual role: They increase blood pressure, which can lead to hypertension, and also joint inflammation, which promotes osteoarthritis.

tis (7). Potassium-sparing diuretics and aldosterone antagonists (such as spironolactone) significantly correlated with a reduced risk of osteoarthritis. The AVP-driven brain-joint axis identified by Kim et al. offers a plausible explanation for this protective effect: Because these diuretics inhibit sodium reabsorption in the kidney's late-distal tubules, they correct fluid overload, thereby reducing the molecular signals that trigger AVP release from the hypothalamus. Notably, previous small-scale trials of spironolactone in patients with knee osteoarthritis have had mixed results. Therefore, larger-scale clinical investigations are warranted to explore the potential benefits of these drugs in this patient population.

The renin-angiotensin-aldosterone system (RAAS) and endothelin-1-driven signaling, which are hormonal networks that control vascular tone, have also been implicated in hypertension-associated osteoarthritis. For example, the RAAS-targeting drug captopril counteracted cartilage loss in hypertensive rats $(8)$ . However, Kim et al. observed that directly blocking RAAS function using losartan did not protect cartilage in their model. This discrepancy suggests that the RAAS might not be the primary mediator of joint degeneration and that captopril probably exerts its protective effects through alternative, RAAS-independent pathways. Endothelin-1 shares potent vasoconstrictor and proinflammatory properties with AVP and triggered joint pain and cartilage breakdown in a mouse model of posttraumatic osteoarthritis $(9)$ . Kim et al. also observed that endothelin-1 expression is markedly up-regulated in osteoarthritic chondrocytes, which suggests that it may act as a secondary or synergistic disease-promoting factor alongside AVP.

In addition to the AVP-AVPR1A pathway, the brain-joint axis is also modulated by the autonomic nervous system, which controls vascular tone and joint homeostasis through a delicate balance between sympathetic (adrenergic) and parasympathetic (cholinergic) signaling $(10, 11)$ . Corticotropin-releasing neurons in the hypothalamus project directly to the joint's synovium to communicate with local cells, and dampening this sympathetic activity improves joint health $(12)$ . The parasympathetic system counterbalances these effects. Stimulation of the vagus nerve, which connects the brain to the intestine, activates cholinergic anti-inflammatory pathways that help to normalize systemic and local vascular tone $(13)$ . This counteracts hypertension-induced hypoxia (low oxygen) in the joint and attenuates synovial inflammation.

Kim et al.'s findings reinforce the idea that hypertension is a key driver of osteoarthritis (14). The vasoconstriction that characterizes hypertension may indirectly induce hypoxia and thus inflammation in local joint tissues and aggravate cartilage damage (15). Furthermore, compounding the effects of sympathetic overactivation, vasoconstrictors such as AVP and endothelin-1 may also act directly on osteoarthritic chondrocytes to trigger catabolic responses (see the figure).

Recognizing osteoarthritis as a vascular disease rather than a simple mechanical problem opens avenues for therapeutic discovery. Repurposing antihypertensive drugs to treat osteoarthritis should be prioritized for patients who also have hypertension. Mapping the complex interorgan networks that affect joint health will usher in an era of precision rheumatology that combines local biomechanical care with systemic therapies to restore vascular health and joint integrity. □

---

REFERENCES AND NOTES

1. S. Tang et al., Nat. Rev. Dis. Primers 11, 10 (2025).

2. S.-J. Kim et al., Science 394, 80 (2026).

3. H. Bliddal et al., N. Engl. J. Med. 391, 1573 (2024).

4. J. Yi et al., Signal Transduct. Target. Ther. 9, 109 (2024).

5. M.T.Auet al., Nat. Microbiol. 9, 2538 (2024).

6. J. G. Niet al., Cell Rep. 45, 117841 (2026).

7. Y.Zhang et al., Aging Clin. Exp. Res. 35, 2437 (2023).

8. Y.Zhang et al., J. Am. Heart Assoc. 15, e042132 (2026).

9. M. Au, Z. Liu, L. Rong, Y. Zheng, C. Wen, Osteoarthritis Cartilage 28, 1559 (2020).

10. Z. Jenei-Lanzl, S. Grässel, H.-G. Schaible, R. H. Straub, Osteoarthritis Cartilage 34, 511 (2026).

11. M. Lauwers, A. Courties, J. Sellam, C. Wen, Osteoarthritis Cartilage 29, 643 (2021).

12. K. Huo et al., Ann. Rheum. Dis. 85, 743 (2026).

13. S. Yadav et al., JBMR Plus 10, zieg078 (2026).

14. K. Ching, X. Houard, F. Berenbaum, C. Wen, Nat. Rev. Rheumatol. 17, 533 (2021).

15. W. Gong et al., Aging Cell 25, e70606 (2026).

---

ACKNOWLEDGMENTS

The authors acknowledge financial support from the Hong Kong Research Grant Council (General Research Grant 15100324 and Collaborative Research Grant C5117-25GF) and the French National Research Agency (Agence Nationale de la Recherche) and Research Grant Council Joint Research Scheme (grant A-PolyU503/24).

10.1126/sciece.ael3784

Department of Biomedical Engineering, The Hong Kong Polytechnic University, Hong Kong, China. Email: chunyi.wen@polyu.edu.hk

DEVELOPMENT

A tumor suppressor shuts down a signal gradient

p53 shuts down the spatial signaling of a morphogen and prevents clonal cell expansion

Dan Engelman $^{1}$ and Cédric Blanpain $^{1,2}$

The most frequently mutated tumor suppressor gene in human cancers is TP53 (1). Loss-of-function mutations lead to expansion of a mutated cell and its progeny in epithelial tissues, including sun-exposed regions of the skin, which can progress to invasive cancers. Upon stress, such as DNA

damage, p53 becomes phosphorylated or acetylated, modifications that promote its stabilization and the activation of genes that cause cell cycle arrest, cellular senescence, or cell death (1). One might expect that clonal expansion triggered by the loss of p53 results from dysregulation of these canonical functions. On page 78 of this issue, Gan et al. (2) report that this expansion is instead driven by activation of a gradient of Wnt (wingless-related integration site) signals that promotes epidermal stem cell self-renewal. Controlling signaling gradients, rather than the absolute amount of a signal, may be a mechanism through which cancer-associated mutations achieve long-term tissue colonization.

Despite extensive knowledge about p53 functions, the molecular mechanisms by which it suppresses tumorigenesis remain unresolved (3). In response to stress, p53 is acetylated at multiple lysine residues, and mutations at the three acetylation sites in the DNA-binding domain prevent it from mediating cell cycle arrest, programmed cell death (apoptosis), and senescence. This mutation in mice ( $Trp53^{3KR/3KR}$ ; Trp53 is the mouse homolog of human TP53) also prevents tumor formation because mice lacking Trp53 ( $Trp53^{-/-}$ ), but not $Trp53^{3KR/3KR}$ , develop early-onset spontaneous lymphoma (4). This raises the possibility that tumor suppression does not depend on p53 canonical functions.

During tissue homeostasis, the balance between cell self-renewal and cell differentiation maintains a constant cell number (5). Oncogenic mutations can impair this balance, leading to the expansion of a mutated cell into a clonal population as the result of excess stem cell divisions and a failure to differentiate. For example, during basal cell carcinoma initiation, cells harboring a mutation in Smo (which encodes the Smoothened protein) exhibit enhanced self-renewing divisions and a block of differentiation, leading to the establishment of preneoplastic lesions (6). Although Trp53 deletion or $Trp53^{R248W}$ expression (a dominant-negative mutation that inactivates p53 protein) (7) do not increase the rate of mouse epidermal stem cell proliferation, they promote self-renewal, giving rise to more stem cells (without causing cells to divide faster or more frequently). Gan et al. did not observe this in mouse epidermal $Trp53^{3KR/3KR}$ clones. This suggests that $Trp53^{-/-}$ -mediated clonal expansion is likely the consequence of an imbalance toward self-renewing cell divisions rather than uncontrolled proliferation, through mechanisms independent of the loss of p53's canonical functions.

To decipher the molecular mechanism by which p53 controls epidermal cell self-renewal, Gan et al. hypothesized that genes controlling self-renewal would be shared between wild-type p53 (Trp53 $^{WT}$ ) and Trp53 $^{3KR}$ . Upon identifying common, differentially regulated genes by Trp53 $^{WT}$ and Trp53 $^{3KR/3KR}$ in epidermal cells from both mouse strains (through epigenetic and transcriptional analyses), the authors performed an in vivo genetic screen by molecularly perturbing each of 1113 candidate genes and assessing the impact on self-renewal. They narrowed the list to three genes: Sfrp1 (secreted frizzled-related protein 1), Lrp1 (low-density lipoprotein receptor-related protein 1), and Usp22 (ubiquitin-specific protease 22). The expression of these genes was up-regulated in Trp53 $^{WT}$ and Trp53 $^{3KR/3KR}$ epidermal cells, where there was decreased self-renewal. Notably, all three genes had been previously reported as potential negative regulators of the Wnt signaling pathway.

In the mouse skin, Wnt signaling is best known for regulating hair follicle stem cells and sebaceous gland function (8). It acts as a morphogen that directs spatial patterning of cells, cell fate, and tissue morphogenesis of the skin. However, loss of $\beta$ -catenin or wntless (Wls), constituents of the Wnt signaling pathway, also decreases interfollicular epidermal stem cell proliferation (9, 10). Consistent with the idea that p53 up-regulates the expression of inhibitors of Wnt signaling in the epidermis, Gan et al. demonstrated that p53 deletion or silencing the expression of Sfrp1, Lrp1, or Usp22 in mouse epidermal cells increased activity of Wnt signaling and clonal expansion. Conversely, overexpression of Sfrp1, Lrp1, or Usp22 in $Trp53^{-/-}$ epidermal cells decreased Wnt activity and clonal expansion. Thus, p53 regulates Wnt activity in the epidermis by controlling the expression of these three genes.

Gan et al. further found that p53 deletion leads to a gradient of Wnt activity within the expansion of mutated clones, with the highest activity in the center of the expansion and decreasing radially toward the edges. This spatial distribution was established only by suppressing the expression of the soluble Wnt inhibitor Sfrp1. The authors hypothesize that the Wnt gradient, rather than a specific level of Wnt signaling, is responsible for the rapid expansion of p53-deficient clones.

Gan et al. uncover a new role for p53 in gatekeeping aberrant cell renewal. In this model, the unrestrained expansion of p53 mutant cells is driven by the generation of a Wnt activity gradient within the mutant clones. It is not clear whether p53 loss in other tissues, or other oncogene or tumor suppressor gene mutations, also modulate Wnt signaling in the skin. Yet to be determined, as well, are the mechanisms by which Wnt signaling itself promotes renewal in this context. p53 puts the brakes on establishing a gradient by activating three inhibitors of Wnt action. However, even in the absence of a gradient, Wnt might still promote clonal expansion. Too much Wnt signaling can decrease renewal and lead to cell differentiation, as observed during hair (11) or Paneth cell differentiation in the intestine (12). This means that the promotion of stem cell renewal requires the right dose of Wnt signaling.

Sometimes a stem cell mutates and turns on its Wnt signaling pathway permanently, as observed in intestinal crypts. However, clonal expansion of such cells has been attributed to their secretion of Wnt inhibitors to eliminate wild-type neighbors (13, 14). This is a key step in the early stages of intestinal cancer. Whether a similar mechanism occurs in the epidermis remains to be determined. The approach of Gan et al. opens the possibility of unraveling clonal expansion and cell competition in response to other oncogenic hits and in other locations.

---

REFERENCES AND NOTES

1. A. M. Boutelle et al., Trends Cell Biol. 31, 298 (2021).

2. Q. Gan et al., Science 394, eaed3065 (2026).

3. M. Oren et al., Cell 187, 1569 (2024).

4. T. Li et al., Cell 149, 1269 (2012).

5. C. Blanpain et al., Nat. Rev. Mol. Cell Biol. 14, 489 (2013).

6. A. Sánchez-Danés et al., Nature 536, 298 (2016).

7. K. Murai et al., Cell Stem Cell 23, 687 (2018).

8. C. Blanpain et al., Annu. Rev. Cell Dev. Biol. 22, 339 (2006).

---

ACKNOWLEDGMENTS

9. Y. S. Choi et al., Cell Stem Cell 13, 720 (2013).

10. X. Lim et al., Science 342, 1226 (2013).

The authors acknowledge support of Fonds National de la Recherche Scientifique, Walloon Excellence Research Institute, Télévie, Fonds Erasme, Fondation Contre le Cancer, Fondation Université Libre de Bruxelles, Fondation Julie et Françoise Drion, and the European Research Council.

10.1126/science.ael7341

11. R. DasGupta et al., Development 126, 4557 (1999).

12. J. H. Van Es et al., Nat. Cell Biol. 7, 381 (2005).

13. D.J. Flanagan et al., Nature 594, 430 (2021).

14. S. M. van Neerven et al., Nature 594, 436 (2021).

REGULATION

Longevity products: Innovation through regulation

Longevity is taking off, but regulatory oversight is essential to determining what works—and what doesn't

Camila Strassle, $^{1}$ Holly Fernandez Lynch, $^{1}$ Patricia J. Zettler $^{2,3}$

The field popularly referred to as “human longevity” is booming, incorporating a variety of approaches aimed at optimizing human health by reversing cellular aging, prolonging life, and extending “healthspan” (the duration of life spent in good health). With longevity companies rapidly proliferating

globally, and, in the United States, drawing billions of dollars in investments and interest from government officials (1), some commentators have raised concern that longevity interventions fall in a gray area for regulators. Using the US as an example, we dispel that notion and discuss how longevity products generally fall squarely inside the existing regulatory regime. To the extent that there are challenges to bringing longevity products legally to market, they are largely scientific rather than regulatory. The most important priorities to support safe and effective interventions, promote innovation, and avoid grift are to develop and validate endpoints capable of predicting longevity benefit and to enforce appropriate legal requirements.

As proponents of the longevity field see it, progress against chronic diseases has lagged because traditional medical interventions have focused on symptom management once disease has occurred. But, their argument continues, if age is the major risk factor for most chronic diseases, manipulating the aging process itself could delay disease development and progression. The hope is to prevent age-related chronic diseases rather than tackling them individually as they arise.

This vision has attracted government interest. For example, in July 2026, the US Food and Drug Administration (FDA) held an Advisory Committee meeting regarding whether to allow the sale of several compounded peptides, including those used for longevity purposes, without first securing FDA approval (2). Amid this interest, questions have arisen regarding whether FDA has an approval pathway that can be applied to “longevity” products, in part because it does not recognize aging as a disease (3, 4). Other commentators seem to recognize that FDA has an applicable approval pathway but bemoan the lack of a pathway specifically tailored for longevity, asserting that regulatory ambiguity impedes the ability to legally bring products to market (5, 6). Yet the law is quite clear that longevity products fit under the existing regulatory regime, despite the current dearth of FDA enforcement. Moreover, although FDA could formally recognize aging as a disease, doing so would not necessarily make the pathway to approval clearer or more efficient.

The real challenge is neither new nor specific to the longevity field: It is difficult to directly measure clinical benefit when the outcome of interest is long-term, such as extended survival. Thus, drug approval is often based on demonstrated effects on surrogate endpoints thought to predict clinical benefit. Although specific details of regulatory approaches to longevity are country dependent, the bottom line will be the same for all policy-makers, science funders, and regulators interested in encouraging true progress in this sector. Rather than leaning into deregulation, dedicating resources to validating endpoints and to enforcing existing regulation is what will best serve the public, advance science, and promote innovation.

---

THE LONGEVITY FIELD AND US FDA JURISDICTION

Several longevity products have been identified as promising because of their connection to cellular and molecular signaling pathways implicated in aging, including those related to inflammation, oxidative stress, DNA stability, mitochondrial dysfunction, and cellular senescence (7). Among products already widely marketed to consumers, many are promoted for claims related to muscle strength, anti-inflammatory effects, improved cognition, and tissue repair or regeneration. To date, none of these products are FDA approved specifically for anti-aging indications.

High-profile longevity products include metformin (glycemic control), rapamycin (immunosuppression), sodium-glucose cotransporter 2 inhibitors (inhibition of glucose reabsorption), nicotinamide adenine dinucleotide (NAD $^{+}$ ) precursors (promoted for energy metabolism), resveratrol (promoted for anti-inflammation), senolytics fisetin and quercetin (promoted for senescent cell clearance), and peptides (short amino acid sequences promoted for various uses) (7, 8). Some of these products are FDA approved for uses unrelated to biological aging, such as diabetes, weight control, immunosuppression for organ transplant, and chronic kidney disease. However, many products popular for longevity have never been FDA approved for any use.

Some longevity products are being studied in ongoing trials. For instance, the US National Institute on Aging created the Translational Geroscience Network (TGN) to standardize research and speed development of therapeutics, with more than 80 affiliated trials. Additionally, the XPRIZE Foundation is hosting a “healthspan competition” through which it awarded funds in August 2026 to research teams running clinical trials. Gene editing and stem cell interventions are also being investigated clinically. Life Biosciences, for example, has started a trial of ER-100, a cellular reprogramming technology, with the hope of rejuvenating aging organs (9). Many companies, however, are already marketing products for anti-aging effects without conducting studies to demonstrate safety or effectiveness for that purpose. This is true even for some products that lack FDA approval for any use.

Longevity products fall within already-defined legal categories in the US. Indeed, most of these products clearly meet the definition of a drug, which the Federal Food, Drug, and Cosmetic Act (FDCA) defines to include products intended for use not only in the diagnosis, cure, mitigation, treatment, or prevention of disease, but also products intended to affect the structure or any function of the body. This “structure/function” element of the statutory definition means two things: Products can still be drugs even if not intended for therapeutic use, and whether FDA recognizes aging as a disease does not determine whether these products fall within FDA's drug authorities nor whether FDA can approve them for anti-aging. As a result, the notion that there is no clear regulatory pathway for longevity drugs because FDA does not classify aging as a disease is mistaken. For example, FDA regulates hair growth products and onabotulinumtoxinA (i.e., Botox) for reducing facial lines, both of which have been approved as drugs for nontherapeutic uses despite the lack of any targeted disease (10). FDA also recently signaled willingness to approve two canine drugs (LOY-001 and LOY-002) for longevity indications, independent of any disease claim, indicating that FDA will also view many human longevity products as drugs given the FDCA's identical definition of “drug” for animals and humans.

---

LEGAL PATHWAYS AND THEIR CONSTRAINTS

The many longevity products that meet the statutory drug definition are subject to the same requirements that apply to all FDA-regulated drugs. Generally, these products cannot be legally marketed for longevity or other uses (e.g., athletic performance, sexual function) in the United States without first securing FDA approval based on demonstrating safety and effectiveness, and they must comply with Current Good Manufacturing Practice requirements. Notably, exceptions are either limited in scope or legally unavailable, at least for now, to longevity drugs. Longevity products are nevertheless widespread on the US market because FDA has inconsistently enforced relevant requirements, and the financial incentives for companies to market products, despite the legal risk, are substantial. However, this does not mean that FDA lacks enforcement authority or that requirements are being met.

Legitimate industry should welcome such enforcement, and the public should demand it.

Some US states have recently expressed interest in expanding their “right to try” laws in ways that would purportedly permit the distribution and administration of unapproved longevity products. These laws, including the federal right to try law, previously have been limited to patients facing life-threatening conditions who seek use of investigational agents outside clinical trials—but in some cases, they are now being

extended to everyone, even individuals who are currently healthy. For example, Montana enacted such an expansion of its law in 2025 (SB535) and published regulations to govern the state's experimental treatment centers in 2026, while New Hampshire has proposed similar bills (HB1734 and HB1735). However, even when enacted, these state-level efforts have limited legal effect because states do not generally have the authority to exempt parties from compliance with federal law. Thus, when distributing longevity products in the United States, companies must follow applicable FDA requirements.

In addition to state efforts, many longevity companies, clinics, and preventive care practices

have emerged that facilitate the sale of longevity products through pathways other than traditional drug approval (11). However, these business models are often premised on a misunderstanding of the extensive reach of FDA requirements. For example, in some cases, companies sell longevity products directly to consumers as “research chemicals” or with the disclaimer “for research purposes only.” Despite their pervasiveness, such businesses generally run afoul of US law. Investigational drugs may be distributed to research participants enrolled in clinical trials under an FDA-authorized Investigational New Drug (IND) Application. But companies cannot sell unapproved drugs intended for human use under the guise of research and without an IND. Even for permissible research under an IND, there are strict limits on what companies may charge research participants; they may do so only with prior FDA authorization and may not profit from unproven interventions.

Compounding is another strategy sometimes used to distribute unapproved longevity products. Compounded drugs, which are ostensibly custom-made by pharmacies or specialized facilities, are not FDA approved, but the FDCA nevertheless allows compounding when certain criteria are met. Most relevant for longevity products, section 503A of the FDCA allows a pharmacy to make a compounded drug for an individual patient in response to a valid prescription order—for example, when a patient is allergic to an inactive ingredient in an approved product and needs a version without it. This kind of compounding is permitted only for distinct patient needs and only for certain substances, like those that are components of FDA-approved drug products or appear on an FDA-created list of permissible substances. Many longevity products do not meet these conditions, however. For example, the Secretary of the Department of Health and Human Services (HHS) recently directed FDA to remove several unapproved peptides from the list of substances disallowed for compounding owing to potentially serious safety risks, despite the absence of data to address the safety concerns that landed the peptides on that list in 2023 (12).

At the FDA Advisory Committee meeting in July 2026, panelists, many of whom had financial interests in peptide-related businesses, narrowly voted to recommend the further step of adding several peptides to FDA's list of permissible compounding substances despite the lack of demonstrated safety and effectiveness and serious concerns raised by FDA staff. At least until FDA further considers the issue through a rulemaking process, however, none of the wholly unapproved products popular for longevity uses may legally be compounded and sold. And regardless of what FDA decides on peptides, even for longevity drugs compounded using components of FDA-approved drugs or other permitted substances, the law does not allow mass production that resembles conventional drug manufacturing.

Some longevity products are marketed as “dietary supplements,” which—unlike substances defined as drugs under the FDCA—are not subject to premarket approval requirements. Yet many of these products are in fact drugs under the law because they do not fall within the definition of a dietary supplement. To be a dietary supplement, a product must contain a dietary ingredient, be intended for ingestion,

and must not contain approved or investigational drug ingredients or be intended for disease uses. Products using injectable, topical, transdermal, or nasal administration, as many longevity products do, cannot be dietary supplements because they are not ingested. Products that are intended to diagnose, cure, mitigate, treat, or prevent diseases related to aging are not dietary supplements, even if labeled as such.

“Off-label prescribing” of already-approved drugs to patients for unapproved longevity uses is the approach on the strongest legal footing, short of FDA approval explicitly for longevity uses. But many products currently used for longevity purposes are not already FDA approved for other uses. Moreover, bu-

siness models premised on off-label use are not completely unrestricted. Critically, drug manufacturers, and those speaking on their behalf, generally cannot directly promote or advertise off-label uses without violating the FDCA.

To be sure, there are limits to FDA's reach. For example, people living in the United States sometimes access experimental products through medical and wellness tourism, with some longevity research done offshore to evade FDA requirements and commercialize products with minimal safety and effectiveness data.

The upshot is that companies marketing longevity agents claiming to prevent or reverse aging or address age-related diseases and conditions generally will need to follow FDA's drug-approval process, including generating rigorous evidence of safety and effectiveness. Drug approvals could be for nondisease “structure/function” uses (where labeling and promotional materials describe the drug's role in altering the body's structure or function) or “disease” uses (where materials describe the drug's role in preventing, mitigating, or reversing age-related diseases). Either way, there are existing paths to approval—and approval is both legally necessary and appropriate.

---

ENFORCEMENT

Despite these legal requirements for longevity products, in many cases FDA has “enforcement discretion,” flexibility to decide not to take formal action against companies that ignore requirements. Under leadership friendly to the longevity industry, FDA might choose to deprioritize longevity products as a subject of enforcement action and seems poised to do so for certain peptides.

Yet FDA, and its global regulatory counterparts, have strong reasons to enforce regulatory requirements for longevity products. Legitimate industry should welcome such enforcement, and the public should demand it. Enforcing drug-approval requirements helps protect consumers against deceptively promoted and potentially unsafe products, while ensuring that minimum quality and manufacturing standards have been met, including that products are free of impurities, contaminants, and toxins and contain only what they claim and nothing else, at the claimed concentration. Experience has shown that medical products lacking FDA oversight can cause death and other serious harm: for example, the 2012 fungal meningitis outbreak traced to compounded steroid injections killed over 50 people and sickened hundreds (13). Enforcement helps ensure that drugs are rigorously studied for safety and effectiveness, which benefits patients and promotes innovation by ensuring that companies investing in the production of such evidence will not be undermined by competitors who are allowed to sell unproven interventions. FDA's role in protecting consumers may be especially important for products marketed for nontherapeutic, structure/function uses because these drugs are unlikely to be considered medically necessary and covered by payers.

Given the scope of the longevity industry, absent a budget increase, enforcement may require FDA to redirect considerable resources away from other priorities. If FDA fails to step in, states might enforce consumer protection laws that prohibit deceptive promotion against clinics and clinicians, as some have done in the context of those marketing stem cell interventions without reliable evidence to support their claims. Clinicians may also be disciplined by state medical boards or sued for malpractice by harmed consumers. However, these are poor replacements for FDA's expert premarket oversight (14).

---

CHALLENGES AND OPPORTUNITIES

The most straightforward path in the United States to meet the standards imposed by the FDA is for companies to seek approval for claims about the effects of longevity drugs on the structure or function of the body. Such approvals are important because consumers should at least be assured that longevity products effectuate the intended physiological or biochemical change in the body, even if it remains unproven whether that change ultimately makes a clinically meaningful difference in lifespan or healthspan. Scientifically, these structure/function claims can be supported by traditional clinical studies.

However, to realize the potential of the field, longevity products ultimately will need to be studied and proven safe and effective for outcomes more directly related to clinical benefit, including preventing and treating disease. This poses an added scientific challenge because demonstrating a preventive effect on age-related diseases and conditions, as well as long-term safety, by definition, will take a long time. The central difficulty will be identifying suitable surrogate endpoints for longevity studies, where the primary outcome measure is a marker that does not directly measure clinical benefit but is known to predict a clinically meaningful outcome and can be studied more quickly. Whether longevity products could qualify for FDA's accelerated approval pathway, under which surrogate endpoints need only be reasonably likely to predict clinical benefit (rather than proven to do so) at the point of approval, with clinical benefit confirmed thereafter, is an open question. That pathway is available only for products intended to treat serious or life-threatening conditions with unmet medical needs. It is not clear that aging itself meets that standard, although specific age-related diseases certainly could. It is also not clear whether use of accelerated approval or a similar pathway would be desirable from a policy perspective, given known challenges in the conduct of post-approval confirmatory studies $15$ . Regardless, testing products for clinical benefit will be essential to making sure that longevity products in fact work to slow, pause, or reverse age-related decline.

FDA, its global regulatory counterparts, and research funders therefore should engage with longevity communities to evaluate what study designs and kinds of evidence are both feasible to produce and sufficient to meet a rigorous standard of safety and effectiveness. Currently, there is no consensus about which surrogate endpoints to use for longevity trials, but there are candidate measures that could be linked to the biological processes of aging, several of which were discussed at a public meeting in May 2026 convened by the US Reagan-Udall Foundation for the FDA with the Advanced Research Projects Agency for Health (ARPA-H). Longevity stakeholders should continue to seek opportunities for identifying scientifically valid and regulatorily acceptable endpoints, whether through bodies like Reagan-Udall and the TGN (or similar efforts in other countries, like the Canadian Translational Geroscience Network), engagement with FDA (e.g., through formal meetings and the Biomarker Qualification Program), or private-led initiatives like the “healthspan competition.”

Regulatory regimes like FDA's both prevent companies profiting from drugs that do not work and help health care professionals and the public make informed decisions about what drugs to pursue or avoid based on reliable information regarding safety and effectiveness. In this way, although enforcement will be challenging and may narrow the longevity product market, at least for a time, it will be essential for advancing scientific understanding and supporting personal autonomy by ensuring that people seeking longevity products are no longer forced to rely exclusively on the claims of interested companies, influencers, and fellow users. Satisfying FDA's regulatory requirements may not be easy but it is the best path to ensure that the longevity movement lives up to its promise and offers more than just snake oil.

---

REFERENCES AND NOTES

1. The global longevity investment landscape: Leading investors by deal count, Longevity Investors (4 July 2025); https://www.longevityinvestors.ch/post/the-global-longevity-investment-landscape-leading-investors-by-deal-count.

2. M. Fu et al., FDA advisory committee's vote may open a drug-compounding back door for unapproved peptides. Health Affairs (4 August 2026).

3. M. Molteni, As billionaires race to fund anti-aging projects, a much discussed trial goes overlooked, STAT (9 August 2022); https://www.statnews.com/2022/08/09/anti-aging-projects-funding-much-discussed-trial-overlooked/.

4. T. Thadani, “These are the treatments dominating the business of living longer, Washington Post,” 12 January 2026; https://www.washingtonpost.com/health/2026/01/12/longevity-maha-antiaging-health-rfk/.

5. L. M. DeVito et al., Ann. N. Y. Acad. Sci. 1507, 70 (2022).

6. "Optimizing longevity: From research to action" (United States Senate Special Committee on Aging, 2025), serial no. 119-03.

7. A. Martinović et al., Front. Aging 5, e1495029 (2024).

8. R. LaFountain, Top 10 longevity & anti-aging breakthroughs of 2025: What the year taught us about extending human healthspan, Healthspan (29 December 2025); https://www.gethealthspan.com/research/article/top-ten-longevity-anti-aging-breakthroughs-of-2025?srsltid=AfmBOopACzkmXSnipyAib2OiySBilYOyi-uY1BFomJ64fqzA9DbiKGB2&zp\_type=article&zp\_slug=top-ten-longevity-anti-aging-breakthroughs-of-2025.

9. P. Knoepfler, FDA OKs risky, pioneering OSK rejuvenation trial with Sinclair's ER-100, The Niche (11 February 2026); https://ipscell.com/2026/02/fda-oks-risky-pioneering-osk-rejuvenation-trial-with-sinclairs-er-100/.

10. P.J. Zettler, Wash. Lee Law Rev. 78, 379 (2021).

11. M. Demaria, Aging (Albany NY) 17, 2452 (2025).

12. C. Jewett, "Heeding Kennedy's wishes, F.D.A. is expected to lift restriction on peptides," New York Times, 31 March 2026; https://www.nytimes.com/2026/03/31/health/peptide-banfda-rfk-jr.html.

13. E. S. Kastango, Lessons learned from compounding tragedies. Can. J. Hosp. Pharm. 66, 152 (2013).

14. R. Sachs, Regulating healthcare technologies in a post-FDA world, Washington University in St. Louis Legal Studies Research Paper no. 26-02-01 (16 February 2026).

15. H. Fernandez Lynch, C.T. Robertson, Science 374, 1205 (2021).

---

ACKNOWLEDGMENTS

C.S. served as assistant chief counsel to the FDA from 2024 to 2025. H.F.L. reports research funding outside this work from Arnold Ventures, the Greenwall Foundation, National Institutes of Health (NIH), and ARPA-H, and an honorarium and travel support from Eli Lilly for a 2024 lecture. P.J.Z. served as deputy general counsel to the HHS from 2023 to 2025; reports grant funding from Arnold Ventures, NIH, and FDA all outside the submitted work; and reports travel support from AbbVie outside the submitted work.

10.1126/science.aei2427

MEDICINE

A medical mystery unfolds in middle America

As a rare form of dementia claims one sibling after another, a journalist captures the profound costs of caregiving

Nancy Berlinger

Dementia is nearly always a family disease, because most dementia caregivers are family members of the person in need of care. In The Vanishing Family, journalist Robert Kolker provides an account of what dementia caregiving is like that is as breathtaking as the scientific “quest to end dementia” that the subtitle proclaims.

This is Kolker's second book about a large postwar American family with bad medical luck. In his 2020 bestseller Hidden Valley Road, six sons in a family of 12 are diagnosed with schizophrenia. In The Vanishing Family, five siblings in a family of nine are diagnosed with frontotemporal dementia (FTD), a relatively rare type within the group of conditions known as Alzheimer's disease and related dementias (ADRD) or, more commonly, as dementia.

Boomer and Gen X readers who grew up in the suburbs are likely to recognize the postwar Catholic family introduced in the early chapters. (Bookmark the page with the family tree.) The parents, Ollie and Jean, meet on a train from St. Louis to South Bend, Indiana, in 1949; marry in 1950; and have their first two children within 2 years. Their life in St. Louis, and briefly in Canton, Ohio, is shaped by Jean's deep commitment to a highly organized household and the sanctity of cocktail hour: old-fashioneds for Ollie, Manhattans for Jean. So far, so Mad Men.

Between the mid-1960s and the mid-1970s, several things happen to this family (their surname is omitted and several names are changed). They move to Pittsburgh, where Ollie, a civil engineer, is pursuing opportunities in a not-yet-postindustrial city. Jean gives birth at 48 to her ninth

The Vanishing Family: Love, Fate, and the Quest to End Dementia

Robert Kolker Doubleday, 2026.

368 pp.

child, Barb, 10 years younger than her nearest sibling. And Jean changes, spending her days drinking and watching game shows, devoid of affect, barely speaking. Barb is the closest observer of Jean's behavior and the child most affected by it; her siblings have left home or are on their way out, Ollie spends most of his time at work or with other women.

The family's initial explanation for Jean's behavior—alcoholism, perhaps related to Ollie's infidelities—leads to an intervention. Jean quits drinking, but neither her affect nor her behavior changes. She dies at age 62 of breast cancer, which had widely metastasized before she disclosed that anything was wrong.

Kolker intersperses the family narrative with chapters about the history of brain science, beginning with the story of Phineas Gage, who survived an industrial accident in 1848 that drove a pointed iron rod through his frontal lobe. (Visitors to the Countway Library at

Harvard Medical School may spot Gage's skull, with its two immense holes, in a display case.) Kolker is expert at summarizing scientific developments and placing them in historical and social context: The impact of Freud's theory of human personality made brain science look old-fashioned, readers learn, slowing investigation into neurological factors in personality and behavior.

As five of Jean's children begin, in their early 40s, to behave like their mother—exhibiting apathy toward their own families, loss of executive function, excessive drinking or other loss of social inhibitions—the scientific and family narratives intersect. Christy, the first sibling to become ill, is initially misdiagnosed with Pick's disease, before, over many years, false starts, and family reunions at the Mayo

Clinic, the cause is identified as V337M.

This autosomal-dominant (inherited) mutation of the MAPT gene leads to overproduction of tau protein and degeneration of the frontal and temporal lobes. It affects speech, executive function, and social emotions and intuitions. By the book's end, several members of the next generation have tested positive for V337M, and the disease has been traced to Jean's grandfather and other branches of the family. None of the new generation of dementia drugs targets FTD.

This book is also a story about family caregiving, represented by Sue, who dedicates 30 years of her life to her sister Christy: relocating and changing jobs, establishing a household and routines built around Christy's substantial needs and also her flourishing, which includes eating at Wendy's and being around people. Sue also periodically

cares for three other siblings with FTD; she is the one who knows how disability insurance works. Dementia caregiving, like all care work, is highly gendered; most caregivers are women within the family. Yet The Vanishing Family also shows how the husbands, exes, partners, sons, and nephews of the five siblings with FTD are involved in their care.

If there is one weakness in this outstanding book, which is paced like a thriller and will stay with the reader for a long time, it is the title. This family does not vanish. Its members show up for each other, as so many families must do, with so little help, in America's aging society. □

10.1126/science.ael1175

HISTORY OF SCIENCE

Walden and the wild roots of conservation science

The famous work of transcendentalism holds lessons for modern ecologists, maintains a Thoreau scholar Renée Bergland

The Walden Experiments: The Science of Henry David Thoreau

Robert Thorson

Princeton University Press, 2026. 328 pp.

Henry David Thoreau resisted social conventions. In his lifetime, he was as famous for his refusal to conform as he was for his published works. After he died in 1862, Ralph Waldo Emerson listed everything his friend had refused to do: Thoreau had “no profession; he never married; he lived alone; he never went to church; he never voted; he refused to pay a tax to the State” (I). Emerson was disappointed that Thoreau had devoted his time to woodcraft and wordcraft instead of paying work. He complained that his friend preferred being “the captain of a huckleberry party” to any other, more serious occupation.

Today, Thoreau's steadfast dedication to wild places looks more like discipline than laziness. He was a remarkably prolific writer. As it

turned out, his published speeches, essays, poems, and books were just the tip of the iceberg; his unpublished journal and detailed logs of observations of the natural world have proved as valuable as his publications.

Thoreau's work is a little hard to categorize. Philosophers claim him as a philosopher. Political activists see his writings as cornerstones of social and environmental justice movements. Writers admire the beauty of his language. Scientists—particularly ecologists—use his careful measurements of 19th-century natural phenomena to document 21st-century changes to our ecosystems.

The Walden Experiments: The Science of Henry David Thoreau is also hard to categorize. Author Robert Thorson is a geologist by training, a professor of earth sciences, and a widely re-

spected Thoreau scholar who has worked collaboratively with theologians, historians, philosophers, and literary scholars over the years. The Walden Experiments is his fourth book on Thoreau. The first of them, Walden's Shore: Henry David Thoreau and Nineteenth-Century Science, is a multifaceted work of literary criticism, biography, and intellectual history that argues that Thoreau was a skilled practitioner of 19th-century inductive science. The Walden Experiments leaps from that early work toward a new set of questions more centered on the 21st century. Here, Thorson invites readers to consult Walden as a “mid-nineteenth-century oracle” that speaks to the current global sustainability crisis.

Thoreau's environmental ethics were informed by Charles Darwin's incisive and inductive observations about the natural world.

Thoreau could certainly be oracular. His proclamation “In wildness is the preservation of the world” continues to inspire readers while resisting easy interpretation. Thoreau defined the wild as “whatever has not come under the sway of man.” Perhaps his refusal to be part of social institutions—not only state, church, family, and profession but also intellectual discipline—was an attempt to preserve as much of his own wildness as he could. It might also explain his lean toward the sciences. In the mid-19th century, there was no field of study quite as wild as the dynamic and shifting world of natural and physical science.

It is common knowledge that Thoreau's interest in wildness prompted his choice to build a small house near Walden Pond, just outside of Concord, Massachusetts; live in it for 2 years; and write a book about his experience. Thorson goes deeper into the story, explaining that Walden took 9 years to write.

Thoreau was not satisfied with the first draft. He put it aside for about 2 years, then returned to it when he discovered Charles Darwin's journal of his own observations of the natural world, The Voyage of the Beagle. Inspired by Darwin's detailed observation, his inductive approach, and his beautiful prose, Thoreau “transformed an otherwise inauspicious lake into a powerful and globally famous symbol for humanity’s rightful place in nature and our responsibility to it.”

Thorson argues that Thoreau did not impose his philosophy on the land. On the contrary, he developed his environmental ethics inductively, from the ground up, starting with geology and then gradually expanding his vision. According to Thorson, "Thoreau—the young man keenly interested in glaciology, climate change and optics—is what we would today call a scientist." He explains that "all but one of Walden's eighteen chapters engage with outdoor field observations and interpretations involving physics, chemistry, astronomy, geology, hydrology, botany, zoology, ecology, meteorology, climatology, and limnology (the study of lakes)."

Thorson's book is also divided into 18 chapters, each of which is a wide-ranging freestanding essay that jumps off from a Walden chapter to recontextualize Thoreau's scientific thought for 21st-century readers.

If Walden is a wild ride, The Walden Experiments is a little wilder. It invites readers to shake off their disciplinary shackles, jump out of their silos, and try to reconnect with the wild roots of the sciences, in hope that embracing wildness will give us the intellectual courage we need for the preservation of the world. □

---

REFERENCES AND NOTES

1. R. W. Emerson, "Thoreau," The Atlantic, August 1862; https://www.theatlantic.com/magazine/archive/1862/08/thoreau/306418/.

10.1126/science.aej3535

Edited by

Jennifer Sills

Gene-editing research is promising, but guardrails can help to prevent tragic results.

---

Epistemic justice in gene-editing research

A Science investigation into a fatal individualized gene-editing trial exposed failures of safety oversight, consent, and disclosure (“A fatal reaction,” B. Borrell, News Feature, 23 July, 10.1126/science.aek8000). The case shows how epistemic injustice—the unfair exclusion of people from producing, interpreting, or validating knowledge (1)—increases the vulnerability of patients and compromises scientific progress. Additional patient protections and advocacy can address this imbalance in power.

The patient, referred to as Mei in the News story, and her parents were indispensable to knowledge production. They supplied genetic and clinical information, financed preclinical research, and documented conversations. After Mei's death, her parents sought to ensure that the research's advances and setbacks were shared with the community, as did the family of Jesse Gelsinger, who died as a result of gene therapy in 1999 (2). Yet their contribution, as in the Gelsinger case (2), was not fully recognized: Their concerns carried less authority, and they lacked independent means to judge whether the experiment constituted ethically defensible human research.

Patient engagement is increasingly recognized in clinical research, and China's Regulations on the Administration of Clinical Research and Clinical Translational Application of New Biomedical Technologies (State Council Order No. 818) (3) strengthens research-participant protections. However, patients are not technical experts. Families hold essential knowledge about conditions, values, and acceptable burdens, but they cannot be expected to adjudicate vector dose, toxicology, or evidentiary sufficiency.

Empowering patients and families requires access to independent scientific advice for families, meaningful patient representation in ethics and safety review, full consideration of adverse preclinical findings, and prompt reporting of serious outcomes. Patients must be recognized not merely as sources of data, funding, and hope, but as participants in knowledge governance.

---

Dong Dong $^{1,2,3}$ , Yiou Wang $^{1}$ , Tao Ma $^{1}$

$^{1}$ The Illness Challenge Foundation, Beijing, China. $^{2}$ The Chinese University of Hong Kong Centre for Bioethics, The Chinese University of Hong Kong, Hong Kong SAR, China. $^{3}$ JC School of Public Health and Primary Care, The Chinese University of Hong Kong, Hong Kong SAR, China. Email: dongdong@cuhk.edu.hk

---

REFERENCES AND NOTES

1. M. Fricker, "IX—Rational authority and social power: Towards a truly social epistemology" in Proceedings of the Aristotelian Society vol. 98 (1998), pp. 159–178.

2. P. Gelsinger, "Jesse's intent" (2001); https://www.jesse-gelsinger.com/jesses-intent.html.

3. State Council of the People's Republic of China, Regulation on the Administration of New Biomedical Technology Clinical Research and Translational Clinical Application, Order No. 818 (28 September 2025); https://www.lawinfochina.com/display.aspx?id=45801&lib=law.

10.1126/science.aek9401

---

Recording results from single-patient trials

In the News Feature “A fatal reaction” (23 July, 10.1126/science.aek8000), B. Borrell reports that a 6-year-old child died 7 days after receiving an individualized base-editing treatment. Although the girl was the only participant that the trial intended to enroll, the registry continued to describe the study as “recruiting” more than a year later (1). During the same period, the related pre-clinical study cited nonhuman-primate delivery as evidence of translational feasibility (2) but did not mention the human trial. To ensure that the results of single-patient research are preserved, relevant protocols and major human outcomes should be recorded and kept current in a public registry.

In a conventional early-phase trial, an individual outcome is interpreted alongside those of other participants. In a single-patient study, however, the first outcome may be the only human evidence available. A registry that remains unchanged after the reported death of the sole participant is therefore more than an administrative oversight. Families, clinicians, ethics committees, and other

research groups may rely on that entry when judging whether human testing should continue and what risks have emerged.

Safety reports, trial registries, and journal articles serve different purposes and are updated on different schedules. As a result, important clinical information can reach regulators and ethics committees without becoming part of the public account of translation.

To remedy the potential for reporting gaps, investigators, trial registries, and journals should work together to keep the translational record current for individualized genetic therapies. Authors submitting preclinical studies should disclose any directly related human protocol and update that information before acceptance. Registries should include a field for treatment-emergent deaths and life-threatening events. Such details could distinguish occurrence from suspected or confirmed causality, as World Health Organization registration standards already allow (3). Preclinical articles and trial records should also link to one another, with dated notices added when major human outcomes arise after publication.

Such notices would not assign blame or settle causality. Rather, they would serve to ensure that claims of clinical promise are not separated from the human experience that followed.

Ge Zhang, Honglin Zheng, Qiang Li, Taiqi Zhao, Teng Li

Henan Medical College, Zhengzhou University, Zhengzhou, Henan, China. Email: liteng981120@126.com

---

REFERENCES AND NOTES

1. ClinicalTrials.gov, "Clinical trial of the dual vector base editor for the treatment of the CHD3-R1025W mutation," no. NCT06860672; https://clinicaltrials.gov/study/NCT06860672

2. K. Yang et al., Nature 651, 785 (2026).

3. World Health Organization, WHO trial registration data set, version 1.3.1.; https://www.who.int/tools/clinical-trials-registry-platform/network/who-data-set.

10.1126/science.aek9583

---

Streamlining oversight in gene-editing trials

In the News Feature “A fatal reaction” (23 July, 10.1126/science.aek8000), B. Borrell reports that a 6-year-old girl died after the world’s first-in-human brain base-editing trial in China. This case highlights a governance paradox: A regulatory system with detailed ethical rules failed to prevent a fatal outcome. Institutional fragmentation may be at fault. As countries expand access to highly individualized gene-editing therapies, they must also strengthen continuous, cross-institutional oversight.

In China, gene-therapy products developed through the investigational new drug (IND) pathway generally undergo national regulatory review, including assessment of nonclinical safety data and requirements for long-term follow-up (1). Gene- and cell-based interventions may also be conducted as investigator-initiated trials (IITs) under institutional oversight, creating a dual-track regulatory structure. The reported intervention followed the IIT route and therefore did not receive national-level product review. The limitations of institution-centered oversight became particularly consequential in this case: The ethics committee approved the study before the final primate toxicology report was available, and the subsequent findings of liver injury in monkeys did not lead to a documented rereview before the child was treated. Although closer adherence to IIT requirements might have prevented some of these failures, the case raises a broader question about whether institutional review alone provides sufficient independent oversight for high-risk interventions.

Other countries also have multiple approval paths. The European Union's regulatory framework for Advanced Therapy Medicinal Products (ATMPs) includes a controversial hospital exemption that permits certain custom-made ATMPs to be used without centralized marketing authorization (2). US regulators have proposed a “plausible mechanism” framework for individualized gene-editing therapies that would remain within the federal regulatory system (3).

The News story described compartmentalization beyond ethics approval. China's Civil Code prohibits charging research participants fees related to a clinical trial (4), yet the family paid for the study by donating to a company controlled by the lead investigator. The university limited its review to whether the published article was connected to the funded experiment, the journal regarded the child's death as a matter for institutional investigation, and regulators confined enforcement to narrowly defined violations.

When responsibility becomes fragmented across multiple actors, accountability becomes difficult to assign (5). China has made changes to address the governance weaknesses revealed by this case, such as establishing a statutory framework for high-risk biomedical technologies (6). However, strengthening regulatory pathways alone is insufficient. Ethical review should be a continuous process rather than a single authorization decision. Financial oversight should encompass affiliated companies and other entities under investigators' control. Safety governance should extend across institutional boundaries. Major adverse events and emerging safety concerns should be communicated through a traceable reporting system linking investigators, ethics committees, regulators, and journals with one designated body responsible for determining whether new evidence warrants modification, suspension, or termination of an ongoing study. In China and beyond, robust oversight is what sustains scientific credibility, protects research participants, and safeguards responsible investigators from avoidable ethical and legal risk.

---

Xiaojie Li $^{1}$ and Ruishuang Liu $^{2}$

$^{1}$ Department of Situation and Policy, University of International Business and Economics, Beijing, China. $^{2}$ Department of Medical Ethics and Law, Peking University Health Science Center, Beijing, China. Email: Irs1187@vip.sina.com

---

REFERENCES AND NOTES

1. National Medical Products Administration, Center for Drug Evaluation, Technical Guidelines for Nonclinical Research and Evaluation of Gene Therapy Products (Trial), Announcement no. 49 (2021) [in Chinese]; https://www.cde.org.cn/main/news/viewInfoCommon/41bc557bec23a6ebfb0e148cc989f041.

2. European Commission: European Health and Digital Executive Agency, Rodés Sánchez et al., "Study on the hospital exemption – Member State reports. Annex 1 to the final report" (Publications Office of the European Union, 2025).

3. US Food and Drug Administration FDA, "Considerations for the use of the plausible mechanism framework to develop individualized therapies that target specific genetic conditions with known biological cause: Draft guidance for industry" (FDA, 2026); https://www.fda.gov/regulatory-information/search-fda-guidance-documents/considerations-use-plausible-mechanism-framework-develop-individualized-therapies-target-specific.

4. Civil Code of the People's Republic of China, Article 1008 (2020) [in Chinese]; https://www.court.gov.cn/zixun/xiangqing/233181.html.

5. D. F. Thompson, Am. Polit. Sci. Rev. 74, 905 (1980).

6. State Council of the People's Republic of China, State Council Decree no. 818, Regulations on the Administration of Clinical Research and Clinical Translation of Novel Biomedical Technologies, effective 1 May 2026 (2025).

10.1126/science.ael0421

---

The utilitarian temptation in gene editing

The recent gene-editing trial led by Zilong Qiu and colleagues in Shanghai (“A fatal reaction,” B. Borrell, News Feature, 23 July, 10.1126/science.aek8000) was not only a case of failed oversight; it also illustrates a recurring moral danger in biomedical research. Such trials can be driven by two mutually reinforcing motivations: the pursuit of professional recognition and a crude utilitarian rationale that risky experiments are justified because, if successful, they may benefit large numbers of present and future patients. This second motivation, which is especially powerful in biomedicine, can cloak self-interest in moral language and encourage

PUT HUMAN HEALTH AT THE HEART OF YOUR RESEARCH?

Submit your research: cts.ScienceMag.org

Science

Translational

Medicine

NAAAS

Science.org/journal/stm

researchers to discount evidence of danger to the individuals enrolled as subjects. Gene-editing interventions, in particular, may appear morally elevated because their promised benefits aim to enhance general welfare for all, including future generations, regardless of nationality or group identity. For this reason, the appeal to future benefit requires strict ethical limits.

He Jiankui, whose work producing genetically altered babies led to jail time (1), invoked such an ideal when he defended his embryo-editing experiment, saying that he was “proud” of work intended to protect children from HIV (2). Qiu, who later joined more than 100 Chinese scientists in condemning He’s experiment for its ethical failures and its “far-reaching and profound social consequences” (3), may have allowed hoped-for therapeutic benefit to overshadow foreseeable risks to the human subject. Such tragedies may not be products of ambition or greed—they may be driven instead by the moral force of utilitarian rhetoric.

The crucial distinction lies between minimal moral requirements and the pursuit of maximal general welfare. The former must be satisfied before the latter can be invoked. Foreseeable harm to human subjects must be identified, assessed, and minimized before an ethics committee approves a trial. In Qiu's case, the committee reportedly approved the human trial without reviewing the final primate-safety report. That failure was not a procedural detail; it was a bioethical error. It allowed speculative future benefit to override the prior duty to protect the subject.

Bioethics education should therefore emphasize that “do no harm” is a minimal and universal moral requirement, whereas “enhancing general welfare” is culturally and contextually specified (4, 5). For example, biomedical trainees, clinicians, and administrators should all be taught that a hospital ethics committee must require and review relevant primate safety studies before approving a human-subject trial. No promise of future benefit, however broad or humane it sounds, can justify overriding the minimal duty not to expose a human subject to foreseeable and preventable harm. Otherwise, the utilitarian temptation cannot be resisted.

---

Ruiping Fan

Department of Public and International Affairs, City University of Hong Kong, Hong Kong, China. Email: safan@cityu.edu.hk

---

REFERENCES AND NOTES

1. A. Joseph, "CRISPR babies scientist sentenced to 3 years in prison," Scientific American, 30 December 2019.

2. S. Sataline, I. Sample, "Scientist in China defends human embryo gene editing: He Jiankui uses Hong Kong summit to reply to critics of his Crispr-Cas9 trials altering baby DNA for HIV resistance," The Guardian, 28 November 2018.

3. “Joint statement by scientists: Firm opposition and strong condemnation” [in Chinese], China Digital Times, 26 November 2018; https://chinadigitaltimes.net/chinese/600448.html?amp#.

4. R. Fan, Dao 23, 353 (2024).

5. R. Fan, J. Med. Philos. 10.1093/jmp/jhae029 (2024).

10.1126/science.ael2357

EDITOR'S NOTE on the Research Article "China's solar expansion policy reduces bird diversity"

On 20 August 2026, Science ran the Research Article “China’s solar expansion policy reduces bird diversity” (1). The editors have received multiple eLetters expressing technical concerns about the paper. We are currently in the process of evaluating the validity of the concerns and determining whether any further action is needed on the paper. H. Holden Thorp, Editor-in-Chief, Science

---

REFERENCES AND NOTES

1. H. Zhang et al., Science 393, 831 (2026).

Receive Prize Money

Publish in Science Journals

Achieve International Recognition

In partnership with our generous sponsors, Science Awards and Prizes is celebrating the winning essays so far this year from five of our international prize programs, each recognizing exceptional early-career researchers whose pioneering work is advancing knowledge and making a meaningful impact across their respective fields.

Through the Science Awards and Prizes program, more than \$300,000 in prize funding is awarded annually, providing recipients with critical resources to propel their research forward, establish independent laboratories, pursue ambitious new lines of inquiry, and build the foundations for the next stage of their scientific careers.

ASU Science

Grand Prize Winner

Satellite data can help transform food systems

Published 5 February 2026

Grand Prize Winner

Courtney A. Schreiber

Changing practice and policy for patients with early pregnancy loss

Published 1 April 2026

Grand Prize Winner

Fábio F. Rosa

Turning tumors against themselves

Published 2 April 2 2026

Grand Prize Winner

Sergey Stavisky

Regaining your voice

Published 16 July 16 2026

Grand Prize Winner

Jan Elaine Soriano

Rewiring the paralyzed body's false alarm

Published 30 July 2026

To see our full list of awards and prizes including those that are currently accepting submissions visit science/org/prizes.

SPECIAL ISSUE

A LIFELONG FOCUS ON

WOMEN'S HEALTH

YEVGENIYA NUSINOVICH, SACHA VIGNIERI, SARAH ROSS, and JOANA OSÓRIO

istorically, most human studies have involved male participants, and their findings were assumed to apply to women and children as well. Meanwhile, women's health typically came up in the context of reproduction and childbearing, if at all. However, biomedical science has come a long way over the years, and it has become increasingly clear that there are sex-dependent differences in many diseases that do not involve reproduction and that occur at all ages. Across all areas of women's health, including both reproductive and nonreproductive conditions, much remains to be learned about the underlying biology, effects of hormones and genetics, and potential therapeutic interventions.

Each person has a specific set of sex chromosomes, which not only play a role in determining their sex at birth (which some genetic mutations and hormone signaling abnormalities can also affect) but also influence disease risk throughout their life. For example, women's cells usually have one active and one inactive X chromosome, but there are cases when an X chromosome escapes from inactivation or some cells lose one of their X chromosomes. These changes influence the risk of age-related conditions such as cardiovascular disease, cancer, and autoimmunity. Pregnancy and lactation stimulate extensive remodeling of physiology, with dynamic changes in the immune system helping to support restructuring of tissues; yet we are just beginning to consider how this immune adaptation may influence health after childbearing. Even the ovaries, which are typically studied in the context of reproduction, have biological functions that last well beyond the reproductive years. Incomplete knowledge that led to myths about potential hormonal interventions after the reproductive years, during and after menopause, has been supplanted by more recent research, and we now understand much more about how to improve women's health and well-being during life transitions. Pain biology is a vital topic that remains poorly understood: Empirical data show clear differences between the sexes in pain perception and rates of specific painful conditions, but mechanistic understanding is often lacking.

Overall, it is becoming increasingly clear that women's health and sex-dependent biological differences go well beyond reproduction and affect people throughout their lifespan and across different medical conditions. There is still much to learn in this area, and continuing research efforts have the potential to improve the well-being of all patients. 10.1126/science.aem4087

PERSPECTIVES

Does the postmenopausal ovary still matter?

Ovaries may influence health long after their reproductive function ends Michael B. Stout $^{1,2}$ , and Francesca E. Duncan $^{3,4}$

Women can spend decades in the postmenopausal phase of life. However, the organ that defines the transition to this stage, the ovary, has been largely ignored after its role in reproduction ceases. A common assumption is that the postmenopausal ovary, which

no longer contains sufficient follicles to sustain fertility, is a dormant remnant. The postreproductive ovary has also been referred to as a senile organ, a gyratum, or a wrinkled and shriveled husk. However, investigators and clinicians have suspected for years that the human ovary retains biological activity after the reproductive period and have pointed out that “the contribution of this organ to the overall well-being of the postmenopausal woman remains to be defined” (1). Historical and emerging data support the idea that the postreproductive ovary is active, which has clinical implications.

In the reproductive years (generally 12 to 50 years of age), ovaries produce eggs and hormones that are essential for fertility and overall health. Female mammals are born with a finite number of eggs, which are housed within primordial follicles. As follicles grow and develop, they produce peptide hormones, including inhibins A and B, activins, follistatin, and anti-Müllerian hormone, as well as steroid hormones, including estrogen, progesterone, and androgens (including testosterone). The reserve of follicles decreases with time and dictates reproductive lifespan. Female individuals have millions of primordial follicles at birth, hundreds of thousands at puberty, thousands at early menopause, and hundreds to none after menopause. Menopause, which is defined as the permanent cessation of menstruation for 12 months, is a direct

consequence of the loss of follicles.

Understanding the functions of the postmenopausal ovary is essential because women's lifespan is increasing globally, whereas the age of menopause has stayed relatively constant. An extended postreproductive period is a conserved feature across mammalian species (see the figure), and today, many women can expect to spend nearly one-third of their life in the postmenopausal state. Analysis of ovarian tissue has shown that the postmenopausal ovary is distinct morphologically and structurally from its premenopausal counterpart (2). Because of the loss of follicles, the postmenopausal ovary is \~30% smaller. Furthermore, its stroma, the scaffold of connective tissue that surrounds and supports it, proliferates and becomes the dominant feature of the organ. Stromal fibrosis, scar-like tissue, and remnants of ovulation known as corpora albicantia are also prominent. Granulomata composed of macrophages and lymphocytes may be present, and sclerotic (hard and thick) blood vessels are abundant. The tunica albuginea, a protective layer of connective tissue that surrounds the ovary, thickens, whereas the surface epithelium forms papillary and cord-like growths and rolls inward to form inclusion cysts.

Nevertheless, the postmenopausal ovary can still produce hormones. The stroma, in particular, produces most of the steroid hormones secreted by the ovary at this stage. Measurements of hormone gradients from the ovarian vein to peripheral blood showed that the postmenopausal ovary primarily produces and secretes testosterone, as well as smaller amounts of androstenedione, estradiol, estrone, and progesterone (3). Notably, stromal cells from the postreproductive ovary

Female reproductive activity across the mammalian lifespan

Although most female mammals do not menstruate, unlike humans, they still spend a substantial portion of their lifespan in a postreproductive phase. This phase is preceded by a period of progressively irregular reproductive cycles, which eventually cease. The parenthetical ages on the left are average lifespans from population-based studies (human) and breeding colony data (animals).

also express receptors for luteinizing hormone and follicle stimulating hormone and can respond to gonadotropin stimulation (3). The observation that the postmenopausal ovary maintains some hormone production is a reminder that it is not dormant. Whether the aged ovary continuously secretes factors—hormones or other molecules—that meaningfully influence whole-body physiology is unclear. This knowledge gap is partly due to a lack of appropriate technologies to assess ovarian functions beyond endocrine activity. Moreover, obtaining postmenopausal ovarian tissue from healthy women for research purposes is challenging. Such studies usually only include tissue from a small number of women and are focused mostly on the menopausal transition itself rather than on the functional changes that occur in the following decades. However, molecular studies in mice, monkeys, and humans suggest that the postreproductive ovary continues to undergo dynamic changes, especially related to inflammation and cellular senescence (4–6).

Ovarian aging is characterized by the progressive accumulation of immune cells and an increase in the concentration of inflammatory molecules. In postreproductive mouse ovaries, T cells, macrophages, and multinucleated giant cells, which are associated with chronic inflammation, continue to expand with age (4). Furthermore, transcriptomic profiling revealed a near-complete loss of gene expression associated with reproductive cellular identity and an up-regulation of gene expression programs associated with antigen presentation, leukocyte activation, and immunoglobulin production (4). These findings demonstrate that the postreproductive mouse ovary develops a sustained proinflammatory phenotype in the absence of acute infection (5). Moreover, in monkeys approaching the end of their reproductive lifespan, spatial transcriptomic profiling has identified discrete proinflammatory foci within the ovarian stroma whose gene expression profile suggests the release of senescence-associated signals (6). Thus, aged ovaries may amplify inflammation locally and potentially systemically.

With age, the ovary also shows signs of cellular senescence. Findings from a proteomic analysis of healthy ovaries from women aged 50 to 75 years suggest that gene expression programs change markedly across the postmenopausal decades of life (7). In this study, younger postmenopausal ovaries expressed more proteins involved in RNA processing and gene regulation. By contrast, older ovaries predominantly expressed proteins involved in complement activation, cellular senescence, tissue remodeling, and extracellular matrix organization. Notably, exposure of postmenopausal human ovarian tissue in culture to doxorubicin, a chemotherapeutic agent that induces cellular senescence, resulted in a secretory phenotype that substantially overlaps with that of healthy (untreated) aged ovaries (8). This similarity suggests that two different stressors, chemotherapy and chronological aging, elicit similar molecular programs in the human ovary.

Not all human data, however, indicate the presence of inflammation and senescence in the postmenopausal ovary. For example, an atlas of individual cells and their accessible DNA in the postmenopausal human ovary revealed a stroma-dominated, transcriptionally active organ expressing multiple hormone receptors, but with relatively few immune cells and senescence-associated molecules (9). The reasons underlying the divergences across studies of the postmenopausal human ovary may reflect differences between the gene and protein expression, the source of ovarian tissue, or the methodology used. However, most observations across species are consistent with an enrichment in senescent stromal cells and immune cells in the postreproductive ovary.

A subset of molecules expressed in the postreproductive ovary are thought to be secreted factors that could reach the bloodstream and have systemic effects $(4, 7)$ . Therefore, similar to adipose tissue, which is now recognized as an active endocrine organ that affects whole-body physiology, the postmenopausal ovary could influence an individual's disease burden by releasing factors that contribute to inflammation and aging long after its reproductive function ceases.

The concept that the postmenopausal ovary contributes to whole-body inflammation and aging is only inferential because it is built on observations that this organ has a secretory phenotype and that other aging tissues drive systemic disease through similar mechanisms. Therefore, direct measurements of the concentrations of ovary-derived factors in the blood of postmenopausal women and investigations into the effects of ovarian secretory outputs on distal organs are priorities for the field. Notably, senescent cells and the molecules they produce, which seem to characterize the postmenopausal ovary, are major drivers of “inflammaging,” the chronic, low-grade, sterile inflammation that emerges with age and is a key driver of cardiovascular disease, neurodegeneration, loss of muscle mass and function, and metabolic dysfunction (10). Therefore, the postmenopausal ovary could be a previously unrecognized contributor to whole-body inflammaging in older women.

If the postmenopausal ovary contributes to systemic aging through the factors it secretes, its surgical removal (oophorectomy)

after menopause could potentially improve health outcomes in women. However, very little data exist about the impact of oophorectomy after natural menopause, especially in the absence of ovarian disease. Observational studies such as the Mayo Clinic Cohort Study of Oophorectomy and Aging (11) and the Nurses' Health Study (12) examined women undergoing oophorectomy before menopause in whom the abrupt loss of estrogen and androgen production had clear adverse health effects, including increased cardiovascular mortality, cognitive decline, and all-cause mortality. The strongest harms were observed in the youngest women, which indicates that they stem from premature loss of ovarian endocrine function rather than from removal of a postreproductive ovary. A 2005 analysis of the possible consequences of postmenopausal oophorectomy used a Markov decision model based on aggregated age-specific risk estimates from the published literature (13). The outcomes were projected through age 80, and the analysis concluded that preserving the ovaries maximized survival until \~65 years of age. However, the survival benefit of postmenopausal ovarian conservation is now less clear than previously assumed. An updated Markov analysis incorporating hazard ratios from the Mayo Clinic and Nurses' Health Study cohorts concluded that for women undergoing hysterectomy at 50 years old or older, concurrent removal of the fallopian tubes and ovaries (salpingo-oophorectomy) did not increase mortality compared with conservation of these organs through age 80 years (14). More rigorous studies are needed to determine whether there are specific time windows in which postmenopausal salpingo-oophorectomy may confer a net health benefit, although caution is required, as ovarian androgens also influence bone density, libido, and metabolic function (15).

The emerging molecular signatures of the postmenopausal ovary show that this organ is far from dormant. Research is needed to fully determine the nature of these ovarian molecules, how they send signals to the rest of the body, and whether they contribute to or protect against systemic aging. This knowledge will enable the development of biomarkers to define healthy and disease states and could uncover therapeutic targets to improve women's health during the postreproductive period, which will continue to widen as humans live longer.

---

REFERENCES AND NOTES

1. S. V. Nicosia, Med. Clin. North Am. 71, 1 (1987).

2. M. Laszczynska, A. Brodowska, A. Starczewski, M. Masiuk, J. Brodowski, Histol. Histopathol. 23, 219 (2008).

3. C. Longcope, J. Soc. Gynecol. Investig. 8, S67 (2001).

4. A. Converse et al., Mol. Hum. Reprod. 32, gaag038 (2026).

5. A. Converse et al., PLoS Biol. 23, e3003204 (2025).

6. H. Lu et al., Protein Cell 15, 364 (2024).

7. M.A. Watson et al., bioRxiv doi:10.64898/2026.03.26.714635 (2026).

8. P. R. Devrukhkar et al., Aging Cell 24, e70111 (2025).

9. E. Lengyel et al., Cell Rep. 41, 111838 (2022).

10. C. Franceschi, J. Campisi, J. Gerontol. A Biol. Sci. Med. Sci. 69, S4 (2014).

11. C. M. Rivera et al., Menopause 16, 15 (2009).

12. W. H. Parker et al., Obstet. Gynecol. 121, 709 (2013).

13. W. H. Parker et al., Obstet. Gynecol. 106, 219 (2005).

14. S. K. Rush, X. Ma, M. A. Newton, S. L. Rose, Obstet. Gynecol. 139, 735 (2022).

15. S. R. Davis, S. Wahlin-Jacobsen, Lancet Diabetes Endocrinol. 3, 980 (2015).

---

ACKNOWLEDGMENTS

We thank the participants of the 2025 National Institute of Aging Workshop on Sex Differences Impacting Human Health Across the Lifespan, which inspired this Perspective, and M. Beestrum from the Galter Health Sciences Library and Learning Center. We acknowledge support from the National Institutes of Health (grants R01AG099844, R01AG069742, and R33AG072137 to M.B.S.; grants U54AG075932 and UH3CA268105 to F.E.D.).

10.1126/science.aeh7496

$^{1}$ Aging and Metabolism Research Program, Oklahoma Medical Research Foundation, Oklahoma City, OK, USA. $^{2}$ Oklahoma City Veterans Affairs Medical Center, Oklahoma City, OK, USA. $^{3}$ Department of Obstetrics and Gynecology, Feinberg School of Medicine, Northwestern University, Chicago, IL, USA. $^{4}$ Buck Institute for Research on Aging, Novato, CA, USA. Email: f-duncan@northwestern.edu

WOMEN'S HEALTH

REVIEW

Shaping maternal physiology: Immune control of reproductive adaptation

Dan Corral $^{1*}$ and Yasmine Belkaid $^{2*}$

Pregnancy is one of the most extensive physiological transitions of adult life, requiring coordinated adaptation across virtually every maternal organ to support fetal growth while preserving maternal health. Although these changes have traditionally been viewed as endocrine driven, emerging evidence identifies the immune system as an active regulator of maternal physiology. The immune system is increasingly recognized as a broad regulator of host physiology, positioning it as a key driver of maternal adaptation. In this work, we discuss how pregnancy reprograms maternal immunity to orchestrate physiological reprogramming of tissues. This framework shifts the view of maternal immunity from one centered on immune homeostasis and fetal tolerance to one in which immune cells actively instruct physiological short- and long-term adaptation across maternal organs to shape health beyond reproduction.

Pregnancy requires one of the most extensive episodes of physiological remodeling in adult life. Coordinated adaptations across multiple maternal organs support fetal development, prepare for lactation, and preserve maternal health, yet the mechanisms orchestrating these changes remain incompletely understood. Despite its importance, pregnancy has historically remained underexplored. Deciphering the biology of this distinctive physiological state is one of the most important challenges in biomedical research and promises fundamental insights with broad implications for human health and disease.

Within the first few months of pregnancy, virtually every maternal organ adapts its structure and function to support fetal growth, protect the developing fetus and prepare for lactation (1). Cardiovascular output increases, hematopoiesis is redirected, nutrient absorption and metabolism are reprioritized, and barrier tissues undergo profound remodeling. Together, these coordinated adaptations redistribute maternal resources to meet the changing demands of reproduction, a process captured by the concept of homeorhesis. Homeorhesis describes the coordinated reprogramming of multiple physiological systems to establish a new physiological state appropriate for reproduction rather than maintaining a stable internal set point, as in homeostasis (2) (Fig. 1). Although many of these adaptations resolve after parturition, others persist well beyond delivery, indicating that pregnancy leaves lasting physiological imprints on maternal health (3). At the same time, the quality of maternal adaptation shapes fetal development and lifelong offspring health, making pregnancy a critical window for the future health of both mother and child. Despite these far-reaching consequences, the physiological mechanisms that actively remodel maternal tissues have received far less attention than fetal development and maternal-fetal tolerance.

Maternal adaptation has traditionally been viewed through an endocrine lens, whereby maternal and placental hormones coordinate tissue growth, metabolism, and vascular adaptation to sustain fetal development while preserving maternal physiology (1). Although these discoveries transformed our understanding of pregnancy, they also reinforced the view that gestation is primarily characterized by controlled immune suppression (4). This framework, however, could not readily explain how mothers simultaneously maintain protective immunity while supporting the extensive tissue remodeling required for reproduction. Longitudinal immune profiling has demonstrated that pregnancy is not a state of generalized immunosuppression but a dynamic immunological process, with distinct immune programs accompanying successive stages of gestation (5, 6).

The recognition that maternal immunity is dynamically remodeled throughout pregnancy also raises a broader possibility: that the immune system may actively contribute to maternal physiological adaptation. Indeed, immune cells are now recognized as fundamental regulators of host physiology, controlling processes that extend well beyond host defense and immune tolerance, including metabolism, vascular remodelling, stem and progenitor cell differentiation, epithelial development, and tissue repair (7). These physiological functions have been described across development, tissue homeostasis, and regeneration, establishing the immune system as an integral component of all organs function throughout life (7). Pregnancy provides a specific physiological context in which diverse and complex physiological programs are synchronously deployed across multiple adult organs. Whether pregnancy primarily repurposes existing immune circuits or establishes specialized programs remains an important unresolved question.

In humans, most knowledge of maternal immunity derives from studies of peripheral blood mononuclear cells. Yet the physiological adaptations that sustain pregnancy occur within tissues, where immune cells establish intimate interactions with epithelial, stromal, and endothelial cells. Although peripheral blood analyses brought new insights (5, 8), circulating leukocytes provide only a partial view of pregnancy-induced immune remodeling because immune cell phenotype is shaped by local tissue-specific signals. The decidua exemplifies this principle, containing specialized immune populations whose composition and function are instructed by local microenvironment (9). Consequently, most mechanistic insights into immune-mediated maternal adaptation come from animal models, which enable direct investigation of tissue-resident immune populations. However, translating these findings to humans will require careful consideration of species-specific differences in reproductive anatomy and physiology.

On the basis of recent evidence, we propose that maternal immunity is an integral component of the homeorhetic program of pregnancy, coordinating the tissue remodeling and physiological adaptations that establish and maintain the maternal state required for successful reproduction. We examine how pregnancy reprograms the immune system before discussing how specialized immune populations coordinate reproductive, epithelial, and metabolic adaptations. Lastly, we explore how these programs generate long-lasting immune and tissue memory that can enhance maternal fitness or, when disrupted, contribute to disease. Viewed through this perspective, pregnancy emerges not simply as a state that the immune system accommodates but as a physiological program that it actively helps construct.

---

Pregnancy reprograms hematopoiesis and lymphopoiesis for physiological adaptation

During pregnancy, the maternal immune system is precisely reprogrammed to preserve host defense, establish fetal tolerance and support tissue remodeling across multiple organs. This transition begins within the primary lymphoid organs, where endocrine and immune signaling pathways redirect hematopoiesis and lymphopoiesis, enhancing erythroid and myeloid output while reshaping lymphocyte development before immune cells are deployed to peripheral tissues (Fig. 2).

Fig. 1. Homeostasis and homeorhesis: Distinct modes of physiological regulation. (Left) Homeostasis maintains physiological variables around a relatively stable set point through compensatory responses that restore equilibrium following transient perturbations, such as thermogenesis or glycemia regulation. (Right) By contrast, homeorhesis establishes a new dynamic equilibrium by coordinately reprioritizing physiological processes across multiple organs to support a changing biological state, such as pregnancy and lactation. Rather than preserving preexisting physiological set points, homeorhesis redirects metabolism and organ function to meet the evolving demands of reproduction while maintaining overall organismal fitness. Pregnancy therefore represents a paradigmatic example of homeorhetic regulation in which coordinated adaptations across endocrine, metabolic, cardiovascular, and immune systems support fetal development, maternal health, and preparation for lactation. The dashed arrows within the curves represent regulatory mechanisms that direct physiological variables towards the set point; in homeostasis, these mechanisms act towards a relatively stable set point, whereas in homeorhesis, they act towards a set point that itself changes over time.

One manifestation of maternal immune reprogramming occurs in the bone marrow, where pregnancy reshapes hematopoietic output to support the evolving physiological demands of the growing fetoplacental unit (10). Classical histological studies in humans demonstrate progressive normoblastic erythropoietic hyperplasia accompanied by granulopoietic expansion and reduced marrow lymphocytes that persist into the postpartum period (10). These changes are accompanied by profound rewiring of hematopoietic stem cell (HSC) biology. Experimental studies in mice indicate that endocrine cues contribute to these adaptations. Estradiol promotes HSC self-renewal through estrogen receptor $\alpha$ -dependent signaling, whereas the cholesterol metabolite 27-hydroxycholesterol (27-HC) cooperates with estrogen to drive HSC mobilization, pregnancy-associated extramedullary hematopoiesis, and expansion of erythroid and granulocyte-macrophage progenitor populations (11, 12). In parallel, pregnancy-induced derepression of endogenous retrotransposons activates innate immune cyclic GMP-AMP synthase (cGAS)-stimulator of interferon genes (STING) signaling and interferon (IFN) production within HSCs, promoting their proliferation and erythroid differentiation (13). These findings are supported by observations that reverse transcriptase inhibitor-based antiretroviral therapy during pregnancy suppresses endogenous retrotransposon signaling, reduces pathway activation, and is associated with an increased risk of gestational anemia, suggesting a role for endogenous retrotransposons in pregnancy-associated erythropoiesis (13). Although the mechanisms underlying maternal hematopoietic adaptation remain incompletely understood, these observations reveal that pregnancy remodels hematopoiesis through coordinated endocrine and immune signaling, resetting erythroid, and immune cell production to support gestation.

The thymus undergoes an equally notable but functionally distinct transformation. Pregnancy induces thymic involution, a phenomenon once interpreted as evidence of diminished immune competence but now recognized as a process of selective immune remodeling. Across mammalian species, progesterone signaling within cortical thymic epithelial cells (TECs) reduces thymic cellularity by $>95\%$ (14–16). In mice, T cell development is transiently arrested at the earliest precursor stages, resulting in marked depletion of double-positive thymocytes, whereas mature single-positive populations are relatively preserved (17). Progesterone simultaneously suppresses chemokine expression required for thymic homing of lymphoid progenitors, thereby limiting de novo T cell production during gestation (18). Despite this great reduction in size, ultrastructural studies demonstrate continued thymocyte selection, epithelial remodeling, and phagocytic clearance of self-reactive cells throughout pregnancy (19). How pregnancy-induced remodeling translates into lineage and functional alterations remains a largely open question.

Pregnancy redirects thymic differentiation toward specialized immune populations that fulfil reproductive functions. Progesterone-dependent activation of receptor activator of nuclear factor $\kappa$ B (RANK) signaling within autoimmune regulator (AIRE)-expressing medullary TECs (mTECs) selectively promotes the generation of regulatory T cells that subsequently populate the maternal-fetal interface and the visceral adipose tissue, where they contribute to fetal tolerance and metabolic adaptation. Disruption of this pathway results in fetal loss and gestational diabetes in mice (20). More recently, pregnancy-induced thymic involution has also been associated with enrichment of non-classical T cell precursors. Among these, intraepithelial lymphocyte precursors (IELps) are redirected toward tissue-specific physiological functions outside the thymus (21).

Thus, during pregnancy, coordinated changes in hematopoiesis, thymic differentiation, and specialized immune cell production establish an immune landscape adapted to gestation. How endocrine cues are integrated with immune-intrinsic signaling pathways to coordinate such reprogramming remains poorly understood. Likewise, how pregnancy remodels primary lymphoid niches to selectively control immune cell production and lineage commitment is only beginning to emerge.

---

Immune regulation of the maternal-fetal interface.

The maternal-fetal interface provides the best-characterized example of immune cells regulating maternal tissue morphogenesis (Fig. 3).

Fig. 2. Pregnancy reprograms immune cell development to support physiological adaptation. Pregnancy induces coordinated changes in the bone marrow, spleen, and thymus that reshape immune cell development to meet the physiological demands of gestation. In the bone marrow, estrogen promotes HSC proliferation, whereas pregnancy-associated derepression of endogenous retroviruses (ERVs) activates cGAS-STING signaling and type I interferon responses to further stimulate HSC expansion. Pregnancy also promotes HSC mobilisation and extramedullary haematopoiesis in the spleen, increasing erythroid and myeloid output to support the heightened oxygen transport and innate immune requirements of gestation. Progesterone directly signals through progesterone receptor-expressing cortical TECs (cTECs), driving pregnancy-associated thymic involution. In parallel, progesterone promotes the RANK-dependent expansion of AIRE-expressing mTECs, enhancing thymic $T_{reg}$ generation. $T_{reg}$ cells subsequently contribute to fetal tolerance and maternal metabolic adaptation. Despite thymic involution, IELps are enriched during pregnancy and are preferentially home to the mammary gland, where they support epithelial differentiation and lactogenesis. Together, these coordinated changes illustrate how pregnancy reprograms immune cell development to orchestrate maternal physiological adaptation rather than solely host defense.

Establishment of efficient placental circulation requires extensive remodeling of uterine spiral arteries into large, low-resistance vessels capable of sustaining fetal growth. This process requires coordinated interactions between invading extravillous trophoblasts, stromal cells, and specialized immune populations. Spatial transcriptomic analyses reveal a highly organized decidual immune landscape in which uterine natural killer (NK) cells, macrophages, dendritic cells, and T cells occupy distinct anatomical niches that evolve throughout the first trimester (22).

Among these populations, decidual NK (dNK) cells represent the dominant leukocyte population during early pregnancy, accounting for up to 70% of decidual immune cells. Unlike circulating NK cells, dNK cells are specialized for tissue remodeling rather than cytotoxicity accumulating around spiral arteries prior to trophoblast invasion and initiating vascular remodeling through the secretion of angiogenic mediators and IFN- $\gamma$ (23, 24). In this context, IFN- $\gamma$ functions not as a proinflammatory cytokine but as an essential developmental signal (24). Beyond vascular remodeling, dNK cells also produce pleiotrophin, osteoglycin, and osteopontin, directly promoting fetal development independently of their vascular effects (25). In humans, genetic interactions between maternal killer cell immunoglobulin-like receptors (KIRs) expressed by uterine NK cells and fetal HLA-C alleles influence trophoblast invasion and are associated with susceptibility to preeclampsia and fetal growth restriction, supporting a conserved role for uterine NK cells in regulating placentation (26, 27).

Other innate immune populations perform complementary functions within this developmental program. Mast cells promote implantation by supporting tissue remodeling and spiral artery remodeling at the maternal-fetal interface, potentially through the production of tissue remodeling mediators, including galectin-1, transforming growth factor- $\beta$ , and mast cell proteases (28). Early depletion studies also implicated uterine CD11c $^{+}$ cells but not conventional dendritic cells

in decidualization and angiogenesis (29, 30). In parallel, ovarian macrophages sustain corpus luteum vascularisation and progesterone production, thereby ensuring uterine receptivity for implantation (31). As gestation progresses, decidual macrophages adopt a tissue remodeling phenotype that promotes trophoblast invasion, extracellular matrix remodeling and vascular adaptation while simultaneously contributing to maternal-fetal tolerance (32). Adaptive immune cells further refine these processes. In both humans and mice, neutrophils contribute to placental vascular development by inducing a proangiogenic T cell population that secretes vascular endothelial growth factor (VEGF) and promotes endothelial sprouting, highlighting how innate and adaptive immune cells cooperate to achieve a common physiological outcome (33). Within the adaptive immune compartment, regulatory T cells ( $T_{reg}$ cells), although comprising only a small fraction of decidual leukocytes, are indispensable for establishing the immunological environment that supports implantation and early placental development (4). In mice, transient $T_{reg}$ cell depletion during the peri-implantation period impairs spiral artery remodeling, alters uterine NK cell function, and results in fetal loss and fetal growth restriction (34). However, whether implantation failure reflects a specialized developmental function of $T_{reg}$ cells or the loss of their broader anti-inflammatory activity remains unclear.

The same immune circuits required to establish pregnancy are subsequently repurposed to terminate it. As gestation reaches term, the decidual immune landscape shifts from a tissue remodeling program to a controlled inflammatory state that initiates labor. Neutrophils and macrophages accumulate in the myometrium during labor, accompanied by increased production of proinflammatory cytokines including interleukin-1β (IL-1β), IL-8, and tumor necrosis factor α (35). In mice, IL-33 released by uterine fibroblasts activates type 2 innate lymphoid cells (ILC2s), whereas IL-5-dependent eosinophil responses also contribute to efficient labor (36).

Thus, the immune system can be repurposed to support distinct physiological programs across gestation. How tissue microenvironmental cues are integrated to confer these stage-specific immune functions remains poorly understood.

---

Programming mammary epithelial differentiation

The mammary gland provides another example in which immune cells regulate tissue development rather than simply protecting mature organs (Fig. 3). During pregnancy, prolactin and progesterone initiate lobuloalveolar expansion and prepare the gland for milk production (37). Increasing evidence indicates that immune cells also contribute to the establishment of local developmental programs that promote epithelial differentiation.

Macrophages, eosinophils, and mast cells are well-established regulators of mammary morphogenesis during puberty and tissue remodeling during involution (38). However, little is known about their physiological contribution during pregnancy, although current evidence supports a temporally ordered sequence of immune programs that coordinate pregnancy-associated epithelial differentiation. During early and midpregnancy in mice, activation of signal transducer and activator of transcription 6 (STAT6) signaling within mammary epithelial cells coincides with local production of IL-4 and IL-13, promoting alveolar differentiation. Although genetic deletion of both cytokines delays epithelial development, differentiation recovers by late gestation, indicating that type 2 immune signals accelerate but are not solely responsible for this program (39). As pregnancy progresses, the immune landscape shifts toward type 1 immunity, characterized by accumulation of T-bet-expressing lymphocytes (21, 40). Mice lacking T-bet exhibit impaired mammary epithelial differentiation, defective lactogenesis, and reduced neonatal growth (21). A specialized population of CD8αα+ intraepithelial lymphocytes directly promotes differentiation of both secretory luminal and myoepithelial cells and is enriched in trophic factors capable of acting on epithelial progenitors, suggesting that immune-derived signals directly regulate mammary morphogenesis (21).

Evidence for analogous immune-epithelial interactions in humans is beginning to emerge. Independent single-cell analyses of human breast milk identify diverse epithelial and immune populations, including macrophages, T cells, NK cells, and tissue-resident IEL-like lymphocytes that resemble those described in mice (21, 41, 42). Moreover, alterations in immune composition correlate with lactation performance, with females exhibiting low milk production displaying altered macrophage inflammatory signatures together with altered epithelial cell states (43). Although mechanistic evidence in humans remains limited, these findings support conserved immune-epithelial interactions during mammary differentiation. Whether immune cells similarly integrate endocrine signals to regulate epithelial differentiation across other maternal barrier tissues remains an important question.

---

Reallocating maternal metabolism via tissue and immune adaptation

Successful pregnancy requires coordinated redistribution of maternal metabolic resources to sustain fetal growth while preserving maternal health and preparing for lactation. This homeorhetic transition has traditionally been attributed to endocrine regulation, but accumulating evidence indicates that immune cells also remodel metabolic organs to accommodate the changing energetic demands of gestation (Fig. 3).

The small intestine undergoes rapid structural expansion early in pregnancy, increasing absorptive surface area before maternal food intake rises, indicating that intestinal remodeling is anticipatory rather than a response to increased nutritional demand (44, 45). Neural regulation may provide an additional layer of reproductive intestinal adaptation: Reproductive hormones remodel enteric neuronal activity to support maternal feeding in Drosophila (46), whereas the extensive bidirectional communication between enteric neurons and intestinal immune populations (47) raises the possibility that neuroimmune circuits similarly contribute to maternal intestinal adaptation in mammals. T cell immune signals also directly regulate intestinal remodeling during pregnancy. RANK ligand (RANKL) produced by T cells and mesenchymal cells promotes epithelial differentiation and enterocyte survival. Mice lacking epithelial RANK fail to undergo normal villous expansion during pregnancy, resulting in impaired nutrient absorption and altered metabolic health in the offspring, demonstrating that RANK-dependent intestinal remodeling contributes directly to reproductive success (44). A complementary immune program remodels the intestinal barrier during reproduction. In mice, eosinophils accumulate in the small intestine from pregnancy through lactation and promote goblet cell differentiation and mucus production, strengthening innate protection against enteric infection (48). Although pregnancy-induced villus expansion occurs independently of microbiota, pregnancy is accompanied by marked remodeling of the maternal gut microbiota (49). Transfer of third-trimester microbiota into germ-free mice is sufficient to induce increased adiposity, insulin resistance, and inflammatory changes (49), demonstrating that host-microbiota interactions constitute an additional physiological mechanism that reallocates maternal metabolism during pregnancy.

Pregnancy also remodels maternal energy storage. Beyond their established role at the maternal-fetal interface, thymus-derived regulatory T cells accumulate in visceral adipose tissue during gestation, where they limit inflammation, preserve glucose tolerance, and support maternal metabolic homeostasis. Their depletion results in gestational diabetes, fetal macrosomia, and persistent metabolic abnormalities in the offspring (20).

The liver undergoes equally profound remodeling. Pregnancy induces physiological hepatomegaly through coordinated hepatocyte proliferation and metabolic reprogramming, increasing biosynthetic capacity to support fetal growth and prepare for lactation (50). Genetic disruption of hepatocyte gp130, the shared signaling coreceptor for the IL-6 cytokine family, selectively impairs late gestational liver remodeling and restricts fetal growth without affecting earlier stages of pregnancy, demonstrating that IL-6 family cytokine signaling is required for normal hepatic adaptation (50). Although the cellular sources of these cytokines remain unresolved, these findings further support the concept that cytokine signaling is an integral component of maternal metabolic adaptation.

Fig. 3. Maternal immune cells coordinate physiological adaptation across tissues. Immune cells contribute to maternal tissue adaptation during pregnancy in mice by regulating vascular, epithelial, and metabolic programs. At the maternal-fetal interface, decidual NK cells, macrophages, $T_{reg}$ cells, and other immune populations support implantation, placentation, fetal growth, and labor. In the mammary gland, IL-4 and IL-13 signaling and T-bet $^{+}$ lymphocytes, including CD8αα $^{+}$ intraepithelial lymphocytes, promote epithelial differentiation and lactogenesis. In the intestine, RANKL produced by T cells and mesenchymal cells drives epithelial expansion and nutrient absorption during pregnancy, whereas eosinophils accumulating during pregnancy and lactation promote goblet cell differentiation and mucus production. $T_{reg}$ cells support metabolic homeostasis in visceral adipose tissue. In the liver, gp130-dependent IL-6 family signaling promotes hepatocyte proliferation and metabolic adaptation. Together, these tissue-specific programs demonstrate how immune cells and immune-associated signals contribute to maternal physiological adaptation during reproduction. Question marks indicate cellular sources or mediators that remain unresolved.

Rather than endocrine regulation alone, immune cells and immune-derived signals play an important role in to redirecting nutrient acquisition, storage, and allocation to meet the physiological demands of reproduction. These findings also raise the intriguing possibility that immune cells do not simply respond to changing metabolic demands but actively anticipate them. Defining how immune cells integrate endocrine, metabolic, and tissue-derived signals to coordinate these anticipatory adaptations across organs remains an important challenge for the field.

---

Pregnancy leaves a lasting physiological memory

Pregnancy-induced immune remodeling extends beyond childbirth. Evidence from both mice and humans indicates that reproductive experience leaves persistent alterations in immune composition, epigenetic programs and tissue organization. These observations support the emerging concept of a physiological memory of pregnancy, in which previous gestational experience modifies maternal responses to subsequent pregnancies and may influence disease susceptibility later in life.

The maternal-fetal interface provides one of the clearest examples of persistent immunological adaptation. In humans, repeated pregnancies are associated with a pregnancy-trained decidual NK cell population characterized by increased production of IFN- $\gamma$ and VEGFA, the latter promoting vascular sprouting (51). In parallel, fetal microchimeric cells established during a first pregnancy can persist within maternal tissues after delivery (52), providing continued antigenic stimulation that supports the maintenance of fetal-specific $T_{reg}$ cells across successive pregnancies despite the progressive replacement of microchimeric cells by those derived from each new gestation (53). Beyond maintaining immune tolerance, fetal microchimeric cells have also been implicated in maternal tissue repair, angiogenesis, and organ homeostasis in mice, with accumulating correlative evidence in humans (54). Whether fetal microchimeric cells contribute to the long-term physiological memory of pregnancy remains an important unanswered question.

Persistent effects of reproductive experience are also apparent in labor that is consistently shorter in multiparous than in nulliparous females who have not previously given birth (55). Multiparous females also exhibit distinct inflammatory trajectories before and during labor, characterized by higher baseline concentrations of CCL4, CXCL9, and CXCL10 before labor onset but a more restrained innate inflammatory response during active labor (56). Together, these observations raise the possibility that prior pregnancy may optimize the inflammatory program governing parturition, although the mechanisms linking these immune changes to labor progression remain to be established.

The mammary gland provides another example in which immune and tissue memory become intimately intertwined. In mice, following a complete pregnancy-lactation-involution cycle, mammary epithelial cells undergo widespread and stable DNA demethylation at STAT5-bound regulatory regions controlling milk synthesis, proliferation, and cell survival (57). These epigenetic modifications persist throughout reproductive life and enable more rapid transcriptional reactivation during subsequent pregnancies, providing a mechanistic basis for the accelerated mammary response observed in multiparous females (57). The immune landscape of the mammary gland is similarly remodeled by reproductive experience. In mice, pregnancy-induced CD8αα $^{+}$ intraepithelial lymphocytes persist following involution while maintaining tissue residency and cytotoxic potential (21). Consistent with these findings, in humans, parity is associated with increased T cell abundance in normal human breast tissue, suggesting that reproductive history is accompanied by lasting changes in the mammary immune landscape (58, 59). More notably, completion of the pregnancy-lactation-involution cycle drives long-term accumulation of CD8 $^{+}$ T cells that suppress mammary tumor growth and enhance antitumor immunity in mice, an effect absent after pregnancy alone. These observations raise the intriguing possibility that immune adaptations acquired during reproduction may be redeployed beyond lactation, simultaneously supporting tissue surveillance and enhancing mammary adaptation during subsequent pregnancies.

These findings establish the concept of pregnancy-associated physiological memory. Understanding how this memory is established, maintained, and disrupted across organs may explain why healthy pregnancies can promote long-term health, whereas complicated pregnancies increase the risk of chronic cardiovascular, metabolic, and inflammatory diseases later in life (60).

---

Could adaptation become maladaptation?

The immune program of pregnancy is not fixed but continuously shaped by environmental cues. Pregnancy must therefore be considered in the context of the physiological and pathological stressors that accompany gestation, including infections, obesity, and metabolic disorders, all of which can shape maternal adaptation and pregnancy outcomes.

Nutritional, metabolic, and infectious exposures all have the potential to influence this adaptive program. Maternal obesity and Western-style diets promote systemic inflammation, alter gut microbial communities, and impair intestinal barrier integrity, leading to increased circulating inflammatory mediators and placental dysfunction, including abnormal vascular development and heightened inflammatory responses (61). Infection encountered during pregnancy can similarly shift this physiological equilibrium toward inflammatory defense. Although pregnancy is not a state of generalized immunosuppression, the endocrine and immune adaptations that support gestation can increase susceptibility to selected infections while amplifying inflammatory responses once infection occurs. Infection-induced cytokine production and complement activation disrupt the immune and angiogenic environment of the maternal-fetal interface and impair placental vascular function and are associated with miscarriage, fetal growth restriction, stillbirth, and preterm birth, even in the absence of direct fetal infection (62). Thus, environmental perturbations might reshape the immune-tissue interactions that coordinate maternal physiological adaptation. Whether these changes represent transient deviations from physiological adaptation or trigger persistent maladaptive remodeling remains unknown. Recent longitudinal multiomic analyses further demonstrate that immunological maladaptation is detectable weeks to months before the onset of spontaneous preterm birth, supporting the concept that pregnancy complications are preceded by abnormal maternal immune adaptation rather than only inflammatory events at labor (6).

If successful pregnancy establishes adaptive programs that improve subsequent reproductive performance, then immune dysregulation during gestation may conversely leave persistent maladaptive alterations. Pregnancy complications, including preeclampsia, fetal growth restriction, recurrent miscarriage, and spontaneous preterm birth, substantially increase the risk of similar complications in subsequent pregnancies (63). Women with previous preeclampsia also exhibit long-lasting alterations in circulating immune cell composition and DNA methylation programs affecting genes involved in T and NK cell function (64). Pregnancy complications are further associated with impaired regulatory T cell responses, altered decidual NK cell function, and persistent inflammatory activation (65). Whether these persistent immune alterations contribute directly to recurrent pregnancy complications or instead reflect underlying disease susceptibility remains unresolved.

Beyond pregnancy, women with previous preeclampsia or gestational diabetes face substantially increased long-term risks of cardiovascular disease and metabolic dysfunction (60). Rather than representing isolated obstetric events, these conditions raise the possibility that physiological adaptations established during pregnancy are not always fully resolved after delivery. Environmental perturbations, including infection and obesity, may similarly leave lasting consequences by altering immune-tissue interactions during gestation, although whether these effects converge on common cellular and epigenetic mechanisms has yet to be established. Defining how physiological immune adaptation is maintained within its adaptive range and identifying the factors that redirect this program toward persistent maladaptation represents one of the central challenges in reproductive immunology, with implications for health that extend far beyond pregnancy.

---

Conclusions

The coordinated homeorhetic adaptation that underpins pregnancy depends on the close integration of endocrine and immune signaling.

Although reproductive hormones orchestrate profound physiological adaptations across maternal tissues, the mechanisms by which endocrine cues are decoded by immune cells and tissue niches to generate organ-specific physiological programs remain poorly understood. Determining how endocrine and immune signaling pathways are integrated to coordinate tissue-specific adaptation across multiple organs represents one of the central challenges for understanding maternal physiology during reproduction.

A second major challenge will be to determine how immune adaptation is temporally coordinated across organs and reproductive stages. Pregnancy requires the simultaneous remodelling of multiple maternal tissues, yet most studies have examined immune regulation within individual organs in isolation. Whether common immune circuits operate across the mammary gland, uterus, adipose tissue, liver, and intestine or whether each tissue interprets reproductive cues through distinct local programs remain unknown. Likewise, lactation remains comparatively unexplored, and whether it represents a distinct systemic immunological state or the continuation of pregnancy-associated immune programs has yet to be resolved. Integrating multiorgan and longitudinal approaches will therefore be essential for establishing a unified framework of maternal physiological adaptation.

Lastly, an important unresolved question is which immune adaptations established during reproduction are fully reversible and which leave lasting physiological consequences. Defining which components of these adaptations are beneficial and under what circumstances pregnancy complications or environmental perturbations become maladaptive represents a central challenge for the field. Addressing these questions will not only advance our understanding of reproductive biology but may also reveal fundamental principles governing how transient physiological states shape women's health across the lifespan.

Together, these questions establish pregnancy as a distinctive framework for understanding how immune cells integrate endocrine, stromal, and environmental signals to coordinate tissue adaptation. Addressing these challenges will require moving beyond organ-specific and descriptive studies toward mechanistic, multiorgan approaches that capture the dynamic nature of maternal physiology. Understanding how immune cells coordinate this adaptive program may reveal new opportunities to protect maternal and child health, improve resilience to pregnancy complications, and reduce the long-term burden of reproductive disorders.

---

REFERENCES AND NOTES

1. R. Haddad-Tóvolli, M. Claret, Nat. Rev. Endocrinol. 19, 564–580 (2023).

2. D. E. Bauman, W. B. Currie, J. Dairy Sci. 63, 1514–1529 (1980).

3. A. Bar et al., Sci. Adv. 11, eadr7922 (2025).

4. S. E. Ander, M. S. Diamond, C. B. Coyne, Sci. Immunol. 4, eaat6114 (2019).

5. N. Aghaeepour et al., Sci. Immunol. 2, eaan2946 (2017).

6. I.A. Stelzer et al., Nat. Commun. 17, 7121 (2026).

7. M. Nahrendorf, F. Ginhoux, F. K. Swirski, Science 389, 594–599 (2025).

8. C. A. DeBolt et al., Am. J. Reprod. Immunol. 94, e70166 (2025).

9. A. Erlebacher, Annu. Rev. Immunol. 31, 387–411 (2013).

10. L. Lowenstein, C. A. Bramlage, M. Lalonde, Blood 12, 261–277 (1957).

11. D. Nakada et al., Nature 505, 555–558 (2014).

12. H. Oguro et al., J. Clin. Invest. 127, 3392–3401 (2017).

13. J. Phan et al., Science 386, eado6836 (2024).

14. A. G. Rijhsinghani, S. K. Bhatia, L. T. Tygrett, T. J. Waldschmidt, Am. J. Reprod. Immunol. 35, 523–528 (1996).

15. T. A. Tibbetts, F. DeMayo, S. Rich, O. M. Conneely, B. W. O'Malley, Proc. Natl. Acad. Sci. U.S.A. 96, 12021–12026 (1999).

16. S. H. Ahn et al., Front. Endocrinol. (Lausanne) 13, 846226 (2022).

17. A. L. Zoller, F. J. Schnell, G. J. Kersh, Immunology 121, 207–215 (2007).

18. M. Laan, U. Haljasorg, K. Kisand, A. Salumets, P. Peterson, Eur. J. Immunol. 46, 2008–2017 (2016).

19. M. D. Kendall, A. G. Clarke, J. Anat. 197, 393–411 (2000).

20. M. Paolino et al., Nature 589, 442–447 (2021).

21. D. Corral et al., Cell 188, 1662–1680.e24 (2025).

22. S. Greenbaum et al., Nature 619, 595–605 (2023).

23. J. Hanna et al., Nat. Med. 12, 1065–1074 (2006).

24. A. A. Ashkar, J. P. Di Santo, B. A. Croy, J. Exp. Med. 192, 259–270 (2000).

25. B. Fu et al., Immunity 47, 1100–1113.e6 (2017).

26. S. E. Hiby et al., J. Exp. Med. 200, 957–965 (2004).

27. S. Xiong et al., J. Clin. Invest. 123, 4264–4272 (2013).

28. K. Woidacki et al., Cell Death Dis. 4, e462 (2013).

29. V. Plaks et al., J. Clin. Invest. 118, 3954–3965 (2008).

30. E. Tagliani et al., J. Exp. Med. 208, 1901–1916 (2011).

31. A. S. Care et al., J. Clin. Invest. 123, 3472–3487 (2013).

32. F. Sun, S. Wang, M. Du, J. Reprod. Immunol. 143, 103264 (2021).

33. S. Nadkarni et al., Proc. Natl. Acad. Sci. U.S.A. 113, E8415–E8424 (2016).

34. S. L. Hosking et al., JCI Insight 10, e169836 (2024).

35. K. Ji et al., Clin. Transl. Med. 13, e1234 (2023).

36. J. Siewiera et al., Immunity 56, 606–619.e7 (2023).

37. F. M. Hannan, T. Elajnaf, L. N. Vandenberg, S. H. Kennedy, R. V. Thakker, Nat. Rev. Endocrinol. 19, 46–61 (2023).

38. R. Vickers, W. Porter, J. Mammary Gland Biol. Neoplasia 29, 16 (2024).

39. W. T. Khaled et al., Development 134, 2739–2750 (2007).

40. A. Jaquish et al., Nat. Immunol. 26, 1411–1422 (2025).

41. D. Cansever et al., Nat. Immunol. 24, 1098–1109 (2023).

42. S. K. Nyquist et al., Proc. Natl. Acad. Sci. U.S.A. 119, e2121720119 (2022).

43. Y. Golan et al., Sci. Adv. 11, eadr7174 (2025).

44. M. Onji et al., Nature 637, 156–166 (2025).

45. T. Ameku et al., Cell 188, 2738–2756.e22 (2025).

46. D. Hadjieconomou et al., Nature 587, 455–459 (2020).

47. Y. Qu, Y. Zhang, S. Zhou, T. Yang, C. Chu, Trends Neurosci. 49, 551–565 (2026).

48. C. Huang et al., Nature 654, 209–218 (2026).

49. O. Koren et al., Cell 150, 470–480 (2012).

50. L. Yang et al., Cell 188, 6572–6590.e23 (2025).

51. M. Gamliel et al., Immunity 48, 951–962.e5 (2018).

52. D. W. Bianchi, G. K. Zickwolf, G. J. Weil, S. Sylvester, M. A. DeMaria, Proc. Natl. Acad. Sci. U.S.A. 93, 705–708 (1996).

53. T. Y. Shao et al., Science 381, 1324–1330 (2023).

54. E. Sedov, J. McCarthy, E. Koren, Y. Fuchs, Dev. Cell 57, 1442–1452 (2022).

55. J. Zhang et al., Obstet. Gynecol. 116, 1281–1287 (2010).

56. W. Ding et al., Am. J. Obstet. Gynecol. 227, 627.e1–627.e23 (2022).

57. C. O. Dos Santos, E. Dolzhenko, E. Hodges, A. D. Smith, G. J. Hannon, Cell Rep. 11, 1102–1109 (2015).

58. T. Kumar et al., Nature 620, 181–191 (2023).

59. B. Virassamy et al., Nature 649, 449–459 (2026).

60. S. M. Grandi et al., Circulation 139, 1069–1079 (2019).

61. W. Gohir et al., J. Physiol. 597, 3029–3051 (2019).

62. A. M. Weckman, M. Ngai, J. Wright, C. R. McDonald, K. C. Kain, Front. Microbiol. 10, 1924 (2019).

63. H. Deshmukh, S. S. Way, Annu. Rev. Pathol. 14, 185–210 (2019).

64. X. Huang et al., Front. Immunol. 12, 686676 (2021).

65. J. Weng, C. Couture, S. Girard, Biology 12, 402 (2023).

---

ACKNOWLEDGMENTS

We thank all past and present members of our laboratories for helpful discussions and critical feedback throughout the preparation of this Review. Figures drafts were designed and prepared by the authors using BioRender.com. Funding: D.C. was supported by institutional start-up funding from INFINITY (Inserm, CNRS, and Université de Toulouse). Y.B. was supported by the Institut Pasteur and the Agence Nationale de la Recherche Chaire d'excellence en biologie/santé (ANR-23-CHBS-0008). Author contributions: D.C. and Y.B. both researched the literature and wrote the manuscript. Competing interests: The authors declare that they have no competing interests. License information: Copyright © 2026 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www.science.org/about/science-licenses-journal-article-reuse

Submitted 15 July 2026; accepted 6 August 2026

10.1126/science.aeh1405

REVIEW

X and Y chromosomes as determinants of aging and disease

Dena B. Dubal $^{1,2*}$ , Fabrisia Ambrosio $^{3,4}$ , Michael B. Stout $^{5,6}$ , Dan Theodorescu $^{7,8*}$

Sex differences have traditionally been attributed to gonadal hormones, but accumulating evidence shows that sex chromosomes exert distinct, cell-autonomous effects on health, aging, and disease. XX and XY complements alter gene dosage, epigenetic programming, immune signaling, metabolism, and cellular stress responses, which shape aging and disease trajectories in women and men. This Review integrates mechanistic advances in sex chromosome biology—including escape from X chromosome inactivation, X reactivation, parent-of-X-origin, X-skew, X- and Y-linked gene functions, and Y loss—with aging, neurologic disease, immunity, cancer, and cardiometabolic disease. We highlight insights from mouse models, human studies, and emerging technologies that disentangle hormone-driven effects from chromosome-mediated effects and discuss translational implications for diagnostics, biomarkers, therapeutics, and clinical trial design, emphasizing the integration of sex chromosome biology into precision medicine.

Biological sex is defined by chromosomal complement (XX or XY), gonadal differentiation, and physiology, all of which profoundly affect human health. It is conceptually separated from gender, which reflects sociocultural identity and lived experience, which we do not discuss here. Biological sex directly influences cellular function, immunity, and metabolism, contributing to mechanisms of aging and disease. Although research has long focused on gonadal hormones, emerging evidence shows that sex chromosomes exert independent and interactive effects through genetic and epigenetic mechanisms across organ systems (1–4).

The XX and XY chromosome complements shape cellular phenotypes through gene dosage, gene expression, and epigenetic processes. The X chromosome regulates fundamental cellular processes in both sexes. In XX individuals, a second X introduces additional biology through X chromosome inactivation (XCI), generating cellular mosaicism and differential gene expression (5). In XY individuals, the Y chromosome directs male sex determination through SRY and carries genes essential for fertility. Beyond reproduction, many Y-linked genes have broader biological functions, including X chromosome paralogs that regulate transcription, chromatin biology, RNA processing, and other cellular processes (6). Consequently, the Y chromosome is increasingly recognized as an important modulator of physiology, aging, and disease (1, 7).

This Review examines how X and Y chromosomes mechanistically shape sex biology across the lifespan, with an emphasis on aging, neurologic and psychiatric conditions, immunity, cancer risk, and cardiometabolic disease. The field is remarkably rich and rapidly advancing; we apologize to authors whose important contributions to these areas could not be cited owing to word count and reference limitations. Although social and environmental determinants of health also contribute to sex differences in aging and disease and may interact with sex chromosome biology, they are beyond the scope of this Review. We consider implications of sex chromosome biology for research design, clinical trials, and personalized therapy. Integrating genetics, genomics, cell biology, and clinical medicine is essential to better understand sex-biased diseases, inform sex-aware precision medicine, and develop new treatments.

---

Mechanistic framework: How sex chromosomes affect biology How a second X chromosome shapes biology

With \~1500 genes making up 5% of the genome in both sexes, the X chromosome is a major contributor to sex differences, yet it remains understudied in aging and disease. In female mammals with two X chromosomes, one is silenced in embryonic development through X-inactive specific transcript (XIST)-mediated XCI, generating an active X (Xa) and an inactive X (Xi) (8). XCI, first proposed by Mary Lyon in 1961, was long viewed as a stable mechanism to equalize X-linked gene expression between XX and XY cells. However, we now know that \~15 to 30% of Xi genes constitutively escape XCI in humans (and 3 to 7% in mice) with variability across genes, cell types, tissues, individuals, and species (9). Constitutive XCI escape enables expression from both the Xa and Xi (Fig. 1), increasing X-linked gene dose in female cells, thus providing a mechanism by which two X chromosomes influence resilience and vulnerability in aging and disease.

One example is KDM6A, a histone demethylase that constitutively escapes XCI and is expressed at higher levels in XX compared with XY brains of both mice and humans. Throughout this Review, findings on KDM6A illustrate how increased X-linked gene dosage influences diseases across organ systems. In humans, increased neural expression of KDM6A, through a genetic variant, is associated with cognitive resilience in Alzheimer's disease (10). In mouse models of Alzheimer's disease, KDM6A overexpression reverses cellular and cognitive impairments (10). Notably, these protective effects extend to males, which highlights that understanding mechanisms of XCI can benefit both sexes.

Reactivation of the Xi chromosome [X chromosome reactivation (XCR)] has emerged as a dynamic, lifespan-dependent process and a source of sex differences that may contribute to aging and disease (Fig. 1). Advances in single-cell and single-nucleus RNA sequencing, combined with allele-specific approaches, now enable precise measurement of gene expression from the Xi versus the Xa in mice (11, 12) and humans (13). These studies show that XCI is not static. Across multiple organ systems in mice and humans, aging promotes XCR by relaxing Xi silencing while also further repressing some Xi genes (11–13). Because XCI depends on DNA methylation, chromatin structure, and noncoding RNAs, age-related disruption of these programs can dynamically alter Xi gene expression.

In XX individuals, XCR provides an additional mechanism for increased X-linked gene expression with aging and potentially in disease. One example is Plp1, a key myelin gene that undergoes XCR in the aging mouse brain. As a result, Plp1 expression is higher in aged XX versus XY brains, mirroring sex differences observed in certain human brain regions (11). Elevating Plp1 improves cognition in aged mice of both sexes (11), which highlights the therapeutic potential of XCR.

XCI during development creates cellular mosaicism in XX individuals, with each cell expressing either the maternal (Xm) or paternal (Xp) X as the active X (Fig. 1). By contrast, XY cells express only Xm, a fundamental source of sex difference. Mosaicism ranges from balanced Xm/Xp expression to X-skew, in which most cells preferentially express either the Xm or the Xp (14) (Fig. 1). Mosaicism may buffer dysfunction during aging and disease by increasing genetic and epigenetic diversity, whereas X-skew may increase vulnerability. For example, in Duchenne muscular dystrophy, X-skew toward the mutant X drives disease manifestation in affected females (15). In humans, age-acquired X-skew is prevalent with advancing age in the blood of women and associates with adverse health outcomes (16), which raises the possibility that an individual woman's X-skew could serve as a biomarker of her disease risk.

Fig. 1. Mechanisms of two X chromosomes. Major features of how two X chromosomes shape biology include constitutive escape from XCI, XCR, parent-of-X-origin effects (including X-skew), X chromosome aneuploidies, and mLOX. These X mechanisms influence aging, brain disorders, cancer, immunity, and cardiovascular disease. Representative findings and proposed mechanisms linking biology of two X chromosomes to sex differences are highlighted. AD, Alzheimer's disease.

Parent-of-X-origin refers to the biological consequences of expressing the maternal (Xm) or the paternal (Xp) X chromosome. Alongside XCI escape and XCR, it represents a distinct mechanism by which a second X chromosome shapes health, aging, and disease. Parent-of-X-origin influences DNA methylation and gene expression, including X-linked imprinting, in which genes are selectively silenced on either Xm or Xp, as demonstrated in mice $17, 18$ . Also in mice, preferential activation of Xm—causing X-skew—impairs cognition and accelerates brain aging, in part through the silencing of Xm-linked cognition-related genes $17$ . Consistent with this, up-regulation of these Xm-silenced genes improves cognition in aged XX mice $17$ . Emerging allele-specific genomic approaches combined with parent-offspring trio studies may enable precise mapping of parent-of-X-origin effects on gene expression and imprinting in humans.

Mosaic loss of the X chromosome (mLOX) (Fig. 1) is an acquired somatic event in which one X chromosome is physically lost in a subset of XX cells. mLOX increases with age in female leukocytes (rising to $>35\%$ after age 80) (19), preferentially affects Xi (20), and reduces expression of XCI escape genes. Loss of either Xm or Xp in mLOX may also generate parent-of-X-origin effects analogous to X-skew. mLOX associates with smoking and increased leukemia risk (19), but its broader effects in female aging and disease remain largely unknown.

---

How the Y chromosome shapes biology

The Y chromosome, despite its low gene content, contains several functionally important loci extending beyond the sex-determining region

SRY, which induces testicular development (Fig. 2) (21). Mutations in several Y gene loci cause testicular disorders (not reviewed here). Y-linked genes, including TSPY, ZFY, and UTY (KDM6C), regulate chromatin, immune signaling, transcription, and other fundamental processes, implicating the Y chromosome in aging, cancer, and other nonreproductive diseases (7). Several of these Y genes are dosage-sensitive paralogs of X-linked genes. For example, UTY (KDM6C) and KDM5D regulate chromatin through enzymatic and scaffolding functions, shaping methylation landscapes and transcriptional output. Although UTY has limited demethylase activity relative to its X paralog UTX (KDM6A), it retains functional relevance through interactions with transcriptional core-pressors such as TLE1, which affects cell lineage programs and immune gene expression implicated in sex-biased disease biology (22).

With aging, many of these Y-linked genes are lost through mosaic loss of the Y chromosome (mLOY), an important mechanism linking Y chromosome biology to aging and disease. For clarity, we use LOY to denote Y chromosome loss in individual cells and mLOY for mosaic loss in cell populations. mLOY (Fig. 2) in peripheral blood mononuclear cells (PBMCs) is the most common somatic chromosomal alteration in aging men (23), becoming common by the seventh decade (7). This builds on earlier observations that age-related clonal mosaicism accumulates with aging and associates with cancer and other age-related diseases (24, 25). Smoking is the strongest known, dose-dependent, and partially reversible environmental risk factor for mLOY (26), with additional contributions from inherited genetic susceptibility, DNA-damaging environmental exposures, and genomic instability. mLOY frequently co-occurs with clonal hematopoiesis of indeterminate potential (CHIP), affecting nearly 40% of men with CHIP (27), suggesting shared age-related mechanisms of clonal expansion and contributing to adverse health outcomes associated with both conditions (28). Single-cell and lineage-sorted analyses detect mLOY across normal tissues and link it to immune phenotypes (29). Its prevalence is increased in men with cardiovascular disease, cancers, severe infections, and Alzheimer's disease, which supports its role as both a biomarker and potential contributor to age-related disease (30, 31).

Experimental studies support a causal role for mLOY in disease. In mouse models, mLOY accelerates cardiomyopathy and reduces lifespan, implicating macrophage-mediated mechanisms in cardiovascular disease (32). The exact mechanism of LOY is not fully understood but distinctive features of the human Y chromosome centromere likely contribute (33). In tumors, LOY remodels the microenvironment (34), promotes CD8+ T cell exhaustion, and associates with worse prognosis but greater responsiveness to immune checkpoint inhibitors (28, 34).

Mechanistically, loss of the Y genes disrupts the X-Y paralog balance and remodels chromatin, immune, and stromal programs. Whereas KDM5D and UTY generally act as tumor suppressors (28, 34), their effects are context dependent, with Kdm5d overexpression in a mouse colon cancer model making tumors more invasive in vivo (35). Likewise, loss of Y genes, particularly UTY in mouse and human macrophages, associates with transcriptional and paracrine changes that promote activation of cardiac fibroblasts and fibrosis through transforming growth factor- $\beta$ (TGF- $\beta$ )-dependent mechanisms (36). These findings provide a mechanistic link between mLOY and cardiovascular disease (32).

---

Dissection of X and Y chromosome biology: Mouse models and human variation

The four core genotypes (FCG) model, originally developed in mice (37) and now extended to rats (38), genetically uncouples the effects of sex chromosomes from gonadal sex. Deletion of the testis-determining gene (Sry) from the Y chromosome and its transgene insertion on an autosome $[XY^{-Sry}(Sry^{+})]$ generates XX and XY animals with either ovaries or testes (37). Crossing $XY^{-Sry}(Sry^{+})$ males with wild-type XX females yields four genotypes: XX and $XY^{-Sry}$ with ovaries and XX and $XY^{-Sry}$ with testes. This design distinguishes sex chromosome effects (XX versus XY with matched gonads) and sex hormone effects (ovaries versus testes with identical sex chromosomes) (37). Sex chromosome effects (XX versus XY) may result from differences in X chromosome number (X dosage, X imprinting, and indirect effects of XCI) or from the presence of the Y chromosome, which can be further dissected using complementary models outlined below. FCG mice have been widely used to define how chromosomal and gonadal sex shape aging, neurological disorders, autoimmunity, cancer, and cardiovascular and metabolic diseases (39).

Sox9 enhancer mouse models provide an alternative strategy to uncouple gonadal and chromosomal sex while preserving the endogenous Sry locus, given that Sry may have functions beyond testis determination. Rather than relocating Sry, these models manipulate the autosomal Sox9 enhancer (Enh13), Sry's principal downstream effector, generating XY mice with ovaries (40) or XX mice with testes (41).

The XY\* mouse model complements the FCG and Sox9 enhancer models by dissecting the contributions of X chromosome dosage (including X genes, X imprinting, and indirect effects of XCI) and Y chromosome presence (including Y-linked gene effects). XY\* males carry a Y chromosome with an altered pseudoautosomal region that undergoes abnormal meiotic combination with the X (37). Crossing XY\* males with wild-type XX females yields four genotypes: XO and XX with ovaries and XY and XXY mice with testes. The XY\* model has become a standard tool to define how sex chromosome dosage contributes to diverse disease processes (39).

To model acquired Y chromosome loss, mouse models use CRISPR-Cas9-mediated Y chromosome deletion in donor hematopoietic stem cells before transplantation into irradiated recipients, generating chimeras with $>50\%$ mLOY in blood cells (32, 42). Because mLOY also occurs in nonhematopoietic tissues, inducible tissue-specific mLOY models are needed to further define its causal roles in aging and disease.

Parallel advances have expanded the toolkit for studying X chromosome biology. Mouse models now enable mechanistic dissection of constitutive XCI escape (11), XCR (11), parent-of-X-origin effects (17, 43), and X-skew (17). When paired with aging or disease models, they can reveal causal mechanisms underlying female heterogeneity and sex differences. Key approaches include allele-specific systems that distinguish expression of Xa from Xi, enabling precise study of XCI escape and XCR (11); dual-reporter models that fluorescently label Xm and Xp to visualize and probe cellular mosaicism (43); and inducible X-skew models (17). Additional strategies use small molecules, adeno-associated virus (AAV), and CRISPR to modulate X-linked gene expression. Integrated with bioinformatics and functional assays, these tools provide high-resolution study of X-linked causative mechanisms in aging and disease and may uncover therapeutic targets benefiting both women and men.

Human sex chromosomal aneuploidies (SCAs) provide human-derived evidence for how altered X and Y chromosome dosage influences fundamental biology, complementing findings from mouse models. SCAs are prevalent yet underdiagnosed, affecting \~1 in 400 newborns (44). They arise from gain or loss of whole or partial sex chromosomes, often owing to meiotic nondisjunction (45). Well-characterized SCAs include Turner syndrome (XO), Klinefelter syndrome (XXY), trisomy X (XXX), XYY, and XYYY (45). These conditions disrupt endocrine, cardiometabolic, musculoskeletal, and neurological systems through altered genome function (46). Mechanistically, SCA phenotypes reflect dysregulated expression of dosage-sensitive genes, particularly those escaping XCI (including KDM6A, KDM5C, and ZFX) and pseudoautosomal genes (including SLC25A6, ASMTL, and GTPBP6). These expression changes perturb chromatin remodeling, translation, meiotic processes, and telomere maintenance, identifying candidate drivers of SCA-associated phenotypes (46).

---

Aging and disease: Differences attributable to sex chromosomes Aging

Worldwide, women live longer than men (47, 48). This female survival advantage persists even under extreme conditions, including famine and epidemics, where girls outlive boys (47). It also holds across major diseases—particularly neurodegenerative disorders, including Alzheimer's disease (49)—and extends across much of the animal kingdom (50), which suggests a conserved biological basis. One evolutionary explanation posits that longer-lived mothers increase offspring survival and reproductive success. Identifying mechanisms that promote longevity may reveal pathways that extend health span, or years lived in good health.

Experimental models support a sex chromosomal contribution to longevity and resilience. In FCG mice, the XX complement increases survival, regardless of gonadal sex (51). However, the underlying mechanisms, whether XCI escape, XCR, parent-of-X-origin effects, or others, remain unresolved. In mouse studies of brain aging, Xm accelerates neuronal aging (17), which raises the possibility that female mosaicism of Xm and Xp confers resilience compared with the single Xm in males. In parallel, mLOY in men associates with increased mortality from midlife age onward, although, paradoxically, it is enriched in centenarians (52), which leaves unresolved whether mLOY is a driver, biomarker, or adaptive consequence of aging.

Aging is the strongest risk factor for diseases across organ systems. Women exhibit resilience during normal brain aging yet remain disproportionately affected by Alzheimer's disease (53, 54). These observations suggest that mechanisms promoting longevity may also underlie resilience during healthy aging and early disease but become insufficient as disease-related pathology accumulates. Defining X-linked mechanisms underlying longevity may uncover resilience pathways that can be harnessed to boost health span and reduce age-related disease.

Fig. 2. The Y chromosome in human health and disease. The Y chromosome contributes to male health and disease through many mechanisms, including key Y genes, mLOY, and Y aneuploidies. Representative associations of mLOY with cancer, cardiac fibrosis, Alzheimer's disease, and immune dysfunction are shown.

---

Neurological, neurodevelopmental, and psychiatric disorders

The brain is enriched for X-linked genes that are more highly expressed than autosomal genes (55), which makes it particularly sensitive to X chromosome biology. Consistent with this, mutations in more than 90 X-linked genes cause intellectual disability, and many more contribute to other neurological phenotypes, most of which are not reviewed here. In normal aging without dementia, women show slower cognitive decline than men, likely influenced by X effects (53) because a second X improves cognition in aging mice of both sexes (56). In mice, aging induces XCR in XX brains, increasing expression of genes such as Plp1, enhancing cognition in both sexes (11). Meanwhile, Xm-skew accelerates mouse brain aging (17), which suggests that Xm and Xp mosaicism in females may confer resilience compared with the single Xm in males.

In neurodegenerative diseases, X chromosome mechanisms promote both resilience and vulnerability, depending on disease stage and context. In Alzheimer's disease, women live longer and show early clinical resilience, despite accumulating more neuropathology, but they later experience steeper cognitive decline (54, 57, 58). FCG studies of Alzheimer's amyloid mouse models show that an XX complement improves survival and cognition, independent of gonadal sex (10). XY\* studies attribute these benefits to increased X dosage, partially owing to higher KDM6A levels, and to X-Y interactions (10), implicating X chromosome mechanisms in resilience.

By contrast, during the later tau phase of the disease, X-linked genes may promote pathology; for example, Usp11 promotes tau accumulation (59) and Tlr7 contributes to tau-induced demyelination (60) in mice. What determines whether X chromosome biology is protective or pathogenic—including disease state, cell type specificity, and the underlying proteinopathy—and how these mechanisms contribute to women's greater vulnerability to tauopathy remain largely unknown.

Human genetic studies implicate sex chromosomes across neurodegenerative diseases. X chromosome-wide association studies identify X-linked loci in Alzheimer's disease, Parkinson's disease, amyotrophic lateral sclerosis, and other disorders (61), whereas mLOY associates with increased Alzheimer's disease (1) and amyotrophic lateral sclerosis risk (62) in men.

Sex chromosomes affect susceptibility and disease severity across neurologic diseases. Stroke disproportionately affects women. In FCG mice, an XX complement increases infarct size with aging, independent of gonads (63), whereas in humans, mLOY associates with worse outcomes (64). In multiple sclerosis, women show higher susceptibility, whereas men experience more severe neurodegeneration (65). Consistent with this, FCG studies of multiple sclerosis models such as experimental autoimmune encephalitis show that two X chromosomes exacerbate immune activation and susceptibility, partly thru lymphocytic KDM6A, whereas the XY complement worsens neurodegeneration—mirroring human sex differences (65). Roles of sex chromosomes in pain, epilepsy, neuromuscular, and movement disorders remain important areas for investigation.

In neurodevelopmental and psychiatric disorders, X-linked genes influence brain development and circuit function. For example, mutations in MECP2 (66) and FMR1 cause distinct neurodevelopmental syndromes with cognitive impairment and autism-related features. Autism in general is male biased, consistent with a protective effect of a second X in females (67), possibly through Xi and cellular mosaicism. By contrast, depression and anxiety are female biased. Although SCAs associate with elevated psychiatric comorbidity (68), the specific X- and Y-linked mechanisms underlying psychiatric risk and resilience across disorders—depression, anxiety, schizophrenia, bipolar, posttraumatic stress, and beyond—merit future study.

---

Autoimmunity and infection

Women mount stronger innate and adaptive immune responses than men (4), which enhances their resistance to infection and vaccine response but also increases their susceptibility to autoimmune diseases, such as systemic lupus erythematosus, rheumatoid arthritis, and multiple sclerosis (MS) (69). Autoimmune diseases arise from loss of immune tolerance and autoreactive T and B cells, leading to chronic inflammation and organ damage. X chromosome dosage contributes directly to these differences.

The X chromosome is enriched for immune-related genes, such as TLR7 and FOXP3, and some immune cells display noncanonical epigenetic XCI maintenance features (4, 70). Autoimmune disease scales with X chromosome number in FCG mice and in individuals with Klinefelter syndrome and trisomy X (4, 71). Conversely, certain diseases, including autoimmune thyroiditis and type 1 diabetes, are increased in Turner syndrome (72), consistent with haploinsufficiency of X-linked immune genes and reduced immune regulation.

In individuals with more than one X chromosome, XCI escape of immune genes provides a plausible mechanism for autoimmunity (2). During XCI, XIST forms ribonucleoprotein complexes that can act as autoantigens in mice and humans (73). In mice, Tlr7 recognizes RNA within these complexes (73, 74), and the resulting Toll-like receptor (TLR) signaling promotes atypical memory B cell expansion and type 1 interferon production, thereby promoting disease activity (73, 75, 76). Autoantibodies to XIST ribonucleoprotein complexes are detectable in patient sera (73), which highlights their potential as diagnostic biomarkers for early disease detection.

Though less studied, LOY has also been implicated in sex-based differences in autoimmune diseases, such as autoimmune thyroiditis and primary biliary cirrhosis (1). Moreover, mLOY in hematopoietic cells associates with increased risk of infectious diseases and may contribute to elevated risk and severity of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection in men (1).

---

Cancer

Men show higher incidence and mortality across many common cancers, even after accounting for behavioral risk factors. Epidemiologic and molecular studies attribute this to both endocrine influences and intrinsic effects of sex chromosomes on tumor cells and the immune microenvironment. Male predominance occurs in bladder, hepatocellular, gastric, and esophageal cancers, and sex-skewed outcomes in bladder cancer, melanoma, and glioblastoma remain incompletely explained by the environment (7).

mLOY provides one mechanistic link between cancers and sex biology. mLOY in PBMCs associates with increased cancer risk and mortality and frequently co-occurs with CHIP mutations (7), reflecting shared influences of aging, DNA damage, and selection of advantaged hematopoietic clones. In tumors, pan-cancer analyses associate mLOY with poor prognosis (34, 77, 78), though its role as driver versus passenger of this aggressive biology was previously unclear. Recent work mechanistically demonstrated that LOY drives a global loss of tumor suppressive and immune-regulatory functions. Y-deficient tumor cells remodel the immune microenvironment, promote CD8+ T cell exhaustion, and enhance responses to PD-1 checkpoint blockade in both animal models and patients (34). Loss of UTY and KDM5D contributes substantially to these LOY effects. LOY was also discovered in nonmalignant hematopoietic and nonhematopoietic cells within the tumor microenvironment, particularly in mLOY tumors, and, when present, predicts worse outcomes (28). Recently, LOY was found in histologically normal tissue adjacent to epithelial cancers, which suggests that LOY may represent an early hallmark of male oncogenesis and a potential biomarker for future or impending malignant transformation in specific tissues (79).

Mutations of X-linked genes also contribute to cancers in both sexes. The X-linked DDX3X and its Y-linked paralog DDX3Y regulate transcriptional programs and can function as tumor suppressors, oncogenes, or cancer modifiers (80). Likewise, the XCI escape gene KDM6A (UTX) is a tissue-specific tumor suppressor frequently mutated or deleted in bladder cancer, renal cell carcinoma, and hematologic cancers (81). Its higher dosage may partially protect women (82), whereas KDM6A loss fosters malignant transformation and disease progression through catalytic and scaffolding functions (83). Limited compensation by its Y-linked paralog UTY (KDM6C) may increase vulnerability in XY cells, helping explain why LOY or UTY deletion promotes transformation or progression in established tumors (83, 84).

Sex differences also extend to antitumor immunity. In glioblastoma models and patients, KDM6A and XCI escape support female T cell resilience, whereas men show greater progenitor-exhausted CD8+ T cells but sometimes stronger PD-1 responsiveness (85).

---

Cardiovascular and metabolic diseases

Sex differences in cardiovascular disease and cardiometabolic traits evolve across the lifespan. Men have higher myocardial infarction incidence at younger ages, but this gap narrows with age, and women experience worse outcomes later in life, including higher mortality after myocardial infarction and heart failure (86, 87). Cardiometabolic risk factors, including adiposity, lipid profiles, and cardiac function, also diverge by sex with aging (88, 89). Although hormonal influences contribute, experimental and population data clearly support independent roles for the sex chromosomes.

In FCG and XY\* mouse models, the XX complement drives cardiac vulnerability; XX mice show greater ischemic injury than XY mice, independent of gonads and driven by increased X dosage (90). XX complement also regulates metabolic traits (3) and promotes atherosclerosis (3, 91). In FCG mice, an XX complement promotes adiposity, weight gain, fatty liver, hyperlipidemia, and hyperinsulinemia under high-fat conditions (92, 93). X effects on adiposity are recapitulated in XY\* models, caused by increased X dosage rather than Y presence (93).

Mechanistically, XCI escapee KDM5C drives adiposity in mice and adipocyte KDM5C expression and genetic variation associate with body mass in humans (92). Age-related X-skew also associates with cardiovascular risk (16). Human data in individuals with SCA further support these links, including dyslipidemia in boys with Klinefelter syndrome (94).

The Y chromosome also modifies cardiovascular risk. In mice and humans, mLOY associates with cardiomyopathy, cardiac fibrosis, and increased mortality (32). In mice, loss of Uty promotes profibrotic macrophage activation and TGF- $\beta$ 1-driven cardiac remodeling (36). Together, these findings identify sex chromosome biology as a key determinant of cardiovascular disease and reveal clinical opportunities, including mLOY-based risk stratification, TGF- $\beta$ 1 inhibition, and targeting XCI escape genes.

---

Clinical and translational implications

Integrating sex chromosome biology into the clinical realm offers tremendous potential for precision diagnostics and therapeutics (Fig. 3). Realizing this potential will require addressing key knowledge gaps, including how sex chromosome states in blood reflect disease-relevant tissues, such as brain, heart, and immune cells, and how specific X or Y chromosome genes can be therapeutically targeted.

Diagnostics will require robust genomic, transcriptomic, and potentially protein assays to detect mLOY, quantify XCI escape, and assess X-skew and XCR where feasible (Fig. 3). Whole-genome sequencing and high-density genotyping arrays sensitively detect mLOY and estimate clonal fraction, which correlates with disease risk and outcomes (23, 28, 34). Clinical implementation will require standardizing mLOY, XCI escape, and other sex chromosomal measures for integration into electronic health records and clinical decision support systems (95).

Sex chromosome biomarkers hold promise for risk prediction and surveillance. mLOY is emerging as a noninvasive biomarker linked to malignancy, cardiovascular disease, and neurodegenerative diseases, including Alzheimer's disease (1, 7). High mLOY fractions often co-occur with CHIP mutations, which suggests that combined assays may improve cancer surveillance and risk stratification (27, 96, 97). Likewise, mLOY and XCI escape alter tumor immune microenvironments and predict responsiveness to immune checkpoint blockade, which supports development of their use in treatment stratification (34, 85). More broadly, establishing associations of mLOY, XCI escape, XCR, X-skew, and mLOX with aging and disease outcomes—and then integrating them with established biomarkers—may improve disease prediction and therapeutic selection. Clinical translation will require harmonized detection thresholds, prospective validation in diverse populations, and demonstration of added predictive value beyond existing biomarkers and risk factors.

Fig. 3. Clinical and translational implications of sex chromosome biology. Integrating X and Y chromosome biology into diagnostics, therapeutics, and clinical trial design is needed to improve patient risk stratification, develop targeted therapies, and advance precision medicine across aging and disease.

Sex chromosome mechanisms offer compelling therapeutic targets that could benefit both sexes (Fig. 3). When we understand the molecular mechanisms of what makes one sex more resilient or vulnerable in aging or disease, we unravel pathways that may be targeted for therapy. Potential strategies include simulating X chromosome effects to promote longevity, stimulating XCR and increasing X-linked PLP1 in the aging brain to improve cognition, inhibiting XIST-driven autoantigen pathways in autoimmunity, boosting KDM6A-mediated antitumor immunity, and modulating XCI escape in cardiovascular disease. Because Y-deficient tumor cells promote CD8+ T cell exhaustion, these tumors are more responsive to immune checkpoint blockade, which provides a way to personalize therapy on the basis of the tumor Y chromosome status (34). These examples represent a small subset of emerging opportunities. Deeper understanding of X and Y chromosome biology opens a frontier for therapeutic development and offers a path to innovative medicines that improve human health.

Precision medicine should leverage X and Y chromosome gene status as biomarkers (Fig. 3). Precision oncology already uses features such as KDM6A mutations and mLOY fraction to stratify patients in selected immunotherapy trial settings, supporting their potential clinical utility. Future studies and trials should therefore incorporate sex chromosome genotyping, mLOY assessment, and immune profiling while treating biological sex as a mechanistic modifier of therapeutic response—given the growing evidence that X- and Y-linked biology and hormone-dependent mechanisms influence outcomes (98–100).

Trials should also integrate sex chromosome-related biomarker endpoints and be powered to detect sex-by-treatment interactions. Cell lines and biobanked tissues should be annotated for donor sex and chromosomal status, including tissue karyotype, LOY clonal fraction when feasible, and data on XCI escape to permit downstream association analyses and cross-cohort analysis. Finally, precision medicine frameworks must integrate sex chromosomes, circulating sex hormones, and environmental exposures because these factors jointly shape therapeutic response.

---

Limitations, open questions, and future directions

Sex chromosome effects on aging and disease are context dependent, varying by cell type, developmental stage, and environment. However, the field lacks comprehensive cross-tissue, cell type-specific maps of XCI escape, XCR, parent-of-X-origin, X-skew, mLOX, mLOY, and X-linked and Y-linked gene expression across development, aging, and disease. Broad allele-specific maps of XCI escape and XCR are also needed to link X chromosomal regulation with functional outcomes across tissues.

mLOY and mLOX increase with age through genetic and environmental influences, but their temporal dynamics and causal roles remain incompletely understood. Although single-cell studies suggest that these processes are dynamic and potentially reversible, most human epidemiological studies remain observational and cannot distinguish whether mLOY is a causal driver or simply a biomarker of biological aging (101). Interpretation is further limited by heterogeneous detection methods, single-time point measurements, frequent co-occurrence with CHIP, and predominantly European ancestry cohorts, all of which limit generalizability. Longitudinal studies combining standardized assays; repeated sampling; diverse cohorts; and integrated genomic, chromatin, and epigenomic profiling are needed to define the trajectories, biological consequences, and clinical relevance of mLOY, mLOX, XCI escape, XCR, X-skew, and other age- and disease-related changes in sex chromosome biology. Studies should also clarify how race, ethnicity, social determinants of health, and environmental exposures influence these processes.

Future studies should define functions of individual X- and Y-linked genes, X-linked imprinting, X-X and X-Y interactions across tissues, aging, and diseases while incorporating sex chromosomes into genome-wide association studies (GWASs) and other genomic analyses. Defining the consequences of XCI escape, mLOX, and X-skew will improve our understanding of heterogeneity and individual trajectories in female aging and disease. Given convergent evidence implicating KDM6A across diseases, cell type-specific targeting of this pathway may offer new medicines. Finally, clinical trials must be designed to detect whether treatments differ in XX and XY individuals and establish standards for sex-aware biomarker validation, reporting, consent, and equitable implementation.

---

Conclusions

The X and Y chromosomes govern fundamental biology throughout the lifespan and across diseases. Although we have begun to uncover their mechanisms, much remains unexplored. Substantial advances would be possible if we rigorously tackled sex chromosomal-driven mechanisms of aging and disease—expanding cutting-edge toolkits in genomics, single-cell biology, functional studies, informatics, and machine learning—and deeply interrogated meaningful sex differences in the human condition. These discoveries may reveal robust biomarkers and therapeutic targets across aging, neurologic and psychiatric disease, neurodevelopmental disorders, autoimmunity, cardiovascular disease, cancer, and beyond in a sex-aware manner. We would advance clinical precision, improve equity and health outcomes, and deepen our understanding of biology to drive transformative gains for science, medicine, and the human condition.

---

REFERENCES AND NOTES

1. B. Bruhn-Olszewska et al., Nat. Rev. Genet. 26, 320–335 (2025).

2. C. Libert, L. Dejager, I. Pinheiro, Nat. Rev. Immunol. 10, 594–604 (2010).

3. C. B. Wiese, R. Avetisyan, K. Reue, Trends Endocrinol. Metab. 34, 652–665 (2023).

4. K. S. Forsyth, N. Jiwrajka, C. D. Lovell, N. E. Toothacre, M. C. Anguera, Nat. Rev. Immunol. 24, 487–502 (2024).

5. X. Deng, J. B. Berletch, D. K. Nguyen, C. M. Disteche, Nat. Rev. Genet. 15, 367–378 (2014).

6. D. W. Bellott et al., Nature 508, 494–499 (2014).

7. H. A. Abdel-Hafiz, L. Hoelzen, D. Theodorescu, Nat. Rev. Cancer 26, 586–603 (2026).

8. J. T. Lee, Nat. Rev. Mol. Cell Biol. 12, 815–826 (2011).

9. T. Tukiainen et al., Nature 550, 244–248 (2017).

10. E. J. Davis et al., Sci. Transl. Med. 12, eaaz5677 (2020).

11. M. Gadek et al., Sci. Adv. 11, eads8169 (2025).

12. S. Hoelzl, T. P. Hasenbein, S. Engelhardt, D. Andergassen, Nat. Aging 5, 984–996 (2025).

13. C. Rocca et al., bioRxiv 2025.12.09.693282 [Preprint] (2026).

14. J. M. Amos-Landgraf et al., Am. J. Hum. Genet. 79, 493–499 (2006).

15. E. Viggiano, M. Ergoli, E. Picillo, L. Politano, Hum. Genet. 135, 685–698 (2016).

16. A. L. Roberts et al., eLife 11, e78263 (2022).

17. S. Abdulai-Saiku et al., Nature 638, 152–159 (2025).

18. W. Davies et al., Nat. Genet. 37, 625–629 (2005).

19. A. Liu et al., Nature 631, 134–141 (2024).

20. M. J. Machiela et al., Nat. Commun. 7, 11843 (2016).

21. H. Skaletsky et al., Nature 423, 825–837 (2003).

22. G. Özden-Yılmaz et al., Cells 12, 836 (2023).

23. L. A. Forsberg et al., Nat. Genet. 46, 624–628 (2014).

24. K. B. Jacobs et al., Nat. Genet. 44, 651–658 (2012).

25. C. C. Laurie et al., Nat. Genet. 44, 642–650 (2012).

26. J. P. Dumanski et al., Science 347, 81–83 (2015).

28. X. Chen et al., Nature 642, 1041–1050 (2025).

29. J. Mattisson et al., BMC Genomics 25, 243 (2024).

27. S. Fawaz et al., eLife 13, RP96150 (2024).

30. X. Guo, Cell. Mol. Life Sci. 78, 7081–7084 (2021).

31. E. Berson et al., Nat. Commun. 14, 4947 (2023).

32. S. Sano et al., Science 377, 292–297 (2022).

33. P. Ly et al., Nat. Cell Biol. 19, 68–75 (2017).

34. H. A. Abdel-Hafiz et al., Nature 619, 624–631 (2023).

35. J. Li et al., Nature 619, 632–639 (2023).

36. K. Horitani et al., Nat. Cardiovasc. Res. 3, 343–355 (2024).

37. P. S. Burgoyne, A. P. Arnold, Biol. Sex Differ. 7, 68 (2016).

38. A. P. Arnold et al., Biol. Sex Differ. 17, 75 (2026).

39. A. P. Arnold, Neurosci. Biobehav. Rev. 119, 1–8 (2020).

40. M. Ridnik et al., Nucleic Acids Res. 52, 5514–5528 (2024).

41. E. Abberbock et al., Nat. Commun. 17, 3186 (2026).

42. Q. Zhang et al., JCI Insight 7, e153768 (2022).

43. H. Wu et al., Neuron 81, 103–119 (2014).

44. A. Berglund, K. Stochholm, C. H. Gravholt, Am. J. Med. Genet. Part C 184, 202–215 (2020).

45. D. Skuse, F. Printzlau, J. Wolstencroft, in Neurogenetics, Part I, D. H. Geschwind, H. L. Paulson, C. Klein, Eds., vol. 147 of Handbook of Clinical Neurology (Elsevier, 2018), pp. 355–376.

C. Klein, Eds., vol. 147 of Handbook of Clinical Neurology (Elsevier, 2018), pp. 355–376.

46 M. Legue et al. Mol Med. 32, 11 (2026)

47. V. Zarulli et al., Proc. Natl. Acad. Sci. U.S.A. 115, E832–E840 (2018).

48. United Nations, World Population Prospects, Online edition, “Female life expectancy at birth - UN WPP,” processed by Our World in Data (2024); https://ourworldindata.org/grapher/life-expectancy-of-women-vs-life-expectancy-of-men.

49. J. B. Lusk et al., JAMA Neurol. 82, 1048–1056 (2025).

50. A. M. Bronikowski et al., Aging Cell 21, e13542 (2022).

51. E. J. Davis, I. Lobach, D. B. Dubal, Aging Cell 18, e12871 (2019).

52. D. M. Hozakowska-Roszkowska et al., Mech. Ageing Dev. 222, 111979 (2024).

53. D. B. Dubal, C. T. Murphy, Y. Suh, B. A. Benayoun, Neuron 113, 2–6 (2025).

54. K. B. Casaletto, D. B. Dubal, JAMA 336, 106–108 (2026).

55. D. K. Nguyen, C. M. Disteche, Nat. Genet. 38, 47–53 (2006).

56. F. Marino, D. Wang, G. E. Merrihew, M. J. MacCoss, D. B. Dubal, bioRxiv 2024.07.26.605328 [Preprint] (2024).

57. E. M. Arenaza-Urquijo et al., Alzheimers Dement. 20, 5695–5719 (2024).

58. D. B. Dubal, in Sex Differences in Neurology and Psychiatry, R. Lanzenberger, G. S. Kranz, I. Savic, Eds., vol. 175 of Handbook of Clinical Neurology (Elsevier, 2020), pp. 261–273.

59. Y. Yan et al., Cell 185, 3913–3930.e19 (2022).

60. C. Lopez-Lee et al., Science 386, eadk7844 (2024).

61. K. Step et al., Front. Genet. 16, 1650259 (2025).

62. W.-M. Su et al., BMC Med. 24, 48 (2026).

63. L. D. McCullough et al., Aging 8, 1432–1441 (2016).

64. M. Dorvall et al., Stroke 54, 2434–2437 (2023).

65. R. Voskuhl, Y. Itoh, J. Exp. Med. 219, e20211488 (2022).

66. A. J. Sandweiss, V. L. Brandt, H. Y. Zoghbi, Lancet Neurol. 19, 689–698 (2020).

67. M. Talukdar, D. C. Page, Nat. Genet. 58, 687–694 (2026).

68. L. Schaffer et al., J. Neurodev. Disord. 16, 56 (2024).

69. W. W. Eaton, N. R. Rose, A. Kalaydjian, M. G. Pedersen, P. B. Mortensen, J. Autoimmun. 29, 1–9 (2007).

70. B. Yu et al., Cell 184, 1790–1803.e17 (2021).

71. K. Liu et al., Arthritis Rheumatol. 68, 1290–1300 (2016).

72. S. Hwang et al., Ann. Med. 57, 2573143 (2025).

73. D. R. Dou et al., Cell 187, 733–749.e16 (2024).

74. C. M. Lau et al., J. Exp. Med. 202, 1171–1177 (2005).

75. T. Kawai, S. Akira, Immunity 34, 637–650 (2011).

76. G. J. Brown et al., Nature 605, 349–356 (2022).

77. P. Müller et al., Comput. Struct. Biotechnol. J. 21, 1573–1583 (2023).

78. M. Qi, J. Pang, I. Mitsiades, A. A. Lane, E. Rheinbay, Cell 186, 3125–3136.e11 (2023).

79. A. Gertych et al., JCI Insight 11, e201996 (2026).

80. M. Gadek, E. H. Sherr, S. N. Floor, Trends Mol. Med. 29, 726–739 (2023).

81. L. Wang, A. Shilatifard, Cancer Cell 35, 168–176 (2019).

82. I. Gažová, A. Lengeling, K. M. Summers, Mol. Genet. Metab. 127, 31–44 (2019).

83. J. Andricovich et al., Cancer Cell 33, 512–526.e8 (2018).

84. L. J. Walport et al., J. Biol. Chem. 289, 18302–18313 (2014).

85. J. Lee et al., Cancer Discov. 13, 2090–2105 (2023).

86. G. Albrektsen et al., JAMA Intern. Med. 176, 1673–1679 (2016).

87. Z. Gao, Z. Chen, A. Sun, X. Deng, Med. Novel Technol. Devices 4, 100025 (2019).

88. H. Ji et al., Circ. Res. 130, 566–577 (2022).

89. J. A. Bell et al., BMC Med. 19, 58 (2021).

90. J. Li et al., Cardiovasc. Res. 102, 375–384 (2014).

91. Y. AISiraj et al., Nat. Commun. 10, 2631 (2019).

92. J. C. Link et al., J. Clin. Invest. 130, 5688–5702 (2020).

93. X. Chen et al., PLOS Genet. 8, e1002709 (2012).

94. S. Davis et al., Andrology 4, 1169–1177 (2016).

95. Y. Tomofuji et al., Cell Genom. 4, 100625 (2024).

96. C.-M. Zhang et al., Front. Immunol. 16, 1642956 (2025).

97. A. A. Z. Dawoud, W. J. Tapper, N. C. P. Cross, Sci. Adv. 9, eade9746 (2023).

98. S. L. Klein, K. L. Flanagan, Nat. Rev. Immunol. 16, 626–638 (2016).

99. S. Chaulagain, J. A. Liu, J. E. McCombs, S. L. Klein, Nat. Immunol. 27, 660–673 (2026).

100. J. C. Redondo-González et al., Clin. Transl. Oncol. 28, 1586–1594 (2026).

101. V. Ljungström et al., Leukemia 36, 889–891 (2022).

---

ACKNOWLEDGMENTS

We acknowledge that space limitations have prevented acknowledgement and recognition of many important contributions to this field, and we apologize to those authors whose notable works we were unable to cite. We thank the organizers and participants of the 2025 National Institute on Aging Workshop on Sex Differences Impacting Human Health Across the Lifespan, which inspired this Review. We are grateful to F. Duncan for her key role in making this Review possible. We also thank E. Kakoshina for creating the original scientific illustrations that served as the basis for the figures in this Review. Funding: This Review was supported by National Institutes of Health (NIH) grants R01AG079176 (D.B.D.), R01AG068325 (D.B.D.), R01AG070035 (M.B.S.), R33AG072137 (M.B.S.), R01AG089455 (F.A.), R01AG087565 (F.A.), and R35CA294022 (D.T.); the US Department of Veteran Affairs, I01BX006239 (M.B.S.); the Simons Foundation (D.B.D.); and the American Federation for Aging Research (D.B.D.). Author contributions: Conceptualization: D.B.D., D.T.; Writing – review & editing: D.B.D., F.A., M.B.S., D.T. Competing interests: D.B.D. has a patent pending for X Chromosome Gene Products that Improve Cognition, is an associate editor for JAMA Neurology, and serves on the JAMA Women's Health Brain Trust. The other authors declare that they have no competing interests. License information: Copyright © 2026 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www.science.org/about/science-licenses-journal-article-reuse

Submitted 2 June 2026; accepted 22 July 2026

10.1126/science.aeh0145

REVIEW

Reframing menopause and menopause hormone therapy as opportunities for healthy aging

Laura L. Gravelsins $^{1\dagger}$ , Mateja Perović $^{1\dagger}$ ,

Madeline Wood Alexander $^{2,3}$ , Tallinn F. L. Splinter $^{2,3}$ ,

Andrew J. McGovern $^{1}$ , Jennifer S. Rabin $^{2,3}$ , Liisa A. M. Galea $^{1,4*}$

Menopause is an important biological transition with up to 72 symptoms that shapes aging trajectories across organs and encompasses \~40% of the female lifespan. Although menopause awareness is increasing, research on strategies for menopause symptom management and optimizing health in the postmenopausal period is lacking. This Review highlights that menopause characteristics, such as type, timing, and symptom burden, influence health outcomes and later-life disease risk. Similarly, menopause hormone therapy comes in >200 varieties that differ by dose, formulation of estrogens and progestogens, and timing, each with their own distinct risk-benefit profiles. These distinctions are often missed in the literature but have considerable repercussions for outcomes. Optimized menopause care necessitates a precision medicine approach that considers menopause symptoms, genetics, and previous hormone exposures.

Menopause occupies up to 40% of the average female lifespan, yet cultural silence around female aging and a paucity of dedicated research have impeded scientific scrutiny of menopause and allowed dogma to substitute for evidence. Menopause is accompanied by decreases in most ovarian hormones, such as estradiol and progesterone, and increases in gonadotropin hormones such as follicle stimulating hormone (FSH), due to interruptions in the negative feedback loop of the hypothalamic pituitary gonadal (HPG) axis. The transition to menopause represents a critical inflection point in aging, characterized by systematic neurological (1), inflammatory (2), and metabolic changes (3) that unfold alongside dynamic, organ-specific changes (4). Menopause, independent of chronological aging, is associated with accelerated aging clocks across multiple organs (1). Despite this, menopause research remains neglected. Analyses show that 0.2% of the total federal investments in health research over 15 years went to menopause-related questions in North America (5, 6). Although research is scarce, extant data show that menopause is not a uniform experience, and these differences can affect health beyond the menopausal transition. Menopause differs in age of onset, type (induced versus spontaneous), length of transition (perimenopausal period), and symptoms (7). Furthermore, menopause hormone therapies (MHT)—first-line medications for treatment of menopause symptoms—vary in timing, formulation, dose, and route of administration (8). Both menopause and MHT heterogeneity matter for health outcomes but remain under-researched. Forgoing study of this heterogeneity does a disservice to the health of the community and to science. This, coupled with the limited scientific investment and gendered taboos surrounding aging (9, 10), has hindered progress, and allowed long-standing assumptions of menopause health to persist largely unchallenged. In this Review, the use of female is used to describe those with biological attributes associated with the female sex (XX chromosomes, ovaries), whereas woman is a gendered term that reflects social categorization including self-identification, and is used where consistent with the sampling characteristics of the cited studies.

The symptoms accompanying menopause are diverse and have a pronounced effect on quality of life. Vasomotor symptoms (VMS), such as hot flashes and night sweats, are experienced by 80% of the population during the menopause transition, but more than 70 additional symptoms have been reported, with an average of seven symptoms experienced at a given time (11–13). The diversity and combination of symptoms have fueled debate over whether they are attributable to menopause or natural aging. However, the abruptness of symptom onset and its close temporal relation with the menopause transition signals menopause-specific effects beyond pure association with age. Given that estrogen receptors (ERs) and the progesterone receptor (PR) exist on every organ (e.g., skin, lung, kidney, heart, brain) and most tissues, it is not surprising that the abrupt hormonal changes accompanying menopause can produce such a diverse range of symptoms. Individual differences in receptor type [ERα, ERβ, G protein-coupled ER (GPER), PR isoforms], density, and sensitivity of these receptors likely contribute to the marked variability in symptom burden. For some, the menopausal transition is barely noticeable, but for others, it severely affects their quality of life. Importantly, even in those without symptoms, menopausal timing can influence future disease risk (14, 15).

Despite the availability of highly effective treatments for VMS, including MHT and neurokinin B antagonists, fewer than 15% of symptomatic women receive treatment (13, 16). This treatment gap likely reflects a broader lack of menopause education and support. In Canada, one in two women feel unprepared for the menopause transition and 38% report feeling alone while navigating it (13). The consequences extend beyond health, as \~10% of women leave the workforce because of menopause-related symptoms, including mental health challenges (13) and sleep disturbances (17). For up to 80% of individuals, menopause symptoms include changes in mood and anxiety levels, which can exacerbate existing mental health challenges or trigger first onset of certain psychiatric disorders (18, 19).

Despite its biological and social impact, menopause is rarely treated as a key opportunity to support healthy aging and quality of life. Human females will spend \~40% of their lifespan in postmenopause, with up to 10 years spent in the menopausal transition. There is growing recognition that menopause is not simply a transition but a pivotal period that affects future health (14). Menopause features such as its timing, type, symptom profile, and burden provide valuable information about future risks for osteoporosis, metabolic syndrome, cardiovascular disease (CVD), and dementia, and may help identify those more likely to benefit from early prevention and targeted interventions (3, 20–22). Thus, menopause, in all its complexity, represents an important inflection point for future aging-related concerns. This Review outlines the diverse characteristics of menopause and their differential contribution to brain-related health. We explore the need to embrace heterogeneity of MHT (type, timing, administration) to reduce menopause symptom burden and optimize healthy aging, and to highlight opportunities for personalized MHT treatments that consider genetics and past reproductive history.

---

Menopause characteristics: type, timing, and symptoms

Spontaneous menopause is characterized by declines in estradiol and progesterone levels and increases in gonadotropic releasing hormones (GnRH) such as FSH and luteinizing hormone (LH), due to disruptions in negative feedback from the HPG axis (Fig. 1 and Box 1). Menopause can occur spontaneously, or it can be induced either surgically (bilateral oophorectomy) or chemically (GnRH agonists, chemotherapy) (7, 23). Although induced menopause is defined by the time of treatment/surgery, spontaneous menopause is defined retrospectively after 12 consecutive months without menstruation. The average age of spontaneous menopause is \~51 years (24). The age at spontaneous menopause is determined by both genetic, and environmental factors, e.g., earlier age at first pregnancy and smoking are associated with earlier spontaneous menopause onset (24, 25). Limited evidence also suggests that earlier menarche may be linked with earlier onset of spontaneous menopause (26).

A

B

C

Menopause in context: Life history and menopause symptoms matter for cognitive aging

Many cognitive aging trajectories

● Maintenance (optimal MHT)

● Typical Aging ● Delayed-onset decline

● Early-onset decline ● Accelerated decline

Fig. 1. A broad overview of factors across the lifespan that need to be considered to understand menopause. (A) Nearly all major organs and body systems express receptors for estrogens and progesterone. (B) Menopause is a whole-body transition, with more than 30 recognized symptoms. Given the widespread receptor distribution for ovarian hormones, it is not surprising that the effects of menopause extend to the brain (memory, mood, AD-related pathology), cardiometabolic changes, immune system, sleep disturbances, kidney function and urogenital system, musculoskeletal system, gastrointestinal system, mucosal membranes, and skin. These effects are independent of chronological age. This figure provides a general overview of when these symptoms occur; however, there is interindividual variability in the onset, duration, severity, frequency, and type of symptoms experienced. In contrast to common belief, these symptoms can persist well into the postmenopausal period. (C) Menopause cannot be understood in isolation. Consideration of biological, psychological, and social factors is needed.

In North America \~10% of menopause is surgical, 2% chemical, \~74% spontaneous, and 10% of cases are spontaneously early (prior to the age of 45) or premature (prior to the age of 40) (23). These menopause types differ in both age of occurrence and abruptness of ovarian hormone decline. Even when spontaneous and surgical menopause are compared at the same age, they have both shared and distinct biological consequences (7, 27, 28). Other reproductive surgeries, such as hysterectomy (uterine removal), do not cause menopause but are associated with earlier menopause onset (\~4 years) and a greater risk for early-onset dementia (29). The combination of hysterectomy with oophorectomy is associated with an even greater risk of dementia than either surgery alone (30, 31), consistent with rodent studies (32). These findings point to distinct biological signatures of oophorectomy, hysterectomy, and spontaneous menopause.

The ovary, an endocrine organ, not only produces hormones but also acts as an immune reservoir (33). The postmenopausal ovary is not quiescent as it still produces androgens and inflammatory signals that differ by past parity (34, 35). The mechanisms of spontaneous menopause are still debated (see Box 2).

Menopause timing carries important health implications. Early surgical menopause, premature menopause and primary ovarian insufficiency are reliably associated with increased risk of mortality (36), dementia (37, 38), CVD (39), osteoporosis/fracture (40), gynecological cancers (41, 42), rheumatoid arthritis (43),

---

Box1. Beyond ovarian hormones

Multiple steroid, peptide, and protein hormone levels change considerably during the menopause transition (Fig. 1). Because of negative feedback from the HPG axis, it becomes challenging to pinpoint which hormone(s) may cause which menopause symptoms. Estrogens and progesterone have received considerable attention, but FSH and LH levels rise markedly across the transition. Estradiol and FSH levels vary substantially across the transition by ethnicity and body mass index (191). Some have suggested that targeting FSH may prove fruitful for healthy aging. However, GnRH agonists reduce FSH and LH levels but paradoxically increase depressive symptoms, bone density loss, and vasomotor symptoms (192). Nonetheless, the dynamic interplay between FSH and estradiol deserves more attention.

Other sources of hormones. Steroid hormones can be locally derived in tissues or created de novo through cholesterol, provided the required enzymes are present (193). Estrogens can be produced through the ovaries, adipose tissue (primarily estrone synthesis), muscle, and the gut microbiome through the estrobiome (194). Few studies have investigated the effects of different sources of estrogens on brain health, with some finding more correlations with estrogens found in cerebral spinal fluid compared with those in blood (195). Steroid hormones are lipophilic and hydrophobic and thus present in greater concentrations in lipid-rich tissue such as the brain. Some attention has been paid to quantifying available steroid hormones as bound versus free steroid hormones. However, steroid hormones require a carrier protein when contained within plasma or serum. Steroid and peptide hormones can be challenging and expensive to assay, and caution must be exercised when comparing across studies for absolute measurements. Furthermore, greater value may lie in understanding individual responses to hormone shifts rather than absolute levels.

chronic pulmonary disease (44), and lung cancer (45). Beyond incident dementia, earlier menopause is associated with faster cognitive decline, regional brain atrophy, increases in Alzheimer's disease (AD) pathology, decreased global brain activity, cerebrovascular changes, and increased risks of other neurological disorders such as Parkinson's disease and amyotrophic lateral sclerosis (45). Notably, AD neurodegenerative pathology begins decades before clinical presentation, making the menopause transition a potential window of opportunity for interventions aimed at supporting later-life brain health and reducing neurodegenerative disease risk (Fig. 1).

Despite consistent evidence linking menopause timing to later-life dementia risk, the specific mechanisms underlying these associations remain unclear. Accumulating data report that earlier menopause is associated with elevated susceptibility to multiple dementia risk factors, including both AD and non-AD pathology, cardiovascular risk burden, and synaptic dysfunction (20, 46–48). Earlier menopause is also associated with accelerated biological aging on epigenetic and proteomic clocks (28, 49). Plasma proteomic signatures of earlier menopause reflect up-regulations in inflammatory signaling, extracellular matrix remodeling/catabolism, and apoptosis (28). It is unknown whether earlier menopause causally induces faster rates of aging to increase dementia risk, or whether earlier menopause and elevated neurodegenerative risks are both consequences of an upstream factor that accelerates aging.

---

Menopause symptoms

Despite a predominant focus on VMS, menopause encompasses at least 72 symptoms, with an average of seven symptoms experienced concurrently (11, 13). Although the average duration of menopause symptoms is 7.4 years (50), at least \~33% of older postmenopausal

---

Box2. The postmenopausal ovary is not quiescent

Previously, the postmenopausal ovary was deemed quiescent, and the mechanisms regulating spontaneous menopause timing were thought to be solely related to the depletion of ovarian follicles. However, signals of cellular senescence (22,25), vascular changes, and shifts in immune cell populations occur with menopause in the ovary (34,35). Anti-Mullerian hormone and inhibin B decline across the transition with different timescales, the former of which is used as a marker of ovarian follicle reserve. However, the exact mechanisms of menopause are still not well understood, but are likely multifactorial, encompassing both genetic factors and life-course environmental, structural, and health exposures (e.g., smoking is associated with earlier menopause) (22,25). Although ovaries have a reservoir of egg follicles at birth, only 0.1% are released across the lifespan (196), with degradation of follicles occurring from birth, suggesting other mechanisms may be at play. The discovery of ovarian stem cells (OSC) across species, including humans, has incited much discussion in the field (197). This discovery was initially met with skepticism, but multiple researchers across the globe have provided irrefutable evidence that OSC exist, with research focused on their application for improving fertility. Single-cell RNA sequencing studies show that the aging ovary across species shows changes in immune cell populations as well as extracellular matrix remodeling (198,199). Parity can affect menopause age (154), the trajectory of aging (154) and influences the ovarian transcriptome in middle age (34), providing yet another clue that the ovary has multiple signaling pathways. However, might OSC hold the key for menopause initiation? How does menopause affect OSC and other stem cells across the body? There are many unanswered questions but one thing is clear, the ovary is not just a reservoir of eggs, providing fertility capability, and its role in health does not end with menopause.

people (ages 65 to 79) will continue to experience VMS (51, 52). In North America, both brain fog (difficulties in memory, attention, and concentration) and irritability/mood/emotional disturbances are common during the menopause transition, reported in up to 80% of cases (19, 53). Depressive symptoms associated with menopause are experienced by \~10 to 30% of individuals with no previous history of depression compared with 45 to 65% of those with a previous history of depression (19). In those with bipolar disorder, symptom worsening has been reported during the menopausal transition (19) and evidence suggests a bidirectional relationship between anxiety and increased VMS (19). Moreover, \~64% of individuals in the menopause transition report “not feeling like myself” more than 50% of the time, reflecting the broader mental health impacts of menopause (54). Despite the high prevalence of mental health symptoms associated with menopause, they are under-recognized by healthcare practitioners, who prioritize management of physical symptoms (55). Sleep disturbances, which often co-occur with brain fog and mood disturbances, are also common and can be exacerbated by nighttime VMS symptoms, which fragment sleep (12, 56). Overall, the most frequently reported menopausal symptoms involve the brain.

Studies have found nonclinical declines in a range of cognitive domains across the menopausal transition, including episodic memory (57–59) and executive function (57, 60). These declines in cognitive performance are temporary, stabilizing after perimenopause (61). However, higher menopause symptom burden (53, 62) and more severe VMS, particularly at night have been associated with worse cognitive performance (63, 64) as well as increased plasma amyloid beta ratio, white matter hyperintensities, and decreased quality of life (65, 66). Finally, severe and frequent menopausal symptoms are associated with increased odds for developing mild cognitive impairment (MCI), a prodromal state to dementia (67). Experiences of menopause symptoms, cognitive or otherwise, may thus have important implications for long-term brain health. Given the link between VMS, MCI, and brain pathology, increased education of both the public and health care practitioners is needed, alongside recognition of the potential for menopause therapy in improving quality of life and preserving brain health through VMS reduction.

Most women are not aware of the diversity of symptoms that may accompany the menopause transition. Although most (77 to 84%) report recognizing VMS as symptoms of menopause, less than half are aware that urogenital symptoms, memory issues, anxiety, and body aches may also accompany the transition (13). How we query menopause symptoms matters not only for research but also patient awareness of potential symptoms. The menopause rating scale, a common tool for measuring menopausal symptoms, lists only 11 symptoms. Reproductive health tracker apps may help identify other symptoms and can be useful in menopausal staging (12). Those with more severe VMS are up to three times as likely to experience depressive symptoms (68), metabolic disturbances, including earlier onset of diabetes (69), and joint pain (69). This suggests that symptoms are interconnected and likely do not operate in isolation, making accurate tracking and clustering of symptoms especially important.

The menopause transition is also marked by weight gain, a shift to more visceral fat, decreases in lean muscle mass, and decreases in bone density, changes that increase systemic inflammation and risk for metabolic syndrome (70–72). Visceral fat is linked to reduced cognition and brain network connectivity across domains postmenopause (73, 74). Those with polyendocrine metabolic ovarian syndrome may be at increased risk of negative metabolic, inflammatory, and cognitive changes during the menopausal transition (75, 76), even though they report fewer menopause symptoms than unaffected individuals (77). Menopause-associated metabolic changes are critical to consider as a part of menopause care for maintained brain health.

Ethnic and cultural considerations of menopause are also needed for informed menopause care. Black and South Asian populations in North America experience earlier ages of spontaneous menopause and Black individuals are more likely to undergo oophorectomies and hysterectomies (78, 79). Furthermore, Black Americans experience more frequent and severe VMS during the menopause transition compared with their white counterparts, and receive fewer prescriptions for MHT (78–80). East Asian populations are more likely to report musculoskeletal symptoms, commonly frozen shoulder, compared with white individuals (81, 82). Overlooked in the literature is that there are cultural differences in the attitudes surrounding menopause, or openness to disclose symptoms (83), as well as cultural differences in menopause education and treatment approaches (84). Less is known about racial/ethnic variations in menopausal symptoms in multicultural countries (84). Recent findings have contested earlier assumptions that East Asian populations are less likely to report VMS (81, 82). There may also be ethnic differences in symptom profiles across the menopause transition (85), either reflective of population differences or psychosocial factors (86) that affect symptom reporting and conceptualization. Notably, experience of menopause symptoms can be shaped by gendered expectations, attitudes, and lived experience, which vary across cultural contexts as well as between individuals (87, 88).

---

Menopause hormone therapy

MHT consists of estrogens and/or progestogens to supplement the decline in ovarian hormones associated with menopause. The first-line treatment for VMS reductions is MHT. VMS are directly related to kisspeptin -neurokinin b-dynorphin (KNDY) neurons in the infundibular bundle of the hypothalamus (89). Thus, neurokinin B (NKB) antagonists also reduce VMS symptoms and offer a nonhormonal option when hormones are contraindicated/not desired (89, 90). Although there have been no RCTs directly comparing NKB antagonists with MHT formulations, network meta analyses found that NKBs were moderately effective compared with MHTs, which were highly effective in reducing VMS severity (91, 92). Another potential use for MHT is preserving bone density for osteoporosis prevention, a condition that affects one in three women after age 50 (93). Systemic MHT-containing estrogens prevent fractures by 20 to 40%, with greater benefits observed with higher doses (93). However, the overall uptake of MHT remains low [USA: 1.7% of women aged 40 or older (\~13M) using data from medical/pharmacy insurance claims in 2023; Sweden: 3.8% of women aged 45 to 60 years in a population cohort study (\~1M) in 2020; (94, 95)]. Research and education remain at the forefront of bridging this gap for both the public and physicians.

MHT has a controversial history as a result of the Women's Health Initiative trial. As the largest randomized control trial (RCT) of MHT, the findings continue to have a major influence on standards of menopause care. They found that oral conjugated equine estrogens (CEE) combined with medroxyprogesterone acetate (MPA), but not oral CEE alone, led to an increased risk of CVD, breast cancer, and dementia compared with placebo, leading to the premature termination of the trial due to these negative outcomes (96). Although this trial was designed to examine whether MHT promoted optimal health in older women (10+ years postmenopause), considerable criticism has been levied toward its approach centered on three major areas: the critical window hypothesis, healthy cell bias, and type of MHT used [reviewed elsewhere (8, 97–99)]. However, the scientific and medical community has applied these negative findings from one type of hormone therapy (oral CEE with or without MPA) to all MHTs, leading to a >60% decline in MHT prescriptions and the addition of blackbox warnings on all MHT products cautioning against their use due to risk of CVD, cancers and dementia (100). Notably, there are >200 MHTs that vary in hormonal formulation, dose, route of administration as well as timing (duration, initiation relative to menopause) and how they are used (cyclically versus continuously; see Fig. 2). Research shows that all these variables can contribute differently to health risks and benefits, illustrating that there is not a simple one-size-fits-all approach to menopause care.

The federal food and drug administration in the US announced the removal of MHT blackbox warning in 2025 (101). This decision has received some criticism from the International Menopause Society, which supports the removal of warnings from vaginal estrogens but not systemic MHTs. Other RCTs since the Women's Health Initiative have generally not found clinically significant effects of MHT on cognitive performance (102–104), or CVD risk (105, 106), but do find positive effects on bone density (107). Although some trials show increased risk or no change in CVD events with CEE or estradiol, evidence suggests that initiation closer to the menopausal transition may reduce CVD risk and subclinical atherosclerosis (108–111). Additionally, prolonged VMS are linked to increased CVD risk factors such as hypertension, insulin resistance, and adverse lipid profiles, as well as increased later-life CVD risk, highlighting the importance of effective management (112, 113). Moreover, these RCTs have not evaluated various doses and formulations of MHT, including low-dose vaginal estrogens and transdermal gels and sprays, among younger, symptomatic postmenopausal individuals or racially diverse cohorts (114, 115, 116). More attention is needed to understand the risk-benefit profiles of the full breadth of available MHT options.

Routes of administration: Oral routes of administration are subject to hepatic first-pass metabolism as well as breakdown by the gastrointestinal tract. This has consequences for not only the types and levels of estrogens that reach circulation, but also clotting factors associated with increased risk of venous thrombosis (117). Approximately 90% of oral estradiol, depending on how it is conjugated, is converted to estrone, a relatively weaker estrogen. By contrast, transdermal estrogens do

---

Menopause hormone therapy (MHT) comes in many varieties with distinct brain health benefits

---

A

Variation in therapies

Optimal MHT traits

\- Route of administration

\- Dose

\- Cyclicity

\- Type of hormones

• Combination of hormones

• Transdermal and vaginal routes

\- Time of initiation - Lowest dose for symptom management

• Cyclic administration

\- Duration

\- Estradiol-based

\- Switches

\- Avoid MPA

\- Past use

\- Initiate close to menopause transition

\- Use as long as needed for symptom management

• Past use has lasting benefits

---

B

---

Medroxyprogesterone acetate versus progesterone

↓ Spatial working memory and reference memory

↓ BDNF — expression

↑ Venous thrombo-embolism

Fig. 2. Menopause hormone therapies are diverse in formulation, dose and timing.

(A) There are many varieties of MHT, which vary in their risk-benefit profiles and resulting health benefits. Major concerns associated with MHT use are breast cancer and CVD, but these risks vary according to MHT type. Optimal MHT features are outlined, including transdermal and vaginal routes of administration that result in less blood clotting factors that cause inflammation and risk for venous thromboembolism. (B) Both human and rodent research suggests that medroxyprogesterone acetate, the progestin used in the Women's Health Initiative trial, has an unfavorable risk-benefit profile. It has been associated with negative brain health outcomes. (C) In practice, there are many possible MHT combinations. For example, considering only estradiol with one route of administration (i.e., estradiol patch), there are at least 70 possible combinations given differences in doses, type of progesterone/progestin, type, dose, and cyclicity. However, these many possibilities are not well-captured in research. All MHT research must contextualize findings according to the MHT features that were investigated and the population that was studied.

not undergo hepatic metabolism, require a lower dose to achieve the same estrogenic effects, and show a two- to fourfold decrease in blood clotting risk, compared with oral routes (118). Vaginal estrogens similarly circumvent first-pass metabolism and are generally used to treat urogenital symptoms. Although vaginal estrogens are a localized form of hormone therapy, they can exert systemic benefits, particularly with the vaginal ring (119), and through downstream reductions in systemic inflammation caused by recurrent urinary tract infections (120). Low-dose vaginal estrogens beyond age 65 have been associated with reduced mortality, ovarian cancer risk, dementia risk, and no increased risk in recurrence of ER+ breast cancer (110). Thus, route of administration affects levels of different estrogens with important health consequences.

Importance of MHT composition and route of administration extends to cognitive aging. Estradiol-based therapies benefit episodic memory (121) and executive function (121) in smaller placebo-controlled studies using sensitive cognitive assessments rather than global cognition and shorter or cyclical treatment durations (122–124), whereas data using CEE is less consistent. Although the combination of CEE and MPA has been associated with either neutral or negative cognitive outcomes, depending on age of initiation (125, 126), positive outcomes on cognition have been seen with combination therapies involving micronized progesterone (127, 128). Route of administration also matters, with recent work showing more pronounced positive effects of transdermal estradiol on episodic memory (129). These differences in MHT effects by type and route of administration may help explain the overall mixed findings in the literature on MHT and cognition as many studies combine MHT types into a single user group (121).

Estrogens and their receptors: Estrogens are released from the ovaries, but there are other sources (Box 1). Three types of ER exist which are located either in the cell membrane or in the nucleus/cytoplasm. Different densities of ERs reside across tissue types (130), and both endogenous and exogenous hormone exposure can influence ER density (131, 132). Genomic signaling occurs after estrogens bind to nuclear ERα, ERβ, which then homo- or heterodimerize, translocate to the nucleus, and bind to an estrogen response element (ERE). Non-genomic signaling occurs when estrogens bind to membrane bound ERs, which results in downstream cell signaling proteins that may involve subsequent genomic mechanisms. Growth factors [such as brain-derived neurotrophic factor (BDNF)] can independently bind to EREs, complicating signaling. Both estrogens and progesterone influence monoamine and cholinergic transmission and growth factors (e.g., BDNF), thus expanding the multiple signaling pathways by which estrogens can exert their effects (133, 134).

Different estrogens have different binding affinity to the different ERs (130, 135, 136) Table 1). Of the four estrogens endogenous to humans, estradiol has a high affinity for both ERs and GPER, whereas estrone, which becomes the predominant estrogen after menopause (Fig. 1), has lower binding affinity to ERα and ERβ compared with estradiol (130). Both estriol and esterol have low affinities for ERα, ERβ, and estriol is a GPER antagonist (135, 137). CEE contains 10 equine estrogens with the highest concentration of estrone and minimal levels of estradiol (0.5 to 3%). Of the estrogens contained in CEE that are not endogenous to humans, 17b-dihydroequilin has the highest binding affinity to ERα and β, but negligible binding to GPER. All of these different estrogens have the potential to invoke different signaling pathways and subsequent downstream health outcomes, but these differences are rarely considered.

---

Role of progesterone

Progestogens are prescribed for those with an intact uterus to protect against endometrial hyperplasia associated with unopposed estrogens (138). Similar to estrogens, progestogens are often reduced to a single hormone but come in many types, with distinct influences on hormone receptors and brain health. Progesterone binds to two different isoforms of the PR (A or B) and is metabolized to allopregnanolone which is a GABA $_{A}$ agonist (133). Progesterone initially augments the influence of estrogens, but after time, progesterone starts to reverse the effects of estrogens, in part by decreasing the density of ERs and PR across some tissues (139). Given these interactions, it is critical to consider that the effects of MHTs containing unopposed estrogens may differ from those of MHTs containing both estrogens and progestogens. Cyclic versus continuous use of progestogens also matters. In rodents, cyclic progesterone with estradiol enhanced reductions in neuropathology outcomes (140), whereas continuous progesterone blunted the neuroplastic effects of estradiol (141, 142). Synthetic progestins have different binding properties to PR, androgen receptors, and even glucocorticoid receptors depending on their derivation. Pregnane i.e. (progesterone-derived), including MPA, have antiandrogenic properties, whereas gonanes derivatives (i.e. 19-nortestosterone derived), including levonorgestrel, have androgenic properties (143). MPA, which was used in the Women's Health Initiative, has been related to most of the poorer health outcomes in that RCT (144) (Fig. 2). Finally, some individuals are more sensitive to progesterone fluctuations across the menstrual cycle (145), and this sensitivity likely extends into mid-to-later-life hormonal transitions (see below).

Table 1. Binding affinity of different estrogens to ERα, ERβ, and GPER. Source reflects whether the type of estrogens is endogenous to humans or equine-derived and where the primary source of the estrogens is produced in humans. However, steroidogenesis or de novo production of hormones (from cholesterol) or local production (through aromatization of testosterone) of hormones can also occur throughout the body or brain if the catalyzing enzymes are present. Percent contained in CEE reflects the approximate percentage of the estrogens present in the CEE composition. Relative binding (ERα versus ERβ) reflects whether the estrogens preferentially bind to the α versus β receptor subtype. Both ERs are present in most tissues and organs, including the brain. CEE, conjugated equine estrogens; ER, estrogen receptor; GPER, G protein–coupled receptor. | Type of estrogens | Source | Percent contained in CEE | ERα binding affinity | ERβ binding affinity | GPER binding affinity | Relative binding (ERα versus ERβ) | 17β-Estradiol (E2) | Human: ovaries | 1% | High | High | High | ERα = ERβ | 17β-Dihydroequilin | Equine: derived | 10 to 15% | High | High | Limited data | ERα < ERβ | Estriol (E3) | Human: placenta, liver | 0% | Moderately weak | Moderately weak | Very weak | ERα < ERβ | Estrone (E1) | Human: adipose, ovaries | 50% | Moderately weak | Moderately weak | Negligible | ERα > ERβ | Equilin | Equine: derived | 22-25% | Moderately weak | Moderately weak | Limited data | ERα > ERβ | 17β-Dihydroequilenin | Equine: derived | 1-3% | Weak | Weak | Limited data | ERα < ERβ | Equilenin | Equine: derived | 1-5% | Weak | Weak | Limited data | ERα = ERβ | Estetrol (E4) | Human: fetal | 0% | Weak | Very weak | Negligible | ERα > ERβ | ---

What can influence the effects of different MHTs?

In addition to receptor-level effects, history of endogenous or exogenous hormones and individual hormone sensitivity may affect response to MHT (146, 147). The Kronos Early Estrogen Prevention Study (KEEPS) study found lasting volumetric increases in the dorsolateral prefrontal cortex that persisted several years after transdermal estradiol use (148). Past MHT use has also been associated with greater hippocampal and parahippocampal volume relative to never-use (149, 150). However, prior use of hormone therapy is not controlled for in most studies, with 20 to 80% of placebo groups having previously been on hormone therapies, which could affect interpretation of findings (102, 103, 116).

Exogenous use of hormones (e.g., hormonal contraceptives) beyond MHT may also influence the effects of MHT in later life; however, the history of use is not well-captured in studies. Although lacking in detail regarding past use, studies show that prior hormonal contraceptive use is associated with reduced risk of later-life cognitive impairment and dementia (146). Prior hormonal contraceptive use may modify the effects of MHT, with dual exposure associated with greater frontal cortical volumes compared with MHT alone (146). Hormonal contraceptives are widely prescribed but are overlooked as a factor in menopause research. Future research should consider the variations in hormonal contraceptive type, age, and duration of use on brain aging.

Switching hormone medications may also play a role in MHT outcomes. Switching is common among hormonal contraceptive users with 33 to 44% of individuals switching/discontinuing contraceptives within the first year (151, 152), and one study found that people switch MHTs 2.4 times on average within a 10 year period (153). Consideration of these details may be key for studying the effects of MHT on brain health, especially as past history of exogenous hormone use can have lasting impacts on the brain.

Endogenous hormone exposures such as pregnancy may also play a role (35, 154, 155). Rodent studies show that past pregnancies modulate effects of acute and chronic estrogen exposure on hippocampal plasticity and cognition in middle age (156, 157). Pregnancy characteristics also influence VMS and the postmenopausal ovarian transcriptome (35). Individuals with preeclampsia, gestational hypertension, or who experience preterm birth and small for gestational age experience more VMS and are at an elevated risk for developing later-life CVD and vascular dementia. Although heavily under-researched, puberty is another major endogenous hormone transition that may set the stage for menopausal health (158). Thus, consideration of past reproductive experiences throughout the lifespan provides meaningful clinical insight for menopause care.

---

Toward personalized MHT: influence of genotype

Genetic factors, such as single nucleotide polymorphisms, contribute to variance in hormonal effects on cognition. Across the menstrual cycle, BDNF Val66Met modulates learning and memory performance, with individuals who carry the Met allele showing increased sensitivity to estradiol (159). Similarly, the catechol-o-methyltransferase (COMT) COMT Val158Met polymorphism interacts with estradiol and hormonal contraceptives to predict executive function performance, such that Met carriers show improved performance in low estradiol phases and with hormonal contraceptive use, whereas Val carriers show improved performance in high estradiol phases (160–162). Given these genotype-specific effects of estradiol fluctuations on cognition, genetic effects of MHT on brain health may also be present but have been unexplored.

There is limited work on how genotype can influence MHT effects. Neuroimaging work shows more pronounced positive effects of MHT use in carriers of the APOEε4 allele, the strongest genetic risk predictor for Alzheimer's disease (163), including fewer white matter lesions (165), larger hippocampal volume (164, 165), and less AD-related neuropathology (166, 167). Genes involved in the production or metabolism of estrogens are strong candidates for exploring how MHT or menopause can have diverse outcomes across individuals (168, 169). Although there are limited studies, the KEEPS trial found that a genetic variation in SLCO1B1, which affects transport of estrogens from the blood into hepatocytes, led to a greater decrease in nighttime VMS with transdermal estradiol (169), and that genetic variations in COMT and CYP3A4, which affect the metabolism (methylation and hydroxylation, respectively) of estrogens, were associated with severity of somatic menopause symptoms (168). These genetic differences may be important in deciphering the mixed literature on neural effects of MHT.

---

Hormonal sensitivities across hormonal transitions

Menopause is often studied as an isolated reproductive/hormonal transition without consideration of past experiences. Hormonal sensitivity is a term that encapsulates how sensitive an individual is to fluctuating gonadal hormones around hormonal transitions (e.g., menstrual cycle, pregnancy). Individual responses to hormonal dynamics can provide an important clue for later experiences of menopause (170, 171). For example, an earlier age of menopause is noted in those with premenstrual disorders (172). Notably, individuals with more severe premenstrual symptoms are more likely to report experiencing more severe symptoms during perimenopause (172, 173). The pattern of consistent mood exacerbations with reproductive transitions across the lifespan suggests that surveying experiences of earlier reproductive/hormonal transitions may be meaningful when providing menopause care. Heterogeneity in response to hormonal fluctuations cannot be overlooked; Some people are sensitive to the rise, and others to the fall, in hormones, suggesting that there are differing subtypes of hormonal sensitivity (174). Although this is a burgeoning field, embracing this complexity and greater investment in understanding how hormonal sensitivities influence health long-term could hold the key to discovering early interventions for later-life disease.

Continued menopause care is essential: Menopause symptoms can persist past the menopausal transition, affecting quality of life postmenopause. However, treatments such as MHT are not usually prescribed long term. Continued use of MHT into later life should not be confused with MHT initiation in later life. Fixed time limits for MHT should be reconsidered, as up to 45% of people over 65 still experience VMS (175). Beyond VMS, MHT is an effective strategy for preventing bone density loss associated with the menopausal transition, with MHT discontinuation associated with an accelerated decline in bone density (176, 177). Observational studies show that MHT use, particularly estradiol-based therapies, is associated with reduced mortality (178). Given the substantial burden of symptoms that persist beyond menopause, the role of continued treatment may need to be reevaluated to better support quality of life in the postmenopausal years. Several medical societies already recommend continued MHT for menopause symptom management, noting that chronological age should not dictate when MHT is stopped (178).

Politicized menopause research: Menopause is having a moment across social media, marketing, and commercialization. Commercial and public interest in supplements, apps, and natural remedies increased by \~15 to 20% across the US, UK, and Australia from 2005 to 2025, and the market share is projected to increase further over the next 10 years (179–181). However, effective, informed menopause care and products depend on quality research. Surveys converge on a consistent message: Women want researchers to prioritize menopause (182, 183), yet menopause research funding has remained at only a fraction of a percent for decades (184). One deterrent is the politicization of menopause science, from those who talk about menopause as a diagnosable “condition” that requires treatment to suggestions that any research into menopause management implies pathologization. Menopause is a life stage that all females will experience if they live long enough, yet stigma around female aging has let dogma persist surrounding menopause care. Most medical schools lack a dedicated menopause curriculum (185). However, training requires evidence, and the evidence gap is profound. It may be no wonder that people feel dismissed by healthcare practitioners when raising menopause concerns (55) and turn to health advocates outside of formal medicine to find answers and relief, fueling a booming supplement market and non-regulated forms of MHT. Most medical societies and healthcare practitioners look to RCTs to inform standards of care. However, RCTs are expensive to conduct, are limited in scope, and are only able to capture a snapshot of the \~40% of the lifespan spent in menopause. In fact, RCTs on MHT and chronic health conditions have limited treatment durations (maximum of 8.1 years, average of \~4.3 years; data from 20 RCTs) (186) that do not align with treatment durations in practice (maximum of 14.6 years, average of \~5.4 years based on data from 69 observational studies) (187). The consequences of long-term MHT use on health outcomes have not been fully explored, particularly across diverse MHT types used over the lifespan. Furthermore, true double-blinding is challenging for the 80% who experience VMS, which has historically led to exclusion of symptomatic participants who may be more likely to benefit from MHT treatment. Thus, RCTs may exclude symptomatic individuals with VMS who are at greater risk for mood dysregulation, CVD, neurodegenerative disorders, and musculoskeletal disease, and who may be more likely to benefit from MHT treatment. Although RCTs are certainly preferred, there is much to be gleaned from observational and longitudinal cohort studies, especially through new statistical tools (186).

Advancing precision care in menopause requires sustained research across menopause stages, embracing the complexity of menopause and MHT characteristics. We encourage clinicians to reframe menopause as a critical life stage that may be accompanied by both persistent physical and mental health changes, requiring continued follow-up care analogous to care devoted to pregnancy and the postpartum period. We encourage researchers to reframe menopause research. Are there really only 11 key symptoms? Do they stop after 5 years? We also encourage researchers to be precise in their language and not overgeneralize terms such as estrogen and MHT, which are not single entities but specific formulations which act differently on hormone receptors. Rigorous menopause research is only impactful if it is translated into accurate and accessible science communication. Menopause and MHT are not one thing, and papers and headlines need to make that clear. Data collection must include more information on hormonal histories and medication use across the lifespan, capturing the diversity of MHT types and prior exogenous hormone exposure. Ethnicity, geographic location, and the broader social context in which a person is going through menopause also need to be considered in both biomedical research and clinical practice. Foundations and federal funding agencies must stimulate the field with funding competitions for paradigm-shifting menopause research.

---

Conclusions

Women consistently rank menopause as one of the highest priorities for improving women's health (183). Menopause research is not a niche area of study as it spans nearly 40% of the lifespan for half of the global population. Its impact extends beyond women, informing a broader understanding of aging and health that can benefit society and families, as well as individuals of all genders, including men (188) and transgender individuals (189).

However, details matter. Broad strokes in health research seed misinformation and setbacks in science. Diminishing menopause to “just aging” is a disservice to women. Menopause is a full-body, hormonally driven change that, like other endogenous hormonal transitions, affects individuals differently based on their individual biology and life experience. The goal of MHT treatment is not to pathologize or “reverse” menopause, but to support healthy aging and improve quality of life. Critically, exogenous hormones have different effects depending on type, route, age of initiation, pattern of use, and reproductive health history. This information, along with genotype and sociocultural context, determines menopause characteristics and likely efficacy of treatments. Precision medicine has already transformed breast oncology and is advancing depression treatment (190); applying the same principles to menopause is the next frontier in women's health.

---

REFERENCES AND NOTES

1. R. D. Brinton, J. Yao, F. Yin, W. J. Mack, E. Cadenas, Nat. Rev. Endocrinol. 11, 393–405 (2015).

2. A. Han, J. Y. Kim, J. Kwak-Kim, S. K. Lee, J. Reprod. Immunol. 146, 103346 (2021).

3. Y. Wang, A. Mishra, R. D. Brinton, F1000 Res. 9, 68 (2020).

4. O. Soldatkina et al., Nat. Aging 6, 1177–1192 (2026).

5. L. L. Gravelsins et al., What Gets Funded Shapes What We Know: 15 Years of Canadian Women's Health Research. medRxiv. 2026.02.17.26346472v2 [Preprint] (2026); https://doi.org/10.1101/2026.02.17.26346472v2.

6. National Academies of Sciences, Engineering, and Medicine; Health and Medicine Division; Board on Population Health and Public Health Practice; Committee on the Assessment of NIH Research on Women's Health; "Overview of the National Institutes of Health Investment in Women's Health Research" in A New Vision for Women's Health Research: Transformative Change at the National Institutes of Health. A. Geller, A. Salganicoff, S. P. Burke, Eds. (National Academies Press, 2025); ch. 4, pp. 145–171.

7. H. Edwards, A. Duchesne, A. S. Au, G. Einstein, Menopause 26, 45–65 (2019).

8. L. L. Gravelsins, L. A. M. Galea, Br. J. Psychiatry 226, 369–382 (2025).

9. S. Westwood, J. Women Aging 35, 557–572 (2023).

10. P. A. Rochon, S. Kalia, P. Higgs, Lancet 398, 648–649 (2021).

11. R. R. Austin, S. Alexander, S. Tupper, J. M. Holt, Comput. Inform. Nurs. 43, e01289 (2025).

12. S. G. Aras, A. D. Grant, J. P. Konhilas, Sci. Rep. 15, 640 (2025).

13. Menopause Foundation of Canada, "The Silence and the Stigma: Menopause in Canada" (Menopause Foundation of Canada, 2022); https://menopausefoundationcanada.ca/menopause-in-canada-report/.

14. R. A. Lobo et al., Climacteric 17, 540–556 (2014).

15. G. D. Mishra et al., Lancet 403, 958–968 (2024).

16. A. J. Huang, S. Faubion, D. Grady, JAMA Intern. Med. 185, 874–875 (2025).

17. N. Safwan et al., Maturitas 185, 107983 (2024).

18. R. C. Thurston, H. N. Thomas, A. J. Castle, C. J. Gibson, Nat. Rev. Psychol. 4, 530–543 (2025).

19. L. Brown et al., Lancet 403, 969–983 (2024).

20. M. Wood Alexander et al., Ann. Neurol. 98, 1014–1026 (2025).

21. P. Anagnostis et al., Endocrine 63, 213–224 (2019).

22. J. Gong, K. Harris, S. A. E. Peters, M. Woodward, PLOS Med. 19, e1003955 (2022).

23. W. A. Rocca et al., Maturitas 170, 213–224 (2019).

24. E. B. Gold, Obstet. Gynecol. Clin. North Am. 38, 425–440 (2011).

25. D. Peycheva et al., BMC Womens Health 22, 438 (2022).

26. N. Bazyar, Z. Moradi, A. Khani Jeihooni, A. Dehghan, Sci. Rep. 15, 18311 (2025).

27. N. J. Gervais et al., Front. Endocrinol. 14, 1265470 (2023).

28. M. Wood Alexander et al., Plasma proteomics link menopause timing to brain aging and dementia risk. medRxiv. 2026.04.23.26351500 [Preprint] (2023); https://doi.org/10.3389/fendo.2023.1265470.

29. Y. Huang et al., J. Ovarian Res. 16, 35 (2023).

30. W. A. Rocca, B. R. Grossardt, L. T. Shuster, E. A. Stewart, Neurodegener. Dis. 10, 175–178 (2012).

31. T. K. T. Phung et al., Dement. Geriatr. Cogn. Disord. 30, 43–50 (2010).

32. S. V. Koebele et al., Horm. Behav. 155, 105411 (2023).

33. M. E. Zardeneta, K. A. Pickle, F. Sohrabji, Front. Neuroendocrinol. 80, 101234 (2026).

34. M. A. Watson et al., The human ovary exhibits dynamic molecular remodeling in the decades post-menopause. bioRxiv 2026.03.26.714635v1 [Preprint] (2026); https://doi.org/10.64898/2026.03.26.714635.

35. J. Byrne et al., Single-nucleus profiling reveals age-associated remodeling opposed by parity in the postmenopausal human ovary. bioRxiv 2026.05.11.724381 [Preprint] (2026); https://doi.org/10.64898/2026.05.11.724381v1.

36. M. C. Cusimano et al., BMJ 375, e067528 (2021).

37. N. Calvo et al., J. Alzheimers Dis. 102, 119–128 (2024).

38. H. Liao et al., EClinicalMedicine 60, 102033 (2023).

39. T. Muka et al., JAMA Cardiol. 1, 767–776 (2016).

40. Y. Abulajiang, T. Liu, M. Wang, A. Abulai, Y. Wu, Front. Oncol. 15, 1541585 (2025).

41. Z.-M. Ouyang et al., BMC Womens Health 25, 138 (2025).

42. N. Namavari, M. Jokar, A. Ghodsian, H. K. Jahromi, V. Rahmanian, BMC Rheumatol. 8, 48 (2024).

43. C. Liang et al., Female reproductive histories and the risk of chronic obstructive pulmonary disease. Thorax 79, 6 (2024).

44. H.-F. Chung, D. G. Gete, G. D. Mishra, Maturitas 153, 1–10 (2021).

45. J. Raymond, P. Mehta, T. Larson, E. P. Pioro, D. K. Horton, Neuroepidemiology 55, 416–424 (2021).

46. G. T. Coughlan et al., JAMA Neurol. 80, 462–473 (2023).

47. M. Wood Alexander et al., Neurology 102, e209298 (2024).

48. M. Wood Alexander et al., Sci. Adv. 11, eadt0757 (2025).

49. M. E. Levine et al., Proc. Natl. Acad. Sci. U.S.A. 113, 9327–9332 (2016).

50. N. E. Avis et al., JAMA Intern. Med. 175, 531–539 (2015).

51. B. M. Zeleke, R. J. Bell, B. Billah, S. R. Davis, Fertil. Steril. 105, 149–55.e1 (2016).

52. C. J. Gibson et al., Women Health 64, 1–17 (2024).

53. C. Gurvich, A. Spector, M. Hickey, Lancet Obstet. Gynaecol. Women's Health 2, e335–e345 (2026).

54. N. Coslov, M. K. Richardson, N. F. Woods, Menopause 31, 390–398 (2024).

55. K. Thavabalan et al., Maturitas 201, 108683 (2025).

56. A. M. C. Brown, N. J. Gervais, Endocrinology 161, bqaa128 (2020).

57. G. A. Greendale et al., Neurology 72, 1850–1857 (2009).

58. F. Kilpi et al., BMC Womens Health 20, 177 (2020).

59. C. N. Epperson, M. D. Sammel, E. W. Freeman, J. Clin. Endocrinol. Metab. 98, 3829–3838 (2013).

60. A. Bangle, D. Williams, J. Walters, L. Nguyen, Psychol. Aging 41, 303–318 (2026).

61. K. Zhou, M. Cohn, R. Novik, P. Batur, C. Just, Obstet. Gynecol. Clin. 53 449–461 (2026).

62. J. F. E. Crockford et al., PLOS ONE 20, e0301165 (2025).

63. P. M. Maki, M. T. Weber, Climacteric 24, 382–388 (2021).

64. P. M. Maki, R. C. Thurston, Front. Neurol. 11, 562275 (2020).

65. R. C. Thurston et al., Neurology 100, e133–e141 (2023).

66. R. C. Thurston et al., Am. J. Obstet. Gynecol. 230, 342.e1–342.e8 (2024).

67. A. Calle et al., Menopause 31, 959–965 (2024).

68. A. N. Reeves et al., Ann. Epidemiol. 58, 48–55 (2021).

69. J. E. Santo et al., Int. J. Gynaecol. Obstet. 165, 138–147 (2024).

70. A. Fenton, J Midlife Health 12, 187–192 (2021).

71. G. A. Greendale et al., JCI Insight 7, e124865 (2019).

72. A. A. Opoku, M. Abushama, J. C. Konje, Best Pract. Res. Clin. Obstet. Gynaecol. 88, 102348 (2023).

73. T. T. James et al., Menopause 33, 151–160 (2026).

74. R. G. Zsido et al., JAMA Netw. Open 2, e196126 (2019).

75. M. Millán-de-Meer, M. Luque-Ramírez, L. Nattero-Chávez, H. F. Escobar-Morreale, Hum. Reprod. Update 29, 741–772 (2023).

76. M. Perović, K. Wugalter, G. Einstein, Front. Neuroendocrinol. 67, 101038 (2022).

77. J. Lavi et al., Acta Obstet. Gynecol. Scand. 105, 836–846 (2026).

78. X. Lyu et al., Climacteric 28, 553–560 (2025).

79. S. A. Kingsberg et al., Climacteric 26, 537–549 (2023).

80. S. D. Harlow et al., Womens Midlife Health 8, 3 (2022).

81. S. R. Davis et al., Climacteric 26, 517–536 (2023).

82. Q. Yu et al., Menopause 29, 553–563 (2022).

83. S. Shorey, E. D. Ng, Menopause 26, 557–569 (2019).

84. M. Hickey et al., Lancet 403, 947–957 (2024).

85. J. Wang et al., Menopause 32, 864–876 (2025).

86. A. N. Reeves et al., Menopause 31, 484–493 (2024).

87. H. Dillaway, in The Palgrave Handbook of Critical Menstruation Studies, C. Bobel et al., Eds., 2020), pp. 253–268, http://link.springer.com/10.1007/978-981-15-0614-7\_21.

88. A. Brown, L. Karkaby, M. Perovic, R. Shafi, G. Einstein, in Sex Differences in Brain Function and Dysfunction, C. Gibson, L. A. M. Galea, Eds. (Springer, 2022); vol. 62 of Current Topics in Behavioral Neurosciences, pp. 3–25.

89. K. Schaudig et al., BMJ 387, e079525 (2024).

90. A. H. Patel et al., BJOG Int. J. Obstet. Gynaecol. 0, 1–21 (2025).

91. W. F. Oliveira Amador et al., Eur. J. Obstet. Gynecol. Reprod. Biol. 312, 114552 (2025).

92. A. Morga et al., Menopause 31, 68–76 (2024).

93. A. Gosset, J.-M. Pouillès, F. Trémollieres, Best Pract. Res. Clin. Endocrinol. Metab. 35, 101551 (2021).

94. S. S. Faubion et al., Mayo Clin. Proc. 101, 946–954 (2026).

95. R. Götze Eriksson et al., Acta Obstet. Gynecol. Scand. 105, 1454–1467 (2026).

96. J. E. Rossouw et al., JAMA 288, 321–333 (2002).

97. J. M. Mehta, R. C. Chester, J. M. Kling, J. Womens Health 28, 705–711 (2019).

98. J. Mehta, J. M. Kling, J. E. Manson, Front. Endocrinol. 12, 2021.564781 (2021).

99. P. M. Maki, Menopause 20, 695–709 (2013).

100. A. L. Hersh, M. L. Stefanick, R. S. Stafford, JAMA 291, 47–53 (2004).

101. D. Bartz, A. Tadikonda, J. E. Manson, JAMA 335, 1297–1299 (2026).

102. C. E. Gleason et al., PLOS Med. 12, e1001833, discussion e1001833 (2015).

103. V. W. Henderson et al., Neurology 87, 699–708 (2016).

104. L. H. Coker et al., J. Steroid Biochem. Mol. Biol. 118, 304–310 (2010).

105. D. Grady et al., JAMA 288, 49–57 (2002).

106. V. M. Miller et al., Menopause 26, 1071–1084 (2019).

107. M. Lorentzon et al., Osteoporos. Int. 33, 2297–2305 (2022).

108. H. N. Hodis et al., N. Engl. J. Med. 374, 1221–1231 (2016).

109. H. N. Hodis, W. J. Mack, Cancer J. 28, 208–223 (2022).

---

REVIEWS

110. S. H. Baik, F. Baye, C. J. McDonald, Menopause 31, 363–371 (2024).

111. H. M. P. Boardman et al., Cochrane Database Syst. Rev. 2015, CD002229 (2015).

112. M. Y. Carson, R. C. Thurston, Curr. Opin. Endocr. Metab. Res. 30, 100448 (2023).

113. R. C. Thurston et al., J. Am. Heart Assoc. 10, e017416 (2021).

114. A. Reeves, M. R. Elliott, C. A. Karvonen-Gutierrez, S. D. Harlow, Int. J. Epidemiol. 52, 1612–1623 (2023).

115. C. J. Crandall, J. M. Mehta, J. E. Manson, JAMA 329, 405–420 (2023).

116. J. S. Rabin et al., Menopausal Hormone Therapy and Brain Health: Current Evidence, Knowledge Gaps, and Future Directions; http://dx.doi.org/10.2139/ssrn.7196839.

117. M. Š. Goldštajn et al., Arch. Gynecol. Obstet. 307, 1727–1745 (2023).

118. B. von Schoultz, Gynecol. Endocrinol. 25, 551–553 (2009).

119. R. J. Santen, Climacteric 18, 121–134 (2015).

120. I. Eriksson, Y. Gustafson, L. Fagerström, B. Olofsson, Arch. Gerontol. Geriatr. 50, 132–135 (2010).

121. M. Melville et al., Lancet Healthy Longev. 6, 100803 (2025).

122. T. Duka, R. Tasker, J. F. McGowan, Psychopharmacology 149, 129–139 (2000).

123. H. Joffe et al., Menopause 13, 411–422 (2006).

124. R. Krug, J. Born, B. Rasch, Psychoneuroendocrinology 31, 965–975 (2006).

125. M. A. Espeland et al., J. Am. Geriatr. Soc. 58, 1263–1271 (2010).

126. S. A. Shumaker et al., JAMA 291, 2947–2958 (2004).

127. B. B. Sherwin, M. Grigorova, Fertil. Steril. 96, 399–403 (2011).

128. V. W. Henderson, Climacteric 21, 333–340 (2018).

129. T. A. Puri et al., Neurology 105, e213995 (2025).

130. A. Biason-Lauber, M. Lang-Muritano, Mol. Cell. Endocrinol. 554, 111710 (2022).

131. N. Fuentes, P. Silveyra, Adv. Protein Chem. Struct. Biol. 116, 135–170 (2019).

132. L. Mosconi et al., Sci. Rep. 14, 12680 (2024).

133. C. Barth, A. Villringer, J. Sacher, Front. Neurosci. 19, 2015.00037 (2015).

134. R. B. Gibbs, Endocr. Rev. 31, 224–253 (2010).

135. R. S. Kumar, N. Goyal, Life Sci. 269, 119091 (2021).

136. J. K. Russell, C. K. Jones, P. A. Newhouse, Neurotherapeutics 16, 649–665 (2019).

137. M. Visser, J.-M. Foidart, H. J. T. Coelingh Bennink, Climacteric 11, 64–68 (2008).

138. S. R. Goldstein, Endometrial Cancer and Hormone Therapy Regimens: Do They Matter? International Menopause Society (2020); https://www.imsociety.org/2020/09/21/endometrial-cancer-and-hormone-therapy-regimens-do-they-matter/.

139. J. Carlson, J. Kulkarni, Neurosci. Biobehav. Rev. 187, 106760 (2026).

140. J. C. Carroll, E. R. Rosario, A. Villamagna, C. J. Pike, Endocrinology 151, 2713–2722 (2010).

141. C. Aguirre, A. Jayaraman, C. Pike, M. Baudry, J. Neurochem. 115, 1277–1287 (2010).

142. D. D. Murphy, M. Segal, Neuroendocrinology 72, 133–143 (2000).

143. B. Pletzer, K. Winkler-Crepaz, K. Maria Hillerer, Front. Neuroendocrinol. 69, 101060 (2023).

144. S. S. M. Faubion et al., Menopause 29, 767–794 (2022).

145. J. R. Peters et al., Mol. Psychiatry 30, 251–262 (2025).

146. E. Schelbaum et al., Neurology 97, e2328–e2339 (2021).

147. D. A. Swales et al., Psychoneuroendocrinology 156, 106339 (2023).

148. K. Kantarci et al., Neurology 90, e1404–e1412 (2018).

149. M. Boccardi et al., Menopause 13, 584–591 (2006).

150. J. J. Steventon et al., Psychoneuroendocrinology 158, 106393 (2023).

151. R. G. Simmons et al., Am. J. Obstet. Gynecol. 220, 376.e1–376.e12 (2019).

152. B. Y. Miao et al., NPJ Digit. Med. 8, 221 (2025).

153. M. D. Read, K. A. Edey, J. Hapeshi, C. Foy, Menopause Int. 16, 60–64 (2010).

154. T. A. Puri, J. E. Richard, L. A. M. Galea, Trends Neurosci. 46, 459–471 (2023).

155. T. A. Puri, S. E. Lieblich, M. Ibrahim, L. A. M. Galea, Horm. Behav. 165, 105616 (2024).

156. C. K. Barha, L. A. M. Galea, Neurobiol. Aging 32, 2091–2095 (2011).

157. C. K. Barha, L. A. M. Galea, Neurobiol. Aging 34, 986–1004 (2013).

158. L. T. Hoyt, A. M. Falconi, Soc. Sci. Med. 132, 103–112 (2015).

159. M. Perović, J. Hou, M. L. Mack, Biol. Sex Differ. 17, 65 (2026).

160. L. Gravelsins, K. Duncan, G. Einstein, PLOS ONE 16, e0252807 (2021).

161. E. Jacobs, M. D'Esposito, J. Neurosci. 31, 5286–5293 (2011).

162. C. C. Louis, E. Jacobs, M. D'Esposito, J. Moser, J. Cogn. Neurosci. 35, 1144–1153 (2023).

163. E. H. Corder et al., Science 261, 921–923 (1993).

164. R. N. M. Saleh, M. Hornberger, C. W. Ritchie, A. M. Minihane, Alzheimers Res. Ther. 15, 10 (2023).

165. Y. Yue et al., Acta Pharmacol. Sin. 28, 1129–1135 (2007).

166. K. Kantarci et al., Neurology 87, 887–896 (2016).

167. H. Depypere et al., Alzheimers Dement. 19, 1320–1330 (2023).

168. E. Kapoor et al., Maturitas 176, 107797 (2023).

169. A. M. Moyer et al., Menopause 25, 877–882 (2018).

170. D. Pereira, A. R. Pessoa, N. Madeira, A. Macedo, A. T. Pereira, Arch. Womens Ment. Health 25, 61–70 (2022).

171. E. Schleimann-Jensen et al., Br. J. Psychiatry 226, 401–409 (2025).

172. Y. Yang et al., JAMA Netw. Open 6, e2334545 (2023).

173. T. J. Reilly et al., J. Affect. Disord. 349, 534–540 (2024).

174. T. A. Eisenlohr-Moul et al., Psychol. Med. 50, 964–972 (2020).

175. A. N. Sheedy, J. Wactawski-Wende, K. M. Hovey, M. J. LaMonte, Menopause 30, 1199–1205 (2023).

176. J. LaClair et al., Urology 215, 43–48 (2026).

177. J. Morin, Y. Rolland, H. A. Bischoff-Ferrari, A. Ocampo, K. Perez, Aging Cell 23, e14334 (2024).

178. L. Cho et al., Circulation 147, 597–610 (2023).

179. F. R. Farina et al., JAMA Netw. Open 9, e2616596 (2026).

180. K. Schweitzer, JAMA 336, 535–538 (2026).

181. United States Menopause Market Size, Share, Trends, Growth. Spherical Insights (2025); https://www.sphericalinsights.com/reports/united-states-menopause-market.

182. T. Hanton et al., Lancet Obstet. Gynaecol. Women's Health 1, e249–e250 (2025).

183. IWK Foundation, The Voice of Maritime Women: The Unspoken Burden of Women's Health Survey Summary. IWK Foundation (2025); https://iwkfoundation.org/sites/default/files/users/2025-iwk-foundation-the-unspoken-burden-of-womens-health-survey-summary-october-2025.pdf.

184. J. T. Allen et al., Menopause 30, 1002–1005 (2023).

185. M. A. Hernán, W. Wang, D. E. Leaf, JAMA 328, 2446–2447 (2022).

186. US Preventive Services Task Force, JAMA 328, 1740 (2022).

187. S. Bunnewell, S. Keating, J. Parsons, M. Hickey, S. Hillman, BJOG Int. J. Obstet. Gynaecol., 2025, 70023 (2025).

188. R. E. Langley et al., N. Engl. J. Med. 394, 1595–1607 (2026).

189. A. S. Cheung, B. J. Nolan, S. Zwickl, Climacteric 26, 256–262 (2023).

190. R. Deif, M. Salama, Front. Psychiatry 12, 2021.650318 (2021).

191. P. G. Tepper et al., J. Clin. Endocrinol. Metab. 97, 2872–2880 (2012).

192. S. Khosla, Endocrinology 161, bqaa111 (2020).

193. Y. Jin, T. Wang, D. Liao, X. Jin, Int. J. Womens Health 18, 589117 (2026).

194. N. Kim, "Sex Difference of Gut Microbiota" in Sex/Gender-Specific Medicine in the

195. E. E. Sundermann, Alzheimers Dement. 21, e107385 (2025).

196. E. E. Telfer et al., Physiol Rev. 103, 2623–2677 (2023).

197. D. C. Woods, J. L. Tilly, Stem Cells 41, 200–204 (2023).

198. A. S. K. Jones et al., Sci. Adv. 10, eadm7506 (2024).

199. A. Converse et al., Mol. Hum. Reprod. 32, gaag038 (2026).

---

ACKNOWLEDGMENTS

Author contributions: Conceptualization: L.L.G., M.P., T.F.L.S., M.A.W. A.J.M., J.S.R., and L.A.M.G. Investigation – L.L.G., M.P., T.F.L.S., & M.A.W. Writing – original draft: L.L.G., M.P., T.F.L.S., M.A.W., and L.A.M.G. Writing – review and editing: L.L.G., M.P., T.F.L.S., M.A.W., A.J.M., J.S.R., and L.A.M.G. Visualization: L.L.G. and M.P. Supervision: L.A.M.G. Project Administration: L.L.G. Funding: L.A.M.G. and J.S.R. gratefully acknowledge funding by Wellcome Leap CARE (CARE2025). L.A.M.G. also receives funding from Cure for Alzheimer's Fund, NSERC (RGPIN-2024-05744.) and CIHR (PJT-186290). M.P. is supported by a Canadian Institutes of Health Research Postdoctoral Fellow (MFE-201012). L.L.G. and AM are funded in part by womenmind (L.A.M.G.: CAMHF-1197). MWA receives funding from CIHR and Alzheimer Society of Canada. T.F.L.S. is funded by an Ontario Graduate Scholarship. Competing interests: The authors declare no competing interests. L.A.M.G. serves on a Strategic Advisory Committee for Bioinnovation Fund. License information: Copyright © 2026 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www.science.org/content/page/science-licenses-journal-article-reuse

Submitted 7 July 2026; accepted 5 August 2026

10.1126/science.aeg5004

REVIEW

Sex-specific mechanisms of chronic pain

Archana Venkataraman $^{1\dagger}$ , Élora Midavaine $^{2\dagger}$ , Holly A. Ingraham $^{1*}$

For decades, we have known that women, especially younger women, are disproportionately affected by pain disorders, including irritable bowel syndrome, endometriosis, menstrual migraines, fibromyalgia, osteoarthritis, and neuropathic pain, all of which degrade quality of life. The adoption of sex as a biological variable in 2016 has begun to transform the field, revealing distinct sex-specific pathways that initiate, amplify, and resolve pain. In this Review, we synthesize recent advances across visceral and somatic pain, highlighting the impact of sex hormones, immune-neural cross-talk, the gut microbiome, and endogenous analgesic circuits. Rather than clinging to the historical standard that simply described the female bias in pain symptoms, these studies begin to explain why females experience heightened pain, paving the path for mechanism-based precision therapeutics.

Epidemiological data on chronic pain indicate that women will suffer from functional pain disorders at a rate of 2:1 compared with men and that in many instances, this ratio can climb much higher. For women, persistent, unresolved pain that outlives its utility as an essential warning system is debilitating. Perhaps the most striking example is chronic pelvic pain in women, affecting three major visceral organs: the gut, bladder, and reproductive organs. Many of these insidious pelvic pain conditions, including irritable bowel syndrome (IBS), endometriosis, interstitial cystitis, fibroid-associated pain, and vulvodynia, are episodic, unpredictable, and affect women during their reproductive years. Menstrual migraines and fibromyalgia experienced by younger women are equally debilitating. Later in life, the higher prevalence of osteoarthritis and bone loss in women compromises their mobility, leads to frailty, and accelerates aging. Although multiple factors initiate pain sensations in women, the fluctuation of ovarian hormones beginning at puberty appears to initiate many of these female-specific and female-prevalent pain conditions. Given the relatively short length of this Review, we have intentionally focused on recent progress that has occurred or is likely to emerge over the next 5 years. We have chosen to highlight progress in two areas of research, visceral pain and somatic pain (Fig. 1), that begin to define how hormonal signaling intersects with epithelial, immune, microbial, and neuronal pathways to generate female bias in pain disorders. Unfortunately, there continues to be a large disconnect between public awareness of pain experienced by women and basic research efforts aimed at treating female-biased debilitating pain disorders. Our goal is to shine some light into the large, unexplored black box of chronic pain experienced by millions of younger and older women worldwide.

---

Sex differences in visceral pain

Shared pain across pelvic organs is one of the signature features of visceral pain disorders in females. Notably, two visceral pain disorders—IBS and interstitial cystitis/bladder pain syndrome (IC/BPS)—are prevalent in women (Fig. 1, bottom right). These two disorders are often linked to endometriosis, a female-specific condition characterized by initial episodic bouts of inflammatory pain that transform into chronic, persistent pelvic pain. Fifty percent of women with IC/BPS and roughly 30% of women with IBS also suffer from endometriosis (1). Menstrual cycle-related changes in visceral pain sensitivity are well documented in IBS (2). Experimental studies show that whereas colonic and rectal pain sensitivity remain largely stable across cycle phases in healthy controls, women with IBS exhibit clear cyclical variation in visceral pain thresholds (3). Similar episodic patterns are observed in endometriosis (1), with pelvic pain often intensifying during menstruation and in some patients during ovulation. This close link underlies clinical management that suppresses the hypothalamic-reproductive axis, counteracting estrogen excess and progesterone resistance (4). The transition from episodic to chronic endometriotic pain is potentially driven by peripheral and central sensitization (5). That ovarian hormones influence sensory processing across multiple visceral organs is supported by the fact that women with IC/BPS also report perimenstrual increases in bladder pain (6). The frequent overlap among IBS, IC/BPS, and endometriosis raises the possibility that common hormone-sensitive cellular pathways contribute to pain across pelvic organs. To that point, no obvious sex differences were noted when comparing visceral sensitivity between healthy women taking oral contraceptives and men (7). Deciphering the mechanistic underpinnings and common logic of these pelvic pain-modulatory pathways represents a research area that will benefit from interdisciplinary approaches spanning multiple fields.

---

Sex-specific mechanisms driving visceral sensitivity

Female and male mice exhibit distinct baseline states of visceral sensory activity. Intravital calcium imaging revealed higher baseline activity in colon-innervating dorsal root ganglion (DRG) neurons in females versus males (8). In males, isovalerate (ISV), a bacterially derived short-chain fatty acid (SCFA), increased the activity of sensory enterochromaffin (EC) cells in the gut epithelium to sensitize downstream mucosal afferents (Fig. 2). EC cells then release serotonin to excite nearby 5-HT $_{3}$ receptor (5-HT $_{3}$ -R)-expressing sensory nerve fibers. Females, however, were largely insensitive to acute ISV treatment, and chemogenetic activation of these serotonergic EC cells produced greater visceral hypersensitivity in males than females (8). By comparison, silencing EC cells elicited a greater reduction in pain responses in females. Collectively, these findings suggest that females operate from a heightened basal sensory state, thereby limiting the dynamic range available for further acute sensitization, whereas males retain greater capacity for stimulus-induced sensitization, such as that observed with ISV. Contribution of serotonergic signaling on sex-specific visceral pain appears to depend not only on stimulus-evoked serotonin release but also on mechanisms regulating serotonin availability. In serotonin transporter (SERT)-deficient rats, impaired serotonin reuptake produced female-specific visceral hypersensitivity driven by enhanced 5-HT $_{3}$ -R signaling (9). Repeated or sustained serotonergic signaling by EC cells may therefore drive long-term plasticity within visceral sensory circuits, transforming transient changes in afferent excitability into persistent visceral hypersensitivity.

Beyond epithelial-derived serotonergic signaling, sex differences in visceral pain may stem from the intrinsic properties of sensory afferents, specifically through colonic mechanosensation. Even though expression of the mechanosensory ion channel PIEZO2 (piezo-type mechanosensitive ion channel component 2) is modestly higher in female sensory afferents, chemogenetic activation of Piezo2-expressing colonic sensory neurons induced hypersensitivity only in males (10). Together, these findings suggest that serotonergic signaling in the epithelium and PIEZO2 mechanotransduction in sensory afferents are complementary regulators of visceral sensitivity. Whether and how serotonergic and PIEZO2-dependent signaling converge within the same sensory neurons or operate in parallel afferent circuits remains an important question for future studies.

The use of chemically induced colonic injury models in male and female mice reveals marked sex differences in disease progression, pain sensitivity, and recovery. For example, intracolonic capsaicin administration resulted in acute hypersensitivity in female mice. However, dextran sodium sulfate (DSS)-induced colitis was far more severe and persistent in males $11$ . Extending these observations, males and females developed distinct patterns of gut microbial dysbiosis and persistent visceral hypersensitivity even after the resolution of DSS-induced colitis $12$ . In females, postinflammatory dysbiosis was accompanied by chronic visceral pain that could be transferred to healthy mice of both sexes by fecal transplantation, implicating the gut microbiome as a mediator of persistent pain. Using the zymosan-induced model of colitis, which activates the innate immune system, hypersensitivity was observed only in males $13$ . These divergent findings using various colonic injury models strongly suggest that sex differences in visceral pain depend on the nature of the initiating insult. Knowing how hormonal signaling pathways intersect with inflammatory pathways may explain why some episodes of intestinal inflammation resolve, whereas others progress to persistent visceral pain.

---

Hormonal modulation of visceral sensitivity

Recent work has identified a cellular mechanism that may explain the heightened visceral sensitivity observed in females (14). Estrogen receptor $\alpha$ (ER $\alpha$ ) signaling in the distal colonic epithelium triggers paracrine cross-talk between sparse enteroendocrine cells and nearby sensory neurons, thereby creating a female-specific sensory pain pathway. Specifically, estrogen increased peptide YY (PYY) release from colonic L cells and up-regulated expression of the SCFA receptor Olfr78, enhancing L cell responsiveness to bacterially derived metabolites, which reach exceedingly high levels in the distal colon after consuming foods high in fermentable carbohydrates (15) (Fig. 2). PYY subsequently increased serotonergic signaling from neighboring EC cells, thereby activating nociceptive 5-HT $_{3}$ -R-expressing mucosal afferents. These findings not only identify molecular targets through which estrogen can amplify visceral sensory signaling but may also explain why a low-carbohydrate diet is often prescribed for IBS (15). Whether similar hormone-sensitive epithelial pathways exist in adjacent pelvic organs to enhance visceral pain remains an open question.

Fig. 1. Visceral and somatic pain syndromes with female prevalence. (Bottom right) Common pelvic pain disorders experienced by women of reproductive age are shown and listed, with the anatomical proximity of these visceral organs shown above the pelvic floor. (Top left) Common somatic pain disorders with higher prevalence in women. Of those listed, osteoarthritis is the only pain disorder specific to postmenopausal women, whereas the other listed somatic pain disorders are experienced by women of reproductive age. Although temporomandibular joint (TMJ) disorder is not formally discussed, this type of pain can lead to migraines, facial pain, earaches, and eventually degradation of the oral cavity due to teeth grinding.

Estrogen also directly sensitizes sensory neurons. Using a mouse model of uterine pain, it was demonstrated that catechol estrogens, which are metabolized from estradiol (E2) and estrone, induce nociceptor hyperexcitability by activating the transient receptor potential ankyrin 1 (TRPA1) channel (16), leading to increased calcium influx and neuronal firing in TRPA1-expressing sensory neurons. Pharmacological inhibition or genetic deletion of TRPA1 markedly attenuated this hormone-induced hypersensitivity. These findings provide a direct mechanism by which estrogen metabolism can rapidly enhance nociceptor excitability, contrasting the slower genomic actions of estrogen in the sensory epithelium (14).

Another dimension of hormone signaling and visceral gut pain includes the role of the gut microbiome in regulating hormone availability (Fig. 2). Indeed, eliminating the microbiota using germ-free mice abolished the changes in visceral sensitivity associated with the estrous cycle and ovariectomy, indicating that microbial communities are required for normal hormone-mediated modulation of nociception (17). Supporting this intimate relationship between estrogen and the microbiome, it was recently shown that 3β-hydroxysteroid dehydrogenase (3β-HSD) derived from bacteria directly facilitates conversion to E2, increasing bioactive estrogens in the host (18). Furthermore, gut bacteria can perform unexpectedly complex steroid transformations, converting host glucocorticoids into progestin-like compounds under specific metabolic conditions (19). In addition, transplanting gut microbiota from women with fibromyalgia—a highly female-prevalent disorder leading to chronic, debilitating musculoskeletal pain—into germ-free mice induces pain and immune changes that mimic those observed in human patients (20). Thus, microbial regulation of steroid metabolism represents an additional facet of hormone regulation of visceral pain, especially when considered alongside changing lifestyles and life stages or hormone therapies.

In contrast to estrogen, the role of androgens in visceral pain remains less defined, but emerging evidence suggests that, unlike estrogens, this sex steroid dampens rather than drives heightened visceral sensitivity. Lower circulating concentrations of free testosterone correlate with greater symptom severity in patients with IBS, which is consistent with the finding that androgen depletion induces bowel dysfunction in mice, which is then normalized by androgen replacement (21). Mechanistically, androgens signal through the androgen receptor (AR) in enteric neurons and spinal afferents, a signaling pathway required for normal intestinal transit (22). The same study also found that $\beta$ -glucuronidase produced by gut bacteria generates active androgens that signal to enteric neurons, thereby stimulating motility and restoring intestinal homeostasis. To date, it remains uncertain whether AR is present in the colonic epithelium, as was recently shown for ER $\alpha$ (14).

The high female bias of visceral pain disorders has been leveraged to develop sex-specific therapies. After stratifying therapeutic outcomes by sex, randomized controlled trials revealed sex-dependent efficacy of the 5-HT $_{3}$ -R antagonist alosetron, the neurokinin 2 receptor antagonist ibodutant, and the CFTR chloride channel inhibitor crofelemer in the treatment of IBS (23). Furthermore, cognitive behavioral therapy, combined with medical management, was more effective in women than in men. By contrast, evidence for sex-specific treatment effects in IC/BPS remains limited despite the disorder's strong female predominance. For endometriosis, current effective treatments include hormone-based therapies that utilize oral contraceptives, progestins, or gonadotropin-releasing hormone (GnRH) drugs, all of which reduce estrogen-driven lesion activity and pain (1, 4). Surgical excision also provides symptom relief, although in some patients this is transient because pain persists when central sensitization or overlapping visceral pain disorders such as IBS and IC/BPS are present (1, 5). Regardless, the observation that symptoms improve after natural or surgical menopause further supports the critical role of sex hormones in both the pathophysiology and treatment responsiveness of endometriosis. Identifying the molecular aspects of hormone-sensitive pathways in the gut and other pelvic organs may provide opportunities for mechanism-based therapies that can selectively target pain-generating circuits while preserving the physiological functions of systemic sex hormones.

Fig. 2. Sex-specific chronic visceral pain. The illustration depicts recent progress in understanding how sex hormones regulate gut-microbiome-neural communication, thereby influencing visceral sensitivity. Endocrine signals—including estrogen, androgens, and stress hormones of the hypothalamic-pituitary-adrenal neuroendocrine axis—indirectly or directly regulate the gut-brain axis. Within the colonic lumen, the gut microbiota, microbial metabolites, and steroid-transforming microbial activity interact with the intestinal mucosa to modulate epithelial function. Specialized sensory enteroendocrine cells—including enterochromaffin (EC) cells (red) and L cells (blue)—functionally interact, with EC cells directly engaging 5-HT $_{3}$ -R $^{+}$ mucosal sensory nerve endings that send information to the spinal cord. Other visceral afferent populations that innervate the colonic muscle layers also contribute to nociceptive signaling. Adjacent immune mediators further contribute to inflammatory signaling through neuroimmune interactions. Visceral sensory signals, mediated by PIEZ02 $^{+}$ and TRPV1 $^{+}$ nociceptors, can propagate sex-biased visceral hypersensitivity via pelvic and splanchnic nerve input to the dorsal root ganglia, ultimately reaching the spinal cord and brain.

---

Central and peripheral neuronal differences associated with visceral pain

Limited evidence suggests that sex likely influences pain processing across multiple levels of the nervous system. Single-cell analyses of sensory neurons have identified sex-dependent differences in nociceptor gene expression and translational responses, indicating that male and female afferents may engage distinct molecular programs in response to inflammatory signals (24, 25). At the systems level, neuroimaging studies reveal sex-specific alterations in brain regions and networks involved in pain integration, emotional regulation, and stress responses, including the insula, anterior cingulate cortex, amygdala, and cortico-striatal-thalamic pathways (26, 27). Female-specific epigenetic regulation of glucocorticoid signaling in the amygdala has also been linked to stress-induced visceral hypersensitivity, suggesting that dysregulation of the hypothalamic-pituitary-adrenal stress axis contributes to persistent, unresolved pain (28). Beyond pain perception, the fact that central circuits exert descending control over visceral sensations and organ physiology suggests that sex differences in the brain can also influence visceral pain. With newer technologies, it should be possible to generate a comprehensive molecular and cellular framework to understand the complex brain-body dialog between the brain and pelvic organs.

---

Neuroimmune mechanisms of sex differences in pain

Neuroimmune interactions within the nervous system and its surrounding tissues represent another important mechanism through which sex differences influence chronic pain. The exchange of information between sensory neurons and their microenvironment shapes inflammatory responses and modulates pain processing involving the innate and adaptive immune systems. In this setting, the immune system controls tissue homeostasis and regulates neuronal activity after injury, modulating responses in the DRG and the spinal cord (Fig. 3A). Early studies using intrathecal injections of the endotoxin lipopolysaccharide (LPS) to generate mechanical allodynia (a condition in which innocuous stimuli such as light touch become painful) affected only male mice. This sex-specific pain phenotype required Toll-like receptor 4 (TLR4) activation and was also observed in an inflammatory pain model involving injection of complete Freund's adjuvant (CFA) or in neuropathic pain models involving spared nerve injury (SNI) or transection of sciatic nerve branches (29). A subsequent study showed that male mice depend predominantly on spinal microglia to develop and maintain pain hypersensitivity, whereas female mice rely on adaptive immune cells, particularly circulating T lymphocytes (30). These studies represented a paradigm-shifting discovery in pain, challenging the long-standing assumption that mechanisms identified in male animals are universally applicable across sexes, and prompting others to begin asking how distinct neuroimmune pathways govern pain processing in both sexes. As discussed below, the field has since evolved beyond the original microglia-versus–T cell paradigm, revealing multiple immune cell populations that act within distinct anatomical compartments to promote, modulate, and resolve pain.

---

Sex-specific neuroimmune pathways in chronic pain

Adaptive immunity, in which T and B lymphocytes elicit antigen-specific responses, has recently been linked to nociception in females (30–33). Indeed, female mice exhibited higher numbers of circulating CD4 $^{+}$ and CD8 $^{+}$ T cells in the blood and higher expression of T cell markers in the spinal cord after nerve injury, implying that adaptive immunity plays a major role in neuropathic pain for females (30). More recently, a female-specific role for CD8 $^{+}$ T cells was identified in this model, whereby pannexin-1 (PANX1), a large-pore membrane channel, was implicated in mechanical allodynia selectively in female mice and rats (31). In addition to PANX1-induced release of excitatory transmitters such as adenosine triphosphate (ATP) and glutamate from CD8 $^{+}$ T cells, leptin release from these cells has also been proposed to drive this pain response (31). Despite these insights, the anatomical site where T cells exert their pronociceptive effects in females remains unresolved. Whether lymphocytes infiltrate the spinal cord parenchyma or act within the surrounding meninges is unclear. This distinction is particularly relevant given the growing consensus that immune cell infiltration into the spinal cord parenchyma is uncommon in most models of chronic pain.

In contrast to effector T cells that promote pain, regulatory T cells ( $T_{regs}$ ) suppress pain and exert robust antinociceptive functions in females. $T_{regs}$ are a specialized subset of CD4 $^{+}$ T cells characterized by expression of the transcription factor FOXP3 and are essential for maintaining immune tolerance and limiting excessive inflammatory responses. The role of $T_{regs}$ in sex-specific pain regulation was first demonstrated by a functional interaction between $T_{regs}$ and microglia (32, 34). After peripheral nerve injury, colony-stimulating factor 1 (CSF-1) is released in the spinal cord from the central terminals of injured primary afferent neurons (Fig. 3A). Administering exogenous CSF-1 induced robust microglial activation and mechanical hypersensitivity in male mice, whereas in females, CSF-1 promoted the expansion of $T_{regs}$ within the spinal meninges. These meningeal $T_{regs}$ suppress microglial activation and prevent the development of CSF-1-induced pain hypersensitivity, potentially revealing a female-specific neuroimmune mechanism that restrains the maladaptive transition from acute to chronic pain (32). As a key component of an endogenous female-specific analgesic circuit, meningeal $T_{regs}$ are regulated by ovarian hormones, which drive their expansion and activation through the production of the endogenous opioid peptide enkephalin (33). $T_{reg}$ -derived enkephalin activates $\delta$ -opioid receptors expressed by Mas-related G protein-coupled receptor D (MRGPRD)-positive sensory neurons in DRG, reducing neuronal excitability and attenuating neuropathic pain in female mice (Fig. 3B). Notably, this analgesic function was independent of the canonical immunosuppressive properties of $T_{regs}$ , revealing an unexpected female-specific neuroregulatory role for adaptive immune cells that would be predicted to modulate chronic pain conditions during different life stages, such as pregnancy.

A

---

Macrophage-neuron cross-talk in females

In addition to adaptive immune cells, the innate immune system provides rapid, nonspecific defense through macrophages, neutrophils, and dendritic cells. Specifically, macrophages contribute to female-specific pain mechanisms by directly contacting sensory neurons (35). An IL-23/IL-17A/TRPV1 (IL-23, interleukin-23; IL-17A, interleukin-17A; TRPV1, transient receptor potential vanilloid 1) signaling axis has been identified in which the macrophage-derived proinflammatory cytokine IL-23 promotes IL-17A-dependent mechanical hypersensitivity selectively in female mice (35) (Fig. 3C). This pathway requires ERα signaling in TRPV1 $^{+}$ nociceptors and is suppressed by androgens in males, highlighting the critical influence of sex hormones on neuroimmune communication. Furthermore, use of the chemotherapy drug paclitaxel to induce neuropathic pain increased IL-23 and IL-17A cytokines in females, implicating macrophage-neuron cross-talk as a key driver of sex-specific chemotherapy-induced pain. The identification of this female-specific pathway suggests that sex-dependent neuroimmune mechanisms may shape therapeutic responses, raising the notion that IL-23- and IL-17-targeting biologics could be repurposed for sex-informed pain management.

C

Nociception and adaptive immunity

Nociception and innate immunity

Fig. 3. Sex-specific adaptive and innate neuroimmune circuits underlying chronic pain. (A) After peripheral nerve injury, injured sensory neurons release colony-stimulating factor 1 (CSF-1), which engages distinct neuroimmune pathways in males and females. A microglia-dependent pain program leads to mechanical hypersensitivity in males, which is suppressed in females by meningeal regulatory T cells ( $T_{regs}$ ). (B) In females, the ovarian hormones (E2 and P4) suppress nociception by promoting the expansion and activation of $T_{regs}$ , which release enkephalin acting on $\delta$ -opioid receptors expressed by MRGPRD $^{+}$ sensory neurons. (C) In females, macrophages drive mechanical hypersensitivity through an IL-23/IL-17A/TRPV1 signaling axis that requires estrogen receptor $\alpha$ (ER $\alpha$ ) signaling. (D) In males, testosterone increases IL-10–producing monocytes and accelerates pain resolution. Together, these findings demonstrate that sex hormones orchestrate distinct neuroimmune circuits that converge on sensory neurons to either promote or restrain chronic pain.

Recent evidence suggests that in addition to the generation of pain, the innate immune system also plays an important role in pain resolution, a process that is slower in women (36). When CFA was used to induce skin inflammation, female mice exhibited delayed resolution of inflammatory pain compared with males, along with a concomitant reduction in the number of anti-inflammatory IL-10-producing monocytes at the site of inflammation (37). Notably, administration of dihydrotestosterone to ovariectomized females increased the abundance of IL-10-producing monocytes and accelerated pain resolution to levels observed in males, implicating androgen-dependent regulation of monocyte function as a determinant of sex differences in pain duration (Fig. 3D). The authors further demonstrated similar sex-dependent associations in a cohort of patients after traumatic injury. Comparison between men and women after traumatic injuries (car accidents) revealed that reported pain severity was lower in men several weeks postaccident, along with the detection of higher circulating monocyte counts and IL-10 concentrations (37). Thus, for pain resolution, monocyte-derived IL-10 appears to serve as a key endogenous analgesic signal in both mice and humans. Although IL-10 has long been considered a promising therapeutic target because of its potent anti-inflammatory and proresolving properties, its short half-life and the limited efficacy of IL-10-based therapies have hindered direct clinical application. These findings instead suggest that harnessing endogenous IL-10 production through monocyte recruitment may represent a more effective strategy for promoting pain resolution. More broadly, they imply that stimulating endogenous immune-mediated analgesic pathways, rather than administering anti-inflammatory cytokines exogenously, may provide a therapeutic framework for chronic pain.

---

Pain hypersensitivity in males

In contrast to females, whose pain signaling relies heavily on adaptive immune cells, testosterone-mediated regulation of microglia has emerged as a central mediator of pain hypersensitivity in male rodents (31, 38), contributing to central sensitization through the release of neuromodulatory and proinflammatory factors. Accordingly, microglia-mediated pain mechanisms can be induced in ovariectomized females supplemented with testosterone and in females lacking adaptive immune cells in both mouse models of CFA-induced inflammatory pain and chronic constriction injury (CCI)-induced neuropathic pain models, indicating that both hormonal and immune factors shape sex-specific pain pathways (30, 38). A key mediator of this male-specific response is the metabotropic purinergic receptor P2X4, which is selectively up-regulated in spinal microglia after nerve injury (30). In the rat CCI model of neuropathic pain, activation of P2X4 triggers phosphorylation of p38 mitogen-activated protein kinase (MAPK) and subsequent release of brain-derived neurotrophic factor, promoting central sensitization and mechanical hypersensitivity. Sex differences in this signaling pathway appear to be regulated by the transcription factor interferon regulatory factor 5 (IRF5), which directly drives P2X4 expression in male microglia (38). However, upstream signals responsible for IRF5 induction after nerve injury have yet to be identified. TLR4 signaling has emerged as a potential upstream mediator of this male-specific microglial activation. As discussed above, intrathecal administration of LPS, a canonical TLR4 agonist, induces mechanical hypersensitivity in male but not female mice, an effect dependent on testosterone (29). Notably, this sexual dimorphism appears to be restricted to the spinal cord, because both sexes develop nociceptive responses after intracerebroventricular or intraplantar LPS administration. In males, TLR4 signaling contributes to inflammatory and neuropathic pain, highlighting its role as a key regulator of male-specific pain processing. However, the cellular target of TLR4 signaling that initiates this pathway and the question of whether TLR4 acts upstream of the IRF5-P2X4 signaling axis remain unresolved. The spinal cord–restricted nature of this sexual dimorphism suggests that the critical determinant may extend beyond TLR4 expression itself to the organization of sex-specific neuroimmune interactions within the spinal cord.

After SNI in mice, CSF-1 is up-regulated in sensory neurons and transported to the spinal dorsal horn, where it activates microglia and promotes the development of neuropathic pain in males. CSF-1 receptor (CSF1R) signaling activates the microglial adaptor protein DAP12, which is required for the development of pain hypersensitivity. In parallel, CSF1R signaling also drives microglial proliferation through a DAP12-independent pathway, indicating that microglial expansion and pain sensitization are mediated by distinct downstream mechanisms $34$ . By contrast, as discussed above, female mice are protected from CSF-1-induced microglial activation and pain hypersensitivity $32$ . CSF-1 thus emerges as a critical determinant of sex-specific pain pathways, promoting microglial activation in males while driving meningeal $T_{reg}$ expansion in females (Fig. 3, A and B). Understanding how this divergence is established may reveal fundamental mechanisms governing sex-specific pain biology and identify potential therapeutic targets.

---

Head, body, and joint pain

Three chronic somatic pain conditions affecting women include menstrual migraines, fibromyalgia, and osteoarthritis (Fig. 1). Even though our working knowledge of these three conditions remains inadequate, we briefly highlight each as an area for future research. Among these conditions, migraine affects younger women two to three times more frequently than men. This female predominance is strongly influenced by fluctuating hormones, with symptoms peaking just before or during menses and beginning at puberty. The frequency and severity of these so-called menstrual migraines dissipate during the second and third trimesters of pregnancy, as placental hormones and sex steroids, including estrogen, rise dramatically (39, 40). Calcitonin gene-related peptide (CGRP) is a central mediator of migraine pathophysiology and a major therapeutic target, with CGRP-blocking monoclonal antibodies demonstrating clinical efficacy, particularly in women (41). Preclinical studies in mice revealed that CGRP application to the dura mater induces facial mechanical hypersensitivity in female mice but has little effect in males (42, 43). Female mice also exhibit a greater abundance of CGRP-expressing neurons within the ophthalmic (V1) division of the trigeminal ganglion, providing a potential anatomical substrate for enhanced CGRP sensitivity and migraine susceptibility (43).

Despite the central role of CGRP in migraine pathophysiology, few mechanisms have emerged to explain the female-biased CGRP responsiveness and migraine susceptibility. Notably, this effect appears to be independent of ovarian hormones. However, the anterior pituitary hormone prolactin (PRL) amplifies neuronal excitability and migraine-like behaviors in mice and is indirectly linked to CGRP release (42, 44). PRL evokes greater nociceptive responses in females than in males (42), which is consistent with higher PRL receptor (Prlr) gene expression in female sensory neurons. Indeed, conditional deletion of Prlr from sensory neurons attenuates CGRP-induced migraine-like behaviors in females, suggesting that PRL participates in sex-specific CGRP-mediated nociception (42). The only wrinkle in this link is that PRL itself is regulated by estrogen, whose concentrations rise during the luteinizing hormone surge of the estrous cycle and just before birth (45). Relevant to pain, PRL concentrations also rise with stress or endogenous opioids (46), raising the question of why it would be beneficial to amplify pain responses during these female life stages. Clearly, more work is needed to understand the interplay between estrogen, PRL, and CGRP signaling across distinct physiological states to understand the female predominance of headache.

Unfortunately, our current understanding of fibromyalgia offers little insight for the millions of mostly younger women suffering from this chronic somatic pain condition. A recent genome-wide association study (GWAS) of more than 2.5 million individuals, including more than 50,000 affected individuals, found that fibromyalgia variants showed a strong association with IBS and enrichment in enteric nerves but not with biological sex or classical autoimmune disorders (47). Findings from this study mirror those of other GWASs that identified overlapping risk pathways beyond autoimmunity but shared across pelvic pain disorders, migraine, and osteoarthritis (48, 49).

Osteoarthritis (OA) is a degenerative joint disease characterized by progressive cartilage loss, impaired joint function, and chronic pain. As women age and become estrogen-depleted, their risk of developing OA increases, especially in the knee, where rates can be twice that of men (50). Mechanisms underlying this hormone-related sex difference are poorly understood. Concentrations of the chemokines (CCL2 and IL-8) in the blood, cerebrospinal fluid, and synovial fluid of the knee joint all correlate with knee pain in women but not men, indicating that OA pain arises from neuroimmune interactions (51). Evidence from preclinical models also points to enhanced inflammatory responses in females. In the context of OA of the temporomandibular joint, female rats developed more severe cartilage degeneration and bone remodeling than males, accompanied by increased macrophage infiltration and elevated IL-1β expression in the synovial membrane (52). Moreover, silencing estrogen receptor reduces pathological joint remodeling, implicating estrogen-dependent inflammatory pain pathways. Together, these studies suggest that OA pain and joint damage in females are shaped by hormonal status and neuroimmune interactions. Understanding how immune cells and sensory neurons communicate within the arthritic joint may shed light on the female predominance of OA pain.

---

Summary

The close link between fluctuating hormones and symptoms is a common feature of many types of female pain, perhaps best illustrated by the sudden onset of menstrual migraine headaches. This association raises two important questions. First, can hormone-responsive target genes be identified in affected nerves and tissues to begin explaining how chronic pain develops? Second, why do only some women develop symptoms? We would speculate that the higher rates of pain experienced by reproductive-aged women may confer an evolutionary advantage for reproductive fitness by heightening sensitivity during different life stages, as would be the case for pelvic organs during pregnancy. Future breakthroughs in treating female chronic pain disorders will only come when mechanistic frameworks emerge from preclinical and clinical studies, a goal that will require focused team efforts and adequate funding. Women are desperately waiting for concrete answers that might help eliminate the common “all in your head” approach to their illnesses, shorten the long wait between symptoms and diagnosis/treatment, and ultimately provide mechanism-based strategies for disease treatment.

---

REFERENCES AND NOTES

1. A. W. Horne, S. A. Missmer, BMJ 379, e070750 (2022).

2. R. P. Sarnoff et al., Neurogastroenterol. Motil. 37, e14977 (2025).

3. L. A. Houghton, R. Lea, N. Jackson, P. J. Whorwell, Gut 50, 471–474 (2002).

4. S. E. Bulun, J. Clin. Invest. 135, e188787 (2025).

5. J. Maddern, L. Grundy, J. Castro, S. M. Brierley, Front. Cell. Neurosci. 14, 590823 (2020).

6. M. L. Schomacker, K. E. Hansen, C. H. Ramlau-Hansen, A. Forman, Eur. J. Obstet. Gynecol. Reprod. Biol. 231, 65–69 (2018).

7. A. Icenhour et al., Neurogastroenterol. Motil. 31, e13664 (2019).

8. J. R. Bayrer et al., Nature 616, 137–142 (2023).

9. N. El-Ayache, J. J. Galligan, Am. J. Physiol. Gastrointest. Liver Physiol. 316, G132–G143 (2019).

10. J. Madar et al., Nat. Commun. 14, 2158 (2023).

11. A. M. Francis-Malavé et al., Pain 164, 197–215 (2023).

12. M. J. Arzamendi et al., Gut Microbes 16, 2409207 (2024).

13. T. Guo et al., Am. J. Physiol. Gastrointest. Liver Physiol. 326, G133–G146 (2024).

14. A. Venkataraman et al., Science 390, 1285–1291 (2025).

15. H. M. Staudacher, K. Whelan, Gut 66, 1517–1527 (2017).

16. Z. Xie et al., JCI Insight 7, e149107 (2022).

17. M. Tramullas et al., iScience 24, 102850 (2021).

18. D. Li et al., Cell Metab. 35, 685–694.e5 (2023).

19. M. D. McCurry et al., Cell 187, 2952–2968.e13 (2024).

20. W. Cai et al., Neuron 113, 2161–2175.e13 (2025).

21. D. Rastelli et al., J. Clin. Invest. 132, e150789 (2022).

22. V. N. Lagomarsino et al., Nat. Neurosci. 29, 1791–1800 (2026).

23. L. van Kessel, D. Teunissen, T. Lagro-Janssen, Int. J. Gen. Med. 14, 867–884 (2021).

24. J. Mecklenburg et al., Sci. Rep. 10, 15278 (2020).

25. D. Tavares-Ferreira et al., Biol. Psychiatry 91, 129–140 (2022).

26. J. S. Labus et al., Pain 164, 292–304 (2023).

27. N. R. Osborne et al., Pain 162, 97–108 (2021).

28. T. Louwies, B. Greenwood-Van Meerveld, Neurogastroenterol. Motil. 34, e14377 (2022).

29. R. E. Sorge et al., J. Neurosci. 31, 15450–15454 (2011).

30. R. E. Sorge et al., Nat. Neurosci. 18, 1081–1083 (2015).

31. C. Y. Fan et al., Neuron 113, 896–911.e9 (2025).

32. J. A. Kuhn et al., eLife 10, e69056 (2021).

33. É. Midavaine et al., Science 388, 96–104 (2025).

34. Z. Guan et al., Nat. Neurosci. 19, 94–101 (2016).

35. X. Luo et al., Neuron 109, 2691–2706.e5 (2021).

36. R. B. Fillingim, C. D. King, M. C. Ribeiro-Dasilva, B. Rahim-Williams, J. L. Riley III, J. Pain 10, 447–485 (2009).

37. J. Sim et al., Sci. Immunol. 11, eadx0292 (2026).

38. J. C. S. Mapplebeck et al., Pain 159, 1752–1763 (2018).

39. E. A. MacGregor, J. Fam. Plann. Reprod. Health Care 33, 83–93 (2007).

40. S. S. Faubion, P. Batur, A. H. Calhoun, Mayo Clin. Proc. 93, 639–645 (2018).

41. F. Porreca et al., Cephalalgia 44, 03331024241238153 (2024).

42. A. Avona et al., Ann. Neurol. 89, 1129–1144 (2021).

43. G. Guzman et al., Cephalalgia 45, 03331024251317446 (2025).

44. S. Singh et al., J. Headache Pain 25, 63 (2024).

45. H. R. Phillipps, Z. Khant Aung, D. R. Grattan, J. Neuroendocrinol. 34, e13129 (2022).

46. J. Rossier et al., Proc. Natl. Acad. Sci. U.S.A. 77, 666–669 (1980).

47. I. Kerrebijn et al., Nat. Med. 32, 3060–3070 (2026).

48. N. Rahmioglu et al., Nat. Genet. 55, 423–436 (2023).

49. F. Yang et al., Cell Rep. Med. 4, 101250 (2023).

50. J. D. Steinmetz et al., Lancet Rheumatol. 5, e508–e522 (2023).

51. E. Kosek et al., J. Neuroimmunol. 321, 49–60 (2018).

52. X. T. Xue et al., Oral Dis. 24, 1503–1513 (2018).

---

ACKNOWLEDGMENTS

We thank W. Yue, D. Julius, and S. Kashem for many helpful suggestions and critical comments on the manuscript. Funding: This work was supported by National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) 1K99DK139516 (A.V.), National Institute of Neurological Disorders and Stroke (NINDS) 1K99NS142248 (E.M.), a Banting Postdoctoral Fellowship [Canadian Institutes of Health Research (CIHR) award no. BPF198457] (E.M.), and NIDDK R01 DK135714 (H.A.I.). Author contributions: Conceptualization: A.V., E.M., H.A.I.; Visualization: A.V., E.M., H.A.I.; Writing – original draft, review & editing: A.V., E.M., H.A.I. Competing interests: The authors declare that they have no competing interests. License information: Copyright © 2026 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www.science.org/about/science-licenses-journal-article-reuse

Submitted 17 July 2026; accepted 7 August 2026

10.1126/science.aeh4468

RESEARCH

IN SCIENCE JOURNALS

Edited by Michael Funk

BIOMATERIALS

---

An evolving structure increases strength

Many composites use a brick-and-mortar structure to increase toughness. However, these materials lack stiffness, and most methods to improve this property result in a lower toughness. Chen et al. explored the structural origins of the unusual combination of stiffness and toughness in the nacre of mature Cristaria plicata, a bivalve mussel. They found that the stiffness comes from a hierarchical crystalline organic-inorganic framework composed of aragonite nanorods and nanoconfined biomolecular crystallites that evolves from the brick-and-mortar arrangement as the mussel matures. The authors created a semisynthetic nacre based on isolated aragonite tablets, which they infiltrated with a synthetic polymer and reprocessed into films. —Marc S. Lavine Science p. 81, 10.1126/science.aeg1070

Mussel nacre contains overlapping inorganic crystallites, pictured here in a scanning electron microscopy image, that simultaneously provide toughness and stiffness.

---

POLAR WANDER

---

Getting around quickly

True polar wander (TPW), which occurs when the solid outer Earth shifts relative to its spin axis, is difficult to detect with standard approaches using paleomagnetic measurements. Domeier et al. used a different strategy to detect TPW in the Mesozoic Era using reconstructions of sea level as a proxy. They found a robust quadru polar pattern that reflects global sea-level fluctuations associated with relative changes in centrifugal potential that occurred due to rapid TPW. This demonstrates that coupling in Earth's rotational dynamics, mantle processes, and surface environments can be highly episodic. —Jesse Smith

Science p. 87, 10.1126/science.aec8423

---

ELECTROCHEMISTRY Greener route to adiponitrile synthesis

Electrochemical hydrodimerization of acrylonitrile offers a cleaner route to adiponitrile, a key precursor for nylon 6,6, but current systems rely heavily on toxic lead or cadmium electrodes and suffer from

poor energy efficiency. Du et al. developed a carbon-based electrosynthesis strategy that uses benzyl-substituted quaternary ammonium cations to engineer the electrode-electrolyte interface. They found multiple noncovalent interactions between benzyl groups and carbon, which enriched acrylonitrile near the electrode, promoted radical coupling, and suppressed side reactions. Even in scaled-up electrolyzers, the system achieved selectivity comparable to lead electrodes but at substantially lower overpotentials. —Jack Huang

Science p. 91, 10.1126/science.aef5175

---

PLANETARY ACCRETION Early gases hitched rides on metal

Earth and other planets formed by collision of smaller bodies. However, it is not clear how these smaller bodies were able to gravitationally retain and deliver the light, volatile building blocks of our atmosphere. Gaillard et al. ran high-pressure experiments on a starting material representative of the “magma oceans” on small, early planetary bodies. They observed that bubbles latched onto metal droplets, forming coalescing clusters that sank

or rose with net buoyancy. The lower gravity on asteroid-scale bodies would have allowed small bubble-metal clusters to sink into the interior, thereby sequestering volatiles for later collision and accretion.

—Angela Hessler

Science p. 97, 10.1126/science.aef2542

---

BACTERIAL IMMUNITY Viruses make the cut

When invaded by a virus, bacterial cells synthesize cyclic nucleotides as signaling molecules to initiate an immune response program that often ends in cell death as a last resort. Hobbs and Kranzusch found that proteases expressed by the phage itself activate the nucleotide-generating enzymes in the host bacterium. This phage-driven cleavage mechanism clarifies how activation of this signaling pathway is timed to late in the viral infection cycle. Cleavable enzymes found in a range of bacteria suggest that prokaryotic cyclic oligonucleotide-based antiphage signaling is regulated though proteolysis similarly to eukaryotic R proteins and inflammasomes. —Cheri Sirois

Science p.102, 10.1126/science.aeg3949

---

MATERIALS SCIENCE Creating selective gradients

Transition metal perovskite oxides are class of materials with a diverse range of functionalities. Their electronic properties are largely determined by the transition metal atom (usually denoted “B” in the general “ABO $_{3}$ ” chemical formula). By contrast, the “A” atoms, which are nontransition metals, determine the lattice size. Kim et al. fabricated films consisting of layers of perovskites in which the composition of the A-site atoms gradually changed from the top to the bottom layer. By doing so, the researchers introduced a strain gradient that led to the displacement of B-site atoms and strong polarity. The technique created an unusual combination of functionalities in the resulting iridate and titanate films and is expected to be generalizable to other similar systems. —Jelena Stajic

Science p. 107, 10.1126/science.adv0235

---

APPLIED PHYSICS Sharper lifetime maps

Fluorescence lifetime imaging distinguishes fluorophores with overlapping colors by measuring how quickly their emission fades after excitation, but optical diffraction still blurs nearby features. Hwang et al. sharpened these maps using deblurring by pixel reassignment (DPR). DPR is a postprocessing computational technique that moves—i.e., reassigns—measured light toward local intensity peaks. Rearranging the recorded signal in this way makes the spatial structure of these peaks easier to see. With enough photons, DPR narrows blurred spots. At low photon counts, where noisy shifts can produce unstable peaks, DPR clarifies features supported by the spatial and lifetime measurements.

—Randy A. Bartels

Sci. Adv. (2026) 10.1126/sciadv.aec7970

---

BIOCHEMISTRY

---

A nuanced phosphomimic

The atypical protein kinase C (aPKC) family has opposing functions in different contexts. Both phosphorylation by the kinase PDK1 and binding by the sphingolipid S1P activate aPKC. Kajimoto et al. found that the phosphate in S1P bound to the same site as the PDK1-conjugated phosphate but induced an opposite downstream effect. Whereas S1P binding promoted the basal activity of aPKC and opposed apoptosis, PDK1-dependent phosphorylation hindered S1P binding and promoted apoptosis. These findings show that this bioactive lipid activates aPKC-like phosphorylation but switches its effect on apoptosis.

—Leslie Ferrarelli

Sci. Signal. (2026)

10.1126/scisignal.adw7297

---

IN OTHER JOURNALS

Edited by Corinne Simonti and Jesse Smith

---

CLIMATE CHANGE A bacterial ally for drought tolerance

Climate change is predicted to increase drought frequency and severity in the future, making the development of drought-resistant crop varieties increasingly important. Although both root hairs and root colonization by growth-promoting bacteria are known to enhance plant drought tolerance, the molecular mechanisms coordinating them during water deficit have been underexplored. Rahimi et al. found that the bacterial root endophyte Flavobacterium sp. 98 (Flavo98) improves drought tolerance across multiple plant species by modifying the root system architecture. This modulation is dependent on enhanced ethylene biosynthesis and induction of ERF114 and ERF115. ERF114, and ERF115 in turn target the small signaling peptide gene C-TERMINALLY ENCODED PEPTIDE5 (CEP5), which promotes root hair development.—Unnati Sonawala

Nat. Plants (2026)

10.1038/s41477-026-02350-4

---

PLANT BIOLOGY Delivering pollen in a scoop

Pollination is important for flowering plants to reproduce, and many use animals to transfer pollen between flowers. In poricidal flowers, pollen is released through small pores when the flower is mechanically shaken, often by buzzing bees. Lazarus et al. investigated whether the shape of these pores controls how pollen is released. By examining 523 plant species, they found that scoop-shaped pores have evolved repeatedly and produce narrower,

Binary stars (such as Alpha Centauri A and Alpha Centauri B, shown above) can differ in composition if one has consumed a planet.

faster pollen streams than

nonscoop pores. Experiments

showed that scoop-shaped

pores release pollen in

directional bursts, whereas

low-frequency vibrations can

make pollen release more

dispersed. These findings

suggest that pore shape is an

overlooked floral feature that

may help plants control where

pollen is placed on pollina-

tors, and this may influence

pollination success. —Di Jiang

Nat. Commun. (2026)

10.1038/s41467-026-76766-z

CELL BIOLOGY

---

Breaking and entering

The delivery of therapeutic antisense oligonucleotides (ASOs) critically depends on targeting to the correct cells and trafficking out of endosomes into the cytosol. Marco et al. identified a pathway in which new-generation ASOs bind a receptor (CD44)

---

STELLAR ASTROPHYSICS A star that consumed a planet

Binary stars form by collapse of a single parent gas cloud, so the two member stars typically have the same composition. If these stars also are of the same mass, then they experience identical internal evolution, so any differences indicate external perturbations. Rathsam et al. investigated the same-mass binary star system HD 129171/HD 129209, finding that HD 129171 has higher abundances of the refractory elements expected in rocky planets, particularly lithium, but not in volatile elements such as sulfur or oxygen. The abundance pattern is consistent with engulfment of at least 11 Earth masses of planetary material. —Keith T. Smith Astron. Astrophys. (2026) 10.1051/0004-6361/202659556

on human pancreatic ductal carcinoma cells and exploit a receptor tyrosine kinase (EPHA2) to drive ASO uptake and trafficking to nuclear-proximal endosomes. These nuclear-captured compartments become leaky, allowing ASO escape into the cytoplasm. Inhibition of stress granule-mediated membrane repair enhances ASO efficacy. Thus, a receptor that supports tumor stemness and growth can be exploited to access an endocytic pathway capable of delivering therapeutic molecules to aggressive tumors.

—Stella M. Hurtley

J. Cell Biol. (2026) 10.1083/jcb.202507217 symptoms of Lyme disease, including the longer-term complications that may persist after antibiotic treatment. In mouse models, Sachan et al. found that after skin infection, B. burgdorferi disseminated to the gut, causing inflammation within the small intestine. This led to an increase in the permeability of the gut, allowing microbial components into the circulation to promote systemic inflammation. These changes were linked to impaired antibody responses toward the gut microbiota and a reduced quantity of antibodies in the blood plasma.

—Sarah H. Ross

12 heptagonal rings and has a triangular cyclic structure. Zhang et al. synthesized this molecule using a nickel-mediated Yamamoto coupling from a halogen-substituted nanographene monomer that already had a negative curvature. Each macrocycle consists of three saddle-shaped units linked together. The binding of $C_{60}$ and $C_{70}$ , fullerenes with the electron-rich central cavity, was stronger than that observed for the parent nanographene monomers.

—Phil Szuromi

J. Am. Chem. Soc. (2026)

10.1021/jacs.6c11599

---

IMMUNOLOGY A leaky gut after skin infection

Dysregulation of the immune system after infection by Borrelia burgdorferi is thought to underpin many of the

J. Exp. Med. (2026)

10.1084/jem.20252380

---

MACROCYCLES Curvature in the triple negative A triply negatively curved graphenic macrocycle contains

---

STEM WORKFORCE Disparities in grant cancellations

The Trump administration has reshaped US biomedical research through grant terminations, budget cuts, and ideological grant screening,

ultimately destabilizing traditional scientific institutions. Nguyen et al. conducted a retrospective trend analysis of publicly available NIH data (2016 to 2025) comprising annual counts of individuals funded through research grants and fellowship grants. In 2025, the number of NIH-funded research and fellowship grants declined for the first time in a decade, by 4.0 and 9.9%, respectively, with historically underrepresented individuals experiencing disproportionately larger declines. This disparity was pronounced among fellowship grants, with Black and Hispanic representation declining by 40.5 and 33.1%, respectively. As this destabilization continues, it is critical to prioritize structural reform centered in equity to protect current and emerging talent from future disruptions.

—Melissa McCartney

JAMA(2026)10.1001/jama.2026.7796

---

CELL BIOLOGY Unlocking unrestricted growth

Healthy epithelia contain cells with cancer-associated mutations, but most of these remain growth restricted. By contrast, cells lacking the tumor suppressor p53 can expand and occupy large tissue areas. Gan et. al. found that p53 loss promoted the expansion of growth-sustaining epidermal progenitor cells, and they identified a regulatory network that removed several brakes on Wnt signaling (see the Perspective by Engelman and Blanpain). This produced a radial pattern of Wnt activity with stronger signaling at the center of patches of mutant cells than at the edges. Clones exposed to this gradient expanded further than those with uniformly high Wnt activity. Thus, the spatial organization of a growth signal can influence whether mutant cells remain contained or progressively colonize normal tissue.—Stella M. Hurtley

Science p. 78, 10.1126/science.aed3065; see also p. 32, 10.1126/science.ael7341

ARTHRITIS

---

A joint risk for osteoarthritis

Both osteoarthritis and hypertension are common in older adults. Traditionally, osteoarthritis was thought to be due to age-associated wear and tear on the joints, but accumulating evidence points to metabolic and inflammatory influences on this condition. Using data from a nationwide longitudinal clinical cohort, Kim et al. identified a correlation between hypertension and osteoarthritis in older participants (see the Perspective by Wen and Au). The authors then turned to models to work out the underlying mechanism, using two mouse models of hypertension with or without osteoarthritis. The authors demonstrated that arginine

Edited by Michael Funk

vasopressin, a peptide hormone up-regulated in the setting of hypertension, promotes osteoarthritis progression, but only in joints that are already primed by injury or aging and that express the appropriate receptor, providing a mechanistic link between the two conditions. —Yevgeniya Nusinovich

Science p. 80, 10.1126/science.aed8207; see also p. 30, 10.1126/science.ael3784

---

DYNAMIC POLYMERS Negative domains stimulate neurite growth

A self-capping supramolecular polymerization yielded peptide-containing assemblies with precisely tuned segments that could exceed 30 micrometers in length (about 1 billion Daltons in mass). Dore et al. created assemblies with a central cationic segment that anchored longer negative domains (see the Perspective by Kumar). The central core bound to the negatively charged surfaces of neurons, and the long negative domains, held close near the surface in a highly dynamic state, stimulated neurite growth in the absence of any receptor signals in the peptide sequence. —Phil Szuromi

Science p. 125, 10.1126/science.aeg5583; see also p. 27, 10.1126/science.ael4456

---

PIEZOELECTRICS Fast, high-magnitude piezoelectricity

The introduction of polaron-defect complexes in potassium-doped silver niobate films enables high piezoelectric performance at kilohertz frequencies. Shu et al. overcame the trade-off between switching speed and piezoelectric magnitude in conventional piezoelectrics by creating giant defect polarization. This effect, driven by multiunit-cell electron polaron hopping that enables collective polaron-defect reconfiguration, exceeds the limited electric-dipole extension of single-unit

cells. The materials exhibited an effective piezoelectric coefficient of about 7170 picometers per volt and a strain of about 6.2%. —Phil Szuromi

Science p. 118, 10.1126/science.aeg4972

---

QUANTUM SYSTEMS Entangled levitation

The ability to cool levitated nanoparticles to ultralow temperatures for accessing mechanical quantum states of the system provides a platform for probing fundamental physics. Deplano et al. demonstrated the generation of entanglement between a macroscopic mechanical oscillator and a freely propagating optical field. Such optomechanical entanglement at room temperature involving a levitated nanoparticle and a propagating optical mode and could also be used to develop quantum sensing technologies. —Ian S. Osborne

Science p. 113, 10.1126/science.aeh1375

---

NEUTROPHILS

---

Nurturing neutrophils

During early human pregnancy, placenta-derived extravillous trophoblasts invade the decidua and remodel the maternal-fetal interface, but the contribution of maternal neutrophils to this process is poorly understood. Höbler et al. analyzed tissue and blood samples from pregnant individuals, identifying a neutrophil subset localized to the Rohr's layer of the maternal-fetal interface that was phenotypically and functionally distinct from peripheral blood neutrophils. This decidual fibrinoid-associated neutrophil subset supported placental invasion during gestation by secreting matrix metalloproteinase-9, which simultaneously activated latent transforming growth factor- $\beta$ and reprogrammed peripheral blood neutrophils toward a state supporting placental invasion.

—Leoma Bere

Sci. Immunol. (2026)

10.1126/sciimmunol.aea4118

---

CHEMISTRY Silver catalysis for protein synthesis

Amide bonds provide a remarkably stable backbone to proteins, but this inertness also makes selective manipulation difficult. Sun et al. developed a programmable and mild amide bond activation strategy starting with N,S-benzylidene thioacetal modified peptides. Silver(I) selectively converts backbone amide into a salicylaldehyde ester within fully unprotected peptides at room temperature without detectable epimerization. The transformation is compatible with diverse peptide sequences and established peptide chemistries. Combined with established peptide ligation chemistry, this strategy facilitated the total synthesis of a mirror-image histone octamer and provides a versatile route to other challenging synthetic proteins.

—Steven D. Townsend

Sci. Adv. (2026) 10.1126/sciadv.aej5227

---

METABOLISM Transporter with a central role

Regulation of lipid metabolism in mitochondria of adipocytes requires integration of information about time of day from the circadian clock, environmental temperature, and nutrient availability. Karavaeva et al. found an important role for the transporter protein SLC25A34 in such control of lipid metabolism and thermogenesis in brown adipocytes from mice and humans (see the Perspective by van der Laan). Inputs from the circadian clock, cold temperatures, and diet all regulated the expression of SLC25A34, allowing control of lipogenesis and mitochondrial biogenesis, lipid oxidation, and thermogenesis. Such regulation may offer insights into inflammation associated with obesity and metabolic diseases.

—L. Bryan Ray

Science p. 79, 10.1126/science.adz4797; see also p. 29, 10.1126/science.ael7271

LET YOUR KNOWLEDGE BLOSSOM

Visit the Science Custom Publishing sites today and grow your knowledge with a wide selection of booklets, podcasts, posters, sponsored features, and webinars!

Booklets

Podcasts

Posters

Sponsored Features

Webinars

Brought to you by the Science/AAAS Custom Publishing Office.

Science

AAAS

Scan the code and start exploring the latest advances in science and technology innovation!

Science.org/custom-publishing

Loss of the tumor suppressor p53 generates a signaling gradient that drives epithelial clonal expansion

Qiwen Gan, Wei Li, Rachel K. Lex, Zhe Ying\*, Slobodan Beronja\*

Full article and list of author affiliations: https://doi.org/10.1126/science.aed3065

INTRODUCTION: Normal tissues accumulate mutations throughout life, but most mutant cells remain contained. Loss of the tumor suppressor p53 is common in epithelial cancers and can allow mutant cell populations to occupy large areas of normal tissue. How p53 loss changes cell behavior to permit this expansion has remained unclear.

RATIONALE: We used mouse skin as a model to determine how p53 restrains the growth of epithelial clones. By combining lineage tracing with quantitative analysis of daughter cell fate, we found that p53 loss drives clonal expansion by enlarging the basal progenitor population rather than by disabling canonical cytostatic pathways. This observation led us to identify direct p53 targets that govern progenitor renewal and differentiation through a combination of gene expression and chromatin analyses, large-scale in vivo genetic screens, and single-cell imaging of Wnt activity.

RESULTS: Loss of p53 shifted epidermal progenitor cells toward self-renewal and away from differentiation, allowing the progenitor pool to expand over successive cell divisions. A p53 variant that cannot activate cell cycle arrest, senescence, and cell death programs still restrained progenitor renewal, indicating that noncanonical p53 targets suppress clonal expansion.

By integrating transcriptional profiling, p53 binding maps, and functional screening in mouse epidermis, we identified three direct p53 targets—secreted frizzled-related protein 1 (Sfrp1), low-density lipoprotein receptor-related protein 1 (Lrp1), and ubiquitin-specific peptidase 22 (Usp22)—that act as the most genetically upstream regulators of the p53-governed differentiation and renewal network. These three regulators suppress Wnt signaling at both the ligand and intracellular levels, and upon p53 inactivation, their expression was reduced and Wnt activity increased. Owing to the multilevel nature of Wnt suppression, increase in Wnt activity across the p53-deficient clones was not uniform; it was lowest at the edge of the clones and rose progressively toward their centers, creating a radial gradient that persisted across different stages of clonal expansion regardless of clonal density.

Genetic experiments that manipulated the pattern of Wnt activity showed that gradient Wnt activation was associated with robust clonal expansion. Additional manipulations that produced uniformly elevated Wnt activity, including conditions where average Wnt activity was higher than in p53-deficient clones, resulted in clones that expanded more slowly.

CONCLUSION: p53 restrains epithelial colonization by maintaining a network that controls both the strength and spatial organization of Wnt signaling. Its loss creates a radial Wnt activity gradient that is associated with sustained progenitor self-renewal and rapid clonal expansion. These findings show that the spatial pattern of a growth signal, not only its overall level, can determine how mutant cells spread through normal tissue. □

Spatial organization of Wnt signaling after p53 loss. In normal epidermis, p53 maintains expression of several Wnt suppressors and balances progenitor renewal with differentiation. Loss of p53 weakens this network, producing a radial increase in Wnt activity from the clone edge toward its center. The resulting gradient supports sustained progenitor renewal and rapid expansion through the surrounding tissue.

Loss of the tumor suppressor p53 generates a signaling gradient that drives epithelial clonal expansion

Qiwen Gan $^{1}$ , Wei Li $^{2}$ , Rachel K. Lex $^{2}$ , Zhe Ying $^{1*}$ , Slobodan Beronja $^{2*}$

Although tumor protein p53 (TP53) mutations are among the most common lesions in epithelial cancers, how p53 loss drives unrestrained clonal expansion remains unclear. Working with mouse epidermis, we found that p53 suppresses clonal expansion by limiting progenitor self-renewal. Integrating chromatin immunoprecipitation sequencing, transcriptional analyses, and genetic screens, we identified a p53-dependent network controlling progenitor renewal and differentiation, in which secreted frizzled-related protein 1 (Sfrp1), low-density lipoprotein receptor-related protein 1 (Lrp1), and ubiquitin-specific peptidase 22 (Usp22) were direct targets of p53 and the most genetically upstream components. Their suppression after p53 loss generated a radial gradient of Wnt activity across mutant clones. Genetic manipulation showed that a Wnt activity gradient, rather than uniform elevation, was associated with sustained clonal expansion. These findings identify spatial organization as a determinant of clonal behavior and a mechanism by which TP53 inactivation drives tissue colonization.

Human tissues acquire a high burden of genetic lesions over a lifetime yet maintain normal morphology and function (1-4). Most cancer-associated mutations remain growth stabilized in the tissue, whereas a limited subset is associated with clonal expansion (5-8). Mouse studies have identified several mechanisms that constrain cancer-associated lesions, including clonal elimination (9-13), growth restriction (14), and intra- (15) or interclonal competition (16-19). Yet some lesions escape these constraints and progressively colonize large areas of normal epithelium (20). Examples include loss of notch receptor 1 (Notch1) in the esophagus (21), loss of FAT atypical cadherin 1 (Fat1) in the skin (22), and inactivation of tumor protein p53 (5, 13). Among these lesions, p53 loss is particularly consequential because it is a common founder event across epithelial cancers (23, 24). How p53 inactivation converts a normally constrained mutant clone into one capable of sustained tissue colonization remains unresolved.

p53 activity responds to changes in cellular conditions (23–26). These range from normal physiological processes, including replication stress and mitochondrial dynamics, to pathological conditions such as acute DNA damage and oncogene activation (23–26). Its best-characterized outputs are cytostatic pathways that induce senescence, apoptosis, and cell cycle arrest (23–26). The growth advantage after p53 loss has therefore been attributed largely to suppression of these pathways (23–26). However, p53 controls a broader and context-dependent signaling network, and the role of canonical cytostatic pathways in restricting tissue expansion under nonstressed conditions has been questioned (27–32). Reintroduction of p53 after DNA damage restores tissue homeostasis in p53-null animals without activating canonical cytostatic pathways (30, 31). Mice carrying Trp53 alleles defective in these responses also largely maintain long-term tissue homeostasis (28, 29). Similarly, combined deletion of the core cytostatic effectors p21, Puma, and Noxa does not increase tissue expansion (27). These findings suggest that p53 can regulate tissue growth independently of canonical cytostatic pathways.

---

p53 restricted epidermal clonal expansion by balancing progenitor renewal and differentiation

Mutations in p53 have been identified as founder lesions of clonal expansion, and inactivation of p53 can act as a robust driver of this process across a spectrum of epithelial tissues, including the epidermis (5, 8, 33, 34). To compare canonical cytostatic functions with other p53 functions that might restrict clonal expansion, we used the $Trp53^{3KR/3KR}$ mouse line (29, 35). Mutation of three critical acetylation sites prevented p53 $^{3KR}$ from transactivating apoptosis, senescence, and cell cycle arrest genes while preserving its ability to induce ferroptosis (29, 35) (Fig. 1A). To assess clonal expansion in wild-type (WT), $Trp53^{-/-}$ , and $Trp53^{3KR/3KR}$ epidermis, we crossed the three Trp53 alleles into a K14-CreER;mTmG Cre reporter background. At P19 (postnatal day 19), we clonally labeled basal progenitors with administration of a low dose of tamoxifen and imaged the clones in the interfollicular epidermis at P21 and P49 with whole-mount confocal microscopy (Fig. 1B) (36). At P21, approximately one cell cycle after tamoxifen treatment (13), we observed sparse labeling throughout the epidermis and no significant size difference between WT, $Trp53^{-/-}$ , and $Trp53^{3KR/3KR}$ clones (Fig. 1, B and C). In P49 animals, however, $Trp53^{-/-}$ clones exhibited a significant growth advantage, whereas $Trp53^{3KR/3KR}$ clones maintained a size similar to wild type (Fig. 1, B and C). When our analysis was focused on the basal progenitor layer, we quantified significant basal cell expansion in $Trp53^{-/-}$ clones relative to wild type and $Trp53^{3KR/3KR}$ in both P21 and P49 animals (Fig. 1, B and D). These observations suggested that canonical p53 functions may not be required to restrict clonal expansion. To directly investigate the underlying cellular mechanism of clonal expansion in $Trp53^{-/-}$ epithelium, we assessed basal cell density, apoptosis via cleaved caspase-3 staining, cell cycle dynamics using 5-ethynyl-2'-deoxyuridine/5-bromo-2'-deoxyuridine (EdU/BrdU) pulse-chase assay (13, 15, 37), senescence with quantitative polymerase chain reaction (qPCR) of cyclin-dependent kinase inhibitor 2A ( $Cdkn2a$ ) and senescence-associated beta-galactosidase (SA-β-gal) staining, and p53 mediated ferroptosis through qPCR of solute carrier family 7 member 11 ( $Slc7a11$ ) and 4-hydroxynonenal (4-HNE)/TUNEL co-staining (23, 35) (fig. S1, A to J). Apart from significant reduction in cleaved caspase-3-positive cells and $Cdkn2a$ expression in both $Trp53^{-/-}$ and $Trp53^{3KR/3KR}$ tissues compared with wild type, no significant changes in basal cell density, proliferation, SA-β-gal staining, or ferroptosis were detected (fig. S1, A to J). Because none of the measured phenotypes correlated with the growth advantage of $Trp53^{-/-}$ over $Trp53^{3KR/3KR}$ and WT clones (Fig. 1, B to D, and fig. S1, A to J), we hypothesized that there was an alternative mechanism of how p53 restricts clonal expansion in the epidermis.

Oncogene-induced differentiation of progenitors into postmitotic progeny can potently restrict clonal expansion (13, 15, 37). Our data suggested a basal layer-specific expansion in $Trp53^{-/-}$ clones (Fig. 1, B to D). To test whether p53 loss may drive clonal expansion by controlling renewal and differentiation in the epidermis, we quantified the rate of progenitor renewal (the proportion of basal progenitors among all progeny produced in one round of cell cycle) using our EdU/BrdU pulse-chase assay (13, 15, 37) (fig. S1, K to M). $Trp53^{3KR/3KR}$ and WT progenitors had a renewal rate of \~0.53, suggesting that just over half the newly generated daughter cells maintained the progenitor potential of the mother, consistent with previously reported rates required for long-term maintenance of the progenitor cell pool and tissue homeostasis (13, 15, 37–39) (Fig. 1E). In contrast, loss of p53 significantly increased the progenitor renewal above the homeostatic rate (Fig. 1E) and reduced the proportion of K10-expressing basal cells within clones (fig. S1, N and O). To determine how p53 loss elevated the renewal rate, we scored the fates of daughter-cell doublets at P21 derived from singly labeled progenitors at P19 (13, 15) (Fig. 1F). In $Trp53^{-/-}$ progenitors, we observed a significant increase in the frequency of symmetric renewal and a concomitant decrease in asymmetric divisions (Fig. 1G). Observed frequencies were consistent with an increase in the renewal rate to \~0.63 (Fig. 1E) and generation of additional proliferative progenitors with each cell division, which can result in geometric expansion of clones over several cell cycles (13, 39). Loss of canonical p53 functions in the $Trp53^{3KR/3KR}$ epidermis alone had no effect on progenitor renewal and differentiation (Fig. 1, E to G).

B

E

F

C

G

Fig. 1. p53 loss promoted epidermal clonal expansion by increasing progenitor renewal. (A) Allele-specific p53 functions: $Trp53^{-/-}$ abolished all p53 functions; $Trp53^{3KR/3KR}$ retained noncanonical activities but lacked induction of senescence, apoptosis, and cell cycle arrest. (B to D) Representative maximum-intensity projections and basal optical sections (B), total clone size (C), and basal-cell number (D) in WT, $Trp53^{-/-}$ , and $Trp53^{3KR/3KR}$ K14-CreER; mTmG epidermis. Basal progenitors were labeled at single-cell density with low dose of tamoxifen at P19 and analyzed at P21 or P49. Scale bar, 200 $\mu$ m. Clone numbers at P21/P49: WT, 63/33; $Trp53^{-/-}$ , 49/21; $Trp53^{3KR/3KR}$ , 81/24. (E) Progenitor renewal measured by EdU/BrdU pulse chase (13,15). $n = 3$ animals per genotype. (F and G) Representative daughter-cell doublets (F) and frequencies of symmetric renewal (SR), asymmetric division (AS), and symmetric differentiation (SD) 48 hours after labeling (G). Scale bar, 50 $\mu$ m; $n = 3$ animals per genotype. (H) Constructs and induction strategy for p53 deletion followed by $p53^{3KR}$ expression. LTR, long terminal repeat. (I) Immunoblot and quantification of p53 at P34 and P40. (J) Progenitor renewal before and after $p53^{3KR}$ reexpression. $n = 3$ animals per condition [(I) and (J)]. Data are mean $\pm$ SD. Comparisons used two-tailed Student's t tests; exact $P$ values are shown.

We set out to determine whether epidermal progenitors were being reprogramed after p53 loss into a heightened stemlike state, or whether ongoing p53 transcriptional activity may be required to guide renewal and differentiation on a continual basis. We transduced $Trp53^{fl/fl}$ mice with two lentiviruses: one expressing rtTA (reverse tetracycline-controlled transactivator) and the other carrying a TRE (tetracycline response element)-driven p53 $^{3KR}$ -T2A-Cre cassette (Fig. 1H). Transduced animals were given a pulse injection of doxycycline at P21 to delete p53 and then switched to a continuous supply of doxycycline to express p53 $^{3KR}$ at P35 (Fig. 1, H and I). Quantification of the rate of renewal at P34 and P40 showed that reexpression of p53 $^{3KR}$ was sufficient to reduce the rate of renewal back to WT levels (Fig. 1J). Thus, clonal expansion in skin epithelium is restricted through p53 activity that regulated progenitor renewal and differentiation to suppress progenitor renewal.

---

A p53 transcriptional program in progenitor renewal and differentiation

Growth-promoting renewal and differentiation dynamics that follow p53 loss, as well as the inability of canonical cytostatic p53 functions to restrict clonal expansion (Fig. 1 and fig. S1), suggested the existence of a p53 transcriptional program that controlled key renewal and differentiation regulators.

We hypothesized that these renewal and differentiation genes are among the shared targets of p53 $^{WT}$ and cytostatic function-deficient p53 $^{3KR}$ (Fig. 2A). To test this hypothesis and identify shared renewal and differentiation targets of p53, we first measured the expression of canonical p53 target genes (40) cyclin-dependent kinase inhibitor 1A (Cdkn1a; p21), BCL2-binding component 3 (Bbc3; Puma), and BCL2-associated X apoptosis regulator (Bax) and observed significant discovery rate (FDR) < 0.01] shared by both p53 $^{WT}$ and p53 $^{3KR}$ (Fig. 2C). To quantify the accessibility of these genes, we performed assay for transposase-accessible chromatin with high-throughput sequencing (ATAC-seq) in basal progenitors and confirmed that p53 peaks were located in regulatable open chromatin regions under nontreated conditions (Fig. 2C). Consistent with the specificity of our assays, we identified ChIP-seq traces on known p53 $^{WT}$ /p53 $^{3KR}$ shared targets, such as Mdm2 (29) which showed similar levels of p53 enrichment at the locus (Fig. 2D). In contrast, ChIP traces on known canonical p53 $^{WT}$ -specific targets Cdkn1a, Bbc3, and Bax (29) showed reduced p53 $^{3KR}$ binding at the respective loci (Fig. 2D).

Fig. 2. Identification of transcriptional targets of p53 that could act as epithelial renewal and differentiation regulators. (A) Strategy to identify renewal and differentiation regulators among transcriptional targets shared by $\mathrm{p53}^{\mathrm{WT}}$ and $\mathrm{p53}^{3\mathrm{KR}}$ . (B) Genes differentially expressed in $Trp53^{-/-}$ basal progenitors relative to WT and $Trp53^{3\mathrm{KR}/3\mathrm{KR}}$ cells. $n = 3$ biological replicates per genotype; genes were identified by pattern matching (MeV, $P < 0.01$ ). (C) Heatmaps of p53 ChIP-seq and ATAC-seq signals centered on p53-binding sites in WT and $Trp53^{3\mathrm{KR}/3\mathrm{KR}}$ basal progenitors. High-confidence ChIP-seq peaks were called with MACS2 (FDR < 0.01). ChIP-seq and ATAC-seq signals combined three replicates containing $1 \times 10^{7}$ and $5 \times 10^{5}$ basal cells, respectively. (D) Representative ChIP-seq and ATAC-seq tracks at the shared target Mdm2 and the $p53^{\mathrm{WT}}$ -selective targets Cdkn1a, Bbc3, and Bax. (E) Genomic distribution of p53 peaks and enrichment of the top p53-binding motif. (F) Intersection of differentially expressed genes with shared $p53^{\mathrm{WT}}/p53^{3\mathrm{KR}}$ ChIP-seq peaks identified direct p53 targets for functional testing.

To our surprise, the heatmap surrounding the ChIP peaks indicated that most of the p53 targets were shared between $p53^{WT}$ and $p53^{3KR}$ (\~20-fold more than the $p53^{WT}$ -specific targets; Fig. 2C). We further characterized the pattern of $p53^{WT}/p53^{3KR}$ shared and $p53^{WT}$ -specific peaks using ChIPseeker (42) and scored the p53 motif enrichment using Homer (43) (Fig. 2E). Our results showed no apparent difference in peak distribution patterns and p53 motif enrichment between $p53^{3KR}/p53^{WT}$ shared and $p53^{WT}$ -specific loci (Fig. 2E). Moreover, the $p53^{WT}/p53^{3KR}$ DNA binding specificity was similar to that identified by published p53/63/73 ChIP-seq analyses across different cell types in Homer datasets (43) (fig. S2, D and E).

To identify p53 $^{WT}$ /p53 $^{3KR}$ targets that were also transcriptionally regulated by p53 in the basal progenitors, we intersected our RNA-seq and ChIP-Seq analyses to identify 510 p53 sustained and 603 p53 down-regulated genes (Fig. 2F). We hypothesized that among them were key targets of a p53 transcriptional reduction in isolated $Trp53^{3KR/3KR}$ and $Trp53^{-/-}$ basal progenitors compared with WT (fig. S2A). We then performed RNA sequencing (RNA-seq) in these cells (Fig. 2B) and identified 793 sustained and 1497 down-regulated genes that were differentially expressed in $Trp53^{-/-}$ relative to WT and $Trp53^{3KR/3KR}$ progenitors [pattern match algorithm, MeV (4I), P < 0.01] (Fig. 2B).

To identify genes that were directly occupied by both p53 and $p53^{3KR}$ , we performed p53 chromatin immunoprecipitation sequencing (ChIP-seq) in WT and $Trp53^{3KR/3KR}$ basal progenitors. To minimize bias toward the cytostatic genes while sufficiently stabilizing p53, we digested the tissues for progenitor cell isolation in the presence of 25 $\mu$ M of nutlin-3a, a dose that did not induce significant cytostatic responses, as indicated by the absence of activated caspase-3 staining (fig. S2, B and C). Our p53 ChIP-seq identified peaks [MSCS2, false program governing renewal and differentiation dynamics.

---

An in vivo genetic screen for components of the p53-regulated renewal and differentiation program

To identify genes that may execute the p53 renewal and differentiation program among the 1113 p53 $^{WT}$ /p53 $^{3KR}$ shared targets, we constructed a lentiviral pool of 7021 short hairpin RNAs (shRNAs) to perturb each gene with five to seven distinct shRNAs (Fig. 3A). We used an established in vivo screening strategy (13, 15, 37) to test the candidates for their ability to regulate progenitor renewal and differentiation in WT, Trp53 $^{3KR/3KR}$ , and Trp53 $^{-/-}$ epidermis (Fig. 3A). Briefly, we introduced the pool into E9.5 epidermis using in utero transduction [n = 30 injected embryos, >50× coverage with a multiplicity of infection (MOI) of \~0.11] and harvested epidermis from P1 animals. On the basis of prior findings (13, 15, 37), we anticipated pro-renewal shRNAs to be enriched in basal $\alpha 6$ integrin-high progenitor cells and pro-differentiation shRNAs to be enriched in suprabasal $\alpha 6$ integrin-low cells, which we isolated with fluorescence-activated cell sorting (FACS) (Fig. 3A). Using a published statistical pipeline (13, 15, 44, 45) [Deseq2 (46), $P < 0.01$ ; for each gene at least two shRNAs with the same significant trend and no shRNAs with the opposite trend] on three genetic backgrounds with distinct transcriptional states (WT, Trp53 $^{3\mathrm{KR} / 3\mathrm{KR}}$ , and Trp53 $^{-/-}$ ), we expected that we may achieve a screen with heightened stringency. Specifically, our scoring strategy (Fig. 3B) was based on the expectation that shRNAs targeting p53 sustained candidates may only promote renewal in WT and Trp53 $^{3\mathrm{KR} / 3\mathrm{KR}}$ epidermis, where the candidates are transactivated, but not in Trp53 $^{-/-}$ epidermis, where they are not. Conversely, for p53 repressed genes, the targeting shRNAs could only be differentiation promoters in Trp53 $^{-/-}$ epidermis, where p53-mediated candidate suppression was lost (Fig. 3B). Following the criteria above, we performed the three sets of genetic screens in triplicate (Fig. 3, C to E) and identified six p53 sustained genes and four p53 down-regulated genes among the 1113 tested as regulators of renewal and differentiation dynamics (Fig. 3F).

B

D

Fig. 3. Functional identification of a p53 governed epithelial renewal and differentiation program. (A) Screening strategy for testing shared p53 $^{WT}$ /p53 $^{3KR}$ transcriptional targets as regulators of progenitor renewal and differentiation in WT, Trp53 $^{3KR/3KR}$ , and Trp53 $^{-/-}$ epidermis. Basal and suprabasal cells were isolated by FACS and analyzed for shRNA enrichment. (B) Genotype-specific criteria used to identify p53-dependent renewal and differentiation regulators. (C to E) Enrichment of pro-renewal

Rather than prioritizing screen hits for downstream analyses on the basis of fold change, we used the limited number of identified candidates to design a test to probe for synergy across all 10 in promoting progenitor renewal (Fig. 3G). We constructed a mini pool containing shRNAs against the six p53 sustained genes and open reading frames (ORFs) of the four p53 downregulated genes. The mini pool was then mixed with high-titer virus of each of the 10 constructs, and these were transduced into the epidermis of 10 sets of WT mice ( $n = 3$ injected embryos per set, $>3750 \times$ coverage with MOI of $\sim 0.11$ ). The ability of each construct to promote renewal or differentiation was scored by relative enrichment in basal progenitors versus differentiated suprabasal cells, as before (13, 15) (Fig. 3G). After functionally profiling all $10 \times 10$ interactions, we scored significant [Deseq2 (46), $P < 0.01$ ] synergistic effect in promoting renewal and mapped the implied epistatic relationship accordingly (Fig. 3, H and I). Among the 10, we found three p53 transactivated genes—secreted frizzled-related protein 1 ( $Sfrp1$ ), low-density lipoprotein receptor-related protein 1 ( $Lrp1$ ), and ubiquitin-specific peptidase 22 ( $Usp22$ )—that did not have a significant synergistic interaction with any other, indicating that they represented the upstream-most signaling genes that regulate renewal and differentiation dynamics (Fig. 3, H and I).

---

The p53-regulated renewal and differentiation program suppresses Wnt signaling

We first set out to validate Sfrp1, Lrp1, and Usp22 as p53 transactivated targets. Analyses of 25 $\mu$ M nutlin-3a-stabilized ChIP-seq and nontreated ATAC-seq data showed robust p53 binding in open chromatin regions of Sfrp1, Lrp1, and Usp22 gene loci in both p53 $^{WT}$ and p53 $^{3KR}$ epidermal progenitors (Fig. 4A). To test whether Sfrp1, Lrp1, and Usp22 were directly occupied by p53 $^{WT}$ and p53 $^{3KR}$ without p53 stabilization, we performed ChIP-qPCR in untreated basal progenitors isolated

(orange) and pro-differentiation (blue) shRNAs in WT (C), $Trp53^{3KR/3KR}$ (D), and $Trp53^{-/-}$ (E) epidermis. Scrambled shRNAs served as negative controls. (F) Effects of the two highest-scoring shRNAs for each candidate. (G) Design of the combinatorial 10×10 interaction screen. (H) Synergistic effects of candidate pairs on renewal (orange) or differentiation (blue). shSfrp1, shLrp1, and shUsp22 showed no significant synergy with other candidates (yellow outlines). (I) Interaction network inferred from the combinatorial screen. Values in (C) to (F) and (H) are means from three biological replicates. Enrichment and interaction statistics were calculated with DESeq2; P < 0.01.

from WT, $Trp53^{3KR/3KR}$ , and $Trp53^{-/-}$ epidermis (Fig. 4B). Significant p53 and $p53^{3KR}$ enrichment was observed across the ChIP-seq peaks for all three genes and was substantially reduced by p53 knockout, suggesting that p53 binds to the Sfrp1, Lrp1, and Usp22 loci (Fig. 4B). In addition, qPCR analyses showed that all three genes were expressed at similar levels in WT and $Trp53^{3KR/3KR}$ cells but were drastically down-regulated in $Trp53^{-/-}$ basal progenitors, further confirming that Sfrp1, Lrp1, and Usp22 are sustained by p53 (Fig. 4C).

Sfrp1 (47–50) and Lrp1 (51–53) are known as robust suppressors of Wnt signaling, and our epistatic analyses suggested that Usp22 may work through the same pathway (Fig. 3, G to I). We thus set out to test whether p53 loss may affect Wnt signaling in epidermal progenitors. We began by evaluating the known epidermal Wnt response signature (54) in WT and p53 mutant clones (Fig. 4, D and E). Gene set enrichment analysis (GSEA) suggested that β-catenin-induced genes were robustly enriched whereas β-catenin-suppressed genes were significantly depleted in $Trp53^{-/-}$ basal progenitors, in comparison to WT (Fig. 4D) and $Trp53^{3KR/3KR}$ (Fig. 4E) controls.

We next measured Wnt activity using the H2B-GFP Wnt reporter mouse line (55). We combined it with the tdTomato $^{Isl/Isl}$ Cre reporter allele (56), so that tdTomato would mark clones transduced with perturbation lentivirus expressing Cre-recombinase and gene-targeting elements (Fig. 4, F to H). Because nuclear $\beta$ -catenin and Wnt activity are mostly restricted to basal progenitors (57) (fig. S3A), we focused our analyses on the H2B-GFP signal from the basal layer (fig. S3B). Using whole-mount confocal imaging and three-dimensional (3D) quantification with Imaris (58), we were able to assess Wnt activity, based on the level of nuclear H2B-GFP, at single-cell resolution and across the transduced clones and their untransduced neighbors (Fig. 4, G and H). Control clones induced in H2B-GFP Wnt reporter;tdTomato $^{Isl/Isl}$ epidermis by lentivirus expressing Scrambled shRNA and Cre (shScram-Cre), showed no significant Wnt activity difference between basal cells in and outside of the clones (Fig. 4G). Thus, lentiviral transduction did not alter Wnt signaling. In contrast, control shScram-Cre transduction of the Trp53 $^{fl/fl}$ ;H2B-GFP Wnt reporter;tdTomato $^{Isl/Isl}$ mice showed robust expression of the H2B-GFP Wnt reporter among basal cells (fold change = 6.85; $P = 5.2 \times 10^{-3}$ ; Fig. 4G) together with highly efficient Cre-mediated knockout (fig. S3, C and D), confirming that p53 loss resulted in activation of the Wnt pathway in the clones relative to the surrounding WT field. Nuclear H2B-GFP signal was unchanged in Trp53 $^{3KR/3KR}$ ;tdTomato clones and only increased upon clonal shTrp53-mediated depletion of p53 $^{3KR}$ (fig. S3E), indicating that p53 $^{3KR}$ was a sufficient suppressor of Wnt activity. To establish the sensitivity and specificity of the Wnt/ $\beta$ -catenin reporter system, we quantified Cre recombination efficiency (fig. S3, C and D) and H2B-GFP levels in Ctnnb1 $^{-/-}$ (catenin beta 1) clones and observed fluorescence levels below those seen in WT and Trp53 $^{-/-}$ clones (Fig. 4, F and G).

Next, we explored the contribution of p53-regulated renewal and differentiation genes in Wnt reporter activity. We first clonally depleted Sfrp1, Lrp1, and Usp22 in the WT and $Trp53^{3KR/3KR}$ epidermis. All three conditions resulted in significant elevation of Wnt signal in transduced clones when compared with WT fields or clones, indicating that Sfrp1, Lrp1, and Usp22 could suppress Wnt/ $\beta$ -catenin signaling (Fig. 4H and fig. S3, F to H). To understand the individual contribution of Sfrp1, Lrp1, and Usp22 to p53-driven suppression of Wnt activity, we next transduced $Trp53^{fl/fl}$ epidermis with lentiviruses coexpressing Cre and ORFs of Sfrp1, Lrp1-mini [truncated Lrp1 that is short enough to be packaged into lentivirus but can still interact with and suppress Wnt receptors (52, 53)], and Usp22, respectively (Fig. 4H). All three constructs significantly reduced Wnt reporter levels in $Trp53^{-/-}$ clones (Fig. 4H and fig. S3I), suggesting that p53's ability to sustain Sfrp1, Lrp1, and Usp22 expression was critical for its suppression of Wnt signaling in epidermal basal cells.

---

p53 inactivation generated a radial gradient of Wnt activity

A curious feature of the p53-governed Wnt-suppression network is that, by transactivating Sfrp1, Lrp1, and Usp22, p53 could suppress Wnt signaling at the ligand (47–50), receptor (51–53), and nucleus (59) levels, respectively (Fig. 5A). Such multilevel control of Wnt signaling implied a mixed suppression network of unclear significance. We hypothesized that, rather than representing a simple redundancy, a multilevel control may create a qualitatively distinct form of Wnt suppression and that p53 loss may result in a gradient of Wnt activation across the radius of $Trp53^{-/-}$ clones (Fig. 5A). To test this hypothesis, we calculated the distance between individual $Trp53^{-/-}$ cells and the closest WT neighbors using Imaris (58) and plotted the intensity of H2B-GFP Wnt reporter signal relative to the distance from the WT/ $Trp53^{-/-}$ interface (Fig. 5B and fig. S3B). Nonlinear regression analysis showed a gradient of Wnt activity, which gradually increased from the outer edge toward the center of $Trp53^{-/-}$ clones (Fig. 5B and fig. S4A). Accompanying this gradient, the rate of progenitor renewal also showed a gradual increase along the radius of the clones that were undergoing rapid expansion (Fig. 5, C to E). Such a phenotype is consistent with the role of Wnt as a driver of progenitor renewal (57, 60) and our observation of gradual reduction of Sfrp1 and sharp decrease of Lrp1 and Usp22 staining within the clones (fig. S4, B and C). A similar Wnt activity gradient was seen upon shTrp53 expression in $Trp53^{3KR/3KR}$ clones (fig. S4D). In contrast, knockout of β-catenin in $Trp53^{-/-}$ clones flattened the Wnt activity gradient (Fig. 5B), lowered the renewal rate to uniform across the clones (Fig. 5C), and resulted in a significant loss of long-term clonal expansion alongside substantial reduction of Wnt activity (Fig. 5, D to F). These observations confirmed that increased progenitor renewal and clonal expansion after p53 loss were dependent on activation of Wnt/β-catenin signaling.

Imaging of clones at different stages of expansion revealed that this robust Wnt activity gradient was maintained not only in well-separated clones (fig. S4, E and F) but also in densely populated regions where clonal fusion was occurring (fig. S4, G and H). Indeed, Wnt activity within the 0-to-50- $\mu$ m and 50-to-100- $\mu$ m bins from the clone border was comparable between clones seeded at low and normal densities as well as the areas of clonal fusion (fig. S4I), indicating a similar ability to establish a radial Wnt activity gradient. Thus, the Wnt activity gradient was robust across a broad range of clonal densities and was maintained in both isolated and fused mutant clones, suggesting that it represented an intrinsic property of p53-deficient epidermal expansion rather than a consequence of clone spacing or local clone geometry.

---

A Wnt activity gradient was associated with rapid epithelial clonal expansion

Ctnnb1 deletion reduced both the gradient and amplitude of Wnt activity (Fig. 5, B to F), leaving it unclear whether one or both of these features contribute to clonal expansion. Thus, we systemically tested association of radial gradient- versus plateau-shaped Wnt activation with clonal expansion. We hypothesized that, because Sfrp1 is a secreted Wnt antagonist (47–50), clones that maintained Sfrp1 suppression (i.e., $Trp53^{-/-}$ , shSfrp1, $Trp53^{-/-}$ Usp22-ORF, and $Trp53^{-/-}$ Lrp1-ORF) may induce distinct pattern of Wnt activation from clones with depletion of Lrp1 (51–53) or Usp22 (59) or from constitutive Wnt activation using

Fig. 4. p53 governed renewal and differentiation program controls Wnt activity in epidermis. (A) p53 ChIP-seq and ATAC-seq tracks at Sfrp1, Lrp1, and Usp22 in WT and Trp53 $^{3KR/3KR}$ basal progenitors. (B) ChIP-qPCR at p53-binding sites in untreated WT, Trp53 $^{3KR/3KR}$ , and Trp53 $^{-/-}$ basal progenitors. n = 3 animals per genotype. (C) Sfrp1, Lrp1, and Usp22 transcript abundance in the three genotypes. n = 3 biological replicates per genotype. (D and E) GSEA of published β-catenin-responsive transcriptional signatures (54) in Trp53 $^{-/-}$ relative to WT (D) or Trp53 $^{3KR/3KR}$ (E) basal progenitors. (F) Mouse strains, perturbations, and imaging strategy used to measure clonal Wnt activity.

(G and H) Representative maximum-intensity projections and mean basal H2B-GFP reporter intensity inside and outside tdTomato-positive clones. p53 loss increased Wnt activity, whereas Ctnnb1 deletion suppressed it (G). Depletion of Sfrp1, Lrp1, or Usp22 increased reporter activity; expression of the corresponding ORFs reduced reporter activity in $Trp53^{-/-}$ clones (H). Scale bars, 100 $\mu$ m. Data are mean ± SD. GSEA statistics are shown in (D) and (E). Other comparisons used two-tailed Student's t tests; exact P values and sample sizes are shown.

Fig. 5. p53 inactivation generated a radial gradient of Wnt activity. (A) Model for spatial Wnt regulation by the secreted antagonist Sfrp1 and the cell-associated regulators Lrp1 and Usp22. (B) Basal H2B-GFP reporter intensity plotted against distance from the clone boundary. Positive and negative distances indicate positions inside and outside the clone, respectively. Nonlinear regression (red) showed increasing Wnt activity from the edge toward the center of $Trp53^{-/-}$ clones. Ctnnb1 deletion reduced reporter activity and abolished the gradient. (C) Progenitor renewal measured by EdU/BrdU pulse chase (13,15) across distance bins from the clone boundary. $n = 3$ animals per condition. (D to F) Representative clones (D), clonal expansion (E), and Wnt reporter activity (F) after Ctnnb1 deletion in $Trp53^{-/-}$ clones. Scale bar, $200\mu m$ . $n = 5Trp53^{-/-}$ and $4Trp53^{-/-};Ctnnb1^{-/-}$ animals. Data are mean ± SD. Curves in (B) were fitted by nonlinear regression; other comparisons used two-tailed Student's t tests. Exact P values are shown.

---

Discussion

$\Delta$ N90 Ctnnb1-ORF (61). Indeed, nonlinear regression showed that depletion of Sfrp1 was sufficient to generate a Wnt activity gradient in both WT and Trp53 $^{3KR/3KR}$ backgrounds, whereas shLrp1, shUsp22, and $\Delta$ N90 Ctnnb1-ORF resulted in a leveled Wnt activation across the clones (Fig. 6A and fig. S5, A to C). Specific to the gradient pattern of Wnt activity were the radial gradient of renewal (fig. S5D) and a significantly higher long-term expansion rate (Fig. 6, B and C). Notably, even in $\Delta$ N90 Ctnnb1-ORF clones (Fig. 6A and fig. S5, B and C), where the average Wnt activity was higher than is observed upon p53 loss and significantly higher than in Sfrp1-depleted clones (Fig. 6D), we recorded lower rates of progenitor renewal (fig. S5D) and clonal expansion (Fig. 6, B and C) relative to the Wnt gradient clones. Thus, the level of Wnt activity alone did not account for clonal growth. Measurements of clone roundness, border roughness, and circularity showed that gradient-forming clones exhibited more irregular shapes and rougher boundaries than did Wnt plateau clones, suggesting that the former may integrate more effectively with the surrounding tissue (Fig. 6, E and F, and fig. S5, E and F). Together, these analyses associated a radial Wnt activity gradient, rather than uniform elevation, with rapid expansion of p53-deficient clones.

The connection between p53 and the renewal of stem and progenitor cells was first suggested three decades ago, with observation that p53 expression shifts the growth of mammary epithelial cells from geometric to linear, consistent with an increase in asymmetric pattern of division $(62)$ . The p53-dependent effect on stem cell divisions is also documented in daughter-cell doublets in both WT and Erbb2 transgenic mammary glands $(63)$ and in the planarian epidermis $(64)$ . Yet the underlying mechanism remained unclear. Studies of induced pluripotency provide another link: p53 depletion increases reprogramming efficiency $(65–69)$ . However, those reports suggest that p53 suppresses reprogramming via canonical cytostatic pathways and induction of p21, p16INK4a, and p14arf $(65–69)$ . Our unbiased identification of p53 targets in basal progenitors of the adult epidermis followed by an in vivo functional screen for key regulators of progenitor renewal among them led to the discovery of Sfrp1, Lrp1, and Usp22 as direct targets of p53. These genes occupy upstream positions in a p53-governed Wnt-suppression network that restricts clonal expansion by maintaining a homeostatic rate of progenitor renewal.

Early ChIP-seq and expression profiling identify an array of Wnt ligands and receptors as transcriptionally activated by p53 in embryonic stem cells (70). During embryonic development, the p53/63/73 family drives Wnt activation to promote mesendodermal development (71). In contrast, work in cultured fibroblasts showed that p53 can suppress both the protein level and the transcriptional activity of $\beta$ -catenin through glycogen synthase kinase-3 beta-triggered proteasome degradation (72). However, further studies in fibroblasts and cancer cell lines show that p53 transactivates the miR-34 family of micro RNAs, which collectively target the 3' untranslated regions of Wnt1/3, Lrp6, Ctnnb1, and Lef1, thereby acting as suppressors of Wnt activity (73). In this study, we did not observe significant p53 ChIP-seq peaks of previously reported p53 targets within the Wnt pathway. In addition, apart from down-regulation of Fzd10 and Wnt3a transcripts in Trp53 $^{3KR/3KR}$ and Trp53 $^{-/-}$ basal progenitors, which does not correlate with increased Wnt activity measured by both transcriptional signature and in vivo reporter, we did not observe a significant change in the majority of reported p53 targets of the Wnt pathway.

C

B

D

E aligned and normalized clonal boundaries stacked along the long axis

F

Fig. 6. A radial Wnt activity gradient was associated with rapid clonal expansion. (A) Basal H2B-GFP reporter intensity plotted against distance from the clone boundary. Positive and negative distances indicate positions inside and outside the clone, respectively. Nonlinear regression (red) identified gradient-shaped Wnt activation in shSfrp1, $Trp53^{-/-}$ ; Usp22-ORF, and $Trp53^{-/-}$ ; Lrp1-ORF clones and plateau-shaped activation in $Trp53^{-/-}$ ; Sfrp1-ORF, shLrp1, shUsp22, and $\Delta$ N90-Ctnnb1-ORF clones. (B to D) Representative clones (B), clonal expansion (C), and mean Wnt reporter activity (D) under gradient- and plateau-forming conditions. Scale bar, 200 $\mu$ m. (E) Clone boundaries were normalized for size, aligned along the major axis, and superimposed. (F) Clone circularity; higher values indicate more isotropic shapes with smoother borders. Green denotes gradient-forming conditions; blue denotes plateau-forming conditions. Individual points represent clones; bars show mean $\pm$ SD. Curves in (A) were fitted by nonlinear regression; comparisons in (C), (D), and (F) used two-tailed Student's t tests. Exact P values and sample sizes are shown.

Our findings suggest that p53 restricts clonal expansion in the adult epidermis, similar to its proposed role in preventing cheater clones during early embryogenesis (74). It does so through multimodal suppression of Wnt activity, and its loss generates a radial gradient over the normally uniform epidermal Wnt field (57, 75). The importance of this spatial pattern is supported by the more restrained growth of clones with uniform Wnt activity (21, 22, 76), which may be vulnerable to intraclonal crowding or competition with neighboring cells (21, 22, 76, 77). Apc-mutant clones in the intestinal epithelium achieve robust growth by coupling clonal Wnt activation with increased secretion of the Wnt inhibitor Notum, which suppresses Wnt activity in adjacent WT tissue without affecting the ligand-independent Wnt activity in the mutant clone (78, 79). This non-cell-autonomous suppression may help Apc mutant clones overcome competition from neighboring cells but does not inform what may be happening during clonal expansion mediated by an intact Wnt pathway we document in $Trp53^{-/-}$ clones. It is tempting to speculate that a radial Wnt activity gradient may blur the boundary between mutant and WT fields, avoiding or minimizing competition. In this model, $Trp53^{-/-}$ clones sustain a higher rate of progenitor renewal at the center than at the edges and in surrounding WT cells, enabling sustained expansion over time. This mode of growth contrasts with that of $Hras^{G12V}$ mutant clones, which exhibit lower renewal at the center than at the edge (15) and undergo growth stabilization owing to intraclonal competition, an outcome not observed in $Trp53^{-/-}$ clones, which rapidly overtake the surrounding tissue. Together, these results establish a model wherein spatial architecture of a signaling pattern can determine how mutant clones expand within normal epithelium.

---

Materials and methods Animals

We used approximately equal numbers of male and female animals throughout the study. Trp53 $^{3KR/3KR}$ [donated by W. Gu (29, 35)]; Trp53 $^{fl/fl}$ (JAX008462); Ctnnb1 $^{fl/fl}$ (JAX 004152); R26 $^{mTmG}$ (JAX007576), R26 $^{tdTomato}$ (JAX007914) Cre-reporter; and H2B-GFP Wnt-reporter mice (JAX 013752) were all on the C57BL/6 background. Mice were housed and cared for in AAALAC-accredited facilities at the Fred Hutchinson Cancer Center and Icahn School of Medicine at Mount Sinai. All animal experiments were conducted in accordance with Fred Hutchinson Cancer Center, Icahn School of Medicine at Mount Sinai, and ethical regulations and IACUC-approved protocols. Project license numbers: 50814 (Fred Hutchinson Cancer Center), IPROTO202100000070 (Icahn School of Medicine at Mount Sinai).

---

Lentiviral constructs

To achieve RNA interference-mediated gene depletion and minimize off-target effect in the p53-specific perturbation pool, we subcloned five to seven top scored and/or validated shRNAs from the TRCportal database of the Broad Institute (80) to target each candidate as previously described (13, 44). In the $10 \times 10$ two-combination complementary test, ORF overexpression was achieved using pLX EF1 Barcode vector as we reported $(13, 44)$ . To ensure equal lentivirus representation within lentiviral pools, plasmids were mixed in equimolar ratios. Expression of Cre recombinase in combination with either constitutive or tetracycline-inducible ORFs was achieved using the pLX-EF1α-Cre and pLKO-TRE-Cre vectors, respectively. Similarly, coexpression of Cre recombinase with constitutive shRNAs was accomplished using the pLKO-Cre vectors $(13, 44, 45, 81, 82)$ .

---

Lentivirus production and in utero injection

Large-scale production, concentration, and titration of lentivirus were performed as described before (81, 83). Briefly, 120 ml viral supernatant produced by two T300 flasks of 293TN (Invitrogen) cells was concentrated using Centricon Plus-70 centrifugal filter unit (100 kDa, Millipore). Concentrated supernatant was subsequently pelleted through ultracentrifugation (Beckman Optima MAX-TL ultracentrifuge with MLS-50 rotor; 45,000 rpm, 1.5 hour). Lentiviral pellet was resuspended in 70 $\mu$ l of viral resuspension buffer (VRB; 20 mM Tris pH 8.0, 250 mM NaCl, 10 mM MgCl $_{2}$ , 5% sorbitol) as stock. Lentivirus stocks were serially diluted and titrated using genomic DNA from lentivirus expressing green fluorescent protein–infected epidermis as qPCR standard. Barcoded virus was diluted in VRB according to titration results to ensure in utero transduction of epidermis at an \~10% rate. Lentiviral transduction into primary keratinocyte cultures or in vivo through ultrasound-guided in utero microinjection was performed as previously reported (45, 81, 83). Briefly, developmental stage of mouse embryos was determined by Vevo 2100 or 3100 LT animal ultrasound imager using MS550S/D transducer (Fujifilm). One microliter of diluted virus was injected into amniotic cavity of E9.25–E9.75 embryos using glass capillary needle fitted on Celltram Vario micro injector (Eppendorf).

---

In vivo genetic screens

In vivo differentiation screens were performed as previously reported $(13, 15)$ . Briefly, whole skin of mice at P1 was digested in 2 mg/ml dispase at $37^{\circ}$ C for 1 hour to separate epidermis from dermis. Epidermal tissue was further digested into single cells with 0.25% trypsin for 30 min at $37^{\circ}$ C. Epidermal cell suspension from 30 animals was pooled together to make one biological replicate of >50-fold coverage for each shRNA. To isolate basal progenitors from differentiated suprabasal cells, epidermal cells were stained with CD326/EpCAM-APC (G8.8, 1:50; BD Biosciences) and CD49f/ $\alpha$ 6-Integrin-PerCP/Cy5.5 (GoH3, 1:50; Biolegend), both cell populations were isolated using a BD FACSAria II machine (BD Biosciences). Genomic DNA from all samples was extracted using QIAamp DNA tissue mini kit (Qiagen). Barcode preamplification, sequencing, and data processing were performed as previously described $(13, 15, 44, 45, 84)$ . Briefly, PCR preamplification of the barcode region generated products that were made into sequencing libraries using NEBNext Ultra II DNA Library Prep Kit (NEB). Sequencing of barcode libraries was performed on Illumina Nextseq 550 at the Fred Hutchinson Cancer Center Genomic Core facility. Short reads were trimmed down to barcode region by FASTX-Toolkit $(85)$ and mapped to predetermined barcode list using BWA $(86)$ . Relative enrichment of barcodes was calculated using Deseq2 program $(46)$ from raw barcode counts of all replicates, and individual shRNA constructs were scored as significant with a P value cutoff at 0.01. To identify individual genes that affected differentiation, we used the following criteria: (i) More than two shRNAs targeting the same gene had a significant impact on differentiation and (ii) all significant shRNAs targeting the same gene affected differentiation in the same direction (e.g., pro-differentiation or pro-renewal).

---

Immunofluorescence and immunoblot analyses

The following primary antibodies were used: chicken anti-GFP (ab13970, 1:1000 for immunofluorescence (IF); Abcam); mouse anti-BrdU (MoBU-1, 1:100 for IF; Invitrogen); mouse anti- $\beta$ -Actin (66009, 1:3000 for WB,

1:500 for IF; Proteintech); rabbit anti-Keratin10 (Poly19054, 1:1000 for IF; BioLegend); rabbit anti-mouse p53 (NCL-L-p53-CM5p, 1:500 for WB; Leica Biosystem); rabbit anti-active Caspase-3 (C92-605, 1:1000 for IF; BD Bioscience); mouse anti 4-HNE (HNEJ-2, 1:100 for IF; Abcam); mouse anti- $\beta$ -Catenin (15B8, 1:2000 for WB, Sigma); rabbit anti-Lamin A/C (#2032, 1:1000 for WB, Cell Signaling); rabbit anti-Gapdh (14C10, 1:1000 for WB, Cell Signaling); rabbit anti-Sfrp1 (PA5119647, 1:200 for IF; Invitrogen); rabbit anti-Lrp1 (E2Q7S, 1:200 for IF, Cell Signaling); rabbit anti-Usp22 (55110-1-AP, 1:500 for IF, Proteintech). Tissues were processed for immunostaining as previously described (13, 44, 45, 81) and mounted in ProLong Gold with or without 4',6-diamidino-2-phenylindole (Life Technologies). EdU-BrdU pulse-chase assay was performed as we described (13, 15), tissue sections were first processed for EdU on the basis of Click-iT chemistry before subsequent Keratin10 and GFP detection. Lastly, tissue sections were incubated in 2N HCl at 37°C for 30 min to denature DNA, quenched with 0.1M pH 8.5 sodium borate twice, and proceeded for BrdU immunofluorescence detection. Confocal images were taken on Zeiss LSM700 and Andor BC43 systems, using Plan-Apochromat 40×/NA1.4 oil objective (Zeiss) and Plan Apochromat Lambda 60×/NA1.42 oil objective (Nikon). Images were processed using Zeiss Zen and ImageJ software. Immunoblot was performed using Novex system (Invitrogen), and fluorescent signal was captured using Odyssey Fc system (LI-COR).

---

EdU-BrdU pulse-chase proliferation and differentiation assays and quantification

We used previously described assays $(13, 15)$ as follows: We first administered EdU to animals, followed by BrdU injection 2 hours later. Twenty-four hours later, we harvested epidermis and detected EdU and BrdU signal by IF. To calculate cell division interval as a direct measurement of proliferation, we defined the number of $EdU^{+}$ -only cells as E. Because one cell division gives rise to two daughter cells, the number of cell divisions that give rise to $EdU^{+}$ -only cells is E/2. Because only basal epidermal cells divide, the fraction of basal cells that divide within 2 hours to give rise to $EdU^{+}$ -only cells is $F = (E/2)/B$ (where B is total number of basal cells). We next extrapolate this measurement to the whole basal cell population using the following ratio: D/1 = 2 hours/F, where D stands for the average cell division interval (in hours). To directly quantify the rate of progenitor renewal, tissues were also stained for expression of Keratin10 to mark committed progeny. We calculated the rate of progenitor cell renewal of the $EdU^{+}$ -only cells (E) using the following equation: rate of renewal = (number of Keratin10 $^{-}$ E cells) / (total number of E cells).

---

ChIP-seq, ChIP-qPCR, and ATAC-seq analysis

For ChIP-seq analysis of mouse p53, $1 \times 10^{7}$ basal progenitor cells were isolated from P21 WT or $Trp53^{3KR/3KR}$ animals using CD326/EpCAM-APC (G8.8, 1:50; BD Biosciences) and CD49f/ $\alpha$ 6-Integrin-PerCP/Cy5.5 (GoH3, 1:50; Biolegend) staining as described (13, 15). Nutlin-3a (25 $\mu$ M; Sigma Aldrich) was added to digestion medium to stabilize p53 for 1 hour. For ChIP-qPCR, $1 \times 10^{7}$ basal progenitor cells were isolated from P21 WT, $Trp53^{3KR/3KR}$ , and $Trp53^{-/-}$ animals without treatment. Cells were fixed and processed for ChIP using Pierce Magnetic ChIP Kit (Thermo Fisher) and rabbit anti-mouse p53 (NCL-L-p53-CM5p, 5ug/ChIP; Leica Biosystem). For ChIP-seq, purified ChIP DNA bound by nutlin-stabilized p53 was prepped into sequencing library with NEBNext Ultra II DNA Library Prep Kit (NEB). For ChIP-qPCR, purified ChIP DNA bound by native level of p53 was subjected for qPCR with the following primer pairs detecting p53 peaks identified by ChIP-seq: Sfrp1 peak1 (TTACA-GCGTCCAACTCCGAC, CGGCCAGAAGGATCGGTTTA), Sfrp1 peak2 (TCGCAAATCGGAAAGGCATC, TCGGCTAGCATGATCCTCCA), Sfrp1 peak3 (ATACCGGAAGGTAAGCCCCT, CACACAGGCAAACTGCTGAC), Lrp1 peak1 (TCTTCTTGCCTTCACGTCCC, TGTGGCCCTTCATGGCTTAG), Lrp1 peak2 (ACCAGCGACCCAATTAAGGG, AAAAGACAGCGAAGGCGGTA), Lrp1 peak3 (TCTCAGTCGGTTTGGGAACG, CTTCAAACGTCGCCCACAAG), and Usp22 peak1 (CTGTCCGCGGGAAAGGATAA, GCCTTCCAAGGA-GAGACTCG). ATAC-seq was performed on $5 \times 10^{5}$ basal progenitor cells using Illumina Tagment DNA TDE1 Enzyme and Buffer Kit. Both ChIP and ATAC libraries were sequenced on Illumina Nextseq 550. Short reads were mapped via BWA (86); ChIP and ATAC peaks were called using MACS2 (87); peak annotation, peak distribution analysis, and motif enrichment were performed using ChIP-seeker (42) and Homer (43); heatmaps around transcription start sites were generated using seqMINER (88).

---

RNA-seq and gene set analyses

We separated epidermis from the stroma using collagenase-dispase digestion (13, 15, 45, 83). Single cells were obtained after 15 min trypsin digestion, and stained for basal integrin expression (13, 15, 45, 83). We FACS isolated $\alpha 6$ -Integrin $^{\mathrm{high}}$ epidermal progenitors and extracted their RNA using QIAamp RNA mini kit (Qiagen). RNA quality was assessed using Agilent 2100 Bioanalyzer, with all samples passing the quality threshold of RNA integrity numbers $>8$ . Library was prepared using NEBNext mRNA sample preparation kit, and cDNA was sequenced on Illumina Nextseq 550. Reads were mapped to mm10 build of the mouse genome using STAR (89) and summarized with feature-Counts (90) and normalized using Deseq2 (46) and cufflinks (91). Differential expression between multiple groups following distinct pattern were achieved with pattern match algorithm of the MeV package (41) with a cutoff at $P < 0.01$ . GSEA (92, 93) analyses were performed using normalized expression data with GSEA 4.3.2, and Gene Ontology (GO) enrichment analyses were performed using gProfiler (94).

---

Whole-mount imaging and image analysis

Freshly harvested epidermis was mounted in phosphate-buffered saline, sealed in chamber slides, and imaged using high-speed spinning disc confocal (BC-43, Andor/Oxford instrument) with $20 \times 0.8$ Plan Apo LD air objective (Nikon) to minimize potential signal attenuation. In mTmG Cre reporter animals, cells per clone were counted using Imaris (Oxford instrument), and the three mode of division types were assessed as we previously described (13, 15). In H2B-GFP Wnt reporter; TdT Cre reporter animals, 3D image analysis was achieved using Imaris (Oxford instrument). Specifically, local suprabasal Z-planes were removed, individual basal H2B-GFP Wnt reporter nuclei were then identified, and 3D sum of intensity was quantified using spot detection function; clone border was determined using surface reconstruction function according to TdT Cre reporter expression; minimal distance between clone border and individual H2B-GFP Wnt reporter nuclei were calculated using distance transformation function (fig. S3B). Clonal boundaries were outlined as regions of interest and aligned by Image J (95). Clonal roundness was calculated as $(4 \times A)/( \pi \times MA^{2})$ , where A represents clone area and MA represents major axis length; normalized radial variance as $[(1/N) \times \Sigma(\mathrm{ri} - \bar{\mathrm{r}})^{2}] / \bar{\mathrm{r}}^{2}$ , where N represents the total number of boundary points, ri represents the distance of the i-th boundary point from the clone centroid, and $\bar{r}$ represents the mean radial distance; circularity as $(4\pi \times A)/P^{2}$ , where A represents clone area and P represents clonal perimeter.

---

Statistics and reproducibility

All quantitative data were expressed as mean $\pm$ SD. Differences between groups were assayed using two-tailed Student's t test. Significant differences were considered when P < 0.05. Nonlinear regression was performed using Prism 6 (GraphPad software). No statistical methods were used to predetermine sample sizes. For the in vivo genetic screens, sample sizes were selected to provide >50-fold coverage of each perturbation. All other quantitative in vivo experiments included at least three animals or biological replicates. Exact sample sizes are reported in the figure legends. Experiments were not randomized, and investigators were not blinded to genotype or experimental condition during data collection or outcome assessment. No animals or data points were excluded from the analyses, and no predefined inclusion or exclusion criteria were applied.

---

REFERENCES AND NOTES

1. F. Blokzijl et al., Tissue-specific mutation accumulation in human adult stem cells during life. Nature 538, 260–264 (2016). doi: 10.1038/nature19768; pmid: 27698416

2. T. H. H. Coorens et al., Extensive phylogenies of human development inferred from somatic mutations. Nature 597, 387–392 (2021). doi: 10.1038/s41586-021-03790-y; pmid: 34433963

3. S. Park et al., Clonal dynamics in early human embryogenesis inferred from somatic mutation. Nature 597, 393–397 (2021). doi: 10.1038/s41586-021-03786-8; pmid: 34433967

4. Y. S. Ju et al., Somatic mutations reveal asymmetric cellular dynamics in the early human embryo. Nature 543, 714–718 (2017). doi: 10.1038/nature21703; pmid: 28329761

5. I. Martincorena et al., Somatic mutant clones colonize the human esophagus with age. Science 362, 911–917 (2018). doi: 10.1126/science.aau3879; pmid: 30337457

6. A. R. J. Lawson et al., Extensive heterogeneity in somatic mutation and selection in the human bladder. Science 370, 75–82 (2020). doi: 10.1126/science.aba8347; pmid: 33004514

7. A. Cagan et al., Somatic mutation rates scale with lifespan across mammals. Nature 604, 517–524 (2022). doi: 10.1038/s41586-022-04618-z; pmid: 35418684

8. I. Martincorena et al., High burden and pervasive positive selection of somatic mutations in normal human skin. Science 348, 880–886 (2015). doi: 10.1126/science.aaa6806; pmid: 25999502

9. M. Bernabé-Rubio et al., Myc-dependent dedifferentiation of Gata6 $^{+}$ epidermal cells resembles reversal of terminal differentiation. Nat. Cell Biol. 25, 1426–1438 (2023). doi: 10.1038/s41556-023-01234-5; pmid: 37735598

10. F. M. Watt, M. Frye, S. A. Benitah, MYC in mammalian epidermis: How can an oncogene stimulate differentiation? Nat. Rev. Cancer 8, 234–242 (2008). doi: 10.1038/nrc2328; pmid: 18292777

11. I. Arnold, F. M. Watt, c-Myc activation in transgenic mouse epidermis results in mobilization of stem cells and differentiation of their progeny. Curr. Biol. 11, 558–568 (2001). doi: 10.1016/S0960-9822(01)00154-3; pmid: 11369200

12. A. Gandarillas, F. M. Watt, c-Myc promotes differentiation of human epidermal stem cells. Genes Dev. 11, 2869–2882 (1997). doi: 10.1101/gad.11.21.2869; pmid: 9353256

13. Z. Ying, M. Sandoval, S. Beronja, Oncogenic activation of PI3K induces progenitor cell differentiation to suppress epidermal growth. Nat. Cell Biol. 20, 1256–1266 (2018). doi: 10.1038/s41556-018-0218-9; pmid: 30361695

14. S. Brown et al., Correction of aberrant growth preserves tissue homeostasis. Nature 548, 334–337 (2017). doi: 10.1038/nature23304; pmid: 28783732

15. M. Sandoval, Z. Ying, S. Beronja, Interplay of opposing fate choices stalls oncogenic growth in murine skin epithelium. eLife 10, e54618 (2021). doi: 10.7554/eLife.54618; pmid: 33393458

16. S. Gallini et al., Injury prevents Ras mutant cell expansion in mosaic skin. Nature 619, 167–175 (2023). doi: 10.1038/s41586-023-06198-y; pmid: 37344586

17. S. J. Ellis et al., Distinct modes of cell competition shape mammalian tissue morphogenesis. Nature 569, 497–502 (2019). doi: 10.1038/s41586-019-1199-y; pmid: 31092920

18. D. Fernandez-Antoran et al., Outcompeting p53-mutant cells in the normal esophagus by redox manipulation. Cell Stem Cell 25, 329–341.e6 (2019). doi: 10.1016/j.stem.2019.06.011; pmid: 31327664

19. K. Murai et al., p53 mutation in normal esophagus promotes multiple stages of carcinogenesis but is constrained by clonal competition. Nat. Commun. 13, 6206 (2022). doi: 10.1038/s41467-022-33945-y; pmid: 36266286

20. K. Curtius, N. A. Wright, T. A. Graham, An evolutionary perspective on field cancerization. Nat. Rev. Cancer 18, 19–32 (2018). doi: 10.1038/nrc.2017.102; pmid: 29217838

21. M. P. Alcolea et al., Differentiation imbalance in single oesophageal progenitor cells causes clonal immortalization and field change. Nat. Cell Biol. 16, 612–619 (2014). doi: 10.1038/ncb2963; pmid: 24814514

22. I. Pastushenko et al., Fat1 deletion promotes hybrid EMT state, tumour stemness and metastasis. Nature 589, 448–455 (2021). doi: 10.1038/s41586-020-03046-1; pmid: 33328637

23. E. R. Kastenhuber, S. W. Lowe, Putting p53 in context. Cell 170, 1062–1078 (2017). doi: 10.1016/j.cell.2017.08.028; pmid: 28886379

24. S. S. Mello, L. D. Attardi, Deciphering p53 signaling in tumor suppression. Curr. Opin. Cell Biol. 51, 65–72 (2018). doi: 10.1016/j.ceb.2017.11.005; pmid: 29195118

25. V. A. Belyi et al., The origins and evolution of the p53 family of genes. Cold Spring Harb. Perspect. Biol. 2, a001198 (2010). doi: 10.1101/cshperspect.a001198; pmid: 20516129

26. A. B. Williams, B. Schumacher, p53 in the DNA-damage-repair process. Cold Spring Harb. Perspect. Med. 6, a026070 (2016). doi: 10.1101/cshperspect.a026070; pmid: 27048304

27. L. J. Valente et al., p53 efficiently suppresses tumor development in the complete absence of its cell-cycle inhibitory and proapoptotic effectors p21, Puma, and Noxa. Cell Rep. 3, 1339–1345 (2013). doi: 10.1016/j.celrep.2013.04.012; pmid: 23665218

28. C. A. Brady et al., Distinct p53 transcriptional programs dictate acute DNA-damage responses and tumor suppression. Cell 145, 571–583 (2011). doi: 10.1016/j.cell.2011.03.035; pmid: 21565614

29. T. Li et al., Tumor suppression in the absence of p53-mediated cell-cycle arrest, apoptosis, and senescence. Cell 149, 1269–1283 (2012). doi: 10.1016/j.cell.2012.04.026; pmid: 22682249

30. G. Hinkal, N. Parikh, L. A. Donehower, Timed somatic deletion of p53 in mice reveals age-associated differences in tumor progression. PLOS ONE 4, e6654 (2009). doi: 10.1371/journal.pone.0006654; pmid: 19680549

31. M. A. Christophorou, I. Ringshausen, A. J. Finch, L. B. Swigart, G. I. Evan, The pathological response to DNA damage does not contribute to p53-mediated tumour suppression. Nature 443, 214–217 (2006). doi: 10.1038/nature05077; pmid: 16957739

32. C. Pitolli et al., p53-mediated tumor suppression: DNA-damage response and alternative mechanisms. Cancers 11, 1983 (2019). doi: 10.3390/cancers11121983; pmid: 31835405

33. L. Moore et al., The mutational landscape of normal human endometrial epithelium. Nature 580, 640–646 (2020). doi: 10.1038/s41586-020-2214-z; pmid: 32350471

34. J. J. Hao et al., Spatial intratumoral heterogeneity and temporal clonal evolution in esophageal squamous cell carcinoma. Nat. Genet. 48, 1500–1507 (2016). doi: 10.1038/ng.3683; pmid: 27749841

35. L. Jiang et al., Ferroptosis as a p53-mediated activity during tumour suppression. Nature 520, 57–62 (2015). doi: 10.1038/nature14344; pmid: 25799988

36. Q. Gan, W. Li, R. Lex, Z. Ying, S. Beronja, Data from: Inactivation of p53 initiates a radial gradient of Wnt activity and unrestrained epithelial expansion [Dataset], Dryad (2026); https://doi.org/10.5061/dryad.h44j0zq2v.doi:10.5061/dryad.h44j0zq2v

37. E. Y. Cai et al., Selective translation of cell fate regulators mediates tolerance to broad oncogenic stress. Cell Stem Cell 27, 270–283.e7 (2020). doi: 10.1016/j.stem.2020.05.007; pmid: 32516567

38. S. J. Morrison, J. Kimble, Asymmetric and symmetric stem-cell divisions in development and cancer. Nature 441, 1068–1074 (2006). doi: 10.1038/nature04956; pmid: 16810241

39. C. Tomasetti, D. Levy, Role of symmetric and asymmetric division of stem cells in developing drug resistance. Proc. Natl. Acad. Sci. U.S.A. 107, 16766–16771 (2010). doi: 10.1073/pnas.1007726107; pmid: 20826440

40. K. Pappas et al., p53 maintains baseline expression of multiple tumor suppressor genes. Mol. Cancer Res. 15, 1051–1062 (2017). doi: 10.1158/1541-7786.MCR-17-0089; pmid: 28483946

41. E. A. Howe, R. Sinha, D. Schlauch, J. Quackenbush, RNA-Seq analysis in MeV. Bioinformatics 27, 3209–3210 (2011). doi: 10.1093/bioinformatics/btr490; pmid: 21976420

42. G. Yu, L. G. Wang, Q. Y. He, ChIPseeker: An R/Bioconductor package for ChIP peak annotation, comparison and visualization. Bioinformatics 31, 2382–2383 (2015). doi: 10.1093/bioinformatics/btv145; pmid: 25765347

43. S. Heinz et al., Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities. Mol. Cell 38, 576–589 (2010). doi: 10.1016/j.molcel.2010.05.004; pmid: 20513432

44. Z. Ying, S. Beronja, Embryonic barcoding of equipotent mammary progenitors functionally identifies breast cancer drivers. Cell Stem Cell 26, 403–419.e4 (2020). doi: 10.1016/j.stem.2020.01.009; pmid: 32059806

45. S. Beronja et al., RNAi screens in mice identify physiological regulators of oncogenic growth. Nature 501, 185–190 (2013). doi: 10.1038/nature12464; pmid: 23945586

46. M. I. Love, W. Huber, S. Anders, Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 15, 550 (2014). doi: 10.1186/s13059-014-0550-8; pmid: 25516281

47. L. Leyns, T. Bouwmeester, S. H. Kim, S. Piccolo, E. M. De Robertis, Frzb-1 is a secreted antagonist of Wnt signaling expressed in the Spemann organizer. Cell 88, 747–756 (1997). doi: 10.1016/S0092-8674(00)81921-2; pmid: 9118218

48. S. Wang, M. Krinks, K. Lin, F. P. Luyten, M. Moos Jr., Frzb, a secreted protein expressed in the Spemann organizer, binds and inhibits Wnt-8. Cell 88, 757–766 (1997). doi: 10.1016/S0092-8674(00)81922-4; pmid: 9118219

49. H. Suzuki et al., Epigenetic inactivation of SFRP genes allows constitutive WNT signaling in colorectal cancer. Nat. Genet. 36, 417–422 (2004). doi: 10.1038/ng1330; pmid: 15034581

50. C. E. Dann et al., Insights into Wnt binding and signalling from the structures of two Frizzled cysteine-rich domains. Nature 412, 86–90 (2001). doi: 10.1038/35083601; pmid: 11452312

51. A. Zilberberg, A. Yaniv, A. Gazit, The low density lipoprotein receptor-1, LRP1, interacts with the human frizzled-1 (HFz1) and down-regulates the canonical Wnt signaling pathway. J. Biol. Chem. 279, 17535–17542 (2004). doi: 10.1074/jbc.M311292200; pmid: 14739301

52. L. M. Obermoeller-McCormick et al., Dissection of receptor folding and ligand-binding property with functional minireceptors of LDL receptor-related protein. J. Cell Sci. 114, 899–908 (2001). doi: 10.1242/jcs.114.5.899; pmid: 11181173

53. I. Kadurin, S. W. Rothwell, B. Lana, M. Nieto-Rostro, A. C. Dolphin, LRP1 influences trafficking of N-type calcium channels via interaction with the auxiliary $\alpha_{2}\delta$ -1 subunit. Sci. Rep. 7, 43802 (2017). doi: 10.1038/srep43802; pmid: 28256585

54. V. Silva-Vargas et al., $\beta$ -catenin and Hedgehog signal strength can specify number and location of hair follicles in adult epidermis without recruitment of bulge stem cells. Dev. Cell 9, 121–131 (2005). doi: 10.1016/j.devcel.2005.04.013; pmid: 15992546

55. A. Ferrer-Vaquer et al., A sensitive and bright single-cell resolution live imaging reporter of Wnt/ $\beta$ -catenin signaling in the mouse. BMC Dev. Biol. 10, 121 (2010). doi: 10.1186/1471-213X-10-121; pmid: 21176145

56. L. Madisen et al., A robust and high-throughput Cre reporting and characterization system for the whole mouse brain. Nat. Neurosci. 13, 133–140 (2010). doi: 10.1038/nn.2467; pmid: 20023653

57. X. Lim et al., Interfollicular epidermal stem cells self-renew via autocrine Wnt signaling. Science 342, 1226–1230 (2013). doi: 10.1126/science.1239730; pmid: 24311688

58. Oxford Instruments, Imaris; https://imaris.oxinst.com/packages.

59. J. Xiong, Y. Wang, Z. Gong, J. Liu, W. Li, Identification of a functional nuclear localization signal within the human USP22 protein. Biochem. Biophys. Res. Commun. 449, 14–18 (2014). doi: 10.1016/j.bbrc.2014.04.133; pmid: 24802393

60. Y. S. Choi et al., Distinct functions for Wnt/ $\beta$ -catenin in hair follicle stem cell proliferation and survival and interfollicular epidermal homeostasis. Cell Stem Cell 13, 720–733 (2013). doi: 10.1016/j.stem.2013.10.003; pmid: 24315444

61. A. D. Tward et al., Distinct pathways of genomic progression to benign and malignant tumors of the liver. Proc. Natl. Acad. Sci. U.S.A. 104, 14771–14776 (2007). doi: 10.1073/pnas.0706578104; pmid: 17785413

62. J. L. Sherley, P. B. Stadler, D. R. Johnson, Expression of the wild-type p53 antioncogene induces guanine nucleotide-dependent stem cell division kinetics. Proc. Natl. Acad. Sci. U.S.A. 92, 136–140 (1995). doi: 10.1073/pnas.92.1.136; pmid: 7816803

63. A. Cicalese et al., The tumor suppressor p53 regulates polarity of self-renewing divisions in mammary stem cells. Cell 138, 1083–1095 (2009). doi: 10.1016/j.cell.2009.06.048; pmid: 19766563

64. B. J. Pearson, A. Sánchez Alvarado, A planarian p53 homolog regulates proliferation and self-renewal in adult stem cell lineages. Development 137, 213–221 (2010). doi: 10.1242/dev.044297; pmid: 20040488

65. H. Hong et al., Suppression of induced pluripotent stem cell generation by the p53–p21 pathway. Nature 460, 1132–1135 (2009). doi: 10.1038/nature08235; pmid: 19668191

66. H. Li et al., The Ink4/Arf locus is a barrier for iPS cell reprogramming. Nature 460, 1136–1139 (2009). doi: 10.1038/nature08290; pmid: 19668188

67. T. Kawamura et al., Linking the p53 tumour suppressor pathway to somatic cell reprogramming. Nature 460, 1140–1144 (2009). doi: 10.1038/nature08311; pmid: 19668186

68. R. M. Marión et al., A p53-mediated DNA damage response limits reprogramming to ensure iPS cell genomic integrity. Nature 460, 1149–1153 (2009). doi: 10.1038/nature08287; pmid: 19668189

69. J. Utikal et al., Immortalization eliminates a roadblock during cellular reprogramming into iPS cells. Nature 460, 1145–1148 (2009). doi: 10.1038/nature08285; pmid: 19668190

70. K. H. Lee et al., A genomewide study identifies the Wnt signaling pathway as a major target of p53 in murine embryonic stem cells. Proc. Natl. Acad. Sci. U.S.A. 107, 69–74 (2010). doi: 10.1073/pnas.0909734107; pmid: 20018659

71. Q. Wang et al., The p53 family coordinates Wnt and nodal inputs in mesendodermal differentiation of embryonic stem cells. Cell Stem Cell 20, 70–86 (2017). doi: 10.1016/j.stem.2016.10.002; pmid: 27889317

72. E. Sadot, B. Geiger, M. Oren, A. Ben-Ze'ev, Down-regulation of $\beta$ -catenin by activated p53. Mol. Cell. Biol. 21, 6768–6781 (2001). doi: 10.1128/MCB.21.20.6768-6781.2001; pmid: 11564862

73. N. H. Kim et al., p53 and microRNA-34 are suppressors of canonical Wnt signaling. Sci. Signal. 4, ra71 (2011). doi: 10.1126/scisignal.2001744; pmid: 22045851

74. M. Dejosez, H. Ura, V. L. Brandt, T. P. Zwaka, Safeguards for cell cooperation in mouse embryogenesis shown by genome-wide cheater screen. Science 341, 1511–1514 (2013). doi: 10.1126/science.1241628; pmid: 24030493

75. E. R. Deschene et al., $\beta$ -Catenin activation regulates tissue growth non-cell autonomously in the hair stem cell niche. Science 343, 1353–1356 (2014). doi: 10.1126/science.1248373; pmid: 24653033

76. M. D. Lynch et al., Spatial constraints govern competition of mutant clones in human epidermis. Nat. Commun. 8, 1119 (2017). doi: 10.1038/s41467-017-00993-8; pmid: 29066762

77. E. Abby et al., Notch1 mutations drive clonal expansion in normal esophageal epithelium but impair tumor growth. Nat. Genet. 55, 232–245 (2023). doi: 10.1038/s41588-022-01280-z; pmid: 36658434

78. D. J. Flanagan et al., NOTUM from Apc-mutant cells biases clonal competition to initiate cancer. Nature 594, 430–435 (2021). doi: 10.1038/s41586-021-03525-z; pmid: 34079124

79. S. M. van Neerven et al., Apc-mutant cells act as supercompetitors in intestinal tumour initiation. Nature 594, 436–441 (2021). doi: 10.1038/s41586-021-03558-4; pmid: 34079128

80. Broad Institute, Genetic Perturbation Platform (GPP) Web Portal; https://portals.broadinstitute.org/gpp/public/.

81. S. Beronja, G. Livshits, S. Williams, E. Fuchs, Rapid functional dissection of genetic networks via tissue-specific transduction and RNAi in mouse embryos. Nat. Med. 16, 821–827 (2010). doi: 10.1038/nm.2167; pmid: 20526348

82. N. Oshimori, D. Oristian, E. Fuchs, TGF- $\beta$ promotes heterogeneity and drug resistance in squamous cell carcinoma. Cell 160, 963–976 (2015). doi: 10.1016/j.cell.2015.01.043; pmid: 25723170

83. S. Beronja, E. Fuchs, RNAi-mediated gene function analysis in skin. Methods Mol. Biol. 961, 351–361 (2013). doi: 10.1007/978-1-62703-227-8\_23; pmid: 23325656

84. D. Schramek et al., Direct in vivo RNAi screen unveils myosin IIa as a tumor suppressor of squamous cell carcinomas. Science 343, 309–313 (2014). doi: 10.1126/science.1248627; pmid: 24436421

85. FASTX-Toolkit; http://hannonlab.cshl.edu/fastx\_toolkit/.

86. H. Li, R. Durbin, Fast and accurate short read alignment with Burrows-Wheeler transform. Bioinformatics 25, 1754–1760 (2009). doi: 10.1093/bioinformatics/btp324; pmid: 19451168

87. Y. Zhang et al., Model-based analysis of ChIP-Seq (MACS). Genome Biol. 9, R137 (2008). doi: 10.1186/gb-2008-9-9-r137; pmid: 18798982

88. T. Ye et al., seqMINER: An integrated ChIP-seq data interpretation platform. Nucleic Acids Res. 39, e35 (2011). doi: 10.1093/nar/gkq1287; pmid: 21177645

89. A. Dobin et al., STAR: Ultrafast universal RNA-seq aligner. Bioinformatics 29, 15–21 (2013). doi: 10.1093/bioinformatics/bts635; pmid: 23104886

90. Y. Liao, G. K. Smyth, W. Shi, featureCounts: An efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 30, 923–930 (2014). doi: 10.1093/bioinformatics/btt656; pmid: 24227677

91. C. Trapnell et al., Transcript assembly and quantification by RNA-Seq reveals unannotated transcripts and isoform switching during cell differentiation. Nat. Biotechnol. 28, 511–515 (2010). doi: 10.1038/nbt.1621; pmid: 20436464

92. J. Kao et al., Molecular profiling of breast cancer cell lines defines relevant tumor models and provides a resource for cancer gene discovery. PLOS ONE 4, e6146 (2009). doi: 10.1371/journal.pone.0006146; pmid: 19582160

93. A. Subramanian et al., Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl. Acad. Sci. U.S.A. 102, 15545–15550 (2005). doi: 10.1073/pnas.0506580102; pmid: 16199517

94. L. Kolberg et al., g:Profiler—interoperable web service for functional enrichment analysis and gene identifier mapping (2023 update). Nucleic Acids Res. 51, W207–W212 (2023). doi: 10.1093/nar/gkad347; pmid: 37144459

95. J. Schindelin et al., Fiji: An open-source platform for biological-image analysis. Nat. Methods 9, 676–682 (2012). doi: 10.1038/nmeth.2019; pmid: 22743772

---

ACKNOWLEDGMENTS

We thank W. Gu for sharing the $Trp53^{3KR/3KR}$ mouse; the Fred Hutchinson Cancer Center Comparative Medicine (AAALAC accredited; G. Roble, director) and the Center for Comparative Medicine and Surgery of Icahn School of Medicine at Mount Sinai (AAALAC accredited; J. Cohen, director) for care of mice in accordance with National Institutes of Health (NIH) guidelines; the Fred Hutchinson Cancer Center Genomics Core (J. Delrow, director) for sequencing; and the Fred Hutchinson Cancer Center Flow Cytometry Core (A. Berger, director) for flow cytometry and FACS. Funding: This research was funded in part by NIH/NCI Cancer Center Support Grant P30 CA015704, NIH/NIAMS SBDRC Support Grant P30 5P30AR079200; NIH grants R01-AR070780 (S.B.), R00-DE029229 (Z.Y.), and R01-DE034100 (Z.Y.); Department of Defense grant HT9425-23-1-0038 (Z.Y.); and the RSG-19-074-01-TBE award (S.B.) and CSCC-PF-23-994844 fellowship (R.K.L.) from the American Cancer Society. Author contributions: All authors conceived of the study and designed the experiments. Q.G. performed all major experiments and data analysis. W.L. and R.K.L. contributed to data generation and analysis. Z.Y. performed epigenetic analysis and genetic screens. Z.Y. and S.B. wrote the paper. All authors provided intellectual input and vetted and approved the final manuscript. Competing interests: The authors declare that they have no competing interests. Data, code, and materials availability: ChIP-seq data that support the findings of this study have been deposited in the Gene Expression Omnibus (GEO) under accession code GSE275726. Previously published microarray data that were reanalyzed here are available under accession code GSE1579. All quantification data presented in figures are available at Dryad (36). Lentiviral constructs generated in this study are available from the corresponding authors upon request and subject to completion of a standard material transfer agreement. There are no additional restrictions on access to materials generated in this study. License information: Copyright © 2026 the authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original US government works. https://www.science.org/about/science-licenses-journal-article-reuse

---

SUPPLEMENTARY MATERIALS

science.org/doi/10.1126/science.aed3065

Figs. S1 to S6; References (96, 97); MDAR Reproducibility Checklist

Submitted 22 October 2025; resubmitted 22 June 2026; accepted 19 August 2026

10.1126/science.aed3065

Mitochondrial carrier SLC25A34 links clock, diet, and temperature control of adipocyte lipid cycling

Iuliia Karavaeva et al.

INTRODUCTION: Brown and beige thermogenic adipocytes (fat cells) have uniquely high energy expenditure capacity. This activity is critical for defending body temperature in small mammals and is linked to cardiometabolic health in humans. Energy expenditure from thermogenic fat cycles daily on a rhythm set by the circadian clock, rising after waking and dropping during sleep. At any point, however, this activity can be suddenly induced by exposure to cold temperature or by ingestion of a calorie-dense, high-fat meal. How thermogenic fat can seamlessly achieve both rhythmic continuity and acute responsiveness is a fundamental question of energy homeostasis that has remained unknown.

RATIONALE: We set out to understand how diverse cues from the body's clock, environmental temperature, and diet are molecularly integrated in adipocytes to facilitate both anticipatory and adaptive regulation. By combining unbiased analysis of DNA-binding patterns, gene expression, and protein levels in thermogenic fat from mice that were subject to different circadian, environmental, and dietary perturbations, we identified the orphan mitochondrial metabolite transporter SLC25A34 as the convergence point of all three physiological paradigms.

RESULTS: Similar to fat's thermogenic activity, the expression of Slc25a34 mRNA exhibited a daily rhythm, which was mediated by the nuclear receptors REV-ERBα and REV-ERBβ. REV-ERBs are transcriptional repressors in the circadian clock that recruit histone deacetylase 3 to the Slc25a34 promoter during sleep to shut off transcription. When animals wake and REV-ERB circadian repression is lifted, the peroxisome proliferator-activated receptors α and γ (PPARα and PPARγ) are able to bind the Slc25a34 promoter and restore expression. However, at any time, when animals are suddenly confronted with an unanticipated need for adipocyte

Full article and list of author affiliations: https://doi.org/10.1126/science.adz4797

The circadian clock, diet, and temperature control lipid homeostasis through SLC25A34. During sleep, REV-ERBs repress Slc25a34. After waking, cold exposure, or lipid-rich meals, PPARs rapidly induce expression. We hypothesize that SLC25A34 imports oxaloacetate (OAA) into the mitochondria to support fatty acid oxidation and lipid synthesis and maximize TCA cycle and electron transport chain (ETC) activity. Cit, citrate; NAA, N-acetylaspartate. [Figure created with BioRender.com]

energy expenditure such as exposure to cold temperatures or eating a lipid-rich meal, REV-ERB repression is rapidly overridden and PPARs are activated by lipolytic signals to boost SLC25A34 levels. When the energy demand is met, the REV-ERB-mediated rhythmicity of Slc25a34 is reestablished.

Our biochemical and metabolic data support a functional model in which SLC25A34 transports oxaloacetate into mitochondria. In doing so, SLC25A34 helps to maintain high tricarboxylic acid (TCA) cycle activity, which is required for powering energy expenditure in thermogenic adipocytes. The need for SLC25A34 during times of high energetic demand arises because thermogenic adipocytes not only oxidize fatty acids in the mitochondria but also simultaneously synthesize lipids in the cytosol, a process collectively known as lipid cycling. Lipid synthesis requires continuous production of acetyl-coenzyme A (acetyl-CoA) in the cytosol, and the two major pathways responsible for this acetyl-CoA pool also generate oxaloacetate as a by-product, which then needs to be transported back into the mitochondria. Increased cytosolic acetyl-CoA additionally promotes the transcription of genes linked to mitochondrial respiration. Thus, SLC25A34 supports both sides of this self-reinforcing cycle in which lipids are continually broken down and resynthesized.

CONCLUSION: These findings identify SLC25A34 as a molecular node that integrates circadian, temperature, and dietary cues and reveal how thermogenic fat activity can sustain a daily rhythm while remaining acutely responsive to sudden energy demands. This mechanism may offer a new entry point for therapeutically enhancing energy expenditure in metabolic disease. □

Mitochondrial carrier SLC25A34 links clock, diet, and temperature control of adipocyte lipid cycling

Iuliia Karavaeva $^{1}$ , Astrid Linde Basse $^{1,2}$ , Samuel A. J. Trammell $^{3}$ , Mohammed Faiz Hussain $^{4,5}$ , Lasse Kruse Markussen $^{2,6}$ , Jesper F. Havelund $^{7}$ , Marie Sophie Isidor $^{1}$ , Adam Chatoff $^{8,9}$ , Andrea Andress Huacachino $^{8}$ , Hannah J. Richter $^{10,11,12}$ , Meghna Birla $^{13}$ , Sabina Chubanava $^{1,14}$ , Yann Deleye $^{15}$ , Rini Arianti $^{16}$ , David Tandio $^{1}$ , Sarah E. Utzon $^{1}$ , Zafir Kaiser $^{4,5}$ , Yachen Shen $^{10,17}$ , Ditte Neess $^{7}$ , Hannes Embring $^{1}$ , Dilip Menon $^{13}$ , Olivia J. Conway $^{13}$ , Frederike Sass $^{1,2}$ , Fabian Finger $^{1,2}$ , Lidia Argemi-Muntadas $^{1}$ , Tao Ma $^{1,2}$ , Elahu G. Sustarsic $^{1}$ , Cecilie Kynding Kristensen $^{1}$ , Rebecca L. McIntyre $^{1,2}$ , Genesee J. Martinez $^{18}$ , Anna Sofie Husted $^{1}$ , Matthew J. Emmett $^{11,19,20,21}$ , Zachary A. Kipp $^{18}$ , Mikkel Frost $^{1}$ , Mark P. Jedrychowski $^{22}$ , Michel van Weeghel $^{23}$ , Homa Majd $^{24}$ , Ekaterina Zhuravleva $^{25}$ , Robert W. McGarrah $^{15}$ , Kaja Plucińska $^{26}$ , Mohit K. Midha $^{10,11}$ , Andreas Prokesch $^{27,28}$ , Paul Cohen $^{26}$ , James G. Granneman $^{29}$ , Patrick Seale $^{10}$ , Riekelt H. Houtkooper $^{23}$ , Jacob B. Hansen $^{30}$ , Steven P. Gygi $^{22}$ , Thue W. Schwartz $^{1}$ , Matthew P. Gillum $^{3}$ , Terry D. Hinds, Jr. $^{18}$ , Raymond E. Soccio $^{10}$ , Phillip J. White $^{15}$ , Edmund R. S. Kunji $^{24}$ , Thomas Moritz $^{1}$ , Jonas T. Treebak $^{1}$ , Endre Kristóf $^{16}$ , Susanne Mandrup $^{2,6}$ , Brice Emanuelli $^{1}$ , Nathaniel W. Snyder $^{8,9}$ , Daniel J. Fazakerley $^{13}$ , Lawrence Kazak $^{4,5}$ , Nils J. Færgeman $^{7}$ , Mitchell A. Lazar $^{10,11}$ , Zachary Gerhart-Hines $^{1,2*}$

Adipocyte lipid metabolism is coordinated by circadian rhythms, diet, and environmental temperature, but how these diverse signals are molecularly integrated remains unknown. We showed that these cues converge on the orphan mitochondrial transporter SLC25A34 to orchestrate lipid cycling. During the sleep phase, the adipocyte clock suppresses Slc25a34 expression through the REV-ERB transcriptional repressors. Entering the active phase, consuming lipid-rich diets, or exposure to cold abolishes REV-ERB repression, and lipolytic signals stimulate Slc25a34 transcription through the peroxisome proliferator–activated receptors. SLC25A34 is proposed to import oxaloacetate into mitochondria, dually supporting the tricarboxylic acid cycle and cytosolic acetyl–coenzyme A (acetyl-CoA) production. Elevated cytosolic acetyl-CoA then fuels the synthesis of lipids and promotes the transcription of genes enhancing mitochondrial oxidation. Thus, SLC25A34 confers circadian, dietary, and temperature control of adipocyte lipid metabolism.

The circadian clock synchronizes physiological processes with environmental cues, integrating signals from sunlight, diet, physical activity, and temperature to optimize metabolic function across tissues (1-3). Circadian metabolic control must balance anticipated daily energetic needs with acute adaptation to unexpected perturbations. Adipose thermogenesis is an ideal paradigm in which to study this coordinated anticipatory and adaptive nature of the clock. In mice and humans, the energy-expending activity of thermogenic adipocytes is regulated by exposure to cold temperature (4-6), diet composition and quantity consumed (7, 8), and the daily oscillations of the circadian clock (9-11). In rodents, these energy-dissipating cells are critical for defending body temperature against environmental cold exposure (12), whereas in humans, brown adipose tissue (BAT) appears to improve lipid and glucose homeostasis and is associated with cardiometabolic protection $(13)$ . In mice and humans, BAT metabolic activity begins to increase just before waking $(9, 14, 15)$ , peaks late in the waking period, and then declines to its lowest point during sleep. Given the energetic cost of BAT activity, clock repression of thermogenesis during sleep may have conferred an evolutionary advantage by reducing unnecessary energy expenditure.

Regardless of the time of day, acute exposure to cold temperature can override circadian control and rapidly stimulate adipose thermogenesis through the sympathetic nervous system $(4)$ . Although the specific metabolic pathways that mediate the interaction between the clock and temperature cues are unresolved, lipid utilization likely contributes because numerous elements of adipocyte lipid metabolism are regulated by both programs $(4, 16, 17)$ . A unique hallmark of BAT lipid metabolism is the ability to simultaneously increase lipid synthesis in the cytosol and fatty acid $\beta$ -oxidation in the mitochondria $(18–20)$ . Such synthesis and breakdown fuel an energy-dissipating lipid cycle that promotes thermogenesis and represents a possible therapeutic strategy against obesity and metabolic disease $(21, 22)$ . Both environmental cold $(18–20)$ and the circadian clock $(23)$ can influence this dual activation of anabolic and catabolic lipid pathways. However, how continuous circadian rhythmicity and acute cold responsiveness are molecularly integrated to coordinate thermogenic metabolism remains unknown.

---

Clock and temperature control of the orphan mitochondrial transporter Slc25a34

To understand how circadian and cold-regulated signaling pathways orchestrate thermogenic metabolism in adipocytes, we used a combination of proteomics, transcriptomics, and genomic binding to identify candidate factors in BAT (Fig. 1A and table S1). We searched for (i) proteins that increased in level in the BAT of mice exposed to 4°C cold exposure for 3 weeks versus thermoneutrality [fold change (FC) >2] (24); (ii) proteins that increased in level in the BAT of whole-body Rev-erba knockout (KO) mice compared with controls at 29°C (FC >2) (table S2); (iii) genes with a Rev-erba-binding site within 10 kilobases (kb) of the transcriptional start site that showed higher occupancy by REV-ERBα at thermoneutrality than in cold conditions (9); and (iv) genes that were regulated both by cold and REV-ERBα and REV-ERBβ at the transcriptional level [FC >2 in wild-type (WT) mice upon cold exposure and in BAT-specific Rev-erba and β double KOs (Rev-erba/β BKO) compared with WT mice, false discovery rate (FDR) <0.05 in all cases] (table S3). Only two factors satisfied all criteria: the well-established, canonical thermogenic effector uncoupling protein 1 (UCP1, also known as SLC25A7) and the uncharacterized, orphan mitochondrial metabolite transporter SLC25A34. Both proteins are members of the SLC25A family of mitochondrial solute carriers (25, 26) highlighting the critical importance of metabolite and ion exchange across the inner mitochondrial membrane (IMM) for adipocyte function.

At thermoneutrality (29°C), Slc25a34 expression was the highest in the heart, followed by slow-twitch skeletal muscle, kidney, and multiple brain regions (Fig. 1B). BAT was among the organs with the lowest amounts of Slc25a34 mRNA. However, upon 24 hours of cold exposure (4°C), Slc25a34 mRNA accumulated in BAT 90-fold over the levels at thermoneutrality, causing BAT to supplant heart as the most Slc25a34-enriched tissue (Fig. 1B). Transcription of Slc25a34 increased in BAT after only 8 hours of cold exposure and remained elevated throughout 3 weeks of cold adaptation (Fig. 1C). At the protein level, the abundance of SLC25A34 was significantly increased after 3 days of cold exposure and then for the duration of cold acclimation (Fig. 1D). We next examined SLC25A34 protein localization and stability in thermogenically active adipocytes. Using the $\beta_{3}$ -adrenergic receptor agonist CL-316,243 to mimic cold exposure in cultured brown adipocytes, we could confirm mitochondrial localization of an exogenously expressed

B

Protein abundance in BAT during cold adaptation

SLC25A34

SLC25A35

C

mRNA expression in BAT during cold adaptation

Slc25a34 Slc25a35

D

F

E

Slc25a34 expression in scWAT during cold adaptation

H

G Protein abundance in BAT

J

K

REV-ERBα-HDAC3 repress Slc25a34 expression

Fig. 1. Clock and temperature control of the orphan mitochondrial transporter Slc25a34 through the REV-ERB circadian repressors. (A) Schematic of the multi-omic approach to discover Slc25a34 comparing the proteome of BAT from (i) cold-exposed mice (top left, 3 weeks cold exposure versus thermoneutrality, FC > 2) and (ii) Rev-erbα WKO mice at thermoneutrality (bottom left, Rev-erbα WKO versus control, FC > 2), (iii) the BAT cistrome from control and Rev-erbα WKO mice (top right, REV-ERBα-binding sites located within 10 kb of the nearest gene that show higher occupancy in cold conditions than at thermoneutrality), and (iv) the BAT transcriptome from cold-exposed

$^{1}$ Novo Nordisk Foundation Center for Basic Metabolic Research, University of Copenhagen, Copenhagen, Denmark. $^{2}$ Center for Adipocyte Signaling (ADIPOSIGN), University of Southern Denmark, Odense, Denmark. $^{3}$ Department of Biomedical Sciences, University of Copenhagen, Copenhagen, Denmark. $^{4}$ Rosalind & Morris Goodman Cancer Institute, McGill University, Montreal, Quebec, Canada. $^{5}$ Department of Biochemistry, McGill University, Montreal, Quebec, Canada. $^{6}$ Functional Genomics & Metabolism Research Unit, University of Southern Denmark, Odense, Denmark. $^{7}$ Department of Biochemistry and Molecular Biology, University of Southern Denmark, Odense, Denmark. $^{8}$ Aging and Cardiovascular Discovery Center, Temple University, Philadelphia, PA, USA. $^{9}$ Cancer Signaling and Tumor Microenvironment, Fox Chase Cancer Center, Philadelphia, PA, USA. $^{10}$ Institute for Diabetes, Obesity, and Metabolism, University of Pennsylvania, Philadelphia, PA, USA. $^{11}$ Division of Endocrinology, Diabetes, and Metabolism, University of Pennsylvania, Philadelphia, PA, USA. $^{12}$ Gladstone Institute of Data Science and Biotechnology, San Francisco, CA, USA. $^{13}$ Institute of Metabolic Science–Metabolic Research Laboratories, University of Cambridge, Cambridge, UK. $^{14}$ Institute for Diabetes and Obesity, Helmholtz Diabetes Center, Munich, Germany. $^{15}$ Duke Molecular Physiology Institute, Department of Medicine, Duke University, Durham, NC, USA. $^{16}$ University of Debrecen, Debrecen, Hungary. $^{17}$ Department of Pathology and Laboratory Medicine, University of Pennsylvania, Philadelphia, PA, USA. $^{18}$ University of Kentucky College of Medicine, Lexington, KY, USA. $^{19}$ Broad Institute of MIT and Harvard, Cambridge, MA, USA. $^{20}$ Massachusetts General Hospital/Harvard Medical School, Boston, MA, USA. $^{21}$ Dana-Farber Cancer Institute, Harvard Medical School, Boston, MA, USA. $^{22}$ Department of Cell Biology, Harvard Medical School, Boston, MA, USA. $^{23}$ Laboratory Genetic Metabolic Diseases, Amsterdam Gastroenterology and Metabolism; Amsterdam Cardiovascular Sciences, Amsterdam UMC, University of Amsterdam, Amsterdam, Netherlands. $^{24}$ MRC Mitochondrial Biology Unit, University of Cambridge, Cambridge, UK. $^{25}$ LEO Foundation Skin Immunology Research Center, University of Copenhagen, Copenhagen, Denmark. $^{26}$ Laboratory of Molecular Metabolism, The Rockefeller University, New York, NY, USA. $^{27}$ Gottfried Schatz Research Center for Cell Signaling, Metabolism & Aging, Division of Cell Biology, Histology and Embryology, Medical University of Graz, Graz, Austria. $^{28}$ BioTechMed-Graz, Graz, Austria. $^{29}$ Wayne State University School of Medicine, Detroit, MI, USA. $^{30}$ Department of Biology, University of Copenhagen, Copenhagen, Denmark. \*Corresponding author. Email: zpg@sund.ku.dk

Rev-erbα/β BKO mice (bottom right, genes induced by cold in control mice (FC > 2, FDR < 0.05) and genes induced by double Rev-erbα/β KO in BAT at thermoneutrality (FC > 2, FDR < 0.05). (B) Slc25a34 expression across tissues expressed as the fold change relative to the expression in the tissue at 29°C. (C and D) Slc25a34 and Slc25a35 mRNA expression (C) and protein abundance (D) in BAT during cold adaptation. Significance indicates effects of cold exposure compared with 29°C. (E) Slc25a34 mRNA expression in scWAT during cold adaptation. Significance indicates effects of cold exposure compared with 29°C. (F) Circadian expression of Slc25a34 and Slc25a35 mRNA in control and Rev-erbα WKO mice acclimated to 29°C. (G) SLC25A34 and SLC25A35 protein abundances in BAT of control and Rev-erbα WKO mice at 22°C at ZT10. (H) Slc25a34 mRNA abundance in BAT of control and Rev-erbα WKO mice at 29°C and after 1.5 hours of 4°C exposure (ZT4 to ZT5.5). (I) REV-ERBα occupancy in the proximal Slc25a34 promoter in BAT at 29°C and after 6 hours (ZT4 to ZT10) of 4°C exposure of control and Rev-erbα WKO mice. (J) Slc25a34 mRNA abundance in BAT from control and Hdac3 AKO mice at 22°C and after 3 hours at 4°C (ZT7 to ZT10). (K) Schematic depicting the role of REV-ERBs in coordinating circadian and cold-mediated regulation of Slc25a34 expression. For all panels, data are represented as mean ± SEM. \*P < 0.05, \*\*P < 0.01, \*\*\*P < 0.001, \*\*\*\*P < 0.0001, unpaired two-tailed multiple t tests with the Benjamini, Krieger, and Yekutieli two-stage linear step-up procedure for multiple-comparisons correction [(B) and (H)]; one-way ANOVA with the Benjamini, Krieger, and Yekutieli two-stage linear step-up procedure for multiple comparisons correction [(C) to (E)]; or two-way ANOVA with the Benjamini, Krieger, and Yekutieli two-stage linear step-up procedure for multiple comparisons correction [(F), (G), and (J)].

hemagglutinin (HA)-tagged SLC25A34 (fig. S1, A and B), consistent with the SLC25A carrier family. However, unlike other canonical adipocyte carriers, such as UCP1 and SLC25A1 (the citrate transporter), SLC25A34 was markedly less stable, as evidenced by its decreased abundance after inhibition of protein synthesis with cycloheximide (fig. S1, C and D).

Cold exposure did not cause accumulation of SLC25A35, the closest homolog of SLC25A34, which is proposed to have a similar transport function based on sequence conservation in key substrate-binding residues (27, 28). In fact, the abundance of Slc25a35 mRNA was acutely suppressed by cold exposure (fig. S1E and Fig. 1C), and protein concentrations remained unchanged throughout cold acclimation (Fig. 1D). In subcutaneous white adipose tissue (scWAT), which undergoes recruitment of energy-expending beige adipocytes during cold adaptation, both transporters accumulated after cold exposure but Slc25a34 to a nearly 40-fold greater extent (Fig. 1E and fig. S1F). Thus, the abundance of SLC25A34 is strongly and sustainably increased in thermogenic adipose tissues after exposure to environmental cold temperature.

---

Circadian and cold control of Slc25a34 by REV-ERBs and HDAC3-mediated transcriptional repression

In addition to regulation by environmental temperature, expression of Slc25a34 in BAT also oscillates over 24 hours (Fig. 1F). Whole-body genetic KO of the circadian nuclear receptor Rev-erba (hereby referred to as Rev-erba WKO) not only abolished this rhythmicity but also resulted in constitutively increased Slc25a34 expression compared with that in control littermates (Fig. 1F), demonstrating that REV-ERBα plays a critical role in clock-mediated repression of Slc25a34 transcription. Consistent with this, the abundance of SLC25A34 protein was increased in BAT from Rev-erba WKO mice (Fig. 1G and table S2). Conversely, Slc25a35 expression was not circadian nor was its protein abundance changed upon Rev-erba deletion (Fig. 1, F and G). Given that the circadian clock, through REV-ERBα, modulates temperature-dependent thermogenic control (9), we investigated the interaction between REV-ERBα suppression and cold enhancement of Slc25a34 transcription. We measured Slc25a34 and Slc25a35 expression in BAT from thermoneutral-acclimated and cold-exposed Rev-erba WKOs and control littermates. Cold exposure increased Slc25a34 expression in the control mice. However, in Rev-erba WKO BAT, the amounts of Slc25a34 were already maximally increased at thermoneutrality and could not be further increased upon cold exposure (Fig. 1H), suggesting that derepression of REV-ERBα control is a critical step in the effects of cold on transporter abundance. Indeed, REV-ERBα binds to two sites in the proximal promoter and the first intron of Slc25a34 under thermoneutral conditions, and this binding was abolished by cold temperature (Fig. 1I). In contrast to Slc25a34, Slc25a35 was downregulated by cold to the same extent in both Rev-erba WKOs and control littermates (fig. S1G) despite the presence of the REV-ERBα-binding site in its proximal promoter (fig. S1H).

Because of the global nature of the Rev-erba WKO, we could not rule out that REV-ERBα-dependent regulation in the central nervous system or non-BAT peripheral tissues might indirectly influence BAT Slc25a34 expression. Therefore, we conditionally deleted Rev-erba and $\beta$ in brown adipocytes (Rev-erba/ $\beta$ BKOs) to determine whether the observed regulation of Slc25a34 expression in BAT was cell autonomous. We depleted both REV-ERBs because they share the same consensus DNA-binding sites, and Rev-erb $\beta$ has been shown to compensate for the loss of Rev-erba in other tissue-specific KOs (29). Moreover, although Rev-erba was previously shown to be down-regulated by cold exposure (9), we found that cold similarly decreased Rev-erb $\beta$ in BAT (fig. S2, A and B), suggesting that both REV-ERBs might contribute to the cold regulation of Slc25a34. Indeed, BAT Slc25a34 expression was increased in Rev-erba/ $\beta$ BKO mice compared with littermate controls both at thermoneutrality and in the cold (fig. S2C and table S3). Conversely, Slc25a35 expression was similar between control and Rev-erba/ $\beta$ BKO mice and was significantly decreased by cold exposure in both genotypes (fig. S2D). Circadian rhythmicity of Slc25a34 expression was additionally blunted in Rev-erba/ $\beta$ BKO, along with that of Ucp1 and core clock REV-ERB target brain and muscle ARNT-like 1 (Bmal1) (fig. S2, E to G). Collectively, these data show that REV-ERB transcriptional repressors cell autonomously control both cold and circadian regulation of Slc25a34.

REV-ERBs mediate transcriptional repression through the recruitment of histone deacetylase 3 (HDAC3) and subsequent histone deacetylation (30). To address whether HDAC3 contributes to cold regulation of Slc25a34, we measured its expression in the BAT of mice with adipose-specific genetic deletion of Hdac3 (Hdac3 AKO) and littermate controls at room temperature (RT, 22°C) and after acute exposure to cold. BAT Slc25a34 expression in Hdac3 AKO mice was increased at RT compared with that of control mice and was not further increased by cold exposure (Fig. 1J), indicating that HDAC3 acts as the REV-ERB-directed corepressor of Slc25a34. Conversely, BAT Slc25a35 expression was decreased in the Hdac3 AKO compared with control animals at RT and was further suppressed by cold in the Hdac3 AKO mice (fig. S2H). In a previously published RNA-sequencing (RNA-seq) dataset of BAT from Hdac3 AKO and control mice at thermoneutrality (31), Slc25a34 was the third most up-regulated gene when Hdac3 was deleted (fig. S2, I and J). Together, these data show that REV-ERBα, REV-ERBβ, and HDAC3 coordinate circadian and temperature-dependent repression of Slc25a34 at thermoneutrality and during sleep, which is relieved by waking and acute cold exposure (Fig. 1K).

---

Cold- and diet-induced lipid metabolism stimulate Slc25a34 expression through peroxisome proliferator-activated receptor $\alpha$ and $\gamma$

Although our findings establish REV-ERBα- and REV-ERBβ-dependent repression as a key event in the regulation of Slc25a34 abundance, the cell-autonomous derepression of REV-ERBs could not explain the entirety of the transcriptional control by cold exposure. Therefore, we next sought to identify the transcription factor or factors contributing to the activation of Slc25a34 expression. Given the numerous transcriptional regulators involved in thermogenic metabolism in adipocytes (32), we sought to identify candidate activators by performing pathway enrichment analysis of the top 50 proteins that exhibited similar expression patterns as SLC25A34 throughout cold adaptation (24) (fig. S3A and table S4). We hypothesized that determining the pathways co-regulated with SLC25A34 might reveal the factors responsible for the transcription control. We found that nearly all proteins co-regulated with SLC25A34 were linked to biological processes related to lipid metabolism and biosynthesis (Fig. 2A). Therefore, we tested whether peroxisome proliferator-activated receptors (PPARs), the canonical regulators of adipocyte lipid metabolism (33), could be transcriptional activators of Slc25a34. Indeed, both PPARα and PPARγ bind to a regulatory region near Slc25a34 that is enriched for active histone marks (34). Agonists for PPARα (GW7647) or PPARγ (rosiglitazone) cell autonomously increased Slc25a34 expression in brown adipocytes alongside their canonical targets Ucp1 and Elovl3 (Fig. 2, B to E, and fig. S3, B and C). Small interfering RNAs (siRNAs) targeting either factor significantly decreased transporter mRNA abundance (Fig. 2, F and G). Thus, both PPARα and PPARγ appear to regulate Slc25a34 expression in vitro; however, PPARγ can both control PPARα expression (35) and compensate for its loss (36), confounding assignment of specific PPAR dependency.

Therefore, we performed RNA sequencing of BAT from mice with adipose-specific genetic depletion of Pparα (Pparα AKO; targeting exon 5) and control littermates. Deletion of Pparα resulted in only 10 significantly down-regulated mRNAs (Fig. 2H, fig. S3D, and table S5). Among these, Slc25a34 showed the greatest reduction (Figs. 2, H and I, and fig. S3D), revealing it to be one of the most robust and selective PPARα targets. By contrast, neither Ucp1 nor Slc25a35 expression was affected by Pparα deletion (Fig. 2J and fig. S3E). This is most likely because Pparγ, which was expressed similarly in BAT from Pparα KOs compared with control animals (fig. S3F), could compensate for some genes such as Ucp1 but not Slc25a34 in the absence of Pparα. We further confirmed an essential role for PPARα in regulating Slc25a34 expression in vivo using a second adipose-specific KO model in which exon 4 was deleted (37) (fig. S3, G to J). To address the effect of acute Pparα depletion on physiological BAT Slc25a34 regulation, we used a tamoxifen-inducible Pparα genetic depletion mouse model expressing CreERT2 under control of the Ucp1 promoter (Pparα iBKO). Pparα iBKO and littermate control mice acclimated to RT were exposed to thermoneutrality or cold for 3 days concurrent with the tamoxifen-induced depletion of Pparα. Pparα expression was decreased in Pparα iBKO mice both at thermoneutrality and after cold exposure compared with that in control animals (fig. S3K). This acute ablation of Pparα in BAT significantly blunted the cold-induced accumulation of Slc25a34, demonstrating the importance of PPARα for physiological temperature-sensitive regulation of Slc25a34 (Fig. 2K). Conversely, cold-induced Ucp1 expression was unaffected by Pparα ablation, suggesting full compensation by PPARγ (Fig. 2L). Thus, whereas both PPARα and PPARγ can increase Slc25a34 expression, PPARα appears to be the predominant physiological regulator of this transporter.

In addition to cold, the transcription-activating functions of PPARs can also be modulated by products of lipolysis, which act as PPAR agonists (38, 39). Given that lipid catabolism is increased in BAT during cold exposure (6), we evaluated whether lipolysis influences Slc25a34 expression. Brown adipocytes were treated with SR-3420, a small molecule that directly enhances adipose triglyceride lipase (ATGL)-dependent triglyceride lipolysis by disrupting the interaction between its coactivator $\alpha\beta$ hydrolase domain-containing protein 5 (ABHD5) and Perilipin 1 (40, 41). SR-3420 increased the expression of Slc25a34 mRNA in the brown adipocytes (Fig. 2M). Furthermore, the lipolytic induction of Slc25a34 expression was completely prevented by blocking fatty acid activation with the long-chain acyl-coenzyme A synthetase (ACSL) inhibitor triacsin C or by preventing the transport of fatty acids into the mitochondria with the carnitine palmitoyltransferase 1a (CPT1a) inhibitor etomoxir (Fig. 2M). These data suggest that endogenous lipid products from mitochondrial fatty acid $\beta$ -oxidation activate Slc25a34 expression in brown adipocytes, likely under PPAR $\alpha$ -dependent control.

Exogenous dietary lipids also activate PPARs and BAT thermogenic capacity similarly to cold exposure (42). Therefore, we compared the effects of a 60% high-fat diet (HFD) and a ketogenic diet with 93% fat content with chow diet (12% fat) (Fig. 2N) on the expression of the transporter in BAT of mice acclimated to RT (22°C). A HFD modestly increased Slc25a34 mRNA abundance, but the ketogenic diet increased mRNA abundance of the transporter to a much greater extent (Fig. 2O). Bulk RNA sequencing further revealed that Slc25a34 was the second most induced mRNA in BAT after ketogenic diet feeding (Fig. 2P and table S6). Consistent with the ketogenic diet results, fasting also increased the expression of Slc25a34 in BAT, skeletal muscle, and heart, but not liver (Fig. 2Q), whereas Slc25a35 and Ucp1 expression was decreased (fig. S3, L and M). Therefore, Slc25a34 is dually controlled by circadian REV-ERB-HDAC3 repression and cold-induced, lipid-liganded PPARα/γ activation (Fig. 2R).

---

SLC25A34 controls mitochondrial respiration in thermogenic adipocytes in vitro and in vivo

To investigate the role of the transporter in mitochondrial function, Slc25a34 was acutely depleted in immortalized murine brown adipocytes. Three different siRNAs against the transporter (fig. S4, A to D) significantly decreased respiratory capacity, particularly norepinephrine (NE)-stimulated and carbonyl cyanide p-trifluoromethoxyphenylhydrazone (FCCP)-dependent maximal respiration (Fig. 3A and fig. S4E). Conversely, continuous measurement of basal respiration over the course of siRNA knockdown (KD) revealed that the effects on basal respiration were periodic and not sustained (fig. S4F). Depletion of Slc25a34 also reduced maximal respiratory capacity of the brown adipocytes not treated with NE (Fig. 3B) and NE-stimulated and maximal respiration in primary brown adipocytes (Fig. 3C). Despite the significant reduction in respiratory capacity, Slc25a34 depletion did not affect adipocyte differentiation (fig. S4G). Conversely, depleting Slc25a35 had little or no impact on NE-stimulated respiratory capacity (Fig. 3D and fig. S4H), highlighting a functional distinction between the two homologs.

SLC25 carriers transport a diverse array of metabolites across the IMM, including amino acids, fatty acids, nucleotides, and cofactors (25, 26). Sequence analysis suggests that SLC25A34 may be a carboxylic acid transporter (25, 26). Given that several carboxylic acids serve as mitochondrial fuel, we evaluated how mitochondrial respiration in brown adipocytes with or without SLC25A34 changed in response to different substrates. We permeabilized cultured brown adipocytes with a plasma membrane permeabilizer that serves to dilute out the cytosolic metabolites and endogenous respiratory substrates. We then supplemented the permeabilized cells with different substrates for respiratory complex I (malate, pyruvate, and glutamate), complex II (succinate), and donating electrons into the Q-pool (octanoylcarnitine, palmitoylcarnitine, and glycerol-3-phosphate) alongside adenosine 5'-diphosphate (ADP) and magnesium and measured their respiratory capacity (Fig. 3E). We observed lower coupled respiration in the cells depleted of Slc25a34 on nearly all the tested substrates (Fig. 3F and fig. S4I), indicating that the mitochondrial deficiency was likely the result of a broader disruption of tricarboxylic acid (TCA) cycle activity or respiratory capacity. Thus, our findings indicate that SLC25A34 is required for the mitochondrial function of brown adipocytes.

To test the role of Slc25a34 in vivo, we generated mice in which Slc25a34 can be deleted in brown adipocytes in an inducible manner (Slc25a34 iBKO). We directly assessed BAT thermogenic function using an experimental paradigm in which animals were anesthetized at thermoneutrality, injected with NE, and changes in oxygen consumption were measured in indirect calorimetric cages (43) (Fig. 3G). NE-induced respiration was significantly lower in Slc25a34 iBKO mice compared with littermate controls (Fig. 3, H to J). This occurred despite unchanged Ucp1 mRNA levels between genotypes (fig. S4J), indicating that SLC25A34 mediates adrenergic-induced mitochondrial respiratory activity in vitro and in vivo.

A Pathway enrichment analysis of proteins temporally co-regulated with SLC25A34 during cold adaptation

B

3 GW7647-induced Slc25a34 expression

C

G

GW7647-induced Ucp1 expression

D ROSI-induced Slc25a34 expression

F

H

E

ROSI-induced Ucp1 expression

Genes regulated by Pparα KO

|

BAT RNAseq Slc25a34 expression

Ppar expression

PPAR target gene expression

J

BAT RNAseq Ucp1 expression

K

BAT Slc25a34 expression

L

BAT Ucp1 expression

M

Slc25a34 expression is induced by lipolysis

N Diet kCal%

P

Q

R

0 Slc25a34 expression in BAT

Genes regulated by ketogenic diet

Slc25a34 expression

Transcriptional control of SLC25A34 expression

Fig. 2. Cold- and diet-induced lipid metabolism activate PPARα and PPARγ to mediate temperature control of Slc25a34. (A) Pathway enrichment of proteins temporally co-regulated with SLC25A34 in BAT during cold adaptation. (B to E) Regulation of Slc25a34 and Ucp1 mRNA by the PPARα agonist GW7647 [(B) and (C)] or the PPARγ agonist rosiglitazone [(D) and (E)]. (F and G) Gene expression in brown adipocytes after siRNA KD of Pparα and Pparγ. (H) Volcano plot comparing BAT transcriptomes of control and Pparα AKO (targeted exon 5) mice acclimated to 22°C. (I and J) Gene expression in BAT of control and Pparα AKO (targeted exon 5) mice quantified in RNA-seq analysis. (K and L) Gene expression in BAT from control and Pparα iBKO mice acclimated to 22°C and exposed to 29°C or 4°C for 3 days. (M) Slc25a34 mRNA in brown adipocytes after treatment with the lipolytic activator SR-3420 and the ACSL inhibitor Triacsin C or the CPT1a inhibitor etomoxir. (N and O) Diet composition and effects of diet challenges on BAT Slc25a34 mRNA (harvested between ZT4 and ZT6). (P) Volcano plot depicting genes from BAT RNA-seq that are regulated by ketogenic diet. (Q) Slc25a34 mRNA expression analysis across tissues of WT mice fed or fasted for 24 hours. (R) Schematic representation of how PPARα coordinates Slc25a34 expression with REV-ERBs in BAT. For all panels, data are represented as mean ± SEM. \*P < 0.05, \*\*P < 0.01, \*\*\*P < 0.001, \*\*\*\*P < 0.0001, two-way ANOVA with the Benjamini, Krieger, and Yekutieli two-stage linear step-up procedure for multiple-comparisons correction [(B) to (G) and (K) to (M)]. For (I) and (J), the P value was computed using Bioconductor software's package and corrected for multiple testing using the Benjamini-Hochberg mode of the R function p.adjust to compute a FDR. For (O), one-way ANOVA with the Benjamini, Krieger, and Yekutieli two-stage linear step-up procedure for multiple-comparisons correction was used. For (Q), unpaired two-tailed multiple t tests with the Benjamini, Krieger, and Yekutieli two-stage linear step-up procedure for multiple-comparisons correction was used.

A Brown adipocyte NE-stimulated respiration

C Primary brown adipocyte NE-stimulated respiration

B

B Brown adipocyte FCCP-stimulated respiration

D Brown adipocyte NE-stimulated respiration, Slc25a35 KD

F

E Adipocyte permeabilization

Respiration of permeabilized brown adipocytes

H

|

G

J BAT Slc25a34 expression

Fig. 3. SLC25A34 controls mitochondrial respiratory capacity in thermogenic adipocytes in vivo and in vitro. (A and B) NE-induced (A) and maximal (B) OCR of brown adipocytes with control or siRNA-mediated Slc25a34 KD. Shown are a representative trace (left) and quantification of maximal OCR values reached during each treatment in n = 9 (A) and n = 3 (B) independent experiments (right). (C) OCR in primary brown adipocytes with control or siRNA-mediated Slc25a34 KD. Shown are a representative trace (left) and quantification of maximal OCR values reached during each treatment in two independent experiments (right). (D) OCR in brown adipocytes with control or siRNA-mediated Slc25a35 KD. Shown are a representative trace (left) and quantification of maximal values reached during each treatment in four independent experiments (right). (E) Schematic of experimental design for assessing respiration in permeabilized brown adipocytes. (F) Respiration of permeabilized brown adipocytes after 4 days of siRNA-mediated Slc25a34 KD and addition of specific respiratory substrates (from left to right): pyruvate and malate, succinate with addition of rotenone, and glycerol-3-phosphate. (G) Schematic of experimental design for the in vivo NE-stimulated thermogenesis test in Slc25a34 iBKO and control mice. (H and I) Oxygen consumption (H) and corresponding area under the curve (I) of anaesthetized mice during the NE-stimulated thermogenesis test. (J) Slc25a34 mRNA expression in BAT mice from the NE-stimulated thermogenesis test. KD of Slc25a34 in (A) to (C) and (F) was performed with siRNA #1; measurements were made after 4 days of KD induction. For all panels, data are represented as mean ± SEM. \*P < 0.05, \*\*P < 0.01, \*\*\*P < 0.001, \*\*\*\*P < 0.0001, two-way repeated-measures ANOVA with the Šidák correction for multiple comparisons [(A) to (D)] or unpaired two-tailed t test [(I) and (J)].

---

SLC25A34 defends mitochondrial oxaloacetate availability and supports cytosolic acetyl-CoA production

To obtain deeper insights into the potential function of SLC25A34, we profiled the metabolomes of brown adipocytes at steady state 4 days after siRNA-mediated depletion of the transporter. Of the 82 differentially regulated metabolites (table S7), the most significant change was the accumulation of N-acetylaspartate (Fig. 4, A and B). N-acetylaspartate is synthesized in mitochondria from oxaloacetate (OAA)-derived aspartate and exported to the cytosol, where sequential hydrolysis by aspartoacylase (ASPA) and acetyl-coenzyme A (acetyl-CoA) synthetase 2 (ACSS2) generates cytosolic acetyl-CoA (44). Consistent with disrupted utilization of N-acetylaspartate, whole-cell concentrations of acetyl-CoA, which largely reflect the cytosolic and nuclear concentrations (45), were significantly decreased after 4 days of Slc25a34 depletion (Fig. 4C). Thus, SLC25A34 may help to maintain the acetyl-CoA pool in thermogenic adipocytes.

Given the large number of metabolites changed at 4 days, we next performed a time course with earlier time points to distinguish primary from secondary metabolite changes (Fig. 4, D and E). Even after just 1 day of depletion, before any changes in cellular respiratory capacity manifested, N-acetylaspartate was still the most dysregulated among only five significantly altered metabolites (Fig. 4, F to I, and tables S8 and S9), indicating that it may be directly affected by the loss of SLC25A34. These rapid metabolite changes after 1 day of siRNA are consistent with the relatively low protein stability that we observed for SLC25A34 (fig. S1, C and D). The other acute metabolite changes included dysregulation of aspartate, $\alpha$ -ketoglutarate, proline, and betaine (Fig. 4, I to M). Thus, ablation of Slc25a34 may disrupt the malate-aspartate shuttle (MAS) and/or the TCA cycle. N-acetylaspartate metabolism by ASPA and ACSS2 is not the only route to generate cytosolic acetyl-CoA in adipocytes. Citrate exported from the mitochondria is metabolized to acetyl-CoA by ATP citrate lyase (ACLY) (46, 47). Both ACLY and ACSS2, the key enzymes in the two acetyl-CoA-generating pathways, are regulated by REV-ERBα and cold temperature in BAT, similarly to SLC25A34 (Fig. 4N), providing a potential link between these pathways. Moreover, the N-acetylaspartate and citrate routes of cytosolic acetyl-CoA production each result in the generation of OAA in the cytosol (Fig. 4O), which is the proposed substrate of the yeast homolog of SLC25A34 (28, 48).

A Metabolites dysregulated in BA with Slc25a34 KD

B

C

F

D

Knockdown time course experimental design

E

G

|

J

K α-Ketoglutarate

L

M

N ACLY and ACSS2 regulation by cold and REV-ERBα

0 Acetyl-CoA shuttling pathways

Fig. 4. SLC25A34 promotes cytosolic acetyl-CoA production. (A) Global differential metabolite changes between brown adipocytes with control or Slc25a34 KD after 4 days of KD induction. BA, brown adipocytes. (B and C) Relative abundance of N-acetylaspartate (NAA) (B) and acetyl-CoA (C) in brown adipocytes with control or Slc25a34 KD after 4 days of the KD induction. (D) Schematic of the experimental design. (E) Slc25a34 expression. (F) Mitochondrial respiration. (G) Heatmap of metabolite changes. (H) Overview of metabolite changes. (I to M) Specific metabolite abundances from an Slc25a34 siRNA-mediated KD time course in brown adipocytes. In (G), the metabolites outlined in black boxes represent the five significantly changed metabolites after 1 day of Slc25a34 KD. Slc25a34 KD on all panels was performed with siRNA #1. (N) Regulation of ACSS2 and ACLY protein abundance by cold in WT mice (left) and by REV-ERBα in Rev-erbα WKO mice (right). (O) Model showing how the generation of cytosolic acetyl-CoA by mitochondrial export of citrate (Cit) and NAA both lead to the production of OAA in the cytosol. For all panels, data are represented as mean ± SEM. \*P < 0.05, \*\*P < 0.01, \*\*\*P < 0.001, \*\*\*\*P < 0.0001, unpaired two-tailed Student's t test [(B) and (C)]; two-way ANOVA with the Benjamini, Krieger, and Yekutieli two-stage linear step-up procedure for multiple comparisons correction [(E), (I) to (M), and (N), left]; or Wilcoxon rank-sum test [(N), right].

Therefore, we tested whether SLC25A34 played a role in balancing the cytosolic and mitochondrial pools of OAA. Because of the labile nature of OAA, we chemically derivatized samples with sodium borohydride, converting OAA to a more stable compound, deuterated malate, for quantification. We purified mitochondria with a rapid HA-tag-based protocol (49) from brown adipocytes with or without depletion of Slc25a34 and performed targeted metabolomics analysis of purified mitochondria and of the mitochondria-depleted (cytosolic) fraction and total cell lysates (Fig. 5A and fig. S5A). In control cells, OAA levels were higher in the cytosol relative to the mitochondria, establishing a concentration gradient or ratio consistent with potential SLC25A34-mediated OAA import into mitochondria (Fig. 5B). Depletion of Slc25a34 further increased the cytosolic-to-mitochondria OAA ratio by raising cytosolic levels and lowering the mitochondrial pool without altering total cellular OAA (Fig. 5B). We also observed dysregulation of the MAS metabolites aspartate, glutamate, and malate (fig. S5, B to D), consistent with the established role of OAA in this circuit. Collectively, these findings support a model in which SLC25A34 facilitates the import of OAA into mitochondria.

To further explore the biochemical feasibility of OAA as a substrate for SLC25A34, we performed docking simulations of OAA with human SLC25A34 in Autodock Vina with the SLC25A34 structure predicted by AlphaFold. These simulations were done with hydrated substrate and flexible residues in the conserved binding sites and matrix salt bridge network. OAA could dock on SLC25A34 with high affinity (-15.42 kcal/mol) in a binding pose in which it formed polar contacts between its carboxyl and carbonyl groups and two arginine residues (R176 and R277) in the conserved common substrate-binding site characteristic of SLC25 carriers (fig. S5E). This pose also permitted the formation of polar contacts between the carbonyl and carboxyl groups of OAA and the four residues that make up the matrix salt bridge network (K31, D228, E28, and K128), which braced SLC25A34 in a matrix closed conformation. Docking to the conserved binding site in a manner that can disrupt the matrix salt bridge network is necessary for substrate transport by SLC25A carriers (50).

Because a lower mitochondrial OAA pool could result from reduced glucose- or glutamine-dependent anaplerosis, we next performed multi-isotope tracing by incubating control and Slc25a34-depleted brown adipocytes with $[^{13}C_{6}]$ -glucose and $[^{13}C_{5}]$ -glutamine for 5, 30, and 60 min (see Fig. 5C for the 60-min time point and table S10 for all time points). We additionally included $[^{13}C_{2},^{2}H_{3}]$ -acetate, which enters the TCA cycle directly as acetyl-CoA, allowing us to determine OAA availability independently of anaplerotic flux. Consistent with our previous analysis, we found that depletion of Slc25a34 significantly increased the total pool of N-acetylaspartate (Fig. 5D). However, in contrast to our earlier metabolomic analysis (Fig. 4J and fig. S5B), global aspartate abundance was decreased in the tracing dataset, potentially due to differences in media composition and analytical platforms (see the Materials and methods). Therefore, we prioritized analysis of the isotopologue enrichment to gain insight into carbon routing in the presence and absence of SLC25A34. $[^{13}C_{6}]$ -glucose tracing revealed increased m+2 labeling in acetyl-CoA and increased m+2 and m+3 labeling in citrate, fumarate, malate, and aspartate in the absence of Slc25a34 (Fig. 5, E and F, and fig. S6B), suggesting elevated glucose-derived anaplerosis. Glutamine additionally supplied more TCA carbons after loss of Slc25a34, as $[^{13}C_{5}]$ -glutamine tracing showed a shift toward m+4 labeling in TCA intermediates and aspartate and increased glutamine-derived acetyl-CoA (Fig. 5, G and H, and fig. S6C).

However, $[^{13}C_{2},^{2}H_{3}]$ -acetate labeling of acetyl-CoA, TCA intermediates, and N-acetylaspartate, which is dependent on the OAA pool, was broadly diminished by Slc25a34 depletion (Fig. 5, I and J, and fig. S6D), consistent with the limiting OAA levels due to the absence of SLC25A34. Collectively, the isotope-tracing metabolomic data are consistent with the hypothesis that SLC25A34 maintains TCA flux by importing OAA from the cytosol to sustain its mitochondrial availability. When the transporter is depleted, glucose- and glutamine-driven anaplerosis are elevated to partially compensate for TCA carbon supply, whereas acetate entry into the TCA cycle is constrained. In this model, disrupted OAA transport leads to impaired N-acetylaspartate turnover and decreased total citrate, functionally linking SLC25A34 to two major pathways for generating cytosolic acetyl-CoA.

---

SLC25A34 couples lipid cycling with transcriptional control of mitochondrial oxidation

In adipocytes, one of the primary metabolic fates of cytosolic acetyl-CoA is to support de novo lipogenesis (DNL) $(51)$ . We assessed DNL through the incorporation of radiolabeled carbons from ${}^{14}$ C-glucose into triglycerides after control or siRNA-mediated depletion of Slc25a34 (Fig. 6A). ${}^{14}$ C-glucose was used because it radiolabels cytosolic acetyl-CoA, and subsequently triglycerides, through TCA-derived citrate through ACLY $(52)$ and N-acetylaspartate through ACSS2 $(53)$ , the two rate-limiting enzymes that we found to be cold and circadian regulated (Fig. 4N), similar to SLC25A34. Consistent with the acetyl-CoA measurements, loss of Slc25a34 significantly diminished glucose-supported triglyceride production in brown adipocytes (Fig. 6B). In further support of the role for SLC25A34 in DNL, we found that insulin, one of the key lipogenic activators in adipocytes, also significantly induced Slc25a34 expression similarly to known DNL mediators (fig. S7, A to D).

In addition to fueling DNL, cytosolic acetyl-CoA is also a critical mediator of transcriptional control of nuclear-encoded genes (46, 54). To assess the potential impact of Slc25a34-dependent acetyl-CoA production on nuclear transcription, we globally profiled gene expression from brown adipocytes with or without Slc25a34 depletion using an siRNA construct #1 as well as a siRNA construct pool (#2) (fig. S4A). When comparing the expression profiles of the two siRNAs, the shared changes revealed that transcription of 1489 genes was increased and that of 1685 genes was decreased (fig. S7E, Fig. 6C, and table S11). Despite causing a general decrease in the abundance of acetyl-CoA, Slc25a34 ablation did not globally affect histone or total protein acetylation (fig. S7, F to H) but instead specifically affected the expression of particular gene networks. Pathway enrichment analysis revealed that the up-regulated genes were related to inflammatory and interferon pathways, underscoring the importance of SLC25A34 in maintaining cellular homeostasis (fig. S7I). To define the temporal sequence of these changes, we profiled a panel of inflammatory and interferon-responsive genes over a 4-day Slc25a34 KD time course. Il6, Cxcl1, and Icam1 were induced from day 1 and remained elevated, indicating an early and sustained nuclear factor $\kappa$ B (NF- $\kappa$ B)-driven proinflammatory state (fig. S7, J to L). Il16, a CD4-binding chemoattractant cytokine, was induced from day 2 onward, consistent with a transition to broader immune communication (fig. S7M). Classical interferon-stimulated genes (e.g., Ifit1, Cxcl10, Isg15, Ifi44, and Gbp4) showed coordinated up-regulation only at later time points, consistent with a delayed interferon response after prolonged mitochondrial dysfunction (fig. S7, N to R).

The genes down-regulated by Slc25a34 depletion were the ones significantly linked to the mitochondrial TCA cycle and the respiratory chain (Fig. 6D). Consistent with this transcriptional down-regulation, we observed a significant decrease in citrate synthase (CS) activity upon loss of Slc25a34 (Fig. 6E). However, this defect did not appear to reflect a reduction in mitochondrial number because the ratio of mitochondrial DNA to nuclear DNA remained unchanged (fig. S8A).

C Stable isotope tracing experiment design

Acetyl-CoA Citrate α-Ketoglutarate Fumarate

A

D Citrate

B Relative abundance of oxaloacetate in cellular compartments

Whole cell lysate

N-acetylaspartate

E

F

$^{13}$ C glucose tracer

G

H

$^{13}$ C glutamine tracer

|

J

$^{13}$ C acetate tracer

Fig. 5. Triple-isotope tracing implicates SLC25A34 in routing substrates into the TCA cycle and OAA-to-NAA cataplerosis. (A and B) Schematic for organelle-specific metabolomics (A) and abundance of OAA (B) across cellular compartments. Data shown are the average of four independent experiments. (C) Schematic of the experimental design of stable isotope tracing in brown adipocytes. (D) Relative abundance of unlabeled aspartate, citrate, acetyl-CoA, succinyl-CoA, and NAA in Slc25a34 KD and control cells. (E and F) Schematic overview (E) and isotopologue enrichment percentage from $[^{13}C_{6}]$ -glucose tracing (F) for different metabolites. (G and H) Schematic overview (G) and isotopologue enrichment percentage from $[^{13}C_{5}]$ -glutamine tracing (H) for different metabolites. (I and J) Schematic overview (I) and isotopologue enrichment percentage from $[^{13}C_{2},^{2}H_{3}]$ -acetate tracing (J) for different metabolites. Ctrl, Control cells; KD, cells with Slc25a34 KD. Slc25a34 KD on all panels was performed with siRNA #1, and measurements are made after 4 days of KD induction. The tracing experiment was repeated three times, all measurements were combined, and the average values are shown. For all panels, data are represented as mean ± SEM. \*\*P < 0.01, \*\*\*P < 0.001, \*\*\*\*P < 0.0001, unpaired two-tailed Student's t test [(B) and (D)], #P < 0.05 siControl versus siSlc25a34, multiple paired t test with the Benjamini, Krieger, and Yekutieli two-stage linear step-up procedure for multiple-comparisons correction [(F), (H), and (J)].

G

J

Fig. 6. SLC25A34 couples lipid cycling with transcriptional control of mitochondrial oxidation. (A and B) Schematic and assessment of radiolabeled glucose-derived DNL in brown adipocytes with control or siRNA-mediated Slc25a34 KD after 4 days of KD. (C) Comparison of genes significantly down-regulated after Slc25a34 KD using two different siRNAs (siRNA #1 and siRNA #2 (pool)). (D) Pathway enrichment for the conserved gene signature down-regulated by both siRNAs. (E) CS activity of brown adipocytes with control or Slc25a34 KD after 4 days of KD. (F and G) Promoter motif analysis of the genes down-regulated by loss of Slc25a34 for candidate transcription factors (F) and pathway enrichment of factor-specific gene clusters (G). (H) Effect of acetate supplementation on respiration and CS activity of brown adipocytes with siRNA-mediated Slc25a34 KD. Shown are a representative seahorse trace (left) and CS activity (right). Slc25a34 KD was performed with siRNA #1, measurements done after 4 days of KD induction. The experiment was repeated five times (see fig. S8G for quantification). (I) Classification of the protein cluster jointly regulated by cold and the circadian clock. (J) Schematic showing circadian and cold- and diet-dependent transcriptional control of Slc25a34 by REV-ERBs and PPARα/γ and a proposed model by which the SLC25A34-mediated import of OAA into mitochondria facilitates cytosolic acetyl-CoA production and intracellular lipid cycling. For all panels, data are represented as mean ± SEM. \*P < 0.05, \*\*\*P < 0.001, \*\*\*\*P < 0.0001, unpaired two-tailed Student's t test [(B) and (E)] or one-way ANOVA with the Benjamini, Krieger, and Yekutieli two-stage linear step-up procedure for multiple-comparisons correction [(H), right].

Therefore, our findings demonstrate that SLC25A34 affects mitochondrial function at least in part through transcriptional control of nuclear genes linked to specific mitochondrial pathways.

As a transporter embedded in the IMM, SLC25A34 cannot directly regulate nuclear genes. To understand the downstream transcriptional control, we examined the promoters of SLC25A34-regulated genes for transcription factor motifs. To refine our focus to the most relevant candidates, we further assessed how highly expressed each transcriptional regulator was in brown adipocytes and whether its expression was changed after depletion of Slc25a34. Among the genes that showed increased expression after Slc25a34 ablation, promoters in the cluster associated with genes implicated in inflammatory pathways were most enriched in the signal transducers and activators of transcription (STAT) and interferon regulatory factor (IRF) transcription factor motifs (fig. S8B and table S12). These factors were also highly expressed basally in brown adipocytes and were increased after Slc25a34 depletion. Thus, STATs and IRFs appear to mediate the induction of inflammatory pathways after loss of the transporter (fig. S8C). Analysis of the genes that showed decreased mRNA expression after loss of Slc25a34 revealed seven clusters linked to transcription factors that satisfied the criteria for brown adipocyte expression and SLC25A34 dependency (Fig. 6F and table S12). Pathway enrichment analysis found significant associations with three of the seven clusters. Clusters 1 and 3 represent targets of PPARα and PPARγ and comprise genes linked to lipid synthesis and catabolism (Fig. 6G and fig. S8D). This SLC25A34-dependent association with PPAR lipid metabolism genes may reveal a positive reinforcement mechanism, given that PPARα and PPARγ also drive Slc25a34 expression. The genes in cluster 7 are related to mitochondrial respiratory function and were most strongly linked to estrogen-related receptor α (ERRα) control (Fig. 6G). Thus, the resulting decrease in cytosolic acetyl-CoA appears to act through specific transcription factor pathways.

We investigated whether the dysfunction caused by depleting Slc25a34 could be alleviated by exogenously boosting cytosolic acetyl-CoA (fig. S8E). This was done by supplementing the cell media with acetate (fig. S8F), which is metabolized in the cytosol by ACSS2 to produce acetyl-CoA, theoretically circumventing the requirement for mitochondrial export of N-acetylaspartate or citrate. Indeed, culturing brown adipocytes with acetate mitigated the respiratory defects caused by Slc25a34 KD (Fig. 6H and fig. S8G). Furthermore, acetate supplementation restored mitochondrial CS activity in Slc25a34-depleted cells (Fig. 6H). These findings support the idea that SLC25A34 influences mitochondrial function partly through cytosolic acetyl-CoA production.

Our data demonstrate that SLC25A34 has an essential role in both cytosolic DNL and mitochondrial respiration, thereby coordinating both anabolic and catabolic lipid processes across organelles. This role is further highlighted by mapping proteins co-regulated by cold and the circadian clock onto their metabolic pathways: Nearly all support lipid synthesis, shuttling, re-esterification, or mitochondrial thermogenesis and may be functionally coordinated by SLC25A34 (Fig. 6I and fig. S8H). Collectively, our findings are consistent with a model in which the circadian clock, environmental temperature, and diet leverage mitochondrial import of OAA through SLC25A34 to couple and accelerate the citrate and N-acetylaspartate pathways and generate cytosolic acetyl-CoA (Fig. 6J). This acetyl-CoA pool fuels DNL and the transcriptional control of mitochondrial respiratory capacity, both of which serve to increase mitochondrial lipid oxidation and thermogenesis and, ultimately, complete an SLC25A34-dependent circuit of lipid synthesis and breakdown.

To investigate the role of SLC25A34 in human biology, we performed siRNA KD of the transporter in primary in vitro differentiated brown adipocytes from four independent donors. Loss of SLC25A34 reduced cyclic adenosine 3',5'-monophosphate (cAMP)-inducible respiratory capacity in three of the four donor adipocytes (Fig. 7A)

and transcription of genes linked to thermogenic function in adipocytes from all donors (Fig. 7B). In humans, SLC25A34 may contribute to adipose and whole-body energy homeostasis beyond thermogenic BAT (13). A meta-analysis of 24 clinical studies (https://adiposetissue.org/) (55–78) revealed that expression of SLC25A34 in scWAT was negatively correlated with circulating C-reactive protein, triglycerides, and low-density lipoprotein cholesterol, Homeostasis Model Assessment of Insulin Resistance (HOMA-IR), body-mass index (BMI), waist circumference, and waist-to-hip ratio (WHR), and positively correlated with circulating high-density lipoprotein cholesterol and the ratio of stimulated versus basal lipolysis (Fig. 7C). We found no significant associations with SLC25A34 expression in omental adipose tissue, possibly indicating that SLC25A34 may play a depot-specific beneficial role in promoting systemic metabolic health.

---

Discussion

The circadian clock has been evolutionarily fine-tuned over hundreds of millions of years to anticipate recurring daily metabolic needs and coordinate tissue-specific homeostatic programs accordingly $(79)$ . This rhythmic framework must also integrate rapid responses to acute environmental and dietary changes. Chronic modern-day perturbations, such as irregular eating patterns, high-calorie diets, and reduced exposure to temperature fluctuations, can overwhelm this capacity, contributing to neurologic, psychiatric, cardiometabolic, and immune disorders $(79–82)$ . Delineating the mechanisms that integrate circadian machinery with tissue-specific metabolic control is therefore critical for understanding evolutionary biology and identifying therapeutic targets.

Adipose thermogenesis exemplifies the integration of circadian, environmental, and dietary control of metabolism. Thermogenic function is suppressed by the clock during sleep and rises before awakening to peak during the waking period, thereby governing daily fluctuations in systemic lipid and glucose homeostasis (9, 14, 16). However, this rhythmicity must also be coordinated with rapid BAT responsiveness to cold (4, 83), which itself influences glucose and lipid metabolism (17), whereas dietary composition and caloric density further shape thermogenic output (84). We discovered that the orphan mitochondrial metabolite carrier SLC25A34 represents a key regulatory node in BAT downstream of all three inputs by integrating circadian control through REV-ERBα and β-HDAC3 repression with cold temperature and dietary regulation through lipid-mediated PPARα and PPARγ activation. The individual contributions of REV-ERBα and REV-ERBβ remain unclear, but given their functional redundancy (85), shared DNA-binding sites (29), and parallel cold down-regulation in BAT, they likely play similarly important roles. The stronger effect observed in the Rev-erba WKO probably reflects indirect contributions from other tissues such as the liver and WAT.

A key hallmark of BAT activation is the simultaneous engagement of mitochondrial fatty acid $\beta$ -oxidation and cytosolic DNL, enabling lipid cycling (86). Fatty acid $\beta$ -oxidation and glucose catabolism (52) feed the mitochondrial acetyl-CoA pool and TCA cycle, and the resulting metabolites citrate and N-acetylaspartate are exported to the cytosol. Their catabolism yields acetyl-CoA that primes DNL, the lipid products of which can be reincorporated into triglycerides or returned to mitochondria for $\beta$ -oxidation, thus completing the cycle. Notably, generating cytosolic acetyl-CoA from citrate or N-acetylaspartate also produces OAA, the cytosolic fate of which has remained unresolved. In Slc25a34-depleted brown adipocytes, we observed N-acetylaspartate accumulation, dysregulated TCA intermediates, a shift of OAA from mitochondria to cytosol, and reduced $[^{13}C_{2},^{2}H_{3}]$ -acetate incorporation into citrate and downstream TCA intermediates, suggesting that mitochondrial OAA becomes limiting without SLC25A34. These data are consistent with the observed decrease in CS activity and may reflect substrate feedback regulation.

Therefore, we propose a working model in which SLC25A34 imports cytosolic OAA into mitochondria, supporting a shuttle that couples citrate and N-acetylaspartate export to cytosolic acetyl-CoA production. This would supply substrate for DNL and re-esterification while sustaining mitochondrial OAA for continued TCA cycle activity. SLC25A34 also appears to help maintain an anti-inflammatory state in brown adipocytes, because its loss triggers an early, sustained NF- $\kappa$ B-driven response followed by the induction of interferon-stimulated genes, although the mechanisms linking SLC25A34 to protection from mitochondrial stress and inflammation remain unknown. SLC25A34-mediated lipid cycling is also sensitive to systemic nutritional and hormonal cues: Fasting and a ketogenic diet, which favor lipolysis and $\beta$ -oxidation, and insulin, which favors DNL, all increase Slc25a34 expression in BAT. These data are consistent with a role for SLC25A34 in both catabolic and anabolic lipid metabolism, with the balance between these arms likely prioritized by biological context.

A

B mRNA expression in human primary adipocytes

Seahorse respirometry of primary human brown adipocytes

Donor 2 Donor 3

c

Correlations between SLC25A34 abundance in adipose depot and parameters of metabolic health

Fig. 7. SLC25A34 shapes human adipocyte respiration and is associated with cardiometabolic health. (A and B) Basal and cAMP-inducible OCR of primary human brown adipocytes from four donors with SLC25A34 KD (A) and expression of thermogenesis-related genes (B). Each data point in (B) represents one of the four donors. db-cAMP, dibutyryl-cAMP. (C) Meta-analysis of 24 studies showing correlations between SLC25A34 expression in subcutaneous WAT (scWAT) and omental WAT (omWAT) and metabolic health parameters in human patients. Shown are the ratio of stimulated versus basal lipolysis (iso/basal), circulating high-density lipoprotein (circ-HDL), circulating C-reactive protein (circ-CRP), circulating triglycerides (circ-TG), circulating low-density lipoprotein (circ-LDL), HOMA-IR, BMI, circulating insulin (circ-insulin), waist circumference (waist), and WHR. For all panels, data are represented as mean ± SEM. \*P < 0.05, \*\*P < 0.01, \*\*\*P < 0.001, for siControl db-cAMP versus siSLC25A34 db-cAMP, repeated-measures two-way ANOVA with the Geisser-Greenhouse correction with Šidák multiple-comparisons test (A) or one-way ANOVA with Tukey's multiple-comparisons test (B).

Members of the SLC25 family often transport multiple substrates across the IMM (87). In our work, the most significant metabolite change observed upon acute Slc25a34 depletion was increased total N-acetylaspartate, raising the possibility that it is a substrate of SLC25A34, with its accumulation reflecting the loss of its counterexchange with OAA and entrapment within mitochondria. SLC25A34 may also transport, or be part of a circuit for, other related dicarboxylates such as malate, consistent with the substrate selectivity of its closest deorphanized homolog, the yeast OAA carrier OAC1 (48). Testing direct transport will ultimately require in vitro assays with recombinant protein in reconstituted liposomes, which may be challenging given the low stability of both OAA and SLC25A34.

Notably, both SLC25A34 and its paralog SLC25A35 share common ancestry with the yeast OAA carrier gene OAC1 (28) and may have retained shared transport capacity while acquiring divergent regulation:

Slc25a35 is highly expressed at baseline in BAT, suppressed by cold, and not circadian regulated, suggesting a role in routine maintenance, whereas Slc25a34 is cold induced and subject to strong circadian and dietary control, consistent with a role under specific energetic demands. Recent work showed that SLC25A35 mediates mitochondrial phosphoenolpyruvate (PEP) efflux to support glycerolipid synthesis $88$ , suggesting that both homologous transporters contribute to lipid metabolism through distinct mechanisms. Although both can partially compensate for Oac1 in yeast $28$ , only Slc25a34 depletion disrupted respiration in brown adipocytes in our studies, indicating that the two carriers have become specialized for different cellular contexts within the mammalian mitochondrial carrier network.

SLC25A34 also influences the transcription of nuclear-encoded genes linked to lipid metabolism and mitochondrial respiration. Whether this occurs directly through cytosolic acetyl-CoA as a histone acetyltransferase substrate or through a lipid-derived signaling molecule remains unknown, although motif enrichment implicating PPARα and PPARγ in SLC25A34 target genes favors the latter. Such a mechanism would connect circadian, dietary, and temperature-sensing pathways to coordinate cytosolic lipid synthesis and mitochondrial lipid oxidation. Although we focused on adipose thermogenesis, REVERBα/β and PPARα/γ, the primary regulators of Slc25a34, have key roles in other tissues (85, 89, 90), suggesting that SLC25A34's impact likely extends beyond adipocytes. For example, mitochondrial OAA transport has been observed in heart, kidney, and liver (91–96), with the heart having the highest Slc25a34 expression. Furthermore, N-acetylaspartate, the metabolite most affected by Slc25a34 loss in BAT, is also the brain's second most abundant metabolite and supports myelination and neuronal integrity (97), raising the possibility of a role in neuronal health. If and to what extent the metabolic signature of SLC25A34 in thermogenic adipocytes is conserved across these functionally distinct cell and tissue types remains to be determined.

---

Materials and methods Animal models

The WT (C57Bl/6JBomTac) mouse studies [tissue panel (Fig. 1B and fig. S1A)], ketogenic diet study (Fig. 2, N and O), cold acclimation study (Fig. 1, C to E, and fig. S1B), and primary adipocyte culture from mice were performed with approved protocols from The Danish Animal Experiments Inspectorate (permit number: 2018-15-0201-01441) and the University of Copenhagen (project numbers P18-312 and P19-374). Mice were housed in an enriched environment with ad libitum access to a standard chow diet (Altromin, 1310) and water. Light in the facility was set to a 12 hour light/dark cycle (light: 6:00 a.m. to 6:00 p.m. and dark: 6:00 p.m. to 6:00 a.m.) at $22 \pm 2^{\circ}$ C. Studies were performed in male and female mice between 8 and 20 weeks of age. Mice were housed at RT ( $22 \pm 2^{\circ}$ C) unless otherwise stated. Housing temperatures used in the studies: thermoneutrality, $29^{\circ}$ C; RT, $22^{\circ} \pm 2^{\circ}$ C; and cold, $4^{\circ}$ to $5^{\circ}$ C.

Details of the housing conditions and studies performed on Rev-erba WKO, Rev-erba/ $\beta$ BKO, Hdac3 AKO, Ppara AKO, Ppara iBKO, and Slc25a34 iBKO mice are specified below.

Tissue panel: The tissue panel experiment was performed as described in (98). Briefly, 20-week-old C57Bl/6JBomTac (Taconic Biosciences) mice were acclimatized to thermoneutrality for 3 weeks before the 24 hours cold challenge at 4°C. Tissues were harvested at the end of cold challenge between 9:00 a.m. [Zeitgeber time 3 (ZT3)] and 1:00 p.m. (ZT7).

Cold acclimation study: This study is published in (24). For cold exposure experiments, 12- to 16-week-old mice were individually housed and placed into climate-controlled rodent incubators (Memmert HPP750Life) set to 29°C to acclimate to thermoneutrality for 2 weeks. Cold exposure mice were then moved to an incubator set to 4°C and kept there until dissection. Tissues were harvested between 9:00 a.m. (ZT3) and 1:00 p.m. (ZT7). BAT was used for proteomics analysis and gene expression analysis with quantitative polymerase chain reaction (qPCR).

High-fat diet and ketogenic diet study (Fig. 20): Adult (8 weeks) male C57BL/6J mice were purchased from Janvier Breeding Centre (France), and either placed on a standard chow diet (Brogaarden, Denmark) or a 60% high-fat diet (HFD) (Research Diets) for 24 weeks. The mice were group-housed with up to 8 mice per cage at $22^{\circ}$ C on a 12:12 hour light-dark cycle. The mice had free access to food and water. Before the ketogenic diet study all mice were singly housed with 2 weeks for acclimatization. For the consecutive 3 weeks the mice from the chow diet either continued receiving chow diet (5 mice) or were given a ketogenic diet (BioServ F3666); fresh ketogenic diet was given to the mice every second day. HFD group (6 mice) continued receiving HFD for the 3 weeks. All mice were fed ad libitum and therefore the food intake was not isocaloric. After 27 weeks since the beginning of the diet intervention the mice were sacrificed, and BAT was used to measure gene expression.

Ketogenic diet study (Fig.2P): Adult (14-weeks old) C57Bl/6N male mice acclimated to $22^{\circ}$ C and fed with standard chow diet (Brogaarden, Denmark) were singly housed for 2 weeks before the experiment. For 3 weeks, the mice were given either a chow diet (4 mice) or a ketogenic diet from BioServ (F3666, NJ, USA) (4 mice) ad libitum. Fresh ketogenic diet was given to the mice every second day. The food intake between the two groups was not isocaloric. After 3 weeks of diet intervention, the mice were sacrificed and tissues harvested.

Fasting study (Fig. 2Q and fig. S2, L and M): For the fasting study, 2- to 16-week-old mice were individually housed at RT. Mice were housed on a 12:12-hours light-dark cycle (lights on at 6:00 a.m., lights off at 6:00 p.m.). For the fasting, the food was removed from the fasting group at 9:00 a.m. (ZT 3) and both fasted and control groups were sacrificed and tissues harvested in 24 hours at 9:00 a.m. (ZT 10).

Rev-erba WKO: Studies with the Rev-erba whole body KO mice (Rev-erba WKO) were performed with an approved protocol from the University of Pennsylvania Perelman School of Medicine Institutional Animal Care and Use Committee as described in (9). The Rev-erba WKO mice were obtained from B. Vennström and backcrossed seven or more generations with C57Bl/6 mice. Mice were housed on a 12:12-hours light-dark cycle (lights on at 7:00 a.m., lights off at 7:00 p.m.). Gene expression and protein analysis were performed on 12- to 16-week-old male Rev-erba WKO mice and WT littermates.

For the circadian gene expression experiment (Fig. 1F) mice were singly housed at $29^{\circ}$ C before the experiment. Tissues were harvested at ZT-1, ZT3, ZT7, ZT11, ZT15, ZT19, ZT23, and ZT27 (2 to 4 animals per group per time point). For the proteomics study (Fig. 1G), mice were acclimated to thermoneutrality at $29^{\circ}$ C, and tissues were harvested at 5:00 p.m. (ZT 10) (table S2). The cold exposure study (Fig. 1H), was performed in climate-controlled rodent incubators. All WT and Rev-erba WKO mice used in the studies were first placed in individual cages with access to food and water and allowed to acclimate to $29^{\circ}$ C for 2 weeks before cold challenge. For the cold challenge mice were exposed to cold at $4^{\circ}$ C at 11:00 a.m. (ZT 4) for 1.5 hours and tissues were harvested at ZT5.5.

Rev-erba/β BKO: All animal work was approved by the University of Pennsylvania Perelman School of Medicine Institutional Animal Care and Use Committee (IACUC protocol number 804747 issued to Mitchell Lazar). Male Rev-erba $^{fl/fl}$ /Rev-erbβ $^{fl/fl}$ (99) mice were maintained on a C57Bl6/J background (Jackson Labs, Stock 008661). To produce Rev-erba/β BKO, Rev-erbα $^{fl/fl}$ /Rev-erbβ $^{fl/fl}$ mice were bred to Ucp1-Cre mice maintained on a C57BL/6J background (Jackson Labs Technologies, Inc., B6.FVB-Tg[Ucp1-Cre]1Evdr/J, Stock 024670) (23). Mice were bred and maintained at 20° to 22°C on a 12:12 hours light:dark cycle with ad libitum access to food and water unless otherwise noted. The normal chow diet (NCD, LabDiet, 5010) consisted of 12.7% kcal fat, 28.7% kcal protein, and 58.5% kcal carbohydrates. Gene expression was performed on 11–16-week-old male Rev-erba/β BKO mice and WT littermates. Cold exposure experiments were performed in climate-controlled rodent incubators set to 29° and 4°C. Mice from the thermoneutrality acclimated group (5 WT and 5 KO mice) were housed at 29°C for 2 weeks and the tissues were harvested at 5:00 p.m. (ZT10). Mice from the cold challenge group (5 WT and 6 KO mice) were singly housed at 29°C for 4 days and acutely exposed to 4°C cold in prechilled cages at 4°C with bedding and free access to standard chow, without a cotton nestlet or access to water, and the cage lid partly open. Cold challenge lasted for 6 hours from 11:00 a.m. (ZT 4) to 5:00 p.m. (ZT 10). Tissues were harvested at 5:00 p.m. (ZT10) and BAT was used for RNA sequencing (table S3).

For the circadian study, mice were maintained under thermoneutral conditions (29°C) for 14 days before tissue collection. To assess circadian gene expression, BAT was harvested at six time points, 4 hours apart over 24 hours (ZT2, ZT6, ZT10, ZT14, ZT18, and ZT22) from n = 3 mice per group per genotype. BAT RNA was used for RNA sequencing. The accession number for the data is GEO no. GSE331054.

Hdac3 AKO: Studies with Hdac3 AKO mice were performed according to ethical regulations and protocols approved by the Institutional Animal Care and Use Committee (IACUC) of the Perelman School of Medicine at the University of Pennsylvania and the Children's Hospital of Philadelphia as described in (31). Mice were group housed with enrichment in a temperature and humidity controlled, specific-pathogen free animal facility at 22°C under a 12:12-hour light-dark cycle with free access to standard chow (LabDiet, 5010) and water. To generate the mouse strain with the adipose-specific Hdac3 KO (Hdac3 AKO), $Hdac3^{f/f}$ mice maintained on a C57BL/6 background were bred to B6;FVB-Tg(Adipoq-Cre)1Evdr/J mice from the Jackson Laboratory (stock no.: 010803) (31). All experiments used male age-matched littermates group housed at 4–5 animals per cage. Gene expression studies were performed on 10- to 12-week-old male mice at ZT 10. Cold-exposure experiments were performed in climate-controlled rodent incubators (Powers Scientific) maintained at 22°C or 4° to 5°C, with interior temperature monitoring by digital thermometers. For the cold exposure experiment, mice were allowed to acclimate to 22°C for 2 weeks. Cold exposure mice were placed in pre-chilled cages at 4° to 5°C with bedding, a cotton nestlet, free access to standard chow and water, and the cage lid partly open at 2:00 p.m. (ZT7). Mice were sacrificed and tissues harvested after 3 hours of cold exposure 5:00 p.m. (ZT10). Animal numbers were 4 WT and 3 KO in the RT group, 5 WT and 6 KO in the cold group).

Pparα AKO (exon 5 targeted): All experiments with Pparα AKO mice with targeted exon 5 were performed according to procedures approved by the University of Pennsylvania's Institutional Animal Care and Use Committee as described in (34). Male WT C57BL/6J (stock 000664) mice, as well as F1 intercrossed progeny with 129S1/SvImJ (B6129SF1/J, stock 101043), were purchased from Jackson Laboratory. Ppara $^{flox/flox}$ mice were kindly provided by F.J. Gonzales (National Institutes of Health, Bethesda) and have been described previously (100). To generate Pparα AKO mice, we crossed Adiponectin-Cre mice with Ppara $^{flox/flox}$ mice. All mice were housed, 5 per cage, in a temperature-controlled, specific pathogen-free facility with a 12-hour light / 12-hour dark cycle and ad libitum access to water and food at RT (22 ± 2°C). 1Twelve-week-old male mice were sacrificed between ZT3 and ZT4, and BAT was used for RNA sequencing (3 WT and 3 KO mice).

Pparα AKO (exon 4 targeted): All experimental procedures and protocols for Pparα AKO (exon 4) mice complied with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and were approved by the Institutional Animal Care and Use Committee of the University of Mississippi Medical Center. The Pparα AKO (exon 4) mice were previously generated and described in (37). In short, homozygous $Ppara^{flox/flox}$ (from Dr. Walter Wahli, first published in (101)) and heterozygous Adipoq-Cre mice were crossed to create Pparα AKO (exon 4) mice and $Ppara^{flox/flox}$ littermate controls. Male and female mice were housed under standard conditions at RT with full access to water until 8 weeks of age. Mice were then switched to a normal chow (NCD) consisting of a 17% fat diet (Teklad 22/5 rodent diet, #860, Harland Laboratories, Inc., Indianapolis, IN, USA) for 30 weeks and had access to food and water ad libitum. After 30 weeks, mice were euthanized by overdose of isoflurane anesthesia at ZT3 to ZT5, and tissues were immediately collected and frozen in liquid nitrogen. Tissue samples were stored at -80°C until use.

Pparα iBKO: The study with Pparα iBKO mice were approved by the Danish Animal Experiment Inspectorate (approval#2018-15-0201-01459) and complied with the ARRIVE guidelines. Male mice of 7-8 weeks old were housed at 22°C, under a 12:12-hours light-dark cycle, with free access to food and drinking water. To generate mice with brown adipose-specific Pparα KO (Pparα iBKO), Pparα $^{f/f}$ mice (kindly provided by F.J. Gonzales, National Institutes of Health, Bethesda) (100) were crossed with the tamoxifen-inducible B6-Tg(Ucp1-cre/ERT2)426Biat mouse line (kindly provided by Prof. Christian Wolfrum) (102). For the cold exposure experiment, RT acclimated mice were given 2 mg of tamoxifen (Sigma-Aldrich T5648) in 100 μl of corn oil (Sigma-Aldrich C8267) per mouse by oral gavage once per day for 2 consecutive days. Then the mice were split into 2 groups and transferred to 29° or 4°C for 3 days, after which the tissue was harvested. One more tamoxifen gavage was given to all mice on the first day of the temperature challenge. Control mice were Pparα $^{f/f}$ without Ucp1-Cre/ERT2 and were given an identical dosage of tamoxifen. Mice of all groups were sacrificed between ZT3-ZT5.

Slc25a34 iBKO: The study with Slc25a34 iBKO mice was performed with approved protocols from The Danish Animal Experiments Inspectorate (permit number: A2025-15-0202-00236) and the University of Copenhagen (project number: P25-131).

Conditional Slc25a34 KO mice (Slc25a34 $^{f/f}$ ) on the C57BL/6N genetic background were generated by Genoway (France) by flanking exon 1, which contains the start codon, with loxP sites. To generate mice with brown adipose-specific Slc25a34 KO (Slc25a34 iBKO), Slc25a34 $^{f/f}$ mice were crossed with the tamoxifen-inducible B6-Tg(Ucp1-cre/ERT2)426Biat mouse line (kindly provided by Christian Wolfrum) (102). To induce KO in this model, 2 mg of tamoxifen (Sigma-Aldrich T5648) in 100 $\mu$ l of corn oil (Sigma-Aldrich C8267) per mouse was delivered by oral gavage once per day for 3 consecutive days. Control mice were Slc25a34 $^{f/f}$ without Ucp1-Cre/ERT2 and were given an identical dosage of tamoxifen.

In vivo thermogenesis test: The NE challenge presented in Fig. 3, G and I, was performed on 14- to 18-week-old Slc25a34 iBKO mice and littermate control female mice (8+8) maintained on a chow diet. Mice were acclimated to RT. The experiment was performed 2 weeks after tamoxifen treatment to induce the KO. Food was removed from the cages 2 hours before the NE test. NE response was assessed in anesthetized mice (intraperitoneal injection of 75 mg/kg BW pentobarbital with 7,5 mg/kg lidocaine) after an intraperitoneal injection of 1 mg/kg BW NE (Sigma-Aldrich, A9512). Oxygen consumption and carbon dioxide production were recorded using the Promethion core system (Sable Systems International) with data processing using OneClickMacro and Macrointerpretersetup. The NE challenge was performed at 33°C, and mice were sacrificed after the study. NE was injected at 10:30 a.m. (ZT 4.5) and mice were sacrificed at 12:30 p.m. (ZT6.5).

---

Cell models

Immortalized brown adipocyte (BMC) culture: Murine brown preadipocytes immortalized with SV40 large T antigen (referred to as BMC) were kindly provided by Patrick Seale (103). Preadipocytes were cultured in basal DMEM (Gibco, 31966) containing 10% FBS (Sigma-Aldrich, F7524) and 1% penicillin/streptomycin. Cells were passaged when they reached 60-70% of confluence. Media was changed every 2nd day. On the day of 100% confluency, cell differentiation was induced by supplementing of standard media with the differentiation cocktail (insulin (20 nM) (Sigma-Aldrich, I9278), dexamethasone (1 mM) (Sigma-Aldrich, D1756), rosiglitazone (0.5 mM) (Cayman Chemicals, 71740), T3 (1 nM) (Sigma-Aldrich, T6397), and 3-isobutyl-1-methylxanthine (IBMX) (250 μM) (Sigma-Aldrich, I5879)). On day 2 of differentiation, media was switched to maintenance media containing insulin and T3. Maintenance media was changed every 2nd day. Cells were harvested or assayed on day 7 of differentiation unless stated otherwise in the figure legends. The cells were maintained at 37°C in a humidified atmosphere with 5% CO2.

Generation of an HA-SLC25A34 overexpressing stable cell line: HA-SLC25A34 overexpressing BMCs were generated by retroviral transduction (pMMLV; VectorBuilder). Retrovirus was generated using the Plat-E packaging cell line $104$ . Cells were transfected with 10 $\mu$ g of DNA using Lipofectamine 2000 (Invitrogen, #52887) after the manufacturer's instructions. Media was changed after 12-16 hours, and after 48 hours, media was collected and filtered using a 0.45 $\mu$ m filter. 1 ml of the viral supernatant was added drop wise to a 10cm dish containing 5 ml of DMEM with 10% FBS and 4 $\mu$ g/ $\mu$ l polybrene. Media was changed after 24 hours, and transduced cells were selected with media containing 2 $\mu$ g/ml puromycin.

Primary murine brown adipocyte culture: Primary brown preadipocytes were isolated from WT C57BL/6J male and female mice as described in (98). Male and female tissues were pooled for the generation of primary cell culture. BATs were harvested from 6-8 weeks old mice and put into ice-cold DMEM (Gibco, 31966) media. Tissues were minced and subsequently digested with DMEM (Gibco, 31966) supplemented with 2% BSA (Sigma-Aldrich, A7030) and 0.2% collagenase type I (Worthington Biochemical Corp., LS004197) at 37°C for 40 min. The digests were centrifuged (5 min, 400 g) to obtain the stromal vascular fraction (SVF) and pellets were resuspended in DMEM with 10% FBS (Sigma-Aldrich, F7524) and 1% penicillin/streptomycin. The cell suspensions were passed through a 70 μm strainer and distributed onto the plate-format of interest. On day 1 after cell harvest, the isolated preadipocytes were rinsed with DMEM containing 10% FBS and 1% penicillin/streptomycin and thereafter, media was changed every 2 $^{nd}$ day. The cells were passaged up to one time before differentiation. On the day of 100% confluency, cell differentiation was induced by supplementing of standard media with the differentiation cocktail (insulin (86 nM) (Sigma-Aldrich, I9278), dexamethasone (0.1 μM) (Sigma-Aldrich, D4902), rosiglitazone (0.5 mM) (Cayman Chemicals, 71740), T3 (1 nM) (Sigma-Aldrich, T6397), and 3-isobutyl-1-methylxanthine (IBMX) (250 μM) (Sigma-Aldrich, I5879)). On day 2 of differentiation, media was switched to maintenance media containing insulin and T3. Maintenance media was changed every second day. Cells were harvested or assayed on day 7 of differentiation. The cells were maintained at 37°C in a humidified atmosphere with 5% CO2.

Primary human brown adipocyte cell culture: Deep cervical adipose tissue collection was approved by the Medical Research Council of Hungary (20571-2/2017/EKU) followed by the EU Member States' Directive 2004/23/EC on presumed consent practice for tissue collection. All experiments were performed in accordance with the guidelines of the Helsinki Declaration. Written informed consent was obtained from all participants before the thyroid surgery. Patients with known diabetes, malignant tumor, or with abnormal thyroid hormone levels at the time of surgery were excluded. SVF was isolated from each biopsy by digestion in collagenase from Clostridium histolyticum (120 U/ml) (Sigma-Aldrich, C0773), filtering (pore size 140 $\mu$ m), and centrifugation at 330 g for 7 min. Pellets were resuspended in Dulbecco's modified Eagle's medium/nutrient F-12 Ham (DMEM-F12, Sigma-Aldrich, T2252) medium supplemented with 10% FBS (Thermo Fisher Scientific, A5256701SA), penicillin/streptomycin (100 U/ml) (Roche, 11074440001), biotin (33 $\mu$ M) (Sigma-Aldrich, B4639), and pantothenic acid (17 $\mu$ M) (Sigma-Aldrich, P5155). After 24 hours, nonadherent cells were removed by washing with PBS, and adherent cells were cultured further until reaching 90% confluency. The absence of Mycoplasma was verified using a PCR Mycoplasma Test Kit I/C (Promocell, catalog no. PK-CA91).

Primary human brown adipocyte differentiation from SVF was initiated for 3 days with DMEM/F-12 HAM medium containing penicillin/streptomycin (100 U/ml), biotin (33 $\mu$ M), pantothenic acid (17 $\mu$ M), human apo-transferrin (10 $\mu$ g/ml) (Sigma-Aldrich, T2252), 20 nM human insulin (Sigma-Aldrich, I9278), 100 nM hydrocortisone (Sigma-Aldrich, H0888), 0.2 nM T3 (Sigma-Aldrich, T6397), rosiglitazone (2 $\mu$ M) (Cayman Chemicals, 71740), dexamethasone (25 nM) (Sigma-Aldrich, D1756), and 500 $\mu$ M IBMX (Sigma-Aldrich, I5879). After the initial 3-day induction period, rosiglitazone, dexamethasone, and IBMX were omitted, and then the media was changed every 2 $^{nd}$ day. The cells were maintained at 37°C in a humidified atmosphere with 5% CO $_{2}$ (105).

siRNA mediated gene KD in murine brown adipocytes: Primary and immortalized murine brown preadipocytes were differentiated as stated above. On day 3 of differentiation, cells were reverse transfected according to (106). In brief, siRNA targeting the gene of interest or relevant control siRNA were diluted in Opti-MEM (Gibco, 51985) to a final concentration of $300\mathrm{nM}$ . RNAiMAX (Invitrogen, 13778-150) was added to a final concentration of $30~\mu \mathrm{l / ml}$ . The siRNA mix was added to the bottom of the wells in the plate-format of interest and allowed to incubate at RT for 30 min. In the meantime, cells were trypsinized, counted, and resuspended in culture media. Finally, the cell suspension was distributed on top of the siRNA mix to a final siRNA concentration of 50 nM. Media was changed 2 days after transfection and cells were harvested or assayed on day 7 of the differentiation unless otherwise is stated in the figure legends. siRNA constructs used in the study are listed in table S14.

siRNA-mediated knock-down of SLC25A34 in human brown adipocytes: Transfection and gene silencing experiments were performed using DharmaFECT1 transfection reagent (Horizon Discovery, T-2001) on day 12 of differentiation by following the manufacturer's instructions. Mature brown adipocytes were incubated with a mixture of DharmaFECT1 and SLC25A34-targeting siRNA (100 nM) (Dharmacon ON-TARGETplus SMARTpool Human SLC25A34 siRNA, Horizon Discovery, L-032041-02) or nontargeting (scrambled) siRNA (100 nM) (Dharmacon plus nontargeting control pool, Horizon Discovery, D-001810-10) for 48 hours (qPCR) or 96 hours (respirometry) at 37°C in a humidified atmosphere with 5% CO₂. After the incubation time, adipocytes were activated with dibutyrylcAMP (db-cAMP, 500 μM) (Sigma-Aldrich, D0260) for 8 hours in DMEM/F-12 HAM medium supplemented with penicillin/streptomycin (100 U/ml), biotin (33 μM), and pantothenic acid (17 μM) (107).

Acetate rescue study: For the acetate rescue study, immortalized brown adipocytes were grown on 10 cm cell culture dishes. On the day of confluency, the culture medium was exchanged to differentiation medium for the control dish and to differentiation media supplemented with 10 mM sodium acetate for the acetate-treated dish. The differentiation protocol was standard for both groups. On day 3 of differentiation, siRNA transfection was performed as described above. The acetate-treated group always received culture media freshly supplemented with 10 mM sodium acetate during transfection and regular media changes. Seahorse respirometry was performed on day 7 in seahorse media containing standard respiratory substrates, without sodium acetate supplementation. Cells were harvested on day 7 for CS activity assay and mtDNA/nDNA quantification. Experiment was repeated 5 times.

Insulin treatment study: BMC brown adipocytes were differentiated according to the protocol described above. On day 6 of differentiation, insulin was omitted from the medium for 24 hours. On day 7, cells were treated with 2 nM or 20 nM insulin or left untreated as controls. Cells were harvested after 4 and 24 hours of treatment. The experiment was repeated once.

---

In vitro assays

Immunofluorescence assay: Cells were reseeded on day 3 of differentiation onto Matrigel-coated glass coverslips. On day 10, cells were serum-starved in DMEM containing 0.2% BSA for 2 hours and stimulated with 1 nM CL-316,243 for 1 hour. Cells were washed with PBS and fixed with 4% paraformaldehyde for 15 min, followed by quenching with 50 $\mu$ M glycine. Cells were blocked and permeabilised in 5% normal swine serum (NSS) containing 0.1% saponin in PBS for 30 min, before 1 hour incubation with primary antibodies (TOMM20, ProteinTech, #11802-1-AP; HA, BioLegend, #901515) diluted in 2% NSS containing 0.1% saponin for 1 hour at RT. After PBS washes, cells were incubated with appropriate secondary antibodies for 1 hour at RT. Coverslips were washed in PBS and distilled water before mounting using ProLong Gold antifade mounting medium (Invitrogen, #P36930). Images were acquired on a Leica SP8 laser scanning confocal microscope with a 63×, NA 1.4 Plan-Apochromat oil immersion objective. Acquired images were median de-noised using Leica Application Suite software (v3.5.7.2). Fluorescence intensity across indicated ROIs were plotted as line profiles.

Subcellular fractionation for isolation of mitochondria from adipocytes: Mitochondria were isolated from cultured brown adipocytes following the differential centrifugation protocol described by Frezza et al. (108). Cells were washed with ice-cold PBS and scraped in ice-cold homogenisation buffer supplemented with protease inhibitors. Cells were pelleted by centrifugation at 600g for 10 min at 4°C and resuspended in homogenisation buffer. Cell disruption was performed using a glass teflon douncer (18 strokes). The homogenate was centrifuged at 600g for 10 min twice at 4°C to remove unlysed cells and debris. The resulting supernatant was centrifuged at 7000g for 10 min at 4°C to pellet mitochondria. A post-mitochondrial supernatant was collected. The mitochondrial pellet was washed with isolation buffer and re-centrifuged at 7000g for 10 min at 4°C before final resuspension in 100-150 μl buffer. Protein concentration was determined using BCA assay followed by immunoblotting of the indicated proteins. Antibodies used for immunoblotting: PMP70 (Abcam, #ab3421); calnexin, CXN (Abcam, #ab22595); total OXPHOS rodent WB antibody cocktail, OXPHOS (Abcam, ab110413); vinculin, VCL (Cell Signaling Technology, #13901).

Cycloheximide chase assay: Cells were reseeded into 24-well plates on day 3 of differentiation. Cells were pretreated with 10 nM CL-316,243 for 24 hours before the start of the experiment. On day 10, protein synthesis was inhibited by addition of cycloheximide (50 $\mu$ g/ml), and cells were harvested at 0, 3, 6, 9, and 24 hours after treatment. Cells were washed twice with ice-cold PBS and lysed in RIPA buffer (50 mM Tris pH 7.5, 150 mM NaCl, 1% Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS, 1 mM EDTA) supplemented with protease and phosphatase inhibitors (cOmplete Mini, Roche; PhosSTOP, Roche). Lysates were sonicated and centrifuged at 16,000g for 20 min at 4°C. Protein concentration was determined using BCA assay.

Immunoblotting: (fig. S1, B and C) Samples were denatured in Laemmli buffer containing TCEP and heated at 65°C for 10 min. Samples were resolved by SDS-PAGE on 4-20% gradient gels (Bio-Rad) and transferred to nitrocellulose membranes using a Trans-Blot Turbo system (Bio-Rad). Membranes were blocked using 5% milk in TBS-T (0.1% Tween-20) for 1 hour at RT and incubated with primary antibodies in 5% BSA TBS-T overnight at 4°C. Samples were analyzed by immunoblotting using antibodies against HA (BioLegend, #901515, 1:2000), SLC25A1 (ProteinTech, #15235-1-AP, 1:1000), UCP1 (ProteinTech, #83870-1-RR, 1:1000) and TOMM20 (ProteinTech, #11802-1-AP, 1:1000). GAPDH (Cell Signaling Technology, #97166S) was used as a loading control. After washing, membranes were incubated with HRP- or Alexa488-conjugated secondary antibodies (1:2500) in 5% milk in TBS-T for 1 hour at RT. Signal was detected using enhanced chemiluminescence (Thermo Fisher) or fluorescence on a ChemiDoc system (Bio-Rad). (figs. S4, C and D, and S7, F to H)

Cells were lysed in ice-cold RIPA buffer containing 0.1% SDS, protease inhibitors (Merck, #11836153001). Total protein concentration was measured using the Pierce Bicinchoninic Acid (BCA) Assay Kit (Thermo Scientific, #23225), with BSA as the standard. Proteins were separated by SDS-PAGE and transferred to a polyvinylidene fluoride (PVDF) membrane. For total protein quantification, the membranes were stained with Revert 700 Total Protein Stain for Western Blot Normalization (Li-Cor, #926-11011) and developed using Odyssey system (700 channel). Then the membranes were destained with Tris-buffered saline containing 0.1% Tween 20 (TBS-T) until clear. Membranes were blocked with 5% BSA or nonfat dry milk in TBS-T and incubated with primary antibodies, diluted in TBS-T with 5% nonfat dry milk, overnight at 4°C. Afterward, horseradish peroxidase-conjugated secondary antibodies were applied. Protein bands were detected after chemiluminescent substrate (Cytiva, RPN2232) addition using an Odyssey XF Imaging System. Protein band intensities were quantified with ImageStudio Lite Version 5.2 software. Specific signals with subtracted background signals for each blot were normalized to the housekeeping protein histone H3 (for S4C, S4D, S7F, S7H-right) or to total protein intensity for the protein pan-acetylation blot (S7G, S7H-left). Relative protein abundances were calculated by dividing the measured signals by their respective normalization factors, enabling relative quantitative comparisons. Antibodies used in the assay: SLC25A34 (Sigma-Aldrich, #HPA069806), H3 (Cell Signaling Technology, #9715), H3K27Ac (Cell Signaling Technology, #4353), pan Ac-Lys (Cell Signaling Technology, #9441).

DNA extraction and qPCR for mtDNA/nDNA quantification: DNA was extracted from cultured murine brown adipocytes (BMC) using AllPrep DNA/RNA/Protein Mini Kit (QIAGEN, 80204) according to the manufacturer's protocol. DNA concentration as measured with a NanoDrop spectrophotometer (Thermo Fisher Scientific). Gene expression was determined based on real-time qPCR using SYBR green (PP00259, Primerdesign) using LightCycler 480II (Roche) according to the manufacturer's protocol. The data was analyzed with the $\Delta \Delta$ CT method. mtDNA was quantified by the expression of Nd1, and nDNA by the expression of Hk2. All primers are listed in table S13.

DNL assay in cultured mouse brown adipocytes: Radiolabeling with $[^{14}C]$ -glucose: Murine brown adipocytes were cultured and siRNA mediated KD was performed as described above in 12 well plates. On day 7 of differentiation, culture media was exchanged to 0.5 ml per well of standard DMEM media supplemented with T3 and Insulin, containing ${}^{14}C$ -labeled glucose (D- $[{}^{14}C(U)]$ Revvity Health Sciences Inc NEC042V250UC, Lot 3106342; 250 $\mu$ Ci (9.25 MBq); 265 mCi/mmol (9.805 GBq/mmol); 1.25 ml ethanol:water (3:97); 0.5 $\mu$ Ci per 0.6 ml of media. A few wells were left untreated as a negative control. Cells were cultivated in the presence of ${}^{14}C$ -labeled glucose for 6 hours. After 6 hours, cells were rinsed twice with ice-cold PBS and frozen on plates at -70°C until subsequent analysis. Lipid extraction and scintillation counting: Lipid extraction was performed as described in (109) with modifications. 0.4 ml of $H_{2}O$ was added to each well and cells were harvested using a rubber policeman into 15 ml conical Falcon tubes. 0.5 ml chloroform and 1.0 ml methanol were added to each sample, followed by vortexing. Next, 0.5 ml chloroform and 0.5 ml of 0.2 N HCl were added, and samples were vortexed again. Phase separation was achieved by centrifugation at 2500 rpm for 5 min at RT. The lower (organic) phase was transferred to a clean 15 ml tube and washed once with a pre-prepared “blank upper phase” (generated by mixing 1 ml chloroform, 1.0 ml methanol, 0.5 ml of 0.2 N HCl, and 0.4 ml water per sample). After a second centrifugation, the cleaned lower phase was transferred to a 2 ml glass vial and organic solvents were evaporated to dryness under nitrogen flow. Dried lipids were resuspended in 75 $\mu$ l chloroform:methanol (2:1, v/v), and 60 $\mu$ l was transferred to scintillation vials containing 5 ml scintillation fluid (Ultima Gold, Revvity Health Sciences Inc 6013329). Scintillation was counted using Hidex 600 SLe Automatic TDCR Liquid Scintillation Counter.

CS activity assay: The mitochondrial function of cultured brown adipocytes (BMC) was estimated by measuring the activity of CS in the cellular extracts. The assay is based on the protocol from $(110)$ and was performed as described in $(111)$ with modifications. Briefly, the cells were grown in 6 well plates (1000000 cells per well) and harvested by adding 400 $\mu$ l of the lysis buffer (50 mmol/l Hepes, 137 mmol/l NaCl, 10 mmol/l Na4P2O7, 20 mmol/l NaF, 5 mmol/l EDTA, 1 mmol/l MgCl2, 1 mmol/l CaCl2, 2 mmol/l Na3VO4, 5 mmol/l nicotinamide, 10 $\mu$ mol/l trichostatin A, Halt protease inhibitor cocktail (Thermo Scientific, Waltham, MA, USA), 1% (v/v) NP-40, 10% (v/v) glycerol, demineralized water). The extracts were mixed with a reaction mix containing 0.1 m Tris-HCl, pH 8.1, 0.4 mm acetyl-CoA sodium salt (Sigma-Aldrich) and 1 $\mu$ m 5,5'-dithiobis(2-nitrobenzoic acid) (Sigma-Aldrich), and absorbance at 412 nm was measured every 30 s for 5 min (basal slope). Oxaloacetic acid at 10 mm was used as the reaction substrate, and the change in absorbance after its addition to the mix was measured every 30 s for another 5 min (reaction slope). CS activity was then calculated from the delta of the reaction and basal slopes and presented in $\mu$ mol/min/g protein. Protein content was measured in a separate set of samples produced in the same experiment using Bradford assay (Quick Start Bradford 1x Dye Reagent #5000205m Bio-Rad) according to manufacturer protocol.

---