Dear Scott Gordon Kenneth MacLeod,
I hope you all got some good rest last night and enjoyed the last day of ARDD 2026!
ARDD 2026 — Day Three Recap
Aging Research and Drug Discovery Meeting · Saturday, October 3, 2026 · David Rubenstein Treehouse, Harvard University, Cambridge, MA
Chair Opening Introduction
09:00–09:15 · Canopy Hall
Chairs: Alex Zhavoronkov, Vadim Gladyshev, Morten Scheibye-Knudsen, Daniela Bakula and Jesse Poganik. (Not captured in the notes.)
Morning Talks
09:15–10:15 · Canopy Hall
The 09:15–09:35 slot is listed on the agenda as "Speaker TBC"; no notes match it (see the cross-check section at the end).
Vadim Gladyshev — Professor, Harvard Medical School
09:35–09:55 · Measuring, targeting and understanding aging (no title on the agenda; topic from the notes)
Key takeaway: Aging should be modeled as systemic biological states using multimodal, multispecies, longitudinal data; actionable biomarkers such as transcriptomic clocks can find interventions that lower biological age even while the underlying trajectory continues.
• Three intervention tiers: physiological remodeling for modest lifespan gains (near term); time-targeted, organ-specific interventions before disease; radical extension by breaking systemic barriers (e.g., coordinated tissue replacement).
• In silico-to-in vivo pipeline: ~4,000 compounds screened computationally, profiled in cells, then tested in mice; of 10 advanced, 5 extended lifespan, and short 5-day dosing improved frailty (Anastasia Shindyapina's work, under review).
• Transcriptomic clocks predict age and time-to-death and replicate across mouse, rat, macaque and human; TACO is a public tool that runs them across datasets with cross-platform normalization.
• Multi-agent mining of GEO-scale data: most perturbations raise biological age; rare rejuvenating hits were validated, including a compound that lowered biological age and frailty over 3 months in old mice.
• Heterochronic heart transplantation: a young heart in an old host ages faster, an old heart in a young host rejuvenates; human donor data agree, supporting wider use of older donor hearts. One transplanted organ barely changed systemic age.
• ~50 clocks for organs, cell types and organelles; models now span up to eight modalities (transcriptomics, proteomics, metabolomics, MRI, DEXA, retinal imaging, EHR, plus voice, photo and video).
• Menopause time-to-event clocks show a sharp transition around age 50, with no analogous pattern in men.
• Q&A: cross-platform bias persists but shrinks with larger datasets; immune infiltration is a fair concern for the heart data, but single-cell imaging suggests the heart cells themselves rejuvenate.
Mark Cobbold — SVP IO Discovery & Cell Therapy Oncology, AstraZeneca
09:55–10:15 · Genetic strategies to prevent T cell senescence for scalable allogeneic CAR T (topic from the notes)
Key takeaway: Knocking out CDKN2A, CDKN2B and MTAP together prevents replicative senescence in T and NK cells, potentially enabling off-the-shelf CAR T at massive scale from a single donor.
• Autologous CAR T has treated ~80–90k patients but is too infrastructure-heavy for common autoimmune diseases (lupus, RA, MS) now in CAR T trials.
• Borrowing from tumor biology: genomics of 21 T-cell lines and lentiviral screens in donor T cells pointed to the chromosome 9p locus (CDKN2A/B, MTAP), often co-deleted in T-cell malignancies.
• Triple knockout in CD8, CD4, γδ T and NK cells turned limited proliferators into cells expanding up to ~10^20-fold; lower β-gal, higher Ki-67; all three knockouts together performed best.
• Combined with allogeneic safety edits (HLA and TCR knockouts) in a "universal" cell platform.
• Safety: no accumulation of deleterious mutations across runs; in mice, edited cells did not persist without cytokines and were rejected after weeks.
• HER2 (solid tumor) and BCMA (myeloma) CARs controlled patient-derived xenografts as well as standard CAR T; ~10^9-fold expansion could treat thousands per batch, ~10^20-fold potentially millions.
• Q&A: iPSC-derived T cells are hard to turn into authentic cytotoxic T cells, so gene editing was chosen over Yamanaka reprogramming; every T-cell subset tested responded to the edits.
Search & Evaluation Strategies for Longevity Therapeutics
10:15–11:00 · Canopy Hall · Panel
Lisa Melton (moderator) — Senior News Editor, Nature Biotechnology
Key takeaway: A "longevity" indication does not exist yet; the practical path is to win approval in a specific age-related disease and use it as a foothold to gather evidence of broader effects on aging.
• Panelists: Thomas Marron (Eli Lilly), Jason Sanders (GSK), Tala Fakhouri (Parexel) and Zachary Cary (Takeda).
• Tala Fakhouri:
– Regulators are more engaged and open to real-world evidence for confirmatory purposes; aging endpoints need pre-competitive consortia.
– AI can help build new biomarkers, ingest multimodal data and run digital-twin simulations, since 60-year studies are not feasible.
– The science can take a systems view, but regulation still requires a single approvable indication first.
• Jason Sanders:
– Useful biomarkers must predict dynamic clinical outcomes (like the ASCVD risk score), not chronological age.
– Even systemic approaches need specificity (senescent-cell clearance acts body-wide but targets one cell type); claims of working "all over the body" are a red flag.
– Understanding healthy aging means deeply phenotyping adults aged ~40–90, not studying centenarians; the field lacks a strong model of "normal aging" and access to living human tissue is a bottleneck.
• Thomas Marron: regulators and payers will get comfortable with new biomarkers gradually, from data collected alongside trials (as in IPF); for senolytics the question is still which indication can be approved and reimbursed.
• Zachary Cary: unvalidated biomarkers add risk to an already expensive process; a drug must meet a clear unmet need to be reimbursed.
• Target-based vs. systems biology is "and, not or"; advice to startups was to follow human biology (e.g., telomere disorders hit specific organs such as marrow and lung).
Panel: Metabolic Health and Longevity by Nestlé Research
11:30–12:10 · Canopy Hall · Panel
Philipp Gut (moderator) — Adult Health Lead and Research Lead Healthy Longevity, Nestlé Research
Key takeaway: GLP-1 drugs are transforming metabolic health, but nutrition, resistance exercise and individualized digital measurement are needed to protect lean mass and personalize response.
• Panelists: Ryan Carvalho (Head of Nestlé Research), Calum MacRae (Mass General Brigham), Giulia Menichetti (Harvard Medical School) and Brianna Stubbs (Buck Institute).
• Calum MacRae:
– GLP-1 benefits reach beyond weight (cardiac, renal, inflammatory) and some appear before weight loss; one patient reported a 98% improvement in arthritis within a week.
– Responses vary widely (lean vs. fat loss), so mechanism must be understood at the individual level; phones, wearables and even purchases can quantify health reserves.
• Ryan Carvalho:
– Nutrient needs persist or rise on GLP-1s; protein and amino-acid advice should be stratified (e.g., older adults with sarcopenia vs. postmenopausal women).
– Million Veteran Program data (2026, Lancet family journal): healthy diet and lifestyle further reduce cardiovascular events during GLP-1 weight loss.
– "Nutritional dark matter": food compounds missing from databases, some modulating GLP-1 pathways via bitter and xenobiotic receptors; microbiome function matters more than diversity.
– With ~1 dietitian per 4,000 people in the U.S., AI tools that infer diet from blood metabolomics are needed.
• Brianna Stubbs: resistance training attenuates GLP-1 lean-mass loss; an ARPA-H ProSPR-funded project (with Mike Snyder and David Furman) is building at-home intrinsic capacity measures to catch declines as early as the mid-40s, framing intrinsic capacity as a "vital sign."
• Giulia Menichetti asked whether food labels could one day describe effects on molecular pathways; Carvalho said the data gap between model foods and branded products, plus the food "matrix effect," makes that hard.
Semaglutide and Healthspan: Decoding the Emerging Evidence
12:10–12:30 · Canopy Hall
Alejandro Aguayo-Orozco — Senior Scientific Director Diabetes Insights, Novo Nordisk
"From clinical outcomes to organ-specific biological age"
Key takeaway: Across major trials, semaglutide lowered proteomic organ-age scores, and the later effects are only partly explained by weight loss.
• Outcome data from ~32,000 patients (SELECT, STEP-HFpEF, FLOW, SOUL): 20% MACE reduction in non-diabetic SELECT; 24% kidney composite reduction in FLOW (stopped early for efficacy); 18% lower all-cause mortality in a pooled six-trial analysis.
• Proteomics from five trials: lower heart and kidney age scores from the earliest timepoint, broadening to multiple organs over time.
• Organ-age reductions were not mediated by weight, HbA1c or hsCRP; the heart-age reduction mediated up to 40% of the reduction in cardiovascular death.
• Early adipose signature paradoxically looks "older" during rapid weight loss and up-titration (leptin, FABP4), then turns neutral or better; a proteomic frailty score was lower from one year onward.
• No clear dose dependence across doses and formulations (including oral); no data on microdosing; EVOKE (Alzheimer's) lean-cohort data not yet analyzed; glucose homeostasis as a mediator under study.
• Sensitivity analyses show most clock proteins move coherently, not a few heavily weighted ones.
Panel: Aging Research in Pharma
13:30–14:10 · Canopy Hall · Panel
Sri Devi Narasimhan (moderator) — Deputy Editor, Cell Press
Key takeaway: Geroscience needs shared definitions, foundational discovery, functional trial endpoints and close academia–industry collaboration to overcome biological heterogeneity.
• Panelists: Sophie Lemire-Brachat (Novartis), Maria Dermit (Novo Nordisk) and William Mair (Harvard).
• Definitions: Dermit separates "aging" (biology) from "health" (patient-centric); Lemire-Brachat frames aging as functional decline across systems and calls for aligned data ontologies; Mair distinguishes gerontology from geroscience, which treats aging as a modifiable risk factor.
• William Mair:
– Discovery science not tied to profit is indispensable (GLP-1 traced to venom proteins, CRISPR to bacterial defense).
– Preclinical results fail to translate because of heterogeneity (microbiome, environment, sex, strain) and incentives that reward discovery over validation.
– Even mTOR/rapamycin is poorly understood; chronic inhibition may "break the system" rather than mimic youth. Psychosocial factors robustly affect aging biology. There is no single "golden window" for intervening.
• Sophie Lemire-Brachat: human genetics and ultra-rare protective variants will reveal new mechanisms; multimodal data enables digital twins; objective endpoints need work (e.g., refining the six-minute walk test).
• Maria Dermit: longevity claims need mortality, functional and validated clock evidence; AI speeds preclinical work but needs human oversight; personalization means matching indication, population and modality.
• Consensus: Phase 2 success requires hard functional outcomes, surrogates alone are insufficient, and non-competitive consortia should be encouraged.
Therapeutic Assets & Licensing (Late Stage)
14:10–14:50 · Canopy Hall · Panel
Cailin McGurk (moderator) — Managing Director, Morgan Stanley
Key takeaway: Large longevity deals will follow translational confidence and regulatory clarity; until then, upfront payments stay modest.
• Panelists: Mark Cobbold (AstraZeneca) and Phil McGurk (SVP, Head of Global Business Development, Takeda).
• Mark Cobbold:
– Early in vivo CAR T data showing robust expansion without lymphodepletion "rewrote the rules" and drove a rapid acquisition; internal champions mobilize organizations.
– Expects genomic and epigenomic medicines to matter more than small molecules for aging; called aging the "mother of all diseases" and predicted a GLP-1-style frenzy for assets once the science matures.
– Thinks China is likely to lead, given scale, sustained investment and long-term planning; prevention does not break pharma's model (the vaccine precedent).
• Phil McGurk:
– Allocate each risk to the party best placed to manage it; contrasted M&A (acquirer absorbs risk) with licensing (shared risk, milestones, royalties).
– Best deals are championed by credible R&D leaders, not BD alone or top-down mandates.
– Cited Takeda's orexin agonist program as a long academia–industry effort (described as the first orexin agonist approval after more than 20 years of research, starting in narcolepsy type 1 with broader indications possible).
• Cailin McGurk: pharma partners in a "barbell," very early or very late; big upfronts go to validated biology (e.g., KRAS molecular glues, oral GLP-1s); for longevity, translation confidence is the main barrier.
• Q&A: pharma–pharma pre-competitive consortia exist; results are mixed but they produce useful side insights.
In Vivo Validation Across the Target Landscape
14:50–15:10 · Canopy Hall
Martin Borch Jensen — Co-Founder & CSO, Gordian Biotechnology
Key takeaway: Mosaic in vivo screens with single-cell readouts map each target's effects across organs, de-risking target selection before committing to a drug.
• AI lacks ground-truth biology; Gordian's aim is to scale in vivo validation.
• Mosaic screens: low-dose AAV libraries up- or down-regulate single genes in scattered cells within an intact animal; single-cell transcriptomics compares perturbed cells with internal controls, then overlays human disease and healthy signatures.
• Osteoarthritis: screens in horses with spontaneous OA (hundreds of targets per animal); validated in human tissue co-cultures and aged, obese mice; program now in IND-enabling studies.
• Weight loss: one target gave ~10% weight loss alone; with semaglutide it deepened loss and prevented rebound for at least two months after semaglutide stopped; now in small-molecule lead optimization. Mechanism (increased lipolysis) confirmed in human adipocytes.
• Cross-organ safety: the same target inhibited glycolysis in cardiomyocytes, flagging possible fibrosis risk and informing systemic vs. targeted delivery.
• Q&A: robots for in vivo handling likely within ~3 years; a 10x jump in research capacity possible in the early 2030s.
Reversing Metabolic Decline Through AMPK Network Activation in Humans
15:10–15:30 · Canopy Hall
James Peyer — Cambrian Bio (listed on the agenda as "CB, Cambrian Bio")
Key takeaway: ATX-304 was presented as the first AMPK activator to translate into humans, raising metabolic rate and improving cardiometabolic markers with placebo-like safety.
• Earlier AMPK approaches failed: metformin (complex I stress), Merck's MK-8722 (cardiac glycogen buildup), DNP-type uncouplers (metabolic collapse), AICAR (poor pharmacokinetics).
• ATX-304, from a phenotypic screen in UmeÃ¥, raises both glucose uptake and mitochondrial demand without lowering ATP or building glycogen; drives mitochondrial biogenesis (believed ANT1-dependent).
• Mice: in aged lean animals, ~30% higher cardiac output and more than double treadmill endurance after 8 weeks; in obese mice, ~20% weight loss in a month without eating less or losing lean mass; 30% in 11 days with semaglutide; continued loss after semaglutide stopped despite eating 30% more.
• Phase 1: 23 subjects (2:1), 8 weeks plus 8-week open-label at 400 mg/day; placebo-like safety, no overheating; higher resting metabolic rate, large adiponectin rise, lower triglycerides, liver fat and visceral fat.
• Up to six trials planned for 2027 (monotherapy, with semaglutide, fatty liver), then intrinsic capacity and broader metabolic decline.
• Q&A: almost no blood-brain barrier penetration; no human weight-loss data yet.
Therapeutic Assets & Licensing (Early Stage)
16:00–16:40 · Canopy Hall · Panel
Brianna Stubbs (moderator) — Research Assistant Professor & Director of Translational Science, Buck Institute
Key takeaway: For pharma, "proof of concept" is not one milestone but a body of evidence that an asset can be best-in-class against a specific, high-value target product profile.
• Panelists: Jonathan Levi (AstraZeneca), Claris Munksgaard (EMD Serono), Nikolaj Kulahin Roed (Novo Nordisk), David Berry (Averin Capital) and Avi Spier (Novartis).
• Jonathan Levi: "validated" must be specific; a hit from a transcriptomic clock is not enough without a clear line of sight to an indication (AstraZeneca's "5R" framework). Study the partner's pipeline and propose concrete combination plays.
• Avi Spier: proof of concept means showing an asset can beat the competition and become the new standard of care.
• David Berry:
– VCs need 10x+ returns; a preclinical partnership is a ~5-year "wedding," so trust and cultural fit matter.
– IP is paramount, pharma will exploit any holes; bring both cutting-edge human data and the conventional mouse data conservative reviewers still expect.
– Proposed a shared, AI-enabled data room for dynamic asset comparisons, since each static diligence can cost $1–3M.
• Nikolaj Kulahin Roed: the "totality of data" (human genetics, animal data, safety) decides; package longevity assets inside established indications and flag cross-aging signals (e.g., pain, hearing loss) without spreading too thin.
• Claris Munksgaard: rigor (gold-standard models, no cherry-picking), clean PK, safety and target engagement, clear differentiation and a credible path to IND and label; indications must be commercially compatible (a high-price rare disease and a low-price broad indication do not add up). AI tools such as AlphaSense speed commercial assessment, but AI-integrated data rooms remain aspirational.
What Can Be Done to the Aging Brain: Why Live Long When the Brain Is Gone?
16:40–17:00 · Canopy Hall
Johan Luthman — EVP R&D, Lundbeck
Key takeaway: Amyloid clearance and behavioral-symptom drugs are progress, but disease modification, prevention and biomarkers for other proteinopathies remain major unmet needs.
• The brain ages on a largely fixed set of neurons; cortical loss and microbleeds accelerate around 60–65; no drugs target the "frail brain," and sleep is a key modifiable factor.
• Anti-amyloid antibodies clear amyloid but slow progression only ~25–35%, a benefit patients do not readily feel; prevention trials in biomarker-positive, asymptomatic people are under way.
• Blood biomarkers are transforming Alzheimer's diagnosis, reducing reliance on PET and CSF.
• A program with Otsuka produced the first approved drug for agitation in Alzheimer's.
• Biomarkers are lacking for tau and α-synuclein; tau antibodies disappointed while antisense shows promise; antisense works strikingly in SOD1 ALS; an anti-α-synuclein antibody is in Phase 3 for multiple system atrophy, chosen for its relatively pure pathology and fast progression.
• Advocated Bayesian adaptive trial designs; credited Oskar Fischer and Solomon Carter Fuller as co-discoverers of Alzheimer's pathology.
Preserving and Restoring the Thymus for Healthspan Extension
17:00–17:20 · Canopy Hall
Francisco Leon — Tolerance Bio (listed on the agenda as "TB, Tolerance Bio")
Key takeaway: Thymic involution is a key driver of immune aging; Tolerance Bio aims to preserve the thymus with anti-IL-33 and regenerate it with IL-7.
• A 2023 Harvard study found more cancer and autoimmunity after adult thymectomy.
• Residual thymic function correlates with survival in older people and with epigenetic clocks (GrimAge); centenarians often keep good thymic function, and naked mole rats keep their thymus; an AI model scores thymus health from CT scans.
• Preservation: an anti-IL-33 antibody for people with a functioning thymus. Regeneration: IL-7 for those with little left.
• In HIV (a model of accelerated immune aging), IL-7 raises naive T cells, recent thymic emigrants and TCR diversity; restoring CD4 counts in HIV is a near-term approvable indication, run in parallel with aging trials.
• In Alzheimer's mouse models, boosting thymic function improved cognition ~50% by helping T cells clear plaques; exploring iPSC-derived thymus and LNP delivery.
Removing Oxidized Cholesterol
17:20–17:40 · Canopy Hall
Matthew O'Connor — CEO, Cyclarity Therapeutics
"Clinical Trials for a New Approach to Cardiometabolic and Age-Related Disease"
Key takeaway: An engineered cyclodextrin dimer selectively clears oxidized cholesterol and reverses foam cells; Phase 1 showed excellent safety and dose-dependent target engagement.
• Hypothesis: 7-ketocholesterol accumulates with age and turns macrophages into plaque-building foam cells.
• The drug is a cyclodextrin dimer (>1,000 Da but regulated as a small molecule), designed with computational chemistry and AI to prefer oxidized over native cholesterol; the complex is cleared by the kidneys.
• Preclinically, it reverses the foam-cell phenotype and restores reverse cholesterol transport.
• Phase 1a (SAD/MAD, 72 healthy volunteers): "extraordinarily safe," ~3-hour half-life, dose-dependent rise in urinary oxidized cholesterol.
• Phase 1b is dosing coronary artery disease patients with CT angiography; an international Phase 2 is planned. A 1% plaque reduction may mean ~20% fewer cardiovascular events; the goal is 5–10%.
• Unpublished: partial rescue of fatty liver in rodents, doubled with another approved drug; a fatty-liver Phase 1b is under consideration.
ARTAN Bio: One Codon Therapy for Multiple Aging Diseases
17:40–18:00 · Canopy Hall
Brian Bodemann — CEO, ARTAN Bio
Key takeaway: One engineered suppressor tRNA could restore full-length protein at tens of thousands of age-associated CGA-to-stop mutation sites, treating a whole class of somatic mutations at once.
• Deamination of methylated cytosine turns CGA (arginine) into TGA (stop): ~68,000 susceptible sites in ~16,000 genes, ~30% of somatic premature stops, reaching ~1 in 9 cells by age 80; UV accelerates it.
• A suppressor tRNA inserts arginine at the premature UGA; tRNAs need no tissue-specific promoters.
• AAV9 test-bed program reached all six tested mouse organs without significant toxicity and restored full-length p53 in Calu-6 cells.
• Lentiviral HSC program: up to ~90% transduction of primary human HSCs; aiming for IND-enabling studies in 2027 in clonal hematopoiesis (~240,000 identifiable U.S. patients; drivers enriched 3.7x after chemotherapy).
• A topical LNP for sun-exposed skin is planned as a cosmetic product.
• Early science funded through VitaDAO's VitaRNA token (~$1M raised from the community).
Talk: David Sinclair
18:00–18:20 · Canopy Hall
David Sinclair — Professor, Harvard Medical School
NAD boosting and epigenetic restoration (no title on the agenda; topic from the notes)
Key takeaway: Two fronts: newly described NAD-consuming structures in aging cells, and OSK gene therapy plus the chemical cocktail SL100 reversing aging markers in animals.
• Information theory of aging: DNA damage redeploys epigenetic modifiers, eroding cell identity; cells retain a backup of their youthful state.
• NAD: the lab describes a new structure in aging cells (transcribed as "inflammasomes") that consumes NAD via PARP14 and drives inflammation; inhibiting PARP14 is a candidate way to raise NAD.
• OSK restored vision in old mice and primates; first results from the ER-100 trial in glaucoma and NAION were expected around October 9, 2026.
• Unpublished: OSK slowed and killed several cancer cell lines by re-differentiating them; xenograft tumors shrank or grew more slowly.
• SL100, from a screen described as 80 billion molecules, preserved neurite length in ALS models and, in a progeroid mouse model, extended mean lifespan 25% and maximum lifespan 33%.
• Q&A: SL100's downstream pathways overlap little with OSK's; the lab studies ECM glycation; pairing a senolytic with an NAD booster is logical but timing is unknown; a forthcoming paper shows chronic oral NMN extended lifespan in female mice.
Talk: Morten Scheibye-Knudsen
18:20–18:40 · Canopy Hall
Morten Scheibye-Knudsen — Professor, University of Copenhagen
Quantifying and targeting aging phenotypes driven by DNA damage (topic from the notes)
Key takeaway: Phenotypic biomarkers from photos and cell morphology predict hard outcomes, and compounds that trigger DNA-damage signaling without damage reversed senescence and rescued Alzheimer's-model mice.
• Facial photos predict survival better than some clinical measures; a morphology classifier identifies senescent cells in skin and blood; very low skin senescence predicted higher future cancer risk.
• People living at higher altitude in Ethiopia had fewer circulating senescent cells.
• Single-cell data suggest most somatic mutations arise during development, so point mutations may not drive degenerative aging outside cancer.
• DNA-damage signaling depletes NAD (countered by PARP1 inhibition or NR); in a COPD pilot, NR lowered lung inflammatory cytokines.
• A screen for compounds that stimulate the damage response without damage found "Drug F," which reversed senescence markers and improved motor function in flies; a spinout is advancing it.
• In 5xFAD and APP/PS1 mice, Drug F rescued learning and motor deficits and reversed weight loss; clinical trials may follow.
Closing Remarks
18:40–19:00 · Canopy Hall
Not captured in the notes. Followed by open-bar networking from 19:00.
Pet and Animal Longevity Forum — Session I: Canine Aging
09:00–10:30 · Cedar Grove · Parallel track
Opening Remarks & Welcome (09:00–09:10): Max Unfried, Scientific Director, The Thalion Initiative. (Not captured in the notes.)
Dan Promislow — Senior Scientist and Scientific Advisor, HNRCA Tufts University
09:10–09:30 · "The Companion Dog as a Translational Model of Aging"
Key takeaway: The Dog Aging Project bridges lab models and humans, and its findings translate: dog mortality markers predicted human mortality from far less follow-up.
• More than 55,000 dogs enrolled since 2020; a Precision Cohort of ~1,000 dogs gives annual biospecimens; data shared via the Broad's Terra.
• NIA renewed funding for five more years; a new cohort of ~1,000 of the oldest U.S. dogs will get 30x genomes; the Precision Cohort could grow to 3,000 and add transcriptome, proteome and lipidome with more support.
• Dogs with lower epigenetic age live longer; genes hypomethylated with age in dogs tend to rise in expression in aging humans.
• 25 metabolites (enriched for kidney markers) predict mortality in dogs and in the same direction in humans, found with 2.6 years of follow-up vs. 8–25 in human studies.
• GWAS linked a chromosome 25 SNP to kynurenic acid; coat "furnishing" genes (RSPO2, FGF5) also shift cystathionine, a cysteine precursor.
• TRIAD rapamycin trial: target 580 dogs (≥7 years, >40 lb), over half enrolled, one year of drug and three years of follow-up, neurology and cardiology arms; safety excellent, efficacy in ~4 years.
• Q&A: U.S. dogs are mostly sterilized, so intact-dog studies may suit France or Germany; owners skew white, wealthy and educated; no experimental drugs in owners' pets without lab-dog safety data.
Steve Horvath — Professor, UCLA
09:30–09:50 · "Surrogate endpoints for veterinary geroprotector trials: the case for methylation clocks"
Key takeaway: DNA methylation clocks are the strongest candidates for surrogate endpoints in veterinary longevity trials, which could speed conditional approval of geroprotectors for dogs.
• Lifespan endpoints are impractical (the 14-year Purina Labrador calorie-restriction study); FDA's Center for Veterinary Medicine allows conditional approval on a "reasonable expectation of effectiveness."
• A surrogate must predict survival and mediate the treatment effect (like LDL); methylation clocks are accurate, stable, mortality-predictive, responsive (GLP-1, calorie restriction) and conserved.
• The Mammalian Methylation Array measures ~36,000 conserved CpGs; >15,000 samples from 348 species are public; dual human–dog clocks estimate relative age across species.
• In the Purina study, clocks detected the benefit of calorie restriction seven years before the survival readout, more strongly in females.
• A clock built on 742 dogs from 93 breeds was accurate but showed no breed differences in aging rate, perhaps because the array targets ultra-conserved sites.
• Castration lowers epigenetic age in sheep; a single-CpG "androgen clock" tracks cumulative androgen exposure.
• Q&A: the field needs "causal clocks" (a Mendelian-randomization-style approach); CpGs that predict a species' maximum lifespan do not overlap with those that track age; a turtle clock paper was days from publication.
Vicky Yang — Associate Professor, Tufts University
09:50–10:10 · "Mitral valve prolapse: studying aging-associated cardiac changes in the dog model"
Key takeaway: Canine myxomatous mitral valve disease is driven by a macrophage–fibroblast feedback loop, offering targets that translate to human mitral valve prolapse.
• Common in small breeds: murmurs at 3–5 years, heart failure at 8–10, survival ~1 year after; surgery is costly and rare for dogs.
• Valve interstitial cells become myofibroblasts; the disease is polygenic in dogs and humans.
• Pro-fibrotic M2 (CD206+) macrophages increase, with TGF-β-linked signaling and higher pERK.
• Extracellular vesicles carry miR-145 (from diseased valves) and miR-125b (from M2 macrophages), both activating fibroblasts.
• Targets: M2 macrophages, EV signaling, MEK–ERK (current MEK inhibitors are too toxic).
• Q&A: mechanical stress likely recruits monocytes; the disease is distinct from checkpoint-inhibitor myocarditis.
Philipp Gut — Adult Health Lead, Nestlé Research
10:10–10:30 · "Nutrition to slow biological aging in dogs"
Key takeaway: Phenotypic and epigenetic clocks respond to diet and predict outcomes in dogs, making them practical endpoints for longevity nutrition trials.
• Used biobanked samples from Purina's 14-year calorie-restriction study.
• A dog PhenoAge from routine chemistry predicts mortality and flags accelerated aging even within normal ranges; lean-fed dogs were biologically younger at mid-life.
• Golden Retriever Lifetime Study (Morris Animal Foundation): more time overweight meant older biological age and more osteoarthritis; in an unpublished trial, a weight-loss diet normalized biological age within six months.
• An EM-seq epigenetic clock tracks age and mortality, classifies breeds, detected calorie restriction earlier than PhenoAge, and was already elevated in dogs that later developed osteoarthritis; healthy "super seniors" were younger.
• Moving from blood to buccal swabs; effects were significant in both sexes, stronger in females.
• Urged industry to agree on "good enough" standard clocks for trials, with next-generation clocks as exploratory endpoints.
Pet and Animal Longevity Forum — Session II: Commercializing Pet Longevity
10:45–12:20 · Cedar Grove · Parallel track
Anna Skaya & Garri Zmudze — General Partners, AniVC
10:45–11:05 · Fireside chat: "From Shark Tank to a 9-Figure Exit: Building, Selling, and Backing Animal Longevity"
Key takeaway: Pet "humanization" is driving demand, but early-stage animal health remains underfunded; AniVC backs strong founders with human-grade products and therapeutics.
• Skaya founded Basepaws (pet genomics, 2016), appeared on Shark Tank and sold it to Zoetis for "close to nine figures"; Zmudze was an early backer. She reinvested proceeds into AniVC.
• Theses: founders first; "pet parenting" over pet ownership; real longevity innovation (therapeutics, AI drug discovery) without overclaiming; the animal regulatory path is an advantage.
• Cancer needs diagnostics first; many clinics lack imaging, so human diagnostic tech could translate.
• Audience: Lou Hawthorne said pet owners consistently ranked "fix cancer" above cloning or general longevity.
• Practical: daily brushing remains the best dental prevention; no diabetes investments yet for lack of innovation.
Kevin Esch — EVP and President R&D, Zoetis
11:05–11:25 · "A Lifetime of Care for More Healthy Years"
Key takeaway: Pets are living longer, so veterinary medicine must shift to earlier detection and long-term management of chronic disease through genetics, diagnostics and therapeutics.
• Zoetis "Prevent, Detect, Treat" continuum, with genetics from the Basepaws acquisition; goals include predicting progression (e.g., feline CKD) and treatment response.
• Osteoarthritis pain affects ~40% of dogs and ~60% of cats (Librela, Solensia); longer-acting antibodies and later disease-modifying agents in development.
• CKD is underdiagnosed early; cancer is the leading cause of death in dogs; obesity needs holistic management.
• Anxiety drives pet surrenders; a non-sedative anxiolytic is planned for 2027.
• Markets are built by demonstrating superior care, as JAK inhibitors did in dermatology.
Graeme Bainbridge — Senior Principal Scientist, Dechra
11:25–11:45 · "Feline Longevity: How an Animal Pharma company thinks about chronic diseases and Quality of Life"
Key takeaway: For cats, the goal is healthspan: shifting from symptom management to disease-modifying treatment of chronic disease.
• More than 80 million pet cats in the U.S.; CKD affects 50% of cats over 10, hyperthyroidism 10% of seniors, and 60% of cats over 6 show osteoarthritis.
• Barriers: multimorbidity, subtle onset, symptom-focused drugs, few validated biomarkers vs. dogs, and difficulty getting cats to the vet.
• Monoclonal antibodies are promising; biological-age biomarkers could transform trials; oral inflammation is an overlooked contributor to CKD.
• Q&A: feline CKD looks different from human and canine disease (possibly inflammation- or infection-driven); finding progressors for trials is key.
Panel: The Pet Longevity Revolution
11:45–12:20 · Cedar Grove · Panel
Anna Skaya (moderator) — General Partner, AniVC
Key takeaway: Pet longevity is large and underserved, but vet time, trial costs and public awareness are the bottlenecks.
• Panelists: Dan Promislow (Tufts), Julie K. Spears (Nestlé Purina PetCare), Kevin Esch (Zoetis) and Alex Voda (CEO, Cat Health Company).
• Alex Voda: in an early CKD proof-of-concept, a 22-year-old cat's SDMA fell from ~24–25 to ~13–14 within weeks (preprint coming); a sarcopenia program is licensed to pharma; Matt Kaeberlein joined the advisory board. Criticized the pace of the field (no product on shelves ~7 years after Loyal's founding) and advised working directly with vets instead of CROs.
• Julie K. Spears: nutrition is foundational (Purina's lean-feeding lifespan study); connected devices such as smart collars and litter-box sensors detected micro-changes across feline CKD stages.
• Are vets ready for longevity? Esch said no (disease-focused training); Promislow said yes but overworked; 10–15 minute appointments are the barrier, and at-home diagnostics can help.
• With $100M: Promislow would scale the Dog Aging Project; Esch would fund early innovation; Spears would run more trials; Voda would cut trial costs.
• Promislow proposed a national search for "supercentenarian" pets to drive shared biomarker work; Voda noted the pet market is cash-pay, aligning company and owner incentives.
Virtual Cell in Time Forum — Session I
13:30–15:00 · Cedar Grove · Parallel track · Chair: Max Unfried, The Thalion Initiative
Opening Remarks & Welcome (13:30–13:40): not captured in the notes.
Omar Abudayyeh — Professor, Harvard Medical School
13:40–14:00 · "Building virtual models of biology and health"
Key takeaway: Population-level generative models plus biology-trained language-model agents can predict perturbations and have already produced validated rejuvenation hits.
• Many virtual-cell models fail to beat linear baselines and are judged only on mean expression, not distribution shape or cell-to-cell structure.
• A population-encoding model (transcribed as "OmniCell") decodes back to realistic single-cell counts and folds in bulk data (~10x more training data).
• New data on 731 transcription-factor perturbations in two primary cell types raised the top-20 DE-gene hit rate from ~53% to ~62%.
• Generic LLMs get ~2% of top-20 gene changes right; fine-tuning with reinforcement learning on biology QA reached ~20%, with interpretable reasoning and no sign of saturation with scale.
• With Google DeepMind's AI co-scientist and an in-house agent, genome-wide in silico screens for rejuvenating old fibroblasts validated ~2 of ~20 predictions (KDM2A, HSF1).
• A co-scientist-nominated epigenetic factor (unnamed) cut DNA breaks and senescence and rejuvenated cells by ~15–25 years across eight clocks without pluripotency.
• Next: "virtual humans" built from protein QTLs via Mendelian randomization.
Yuancheng (Ryan) Lu — K99 Research Fellow
14:00–14:20 · "Functional Genomics and Virtual Screening for Rejuvenation Factor Discovery"
Key takeaway: Functional genomics plus single-cell foundation models are finding rejuvenation factors safer than OSK that restore function in the eye and beyond.
• The eye suits this work: ~300M patients with AMD or glaucoma, efficient AAV delivery and non-invasive vision readouts.
• A downstream OSK factor that clears the lipid-peroxidation product 4-HNE (likely GSTA4; transcribed "TSG4") restored vision tracking in 13 of 14 mice; a small-molecule enhancer was found.
• An ORF screen (~13,000 ORFs in ~160M RPE cells under oxidative stress) yielded nine hits; the cardiac factor NKX2-5 beat OSK in two protection assays without changing cell identity.
• Long-term NKX2-5 was toxic; deleting its homeodomain kept protection with less toxicity, and 21-month-old mice regained vision and ERG comparable to 3-month-olds.
• Systemic NKX2-5 improved grip strength and frailty index (including kyphosis).
• A model trained on ~14M cells, using paired fibroblasts from the same donors ~20 years apart (Baltimore Longitudinal Study of Aging), gave ~2x enrichment of hits orthogonal to OSK.
Hao Pei — Founder & CEO, MobiDrop
14:20–14:40 · "Aging Is Multidimensional — So Must Be the Data: Co-registered Single-cell Readouts for Longevity Drug Discovery"
Key takeaway: Aging is dynamic and heterogeneous, so MobiDrop co-registers multiple readouts on the same single cells over time to generate AI-ready data.
• Current assays capture one dimension at one timepoint, averaged across millions of cells.
• The platform (transcribed as "NovaCell") stacks five layers on the same cells: live imaging, electrophysiology, Cell Painting, whole transcriptome and proteome panels (up to 6,000 proteins); epigenomics is next.
• Plugs into a lab's existing incubators, microscopes and sequencers.
• Uses: senolytic and inflammaging screens, an intermediate funnel before animal studies, and gene-therapy QC.
• Goal: virtual cells trained on in vitro perturbations linked to aged-mouse outcomes; preliminary models were not shown.
Kejun Ying — Postdoctoral Scholar, Stanford University School of Medicine
14:40–15:00 · "An open world model for human health and mortality"
Key takeaway: World models borrowed from video generation and protein language models can forecast individual disease trajectories and link protein structure to species lifespan.
• Treats health as a "movie," not a sentence: 11 data domains (EHR, labs, MRI, proteomics, genetics and more) merge into a joint health state evolving in continuous time.
• State-of-the-art prediction for more than 1,000 diseases up to 20 years out; predicts unseen diseases from a text description; simulated drug effects (e.g., statins) were checked against 286 real trials.
• Structural model of 2M proteins from 287 species found 170 proteins whose structural features track maximum lifespan, independent of expression; it predicts lifespan about as well as methylation clocks (Cape golden mole predicted at ~30 years).
• Starting his own lab in January 2027.
Virtual Cell in Time Forum — Session II
15:15–16:45 · Cedar Grove · Parallel track · Chair: Max Unfried, The Thalion Initiative
Le Song — Professor, Mohamed bin Zayed University of Artificial Intelligence
15:15–15:35 · "A World Model of the Virtual Cell"
Key takeaway: A true virtual cell should be multiscale, multimodal and stateful, and should drive generative design toward desired cell states.
• Current models handle one perturbation and one response in one cell type.
• Needed: multiscale simulation (DNA to cell), a persistent state for sequential perturbations, and "inverse design" of molecules for a target cell state rather than a binding target.
• Hybrid architecture: graph neural networks on pathway knowledge (Reactome, KEGG) plus transformers to learn what is missing.
• Multi-stage training on single modalities, then paired data; temporal live imaging and perturbation data are crucial.
• The field is at the "GPT-1 or AlphaFold 1" stage; GenBio AI released a model (transcribed "iCell") pitched as the first world model of a human cell.
Petrina Kamya — Global Head of AI Platforms, Insilico Medicine
15:35–15:55 · "Introducing Insilico Medicine's Virtual Aging Cell"
Key takeaway: The Virtual Aging Cell is a multi-agent system that simulates interventions across five biological scales, now undergoing prospective lab validation.
• Insilico has 10 clinical-stage programs (one in Phase 3), 18 IND-enabling studies and more than 47 internal programs.
• Frontier LLMs do well at single-step retrosynthesis but poorly at ab initio tasks; Insilico's Pharmaceutical Superintelligence framework trains and benchmarks models on 12 years of data.
• Lineage: 2016 embryonic-development AI (COX7A1) and the Precious1GPT–3GPT models.
• ~43 specialized agents per layer across molecular, intracellular, intercellular, tissue/organ and organism scales; a patient-seeded baseline graph is compared with the post-intervention graph.
• Q&A: prospective validation on molecules with unknown mechanism, in Insilico's robotic lab, is the only way to rule out literature leakage.
Daniel Ives — CEO & Founder, Shift Bioscience
15:55–16:15 · "A single-cell clock and perturbation-model reveals anti-fibrotic targets for epigenetic rejuvenation"
Key takeaway: A single-cell aging clock combined with virtual-cell perturbations finds single-gene knockdowns that both rejuvenate cells and reduce fibrosis.
• Clock trained on >300 aged dermal fibroblast lines with methylation and single-cell data (correlation >0.9).
• Virtual perturbations scored by the clock showed single genes dominating top combinations, and that knockdowns (not just OSK-style overexpression) can rejuvenate.
• In the lab, ~150 of ~190 targets lowered cellular age and ~40 raised it; mTOR inhibition was the positive control.
• Lead target in the TGF-β pathway: dose-dependent age reduction in fibroblasts (stronger in keratinocytes) and anti-fibrotic effects in liver, lung and heart fibroblasts; expression tracks fibrosis in MASH, IPF, heart failure and systemic sclerosis.
• Partnering on LNP/siRNA delivery; mouse fibrosis and lifespan studies under way.
• Q&A: acknowledged circularity risk in training on methylation clocks; mitigated with aging hallmarks outside the clock and OSK benchmarks.
Science Panel: Aging as a Stress Test of the Virtual Cell
16:15–16:45 · Cedar Grove · Panel
Julian Issa (moderator) — Host, The Beyond Tomorrow Podcast
"From Perturbation Response to Health and Mortality"
Key takeaway: Panelists agree virtual cells are worth building, but success depends on new longitudinal, perturbational data and on experimental validation that convinces skeptics.
• Panelists: Petrina Kamya (Insilico), Omar Abudayyeh (Harvard Medical School), Kejun Ying (Stanford), Le Song (MBZUAI) and Morgan Levine.
• Definitions converged on a dynamic, multimodal, multiscale, stateful simulator that predicts a cell's next state and supports "what-if" experiments.
• Missing data: longitudinal perturbations of aged cells or young cells in aged environments (Levine), longitudinal human proteomics, and donor heterogeneity (Kamya).
• Synthetic data split the panel: Abudayyeh was skeptical (hard to verify, amplifies noise); Song saw a role (AlphaFold 2 precedent) and called the challenge engineering and economics.
• Sharing: federated learning (Kamya) and pooled data such as Illumina's billion-cell project (Song).
• Convincing skeptics takes experimental proof, e.g., a correctly predicted novel mechanism; protein design earned credibility the same way. Levine: an accurate model can be inverted to design interventions.
• Asked which aging target they would edit first in a person, a panelist named APOE, GDF15 and Klotho as interesting but urged starting with a small molecule to establish safety.
Virtual Cell in Time Forum — Session III
17:00–18:00 · Cedar Grove · Parallel track · Chair: Dominika Wilczok, Qiming Venture Partners
Morgan Levine — Computational scientist and entrepreneur
17:00–17:20 · "Biology is Multiscale: Why Virtual Cells Are Not Enough"
Key takeaway: Generative models that reproduce noisy RNA-seq are not enough; biology needs models that learn a cell's latent state as the basis for reasoning and planning.
• Most virtual cells are in silico Perturb-seq, trained on narrow data, failing out of distribution, and seeing the cell through only the RNA "lens."
• Benchmarks are weak; a recent virtual cell challenge was "hackable."
• Proposes a general "perception layer" that integrates modalities into an internal representation (the "sculpture," not its "shadows"); this is the focus of her new startup (formerly at Altos Labs).
• Q&A: impose as little human prior knowledge as possible and let data drive the model.
Eric Sun — Assistant Professor, MIT
17:20–17:40 · "Measuring and Modeling Cell Aging"
Key takeaway: Single-cell and spatial aging clocks reveal which cells drive aging in their neighbors, guiding targeted interventions.
• A single biological age hides variation across cell types and organs.
• Spatial transcriptomic map of mouse brain across 20 ages (3–35 months) with spatial aging clocks.
• T cells are broadly pro-aging to nearby cells, even controlling for inflammation; neural stem cells are pro-rejuvenating. The lab is engineering cytokines against the T-cell effect.
• A spatial perturbation model (graph neural network) learns from natural microenvironment variation and distinguished causal sender–receiver signaling.
• The field should fill the "off-diagonal": tissue-level consequences of cell-level perturbations.
Ava Amini — Principal Researcher, Microsoft
17:40–18:00 · "Multimodal models to understand and target cell states"
Key takeaway: AI world models can prioritize which experiments to run; Microsoft's Quine outperformed human experts at nominating compounds that shift cancer cell states.
• Quine is a multiscale, multimodal world model that also draws on the literature; the goal is to guide experiments, not replace them.
• Pancreatic cancer: ranked ~8,000 FDA-approved compounds for shifting classical vs. basal states; the top 20 produced the largest shifts tested and beat an expert-chosen set.
• It predicted a third, novel cell state, and compounds predicted to induce it worked as expected.
• A fellowship program will give researchers access; spatial data integration is next.
Panel: Virtual Cells — A Venture Capital Perspective
18:00–18:40 · Cedar Grove · Panel
Dominika Wilczok (moderator) — Entrepreneur in Residence, Qiming Venture Partners
Key takeaway: Investors want focused, problem-specific uses of virtual cells built on high-fidelity human data and a de-risked path to an asset, not an "all-seeing eye" platform.
• Panelists: Anthony Philippakis (GV), Avak Kahvejian (Flagship Pioneering), Manish Sharma (Warburg Pincus) and Andrew Marshall (Haystack Sciences). Philippakis's remarks were not separately captured.
• Avak Kahvejian: much high-throughput data is "contrived" and may not translate; proprietary human-relevant data is needed for training and validation. Cited Cellarity (fetal hemoglobin for sickle cell, now in the clinic); 99% of investors still want a defined target.
• Andrew Marshall: start with a niche problem where data comes fast; a virtual cell may be a tool rather than a company.
• Manish Sharma: private equity needs a "whole product" (workflow, proprietary data, IP); institutional exposure to aging will be staggered.
• Discussion: causality over exhaustive mechanism; agreed benchmarks; near-term uses in diagnostics, cell-therapy manufacturing and engineered microbes; omics layers are correlated, so orthogonal data (new 3D microscopy, the "dark proteome") is needed; partnerships can bootstrap a pipeline with $20–50M rather than $1B, but unvalidated models will not sell.
Closing Remarks (18:40–19:00): Max Unfried, The Thalion Initiative. (Not captured in the notes.)
Bewerbungsschluss: 22. OKT 2026
That's absolutely wild - imagine zooming from your neighborhood straight into your DNA. This AI + biotech combo is exactly where breakthroughs happen. #HealthTech
Posts
