Our belief is simple: health is the most important compounding asset in your life, and for the first time, agentic AI makes it possible to manage it the way it deserves — continuously, deeply, personally. A new kind of relationship with your health. An intelligence that holds your complete biological context, life patterns, reasons across all of it, and grows with you over time. This is Compound.

What if we could simulate how health unfolds over a lifetime and explore what might change its course? HealthFlux takes a promising step in that direction. Congratulations on the work!

The search for longevity drugs may get an important lead from a lung fibrosis trial. Insilico’s rentosertib targets TNIK, a protein involved in fibrosis and several aging-related processes. In a new Nature Biotechnology analysis of 42 patients followed for 12 weeks, six proteomic aging clocks detected shifts toward younger predicted biological ages with treatment. One detail makes this particularly interesting for longevity. The dosing regimen that produced the greatest improvement in lung function was different from the one with the broadest agreement across aging clocks. That leaves open the possibility that the drug is affecting biology beyond the lung’s response to treatment. There is still plenty to resolve. Fibrosis-related proteins influence these clocks, and this study cannot establish broader aging benefits or effects in healthy people. But it gives longevity research a concrete question to pursue. If TNIK inhibition affects processes shared across age-related diseases, could its benefits eventually extend beyond the condition it was developed to treat? Zhavoronkov et al., Nature Biotechnology, 2026

Old blood makes young intestinal stem cells act old. This has been shown before through parabiosis, where two mice share a circulation. What wasn't clear was whether the aged environment was instructing the cells or just correlating with their decline. A new study in Nature Aging removed one receptor, TNFR1, from young mouse intestinal cells before joining them to old mice. Their stem cells stayed completely unaffected by the aged circulation. The same receptor knockout also explained why rapamycin worked in aged mice for reasons that had nothing to do with its famous mTOR pathway. It was acting as an anti-inflammatory. This reframes inflammaging from a vague descriptive term into a single addressable signal. The stem cells were never broken. They were listening to an instruction that arrived through one receptor, and removing that receptor removed the instruction entirely.🧬 nature.com/articles/s43587-0…

Endurance athletes carry high intramyocellular fat, and it doesn't hurt them. Their muscle cells oxidize it efficiently. This is well documented, and it has quietly trained an entire generation of health-conscious people to think all muscle fat is fine as long as you train hard. There is a second fat depot in muscle that does not follow this rule. It sits between the fiber bundles rather than inside them, and it behaves like an independent endocrine tissue. It secretes MCP-1, TNF-alpha and resistin directly into the surrounding muscle environment, disrupting insulin signaling regardless of how fit the person is or how much body fat they carry overall. Most fitness tracking never distinguishes between the two. A DEXA scan or a standard body composition report cannot tell you which kind of fat is in your muscle. The measurement gap is the story here, not the fat itself. sciencedirect.com/science/ar…

Most people treat fitness as the output of training. There's a result in Nature Aging from July that makes a stronger case for reading it the other way, as the input that decides what your next session is actually worth. Janssens and colleagues at Amsterdam UMC and Maastricht took muscle biopsies from older adults at different levels of physical function, ran them through one bout of submaximal exercise, and measured what moved. Everybody responded. But among the older participants, the magnitude of the transcriptional stress and immune response scaled with how fit they already were. Same hour of work, different amount of biological signal received. This is cross-sectional and the cohort skews old, so treat it as a correlation with a plausible mechanism rather than a settled causal story. But if it holds, the returns on training don't accumulate, they compound. Your first year buys you the machinery to extract more from your second. It also means an identical protocol handed to two people delivers two different doses, and almost nobody knows which dose they're on nature.com/articles/s43587-0…

60% of carotid plaques contain polyethylene or PVC (Marfella et al., NEJM 2024). Patients with plastic present had roughly 4.5x the risk of heart attack, stroke, or death. The association is observational, not causal. Cristina et al. (Nanomaterials 2026) map why it is biologically plausible: nanoplastics cross into mitochondria and nuclei, activate cGAS-STING, and accelerate cardiac senescence from inside the cell. The particles are physically there. The mechanism fits. What is missing is a clinical tool to detect or act on any of this at the individual level.A strong signal with no infrastructure. That is the gap. 🧬 academic.oup.com/eurheartj/a…

The protein behind almost every ALS case has been called undruggable for a decade. Drug designers looked at TDP-43's toxic region and saw a shapeless mess with nothing to target. This week in Nature Aging, someone found the target. Inside that disordered mess, Gao et al. identified a conserved 21-amino-acid segment that forms a druggable pocket. Their compound XL20 binds that pocket, crosses the blood-brain barrier, extends survival in ALS mice, and rescues human iPSC-derived motor neurons. Preclinical still. But the payload here is conceptual. "Undruggable" was never a measurement. It was a guess about what disorder means. The measurement just replaced the guess with an address. 🧬 nature.com/articles/s43587-0…

The mice that lost the most fat weren't eating less than the others. They were eating the same, sometimes more. In Longo's new Cell Metabolism work, the lever wasn't calories, and it wasn't even total protein. It was the dose of a single amino acid, methionine. Modulating it alone moved an entire hormone panel. IGF-1 fell while growth hormone, GLP-1 and FGF21 rose, and that shift was the mechanism rather than a byproduct. FGF21 turned out to be required for the fat loss and the insulin sensitivity to happen at all. Too little methionine caused frailty and too much erased the benefit. The useful range was narrow. None of this is a diet to copy. It's aged mice, and the human side is a 200,000-person cross-sectional read where the highest animal-protein intake tracked with twice the rate of diabetes. The trial in people hasn't run. What the paper actually establishes is quieter and more useful. The longevity value of how you eat becomes legible as a hormone signature you can measure, and the window between too little and too much is individual. Dietary rules give you the population average. They can't tell you where your own IGF-1 and FGF21 sit, or which way they're moving. That's not a question you settle at the dinner table.

Nearly 1 in 5 healthy adults have at least one organ aging years faster than the rest of their body, with no symptoms to show for it. Wyss-Coray's lab at Stanford measured 11 organs across 5,676 people from a single blood draw. The ones with an accelerated heart carried a 250% higher risk of heart failure. Every one of them would pass a standard physical as normal. You don't have one biological age. You have eleven, and a normal checkup is blind to the gap.

Gene therapy already extends lifespan in animals by targeting aging mechanisms directly. The bottleneck isn't biology, it's whole-body delivery in humans, where use is still mostly rare-disease. The frontier keeps rising. The layer you can act on today is measurement.

APOE4 is widely understood as the strongest genetic risk factor for Alzheimer's. A new study from @wysscoray 's lab at Stanford shows the picture is more complicated than the genotype alone suggests. Using plasma proteomics across 60,542 people, the team built biological aging clocks for over 40 individual cell types. They found APOE4 does two things at once at cellular resolution: it ages astrocytes faster while keeping macrophages younger. The same allele accelerates brain aging and enhances immune function simultaneously. Evolutionary biologists have theorized this trade-off for decades. This is the first time it has been directly measured, cell type by cell type, in living humans. The clinical finding is striking. Among APOE4/4 homozygotes with extreme astrocyte aging, 38% developed Alzheimer's over 15 years. Among those whose astrocytes were aging well, zero did. Same genotype. Different cellular trajectory. Different outcome.

The aging biomarker field has a measurement problem that nobody talks about. Over the past decade, we've accumulated dozens of epigenetic clocks, each trained on different cohorts, different preprocessing pipelines, different endpoints. Horvath, GrimAge, PhenoAge, DunedinPACE. All claim to measure biological age. But because none were ever tested head to head on the same dataset, the honest answer to "which one works best" has been: we don't actually know. The Biomarkers of Aging Challenge, now published in Nature Aging, took the same approach that transformed protein structure prediction. CASP gave us AlphaFold. This challenge gives aging measurement its first real benchmarking arena. Five hundred individuals, ages 18 to 99, DNA methylation profiles paired with real health outcomes, all run through a single standardized platform. Open to anyone. The winning models identified novel methylation signatures and outperformed existing biomarkers by combining advanced machine learning with biological domain knowledge that no single lab would have deployed alone. The question that matters most in longevity science right now isn't whether we can measure aging. It's whether we can agree on how to measure it. This is the infrastructure that makes agreement possible. 🧬 pubmed.ncbi.nlm.nih.gov/4227…

Do mice and humans actually age by the same dynamics? A large share of longevity intervention research is built on mouse models, and the working assumption is that the same basic process plays out across species, just at different speeds. A new analysis in Nature Aging from Uri Alon's group at Weizmann suggests the picture is more nuanced. They fit a damage accumulation model across nine species and found two distinct regimes. In mice, damage production overwhelms clearance almost from the start, so damage accelerates like free fall. In humans, production and clearance stay in dynamic balance for decades. The system only tips at very old age. If the dynamics are that different, how confidently can we extrapolate a mouse longevity result to ourselves? The paper doesn't answer that. But it's a question worth sitting with.

Statins work because we solved blood-based risk stratification for cardiovascular disease decades ago. Measure LDL, treat if it's high, prevent the event. Cancer prevention has never had that kind of molecular handle. A 14-protein plasma signature, trained on 48,000+ UK Biobank participants and validated across eight independent cohorts, now predicts lung cancer more than five years before diagnosis. The proteins are not shed by a tumor. They reflect an inflammatory pre-disease state in lung tissue, upstream of cancer itself. When Swanton's team re-analyzed the CANTOS trial, identifying high-signature individuals for anti-inflammatory treatment cut the number needed to treat to 55. That is statin-level precision applied to cancer prevention. The pattern keeps repeating across disease categories: blood-based molecular surveillance, applied early enough, turns detection into prevention. The measurement science is arriving. The system to deliver it to individuals is not. Pandya, Zagorulya et al. Cell, 2026. 🧬

Gladyshev Lab just published the largest transcriptomic aging study to date in Nature. 11,000+ gene expression profiles across four species (mouse, rat, macaque, human), 25+ tissue types, 79 interventions. What they built: a set of multi-species, multi-tissue clocks that predict not just chronological age, but expected mortality. The mortality clock outperformed dedicated lifespan clocks at distinguishing animals on longevity interventions from controls. Aging signatures were conserved across species and across 48 of 49 cell types tested. Two genes, CDKN1A and LGALS3, were validated against mortality and multimorbidity in the UK Biobank. What changes the picture: aging is modular. Gene expression clusters into 23 co-regulated modules, each with its own clock, each responding to different interventions. Caloric restriction slows metabolic modules. Chronic disease accelerates inflammatory ones. Klotho deficiency hits mitochondria. One biological age number was always a compression of 23 different trajectories. nature.com/articles/s41586-0…