Weekly Issue — 2025-08-03

Start with the head-to-head data clinicians actually care about. A real-world analysis of U.S. claims and electronic medical records compared once-weekly semaglutide against SGLT2 inhibitors in adults with uncontrolled type 2 diabetes. After inverse-probability weighting, the semaglutide cohort showed mean weight reductions of roughly 4–5 kg at one year, alongside HbA1c improvements — outcomes the authors describe as significant relative to the SGLT2 comparator, though with the usual caveats of observational design. It is the kind of evidence biohackers like to cross-reference against their own continuous glucose traces: directional, large-sample, but not a randomized trial. The PAUSE study sits squarely in the "useful but read carefully" pile.

Statins reliably crush LDL, but a stubborn fraction of cardiovascular risk hides in remnant-like particles — triglyceride-rich lipoproteins that statins barely touch. A 2025 before-and-after study in 41 patients with ischemic heart disease, all already on statins, tracked remnant-like lipoprotein (RLP) cholesterol three months after starting oral semaglutide. Mean RLP cholesterol fell from 8.52 mg/dL to 5.46 mg/dL, a statistically significant drop, with reductions also observed in the subgroup switched from DPP-4 inhibitors. The single-center trial is small and uncontrolled, but it points at a mechanism the cardiology community has been chasing: an add-on lever for residual lipid risk in patients who have already done everything their guideline-directed therapy asks of them.

This is the kind of data point that lands well with readers who treat their lipid panels like telemetry. It is also exactly the kind of finding that needs replication before anyone starts retitrating their regimen around it.

One of the most interesting papers of the year is not about humans at all. Researchers built single-cell atlases of mouse white adipose tissue across four states: lean, diet-induced obese, post-vertical-sleeve-gastrectomy, and post-semaglutide. The headline finding is that semaglutide and bariatric surgery produce different cellular signatures, even when the scale moves comparably. Some populations return toward a lean-like phenotype; others persist in an obese-like state. The comparison implies that the route to weight loss may matter for what fat tissue ends up looking like — and, by extension, for downstream metabolic behavior.

For self-experimenters, the takeaway is conceptual rather than actionable: a kilogram lost on semaglutide may not be metabolically identical to a kilogram lost via surgery or, by inference, via diet alone. It is a mouse study. It is also a useful corrective to the assumption that endpoint weight is the only variable worth tracking.

Adults with congenital heart disease (ACHD) are a population that rarely gets clean trial data — they are heterogeneous, complex, and small in number. A Mayo Clinic retrospective cohort of 70 ACHD patients prescribed semaglutide or liraglutide found that 42.9% achieved more than 5% weight loss over a mean of roughly 21 months, with better results in those with higher baseline BMI and younger age. HbA1c trended down by 0.6%. There were no significant changes in NYHA functional class or estimated glomerular filtration rate; one-third experienced side effects, mostly gastrointestinal, and about 11% discontinued or were hospitalized for adverse effects. The analysis is the first systematic look at GLP-1 use in this group and reads as cautiously encouraging: feasible and reasonably safe, with weight loss but no proven functional benefit yet.

Cardiac surgery is murkier. A 2025 review of novel antidiabetic agents in heart failure patients undergoing cardiac surgery concluded that GLP-1 receptor agonists, DPP-4 inhibitors, and SGLT2 inhibitors all appeared safe perioperatively in the available evidence, with SGLT2 inhibitors showing the most heart-failure-specific benefit. The authors are explicit that the literature is thin and that larger, more rigorous studies are needed before strong perioperative recommendations land. Anyone weighing this in their own care should be talking to their surgical and anesthesia teams, not to a wearable.

GLP-1 receptors are not confined to the gut and brain. A 2025 review in clinical and aesthetic dermatology surveyed the literature on GLP-1RAs and skin, and reports preliminary anti-inflammatory effects, with signals in psoriasis, hidradenitis suppurativa, and wound healing. The same review catalogs cutaneous adverse reactions — injection-site pruritus, erythema, rash — and is candid that current evidence rests on case reports and small studies. It is a frontier worth watching, not a treatment plan.

A broader review of semaglutide in obesity pulls the SUSTAIN, PIONEER, and STEP programs into one frame and emphasizes both the magnitude of benefit and the known limitations: weight regain after discontinuation, GI side effects, and substantial inter-individual variability. The synthesis is a reminder that the durability problem has not been solved — and that maintenance, not initiation, is where the next generation of evidence needs to land.

For the quantified-self reader, the honest framing is this: semaglutide is doing more than it was designed to do, in more tissues than its original label suggests, and the 2025 literature is starting to map where. The map is still rough. Edges are dotted, scale bars are approximate, and several promising regions are sketched from a single small study. That is not a reason to dismiss the work — it is a reason to read it with the same skepticism you would apply to any new dataset, and to wait for the replications that turn moderate evidence into something stronger.

Epigenetic clocks work by reading DNA methylation: small chemical tags on the genome that shift in predictable patterns as we age. The standard lab workflow uses bisulfite conversion, a chemistry step designed to distinguish methylated from unmethylated cytosines so an array can quantify the difference. Feed enough of those readings into a regression model, train it against the known ages of thousands of donors, and you get a clock — a number that tracks chronological age with uncanny accuracy and, in some configurations, claims to predict biological age too.

The new analysis, published in GeroScience in July 2025, pulls on a loose thread in that workflow. The researchers report that methylation array signals can predict chronological age even without the bisulfite conversion step — the step that is supposed to make the measurement about methylation in the first place. In other words, something else in the signal is carrying age information, and the clocks have been quietly riding along on it.

The team calls these confounders "pseudomethylation" signals: array readings shaped by non-methylation factors that nonetheless correlate with age. Some of this appears tied to sequence variation — genotype-dependent quirks in how probes bind to DNA. The authors note that epigenetic clock sites are overrepresented near genomic regions whose methylation state depends on sequence variants, suggesting that part of what clocks have been learning is genetic, not epigenetic.

If you have spent any time in the longevity corner of the internet, you have seen the screenshots: a user posts their epigenetic age before a protocol and again six months later, triumphantly two or three years younger. That delta is the entire emotional product of a consumer clock. It is also the part most vulnerable to the issues the new paper raises.

Two implications stand out. First, if non-methylation factors are partly driving the age signal, then small technical shifts — a different array lot, a different lab, a different sample type — could nudge a result in ways that have nothing to do with how well you slept or how clean your diet got. Second, if genotype is baked into the signal, then a portion of the "age" being reported is, in effect, a fixed feature of the person being tested, not a dynamic readout of how they are living. Neither point invalidates epigenetic clocks. Both complicate the way their outputs are currently marketed.

The authors are careful, and so should we be. They do not argue that methylation is irrelevant to aging — the broader literature on that point is substantial. They argue that quantifying these covariates will be critical to building better clocks and designing appropriate studies of epigenetic aging. That is a methods critique, not a demolition. But methods critiques are exactly the kind of thing that should travel from the journal to the product page before a consumer pays two or three hundred dollars for a number.

Here is the interesting twist. The same paper that flags pseudomethylation as a confounder also reports that those signals are uniquely age predictive in their own right. That is not just a bug; it is potentially a new lane. If non-methylation array signals carry independent information about chronological age, future clocks could explicitly incorporate them — or, better, separate them — to produce a cleaner read on what is actually changing as a person ages versus what is fixed by their genome or introduced by the assay.

That is how the field tends to move. A first generation of clocks proved the concept. A second generation chased biological age, mortality risk, and pace-of-aging metrics. A third generation, if this line of work holds up, may be defined by how rigorously it strips out the confounders the early clocks were inadvertently leaning on. The consumer market will follow, eventually. It usually does.

For now, the practical posture is the unglamorous one. If you are tracking your own optimization, an epigenetic age report is a reasonable data point to collect, especially over long intervals and from the same provider using the same sample type. It is a poor scoreboard for a six-week protocol. And it is not, despite the framing on most product pages, a verdict on how well you are aging. The clocks are real instruments. They are also, as of mid-2025, instruments whose calibration is openly being revised in the literature.

The looksmaxing instinct — measure everything, optimize relentlessly — is, on balance, a good one. It is what turns vague intentions into routines that compound. But measurement only pays off when the instrument is honest about what it is measuring. The GeroScience paper is a useful reminder that the most prestigious number in longevity is still, in important ways, under construction. The smartest move is not to abandon the tools. It is to hold them at the confidence level the evidence currently supports — and to keep an eye on the next generation of clocks, which may finally tell us something the first generation only implied.

The premise behind these clocks is by now familiar to anyone tracking the longevity beat. DNA methylation patterns at specific genomic sites shift with age in ways consistent enough to be read like a chemical timestamp. Train an algorithm on enough samples, and you can estimate how old a person's tissues appear to be — and, more interestingly, whether they appear older or younger than the birthday on their passport. That gap, often called biological age acceleration, is the variable the new research is finally putting to work.

The question is no longer whether these clocks tick. It is whether their readings translate into anything a physician — or a thoughtfully self-directed reader — could meaningfully use.

Among the body's organs, the kidneys are an unusually honest reporter of systemic aging. They filter, they accumulate damage, and they decline quietly until the decline is no longer quiet. A 2024 analysis in GeroScience drew on the Future of Families and Child Wellbeing Study to ask whether a panel of aging biomarkers could predict kidney health better than chronological age alone. The researchers combined epigenetic clocks (GrimAge and Horvath), immune-function markers, and metabolic indicators across nearly 4,900 participants, then used machine learning to predict Cystatin C, a sensitive readout of kidney filtration.

The model explained roughly 86% of the variance in Cystatin C levels, with GrimAge, pack-years of smoking, and immune-function markers emerging as the strongest contributors. The authors also identified three biologically distinct clusters in the data — one with notably younger biomarker profiles, and one in which both GrimAge and the stress-response protein GDF-15 were elevated, marking what they described as elevated risk for age-related disease.

The takeaway is narrower than a headline-ready miracle, and more interesting for it. Epigenetic age, on its own, is informative. Epigenetic age plus immune aging is more informative still. The body's defensive cells are not bystanders in organ decline; they appear to be participants, and clocks that ignore them may be leaving signal on the table.

If kidney filtration is the quiet reporter, physical capacity is the loud one. Can you stand up from a chair five times without using your hands? How much oxygen can your body actually use under load? These are not abstractions; they are some of the most durable predictors of independence in later life. A 2025 cross-sectional analysis from the INSPIRE-T cohort in south-west France — 1,014 adults aged 20 to 104 — tested whether biological age acceleration measured by Horvath's, Hannum's, PhenoAge, GrimAge and the inflammation-based iAge clocks tracked with these real-world capacities.

GrimAge was the standout. Higher GrimAge acceleration was associated with slower five-time sit-to-stand times, lower Short Physical Performance Battery scores, and reduced VO2max, with effect sizes that were modest but statistically robust across the lifespan. The other clocks performed less consistently, a reminder that not all biological-age scores are interchangeable — a point the supplement industry has been slow to absorb.

What makes the INSPIRE-T result useful is its breadth. The cohort spans more than eight decades of adulthood, which lets researchers ask whether a clock's signal holds in a 30-year-old as well as a 90-year-old. GrimAge, it turns out, behaves more like a stable instrument than a curiosity tuned to one end of life.

Cardiovascular medicine has historically been comfortable with risk scores — cholesterol, blood pressure, smoking status, age. A 2025 review in Cardiovascular Research argues that the next generation of risk modelling should look beneath those surface variables to the molecular machinery of aging itself. The authors map four interlocking mechanisms — telomere attrition, cellular senescence, epigenetic modification, and mitochondrial dysfunction — onto the endothelial dysfunction and systemic inflammation that drive most acquired cardiovascular disease.

The review is candid about the challenge. Each of these mechanisms can be measured in clinical or laboratory settings, but the measurements do not yet add up to a single, validated score that a cardiologist can act on. What they may add up to, the authors suggest, is a feedstock for machine-learning models capable of integrating disparate data — imaging, bloods, methylation, functional tests — into a more personalised picture of cardiovascular age. That is a careful claim, and the appropriate one. We are at the point of building the dashboards, not yet at the point of trusting them with treatment decisions.

For readers tracking the cutting edge, the temptation is to read these results as a green light to start optimising a GrimAge score. That would overshoot what the evidence currently supports. These studies are largely cross-sectional or observational; they show association, not that intervening on the clock changes the downstream outcome. Whether lowering biological age acceleration through lifestyle or pharmacology actually preserves kidney filtration, sit-to-stand speed, or coronary health is a separate question, and the answer requires randomised trials that are only now beginning.

What has shifted is the credibility of the underlying instruments. Three independent lines of work — kidney health, functional capacity, and cardiovascular biology — now converge on the same broad conclusion: GrimAge in particular, and inflammation-aware clocks more generally, carry information that chronological age does not. That is not a revolution. It is, more usefully, the moment when a field stops debating whether its main tool works and starts debating how to use it.

The honest framing, for now, is this: biological age is no longer a parlour trick, but it is not yet a prescription pad. It is a measurement — increasingly trustworthy, increasingly tied to outcomes that matter — and like every measurement before it, its value will depend on what we choose to do with it next.