Loss of proteostasis

Your cells run a constant quality-control operation on their own proteins — folding them correctly, and shredding the ones that come out wrong — and the machinery for this is thought to weaken with age, letting damaged proteins build up. It is one of the most elegant stories in aging biology and one of the least actionable: the evidence that it drives normal human aging is thinner than the confident textbook diagrams suggest, and outside of specific diseases there is no proven way to improve it.

Loss of proteostasis is the fourth of the twelve hallmarks of aging. "Proteostasis" is shorthand for protein homeostasis — keeping the cell's roughly ten thousand different proteins correctly folded, correctly located, and correctly disposed of when they wear out. A protein is a chain of amino acids that only works if it folds into a precise three-dimensional shape; a misfolded one is at best useless and at worst sticky, clumping together with others into aggregates the cell cannot clear. The failure of this system is the accepted explanation for Alzheimer's disease, Parkinson's disease, and the amyloidoses — diseases in which a normally soluble protein turns into deposits that accumulate in tissue.^[1]

How confident should you be

Descriptive — Moderate. The machinery is real and its failure is well characterised in cultured cells. But the human evidence for a steady lifelong decline is thin: most of it is cells in a dish, the decline looks biphasic rather than steady, and one animal study found no universal decline at all.

Intervention — none proven. No mammalian experiment has extended lifespan by boosting protein quality control. The drugs that genuinely repair a folding defect work only against a single named protein in a single defined disease. The two supplements sold for this that got proper trials both missed.

The three arms of quality control

The cell defends its proteins on several fronts at once, and the arms are worth separating because they fail — and are studied — differently.

Chaperones fold and refold. Molecular chaperones are proteins whose job is to shepherd other proteins into the right shape and rescue them when heat, oxidation, or crowding knocks them out of it. The best-known are the heat-shock proteins (HSP70, HSP90, and a family of small ones), so named because they surge when a cell is heated. Their master switch is HSF1, a protein that controls which genes get switched on; it sits dormant until stress activates it and then turns on the whole chaperone program.^[2]

The proteasome shreds. Proteins that are too far gone to rescue get tagged with a small marker called ubiquitin and fed into the proteasome, a barrel-shaped molecular shredder that chops them back into amino acids. This is the ubiquitin-proteasome system, and it handles most of the cell's routine protein turnover.

The unfolded-protein response sounds the alarm. When misfolded proteins pile up inside the endoplasmic reticulum — the compartment where proteins destined for export are made — a set of sensors detects the backlog and throttles new protein production while boosting chaperone output, buying the cell time to catch up.^[3]

A fourth arm, autophagy, digests aggregates and worn-out organelles wholesale. It is important enough, and separate enough, to have its own hallmark and its own page: see disabled macroautophagy. This article deals with the other three.

What actually declines in humans — and how thin the evidence is

Weak — almost all of it is cell culture, with one 2026 human-tissue exception. Here the honest answer is uncomfortable, and it is the reason this article is more cautious than most accounts you will read.

The cleanest study took skin fibroblasts from donors aged 20 to 82 and measured their proteasome activity. Capacity did fall with age — and the shortfall was real rather than merely an assay artefact, because oxidised and ubiquitin-tagged proteins piled up in exactly the cells whose shredders had slowed, which is what proteostasis failure is supposed to look like. But the decline was biphasic: it fell significantly up to about age 50, then showed no further significant change between 50 and 78.^[4] Whatever you lose, on this evidence, you have largely lost by midlife. Worth knowing whose question this was answering: the work came from the research arm of a cosmetics company, and its stated conclusion is about delaying skin aging. That does not make it wrong — it remains the best human dataset available — but the field's anchor study was built around skin aging, not human aging in general.

The chaperone arm shows a similar pattern with a specific lesion. In senescent human fibroblasts — cells that have stopped dividing but refuse to die — the heat-shock response is broken at HSF1 itself: after a heat challenge, senescent cells got roughly a third as much activated HSF1 into the nucleus, failed to switch on a 161-gene stress program including 27 chaperones, and lost 31% of their proteasome activity during the challenge — a deficit that deepened to 45% during what should have been the recovery period, while young cells were unaffected throughout. Notably, the cells could still sense the stress; what they had lost was the ability to mount a response to it.^[5]

Now the caveats, which are load-bearing:

Both findings are cultured cells, one of them fetal lung fibroblasts pushed into senescence by repeated division and then heated to 44 °C. That is not a 60-year-old's physiology, and it cannot tell you how much folding capacity you personally have lost.
The decline is not universal. Aged rat brain retains normal capacity to degrade ubiquitin-tagged proteins — if anything slightly more — even as its ability to cut simpler test substrates falls.^[6] And in female mice, proteasome activity changed in different directions in different tissues, leading the authors to conclude flatly that there is no universal decline in proteasome function with age.^[7]
Some of the longest-lived subjects hold on to their capacity. In that same mouse study, the exceptionally old animals had the best-maintained proteasome activity, not the worst. The same pattern appears in humans: fibroblasts from four healthy centenarians looked like those of young donors rather than elderly ones on every assay tried.^[8] Four cell lines is not much to build on, but it is hard to square with a simple "it wears out" model.

So the flat claim "your proteostasis is failing with age" is not something the human data can support. What it supports is narrower: protein quality control can fail, it demonstrably does so in cultured aged and senescent cells, and the failure is real enough to leave a trail of undegraded protein behind it.

One caveat on that caveat, from 2026. The "it's all cell culture" objection is beginning to weaken. A preprint from the Buck Institute measured the insoluble protein fraction directly in aging human brain tissue and found something more interesting than steady accumulation: the aggregate population remodels, starting in midlife. The most stable aggregates decline sharply in old age while intermediate-stability ones — the fraction enriched for Alzheimer's plaque and tangle constituents — build gradually and then accelerate after 80. Most relevant here, the amount of proteasome and chaperone machinery in a given brain predicted that person's aggregate burden about as well as their age did, which is the first human tissue evidence tying the apparatus to the outcome it is supposed to control.^[9] Two things to hold onto: this is not yet peer-reviewed, and it independently lands on the same midlife timing the fibroblast work did. It is a measurement, not a lever, and it complicates the naive picture of aggregates simply piling up rather than confirming it.

The species gap is the headline

Strong for lifespan in worms and flies; no lifespan result in mammals at all. The evidence that boosting proteostasis extends life is real, robust, and entirely invertebrate.

In the worm C. elegans, adding extra copies of the proteasome subunit rpn-6 is sufficient to make animals resistant to protein-folding stress and to extend their lifespan.^[10] Overexpressing the chaperone master switch hsf-1 — the worm's spelling of the same HSF1 introduced above — extends lifespan too.^[11] In fruit flies, switching on a proteasome subunit in adult neurons alone produced a robust extension of lifespan.^[12]

That fly experiment also contains the most under-quoted result in this literature. The same paper switched the subunit on everywhere in the body rather than only in neurons — and lifespan did not budge, while some healthspan measures got worse. More protein quality control is not simply better; where it helped, it helped in one tissue, and generalising it did harm.

Then you cross into mammals, and the lifespan results stop.

Mice engineered to overexpress human HSF1 do get a better stress response — but the gain is almost entirely in the inducible reaction to a challenge, with baseline chaperone levels barely shifted, and no aging or lifespan endpoint was measured at all.^[13] The most encouraging mammalian result is recent: aged mouse brain loses about 70% of its working shredder capacity in one proteasome form and half in the other, and boosting a proteasome subunit specifically in neurons slowed the age-related decline in spatial learning and memory in middle-aged and old mice — with no effect whatever in young ones, exactly the age-restricted pattern you would predict if the deficit were what limited them.^[14] That is a genuine result. But it slowed a decline rather than restoring a capacity, the endpoint was cognition rather than lifespan, and it came from a single laboratory — the same laboratory that ran the fly experiment above. The fly step and the mouse step of the chain below share a senior author.

There is, as of now, no mammalian experiment in which boosting proteostasis extended lifespan. The chain of evidence runs: human cells decline in a dish → worms and flies live longer if you boost them, and in flies only in the right tissue → mice think better if you boost them → and there it stops.

Two further complications deserve stating plainly. First, even in worms the mechanism is not what it appears: an engineered version of HSF-1 extended lifespan without enhancing chaperone induction at all, working instead through the cell's structural scaffolding.^[15] So "more chaperones, longer life" is not established even in the animals where the lifespan effect is real. Second, HSF1 is not a free lunch — Caution: deleting it protects mice from tumours driven by mutations in RAS, one of the commonest cancer-causing genes, because malignant cells lean on the chaperone system to survive their own chaos.^[16] Anything that durably cranks up HSF1 in a healthy adult is dialling a knob that cancers also like to see turned up.

The neurodegeneration trap

Moderate — the strongest-sounding argument for proteostasis, and the data resist it. Its failure is Alzheimer's, Parkinson's, Huntington's, and amyotrophic lateral sclerosis — all diseases of a specific protein misfolding and clumping. If aggregation destroys brains, surely aggregation is destroying yours a little every year?

The data resist this move in two ways.

Aggregation is common, and often silent. Pooling 55 studies and nearly 3,000 people with entirely normal cognition, the prevalence of Alzheimer's-type amyloid in the brain rises from about 10% at age 50 to 44% by age 90.^[17] Nearly half of cognitively normal ninety-year-olds are carrying the pathology while testing normal. That does not mean all of them are in the clear — the same paper infers a 20-to-30-year gap between becoming amyloid-positive and developing dementia, so some are early rather than exempt. But a lag that long is itself the point: aggregate burden and clinical disease are dissociable over decades, which is precisely what you would not expect if protein aggregation were a straightforward dose-dependent poison.

Clearing the aggregate barely helps the patient. This is the most instructive test the proteostasis hypothesis has been given, and it was run at enormous expense. Pooled across randomised trials, the anti-amyloid antibodies do what they were designed to do: they strip amyloid out of the brain, and by the standard yardstick for comparing effects measured on different scales, the effect is very large — a shift of more than one full standard deviation, which counts as large by any conventional reckoning. The effect on how patients actually are is a shift of 0.06 standard deviations on the primary measure, the Clinical Dementia Rating sum-of-boxes — an 18-point clinician-rated scale of dementia severity. That is far below the threshold at which a patient or clinician would notice anything. Even donanemab, the best performer, moved that scale by about 0.70 points against the 1-point bar for a noticeable difference. Meanwhile brain swelling became almost eight times more likely — a relative increase; the pooled analysis does not report how common it was in absolute terms.^[18]

Hold those two results side by side: the target moved enormously and the patient barely moved at all. You can read that charitably — and the authors do, arguing that treating earlier and longer might let a small effect compound into a real one. That reading is not unreasonable. But you cannot read it as a vindication. The cleanest test of "remove the misfolded protein, restore the function" came back with a resounding sort of, barely.

The one drug class that genuinely works — and what it does not prove

Strong, for one disease. There is a real proteostasis success story, and it is worth understanding precisely because of how narrow it is.

In transthyretin amyloid cardiomyopathy, a blood protein called transthyretin misfolds and deposits in heart muscle, stiffening it. Tafamidis is a small molecule that clamps transthyretin into its correct shape so it cannot misfold in the first place — proteostasis repair in its purest available form. In a 441-patient randomised trial over 30 months, it cut deaths from any cause from 42.9% to 29.5% — about 13 fewer deaths per 100 patients treated, or roughly 30% lower risk (hazard ratio 0.70, 95% CI 0.51–0.96). That bracketed range is where the true effect most plausibly lies; when such a range crosses 1.0 the result is compatible with no effect at all, and when it sits entirely below 1.0, as here, the benefit is unlikely to be chance. The drug also reduced cardiovascular hospitalisations from 0.70 to 0.48 per year and slowed the decline in both walking distance and quality of life.^[19]

The principle has since held up twice more, which matters, because a single trial of a single molecule would be a thin foundation. Acoramidis, a second stabiliser working the same way, beat placebo in 632 patients on a combined ranking of death, hospitalisation and function — not on mortality specifically.^[20] And vutrisiran attacks the problem from the other end: instead of fixing the fold, it shuts down production of the protein, cutting deaths from any cause over 42 months by about 35% (hazard ratio 0.65, 95% CI 0.46–0.90).^[21] All three trials were funded by the companies selling the drugs, which is normal for cardiology and worth knowing anyway.

That is a large mortality benefit from stabilising a misfolding protein, and a principle that has now survived three trials of two different strategies. It proves the principle — and it also shows exactly what the principle requires. These drugs work because there is one culprit protein, with a known misfolding mechanism, causing a specific disease, in patients selected because they have it. None of those conditions holds for ordinary aging, where the hypothesised problem is a diffuse, gradual loss of general folding capacity with no single protein to stabilise and no drug that could stabilise all of them at once. Note that even the silencer, which sidesteps folding entirely, needs the same thing the stabilisers do: a named protein to remove. This drug class validates proteostasis as a target for a defined proteinopathy. It says nothing about whether a healthy 50-year-old should be trying to boost their chaperones.

What you can actually do (very little, honestly)

Sauna — Weak; the mechanism is weaker than advertised. The logic runs: heat induces heat-shock proteins, heat-shock proteins are chaperones, therefore sauna improves proteostasis. The human data supporting the first link are slim. The most-cited study is not a sauna study at all: 25 healthy university-age adults sat in a heat chamber at 73 °C for 30 minutes, and in the 13 who gave blood samples, a circulating heat-shock protein (HSP72, from the HSP70 family) rose by 48.7% — but with a standard deviation of ±53.9%, meaning the variation between individuals exceeded the average effect.^[22] And there is a deeper problem: what rises in the blood is extracellular heat-shock protein, which acts as a stress and immune signal. It is not evidence that the chaperones inside your cells are folding proteins better. The bridge from "sauna raises a blood marker in 22-year-olds" to "sauna improves your protein quality control" is, at present, an inference and not a finding.

None of which means don't use the sauna. It means use it for the reasons that actually have outcome data behind them — see sauna — and not because you are tuning your proteostasis.

The supplements aimed at this have missed — Weak. Spermidine got the trial the field wanted: 100 older adults with subjective cognitive decline, 12 months, randomised and placebo-controlled. It did not improve memory, and it did not move the biomarkers.^[23] One honest caveat on the strength of that null: the dose was 0.9 mg per day of a wheat-germ extract, and the authors themselves call for trials at higher doses. Urolithin A is often marketed in this space, though its mechanism is really mitochondrial recycling rather than protein folding; in 66 older adults it failed both of its primary endpoints, and the muscle-endurance benefit that did appear at 2 months had evaporated by 4 months as the placebo group caught up. That trial was run with the manufacturer's involvement, so the null is if anything the conservative reading.^[24] Both are covered further under geroprotectors.

You cannot measure it — Caution on anyone who offers to. There is no validated blood test of "proteostasis capacity." The aggregation-related blood markers that do exist and work well — p-tau217 and p-tau181 (fragments of the tau protein that tangles inside neurons), neurofilament light (a structural protein released when nerve fibres are damaged), and glial fibrillary acidic protein (released by the brain's support cells when they react to injury) — are diagnostic tools for people who already have symptoms. The 2025 Alzheimer's Association clinical practice guideline restricts them to patients with objective cognitive impairment in specialist care, and states explicitly that it "does not extend to cognitively unimpaired individuals, given the current lack of clinical relevance."^[25] The same guideline warns that many commercially available versions of these tests do not meet its accuracy thresholds. If you are well and someone offers to measure your amyloid biology, the profession's own guideline says there is no established reason to.

What this does and doesn't tell you

What it tells you: protein quality control is real machinery, it demonstrably fails in aged and senescent human cells, and repairing a specific folding defect can save lives — the transthyretin drugs are proof, three times over. The mechanism deserves its place among the hallmarks.

What it doesn't tell you: that your proteostasis is measurably declining (the human data are mostly cell culture, contested, and suggest most of the change happens before 50 — the one human-tissue study is a preprint); that boosting it would extend a mammal's life (no such experiment has succeeded, and the one animal experiment that boosted it body-wide made things slightly worse); that clearing aggregated protein reliably restores function (the antibody trials moved the target enormously and the patients barely at all); or that any available supplement, sauna protocol, or blood test lets you act on it. The honest position is that this is the hallmark with the widest gap between the beauty of the mechanism and the emptiness of the shelf — and the correct response to an empty shelf is to notice that it is empty, not to buy something anyway.

There is no proteostasis-specific lever worth naming. The general ones — exercise, sleep, staying metabolically uninflamed — are the best available bet, and the deep-sleep brain-clearance story is often invoked here, but all of that is inference from what those habits do elsewhere in the body rather than any finding about protein quality control. Do them for the reasons that have outcome data. See genomic instability for the same argument in a different hallmark, and dementia prevention for what is genuinely known about protecting the aging brain.