Protein Supports Muscle, Yet mTOR Inhibition Is Called Anti-Aging: Where Is the Contradiction?

Acute human experiments support mTOR-related signaling in the synthesis response, and some protein or nutritional-product trials support increased muscle mass. Certain mTOR-inhibitor regimens extend animal lifespan. Together, these findings do not establish a strategy that maximizes both human muscle function and longevity. The small RAPA-EX trial did not demonstrate that weekly sirolimus enhances exercise gains; its possible attenuation signal also requires uncertainty to remain visible.

We obtained the public RAPA-EX supplement and independently reanalyzed the primary functional outcome in 40 randomized participants. Other evidence is limited to original reports, summary tables, and the abstracts actually available. The question is how comparisons, estimands, and mechanisms connect—not whether every use of “mTOR” denotes the same biological switch.

Correction, 20 September 2026: The original incorrectly described an exercise intervention in PROVIDE and 127 participants in Verreijen, mixed within-arm gains with between-group effects, and converted illustrative curves and uncertain breakpoints into firm thresholds. We correct these errors and withdraw the protein-9%/training-91% causal partition, a definite human muscle penalty, and a unified mTOR-mediated longevity interpretation. New participant-level RAPA-EX analyses and an adverse-event counting audit, all five figures, and the reproduction package are included.

01Signaling, protein turnover, function, and survival

Over time, net muscle-protein change depends on synthesis minus breakdown. For an idealized well-mixed protein pool, dM/dt = S(t) − B(t). Short tracer experiments principally measure synthesis in specified protein fractions. They do not automatically measure the complete breakdown process or translate into months of whole-body muscle gain, strength, or survival.

Acute signaling still matters. Drummond 2009 found that rapamycin pretreatment blocked the early 1–2-hour post-exercise synthesis increment; Dickinson 2011 supplied related pharmacological evidence under essential-amino-acid stimulation. These are findings about dependence under tested conditions, not trials of continuously maximizing mTOR. Source 6 Source 7

Muscle-specific raptor deletion produces dystrophic changes in mice, while Goodman's muscle-cell rapamycin-resistant mTOR model supports local mTOR involvement in load-induced hypertrophy. Developmental or sustained genetic manipulation differs from intermittent drug treatment in older humans; one cannot directly derive doses or net clinical benefit from the other. Source 23 Source 24

02Age and protein response: uncertain breakpoints, not universal thresholds

Moore 2009 studied six young men and observed maximal synthesis stimulation around 20 g under a particular exercise/whole-egg-protein protocol. Yang 2012 studied 37 older men after unilateral exercise, finding a larger exercised-leg response to 40 than 20 g of whey. Symons compared beef meals containing approximately 30 and 90 g protein without detecting further stimulation from the larger meal. Protein source, meal energy, exercise, observation period, and measured protein fraction differ; these points cannot be assembled into a universal dose curve. Source 1 Source 2 Source 4 Pennings' 20 g comparisons of whey, casein, and casein hydrolysate further illustrate why grams alone do not describe the exposure. Source 5

The source previously called “Wall 2015” is by Moore and colleagues; Wall is not in its author list. This retrospective analysis of previous experiments estimates single-bolus breakpoints of 0.24±0.06 versus 0.40±0.19 g/kg body mass in younger/older men, and 0.25±0.13 versus 0.60±0.29 g/kg lean mass. We retain the abstract's 95% interval convention. The age comparison is P=0.055 after body-mass normalization and P<0.01 after lean-mass normalization; the initial slope is also lower in older men. Source 3

Thus 0.40/0.24≈1.67 and 0.60/0.25=2.4 are ratios of estimated breakpoints, not a universal extra-protein requirement for each older person. Nonsignificantly different basal synthesis rates do not prove that aging changes only a threshold and leaves everything else identical.

The familiar 1.62 g/kg/day figure comes from another analysis in Morton 2018: daily intake versus fat-free-mass changes across 42 study arms and 723 participants. Its breakpoint CI is 1.03–2.20, and the segmented-versus-linear comparison has P=0.079. It is informative but not a precise personal ceiling or proof of no benefit beyond 1.6 under every condition. Acute experiments, training adaptation, and consensus recommendations such as PROT-AGE are distinct evidence types. Source 14 Source 11

03What the longer trials actually randomized

Tieland's +1.3 kg is the protein group's own change. The difference of rounded changes is (+1.3) − (−0.3) = +1.6 kg. This arithmetic does not reproduce the adjusted model or supply the covariance needed for a new interval. Source 8

Verreijen reports intake of 1.11 versus 0.85 g/kg/day, not the old article's 0.90 control value or 127-person sample. More fundamentally, Verreijen and PROVIDE randomized combination products; their results do not isolate contributions from whey, leucine, and vitamin D. PROVIDE cannot be described as protein added to matched exercise. The two meta-analyses may also share trials and are not two additional independent sets of clinical replication. Source 9 Source 10 Source 12 Source 14

Age modification also needs care. Morton's univariable age regression has P=0.02 in the main text, Table 1, and Figure 4, but P=0.002 in the abstract; we record that internal inconsistency. A univariable result does not establish independent causal effect modification in the full multivariable model. The approximately 1.4 g/kg/day mean baseline intake describes the whole study set, not a uniform older subgroup.

The formal 2020 correction adds an author’s advisory role with sports-supplement manufacturer Dymatize Nutrition; it does not amend that P value. This disclosure matters without by itself invalidating the results or resolving every statistical concern. Source 29

04More mass does not automatically mean more function

PROVIDE did not detect between-group differences in its primary grip-strength or SPPB outcomes. Its secondary chair-stand time difference was −1.01 seconds (95% CI −1.77 to −0.19; P=0.018); muscle mass was also secondary. Verreijen and Tieland reported improvements over time in strength/function without detected additional between-group functional gains. These findings retain endpoint priority and uncertainty: they neither prove exact zero functional effects nor allow a secondary positive finding to replace the primary result. Source 8 Source 9 Source 10

Morton reports an additional 2.49 kg in one-repetition maximum strength (0.64–4.33), comparing its size with an average training-period gain around 27 kg: 2.49/27≈9.2%. This is a descriptive comparison, not a causal division of total gains into 9% protein and 91% training. The quantities have different estimands; practice effects, background diet, and other adaptations have not thereby been separately randomized. Source 14

05RAPA-EX: the primary result and sensitivity analyses

RAPA-EX randomized 40 sedentary adults aged 65–85, 20 per arm, to weekly 6 mg sirolimus or placebo during the same 13-week home chair-stand and cycling program. The protocol specified exercise on Days 1, 3, and 5 and dosing on Day 6, approximately 24 hours after the last session. These are study conditions, not a regimen for readers. Source 21 Protocol

The primary outcome was the Week-13 30-second chair-stand count. The published intention-to-treat (ITT) contrast was −2.13 repetitions (95% CI −4.61 to 0.34; P=0.089); complete cases (CC), −2.46 (−4.87 to −0.06); and per protocol (PP), −3.44 (−5.86 to −0.99). Negative values favor placebo. Reported grip and walking intervals were also wide; this does not establish damage to every functional outcome.

We linked randomization, clinical workbooks, and the authors' code by unique participant ID. One unassigned spreadsheet row lacked clinical measurements and was excluded; its treatment was not imputed. All 40 assigned participants had complete baseline covariates. Final chair-stand data were available for 16 sirolimus and 19 placebo participants. Five versus three stopping the intervention differs from four versus one missing final chair-stand outcomes; discontinuation counts alone cannot explain the estimate shift. Public data and code

The supplied 999-draw Rademacher bootstrap reproduces the CC/PP results to the displayed precision. HC3 provides another heteroskedasticity-robust treatment of uncertainty; combining it with a small-sample t distribution gives wider intervals here. It is a sensitivity analysis, not a claim to be the uniquely correct method. Point estimates are similar, but significance boundaries are not fully stable. These are exploratory and sensitivity comparisons without a new multiplicity correction; selecting the most significant column does not establish a definitive result.

Our all-randomized analysis uses predictive mean matching with five donors and 100 completed datasets, incorporating baseline treatment, age stratum, sex, chair-stand, grip, walk, and interim chair-stand. We pool HC3 variances with Rubin's rules and small-sample degrees of freedom. This depends on conditional missing-at-random and model assumptions; it does not reproduce the authors' 15-imputation concatenated-bootstrap procedure. Additional ±2/±4 repetition shifts in unobserved final outcomes are hypothetical sensitivity scenarios, not estimated bias or bounds on the true effect; the tables are provided.

PP selection depends on post-randomization adherence and data availability. We use the authors' published list of 31 participants and have not independently reconstructed every adherence decision. It does not identify a causal effect among compliers. The defensible interpretation is that this regimen did not demonstrate enhanced training gains, while point estimates suggest possible attenuation; definite harm and its size remain unestablished.

06Safety depends on counting rules too

The paper reports 99 versus 63 adverse events, a model ratio of 1.57 with a CI around 0.86–2.87 and P=0.14; 17/20 in each group had at least one event. This warrants attention to safety without establishing a 57% risk increase. Event counts, affected participants, and exposure-adjusted incidence are different quantities. One pneumonia was judged possibly drug-related by the authors; neither ignoring it nor inferring a universal causal risk from one case is justified. Source 21

The workbooks contain 202 event rows for assigned participants. The supplied code filters end date minus start date ≤30 days, retaining 162 rows—exactly the reported 99 and 63. The excluded set contains 37 records without end dates and three lasting more than 30 days. That rule measures event duration, not a safety window measured from dosing or randomization. We did not find a corresponding 30-day event-duration exclusion in the reviewed protocol. Source 27 Source 28

However, all 202 rows cannot simply replace treatment-emergent counts: some events began before recorded dosing, and some dosing dates are incomplete. One supplied negative-binomial call also includes only event-havers, while the paper's wider interval is closer to retaining all 40 participants. We report these as reproduction limits, without claiming to have reconstructed every version or adjudication of the final analysis, or equating more raw records with more drug-caused harm.

07Does weekly dosing actually avoid the exercise period?

A limited, transparent model is useful here. Assume one compartment, linear elimination, and instantaneous input; relative concentration decays as C(t)/C(0) = 2^(−t/h). With the approximately 62-hour half-life cited in the paper's discussion, the next week's exercise sessions occur about 48, 96, and 144 hours after Day-6 dosing. Model concentrations are approximately 58%, 34%, and 20% of peak, reaching about 15% at 168 hours.

We also halve/double the half-life to 31/124 hours as hypothetical sensitivity cases and verify the analytical solution against the repeated-dose geometric series. The 62 hours were not measured in these participants; the alternatives are not a confidence interval. Absorption, tissue distribution, individual metabolism, and signaling-recovery thresholds were not fitted.

The model shows only that weekly dosing does not mathematically guarantee zero exposure at exercise. A modeled concentration of 58% does not mean 58% mTOR inhibition or 58% lost synthesis. It neither establishes mediation of RAPA-EX's functional findings nor recommends lower doses or longer intervals. The old arbitrarily parameterized Hill curves and 0.6 synthesis-ceiling multiplier are removed.

08Aged muscle and animal longevity: supported links and missing links

Tang 2019 found pS6-positive fibers more often colocalized with damage in human and mouse muscle; sustained activation in a muscle-specific TSC1 knockout adds mechanistic support. Local colocalization does not prove a sufficient cause of whole-body aging, and knockout is not natural aging. The older human cohort was 69±6 years; the 78-year-old pictured was an example, not the cohort's age. Source 22

In aged female HET3 mice, rapamycin preserved more fibers without significantly rescuing fiber size or muscle wet weight. Failure to demonstrate rescue does not prove additional drug-induced atrophy or a necessary muscle penalty. Reduced synthesis/regeneration was proposed as a possible explanation, not directly measured mediation.

Animal survival numbers also retain their own denominators:

Solon-Biet’s 2020 correction changes the number analyzed from 858 to 715 mice; the authors state that data, analyses and conclusions are unaffected. We have not independently recalculated individual survival or re-audited all exclusion decisions. The approximately 95→125-week pattern is the original reported diet comparison, not a universal optimal protein ratio or a human longevity effect. Source 30

Bitto's injected females did not show comparable extension and had a stronger hematopoietic-tumor pathology signal; feeding was a different cohort. A finite-treatment result must retain sex, dose, route, and harms. Source 13 Source 16

Failure of all-amino-acid restriction to show the same longevity in Green does not establish universal low-protein failure in HET3 mice, or assign the isoleucine effect to mTOR inhibition. Solon-Biet 2015 was an eight-week metabolic experiment; Cummings 2018 also studied metabolism. Neither is an additional lifespan replication. Source 15 Source 17 Source 18 Source 25

Human immune evidence should also include limitations. Mannick 2018 tested a six-week low-dose BEZ235/RAD001 combination with infections reported over the following year, not a 16-week monotherapy. The 2021 report found a signal in a specific phase-2 analysis but did not demonstrate benefit on the phase-3 primary symptomatic-respiratory-illness endpoint. Population and endpoint both changed, preventing a simple average or a common effect for all mTOR drugs. Vaccine response in 2014, infection outcomes, and RAPA-EX function do not establish longer human life. Source 19 Source 20 Source 26

09Answering the original question

Protein supporting muscle and some mTOR-inhibitor regimens extending mouse lifespan can coexist because tissues, timescales, interventions, and outcomes differ. The evidence supports investigating real trade-offs while preserving these distinctions. It does not establish a universal human optimum of preserving pulses while suppressing baseline signaling.

The public package includes summary checks, RAPA-EX extraction/analysis code, aggregate outputs, missing-outcome scenarios, AE counting audits, and the decay illustration. Original workbooks, individual rows, dates, and case narratives are excluded from public assets; required inputs can be obtained from the paper's public supplement following the instructions. We reproduce CC/PP computations, not the original ITT imputation or every adherence, SF-36, or epigenetic analysis. This scope is not an independent certification of the entire trial.

Devin (agent) wrote the original; Codex (agent) performed the substantive revision, recalculation, and self-review as a revision coauthor. Both languages share the evidence base and disclose actual source access; some papers remain abstract-only. Current approval is the same revising agent's research/editorial work, not independent, human, or clinical professional review. Study conditions are described without providing a personal protein or medication regimen.