Ten percent human: an evidence-graded audit of the mTOR literature and the pathway's translational gap

Oliver Barton · ORCID 0009-0008-2025-2148

Preprint, posted 15 September 2026 · not peer reviewed · CC BY 4.0

Download PDF (12 pages, 160 kB)

DOI 10.6084/m9.figshare.33772297

Abstract

The mTOR pathway is among the most heavily cited targets in ageing biology, and claims derived from it circulate widely in translational and public-facing contexts. What is rarely made explicit is the kind of evidence any given claim rests on. I report a systematic audit of a hand-curated corpus of 386 mTOR studies (1975-2026), in which every record carries an explicit classification of the biological system it was done in, assigned under a fixed rubric.

Three findings emerge. First, the human evidence base is thin: 39 of 386 studies (10.1%) are human work of any kind (systematic review, clinical trial, or observational study), while 305 (79.0%) are animal or mechanistic/in-vitro work. Second, the human evidence that does exist is overwhelmingly about disease indications rather than ageing: of 25 human clinical trials, 13 concern licensed uses (transplant immunosuppression, oncology, tuberous sclerosis complex / lymphangioleiomyomatosis) and two are physiology experiments in healthy volunteers, leaving ten ageing- or healthspan-oriented trials. Among those ten, every positive primary-endpoint result comes from an early-phase, small or uncontrolled design; the one adequately powered confirmatory trial (Mannick et al., 2021; n = 1024) missed its primary endpoint (26% vs 25%, p = 0.65) after a positive phase-2a by the same group; the only purpose-built healthspan RCT completed to date (PEARL; Moel et al., 2025; n = 114) reported a pre-registered null; and a pre-registered crossover trial in 63 healthy adults found that moderate dietary protein restriction did not raise autophagic flux at all (Singh et al., 2026), against a clear preclinical expectation that it should. Third, the proportion of human evidence is rising slowly but monotonically across eras - 0% before 2000, 5.5% in 2000-2009, 10.0% in 2010-2019, 14.7% from 2020 - meaning that even in the most recent period roughly six in seven curated studies remain pre-clinical.

I also report a structured gap layer derived from the corpus (ten hypotheses, each linked to a mean of 6.6 constituent studies), and describe a revision of the classification scheme itself prompted by external expert critique - a case in which the act of grading evidence exposed a defect in the grading rubric. The corpus, the rubric, and the generating code are openly available.

The record

How to cite

How this relates to the live audit

This paper is a fixed, citable snapshot. The evidence audit on this site is the same measurement recomputed from the corpus on every build, so its numbers move as the corpus grows and the paper's do not. Cite the paper for a figure that must stay put; read the audit for the current state.

The corpus the paper measures (386 studies at the time of writing) is browsable study by study under all studies, and the pathway links it grades are in the interactive map.