mTOR Core · Lesson 10 · Research · 30 min

How to Read an mTOR Paper

Six habits that survive contact with the literature

The question

What separates what a paper measured from what its abstract says?

What you should be able to do

After this lesson you should be able to

Separate any reported result into its model system, perturbation, readout and claim.
Evaluate a selectivity claim in terms of dose and duration rather than of the molecule.
Distinguish a pre-registered primary endpoint from a secondary finding, and weight them accordingly.
Propose the experiment that would close the gap between a readout and the claim made from it.

Research skill Evidence evaluation

The core idea

Nine lessons of mechanism are worth less than they should be without one more skill: the ability to read a paper and work out what it actually showed. This lesson has no new pathway in it. It has a set of habits, each attached to a study in this corpus that makes the habit concrete.

The organising idea is simple. Every result has four parts: a model system, a perturbation, a readout, and a claim. The first three are what happened in the laboratory. The fourth is a sentence written afterwards. Most of the work of reading is measuring the distance between the third and the fourth, and that distance is not usually marked.

None of what follows is a reason for cynicism about the literature. The studies used here as examples are good studies; several are landmarks. The point is the opposite of dismissal — a careful reader gets more out of a strong paper, not less, because they can see what it bought and what it left for someone else.

Nine lessons of mechanism are worth less than they should be without one more skill: working out what a paper actually showed.

Every result has four parts: what was studied, what was done to it, what was measured, and what the authors say it means. The first three happened in a lab. The fourth is a sentence written afterwards.

Most of reading is measuring the distance between the third and the fourth. None of this means the literature is untrustworthy — the examples here are good papers.

The anatomy of a claim

Model system. A cancer cell line, a primary cell, a fly, a worm, a mouse, a dog, a person. Each answers a different question and none is a small version of the next one. The Atlas's evidence tiers are a compressed statement of this: tier D is mechanistic or in-vitro work, tier C animal, tier B human, tier A systematic review. Tier is study design, not quality. A careful biochemistry paper is not a weak trial; it is a different experiment.

Perturbation. What was actually done — a knockout, a knockdown, an overexpression, a drug at some dose for some duration, a diet. Overexpression tells you what a protein can do; a knockout tells you what the system does without it, in the presence of whatever compensation occurred. Neither is the same as watching the endogenous protein work.

Readout. What was measured, when, and in what units. A phosphorylation on a blot, a lifespan curve, a tumour volume, a body-composition scan. The readout is the only quantitative fact in the paper; everything else is interpretation of it.

Claim. The sentence in the abstract. Read it last, deliberately, after you have the other three — and then ask what would have to be true for it to follow.

What was studied. A cell line, a fly, a mouse, a dog, a person. The Atlas tiers record this: D is lab-bench work, C animal, B human, A a review of many human studies. Tier is the type of study, not a score.

What was done. A gene removed, a gene added, a drug at some dose for some time, a diet. Adding extra protein shows what it can do; removing it shows what the system does without it.

What was measured. A band on a gel, a survival curve, a tumour size. This is the only number in the paper.

What is claimed. Read this last, on purpose, and ask what would have to be true for it to follow.

Interactive model

The tool and the readout both write the conclusion

Tool

Readout measured

The same two complexes and the same three readouts as Lesson 02, with a different question attached: which tool did the experiment use, and which readout did it choose? What lights up here is what that combination lets a reader see — not what changed in the cell. The biology is the same in all nine; the conclusions are not.

p-S6K1 collapses

The most-published combination in the field, and the one that produced the shorthand 'rapamycin inhibits mTORC1'. Read on its own it supports exactly one sentence: this drug lowers this substrate's phosphorylation.

mTORC1: mTOR–Raptor–mLST8 (+PRAS40, DEPTOR); switched on only at the lysosome.The growth decision itself.Coincidence detector: nutrients supply location via the Rags, growth factors supply activation via Rheb. Neither alone is sufficient — the single most important idea in the pathway.
mTORC2: mTOR–Rictor–SIN1–mLST8; phosphorylates Akt, SGK1 and PKC.mTOR's second, less famous complex.Largely plasma-membrane associated and PI3K-responsive via the SIN1 PH domain; acutely rapamycin-insensitive, which is the cleanest way to separate mTORC1 from mTORC2 biology experimentally.
S6K1: Ribosomal protein S6 kinase; the standard readout of mTORC1 activity.mTORC1's best-known output kinase.T389 phosphorylation by mTORC1 is rapamycin-sensitive, which is why S6K1 became the field's default assay — and why the field long over-read rapamycin as a complete mTORC1 inhibitor.
4E-BP1: Translational repressor released from eIF4E upon multi-site phosphorylation.A cap on protein-making that mTORC1 removes.Only partially rapamycin-sensitive. This single fact explains the rapalog/Torin discrepancy and drove the whole ATP-competitive inhibitor programme.
Akt/PKB: AGC kinase requiring PIP3 recruitment plus two phosphorylations to be fully active.The main 'grow' relay from growth factors.T308 by PDK1 and S473 by mTORC2. Recruitment and activation are separate events — a distinction the older Atlas diagram blurred.

This is a simplified teaching model with a fixed set of states, not a simulation. It shows the direction each control pushes the pathway, not how much, how fast, or what any particular cell would do.

Every state of this model

Tool	Readout measured	Readout	What it means
rapamycin	p-S6K1	p-S6K1 collapses	The most-published combination in the field, and the one that produced the shorthand 'rapamycin inhibits mTORC1'. Read on its own it supports exactly one sentence: this drug lowers this substrate's phosphorylation.
rapamycin	p-4E-BP1	4E-BP1 phosphorylation largely maintained	Same drug, same cells, different substrate — and the conclusion inverts. A reader who only ever sees the S6K1 blot will believe mTORC1 was switched off; the 4E-BP1 blot says part of its output never stopped.
rapamycin	AKT Ser473	Ser473 present after short exposure, reduced after prolonged exposure	Here the answer depends on a number that is usually in the methods section rather than the figure: how long the drug was present. Two papers reporting opposite results can both be right.
ATP-competitive	p-S6K1	p-S6K1 collapses	The same readout as the first state and a different tool, giving the same answer — which is exactly why a single readout cannot distinguish between the two drugs, and why the field needed the comparison rather than either result alone.
ATP-competitive	p-4E-BP1	4E-BP1 phosphorylation lost	This is the comparison that reset the field: the output rapamycin leaves running stops when the active site is blocked. The finding is not in either experiment on its own, it is in the difference between them.
ATP-competitive	AKT Ser473	Ser473 lost	Both complexes stop, so anything measured here cannot be attributed to mTORC1 alone. A paper using this tool and claiming an mTORC1-specific conclusion is over-reading its own reagent.
bi-steric	p-S6K1	p-S6K1 collapses	Bi-steric compounds were designed to block mTORC1 more completely than rapamycin while sparing mTORC2. On this readout they look like every other tool — the readout cannot see what the compound was designed for.
bi-steric	p-4E-BP1	4E-BP1 phosphorylation lost	This is where the design shows: the rapamycin-resistant output stops, while the mTORC2 readout is left. Whether that separation holds at a given dose over weeks in an animal is a different question, and one the corpus keeps open.
bi-steric	AKT Ser473	Ser473 present	The readout the selectivity claim rests on. Note what would be needed to trust it in a specific paper: the dose, the duration, the cell type, and a measurement rather than an assumption that the target was engaged.

Six habits, each with a worked example

1. The tool defines the result. For years 'mTORC1 inhibition' meant rapamycin, and rapamycin's effects were read as mTORC1's functions. An ATP-competitive inhibitor that blocks the kinase site directly revealed mTORC1 activities rapamycin leaves running — notably 4E-BP1 phosphorylation. Whenever a field's central concept is defined by one reagent, expect the concept to change when a second reagent arrives.

2. Dose and duration are part of the mechanism, not the methods section. Short rapamycin exposure principally affects mTORC1; prolonged treatment also reduces mTORC2 signalling in many cells by interfering with assembly of new complexes. 'Rapamycin inhibits mTORC1 and spares mTORC2' is therefore a statement about a protocol. Two papers can disagree while both being right, because they ran different experiments with the same molecule.

3. Ask which half of the animals it worked in. Deleting S6K1 extended median lifespan by about 19% in female mice, with no significant effect in males. Sex-specific results are common in this literature and are frequently lost in summary. When a lifespan figure is quoted without a sex, the number is incomplete rather than wrong.

4. Notice which species carried the claim. A short course of low-dose rapamycin was associated with improved measures of heart function in healthy pet dogs. Dogs are a genuinely informative model — outbred, sharing a human environment, ageing at a useful rate — and this is still tier C. It is stronger than a mouse result for extrapolation purposes and it is not a result in people.

5. Read the pre-registered primary endpoint, not the highlights. The first completed long-term randomised trial of rapamycin for healthy ageing reported no significant change in its primary endpoint, visceral fat by DXA. A null primary endpoint is a real result and it is the one the trial was designed to answer; secondary findings from the same trial are hypothesis-generating and should be read as such.

6. Prefer the systematic review for the state of a field — and read its limits. A systematic review of rapamycin and its derivatives for ageing screened more than 18,000 records and included 19 studies, reporting improvements in some parameters and no significant effect on others. That is the highest tier in the Atlas, and it is still a summary of a small and heterogeneous literature. Tier A means the method of aggregation was rigorous; it does not manufacture evidence that the underlying studies do not contain.

The tool shapes the answer. For years 'blocking mTORC1' meant rapamycin. A different drug showed that rapamycin leaves some of mTORC1's jobs running.

Dose and time are part of the biology. Short rapamycin hits one assembly; long rapamycin can hit both.

Ask which half of the animals it worked in. Removing S6K1 extended life in female mice, with no significant effect in males.

Notice the species. A trial in pet dogs is more informative than a mouse study and is still not a result in people.

Read the endpoint the trial promised to measure. The first long trial of rapamycin for healthy ageing found no change in the measure it was designed around.

A review of many studies is the best summary of a field — and cannot contain more evidence than the studies it reviews.

Three sentence-shapes that should slow you down

'X is a therapeutic target for Y.' Usually this means an inhibitor changed a readout in a model. It is a hypothesis about a target, stated in the grammar of a conclusion.

'Rapamycin extends lifespan.' In which species, at what dose, from what age, in which sex, on which diet, and by what measure — median or maximum? Each of those has been shown to matter somewhere in this corpus. The bare sentence is a compression of a set of results, several of which disagree with each other in detail.

'Our results show that A causes B.' Check whether the design supports a causal claim at all: was the perturbation randomised, was there a control for the perturbation itself, is there a dose-response, and is the effect reversible? Then check what the paper measured about B — often it is a marker of B rather than B.

'X is a target for treating Y.' Usually this means a drug changed a measurement in a model. It is a hypothesis written in the grammar of a conclusion.

'Rapamycin extends lifespan.' In which animal, at what dose, from what age, in which sex? Each of those changes the answer somewhere in this corpus.

'Our results show A causes B.' Ask whether the design can support a causal claim at all — and whether B was measured, or only a stand-in for it.

The same rules apply to this Atlas

Everything above applies to what you are reading now. These lessons are written from a curated corpus, which means someone chose what to include; the evidence tiers are assigned by that same curation; and a teaching model is a simplification with the errors that simplification implies. Where a lesson names an uncertainty, that is not a disclaimer bolted on at the end — it is part of the content, and the most useful thing you can do with any lesson here is find the place where its account stops being supported.

What does the evidence say?

These are Atlas studies, with the Atlas's own evidence tier. Each card links to the full record — nothing here restates it.

DAn ATP-competitive mammalian target of rapamycin inhibitor reveals rapamycin-resistant functions of mTORC12009

Dropped a bombshell: rapamycin does NOT fully block mTORC1.

Study page →

DProlonged rapamycin treatment inhibits mTORC2 assembly and Akt/PKB2006

The molecular explanation for rapamycin's dark side.

Study page →

CRibosomal protein S6 kinase 1 signaling regulates mammalian life span2009

Deleting S6K1 (a direct mTORC1 effector) extended lifespan in FEMALE mice (+19% median); the effect was not significant in males.

Study page →

CA randomized controlled trial to establish effects of short-term rapamycin treatment in 24 middle-aged companion dogs2017

A short 10-week course of low-dose rapamycin improved heart function measures in healthy pet dogs with no clinical side effects.

Study page →

BInfluence of rapamycin on safety and healthspan metrics after one year: PEARL trial results2025

First completed long-term RCT of rapamycin for healthy human aging (NCT04488601, 48 weeks, n=114).

Study page →

ATargeting ageing with rapamycin and its derivatives in humans: a systematic review2024

The first systematic review of rapamycin/rapalogs in humans for aging.

Study page →

This lesson teaches habits rather than findings, and the habits themselves are a curator's judgement about what has gone wrong most often in this particular literature. Someone reading the same corpus with a different background would emphasise different failure modes, and the studies used as examples were chosen to be illustrative rather than by any systematic procedure.

Work through it

Each of these asks you to commit to something — a prediction, a reading of two studies, a design — before it answers. Everything here is in the page, so nothing is lost if you would rather just read it.

Compare the evidence

Two randomised trials, two species

	C A randomized controlled trial to establish effects of short-term rapamycin treatment in 24 middle-aged companion dogs	B Influence of rapamycin on safety and healthspan metrics after one year: PEARL trial results
Model system	Companion dogs (client-owned, RCT)	Humans, RCT, ages 50-85 (n=114 completed)
Perturbation	A randomised controlled trial in client-owned companion dogs: ten weeks of low-dose rapamycin versus control.	A randomised controlled trial in people aged 50–85, 48 weeks, 114 completing, with a pre-registered primary endpoint.
Readout	Measures of heart function, improved in the treated group, with no clinical side effects reported over that period.	Visceral fat by DXA: no significant change (p=0.942), a null result on the endpoint the trial was designed to answer.

What do both studies support?

That the question can be asked with a real control group in a real population, and that both fields have moved past single-arm observation. Both are randomised; both report what they set out to measure.

Where do they differ?

In what a positive and a null result each buy. The dog trial is tier C and outbred, sharing a human environment — stronger for extrapolation than a mouse, and still not a result in people. The human trial is tier B and returned a null on its primary endpoint, which is a real answer to the question it asked and not a failed experiment. Reading the positive one as the headline and the null one as a disappointment is the single most common way this literature gets misreported.

What experiment would help next?

Neither addresses the endpoint the field actually wants, because no one has agreed what a healthspan endpoint should be. That is why the Atlas carries a biomarker-panel surrogate as an open question — the design problem sits upstream of the trials.

Scientific caution

What this lesson itself supports

What this evidence supports

That the four-part anatomy — model, perturbation, readout, claim — applies to every result in this corpus
That several widely repeated summaries in this field compress a protocol, a species or a sex into a sentence that drops it

What it does not establish

That the six habits here are the complete set, or that they were chosen systematically
That a paper failing one of these checks is wrong — most of the examples used here are landmark studies
That the Atlas itself escapes the same scrutiny

Why?

This lesson teaches habits rather than findings, and the habits are a curator's judgement about what has gone wrong most often in this particular literature. Someone reading the same corpus with a different background would emphasise different failure modes.

Think

Two papers report opposite conclusions about whether rapamycin affects mTORC2. Before deciding that one is wrong, what three features of the experiments would you compare?

Everything in the methods that the abstract left out.

Think first, then reveal

Duration of exposure, first: acute treatment principally affects mTORC1, while prolonged treatment can reduce mTORC2 signalling by interfering with assembly of new complexes, so a six-hour and a six-day experiment can honestly disagree. Cell type second: the assembly effect is not uniform, and complex abundance and turnover differ between cells. Readout third: 'mTORC2 activity' is usually inferred from AKT Ser473, which is also affected by the feedback loop of Lesson 08, so the same blot can move for reasons that have nothing to do with mTORC2 itself. Only after those three match is a genuine contradiction on the table — and this is the ordinary situation in the literature rather than an unusual one.

You read that a trial of an mTOR-directed drug 'improved immune response in older adults' and that another 'showed no benefit'. What would you check first, and what would you refuse to conclude from either alone?

Start with what each trial was designed to answer.

Think first, then reveal

First check what the pre-registered primary endpoint was in each, since a positive secondary finding and a positive primary result are different objects; then the population, the dose and schedule, the duration, and the size. What neither can support alone is a general claim about mTOR inhibition and ageing: one trial is one dose, one schedule, one population and one endpoint. The corpus's own summary of this literature is a systematic review that found improvement in some parameters and not others across 19 studies — and even that is a summary of a small, heterogeneous field rather than a verdict on the strategy.

Check yourself

Three questions, easiest first. Nothing is recorded and there is no pass mark — the explanation after each answer is the part worth reading.

Question 1 · Warm-upIn the Atlas, what does an evidence tier describe?

Show the answer

B — The design of the study — mechanistic, animal, human, or systematic review. Tier is design, not grade. A tier D biochemistry paper and a tier B trial answer different questions, and calling one better than the other is a category error rather than a judgement.

Question 2 · Step upWhy is ‘rapamycin inhibits mTORC1 and spares mTORC2’ a statement about a protocol rather than about the molecule?

Show the answer

B — Because prolonged treatment can also reduce mTORC2 signalling in many cells by interfering with assembly of new complexes. Selectivity here depends on dose and duration. Two studies using the same drug can reach opposite conclusions because they ran different protocols, and neither has to be wrong.

Question 3 · HarderA trial reports a null result on its pre-registered primary endpoint and a promising change in a secondary measure. How should each be read?

Show the answer

C — The primary endpoint is the question the trial was designed and powered to answer; the secondary finding is hypothesis-generating and needs its own test. Pre-registration is what protects a result from being selected after the fact. A null primary endpoint is a real answer to the stated question, and a secondary signal is a reason to design the next study rather than a conclusion.

Go deeper

S6K1 4E-BP1 mTORC1 mTORC2 Rapamycin Longevity Open question: Sex dimorphism in mTOR-longevity responses is pervasive and sometimes direction-flipping Open question: A sensor/autophagy biomarker panel as a surrogate endpoint for human trials

Follow a guided route through the mechanism:

Why doesn't rapamycin switch mTOR off completely?The drug whose protocol details this lesson keeps returning to.
Where does this pathway stop being known?Where the corpus itself says the evidence stops.

Concepts introduced in this lesson

Evidence tiers as design, not grade · Reagent-defined concepts · Primary versus secondary endpoints · Sex-specific effects · Extrapolation across species

← mTOR and AutophagyBack to the curriculum →