Guided routes

A cell cannot start building unless the amino acids are actually present. But amino acids are small molecules with no receptor on the cell surface – so how does the cell measure something it cannot bind from outside?

Follow the highlighted spine: leucine switches off a brake (Sestrin2), which switches off a second brake (GATOR2 on GATOR1), which frees the Rag GTPases to drag mTORC1 to the lysosome. Nothing here activates the kinase – the amino-acid route only decides where mTORC1 sits. Rheb does the switching on.

1. 1. Leucine binds Sestrin2 – and switches a brake off. mechanism ↓

   Sestrin2 carries a pocket that fits leucine with roughly 20 µM affinity. That number is the whole argument: it sits inside the range over which leucine inside a real cell actually rises and falls, so Sestrin2 changes state when leucine changes, rather than being permanently full or permanently empty.

   Leucine-loaded Sestrin2 can no longer hold onto GATOR2. Nothing has been switched on yet – something has been let go of. GATOR2 is now free. Watch what it does with that freedom: it does not activate anything either. It inhibits the next brake.

   Certainty: The binding is structurally resolved and the affinity measured in vitro. What is not established is whether the same 20 µM setpoint holds in tissues with different leucine transport – so this is high mechanistic confidence with unproven human physiological calibration.

   This is where the pathway's logic starts being counter-intuitive. The cell does not detect food and then send a 'grow' signal. It detects food and stops sending a 'do not grow' signal. Most nutrient inputs drawn in this map work that way – count the double negatives yourself rather than taking the word 'almost every' on trust – and it is the reason the pathway is so hard to read off a diagram of arrows.

2. 2. GATOR2 is released – the first brake comes off. mechanism ↓

   Sestrin2 and GATOR2 are mutually exclusive binding partners: leucine-bound Sestrin2 lets go, and free GATOR2 becomes able to act on GATOR1.

   GATOR2 goes from sequestered to available. Its availability, not its abundance, is what the cell regulates. Available GATOR2 now inhibits GATOR1. Count the negatives as you go – you are two into a chain of them.

   Certainty: Mechanistically solid and reproduced. Structures of the GATOR2 cage and its sensor-binding surfaces exist; the cited corpus evidence here is cell-line biochemistry, so human relevance is plausible rather than demonstrated.

   Regulation by sequestration rather than by synthesis is fast and cheap – no transcription, no translation, no degradation. It lets the cell respond to a meal in seconds instead of hours. Evolution reaches for this trick whenever speed matters.

3. 3. GATOR2 shuts down GATOR1 – the second brake comes off. mechanism ↓

   GATOR1 is the machine that switches the Rag GTPases off. GATOR2 inhibits it. So inhibiting GATOR1 means the Rags stop being switched off.

   GATOR1's GAP activity toward RagA/B falls. Two negatives have now cancelled: leucine present → Sestrin2 inhibited → GATOR2 free → GATOR1 inhibited. The Rag GTPases can finally load GTP and stay loaded. That is the state that does something.

   Certainty: Everyone agrees the inhibition happens; nobody has fully resolved how it happens catalytically. This step is graded emerging consensus with medium mechanistic confidence – an honest hole in the middle of a canonical pathway.

   Worth pausing on: this is a textbook step that a textbook will draw as a confident arrow, and the mechanism behind it is genuinely unresolved. A map that hides that is more comfortable and less useful.

4. 4. With GATOR1 suppressed, the Rag GTPases stay loaded with GTP. mechanism ↓

   GATOR1 is a GAP – it forces RagA/B to hydrolyse GTP to GDP. Remove the GAP and RagA/B accumulates in the GTP state, which is its active conformation.

   RagA/B flips from GDP-loaded to GTP-loaded. Note the inversion in this heterodimer: RagA/B is active with GTP, but its partner RagC/D is active with GDP. GTP-loaded RagA/B can now grip Raptor. That grip is what brings mTORC1 in.

   Certainty: The GAP activity is directly demonstrated biochemistry, and human genetics supports its physiological importance – DEPDC5 mutations cause focal epilepsy. The cited corpus evidence is cell-line work, so human relevance is graded plausible.

   Nucleotide state is the pathway's memory. A GTPase holds its answer until something actively changes it, which lets a signal that arrived seconds ago still be true now.

5. 5. Ragulator holds the Rags on the lysosomal membrane. mechanism ↓

   The Rags are not free-floating. Ragulator is lipid-anchored to the lysosome and clamps the Rag heterodimer to that surface, so everything the Rags do, they do at one specific place.

   Nothing about the Rags' activity changes here. What is fixed is their address. Because the Rags are on the lysosome, whatever they recruit arrives on the lysosome too.

   Certainty: The tethering role is well established. Ragulator has also been reported as a GEF for RagA/B; that assignment is less secure than the tethering function, and the model records the tethering claim rather than the GEF claim.

   This is the step that makes the rest of the pathway make sense. Signalling here is not chemistry in free solution – it is a set of mechanisms for putting particular molecules in particular places. Location is the regulated variable.

6. 6. The Rags recruit mTORC1 to the lysosome. They do not switch it on. mechanism ↓

   GTP-loaded RagA/B binds Raptor directly, dragging the whole mTORC1 complex out of the cytosol and onto the lysosomal surface.

   mTORC1's location changes, and only its location. Its kinase activity at this moment is essentially unchanged. Recruitment is not activation – these are two different claims and this map draws them differently on purpose. mTORC1 is now at the surface where the Rheb pool that switches it on in this route is waiting. Meeting Rheb is the event that actually switches it on.

   Certainty: Directly demonstrated and reproduced across labs; the corpus evidence is cell-line biochemistry, so human relevance is graded plausible rather than established.

   If you take one thing from this route, take this: amino acids alone cannot make a cell grow. Starve a cell of growth factors, flood it with leucine, and mTORC1 will sit on the lysosome doing nothing. The nutrient arm answers 'are the parts available?' – it does not answer 'am I allowed to build?'

7. 7. Rheb switches mTORC1 on – and nutrients do not control Rheb. mechanism ↓

   GTP-loaded Rheb binds mTORC1 and physically realigns its active site into a catalytically competent conformation. This is an allosteric activation, a different kind of event from everything upstream in this route.

   mTORC1 becomes an active kinase. Now, and only now, S6K1 and 4E-BP1 start getting phosphorylated. The cell builds. And because Rheb is controlled by the TSC complex, which is controlled by Akt, AMPK and ERK, the growth-factor and energy arms all converge on this single step.

   Certainty: Structurally resolved and mechanistically secure. Cited evidence is mammalian cell work, so human relevance is graded plausible.

   This is coincidence detection, and it is the answer to why the pathway is built the way it is. Two independent conditions – nutrients supplying location, growth factors supplying activation – must both be satisfied at the same place and the same time. A cell that grew on either signal alone would build without materials or build when told not to. The lysosome is where the cell checks both answers against each other.

Key paper: SAN2010 — Reframed nutrient sensing from a chemistry problem into a GEOGRAPHY problem. The Rag–Ragulator complex does not switch mTORC1 on; it moves mTORC1 to the lysosomal surface. Everything about amino-acid sensing turned out to be about location, which is why the answer had eluded people looking for a classical receptor. SAN2008 had already shown the Rags carry the amino-acid signal; this paper said where.

Evidence base: Structural biology and genetic epistasis in human cell lines, plus imaging of mTORC1 translocation. Cell-line work throughout – this arm has no human genetic or clinical evidence in this corpus, which is why almost every step is graded human-relevance *plausible* rather than established.

Still unresolved: How GATOR2 actually inhibits GATOR1 catalytically is still unresolved. Whether Sestrin2's ~20 µM leucine affinity is the operating setpoint in real tissue is untested. And the LARS and glutamine arms remain contested – reproduced in some labs, not others.

Raw material is not permission. A cell in a tissue must not grow just because food is available – it has to be told by the organism that growth is wanted. How does a hormone signal at the cell surface reach a kinase on the lysosome?

Two negatives in a row: Akt disables TSC, TSC stops disabling Rheb, Rheb switches mTORC1 on. Then follow the arrows that come back. mTORC1 switches on S6K1, and S6K1 phosphorylates and represses IRS-1 - cutting the wire between the insulin receptor and PI3K. That is the pathway's principal negative feedback loop, and it is the main reason mTOR inhibitors paradoxically raise Akt activity. Grb10 is a second, parallel arm of the same idea, and mTORC1 inhibition also relieves a brake on MAPK. Note too that ERK enters from the right onto TSC2: growth signalling reaches mTORC1 through Ras-MAPK as well as through Akt, which is why PI3K inhibitors alone rarely close the pathway.

1. 1. A hormone arrives and PI3K is switched on at the membrane. mechanism ↓

   IGF-1 binds its receptor, the receptor autophosphorylates, IRS adaptors dock, and PI3K is recruited to the membrane. This map draws it as one arrow, but it is at least four events.

   PI3K starts converting PIP2 into PIP3 – the cell writes a lipid message into its own membrane. That lipid becomes a docking site. Whatever can read PIP3 will now be pulled to the membrane.

   Certainty: Mechanistically secure and drawn as a long dash precisely because it is compressed. Cited evidence is cell-line work, so human relevance is plausible, not established.

   The cell's answer to 'am I allowed to grow?' does not arrive as a molecule entering the cytosol. It arrives as a change in membrane chemistry. That is why this arm is reversed by a phosphatase rather than switched off by degradation.

2. 2. PIP3 recruits Akt to the membrane – and recruitment is not activation. mechanism ↓

   Akt has a domain that binds PIP3. Arriving at the membrane puts it where two kinases can reach it, but arriving is not the same as being switched on: PDK1 must phosphorylate T308 and mTORC2 must phosphorylate S473.

   Akt's location changes. Its activity changes only once the two phosphorylations happen. Full growth-factor signalling therefore depends on mTORC2 being functional – which is why mTOR sits on both sides of this pathway.

   Certainty: High mechanistic confidence, well replicated, cell-line evidence.

   The same distinction as in the nutrient arm, appearing again: getting a protein to a place is a different act from switching it on. A diagram with one arrow from PI3K to Akt hides three events and one dependency on the other mTOR complex.

3. 3. Akt phosphorylates the TSC complex and takes the brake off. mechanism ↓

   TSC1/TSC2 is the pathway's master brake. Akt phosphorylation inhibits it – partly by changing its activity, substantially by moving it away from where its target sits.

   The brake stops being applied. Nothing has been pushed yet; something has stopped being held back. Whatever the brake was suppressing is now free to act. That target is Rheb.

   Certainty: This is one of the two papers the route's Journey header names as its breakthrough (INO2002). High mechanistic confidence; cell-line evidence, so human relevance plausible – though TSC loss in people is the one place this pathway's causality is established.

   Double-negative logic again, exactly as in the nutrient arm. Growth signals in this pathway overwhelmingly work by removing inhibition rather than adding stimulation. Once you see that pattern you stop being surprised by it.

4. 4. Released from Akt's inhibition, TSC would switch Rheb off – so inhibiting TSC leaves Rheb loaded. mechanism ↓

   TSC2 is a GAP: it forces Rheb to hydrolyse GTP to GDP. With TSC inhibited, Rheb accumulates in its GTP state.

   Rheb flips from GDP-loaded to GTP-loaded. This is the moment the growth-factor signal becomes a switch position. GTP-Rheb can now do what nothing upstream in this route has done: switch the kinase on rather than merely position it.

   Certainty: The second breakthrough paper (INOK2003), with GAR2003 independently. High mechanistic confidence; cell-line evidence.

   Note how far the signal has travelled and how little has been 'activated': a hormone bound a receptor, a lipid was made, a kinase was recruited, a brake was released, and a GTPase changed nucleotide. Four negations and a nucleotide swap. That is what a signalling pathway actually is.

5. 5. GTP-Rheb binds mTORC1 and switches the kinase on. mechanism ↓

   Rheb realigns the mTOR active site into a catalytically competent conformation. This is an allosteric activation – structurally different from everything upstream.

   mTORC1 becomes an active kinase. Now, and only now, its substrates start getting phosphorylated. The cell begins to build. And because this step happens at the lysosome, it can only happen if the nutrient arm has already delivered mTORC1 there.

   Certainty: Structurally resolved, high mechanistic confidence, cell-line evidence.

   This is the convergence point of the whole map. The nutrient arm answers 'are the parts available' by controlling location; this arm answers 'am I allowed' by controlling Rheb. Both must be satisfied at the same membrane at the same time – and this step is where the AND gate is evaluated.

6. 6. Active mTORC1 phosphorylates S6K1. mechanism ↓

   S6K1 carries a TOS motif recognised by Raptor, which presents it to the kinase. T389 phosphorylation activates it.

   S6K1 becomes an active kinase with its own substrates. Two things follow, and they point in opposite directions. S6K1 promotes translation – and it also starts dismantling the signal that created it.

   Certainty: High mechanistic confidence, replicated, and this is the classical rapamycin-sensitive readout. Cell-line evidence.

   S6K1's rapamycin sensitivity is why it became the field's default assay for 'mTORC1 activity' – and why the field systematically overestimated how completely rapamycin inhibits mTORC1 for years. The convenience of a readout shaped what people believed.

7. 7. S6K1 phosphorylates IRS-1 and marks it for destruction. mechanism ↓

   Serine phosphorylation of IRS-1 creates a degradation signal. The adaptor that connected the receptor to PI3K is removed.

   IRS-1 protein levels fall. The input arm of this very route is dismantled. PI3K recruitment drops, Akt activity falls, and the growth signal decays – even though the hormone is still present.

   Certainty: High mechanistic confidence, multiple supporting studies. Cell-line evidence, so human relevance graded plausible, though the clinical consequence is well documented.

   This is negative feedback, and it is the most clinically consequential loop in the pathway. Block mTORC1 with a rapalog and you also block this loop – so IRS-1 survives, PI3K/Akt reactivate, and the tumour you were treating gets a growth signal back. A large part of why rapalog monotherapy underperforms is visible in this single arrow. Note which way round the loop runs: it is mTORC1 ACTIVITY that destroys IRS-1, so the loop is a mechanism of insulin resistance under nutrient excess – the drug relieves it.

8. 8. IRS-1 recruits PI3K – closing the loop back to step one. mechanism ↓

   IRS-1 is the scaffold that brings PI3K to the activated receptor. Its abundance sets how much signal gets through.

   The route returns to where it started. This is not a chain; it is a cycle with a set point. The steady state of growth-factor signalling is determined by the balance between the forward arm and this feedback arm – not by the hormone concentration alone.

   Certainty: High mechanistic confidence; cell-line evidence.

   The question was how a cell learns it is allowed to grow. The answer turns out to be that it never simply learns – it continuously negotiates. The pathway measures its own output and turns its own input down. Any drug that interrupts the loop changes the negotiation, which is why mTOR inhibitors have effects nobody predicted from the linear diagram.

Evidence base: Direct biochemistry and genetic epistasis in mammalian cells, with the TSC arm additionally supported by human disease genetics (tuberous sclerosis complex is the one place this pathway's causality is established in people).

Still unresolved: How much of TSC regulation is phosphorylation changing its activity versus relocation changing its access to Rheb. Which endomembrane pool of Rheb supplies the activating signal. And the relative strength of the two feedback arms (S6K1→IRS-1, mTORC1→Grb10) in any given tissue.

Rapamycin was the drug that discovered this pathway, and for a decade it was treated as *the* mTOR inhibitor. But cells treated with rapamycin keep doing some of the things mTORC1 drives. Why does a drug that clearly hits mTOR fail to stop all of its outputs?

The drug is a two-part molecule: rapamycin alone does nothing, rapamycin+FKBP12 blocks mTORC1 – but only partially, which is why 4E-BP1 stays half-phosphorylated. The dashed arrow to mTORC2 is the one everyone gets wrong: it does not exist acutely, only after days of exposure. That single time-dependence separates the benefit from the side effect.

1. 1. Rapamycin binds FKBP12 – and on its own, does nothing to mTOR. mechanism ↓

   Rapamycin is not an mTOR inhibitor in the way that word is normally used. It has essentially no activity against mTOR by itself. It first binds a small abundant prolyl isomerase, FKBP12, and the drug-protein pair becomes the actual inhibitor.

   A new molecular surface exists that did not exist before: the FKBP12–rapamycin composite. Neither half has that surface alone. Because the inhibitor is a complex, how much inhibition a cell experiences depends on how much FKBP12 that cell expresses – not only on drug concentration.

   Certainty: Structurally resolved and mechanistically secure. Cited evidence is cell-line and structural work, so human relevance is graded plausible rather than established.

   This is the first clue that rapamycin will behave oddly. A drug that must borrow a host protein to work is a drug whose potency varies with the host. It also explains why FKBP12 expression is a determinant of rapalog sensitivity – a fact with no analogue in ordinary ATP-competitive inhibitors.

2. 2. The complex binds the FRB domain and partially blocks the substrate channel. mechanism ↓

   It does not enter the active site. It docks on a domain adjacent to it and gets in the way of substrates arriving. That is a different kind of inhibition from occupying the catalytic pocket – it is steric obstruction, and obstruction can be partial.

   mTORC1 remains a catalytically intact kinase. What changes is which substrates can still reach it. Substrates that need deep, sustained access lose out. Substrates that need less access carry on. The pathway does not switch off – it becomes selectively deaf.

   Certainty: Structurally resolved, high mechanistic confidence. The clinical consequences are supported by trial evidence; this mechanism is not human data.

   Here is the answer to the route's question, and almost the whole field missed it for a decade. Because rapamycin obstructs rather than occupies, S6K1 phosphorylation collapses while 4E-BP1 phosphorylation largely survives. Every experiment that used S6K1 as 'the mTORC1 readout' therefore over-reported how much rapamycin inhibits mTORC1.

3. 3. One output rapamycin does release: mTORC1 stops holding ULK1 down. mechanism ↓

   mTORC1 phosphorylates ULK1 on S757, which blocks ULK1 from being activated by AMPK. Inhibit mTORC1 and that block lifts.

   ULK1 becomes available to AMPK. Autophagy initiation is no longer suppressed. The block on autophagy initiation is lifted. How much recycling actually follows is a separate question, and the answer is less than the textbook suggests.

   Certainty: Direct biochemistry, replicated across labs; mechanistic confidence high, evidence from mammalian cells.

   Here the route has to correct a story it would be easy to tell. THO2009 – the paper this route names as its breakthrough – found autophagy among the mTORC1 outputs that rapamycin leaves largely intact, and ATP-competitive inhibitors induce it far more completely. So rapamycin is partial on BOTH arms, and what it does is not turn 'mTORC1 activity' down by a fixed amount but reshape which outputs stay on. That reshaping is why rapamycin can extend lifespan in mice while being a mediocre anti-proliferative in many tumours – and why 'rapamycin induces autophagy' needs the dose, the cell type and the readout attached before it is true.

4. 4. ULK1 initiates autophagy. mechanism ↓

   Freed and phosphorylated by AMPK, ULK1 nucleates the machinery that builds an autophagosome.

   Bulk degradative recycling begins: damaged proteins and organelles are captured and delivered to lysosomes. The cell buys time and materials. Most of the healthspan claims made for rapamycin route through this step.

   Certainty: Mechanistically solid. But autophagic FLUX is genuinely hard to measure in tissue rather than cells, so quantitative in vivo claims about how much autophagy a given rapamycin dose produces are weaker than they sound.

   This is where the rapamycin story usually stops being told carefully. 'Rapamycin induces autophagy therefore it extends lifespan' skips the part where nobody has cleanly shown autophagy is the required mediator in a mammal.

5. 5. Given long enough, rapamycin also disturbs mTORC2 – sometimes. mechanism ↓

   Chronic exposure can interfere with mTORC2 assembly in some cell types. This is not the acute, direct inhibition seen with mTORC1; it is a slower, indirect effect on complex integrity.

   In susceptible cells, mTORC2 output falls. In others, it does not. The clean textbook statement 'rapamycin inhibits mTORC1 but not mTORC2' is true acutely and unreliable chronically – which matters enormously, because patients take rapalogs chronically.

   Certainty: Contested, low mechanistic confidence, a single supporting study in this corpus. Cell type and duration both change the answer. This is drawn as a dashed line with an amber halo for exactly that reason.

   A route that taught only the tidy version would be teaching a fact with a hidden expiry date. The honest position is that acute and chronic rapamycin are different drugs pharmacologically, and most of what people 'know' about rapamycin comes from acute experiments.

6. 6. Losing mTORC2 is one route to insulin resistance. mechanism ↓

   mTORC2 phosphorylates Akt on S473. Reduce that and insulin signalling degrades, which in mice produces measurable glucose intolerance.

   Whole-body glucose handling worsens – an organism-level consequence, not a cellular one. This is the leading mechanistic explanation for the dysglycaemia seen in patients on rapalogs.

   Certainty: Mouse data (A – animal), medium mechanistic confidence. In humans the relative contributions of mTORC2 loss, S6K1–IRS-1 feedback and direct beta-cell effects are unresolved – so attributing the clinical side effect to this one mechanism overstates what is known.

   The most common serious side effect of the drug may be caused by the arm the textbook says the drug does not hit. If chronic rapamycin does reach mTORC2, then the 'selective mTORC1 inhibitor' framing is not just imprecise, it is clinically misleading.

7. 7. And still, rapamycin extends lifespan in mice. mechanism ↓

   Reproducibly, across genetically heterogeneous strains, at multiple independent sites, including when started late in life.

   Median and maximum lifespan increase. This is one of the most robust pharmacological longevity results in mammals. Everything upstream in this route – partial inhibition, asymmetric outputs, possible mTORC2 disruption, insulin resistance – is the mechanism this outcome sits on. The outcome is solid; the causal chain is not.

   Certainty: Strong for mice: replicated, multi-site (A – animal). For humans: there is no lifespan data of any kind. Human relevance is graded untested, and that grade is not pessimism, it is arithmetic.

   The question this route asked was why rapamycin does not switch mTOR off completely. The answer may be why it works at all. A complete mTOR shutdown is lethal; partial, asymmetric inhibition that suppresses growth signalling while permitting recycling may be exactly the therapeutic window – achieved by accident, through a drug that obstructs rather than occupies.

Key paper: THO2009 — Built an ATP-competitive inhibitor and used it as a ruler. Comparing it against rapamycin exposed a whole class of rapamycin-RESISTANT mTORC1 outputs – most importantly 4E-BP1 phosphorylation, which rapamycin barely touches while collapsing S6K1. That single comparison explained a decade of confusing results and launched the second-generation inhibitor programme that reached trials by 2025.

Evidence base: Pharmacological comparison plus biochemistry in cell lines, with the structural basis (FKBP12–rapamycin occluding the substrate channel rather than the active site) resolved separately. The clinical consequence is supported by trial evidence; the mechanism is not human data.

Still unresolved: Whether chronic rapamycin genuinely disrupts mTORC2 is contested and appears to be cell-type and duration dependent. How much the 4E-BP escape matters in any particular tumour is unresolved, which is precisely the question bi-steric inhibitors are being trialled to answer.

'Promotes growth' is not a mechanism. If mTORC1 switching on has consequences, those consequences are specific molecules being made and specific processes being stopped. Which ones – and does mTORC1 turn everything up equally?

Follow the spine: mTORC1 releases 4E-BP1, 4E-BP1 lets go of eIF4E, translation runs, muscle is built. Then look at what does not follow that line. Rapamycin blocks S6K1 almost completely but 4E-BP1 only partly - which is why 'mTORC1 activity' measured as p-S6K can look abolished while half the real output carries on. And the two lifespan arrows point in opposite directions: losing S6K1 extends life, keeping 4E-BP active extends life. One node is not the pathway.

1. 1. mTORC1 phosphorylates 4E-BP1 and releases a brake on translation. mechanism ↓

   4E-BP1 sits on eIF4E and prevents it from starting translation. Multi-site phosphorylation by mTORC1 makes 4E-BP1 let go.

   eIF4E becomes available. Note the direction: a phosphate was ADDED, and the effect is to STOP an inhibitor – phosphorylation is a mechanism, not a sign. Cap-dependent translation initiation can begin.

   Certainty: High mechanistic confidence, well replicated; cell-line evidence.

   This one substrate carries more consequence than any other in the pathway, because it is only PARTLY rapamycin-sensitive. That single property explains the rapalog/Torin discrepancy and motivated the entire second-generation inhibitor programme.

2. 2. Free of 4E-BP1, eIF4E can bind eIF4G. mechanism ↓

   4E-BP1 and eIF4G compete for the same surface on eIF4E. Removing one lets the other bind – competitive inhibition, not enzymatic.

   The initiation complex can assemble on capped mRNA. Ribosomes begin loading. The cell starts making protein.

   Certainty: High mechanistic confidence, structurally understood; cell-line evidence.

   Competition is a distinct mechanism from catalysis, and it behaves differently: it is concentration-sensitive and instantly reversible. That is why 4E-BP:eIF4E stoichiometry matters as much as mTORC1 activity, and why tissues with different 4E-BP levels respond differently to the same drug.

3. 3. Translation increases – but not uniformly. mechanism ↓

   Ribosome profiling showed mTORC1 does not raise all translation equally. It selectively promotes a specific class of transcripts.

   The composition of what the cell is making changes, not just the amount. Which proteins increase determines which phenotype follows – and those transcripts are enriched for growth and invasion programmes.

   Certainty: This is the route's breakthrough paper (HSI2012). High mechanistic confidence; cancer cell lines, so human relevance plausible.

   'mTORC1 increases protein synthesis' is the summary that hides the actual biology. The regulated variable is transcript CHOICE. Anyone reasoning about mTOR from the summary will predict the wrong consequences, because a uniform increase and a selective one have different phenotypes.

4. 4. In muscle, that translation supports hypertrophy. mechanism ↓

   Load-driven growth requires mTORC1: Raptor-null muscle is dystrophic, and rapamycin blocks overload-induced hypertrophy.

   Muscle fibres grow – over days, not minutes. This is one of the few places where the pathway's output has been tested in people.

   Certainty: The strongest human evidence in this route: DRU2009 is an H – human – interventional study showing rapamycin blocks the contraction-induced increase in muscle protein synthesis. Human relevance established, not merely plausible.

   Worth pausing on, because it is rare. Most of this map is graded human-relevance plausible on cell-line evidence. Here a human intervention closes the loop. It also carries a caveat: mTORC1 activation is NECESSARY for healthy hypertrophy but not sufficient – constitutive activation alone does not build good muscle.

5. 5. At the same time, mTORC1 is holding recycling down. mechanism ↓

   Phosphorylation of ULK1 on S757 blocks the AMPK–ULK1 interaction, preventing autophagy initiation.

   Autophagy is suppressed while building proceeds. The two arms are reciprocal by design: the cell does not build and demolish simultaneously.

   Certainty: High mechanistic confidence, replicated; cell-line evidence.

   This is the answer to what mTORC1 firing actually DOES, stated properly: it is not one action but a coordinated switch between two mutually exclusive programmes. Any account that lists only the build side has described half a switch.

6. 6. Release mTORC1 and autophagy resumes. mechanism ↓

   Unblocked ULK1 nucleates autophagosome formation, and TFEB – released from mTORC1 phosphorylation – transcribes the genes to sustain it.

   The cell shifts from building to recycling, at both the initiation and the transcriptional level. Materials are regenerated. Over longer timescales this arm is where most healthspan claims for mTOR inhibition are made.

   Certainty: High mechanistic confidence for initiation. Autophagic flux in tissue is genuinely hard to measure, so in vivo quantitative claims are weaker than the mechanism.

   Two independent control points – a kinase switch in minutes and a transcriptional programme in hours – on the same process. That is how the pathway gets both a fast response and a sustained one out of a single input.

7. 7. And deleting one output extends lifespan – in female mice. mechanism ↓

   S6K1-null mice live longer and resist age-related pathology, which is the cleanest genetic evidence that a specific mTORC1 output influences lifespan.

   Median lifespan increases, along with metabolic protection. It suggests the longevity effect of mTOR inhibition can be traced to particular outputs rather than to 'less mTOR' in general.

   Certainty: Mouse genetics (A – animal), medium mechanistic confidence, human relevance untested. And the effect is SEX-SPECIFIC – reported in females – a qualification routinely dropped when this result is cited.

   The route asked what a cell does when mTORC1 fires. The honest ending is that we can trace it from a kinase to a phosphosite to a translational programme to a phenotype in one sex of one species – and that no step of that chain has been demonstrated in a human. The map is strongest at the top and weakest exactly where people most want to use it.

Key paper: HSI2012 — Answered the second half, which almost everyone had assumed away. Ribosome profiling showed mTORC1 does not raise translation uniformly – it selectively promotes a specific class of transcripts. 'mTORC1 increases protein synthesis' turned out to be a summary that hides the actual biology, which is transcript choice.

Evidence base: Ribosome profiling and biochemistry in cancer cell lines for the selectivity; genetic knockouts in mice for the phenotypic arms (muscle, mitochondria, lipid). The output-to-phenotype steps are the weakest links in the route, and they are graded accordingly.

Still unresolved: How much of the mTOR-responsive phosphoproteome is functionally relevant rather than incidental. Which outputs matter for which phenotype – the map draws mTORC1 to muscle growth and to longevity, but these are not the same kind of claim and the second has no human evidence at all.

Building is expensive. A cell that starts a growth programme it cannot fuel will damage itself. So there must be a way for energy status to override a growth instruction that has already been given – and it has to work even when the usual brake is broken.

Three different stresses, three different speeds. AMPK reads falling ATP in seconds. REDD1 needs low oxygen to switch on a gene, so it is minutes. The integrated stress response is slower still. All three converge on the same two brakes - and notice AMPK does not only lift the brake on autophagy, it pushes ULK1 directly. That is the pathway doing arithmetic rather than relaying. One caveat on the left-hand corner: metformin is drawn into AMPK because that is the route it is usually credited to, but the dashed arrow running past AMPK straight to mTORC1 is real - metformin still works in AMPK-null and LKB1-null tissue. How the two routes divide the labour in people is unresolved.

1. 1. Falling energy charge activates AMPK directly. mechanism ↓

   AMP and ADP bind the AMPK gamma subunit, which both activates the kinase allosterically and protects its activating phosphorylation from being removed. The cell is not reading 'low ATP' – it is reading the RATIO.

   AMPK becomes active within seconds of the energy charge dropping. A kinase is now running whose entire job is to stop expensive processes and start cheap ones.

   Certainty: High mechanistic confidence, well replicated; cell-line evidence, so human relevance plausible.

   Reading a ratio rather than an absolute is what makes this a sensor rather than a thermometer. A cell with genuinely low but stable ATP is not in trouble; a cell whose ATP is falling is. The ratio distinguishes them.

2. 2. AMPK phosphorylates and activates the TSC complex. mechanism ↓

   Where Akt phosphorylation inhibited TSC, AMPK phosphorylation at different sites activates it. The same brake, driven in the opposite direction by a different kinase.

   TSC GAP activity rises. Rheb starts being switched off. The growth-factor signal is overridden. A cell that was told to grow can now refuse.

   Certainty: High mechanistic confidence, though on a single M – molecular – study in this corpus – the validator flags it, and it is fair to note that the corpus here is thinner than the literature.

   Two opposing inputs converge on one protein, and TSC becomes the place where 'permitted' and 'affordable' are reconciled. Integration in this pathway is not a special mechanism; it is several kinases writing to the same substrate.

3. 3. Activated TSC drives Rheb back to its GDP state. mechanism ↓

   Same GAP reaction as in the growth-factor route, running the other way because TSC is now active rather than inhibited.

   GTP-Rheb falls. The mTORC1 on-switch is being withdrawn. mTORC1 activity declines even if nutrients are plentiful and hormones are still signalling.

   Certainty: High mechanistic confidence; cell-line evidence.

   Energy status wins. Of the four inputs on the overview diagram, this is the one that can veto the others – which makes biological sense, because a cell that builds without fuel destroys itself.

4. 4. With Rheb off, mTORC1 goes quiet. mechanism ↓

   No GTP-Rheb, no allosteric activation, no active kinase – regardless of where mTORC1 is sitting.

   mTORC1 stops phosphorylating its substrates. Building stops. But stopping growth is only half of what an energy-starved cell needs. It also needs to generate resources.

   Certainty: Structurally resolved; high mechanistic confidence.

   This is the same step the growth-factor route ended on, reached from the opposite direction. Seeing one node arrived at by two different arms is how the map teaches convergence – and why 'mTORC1 activity' is never explained by a single upstream signal.

5. 5. In parallel, AMPK phosphorylates ULK1 and switches recycling on. mechanism ↓

   AMPK acts on ULK1 directly, at sites distinct from the inhibitory site mTORC1 uses. And with mTORC1 now quiet, the mTORC1 block on ULK1 has lifted too.

   Autophagy initiation is both released and actively driven – two independent pushes in the same direction. The cell starts digesting its own components to regenerate substrates.

   Certainty: Direct biochemistry, replicated for the phosphorylation itself; the net sign is contested. Park, Lee and Kim (Nat Commun 2023, PMID 37225695) report that under glucose starvation and mitochondrial energy stress AMPK restrains ULK1 activation while protecting the ULK1 machinery.

   This is the elegant part of energy sensing, in its classic reading. One kinase performs both halves of the switch: it stops the expensive programme and starts the recovery programme, simultaneously, without needing a second sensor. Note also that ULK1 phosphorylates AMPK back – so this is a loop with a set point, not a one-way command.

6. 6. Selectively, damaged mitochondria are recycled. mechanism ↓

   AMPK promotes mitophagy, the targeted autophagy of mitochondria – which is both a quality-control mechanism and a way to reclaim material.

   Dysfunctional mitochondria are cleared rather than left to leak. Over longer timescales this shapes mitochondrial quality, and it is one of the arms through which energy stress is proposed to influence ageing.

   Certainty: Medium mechanistic confidence, indirect, and measured largely with reporter mice – so quantitative claims are model-bound. Mouse evidence, human relevance plausible at best.

   The route began with a question about affordability and ends with quality control. That is not a digression: a cell short of energy is usually a cell with failing mitochondria, so the same signal that stops growth is the right signal to trigger repair of the cause. Energy sensing is not a thermostat – it is a diagnostic.

Key paper: GWI2008 — Found the arm nobody expected: AMPK phosphorylates Raptor directly, inhibiting mTORC1 without going through the TSC complex at all. That explained why TSC2-null cells still shut down under energy stress, and it established that this pathway has redundant brakes rather than one master switch.

Evidence base: Direct biochemistry and genetic epistasis in mammalian cells, with the two AMPK arms separable using TSC-null lines. LKB1 dependence means cell lines lacking LKB1 cannot mount the response at all – a boundary condition that invalidates naive comparison across cell types.

Still unresolved: The relative weight of the TSC2 arm versus the Raptor arm in intact tissue is not resolved, and it is cell-type dependent. The metformin route is genuinely contested: several mechanisms are proposed, and the concentrations used in vitro often exceed what clinical dosing achieves.

mTOR is a single protein, yet it does two jobs that respond to different signals, sit in different places, and have different drug sensitivities. Why did evolution not simply use two kinases – and how do you study one of two jobs when your only tool inhibits the other?

mLST8 is the clean argument that these really are two machines: delete it and mTORC2 dies while mTORC1 carries on. Everything below mTORC2 was discovered because rapamycin failed to block it - the actin cytoskeleton first. And the dashed arrow from rapamycin is the whole clinical problem in one line: absent acutely, present after days.

1. 1. Rictor binds mTOR and defines a second complex. mechanism ↓

   The same catalytic subunit, a different partner. Rictor takes the place Raptor occupies in mTORC1, and the resulting complex has different substrates, a different location and – decisively – different drug sensitivity.

   There are now two mTOR complexes in the cell, not one kinase with two moods. Because Rictor confers rapamycin insensitivity, this complex was invisible for a decade to anyone using rapamycin as their probe.

   Certainty: This is the route's breakthrough paper (SAR2004). High mechanistic confidence, biochemistry and complex purification in mammalian cells.

   The answer to the route's question starts here. Evolution did not need two kinases because the catalytic domain is not what specifies a signalling job – the partner is. Substrate choice, location and regulation all come from the accessory subunit, so one kinase gene can serve two pathways.

2. 2. SIN1 joins, and brings a growth-factor antenna with it. mechanism ↓

   SIN1 is required for complex integrity and for Akt S473 kinase activity. Its PH domain inhibits mTORC2 until PIP3 relieves that inhibition.

   mTORC2 becomes assembled, competent, and responsive to membrane lipid state. The complex now has a way to know whether growth factors are present – through the same PIP3 signal Akt uses.

   Certainty: High mechanistic confidence, multiple studies including structural work.

   A subunit doubling as a sensor is an economical piece of design: the same lipid that recruits the substrate also licenses the kinase. It also means PI3K sits upstream of both arms, which is why PI3K inhibition has broader consequences than mTOR inhibition.

3. 3. PIP3 relieves the SIN1 brake and mTORC2 becomes active. mechanism ↓

   Growth-factor-generated PIP3 engages the SIN1 PH domain, releasing its autoinhibition of the complex.

   mTORC2 activity rises in response to growth factors – on a seconds timescale. Both mTOR complexes are now downstream of PI3K, but they read it differently: mTORC1 through Akt→TSC→Rheb, mTORC2 through this direct lipid relief.

   Certainty: Medium mechanistic confidence, emerging consensus, one supporting study in this corpus. Drawn as a long dash because it is a compressed relay, not a single event.

   This is the cleanest available answer to how growth factors reach mTORC2, and it is weaker evidence than the equivalent step in the mTORC1 arm. Worth noticing: the two complexes are not equally well understood, and the map shows that asymmetry rather than smoothing it over.

4. 4. mTORC2 phosphorylates Akt on S473. mechanism ↓

   This is mTORC2's signature reaction. Akt needs both T308 from PDK1 and S473 from mTORC2 for full activity against many substrates.

   Akt becomes fully active – and Akt is what activates mTORC1's upstream arm. Through Akt, mTORC2 is one of the inputs that licenses mTORC1. That is not a simple hierarchy, though: both complexes read PI3K, each has substrates the other never touches, and mTORC1's own S6K1 to IRS-1 feedback runs back into the signalling that feeds mTORC2. Upstream here names a route, not a rank.

   Certainty: High mechanistic confidence, multiple studies, and the genetic dissection in mice is the strongest evidence in this route: Rictor or mLST8 loss abolishes signalling to Akt while sparing S6K1.

   Here is the structural reason the two-complex question matters clinically. Rapamycin hits mTORC1 but not mTORC2 acutely, so it leaves the Akt-activating arm intact – one more reason blocking mTORC1 does not simply shut the pathway down.

5. 5. Disrupting mTORC2 degrades whole-body glucose handling. mechanism ↓

   Less S473 phosphorylation means weaker insulin signalling, which in mice produces measurable glucose intolerance.

   An organism-level metabolic phenotype appears, from a change in one complex. This is the leading explanation for the dysglycaemia patients experience on chronic rapalogs.

   Certainty: Mouse data (A – animal), medium mechanistic confidence. In humans the relative contributions of mTORC2 loss, S6K1–IRS-1 feedback and direct beta-cell effects are unresolved.

   A complex that was invisible because the standard drug did not hit it turns out to explain that drug's most common serious side effect. That is a strong argument for the Atlas's central habit: knowing which arm a claim rests on, and on which species.

6. 6. And chronic rapamycin may reach mTORC2 after all. mechanism ↓

   Prolonged exposure can interfere with mTORC2 assembly in some cell types – slowly, indirectly, and not universally.

   The clean separation that made mTORC2 discoverable becomes unreliable over time. The textbook line 'rapamycin inhibits mTORC1 but not mTORC2' is a statement about acute treatment being applied to chronic therapy.

   Certainty: Contested, low mechanistic confidence, a single supporting study. Cell type and duration both change the answer – which is why this arrow is dashed with an amber halo.

   The route closes on an irony worth sitting with. Rapamycin insensitivity is the property that revealed mTORC2 existed; that same property may not hold under the conditions in which the drug is actually used. The tool that made the discovery possible may not describe the therapy.

Key paper: SAR2004 — Identified Rictor and, with it, a second mTOR complex that is raptor-independent and – decisively – rapamycin-insensitive. That last property is what made mTORC2 studiable at all: it gave the field a way to separate the two jobs experimentally, using the very drug that had previously hidden one of them.

Evidence base: Biochemistry and complex purification in mammalian cells, then genetic dissection in knockout mice (Rictor and mLST8 loss abolishes signalling to Akt and PKCα while sparing S6K1). The mouse genetics is the strongest evidence in this route.

Still unresolved: Whether prolonged rapamycin disrupts mTORC2 assembly is contested. The mTORC2-to-insulin-resistance link rests on mouse data, and in humans the relative contributions of mTORC2 loss, S6K1–IRS-1 feedback and direct β-cell effects are unresolved.

Forty years of mechanism is not a treatment. If mTORC1 drives growth and we have drugs that inhibit it, where does that convert into benefit for a person – and where does it conspicuously fail to?

Read this route next to the amino-acid one and the contrast is the point. There, almost every badge is D. Here they are B - randomised human trials, one of them (temsirolimus in renal cancer) on overall survival. The pathway is not unproven in people. Watch the two arrows into LAM, though: everolimus was tested against the kidney lesion in EXIST-2, while the lung disease itself was treated with sirolimus in MILES. They are different drugs and different endpoints, and collapsing them is a common error. What remains unproven is the longevity claim: every arrow on this page ends in a tumour or an immune endpoint, and not one of them ends in lifespan.

1. 1. Lose the TSC brake and mTORC1 runs unopposed. mechanism ↓

   TSC never touches mTORC1 – it acts on Rheb. This arrow is a deliberate two-step compression, drawn as one link so the clinical story reads cleanly.

   Without TSC, Rheb stays GTP-loaded and mTORC1 stays on even when no growth factor says it should. Note what this does not mean: TSC2-null cells still need amino acids and still respond to energy stress, so the input that has been lost is the growth-factor one, not every input. Growth signalling becomes constitutive rather than conditional.

   Certainty: High mechanistic confidence, and this is the one place in the whole map where human genetics establishes causality: TSC1/TSC2 loss causes disease in people.

   Every other arm of this pathway is inferred from cells and mice. This one is inferred from patients. That difference is why tuberous sclerosis is the setting where mTOR inhibitors work best – the drug is aimed at the actual cause.

2. 2. Constitutive mTORC1 supports tumour growth. mechanism ↓

   Sustained translation of growth and invasion programmes, plus suppressed autophagy, plus the biosynthetic outputs – mTORC1 supplies much of what a proliferating cell needs.

   Proliferation and mass increase. That makes mTORC1 a drug target. It does not make it the target in every tumour.

   Certainty: High mechanistic confidence but INDIRECT, and strongly genotype-dependent: a strong dependency in TSC- and PI3K-pathway-mutant contexts, much weaker elsewhere.

   The gap between 'mTORC1 supports tumour growth' and 'inhibiting mTORC1 treats this tumour' is where most of the clinical disappointment of the last twenty years lives. Dependency is contextual; the arrow is not.

3. 3. Everolimus inhibits mTORC1 – partially, and via FKBP12. mechanism ↓

   Same allosteric mechanism as rapamycin, with better oral pharmacokinetics. It obstructs the substrate channel rather than occupying the active site.

   S6K1 signalling collapses; 4E-BP1 phosphorylation substantially persists. The drug delivers partial, asymmetric inhibition to a tumour that may depend on the arm it does not fully block.

   Certainty: High confidence, and unusually for this map, supported by H – human trial – evidence across several indications.

   Carry the rapamycin route's lesson into the clinic. The incomplete inhibition that is a curiosity in a cell-biology paper is a therapeutic ceiling in a patient – and it is the reason bi-steric and ATP-competitive inhibitors reached trials.

4. 4. In tuberous sclerosis, it works. mechanism ↓

   The disease is caused by loss of the brake this drug substitutes for. Mechanism and treatment are matched.

   Tumours shrink, including subependymal giant-cell astrocytoma and renal angiomyolipoma. This is the pathway's clearest mechanism-to-benefit case.

   Certainty: H – trial evidence in humans; human relevance established. Note the honest scope: these are benign tumours and the benefit is control, not cure – treatment interruption is followed by regrowth.

   The best result in the whole map comes from the one disease where the causal lesion is known and the drug addresses it directly. That is the template, and the rest of oncology has struggled to reproduce it precisely because the causal lesion is usually not so clean.

5. 5. In renal cancer the link is association, not demonstrated causation. mechanism ↓

   Renal cancers frequently carry lesions that leave mTORC1 active. That is a correlation between genotype and pathway state, not evidence that mTORC1 activation initiates the disease.

   Nothing mechanistically. This arrow records a statistical relationship. It explains why the tissue responds to rapalogs at all, and why an exceptional responder could be traced to TSC1 loss.

   Certainty: Typed as ASSOCIATION, directness unresolved, mechanistic confidence low – the lowest-graded link on the route, deliberately. Drawn dotted and thin.

   This step exists to be read sceptically. It sits between two well-evidenced clinical steps, and if it were drawn like them a reader would infer a causal chain that the evidence does not support. Grading it honestly is what stops the route from over-claiming.

6. 6. Temsirolimus improved survival in advanced renal cell carcinoma. mechanism ↓

   A randomised trial in poor-prognosis patients – the result that produced the first mTOR inhibitor approval in oncology.

   Overall survival improved versus interferon alfa. mTOR moved from a laboratory pathway to a licensed drug target.

   Certainty: H – a randomised controlled trial; human relevance established. This is trial evidence: it establishes that the drug changed an outcome, NOT that the mechanism drawn upstream is the reason.

   The distinction in that last sentence is the whole point of the route. A positive trial validates a treatment, not a diagram. Everything above this step remains inferred from cells and mice even after the drug is approved.

7. 7. In lymphangioleiomyomatosis, sirolimus stabilised lung function. mechanism ↓

   LAM involves TSC-pathway lesions, so the same mechanistic logic as tuberous sclerosis applies to a progressive lung disease.

   FEV1 decline stabilised during treatment in the MILES trial. A rare, previously untreatable progressive disease acquired a therapy derived from pathway biology.

   Certainty: H – a human trial; human relevance established. Precision matters here: MILES tested SIROLIMUS in the lung disease. EXIST-2 tested everolimus against renal angiomyolipoma, not the lung disease – a distinction routinely blurred, and an external review of this Atlas caught us blurring it.

   Arguably the strongest answer to the route's question. Not a cancer, not a lifespan claim – a specific progressive disease where understanding the pathway produced a treatment that changed the disease course. It is also the example most people have never heard of.

8. 8. And in older adults, low-dose everolimus improved vaccine responses. mechanism ↓

   The same drug class used as an immunosuppressant at transplant doses improved influenza vaccine responses when given intermittently at low dose.

   Immune function improved rather than degraded – the opposite direction from the drug's classical use. It suggests dose and schedule, not the target, determine whether mTOR inhibition suppresses or rejuvenates immunity.

   Certainty: H – human trials – but typed CONTESTED with medium confidence, because the direction of effect depends on dose and schedule and the finding has not been uniformly replicated at scale.

   The route ends where the field currently is. Forty years of mechanism produced clear wins in rare diseases with known causal lesions, partial wins in cancer limited by feedback and incomplete inhibition, and a genuinely open question about whether intermittent low-dose inhibition can improve ageing physiology in people. No human lifespan data exists. That is not a disappointing ending – it is the accurate one, and it is where the next set of trials is aimed.

Evidence base: Randomised controlled trials and one systematic review – the only route in this section built primarily on human evidence (S and H – synthesis of human data and human studies). Note what that buys and what it does not: trials establish that the drug changes an outcome, not that the mechanism drawn upstream is the reason.

Still unresolved: Which tumours depend on mTOR remains largely unpredictable from genotype. There is no human lifespan data of any kind. And the pattern that rapalogs delay progression without clearly extending overall survival in several indications is unexplained – the feedback loops are the leading suspect.

Muscle grows in response to mechanical load, not to a hormone injection. Somewhere a physical force has to become a molecular signal, and then a decision to build. Where does that conversion happen, and how do we know mTORC1 is required rather than merely present?

The one route in this section whose final claim was tested by interrupting it in living people. Follow mechanical load to muscle growth, and notice where the evidence stops being inference.

1. 1. You lift something heavy, and mTORC1 activity rises in the muscle. mechanism ↓

   Mechanical loading raises mTORC1 signalling in skeletal muscle. Rapamycin given to human volunteers before resistance exercise BLOCKS the contraction-induced rise in muscle protein synthesis – so mTORC1 is not merely present during the response, it is required for it.

   A physical force has become a molecular signal. The muscle has committed to building. Everything downstream is the pathway you already know, running on a stimulus that is not a hormone.

   Certainty: The strongest evidence in this entire section: a human interventional study (DRU2009, H – human), supported by rodent genetics and pharmacology (BOD2001). Human relevance ESTABLISHED, not plausible.

   Pause on what kind of evidence this is. Almost every other step in this Atlas says 'we showed this in cells and infer it in people'. Here someone gave a drug to humans, removed the pathway, and the response disappeared. That is an interruption experiment in a person, and it is worth far more than any amount of correlative human data.

2. 2. Part of the signal does travel the familiar growth-factor route. mechanism ↓

   IGF-1 acting through its receptor and PI3K produces myotube hypertrophy (ROM2001). Mechanical load itself is different: in mouse muscle it activated mTOR signalling independently of PI3K and Akt (Hornberger 2004, outside this corpus), so this is the hormonal contribution, not the mechanical one.

   PIP3 accumulates in the membrane. Akt can be recruited, exactly as in the growth-factor route.

   Certainty: High mechanistic confidence, but compressed and drawn as a long dash. Cell and myotube evidence (ROM2001, CAN2002).

   Only PART of the signal goes this way, and that qualifier matters. Load reaches mTORC1 partly independently of circulating IGF-1, which is why you cannot substitute a growth-factor injection for the mechanical stimulus. The mechanosensor upstream of this has not been identified – the honest gap in an otherwise well-evidenced route.

3. 3. PIP3 recruits Akt to the membrane. mechanism ↓

   Akt binds PIP3 and is then phosphorylated by PDK1 and mTORC2. Recruitment is not activation; both phosphorylations are required.

   Akt becomes active at the membrane. The brake downstream can now be released.

   Certainty: High mechanistic confidence; cell-line evidence, human relevance plausible.

   The same recruitment-versus-activation distinction that runs through the whole map. In a muscle context it also means mTORC2 function is quietly required for a hypertrophy response – a dependency invisible in most exercise-physiology accounts.

4. 4. Akt inhibits the TSC complex. mechanism ↓

   Phosphorylation of TSC2 inhibits the complex and moves it away from Rheb.

   The master brake comes off. Rheb stops being switched off.

   Certainty: High mechanistic confidence; cell-line evidence.

   Double-negative logic again: the growth signal in muscle works by removing inhibition, not by adding stimulation. This is why the pathway can respond within an hour of a training set – there is nothing to synthesise, only something to stop doing.

5. 5. Rheb accumulates in its GTP-loaded state. mechanism ↓

   TSC2 is the GAP that forces Rheb to hydrolyse GTP. Inhibited GAP, loaded Rheb.

   The mTORC1 on-switch moves into position. mTORC1 can be activated wherever the two meet.

   Certainty: High mechanistic confidence; cell-line evidence.

   Nutrient state still gates this, through location. A trained muscle in a fasted state does not build – which is the mechanistic basis for the entire field of post-exercise nutrition, visible here as the AND gate the nutrient route describes.

6. 6. mTORC1 is switched on. mechanism ↓

   GTP-Rheb allosterically activates the kinase.

   mTORC1 begins phosphorylating its substrates. The translational machinery is released.

   Certainty: Structurally resolved, high mechanistic confidence; cell-line evidence.

   Convergence point again, reached from a third direction – nutrients, hormones, and now mechanical load all terminate here. That is the strongest argument for why this one node is worth understanding properly: everything that decides whether a cell grows has to come through it.

7. 7. mTORC1 phosphorylates 4E-BP1 and releases the cap on translation. mechanism ↓

   Multi-site phosphorylation makes 4E-BP1 let go of eIF4E.

   eIF4E is free. A phosphate was added and an inhibitor stopped inhibiting. Cap-dependent initiation can proceed.

   Certainty: High mechanistic confidence; cell-line evidence.

   This is the arm rapamycin only partly blocks – and yet in DRU2009 rapamycin abolished the rise in human muscle protein synthesis after a single bout of resistance exercise, measured 1–2 h afterwards. Read the endpoint carefully: that is acute protein synthesis, not muscle gained, and rapamycin also blunted ERK1/2 in the same volunteers, so the block is not cleanly attributable to this step alone. Even so, partial inhibition was enough to abolish a whole-body acute response, which suggests little reserve in it.

8. 8. eIF4E binds eIF4G and the initiation complex assembles. mechanism ↓

   4E-BP1 and eIF4G compete for the same surface. Remove one and the other binds.

   Ribosomes begin loading onto capped mRNA. Protein synthesis rises.

   Certainty: High mechanistic confidence, structurally understood.

   Competition rather than catalysis, so the response is stoichiometric and immediately reversible. Muscle expresses its own balance of 4E-BP isoforms, which is one reason the same training stimulus produces different responses in different people and different fibre types.

9. 9. Muscle protein synthesis increases. mechanism ↓

   Cap-dependent initiation rises, selectively favouring a particular class of transcripts rather than everything equally.

   The rate and the COMPOSITION of protein synthesis both change. Given repeated stimuli and adequate substrate, net protein accretion follows.

   Certainty: High mechanistic confidence; and this is the exact readout DRU2009 measured in humans and found rapamycin-sensitive.

   This is where the human experiment intersects the molecular chain, and it is the reason this route can claim more than the others. The measured variable in the person is the same variable the cell biology predicts.

10. 10. Over days and repeated sessions, the muscle grows. mechanism ↓

    Repeated bouts of elevated synthesis, exceeding breakdown, produce hypertrophy. Raptor-null muscle is dystrophic; rapamycin blocks overload-induced growth.

    Fibre cross-sectional area increases. The adaptation that the training was for.

    Certainty: Human relevance established, supported by mouse genetics and human pharmacology. Timescale: days – this is the slowest step in the route by two orders of magnitude.

    One honest limit to end on. mTORC1 activation is NECESSARY for healthy hypertrophy but not SUFFICIENT: constitutively activating mTORC1 in muscle does not produce good muscle, it produces inflammation and dysfunction. The signal has to be intermittent. That is a general lesson about this pathway – it is a switch that is meant to be thrown, not held.

Key paper: DRU2009 — Did the interruption experiment in people. Human volunteers were given rapamycin before resistance exercise, and the contraction-induced rise in muscle protein synthesis was blocked. Almost everything else in this Atlas is inferred from cells or mice; this placed mTORC1 causally inside a human physiological response. BOD2001 had established the pathway's necessity in rodent muscle, but necessity in a person is a different claim.

Evidence base: A human interventional study for the causal step, rodent genetics and myotube work for the upstream route. This is the only route in the section whose terminal claim carries human-relevance ESTABLISHED rather than plausible – worth noticing, because it is rare here and it is what an interruption experiment buys you.

Still unresolved: How mechanical force is actually transduced into mTORC1 activation is not settled: load reaches mTORC1 partly independently of IGF-1, and the mechanosensor has not been identified. Whether the same dependence holds in ageing muscle, where anabolic resistance appears, is untested. And mTORC1 activation is necessary but NOT sufficient – constitutive activation alone does not build healthy muscle.

Fasting is usually described as a state – fed or fasted. But a cell does not switch states; it runs a sequence, and the parts of that sequence operate on wildly different timescales. Which responses have happened after a few seconds, and which are still hours away?

Not a clock but an ORDER. Set the BY TIME control to seconds, then minutes, then hours, and watch which parts of the response have happened yet. The pathway does not switch state – it unfolds.

1. 1. You stop eating. mTORC1 signalling falls. mechanism ↓

   Two arms detect it at once: the amino-acid sensors stop reporting sufficiency, and AMPK detects the falling energy charge. Neither alone accounts for the drop.

   mTORC1 output declines across all of its substrates. Everything mTORC1 was suppressing is now released – and that release runs on a slower clock than the suppression did.

   Certainty: High mechanistic confidence for the direction. Evidence spans mouse, rhesus and one human safety trial (ROM2016), so human relevance is graded plausible.

   Set the BY TIME control to seconds and look at the canvas before reading on. Almost nothing downstream has happened yet. That is the point of this route: fasting is not a state the cell enters, it is a sequence the cell runs, and the interesting part is the ordering.

2. 2. Within seconds: AMPK activates. mechanism ↓

   AMP and ADP bind the AMPK gamma subunit directly, activating it allosterically and protecting its activating phosphorylation. No transcription, no translation, no new protein.

   An active kinase exists that did not exist a moment ago. The fastest arm of the response is now running.

   Certainty: High mechanistic confidence, direct biochemistry; cell-line evidence.

   Allosteric activation is the fastest control mechanism a cell has, and the pathway spends it on the energy sensor. That is a priority statement: running out of fuel is the emergency that cannot wait for gene expression.

3. 3. Within minutes: AMPK activates the TSC brake. mechanism ↓

   Phosphorylation at sites distinct from Akt's – same substrate, opposite direction.

   TSC GAP activity rises and Rheb starts being switched off. The mTORC1 on-switch is withdrawn.

   Certainty: High mechanistic confidence, though on a single M – molecular – study in this corpus.

   Notice the timescale step you just took: seconds to minutes. AMPK activation and AMPK's effect on mTORC1 are not simultaneous, and treating them as one event is what makes fasting look like a switch rather than a cascade.

4. 4. mTORC1 goes quiet. mechanism ↓

   No GTP-Rheb, no allosteric activation – regardless of where mTORC1 is sitting.

   The kinase stops phosphorylating substrates. Building stops. The brakes mTORC1 was applying to recycling now come off, one substrate at a time.

   Certainty: Structurally resolved, high mechanistic confidence.

   Stopping is the easy half. Everything from here is the cell constructing a recovery programme, and each subsequent step is slower than the last – which is why short and long fasts are not the same intervention on a different scale, but different interventions.

5. 5. Minutes: the block on autophagy initiation lifts. mechanism ↓

   mTORC1 had been phosphorylating ULK1 on S757, preventing AMPK from activating it. With mTORC1 quiet, that block is gone.

   ULK1 becomes available. Availability is not activity. Something still has to switch it on.

   Certainty: High mechanistic confidence, replicated; cell-line evidence.

   Release and activation as separate events, on the same protein, from two different kinases. On this arm the cell has built something close to an AND gate: ULK1-driven initiation needs mTORC1 down AND AMPK up, which stops recycling from firing on a brief dip in either signal. 'Close to', not exactly: autophagy has inputs this map does not draw, so treat the gate as the logic of these two arrows rather than as the whole control of autophagy.

6. 6. And AMPK provides the activating push. mechanism ↓

   AMPK phosphorylates ULK1 at activating sites – the second half of the gate.

   ULK1 is now both released and driven. Autophagy initiation begins in earnest.

   Certainty: Direct biochemistry for the phosphorylation; the net sign is contested (Park, Lee and Kim, Nat Commun 2023, PMID 37225695: under energy stress AMPK can restrain ULK1 while protecting it).

   In the classic reading, one kinase performing both halves of a switch – stopping the expensive programme and starting the recovery programme – is the economy of this design. And ULK1 phosphorylates AMPK back, so the steady state is a set point rather than a command.

7. 7. Autophagy runs. The cell begins digesting its own components. mechanism ↓

   ULK1 nucleates the machinery that captures cargo and delivers it to lysosomes.

   Damaged proteins and organelles are broken down; amino acids are regenerated. The cell buys both time and materials – and the materials feed the very sensors that started this.

   Certainty: High mechanistic confidence for initiation. Autophagic FLUX in tissue is genuinely hard to measure, so quantitative in vivo claims are weaker than the mechanism.

   This closes a loop the Atlas cannot yet draw: autophagy-derived amino acids re-enter the sensing machinery, so a fasting cell is partly feeding itself. It is declared as an open loop in the model, because no curated edge carries that final step.

8. 8. Also in minutes: TFEB is released from the nucleus's doorstep. mechanism ↓

   mTORC1 had been phosphorylating TFEB on S211, trapping it in the cytosol via 14-3-3. Quiet mTORC1 means TFEB is free to enter the nucleus.

   A transcription factor changes compartment – the slow arm has been armed. Gene expression is about to change, which takes hours rather than minutes.

   Certainty: High mechanistic confidence, and substrate-selective: it depends on FLCN/RagC status, so mTORC1 can be active on S6K1 while TFEB escapes.

   The timescale changes character here. Everything before this was post-translational and reversible in minutes. From here the cell is rewriting which proteins exist, and that cannot be undone quickly. Switch BY TIME to hours and watch the rest of the route appear.

9. 9. Hours: TFEB transcribes the autophagy programme. mechanism ↓

   Nuclear TFEB switches on autophagy and lysosomal genes as one coordinated module.

   The cell now has MORE autophagy machinery, not just active machinery. The response becomes sustainable rather than a burst.

   Certainty: High mechanistic confidence; cell-line evidence, human relevance plausible.

   This is the difference between a short fast and a long one, and it is a difference in kind. The first hour reallocates existing machinery. Later hours build more of it. Any claim that a 16-hour and a 48-hour fast do 'the same thing more' is ignoring this step.

10. 10. The lysosomal compartment itself expands. mechanism ↓

    The same TFEB programme drives lysosomal biogenesis – more lysosomes, not just more autophagosomes.

    The organelle on which mTORC1 is regulated is being rebuilt. The cell is remodelling the platform that controls the signal that started all of this.

    Certainty: High mechanistic confidence; and SET2012 established the lysosome-to-nucleus circuit this arm belongs to.

    This closes a real feedback loop, and the model detects it as one: TFEB to lysosomal biogenesis to lysosome to mTORC1 to TFEB. Fasting does not just lower mTORC1 signalling – it changes the machine that does the signalling. That is why refeeding after a long fast is not simply the reverse of fasting.

11. 11. Over hours: damaged mitochondria are selectively cleared. mechanism ↓

    AMPK promotes mitophagy, the targeted autophagy of mitochondria.

    Mitochondrial quality improves rather than merely mitochondrial number falling. This is one of the arms through which fasting is proposed to influence ageing.

    Certainty: Medium mechanistic confidence, indirect, measured largely with reporter mice – so quantitative claims are model-bound. Human relevance plausible at best.

    Quality control rather than accounting. A fasting cell is usually a cell with strained mitochondria, so the same signal that stopped growth is the right trigger for repairing the cause. Note the grade drop: from here the evidence weakens as the claims get more interesting.

12. 12. And over a lifetime, restricted animals live longer. mechanism ↓

    Caloric restriction improved health and survival in rhesus monkeys, and macronutrient composition altered lifespan in mice.

    Median and in some studies maximum lifespan increase. The molecular sequence you just walked is the leading mechanistic account of why.

    Certainty: Medium mechanistic confidence, INDIRECT, rhesus and mouse. Human relevance UNTESTED – there is no human lifespan data. The two large rhesus studies famously disagreed depending on the control diet, and SOL2014 found macronutrient RATIO mattered more than total calories.

    The route asked what happens and in what order. It can answer that with reasonable confidence for the first few hours in a cell, and it cannot answer it at all for a human lifetime. The gap between those two ends of the same arrow is the single most important thing to carry away from this section – and the reason the Atlas grades human relevance separately from mechanism.

Evidence base: Direct biochemistry in cells for the ordering, and its timescale grading comes from the curated timescale field on each interaction rather than from any single time-course experiment. The organism-level end of the route rests on rhesus and mouse data (MAT2017, SOL2014) plus one human safety trial (ROM2016).

Still unresolved: The ORDER here is defensible; the CLOCK is not. This map holds six ordinal timescale buckets, not rate constants, so it can say what has happened by now but never how fast. Human fasting time-courses for these molecular events do not exist in this corpus. And SOL2014 raises the harder question of whether calories or macronutrient ratio is even the right variable.

Every pathway diagram looks equally confident everywhere. This one is not, and it records where. So: which parts of mTOR biology are contested between labs, which have a mechanism nobody has resolved, and which have never been tested in a human at all?

Every other route shows you what the Atlas holds. This one walks the weakest links on purpose: the interactions this Atlas grades as contested, the ones carrying low mechanistic confidence, and the ones with no human evidence in this corpus. Read it as a guide to reading evidence, not as a confession – and read it as a statement about what has been curated here, not about what the field has done.

1. 1. Two labs proposed two different leucine sensors, and the field has not fully closed the question. mechanism ↓

   Sestrin2 binds leucine with an affinity in the range over which intracellular leucine actually fluctuates. LARS, a leucyl-tRNA synthetase, was independently proposed to moonlight as a leucine sensor acting on the Rag GTPases. Both could operate, but they cannot both be the dominant sensor in one cell, and the two proposals have not been reconciled – it is also possible that which one dominates depends on the cell type.

   Nothing in the cell. What changes is how much weight you should put on either arrow. This map draws both, and marks LARS as contested with a dashed line and an amber halo.

   Certainty: Contested consensus, LOW mechanistic confidence, human relevance untested. The LARS model has not reproduced cleanly across labs.

   Start here because it is the cleanest example of the general problem. A pathway diagram that showed only the winning model would be more comfortable and less true; one that showed both without marking which is disputed would be worse still. The honest option is the one that costs a dashed line and an explanation.

2. 2. The proposed mechanism – LARS acting as a GAP for RagD – is the weakest link in the amino-acid arm. mechanism ↓

   A moonlighting GAP function for a tRNA synthetase is an unusual claim, and the reproduction record is mixed.

   Nothing. This is an unresolved mechanism, not an event. If it is wrong, the amino-acid arm is simpler than this map suggests. If it is right, there is a second sensing route nobody has integrated.

   Certainty: Contested, low mechanistic confidence, untested in humans – and the validator flags that no boundary conditions are stated for it, which is itself a curation gap this route is happy to expose.

   Notice that the Atlas is admitting a hole in its own curation here rather than hiding it. A validator warning left visible is more useful than a warning silenced.

3. 3. Glutamine may reach mTORC1 without the Rag GTPases at all. mechanism ↓

   A Rag-independent, Arf1-dependent route has been reported. If real, it means the lysosomal recruitment story is not the only way in.

   Potentially the architecture of the whole nutrient arm. The map draws it dotted and thin, because the claim is large and the reproduction is not uniform.

   Certainty: Contested, low mechanistic confidence, human relevance untested.

   This one matters disproportionately because of what it would overturn. Most contested edges are details; this is a contested claim about whether the central mechanism is complete. Weight of a claim and strength of its evidence are independent, and this step has high weight with low evidence.

4. 4. A step everyone draws confidently has a mechanism nobody has resolved. mechanism ↓

   That GATOR2 inhibits GATOR1 is not in doubt – it is in every textbook figure. HOW it does so catalytically is still argued, even after the GATOR2 structure was determined.

   Nothing. This is a hole in the middle of a canonical pathway. Every account of amino-acid sensing passes through a step whose mechanism is an open question.

   Certainty: Consensus graded EMERGING with medium mechanistic confidence – deliberately not established, despite how confidently the step is usually drawn.

   This is the most instructive step in the route. It is not contested, not obscure, and not weakly evidenced – it is simply unresolved, in a place where the diagram looks finished. Textbook confidence and mechanistic understanding are different things, and a map that grades them the same teaches the wrong lesson.

5. 5. The most-prescribed drug that touches this pathway has a contested mechanism. mechanism ↓

   Complex I inhibition raising AMP, AMPK-independent Rag inhibition, lysosomal PEN2–ATP6AP1 sensing and gut-microbiome effects have all been proposed. HOW2017 showed the mTORC1 effect is dose-dependent and mechanistically plural.

   Nothing mechanistically. What changes is how confidently anyone can say why metformin works. Claims that metformin acts 'via AMPK' are shorthand for an unsettled question.

   Certainty: Contested, LOW mechanistic confidence. And the boundary condition is decisive: concentrations used in cell culture routinely exceed plasma levels achieved at clinical doses.

   Dose is the whole argument, and it is the most commonly ignored variable in translating cell biology. A mechanism demonstrated at 5 mM in a dish may be irrelevant at 20 uM in a patient. Any in vitro mechanism claim should come with the concentration attached.

6. 6. So the downstream link inherits the uncertainty. mechanism ↓

   If the route from metformin to AMPK is unsettled, the route from metformin to mTORC1 cannot be firmer than its weakest segment.

   Nothing. This is uncertainty propagating along a chain. Metformin's mTOR-lowering effect is real and reproducible; its mechanism is not settled, and the two facts are often conflated.

   Certainty: Contested, low mechanistic confidence, indirect. Cited across mammalian cells and mice.

   Uncertainty compounds along a path, and no pathway diagram shows that. If a route has three steps at medium confidence, the endpoint is not medium confidence. The Atlas grades each step but cannot yet grade a PATH – a genuine limitation of this design, worth naming.

7. 7. Whether chronic rapamycin hits mTORC2 rests on a single study in this corpus. mechanism ↓

   Prolonged exposure was reported to disrupt mTORC2 assembly in some cell types. It is not the acute, direct inhibition seen with mTORC1, and it is not universal.

   Nothing acutely. Over weeks, possibly a great deal. The textbook claim 'rapamycin inhibits mTORC1 but not mTORC2' is a statement about acute treatment being applied to chronic therapy.

   Certainty: Contested, LOW mechanistic confidence, one supporting study. Cell type and duration both change the answer.

   The clinical stakes are inverted relative to the evidence. This is one of the weakest-evidenced edges in the Atlas, and it may explain the most common serious side effect of the drug class. Low confidence does not mean low importance – which is exactly why the two are graded separately.

8. 8. An association presented in a chain of causal claims. mechanism ↓

   Renal cancers frequently carry lesions that leave mTORC1 active. That is a correlation between genotype and pathway state – not evidence that mTORC1 activation initiates the disease.

   Nothing mechanistically. This arrow records a statistical relationship. It explains why the tissue responds to rapalogs, and why an exceptional responder was traceable to TSC1 loss. It does not establish causation.

   Certainty: Typed ASSOCIATION, directness UNRESOLVED, mechanistic confidence low – the lowest grade available, applied deliberately.

   Correlation sitting between two well-evidenced clinical steps is the most dangerous position on any pathway map, because the reader's eye carries causality across it. Typing it as an association and drawing it dotted is the only defence, and it only works if someone reads the grade.

9. 9. The pathway's most famous claim has no human evidence at all. mechanism ↓

   Lowering mTOR signalling extends lifespan in yeast, worms, flies and mice – reproducibly, multi-site, in genetically heterogeneous strains.

   In those organisms, median and sometimes maximum lifespan. In humans: unknown. Not disputed, not negative – simply never measured.

   Certainty: Human relevance UNTESTED. That grade is arithmetic, not pessimism: no human lifespan trial of any mTOR-lowering intervention exists or could have completed.

   Thirteen interactions in this map carry human relevance untested, and every longevity edge is one of them. This is the single most important calibration in the Atlas, because it is the claim most likely to be repeated without its qualifier. Robust in four species is a strong result; it is not a human result.

10. 10. The same gap, for the drug specifically. mechanism ↓

    Rapamycin extends mouse lifespan reproducibly, including when started late in life.

    Mouse lifespan. Effects are sex- and strain-dependent, and healthspan and lifespan do not always move together. It is the strongest pharmacological longevity result in mammals and the basis for current human interest.

    Certainty: Mouse work (A – animal), medium mechanistic confidence, human relevance untested. LEE2024 systematically reviewed what human rapamycin data actually supports – and lifespan is not among it.

    There is a systematic review in this corpus specifically about human rapamycin data, and the honest summary of it is that the human evidence concerns safety and surrogate outcomes, not longevity. When a claim has a systematic review and the review does not support the popular version, the gap is not in the science – it is in the retelling.

11. 11. A case where the direction of effect itself is contested. mechanism ↓

    The same drug class is an immunosuppressant at transplant doses, yet intermittent low-dose everolimus improved influenza vaccine responses in older adults.

    Immune function – in opposite directions depending on dose and schedule. 'mTOR inhibition suppresses immunity' and 'mTOR inhibition rejuvenates immunity' are both supported, under different regimens.

    Certainty: Human trials (H), and still typed CONTESTED with medium confidence – because dose and schedule change the sign of the effect and the finding has not been uniformly replicated at scale.

    Human trial evidence and contested status are not mutually exclusive, and this step exists to make that visible. Good human data can still leave a question open when the effect depends on a variable the trials sampled differently. Dose and schedule are not implementation details here; they are part of the claim.

12. 12. And a result whose crucial qualifier is routinely dropped. mechanism ↓

    S6K1-null mice live longer and resist age-related pathology – in FEMALES. The effect is sex-specific.

    Median lifespan and metabolic protection, in one sex. It is the cleanest genetic evidence that a specific mTORC1 output influences lifespan, and it is narrower than its usual citation.

    Certainty: Mouse genetics (A – animal), medium mechanistic confidence, human relevance untested – and an external review of this Atlas found the sex-specificity stated in two places and omitted in two others. We were making the error this step describes.

    End on that. The route has been about where the pathway stops being known, and the most common failure is not a missing experiment – it is a qualifier lost in transmission. Sex, strain, dose, duration, species: five words that turn a true claim into a false one when dropped, and no pathway diagram has room for them. Which is why this Atlas puts them in the grade instead.

Evidence base: Deliberately the weakest evidence in the Atlas, and the grading is the point: 7 interactions are typed contested, 8 carry low mechanistic confidence, 13 have human relevance untested. Those counts are computed from the model, so this route cannot drift from the data it is complaining about.

Still unresolved: That is the entire route. But the meta-unknown is worth stating: gaps here are computed against THIS corpus, not against the literature. A gap may mean nobody has done the experiment, or it may mean the Atlas has not yet found the paper – and an external review already caught one case of the second kind. Treat every step as a hypothesis about the evidence, testable by finding the study that closes it.

mTORC1 is one of the best-characterised growth pathways in biology, and there are licensed drugs that inhibit it. Yet outside a few genotypes the clinical results are modest: progression delayed, survival rarely extended. If the mechanism is right, why is the treatment only partly right?

Follow one tumour genotype from the lesion that creates it, through the drug that was aimed at it, to the two feedback arms that undo the drug, and on to the molecules designed against that failure. This is the route where mechanism meets the clinic and the clinic answers back.

1. 1. PTEN is lost, and nothing erases the growth signal any more. mechanism ↓

   PTEN is a lipid phosphatase: it converts PIP3 back to PIP2. It does not inhibit the PI3K enzyme – it destroys PI3K's product. Delete PTEN and the product accumulates even at normal PI3K activity.

   PIP3 builds up in the membrane. The signal is no longer being written faster than it is erased; it is simply not being erased. Anything that reads PIP3 now spends far more of its time at the membrane, with no upstream hormone required.

   Certainty: Mouse evidence in this corpus (A – animal), with PTEN among the most frequently inactivated tumour suppressors in human cancer. Graded human-relevance plausible here because the cited studies are mouse, not because the human genetics is weak.

   The first thing to understand about mTOR in cancer is that the lesion is usually not in mTOR. It is in the machinery that decides whether mTOR should be receiving a signal. That distinction is why inhibiting mTOR treats a symptom of the genotype rather than its cause.

2. 2. Accumulated PIP3 recruits Akt continuously. mechanism ↓

   Akt binds PIP3 and is then phosphorylated by PDK1 and mTORC2. With PIP3 chronically elevated, recruitment stops being an event and becomes closer to a standing condition.

   Akt occupies the membrane persistently, and Akt signalling runs high. The growth-permission signal is now generated inside the cell rather than arriving from outside it.

   Certainty: High mechanistic confidence, cell-line evidence.

   This is what oncogenic means in signalling terms: not a stronger signal, but a signal that no longer requires its input. The cell has stopped asking the organism for permission and started granting it to itself.

3. 3. Akt inhibits the TSC complex, persistently now rather than in bursts. mechanism ↓

   Phosphorylation of TSC2 by Akt inhibits the complex and moves it away from where its target sits.

   The pathway's master brake is held off continuously rather than transiently. Rheb stops being switched off.

   Certainty: High mechanistic confidence, cell-line evidence; and TSC loss in people establishes that removing this brake causes disease.

   Notice that the tumour is exploiting the pathway's own logic rather than breaking it. Every step from here on is the normal mechanism running correctly on a false input – which is precisely why the pathway is hard to drug selectively.

4. 4. With TSC inhibited, Rheb stays GTP-loaded. mechanism ↓

   TSC2 is the GAP that forces Rheb to hydrolyse GTP. Inhibit the GAP and Rheb accumulates in its active state.

   The mTORC1 on-switch is held in the on position. mTORC1 will now fire whenever it is at the lysosome – which, given adequate nutrients, is most of the time.

   Certainty: High mechanistic confidence, cell-line evidence.

   The AND gate from the nutrient and growth-factor routes has been half-defeated. Nutrient sensing still controls location, but the permission input is stuck at yes. A coincidence detector with one input jammed is no longer a detector.

5. 5. mTORC1 is constitutively active. mechanism ↓

   GTP-Rheb allosterically activates mTORC1 whenever the two are co-located.

   The kinase runs without regard to whether the organism wants this cell to grow. Its outputs run too: translation of growth and invasion programmes up, autophagy down.

   Certainty: Structurally resolved, high mechanistic confidence, cell-line evidence.

   This is the state the drug will be aimed at. Worth holding onto the fact that mTORC1 itself is entirely normal here – correct protein, correct regulation, wrong input. A drug that inhibits mTORC1 is therefore not correcting an error; it is imposing a second one in the opposite direction.

6. 6. The tumour grows on that output. mechanism ↓

   Sustained selective translation, suppressed autophagy, and the biosynthetic arms together supply much of what a proliferating cell needs.

   Proliferation and mass increase. mTORC1 becomes a rational drug target for this genotype.

   Certainty: High mechanistic confidence but INDIRECT, and strongly genotype-dependent: a real dependency in TSC- and PI3K-pathway-mutant contexts, considerably weaker elsewhere.

   The gap between this arrow and a treatment is where most of the last twenty years of clinical disappointment lives. 'mTORC1 supports tumour growth' is a statement about biology; 'inhibiting mTORC1 treats this tumour' is a statement about dependency – and dependency is contextual in a way the arrow cannot show.

7. 7. Give everolimus. mTORC1 is inhibited – partially. mechanism ↓

   Like rapamycin, everolimus works as a complex with FKBP12 and obstructs the substrate channel rather than occupying the active site. Obstruction is partial by nature.

   S6K1 phosphorylation collapses. 4E-BP1 phosphorylation substantially persists – and 4E-BP1 controls the translation arm that matters most for proliferation. The tumour loses one output and keeps a good part of the other, while the reader's assay says the drug is working.

   Certainty: High confidence with H – human trial – evidence across several indications, including BAS2012 in hormone-receptor-positive breast cancer.

   Two failures compound here. The drug is incomplete, and the standard readout is blind to the part it misses – because S6K1 is rapamycin-sensitive and became the field's default assay. For years the pathway looked more inhibited than it was, in the exact output that mattered.

8. 8. And inhibiting mTORC1 releases a brake the tumour had been living under. mechanism ↓

   Active S6K1 had been phosphorylating IRS-1 and marking it for degradation. Inhibit mTORC1, S6K1 goes quiet, and IRS-1 stops being destroyed.

   IRS-1 protein accumulates. The adaptor that couples receptors to PI3K comes back. PI3K signalling recovers – driven by the drug, not despite it.

   Certainty: High mechanistic confidence, multiple supporting studies, cell-line evidence. ROD2011 additionally showed mTOR kinase inhibition produces biphasic Akt regulation through exactly this kind of feedback.

   This is the sentence that reframes the whole route. The drug does not merely fail to finish the job – it actively removes one of the tumour's own restraints. Any therapy that interrupts a negative feedback loop is partly self-defeating, and this loop was there all along in the growth-factor route.

9. 9. PI3K and Akt reactivate. The tumour has escaped. mechanism ↓

   Restored IRS-1 recruits PI3K to receptors that are still present, regenerating PIP3 and reactivating Akt.

   The upstream arm recovers while mTORC1 remains partly inhibited – the worst of both worlds, since Akt has many targets besides mTORC1. Progression resumes. Clinically, this is one mechanism behind rapalogs delaying progression without clearly extending survival.

   Certainty: High mechanistic confidence, cell-line evidence; ORE2006 showed the parallel arm in which mTOR inhibition raises receptor tyrosine kinase signalling directly.

   The escape route is not a mutation. It requires no new genetic event and no selection time – it is the pathway's normal homeostatic wiring responding correctly to the drug. That is why resistance appears fast and why combination strategies target the loop rather than the kinase.

10. 10. There is a second escape, through MAPK. mechanism ↓

    mTORC1 inhibition activates ERK in a PI3K-dependent manner. CAR2008 traces it through the same S6K1–PI3K–Ras relay that carries the IRS-1 escape, so this is a second OUTPUT of one feedback circuit rather than a separate circuit.

    ERK activity rises, which additionally phosphorylates and inhibits TSC2, feeding back toward mTORC1. The drug now faces two reroutes with a shared root. Because they share it, blocking PI3K upstream can close both – while blocking only the IRS-1 arm leaves the MAPK output open.

    Certainty: High mechanistic confidence, cell-line and mouse evidence (CAR2008). This paper sat in the Atlas corpus with zero edges until an external review flagged that the pathway's most clinically consequential feedback was missing from the graph.

    Redundancy is the theme of this pathway and it cuts both ways. The same architecture that makes the cell robust makes the tumour robust. This is the mechanistic rationale for combining mTOR inhibition with MEK inhibition rather than escalating the mTOR dose.

11. 11. So build a drug that occupies the site instead of obstructing it. mechanism ↓

    ATP-competitive inhibitors compete with ATP at the mTOR active site. They do not need FKBP12 and they suppress the 4E-BP1 phosphorylation that rapalogs leave standing.

    Inhibition becomes deep rather than partial – and extends to mTORC2, because both complexes share the same catalytic site. The 4E-BP1 escape closes. The mTORC2 toxicity opens.

    Certainty: High mechanistic confidence from cell-line pharmacology (THO2009, FEL2009, CHR2009). Clinical development of this class has been limited by toxicity attributed to simultaneous mTORC2 inhibition.

    Selectivity in this pathway comes from accessory subunits, not from the catalytic site – so a drug aimed at the site inherits no selectivity. That is a structural fact rather than a design failure, and it sets up the problem the next generation had to solve.

12. 12. The current attempt: deep inhibition of mTORC1 only. mechanism ↓

    A bivalent molecule engages both an FKBP12-dependent site and the active site, achieving the depth of an active-site inhibitor with selectivity for mTORC1 over mTORC2.

    In principle: 4E-BP1 actually suppressed, without the mTORC2-dependent metabolic toxicity. RMC-5552 has completed a phase 1 trial in advanced solid tumours – the newest clinical evidence anywhere in this Atlas.

    Certainty: One phase 1 trial (SCH2025, 2025). H – human – but phase 1 reports safety and pharmacodynamics, not efficacy. Consensus graded emerging, on a single study. Directness is indirect because, like rapalogs, the mechanism still requires FKBP12.

    This is the honest answer to the route's question, and it is not a failure story. A weakness identified in cell culture in 2009 became a molecular design constraint, then a compound, then a trial in 2025. The pathway was not too complicated to drug – it was drugged with a molecule discovered before anyone knew what it did, and it has taken this long to build one aimed at what we now know. Whether closing the 4E-BP1 escape produces survival benefit is genuinely unanswered, and that is what the current trials are for.

Evidence base: Cell-line pharmacology and biochemistry for the escape mechanisms; H – randomised trials – for the clinical outcomes; one phase 1 trial (SCH2025, 2025) for the newest drug class. Note the asymmetry that runs through this route: the mechanism is cell-line evidence, the disappointment is human evidence.

Still unresolved: Which tumours depend on mTOR remains largely unpredictable from genotype – the single largest open problem here. Whether closing the 4E-BP1 escape translates into survival benefit is exactly what the bi-steric trials are testing and is not yet answered. And why rapalogs delay progression without clearly extending overall survival in several indications has no accepted explanation, with the feedback arms as leading suspects.