Research Challenge 02 · Pharmacology / Mechanism · Core · 35 min
Why rapamycin does not switch mTORC1 off
Cells treated with rapamycin keep doing some of the things mTORC1 drives. Is that because of how the drug inhibits, because of what the experiment measures, because the network answers back, or because the drug is acting somewhere else?
This is a decision environment, not a quiz. Several of the choices below are scientifically defensible, and the feedback tells you what each one buys and what it costs rather than marking it right or wrong.
What you should be able to do
By the end of this challenge you should be able to
- Distinguish a claim about a drug's target from a claim about which of that target's outputs it blocks
- Predict which readout would hide a partial block and which would expose it
- Rank candidate experiments by how well they separate a property of the drug from a response of the network
- Identify what a single-arm dose-escalation trial can and cannot support
Research skills Causal reasoning · Experimental design · Control selection · Evidence evaluation · Alternative explanations · Scientific uncertainty
Recommended preparation
- What is mTOR?What the kinase integrates, and which readouts stand in for its activity.
- mTORC1 vs mTORC2Why a sentence about ‘mTOR inhibition’ is usually a sentence about one of two complexes.
- Feedback RegulationThe loops that make a treated cell something other than an untreated cell with one arrow cut.
- How to Read an mTOR PaperReading a single band as a claim about a whole pathway is the habit this challenge works against.
Step 1
What we know
Every line here is carried by a study in the Atlas, linked beside it. Read the second list as carefully as the first: it is where your own decisions start.
What the Atlas already supports
- Rapamycin does not work alone. It binds the small abundant protein FKBP12, and the pair docks on a site next to the kinase's active site rather than in it, so the inhibition is steric obstruction rather than occupancy of the catalytic pocket. CHO1996 SAB1995
- The S6K1 arm goes quiet within minutes of treatment, and that band is the readout most papers use when they report ‘mTORC1 activity’. CHU1992 PRI1992
- Compared against an inhibitor that occupies the active site, rapamycin leaves a class of mTORC1 outputs largely running — 4E-BP1 phosphorylation most of all, and autophagy induction as well. THO2009 FEL2009
- Time changes the target list. Short exposure leaves the second complex alone; prolonged exposure traps free mTOR and new mTORC2 stops being assembled in the sensitive lines, and the same loss of signalling appears in mouse liver. SAR2006 LAM2012
- Inhibiting mTORC1 releases a brake on the arm above it: S6K1 no longer represses IRS-1, upstream receptor signalling picks up, and Akt activity can rise rather than fall. HAR2004 ORE2006 ROD2011
- A bivalent molecule can reach the substrate rapamycin leaves alone while leaving the second complex largely spared, in resistant breast-cancer models and in a first-in-human dose-escalation trial. MEN2023 SCH2025
Where that stops
- Whether prolonged exposure reaches the second complex in any particular cell depends on the line, the dose and the duration. The Atlas grades that edge as contested for a reason: it is absent acutely and variably present later.
- Two routes lead from mTORC1 inhibition to disturbed glucose handling — loss of the second complex and relief of the S6K1-to-IRS-1 brake — and their relative contribution has never been apportioned in people.
- The lifespan data in this corpus come from rapamycin, and the selective compounds have been read out against tumour endpoints. No study here joins selective inhibition to an ageing outcome, which is the join the last sub-question asks for.
Step 2
Build and break the model
Switch the controls and watch which combinations produce output. Then break it on purpose: each prediction below asks you to commit before it answers. These are predictions about a simplified teaching model, not measurements.
Two chemistries on the left, one kinase in the middle, and three outputs on the right. The drug switch chooses nothing, rapamycin or an active-site inhibitor; the clock switch chooses acute or prolonged exposure. Nodes and arrows come from the Atlas pathway model, so this teaching diagram cannot drift away from the scientific one.
p-S6K1 high · p-4E-BP1 high · autophagy held down · Akt Ser473 phosphorylated
The reference state. Note what an active kinase does to two of the arrows: phosphorylated 4E-BP1 lets go of eIF4E, and phosphorylated ULK1 cannot be switched on, so a growing cell builds protein and keeps recycling down. Both of those are brakes being held off rather than accelerators being pressed.
- Rapamycin
- Allosteric mTORC1 inhibitor that works only as a complex with FKBP12.The drug that made this pathway famous.Sirolimus. Not an active-site inhibitor: the FKBP12–rapamycin complex binds the FRB domain and partially occludes substrate access, which is why 4E-BP1 phosphorylation is only incompletely blocked.
- ATP-competitive mTOR inhibitors
- Torin1, AZD8055 and relatives occupy the mTOR active site directly, inhibiting both complexes.Drugs that block mTOR's engine instead of getting in its way.Designed as a research answer to rapamycin's incompleteness, then developed clinically. Because they compete with ATP at the catalytic site they do not need FKBP12, they suppress 4E-BP1 phosphorylation that rapalogs leave standing, and they hit mTORC2 as well – which is both their advantage and their toxicity problem.
- FKBP12
- Prolyl isomerase that forms the drug-receptor complex with rapamycin.The protein rapamycin must borrow.FKBP1A. Rapamycin has essentially no activity against mTOR without it, which is why FKBP12 expression sets rapalog sensitivity.
- mTORC1
- mTOR–Raptor–mLST8 (+PRAS40, DEPTOR); in the amino-acid route it is switched on at the lysosome.The growth decision itself.Coincidence detector: nutrients supply location via the Rags, growth factors supply activation via Rheb. Neither alone is sufficient – the central idea of the amino-acid route.
- mTORC2
- mTOR–Rictor–SIN1–mLST8; phosphorylates Akt, SGK1 and PKC.mTOR's second, less famous complex.Largely plasma-membrane associated and PI3K-responsive via the SIN1 PH domain; acutely rapamycin-insensitive, which is the cleanest way to separate mTORC1 from mTORC2 biology experimentally.
- S6K1
- Ribosomal protein S6 kinase; the standard readout of mTORC1 activity.mTORC1's best-known output kinase.T389 phosphorylation by mTORC1 is rapamycin-sensitive, which is why S6K1 became the field's default assay – and why the field long over-read rapamycin as a complete mTORC1 inhibitor.
- 4E-BP1
- Translational repressor released from eIF4E upon multi-site phosphorylation.A cap on protein-making that mTORC1 removes.Only partially rapamycin-sensitive. This goes a long way to explain the rapalog/Torin discrepancy and was a main motivation for ATP-competitive inhibitors.
- ULK1
- Autophagy-initiating kinase, inhibited by mTORC1 and activated by AMPK.The switch that starts self-digestion.mTORC1 phosphorylates S757 to block the AMPK–ULK1 interaction; ULK1 also feeds back to phosphorylate and dampen AMPK, making this a closed loop rather than a switch.
- Akt/PKB
- AGC kinase requiring PIP3 recruitment plus two phosphorylations to be fully active.The main 'grow' relay from growth factors.T308 by PDK1 and S473 by mTORC2. Recruitment and activation are separate events – a distinction the older Atlas diagram blurred.
- eIF4E
- Cap-binding translation initiation factor.The clamp that starts reading mRNA.Rate-limiting for cap-dependent initiation; a proto-oncogene in its own right when overexpressed.
- Autophagy
- Bulk degradative recycling, held off by mTORC1 and switched on by AMPK.The cell eating its own worn-out parts.Regulated at initiation (ULK1), at transcription (TFEB) and at fusion; most 'autophagy is required' claims rest on flux measurements that are hard to do in tissue.
- IRS-1 / IRS-2
- Scaffold recruiting PI3K to activated insulin/IGF-1 receptors.The adaptor that connects the insulin receptor to PI3K.Serine phosphorylation by S6K1 (and others) triggers IRS-1 degradation – the molecular basis of the best-characterised mTORC1 feedback loop. Note the direction: the loop degrades IRS-1 when mTORC1 is ACTIVE, so it is a mechanism of insulin resistance under nutrient excess. Inhibiting mTORC1 releases it and raises Akt signalling instead.
- Protein synthesis
- Cap-dependent translation, the most direct mTORC1 output.Making new proteins.Controlled through 4E-BP/eIF4E and S6K1/PDCD4/eIF4A arms; ribosome-profiling shows the response is transcript-selective, not uniform.
This is a simplified teaching model with a fixed set of states, not a simulation. It shows the direction each control pushes the pathway, not how much, how fast, or what any particular cell would do.
Every state of this model
| Drug | Exposure | Readout | What it means |
|---|---|---|---|
| none | acute | p-S6K1 high · p-4E-BP1 high · autophagy held down · Akt Ser473 phosphorylated | The reference state. Note what an active kinase does to two of the arrows: phosphorylated 4E-BP1 lets go of eIF4E, and phosphorylated ULK1 cannot be switched on, so a growing cell builds protein and keeps recycling down. Both of those are brakes being held off rather than accelerators being pressed. |
| none | prolonged | p-S6K1 high · p-4E-BP1 high · autophagy held down · Akt Ser473 phosphorylated | With no drug on, the clock changes nothing. The state is here so you can hold one switch still while you move the other — the comparison that makes a time-dependent effect visible at all. |
| rapamycin | acute | p-S6K1 low · p-4E-BP1 largely retained · autophagy partly released · Akt Ser473 phosphorylated | Sit with this one; it is the state a decade of papers reported as ‘mTORC1 off’. The S6K1 arm is gone, and the kinase is still phosphorylating 4E-BP1, so protein is still being built. IRS-1 lights up because the brake on it went with S6K1 — the treated cell has a livelier insulin arm than the untreated one. |
| rapamycin | prolonged | p-S6K1 low · p-4E-BP1 largely retained · Akt Ser473 lost in the sensitive lines · glucose handling disturbed in mice | Same drug, same dose, different answer to the question ‘what does it block?’. New mTORC2 stops being assembled, Akt loses its second phosphate, and in mice the metabolic cost appears. The Atlas draws this arrow dashed and grades it contested, because it is absent acutely and varies between cell lines. |
| active-site | acute | p-S6K1 low · p-4E-BP1 low · cap-dependent translation down · autophagy induced · Akt Ser473 low | Occupying the active site rather than obstructing the doorway gives a different pattern from the first click: 4E-BP1 stays unphosphorylated, keeps hold of eIF4E, and protein synthesis goes with it. The same site sits in both complexes, so the second complex goes down too — at once, not after days. |
| active-site | prolonged | p-S6K1 low · p-4E-BP1 low · cap-dependent translation down · autophagy induced · Akt Ser473 low | Here the clock is the uninteresting switch. This chemistry hits both complexes from the first minutes, so there is no second phase to wait for — which is the contrast that makes the time axis a real finding for the other drug rather than a detail of dosing. |
Prediction 1
Set the drug to rapamycin and the clock to acute. Two of the three output arms are in the model. Which readout would report the pathway as switched off, and which would report it still running?
The two arms come apart in this state, and which one you blot for decides what you would write in the abstract. That is why a decade of papers using p-S6K1 as the readout of mTORC1 activity over-reported how much of the kinase rapamycin was switching off.
Show the expected answer
A. p-S6K1 would report it off, and p-4E-BP1 would report output continuing
The two arms come apart in this state, and which one you blot for decides what you would write in the abstract. That is why a decade of papers using p-S6K1 as the readout of mTORC1 activity over-reported how much of the kinase rapamycin was switching off.
Prediction 2
Keep the drug at rapamycin and move the clock from acute to prolonged. A part of the diagram lights up that was dark before. What does that do to the sentence ‘rapamycin is an mTORC1 inhibitor’?
The sentence is not wrong so much as unfinished: it needs a duration attached. An acute experiment and a chronic one are asking different questions of the same molecule, and much of the disagreement in this literature is two groups reporting honestly from either side of that clock.
Show the expected answer
B. The second complex joins the picture, so the target list depends on how long the drug has been on
The sentence is not wrong so much as unfinished: it needs a duration attached. An acute experiment and a chronic one are asking different questions of the same molecule, and much of the disagreement in this literature is two groups reporting honestly from either side of that clock.
Prediction 3
Three states in this model give a low p-S6K1 and differ in almost everything else. What follows for a paper that reports only that band?
A readout earns its place in proportion to how differently the candidate explanations predict it. This one is a fine measure of the S6K1 arm and a poor discriminator between the four explanations you are about to choose from, because three quite different states of the cell collapse onto the same band.
Show the expected answer
B. A low p-S6K1 does not say which chemistry or which exposure produced it, so the design needs a second substrate or a second time point
A readout earns its place in proportion to how differently the candidate explanations predict it. This one is a fine measure of the S6K1 arm and a poor discriminator between the four explanations you are about to choose from, because three quite different states of the cell collapse onto the same band.
Step 3
Commit to a working hypothesis
Commit to a working hypothesis before you spend anything. It is meant to be revised later — that is what the second half of this challenge is for.
What each choice commits you to
The block is partial by construction. The drug obstructs the doorway to the kinase rather than filling its catalytic pocket, so substrates that need deep, sustained access lose out and others carry on. It predicts that a chemistry which does occupy the site reaches further, and that the split between outputs is reproducible rather than noise. Its weak point: obstruction and network feedback predict the same blot in a single-time-point experiment.
The incompleteness is in the readout, not the drug. On this reading mTORC1 really is switched off and the persisting signal is an artefact of which band gets blotted, of antibody sensitivity, or of a phosphosite that another kinase also touches. It predicts that a wider substrate panel, or a functional readout instead of a phosphosite, makes the residue disappear. It is the hypothesis most worth holding early and the easiest to hold too long.
The network answers back. Block the kinase and the loops around it re-set: S6K1 stops repressing IRS-1, upstream receptor signalling picks up, and the cell you measure is no longer the untreated cell with one arrow cut. It predicts that the residue grows over time rather than staying flat, and that the arm above the kinase moves in the same lysates. It has to explain why a deeper chemistry, which relieves the same feedback, still gives a deeper block.
The interesting part is the complex the drug was not supposed to touch. Here the incompleteness on mTORC1 is the less consequential half of the story, and what matters is that prolonged exposure reaches the second complex, taking the arm to Akt with it. It predicts that the drug's effects at day ten are not the drug's effects at minute ten. Its weak point as an answer to the question asked: it explains the side effects better than it explains the residue.
Step 4
Run the investigation
This is a research path, not a shopping list. Each step costs what its equipment costs, and each one opens up different next steps — so the order you choose decides which questions you can still afford to answer. You can go back to any earlier step and take another branch; what you have already spent stays spent. The panel above tracks which parts of the research question your results can actually answer.
What you can answer so far 0 of 6
- Does rapamycin block every mTORC1 output to the same degree?Three findings, and that is the point: one band tells you the drug engages its target, two bands tell you the block is uneven, and a functional readout tells you the unevenness is not an antibody story.
- Is a deeper block chemically available, or is this as far as the kinase can be pushed?
- Does the network answer back when mTORC1 is inhibited?
- Does the drug reach the second complex, and does that depend on time?
- Can the metabolic cost be separated from target engagement?Two ways of separating them — by calendar and by molecule — and neither is worth much without the other, because one shows the cost shrinking and the other shows it being designed out.
- Does a block that spares the second complex keep the lifespan benefit without the metabolic cost?Nothing in this corpus yields this. The only step that would address it returns a prediction rather than a result, so this sub-question stays open at any price — and it is the same gap the Atlas records as H2, not an omission from the menu.
Research budget
120 of 120 research units remaining
Each step costs what its equipment costs. Running everything would take 205 units, and answering every sub-question that can be answered would take 165 — so this budget buys four of the five answers at most. Which four is the whole exercise.
120 of 120 research units left
Everything else currently open to you
The research path, without JavaScript
With scripting on, this page walks you through the steps below one decision at a time and keeps the budget. Without it, every step is on the page in full and here is the map that would otherwise live in the script.
Which step answers which sub-question
| Does rapamycin block every mTORC1 output to the same degree? | Treat cells with rapamycin and blot p-S6K1 + Probe the same lysates for p-4E-BP1 + Ask whether the autophagy arm is released the way textbooks say |
|---|---|
| Is a deeper block chemically available, or is this as far as the kinase can be pushed? | Compare rapamycin head to head with an active-site inhibitor |
| Does the network answer back when mTORC1 is inhibited? | Follow the arm above the kinase after mTORC1 is inhibited |
| Does the drug reach the second complex, and does that depend on time? | Hold the drug on for days and read the second complex |
| Can the metabolic cost be separated from target engagement? | Change the schedule rather than the molecule + Build a molecule that blocks mTORC1 deeply and leaves the second complex alone |
| Does a block that spares the second complex keep the lifespan benefit without the metabolic cost? | nothing in this corpus answers this |
What each step opens up
- Treat cells with rapamycin and blot p-S6K1 opens Probe the same lysates for p-4E-BP1, Compare rapamycin head to head with an active-site inhibitor
- Probe the same lysates for p-4E-BP1 opens Compare rapamycin head to head with an active-site inhibitor, Ask whether the autophagy arm is released the way textbooks say
- Compare rapamycin head to head with an active-site inhibitor opens Hold the drug on for days and read the second complex, Ask whether the autophagy arm is released the way textbooks say
- Ask whether the autophagy arm is released the way textbooks say opens Hold the drug on for days and read the second complex
- Follow the arm above the kinase after mTORC1 is inhibited opens Hold the drug on for days and read the second complex
- Hold the drug on for days and read the second complex opens Change the schedule rather than the molecule, Build a molecule that blocks mTORC1 deeply and leaves the second complex alone
- Change the schedule rather than the molecule opens Build a molecule that blocks mTORC1 deeply and leaves the second complex alone
- Build a molecule that blocks mTORC1 deeply and leaves the second complex alone opens Run a selective molecule against an ageing endpoint
- Run a selective molecule against an ageing endpoint opens nothing further
You can start with
Treat cells with rapamycin and blot p-S6K1, Compare rapamycin head to head with an active-site inhibitor, Follow the arm above the kinase after mTORC1 is inhibited
The cheapest routes
- 15 research units → 1 of 6 sub-questions
- 40 research units → 2 of 6 sub-questions
- 70 research units → 3 of 6 sub-questions
- 115 research units → 4 of 6 sub-questions
The best this budget allows is 4 of 6 sub-questions for 115 research units.
10 research units
Treat cells with rapamycin and blot p-S6K1
Addresses Whether the drug reaches mTORC1 at all
Needs Cell culture, one compound, one antibody. The oldest experiment in this field and still the cheapest item on the menu.
The design
| Model | Cultured mammalian cells and T cells |
|---|---|
| Perturbation | Add rapamycin to growing cells for tens of minutes |
| Readout | Phosphorylation of p70 S6 kinase |
| Control | Untreated cells, and vehicle without the compound |
Before you run it: what would change your mind?
Before the blot comes back: which result would weaken the model you are holding?
What each choice commits you to
p-S6K1 is unchanged — the drug never reaches the kinase. That would end the investigation and start a different one. It is the result none of your four hypotheses can absorb, which makes it the honest answer to what would change your mind — and the least likely one to arrive.
p-S6K1 collapses within minutes. What the two source studies report, and what all four of your hypotheses predict. A result everyone expects cannot weaken anybody, so naming it here is a sign the step is background rather than evidence.
p-S6K1 falls only after days of treatment. That would make this a slow, indirect effect and would put the time axis in play far earlier than it belongs. Worth naming, because it is how you would have spotted it.
Nothing here would change my mind. The most defensible answer on this card, and it comes with a cost attached: if no outcome would move you, the ten units are buying a lysate rather than an argument.
Which of these the step returned
B. p-S6K1 collapses within minutes
Run · 10 research units spent
Not enough budget left for this one. What you have already spent stays spent.
What it opened up One lysate, two antibodies: the second substrate is the cheapest next move. The drug comparison becomes worth its price once you know the drug engages its target.
Derived from M Rapamycin-FKBP specifically blocks growth-dependent activation of and signaling by the 70 kd S6 protein kinases 1992 · M Rapamycin-induced inhibition of the 70-kilodalton S6 protein kinase 1992
p-S6K1
Direction and rank order as reported in the two source studies. The bars are a schematic of that direction, not measured values.
Interpret — what does this let you say?
What each choice commits you to
It supports my hypothesis. Be careful what it supports. A result every candidate explanation predicts favours none of them; it agrees with the part nobody was disputing, which feels like progress and is not.
It weakens my hypothesis. Hard to see how. No hypothesis on your list denies that rapamycin takes the S6K1 arm down — they differ about what the rest of the kinase is doing while that happens.
It is inconclusive for my question. That is the reading this result deserves. A clean, interpretable experiment about a step none of your four explanations disagreed on.
It introduces an alternative explanation. Not an alternative so much as a reminder of how the habit started: this band is fast, cheap and dramatic, which is exactly why it became the readout everyone reported.
What this evidence supports
- Rapamycin reaches mTORC1 and takes the S6K1 arm down within minutes.
- The drug has a real molecular target in this pathway, which is why it became the tool the whole field was built on.
What it does not establish
- Anything about the outputs you did not blot for. One band is one substrate.
- Whether mTORC1 as a whole went quiet, which is the question you actually asked.
Information value High for the existence of the target and zero for your question: all four hypotheses predict this band collapsing. Ten units buy confidence in shared background rather than a decision — though these are also the lysates the next step reads, so it is not money thrown away.
15 research units
Probe the same lysates for p-4E-BP1
Addresses Whether the block is the same size on every output
Needs The lysates you already have and one more antibody. No new cells, no new dosing.
The design
| Model | The same cultured cells and the same lysates as the step above |
|---|---|
| Perturbation | None new — this reads a second substrate from material you already have |
| Readout | Phosphorylation of 4E-BP1 alongside p-S6K1 |
| Control | Untreated lysates from the same experiment, run on the same blot |
Before you run it: what would change your mind?
Name the outcome that would weaken your hypothesis, then read the second antibody.
What each choice commits you to
Both substrates collapse together. That would make the drug a clean off-switch and leave the residue everyone argues about as an artefact of somebody's blot. It is the strongest result the readout-artefact reading could have received.
S6K1 collapses while 4E-BP1 largely survives. What the source studies report. It is the observation the rest of this challenge exists to explain, and the first real pressure on a readout-artefact reading.
4E-BP1 collapses while S6K1 survives. The mirror image, and it would overturn a decade of reading. Naming it is useful precisely because it shows which way round the asymmetry runs.
Nothing here would change my mind. Then you are treating the split as decided before you have seen it, which is the habit this step is cheapest at interrupting.
Which of these the step returned
B. S6K1 collapses while 4E-BP1 largely survives
Run · 15 research units spent
Not enough budget left for this one. What you have already spent stays spent.
What it opened up Two ways on from an uneven block: bring in a chemistry that blocks differently, or check whether the unevenness reaches the arm the drug is famous for.
Derived from M An ATP-competitive mammalian target of rapamycin inhibitor reveals rapamycin-resistant functions of mTORC1 2009 · M Active-site inhibitors of mTOR target rapamycin-resistant outputs of mTORC1 and mTORC2 2009
Phosphorylation of two substrates under rapamycin
Direction and rank order as reported in the source studies, which describe 4E-BP1 phosphorylation as largely surviving a treatment that collapses S6K1. Schematic, not measured values.
Interpret — what does this let you say?
What each choice commits you to
It supports my hypothesis. If you chose partial block, this is the observation your hypothesis exists to explain. Notice it is an observation and not yet a mechanism: nothing here says why the two substrates differ.
It weakens my hypothesis. For the readout-artefact reading it is awkward: the residue is on a second antibody, in the same lysate, under the same dose, so it is not a property of one reagent.
It is inconclusive for my question. Too cautious now. The split between arms is the thing that needs explaining, and you have it for 25 units in total.
It introduces an alternative explanation. It does, quietly: if one arm survives, perhaps the kinase was never fully blocked, and perhaps the cell helped. Those are two different follow-ups, and they cost different amounts.
What this evidence supports
- The drug is substrate-selective rather than a general off-switch: one arm goes, the other largely stays.
- A single-substrate readout reports the size of the block on that substrate, and nothing wider.
What it does not establish
- Why the two arms differ. A blot shows the difference; the mechanism behind it needs a second chemistry or a structure.
- Whether the residue is a property of the drug or a response of the network, since both were free to happen in this dish.
Information value The best value on the menu. Fifteen units turn ‘the drug engages its target’ into ‘the drug engages it unevenly’, which is the observation the rest of this challenge is about. It also puts the readout-artefact hypothesis under real pressure for the first time.
25 research units
Compare rapamycin head to head with an active-site inhibitor
Addresses Whether the incompleteness belongs to the drug or to the kinase
Needs A second compound, matched dose-response for both, and a full substrate panel. New chemistry, so a new cost.
The design
| Model | Mouse and human cell lines, including primary cells |
|---|---|
| Perturbation | Treat matched cultures with rapamycin or with an ATP-competitive inhibitor |
| Readout | p-S6K1, p-4E-BP1 and cap-dependent translation |
| Control | Vehicle-treated cultures, and a dose-response for each compound so the comparison is not a dose artefact |
Before you run it: what would change your mind?
This is the discriminating step. Which result would weaken the hypothesis you committed to?
What each choice commits you to
Both compounds stop at the same depth. That would make the residue a property of the kinase rather than of the drug, and a partial-block reading would lose its main claim. It is the result that reading has to be willing to receive.
The active-site compound reaches further than rapamycin. What both source studies report. The residue moves when the molecule changes while the antibodies stay the same, which is the pattern a readout artefact cannot produce.
Rapamycin reaches further than the active-site compound. That would be a surprise worth chasing and would say something is wrong with the dosing or the compound rather than with the model. The dose-response on this card exists to tell those apart.
Nothing here would change my mind. Twenty-five units is a lot to spend on a result you have decided in advance. If that is where you are, the cheaper steps are the better buy.
Which of these the step returned
B. The active-site compound reaches further than rapamycin
Run · 25 research units spent
Not enough budget left for this one. What you have already spent stays spent.
What it opened up Having a ruler makes two further questions affordable: how far the unevenness extends across outputs, and what happens when the drug stays on.
Derived from M An ATP-competitive mammalian target of rapamycin inhibitor reveals rapamycin-resistant functions of mTORC1 2009 · M Active-site inhibitors of mTOR target rapamycin-resistant outputs of mTORC1 and mTORC2 2009
mTORC1 output under two chemistries
Direction and rank order as reported in the two source studies, which used the active-site compound as a ruler against rapamycin. Schematic, not measured values.
Interpret — what does this let you say?
What each choice commits you to
It supports my hypothesis. For partial block, strongly, and with a scope worth stating: two chemistries, one panel of substrates, in cells. What it does not give you is the size of that difference in a tissue.
It weakens my hypothesis. For readout-artefact it is close to decisive: the residue moves when the molecule changes while the antibodies stay the same, which is not how a reagent artefact behaves.
It is inconclusive for my question. Defensible only in a narrow sense — relief of feedback happens under both compounds, so this design does not take that contribution off the table.
It introduces an alternative explanation. It introduces a cost rather than an explanation: a compound that reaches further reaches into the second complex as well, and that is the trade-off the last two steps are about.
What this evidence supports
- A deeper block is chemically available, so what rapamycin leaves running is a property of how rapamycin inhibits rather than a limit of the kinase.
- The rapamycin-resistant outputs are only visible when a second chemistry is there to measure against — which is why the field lived with them for a decade.
What it does not establish
- That the deeper block is the better drug. This design reads substrate phosphorylation, not benefit and not harm.
- Which complex each compound is acting on: the same active site sits in both, so this chemistry has no reason to tell them apart.
Information value The most discriminating step at a middling price. The two chemistries are predicted to behave differently by the partial-block reading and identically by the readout-artefact one, so the result separates them — and it says nothing about the time axis or about the network's reply.
20 research units
Ask whether the autophagy arm is released the way textbooks say
Addresses Whether the uneven block reaches the arm the drug is best known for
Needs Flux assays with lysosomal blockade, imaging and blots. Cheap reagents, fussy controls, and a fortnight of them.
The design
| Model | Mouse and human cells |
|---|---|
| Perturbation | Rapamycin or an active-site inhibitor, each with and without lysosomal blockade |
| Readout | Autophagic flux rather than a single marker at a single time point |
| Control | Untreated cells, plus the blockade-only arm that makes flux readable at all |
Before you run it: what would change your mind?
Before the flux assay: what would weaken your model?
What each choice commits you to
Both compounds induce recycling to the same extent. That would confine the unevenness to translation and leave the drug's best-known effect intact. It is the result that would keep the textbook sentence as it stands.
Rapamycin induces it partly, and the active-site compound more. What the source study reports. A functional readout behaving like the phosphosites leaves a readout-artefact reading with very little room.
Neither compound induces recycling at all. That would say the block on this arm is not being relieved by either chemistry, and it would point the investigation at the assay before it pointed at the biology.
Nothing here would change my mind. Then twenty units buy a wider description of something you already accept. That is a legitimate purchase and a poor one at this stage.
Which of these the step returned
B. Rapamycin induces it partly, and the active-site compound more
Run · 20 research units spent
Not enough budget left for this one. What you have already spent stays spent.
What it opened up Once the unevenness is established across arms, the open question is what the same molecule does when it stays on for days.
Derived from M An ATP-competitive mammalian target of rapamycin inhibitor reveals rapamycin-resistant functions of mTORC1 2009
Autophagy induction
Direction and rank order as reported in the source study, which found autophagy among the mTORC1 outputs rapamycin leaves largely intact. Schematic, not measured values.
Interpret — what does this let you say?
What each choice commits you to
It supports my hypothesis. For partial block, yes, and it widens the claim from one substrate to a functional output — which is a stronger position than a second band on the same blot.
It weakens my hypothesis. For readout-artefact it is the harder result of the two, because this is not a phosphosite. A functional readout behaving like the phosphosites leaves little room for an antibody story.
It is inconclusive for my question. A fair reading if your question was about the residue on 4E-BP1 specifically. This step widens the phenomenon rather than explaining it.
It introduces an alternative explanation. It reframes more than it explains: the drug does not turn mTORC1 output down by a fixed fraction, it reshapes which outputs stay on. That is a different kind of pharmacology from a dial.
What this evidence supports
- ‘Rapamycin induces autophagy’ needs a qualifier: the induction is partial, and the compound that occupies the active site induces more.
- The unevenness is not confined to translation. It shows up on the arm most often quoted as the drug's main effect, which is where it does the most damage to sloppy reading.
What it does not establish
- How much of this carries into a tissue, where the dose at the cell and the supply of nutrients are both different.
- That the difference between compounds on this arm runs through mTORC1 alone, since the active-site compound is not complex-selective.
Information value Moderate for the question as posed and high for unlearning a sentence most readers arrive with. It is also the step whose price is easiest to underestimate: flux needs the blockade arm, and without it you are measuring a marker, not recycling.
15 research units
Follow the arm above the kinase after mTORC1 is inhibited
Addresses Whether the network replies when the kinase is blocked
Needs Cell culture, a time course, and antibodies against p-AKT and IRS-1. Cheap, and the design is in the time points rather than the reagents.
The design
| Model | Human cancer cell lines |
|---|---|
| Perturbation | Inhibit mTORC1 and follow the upstream arm over a time course rather than at one point |
| Readout | p-AKT and IRS-1 levels alongside p-S6K1 |
| Control | Untreated cultures sampled at the same time points |
Before you run it: what would change your mind?
Say what would weaken your model before you follow the arm above the kinase.
What each choice commits you to
Upstream signalling sits still while mTORC1 goes down. That would take the network's reply off the table and leave the residue entirely to the chemistry. It is the cleanest result a partial-block reading could have received here.
p-AKT rises while p-S6K1 falls. What the source studies report. It does not explain the residue by itself, and it does mean the cell you measured is not the untreated cell with one arrow cut.
p-AKT falls together with p-S6K1. That would make the drug look like a general dampener of the whole arm and would need a different mechanism from the loops in the lessons. It is not what came out.
Nothing here would change my mind. This is the only step on the menu that can move the feedback hypothesis at all. Declining to be moved by it makes the other fifteen units the better buy.
Which of these the step returned
B. p-AKT rises while p-S6K1 falls
Run · 15 research units spent
Not enough budget left for this one. What you have already spent stays spent.
What it opened up With the network's reply on the record, the expensive question is what happens when the drug stays on long enough for the cell to rebuild itself around it.
Derived from M mTOR inhibition induces upstream receptor tyrosine kinase signaling and activates Akt 2006 · M The TSC1-2 tumor suppressor controls insulin-PI3K signaling via regulation of IRS proteins 2004 · M mTOR kinase inhibition causes feedback-dependent biphasic regulation of AKT signaling 2011
Upstream signalling after mTORC1 inhibition
Direction as reported across the source studies, which describe relief of feedback onto receptor signalling and a biphasic rise in Akt phosphorylation. Schematic, not measured values.
Interpret — what does this let you say?
What each choice commits you to
It supports my hypothesis. For the feedback reading, directly. Keep the limit in view: showing that a loop re-sets is not the same as showing how much of the residue it accounts for.
It weakens my hypothesis. For partial block in its strictest form it is uncomfortable, because now two things change when you add the drug. Most of the strength survives if you treat feedback as a co-contributor rather than a rival.
It is inconclusive for my question. Reasonable: the design shows a loop moving without apportioning the residue between two mechanisms. Saying what would apportion it is the useful next sentence.
It introduces an alternative explanation. It makes an existing one measurable, and it carries a warning for later: relief of this brake is one of two routes from mTORC1 inhibition to disturbed glucose handling, and the two have never been separated in people.
What this evidence supports
- Inhibiting the kinase releases a brake on the arm above it, so the cell you measure an hour later is not the untreated cell with one arrow cut.
- Part of what reads as an incomplete block is the network pushing back, which is a different thing from a drug obstructing a substrate channel.
What it does not establish
- How much of the surviving 4E-BP1 phosphorylation is feedback and how much is substrate selectivity. Separating those needs the chemistry comparison, not this blot.
- Anything about the second complex, which this design does not read, even though the arm to Akt runs through it.
Information value Cheap and clarifying. It puts a second contributor on the table with a measurement attached rather than leaving it as an objection raised after the talk — and it is the only step that discriminates the feedback hypothesis at all.
30 research units
Hold the drug on for days and read the second complex
Addresses Whether exposure time changes which complex the drug is acting on
Needs Long-term culture across several lines, complex immunoprecipitation and the Ser473 antibody. Weeks of dosing before the first informative point.
The design
| Model | Several human and mouse cell lines |
|---|---|
| Perturbation | Short exposure against prolonged exposure to the same compound at the same dose |
| Readout | Assembly of the second complex, and Akt Ser473 phosphorylation |
| Control | Acutely treated and untreated cultures of the same lines, read on the same day |
Before you run it: what would change your mind?
Before the long exposure comes back: which result would weaken your model?
What each choice commits you to
Prolonged exposure looks exactly like acute exposure. That would remove the time axis from the story and make the drug's target list a fixed property. A wrong-complex reading would lose most of its force, and the challenge would have one axis instead of two.
Prolonged exposure lowers Akt Ser473 while acute exposure leaves it alone. What the source studies report, with the scope the Atlas attaches: sensitive lines, prolonged exposure, an edge graded contested rather than background.
Even acute exposure lowers Akt Ser473. That would make the second complex an immediate target and would sit badly against the acute experiments in the rest of the corpus. Naming it is how you would have noticed a dose or a line behaving unusually.
Nothing here would change my mind. Thirty units on a foregone conclusion is the most expensive way to be certain on this menu. If the time axis cannot move you, spend it on the chemistry instead.
Which of these the step returned
B. Prolonged exposure lowers Akt Ser473 while acute exposure leaves it alone
Run · 30 research units spent
Not enough budget left for this one. What you have already spent stays spent.
What it opened up A time-dependent effect invites two quite different moves: change the schedule, or change the molecule. They cost different amounts and answer different questions.
Derived from M Prolonged rapamycin treatment inhibits mTORC2 assembly and Akt/PKB 2006 · A Rapamycin-induced insulin resistance is mediated by mTORC2 loss and uncoupled from longevity 2012
Second-complex output (Akt Ser473)
Direction and rank order as reported in the source studies, in which prolonged treatment reduced assembly of the second complex in sensitive lines. Schematic, not measured values.
Interpret — what does this let you say?
What each choice commits you to
It supports my hypothesis. For the wrong-complex reading, yes — with the scope the Atlas itself attaches: sensitive lines, prolonged exposure, and an edge graded contested rather than settled background.
It weakens my hypothesis. For any reading that treats the drug as having one fixed target, this is the awkward result. The molecule did not change between the two lanes; only the clock did.
It is inconclusive for my question. Defensible if your question was strictly about the residue on mTORC1 outputs. This step answers a neighbouring question that turns out to matter more for what the drug does to an animal.
It introduces an alternative explanation. It introduces a second axis rather than an alternative: chemistry decides depth, and duration decides breadth. Most confusion in this literature comes from reporting one and reading the other.
What this evidence supports
- The drug's target list depends on how long it has been on: acutely the second complex is left alone, and after prolonged exposure new complex stops being assembled.
- Any description of what this molecule blocks is unfinished without a duration attached, which is most of why two honest groups can report different target lists.
What it does not establish
- That this happens in every cell type or at every dose. The reported sensitivity varies between lines, and the effect is dose- and duration-dependent, which is why the Atlas grades this edge contested.
- That losing the second complex is what produces any whole-body effect — that needs an animal rather than a lysate.
Information value High, and it is the reason this challenge has a second axis at all. It separates the wrong-complex reading from the other three, and it leaves the residue on 4E-BP1 exactly where it was, since that sits inside mTORC1 signalling.
New information
The same loss of second-complex signalling turns up in mouse liver, where prolonged treatment goes together with disturbed glucose handling — and in the same work, reducing mTORC1 signalling genetically, without the drug, extended lifespan in female mice without that metabolic cost. LAM2012
Does this affect your conclusion?
What each choice commits you to
Yes, substantially. This is the reading the design supports. The benefit and the metabolic cost come apart in that experiment, so ‘what does the drug block’ stops being one question and becomes two, with different answers and different consequences.
Possibly. Reasonable if you doubt how far a liver lysate travels. Say what would move you — matching target engagement between the genetic and the drug arms is the comparison that would make the separation readable.
No. Hard to hold once the genetic arm is on the table: it reaches the lifespan endpoint without the glucose phenotype, so attributing both to one undifferentiated ‘mTOR inhibition’ no longer fits what was measured.
Not enough information. Also defensible, and it points at a missing measurement rather than missing knowledge: what is absent is a matched comparison of schedules and molecules against the same endpoints in the same animals.
Two things that had travelled together in the literature separate here. Reduced mTORC1 signalling reached the survival endpoint, while the metabolic cost tracked the loss of the second complex — which reframes the question from ‘how much does the drug inhibit’ to ‘which parts, for how long, and at what cost’. The separation is reported in mice, so it is a strong lead about mechanism rather than a statement about people.
The control that would help Vary the schedule at a matched per-injection dose (which lowers cumulative exposure) and read target engagement and glucose handling in the same animals; or hold the schedule and change the molecule to one that spares the second complex. Both of those are on the menu below.
15 research units
Change the schedule rather than the molecule
Addresses Whether the metabolic cost can be separated from target engagement by timing alone
Needs Mouse cohorts on different dosing intervals, glucose and pyruvate tolerance tests. Cheap only because the cohorts already exist.
The design
| Model | Mice dosed daily, weekly, or every fifth day |
|---|---|
| Perturbation | Vary the dosing interval at a matched per-injection dose (which lowers cumulative exposure) |
| Readout | Glucose and pyruvate tolerance, fasting insulin, beta-cell function, alongside mTORC1 output |
| Control | Untreated littermates, and the daily-dosed group as the comparison that makes the interval the variable |
Before you run it: what would change your mind?
What result here would weaken the reading you are holding?
What each choice commits you to
Spacing the doses changes nothing — the cost tracks the dose, not the exposure. That would tie the metabolic cost to target engagement itself and leave scheduling as a dead end. It is the result that would push the whole question towards a different molecule.
Spaced schedules spare glucose handling while mTORC1 inhibition is kept. What the source study reports, with the endpoint it actually measured: the harm shrinks. Whether the benefit survives is not in this experiment. A follow-up study with the same schedule reported extended lifespan in female C57BL/6J mice (Arriola Apelo 2016), so harm reduction plus benefit has been shown once, in one sex and one strain.
Spacing removes the metabolic cost and the target engagement together. That would make the schedule a blunt dose reduction rather than a separation, which is the alternative reading this design exists to rule out.
Nothing here would change my mind. Then the fifteen units are confirming a schedule you have already chosen, and the missing benefit endpoint will still be missing afterwards.
Which of these the step returned
B. Spaced schedules spare glucose handling while mTORC1 inhibition is kept
Run · 15 research units spent
Not enough budget left for this one. What you have already spent stays spent.
What it opened up If a calendar can separate them, the question is whether a molecule can do it more cleanly.
Derived from A Alternative rapamycin treatment regimens mitigate the impact of rapamycin on glucose homeostasis and the immune system 2015
Disturbance of glucose tolerance
Direction and rank order as reported in the source study, in which spaced schedules largely spared glucose handling while mTORC1 inhibition was retained. Schematic, not measured values.
Interpret — what does this let you say?
What each choice commits you to
It supports my hypothesis. For the wrong-complex reading it supports the practical half: if the cost tracks prolonged exposure, spacing the exposure should reduce it, and it did in these cohorts.
It weakens my hypothesis. It weakens any reading in which the drug's harms follow inevitably from its target engagement, since here the two were separated by a calendar alone.
It is inconclusive for my question. Right for the question as asked. Nothing here touches the residue on mTORC1 outputs; it is about the cost of keeping the drug on.
It introduces an alternative explanation. It introduces a design lesson: an endpoint that was never measured is the easiest thing in the world to read into a result. The benefit endpoint is missing from this study by design.
What this evidence supports
- Spacing the doses kept glucose handling near the untreated range while mTORC1 inhibition was still there, so the two can come apart in a living animal without changing the molecule.
- Schedule is a variable in its own right, alongside dose and chemistry — and it is the one most often left out of the methods sentence.
What it does not establish
- That the benefit survives the change. This study measured side-effect endpoints and no lifespan or healthspan outcome, so it shows the harm shrinking rather than the benefit being kept.
- That the second complex was untouched on the spaced schedules. What was reported is that no effect on it was detectable under that assay, which is a narrower statement.
Information value Cheap and load-bearing for the last sub-question, and the step most likely to be over-read. Notice how easily the honest version — harm reduced, benefit not measured — turns into the sentence people actually quote.
35 research units
Build a molecule that blocks mTORC1 deeply and leaves the second complex alone
Addresses Whether depth of block and complex selectivity can be had at the same time
Needs A bivalent compound, resistant breast-cancer models, patient-derived xenografts, and then a first-in-human dose-escalation trial. The whole apparatus of drug development, and priced accordingly.
The design
| Model | Human breast-cancer lines and mouse patient-derived xenografts; then dose escalation in advanced solid tumours |
|---|---|
| Perturbation | A bi-steric molecule engaging an FKBP12-dependent site and the active site at once |
| Readout | Suppression of p-4E-BP1 and the state of Akt Ser473, plus safety and disease control in the trial |
| Control | Rapalog-treated and untreated models in the preclinical arm; no randomised comparator in the trial, which is the design's main limit |
Before you run it: what would change your mind?
Before the selective molecule: name the result that would weaken your model.
What each choice commits you to
Depth and selectivity trade off — reaching 4E-BP1 costs the second complex. That would make the two properties inseparable and turn the metabolic cost into the price of a deep block. It is the result that would end the selective-inhibitor programme, and it is the one to watch for.
The molecule suppresses 4E-BP1 and leaves Akt Ser473 largely spared. What the source studies report in resistant tumour models and in a single-arm dose-escalation trial. It shows the design problem is soluble; it shows nothing about an ageing endpoint.
The molecule behaves like a rapalog. That would say the bivalent design does not deliver what it promises, and the preclinical comparison on this card is the arm that would have shown it.
Nothing here would change my mind. Thirty-five units is the second-largest item on the menu. Buying it with no result in mind that would move you is the definition of an expensive habit.
Which of these the step returned
B. The molecule suppresses 4E-BP1 and leaves Akt Ser473 largely spared
Run · 35 research units spent
Not enough budget left for this one. What you have already spent stays spent.
What it opened up One step remains, and it is the one nobody in this corpus has taken.
Derived from A A bi-steric mTORC1-selective inhibitor overcomes drug resistance in breast cancer 2023 · H The Bi-steric, mTORC1-Selective Inhibitor, RMC-5552, in Advanced Solid Tumors: A Phase 1 Trial 2025
Depth of mTORC1 block against sparing of the second complex
Direction and rank order as reported in the source studies, which describe over twenty-five-fold selectivity for mTORC1 over the second complex alongside suppression of the rapamycin-resistant substrate. Schematic, not measured values.
Interpret — what does this let you say?
What each choice commits you to
It supports my hypothesis. For partial block, it closes the loop: the residue was real, it was drug-shaped, and a differently shaped drug removes it. For wrong-complex, it supports the practical half — the cost can be engineered away rather than merely rescheduled.
It weakens my hypothesis. For readout-artefact, decisively: a molecule designed against a supposed artefact would have nothing to show, and this one has preclinical and early trial data.
It is inconclusive for my question. Fair for the question this challenge ends on. It shows the chemistry is possible without showing what the chemistry buys where the benefit was claimed.
It introduces an alternative explanation. It introduces a scoping problem instead: the mechanism was read out in one disease area, and the benefit everyone is interested in was measured in another. Nothing joins them yet.
What this evidence supports
- The two properties are not forced to trade off: a molecule can reach the substrate rapamycin leaves alone and still leave the second complex largely spared.
- The residue identified in cells in 2009 was specific enough to design against, which is what a well-posed mechanistic question eventually buys.
What it does not establish
- That sparing the second complex is why the metabolic signal stayed low in the trial. It was open-label and single-arm with no head-to-head comparator, so the low rate sits against historical experience rather than against a control group.
- Anything about ageing: these are tumour models and an oncology trial, and the endpoints are tumour endpoints.
Information value High for the mechanism and deliberately silent on the sub-question left open. It is the step that shows the design problem is soluble, and the step that makes the absence of an ageing readout impossible to miss.
40 research units
Run a selective molecule against an ageing endpoint
Returns a prediction, not a result
Addresses Whether a block that spares the second complex keeps the lifespan benefit without the metabolic cost. No study in this Atlas corpus reports it: the selective molecules have been read out against tumour endpoints, and the survival data come from rapamycin.
Needs Long-lived cohorts of both sexes, a selective compound dosed for most of a lifespan, survival and metabolic endpoints in the same animals. The most equipment on this menu, and part of why nobody in this corpus has it.
The design
| Model | Genetically heterogeneous mice of both sexes, dosed for most of the lifespan |
|---|---|
| Perturbation | A selective mTORC1 inhibitor, against rapamycin and against vehicle |
| Readout | Survival, glucose handling and mTORC1 output in the same animals |
| Control | A vehicle cohort and a rapamycin cohort at matched target engagement, so the comparison is about selectivity rather than about dose |
Run · 40 research units spent
Not enough budget left for this one. What you have already spent stays spent.
Expected, not observed No study in this Atlas corpus reports this experiment. What follows is the outcome the model above predicts, written as a prediction and labelled as one. Treat it as a hypothesis to be tested, not as a result to cite.
Survival and glucose handling (predicted)
Expected, not observed. These bars are what the teaching model predicts if the benefit tracks mTORC1 and the metabolic cost tracks the second complex. Nothing in the Atlas corpus reports this measurement.
Interpret — what does this let you say?
What each choice commits you to
It supports my hypothesis. A prediction cannot support a hypothesis; it is generated by one. If this felt like support while you were reading it, that reflex is what the labelling is here to interrupt.
It weakens my hypothesis. Also not something a prediction can do. What it can do is tell you which hypothesis you would give up if the experiment came out the other way, which is worth writing down before it runs.
It is inconclusive for my question. Correct, and the useful follow-up is: what would you have to see here to change your mind? Naming that in advance is most of what makes an experiment worth its budget.
It introduces an alternative explanation. It introduces a design problem rather than an explanation: lifelong dosing, two sexes and matched engagement are three expensive requirements, and dropping any one of them changes what the survival curve would mean.
What this evidence supports
- Only a survival endpoint would turn the mechanistic separation into an answer about the benefit rather than a story about it.
- Running rapamycin alongside at matched target engagement is what would make the comparison readable, and it is the arm most likely to be cut for cost.
What it does not establish
- Anything at all, until it is done. Reading a predicted pattern as evidence is the specific error this challenge is built to make visible.
- Sex is not a nuisance variable here: the reported effects on lifespan differ between males and females, so a single-sex cohort would answer a narrower question than the one asked.
Information value The highest in principle, and a gap in this corpus rather than a line in it. Choosing it costs 40 units and returns a prediction — which is a fair picture of what committing to an unpublished experiment feels like from the inside.
Step 5
Revise the hypothesis
Your working hypothesis, after the evidence you bought
Changing your mind here is the expected outcome, not a failure state. A first hypothesis that survives every experiment untouched usually means the experiments were not discriminating.
What each choice commits you to
The block is partial by construction. The drug comparison is the strongest support available in this corpus for this reading, and it comes with a scope: what you have is a difference between two chemistries on a panel of substrates in cells. Holding it well means saying which substrates, at which dose, and keeping the feedback arm in view as a second contributor rather than a rival.
The incompleteness is in the readout, not the drug. Hard to hold once a second chemistry is in the experiment, because the residue moves when the drug changes and the antibody does not. Worth keeping a narrow version of it: a phosphosite is still a proxy, and the functional readouts in this challenge behave more like the phosphosites than most people expect.
The network answers back. Well supported as a real contributor and hard to hold as the whole answer: relief of the upstream brake happens under either chemistry, so it cannot by itself explain why one of them reaches further. The version that survives is a joint one, where obstruction sets the floor and feedback decides how fast the signal creeps back up.
The interesting part is the complex the drug was not supposed to touch. Right about the most consequential part of the drug's behaviour and slightly off the question as posed. The second-complex arm is where the metabolic cost lives and where the schedule and the selective molecule earn their place — but the residue on 4E-BP1 sits inside mTORC1 signalling and needs the chemistry comparison rather than the time axis.
Step 6
Compare with published research
Now, and not before, here is what someone actually did.
What the researchers actually tested
What an inhibitor that occupies the active site would reveal about the one everybody was already using — measured as phosphorylation of both mTORC1 substrates, as cap-dependent translation, and as autophagy induction, in matched cultures side by side.
What it answered
- Rapamycin leaves a class of mTORC1 outputs largely running, 4E-BP1 phosphorylation among them.
- Autophagy sits in that class too, so the drug's best-known effect is also its partial one.
- The residue is a property of the chemistry rather than a limit of the kinase, since a different chemistry reaches further.
What it did not answer
- It did not read the second complex over days, so the time axis of this challenge is outside its design.
- The active-site compound is not complex-selective, so it cannot say which complex each part of the deeper effect came from.
- It is cell-line pharmacology: how much of the residue matters in any particular tissue or tumour is not something this design measures.
The published experiment is not the perfect version of your design, and it was never meant to be. It is a real design with its own controls, its own scope and its own limits — and reading it that way rather than as a verdict is the skill this step is for. Compare it with what you chose: which of your unspent steps would you now run alongside it?
Study page →Step 7
Where the question stands
You have committed, spent, revised and compared. Here is what the evidence in this Atlas actually supports today — separated into what was measured, how it is read, and what is still nobody's answer.
The short answer
On the evidence in this Atlas: because of how it inhibits, and then, after long enough, because of where it ends up. The drug obstructs the doorway to the kinase rather than filling its catalytic pocket, so some substrates keep their access — and prolonged exposure adds a second target that acute exposure leaves alone.
Observation — what has actually been measured
- Rapamycin binds FKBP12 and the pair docks on a site adjacent to the kinase's active site rather than in it. CHO1996 SAB1995
- Treatment takes p-S6K1 down within minutes, and that band became the standard readout of mTORC1 activity. CHU1992 PRI1992
- Against an active-site inhibitor as a ruler, rapamycin leaves 4E-BP1 phosphorylation largely in place and induces autophagy only partly. THO2009 FEL2009
- Inhibiting mTORC1 relieves feedback onto receptor signalling, and Akt phosphorylation can rise rather than fall. ORE2006 HAR2004 ROD2011
- Prolonged exposure reduces assembly of the second complex in sensitive lines, and the same loss appears in mouse liver alongside disturbed glucose handling. SAR2006 LAM2012
- Spacing the doses in mice largely spared glucose handling while mTORC1 inhibition was retained; side-effect endpoints only, with no benefit endpoint measured. ARR2015
- A bi-steric molecule suppresses the rapamycin-resistant substrate while sparing the second complex, in resistant tumour models and in a single-arm dose-escalation trial. MEN2023 SCH2025
Interpretation — how the field reads it
Read together, these support a two-axis picture rather than a single number for potency. Chemistry decides the depth of the block on mTORC1 outputs: obstruction of the substrate channel leaves some substrates reachable, and occupying the active site does not. Duration decides its breadth: acutely one complex, after prolonged exposure often both. Most apparent disagreements about ‘how much rapamycin inhibits mTOR’ turn out to be two groups reporting from different corners of those two axes.
The network's reply is a real third contributor and not a rival to the first. Relieving the brake on the arm above the kinase happens under either chemistry, so it cannot be why one reaches further — but it does mean a treated cell is not an untreated cell with one arrow cut, and it sets how fast a partly blocked signal creeps back.
What the drug does, then, is not to turn mTORC1 output down by a fixed fraction but to reshape which outputs stay on. That reshaping is a plausible reason the same molecule can extend lifespan in mice while performing modestly as a single agent against many tumours, and it is why ‘rapamycin induces autophagy’ needs the dose, the cell type and the readout attached before it is a claim.
Still open — what nobody has done
- No study in this corpus joins a second-complex-sparing molecule to an ageing endpoint. The selective compounds were read out against tumour endpoints; the survival data come from rapamycin. That join is the 40-unit item on your menu that returns a prediction, and it is the same gap the Atlas records as H2.
- Two routes lead from mTORC1 inhibition to disturbed glucose handling — loss of the second complex, and relief of the S6K1-to-IRS-1 brake — and their relative contribution in people has not been apportioned. The Atlas's own edge on that feedback arm records the missing apportionment rather than papering over it.
- Whether prolonged exposure reaches the second complex in a given cell depends on line, dose and duration, and what the spaced-schedule study reports is that no effect on it was detectable under its assay. ‘Not detected here’ and ‘not happening’ are different claims, and the difference matters for anyone choosing a schedule.
How each hypothesis comes out
- The block is partial by construction
- The reading the evidence best supports for the question as asked, once it is stated with its scope: steric obstruction rather than occupancy, measured across a substrate panel in cells, with the chemistry comparison as the load-bearing result. It got sharper rather than failing — the qualified version names substrates and a mechanism instead of a potency.
- The incompleteness is in the readout, not the drug
- Does not survive the chemistry comparison. The residue moves when the molecule changes while the antibodies stay the same, and it shows up on a functional readout as well as on phosphosites. A narrow version is still worth keeping: a phosphosite is a proxy, and this challenge is partly about how much weight one band carries.
- The network answers back
- Right about a real contributor, wrong as the whole answer. Relief of the upstream brake is measured, and it happens under either chemistry, so it cannot explain why one reaches further. The version that survives is joint: obstruction sets the floor, feedback sets how quickly the signal comes back.
- The interesting part is the complex the drug was not supposed to touch
- Half-right in a way worth naming precisely. The second-complex arm is where the metabolic cost lives, and it is the half with consequences outside the dish — but it is time-dependent, graded contested, and silent about the residue on 4E-BP1, which sits inside mTORC1 signalling.
Step 8
What would you do next?
What changed your mind, if anything?
Which evidence mattered most to your final position?
The list above is where a next experiment would go. Pick the one you would defend in front of someone holding the budget — not the most interesting one, the one that would change what the field can say.
You have one more opportunity and a limited budget. What would you do next?
What each choice commits you to
Run a second-complex-sparing molecule against a survival endpoint, with rapamycin alongside at matched target engagement. The direct attack on the gap, and the most expensive option on any realistic menu. Budget both sexes and the rapamycin arm from the start: without the comparator the survival curve cannot separate selectivity from dose, and without both sexes it answers a narrower question than the one asked.
Apportion the two routes to disturbed glucose handling in one set of animals. Cheaper and badly needed. Loss of the second complex and relief of the S6K1-to-IRS-1 brake have never been weighed against each other in the same experiment, and the answer would tell you whether a selective molecule can spare the metabolic cost or only part of it.
Map the residue across a wide substrate panel and a dose-response, in more than one cell type. The unglamorous option that would firm up the foundation. Much of what we call rapamycin-resistant output rests on a handful of substrates in a handful of lines; knowing which substrates behave that way, and how reproducibly, would sharpen every design downstream.
Read the second-complex effect across lines, doses and durations to find where it actually appears. Follows the most contested edge in this story rather than the original question. The Atlas grades it contested because sensitivity varies; a systematic map would turn a caveat into a rule of thumb, which is worth more than it sounds for anyone choosing a schedule.
Step 9
Research notebook
Everything you decided on this page, on one page. It is a record of choices rather than a score — what you started with, what you said would change your mind, what you bought with the budget, and where you left the question.
Research notebook
With scripting on, the record of your own decisions collects here as you work through the page. Without it the record cannot assemble itself, because the decisions are yours rather than the page's — the sections it would hold are listed below.
The notebook, without JavaScript
With scripting on, this page keeps a running record of your own decisions and prints it here as one page you can copy out. Nothing is stored anywhere and nothing is scored — it is a record of what you chose, not of how well you chose. Without scripting the record cannot assemble itself, because the decisions are yours rather than the page's. These are the sections it would hold, and every piece of material in them is already above on this page.
- The question — the research question this challenge opens with
- My working hypothesis — the one you committed to before spending anything
- What I said would change my mind — your prediction at each step, and whether the step returned it
- Evidence I bought — each step you ran, what it cost and what it let you conclude
- What it could not settle — the other half of every result you bought
- Where the question stood when I closed it — sub-questions answered, and the ones left open at any price
- What changed my mind — your revised hypothesis and your own answer to why
- My next experiment — the one you said you would defend in front of a budget
Step 10
Challenge complete
No score, and nothing unlocked. What you take away is a shorter list of things you would need to find out, and a clearer sense of which experiment would tell you.
Research skills this challenge practised
- Causal reasoning
- Experimental design
- Control selection
- Evidence evaluation
- Alternative explanations
- Scientific uncertainty
Where to go next
Academy lessons
Guided routes
- Why doesn't rapamycin switch mTOR off completely?The rapamycin route, which is this whole question drawn end to end.
- What does a cell actually do when mTORC1 fires?The output route, where the substrates this challenge argues about do their work.
In the Atlas