Checkpoint failure has more than one address
RAS proved you have to match the lever to the lesion. The immune cycle has the same logic — and one coordinate, the egress barrier, that nobody has claimed.
For a decade, metastatic pancreatic cancer absorbed every immune drug the field had — checkpoint inhibitors, CAR-T, combinations — and none of them moved survival. Then daraxonrasib, a drug that does nothing to the immune system and instead jams the RAS driver the tumour runs on, roughly doubled it. The lesson was about aim: the immune levers were always the wrong tool, because the step that was blocking was never an immune step. The win came from matching the lever to the lesion.
That principle generalises. Treatment works when the lever matches the position that is actually blocking, and fails when it acts at a position the response never reached. The hard part is saying, for a given patient, which position is the one that is blocking — and checkpoint failure is where this breaks down.
A patient fails anti-PD-1 and the tumour is labelled resistant, or cold, or immune-excluded. These words describe an outcome, not a location. They lump together responses that failed at completely different points and send the same next drug at all of them. Read the cycle as a sequence of positions rather than a single arrow and those places become explicit — there are five, and every cancer sits at one of them.
The barriers — the five places a response stalls
| Step | Cycle location | What is blocked | Matched lever |
|---|---|---|---|
| Driver-dominated | off the immune cycle | tumour-intrinsic engine outruns any immune response | hit the driver (RAS, PI3K) |
| Build failure | Potentiality / Construction (pos 5–7) | no deployable response is generated | priming: vaccine, antigen, danger signal |
| Egress barrier | Selection (pos 8, Garrison) | response built, parked, never licensed to leave the node | CD40 licensing, before checkpoint |
| Engaged zone | Discovery to Equilibrium (pos 9–12) | nothing; this is where checkpoint works | anti-PD-1 / anti-PD-L1 |
| Suppression barrier | Conservation (pos 13–16) | response engages, then is actively shut down | de-suppress: anti-stroma, myeloid/Treg reprogramming |
The map
| Cancer | Phase reached | Rate-limiting step (hypothesis) | Why | Matched lever | Evidence status |
|---|---|---|---|---|---|
| MMRd / MSI-H (colorectal, endometrial) | Engaged | none; response gets to the synapse | high neoantigen load, T cells engage | checkpoint | established, checkpoint ~50% |
| Melanoma | Engaged | suppression / terminal exhaustion in non-responders | UV-driven high TMB, engages by default; Garrison absent (0/11) | checkpoint, then de-suppression | established + framework null confirmed |
| Urothelial / bladder | Engaged | partial; trafficking and build in non-responders | moderate TMB, responds to checkpoint and EV+checkpoint | checkpoint + ADC | established (e.g. VOLGA, EV+pembro) |
| MMRp / MSS colorectal | Built, parked | **egress barrier (Garrison)** | response built but stalled before the synapse, low PD-1 | **CD40 before checkpoint** | framework lock, OR 4.64 (Pelka) |
| Breast (subset of non-responders) | Built, parked | **egress barrier (Garrison)** | parked CD8 state present in non-responders | **CD40 before checkpoint** | framework lock, 12/20 vs 0/9 (Bassez) |
| Pancreatic (PDAC) | blocked early | **driver + suppression**, not egress | dense stroma, Tregs / MDSC / TAM, RAS engine re-feeds growth | **hit the driver (RAS)**; de-suppress | established + RAS proven (RASolute 302) |
| Glioblastoma | Build failure | priming + immune exclusion | low TMB, immune-privileged, T cells excluded | priming / vaccine; trafficking | established checkpoint resistance |
| Renal (RCC) | Engaged via vasculature | trafficking / vascular | responds despite low TMB, VEGF-driven | anti-VEGF + checkpoint | established |
| Prostate | cold | build failure + suppression | low TMB, immunosuppressive, largely checkpoint-resistant | priming + de-suppression | established |
| NSCLC (splits) | varies | high-TMB engages; KRAS / EGFR driver-dominated; low-TMB build | depends on driver and mutational load | checkpoint, or driver drug, or PD-1×VEGF | established, splits by subtype |
Which cells are earned. Two of these placements are measured: the egress barrier and the pancreatic suppression block both rest on data. The engaged cancers are nearly self-evident, since they respond to the drug that acts at the engaged step. The rest — melanoma's non-responder mechanism, prostate, glioblastoma — are the lens predicting where to look, and the part that still needs testing. The map is a hypothesis generator, honest about which cells it has earned.
The address that was missing
One position on that list is not in the standard diagram at all: egress. A T cell that finishes its effector programme does not leave the lymph node on its own — it waits for a licensing handshake from a dendritic cell, delivered through CD40, before it is cleared to travel to the tumour. The usual cancer-immunity cycle compresses that release into a single arrow from priming to trafficking, as if exit were automatic. It is a gate, and a finished response can pile up behind it.
The cells that do are built, armed, tumour-specific, and not exhausted. They hold their stem and memory markers, carry little PD-1, and have never fired the engagement signal that marks arrival at the tumour — a response parked one step short of deployment. Anti-PD-1 has no purchase on them, because it releases a brake these cells never applied. A patient can carry a real anti-tumour army, fail checkpoint completely, and look cold on every standard read while doing it.
This state has a fingerprint that was fixed before any data was inspected. It separates checkpoint-failing MMRp colorectal from responding MMRd at an odds ratio of 4.64 (Pelka), repeats without modification in breast at 12 of 20 non-responders versus 0 of 9 responders (Bassez), and is absent in melanoma, 0 of 11 — the negative case the framework predicted, because melanoma reaches the engaged phase by default. One definition, three cancers, three independent cohorts.
The call
The claim is narrow and falsifiable: a large share of checkpoint failure is not the absence of a response but a response stalled at the egress gate, needing the gate opened, not the brake released. The lever is CD40 licensing, given before the checkpoint so the army reaches the synapse first, where the checkpoint then has something to act on.
And it costs nothing new to test. CD40 agonists have already been run with checkpoint inhibitors in completed trials. Score the baseline biopsies for the parked state and ask whether the benefit of adding the agonist concentrated in the patients who carried it — a mixed trial result turned into a stratified subgroup, with no new patient and no new drug.
RAS was the unclaimed win on the driver axis, in plain sight for a decade while the field reached for the wrong tool. The egress barrier is the same shape of opportunity on the immune axis: a definable lesion, a matched lever, and patients mislabelled as cold while carrying the response that would have worked. RAS proved the method. This is where it should pay off next.
This is a research preprint, not medical advice. The predictions described have not been clinically validated and must not guide treatment decisions.