Brian Reed: Eliza, I'm going to open with a genuinely uncomfortable question — not as a setup, I actually want to hear you defend it.
Eliza Ward: Oh, this is already a bad sign — what are we getting into?
Brian Reed: Precision oncology. You keep saying it works. Explain this to me: a lung cancer patient with a KRAS mutation gets a targeted drug and goes into remission. Pancreatic cancer patient, exact same KRAS mutation, exact same drug — fails completely. Same mutation. Same drug. Different body part. How does that happen if the mutation is what's driving the disease?
Eliza Ward: Okay — wait, I'm not sure I'd say precision oncology 'works' without qualification, but let me actually answer the question. The Falcomatà research — Chiara Falcomatà, working with the German Cancer Consortium, Technical University of Munich, University Medical Center Göttingen — that's the documented answer. Same activating oncogene mutation, pancreatic precursor cells versus bile duct precursor cells, fundamentally different cellular outcomes.
Brian Reed: Which means what, for the mutation framework?
Eliza Ward: It means — actually, no, let me not jump ahead — it means the mutation's effect cannot be separated from the cell it lands in. Tissue-specific genetic interactions. The regulatory environment, the existing co-mutations, the cell type itself. That's not a footnote to the KRAS finding. That's a challenge to calling it a 'driver' at all, because driver implies it drives the same thing everywhere.
Brian Reed: And Professor Dieter Saur's team showed that's not what it does.
Eliza Ward: Right — but here's where I want to be precise. Is this the mutation framework being wrong, or is it the mutation framework being applied to the wrong unit of disease?
Brian Reed: That is the question. And I genuinely don't think the DKTK finding resolves it cleanly. Because the case for sequencing as the right foundation — that's not a naive position. Deep sequencing has produced real druggable targets. Targeted therapies exist that work because of that framework. The issue is whether 'KRAS mutation' names one disease or several diseases that share a mutation label.
Eliza Ward: KRAS-in-lung and KRAS-in-pancreas as nearly separate diseases. That's a significant reframe.
Brian Reed: It's significant because the research consequence is massive — you're not adding profiling layers to one predictive model. You're building separate mechanistic maps per tissue type. The cellular context — cell type, tissue of origin, epigenetic state, microenvironmental signals — that becomes the primary variable, and the mutation is almost downstream of it.
Eliza Ward: And we cannot yet predict that for a specific patient. That's the part that's still unmapped — the mechanistic rules explaining why pancreatic precursor cells respond differently haven't been established yet.
Brian Reed: And that unmapped part — that's where I want to go, because I think the word 'unmapped' is doing a lot of work. The mechanistic rules aren't missing because nobody looked. They're missing because the variable isn't in the DNA. It's epigenetic state. Heritable changes in gene activity that don't touch the sequence at all — chromatin accessibility, transcription-factor landscape, metabolic state. None of that shows up on a sequencing panel.
Eliza Ward: Can you just — give me the plain version of that? Because 'chromatin accessibility' is going to lose people.
Brian Reed: Yeah — okay. Think of it like the same light switch wired into two completely different electrical systems. Flipping the switch turns the lights on in one house and blows a fuse in the other. The switch is identical. The wiring behind it determines what happens. That's the mutation versus the epigenetic state. Same KRAS flip, totally different cellular response, because the wiring — the chromatin, the transcription factors — is different in lung tissue than in pancreatic tissue.
Eliza Ward: And none of that wiring is visible on the sequencing report.
Brian Reed: None of it. Two tumors, identical mutation profiles, completely different biological states. The SIRT6 finding makes this concrete — SIRT6 is a longevity gene, and its absence, one gene, not a mutation in the driver, just an absent modifier — that alone can redirect whether a cell dies or survives an oncogenic signal. The apoptosis response just... flips. Cell gets the oncogenic hit, in one context it triggers programmed cell death, in another context it bypasses it entirely and proliferates.
Eliza Ward: Hold on. SIRT6 absence changes the apoptosis outcome — not the mutation, not the drug, just whether this one longevity gene is present.
Brian Reed: That's the finding. And SIRT6 isn't a driver mutation, it's not what you'd sequence for. It's a non-mutational modifier that changes whether the oncogenic signal kills the cell or lets it survive. That's epigenetic regulation in action — heritable, consequential, invisible to the standard panel.
Eliza Ward: Okay, so — wait, I want to make sure I'm not running ahead of this. You're saying two tumors can look genetically identical and be in fundamentally different biological states because the epigenetic state doesn't get captured by DNA sequencing.
Brian Reed: That's exactly the claim. And it's not speculative — the 2023 review on non-genetic heterogeneity formalizes it. Genetically identical cancer cells, distinct phenotypic states, switching under treatment pressure without acquiring a single new mutation.
Eliza Ward: Which is why the resistance happens without a traceable genetic change.
Brian Reed: Right — and here's the part that I find genuinely hard to sit with. If epigenetic state is the hidden variable, the question that follows immediately is: why aren't we measuring it? Why does every oncology panel still lead with DNA?
Eliza Ward: Because we don't have prospective clinical tools to read epigenetic state before treatment. A researcher can characterize chromatin accessibility in a lab setting after the fact. An oncologist with a biopsy on Tuesday morning cannot. There's no validated assay she can run that tells her: in this specific tissue, with this specific epigenetic profile, the KRAS mutation will probably do X. That tool does not exist at the clinical level yet.
Brian Reed: So the sequencing panel isn't wrong — it's just answering an easier question than the one we actually need answered.
Eliza Ward: But that's — wait, that framing is almost too generous to the framework. 'Answering an easier question' implies we just need a harder tool. The 2023 review isn't saying we need better sequencing. It's saying the cells are doing something sequencing was never designed to see.
Brian Reed: That's the punch I want to land. The 2023 review on non-genetic heterogeneity — the finding is that genetically identical cancer cells can switch phenotypic states under treatment pressure without acquiring a single new mutation. So we hit the tumor with a targeted inhibitor, we eliminate the dominant clone, and the remaining cells don't evolve. They just... shift. Different mode, no new DNA, no trace.
Eliza Ward: No genetic trace of the escape.
Brian Reed: None. And that's where the pathologist scenario hits hardest — biopsy comes back, clean KRAS target, drug prescribed, three months later the tumor's back. She re-sequences. Nothing new. The cells didn't evolve under drug pressure. They switched phenotypic state. That's phenotypic plasticity as a resistance mechanism, and it leaves no genetic fingerprint.
Eliza Ward: Okay but — I don't buy that this means sequencing is the wrong foundation. That's too far.
Brian Reed: I'm not saying abandon it. I'm asking what it's actually buying you if resistance can arise without a detectable genetic change. Intra-tumor heterogeneity means there were already epigenetically distinct subpopulations in that tumor before the drug arrived. The drug didn't create the resistant cells. It selected for a pre-existing state.
Eliza Ward: Right — and that subpopulation was invisible on the original panel.
Brian Reed: Invisible. The question is — 'necessary but not sufficient' as a description of what mutations tell us. That sounds like a careful scientific hedge, but actually, no — I think it's describing something more serious. If the cell-state switching happens without new mutations, then sufficiency isn't even the right frame. The mutation might be downstream of the state that determines outcome.
Eliza Ward: That's — I mean, that's the destabilizing version of the claim. And I'm not sure the evidence gets you all the way there yet. Whether resistance is driven primarily by plasticity or by new mutations under selective pressure — that's explicitly unsettled in the same review.
Brian Reed: It is unsettled. But the fact that it's unsettled is itself the problem for precision oncology, right? The whole clinical infrastructure was built on the assumption that resistance meant evolution — new mutation, natural selection, resistant clone emerges. If state-switching can do the same job without any evolution, that causal chain is incomplete.
Eliza Ward: Incomplete — not wrong.
Brian Reed: Incomplete in a way that changes what you'd prescribe. That's not a semantic distinction for the patient whose tumor came back with no new mutations.
Eliza Ward: No, I'll give you that. The clinical consequence is real. And the part we haven't touched yet — the resourcing side of whether this is actually a science gap or a decision about where money goes — it gets uncomfortable in a different way.
Brian Reed: Yeah — the molecular profiling tools beyond genomics exist in research settings. They're not in clinic. And that gap isn't purely a science problem.
Eliza Ward: And that gap — that's the thing I want to push on, because I think there are actually two separate claims living inside it. One is: we don't have the tools yet. The other is: we've made a decision, collectively, not to build them as fast as we could. Those are different problems.
Brian Reed: Yeah, hang on — are you saying the field is under-investing in epigenetic profiling, or that the science isn't ready to be invested in?
Eliza Ward: Both. And they're not separable. Molecular profiling beyond genomics — we're talking epigenetic state, chromatin accessibility, transcription-factor activity, metabolic profile, tumor microenvironment signals — that layer exists in research settings. It's not mysterious. It's just expensive, slow, and hasn't been validated prospectively in clinical trial design.
Brian Reed: Not yet validated, or not yet funded to be validated?
Eliza Ward: Right — that's exactly the uncomfortable version of the question. Picture a trial statistician, let's say she's designing a Phase II study right now, and she has to decide: do we profile these tumors by DNA panel only, or do we add chromatin accessibility assays and transcription-factor mapping? The DNA panel is fast, cheap, reimbursable, and every oncologist knows what to do with the output. The multi-omic layer — I mean, she can't even tell the clinical team what to prescribe based on it yet. There's no validated decision rule.
Brian Reed: So she defaults to the sequencing panel. Not because it answers the question, but because it's the only answer she can act on.
Eliza Ward: Which then generates more sequencing data, which gets cited as the evidentiary base, which is how a resourcing default becomes a scientific paradigm.
Brian Reed: That's — wait, is that too strong? Because sequencing did produce real targets. That's not a funding artifact.
Eliza Ward: No, it's not — and I want to be clear about what I'm actually conceding here. Deep sequencing defined druggable targets. It built the therapies that exist. That's a genuine advance, not in dispute. The claim I'm making is narrower: the same infrastructure that made sequencing clinically usable hasn't been built for epigenetic or microenvironmental profiling. And without that infrastructure, we can't know whether the tissue-context problem — why KRAS-in-pancreas and KRAS-in-lung diverge so completely — is scientifically intractable or just underfunded.
Brian Reed: The DKTK finding is documented. The mechanism behind it — why Chiara Falcomatà's pancreatic precursor cells respond differently from bile duct cells — is still unmapped. That's a science gap.
Eliza Ward: It is. And mapping it per tissue type — actually building that mechanistic model — is vastly more expensive than adding a panel. You're not iterating on one predictive model. You're building separate maps. That's a resource decision with a real price tag, not a rhetorical one.
Brian Reed: So the field has documented that tissue identity may make these nearly separate diseases. It just hasn't funded the work to prove it clinically. And until it does, the oncologist is still prescribing off the mutation match.
Eliza Ward: Which is my actual position, and I don't have a clean way out of it. Because the next frontier everyone points to — prospective maps of microenvironmental and epigenetic features, per tissue, per mutation — that's the stated answer. But we can't do that yet. Not in clinic. Not before treatment.
Brian Reed: And that's the question I can't shake. Whether adding those molecular layers actually solves the prediction problem — or whether tumor behavior is driven by dynamic plasticity so fluid that there's no fixed measurable state to map in the first place. Like, what if we build the epigenetic profiling infrastructure, we get it into clinic, and the cells have already switched by the time the assay runs?
Eliza Ward: That's — yeah. That's the darker version.
Brian Reed: So until we can actually predict when a mutation behaves as a driver versus just... sitting there as a passenger, precision oncology is precise in what it targets. It's not precise in what happens next. And I don't know if that's a missing tool or a missing concept. Because those require completely different responses. A tool problem you can fund. A concept problem means the framework itself has a gap we haven't named yet.
Eliza Ward: I think — I was about to say I think it's the tools. But I'm actually not sure that's defensible right now.
Brian Reed: I don't think we know yet.
Eliza Ward: No. We don't. Good conversation to not have an answer to.
