Dr. Nathan Hayes: Maya, hey — I want to start somewhere uncomfortable today, if that's all right.
Maya Chen: Yeah, I think that's where this one lives, honestly.
Dr. Nathan Hayes: So the thing I keep returning to: Paracelsus writes 'sola dosis facit venenum' around 1538 — the dose alone makes the poison — and we have five centuries of pharmacology and toxicology built on that sentence. And then you pick the most neutral substance you can think of, and the sentence still holds perfectly.
Dr. Nathan Hayes: Six liters consumed over a few hours produces lethal hyponatremia — sodium dilutes in the blood, cells absorb the excess water, intracranial pressure rises. People have died this way. It's not rare.
Maya Chen: And I want to stay there for a second, because — mm, I think our instinct is to file that away as a curiosity and move on. But it's actually the whole argument. Water, oxygen under pressure, table salt — they're all following the same curve.
Dr. Nathan Hayes: The dose-response curve. Log dose on the x-axis, magnitude of biological response on the y — and for most substances you get this sigmoidal shape. A threshold below which nothing measurable happens. Then a steep response region. Then a plateau.
Maya Chen: And every major regulatory body is running on this. The FDA, the EPA, the European Medicines Agency — none of them say 'safe' or 'unsafe' as a category. They set windows. Therapeutic ranges. Acceptable daily intakes.
Dr. Nathan Hayes: Which is the honest position. The window is where the remedy lives. Outside it — in either direction, sometimes — is where harm begins.
Maya Chen: In either direction — wait, can you say more about that? Because I think people hear 'lethal dose' and assume harm is always a matter of too much.
Dr. Nathan Hayes: So below the threshold, you have no effect at all — which for a drug means no benefit. That's also a failure. The curve has two dangerous edges.
Maya Chen: Which means the question 'is this substance dangerous' is sort of... it's not even the right question to ask.
Dr. Nathan Hayes: Right — and that's exactly the reframe. The question isn't dangerous or not. It's: where on the curve is this dose, and how wide is the window around it. Now, picture it concretely. One sip of whiskey — nothing registers. Three drinks — your balance, your judgment, measurably altered. A bottle in an hour — your respiratory system is in crisis. Same molecule. Three points on the same curve.
Maya Chen: That's the whole thing, isn't it. Not a metaphor — that's literally what the curve is.
Dr. Nathan Hayes: Literally. So the vocabulary that comes out of this — threshold is the floor, below it nothing happens. ED50 is the midpoint — the dose that produces the target effect in 50% of a population. And LD50 is the danger ceiling — the dose lethal to 50%. Between ED50 and LD50 sits the therapeutic window. That gap is, I'd argue, the most important number in clinical medicine.
Maya Chen: Most important — why that specifically?
Dr. Nathan Hayes: Because it tells you how much room for error you actually have. Wide window — ibuprofen, say — someone takes three times the label dose and they're uncomfortable, maybe, but they're not in cardiac arrest. Narrow window — and this is where I want to go, because it's not abstract — Aconitum kusnezoffii. Used in traditional Asian medicine for centuries, for joint pain, for inflammation. Milligrams. We are talking about milligrams separating a therapeutic dose from one that stops the heart.
Maya Chen: Hold on. Milligrams.
Dr. Nathan Hayes: That's the window. And what that means practically — a preparation that's slightly more concentrated than expected, a patient who metabolizes it faster, a practitioner who eyeballs the dose — any one of those and you've crossed the line. The curve hasn't changed. The window just doesn't forgive.
Maya Chen: I keep thinking about Diane — the hypothetical patient. She's taking an Aconitum kusnezoffii preparation for chronic joint pain, something her family has used for generations. And she's not thinking about milligrams. She's thinking, this helps me move in the morning.
Dr. Nathan Hayes: And that's where the narrow window becomes — I mean, it becomes a clinical emergency waiting on probability. Because she's not wrong that it can work. The therapeutic effect is real. But potency and window are two separate things, and Aconitum has both: high potency, terrifyingly narrow margin.
Maya Chen: Whereas if Diane were taking ibuprofen for that same pain, the window is so wide that — sort of — the curve is almost forgiving by design.
Dr. Nathan Hayes: That's not accidental. That's what pharmaceutical development optimizes for — pushing the LD50 as far from the ED50 as possible. Many OTC drugs have wide windows precisely so they can be self-administered without clinical supervision. The difference between those two scenarios for Diane isn't semantic. It is clinically life-or-death.
Maya Chen: And yet Aconitum has been used for centuries — so the window was navigated somehow, without any of this vocabulary.
Dr. Nathan Hayes: Navigated, yes, and also — dosing errors with Aconitum are still a documented cause of toxicity. The centuries of use don't mean the window widened. They mean practitioners developed empirical rules for staying inside it. But those rules aren't the same as knowing the curve. And that gap — between empirical tradition and characterized dose-response — that's still unresolved.
Maya Chen: But that gap — between tradition and the characterized curve — that's sort of the clean version of the problem, right? Because there's a version of this where the curve itself... doesn't hold the shape we think it does.
Dr. Nathan Hayes: Say more about that.
Maya Chen: BPA. Bisphenol A. Because everything we've been describing assumes the curve goes one direction — more dose, more effect, up to a plateau. Sigmoidal. But with BPA, that's... mm, that's not what happens.
Dr. Nathan Hayes: Right — and this is where I'd push hard, because it's not a minor caveat. What we're talking about is a nonmonotonic dose-response relationship. The effect doesn't increase consistently with dose. It goes up, and then — at a higher dose — it reverses. Or disappears. Or does something structurally different.
Maya Chen: The model breaks.
Dr. Nathan Hayes: The model breaks. And with BPA specifically, the low-dose endocrine effects — on estrogen receptor signaling — are qualitatively different from the high-dose effects. Not just smaller. Different in kind. A low dose produces a response that a medium dose actually doesn't.
Maya Chen: Wait — so the medium dose is... safer? In that specific sense?
Dr. Nathan Hayes: That's exactly what makes it incoherent from a regulatory standpoint. There's a named pattern here — hormesis. Low-dose stimulation, high-dose inhibition. The inverse of what a threshold model predicts. And the EPA's standard framework — acceptable daily intake, all of it — is built on the assumption that below a threshold, nothing meaningful happens. But if BPA is producing endocrine effects at low doses that vanish at the doses you're actually testing, you will never see them. The test is designed for a curve shape that BPA doesn't follow.
Maya Chen: So the safety tests are — I mean, they're not wrong, exactly. They're just aimed at the wrong part of the curve.
Dr. Nathan Hayes: That's the precise problem. Standard high-dose testing can miss real low-dose harms entirely when the dose-response is nonmonotonic. And this has been formally recognized — there's ongoing scientific debate in the published literature about whether current regulatory thresholds are even sufficient for compounds like BPA. It's not fringe. It's a live disagreement.
Maya Chen: And that's where Paracelsus starts to feel a little — not wrong, but, sort of... it becomes rhetorical cover? Like, 'the dose makes the poison' is technically true but if the curve bends back on itself, saying that doesn't actually tell you where the harm lives.
Dr. Nathan Hayes: Correct. And it gets considerably more uncomfortable when you factor in something we haven't touched yet — which is that plasma concentration, what the EPA measures, what the FDA models on, isn't the same as the dose actually reaching the receptor inside the cell. That's a different problem, and honestly it makes all of this harder.
Maya Chen: Yeah — and I think that's the part I want to pull on, because if you're not the population average, if your metabolism is doing something the ED50 doesn't account for... the window we've been describing may not be your window at all.
Dr. Nathan Hayes: Right — and that's the thing the sigmoidal model papers over completely. Exposure-response analysis exists because the number on the prescription and the number at the receptor are not the same number. Not even close, in some patients.
Maya Chen: Wait — so what are we actually measuring when we set a dose?
Dr. Nathan Hayes: Plasma concentration. What's circulating in the blood over time. And that's — I mean, it's measurable, it's real, but it's a proxy. What drives the biological effect is what arrives at the receptor inside the cell. Those two numbers can diverge significantly depending on how an individual metabolizes the compound.
Maya Chen: Which is why the statin story lands so hard for me. A patient at a standard atorvastatin dose — 40 milligrams, well within the therapeutic window — develops elevated creatine kinase. Actual muscle breakdown.
Dr. Nathan Hayes: And the administered dose was fine. By every population-level metric, she was safe. But intracellularly — her exposure-response relationship was different. The curve the population data built wasn't her curve.
Maya Chen: That's the part that — mm — sort of quietly undoes the reassurance, doesn't it. ED50 describes the 50th percentile. That's the definition. So the outlier, the person whose metabolism runs differently, was never in the model.
Dr. Nathan Hayes: Never in the model by design. LD50, ED50 — population averages. They tell you about the median individual. The hypersensitive outlier can cross into toxicity well below the threshold the curve marks as safe, because the curve was never calibrated to them.
Maya Chen: And ADCs — antibody-drug conjugates — I keep thinking about those here, because they're almost engineered as an answer to this problem, right? Target the delivery directly to tumor cells, widen the therapeutic window.
Dr. Nathan Hayes: They widen it, yes — but they still require meticulous exposure-response analysis. Because even with targeted delivery, the gap between what enters the bloodstream and what actually arrives at the tumor varies person to person. You've solved part of the delivery problem, not the individual variation problem.
Maya Chen: So even the most sophisticated targeting we have still can't fully close the distance between administered dose and receptor dose.
Dr. Nathan Hayes: Not fully. And — actually, this is what the 2022 rhubarb metabolomics study does that I find genuinely clarifying. Researchers applied integrated dose-response analysis to rhubarb — a plant used in traditional medicine for centuries — and identified 0.69 grams per kilogram as the optimal dose. That number hits 90% of the effective dose while keeping three specific adverse reactions at acceptable levels. It took rigorous characterization to find it.
Maya Chen: Centuries of use and you still needed that methodology to find the actual number. Which means — mm, the gap between what you give someone and what safely arrives has always been there. It's not a modern pharmaceutical problem. It predates all of this.
Dr. Nathan Hayes: Paracelsus pointed at the right question in 1538. What he couldn't know — what we're still mapping — is that the dose making the poison isn't the dose in the bottle. It's the dose at the site of action. And for some individuals, those are separated by a gap that population curves were never built to see.
Maya Chen: Water keeps coming up — this is going to sound strange — water. We started there. Six liters, lethal hyponatremia. And that felt like the clean version of the principle, right? The thing that made Paracelsus feel airtight. But now I'm sitting here thinking, the reason water is the clean example is because its curve actually is sigmoidal. It does what the model says. And BPA doesn't. And the person taking atorvastatin whose creatine kinase spikes — her curve didn't either.
Dr. Nathan Hayes: The water case is almost — I mean, it's the one where the FDA and EPA's classical framework works perfectly. Because the dose-response is monotonic. Predictable. But 'sola dosis facit venenum' is 500 years old, and what Paracelsus couldn't have known is that for some compounds — BPA, certain endocrine disruptors — the dose at the receptor and the dose we're actually measuring and regulating aren't even tracking the same phenomenon.
Maya Chen: 'Largely' is doing a lot of work.
Dr. Nathan Hayes: It really is. 'Largely true' — that's where hormesis lives, that's where nonmonotonic responses live, that's where the individual outlier the ED50 was never built to see lives.
Maya Chen: Mm. And the question I don't think we answered — I'm not sure it can be answered yet — is whether personalized medicine, real-time biomonitoring, eventually makes it possible to replace the population threshold with something that actually knows where a specific person sits on the curve. Not the 50th percentile. Them.
Dr. Nathan Hayes: That's the right question to not answer. Because answering it prematurely is how we got here — confident in a model that's sufficient for most people, most of the time, and quietly wrong for the rest.
Maya Chen: Good place to stop, I think. Thank you for going into the uncomfortable parts with me today.
