Before you can locate your place on a dose-response curve, you have to know what kind of curve you’re on. That sounds obvious, and it’s the single most-skipped step in the whole subject. People argue endlessly about whether coffee, sunlight, alcohol, or stress are “good” or “bad” without ever agreeing on the one thing that settles it: the shape of the relationship between dose and effect. Get the shape wrong and every downstream decision — how much, how often, when to stop — inherits the error.
A dose-response curve is just a graph: dose along the bottom, effect up the side. It turns “how much of X, and what does it do to me?” into a line you can point at. And it turns out that almost every dose-response relationship in biology, toxicology, medicine, and economics is one of just three shapes. Learn to tell them apart on sight and you have the master key to this course. This lesson builds that eye — and then shows you the trap that swallows people who forget the curve is curved.
Before you read — take a guess
Before we start — take a guess. A toxicologist and a physiologist stare at data on the same substance (say, a trace mineral) and draw completely different curves: one draws harm rising straight from the first molecule, the other draws a hump that helps at low dose and harms at high dose. What's the most useful thing to say about this disagreement?
Shape one: linear-no-threshold — the pure poison with no safe dose
Picture a tax with no free bracket: every single dollar you earn is taxed at the same rate, right down to the first cent. There’s no exempt amount, no threshold below which you owe nothing. Damage accrues in exact proportion to the amount. That’s the mental image of the first shape.
The linear-no-threshold model (LNT) says the effect is directly proportional to the dose, and — this is the load-bearing part — the line passes through the origin with no safe threshold. Harm begins with the very first molecule and rises in a straight line from there. Double the dose, double the expected harm. There is no dose so small that its risk is exactly zero; there is only smaller risk.
LNT is the conservative default in radiation protection and carcinogen regulation. Agencies like the ICRP and the US EPA use it to set exposure limits precisely because it refuses to assume a safe floor: if you can’t prove a threshold exists, you regulate as if every increment of a genotoxic carcinogen or ionizing radiation carries proportional cancer risk. It’s a policy stance as much as a biological claim — err toward caution when the stakes are irreversible and the low-dose data are too noisy to settle.
Worked example. Suppose a chemical is modeled as LNT with a slope of 5 extra cancer cases per 100,000 people per unit of lifetime dose.
| Lifetime dose (units) | Extra cases per 100,000 (LNT: 5 × dose) |
|---|---|
| 0 | 0 |
| 1 | 5 |
| 2 | 10 |
| 4 | 20 |
| 10 | 50 |
Perfectly straight, straight through zero. Notice there is no “safe” row — even dose 1 carries 5 expected cases. That’s the signature: extrapolate the line back and it hits the origin, not a comfortable plateau.
The pitfall: LNT is an assumption, not a proven law
LNT is genuinely contested at low doses. It’s a defensible, cautious default for regulating irreversible genotoxic harm — but at very low doses the real curve for some agents may bend (a threshold, or even a hormetic dip), and the data are too sparse to tell. Treat LNT as “the shape we assume when we can’t afford to guess wrong,” not as “the proven shape of everything.” Assuming LNT for a substance that is actually hormetic is exactly the linear-extrapolation error this course exists to cure.
Shape two: threshold — nothing, nothing, nothing… then something
Now imagine a bridge with a weight limit. Drive one car, ten cars, a hundred cars across — nothing happens; the bridge shrugs. Keep piling on load and at some critical weight it starts to buckle, and past that point more load means more damage, fast. Below the limit: no effect. Above it: rising effect. That kink is a threshold.
The threshold model says there is a cutoff dose (a threshold) below which the effect is essentially zero, and only above it does the response climb. This is the workhorse shape of classical toxicology — captured in Paracelsus’s line that the dose makes the poison — where most substances have a No Observed Adverse Effect Level (NOAEL): a dose you can absorb with no measurable harm because your body’s repair, detox, and buffering systems mop it up until they’re overwhelmed.
The very same shape, flipped in meaning, describes nutrient adequacy: below a threshold intake you’re deficient and suffer; cross the threshold and you’re adequate and fine. (Push far past it and many nutrients turn toxic — which starts to look like the third shape.)
Worked example. A solvent with a threshold at dose 3 — below it the body clears it completely; above it, harm rises.
| Dose | Threshold model (harm) | LNT model (harm, slope 5) |
|---|---|---|
| 1 | 0 | 5 |
| 2 | 0 | 10 |
| 3 | 0 (at the threshold) | 15 |
| 4 | 8 | 20 |
| 5 | 16 | 25 |
Same substance, two models, wildly different advice at low dose: the threshold model says doses 1–3 are genuinely safe; LNT insists they already carry harm. Which one you believe changes the exposure limit by a lot.
When to use it
Reach for the threshold shape when there’s a plausible buffering or repair mechanism that has spare capacity — detox enzymes, DNA repair, renal clearance, a safety margin — so small insults are neutralized before they register. The failure mode is trusting a threshold that isn’t really there: declaring a dose “under the limit, therefore safe” when the true curve has no flat floor (the LNT worry), or when repeated sub-threshold hits accumulate faster than the system clears them. A threshold protects you per-dose, not necessarily per-lifetime.
A regulator says a food additive is 'safe below 3 mg/day' because studies find no measurable harm up to that intake, with harm appearing only above it. Which dose-response shape is that claim asserting, and what's the key risk in trusting it?
Shape three: hormetic — the hump where a little helps and a lot harms
Here’s the shape this whole course is named for, and the one people’s intuition handles worst. Think of watering a plant. No water: it withers. A sensible amount: it thrives. Drown it every day: the roots rot and it dies. Plot “plant health” against “water” and you don’t get a straight line or a simple cutoff — you get a hump: up, peak, down.
The hormetic curve is biphasic and non-monotonic. Two words worth pinning down precisely, because they are the definition:
- Biphasic means the response has two opposite phases as the dose rises — a stimulatory / beneficial phase at low doses and an inhibitory / harmful phase at high doses. One substance, two opposite effects, sorted by amount.
- Non-monotonic means the curve does not move in only one direction — it rises and then falls (an inverted-U for a benefit) or falls and then rises (a J-shape for harm). Contrast a monotonic curve (LNT, threshold), which only ever goes one way.
So the hormetic shape has something the other two lack entirely: an interior optimum — a best dose sitting in the middle, with worse outcomes on both sides. Too little and you miss the benefit; too much and you tip into harm. The classic biological engine under it is adaptive overcompensation (the next lesson’s subject): a mild stressor triggers a repair-and-defense response that overshoots baseline, leaving you stronger — but a large or relentless dose of the same stressor overwhelms that response and does net damage.
Worked example. A trace-mineral intake (think selenium-like) scored as net health effect, where deficiency hurts, an optimum helps, and excess is toxic:
| Daily intake (µg) | Net health effect | Region |
|---|---|---|
| 0 | −8 | deficiency (harm) |
| 20 | −2 | still deficient |
| 55 | +6 | near the optimum |
| 90 | +5 | past the peak, sliding |
| 150 | 0 | benefit gone |
| 400 | −10 | toxicity |
Read that column top to bottom: harm, less harm, peak benefit, then decline back into harm. No straight line and no single cutoff can reproduce that. Both “none is safest” (you’d stay at −8) and “more is better” (you’d end at −10) are catastrophically wrong. The right answer lives in the middle, and finding it is the entire skill.
How to read an optimum off a hump
A hormetic curve has three parts, and naming them fixes your bearings:
- The rising side — more dose still helps; you’re under-dosed and should (cautiously) add.
- The peak — the optimum; the best dose. More from here makes things worse, not better.
- The falling side — more dose now harms; you’re over-dosed and should back off. The only question that matters is which side am I on? “Good or bad” has no answer on a hump — it has an optimum and two wrong directions.
Let’s make all three shapes concrete in one place. Start the explorer on Hormetic and drag the dose across to watch benefit rise to the peak and then tip into harm. Then toggle the shape over to Linear (harm from the first molecule — shape one) and Threshold (flat, then harm — shape two) to feel how differently the same slider behaves. And switch on the linear-extrapolation ghost: that dashed straight line is what you’d wrongly predict if you assumed the hormetic curve was actually LNT — notice it screams “harm!” at exactly the low doses where the real curve is helping you.
Compare the shapes
Three shapes, one slider
The effect of a stressor is a function of the dose — and that function is usually curved, not straight. Pick a curve, then drag the dose to hunt for the optimum and watch benefit tip into harm.
Dose-response shape
- Response now
- +33
- Optimal dose
- 35
- Net effect
- net benefit
At a dose of 12, the response is 33 — net benefit. The optimum sits at a dose of 35. Past the peak, more is not better; it is worse.
The dashed line is what you get by extrapolating high-dose harm straight down to zero — it wrongly predicts harm at the low doses where the real curve helps.
Dosing
Which single feature does the HORMETIC shape have that neither the linear-no-threshold nor the threshold shape has?
Reading the optimum — and why “none” and “more” both fail on a hump
The reason humans are so bad at hormetic curves is that our default heuristics are monotonic. We reason “less bad thing = better” and “more good thing = better.” Both are straight-line rules, and both shatter on a curve with a middle.
Return to the water-the-plant image and put the two errors side by side:
- “None is always safest.” True on an LNT curve (less is genuinely always better). Fatal on a hormetic curve, where zero dose sits in the deficiency valley. Avoid sunlight entirely and you get vitamin-D deficiency; avoid all mechanical load and your bones demineralize; avoid every stressor and the adaptive machinery never fires. Zero is not the safe end of a hump — it’s one of the two bad ends.
- “More is always better.” True on the rising side only. Fatal past the peak, where the same stressor that built you now breaks you. Train hard with no rest and you get overtraining; take more of the helpful mineral and you hit toxicity; more of a good medicine becomes an overdose.
The optimum is where the marginal effect of one more unit of dose is zero — the top of the hump, where adding stops helping and starts hurting. Practically, you locate it by asking which direction improves things from where you stand, and stepping that way until improvement stalls. That’s it: hill-climbing on a curve you can’t see all of.
Find the peak
Hunt for the optimum
The effect of a stressor is a function of the dose — and that function is usually curved, not straight. Pick a curve, then drag the dose to hunt for the optimum and watch benefit tip into harm.
Dose-response shape
- Response now
- +17
- Optimal dose
- 35
- Net effect
- net benefit
At a dose of 5, the response is 17 — net benefit. The optimum sits at a dose of 35. Past the peak, more is not better; it is worse.
Dosing
A friend on a hormetic curve is currently sitting a bit PAST the optimum (on the falling side). Following the advice 'if a little was good, more must be better,' what happens?
Why averages mislead on a curved response
Here’s the subtlest and most powerful idea in the lesson, and it’s where hormesis shakes hands with the prerequisite model of asymmetry & optionality. When a dose-response curve is curved — bending up (convex) or down (concave) rather than running straight — the response to the average dose is not the same as the average of the responses. The curvature drives a wedge between them. Averaging the dose quietly throws away information about the shape.
The intuition (this is Jensen’s inequality in plain clothes — and per this course’s rules, we keep the math in prose, not in any exercise): on a curve that bends downward like a hump, the straight-line “average” always sits below the curve. So spreading your dosing out to two points and averaging the outcomes lands you lower than the outcome you’d get by sitting steadily at the average dose. On the hump, variability in dose is costly near the peak; steadiness is rewarded. On a curve that bends the other way — convex, where the payoff accelerates — variability is rewarded instead. Same principle, opposite sign; the curvature decides.
Worked example. Take a hormetic response with an optimum at dose 50, scored so that:
| Dose | Response |
|---|---|
| 20 (low) | +2 |
| 50 (medium, the optimum) | +10 |
| 80 (high) | +2 |
Now compare two dosing regimens with the exact same average dose of 50:
- Steady medium: always dose 50 → response +10 every time → average response +10.
- Alternate high/low: dose 20 then dose 80, forever → responses +2 and +2 → average response +2.
Identical average dose. Wildly different outcomes: +10 vs +2. The alternating regimen scored the average of the responses (+2), which sits far below the response to the average dose (+10), because the curve bends downward between the two extremes. Averaging the dose to 50 and expecting +10 from the alternating plan would be a fivefold over-estimate of how well it works.
Flip the curvature and the lesson flips too. On a convex stretch — say the accelerating gain from an intermittent stressor with recovery, where two spaced hard doses plus rest beat the same total delivered as a constant grind — the variable schedule wins and the steady one under-delivers. This is the deep reason the course keeps insisting that “recovery is part of the dose” and that acute-intermittent stress can build while chronic-steady stress of the same total destroys: they land on different points of a curved response, so their averages cannot be compared as if the response were a straight line.
The pitfall: 'on average it's a moderate dose'
The instant a response is curved, the sentence “on average my dose is moderate, so my effect is moderate” becomes a fallacy. A body alternating starvation and feasting is not the same as one eating steadily at the mean, even with identical average intake — because metabolism responds non-linearly. Whenever someone averages a dose to reassure you, ask: is the response curved, and which way does it bend? If it bends, the average of the doses tells you almost nothing about the average of the effects.
Two people take the exact same AVERAGE weekly dose of a hormetic stressor. Alice takes a steady medium dose each day; Bob alternates a tiny dose and a huge dose. On a hump-shaped (downward-bending) response curve, who does better and why?
The two symmetric errors — and telling the shapes apart
Everything so far collapses into a single, beautifully symmetric pair of mistakes. Both come from drawing a straight line through a curved world; they just lean opposite ways.
- Error A — the linear-extrapolation error (“a lot is bad, so a little must be bad”). You observe that a high dose harms, draw a straight line back through the origin (the LNT ghost), and conclude that even a tiny dose harms proportionally. If the true curve is hormetic, you’ve just missed the entire beneficial region — you’ll avoid sunlight, exercise, or an essential nutrient because you extrapolated the toxicity slope down to zero. This is the error the ghost line in the explorer dramatizes.
- Error B — the overshoot error (“a little is good, so more must be better”). You observe that a low dose helps, and conclude that piling on more helps more. If the true curve is hormetic, you sail straight past the optimum onto the falling side — overtraining, over-supplementing, overdosing on a genuinely good thing.
They are mirror images: Error A denies the rising side exists; Error B denies the falling side exists. Both are true only for monotonic curves and both are lethal on a hump. The cure is identical for both: stop reasoning about the substance and start reasoning about its shape. Ask which of the three curves the evidence supports, then ask where on it you stand.
Here’s a quick field guide to telling them apart from data:
| Signature | Linear-no-threshold | Threshold | Hormetic |
|---|---|---|---|
| Effect at very low dose | already harmful, proportional | zero (flat floor) | beneficial |
| Direction of the curve | monotonic (one way) | monotonic (flat, then one way) | non-monotonic (up then down) |
| Is there a best dose > 0? | no (less is always better) | no (below-threshold is fine, more never helps) | yes — an interior optimum |
| Typical home discipline | radiation / carcinogen regulation | classical toxicology; nutrient adequacy | pharmacology, exercise physiology, nutrition (essential agents) |
| Right question | how do I minimize exposure? | am I under the limit? | which side of the peak am I on? |
Now sort some real relationships by shape.
Each item describes a real dose-response relationship. Sort it by its underlying SHAPE: a hormetic curve (helps at a low dose, harms at a high dose, with an optimum in the middle), a monotonic pure-poison / linear-no-threshold curve (harm rises from the first molecule, no beneficial dose), or a threshold curve (no effect until a cutoff, then effect appears).
- A potent genotoxic carcinogen regulated under LNT: no dose assumed safe, risk rises in proportion to exposure
- Cyanide: there is no beneficial dose; harm scales with the amount absorbed from the first exposure upward
- Selenium intake: deficiency causes disease, a moderate intake is essential and protective, high intake is toxic (selenosis)
- A solvent the liver clears completely below a NOAEL, causing measurable harm only once intake exceeds that cutoff
- Botulinum toxin as a poison at high dose: no health benefit from ingesting it, toxicity rises with the amount
- A food additive with no observed adverse effect up to a defined daily limit, above which effects begin to appear
- Sunlight / UV exposure: too little starves vitamin-D synthesis, moderate exposure is beneficial, excess drives burns and skin-cancer risk
- Dietary iron: deficiency causes anemia, an adequate intake is healthy, chronic overload damages the liver and heart
- Mechanical load on bone and muscle: too little wastes them, a moderate load strengthens, crushing or unrelenting load injures
Key takeaways
- A dose-response curve plots effect against dose. Almost every real relationship is one of three shapes — identify the shape before deciding anything.
- Linear-no-threshold (LNT): harm proportional to dose, straight through the origin, no safe amount. The cautious default for radiation and carcinogen regulation — a policy stance as much as a proven biology.
- Threshold: no effect below a cutoff (a NOAEL), then effect rises. The classical toxicology shape, and the mirror shape of nutrient deficiency-to-adequacy. Safe per dose isn’t automatically safe per lifetime.
- Hormetic: biphasic (a beneficial low-dose phase, a harmful high-dose phase) and non-monotonic (rises, then falls). It’s the only shape with an interior optimum — a best dose in the middle, with harm on both sides.
- Reading an optimum: rising side (more helps) → peak (the optimum) → falling side (more harms). “Good or bad” has no answer on a hump; “which side am I on?” does.
- Averages mislead on a curved response (Jensen’s inequality): the response to the average dose ≠ the average of the responses. On a hump, steady beats swinging; on a convex stretch, variability wins. Curvature decides — which is why recovery is part of the dose.
- Two symmetric errors: “a lot is bad, so a little must be” (missing the benefit — the linear-extrapolation error) and “a little is good, so more is better” (overshooting the optimum). Both are straight lines drawn through a curved world.
Shape check
Which pair of words together DEFINES the hormetic shape?
Check your answer to continue.
You can now name the three shapes on sight, read an optimum off a hump, and spot both straight-line errors before they cost you. But we’ve been treating the hormetic hump as a given — why should a low dose of a stressor ever make a living system stronger rather than merely less harmed? The next lesson opens the hood: adaptive overcompensation, the biology of repair systems that overshoot baseline, and the reason recovery isn’t a break from the dose but part of it.