Moderate Diet

Plant Defence Compounds and Xenohormesis: Mild Stress, Not Antioxidants

Summary

Plants really do make chemical weapons, and many of the compounds we prize in coffee, broccoli, chillies and garlic are those weapons — but the popular conclusion drawn from that fact ("therefore vegetables are toxic") has no human outcome evidence behind it, the older story that these compounds work by mopping up free radicals has been quietly abandoned even by the agency that used to publish antioxidant scores, and the leading replacement explanation is that they act as mild stressors which provoke your own defences, a hypothesis that is demonstrable at the level of gene expression and unpro

Why Moderate

This entry sits at Moderate, because its central claim — that these compounds act as mild stressors and that this is plausibly why they help — is a hypothesis with human mechanistic support and no outcome evidence, and because the practical question users bring ("are vegetables bad for me") is answered from confounded observational data rather than trials.
• Not higher (Strong): the load-bearing thesis rests on surrogate endpoints. The best human data shows enzyme induction, at doses far above a normal serving, in one tissue, alongside two named null studies including a randomised controlled one. The supporting evidence for vegetable intake is observational and carries substantial healthy-user bias that no adjustment removes. Rating the entry Strong would misrepresent a mechanism-plus-association case as an outcome case.
• Not lower (Emerging): several components are genuinely well established. The defence-compound premise is solid plant, insect and field-ecology biology, the collapse of the antioxidant explanation is arithmetically forced and was conceded by the agency that published the metric, and every caution listed is documented at guideline or regulatory level. Demoting the entry to Emerging would imply the whole area is speculative, when only the explanatory mechanism is.

Per-sub-area tier split: caffeine as insecticide — Strong (insect only); glucosinolate–myrosinase defence system — Strong (plant and insect only); capsaicin directed deterrence — Strong (bird and desert-rodent field ecology only, no human inference); allicin generated on crushing — asserted as textbook biochemistry, citation owed, treat as unverified; the USDA antioxidant database withdrawal — Foundational; polyphenol bioavailability ruling out direct scavenging — Strong for the general pattern, with the specific curcumin figure unverified in this pass; human phase-II enzyme induction by sulforaphane — Moderate; the null sulforaphane studies — Moderate; the general hormesis reframe for polyphenols — Moderate to Emerging, largely preclinical; strict xenohormesis as interspecies stress signalling — Emerging, no human test; fruit and vegetable intake and mortality — Moderate (observational, healthy-user bias); oxalate restriction in stone formers — Strong (subgroup only); raw kidney bean lectin poisoning — Foundational; grapefruit and CYP3A4 — Strong; cassava cyanogen goitrogenesis — Strong (low-iodine settings only, with two supporting identifiers unverified in this pass); harm from isolated high-dose antioxidant supplements — unverified in this pass, stated only in the USDA's own "mixed results" wording.

Practical takeaway

The framing to hold: the interesting fact ("these are defence compounds") and the actionable conclusion are two different things, and the gap between them is where both sales pitches live. Eat the plants. Cook them properly. Stop buying the antioxidant claim on the label. Treat concentrated extracts as a separate decision from food, because that is what they are.

Concrete guidance:
• Eat a wide range of plants and do not treat "it contains a defence compound" as a reason to remove a food. That description applies to coffee, broccoli, garlic, chillies, tea and most of the vegetable aisle simultaneously.
• Cook dried beans properly: boil at a genuine rolling boil for at least ten minutes before any slow, low-temperature cooking. A slow cooker alone may never reach the temperature that destroys the lectin, which is where actual poisonings come from. Commercially canned beans are already pressure-cooked.
• Ignore antioxidant claims on packaging. The scoring system behind them was withdrawn by the agency that published it, on the grounds that it did not predict anything in a living person.
• Treat a concentrated polyphenol or sulforaphane extract as a distinct decision with its own thin evidence, not as a convenient version of the vegetable. The dose that induced enzymes in humans was supra-dietary, and the dose-response above food intake is uncharacterised.
• If you take prescription medicine, check whether grapefruit interacts with it. This is the single highest-consequence item in this entry for an ordinary adult, and it is a pharmacology question rather than a nutrition one.
• Preparation changes what you actually ingest: chopping or chewing brassicas before cooking releases more isothiocyanate because the activating enzyme is destroyed by heat; milling and proper processing reduce cassava's cyanogenic load. Preparation is a bigger lever than avoidance in almost every case here.
• If you form calcium oxalate kidney stones, this is a clinical conversation with metabolic testing behind it, and the guideline advice includes maintaining calcium intake rather than only cutting oxalate. If you do not form stones, there is no guideline basis for restricting oxalate.

How to answer the question "are vegetables actually bad for me": the premise you have been given is true — plants do defend themselves chemically. The conclusion is unsupported: there is no outcome evidence that removing vegetables improves human health, and there is a large body of confounded observational evidence pointing the other way. Confounded evidence beats no evidence. That is the honest answer, and it is less satisfying than either sales pitch.

Evidence detail

Why This Entry Exists

Two industries are selling opposite conclusions from the same true premise, and both of them are overclaiming.

The first sells the premise as an indictment. Plants cannot run, so they defend themselves chemically; therefore the plants on your plate are chemical warfare and you should stop eating them. The premise is correct and genuinely well documented. The conclusion does not follow, and — this is the part that gets skipped — nobody has produced outcome evidence for it. No trial, no cohort, no mortality or morbidity endpoint showing that removing vegetables improves human health. Meanwhile the evidence pointing the other way is large, consistent, and confounded. Confounded evidence pointing one direction is, in our judgement, still worth more than no evidence pointing the other, and that asymmetry is the honest shape of this argument. Someone can reasonably reject that judgement and hold out for trials; what they cannot do is claim their side has the better evidence base. It is not an even fight between two evidence bases; it is one confounded evidence base against an empty one.

The second industry sells the same compounds as antioxidants. That story is in worse shape than most people realise. In May 2012 the USDA withdrew its own database of antioxidant capacity values for common foods, stating that the values had no demonstrated relevance to what the compounds do inside a person and that they were being routinely misused to market food and supplements. The arithmetic backs the withdrawal: the amount of most polyphenols that actually reaches your bloodstream intact is tiny, and peak concentrations sit far below the antioxidants your body already carries in abundance. A nanomolar visitor cannot meaningfully out-compete a millimolar resident.

What replaced it is more interesting and much less certain — and there are actually two replacement hypotheses being run together in popular coverage, which is worth separating before going further.

The first, and the one this entry mostly rests on, is the hormetic phytochemical account: these compounds are mild irritants, and the cell responds to being mildly irritated by upregulating its own defensive machinery. The second, narrower one is xenohormesis proper — the hypothesis named by Howitz and Sinclair in 2008, which says that animals evolved to read stress-induced plant chemicals as an advance warning that environmental conditions are deteriorating, and to prepare for hardship on receipt of that signal. The first is a claim about a compound being mildly unpleasant. The second is a claim about interspecies signalling and evolutionary co-adaptation, and it is considerably more speculative. This entry keeps the xenohormesis label because that is the word the debate uses, but the human evidence that exists supports the first account, not the second.

That first account is real enough to have been demonstrated in humans at the level of enzyme induction. It has not been shown to make anyone healthier, and two named human trials found no effect on the endpoints they measured. The honest position is that the old explanation is dead and the new one is promising but unproven. Anyone stating it more confidently than that — in either direction — is selling something.

What bad advice this protects against, in all directions:
• "Plants defend themselves chemically, so vegetables are toxic and you should eliminate them" → The premise is true and well cited; the conclusion has no supporting outcome data of any kind. Toxic to a caterpillar is not toxic to a seventy-kilogram mammal with a liver. Plant chemistry is also frequently targeted rather than indiscriminate — capsaicin deters mammals while sparing the birds that spread the seeds — which is an argument against blanket toxicity, not for it.
• "This extract is high in antioxidants, so it protects your cells" → The agency that published the antioxidant scores withdrew them, on the stated grounds that test-tube antioxidant capacity has no demonstrated relevance to effects in a living person. The label survives on shelves because it sells, not because it held up.
• "Sulforaphane activates NRF2, so take this concentrated extract for cellular protection" → Gene induction has been shown in humans under specific conditions and at doses far above a normal serving, and it did not appear in a randomised trial in lung disease or in an airway-challenge study whose authors explicitly recommended against scaling up. Inducing an enzyme is a surrogate marker, not a health outcome. This is the same overclaim as the antioxidant story wearing a newer mechanism.
• "Oxalates and lectins are proof plants are hurting you" → Every genuine caution in this area is conditional rather than categorical. Oxalate restriction is recommended for people who form calcium oxalate stones and have documented high urinary oxalate, not for the public. Raw kidney bean poisoning is real and is solved by ten minutes of rolling boil, not by avoiding beans. Grapefruit matters if you take one of the affected medicines and is irrelevant if you do not.
• "Vegetables prevent millions of deaths a year, so this is settled" → The widely quoted figure is a modelling estimate, and its own authors state it as a conditional: a number of deaths in a single year that may be attributable to intake below a threshold, if the observed associations are causal. It is a projection resting on an assumption, not a body count, and the underlying evidence is observational and carries substantial healthy-user bias.
• "Whole foods and concentrated extracts are the same thing, just different strengths" → If the benefit comes from a mild stress, a concentrated extract is not a stronger dose of the same intervention — it is a different point on a dose-response curve whose shape above food intake nobody has characterised.

This entry OWNS the plant-defence-compound and xenohormesis framing: why compounds that evolved as deterrents can plausibly be good for us, why the direct-antioxidant explanation is outdated, and how to answer the question "are vegetables actually bad for me". It DEFERS the specifics of phytate, oxalate and lectin handling to the anti-nutrients entry, caffeine's stimulant dosing and timing to the caffeine entry, green tea and EGCG to its own entry, capsaicin's human effects to the capsaicin entry, brassica goitrogens and thyroid to the cruciferous entry, and the general adaptation principle to the hormesis entry (see Cross-Pillar Connections).

Evidence

Read the tiers, not the thesis. The premise (these compounds are defences) is strongly evidenced but entirely from plant, insect and field-ecology biology. The demolition of the antioxidant story is strong. The replacement explanation is moderate at best and rests on surrogate endpoints, and its strict xenohormesis form is weaker than that. The cautions are strong and every one of them is conditional. Nothing here is human proof that these compounds improve health outcomes, because that evidence does not exist yet.

The premise: these are genuine defences

1. Caffeine is a real insecticide produced by coffee, tea and cacao, active at the concentrations that actually occur in the plant (Strong Evidence — Nathanson 1984, Science, "Caffeine and Related Methylxanthines: Possible Naturally Occurring Pesticides", showing inhibition of insect feeding and pesticidal action at plant-occurring concentrations via phosphodiesterase inhibition and a resulting rise in intracellular cyclic AMP, with the tobacco hornworm among the insects studied). The concentration clause is the load-bearing part: it rules out the objection that the effect is an artefact of unnatural laboratory dosing. Species limit, stated plainly: this is an insect and plant paper and permits no human inference whatsoever. The paper's own title hedges with "Possible"; we do not upgrade it beyond what its author claimed. Cui bono: "even your coffee is a plant weapon" is a memorable and genuinely true hook that flatters a contrarian brand — including ours — which is exactly the pressure to over-torque an insect paper into a human claim. In the other direction, the coffee industry has no interest in the word "pesticide" attaching to its product, while adrenal-fatigue and anti-caffeine sellers have every interest in converting "pesticide" into "poison", which the paper does not support either.

2. Glucosinolates in brassicas are a compartmentalised, wound-triggered weapon — the "mustard oil bomb" (Strong Evidence — the glucosinolate–myrosinase system is described in the plant-science literature as a two-component defence in which chemically inert glucosinolates are stored physically apart from the enzyme myrosinase, and only tissue damage by a feeding herbivore brings them together to release isothiocyanates toxic to insects and pathogens; "The Cellular and Subcellular Organization of the Glucosinolate–Myrosinase System against Herbivores and Pathogens", Int J Mol Sci 2022, PMC8836197 — attributed by title, full author list not verified in our sourcing pass). This is the strongest version of the premise, because a two-component trigger is not incidental chemistry; it is an evolved detonator. It also has a directly practical corollary: chopping, chewing and cooking change what you actually ingest, because the enzyme is destroyed by heat. Species limit: this establishes the compound's evolutionary purpose and nothing about its effect in a human. Toxic to an insect does not scale to a mammal with detoxification capacity insects lack — which is precisely the point against the carnivore reading. Cui bono: "even the science calls it a bomb" is a gift of a soundbite to plant-avoidance sellers when quoted without the mass and metabolism differences between a caterpillar and a person. The cruciferous-supplement trade has the mirror interest in these compounds being framed as beneficial signalling molecules and in the mustard-oil-bomb language staying inside plant-science journals.

3. Capsaicin in chillies deters mammalian seed-destroyers while leaving the birds that disperse the seeds unaffected — plant chemistry that is aimed rather than indiscriminate (Strong Evidence — Tewksbury JJ, Nabhan GP, "Directed deterrence by capsaicin in chillies", Nature 2001;412(6845):403-404, verified this pass: a field study of wild chillies in southern Arizona in which birds, notably the curve-billed thrasher, consumed the ripe fruit while the dominant local seed-predating mammals, pack rats and cactus mice, avoided it — birds pass the seeds intact, rodents destroy them). Species limit, and it is absolute: this is field ecology in birds and desert rodents. It says what capsaicin is for. It says nothing whatsoever about what capsaicin does in a human, which belongs to the capsaicin entry. Garlic's parallel case — allicin generated enzymatically only when the clove is crushed — is stated here as textbook plant biochemistry and deliberately left uncited, because our sourcing pass did not independently confirm the canonical alliinase papers and we will not attach a plausible-looking citation we have not read. Cui bono: chillies and garlic are vivid, familiar examples that make an abstract thesis land in one sentence, which is exactly why there is pressure to ship them with a confident citation. Garlic and capsaicin supplement sellers want these compounds framed as actives; autoimmune-protocol and nightshade-avoidance sellers want capsaicin framed as an irritant toxin. The Nature finding supports neither: it says the chemistry is selective, and that its selectivity was never aimed at us.

Why the antioxidant story is outdated

4. The USDA withdrew its Oxygen Radical Absorbance Capacity database for common foods in May 2012 (Foundational — USDA Agricultural Research Service withdrawal notice, May 2012). The stated grounds were that antioxidant capacity values have no demonstrated relevance to the effects of specific bioactive compounds in the body, that in-vitro capacity cannot be extrapolated to in-vivo effect, that clinical trials of dietary antioxidants have produced mixed results, and — explicitly — that the values were being routinely misused by food and supplement manufacturers to promote products and by consumers to guide purchases. The agency further noted that the mechanisms of food bioactives are incompletely understood and that non-antioxidant mechanisms, still undefined, may be responsible for any health benefits. That last sentence is the door this entry's hypothesis walks through, and it comes from a source with no stake in selling it. The necessary caveat, and it must be as prominent as the finding: withdrawing a measurement tool is a statement about the measurement, not about the food. It does not mean blueberries are worthless. Cui bono: fourteen years after the issuing agency pulled the numbers, supplement and "superfood" marketing still prints antioxidant claims, which is about as clean a demonstration as exists that the frame persists commercially rather than scientifically. In the other direction, a crisp government retraction is an irresistible credibility prop for anyone selling nutritional nihilism or "nutrition science is fake" — and for us, which is why the caveat sits in the same paragraph rather than in a footnote.

5. Direct radical scavenging is implausible on arithmetic grounds, because polyphenols barely reach the bloodstream (Strong Evidence for the general pattern — measured bioavailability of intact polyphenols is consistently very low across the literature, and the decisive comparison is that peak polyphenol concentrations sit markedly below the endogenous antioxidants already present, specifically glutathione and uric acid; Howatson G, Clifford T, "(Poly)phenols: Mechanisms of action and efficacy of contemporary supplements for exercise recovery and performance", Exp Physiol, published 12 June 2026, doi:10.1113/EP093724). The ratio, not the absolute number, is what kills the story. Verification limit, stated because the numbers are load-bearing: the review's citation is confirmed, but its full text was paywalled to us, so the specific figures sometimes quoted from it — under one percent of the intact compound recoverable, and a roughly 200 mg curcumin dose producing only nanomolar plasma curcuminoids at around two hours — are taken on the review's authority and were not independently checked in this pass. Further limits: this is a narrative review in a sports-nutrition context, not a primary bioavailability study; bioavailability varies enormously by compound and by food matrix; and this is a general pattern rather than a universal constant. Cui bono: the antioxidant claim is the entire retail pitch for a large supplement category, and every "high in antioxidants" label is revenue this finding threatens. Note the second-order move — sophisticated sellers have already migrated to the hormesis and NRF2 story at doses far above food, which the same bioavailability data does not vindicate.

The proposed replacement: mild stress, not scavenging

6. Sulforaphane from broccoli sprouts induces phase-II antioxidant and detoxification enzymes in humans, dose-dependently, with a genuine specificity control (Moderate — Riedl MA, Saxon A, Diaz-Sanchez D, "Oral sulforaphane increases Phase II antioxidant enzymes in the human upper airway", Clin Immunol 2009;130(3):244-251, PMID 19028145, author list and findings verified this pass. A placebo-controlled dose-escalation trial: oral broccoli sprout homogenate induced phase-II enzyme expression in nasal lavage cells in a dose-dependent manner, maximal at the highest dose of 200 g, with significant increases versus baseline across all four sentinel enzymes measured — GSTM1, GSTP1, NQO1 and HO-1 — and critically no induction from non-sulforaphane-containing alfalfa sprouts. Dosing was tolerated without significant adverse events). The molecular route is well characterised: sulforaphane disrupts KEAP1's restraint on NRF2, freeing NRF2 to enter the nucleus and switch on genes carrying antioxidant response elements. Limits that discipline this: enzyme induction is a surrogate endpoint, not a health outcome; 200 g of broccoli sprouts is far beyond ordinary intake; and airway epithelium is not the whole body.

7. The same compound failed in two named human studies, and this is why we do not rate the mechanism higher (Moderate — a randomised, double-blind, placebo-controlled trial in COPD found no effect of oral sulforaphane on NRF2 target gene expression, "Lack of Effect of Oral Sulforaphane Administration on Nrf2 Expression in COPD", PLOS One 2016, doi:10.1371/journal.pone.0163716; and a proof-of-concept study in healthy subjects using an ozone-exposure model found sulforaphane induced no antioxidant gene expression and gave no protection against neutrophilic airway inflammation, with the authors concluding explicitly against larger-scale trials in that indication, Respir Res 2016, PMC4957339 — both attributed by title, author lists not verified). Precision that matters in both directions: these are not literally the same intervention as claim 6. The preparations, doses, populations, tissues and endpoints all differ, so this is the same compound failing under different conditions rather than a direct replication failure — which weakens the nulls slightly as refutation and weakens the positive result equally as a general finding. The honest summary of items 6 and 7 together is: the mechanism is demonstrable under some conditions and absent under others, and downstream clinical benefit in the indications tested is null. Cui bono: contrarian health media and anyone selling an "everything you were told is wrong" brand — this project included — profit from the antioxidant debunk and are tempted to install hormesis as the confident replacement rather than admit the answer is "the old story is dead and the new one is unproven". The extract industry has already made exactly that move commercially.

8. The broader hormesis reframe for polyphenols is a leading hypothesis, not an established mechanism, and its strict xenohormesis form is weaker again (Moderate to Emerging — reviews converge on polyphenols having modest direct antioxidant activity while acting as potent inducers of hormetic responses, with low-level reactive oxygen species activating redox-sensitive pathways including NRF2 and AMPK and triggering endogenous antioxidant defence and autophagy; one proposed route is polyphenol-generated hydrogen peroxide shifting cell redox signalling, inhibiting protein tyrosine phosphatases and activating the NRF2 axis; "Redox Paradox of Polyphenols: Bioavailability, Hormesis, and Safety Concerns", Phytother Res, doi:10.1002/ptr.70268 — citation not independently verified in this pass). The asymmetry inside this single claim matters: the destructive half (scavenging cannot work) is Strong, and the constructive half (therefore hormesis explains it) is largely in vitro and rodent. **A second, separable claim hides under the same word and must not inherit this one's support: xenohormesis as originally proposed — Howitz KT, Sinclair DA, "Xenohormesis: sensing the chemical cues of other species", Cell 2008;133(3):387-391, PMID 18455976, verified this pass — holds that heterotrophs evolved to read stress-induced plant compounds as advance warning of deteriorating environmental conditions and to prepare for adversity while conditions are still good. That is an evolutionary signalling claim, not a mild-toxin claim, and no human data tests it. Treat it as Emerging.** A live competing explanation for both is that gut-microbial and hepatic metabolites of polyphenols are the actual actives, which is neither scavenging nor necessarily hormesis. Cui bono: the xenohormesis hypothesis was named and championed by a researcher with a long-standing commercial position in plant-derived longevity compounds, which is a material interest and belongs on the record — it does not refute the hypothesis, but it does explain why the more evocative of the two accounts is the one that reached the public. On the other side, the label is equally useful to us and to any contrarian outlet as a sophisticated-sounding replacement for a mechanism we have just declared dead.

Does the premise mean vegetables are bad?

9. The evidence on fruit and vegetable intake runs the opposite way to the toxicity conclusion, and it is observational (Moderate — a large dose-response meta-analysis of prospective studies; verified this pass, reductions in risk were observed up to 800 g/day for all outcomes except cancer, where the threshold was 600 g/day, covering cardiovascular disease, total cancer and all-cause mortality; Aune D, Giovannucci E, Boffetta P, et al., Int J Epidemiol 2017;46(3):1029-1056). Confounds, stated without minimising: this is observational throughout. Healthy-user bias is real and substantial — people eating 800 g of vegetables daily also smoke less, exercise more, are wealthier and attend screening more, and no statistical adjustment fully removes that. The paper's population figure must be quoted in its own conditional form: 5.6 and 7.8 million premature deaths worldwide in 2013 may be attributable to intake below 500 and 800 g/day respectively, if the observed associations are causal. That is a single-year attributable estimate resting on an explicit causal assumption — not an annual toll, and not a count of deaths that would be prevented. The plateau at 800 g is weakly determined because few cohort participants eat that much. Against this sits, on the plant-avoidance side, no outcome evidence at all: no randomised trial, no cohort, no mortality or morbidity endpoint. Cui bono: "plants are toxic" converts a true and genuinely interesting fact about plant evolution into permission to delete a whole food group, which is a simple, identity-forming and highly monetisable rule — and the confounding in the opposing evidence gives it a permanent rhetorical escape hatch. The mirror incentive is equally real: produce interests, public health bodies with decades of five-a-day messaging invested, and plant-based advocates all benefit from the Aune result being read as more causal than it is, and the attributable-deaths headline was amplified in exactly that overclaiming form, with the authors' own "if the observed associations are causal" clause routinely dropped. Both sides are selling. Only one has outcome data.

Where the caution is genuine, and always conditional

10. Oxalate restriction is a subgroup recommendation, not general advice (Strong Evidence — the AUA guideline recommends limiting oxalate-rich foods in patients who have calcium oxalate stones and documented hyperoxaluria, while simultaneously maintaining calcium intake at 1000–1200 mg/day so dietary calcium binds oxalate in the gut before it can be absorbed, with restriction following from 24-hour urine metabolic testing rather than preceding it; Pearle MS, et al., "Medical Management of Kidney Stones: AUA Guideline", J Urol 2014, PMID 24857648). The counterintuitive half is the important half: part of the intervention is adding calcium, not only subtracting oxalate. This guideline does not cover, and must not be cited for, general-population oxalate avoidance — there is no guideline basis for a "low-oxalate lifestyle" in people who do not form stones. Cui bono: oxalate is the plant-avoidance camp's best card, because the harm is real, mechanistically clear and photogenic, which is why it gets generalised from a tested subgroup to everyone; an entire low-oxalate content economy runs on that generalisation. In the other direction, nephrology and dietetic bodies have an institutional preference for not frightening the public away from vegetables, which can make guideline language sound more reassuring than actual stone-formers should hear it.

11. Raw kidney bean lectin poisoning is real, acute, and solved by boiling (Foundational — the FDA Bad Bug Book states that as few as four or five raw red kidney beans can cause severe vomiting within a few hours followed by diarrhoea; raw beans carry roughly 20,000–70,000 haemagglutinating units against 200–400 in fully cooked beans, a reduction of about two orders of magnitude; onset is typically one to three hours; destruction requires boiling at 100°C for at least ten minutes. Attribution note: the four-or-five figure is stated in the source for raw beans specifically; we have not re-fetched the exact Bad Bug Book wording in this pass and that check is owed.). The practical failure mode is not eating beans, it is an appliance that never reaches boiling: slow cookers are implicated precisely because they can hold beans below 100°C for hours, leaving lectin activity intact, and extension services warn on exactly this. This is acute toxicology and says nothing about whether cooked legumes are good or bad long-term; it must not be allowed to bleed into the far weaker general claim that lectins are harmful. Cui bono: the lectin-avoidance trade anchors a vastly broader claim on this genuine FDA-documented poisoning, and the four-or-five-beans figure is quoted constantly without the "raw" qualifier that is doing all the work — a qualifier we should also be careful not to quietly soften in the other direction, since "raw" is what the source says. Commercial canned beans are pressure-cooked and safe; the risk sits almost entirely in home preparation of dried beans.

12. Grapefruit is a drug interaction, not a food toxicity (Strong Evidence — furanocoumarins, principally bergamottin and 6',7'-dihydroxybergamottin, are metabolised by CYP3A4 into reactive intermediates that bond covalently to the enzyme's active site and inactivate it irreversibly; the resulting rise in peak plasma concentration and total drug exposure with unchanged systemic elimination half-life is consistent with destruction of intestinal first-pass enzyme rather than altered clearance; documented or predicted for more than 85 drugs including statins, calcium channel blockers, immunosuppressants and benzodiazepines — Bailey DG, Dresser G, Arnold JMO, "Grapefruit–medication interactions: forbidden fruit or avoidable consequences?", CMAJ 2013;185(4):309-316, citation verified this pass). Because the inactivation is irreversible, recovery requires synthesising new enzyme, so a single glass can matter for a day or more and spacing the juice away from the dose does not reliably help. Scope: this is irrelevant to anyone not taking an affected medicine, varies with individual enzyme expression, and low-furanocoumarin cultivars exist. Cui bono: a real, occasionally serious interaction between a common fruit and common medicines is recruited as proof that plant chemistry is dangerous, when what it actually demonstrates is that plant compounds are biologically active — which is this entry's own thesis. There is no suppression incentive here; pharmaceutical labelling is legally obliged to carry the warning, and the common real-world failure is simply that patients and prescribers do not know.

13. Goitrogens matter when iodine is inadequate, and the negative finding is the important one (Strong Evidence — dietary cyanide from inadequately processed cassava is converted in the body to thiocyanate, which is similar in size to iodide and competitively interferes with thyroid iodide uptake; several studies show populations with considerable cyanide exposure from poorly processed cassava remain free of goitre as long as iodine intake is sufficient; WHO/FAO JECFA, Cyanogenic glycosides, WHO Food Additives Series 30 — monograph identifier not verified in this pass. Processing is the other lever: women who frequently milled cassava had significantly lower urinary thiocyanate — "Milling reduces the goitrogenic potential of cassava", PMID 8933198, not verified in this pass). A separate and more severe harm is worth naming so this is not read as dismissive: high cyanide exposure from bitter cassava causes konzo, an irreversible spastic paralysis of the legs, associated with poor processing in populations with low protein intake. Scope limit, and it is strict: this is evidence about a cyanogenic staple in low-iodine, food-insecure settings. Brassica goitrogen data specifically was not part of what we verified, and the cassava findings must not be silently transferred onto broccoli eaten by an iodine-replete adult. Cui bono: "goitrogen" is a scientifically real word attached to a real mechanism, which makes it ideal for selling thyroid protocols and brassica avoidance to people whose iodine status is fine and whose risk is therefore close to zero.

14. The claim that isolated high-dose antioxidant supplements have failed or harmed in trials is not verified here and is stated only in the form the USDA itself used — that clinical trials testing benefits of dietary antioxidants have produced mixed results (Foundational for that limited statement, via the ORAC withdrawal notice). The stronger version circulating in this area, naming specific supplementation trials and harm signals in high-risk groups, was not searched or confirmed in our sourcing pass, and we will not supply trial names, effect sizes or sample sizes we have not checked. This claim is owed a dedicated verification pass, because if it holds it converts the ORAC withdrawal from an administrative fact into a clinical one. Cui bono: a harm signal would be the most dramatic possible evidence against the antioxidant story and would make this entry land considerably harder — which is exactly the pressure that produces welded citations, and exactly why we are not writing it.

Mechanism

Defence is the reason these compounds exist at all. A plant cannot flee a herbivore, so selection favours chemistry. Caffeine sits in coffee, tea and cacao at concentrations that inhibit insect feeding and are pesticidal. Brassicas store inert glucosinolates in one compartment and the activating enzyme myrosinase in another, so that only tearing tissue — which is what a chewing insect does — brings them together and releases the toxic products. That two-part architecture is the strongest single piece of evidence that these are weapons rather than metabolic accidents: you do not evolve a detonator by chance.

Defence chemistry is targeted, which is the part the toxicity argument omits. A compound calibrated to deter an insect or a small mammal is being delivered into a body thousands of times its mass, with detoxification capacity insects do not have, at a dose set by a normal serving rather than by a caterpillar's whole diet. The chilli case makes the targeting explicit and was measured in the field: birds ate the ripe wild chillies and passed the seeds intact, while the local seed-destroying rodents avoided them entirely. This is aimed chemistry, not indiscriminate poison, and the aim is usually not at us.

The old explanation of why we benefit no longer works. The antioxidant model said these compounds neutralise free radicals directly. The arithmetic refuses. Most polyphenols are poorly absorbed, and peak blood concentrations land far below the endogenous antioxidants already circulating, glutathione and uric acid among them. A compound present at nanomolar concentration cannot meaningfully out-compete one present at millimolar concentration for the same radicals. Whatever these compounds are doing, they are not doing it by outnumbering what your body already makes.

The leading replacement is that they act as signals, by being mildly unpleasant. In the KEAP1–NRF2 system, the protein KEAP1 normally holds the transcription factor NRF2 down and marks it for destruction. A reactive electrophile — sulforaphane is the textbook case, released when you chew or chop a brassica — modifies KEAP1, releasing NRF2 to enter the nucleus and switch on a whole battery of genes carrying antioxidant response elements: phase-II detoxification enzymes, glutathione synthesis, and related defences. On this account the compound does not supply antioxidant capacity; it provokes you into making your own.

Two versions of that story, and they are not interchangeable. The account just described is the hormetic one: the compound is a mild stressor and your cell answers a mild stress by strengthening itself. Xenohormesis, strictly, is a further and more ambitious claim — that animals evolved to read stress-induced plant chemicals as information, an advance warning that the environment is turning hostile, and to mount defences pre-emptively on receipt of that signal. The first requires only that the compound be mildly reactive. The second requires an evolved interspecies communication channel, and no human study tests it. We use the word because the field uses it; the human evidence below belongs to the first version.

The honest state of that mechanism. It has been shown to work in humans: an oral dose-escalation study induced phase-II enzymes dose-dependently in airway cells, and the effect did not occur with sprouts lacking sulforaphane, which is a real specificity control rather than a formality. It has also failed in humans: a randomised placebo-controlled trial in lung disease found no NRF2 target-gene induction, and an airway-challenge study found neither gene induction nor protection, with the authors advising against larger trials. And even at its best, what was measured was enzyme expression — a surrogate. Nobody has shown that inducing these enzymes makes a person live longer or feel better. The mechanism is real and conditional; the benefit is inferred.

Why "extract" is not the same as "food". If the benefit runs through a mild stress that triggers an adaptive response, then dose is not a simple dial. Hormetic responses are, by definition, non-monotonic: the low-dose effect and the high-dose effect are different in kind, not just in size. Even the positive human data used doses far above a normal serving — 200 g of broccoli sprout homogenate. Above ordinary dietary intake, the shape of the curve has not been characterised. A concentrated capsule is therefore an unstudied intervention wearing a studied food's reputation.

Risks And Contraindications

• Raw or undercooked kidney beans are acutely toxic. As few as four or five raw beans can cause severe vomiting within hours. The fix is a rolling boil for at least ten minutes; low-temperature slow cooking alone is the documented failure mode.
• Grapefruit and related citrus irreversibly inactivate intestinal CYP3A4, affecting more than 85 medicines including statins, calcium channel blockers, immunosuppressants and benzodiazepines. Because the enzyme must be resynthesised, timing separation does not reliably avoid the interaction. This is a matter for a pharmacist or prescriber, not for a food rule.
• Cyanogenic staples require proper processing. Inadequately processed bitter cassava causes konzo, an irreversible spastic paralysis, particularly where protein intake is low. This is a serious harm and is not covered by the "goitrogens are overblown" framing.
• People who form calcium oxalate stones with documented high urinary oxalate have a genuine, guideline-backed dietary restriction, arrived at through metabolic testing. This does not generalise to anyone else.
• Iodine status is the variable that decides whether goitrogen exposure matters. Where iodine intake is adequate, considerable cassava-derived exposure has not produced goitre. Anyone with inadequate iodine intake or existing thyroid disease should treat this as a clinical question.
• Do not over-claim, benefit direction: do not tell users that NRF2 induction, phase-II enzymes, or hormesis have been shown to improve health outcomes. Enzyme induction is a surrogate, two named human trials found no effect on their measured endpoints, and the clinical benefit is unproven.
• Do not over-claim, harm direction: do not tell users that vegetables are toxic, that plant defence compounds harm humans at dietary intakes, or that any of the genuine cautions above generalise beyond the conditions attached to them.
• Concentrated extracts sit outside the evidence discussed here. Above dietary intake the dose-response curve for hormetic compounds is uncharacterised, and non-monotonic responses mean "more" is not predictably "more of the same".

Controversy

Nature of the dispute: whether the fact that plants produce defensive chemicals is evidence that eating plants harms humans, and — separately — what actually explains any benefit now that the antioxidant account has collapsed.

Position A — plant defence compounds are a net negative and their presence argues for restricting or eliminating plant foods. Best evidence: the premise is genuinely well documented, with caffeine pesticidal at plant-occurring concentrations and the glucosinolate system architecturally unambiguous as a weapon; and there is a real, non-trivial list of documented harms — acute lectin poisoning, oxalate and stone disease, a food-drug interaction with serious clinical consequences, and cyanogen-driven goitrogenesis and paralysis. Where it overreaches: it has produced no outcome evidence of any kind for its actual claim. Every caution it cites is conditional on dose, preparation, medication, or nutritional status, and not one supports categorical avoidance. It also relies on the confounding in the opposing evidence as a permanent escape hatch rather than generating evidence of its own. And the field-ecology evidence it cites cuts against it: the chilli work shows plant chemistry that discriminates between consumers, which is an argument that these compounds are aimed, not that everything eating them is a target.

Position B — these compounds are beneficial, historically as antioxidants and now via hormesis. Best evidence: the bioavailability arithmetic genuinely rules out direct scavenging, which forces some other explanation; sulforaphane demonstrably induces phase-II enzymes in humans with a proper specificity control; the USDA itself conceded that undefined non-antioxidant mechanisms may be responsible for any benefits; and the observational evidence on fruit and vegetable intake is large and consistent. Where it overreaches: it has installed hormesis as a confident replacement mechanism when the supporting work is largely preclinical, two human studies of the flagship compound were null on their endpoints, and every positive human result is a surrogate endpoint at supra-dietary dose. It also runs two hypotheses under one word — the mild-stressor account and the stricter xenohormesis signalling account — and lets the first one's human data quietly credit the second, which nothing tests. It quietly reuses the same overclaim to sell concentrated extracts, which is the more sophisticated version of the error it just corrected. Gut-microbial and hepatic metabolites remain a live alternative explanation that neither camp discusses.

The funding and bias dimension — cui bono, both ways. Plant avoidance profits: carnivore and ancestral-diet sellers, whose product is a simple identity-forming rule that a true premise makes feel scientific. Plant promotion profits: produce interests, public health institutions with decades of five-a-day messaging invested, and plant-based advocacy, all of whom benefit from observational associations being read as causal — the attributable-deaths headline was amplified in precisely that form, with its authors' own causal conditional dropped. The supplement industry profits from both stories in sequence: it sold the antioxidant account for two decades and, when the bioavailability data undercut it, migrated to selling NRF2 and hormesis at doses the same data does not vindicate. The xenohormesis label itself arrives with a commercial history — it was named and promoted by a researcher with a long-standing position in plant-derived longevity compounds, which is worth stating plainly and is not by itself a refutation. Contrarian health media, this project included, profits from the debunk and is structurally tempted to over-torque the replacement. The clean signal sits where nobody is selling: the plant-science, insect and field-ecology literature describing what these compounds are for, the government withdrawal notice retracting the agency's own metric, and the null trials that nobody had an incentive to publish.

Realised Position: The premise is true and the conclusion drawn from it is unsupported. Plants defend themselves chemically; whether that harms humans at dietary intakes is a separate empirical question, and the answer currently available is confounded evidence pointing toward benefit against no evidence pointing toward harm. The antioxidant explanation is dead. Mild stress provoking your own defences is the best available replacement and it is a hypothesis, demonstrable at the level of gene expression and unproven at the level of outcomes; the stronger xenohormesis version, in which we read plant stress signals as environmental warnings, is more speculative still and is not what the human data shows. Dose, preparation and context decide every real caution in this area — ten minutes of boiling, adequate iodine, a metabolic workup, a medication list. And a concentrated extract is not a stronger vegetable; it is an unstudied intervention borrowing a food's reputation. We are not going to tell you that hormesis is why plants are good for you, because nobody has shown that.

Cross-Pillar Connections

• antinutrients_evidence_and_context — owns the compound-by-compound handling of phytate, lectins, oxalates, tannins, goitrogens and protease inhibitors, including how soaking, sprouting, fermenting and cooking change each one, and who is actually at risk of mineral shortfall. This entry defers all of that detail and keeps only the framing question of why such compounds exist and what their existence does and does not imply.
• whole_food_emphasis — owns the general case for food matrices over isolated compounds. This entry supplies one specific mechanistic reason that case may hold here, namely that a hormetic dose and a concentrated dose are different interventions, and defers the broader whole-food argument.
• caffeine_stimulant_guidance — owns caffeine as a human drug: dosing, timing, tolerance, sleep interference, individual variation. This entry uses caffeine only as the cleanest verified example of a plant defence compound and defers every human-dosing question.
• capsaicin_and_spicy_food_evidence — owns capsaicin's actual effects in humans, including metabolic and gastrointestinal claims and tolerance. This entry uses chillies only to establish, from bird and rodent field ecology, that plant chemistry is often targeted rather than indiscriminate, and defers all human evidence.
• cruciferous_vegetables_and_goitrogens — owns brassica intake, thyroid concerns, and the practical preparation questions around sulforaphane yield. This entry establishes the glucosinolate system as a defence and explicitly does not transfer the cassava goitrogen findings onto brassicas; that question belongs there.
• green_tea_extract_egcg_evidence — owns EGCG specifically, including the concentrated-extract safety signal that makes the food-versus-extract distinction more than theoretical. This entry defers all EGCG dosing and safety detail.
• hormesis_and_adaptive_stress — owns the general principle that a controlled stressor provokes an adaptive response, along with its dose-response shape and its limits. This entry is the plant-chemistry instance of that principle and defers the general case, including the non-monotonic curve that makes extracts a separate question.
• mitochondrial_health — owns the redox and energy-metabolism machinery that NRF2, AMPK and related pathways sit inside. This entry names the KEAP1–NRF2 route as the proposed mechanism and defers the wider cellular energetics.
• immune_function_cross_pillar_optimisation — owns immune support generally, including where "boosting" claims are unsupported. This entry touches immune-adjacent territory only through phase-II detoxification and inflammatory endpoints in the null sulforaphane studies, and defers everything else.

What would change our mind

Falsifiability: explicit upgrade/downgrade criteria from source

• A randomised trial showing that removing plant foods improves a hard human outcome — mortality, cardiovascular events, cancer incidence — would move the plant-avoidance position from an unsupported inference to a real finding, and would be the single most important piece of evidence this field lacks.
• A human trial in which inducing NRF2 targets or phase-II enzymes through dietary compounds produced a measurable clinical benefit, not a surrogate marker, would upgrade the mild-stress explanation from leading hypothesis to established mechanism.
• Direct human evidence for the strict xenohormesis claim — that we detect plant stress chemistry as environmental information and mount defences on that signal, rather than simply responding to a mildly reactive compound — would separate the two hypotheses this field runs under one name and would justify the title of this entry rather than merely borrowing the field's vocabulary.
• Characterisation of the dose-response curve above dietary intake for a named compound — showing where the hormetic benefit peaks and where it reverses — would let us give real guidance on extracts instead of the current "this is a separate, unstudied decision".
• Evidence that circulating microbial or hepatic metabolites, rather than the parent compounds or any hormetic signal, account for the observed effects would displace the xenohormesis framing entirely, and we would rewrite this entry around it.
• A large trial or a natural experiment that substantially removes healthy-user confounding from the fruit and vegetable association — in either direction — would let us tier the outcome evidence honestly rather than describing it as confounded-but-one-sided.
• What would NOT move us: more in-vitro antioxidant capacity measurements of any kind (the metric was withdrawn for exactly this reason); more rodent or cell-culture NRF2 induction studies (the preclinical case is already made and is not the bottleneck); more plant, insect and field ecology confirming that defence compounds are defence compounds (the premise is not in dispute); another observational cohort finding that vegetable eaters are healthier (the association is established and the confounding is the problem, so a fifteenth cohort adds nothing); a modelled attributable-deaths figure of any magnitude, since it assumes the causality in question; or any volume of extract marketing citing the hormesis mechanism at supra-dietary dose.

Industry bias note

Structural incentives the evidence base may reflect

Commercial pressure sits at both ends of this topic, and unusually, the same true fact powers both.

The plant-avoidance end. Carnivore, ancestral and autoimmune-protocol sellers take a genuine, citable, interesting fact about plant evolution and convert it into permission to eliminate a food group. The tell is a real plant-science citation deployed without its species boundary — an insect feeding study, a plant-defence review or a field-ecology paper quoted as though it described a human — followed immediately by a dietary rule the citation does not support. The secondary tell is the caution list stripped of its conditions: the four-or-five-beans figure without "raw", oxalate advice without "in stone formers with documented hyperoxaluria", goitrogens without "where iodine is inadequate". Each of those qualifiers is doing all the work in the source, and each is routinely deleted.

The supplement end. This one has already pivoted once. For two decades the pitch was antioxidant capacity, and antioxidant claims still appear on packaging fourteen years after the issuing agency withdrew the database and said the values were being routinely misused in exactly that way. When the bioavailability arithmetic made scavenging untenable, the sophisticated end of the market migrated to the hormesis and NRF2 story — which is scientifically more interesting and commercially convenient, because it justifies higher doses. The tell is a concentrated extract sold on a mechanism demonstrated at 200 g of broccoli sprouts, with the two null human studies absent from the marketing and the surrogate-endpoint problem never mentioned. The longevity-supplement sector is a live interest in this specific vocabulary: the xenohormesis hypothesis was named and popularised by a researcher with a long-standing commercial position in plant-derived longevity compounds, which does not make the hypothesis wrong but does explain why the most evocative version of it is the one that reached the public.

The institutional tilt, which is real even where the conclusion is right. Public health bodies, produce interests and plant-based advocacy all benefit from the fruit and vegetable association being read as causal, and the modelled attributable-deaths estimate was amplified in a form that dropped its authors' own causal conditional. On the iodine question, global public health's interest in "adequate iodine neutralises the risk" happens to align with what the evidence shows — alignment of interest with truth is worth naming rather than assuming away.

Our own exposure. A crisp government retraction and a debunked mainstream story are irresistible credibility props for a contrarian health brand, and the structural temptation is to install hormesis as a confident new mechanism-sell in the vacated slot. We have deliberately not done that, and the reason is written into the tier: the old story is dead and the replacement is unproven.

The clean signal. It sits with the parties who had nothing to sell: the 1984 insect pharmacology paper describing what caffeine does to a hornworm, the plant-science literature describing a two-component detonator, the field study that simply watched which animals ate wild chillies, the USDA notice retracting its own widely used metric, and the two null sulforaphane studies whose authors recommended against continuing. None of those advance anyone's product.

Sources (17)

MODERATE — the premises are strong and the antioxidant debunk is strong, but the mild-stress explanation that replaces it is unproven, and the strict xenohormesis version of it is weaker stillLast updated July 2026
Open in the Library: search, filter, every entry →

We set no cookies and run no ad trackers. We count visits with Cloudflare's cookieless, privacy-first analytics. The only thing stored on your device is which example you last viewed.