The Plant Chemicals

TL;DR. Plants make tens of thousands of compounds that are not nutrients: pigments, defence chemicals, flavour molecules. Whole plant foods are consistently associated with better health, and the attempt to identify which specific compound is responsible has mostly failed. Isolated compounds in supplements have repeatedly underperformed the foods they came from, sometimes badly. The most likely explanation is that the benefit comes from the whole package (fibre, matrix, hundreds of compounds acting weakly and together, plus what those foods displace), and that "antioxidant" was the wrong framing for most of it.

Key takeaways

• The antioxidant hypothesis, as popularly understood, has largely failed in trials. Antioxidant supplements have shown no benefit and occasionally harm.
• The better current explanation is hormesis: many plant compounds act as mild stressors that activate your own protective systems, rather than mopping up radicals directly.
• Bioavailability is the recurring problem. Most polyphenols are poorly absorbed, extensively metabolised, and reach blood at concentrations far below those used in laboratory studies.
• The ORAC scale was withdrawn by the USDA in 2012 because it did not predict anything in humans. It is still used in marketing.
• Sulforaphane from brassicas and lycopene from cooked tomato have the clearest mechanisms and the best, though still not definitive, evidence.

What phytochemicals are

In short: Non-nutrient plant compounds, mostly defensive or structural, numbering in the tens of thousands.

Nutrients (vitamins, minerals, macronutrients) are compounds whose absence causes a defined deficiency disease. Phytochemicals are not nutrients: you will not get a deficiency disease from a diet lacking quercetin. Yet diets rich in the foods that contain them are consistently associated with better outcomes.

Estimates put the number of distinct phytochemicals in the human diet somewhere between five and ten thousand, with perhaps a few hundred studied in any depth and a handful studied properly in humans. A single apple contains several hundred.

The rough rule that colour indicates phytochemical class is genuinely useful for shopping: red usually means lycopene or anthocyanin, orange means carotenoids, deep green means carotenoids masked by chlorophyll plus glucosinolates in brassicas, purple means anthocyanins, white and pale means organosulphur or flavonols. "Eat a variety of colours" is a marketing slogan that happens to be reasonable advice.

The antioxidant story, and why it collapsed

In short: A plausible mechanism, a memorable narrative, an industry built on it, and a run of large trials that did not support it.

The hypothesis

Metabolism generates reactive oxygen species (free radicals and related molecules) as an unavoidable by-product. These damage DNA, proteins, and membrane lipids. Cumulative oxidative damage was proposed as a driver of ageing, cancer, and cardiovascular disease. Plant foods are rich in compounds that neutralise radicals in a test tube. Therefore, the reasoning went, supplementing those compounds should reduce disease.

Every step is plausible. The conclusion failed.

The trials

That is an unusually consistent negative record for a hypothesis with such a strong theoretical basis.

Why it failed

Several explanations, probably all partly true.

Radicals are not simply bad. Reactive oxygen species are signalling molecules. Your immune cells use an oxidative burst to kill pathogens. Exercise adaptation is partly driven by oxidative stress signalling: several studies show that high-dose vitamin C and E supplementation blunts the training adaptations to exercise. Suppressing an important signalling system wholesale is not obviously beneficial.

Concentration. In vitro studies apply micromolar concentrations that dietary intake rarely achieves in blood.

The wrong mechanism. The direct radical-scavenging capacity of dietary polyphenols in the body is small relative to endogenous antioxidants such as glutathione, superoxide dismutase, catalase, and uric acid. Whatever polyphenols do, direct scavenging is probably not the main part.

Isolation removes the context. A tomato is not lycopene; it is lycopene plus fibre plus potassium plus a hundred other compounds in a physical matrix, eaten instead of something else.

What replaced it: hormesis

The better current framework is hormesis: a mild stressor provoking a protective adaptive response that exceeds the harm of the stressor itself.

The best-worked example is the Nrf2 pathway. Compounds including sulforaphane (from broccoli), curcumin, and various polyphenols are mildly reactive and mildly toxic. Cells detect them and activate Nrf2, a transcription factor that switches on dozens of the body's own antioxidant and detoxification enzymes: glutathione S-transferases, NAD(P)H quinone oxidoreductase, heme oxygenase-1. The net effect is a stronger endogenous defence system, lasting far longer than the compound itself.

Under this model, plant compounds are useful because they are mildly toxic, which is exactly what Chapter 1 said they were made to be. It also explains why mega-dosing does not scale: past the hormetic window, the stressor is simply toxic.

Don't be confused: ORAC scores mean nothing for you. The Oxygen Radical Absorbance Capacity scale measured a food's ability to neutralise radicals in a test tube. Foods were ranked by it and marketed on it for years. In 2012 the USDA withdrew its ORAC database entirely, stating that the values had "no relevance to the effects of specific bioactive compounds on human health" and that they were being misused in marketing. The scores are still quoted on packaging.

Bioavailability: the recurring disappointment

In short: Most polyphenols are poorly absorbed, heavily transformed by the liver and gut bacteria, and cleared quickly.

The typical fate of a dietary polyphenol:

1. Poor absorption. Many are bound to sugars or to fibre and pass through largely unabsorbed.
2. Immediate metabolism. What is absorbed is rapidly conjugated in the intestinal wall and liver, so the compound circulating in your blood is not the compound in the food.
3. Low peak concentrations, typically nanomolar to low micromolar, often 10 to 1,000-fold below the concentrations used in cell studies.
4. Fast clearance, often within hours.
5. Microbial transformation. A large share reaches the colon, where bacteria break it into smaller phenolic acids, and those metabolites may be the biologically active agents. This is increasingly seen as the main story rather than a footnote.

The clearest illustration is soy isoflavones and equol. Daidzein, an isoflavone in soy, is converted by certain gut bacteria into equol, which is substantially more oestrogenically active. Only about 25 to 30 percent of Western populations and 50 to 60 percent of East Asian populations harbour the bacteria to do this. So identical soy intake produces different exposure in different people, and trials that ignore equol producer status are averaging over two different populations. This may explain a good deal of the inconsistency in soy research.

Curcumin is the other famous case. It has genuinely interesting activity in cell studies and terrible oral bioavailability: poorly absorbed, rapidly metabolised, and barely detectable in plasma after ordinary doses. The standard workaround is co-administration with piperine from black pepper, which inhibits its metabolism and raises bioavailability substantially, and which is why nearly every curcumin supplement contains it. Curcumin also has a reputation among medicinal chemists as a PAINS compound (pan-assay interference compound): it produces apparent hits in a very wide range of assays through non-specific mechanisms such as membrane disruption and metal chelation, which makes its enormous in-vitro literature much less informative than its volume suggests. The human trial evidence, mostly for osteoarthritis pain, is modest.

The compounds with the best cases

In short: A handful have mechanisms, human data, and plausibility all pointing the same way.

Sulforaphane and the glucosinolates

The mechanism is elegant. Brassicas store glucosinolates (glucoraphanin in broccoli) in one cell compartment and the enzyme myrosinase in another. When an insect chews the leaf, the compartments break and myrosinase converts the harmless storage compound into sulforaphane, a reactive isothiocyanate. It is a chemical weapon that is only armed on damage.

In humans, sulforaphane is one of the most potent known dietary activators of the Nrf2 pathway. Evidence includes human trials showing accelerated excretion of airborne pollutants in a randomised trial in Qidong, China, using broccoli sprout beverage, and consistent observational associations between cruciferous vegetable intake and lower cancer risk.

The practical consequence is about cooking. Myrosinase is a protein and is destroyed by heat. Boiled broccoli produces far less sulforaphane than raw or lightly steamed. Two workarounds have human evidence: chop and wait, cutting the vegetable and leaving it 40 minutes before cooking so the enzyme does its work first, and adding a raw brassica (mustard powder, rocket, radish, wasabi) to cooked brassicas to supply myrosinase. Your gut bacteria can also perform the conversion, less efficiently. Steaming for about three to four minutes appears to be the sweet spot: enough to soften, not enough to destroy the enzyme.

Broccoli sprouts contain 10 to 100 times more glucoraphanin than mature broccoli, which makes them the most concentrated practical source.

Lycopene

The red carotenoid in tomatoes, and unusual in that cooking and processing increase its availability substantially. Heat breaks cell walls and converts lycopene from the trans to the more absorbable cis form. Tomato paste and cooked tomato sauce deliver several times more absorbable lycopene than raw tomato, and oil in the same dish increases absorption further, because lycopene is fat-soluble. Traditional Mediterranean cooking arrived at exactly the right method by taste alone.

The evidence: consistent observational association between tomato product intake and lower prostate cancer risk, plus supporting mechanisms. Lycopene supplements have not replicated this, which is the recurring pattern.

Lutein and zeaxanthin

These two carotenoids selectively accumulate in the macula of the retina, forming the macular pigment, where they filter blue light and act as antioxidants in a tissue exposed to intense light and high oxygen. This is one of the few cases where a phytochemical has a demonstrated, specific, anatomical destination.

The AREDS2 trial is one of the genuinely positive supplement trials in this field: a formulation including lutein and zeaxanthin (replacing beta-carotene, which was removed because of the lung cancer signal in smokers) reduced progression of age-related macular degeneration in people with intermediate disease. It does not prevent AMD in healthy eyes.

Sources: kale, spinach, and other dark greens; maize; and egg yolk, whose lutein is particularly well absorbed because it comes packaged with fat.

Flavanols from cocoa and tea

Cocoa flavanols have reasonably consistent short-term trial evidence for improved endothelial function and modest blood pressure reduction (a few mmHg). The large COSMOS trial of cocoa extract supplements found no reduction in the primary cardiovascular outcome, though a secondary analysis suggested a reduction in cardiovascular death. Modest, real, and not a reason to eat more chocolate: most chocolate contains little flavanol, since the processing that makes cocoa palatable (fermentation, roasting, and especially Dutch alkalising) destroys most of it, and it arrives with sugar and fat.

Nitrate from beetroot and greens

Dietary nitrate from beetroot, rocket, spinach, and lettuce is reduced to nitrite by bacteria on the tongue, then to nitric oxide, which dilates blood vessels. This is a well-characterised pathway with measurable effects: beetroot juice lowers blood pressure by a few mmHg and modestly improves exercise efficiency in trials. It is one of the better-evidenced ergogenic aids in sport.

The wrinkle, worth noting because it seems contradictory: nitrates and nitrites in cured meat are associated with colorectal cancer risk, because in the presence of haem iron and protein they form N-nitroso compounds. In vegetables, accompanied by vitamin C and polyphenols that block nitrosation, the same ions behave differently. Context again.

Resveratrol, and a cautionary tale

Resveratrol from grape skins was the proposed explanation for the "French paradox" and became one of the most heavily marketed supplements of the 2000s, on the back of research on sirtuins and lifespan extension.

Two problems. First, the amount in red wine is tiny: reaching the doses used in mouse studies would require drinking hundreds of litres a day. Second, the field suffered a serious research integrity failure when a leading resveratrol researcher was found by his university to have falsified data in dozens of papers, which were retracted. Human trials have not shown meaningful benefit. It is the clearest example in this chapter of a compound whose reputation was built on marketing rather than evidence.

What this means for eating

In short: Eat a variety of whole plant foods, prepare them in ways that preserve or release what is in them, and be sceptical of extracts.

The consistent finding across nutritional epidemiology is that whole plant foods are associated with lower risk of essentially every major chronic disease. The consistent finding across supplement trials is that isolated compounds from those foods mostly do not reproduce it.

The practical rules that follow:

1. Variety over intensity. Thirty different plants a week beats large amounts of one superfood. Different compound classes, and probably synergy.
2. Colour is a proxy for compound class, so eating across the spectrum is a reasonable heuristic.
3. Preparation matters more than people think: fat with carotenoids, chop-and-wait or light steaming for brassicas, cooking for lycopene, raw for vitamin C and myrosinase.
4. Skins and outer leaves are where polyphenols concentrate. Peeling an apple removes most of its quercetin, and the outer leaves of a lettuce or cabbage are the most nutrient dense.
5. Be sceptical of extracts and "superfoods." The word superfood has no scientific definition and is banned as an unsubstantiated health claim on packaging in the EU. Exotic berries are not better than local ones; they are more profitable.
6. Some interactions are real. Grapefruit inhibits CYP3A4 and interacts dangerously with many drugs (Chapter 84). St John's wort induces the same enzyme and reduces the effectiveness of many drugs including contraceptives. High-dose green tea extract has caused liver injury. Concentrated plant compounds are drugs.

The bottom line

• Plants contain thousands of non-nutrient compounds, mostly defensive, and the foods containing them are consistently associated with better health.
• The simple antioxidant explanation failed in trials, sometimes with harm. Hormesis, mild stress provoking your own defences, is the better current model.
• Bioavailability is the recurring obstacle: what circulates in your blood is usually a metabolite at a fraction of the concentration used in laboratory studies, and gut bacteria do much of the transformation.
• Sulforaphane, lycopene, lutein and zeaxanthin, and dietary nitrate have the clearest mechanisms and best human data. Resveratrol is the cautionary tale.
• Eat varied whole plants, prepare them thoughtfully, and treat concentrated extracts as drugs rather than food.

Sources and notes

Compound classes and dietary distribution follow the Phenol-Explorer database and standard phytochemistry reviews. The antioxidant supplement trial record is the same ATBC, CARET, HOPE, and SELECT set as Chapter 15, plus the Cochrane antioxidant reviews. USDA withdrawal of the ORAC database, with its stated reasoning, was announced in 2012. Nrf2 activation by sulforaphane follows Talalay's and Fahey's work at Johns Hopkins; the Qidong broccoli sprout trial is Egner et al., Cancer Prevention Research, 2014. Polyphenol bioavailability and microbial transformation follow Manach et al., American Journal of Clinical Nutrition, 2005. Equol producer status follows Setchell's work. Curcumin bioavailability and its PAINS characterisation follow Nelson et al., Journal of Medicinal Chemistry, 2017. Lycopene bioavailability and cooking follows Gartner et al. and Rao's reviews. AREDS2 is the JAMA 2013 report. Dietary nitrate and nitric oxide follows Lundberg, Weitzberg, and Gladwin, Nature Reviews Drug Discovery, 2008, and the mouthwash abolition studies. The resveratrol research misconduct case is the University of Connecticut investigation and subsequent retractions.

Open questions. Which polyphenol metabolites are biologically active in humans, and at what concentrations, is largely unknown, which is why almost every mechanistic claim in this area outruns its evidence. Whether hormesis is a useful general framework or a post-hoc explanation is debated.

👉 Next: the natural toxins already in your food, because "natural" was never a safety category.