Evolution is not a recipe. It is a record of what our ancestors survived.

The autopsy

George Mann spent the 1960s in East Africa trying to prove Ancel Keys wrong.

Mann was a Vanderbilt biochemist and a former associate director of the Framingham Heart Study, which meant he knew exactly what he was looking for. The Maasai were his test case. Here was a people living on milk, meat and cattle blood, taking in, by the accounts of the day, an enormous quantity of fat and cholesterol, and not dropping dead of heart attacks. If diet caused heart disease, they should have been falling over.

In 1964 he published his survey of four hundred Maasai men. They were lean. Hypertension was almost absent. Of the 363 men whose blood was pooled for analysis, serum cholesterol averaged about 125 mg/dL, roughly half the American figure of the era. That paper has been cited ever since as proof that dietary fat is not the problem.

The paper also acknowledges support from the National Live Stock and Meat Board.

Hold that thought, because of what Mann did next.

Eight years later he published something else. He and his colleagues had obtained the hearts and aortas of fifty Maasai men at autopsy, and they wrote up what they saw:

Measurements of the aorta showed extensive atherosclerosis with lipid infiltration and fibrous changes but very few complicated lesions. The coronary arteries showed intimal thickening by atherosclerosis which equaled that of old U.S. men.

Equal to old American men. The plaque was there. What was rare was the event, the rupture, the clot, the heart attack, and Mann’s own explanation was that a lifetime of physical work had widened their coronary arteries enough to keep the channel open around the disease.

So the meat industry funded the study that made the Maasai famous, and the same researcher then went and collected the evidence that undercut the use his study was being put to. He didn’t have to. He published it anyway.

That is a better man than the argument against him requires, and it is the standard this essay is trying to meet. The full case, including the Inuit version of the same argument, is taken apart here.

A detail from the 1964 fieldwork rarely travels with it. Of those four hundred adult men, three were over the age of 55. Heart attacks are a disease of older people, and a study of young men is not evidence that a diet protects the old. Mann knew it. His 1964 paper says plainly that there was no autopsy data to confirm the anatomical situation, and calls for exactly the material he would supply eight years later. It also reports hand-dynamometer results showing the Maasai were less powerful than the American comparison group, and notes that the elders seemed sedentary, which complicates the fitness explanation he would later reach for.

The same thing happens further north. When researchers ran CT scanners over 137 preserved bodies from four preindustrial populations, they found atherosclerosis in a third of them, including three of the five Unangan hunter-gatherers of the Aleutian Islands, who ate a marine-mammal diet (Thompson et al. 2013). Five bodies is far too few to compare rates against anything. It is enough to establish that a diet of seal and whale did not prevent the disease.

A note before we go further. I think it is wrong to kill animals for food when we don’t need to. That is a moral position, not a scientific one, and it will not appear again until the last third of this essay, where I will tell you exactly what work I am asking it to do. You will meet the number of animals slaughtered each year before then, counted as a fact of scale alongside acres and tonnes; what that number means morally waits until the end. Everything between here and there is meant to stand without it. If you catch me leaning on it, stop trusting the rest.

There was never one diet

The Maasai case fails on its own evidence. But there is a deeper problem with the family of arguments it belongs to, and it needs addressing directly, because otherwise every refuted example is simply replaced with another.

The argument runs: our ancestors ate a certain way, we are built for that way, therefore that way is optimal. It requires there to have been a way.

Consider three well-studied populations. Maasai women got roughly 42% of their calories from dairy and about 9% from meat, in a diet with essentially no fruit, vegetables or pulses. The Inuit and Unangan of the Arctic ate marine mammals and fish, in an environment that offered almost nothing that grows. The Tsimané of the Bolivian Amazon eat plantains, rice, corn, cassava and fruit with modest lean game and fish, about 72% carbohydrate, 14% fat.

When Hillard Kaplan’s team ran CT scans on 705 Tsimané adults, 85% had no detectable coronary calcium at all. The comparable American figure is 14%. The researchers described the lowest levels of coronary artery disease ever recorded in a studied population (Kaplan et al. 2017).

Before anyone runs that as an ancestral argument, it deserves the same scrutiny the Maasai just got. The Tsimané are also extraordinarily active, lean and non-smoking, and each of those independently lowers cardiovascular risk, so their diet is not the only thing keeping their arteries clean. But notice that the scrutiny cuts differently here. They also carry an enormous burden of infection and inflammation, with C-reactive protein levels that would ring alarm bells in a Western clinic, and that pushes cardiovascular risk up, not down. They have the arteries anyway. So the honest reading of the Tsimané is not a confounded story to be waved off. It is a population that should have had heart disease on every count except diet and activity, and did not.

Which is the actual point, and it is worked through population by population in the ancestral diet argument. The appeal is unfalsifiable by selection, whichever population you name, someone names another. What settles it is not a better population. It is genetics.

These peoples are not interchangeable bodies eating different foods. They carry different machinery.

Adult milk digestion in the Maasai runs on regulatory variants near the lactase gene, chiefly C-14010, that arose in East Africa independently of the European variant, on a different genetic background, in the last several thousand years (Tishkoff et al. 2007). The selection signal there is the strongest in the Maasai genome, and stronger than anything at the equivalent European locus (Schlebusch et al. 2013). Arctic peoples carry a variant in CPT1A, the enzyme that feeds long-chain fats into the mitochondria, at frequencies near 70% and essentially zero everywhere else (Clemente et al. 2014).

Then the detail that ends the argument. The FADS gene cluster builds long-chain fatty acids. In Greenlandic Inuit, selection favored variants that suppress that pathway, sensible when your food already arrives packed with marine omega-3 (Fumagalli et al. 2015). In Europeans, selection favored variants that increase production of the same long-chain fats from the short-chain precursors you get from plants (Buckley et al. 2017).

The same gene cluster, two populations, opposite directions. Tuned in one case to marine animals and in the other to plants. The European story is not simple, the alleles under selection shifted before and after the arrival of farming, but the contrast with the Arctic holds.

Meanwhile the gene for salivary amylase, which starts breaking down starch in your mouth, has been multiplying in our lineage for a very long time. Two 2024 analyses dated the first duplication to possibly 800,000 years ago and found hunter-gatherers 45,000 years ago already carrying four to eight copies, with European farmers rising from around four to around seven in roughly four thousand years (Yilmaz et al. 2024; Bolognini et al. 2024).

When Herman Pontzer and Brian Wood reviewed the forager evidence, they found the plant-to-animal balance varies widely and tracks climate and economy rather than anything about human nature. A roughly even split prevails among warm-climate hunter-gatherers. Cold-climate societies and those relying on fishing or herding eat more meat; warm-climate foragers and part-time farmers eat more plants. Paleolithic diets, they concluded, would have been remarkably variable across time and space (Pontzer & Wood 2021). What foragers share is a pattern, whole foods, enormous variety, a great deal of fiber, constant physical work, not a ratio.

There is also a ceiling on how much protein a human can eat, somewhere around 35 to 40 percent of calories, beyond which Arctic explorers documented what they called rabbit starvation (Speth & Spielmann 1983; Bilsborough & Mann 2006). That is why any all-meat human diet has to be a high-fat diet. It isn’t a preference. It’s the only shape a viable diet can take with those ingredients and this body.

Local adaptation is real, and none of it transfers. The Maasai could drink three liters of milk a day because of an allele most of humanity doesn’t carry.

Evolution was never trying to make you old

Even if there had been a single ancestral diet, it would not follow that the diet was good for you. Natural selection does not optimize for health. It optimizes for reproduction, and its grip loosens sharply once reproduction is done.

Two ideas from evolutionary medicine explain why that leaves us where it does. They are worth learning by name, because once you have them you can see the shape of a great many arguments about food, and you stop being persuaded by the weaker ones.

What is antagonistic pleiotropy?

Pleiotropy just means one gene having more than one effect. Antagonistic pleiotropy is the case where those effects pull in opposite directions: the gene helps you early in life and harms you later.

George Williams put it into its modern form in 1957, and the logic is uncomfortable once you follow it. Selection is a numbers game played in your reproductive years. A gene that helps a twenty-year-old survive, grow and reproduce gets passed on in large numbers. If that same gene quietly damages a seventy-year-old, selection barely notices, because by seventy the gene has already done its work and there is very little selective pressure left to remove it. The damage arrives after the accounting closes.

So a trait can be an evolutionary success and a health liability at the same time. That is not a paradox. It is the expected result of a process that counts grandchildren rather than birthdays.

This is also why lifespan and healthspan come apart. Lifespan is how long you last. Healthspan is how long you last in good working order. Evolution had a stake in getting you to reproduction. It had no stake in the thirty or forty years that come after, which is precisely the stretch most of us are now trying to live well.

What is evolutionary mismatch?

Evolutionary mismatch is the second idea, and it works from the outside in rather than the inside out. It describes what happens when an environment changes faster than the bodies living in it can adapt. A trait that was well matched to the conditions our ancestors faced becomes a liability under conditions they never faced.

The classic illustrations are almost too familiar to notice. A powerful appetite for calorie-dense food is an excellent trait when food is scarce and unpredictable, and a poor one beside a vending machine. A stress response tuned for brief physical emergencies is useful when the threat is a predator and corrosive when the threat is an inbox. Nothing has gone wrong with the body. The world moved.

How do antagonistic pleiotropy and evolutionary mismatch fit together?

Here is the part that matters for food, and it is the hinge of this whole essay.

Antagonistic pleiotropy loads the gun. It leaves us carrying traits with late-life costs that selection never had reason to remove. Evolutionary mismatch pulls the trigger, and it does so in two ways at once. It hands us an environment those traits were never tested against. And, because modern medicine and sanitation now carry most of us decades past reproduction, it keeps us alive long enough to actually pay the late-life bill that selection was blind to.

Far fewer of our ancestors lived long enough to collect on those debts. We do. That is not a failure of our biology; it is a consequence of our success at everything else.

Both of these are frameworks rather than findings, and I am not going to claim that eating animals is bad for you because of them. Neither idea can tell you what to put on a plate. What they do is dismantle a common move, humans evolved eating X, therefore X is good for us. Evolution has no stake in your seventh decade. Whatever our ancestors ate got them to reproduction in the environments they lived in. That is the entire claim the evidence supports.

For what that costs in practice, return to CPT1A, the Arctic fat-processing variant. Selection drove it to roughly 70% in Arctic populations, and it is associated with impaired fasting ketogenesis, hypoglycemia in infants, and elevated infant mortality (Clemente et al. 2014; Gessner et al. 2016). The Inuit FADS alleles carry costs too, effects on the order of two centimeters of height and four kilograms of weight (Fumagalli et al. 2015).

These are the adaptations most often held up as proof that a population was built for its diet. They were. They were also expensive. Selection is not a quality-control department; it is a filter that keeps whatever reproduces, side effects included.

So the past cannot tell us what to eat. What it can tell us is what happens when the environment changes and the genome doesn’t, and we have run that experiment many times, in the lives of people who lived it.

We already ran the experiment

Evolutionary mismatch would be an untestable story if the environment only ever changed slowly. But it has changed abruptly, many times, in populations we can name and follow, and each of those is a natural experiment in mismatch. The genome stays put. The world around it changes inside a generation. Then you watch.

The most instructive evidence in nutrition rarely comes from asking a hundred thousand people what they ate last year. It comes from situations where something changed and you can watch what followed. Arrange those by how well each design controls for the alternatives.

Start with a reversal. In 1984 Kerin O’Dea took ten diabetic Aboriginal Australian adults, all living an urbanized life in northwestern Australia, and returned them to a traditional hunter-gatherer existence in their ancestral country for seven weeks. Fasting glucose fell. Insulin fell. Triglycerides fell dramatically. The metabolic profile of diabetes largely unwound in under two months (O’Dea 1984).

Now the part that has to be said in the same breath. They were eating around 1,200 calories a day and lost about eight kilograms. They were walking, hunting and fishing constantly. And the food they returned to was rich in wild lean meat, shellfish and fish. However, this may not mean animal foods reverse diabetes. It shows that energy restriction, weight loss and hard physical work reverse diabetes, and that a traditional life delivered all three at once.

Now the same experiment in reverse, with the same caveat. The Tarahumara of northern Mexico are famous for long-distance running and a diet built on corn and beans. In 1991 researchers fed thirteen of them an affluent Western diet for five weeks. Total cholesterol rose 31%, LDL rose 39% (McMurry et al. 1991). The affluent diet also supplied about 4,100 calories against their usual 2,700, and they gained weight, so this is a study of Westernization, calories included, not of any single nutrient. Applying that caveat to O’Dea and withholding it here would be exactly the selectivity this essay is complaining about.

Then a natural experiment that controls for genes. The Pima people are split by an international border. Some live in Arizona; a related group farms by hand in the Sierra Madre of Mexico. Type 2 diabetes prevalence was 6.9% among the Mexican Pima and 38% among the Arizona Pima (Schulz et al. 2006). Same ancestry, different environment, a more than fivefold difference.

Then the migrant cohorts. The Ni-Hon-San study followed Japanese men in Japan, Hawaii and California. Coronary heart disease rose along that gradient with the degree of Westernization (Marmot & Syme 1976). Stroke ran the other way, highest in Japan. Hold on to that; it comes back.

And then the population transitions, the weakest design. Among the Maasai, diabetes was essentially absent in the 1930s; by 2006 impaired glucose tolerance or diabetes affected 5.2% of rural men and 7.8% of rural women, rising to 11.8% and 26% in urban samples (Pressler et al. 2022). Among Alaska and Greenland Inuit, diabetes rates rose almost threefold between the 1960s and 1970s, and arterial calcification increased fivefold in families that gave up a nomadic life (Pressler et al. 2022).

Every one of these transitions changed more than food. Calories went up, physical work went down, smoking and alcohol and processing arrived, and so did antibiotics, vaccination and obstetric care, which is why life expectancy in most of these populations rose over the same period. Nobody gets to run that correlation cleanly in either direction.

The review that supplies the Maasai numbers goes further, and it is worth reporting because it cuts against what comes next. Looking across the four populations it reviewed systematically and seven more it examined for bias, Pressler and colleagues concluded that the health outcomes cannot be attributed to saturated fat or total fat consumption. What they implicate is refined carbohydrate and sugar, Maasai carbohydrate intake rose 39% in men and 362% in women.

I think they are right about what their evidence can show, and that it does not reach as far as it appears to. Population transitions cannot isolate saturated fat, because saturated fat moves in lockstep with calories, processing, sugar and inactivity. Ruling it out requires studies that can hold those constant. That is the next section, and it uses a different and stronger class of evidence.

What this section establishes is narrower and solid: the disease was environmental, and it was reversible. The environment is the variable. But you cannot eat an environment.

Which part of the change?

Nutrition arguments usually get settled by whoever cites the most studies. That is the wrong instinct. Evidence-based medicine sorts studies into a hierarchy of evidence, from case reports and ecological comparisons at the bottom, through observational cohorts, up to randomized controlled trials and genetic causal methods at the top, and the ranking exists because each rung fails in a characteristic way that the rung above it fixes.

Every study design fails in a characteristic way. Observational cohorts are confounded. Trials are short. Mechanisms mislead. Genetic studies are indirect. So the useful question is not which study is best, but what each design defeats, and whether designs that fail differently agree anyway.

To defeat reverse causation and confounding entirely, you need genetics. LDL cholesterol does not merely correlate with heart disease; it causes it. People born with variants giving them lifelong low LDL have correspondingly low rates of heart attack. Drugs that lower LDL by unrelated mechanisms lower events in proportion to how much they lower it. The dose-response holds across both (Ference et al. 2017). That is as close to established causation as nutrition gets, and it converts a blood test into a bridge between what trials can measure and what cohorts observe.

It also explains the Tsimané. Their mean LDL is about 91 mg/dL, and the paper’s own conclusion leads with a lifetime of very low LDL rather than with any particular food.

To defeat the “which nutrient” problem, you need controlled feeding. Saturated fat raises LDL, and not all saturated fats do it equally: lauric, myristic and palmitic acids raise it, myristic and palmitic most potently, while stearic acid is roughly neutral because the body rapidly converts it (Bonanome & Grundy 1988; Mensink et al. 2003). Dairy fat is dominated by palmitic and myristic, which is why butter behaves the way it does. Beef fat is more mixed, carrying a substantial stearic fraction alongside the palmitic, so the point holds firmly for dairy and more loosely for meat.

To defeat the “compared to what” problem, you need substitution analyses. Nobody eats in a vacuum. A 2023 meta-analysis of 24 cohorts asked what happens when one food replaces another. Replacing processed meat with whole grains was associated with a hazard ratio of 0.64 for cardiovascular disease, with nuts 0.73, with legumes 0.77; replacing eggs with nuts, 0.83. For diabetes, red meat for whole grains was 0.90 and processed meat for nuts 0.78. Swapping 50 grams a day of processed meat for a handful of nuts was associated with 21% lower all-cause mortality (Neuenschwander et al. 2023).

The same analysis found no clear benefit from replacing fish or poultry with nuts or legumes. The gradient is not uniform across all animal foods. It is steep at the red-and-processed-meat end and flat at the seafood end, and the honest version of the finding says so.

To defeat between-study noise, look inside one dataset where the same researchers classify everyone the same way. In the Adventist cohort, two-year incidence of type 2 diabetes ran 2.12% among non-vegetarians, 1.29% among pescatarians, 1.08% among lacto-ovo vegetarians and 0.54% among vegans. The fifth group breaks the order: semi-vegetarians came in at 0.92%, below both pescatarians and lacto-ovo vegetarians (Tonstad et al. 2013). A gradient with one rung out of place is what real data looks like.

To defeat the macronutrient framing, separate low-carbohydrate diets by what replaced the carbohydrate. Done that way, the animal-based version was associated with higher mortality and the plant-based version with lower (Fung et al. 2010; Seidelmann et al. 2018). Both are observational and the people choosing each differ in other ways, so this does not isolate the food. What it does show is that the macronutrient label on the tin tells you almost nothing.

To defeat genetic confounding, use twins. In 2023 a Stanford team randomized 22 pairs of identical twins, one to a healthy vegan diet, one to a healthy omnivorous diet, for eight weeks. The vegan twins ended with LDL 13.9 mg/dL lower than their siblings, plus lower fasting insulin and body weight (Landry et al. 2023). Eight weeks, 44 people, and not calorie-matched: the vegan arm ate less and lost weight, which drives part of the LDL change.

To defeat self-report, put people in a metabolic ward. Given unlimited access to either a low-fat plant-based diet or an animal-based ketogenic diet, twenty people spontaneously ate 689 fewer calories a day on the plant-based one (Hall et al. 2021). The arms differed in energy density as well as in animal content, 1.1 against 2.2 calories per gram, and body weight actually fell more on the ketogenic arm. What the trial cleanly demonstrates is that a low-energy-density whole-food diet produces spontaneous calorie reduction without instruction.

To defeat the noise around cancer, there is formal review. In 2015 a working group convened by the World Health Organization’s cancer agency classified processed meat as a Group 1 carcinogen for colorectal cancer, red meat one rung below (Bouvard et al. 2015). Group 1 describes the strength of the evidence, not the size of the risk, which is nothing like tobacco’s. Mendelian randomization, which uses genetic variants to approximate a randomized comparison, points the same way (Yun et al. 2023).

And one thing needs no design at all. Fiber comes only from plants. An analysis commissioned for WHO guidelines, pooling 185 prospective studies and 58 trials, found the highest fiber consumers had 15–30% lower all-cause and cardiovascular mortality. The same review’s randomized arm found higher fiber intake lowered body weight, systolic blood pressure and total cholesterol (Reynolds et al. 2019), the interventional evidence people often say is missing from this field.

A diet built entirely on animal foods scores a structural zero on the dietary factor most consistently associated with living longer. That is not an argument about ratios. It is an argument about which foods contain the thing.

These designs fail in different ways and agree anyway. Not one of them names a place to stop.

Where the evidence stops

Now the other side of the ledger, in full, because an essay that presents only one side has told you about its author rather than about the evidence.

Stroke. The EPIC-Oxford cohort followed 48,188 people for eighteen years. Vegetarians had 22% lower ischemic heart disease, about ten fewer cases per thousand over a decade, and 20% higher total stroke, mostly hemorrhagic, about three more cases per thousand (Tong et al. 2019). Two things about that heart-disease number belong here rather than in the previous section: it attenuates to 0.90, with the confidence interval touching 1.00, once you adjust fully for cardiovascular risk factors. The stroke association, by contrast, did not shrink on adjustment.

What limits it: EPIC-Oxford recruited in the 1990s, before supplementation was routine, and its own biomarker substudy found 52% of its vegans and 7% of its vegetarians B12-deficient by a strict cutoff, against one omnivore in 226 (Gilsing et al. 2010). Low B12 raises homocysteine, which is associated with stroke, and the stroke models adjusted for neither serum B12 nor supplement use. A Taiwanese cohort found vegetarians there had roughly half the stroke rate of non-vegetarians, and, more to the point, found that B12 intake modified the association, with a formal interaction test at p = 0.046 (Chiu et al. 2020). That is the strongest direct evidence for the B12 explanation, and it comes from the study that reverses the finding. Remember also that stroke ran opposite to heart disease in the Japanese migrants, in a population eating no shortage of animal food. The direction of the stroke association is unresolved.

Fractures. Vegans in EPIC-Oxford had 2.31 times the hip-fracture rate of meat-eaters, about fifteen extra per thousand over a decade (Tong et al. 2020). It replicates, and this is where I have to be more careful than the usual account.

The usual account points to an earlier analysis of the same cohort in which vegans getting at least 525 mg of calcium a day had a fracture rate ratio of exactly 1.00 (Appleby et al. 2007), and to the Adventist finding that vegan women taking calcium and vitamin D had no excess risk (Thorpe et al. 2021). Both are real. But the later, longer, hospital-recorded analysis did not reproduce that. Adjusting for calcium moved the vegan total-fracture hazard ratio from 1.43 to 1.31, about 28% of the excess. Adjusting for protein moved it to 1.39, about 9%. And vegans already eating adequate protein were not better off; their hip-fracture ratio was 2.71, higher than the overall estimate.

So calcium does roughly three times the work of protein, and it does not do all of it. Something beyond calcium and protein remains unexplained. One candidate is exercise type: vegans doing resistance training had bone microarchitecture indistinguishable from omnivores who trained (Wakolbinger-Habel et al. 2022), but that is a hypothesis, not a finding. A 2025 meta-analysis of hip fracture puts the vegan estimate at 1.75 and grades the certainty very low; it also puts vegetarians at 1.25 at low certainty (Ballarin et al. 2025), which is a better-graded signal and extends the problem beyond veganism. And the same cohort found elevated hip fracture in fish eaters, at 1.26, which no account built on vegan nutrient gaps explains at all.

This one does not resolve. I am not going to pretend it does.

Cancer. A 2026 pooled analysis of 1.8 million people found vegans had a hazard ratio of 1.40 for colorectal cancer against meat-eaters. That rests on 93 cases and loses significance when the first four years of follow-up are excluded; pescatarians in the same pooling came in at 0.85. The same analysis reports vegetarians at 1.93 for squamous-cell esophageal cancer (Dunneram et al. 2026). The paper and its weak points are examined in full in The Real Story on Diet and Bowel Cancer.

Death from dementia and stroke in the very old. The 2024 Adventist follow-up found that among participants aged 85 and over, vegetarians had higher rates of death from stroke, dementia and Parkinson’s disease, with two of three intervals touching or crossing 1.0 (Abris et al. 2024). The category pooled everyone not eating meat, including pescatarians whose omega-3 levels resemble omnivores’, and the signal appeared anyway, which argues against omega-3 as the missing piece. The authors attributed it to possible B12 and long-chain omega-3 gaps.

No vegan trial has measured the thing that matters most. A Cochrane review found no trial of a vegan diet anywhere reporting cardiovascular events; all thirteen measured blood markers (Rees et al. 2021). This is not true of every dietary pattern, PREDIMED and the Lyon Diet Heart Study randomized people to Mediterranean-style diets and counted cardiovascular events (Estruch et al. 2018; de Lorgeril et al. 1999). Those trials exist, they favor a plant-forward pattern, and their absence on the vegan side is a real gap.

Age changes the answer. People aged 50–65 with high protein intake had substantially higher mortality and cancer mortality, attenuated or absent for plant protein. In the same dataset, over 65, the association reversed, higher protein was protective (Levine et al. 2014). Older adults lose muscle and muscle loss kills; nothing here argues for restricting protein in the elderly. The same study found roughly fivefold higher diabetes mortality with high protein intake across every age group. And notice the shape: lower protein looks protective in midlife and harmful in old age, a benefit early and a cost late in the same variable. That is an analogy rather than an example, since antagonistic pleiotropy is about genes rather than nutrients, but it rhymes.

And the people who choose these diets are different. Western vegetarians and vegans smoke less, exercise more, weigh less and earn more. Adjustment cannot eliminate what it cannot measure, and hazard ratios of 0.8–0.9 sit comfortably inside what residual confounding produces. Two things limit the objection: the Adventist cohort compares people inside one community with shared norms on smoking, alcohol and exercise, and the gradient appeared anyway; and research on dietary motivation finds that less restrictive avoidance is more health-driven, while vegans report stronger ethical than health motives (Hopwood et al. 2020). If that holds here, the health-optimizer bias runs toward the fish-eaters.

Now the pattern, stated accurately rather than conveniently. Most of these findings point at nutrients rather than at plants. Stroke points at B12, with an interaction term to support it. The dementia signal was attributed to B12 and omega-3 by its own authors. The colorectal estimate is fragile. That is the argument for planning, and it is why “well-planned” does real work. B12 without exception, from a supplement or fortified foods. Vitamin D, supplemented through the darker months at most latitudes. Iodine, from nori or iodized salt. Calcium, from fortified plant milks, calcium-set tofu, kale or tahini. Iron from legumes, tofu and dark greens, eaten with vitamin C to help absorb it. And adequate protein, from legumes, nuts, seeds, soy foods or seitan, more of it, not less, past sixty-five.

But the fracture signal partly resists that explanation, and the fish-eater result resists it entirely. Something in that finding is not about supplements. I don’t know what it is, and neither does anyone else yet.

And here is where the evidence stops.

No study in this literature identifies a level of animal food whose addition to a well-planned plant-based diet improves a hard outcome. That claim is weaker than it sounds. It does not say a fully plant-based diet beats a mostly-plant one. It says nothing in the data marks a floor, so anyone asserting one is making a claim the evidence does not support, in either direction.

The evidence gives a direction. It declines to name a stopping point. Something else has to.

What we actually hand down

Before that, a question people ask about food that is really a question about time.

Start with a fact of embryology. A woman is born with every egg she will ever have, and those eggs form while she is herself a fetus. The classic germ-cell counts rest on few specimens and are debated (Baker 1963); the timing is not. When your grandmother was pregnant with your mother, the cell that would become half of you was already forming inside your mother’s developing ovaries. Your grandmother’s nutrition during that pregnancy reached three generations at once.

Your DNA is still a fresh shuffle of both parents. But the cellular starting conditions were laid down a generation earlier than most people imagine. That is developmental biology. It requires no epigenetics at all.

What is epigenetics, in plain terms?

Your DNA sequence is the text. Epigenetics is everything that decides which parts of the text get read, and how loudly. The most studied mechanism is DNA methylation: small chemical tags attached to the DNA that turn genes down or off without altering a single letter of the sequence itself. Volume knobs rather than rewrites.

Those settings respond to the environment, including nutrition, which is what makes the idea so appealing when we talk about food. If diet adjusts the volume knobs, and the knob settings can be inherited, then what you eat could echo down generations. That is the claim, and it is where care is needed, because one half of it is well established and the other is not.

What is the difference between intergenerational and transgenerational inheritance?

The line scientists draw here is routinely blurred in popular coverage, and it is the difference between a solid finding and an overreach.

Intergenerational effects reach a generation that was directly exposed. If a pregnant woman is starved, three parties are exposed at once: the woman, the fetus inside her, and the egg and sperm cells already forming inside that fetus. Effects on any of those are impressive but not mysterious. Something was physically there when the exposure happened.

Transgenerational effects reach a generation that was never exposed at all. For a maternal exposure, that means the great-grandchild. This is the claim that would be genuinely startling, and the next section explains why it is so hard to demonstrate.

Hold that distinction and most of the confusion in this area dissolves.

What is transgenerational epigenetic inheritance?

Put the two halves together and you get the full concept. Transgenerational epigenetic inheritance is the claim that an environmental exposure in one generation leaves a chemical mark on the DNA that is copied faithfully through the germ line and still shapes the biology of a descendant who never encountered that exposure, without any change to the DNA sequence itself.

It is a genuinely radical idea, and worth being clear about why. Ordinary inheritance passes down the text. This would pass down the annotations. It would mean a famine, a toxin or a diet could write instructions into a lineage that outlast everyone who lived through the event, which is close to the Lamarckian inheritance of acquired characteristics that biology spent a century ruling out.

The obstacle is mechanical rather than philosophical. A mark has to survive two rounds of near-total erasure: one when eggs and sperm are formed, another in the days after fertilization when the embryo resets its epigenome to build a body from scratch. Both events strip most methylation deliberately, because a cell that is going to become an entire organism cannot arrive carrying its parent’s settings. A handful of regions escape, which is why the idea is not absurd. But anything claiming to have made the trip has to explain how it got through both.

In plants, worms and flies, it demonstrably happens. Their germ lines do not reprogram the way ours do. In laboratory rodents there is real evidence, though it is contested. In humans it is not established, and the reason is worth stating plainly: families pass down far more than chemistry. They pass down genes, neighborhoods, incomes, traumas, kitchens and recipes. Disentangling a methylation mark from a grandmother’s cooking habits, in a species you cannot breed under controlled conditions for four generations, is close to impossible with the tools we have.

So the popular version of this, that your grandmother’s diet reprogrammed your genes, is running well ahead of the evidence. What follows is what the evidence will actually carry.

With that in hand, here is the honest ladder, rung by rung.

That a mother’s nutrition in pregnancy shapes her child’s lifelong metabolic risk: established. People conceived during the Dutch Hunger Winter carried measurably different methylation on a growth-regulating gene six decades later, compared against their own siblings conceived before or after (Heijmans et al. 2008).

That a father’s diet and weight alter his sperm and can reach his children: strong in animals, emerging in humans. The sperm of men with obesity carries a different epigenetic pattern than that of lean men, and the pattern shifts within a year of weight loss after bariatric surgery (Donkin et al. 2016). The molecular change is real. Whether it alters children’s health is not established.

That grandparents’ diets echo in grandchildren: suggestive, mechanism contested. In the Swedish parish of Överkalix, earlier studies, as Vågerö and colleagues summarize them, reported sex-specific patterns linking a grandfather’s childhood food supply to his grandsons’ mortality. A far larger Swedish study tested it: the grandson mortality signal partly held for cancer, but the diabetes and cardiovascular findings did not replicate (Vågerö et al. 2018).

That epigenetic marks pass unaltered through multiple human generations: not established. A 2018 review put it bluntly, confirmed in plants, worms and flies, and in humans nearly impossible to separate from everything else families pass down, including what they cook (Horsthemke 2018).

The claim I wanted to make here was that what you eat reprograms your grandchildren’s biology, and I can’t, because the human evidence doesn’t support it.

What is the Mother’s Curse?

One more piece of biology gets dragged into this and shouldn’t. Mitochondria, the structures that turn food into usable energy, carry their own small loop of DNA, inherited only from your mother. A mutation harming daughters gets weeded out; one harming only sons is nearly invisible to selection, because it never passes through a son to the next generation. Evolutionary biologists call it the Mother’s Curse, and in 2017 a team traced exactly that pattern through French-Canadian genealogies across almost three centuries (Milot et al. 2017). The paper’s own title says the curse neutralizes natural selection against the disease. It is a real thing, it has nothing to do with diet, and no food changes it.

Put the whole ladder together and one thing stands out. Beyond the generation directly exposed, the channel between what we do now and the people who come after us, the one we can actually measure and act on, does not run through the genome.

It runs through the world we hand them.

The second ledger

That world is measurable, and the measurements do not depend on resolving a single nutrition controversy in this essay.

The largest analysis of the food system covers 570 studies, roughly 38,700 farms across 119 countries, and 40 products representing about 90% of global protein and calories. Animal products occupy about 83% of the world’s farmland and generate 56–58% of food’s greenhouse gas emissions, while supplying 37% of our protein and 18% of our calories (Poore & Nemecek 2018).

The obvious objection is that averages hide the best producers, and the paper answers it: even the lowest-impact animal products typically exceed the average impacts of plant substitutes. The gap is not between good farms and bad farms.

That is modeling from production data. The other check is to take what people actually ate. A 2023 study linked the dietary records of 55,504 UK adults to food-level environmental data. Against people eating 100 grams of meat a day or more, vegan diets came in at roughly 25% of the greenhouse gas emissions, 25% of the land use, 46% of the water use, 27% of the eutrophication and 34% of the biodiversity impact (Scarborough et al. 2023). Every indicator rose with animal-food consumption.

It is worth being literal about what that farmland is for. The system runs on two kinds of building: the shed where animals are grown and the plant where they are killed. Nearly 88 billion land animals passed through the second one in 2024 alone (FAO 2026). Everything in this section, the land, the emissions, the cleared forest, is the infrastructure required to keep that throughput going.

Cattle are the single largest commodity behind the carbon emissions from tropical deforestation tied to expanding farms and plantations, and cattle and oilseeds together account for more than half of them (Pendrill et al. 2019). Two-thirds of the deforestation done for agriculture is done to feed livestock, through soy, maize and pasture (Poore & Nemecek 2018). The soy in that sentence is feed, not tofu.

This also answers an argument that gets made in the other direction: that harvesting crops kills animals too, so vegans are not innocent either. They are not, and nobody serious claims they are. But the arithmetic runs the wrong way for the objection. Most crops are grown to feed livestock, so eating animals means more acres harvested, not fewer, the combine runs either way, and it runs more often for meat.

Three concessions, because the counterarguments here are stronger than the ones in nutrition.

The first is that some rangeland cannot grow crops, so grazing animals convert otherwise-unusable land into food. This is the best argument livestock has, and it is true, but it is smaller than it sounds, and it rests on an assumption worth examining.

Start with the numbers, taken from the analysis livestock advocates cite most. By dry weight, 86% of what livestock eat is inedible to humans. By dry weight is doing the work: a tonne of straw counts the same as a tonne of grain. Measured in the currencies that actually feed people, feed crops take 36% of the world’s crop calories and 53% of its plant protein, and livestock consume roughly a third of global cereal production (Cassidy et al. 2013; Mottet et al. 2017). The same analysis reports 2.8 kg of human-edible feed per kilogram of boneless ruminant meat. And the “only 14%” that is human-edible amounts to something like 840 million tonnes a year, on the order of the entire global wheat harvest.

Then the land. By that study’s own accounting, 57% of the land used for feed production is unsuitable for growing food, which means 43% is suitable, and of the two billion hectares of grazed grassland, roughly 700 million hectares could physically be cropland. An area about the size of Australia.

But the deeper problem with the marginal-land argument is what it assumes: that the only alternative use for land is farming it. Land that cannot grow wheat can grow forest, store carbon, and hold species. Freed land does not need to grow crops. It needs to grow ecosystems.

The second concession is regenerative grazing: the case that well-managed cattle build soil carbon and can be climate-neutral or better. Serious people make it. The most thorough independent assessment concluded that grazing does sequester carbon, that the effect is small, time-limited and reversible, and that it is outweighed by the emissions of the animals doing the grazing (Garnett et al. 2017). Grass-fed cattle remain net contributors.

There is also a case study that tests the claim directly. Hart Mountain National Antelope Refuge in Oregon removed all cattle in 1990 and did almost nothing else, no reseeding, no engineered stream repair. Researchers compared 64 matched photographs of refuge streams from before removal against retakes 23 years later. Bare soil fell 90%. Eroding channel fell 63%. Willow cover rose 388%. Aspen recruitment rose more than tenfold, filling the missing age class a century of grazing had eaten. Bird abundance rose about a third, with ground- and understory-nesting species more than doubling, while the same species declined regionally (Batchelor et al. 2015, as summarized in Rewilding the Plate). That post also recounts what the refuge saw beyond the photographs: pronghorn climbing from summer counts in the low hundreds into the thousands, and springs that staff remembered as dry running again.

The refuge did this by buying grazing permits rather than confiscating them, which is what a just transition looks like in practice, one buyout and one fence at a time.

The third concession is that not every plant food is low-impact, and here the picture is more interesting than either side usually admits. Air-freighted produce carries a heavy footprint. So do water-intensive crops in the wrong places, and pesticide load varies enormously between crops and production systems. And heated, climate-controlled greenhouse production is genuinely carbon-expensive: a systematic review of fifty tomato life-cycle studies found median emissions around 80 kg CO₂-equivalent per tonne for open-field production against around 1,709 for climate-controlled protected production (Zhu & Miller 2025). About twenty times worse.

That is the carbon column, and the same review shows where the penalty comes from. Unheated, climate-uncontrolled greenhouses came in around 83, almost exactly the open-field figure. The cost is the heating and lighting, not the glass, and the review found that locally grown heated-greenhouse tomatoes only beat imported field-grown ones when the heat comes from extremely low-carbon sources (Zhu & Miller 2025).

Staple calories, wheat, rice, maize and pulses, come from fields, and that is where the land question is actually decided.

So the honest version is not “greenhouses good, fields bad” or the reverse. It is that produce can be grown under glass without a large carbon penalty when the energy is clean, staples come from open ground, and neither of those is the thing consuming 83% of the world’s farmland.

One more objection deserves an answer, since it comes up constantly: you cannot farm without animals, because crops need manure. Farms that run entirely without animal inputs have existed for decades. The flagship is Tolhurst Organic in England, certified stockfree, which, by the farm’s own account as summarized in Rewilding the Plate, produces on the order of 120 tonnes of vegetables a year on about 20 acres of low-grade land, with fertility built from green-manure rotations, cover crops and woodchip compost from its own hedges. One farm is not a global agricultural system, and I am not going to pretend otherwise. But the claim that it is impossible is answered.

What the land could do instead is the last piece. Modeling a complete global shift to plant-based diets by 2050 estimates that vegetation returning to the freed farmland could sequester 332 to 547 gigatonnes of CO₂, on the order of 99 to 163% of the remaining carbon budget for 1.5 °C (Hayek et al. 2021). That is a projection under a scenario nobody expects to happen in full. It is worth stating anyway, because it sets the order of magnitude of what is on the table.

Two ledgers, pointing the same way. Which returns us to the question the first one refused to answer.

The wager

Grant every uncertainty in this essay, by name.

There is no randomized trial of a vegan diet measuring heart attacks or deaths. The direction of the stroke association is unresolved. Part of the fracture signal is unexplained. The colorectal finding is fragile but real. The cohort effect sizes sit inside the range residual confounding produces, and the headline heart-disease benefit attenuates toward the null under full adjustment. The debate over whether the food matrix of fermented dairy changes how its saturated fat behaves is live. The sequestration figures are models.

Grant all of it. Here is what remains.

The direction survives because it rests on no single study. It rests on a causal mechanism established through genetics and drug trials, a substitution analysis, a gradient inside one cohort, a comparison that strips away the macronutrient label, a trial in identical twins, a metabolic ward, a formal carcinogen review, and the structural fact that fiber exists in one kingdom and not the other. Those designs fail differently and agree anyway.

Stated precisely: shifting your diet toward whole plant foods and away from red and processed meat is well supported. That claim is robust and it does most of the work.

A well-planned vegan diet is a defensible endpoint of that gradient, conditional on B12 without exception, and on attention to calcium, vitamin D, iodine and protein. The support is real but weaker, and it takes the form of an absence: nothing marks a floor below which adding animal food starts helping.

What the evidence does not support is a scientific mandate that everyone must be vegan.

That distinction matters for how the wager is priced, so let me price both sides properly rather than quietly transferring the cheaper cost to the stronger claim.

If you shift toward plants and the nutrition science turns out to have been overstated, you have eaten more legumes and vegetables than you strictly needed to. There is no plausible downside there at all.

If you go fully plant-based and the science turns out overstated, the cost is more than a supplement, it includes the unresolved findings from three sections ago, the ones I said don’t resolve. That cost is manageable with planning and monitoring, and it is not nothing. Anyone telling you it is nothing hasn’t read their own citations.

If we keep expanding animal agriculture and the environmental accounting turns out to have been right, the cost is cleared forest, emitted carbon and species that do not come back.

Those last two are not the same kind of cost, and I should say so rather than let the sentence structure imply it. One is a health risk to an individual, manageable and reversible. The other is a collective loss that is neither. That asymmetry is exactly why the environmental ledger carries more weight in this decision than the nutritional one, and why it is the ledger that survives every uncertainty in the nutrition literature intact.

Something still has to name the stopping point on that gradient, and this is where I said I would come back.

For me it is this. If the health evidence gives a direction and no floor, and the environmental ledger pushes the same way with far greater confidence, then the remaining question is not scientific. It is about what you are willing to have done on your behalf, for a benefit the evidence cannot demonstrate. That is an ethical argument, doing exactly the work the evidence can’t, and it is a different essay. I am not asking it to prove anything. I am telling you where it enters, and that everything before this paragraph stands without it.

What comes out of our mouths

There is a third thing in the title, and it is the one most likely to be delivered badly.

Demand shapes production. That is how agriculture works: what people buy determines what gets planted, cleared, subsidized and built. The 83% of farmland is not a fact about cows. It is a fact about orders.

So advocacy matters. What follows rests on a thinner kind of evidence than the rest of this essay, surveys of people who stopped and what they say about why, collected in the post linked below, and you should weigh it accordingly.

Most people who stop eating animals go back. The reasons they give are overwhelmingly practical and social rather than moral. They were tired and hungry, because nobody told them plant foods are less calorie-dense and they needed to eat more volume. They had no idea what to cook. They were the difficult one at every meal. The identity became exhausting to defend. And a version of this comes up often enough to sit with: a vegan is the reason I stopped listening.

If that is even roughly right, it should change how the case gets made. The person who quietly cooks a good meal for six people has probably moved more than the person who won an argument. Being right and being persuasive are different skills, and we have been optimizing for the wrong one.

The useful version of “what comes out of our mouths” is not volume. It is accuracy about the science, including the parts that cut against you, and enough humility to make the thing easy for someone else to try.

Fields, not feedlots

I opened with George Mann because of what he did, not what he concluded. He took money from the meat board, published a paper the meat board liked, and then went and collected the autopsy data that undercut the use his paper was being put to. He published that too.

That is the standard this essay has tried to meet. It is why the section on grandchildren cut its own best claim, why the fracture signal is left standing unexplained, and why the closing image puts fields first and lets greenhouses in only on clean energy.

What survives is a direction, and a choice about where to stop along it that the evidence hands back to you.

The inheritance is real. It just isn’t in the genome. It is in the soil, the atmosphere and the list of species still alive when your grandchildren are old enough to notice. That channel is the one we can measure, and the one we know works.

Here is the whole land question in one comparison, worked out in full in Two Acres: What Beef Produces vs. What Plants Can, which uses USDA potato yields and a cow-calf pair on good pasture. Two acres of pasture will raise a calf to weaning, around half a million calories of beef, after the pair has eaten twenty million or more calories of forage and feed getting there. Two acres of potatoes, at ordinary American yields, produce something like 32 million calories. Sixty times the food from the same ground.

That ratio is why roughly four-fifths of the world’s farmland produces under a fifth of our calories and a bit over a third of our protein. We route the food through an animal first.

Potatoes are an illustration, not a proposal. You cannot grow them on the same field forever, and calories are not a diet, so rotate, which is better for the soil anyway.

The alternative is not exotic. It is growing food instead of feed on the land we already farm, and letting a great deal of the rest come back.

So picture the two futures honestly, because we are building one of them either way.

One runs on sheds and slaughterhouses. It needs four-fifths of the world’s farmland to deliver under a fifth of its calories, it drives most of the deforestation done for agriculture, and in 2024 alone it sent nearly 88 billion land animals through buildings most of its customers would rather not see. It works. It is not going to collapse on its own. It simply costs what it costs.

The other runs on fields, orchards and greenhouses. It grows food for people instead of feed for animals, on land we are already farming, and it hands back the rest. We know what that handing back looks like, because a refuge in Oregon did it and the willows came back at 388 percent and the springs started running.

The evidence in this essay does not force you to pick the second one. It gives a direction and declines to name a stopping point, and I have tried not to pretend otherwise anywhere in this essay. But it does make the choice legible, and it makes one thing plain: the future is not something that arrives. It is something that gets ordered.

Fields and greenhouses, or sheds and slaughterhouses.

We are choosing between them three times a day.

Further reading

What does evolution actually say about what humans should eat?

• How humans evolved to eat. The fossil, dental, isotopic and genetic record behind the first three sections here.
• The ancestral diet argument. What the Inuit and the Maasai actually prove, and the five errors the argument makes in both of its usual forms.

Is a plant-based diet actually healthier, and how do we know?

• Suppose the science isn’t settled. The short version of this whole argument: even granting every uncertainty, the choice still points one way.
• The full evidence review. Forty-eight studies, screened for peer review, journal quality, funding and retraction status, then sorted by strength of design.
• What the longest-living people actually eat. The longevity evidence climbed rung by rung, from population patterns to identical-twin trials.
• What the Adventist Health Studies tell us. The cohort behind most of the diet gradient in this essay, and what it can and cannot show.
• Are the Blue Zones debunked?. Why this essay leans on cohorts rather than demography, and what the demographic critique gets right.
• Saul Newman versus the Blue Zones, a fact-check. The specific critique that gets cited, checked claim by claim.
• Plants over animals. Ten findings, tier by tier: how a well-planned plant-based diet covers every nutrient, how plant antioxidants reach your arteries, and what animal products cost beyond your plate.

What about the nutrients people worry about on a vegan diet?

• The Great B12 Gamble. The one non-negotiable supplement, why it exists, and what deficiency actually does. Behind most of the adverse findings in this essay.
• Essential nutrients. What essential actually means in nutrition, where essential nutrients come from in the first place, and the fiber that animal foods lack.
• The full protein breakdown. What protein is, the complete-protein myth, how much you need, kidneys, and building muscle on plants.
• Do vegans need DHA and EPA?. The short answer first, then how the body converts ALA to DHA, the Omega-3 Index, and eight experts who disagree.
• Do vegans have weaker bones?. The fracture evidence, whether the studies controlled for confounding, and the calcium and exercise playbooks.
• Raising children vegan. Developmental nutrition, what the pediatric bodies say, and what has to be planned.

Does saturated fat raise cholesterol and heart disease risk?

• Cholesterol, from the ground up. Where cholesterol comes from, how eating it changes your blood, and how plaque actually forms.
• Animal fat, plant fat, and your arteries. Two kingdoms, two kinds of fat: which saturated fats raise cholesterol most, and one fatty meal followed from fork to bloodstream.
• Coconut oil versus animal fat. Butter, tallow, lard and coconut ranked by what they do to LDL, with the food-matrix argument handled fairly.
• Saturated fat and seed oils. How the hierarchy of evidence works, why not all saturated fat is the same, and what each level of evidence says.

What about carbohydrates, weight and the other common objections?

• Carbohydrates are a package deal. What a carbohydrate is, why you can digest starch but not fiber, and the journey from mouth to colon.
• Losing weight on a plant-based diet. Why energy density lets people eat until full and still lose weight.
• The anti-inflammatory diet. A meal followed from mouth to mitochondria, why animal foods stoke inflammation, and what blood tests and trials show.
• Plant toxins and antinutrients. Why the fear of plant toxins and antinutrients does not survive the evidence, why variety matters, and when people need individual help.
• Raw and cooked. What the stove unlocks, what the cutting board protects, and what happens to people on the extremes.
• The real story on diet and bowel cancer. Debunking the vegan scare: why a viral 40 percent figure rests on ninety-three cases. Then what the best evidence says to eat to lower bowel cancer risk, why screening matters most, and what to do after a diagnosis.
• Did cooked meat give us bigger brains?. The brain-size argument, and what the fossil and genetic record supports.
• The Ken Berry fact-check. The specific claims on lipids, insulin, DHA and diabetes reversal, answered against the published evidence.

How much land and how many animals does animal agriculture cost?

• Rewilding the Plate. What a billion freed hectares is worth, and what happened at Hart Mountain when the cattle came off.
• The 86% meme, debunked. The feed-versus-food statistic taken apart, including the 700 million hectares of grazed land that could grow crops.
• Two acres: what beef produces versus what plants can. The same two acres as beef and as potatoes, the protein and B12 math, the health evidence, what happens at scale, and why regenerative grazing does not close the gap.
• Do vegans kill more animals?. Nobody has a reliable count, and the arithmetic still runs the wrong way for the objection.
• The pesticide argument. Where pesticide and persistent-pollutant exposure actually concentrates in the food chain.
• Bees, honey and pollination. The edge case that gets raised whenever the land argument comes up.

Why do people stop eating plant-based, and what helps?

• Why people relapse, and what actually helps. Cravings, energy, protein worries, social pressure, and why the advocacy usually backfires. The material behind the advocacy section of this essay.

References

Populations, evolution and genetics

1. Mann GV, Shaffer RD, Anderson RS, Sandstead HH. Cardiovascular disease in the Masai. J Atheroscler Res. 1964;4:289–312. https://doi.org/10.1016/S0368-1319(64)80041-7
2. Mann GV, Spoerry A, Gray M, Jarashow D. Atherosclerosis in the Masai. Am J Epidemiol. 1972;95(1):26–37. https://academic.oup.com/aje/article-abstract/95/1/26/167903
3. Thompson RC, Allam AH, Lombardi GP, et al. Atherosclerosis across 4000 years of human history: the Horus study of four ancient populations. Lancet. 2013;381:1211–1222. https://doi.org/10.1016/S0140-6736(13)60598-X
4. Kaplan H, Thompson RC, Trumble BC, et al. Coronary atherosclerosis in indigenous South American Tsimane. Lancet. 2017;389:1730–1739. https://doi.org/10.1016/S0140-6736(17)30752-3
5. Tishkoff SA, Reed FA, Ranciaro A, et al. Convergent adaptation of human lactase persistence in Africa and Europe. Nat Genet. 2007;39:31–40. https://doi.org/10.1038/ng1946
6. Schlebusch CM, Sjödin P, Skoglund P, Jakobsson M. Stronger signal of recent selection for lactase persistence in Maasai than in Europeans. Eur J Hum Genet. 2013;21:550–553. https://doi.org/10.1038/ejhg.2012.199
7. Clemente FJ, Cardona A, Inchley CE, et al. A selective sweep on a deleterious mutation in CPT1A in Arctic populations. Am J Hum Genet. 2014;95:584–589. https://doi.org/10.1016/j.ajhg.2014.09.016
8. Gessner BD, Wood T, Johnson MA, Richards CS, Koeller DM. Evidence for an association between infant mortality and homozygosity for the Arctic variant of carnitine palmitoyltransferase 1A. Genet Med. 2016;18:933–939. https://doi.org/10.1038/gim.2015.197
9. Fumagalli M, Moltke I, Grarup N, et al. Greenlandic Inuit show genetic signatures of diet and climate adaptation. Science. 2015;349:1343–1347. https://doi.org/10.1126/science.aab2319
10. Buckley MT, Racimo F, Allentoft ME, et al. Selection in Europeans on fatty acid desaturases associated with dietary changes. Mol Biol Evol. 2017;34:1307–1318. https://doi.org/10.1093/molbev/msx103
11. Yilmaz F, Karageorgiou C, Kim K, et al. Reconstruction of the human amylase locus reveals ancient duplications seeding modern-day variation. Science. 2024;386:eadn0609. https://doi.org/10.1126/science.adn0609
12. Bolognini D, et al. Recurrent evolution and selection shape structural diversity at the amylase locus. Nature. 2024;634:617-625. https://doi.org/10.1038/s41586-024-07911-1
13. Speth JD, Spielmann KA. Energy source, protein metabolism, and hunter-gatherer subsistence strategies. J Anthropol Archaeol. 1983;2(1):1-31. https://doi.org/10.1016/0278-4165(83)90006-5
14. Bilsborough S, Mann N. A review of issues of dietary protein intake in humans. Int J Sport Nutr Exerc Metab. 2006;16(2):129-152. https://doi.org/10.1123/ijsnem.16.2.129
15. Pontzer H, Wood BM. Effects of evolution, ecology, and economy on human diet: insights from hunter-gatherers and other small-scale societies. Annu Rev Nutr. 2021;41:363–385. https://doi.org/10.1146/annurev-nutr-111120-105520
16. Williams GC. Pleiotropy, natural selection, and the evolution of senescence. Evolution. 1957;11:398–411. https://doi.org/10.2307/2406060
17. Milot E, Moreau C, Gagnon A, et al. Mother’s curse neutralizes natural selection against a human genetic disease over three centuries. Nat Ecol Evol. 2017;1:1400–1406. https://doi.org/10.1038/s41559-017-0276-6

Nutrition transitions and natural experiments

1. O’Dea K. Marked improvement in carbohydrate and lipid metabolism in diabetic Australian Aborigines after temporary reversion to traditional lifestyle. Diabetes. 1984;33:596–603. https://doi.org/10.2337/diab.33.6.596
2. McMurry MP, Cerqueira MT, Connor SL, Connor WE. Changes in lipid and lipoprotein levels and body weight in Tarahumara Indians after consumption of an affluent diet. N Engl J Med. 1991;325:1704–1708. https://doi.org/10.1056/NEJM199112123252405
3. Schulz LO, Bennett PH, Ravussin E, et al. Effects of traditional and Western environments on prevalence of type 2 diabetes in Pima Indians in Mexico and the U.S. Diabetes Care. 2006;29:1866–1871. https://doi.org/10.2337/dc06-0138
4. Marmot MG, Syme SL. Acculturation and coronary heart disease in Japanese-Americans. Am J Epidemiol. 1976;104:225–247. https://doi.org/10.1093/oxfordjournals.aje.a112296
5. Pressler M, Devinsky J, Duster M, et al. Dietary transitions and health outcomes in four populations: a systematic review. Front Nutr. 2022;9:748305. https://doi.org/10.3389/fnut.2022.748305

Diet and disease

1. Ference BA, Ginsberg HN, Graham I, et al. Low-density lipoproteins cause atherosclerotic cardiovascular disease. Eur Heart J. 2017;38:2459–2472. https://doi.org/10.1093/eurheartj/ehx144
2. Bonanome A, Grundy SM. Effect of dietary stearic acid on plasma cholesterol and lipoprotein levels. N Engl J Med. 1988;318:1244–1248. https://doi.org/10.1056/NEJM198805123181905
3. Mensink RP, Zock PL, Kester ADM, Katan MB. Effects of dietary fatty acids and carbohydrates on the ratio of serum total to HDL cholesterol: a meta-analysis of 60 controlled trials. Am J Clin Nutr. 2003;77:1146–1155. https://doi.org/10.1093/ajcn/77.5.1146
4. Neuenschwander M, Stadelmaier J, Eble J, et al. Substitution of animal-based with plant-based foods on cardiometabolic health and all-cause mortality. BMC Med. 2023;21:404. https://doi.org/10.1186/s12916-023-03093-1
5. Tonstad S, Stewart K, Oda K, et al. Vegetarian diets and incidence of diabetes in the Adventist Health Study-2. Nutr Metab Cardiovasc Dis. 2013;23:292–299. https://doi.org/10.1016/j.numecd.2011.07.004
6. Fung TT, van Dam RM, Hankinson SE, et al. Low-carbohydrate diets and all-cause and cause-specific mortality: two cohort studies. Ann Intern Med. 2010;153:289–298. https://doi.org/10.7326/0003-4819-153-5-201009070-00003
7. Seidelmann SB, Claggett B, Cheng S, et al. Dietary carbohydrate intake and mortality: a prospective cohort study and meta-analysis. Lancet Public Health. 2018;3:e419–e428. https://doi.org/10.1016/S2468-2667(18)30135-X
8. Landry MJ, Ward CP, Cunanan KM, et al. Cardiometabolic effects of omnivorous vs vegan diets in identical twins: a randomized clinical trial. JAMA Netw Open. 2023;6:e2344457. https://doi.org/10.1001/jamanetworkopen.2023.44457
9. Hall KD, Guo J, Courville AB, et al. Effect of a plant-based, low-fat diet versus an animal-based, ketogenic diet on ad libitum energy intake. Nat Med. 2021;27:344–353. https://doi.org/10.1038/s41591-020-01209-1
10. Bouvard V, Loomis D, Guyton KZ, et al. Carcinogenicity of consumption of red and processed meat. Lancet Oncol. 2015;16:1599–1600. https://doi.org/10.1016/S1470-2045(15)00444-1
11. Yun Z, et al. Dietary factors and colorectal cancer: a Mendelian randomization study. Front Nutr. 2023;10:1078963. https://doi.org/10.3389/fnut.2023.1078963
12. Reynolds A, Mann J, Cummings J, et al. Carbohydrate quality and human health: a series of systematic reviews and meta-analyses. Lancet. 2019;393:434–445. https://doi.org/10.1016/S0140-6736(18)31809-9

Countersignals and limitations

1. Tong TYN, Appleby PN, Bradbury KE, et al. Risks of ischemic heart disease and stroke in meat eaters, fish eaters, and vegetarians (EPIC-Oxford). BMJ. 2019;366:l4897. https://doi.org/10.1136/bmj.l4897
2. Gilsing AMJ, Crowe FL, Lloyd-Wright Z, et al. Serum concentrations of vitamin B12 and folate in British male omnivores, vegetarians and vegans. Eur J Clin Nutr. 2010;64:933–939. https://pubmed.ncbi.nlm.nih.gov/20648045/
3. Chiu THT, Chang HR, Wang LY, et al. Vegetarian diet and incidence of total, ischemic, and hemorrhagic stroke in 2 cohorts in Taiwan. Neurology. 2020;94:e1112–e1121. https://pubmed.ncbi.nlm.nih.gov/32102976/
4. Tong TYN, Appleby PN, Armstrong MEG, et al. Vegetarian and vegan diets and risks of total and site-specific fractures (EPIC-Oxford). BMC Med. 2020;18:353. https://doi.org/10.1186/s12916-020-01815-3
5. Appleby P, Roddam A, Allen N, Key T. Comparative fracture risk in vegetarians and nonvegetarians in EPIC-Oxford. Eur J Clin Nutr. 2007;61:1400–1406. https://pubmed.ncbi.nlm.nih.gov/17299475/
6. Thorpe DL, Beeson WL, Knutsen R, Fraser GE, Knutsen SF. Dietary patterns and hip fracture in the Adventist Health Study 2. Am J Clin Nutr. 2021;114:488–495. https://pubmed.ncbi.nlm.nih.gov/33964850/
7. Ballarin RS, Lazzarin T, Minicucci MF, Azevedo PS, et al. Vegetarian and vegan diets and the risk of hip fracture in adults: a systematic review and meta-analysis. Oxford University Press; 2025. Full text (UNESP repository)
8. Wakolbinger-Habel R, Reinweber M, König J, et al. Self-reported resistance training is associated with better HR-pQCT-derived bone microarchitecture in vegan people. J Clin Endocrinol Metab. 2022. https://pubmed.ncbi.nlm.nih.gov/35924941/
9. Dunneram Y, Lee JY, Watling CZ, et al. Vegetarian diets and cancer risk: pooled analysis of 1.8 million women and men in nine prospective studies on three continents. Br J Cancer. 2026;134:1218-1229. https://doi.org/10.1038/s41416-025-03327-4 Funded by Cancer Research UK and the World Cancer Research Fund.
10. Abris GP, Fraser GE, Knutsen SF, et al. Cause-specific and all-cause mortalities in vegetarian compared with those in nonvegetarian participants from the Adventist Health Study-2 cohort. Am J Clin Nutr. 2024;120(4):907–917. https://pubmed.ncbi.nlm.nih.gov/39098708/
11. Rees K, Al-Khudairy L, Takeda A, Stranges S. Vegan dietary pattern for the primary and secondary prevention of cardiovascular diseases. Cochrane Database Syst Rev. 2021;2:CD013501. https://doi.org/10.1002/14651858.CD013501.pub2
12. Estruch R, Ros E, Salas-Salvadó J, et al. Primary prevention of cardiovascular disease with a Mediterranean diet supplemented with extra-virgin olive oil or nuts. N Engl J Med. 2018;378:e34. https://doi.org/10.1056/NEJMoa1800389
13. de Lorgeril M, Salen P, Martin JL, Monjaud I, Delaye J, Mamelle N. Mediterranean diet, traditional risk factors, and the rate of cardiovascular complications after myocardial infarction: final report of the Lyon Diet Heart Study. Circulation. 1999;99(6):779-785. https://doi.org/10.1161/01.CIR.99.6.779
14. Levine ME, Suarez JA, Brandhorst S, et al. Low protein intake is associated with a major reduction in IGF-1, cancer, and overall mortality in the 65 and younger but not older population. Cell Metab. 2014;19:407–417. https://doi.org/10.1016/j.cmet.2014.02.006
15. Hopwood CJ, Bleidorn W, Schwaba T, Chen S. Health, environmental, and animal rights motives for vegetarian eating. PLOS ONE. 2020;15:e0230609. https://doi.org/10.1371/journal.pone.0230609

Inheritance and developmental programming

1. Baker TG. A quantitative and cytological study of germ cells in human ovaries. Proc R Soc B. 1963;158(972):417-433. https://doi.org/10.1098/rspb.1963.0055
2. Heijmans BT, Tobi EW, Stein AD, et al. Persistent epigenetic differences associated with prenatal exposure to famine in humans. PNAS. 2008;105:17046–17049. https://doi.org/10.1073/pnas.0806560105
3. Donkin I, Versteyhe S, Ingerslev LR, et al. Obesity and bariatric surgery drive epigenetic variation of spermatozoa in humans. Cell Metab. 2016;23:369–378. https://doi.org/10.1016/j.cmet.2015.11.004
4. Vågerö D, Pinger PR, Aronsson V, van den Berg GJ. Paternal grandfather’s access to food predicts all-cause and cancer mortality in grandsons. Nat Commun. 2018;9:5124. https://doi.org/10.1038/s41467-018-07617-9
5. Horsthemke B. A critical view on transgenerational epigenetic inheritance in humans. Nat Commun. 2018;9:2973. https://doi.org/10.1038/s41467-018-05445-5

Environment and agriculture

1. Mottet A, de Haan C, Falcucci A, et al. Livestock: on our plates or eating at our table? A new analysis of the feed/food debate. Glob Food Sec. 2017;14:1–8. https://doi.org/10.1016/j.gfs.2017.01.001
2. Cassidy ES, West PC, Gerber JS, Foley JA. Redefining agricultural yields: from tonnes to people nourished per hectare. Environ Res Lett. 2013;8:034015. https://doi.org/10.1088/1748-9326/8/3/034015
3. Batchelor JL, Ripple WJ, Wilson TM, Painter LE. Restoration of riparian areas following the removal of cattle in the northwestern Great Basin. Environ Manage. 2015;55:930–942. https://doi.org/10.1007/s00267-015-0436-x
4. Food and Agriculture Organization of the United Nations. FAOSTAT livestock primary data, “Producing Animals/Slaughtered,” as compiled in Faunalytics, Global Animal Slaughter Statistics and Charts, April 2026 update. https://faunalytics.org/global-animal-slaughter-statistics-and-charts/
5. Poore J, Nemecek T. Reducing food’s environmental impacts through producers and consumers. Science. 2018;360:987–992. https://doi.org/10.1126/science.aaq0216 PMID 29853680. Erratum: Science. 2019;363(6429):eaaw9908, https://doi.org/10.1126/science.aaw9908; it revises a land-sequestration estimate upward and does not change the figures cited here.
6. Scarborough P, Clark M, Cobiac L, et al. Vegans, vegetarians, fish-eaters and meat-eaters in the UK show discrepant environmental impacts. Nat Food. 2023;4:565–574. https://doi.org/10.1038/s43016-023-00795-w
7. Pendrill F, Persson UM, Godar J, et al. Agricultural and forestry trade drives large share of tropical deforestation emissions. Glob Environ Change. 2019;56:1–10. https://doi.org/10.1016/j.gloenvcha.2019.03.002 Corrigendum: Glob Environ Change. 2020;63:102103, https://doi.org/10.1016/j.gloenvcha.2020.102103; it changes trade shares, not the commodity results cited here.
8. Garnett T, Godde C, Muller A, et al. Grazed and Confused? Food Climate Research Network, University of Oxford, 2017. https://edepot.wur.nl/427016
9. Zhu Z, Miller SA. Greenhouse gas emissions of tomato production and supply: a systematic review. Resour Conserv Recycl. 2025;218:108236. https://doi.org/10.1016/j.resconrec.2025.108236
10. Hayek MN, Harwatt H, Ripple WJ, Mueller ND. The carbon opportunity cost of animal-sourced food production on land. Nat Sustain. 2021;4:21–24. https://doi.org/10.1038/s41893-020-00603-4

This is health information, not medical advice. Anyone with questions about their own diet should consult a physician or registered dietitian, ideally one who practices lifestyle medicine, since that field puts the evidence first and respects a person’s values.

Last updated: October 6, 2026, 3:02 PM ET