Science

Veganism is a set of ethical commitments, but almost every question it raises — is this diet adequate, does it really lower emissions, do fish feel pain, can we grow meat without animals, will the world’s nitrogen economy survive a legume-heavy future — ends up in a laboratory, a field trial, or a model. The “science” pillar is where those questions get answered on their own terms, with the instruments and statistics each field actually uses.

This page is the trunk. It sketches the main scientific disciplines that touch plant-based living, names the landmark studies, and — just as importantly — flags where the evidence is thinner than popular accounts imply. Downstream pages go deep on specific nutrients, technologies, and methods.

How the evidence base is built

Nutrition, environmental, and welfare claims about veganism are supported by different classes of study, each with distinct strengths and blind spots.

Randomised controlled trials (RCTs) are the strongest design for causal questions — usually short, narrow, and focused on a specific outcome such as LDL cholesterol, blood pressure, or fasting glucose. They can answer “does this intervention move this marker?” but rarely run long enough to settle “does this diet prevent this disease over 30 years?”

Prospective cohort studies follow large groups over decades. The four most cited in plant-based nutrition are:

• EPIC-Oxford — around 65,000 UK participants including a large vegetarian and vegan subcohort, recruited specifically to enable diet-pattern comparisons (Key et al., 2009).
• Adventist Health Study-2 (AHS-2) — roughly 96,000 Seventh-day Adventists in North America, a population enriched for vegetarian and vegan dietary patterns and low in confounding behaviours like smoking (Orlich et al., 2013).
• Nurses’ Health Study / Health Professionals Follow-Up Study (NHS/HPFS) — two long-running US cohorts exceeding 200,000 participants combined, with repeated food-frequency questionnaires.
• UK Biobank — approximately 500,000 UK adults, with a one-time dietary assessment and a small self-identified vegan subgroup. Its vegan exposure is cross-sectional and comparator-dependent — a detail that matters when interpreting headline findings.

Cohorts can detect long-horizon associations RCTs cannot reach, but they are observational. Residual confounding — the suspicion that vegetarians and vegans also exercise more, smoke less, and visit doctors more often — is a permanent caveat.

Meta-analyses and systematic reviews pool effect estimates across studies. They are only as good as the input papers and the heterogeneity of the question; a meta-analysis of fifteen underpowered studies on incompatible diets will still produce a confident-looking number.

Life-cycle assessments (LCAs) quantify environmental impacts from cradle to grave. The dominant database in food-system LCA is the Poore & Nemecek (2018) synthesis of 570 studies covering 38,700 farms and 40 products. Results depend heavily on system boundaries, allocation rules (how impact is split between milk and beef in a dairy system, for example), and whether land-use change is included.

Ioannidis (2018) has argued forcefully that nutritional epidemiology has been under-rigorous for decades — small effects, shared confounders, and noisy food-frequency questionnaires generate more signal than the data actually supports. Reading the literature with that caveat in mind is not cynicism; it is method.

Nutrition biochemistry highlights

Protein complementation, the myth. Frances Moore Lappé’s 1971 Diet for a Small Planet popularised the idea that plant proteins had to be combined at the same meal to be “complete.” She retracted the claim in the 1981 edition. Young & Pellett (1994) gave the definitive biochemistry: the liver maintains a free amino acid pool sufficient to complement intakes across a 24-hour window. Mixing cereal and legume proteins matters at the daily scale, not the plate scale.

DIAAS vs PDCAAS. The FAO’s 2013 report replaced the 1991 PDCAAS protein quality score with DIAAS — Digestible Indispensable Amino Acid Score — which uses true ileal digestibility per amino acid and removes the artificial 1.0 cap PDCAAS imposed. Under DIAAS (Herreman et al., 2020) soy qualifies as high quality at 91; wheat, rice, and hemp fall below the 75 threshold for any protein-quality claim. Pea sits at 70. The hierarchy is real, and it is only visible once you stop capping.

Heme vs non-heme iron. Hurrell & Egli (2010) summarise four decades of work on iron bioavailability. Heme iron (animal flesh) is absorbed at 15–35 percent through a dedicated HCP1-like pathway regardless of body iron status. Non-heme iron (plants, fortified foods, dairy) is absorbed at 2–20 percent through DMT1, which the body upregulates when stores are low and downregulates when they are full. Plant-based eaters typically consume more total iron but store less — because the system is doing exactly what it evolved to do.

B12 is bacterial, not animal. Vitamin B12 (cobalamin) is synthesised only by certain bacteria and archaea. Ruminants harbour B12-producing microbes in the rumen; monogastric animals — including humans — do not. Watanabe (2014) and subsequent reviews establish that the B12 in animal foods originates from microbial synthesis upstream; the same synthesis can be done industrially in fermentation tanks, which is where supplement and fortification cobalamin actually comes from. Vegans supplement a bacterial metabolite directly rather than laundering it through an animal first.

ALA to EPA and DHA. Burdge & Wootton (2002) measured the conversion rate of alpha-linolenic acid (ALA, the short-chain omega-3 in flax, chia, and walnuts) to EPA and DHA in young women. Conversion to EPA ran around 21 percent; DHA around 9 percent — rates higher than the often- quoted “about 1 percent” figure. Conversion is lower in men and is suppressed by high linoleic acid intake. The implication for vegans is not that ALA is useless but that algal DHA supplements remain the simplest way to secure long-chain omega-3 status without relying on a metabolic step that is real but variable.

Animal cognition science

Whether non-human animals can suffer is a scientific question, not only a philosophical one — and the empirical literature has moved substantially in the past fifteen years.

The Cambridge Declaration on Consciousness (Low et al., 2012), signed by a group of neuroscientists at Cambridge, stated that non-human animals, including all mammals, birds, and many other creatures including octopuses, possess the neurological substrates that generate consciousness. It was a consensus statement rather than a primary study, but it crystallised a shift in what mainstream neuroscience was willing to say aloud.

The New York Declaration on Animal Consciousness (Andrews et al., 2024) extended that position, with broad signatory support, to include “a realistic possibility of conscious experience” in all vertebrates (reptiles, amphibians, fishes) and many invertebrates (cephalopods, decapod crustaceans, insects), and argued that it is irresponsible to ignore that possibility in decisions that affect them.

Jonathan Birch’s work on invertebrate sentience — a comprehensive UK-commissioned review in 2021 — catalogued the behavioural and neurological markers of pain in cephalopods and decapods that informed UK legislation recognising them as sentient beings under the Animal Welfare (Sentience) Act 2022. Lynne Sneddon’s 2015 Journal of Experimental Biology review synthesised evidence that fish possess nociceptors, exhibit pain-related behaviours, and alter those behaviours in response to analgesics — meeting standard criteria used to establish pain in mammals.

This does not settle every welfare question, but it removes the claim that “fish don’t feel pain” from the category of defensible scientific opinion.

Food systems science

Food systems produce somewhere around one-quarter to one-third of anthropogenic greenhouse-gas emissions, and the share attributable to animal products is the most consequential lever identified in the literature.

Poore & Nemecek (2018) assembled the largest meta-analysis of food LCAs to date — 570 studies, 38,700 farms, 40 products — and reported that moving from current diets to a plant-based diet could reduce food- related land use by about 76 percent and food-related greenhouse-gas emissions by about 49 percent, alongside reductions in freshwater withdrawals and eutrophication. These are population-level averages; individual products and production systems vary widely.

Clark et al. (2020), in Science, modelled that even if fossil fuels were eliminated immediately, emissions from the current food system alone would make the 1.5 and 2 °C Paris targets unreachable without food-system change. Dietary shift toward plant-based patterns was one of five interventions they identified as necessary.

Willett et al. (2019) — the EAT-Lancet Commission — proposed a “planetary health diet” with roughly 300 g/day of vegetables, 200 g of fruit, 232 g of whole grains, 75 g of legumes, and much smaller allowances of animal products than current Western norms. The diet is not vegan, but it reduces ruminant meat by roughly an order of magnitude relative to typical American intakes.

Xu et al. (2021), in Nature Food, estimated that animal-based foods account for roughly 57 percent of food-system greenhouse-gas emissions versus about 29 percent for plant-based foods — the first study to quantify this split at the global scale using a consistent accounting framework.

The IPCC AR6 Working Group III, Chapter 7 (2022), covering agriculture, forestry, and other land uses, identifies dietary change toward plant-based patterns as one of the highest-mitigation-potential demand-side options available, alongside reduced food waste.

Cellular agriculture and precision fermentation

Two technology families promise to deliver animal-protein products without animals.

Cultivated meat grows muscle and fat tissue from animal cells in bioreactors. Tuomisto & Teixeira de Mattos (2011) published the first serious LCA of the concept, suggesting substantial reductions in land, water, and GHG intensity under assumed scaled conditions. Humbird (2021) countered with a techno-economic analysis arguing that even at full industrial scale, bioreactor costs, growth medium, and contamination control would keep cultivated meat well above commodity meat prices without breakthrough advances in media formulation and cell-line performance. Both assessments rely on assumptions a decade ahead of demonstrated operation; reality is still being written.

Precision fermentation uses engineered microbes to produce specific animal proteins — whey (Perfect Day), ovalbumin (EVERY), casein, collagen — without animals. Unlike cultivated meat, it is already commercial: the same platform that has produced recombinant insulin and rennet (chymosin for cheesemaking) since the 1980s has been extended to dairy and egg proteins. Good Food Institute’s annual State of the Industry reports track commercial progress and remaining scientific bottlenecks, particularly in scaffolding, cell-line development, and food-grade growth medium.

The honest state of the science: precision fermentation of individual proteins is technically mature and cost-descending; cultivated whole- tissue meat remains a research-stage technology with unresolved cost and scale questions.

Plant breeding and agronomy

A plant-based food system is not just the current system minus animals. It requires different inputs, different rotations, and different breeding priorities.

Legume renaissance. Pulse crops — lentils, chickpeas, dry peas, faba beans, lupines — are nitrogen-fixing: in symbiosis with Rhizobium bacteria in their root nodules, they convert atmospheric N₂ into ammonia, supplying their own nitrogen and leaving residual nitrogen for the next crop in rotation. Peoples et al. (2009) estimated that globally, legume-rhizobium symbioses fix on the order of 50–70 million tonnes of nitrogen annually in agricultural systems — a substantial share of total reactive nitrogen input. Expanding legumes in both human and rotational roles displaces synthetic nitrogen fertiliser, the Haber-Bosch production of which is responsible for roughly 1–2 percent of global energy use.

Reduced-till and cover-cropping systems preserve soil organic carbon, reduce erosion, and lower fuel use. They pair naturally with legume rotations and are central to the emerging “regenerative plant agriculture” literature. Herrero et al. (2020) in Nature Food catalogue a broader innovation portfolio for sustainable food-system transition — from microbial inputs to improved crop genetics — that is largely independent of whether livestock remain in the system.

The upshot is that the agronomy of a plant-based future is not speculative. Most of the techniques exist and are deployed at scale; the question is how rapidly they can be expanded and how the incentive structure rewards them.

How to read a nutrition study

The last thing a science pillar should provide is a set of reflexes for reading the literature without being misled by headlines.

• Confounding and healthy-user bias. Vegetarians and vegans in Western cohorts systematically differ from the comparator population on exercise, smoking, alcohol, and healthcare use. When a study reports that vegans have lower cardiovascular mortality, ask how much of that association survives adjustment for those covariates.

• Specify the comparator. In UK Biobank, pooled “vegetarians” and “vegans” produce different effect estimates than vegans alone against regular meat-eaters. Headline claims often smuggle in the wider category. Insist on the exact exposure and the exact referent.

• Food-frequency questionnaires (FFQs) are noisy instruments. An FFQ asked once in a lifetime cannot capture decades of dietary variation. Effect sizes for small differences in intake should be treated with corresponding humility.

• Ecological fallacy. Country-level correlations — average meat intake and average life expectancy — cannot be translated to individual-level causation without further evidence.

• Absolute versus relative risk. A 30 percent relative reduction in a rare outcome may be a 0.3 percentage-point absolute reduction. Both numbers are true; only one is useful for decisions.

• Verify citations against primaries. WebFetch summaries of papers paraphrase. Effect sizes, confidence intervals, and comparator definitions should be read in the abstract or full text, not in a downstream summary. Mismatches — including DOI-year divergence — are fabrication signals.

• Pre-registration and replication. Pre-registered analyses and replicated effects are more trustworthy than novel findings from small cohorts. Nutritional epidemiology is still catching up on this front.

What this pillar covers

The science pillar branches in roughly six directions. Downstream pages deepen each.

• Nutrition evidence methods — how RCTs, cohorts, meta-analyses, and Mendelian randomisation studies apply to plant-based diets, and where each has failed.
• Cellular agriculture — cell lines, bioreactors, scaffolds, growth media, techno-economic ceilings, and the LCA debate.
• Precision fermentation — recombinant dairy, egg, and collagen proteins; strain engineering; scale economics; regulatory pathways.
• Cognition science — sentience criteria across vertebrates and invertebrates, neurobiological markers, and the policy implications of the Cambridge and New York declarations.
• LCA methodology — system boundaries, allocation rules, land-use change accounting, and why plant-based LCAs still produce a wide range of estimates.
• Food systems modelling — integrated assessment models, EAT-Lancet and beyond, dietary scenarios against climate and land budgets.

Veganism does not need scientific perfection to be defensible. It needs the same honest relationship with evidence any serious claim does: stating what is known, naming what is not, and refusing to overreach. The rest of this pillar is an effort to hold that line.