The 21-to-479 Leap: Ancient DNA Shows Human Evolution Sped Up After Farming

Before this month, the published ancient-DNA record contained roughly two dozen confirmed examples of directional selection acting on the human genome in the last ten thousand years. A single paper in Nature, released April 15, 2026, raises that count to 479. The scale of the jump is not incremental; it is a phase change in how much of recent human biology can now be read from the bones of people who lived through the dawn of agriculture, the Bronze Age, and the plagues that stitched the early medieval world together.

The paper, led by first author Ali Akbari and senior author David Reich at Harvard, assembles 10,016 newly sequenced ancient West Eurasian genomes alongside 5,820 previously published ancient sequences and 6,438 modern ones, according to the Broad Institute release accompanying publication. The new sample alone more than doubles the prior global catalog of high-quality ancient DNA, a consolidation that Reich summarized bluntly: "This single paper doubles the size of the ancient human DNA literature."

What the authors do with that data matters as much as the data itself. By pairing the expanded catalog with a computational method designed to strip out the confounding fingerprints of migration, population mixing, and genetic drift, the team isolates a pattern that previous studies could only hint at: across the last ten millennia, directional natural selection in West Eurasia did not run at a steady clock. It accelerated — and the acceleration tracks the transition to farming.

Why Scale Flips the Question

The hardest thing to detect in a population-genetic record is something small happening consistently. Individual allele frequencies wobble with every generation of reproduction and every wave of migrants arriving in a region. Most signals a researcher wants to see — the quiet upward drift of a variant that confers a slight survival or fertility advantage — are smaller than the noise.

Reich described the design constraint in the EurekAlert release: detecting frequency changes larger than chance alone would produce, over ten millennia, requires reading subtle effects — and reading subtle effects requires thousands of genomes distributed across that entire span.

That is why the 15,836 ancient genomes assembled here — a figure arrived at by adding the Reich Lab's 10,016 new sequences to the 5,820 already in the literature — change the question being asked. With roughly 16,000 individuals from a single broad region, each of them placed in archaeological time by carbon dating and context, it becomes statistically tractable to ask which variants rose faster than drift alone would predict, in which millennium, and in which direction.

The seven-year collection effort behind the new sequences involved more than 250 archaeologists and anthropologists, per the Broad release. That collaborator count is itself a statement about how ancient-DNA work has changed: the bottleneck is no longer the sequencing chemistry but the curated, contextualized skeletal archive.

The 21-to-479 Leap

The headline number deserves to be handled carefully. Prior to the new paper, roughly 21 confirmed instances of directional selection had been identified in the modern human lineage over the last ten thousand years, a figure reported across the Broad, EurekAlert, phys.org, and Earth.com coverage. The new study identifies 479 alleles that have been strongly selected for or against.

Underneath that number sits a broader tier of candidates: more than 7,600 genetic locations where the authors estimate a greater-than-50 percent probability that directional selection was at work, per the EurekAlert summary. Not all of those candidates would survive a stricter threshold, but the size of the pool tells you how much texture the field has been missing.

Akbari put the methodological shift in plain language: "With these new techniques and large amount of ancient genomic data, we can now watch how selection shaped biology in real time."

Note what that sentence is and is not claiming. It is not a claim that everyone's genome is now legible, or that any single variant causes any single trait in a deterministic sense. It is a claim that, at population scale and over deep time, the lifting and lowering of specific alleles can be tracked as it happens, generation after generation, rather than inferred retrospectively from modern-day frequencies alone.

What 479 Variants Actually Look Like

One of the paper's most useful contributions is the breadth of the trait landscape it touches. Across the Broad and phys.org writeups, the selected-variant list includes pigmentation traits such as light skin tone and red hair; digestive shifts including lactose tolerance after infancy and susceptibility to celiac and Crohn's disease; immune-system rewiring visible in variants linked to HIV infection immunity and leprosy resistance; cardiometabolic traits such as body mass index, body-fat percentage, and waist-to-hip ratio; blood group diversification across the A, B, and O systems; and mental-health signatures including reduced risk for bipolar disorder and schizophrenia.

More than 60 percent of the 479 variants flagged as strongly selected have documented links to present-day traits, per the phys.org coverage. That is a striking overlap with a catalog — the modern biobank-derived genome-wide association literature — that was built for an entirely different purpose.

The trait list is not random. It clusters around three domains that align with what archaeology already tells us about the last ten thousand years of West Eurasian life: what people ate, what they caught from their neighbors (human and animal), and how their metabolism coped with sedentary agrarian life. That clustering is the substantive claim the paper makes — not that selection happened (everyone knew that), but that its thematic signature is visible once the data is large enough.

Why the Farming Transition Shows Up

The interpretive spine of the paper is the observation that selection signals intensify after the Neolithic transition to agriculture. People shifted from a hunter-gatherer foraging diet to carbohydrate-rich cereal staples. They lived in close proximity to domesticated animals. They gathered in denser populations, which concentrated both pathogens and immune-relevant selective pressure. Their children encountered milk past weaning for the first time in species history.

It is not difficult, in principle, to see how each of those environmental changes would reshape the fitness landscape. What was previously missing was the empirical resolution to watch it happen. The phys.org writeup frames the shift in dietary and disease pressure as the proximate cause of intensified selection for blood, immune, inflammatory, and cardiometabolic traits during and after the Bronze Age.

This kind of mapping — from an archaeological process to a selection signature — is where ancient-DNA studies have always aimed, and where small sample sizes have always forced caution. The new paper does not resolve every case to a single mechanism. But by stacking hundreds of variants under a single, consistent statistical framework, it lets the thematic clustering speak.

The 2 Percent Problem

One of the quieter but more important numbers in the paper is the estimate that directional selection accounts for roughly 2 percent of the total gene-frequency change observed over the past ten millennia in West Eurasia, per the EurekAlert release. The other 98 percent is dominated by migration, population mixing, and neutral drift.

This is an important corrective to overheated framings of "accelerated human evolution." The genome is mostly still moving around because populations moved around. Farming-era West Eurasia absorbed an enormous sequence of migration pulses from Anatolian early farmers, Pontic-Caspian steppe pastoralists, and later movements that reshaped the allele frequencies of every settled community they touched. That noise is the majority of the signal.

What the paper establishes is that a minority of the signal is not noise. Two percent is not trivial when the substrate is every single genomic locus, because the 2 percent is concentrated in functionally consequential variants — the ones that change whether you can digest lactose, whether a particular virus can establish infection, whether your immune system over- or under-reacts to gluten. The selected minority is the part of the record that tells you what human life, in that environment, was rewarding or penalizing.

This Is Not Gattaca

The list of selected traits includes several that are socially loaded — intelligence-test scores, years of schooling, and household income appear in the trait-linkage catalog reported by the Broad and EurekAlert coverage. That phrasing requires careful reading.

What is under selection, in those cases, is not "intelligence" or "educational attainment" as any ordinary person would use those words. What is under selection is a statistical covariance between certain genetic variants and polygenic scores that were trained on those modern outcome measures in modern, socially stratified, educationally structured, heavily confounded populations. Polygenic scores for educational attainment are notorious for absorbing environmental confounds; they are not a direct window into any cognitive trait.

The ancient-DNA paper does not claim otherwise. It reports the co-movement. A responsible reading is that some variants whose frequencies have changed over ten millennia happen to also correlate, in modern biobank populations, with traits that are themselves proxies for complex, highly environmentally structured outcomes. That is a long chain, and nothing in the paper shortens it.

Prior Art and the Long Arc

This is not the first multi-thousand-genome ancient-DNA study of West Eurasian selection. An earlier 2022 analysis assembling around 1,000 ancient genomes across roughly 10,000 years of European prehistory established the field's prior benchmark for this kind of scan, and the Broad release itself frames the new work as building on a decade of effort since the recovery of the first genome-wide ancient data in 2010.

What has changed is roughly fifteenfold the sample size and, more importantly, a methodological innovation that Reich credits to his first author: "Ali developed a powerful technique that could zoom in on the patterns that actually mattered." The technique matters because ancient DNA's fundamental obstacle is not sequencing error; it is disentangling selection from the tangled background of human movement.

The narrative outcome — that selection sped up after agriculture — was already the field's working hypothesis. What is new is that the hypothesis is now anchored in a specific, numerate, trait-resolved, time-stamped atlas rather than a handful of canonical examples (lactase persistence, pigmentation loci, immune-adjacent variants) that everyone had been citing for years.

That distinction matters because a single vivid example can be explained away or reinterpreted. A catalog of nearly five hundred variants, each with a time window, a direction, and a trait association, is harder to dislodge. It is also harder to misread in a sensational direction: the individual variant stories are quieter than the aggregate, and the aggregate makes the claim that previously rested on five or six famous loci something that can be stress-tested across hundreds of independent data points. The field moves, in other words, from an anecdotal evidence base to a statistical one.

Implications for Medicine

Akbari raised, in the phys.org coverage, an implication with direct clinical resonance: editing away a variant that has been strongly selected for in humans is probably not a good idea, because something about that variant was rewarding enough to drive its frequency up over thousands of years.

The logic is worth spelling out. Gene therapy, base editing, and CRISPR-adjacent interventions operate on the assumption that one can identify a variant to edit away. Most of the targeting literature focuses on rare, large-effect disease alleles. But the ancient-DNA atlas now gives researchers a quantitative reason to ask, before editing any reasonably common variant, whether that variant has been recently selected for in humans — and, if so, under what environmental context.

A variant that rose in frequency under sustained selection is likely to have been coupled to some fitness-relevant function, even if its trait association in a modern biobank looks benign or undesirable. Editing such a variant away without understanding why it was selected is the kind of intervention whose off-target consequences are most likely to emerge only generations later. The atlas is, among other things, a map of which variants require that caution.

What Happens When This Method Goes Global

Reich has already flagged the geographic limit of the current paper. The 15,836-genome atlas is almost entirely West Eurasian. "To what extent will we see similar patterns in East Asia or East Africa or Native Americans?" he asked, in comments carried by phys.org and Earth.com.

The answer will probably be "similar patterns, different specifics." The thematic domains — diet, pathogens, sedentism — were global transitions, not West Eurasian peculiarities. But the specific variants that rose to prominence in, say, the Yellow River rice agricultural transition, or in the maize-intensive central Andes, or in the millet-and-sorghum belts of early Sahelian agropastoralism, were drawn from local starting frequencies and shaped by local pathogens. Extending the atlas to those regions is not a copy-paste exercise. It is a separate empirical effort, and it will require the same kind of multi-year archaeological collaboration that produced the current paper's 250-plus-contributor author list.

What This Does Not Tell Us — Yet

It is worth enumerating what the paper does not, on its own, resolve.

1. Geographic completeness. The atlas is West Eurasian. Non-European populations, including the majority of humanity alive today, are underrepresented. Any global claim about "how farming changed humans" inherits that sampling asymmetry.

2. Polygenic-score confounds. The correlations between selected variants and modern polygenic scores for socially-mediated outcomes (schooling, income) reflect the structure of modern societies as much as any biology. These are the weakest members of the trait-linkage catalog and should not be read as causal.

3. Variant-to-mechanism causality. A signal of directional selection tells you a variant rose faster than drift predicts. It does not tell you which phenotype was being rewarded, or in which generation the reward was strongest. The trait attributions in the paper are statistical associations with modern biobank data — informative, but not mechanistic.

4. Specific named loci in public coverage. The press releases accompanying the paper describe trait clusters rather than naming individual gene loci at resolution. Any reader who wants to know whether a specific variant of personal interest appears on the 479-allele list will need to consult the supplementary material of the primary paper.

5. Causal direction of the farming-selection link. That selection accelerated after farming is a correlation with a plausible mechanism. Distinguishing, among candidate mechanisms, which contributed most — diet, pathogens, sedentism, population density, or something else — will require follow-up work that integrates archaeological and epidemiological evidence more tightly than a single genomic paper can.

Key Takeaways

• A single paper led by Akbari and Reich (Broad release; EurekAlert) expands the ancient-DNA record from roughly two dozen documented directional-selection signals to 479, using 15,836 ancient West Eurasian genomes plus 6,438 modern sequences.
• Selection intensified after the Neolithic transition, concentrating in blood, immune, inflammatory, and cardiometabolic traits, as the phys.org coverage describes.
• Directional selection accounts for roughly 2 percent of total gene-frequency change over ten millennia (EurekAlert); migration and drift remain the dominant forces.
• More than 60 percent of the strongly-selected variants have documented links to present-day traits (phys.org), concentrated in diet, disease, and cardiometabolic domains.
• The clinical implication Akbari raises — that recently-selected variants deserve extra scrutiny before being edited — is the kind of second-order consequence that only becomes legible once an atlas this size exists.
• The work is West-Eurasia-specific; extending the method to East Asia, East Africa, and the Americas will require comparable archaeological infrastructure, and the patterns found there will not be identical.