So Autism is Genetic: What Does That Settle?

Population as Organism: Epigenetics, Neurodiversity, and the Problem of the Individual

The Western neuroscience community has recently taken to saying that autism is genetic — a corrective, and a welcome one, to public claims, made from positions of real institutional power, that environmental exposure is the primary driver of rising diagnosis rates. The rebuttal leans on heritability: family studies put the heritable component of autism around eighty percent, while twin studies attribute something like forty percent of trait variance to environmental factors [1]. These numbers get read as opposed, as though they’re two explanations competing for the same hundred percent. They aren’t.

Heritability describes how much of the variation between individuals in a population tracks with genetic variation, under whatever environmental conditions that population happens to be living through. It says nothing, however, about how a gene gets expressed. A trait can be highly heritable and still environmentally responsive in its expression, and that isn’t a paradox — it’s the entire premise epigenetics was built to investigate. When the “it’s genetic” statement collapses into a genetic-versus-environmental binary in public discourse, it imports a determinism the underlying research doesn’t support: genes as fixed instruction, disorder as fixed outcome, environment demoted to noise around a signal that was already decided at conception. Epigenetics provides a more nuanced picture of a population responding to environmental pressure.

Western scientific inquiry is good at identifying units and tracking variation across a group of those units. What it is not good at, at least where human populations are concerned, is treating a population as a coherent entity in its own right, rather than a collective of discrete individuals. Epigenetics is the lens that could close this gap, because epigenetics is not about the genome per se — it’s about how the traits within the genome get expressed, generation after generation, in response to what the organism and its lineage actually lived through. Studying the genome tells you what’s possible. Studying trait expression across a population relative to shared conditions tells you how that population is responding.

Here’s a way to think about this: a tree’s rings hold both stories at once: the tree’s own history, and the climate it shared with every other tree that grew through the same timespan. Cross-dating between trees, matching ring patterns across a stand or a region, recovers events no single tree’s rings could prove alone. Human neurodevelopment, read at the population level rather than the individual one, potentially offers something structurally similar — clusters of trait expression legible as a record of what a population was responding to, not just what any one person happened to inherit.

The obstacle here is that human populations mostly don’t hold still long enough to be read as coherent organisms anymore. The method works cleanly on emplaced populations — a group bound to one watershed for many generations, with a stable set of neighbors, a bounded resource base, and enough repeated contact across generations for epigenetic signal to compound rather than scatter. Modern industrial capitalism does not organize people this way. Supply chains, labor markets, and social welfare systems now span the globe, so a disruption in one place ripples into populations that share none of its immediate conditions.

And within any one place, wealth inequality means neighbors under the same sky experience different climates in practice — air conditioning for some, extreme heat for others — so that even physical proximity stops guaranteeing shared exposure. The controls a study needs for coherence simply aren’t there anymore — not because our biology changed significantly, but because the population structure that made the biology legible has been systematically and fundamentally dismantled.

Consider what we would expect to happen, in principle, when an emplaced population experiences a shared disruption — a volcanic eruption, say, creating environmental changes that are sharp in onset and dissipating over time toward a new equilibrium. The epigenetic record should carry a recognizable signature of that event. But the expressions produced in any one individual won’t repeat identically in the next; they’ll appear as a distributed pattern across the population, not a uniform stamp on every member of it. The interesting question here is what that distributed pattern actually represents.

In the current academic paradigm, the field’s default reading treats stress-induced trait variance as damage — a burden to be measured, a departure from an optimum to be corrected. I suspect that at least some of that variance is better read as adaptation: a population’s response to disruption spreading itself across a range of trait expressions rather than converging on one, because a spread of responses is itself the more survivable configuration when the organism — the overall population — is adjusting with a range of specific traits that the new conditions might reward.

Evolutionary biology has language for this — adaptive versus non-adaptive plasticity, and heritable bet hedging — and the honest state of that literature is that the two readings are hard to tell apart from the epigenetic marker alone [2]. A recent transgenerational study of a Daphnia (water flea) species, (which makes disentangling inherited epigenetic effects from direct exposure tractable), found exactly this ambiguity in miniature [3]. Individual-level metrics — survival, growth, maturation rate — dropped significantly in the lineage carrying an ancestral stress exposure. But the population-level reproductive rate showed no significant deficit. Something absorbed the individual cost before it reached the population. The variance in reproductive output went up. The authors named heritable bet hedging as one live explanation for that increased variance and could not rule it out. What’s worth sitting with here is that even the study’s fitness metric was built individual-first — survival, growth, reproduction, then summed upward into a population number — which is exactly the bias I want to name, showing up even in a species chosen because it’s supposed to be observable at the population level.

(Worth noting: the study isn’t working with a coherent population even on its own terms. The eight genotypes defined in the study came from one wild pond, but the exposure itself ran across four lab generations in controlled common-garden conditions — fixed temperature, fixed light, a single defined food source. That’s a stable, simplified substrate standing in for a lake environment that varies on several timescales at once. A bet-hedging strategy tuned for that richer natural variability may not express the same way against an environment that’s had most of its variability engineered out.)

If population-level fitness can hold steady while individual-level fitness declines, then a population absorbing individual cost to preserve its own stability isn’t obviously a malfunction — it might be the mechanism working as intended. This is where the argument gets genuinely contested rather than merely underdeveloped — it edges toward group selection, a framework with a difficult history in evolutionary biology, because pure altruistic restraint at the individual level is usually unstable unless something specific holds the group together: kin proximity, low migration, repeated interaction across generations.

Which is to say: emplacement. The structural condition that makes the epigenetic signal legible in the first place may be the same condition required for population-level trait distribution to function as an adaptive strategy at all. Even the Daphnia lineage was only emplaced in a laboratory, not a lake — a reminder of how easily environmental substitution slips in unnoticed. I don’t think this connection has been made yet, and I’m not going to pretend I’ve made it rigorously here. I’m naming it as a gap to be explored.

There’s a second mechanism worth examining here, that points to why we’ve been able to go so far down the road of pathologizing differences, rather than including them in the catalog of adaptive variation — without the kind of correction that this essay started with. In an emplaced population with overlapping resource boundaries, recognizing someone as fundamentally similar to your own group but of ambiguous relational status — friendly or hostile, a party to shared agreements or a threat to them, horizontal or hierarchical in the standing they hold relative to you — would matter enormously, and the number of groups any one person needed to classify this way was bounded and slow-changing across generations.

Modern life multiplies the number of ambiguous-status “others” a person encounters in a single day past anything that human cognitive architecture was built to resolve, while stripping away the generational information that used to settle the question. A system that can’t sort friend from foe fast enough either stays on high alert or just checks out — and either one, multiplied across a whole population, could easily look like the individual neurological disorders we diagnose today.

This narrative has gone far enough outside of the paper trail of existing research to be considered unsupported. That’s the point. If this were all settled science, my brain would have very little interest in articulating all of this, because the scholarship would speak for itself. I am wired to advocate for processes and solutions that are being overlooked. The bulk of neurodevelopmental research measures individuals because individuals are what current tools can quantify — diagnostic criteria built around individual deviation from a norm, genetic and epigenetic markers drawn from individual tissue samples, fitness metrics that start at individual survival and only later, if at all, get summed into anything population-shaped. That isn’t a neutral limitation. It’s the same units-not-populations bias I started with, inherited by a field that never had the instruments to ask its central question in the form it actually takes: not whether neurodivergence is a disorder in individuals, but whether it’s the visible edge of a population doing what populations under pressure have always done — spreading its bets across a range of trait expression, because no population and no watershed ever knew in advance which single trait the next disruption would reward.

The stable, emplaced population I’ve been using as a reference point throughout is itself a fiction — useful for isolating a mechanism, but never realized in an existing human system. Human groups have migrated, intermixed, and re-sorted themselves for as long as there have been humans, and the pace of that flux has only accelerated relative to any given environment’s own rate of change. So there’s no baseline population sitting somewhere, patiently emplaced, against which modern trait distribution could be measured as a deviation. If there’s a default condition at all, it’s closer to something like adaptation to an urban industrial economic substrate — but even that default doesn’t hold steady, because individuals and groups sit at wildly different distances from those conditions, and that distance itself keeps shifting.

I’ll state my own bias plainly: I think what currently gets diagnosed as a spectrum of disorder is, in large part, a subset of ordinary human adaptive variation, misread through instruments built to find pathology in individuals rather than distribution in populations. This is also why autism in particular resists every attempt to encapsulate it as one condition. A diagnostic category built to describe a single deficit keeps fracturing under the weight of everyone it’s asked to hold, because it was never describing a single thing. It was describing a range.

If that’s correct, the diagnostic model isn’t a neutral tool waiting for better data — it’s the wrong kind of tool for the question. I’d rather see it abandoned in favor of a needs-based model: assessment and accommodation built around what a given person requires to function and contribute, not around how far they sit from a statistical center that was never stable to begin with. That shift only works if it’s accompanied by a parallel shift in social organization — the accommodations have to actually exist and be reachable by the people who need them, not offered in name while remaining structurally unavailable. That’s the harder problem, and it isn’t a scientific one — accommodation keeps getting gated behind proof that a difference deserves it, and that’s a judgment we don’t get to make. Difference can stay visible without having to prove its worth on our timeline first. Then, everyone is taken care of.

---

Sources cited

1. Bey, A. L., Soderling, S., & Dawson, G. (2025). Genetic and environmental influences in autism: guiding the future of tailored early detection and intervention. Journal of Clinical Investigation, 135(22), e201157. https://doi.org/10.1172/JCI201157
2. Ghalambor, C. K., McKay, J. K., Carroll, S. P., & Reznick, D. N. (2007). Adaptive versus non-adaptive phenotypic plasticity and the potential for contemporary adaptation in new environments. Functional Ecology, 21(3), 394–407. https://doi.org/10.1111/j.1365-2435.2007.01283.x
3. Shahmohamadloo, R. S., Fryxell, J. M., & Rudman, S. M. (2025). Transgenerational epigenetic inheritance increases trait variation but is not adaptive. Evolution, 79(6), 1033–1043. https://doi.org/10.1093/evolut/qpaf050

Further reading (background, not directly cited)

These inform the essay’s framing but aren’t quoted or drawn on for specific claims above — worth reading if you want the theoretical scaffolding underneath the group-selection and coalitional-cognition threads.

• Williams, G. C. (1966). Adaptation and Natural Selection: A Critique of Some Current Evolutionary Thought. Princeton University Press. — the classical case against naive group selection.
• Wynne-Edwards, V. C. (1962). Animal Dispersion in Relation to Social Behaviour. Oliver & Boyd. — the group-selection argument Williams was answering.
• Cosmides, L., & Tooby, J. (2001 and later work). On coalitional psychology and the cognitive architecture for in-group/out-group/ambiguous-status classification.
• Barth, F. (Ed.). (1969). Ethnic Groups and Boundaries: The Social Organization of Culture Difference. Little, Brown. — boundary-maintenance as the operative unit of intergroup relation, rather than group “content.”
• Geronimus, A. T. — foundational and ongoing work on the “weathering hypothesis,” population-level biological wear from chronic structural stress.
• Barker, D. J. P., and the broader Developmental Origins of Health and Disease (DOHaD) literature, including studies of the Dutch Hunger Winter cohort — early precedent for population-level, generationally legible response to shared disruption.

Discussion questions

1. Considering the impact of modern industrial social organization on the environment, what might be problematic about using adaptation to that organization as a standard of human fitness?
2. The essay distinguishes heritability from expression — a trait can be highly heritable and still environmentally responsive. Why does collapsing this distinction into a genetic-versus-environmental binary distort public understanding of conditions like autism?
3. What does the tree-ring analogy make visible about population-level epigenetic signal that a purely individual-level genetic analysis would miss? Where does the analogy break down when applied to human populations?
4. The essay argues that emplacement — boundedness to one place across generations — is a precondition for reading epigenetic signal clearly. Is emplacement also a precondition for the underlying biological mechanism to function adaptively, or only for scientists’ ability to detect it? What would settle that question?
5. In the Daphnia study, individual-level fitness metrics declined while population-level reproductive rate held steady. What assumptions would have to hold for this pattern to generalize to human populations, and which of those assumptions seem least secure?
6. The essay raises group selection as a candidate framework, then flags its difficult history in evolutionary biology. What conditions does group-level explanation require to avoid being undermined by individual-level “cheating,” and does anything in modern human social organization supply those conditions?
7. What would a “needs-based” model of assessment have to measure that a diagnostic model does not? Sketch what an intake or evaluation process might look like under each model.
8. The essay claims current neurodevelopmental research is individual-first by instrument design rather than by neutral necessity. What would a population-level instrument for measuring distributed trait covariance actually need to capture, and what data would it require that isn’t currently collected?
9. The coalitional-cognition thread proposes that ambiguous-status classification overload is a plausible mechanism linking loss of emplacement to neurodivergent-coded trait expression. Is this mechanism independent of the epigenetic argument, or could the two be connected? What kind of evidence would link them?
10. The essay states its own bias openly at the close rather than at the outset. What effect does placing the bias statement at the end, after the evidence and its limits have been laid out, have on how a reader is likely to receive it, compared to stating it up front?
11. If neurodivergence is partly a population’s distributed bet against future conditions, what does that framing predict about how the proportion or character of neurodivergent trait expression might shift under a new large-scale disruption — and how would you test that prediction without waiting a generation?
12. The essay ends by calling for accommodations to be not just available in principle but structurally reachable. What social or economic conditions would have to change for a needs-based model to avoid simply reproducing the access inequalities already present in the diagnostic model it replaces?