On the Origins of a Specious Argument about Autism Part 2: Twins

There is a public debate over whether autism rates are increasing. Empirically, rates have dramatically risen over the last two decades. Some scientists argue these rates are due to increased screening and diagnostic criteria changes. Prestigious news outlets have published articles in which genetic researchers argue that the disorder has always been present at these rates; we just didn’t call it autism. These high rates come from new criteria and more screening.

The follow-up question that rarely comes up is this: Why does this matter? If we agree that the rate is roughly 1 in 31 children now, then why argue about what the rate was in the year 2001?

Maybe it’s because the rate points to causes of autism. If autism is increasing, the rate increase cannot be explained by genetics. Genes don’t change on a time scale of decades. Our genes are the same as they were 50 years ago—there hasn’t been a mass mutation to increase any disease. So, if we accept that autism rates are increasing, the increase has to be environmentally driven. That doesn’t mean genetics aren’t important, but it does mean that if we want to reverse this trend, we need to look beyond just genetics. The New York Times has published a number of articles about autism being primarily genetic with environment being of only minor importance even stating: “studies have so far found only weak evidence for environmental factors —like pollution or certain chemicals — that might contribute to autism. They find correlations, but not causal evidence.” As someone who likes to read novels, I always cherish irony whenever I find it.

Correlation ≠ Causation (with apparently 1 exception)

Genetic heritability is an estimate based on correlation alone. In fact, virtually no other study type depends as much on correlation as heritability. No genetic or environmental factors actually need to be measured to calculate heritability (and typically aren’t measured). There are also untestable assumptions that favor genetics over environment. Most heritability studies are done in twins. Because twins share environment along with sharing genetics, assumptions have to be made about whether the diseases twins experience came from their genetic correlations or their environmental correlations. For example, the possibility of joint effects of genes and environment (e.g., interaction, effect modification, synergy, or potentiation) is assumed not to exist. I will say it one more time: heritability is correlational. Nonetheless, whenever heritability is discussed in the press, somehow correlation does indeed equal causation, and high heritability is our proof of genetic causation—especially for autism.

So, in summary—observational environmental research is correlational; observational genetic research is causal—“Got it!” It’s an exception to the rule—like “i” before “e” except after “c.” I apologize for the sarcasm, but it’s tiring to rehash this over and over. Genes and environment interact. Heritability assumes they work independently. This means all the assumptions of heritability as a “Nature versus Nurture” dynamic have no basis in biology.

Everything that Rises Must Converge

Regarding autism, we have two pieces of information that conflict with each other. Rising rates (i.e., it’s the environment) and high heritability (i.e., it’s genetics). Could autism rates be rising and still be compatible with high heritability? Well, this would be a dull blog if I said “no,” wouldn’t it? So yes, it is possible. The underlying problem is the “either/or” nature of how heritability is interpreted. We don’t consider the possibility that both genetics and environment are needed at the same time to cause autism. Instead, we force results to be interpreted within immutable bins of genes vs environment. Genes don’t work in the absence of environment—they always interact with some component of environment. Consider the theory of natural selection. In reality, this is not a genetic theory; it is a theory of gene-environment interaction. All our genes and genetic variants were selected over thousands of years to help us live long enough to reproduce in the environments our ancestors lived in. Our genetics were shaped by environment.

Of Mice Chickens and Men

So how does this play out in heritability? With apologies to Ken Rothman, a famous epidemiologist, I will use chickens to illustrate. I first saw Dr. Rothman illustrate the joint role of environment in genetic traits at a lecture in 1998. A chicken shank is the lower part of the leg near the foot. The color of the shanks can be yellow or white and is a genetic trait following an autosomal dominant pattern. However, the trait will only arise if the chicken with the genetic variant also eats yellow (as opposed to white) corn.

Imagine two farmers—Jones and Smith, who live down the road from each other. Each has a flock of chickens. Farmer Jones’ inbred flock of chickens all carry two copies of the allele for yellow shanks, unbeknownst to him. He feeds them white corn because his parents told him it is healthier. One day the store runs out of white corn, so he has to feed them yellow corn. All his chickens develop yellow shanks. To him, yellow shanks are environmental and are caused by feeding chickens yellow corn.

Down the road, Farmer Smith also has an inbred flock of chickens, none of which have the genetic variant for yellow chicken shanks. He feeds them yellow corn because his parents told him that all corn is the same so use the cheapest. One day one of his chickens has a chick that has yellow shanks, i.e., a spontaneous mutation. He notes that half of the offspring of this chick, in turn, have yellow shanks. Of the ones with white shanks, none of their descendants get yellow shanks. Only the chickens with yellow shanks have offspring with yellow shanks. A pedigree diagram would reveal an autosomal dominant inheritance pattern. To Farmer Smith, yellow shanks are genetic.

Heritability of yellow shanks in Farmer Jones’ flock would be 0%. Heritability in Farmer Smith’s flock would be 100%. Which heritability measure is correct? The same genetic variant and environmental factor are causing yellow shanks in each flock. Biologically, nothing has changed. How can the heritability be so wildly different?

How do Twin Studies Work?

A twin study compares two types of twins: “identical” or monozygotic (MZ) twins with 100% shared genes and “fraternal” or dizygotic (DZ) twins that share about 50% of their genes. The core logic is that both MZ and DZ twin pairs experience equally shared environments (e.g., family, diet, neighborhood) within their two-person set. Therefore, any differences in traits or disease rates across twin sets are attributable to their genetic differences. To estimate the genetic difference, researchers measure the concordance (i.e., correlation) for a specific trait or disorder, within each twin pair. In other words, if an MZ twin has autism, what is the probability that the sibling MZ twin also has autism? And if a DZ twin has autism, what is the probability of the sibling DZ twin having autism? These two measures of concordance (MZ vs DZ) are then compared.

If MZ twins show a significantly higher rate of concordance for a trait than DZ twins, it suggests that genetics plays a significant role. The greater the difference in concordance between MZ and DZ twins, the higher the heritability of the trait. The analysis also assumes that genetics and environment operate independently. This is why the percentage of heritability equals 100% when adding genetic heritability with environmental effects.

Probands are anti-environment

In order to be included in a twin study of heritability, at least one of the two twins must have the trait or disorder—otherwise, they can’t contribute to the heritability calculation. We aren’t analyzing all twins—just the ones with a disease. In genetics, a “proband” is the first person in a family to be identified with a disorder, condition, or trait under investigation. The proband is therefore the entry point for twin pairs into a heritability study. In the case of yellow chicken shanks, an MZ pair of twin chickens with the genotype for yellow shanks who ate white corn won’t be included in a heritability estimate—because there is no proband. Any MZ chicken twins owned by Farmer Jones consumed white corn and will have white shanks despite their genetics. That means that prior to giving them yellow corn, all of Farmer Jones’ twin chickens would never be included in a twin study of yellow shanks. All the probands in a twin study of yellow shanks would come from Farmer Smith, and the calculation would show 100% heritability.

Now, let’s use a human genetic disease as an example. Hemochromatosis is a genetic disease in which hyper-absorption of iron leads to liver disease and diabetes. Let’s assume that the genetic trait is found in a population that is part of a large twin registry that has a highly variable diet. About half of the population is vegan and the other half eats a diet rich in red meat and iron. The vegans would be disease-free even if they have the hemochromatosis genetic trait. In a twin study of hemochromatosis, the probands by definition would be everyone with the genotype for hemochromatosis who also ate red meat, because BOTH high dietary iron and genetic variant are needed to get the disease and all probands have the disease. Twins with the genotype for hemochromatosis but who were vegetarians would not be included in the heritability calculation. Furthermore, under the typical assumptions of heritability, gene-environment interaction doesn’t exist. A gene-environment interaction that is biologically mandatory in order to express the trait/disease will always be assigned to the genetic bin. Put more simply, the design of a twin study forces gene-environment interactions to appear genetic. We would never see any proband twins with the genetic risk factor unless they also had the environmental risk factor, which is shared across twin pairs. That is why the heritability stays 100% in this example, even though about half the people with the hemochromatosis genotype in our population never get hemochromatosis. We assume that environment is equal within each set of twins, but we ignore that environment is not equal across the world’s population of twins with the disease genotype. The end result is that the gene-environment interaction is hidden and appears genetic, and the heritability estimate is inflated (i.e., high). As a thought experiment, what would happen to heritability and to the rate of new cases of hemochromatosis if 100% of people were exposed to high dietary iron? Let’s say iron was added to the food supply at mistakenly high levels. Healthy vegetarians with the hemochromatosis genotype but no disease would now develop the disease. The rate would go up, and heritability would stay the same. Doesn’t that sound similar to autism?

“Everything should be made as simple as possible, but not simpler than possible.” - Albert Einstein

Heritability is easy to understand, but it elevates correlation to causation. At the same time, heritability looks solely under the lamppost for genes versus environment and doesn’t consider other possibilities—like gene-environment interaction. To find gene-environment interactions, you have to look for them. If they are not measured, they will not be found, but they absolutely exist.

Thinking that autism cannot have mandatory environmental causes based on heritability omits highly realistic biological processes. Like dietary interventions in the genetic disease Phenylketonuria, there may be environmental interventions that can prevent autism. Such interventions can even be applicable in the 80% of people who fall into the high heritability category for autism. The “Nature versus Nurture” dynamic makes us assume only one part of that phrase drives a trait or disorder. The truth is that both parts can be mandatory together. Autism could even be 100% heritable and still have a required environmental cause. “Nature versus Nurture” is not a biological argument—it is a belief system. The biggest mistake we can make going forward is using heritability to justify looking solely at genetics and “finish that work first” before doing the environmental research. The missing heritability will never be found with a “finish genetics” approach because what is missing are the gene-environment interactions. Let’s not make that mistake; let’s study Nature and Nurture together as interaction, synergy, potentiation—whatever we want to call it. This isn’t even a new idea—Ken Rothman taught me this in 1998. I’ll end with a quote from another prominent scientist. The late neuroscientist Donald Hebb was once asked, “What contributes more to personality—nature or nurture?” He responded: “Which contributes more to the area of a rectangle, its length or its width?” For the last two decades, we’ve just measured the length—let’s finally add the width now, too.

If you are interested, please consider reading Part I from 2022, which is accessible on the Institute for Exposomic Research website where we first published this Substack blog.