Crosstalk Debates and conversations about timely topics in autism.
Opinion / Cross Talk

What constitutes ‘environmental’ risk for autism?

by  /  11 June 2014

In May, we reported on results from the largest population study to date to assess the heritability of autism.

The researchers concluded from their study that “genetic factors explain half of the risk for autism.” Much of the media interpreted this — incorrectly — to mean that environmental factors make up the rest of autism risk.

Among autism researchers, the paper led to vigorous discussion about what exactly the term ‘environmental’ comprises, and about the challenges of calculating the heritability of complex disorders.

As part of our discussion series, Cross Talk, we asked researchers who study heritability and quantitative genetics to clarify these points.

What do you think? Share your reactions and follow-up questions in the comments section below.

The Experts:

Michael Wigler

Professor, Cold Spring Harbor Laboratory

Greg Gibson

Director, Center for Integrative Genomics, Georgia Institute of Technology

Kevin Mitchell

Associate Professor of Genetics, Trinity College Dublin

Peter Visscher

Professor and Chair of Quantitative Genetics, University of Queensland

Lonnie Zwaigenbaum

Professor, University of Alberta

Random, not environmental

Professor, Cold Spring Harbor Laboratory

Essential caveats: “This important study from the Karolinska Institutet reports incidence data for autism and autism spectrum disorders for the Swedish population, accrued over the past 20 years. Our point of view was not clearly captured in article on this work. These valuable new data offer a comparison of risk between twins, siblings and half-siblings. The authors conclude that the data support the hypothesis that autism is half genetic and half ‘environmental.’

“Unfortunately, the authors appear to confuse randomness (in outcome) with environment — a very loaded term — as a catch-all for ‘incompletely determined by genetic state.’ Moreover, the model the authors use to reach their conclusion may not be appropriate, and they do not evaluate the data in light of recent thought and evidence for the role of singly highly penetrant rare and de novo variants (mutations that are not inherited, but spontaneously occur in a parent’s egg or sperm cells). They also seem to ignore the importance of evaluating the higher moments of sibling riskriskto children born into families that already have two others with autism — as explained in our recent review.”

Alternate conclusions possible: “Full access to the raw data would be valuable to the wider community, who could use other tools to model the data — in which case, a very different set of conclusions might be drawn.”

Heritability is about populations, not individuals

Director, Center for Integrative Genomics, Georgia Institute of Technology

Does it matter?Without detracting from the magnitude of the accomplishment and the analyses, which I find compelling, at some level arguing over whether genetics explains 50 percent or 80 percent of the variance for autism is a little like asking whatamount of global warming can be explained by human activity: The bottom line is that the contribution is enormous, but it is not particularly enlightening with respect to specific cases (of autism or environmental catastrophe). Heritability is a statement about populations, not individuals, and any measure is compatible with markedly different genetic contributions in individual cases.”

What does environment mean? On the other hand, the contention that environment is more important than generally accepted surely only has policy implications if the source of the environmental effect is known. Is it unidentifiable effects, identifiable non-biological effects we can’t do much about (urban living, sunlight exposure), identifiable cultural effects we may be able to change (daily exposure to TV or iPads), or identifiable effects we would probably not change even if known (living as nuclear instead of extended families, perhaps)? More important than this is the question of whether the environmental effects are the same for everyone: Perhaps they are triggers for some children, and general susceptibility factors for others.”

Rare variants may be overhyped: It does not seem that copy number variations, de novo mutations, and segregating rare variants are collectively causal in more than a quarter of cases (causal in the sense that without the variant, the criteria for autism would not be met). This leaves the vast majority of cases unexplained, and there is really strong evidence that thousands of common variants of small effect are contributing to risk. Quite likely they contribute to gene expression profiles that place each of us closer or further from a risk threshold that rare variants or environmental perturbations push some children over. Given this complexity, it is almost certainly the case that environmental factors are also important, so for me, this paper reminds us of the complexity of the disorder.”

Exploring heritability: “Geneticists are finding more and more rare and common variants that influence the likelihood of developing autism. It would be great if there were as much research into the genetics and heritability of the disorder’s progression and response to interventions through genome-wide association studies.”

Work does not suggest external, environmental factors

Associate Professor of Genetics, Trinity College Dublin

Questionable model: “The general conclusions from this new work on the heritability of autism fit with previous studies, in that they show a large effect of genetic variance and no effect of the shared family environment. However, the precise estimates of the mathematical values of these terms depend on a number of assumptions in the statistical models, which are not universally agreed upon. The numbers that emerge are in general of limited value anyway, given that the diagnostic category of autism is really an umbrella term for many distinct genetic etiologies and not a specific genotype.”

Dangerous terminology: “The conclusions of an important effect for the environment should be tempered, especially given how they can be misinterpreted and misrepresented. The fact that monozygotic twins are not always concordant for the disorder suggests there are some non-genetic factors that contribute to whether individuals develop the disorder or not. In the analytical models used here, these sources of variance are bundled into a statistical term called the ‘non-shared environment.’ This term is used here in a technical sense that does not correspond to the colloquial meaning. In particular, it should not be taken as evidence of effects of some causal factors out in the environment.

Unique development: “Brain development is incredibly complex and inherently variable, even for the same genotype and in the same environment. The probabilistic nature of neurodevelopmental events at the cellular level can manifest as quite different phenotypic outcomes at the organismal level. By the time they are born, the brains of monozygotic twins are thus already highly distinct from each other. This does not represent effects of external factors — this variability is entirely intrinsic to the developing organism itself.”

“Given that they are likely to contribute to the already rampant misinformation regarding causes of autism, headlines suggesting that this study demonstrates an important role for external, environmental factors should therefore be avoided.”

Common variation does not contradict rare contributions

Professor of Quantitative Genetics, Queensland Brain Institute, University of Queensland

What it tells us: “This is the largest population-based study to date on autism risk for relatives of individuals with the disorder. The authors use the observed risk to relatives to estimate heritability on the unobserved scale of ‘liability.’ They estimate that about 50 percent of variation in liability to autism in the Swedish population is due to additive genetic factors. The study confirms that autism is a ‘complex trait,’ just like other traits and disorders in humans, all the way from height to schizophrenia.”

New mutations: “The population parameter of heritability is ‘blind’ in terms of the number of genes, their effect size and frequency in the population, so this study is not informative in that respect. The mathematical model that fit the observations best was one for which variation in liability is due to additive genetic effects (which are shared by relatives in proportion to the degree of relationship) and ‘environmental’ factors that are specific to an individual. The study is not informative about what these ‘environmental’ effects might be, other than to conclude that they cannot be attributed to the shared environment of family members and hence are considered to be unique to individuals. But they can also include the effect of new mutations.”

Parsing liability: “The observed risk to relatives of this study is remarkably consistent with the researchers’ hypothesized model of liability, since on that scale the risk halves with each degree of relatedness, all the way from monozygotic twins to cousins. The results are also consistent with similar studies on schizophrenia and bipolar disorder. For those disorders, many common genetic variants have been found that explain some of the heritability, using genome-wide association studies. Notably, the experimental sample sizes for genome-wide association studies on autism have been much smaller than those for other psychiatric disorders, even though evidence for the existence of common genetic variation for autism has been reported. A role of common genetic variation does not contradict a contribution of rare and de novo mutations to risk of autism.”

Highlights challenges of modeling heritability

Professor of Pediatrics, University of Alberta

Recurring insight: “The authors provide estimates of autism recurrence risk for each of the types of relatives (from twins, to siblings, half-siblings and cousins), which may be helpful in counseling families. It is important to note that these risk estimates (for example, 12.9 percent in siblings) are based on risk at age 20 years, in contrast to other recent studies in which recurrencerisk was based on assessment at a much younger age. For instance, Sally Ozonoff and her colleagues reported in 2011 that 18.7 percent of younger siblings of children with autism also have the diagnosis at age 3 years.

“It is intriguing that they do not find any trend towards increasing recurrence risk of autism in relatives over the study period. This is consistent with Danish birth registry data reported last year by Therese Koops Grønborg and her colleagues, despite trends towards higher rates of autism diagnosis in the general community. Such trends are widely regarded to be influenced, at least in part, by the clinical diagnostic criteria for autism broadening over time (that is, to a broader ‘spectrum’). If this is truly the main driver, one might expect to observe trends towards higher recurrence rates as well.”

Challenging model: “It’s important to emphasize, though, that heritability estimates do not have a straightforward interpretation for individual families. For example, there are families in which two affected siblings carry different rare pathogenic variants, a circumstance that certainly challenges how we have modeled heritability.”

About Cross Talk
Discussions among leading experts in the field. Submit your questions to

16 responses to “What constitutes ‘environmental’ risk for autism?”

  1. passionlessDrone says:

    “The observed risk to relatives of this study is remarkably consistent with the researchers’ hypothesized model of liability, since on that scale the risk halves with each degree of relatedness, all the way from monozygotic twins to cousins.”

    OK, am I missing something, or was the liability was much, much, MUCH higher for monozygotic twins compared to *everyone else*? From the abstract:

    “**The RRR and rate per 100 000 person-years for ASD among monozygotic twins was estimated to be 153.0 (95% CI, 56.7-412.8; rate, 6274 for exposed vs 27 for unexposed ); for dizygotic twins, 8.2 (95% CI, 3.7-18.1; rate, 805 for exposed vs 55 for unexposed); for full siblings, 10.3 (95% CI, 9.4-11.3; rate, 829 for exposed vs 49 for unexposed); for maternal half siblings, 3.3 (95% CI, 2.6-4.2; rate, 492 for exposed vs 94 for unexposed); for paternal half siblings, 2.9 (95% CI, 2.2-3.7; rate, 371 for exposed vs 85 for unexposed); and for cousins, 2.0 (95% CI, 1.8-2.2; rate, 155 for exposed vs 49 for unexposed).**”

    Monozygotic twins had a roughly fifteen times increased risk compared to full siblings / dizygotic twins! Forgetting for the moment about which statistical models were used to arrive at ‘half genetic/half environmental’ talking point that landed int he media, the risk between monozygotic twins and the rest of the relationships was *not* stepwise by half, it is exponential.

    Maybe I’m a bit too dim to understand, but I’m really struggling with understanding why this doesn’t color our interpretation previous twin studies which formed the basis of heritability estimates in autism? If this risk relationship is real, how can genetic modifications account for it such a large effect in this one, very particular population?

    As noted in the previous discussion @ SFARI, monozygotic twins share a *placenta* roughly 75% of the time. This is an environmental condition that is unique to this population.

    Can anyone help?

    • RAJensen says:

      Down Syndrome (1-750) and Klinefelter syndrome (1 in 500 newborn males) are the most common genetic syndromes associated with increased autism risk.These syndromes are rarely ‘inherited’ they are caused by reproductive errors (sperm or egg mutations) yet MZ concordance rates are nearly 100% and DZ concordance rates are nearly 0%. Classic twin study design claims it can calculate a ‘heritabilty’ estimate based on the difference betwwen MZ and DZ concordance rates. The ‘heritabilty estimate’ of Down syndrome and Klinefelter syndrome is higher than the ‘heritablity estimates’ for autism.

      The high rates of de novo gene mutations in autism caused by reproductive errors and classic twin study design has inflated the heritabily estimates for autism.

    • Jack says:

      The heritability estimates are based on a model where it is assumed that there is an underlying normal distribution for liability to ASD. If these is an underlying normal distribution, the expectation is that the risk increases exponentially with genetic relatedness. The risk ratios are converted to correlations, and heritability is estimated from those correlations.

    • Mark Davis says:

      Sure … if this was all genetics – then we would have a 100% genetics.
      If it is not genetics and thus environment – then why are babies in essentially the same conditions in the same home have such differences? We are talking of babies less than six months old.

      The answer is the trick assumption – it is neither genetics nor environment – but “something else.”

      Ergo there is genetics – there is environment (such as a twin raised in another place and socio-economic background) and there is “something else” which is used to translate genetics into the way create and program our own brains.

      In this case – genetics creates the same brain but it is the software programming mechanism which means each and every identical twin learns how to think – and this learning mechanism is unique, and thus identical twins must be individual.

      The modelling used in “self programming our brains as babies” seems to be based on fuzzy logic principles. That is – outcome oriented programming with genetics providing both the hardware of the brain – plus the motivation to learn how to learn … and then we learn how to use our brain.

      OUTCOME – The way we self program our brains as babies MUST be unique which means that Identical twins cannot be identical.
      It also says that Autism Spectrum is a self programming problem in the brain.
      It also implies that you can change this programming.

      Which – if you are interested, is what I did. So the theory and the practice unite in the data. You can literally debug your self programming and learn to use other parts of the brain.

      The real question is what sort of genetic damage is done so that 77% of identical twins choose (as babies) to program their brains with ASD instead of normal …

      And of course – how they change from a normal brain to an ASD brain.

  2. Caroline Rodgers says:

    Is it not time to seriously consider the role that prenatal ultrasound may be playing in causing de novo copy number variations and the autism outcomes that are sometimes associated with these genetic findings?

    From as early as five and six weeks gestation, when ultrasound fetal heartbeat monitors are used to confirm pregnancy, followed by the 10-14 week nuchal translucency ultrasound, the second trimester anatomical ultrasound, the electronic fetal monitoring during labor, “keepsake” ultrasounds undertaken to provide fetal portraits and the many, many ultrasound scans that are undertaken in between these times, there are ample opportunities to interfere with early genetic replication and the formation of brain architecture that could have lifelong consequences.

    Prenatal ultrasound would also explain why identical twins do not always have identical outcomes when it comes to autism — it would be easy for one twin to wind up with more ultrasound exposure than the other, even shielding the second fetus from the heat-related effects.

    Certainly, this is a possibility that is worth pursuing.

  3. Steve White says:

    I don’t think ultrasound can cause genetic problems like copy number variations. If there any reason to think so? We know very intense sound is harmful but copy number variations seem unlikely to me.

    • passionlessDrone says:

      Hi Steve White –

      As far as creating CNVs you are likely correct. That being said, there are those that worry that there could be developmental timeframes during which exposure to ultrasound could perturb neuronal migration. For instance,

      Prenatal exposure to ultrasound waves impacts neuronal migration in mice

      might be of interest.

  4. LSwank says:

    My son has mild autism with PANS issues. Strep b treatment, rhogam treatment both in utero. While pregnant, house remodeling with new paint, new carpet, new furniture, fire retardant materials. Early blood work showed fire retardants plus chronic strep issues. Broken immune system with possible neurological assault. Just our story.

  5. N Bishop says:

    MZ twins can differ in chromosome composition. For example, there are increasing numbers of reports of MZ discordance for Down syndrome (trisomy 21), fewer of Klinefelter syndrome (47,XXY), while MZ twins discordant for both sex and Turner syndrome are also well-known.

    Each of these syndromes is considered a ‘syndromic form’ of autism, a syndrome accompanied by a significantly increased risk of fulfilling ASD diagnostic criteria.

    My understanding, of the major contributor to the different frequencies of discordance for these disorders in MZ twins, is that there are differences in the ‘timing’ of the most common causative de novo mutational event(s) for these disorders. Syndromes more commonly resulting from pre-zygotic non-disjunction, will show less discordance, while those more commonly resulting from post-zygotic non-disjunction, will show more discordance.

    Syndromes due to expansion of trinucleotide-repeats, such as FXS, can also have highly complex heritability.

    My understanding, however, is that most studies using twin concordance to examine the heritability of ASD, exclude those individuals where ASD is not their primary diagnosis, but is secondary to a defined genetic syndrome, including aneuploidies.

    It is important that papers specify inclusion and exclusion criteria, in addition to the mathematical analyses used, as it would be expected that different exclusion/inclusion criteria will affect results. However, excluding the increasing numbers of highly-heritable genetically-defined causes of ASD(due to both inherited and de novo mutations), from heritability studies, might lead to an increasing underestimation of the heritability of ASD overall.

    • RAJensen says:

      Not all MZ twins share the same prenatal environment. 2.3rds of MZ twins share the same prenatal environment developing in the same placenta (monochorionic), however 1/3rd of MZ twins do not share the same prenatal environment and develop in separate placentas just like dizygotic twins (dichorionic). Studies have shown that monochorionic MZ twins are more alike than dichorionic twins
      ‘The results indicate that concordant MZ pairs were more likely to have been monochorionic (MC) and to have shared a single placenta, whereas discordant MZ pairs appear more likely to have been dichorionic (DC) with separate placentas’

      A great deal of research has gone into the phenomena of the high male:female (4:1) ratio in autism. Another growing area of interest is the role of skewed X-inactivation found primarily in females. Almost all Rett Syndrome girls have been shown to have skewed X-inactivation patterns. Trejo et al studied 41 MZ twin pairs. 26 with autoimmune disease and 15 newborn females. They found a strong correlation between dichorionic fetal anatomy and differences in X chromosome inactivation patterns between members of an MZ twin pair. In contrast, all monochorionic twin pairs had closely correlated patterns of X chromosome inactivation. Highly similar patterns of X chromosome inactivation among monochorionic twin pairs may result from their shared placental blood supply during intrauterine life. An interesting study that deserves follow-up in autism research particularly what role, if any, skewed X-inactivation may play in autism in females and/or how X-inactivation is expressed in some females meeting strict autism criteria.

  6. Don, Autism resolved says:

    It seems like the mother to be may have the case of an undiagnosed infection, during pregnancy, as well as quite possibly, inflammation build up in the placenta, caused by gestational diabetes, which would do a number of damaging things to the fetus. One very important thing is that the immune system would be sent into a fight or flight mode, keeping in mind 80% of the immune system is the intestines, caused by gestational diabetes, another very important thing I noticed about gestational diabetes is that it disrupts the hormonal balance, one example is insolin, the pancreas is responsible for regulating the levels there of. The root cause to gestational diabetes, and all other forms of diabetes goes beyond the pancreas, to the center of the brain, at the hypothalamus. Autism, Diabetes, Adhd, bipolarism, all can be addressed, by addressing the hormones of the hypathylamus.

  7. Harold L Doherty says:

    SFARI panel discussion title: What constitutes ‘environmental’ risk for autism?

    Wouldn’t it have made sense to at least have 1 panel member actively involved with environmental autism research?

    • gregboustead says:

      Hi Harold,

      You’re right. We should have worked harder to include a researcher specifically studying environmental factors (though we did invite them).

      Our main motivation for hosting a follow-up conversation on the JAMA paper by Sven Sandin and colleagues on the familial risk of autism, however, was to clarify what the authors labeled as contributing risk from the “non-shared environment.” There was much confusion over this ~50% of the pie, which was widely interpreted to mean “environmental risk” — when “unaccounted for risk” is perhaps more accurate. Several researchers expressed concern over this interpretation, and we felt it was important to offer a platform for them to clarify what factors might be included in this category, based on the statistical models used by the researchers.

      This was not intended to be a panel on environmental risk factors in general. With the benefit of hindsight, I think the headline could have better captured that.

      (Note: We also invited the authors of the original paper to respond to these perspectives and offer their interpretation of the “non-shared environment” category of risk, but have yet to receive a response. We’ll gladly post it if we do.)

      Nevertheless, we want this to be a transparent process. We still invite further comment from researchers studying environmental influences in autism. All are welcome to add clarifications or rebuttals to any of the interpretations or assumptions made above. I’ll personally invite some to respond in an effort to balance the perspectives, and all of this could influence future discussion topics for the Cross Talk series.

      Thanks for reading and for your comments!


  8. usethebrainsgodgiveyou says:

    In this PubMed publication, we are made aware that maternal infection with German Measles increased the incidence of Autism >200 fold in affected mothers/fetuses. Activation of the Maternal Immune System Alters Cerebellar Development in the Offspring This is NOT the heritibility we have been led to believe the cause, that arrived coincidentally (or NOT) at the same time scientists were looking to fund the Genome Project.

    I am old enough to remember pregnant women being fearful on contacting rubella or German Measles, as it was known at the time. I had friends who were partially blind and deaf because of it.

    This historical perspective should make us aware that environment, via maternal infection, in the past led to a dramatic increases in the incidence of autism.

    That should tell us something…

  9. usethebrainsgodgiveyou says:

    Thank you, PD, for the numbers.

  10. usethebrainsgodgiveyou says:

    This work at Princeton University gives another environmental possibility, , that is, cerebellar damage. When genetics is pushed over environment, being the cynic I am, all I hear is “it’s not the shots” and possibly, “it’s not the smog.” Sure, if you are a geneticist, you are looking for a genetic cause. But isn’t that a bit myopic?

Leave a Reply

Your email address will not be published. Required fields are marked *