As the number of genes associated with autism increases, so does the number of mouse models based on disruptions of those genes. These genetic models hold great potential for advancing our understanding of the biology of autism. They also tend to incite criticism of the animal models that came before.
I use a mouse model based on in utero exposure to valproic acid (VPA). VPA is a drug that is used to control epileptic seizures and to treat bipolar disorder. Multiple studies have demonstrated that prenatal exposure to VPA in people is associated with a markedly increased risk for autism1,2,3.
Importantly, the risk of autism is substantially higher for children whose mothers took VPA than for those whose mothers took a different anticonvulsant or left their epilepsy untreated. This suggests that the autism risk associated with VPA exposure is not driven entirely by a family history of epilepsy, and is greater for VPA than for other anticonvulsants.
Based on this association, many teams have exposed rodents to VPA during embryonic development and found that the animals show behaviors like those we classically associate with autism. These include diminished social interactions, an increase in repetitive behaviors and anxiety, and changes in sensitivity to sensory stimuli4,5.
In 2016 and 2017, my colleagues and I reported that mice exposed to VPA in utero show changes in patterns of brain activity that are linked to attention and social behavior6,7. So other autism researchers frequently ask me whether this is a ‘good’ model of autism. I believe the model has value for increasing our understanding of the brain circuitry that underlies autism features.
Before explaining why, I think it would be useful for me to outline what I — as a psychiatrist and a neuroscientist — hope to learn from any mouse model of autism.
My overall goal is to understand the neural mechanisms that produce typical behaviors, and to determine how these mechanisms go awry in conditions such as autism. The approach relies on the assumption that the neuronal processes that produce a given behavior are more or less the same from one individual to the next.
This supposition may not hold true in every case. For example, individuals may use different strategies — and different brain circuits — to solve cognitive tasks. But as a first approximation, systems neuroscientists accept the assumption that we can identify the neural mechanisms underlying typical behaviors.
There is much greater resistance, however, to extending this logic to atypical behaviors or deficits.
The problem is that there are many ways to disrupt any given behavioral pathway. To make an analogy, there might be one correct way to use a child’s complicated Lego set to build a robot, say, but there are countless ways to assemble the Lego pieces incorrectly. This abundance of potential disruptions introduces the concern that if we do not study the ‘right’ models, we could end up identifying brain mechanisms that are not actually relevant to autism.
Although such caution is warranted, I would argue that if we start off by identifying mechanisms that are linked to typical behaviors and then study a model in which those behaviors are disrupted, we are bound to learn something valuable about the mechanisms underlying atypical behaviors.
Because autism is so immensely heterogeneous, no single pathway in the brain is likely to underlie all cases. At the same time, any perturbation we discover is likely to be clinically relevant to at least a subset of individuals with autism.
The VPA exposure model is particularly compelling in that it recapitulates a cluster of autism-associated behaviors using a single perturbation. So we can use this model to identify an upstream developmental or neural perturbation that gives rise to the behavioral traits that tend to co-occur in autism.
The fact that animals exposed to VPA bear a behavioral resemblance to people with autism gives this model ‘face validity.’ But the model’s relevance should not be based on this resemblance. Rather it should rest on whether the altered behavior in the mice results from neural circuits that are relevant to autism.
Work from my lab and others’ suggests that the cell types compromised by in utero VPA exposure in mice are homologous to those that have been implicated in autism7. So the social deficits that appear in mice exposed to VPA arise from a perturbation that increases autism risk, co-occur with other autism-associated behaviors and involve brain circuits analogous to those implicated in autism.
This confluence of factors suggests that VPA-exposed mice are a valid model for studying processes relevant to autism.
The more we learn about the genetics of autism, the more the mechanisms through which VPA exposure exerts its effects seem biologically relevant.
VPA inhibits a class of enzymes called histone deacetylases that regulate chromatin packing and gene expression. Many candidate genes for autism, including ARID1B and CHD8, are involved in chromatin regulation8.
What’s more, VPA and autism genes appear to act on the brain during the same developmental period — one that corresponds with 10 to 24 weeks after conception in people9. Mice exposed to VPA during this developmental window show autism-like behaviors, and many autism-linked genes are expressed during this time as well.
Work from my lab and others’ suggests that VPA exposure interferes with brain systems governing social behavior and attention in mice6,7. We are working to validate the brain mechanisms we have identified in VPA-exposed mice by showing that these same pathways are similarly altered in other animal models of autism, including those based on disruptions of high-confidence autism genes.
Although the VPA model for autism has clear clinical relevance, the autism risk associated with prenatal exposure has been declining over time. As our knowledge about the risks of the drug has increased, clinicians may be prescribing lower doses or shorter courses of treatment for pregnant women with epilepsy.
No one can say VPA exposure is the perfect model of autism. But I believe it is a good one. I also believe that trying to extract meaningful biological insights from imperfect models remains an important direction for the field. The key to success may be to avoid focusing on findings that are idiosyncratic to one model. Instead, we should search for results common to multiple models that also intersect with brain mechanisms that contribute to typical behavior.
Vikaas Sohal is associate professor of psychiatry at the University of California, San Francisco.