The research world seems to love “biomarkers.”  Not only does “biomarker” appear to be a very fashionable buzzword in certain circles, but it seems like projects that aim to find biomarkers are very attractive to agencies that fund research.[1]  Among researchers, biomarkers are in the vogue.

What is a “biomarker,” exactly?

Well, properly speaking, it refers to some kind of biological marker – something like a physiological, functional, chemical, or molecular signal.  This biological marker might tell us that a person has a particular neurodevelopmental pattern (like autism), or a medical disease, or some other condition.  The biological marker could be even more specific: it might tell us that a person has a given subtype of autism.  Some biological markers might be used to monitor the effects of treatments and interventions.

Because the word “biomarker” is so fashionable, some people use it to refer to things that aren’t really biological at all.  They might use the term to refer to behavioural evidence that is collected in some kind of technologically advanced way.  An example might be eye-tracking data or behavioural videos that are subjected to some kind of classifying algorithm.  I want to be clear about this: when I’m speaking about a biomarker in this post, I’m not talking about these behavioural markers.  As far as I’m concerned, regardless of the merits of specific projects, there’s nothing fundamentally problematic about the concept of an autism behavioural marker.[2]

But I can’t say the same for the idea of an autism biomarker.

Why We Shouldn’t Waste Time Looking for Autism Biomarkers

Unfortunately, researchers in scientific fields today aren’t necessarily trained to think critically about the ideas and concepts they study.  One consequence of this is that we can have a tendency to reify categories that really exist only in our heads.

Autism is kind of made-up.  It’s a social construction.

Obviously, I don’t mean that autism doesn’t exist.  Autistic symptoms and behaviours are real, and we even have standardized assessments to measure them.

However, the idea of autism is made-up.  It’s a socially constructed category that we’ve imposed over an incredibly diverse range of real behaviours.  There is far more diversity within autism than within typical development.  Instead of defining one autism spectrum, we could easily carve up the diversity within autism in different ways.  That’s why autism prevalence has fluctuated over time and across cultural (and even regional!) contexts.

It’s illuminating to contrast autism with something like, say, Fragile X Syndrome.  Fragile X Syndrome is caused by a specific genetic difference: an unusually-large number of CGG repeats in the X chromosome.  These CGG repeats are, by definition, a biomarker.  Furthermore, they cause protein differences, which cause all sorts of other biological differences – and because there is a common biological cause at the root of it all, these differences are going to be relatively similar in different people.  Thus, there are Fragile X biomarkers.

In autism, the situation is very different.  Autism refers to an extraordinarily heterogeneous range of diverse behaviours.  All the research we’ve done suggests that autism is rarely attributable to some specific cause; usually, it will reflect numerous environmental and polygenetic influences and interactions that somehow end up producing autistic behaviour.  When there’s no clear biological cause, the idea that we can find biomarkers just seems bizarre.  The fact that autism is defined in terms of behaviour means that we can use behaviour to identify autism, if only because we’re the ones who set the behavioural bounds of autism in the first place.  But the heterogeneous biology that underpins the heterogeneous behaviours of autism is completely unconstrained: we’ve set a behavioural boundary, but the biological underpinnings of autistic behaviour can freely cross that boundary.  Thus, heterogeneity at the biological level will be greater than heterogeneity at the behavioural level.  Indeed, our studies generally suggest that the neurobiological heterogeneity in autism may be greater than such heterogeneity in typical development.

To be sure, some things that aren’t caused by specific genetic differences do have biomarkers.  Anxiety is a good example.  But anxiety is a largely coherent construct that is associated with some comparatively specific neurobiological processes.  Autism, in contrast, stubbornly resists attempts to identify some core cognitive or biological factor.  It’s just a heterogeneous mess, because the category is so much a social construction.

Now, I’ll admit that some studies have claimed that they have actually found autism biomarkers: signals in autistic people’s brains (or even signals in things like urine) that computer algorithms and machine learning can use to diagnose autism.  However, I’m hardly the first to point out that these studies generally compare autistic people to neurotypicals, but not to people with a variety of divergent neurotypes.  All they’re doing is saying that a person is atypical in some way.  In the real world, that’s spectacularly useless, because typically-developing people generally don’t get referred for clinical neurodevelopmental evaluations.  We want to be able to distinguish divergent neurotypes from one another.  Fortunately, I don’t think very many people take these studies too seriously.[3]

Biomarkers for Subtypes

We’re also fortunate that serious researchers today have some understanding of the idea that autism is a social construction.  They might not use that terminology, and they might sometimes reify the category in a way that doesn’t make sense, but at some level, they get that autism isn’t a categorically discrete “thing.”  At the very least, everyone knows that autism is on a dimension, a spectrum.  Nobody’s disputing that.

However, one response to the dimensionality we find in autism is to start searching for “autisms”: categorical subtypes of autism that are believed to underlie this heterogeneous dimensionality.

It’s not an entirely mistaken idea.  There are some categorical “autisms”, even if they might account for a minority of cases.  For example, not all people with Fragile X Syndrome have autism (the common biological processes of the syndrome[4] do eventually end in a somewhat heterogeneous behavioural phenotype), but many are autistic, and their common biological roots make them a categorical subtype.  There are other categorical genetic subtypes as well.

But what about the majority of autistic people who don’t have a clear genetic cause to their autism – whose autism is idiopathic?

For them, subtypes aren’t going to be categorical “autisms”.  They’ll be little slices of a heterogeneous, (multi-)dimensional continuum.  We talk as though we can find coherent, categorical subtypes, but mostly, what we’ll get will somewhat narrower subtypes of the same class as autism itself: socially constructed, heterogeneous subgroups.

Now, I will admit that it might sometimes be possible to define these subgroups on the basis of biology, not behaviour.  For example, research suggests that autistic preschoolers with unusually big heads are a meaningful subgroup (Amaral et al., 2017).  But unless we can narrow these groups down to a specific causal process, and find a biomarker of that causal process, these subgroups will only offer probabilistic information: they’ll only be able to predict that someone is more likely to have a given outcome.  That’s far from useless, but it’s not exactly a “marker.”

Furthermore, I suspect that behaviour will probably be more practically useful than biology when it comes to subtyping, particularly in the more immediate future.

Now, I want to be clear here that I do think there is a great deal of valuable information to be found in investigations of the neurological underpinnings of autism![5]  I’m currently working on some ERP research that is exploring the neurological underpinnings of autism; I wouldn’t do that if I didn’t think the research was valuable.  My point is that I don’t think it necessarily makes sense to frame such investigations as a search for “biomarkers.”  I think that label is likely to exaggerate our idea of the sensitivity and specificity of these biological factors.  Most of the time, when we do this research, I believe we’ll be uncovering information about possible neurological correlates of autism, not “markers” that would allow for diagnosis or the monitoring of autism symptom outcomes.[6]

Footnotes

[1] And this is, in my view, an excellent argument in favour of the idea of giving stakeholders like autistic people a role in decisions about research funding.

[2] Assuming that I’m right and that autism biomarkers per se are just a problematic concept, there’s a great deal of irony in the apparent fact that people doing autism research are so incentivized to find “biomarkers” that they would describe a non-biomarker as a biomarker!

[3] Which begs the question of why people are doing the studies at all.  Clearly something is wrong with how we incentivize research…

[4] Okay, the biological processes can sometimes be a bit heterogeneous as well when you take things like mosaicism into account.  But these details are irrelevant to the main point I’m trying to make here.

[5] And to a lesser extent, in investigations of the biological underpinnings – but I think we spend too much on this biology at present.

[6] A really logical person could reply to this post by accusing me of assuming my conclusion.  They could accuse me of arguing that autism has no biomarkers because I say it is a socially constructed category with no biomarkers.  They could say that the question of whether autism has biomarkers is an empirical one, and I can’t assume the truth of my conclusion in my premises.  They could say I’m arguing in circles.

The concern that my argument might be circular deserves to be treated seriously, but I would like to take this opportunity to point out that we’ve been hunting for common biological processes and core cognitive factors in autism for a long time, and we have nothing to show for it.  Empirically, the grounds for believing in autism biomarkers are pretty thin.  Thus, I’m not exactly saying that there is no autism biomarker because there is no autism biomarker.  I’m saying there probably aren’t clinically useful biomarkers because a great deal of empirical searching for biomarkers has not only generally failed to turn up satisfactory evidence of biomarkers, but has shown us that biological heterogeneity in autism is very high.  This high heterogeneity does not encourage us in the search for biomarkers.

Furthermore, we have good reason to believe that autism is a socially constructed category.  There are massive historical and cross-cultural variations in autism prevalence, and at this point we have no grounds to attribute these prevalence variations to anything except social factors.  We also know that at least some of the variations in prevalence can be attributed to social factors.

References

Amaral, D. G., Li, D., Libero, L., Solomon, M., Van de Water, J., Mastergeorge, A., … Nordahl, C. W. (2017). In pursuit of neurophenotypes: The consequences of having autism and a big brain. Autism Research, 10(5), 711–722. https://doi.org/10.1002/aur.1755