Posted at 08:27h in Uncategorized by Eric Turkheimer 3 Comments
In his new book, Blueprint, and elsewhere in his recent writings, Robert Plomin has repeated a striking claim about the causal properties of DNA, in the form of polygenic risk scores:
GPSs are unique predictors in the behavioural sciences. They are an exception to the rule that correlations do not imply causation in the sense that there can be no backward causation when GPSs are correlated with traits. That is, nothing in our brains, behaviour or environment changes inherited differences in our DNA sequence.
There is a lot that could be said about this statement,starting with the fact that it is obviously incorrect, and proceeding to the odd stylistic quality of the theorizing in Plomin’s recent work. He must know that the statement is literally incorrect, but he doesn’t care. I think he means something like, “DNA precedes behavior, so behavioral differences are obviously not causing DNA differences, and that allows one to rule out certain alternative hypotheses about genetic causation, and although we could make that case here, we am not going to bother, because we don’t have to.”
Let’s think about genetic causation for a bit. The obvious model to apply, and the one which Plomin is using without thinking about it, is a “bottom up” model, which comes from biological and medical genetics. Think for a moment about how the Huntington’s gene causes the disease. It has been said a million times that genes aren’t capable of causing dementia and chorea in any kind of direct way. Genes code for protein synthesis, blah blah blah, then those proteins are involved in certain cellular processes, and those cells form organelles, which in turn form organs, and then brains and then motor systems, and eventually the disturbance of the motor and cognitive systems emerge. It’s a hierarchical system, with larger things composed of smaller ones, and causal effects bubbling up from below.
Plomin’s idea is that this is the way everything works.Genes are building blocks that go into the construction of phenotypes, even complex ones, and while it might be complicated to figure out the pathway from gene to phenotype, at least we can be certain that the phenotype isn’t causing the gene, ie, there is no top-down causation.
Except there is top-down causation. Consider the genetic causation of divorce. Genes cause protein synthesis, blah blah blah, and things bubble upward. Eventually you get people having relationships. But then (actually way before then, but I’ll get to that in a minute) something dramatic happens: an unhappy spouse sees a therapist, becomes convinced that her marriage is toxic, and that realization causes her to file for divorce.
As soon as that happens, the whole bottom-up model of genetic causation is broken for good, because now all the “see a therapist” genes are related to “get a divorce” phenotypes. And of course, that whole story is itself a gross exaggeration in a million ways. We know from modern network theory that the best way to conceptualize things like divorce is as a network of high-level phenotypes causing each other: personality has an effect on marital quality,which in turn has an effect on mood, which is related to alcohol use, which changes personality, etc. Every time one of these high-level phenotypic causes takes place, the relationship between the phenotype and the underlying genes gets scrambled.
And even that is a gross oversimplification, because the processes of causation among high level entities doesn’t begin in therapist’s chair: it happens in the gene networks,in the cells, in the organs, in the low-level nervous system. The real hierarchical system is a stack of intra-level networks, with entities causing other entities. Throughout development, entities are causing other entities to do things, and when that happens the smaller things that compose them go along for the ride. Phenotypic networks on top of other phenotypic networks, all the way down. The result is omnigenics, with all genes associated with complex outcomes. It’s not that the correspondence between individual units of DNA and behavioral phenotypes is complex; it’s that the very idea of the existence of a bottom-up correspondence has broken down.
There is a ton of excellent philosophy on the problems of bottom-up and top-down causation. My favorite paper on the subject is by CarlCraver and William Bechtel: “Top Down Causation Without Top Down Causes.” They imagine a hot dog vendor, Ignatius, pushing his hot dog cart across the street. The hot dogs inside the cart duly travel from one side to the other. What causes the hot dogs to move? Iggy does. Iggy’s intention to get from one side to the other causes the hotdogs to move; the chemical composition of the hot dogs has nothing to do with it. The hot dogs are just along for the ride.
So it is with genes associated behavior. When the person in the unhappy marriage decides to see a therapist, the genes come along like hotdogs in the cart. It’s ridiculous to say that the genes are causing her to go to the therapist, much less to get a divorce that is caused by the therapy. In my opinion this is just as true for depression as it is for divorce. The SNPs associated with depression don’t cause depression any more than the molecules in the hot dogs cause Iggy to cross the street. We don’t understand the causal processes that create depression, but they are certain to be high level cognitive and affective entities, not bottom-up consequences of individual genes or SNPs. That is why I am so pessimistic that contemporary blind-pig GWAS (and yes, it has already found a few truffles) is ever going to explain anything complex: it just isn’t possible to explain complex human causal networks in terms of the distant bottom-up correlations that GWAS “discovers”. Trying to do so is like trying to understand the economics of Iggy’s hot dog business by tracing the motions of the hot dogs inside his cart. The business explains the motion of the hot dogs, not the other way around.
Anyway, Plomin and von Stumm are wrong about the simplicity of genetic causation of complex phenotypes. Although it is obviously true that our behaviors aren’t going to cause our genes to change ignoring epigenetics, which is a different problem), variation in our phenotypes causes variation in our genes all the time. And isn’t genetic variation what we are usually trying to explain?