Thursday, July 12, 2018

"Yoking" Stratification. A working theory.

Recently, I've been trying to get a better handle on population stratification.  Some of this has to do with what I believe are erroneous "correlations" made by the use of polygenic risk scores.  I am under the assumption that these so-called correlations for various traits are due more to stratification issues.  Recently, there have been a few studies proposing stratification issues for apparent genetic associations that did not replicate.  I discussed this paper related to genetics and height in my previous post.
In a discussion on the intellectual bastion known as Twitter, someone sent me a link to a study discussing the issues of stratification and polygenic risk related to schizophrenia (hat tip to @Race_Realist) and it occurred to me that there might be at least a partial explanation for some of this "hidden" stratification.  Let me start by looking at this paper.
The authors discuss a bit of a perplexing finding related to using polygenic scores when working with probands of different ancestry than the ones that were used for the original polygenic scores.  In this case, they are contrasting the European subset with those of African descent:
The PRS for schizophrenia varied significantly between ancestral groups (p < 2*10-16) and was much higher in African than European HapMap subjects. The mean difference between these groups was ten times as high as the mean difference between European schizophrenia cases and controls. The distributions of scores for African and European subjects hardly overlapped. 
This is interesting in and of itself, since schizophrenia is fairly uniform worldwide, generally estimated at 1 to 2% of the population.  The above would seem to imply that these two groups have schizophrenia due to very different sets of genes.  It is highly unlikely that two entirely different polygenic mechanisms, would confer identical symptoms, in approximately the same population frequency, with no apparent adaptive purpose.  The authors suggest that this finding is due to population stratification:
The PRS cannot be regarded as simply a measure of the polygenic contribution to risk of schizophrenia and clearly contains a strong ancestry component. It is possible that this could be controlled for to some extent by incorporating principal components as covariates but doubts remain as to how it should be interpreted. The PRS derived from European subjects cannot be applied to non-Europeans, limiting its potential usefulness and raising issues of inequity. Previous studies which have used the PRS should be re-examined in the light of these findings. 

In other words, many of the claimed genetic associations found in the GWAS studies used for deriving polygenic risk scores, might really be genetic markers for particular racial groups rather than having anything to do with schizophrenia.  The authors seem to hold onto the validity of PRS in this case for those of European ancestry but suggest that it might not be transferrable to other cultural groups, even if one tries to correct for stratification.  They otherwise seem perplexed by the finding and aren't really able to offer an explanation for the PRS disparities beyond "ancestry".  After thinking about this for a time, it occurred to me that there might be a simple explanation that could at least partially explain this mystery.  I am referring to this as Yoking Stratification.  The premise is that like attracts like.
Of course, this, by itself, is not a particularly original idea.  It's well known that people from the same cultural/racial backgrounds tend to marry within their culture as do those from the same geographical locations.  This is the basis of the commercial ancestry DNA companies which track genetic markers for racial backgrounds.  So, extending that idea, one might assume that individuals with similar traits will tend to marry.  Say, for example, people with high IQ's will be more likely to marry others with high IQ's, and the same applies to low IQ's.  Tall people tend to marry other tall people,  etc.  I wondered to what extent that might apply to people with mental illness and, as this study shows, that is indeed the case.  Schizophrenics marry other schizophrenics at a higher frequency and, to a significant  but lesser extent, they also marry people with bipolar disorder and other mental illnesses.
Most of these traits, including schizophrenia, have a heritability component.  This is where I think we are being a bit misled by the assumption that the SNP's found in GWAS studies actually confer causation.  If we let go of that idea and assume that, due to the failure of replication of these studies, these genetic associations are not related in any way to the trait itself (I know it's hard for some to put themselves in that mindset, but humor me), then we can look at them almost in the same way that we look at the data from Ancestry.com or 23andMe.
Years of individuals with similar traits being more likely to "interbreed", will lead to them having specific genetic markers that confer no more to the trait than most of the genetic markers for racial heritage.  This would give us a stratification for each trait.  So individuals with schizophrenia will accumulate some common genetic markers that really aren't related to schizophrenia.  This might explain why there is such a disparity between the genetic markers for European and African probands.  In both cases, there was some likely, like attracts like, leaving them with specific genetic markers, but since most still married within their own race, you would have a different sets of genetic markers within each culture.  Again, in both cases, the genetic markers are not causal to the trait.
The same idea might explain the recent issue related to height GWAS studies.  Tall people were breeding with other tall people and the presumed causal genetic loci were simply markers amongst a particular group of tall people, that did not have anything to do with height, per se, but which probably wouldn't even transfer to tall people in other racial backgrounds.
This might not sit well with our current assumptions, but we have GWAS studies that consistently fail to replicate.  We have polygenic risk scores that only work in certain cultures.  This model of "Yoking Stratification" is a possible explanation for at least some of these contradictory results.

Addendum:  Here is another study from the UK that looks at this problem via socioeconomic status, education, and body mass index and found stratification related to geographic location.  This is also consistent with my theory:  Rich people usually marry rich people, heavy people more often marry heavy people and educated people more often marry other educated people and they are more likely to be from the same geographical location.  Again, you have a like attracts like stratification which regionalizes with these genetic markers, but probably has no causal genetic effect.

Also, here is another paper discussing the matter.

No comments:

Post a Comment