The "omnigenic" model for schizophrenia - why it's even worse than you think.

I'm happy to see the article by Boyle, Yang and Pritchard (Boyle et al, 2017) getting so much attention. Quite justifiably, many people are commenting on it and I thought I would add my two pence worth.

To summarise, I think the main claim they are making is that for a complex disease there will be a relatively small number of "core" genes whose functioning is related to disease mechanisms but that a very large number of polymorphisms will impact to a small extent on these core genes through gene expression networks. This then leads to a phenomenon whereby GWAS hits are very widely scattered through the genome. Most of these hits will not be anywhere near genes involved in disease mechanisms, they'll just reflect impacts on generic expression networks in one way or another.

Here, I'd like to add another dimension, taking schizophrenia as an example. My claim will be that as well as the core genes there will be many other genes which do have an effect on risk, but through such an indirect route such that their identification would not really add anything useful to the understanding of disease processes.

Let's begin with some examples of things which are thought to influence risk of schizophrenia. The evidence may be stronger for some than others but I think they're mostly pretty reasonable.

• Maternal famine
• Maternal viral or toxoplasma infection
• Probably anything with an adverse effect on foetal development
• Cannabis smoking
• De novo mutations in vulnerable genes
• Urban living
• Childhood abuse including physical abuse, sexual abuse and bullying

(The last one about childhood abuse may be a bit questionable. I'm keeping it in because psychologists seem keen on it even though I think it's possible that it reflects a genetic susceptibility, in that children who are to go on to develop schizophrenia may be intrinsically more likely to be abused and/or more likely to have abusive parents.)

So, logically, any  genetic variant which affects any of these traits will indirectly impact on schizophrenia risk. The effect may be small, but a main thrust of the Boyle et al paper is that, because GWAS samples are huge, very small effects will  nevertheless be statistically significant. If we accept these risk factors, we can see that a GWAS for schizophrenia should pick up variants associated with any of the following (among many others): impaired nutritional intake; wanting to keep pet cats; wanting to smoke cigarettes; not being nauseated by cannabis; generalised genomic instability with increased mutation rate; susceptibility to hay fever; any mild dysmorphism leading to increased bullying risk.

Thus there may be many genes, aside from "core" genes, which exert an effect on schizophrenia which is mediated through phenotype rather than through gene expression networks.

In fact, I would suggest that the situation is even worse. If we look at variants influencing the susceptibility to each of these traits we might or might not see a correlation with schizophrenia susceptibility. What might make things worse is the notion which Boyle et al promote, which is that large numbers of variants may act indirectly through gene expression networks. This means that individual variants could have diverse effects on schizophrenia risk. For example, an SNP allele might, through a gene expression network, be associated with impaired ability to absorb certain nutrients. Would it then be associated with increased risk of schizophrenia through an effect on maternal malnutrition? Maybe. But what if the same allele were also associated, perhaps through other expression networks, with a genetically mediated distaste for smoking? Then women possessing it might less likely to smoke in pregnancy, with a consequent reduction in risk.

Any given variant might be associated with a different pattern of effects across different risk factors, increasing some but decreasing others. Variants which increased risk of schizophrenia would be the ones which, through their gene expression networks and other indirect relationships, had an overall positive association with all the risk factors, known and unknown, considered together, in addition to any association they might have with "core" genes. Conversely, if one examined all the variants positively associated with any given risk factor one might not find an overall association with schizophrenia risk because their effects overall might be fairly evenly balanced.

I'm not going to give any opinion on the general value or otherwise of the GWAS methodology. I just want to point out that even when one does find significant, replicable associations these need not necessarily be telling us something of much relevance to pathogenesis.