This review in Nature Neuroscience is excellent. I have never seen the issue of gene-environment interactions laid out so eloquently. Unfortunately, it is behind a subscription wall, so those of you not affiliated with a University may have to just live with this excerpt:
The recent history of psychiatric research that has measured genetic differences at the DNA sequence level can be divided into three approaches, each with its own logic and assumptions. The first approach assumes direct linear relations between genes and behaviour (Fig. 1a). The goal of this approach has been to correlate psychiatric disorders with individual differences in DNA sequence. This has been attempted using both linkage analysis and association analysis, with regard to many psychiatric conditions such as depression, schizophrenia and addiction. Although a few genes have accumulated replicated evidence of association with disorder, replication failures are routine and overall progress has been slow. Because of inconsistent findings, many scientists have despaired of the search for a straightforward association between genotype and diagnosis, that is, for direct main effects.
Click to enlarge
The second approach has sought to make more progress by replacing the disorder outcomes with intermediate phenotypes, called 'endophenotypes' (Fig. 1b). Endophenotypes are heritable neurophysiological, biochemical, endocrinological, neuroanatomical or neuropsychological constituents of disorders. Endophenotypes are assumed to have simpler genetic underpinnings than disorders themselves. Therefore, this research approach pursues the hypothesis that it will be easier to identify genes associated with endophenotypes than genes associated with their correlated disorders. Although this approach substitutes the psychiatric diagnosis with an intermediate brain measure, it still searches for direct main effects.
The third approach to psychiatric genetics, unlike the first two approaches, seeks to incorporate information about the environment (Fig. 1c). This gene-environment interaction approach differs fundamentally from the 'main-effect approaches', with regard to the assumptions about the causes of psychiatric disorders. Main-effect approaches assume that genes cause disorder, an assumption carried forward from early work that identified single-gene causes of rare Mendelian conditions. By contrast, the gene-environment interaction approach assumes that environmental pathogens cause disorder, and that genes influence susceptibility to pathogens. In contrast to main-effect studies, there is no necessary expectation of a direct gene-to-behaviour association in the absence of the environmental pathogen. The gene-environment interaction approach has grown out of two observations: first, that mental disorders have environmental causes; second, that people show heterogeneity in their response to those causes. (Citations have been removed.)
Considering that I study schizophrenia and in part the genetics of schizophrenia, I would point out that this new model of disease formation poses a lot of problems for us. It is very difficult sometimes to relate genes to diseases when there are so many other confounding factors. But as my boss likes to say, "The low lying fruit have been picked..."
Jake, thanks for this interesting post. I will definately check out the article when I'm back on campus. Would you be able to cite some main-effects studies on schizophrenia to compare with gene-environment interaction studies on schizophrenia? What confounds have you had to address in your own research?
I am not aware of any main-effects studies on schizophrenia because all the genes thus far identified for schizophrenia are not even close to 100% penetrant. On top of that we simply don't know enough about the etiology of schizophrenia to relate a gene to recognizable clinical features.
This is one of the problems we have doing research on this. First, only two genes identified thus far, COMT and DISC1, show coding mutations. The rest of them are mutations in noncoding regions that presumably act by modifying expression is some way. However, proving this has proven extraordinarily difficult.
Say you have 4 SNPs in one particular gene that are shown by linkage dysequilibrium to be associated with schizophrenia -- a solid piece of evidence. You want to find out whether that gene is differentially expressed in schizophrenia so you go out and get a horde of schizophrenic brains and check their expression. You find that your results are not statistically significant. (I know because I have done this with several of them.) The problem is that only a particular haplotype of those SNPs will be associated with changes in transcription with the other haplotypes being mitigated by epistatic effects or environment of whatever.
So you go back and genotype all those people and then relate their expression for that gene to their particular haplotype -- highlighting in particular the haplotype that is associated with schizophrenia risk. Only this time because you have a four SNPs in your haplotype your risk group has an N = 2 so you still have no statistical significance.
The fundamental problem in behavioral genetics is that we keeping finding noncoding SNPs as genetic risk factors. We think they do something but we can't prove it A) because you have to relate expression to haplotype and our cohorts aren't large enough to test complex haploytpes and B) because they are always mitigated by epistatic and environmental effects unless you have a very inbred population. This is why most of the successful studies for schizophrenia genetics have been done in Iceland.
Does that make sense? It can be pretty frustrating at times.
If you want to understand the logic that goes into these new behavioral genetics experiments there was a good paper that came out this year that showed a haplotype that is related to NRG1 expression. (NRG1 is also a genetic risk factor for schizophrenia.) The paper is Law, 2006. Let me know if you have more questions.