Genetic approaches to schizophrenia have been applied for over 40 years. Even casual readers in this area may be aware that dissecting the genetics of schizophrenia has been challenging. There have been many claims of identification of causal genetic risk factors for schizophrenia, and yet few claims prior to 2008 have stood the test of time and replication (83
). Such claims have occasionally engendered considerable media attention. As one example, evidence from large and carefully conducted studies do not support DISC1
as a risk factor for schizophrenia (27
). A prominent exception that has shown robust and replicable associations with schizophrenia is the 22q11.2 del CNV present in approximately 0.3% to 2% of schizophrenia cases (34
We believe these difficulties and the attendant, unwelcome controversies have arisen for two main reasons. First, as discussed above, the genetic variants that increase risk for schizophrenia differ from what had often been assumed of a genetic architecture, with a prominent role for numerous loci of subtle effect. Despite significant efforts, there are no compelling examples of deterministic genetic variants acting in a Mendelian fashion as yet. From a statistical perspective, this genetic architecture implies that very large sample sizes are required (i.e., tens of thousands of cases) for any type of association study as well as for genome-wide linkage approaches (86
Thus, with the immense benefits of hindsight, the poor past replicability of genetic associations for schizophrenia can be at least partially understood on statistical grounds. If sample sizes are orders of magnitude too small, true associations will almost always be missed due to low power. Similarly, claims of association have an overwhelming probability of arising by chance and thus can be expected not to replicate. Since 2008, with the advent of far larger sample sizes, multiple genomic findings have replicated relatively well across samples.
A second source of controversy has arisen more recently. Multiple authors have argued that elucidation of genetic architectures for complex biomedical diseases (including schizophrenia) that are characterized by large numbers of common variants of subtle effects is effectively unhelpful (88
). The argument rests on two main points. The first point is the “missing heritability” objection that the identified variants explain only a small fraction of the variance in liability to a disorder. The counterargument is that if the genetic architecture has thousands of variants (as appears to be the case for many complex diseases) (15
), models containing a small fraction of the thousands of truly associated variants could only explain small amounts of variation. Indeed, if modeled more inclusively, common variation can explain substantial fractions of the variance in liability, so that, effectively, variation is “hidden” by inadequately powered studies rather than “missing” (15
The second point is that elucidation of complex genetic architectures is unhelpful for attaining the goals of personalized (“precision”) medicine. The crux of this objection is the intention of conducting genomic searches: to generate rational biological starting points for idiopathic psychiatric disorders or to improve disease prediction and tailor treatments? The answer here depends on the researcher. Personalized medicine in regard to schizophrenia will be a difficult goal, given its genetic architecture. We note that this was predictable: the risk of schizophrenia to the co-twin of a monozygotic twin with schizophrenia is only around 50% (7
), and this single observation strongly limits the potential role of genetic variation for personalized medicine. However, for many in the field, the primary goal of carrying out genetic studies has been biology. Each genetic finding is a potential etiological clue. Many of these findings will fall into a discrete number of biological, developmental, or functional pathways whose elucidation and characterization could dramatically advance knowledge of the pathophysiology of schizophrenia. As early examples of this, transcriptional modules with altered gene expression in postmortem brain samples of individuals with schizophrenia (90
) and autism (91
) were enriched for smaller P
values in GWAS.