Modern genotyping technologies and statistical analyses have enabled the discovery of genetic loci related to the etiology of many complex traits [11], such as Type 2 diabetes mellitus, obesity, and Alzheimer's disease. These “discovery science” approaches have been applied to schizophrenia, and are summarized in Figure 2. The 27 samples shown here included from one to 294 multiplex pedigrees (see Glossary) (median 34) containing 32 to 669 (median 101) individuals affected with a narrow definition of schizophrenia. There were 310 to 950 (median 392) genetic markers in the first-stage genome scans.
The x-axis shows the location on the genome, from the telomere of the short arm of Chromosome 1 to the telomere of the long arm of Chromosome 22 (bottom row) along with 303 band chromosomal staining on the second-to-bottom row. The y-axis shows the 27 primary samples that reported first-stage genome scans for schizophrenia (i.e., excluding fine-mapping or partial reports) along with the results of a meta-analysis including most of the primary samples [12] (studies not included are shown with asterisks). Within each row, the height and color of the bars are proportional to the –log10(P-value), and the width of the bar shows the genomic location implicated by a particular sample. A selected set of candidate genes for schizophrenia are also shown. All genomic locations are per the hg16 build (http://genome.ucsc.edu). The physical positions of an inclusive set of the markers showing the best findings in the primary samples were plotted (assuming a confidence interval of ± 10 cM or, if mapping was uncertain, ± 10 megabases; seven markers from the primary samples did not map).
“Hard” replication—implication of the same markers, alleles, and haplotypes in the majority of samples—is elusive. It is evident from Figure 2 that these studies are inconsistent, and no genomic region was implicated in more than four of the 27 samples. The Lewis et al. meta-analysis [12] included most of the studies in Figure 2 and found that one region on Chromosome 2 was stringently significant and several additional regions neared significance. Our focus on first-stage genome scans does not adequately capture the evidence supporting replication for certain regions (e.g., 6p) [13–18]. However, there appears to be “soft” replication across studies.
It is unlikely that all of these linkage findings are true. The regions suggested by the Lewis et al. meta-analysis implicate more than 3,000 genes (18% of all known genes). For the 27 samples in Figure 2, the percentages of all known genes implicated by 0, 1, 2, 3, and 4 linkage studies were 42%, 35%, 14%, 6%, and 3%, respectively. This crude summation suggests that linkage analysis is an imprecise tool—implausibly large numbers of genes are implicated and few genes are consistently identified in more than a small subset of studies.
There are several potential reasons why clear-cut or “hard” replication was not found. With respect to the teams that conducted these enormously effortful studies, it is possible that no study possessed sufficient statistical power to detect the subtle genetic effects suspected for schizophrenia. For example, it would require 4,900 pedigrees to have 80% power to detect a locus accounting for 5% of variance in liability to schizophrenia at α = 0.001. These calculations make highly optimistic assumptions, and less favorable assumptions can lead to sample size requirements above 50,000 sibling pairs. For comparison, the total number of pedigrees in Figure 2 is less than 2,000.
In addition, it is possible that etiological heterogeneity (different combinations of genetic and environmental causes between samples) and technical differences (different ascertainment, assessment, genotyping, and statistical analysis between samples) contributed; however, their impact is uncertain, whereas insufficient power is clear. If correct, the implication is that Figure 2 contains a mix of true and false positive findings.
Answers to all your Biology Questions
