Ann L. Collins

Hypothesis-driven candidate genes for schizophrenia compared to genome-wide association results

BACKGROUND Candidate gene studies have been a key approach to the genetics of schizophrenia (SCZ). However, the results of these studies are confusing and no genes have been unequivocally implicated. The hypothesis-driven candidate gene literature can be appraised by comparison with the results of genome-wide association studies (GWAS). METHOD We describe the characteristics of hypothesis-driven candidate gene studies from the SZGene database, and use pathway analysis to compare hypothesis-driven candidate genes with GWAS results from the International Schizophrenia Consortium (ISC). RESULTS SZGene contained 732 autosomal genes evaluated in 1374 studies. These genes had poor statistical power to detect genetic effects typical for human diseases, assessed only 3.7% of genes in the genome, and had low marker densities per gene. Most genes were assessed once or twice (76.9%), providing minimal ability to evaluate consensus across studies. The ISC studies had 89% power to detect a genetic effect typical for common human diseases and assessed 79% of known autosomal common genetic variation. Pathway analyses did not reveal enrichment of smaller ISC p values in hypothesis-driven candidate genes, nor did a comprehensive evaluation of meta-hypotheses driving candidate gene selection (SCZ as a disease of the synapse or neurodevelopment). The most studied hypothesis-driven candidate genes (COMT, DRD3, DRD2, HTR2A, NRG1, BDNF, DTNBP1 and SLC6A4) had no notable ISC results. CONCLUSIONS We did not find support for the idea that the hypothesis-driven candidate genes studied in the literature are enriched for the common genetic variation involved in the etiology of SCZ. Larger samples are required to evaluate this conclusion definitively.

Transcriptional targets of the schizophrenia risk gene MIR137

Genome-wide association studies (GWAS) have strongly implicated MIR137 (the gene encoding the microRNA miR-137) in schizophrenia. A parsimonious hypothesis is that a pathway regulated by miR-137 is important in the etiology of schizophrenia. Full evaluation of this hypothesis requires more definitive knowledge about biological targets of miR-137, which is currently lacking. Our goals were to expand knowledge of the biology of miR-137 by identifying its empirical targets, and to test whether the resulting lists of direct and indirect targets were enriched for genes and pathways involved in risk for schizophrenia. We overexpressed miR-137 in a human neural stem cell line and analyzed gene expression changes at 24 and 48 h using RNA sequencing. Following miR-137 overexpression, 202 and 428 genes were differentially expressed after 24 and 48 h. Genes differentially expressed at 24 h were enriched for transcription factors and cell cycle genes, and differential expression at 48 h affected a wider variety of pathways. Pathways implicated in schizophrenia were upregulated in the 48 h findings (major histocompatibility complex, synapses, FMRP interacting RNAs and calcium channels). Critically, differentially expressed genes at 48 h were enriched for smaller association P-values in the largest published schizophrenia GWAS. This work provides empirical support for a role of miR-137 in the etiology of schizophrenia.

Comprehensive analysis of copy number variation in monozygotic twins discordant for bipolar disorder or schizophrenia

Copy number variation plays a clear role in the etiology of many psychiatric disorders, particularly schizophrenia. We performed array-CGH to look for copy number variants between five pairs of monozygotic twins discordant for bipolar disorder or schizophrenia. Our study found no differences in copy number variants between the sets of twins. Although alluring, realistic accounting for heterogeneity and chimerism highlights the technological limitations in studying monozygotic twins discordant for psychiatric disorders.

Disruption of the MicroRNA 137 Primary Transcript Results in Early Embryonic Lethality in Mice

To the editors: Multiple lines of evidence support a role for microRNA 137 (miR-137) in the etiology of schizophrenia (1–4) and fundamental neuronal processes (5–7). In the largest genome-wide association meta-analysis for schizophrenia,(8) the second most significant association is in MIR137HG, the gene encoding miR-137 (rs1702294, P=3.4×10−19). Prior reports indicated that genes with predicted miR-137 target sites were enriched for smaller GWAS P-values (2), raising the possibility that miR-137 regulates a gene network involved in schizophrenia. To understand more about the role of miR-137, we describe here our characterization of embryonic development, behavior, and gene expression in mice with targeted disruption of the Mir137 transcript (supplementary detailed information can be found at: http://crowley.web.unc.edu/files/2014/03/mir137.ko_.supplement.pdf). Mir137tm1Mtm mice were created in a project to generate conditional, reporter-tagged targeted mutations in 162 miRNAs (9). These mice were created using a “knockout-first” strategy (10) to produce a knockout at the RNA processing level. We purchased heterozygous (+/−) breeder mice (www.mmrrc.org/catalog/sds.php?mmrrc_id=36301). We confirmed genetic background (~75% C57BL/6J) and backcrossed twice to C57BL/6J to create animals of >95% C57BL/6J ancestry. To our knowledge, these Mir137 mutant mice have not been characterized previously. Seven heterozygous × heterozygous matings produced 35 living offspring, none of which were homozygous (−/−) for the targeted Mir137 allele (Figure 1A, P < 0.001), suggesting complete embryonic lethality in the presence of knockout of Mir137. Figure 1 Mir137 knockout mice (−/−) are embryonic lethal. (A) Heterozygous × heterozygous matings failed to yield a live born (−/−) mouse or embryo, and timed matings revealed an excess of resorbed embryos at E11.5. Heterozygous ... To confirm embryonic lethality and to determine when lethality occurred, we conducted timed matings (six heterozygous × heterozygous and four heterozygous × wildtype). On embryonic day 11.5 (E11.5), all embryos were collected (Figure 1A). Again, no homozygous (−/−) embryos were identified (P < 0.001). We observed significantly more resorbed embryos from heterozygous × heterozygous matings (P < 0.0001), indicating that embryonic lethality must occur after implantation (~E4.5) but before E11.5. Efforts to identify a Mir137 −/− mouse by genotyping resorbed embryos at E11.5 or collecting flushed embryos at E3.5 were unsuccessful, likely due to insufficient and/or impure DNA preparations from small amounts of tissue. We did not observe reduced viability of heterozygous mice (+/−) after genotyping 256 offspring (P=0.14) collected from birth to three weeks of age, indicating that one copy of Mir137 is sufficient for survival (Figure 1A). We subjected wild-type and heterozygous animals to a battery of behavioral tests (e.g., basic sensorium, activity, social behavior, learning/memory, etc.), and observed no consistent and interpretable differences. These results also suggest compensation from the remaining allele. We next examined the expression of Mir137 and the levels of mature miR-137 in the brains of heterozygous and wildtype animals (−/− embryos died too early for collection of sufficient tissue). Using allele-specific reverse transcriptase PCR, we confirmed that the targeted allele leads to complete loss of Mir137 expression downstream of the polyA sequence (Figure 1B). However, mature miR-137 levels were not significantly different between heterozygous and wild-type animals (Figure 1C), suggesting that heterozygous mice have compensatory upregulation or decreased degradation of miR-137. These expression data are consistent with the lack of a viability or behavioral phenotype in the heterozygous mice. To circumvent the embryonic lethality of Mir137 −/− mice, we attempted to make conditional knockouts (9), such that ablation of Mir137 could be limited to a particular tissue. The construct contains a combination of FRT and loxP sites intended to allow conditional mutagenesis by crossing to germline deleter Flp mice (expected to restore the wild-type allele) and then crossing to tissue-specific Cre transgenic mice (expected to generate adult brain-specific knockouts). Crosses between Mir137 +/− mice and an efficient Flp line led to several successfully recombined mice, heterozygous for the rescued allele. Repeated intercrosses of these recombined mice, however, failed to yield any homozygous offspring. Therefore, we were unable to rescue embryonic lethality. We reasoned that there were two major possibilities for the rescue failure. First, the targeting construct could have inserted into the wrong genomic position or could have sequence errors. We have likely excluded this possibility by confirming the location of the integration site and re-sequencing >95% of the construct without identifying an error. Second, Flp may have been unable to restore proper gene function due to issues inherent with the gene construct (e.g., FRT or LoxP sites in critical part of the gene). We favor this possibility because we were unable to detect any reporter gene (beta-galactosidase) activity in Mir137 +/− embryos or adult tissues, suggesting that the endogenous Mir137 promoter was inadvertently inactivated. Furthermore, bioinformatic analysis of the region (using additional data that became available after we were well into this project), suggests that the exogenous gene targeting elements may indeed interfere with endogenous functional elements, and act to prevent conditional mutagenesis (Figure 1D). The LoxP site upstream of Mir137 is in a putative splice donor sequence and the FRT site lies between highly conserved DNase hypersensitivity sites that are active in embryonic brain. In conclusion, these results suggest that at least one functional copy of Mir137 is essential for embryonic development. These data are consistent with miR-137 playing important roles in development and perhaps also in neurodevelopmental disorders like schizophrenia. It appears that miR-137’s biological pathway is capable of homeostatic compensation and, while we do not yet understand the functional impact of Mir137 schizophrenia risk variants, it is conceivable that this homeostatic capacity could be affected in some way. It is clear, however, that this genomic region is more complex than previously thought, and the design of targeted mutations in this region should incorporate all available genomic data.

Identifying bipolar disorder susceptibility loci in a densely affected pedigree

Bipolar disorder (BIP) is a highly heritable disorder with complex patterns of genetic inheritance, and recent genetic findings highlight the role of numerous common variants each with subtle effects.1 The existence of Mendelian subtypes of BIP (rare variants of very strong effects) has been postulated, particularly as such variants could prove to be more tractable for subsequent biological investigation.2 One way to evaluate this hypothesis is via the study of pedigrees densely affected with BIP in which a genetic variant of strong effect inherited from a common ancestor may be more likely. With the advent of high-throughput technologies, we can now search densely-affected pedigrees for specific variants that may contribute to risk for BIP. We therefore evaluated a Spanish multi-generational pedigree with an exceptional prevalence of BIP using multiple complementary genomic techniques (Table S1). This pedigree contains 18 cases with BIP (including a sibship with six of 11 affected) and seven individuals with recurrent major depressive disorder (Figure 1a). The lifetime prevalence of mood disorders in this large pedigree (6 generations; 120 individuals, 30 with known mood disorders, 42 with DNA) makes it a strong candidate for identifying genetic risk factors of near-Mendelian effects. Our search strategy is depicted in Figure 1b. All protocols were IRB approved and all subjects provided written informed consent. Figure 1 (a) The pedigree was ascertained in Spain, and has a high prevalence of mood disorders, particularly bipolar disorder (type 1) and recurrent major depressive disorder. First, we used genome-wide linkage analysis and an affecteds-only approach to identify genomic regions sharing identity-by-descent. Linkage analysis was performed using microsatellite data from 13 individuals.3 We used data from Illumina HumanOmni Quad genotyping arrays to identify shared segments using Beagle,4 Germline,5 PedIBD,6 and runs of identity-by-state.7 Regions identified by linkage or with two sharing methods were considered candidate regions (Table S2, six regions totaling 93.7Mb). Second, we hypothesized that one of these regions contained a novel, rare functional single nucleotide variant (SNV) of high penetrance. We attempted to identify novel SNVs in these six genomic regions shared by BIP cases using exome sequencing (five BIP cases) and whole genome sequencing (three BIP cases). Whole genome sequence data was screened to identify high-quality homozygous or heterozygous SNVs that were within the candidate regions, novel, of predicted functional consequence, and present in all three BIP cases. This procedure identified 26 SNVs in three olfactory receptors (OR4C3, OR9G9, and OR4C45). Because multiple variants were present in each gene, given the existence of other highly similar members of the large olfactory gene family, and manual review of alignment patterns, we believed all to be due to misalignment. We repeated this process for the exome sequencing data, and identified 21 SNVs (17 were also identified by whole genome sequencing), and all were in olfactory genes (OR4C3, OR9G9, and OR8U8/OR8U1), and likely due to incorrect alignment. We next evaluated CNVs in these regions. Using PennCNV8 calls from Illumina arrays, no novel CNVs were identified in these candidate regions that were >1kb, present in ≥8 of 11 BIP cases, and confirmed via whole genome sequencing. Therefore, we were unable to identify any potential SNVs or CNVs within these candidate regions that were promising for follow-up. Third, we then extended our analysis genome-wide. From the whole genome sequencing of three individuals, we identified novel SNVs of predicted functionality that were present in all three BIP cases (Table S3). We excluded SNVs in olfactory receptors, with questionable alignments, or that were not confirmed in exome sequencing of five BIP cases. This resulted in eight SNVs that we sequenced in 11 affected individuals using Sanger sequencing. Five SNVs verified and had plausible inheritance patterns, and were then genotyped in 42 individuals in the pedigree. Two SNVs were potentially interesting (chr12:52452495 C>T in NR4A1 and chr18:47793974 G>C in MBD1). Neither exhibited unequivocal Mendelian inheritance (Figures S2–S3). The NR4A1 (nerve growth factor IB, NGFIB or Nur77) SNV is novel but common in this pedigree (homozygous or heterozygous in 30/34 individuals descended from the founders, excluding married-in individuals) and did not clearly segregate in BIP cases as the SNV was found in 92% of BIP cases and 86% of subjects without a mood disorder (although the number of unaffecteds is small). The MBD1 (methyl-CpG-binding domain protein 1) SNV tracked with BIP (66%), other mood disorders (88%), and less so in unaffecteds (21%). However, this SNV was recently identified by the 1000 Genomes Project in subjects from the United Kingdom. Therefore, this variant is not completely novel, but has an interesting pattern of segregation. Both variants warrant additional follow-up in a larger case-control samples although neither appears to be a strong novel Mendelian variant. Genome-wide expansion of the CNV search identified no novel CNVs that were present in ≥8 BIP cases, >1kb in size, overlapped an exon, and verified by whole genome sequencing. Fourth, given the absence of compelling results in support of a near-Mendelian variant, we evaluated the contribution of common variation in this pedigree. The presence of many common variant risk alleles in a pedigree is a potential explanation for a dense pedigree.9 This can be due to or exacerbated by assortative mating (in which mental illness in the family of a spouse is more likely to be tolerated if one has it in one’s own family) and result in an accumulation of common risk alleles. We therefore calculated risk profile scores, which are the weighted number of BIP risk alleles in each subject. Based on results from the Psychiatric GWAS Consortium10 (7,481 BIP cases and 9,250 controls), risk profile scores were computed for the GAIN BIP cases (N=1080) and controls (N=1058),11 and the 11 BIP cases in the densely-affected pedigree (Figure 1c). As anticipated, the GAIN BIP cases had significantly higher risk profile scores than the GAIN controls (p < 0.001). Risk profile scores for 11 BIP cases from the Spanish pedigree were also significantly greater than controls (p < 0.001) but not significantly different from GAIN BIP cases (p = 0.19). It is particularly notable that the BIP pedigree cases did not have markedly lower risk profile scores (e.g., low common variant profiles might be consistent with the present of a strong mutation). Rather, the BIP pedigree cases appeared to have common variant risk profile scores similar to European-American BIP cases ascertained clinically without regard to family history. In conclusion, we systematically assessed this large pedigree dense with BIP for genetic variants of strong effect. This comprehensive analysis did not conclusively identify any SNVs or CNVs of near-Mendelian effect. However, we cannot exclude the presence of risk variants with more complex inheritance patterns, variants with more cryptic functional effects, or variants missed due to coverage or individuals sequenced. However, the common variant risk profiles of BIP cases in this pedigree are similar to those of BIP cases ascertained without regard to family history. Therefore, it is possible that the etiology of BIP in this pedigree is more related to multiple common risk variants rather than one or a few variants of extremely strong effect.

Deep Sequencing of Three Loci Implicated in Large-Scale Genome-Wide Association Study Smoking Meta-Analyses.

INTRODUCTION Genome-wide association study meta-analyses have robustly implicated three loci that affect susceptibility for smoking: CHRNA5\CHRNA3\CHRNB4, CHRNB3\CHRNA6 and EGLN2\CYP2A6. Functional follow-up studies of these loci are needed to provide insight into biological mechanisms. However, these efforts have been hampered by a lack of knowledge about the specific causal variant(s) involved. In this study, we prioritized variants in terms of the likelihood they account for the reported associations. METHODS We employed targeted capture of the CHRNA5\CHRNA3\CHRNB4, CHRNB3\CHRNA6, and EGLN2\CYP2A6 loci and flanking regions followed by next-generation deep sequencing (mean coverage 78×) to capture genomic variation in 363 individuals. We performed single locus tests to determine if any single variant accounts for the association, and examined if sets of (rare) variants that overlapped with biologically meaningful annotations account for the associations. RESULTS In total, we investigated 963 variants, of which 71.1% were rare (minor allele frequency < 0.01), 6.02% were insertion/deletions, and 51.7% were catalogued in dbSNP141. The single variant results showed that no variant fully accounts for the association in any region. In the variant set results, CHRNB4 accounts for most of the signal with significant sets consisting of directly damaging variants. CHRNA6 explains most of the signal in the CHRNB3\CHRNA6 locus with significant sets indicating a regulatory role for CHRNA6. Significant sets in CYP2A6 involved directly damaging variants while the significant variant sets suggested a regulatory role for EGLN2. CONCLUSIONS We found that multiple variants implicating multiple processes explain the signal. Some variants can be prioritized for functional follow-up.