Targeted sequencing of the LRRTM gene family in suicide attempters with bipolar disorder

Abstract

Glutamatergic signaling is the primary excitatory neurotransmission pathway in the brain, and its relationship to neuropsychiatric disorders is of considerable interest. Our previous attempted suicide genome-wide association study, and numerous studies investigating gene expression, genetic variation, and DNA methylation have implicated aberrant glutamatergic signaling in suicide risk. The glutamatergic pathway gene LRRTM4 was an associated gene identified in our attempted suicide genome-wide association study, with association support seen primarily in females. Recent evidence has also shown that glutamatergic signaling is partly regulated by sex-related hormones. The LRRTM gene family encodes neuronal leucine-rich transmembrane proteins that localize to and promote glutamatergic synapse development. In this study, we sequenced the coding and regulatory regions of all four LRRTM gene members plus a large intronic region of LRRTM4 in 476 bipolar disorder suicide attempters and 473 bipolar disorder nonattempters. We identified two male-specific variants, one female- and five male-specific haplotypes significantly associated with attempted suicide in LRRTM4. Furthermore, variants within significant haplotypes may be brain expression quantitative trait loci for LRRTM4 and some of these variants overlap with predicted hormone response elements. Overall, these results provide supporting evidence for a sex-specific association of genetic variation in LRRTM4 with attempted suicide.

1 |. INTRODUCTION

The glutamatergic signaling pathway is the main excitatory neurotransmission pathway in the brain and is involved in a number of cell processes, including cell survival, proliferation, and intracellular calcium signaling (Hachem, Mothe, & Tator, 2016; Jeong et al., 2017; Magi, Piccirillo, & Amoroso, 2019). There are multiple glutamate transporters, receptors, and downstream effector molecules that amplify glutamate signaling (Macht, 2016; Ohgi, Futamura, & Hashimoto, 2015; Ribeiro, Vieira, Pires, Olmo, & Ferguson, 2017; Rose et al., 2017). The NMDA receptor family, for example, binds glutamate and affects intracellular calcium concentrations, and is also a target for ketamine, a drug known to reduce acute suicidal ideation and relieve symptoms of major depressive disorder (MDD) within minutes of administration (Davies, Alford, Coan, Lester, & Collingridge, 1988; Kraus et al., 2017; Wilkinson et al., 2018). Because glutamatergic signaling is widespread and has a variety of cellular functions depending on brain cell population, brain region, and environmental influences, regulation of glutamatergic signaling is of great interest in psychiatric research.

Known but less well-studied glutamatergic signaling regulators are gonadal hormones, including estrogen and testosterone (Wei et al., 2014; Wright & McCarthy, 2009). Gonadal hormones have been shown to regulate glutamatergic signaling in multiple brain regions including the prefrontal cortex (PFC), amygdala, hypothalamus, cerebellum, hippocampus, and somatosensory cortex (De Jesus-Burgos, Gonzalez-Garcia, Cruz-Santa, & Perez-Acevedo, 2016; Galvin & Ninan, 2014; Hedges, Ebner, Meisel, & Mermelstein, 2012; Oberlander & Woolley, 2016; Qiu et al., 2018; Wei et al., 2014). It was first demonstrated that around 50–60% of neurons expressing the NMDA glutamate receptor also expressed estrogen receptor mRNA, indicating both receptors were present in the same neurons (Kia, Yen, Krebs, & Pfaff, 2002). More recently, estrogen was shown to increase the expression of glutamate receptor GluR1 and increase spine density and glutamatergic synapse formation in both PFC and superficial superior colliculus neurons (Khan, Dhandapani, Zhang, & Brann, 2013). Further, estrogen has been demonstrated to regulate neurotransmission in the cerebellum and increase glutamatergic synaptic transmission in the hippocampus, but through different mechanisms in males and females (Hedges et al., 2018; Oberlander & Woolley, 2016). This sex-specific difference highlights how the same functional endpoint can be achieved in a cell through different pathways, and that the functional endpoint can be perturbed differently based on sex.

Glutamatergic signaling has been implicated in multiple psychiatric disorders including schizophrenia, bipolar disorder, MDD, post-traumatic stress disorder, and obsessive–compulsive disorder (Averill et al., 2017; Gray, Hyde, Deep-Soboslay, Kleinman, & Sodhi, 2015; Ng, Lau, Graham, & Sim, 2009; Rajendram, Kronenberg, Burton, & Arnold, 2017). Our lab is interested in the genetic basis of suicide, in the context of psychiatric illness. In order to better understand genetic influences on suicide risk, our lab previously conducted a genome-wide association study (GWAS) of attempted suicide within bipolar disorder subjects (Willour et al., 2012). Intriguingly, several of our top findings from this GWAS implicated genes involved in glutamatergic signaling. These results are consistent with previous findings that glutamatergic signaling has been associated with suicidal behavior. Dysregulation of the glutamatergic signaling pathway has been implicated in suicidal behavior through gene expression (Gray et al., 2015; Sequeira et al., 2009; Zhao et al., 2018) and epigenetic studies (Nagy et al., 2015). Genetic variation in glutamatergic genes has also been previously associated with suicidal behavior (Sokolowski, Ben-Efraim, Wasserman, & Wasserman, 2013; Willour et al., 2012). It has been shown that glutamatergic genes, although not any LRRTM family genes, are upregulated in postmortem brains of female MDD suicide patients, but upregulated to a lesser extent in male MDD suicide patients, indicating that gene expression in suicide completers can differ based on sex (Gray et al., 2015).

As part of our attempted suicide GWAS, we performed sex-specific analyses of our genotypic data. The top finding from the female-specific analysis implicated the glutamatergic gene LRRTM4 in suicidal behavior. This gene is also found within the 2p11–12 attempted suicide linkage peak that was identified in four different studies investigating attempted suicide within bipolar disorder (Willour et al., 2007), alcoholism (Hesselbrock et al., 2004), and major depression (Butler et al., 2010; Zubenko et al., 2004). LRRTM4 is part of a four-gene family, LRRTM1–4, and genetic variation in the LRRTM gene family has been implicated in several additional psychiatric and neurological disorders including schizophrenia (Francks et al., 2007; Leach, Prefontaine, Hurd, & Crespi, 2014), intellectual disability (Kleffmann et al., 2012), bipolar disorder (Malhotra et al., 2011), autism spectrum disorder (Michaelson et al., 2012; Pinto et al., 2010; Sousa et al., 2010), and Alzheimer’s disease (Majercak et al., 2006; Reitz, Conrad, Roszkowski, Rogers, & Mayeux, 2012).

The LRRTM gene family encodes neuronal leucine-rich repeat transmembrane proteins (Lauren, Airaksinen, Saarma, & Timmusk, 2003). The LRRTM genes are expressed primarily in neurons of the central nervous system, where they localize to excitatory synapses and promote glutamatergic synapse development (Lauren et al., 2003; Linhoff et al., 2009; Siddiqui, Pancaroglu, Kang, Rooyakkers, & Craig, 2010). In the current study, we chose to sequence the four members of the LRRTM gene family as well as the high linkage disequilibrium (LD) region of LRRTM4 that included the top female-specific variants from the suicide GWAS. Our goal was to investigate the contribution of both common and rare genetic variation within the LRRTM gene family to suicidal behavior, and to identify sex-specific variation in this association.

2 |. MATERIALS AND METHODS

2.1 |. Sample set

Our samples were obtained from the National Institute of Mental Health Bipolar Initiative (https://www.nimhgenetics.org/; “Genomic survey of bipolar illness in the NIMH genetics initiative pedigrees: a preliminary report,” 1997) and were a subset of our attempted suicide GWAS samples (Willour et al., 2012). The sample set for this study included 476 bipolar disorder suicide attempters (224 males and 252 females) and originally 476 bipolar disorder nonattempters (224 males and 252 females) of which 473 subjects were included in the final sample set, all of whom were unrelated and of European-American descent (Table S1). All of our subjects were diagnosed with either bipolar I disorder or schizoaffective bipolar disorder. The Diagnostic Interview for Genetic Studies (DIGS; Nurnberger et al., 1994) was used to define suicide attempters based on a positive response to the question, “Have you ever tried to kill yourself?” The DIGS suicide intent score is determined by an interviewer on a rating scale of 1–3. A score of 1 is defined as no or minimal intent, 2 is defined as definite intent, and 3 is defined as serious intent. We only selected attempters with definite to serious intent to die in their suicide attempt according to the DIGS. Institutional Review Board-approved consent was given by all participants prior to enrollment in this study.

2.2 |. Next-generation sequencing

We performed targeted sequencing of the coding and regulatory regions of the LRRTM gene family. Coding regions included all exons of all alternative transcripts, as defined by the GENCODE (Harrow et al., 2012), Ensembl (Cunningham et al., 2015), RefSeq (Pruitt et al., 2014), or “UCSC Genes” tracks in the UCSC Genome Browser (Kent et al., 2002). The regulatory target regions included all alternative promoter regions (defined as 2 kb upstream of the transcription start site), intron–exon boundaries (±50 bp), and any putative regulatory elements within the gene or within 10 kb (upstream or downstream) of the gene. Putative regulatory elements were defined as any region that was overlapping a DNase hypersensitivity site (UCSC track: wgEncodeRegDnaseClusteredV2) and transcription factor binding site (UCSC track: wgEncodeRegTfbsClusteredV3) according to ENCODE (Consortium, 2012). We additionally targeted the portion of intron 3 (∼318 kb) of LRRTM4 containing our previous top female-specific GWAS findings that was within a large LD block (Willour et al., 2012). These regions were targeted using sequencing probes designed by the Agilent SureDesign Custom Design Tool software (Santa Clara, CA).

We sequenced the targeted regions of the LRRTM gene family using the Agilent SureSelect^XT Target Enrichment system. Briefly, we sheared 3 μg of high-quality genomic DNA using a Covaris (Woburn, MA) E220 ultrasonicator. We used a Sciclone Next-Generation Sequencing Robotics Workstation (PerkinElmer: Waltham, MA) to run this sheared DNA through library preparation, hybridization, and target selection using standard protocols. Prepped samples were then multiplexed into pools of 16 samples and sequenced on an Illumina (San Diego, CA) HiSeq2000 as 100 bp paired-end reads at the Iowa Institute of Human Genetics, Genomics Division.

Alignment to the human genome (hg19/GRCh37) was done using the Burrows-Wheeler Aligner (v0.6.2; Li & Durbin, 2009). SAMtools (v0.1.18; Li et al., 2009) was used to create, sort, and index binary alignment (BAM) files and to produce summary statistics. Duplicate reads were removed using Picard (v1.88: http://picard.sourceforge.net), and unpaired, incorrectly paired, and unmapped reads were removed with BAMtools (v2.2.3; Barnett, Garrison, Quinlan, Stromberg, & Marth, 2011). BAMtools was also used to remove low quality reads (mapping quality <20). The Genome Analysis Toolkit (GATK: v3.1; McKenna et al., 2010) was used to measure the depth of coverage and to call indels and single nucleotide polymorphisms (SNPs) with all samples being called as one group. GATK was also used to filter out any variants failing the following quality control tests: quality-to-depth ratio, mapping quality, strand bias, and Hardy–Weinberg equilibrium (p < 1 × 10⁻⁶). Variants were set as “missing” if they had low coverage (<10X) or genotyping quality scores (<20), or if they were heterozygous calls in the X-chromosome of males or Y-chromosome calls in females. Any variants with high levels of “missingness” (>10%) in all subjects were removed from the dataset. Finally, any subjects that had high levels of mismatch with our attempted suicide GWAS calls were removed from the sample set (Willour et al., 2012). This resulted in the exclusion of three nonattempter subjects, so our final sample set consisted of 476 bipolar disorder suicide attempters and 473 bipolar disorder nonattempters (Table S1). After the exclusion of these three samples, the overall dataset from the current study had a high level of concordance (99.75%) with the previous GWAS study in which the female specific variant was identified. The final variant set was annotated with ANNOVAR (version March 22, 2015; Wang, Li, & Hakonarson, 2010), aligned with Ensembl v75 genes (Hubbard et al., 2002), and labeled with dbSNP142 rs numbers (Sherry et al., 2001), if available.

2.3 |. Statistical analyses

We performed three types of statistical analyses on the LRRTM gene family sequencing data: individual-variant, gene-level, and haplotype. Individual-variant tests were performed using Firth’s penalized logistic regression method (Firth, 1993) in the “logistf” R package (v1.21; Heinze & Puhr, 2010), including sex and the first three principal components from our attempted suicide GWAS as covariates. Age was not included as a covariate in this study as there was no significant difference (p-value = .75; Table S1; Gaynor et al., 2016; Willour et al., 2012). Individual variants were analyzed in the overall sample set and in sex-specific subsets (male-only, female-only). The individual-variant tests were permutation-corrected by running the analyses 10,000 times using randomly swapped attempter and nonattempter labels. Four genes: LRRTM1, LRRTM2, LRRTM3, and LRRTM4 were all included in the permutation, and variants were corrected for the all subjects, male, and female analyses. Permutation-corrected p-values less than .05 were considered statistically significant, and CADD scores were included for each of the top variants. The online database CADD v1.4 was used, and results are reported as PHRED scores that represent a log-based scale (Rentzsch, Witten, Cooper, Shendure, & Kircher, 2019). A value of 10 would indicate the value is in the top 10% of all CADD scored bases (more than 8.6 billion scores in their database), 20 would be in the top 1% and so on, for how damaging the variant is. The higher the score, the more likely to have a functional effect. Scores tend to be lower in noncoding areas due to the minimal annotations available for these regions.

We also assessed variation at the gene-level for the LRRTM gene family. These gene-based tests assessed rare (minor allele frequency [MAF] <0.05), functional variants in both coding and regulatory regions of the genes. Functional variants were classified as either disruptive or broad (Purcell et al., 2014). For coding regions, disruptive variants included any stopgain, frameshift, or essential splice site (±2 bp of intron–exon boundary) variants as annotated by ANNOVAR (Wang et al., 2010). Broad coding variants included disruptive variants plus any variant classified as potentially damaging by at least one of these bioinformatic tools: SIFT (Ng & Henikoff, 2003), PolyPhen2 (HVAR and HDIV; Adzhubei et al., 2010), LRT (Chun & Fay, 2009), MutationTaster (Schwarz, Rodelsperger, Schuelke, & Seelow, 2010), or VEST (Carter, Douville, Stenson, Cooper, & Karchin, 2013). For regulatory regions, disruptive variants included any variant that received a score of 1 or 2 from the Regulome database (v1.1; Boyle et al., 2012). These scores indicate how likely the variant is to affect the binding of regulatory elements. The broad regulatory variants included those with any evidence for regulatory potential from the Regulome database as indicated by a score of 1–6.

These rare, functional variants were assessed in the overall sample set and sex-specific subsets using two types of gene-level analyses: the gene-burden test and the sequence kernel association test (SKAT: v1.0.9; Wu et al., 2011). The gene-burden test was done using the CMC collapsing method (Li & Leal, 2008). In this method, any subject containing at least one rare, functional variant within a given gene is given a score of “1,” while subjects with no rare, functional variants are given a score of “0.” These scores are then assessed using Firth’s penalized logistic regression model (Firth, 1993) in the “logistf” R package (v1.21; Heinze & Puhr, 2010) to determine whether the number of attempters with variation in a certain gene differs significantly from the number of nonattempters with variation in that gene. The SKAT method assessed all rare, functional variants for a given gene as independent variables within a multivariate model. All variants in a given gene contributed to this model, regardless of direction of effect, meaning that this test did not generate odds ratios (OR). Sex and the first three principal components from our attempted suicide GWAS (Willour et al., 2012) were included as covariates for both the gene-burden and SKAT analyses. Because we ran a total of eight tests per gene, the threshold for study-wide significance was a p-value of 1.56 × 10⁻³ for the gene-level tests.

We also performed a haplotype analysis on the ∼318 kb targeted region spanning the large targeted region of intron 3 of LRRTM4 where our top female-specific attempted suicide GWAS findings were (Willour et al., 2012). This analysis was done in Haploview v4.2 (Barrett, Fry, Maller, & Daly, 2005) using the Gabriel et al. (2002) method and assessed all variants with MAF ≥0.05. The goal of this haplotype analysis was to determine whether specific haplotypes within this intronic region were enriched within attempters or nonattempters. The pathway and haplotype analyses were each permutation-corrected for multiple testing by running the analyses 10,000 times using randomly swapped attempter and nonattempter labels. Permutation-corrected p-values less than .05 were considered statistically significant.

Finally, we examined the variants contributing to significant haplotypes using Braineac (Trabzuni et al., 2011) to determine whether they are predicted brain expression quantitative trait loci (eQTLs). The Braineac database (Trabzuni et al., 2011; http://braineac.org) contains postmortem gene expression data for 134 brains from individuals without any neurodegenerative disorders and includes data from up to 10 brain regions (cerebellar cortex, frontal cortex, hippocampus, medulla, occipital cortex, putamen, substantia nigra, thalamus, temporal cortex, and intralobular white matter). Braineac used the Affymetrix (Thermo Fisher: Santa Clara, CA) Human Exon 1.0 ST array to assess gene expression of about 26,000 genes. The expression data were adjusted for sex, brain bank, and batch effects. Genotyping was done using the Illumina Infinium Omni-Quad BeadChip and Immunochip on DNA isolated from the cerebellum or occipital cortex. Following quality control and imputation, the final genotyping results included about 5.88 million SNPs and about 577,000 indels with a MAF >5%. For the eQTL analysis, MatrixEQTL (Shabalin, 2012) was used to compare each expression profile against each genetic marker using an additive genetic model (Ramasamy et al., 2013). Braineac only reports eQTLs with a false discovery rate ≤1%. Additionally, to control for potential overrepresentation of eQTLs due to regions of high LD, multiple associations for a specific probe or probeset were considered one signal if the r² > .5 for the associated SNPs. The SNP with the lowest p-value is reported as the “LD-resolved” eQTL for that region (Ramasamy et al., 2013).

2.4 |. Hormone response element alignment in intron 3 of LRRTM4

We used the online transcription factor binding site database JASPAR to compile the predicted binding sites of both estrogen and androgen receptors (Khan et al., 2018). We included the predicted binding sites for estrogen receptors (ESR1, ESR2, ESRRA, ESRRB) and androgen receptors (AR). We created custom tracks in the UCSC genome browser to align these hormone receptor binding sites to the SNPs contributing to both male and female haplotypes previously identified in Haploview from our data.

3 |. RESULTS

We identified a total of 5,986 variants in LRRTM4 in this study (Figure 1, Tables S2–S4). In the overall sample set, 300 variants were nominally significant (p < .05; Table S2), but no variants survived correction for multiple testing. When the female-specific analysis was performed, we identified 117 nominally significant variants, including rs10170138, the top female-specific variant from the previous GWAS (uncorrected p-value = .028; corrected p-value = 1.00, OR = 0.67; Table S3), but similarly found no variants that survived correction for multiple testing. However, the male-specific analysis found 634 variants were nominally significant (Table S4), including two variants that survived correction for multiple testing (Table 1). The two study-wide significant male-specific variants, rs67475248 (uncorrected p-value = 6.16 × 10⁻⁵, corrected p-value = .045, OR = 1.82) and rs7580390 (uncorrected p-value = 7.17 × 10⁻⁵, corrected p-value = .05, OR = 1.81), were located in the large intronic region of LRRTM4 that contained the top female-specific findings from our attempted suicide GWAS (Figure 1; Willour et al., 2012). Both of these male-specific variants occurred more frequently in attempters (MAF = 0.35) than nonattempters (MAF = 0.23). LRRTM1, LRRTM2, and LRRTM3 were also included in the variant sample set and we identified a total of 119, 75, and 150 variants, respectively, but none survived correction for multiple testing (Figure 1, Tables S2–S4).

FIGURE 1.

Individual variants in four human LRRTM family genes. Each gene (1–4) displays the mRNA (exons are blue blocks, and introns are blue hash-marked lines) along the bottom rows, and the variants found from the fine-mapping experiments in this text are displayed as vertical tick-marks at the top of each image. The color of the variants ranges from black to light gray, depending on the significance of the variant (black is most significant). LRRTM1–3 do not have any significant variants that survived multiple correction, nor do they have any significant haplotypes. LRRTM4 has five male-specific haplotype blocks (blue horizontal bars above variants), and one female-specific haplotype block (red horizontal bar above variants). There are also red and blue vertical lines representing the male variants from this study that survived multiple correction, and the female-specific variant from the previous genome-wide association study. There are two blue variants, however their close proximity prevents them from being distinguishable at this scale

TABLE 1.

Top variants within LRRTM Family

Chromosome location	Gene	Sample set	dbSNP142	Position	p-Value	Corrected p-value	Odds ratio	Attempter MAF	Nonattempter MAF	CADD PHRED score
chr2:77091865 (A/G)	LRRTM4	Male	rs67475248	Intronic	6.16E - 05	0.045	1.82	0.35	0.23	2.30
chr2:77092464 (A/G)	LRRTM4	Male	rs7580390	Intronic	7.17E - 05	0.050	1.81	0.35	0.23	0.71
chr2:77091972 (C/CCT)	LRRTM4	Male	rsll2444011	Intronic	9.40E - 05	0.061	1.80	0.35	0.23	0.87
chr2:77083510 (C/A)	LRRTM4	Male	rs7558600	Intronic	9.80E - 05	0.063	1.80	0.34	0.22	1.40
chr2:77090191 (G/A)	LRRTM4	Male	rs6750897	Intronic	.00015	0.082	1.76	0.34	0.23	3.07
chr2:77097065 (T/C)	LRRTM4	Male	rsl7013276	Intronic	.00018	0.091	1.79	0.31	0.20	0.18
chr2:77091311 (C/CT)	LRRTM4	Male	rs398090411	Intronic	.00019	0.093	1.76	0.33	0.22	8.43
chr2:77089765 (G/A)	LRRTM4	Male	rs56270055	Intronic	.00023	0.10	1.74	0.33	0.22	0.90
chr2:77103719 (T/C)	LRRTM4	Male	rsl0520168	Intronic	.00024	0.10	1.78	0.31	0.20	7.52
chr2:77102780 (C/T)	LRRTM4	Male	rs61279588	Intronic	.00024	0.10	1.78	0.31	0.20	1.49

Abbreviation: MAF, minor allele frequency.

We also performed gene-level analyses, including gene burden and SKAT tests, for the LRRTM1–4 gene family in the overall sample set and sex-specific subsets (Tables S5–S8). However, the threshold for study-wide significance for our gene-level results was 1.56 × 10⁻³, and none of our results survived correction for multiple testing.

We wanted to focus on the intronic region that provided the top GWAS and sequencing signals, so we performed a haplotype analysis of the large intronic region of LRRTM4 containing our top female-specific GWAS findings and the significant male-specific variant-level results to determine whether specific haplotypes were enriched in our attempters or nonattempters. We performed the haplotype analysis for the overall sample set and sex-specific subsets (Tables S9–S10). In the overall sample set, no haplotypes survived permutation correction. However, we did identify one female-specific haplotype and five male-specific haplotypes that were study-wide significant (Table 2, Figure 2, Figures S1–S7). The significant female haplotype (uncorrected p-value = 6.00 × 10⁻⁴, corrected p-value = .037, OR = 0.46) was a 16 kb block (chr2:77018564–77036892) that included rs10170138, the top female-specific finding from our attempted suicide GWAS (Willour et al., 2012; Figure 2; Figures S1–S2). There were 80 unique variants contributing to this haplotype, and 60 of those variants were predicted to be brain eQTLs according to Braineac (Trabzuni et al., 2011; Table 2). The most significant male-specific haplotype (uncorrected p-value = 5.00 × 10⁻⁴,corrected p-value = .016, OR = 0.60) was a 30 kb block (chr2:77085351–77116226) that includes the two male-specific study-wide significant individual variants (Figure 2; Figure S4). A total of 48 unique variants contributed to this haplotype, and 40 of those were predicted to be brain eQTLs (Table 2). Interestingly, the vast majority of contributing variants for each of the six significant haplotypes are predicted brain eQTLs according to Braineac (Trabzuni et al., 2011; Table 2). Further, the top male-specific significant variant, rs67475248, that is also found in one of the five significant male haplotypes, is two base-pairs away from a predicted ESRRA/B binding site and seven base-pairs away from a predicted AR binding site.

TABLE 2.

Significant haplotypes and eQTLs within LRRTM4

Subject group	Block	Haplotype alleles	Frequency	Attempter frequency	Nonattempter frequency	p-Value	Corrected p-value	Odds ratio	Total number of variants	Number of predicted eQTLs
Male	Block 13	A_C_A_A_G_G_C_A_G_C_A_G_G_T_C_T_T_G_C_G_T_C_T_C_ C_T_T_AT_GA_G_G_G_A_C_T_T_TA_C_T_G_T_C_G_A_A_ C_A_CAT_CTTCAT	0.33	0.27	0.38	5.00E - 04	.016	0.60	48	40
Male	Block 19	T_ATTGT_GAC_G_C_G_TATGATG_C_G_A_A_AT_C_T_T_C_ C_G_A_C_A_T_C_C_T_G_G_T_T_ATTAT_T_G_C_C_T_ TAAAA_C_G_A_C_A_G_A_T_A_CCTG_G_TATA_C_A_TAG_ T_A_T_C_C_C_C_TTTTC_C_GT_T_T_C_G_G_C_T	0.14	0.18	0.096	5.00E - 04	.018	2.01	65	49
Female	Block 14	C_T_TA_C_C_C_G_C_G_T_G_T_A_T_C_T_CTGAT_C_G_G_G_ T_A_TTATC_A_A_A_C_G_G_G_G_A_A_A_A_G_G_C_A_G_ G_T_G_T_A_G_G_C_A_A_C_G_T_C_A_G_G_G_A_G_G_G_ C_G_A_C_A_T_C_A_T_A_A_G_G_G_G_C_T	0.093	0.062	0.13	6.00E - 04	.037	0.46	80	60
Male	Block 15	C_T_A_T_G_T_T_T_T_C_A_T_G_T_T_GAATAT_GA_C_C_G_A_ C_C_G_C_T_A_T_T_C_A_A_T_G_C_A_T_T_A_A_C_T_C_T_ A_C_AGTGTGTGT_A_G_C_T_G_T_CA_C_A_G_C_T_A_C_A_ G_G_G_G_T_T_G_G_A_G_T_C_G_G_T_A_C_G_G_T_G_T_ G_T_AG_C_C_C_A_GAAGATT_T_G	0.15	0.19	0.11	.0011	.039	1.86	91	74
Male	Block 8	A_A_T_T_T_A_A_C_AATATG_T_G_T_C_T_A_A_G_T_G	0.24	0.28	0.19	.0013	.048	1.67	19	15
Male	Block 20	A_T_C_G_T_C	0.14	0.18	0.11	.0013	.049	1.87	6	5

Abbreviation: eQTLs, expression quantitative trait loci.

FIGURE 2.

Haplotype blocks divided by sex-specificity within LRRTM4 intron 3. Both images display haplotype blocks traced to the LD map. (a) The female haplotype block contains the top female-specific variant from the previous GWAS study and the red lines outline where the haplotype falls within the LD blocks. (b) The five male haplotype blocks are displayed and red vertical lines trace to the LD block to which each haplotype is associated. Both panels display the gene track in blue and conservation in green. The AR and ER HRE tracks are the black tracks below the conservation track (green). All haplotypes are colored to indicate variant significance: The top 10 significant variants are red (p < .05), additional nominally significant variants are green, and nonsignificant variants are black. LD blocks are outlined in black over the red map at the bottom of each panel. AR, androgen receptors; ER, estrogen receptor; GWAS, genome-wide association study; HRE, hormone response element; LD, linkage disequilibrium

4 |. DISCUSSION

Our sequencing study identified two male-specific variants, five male-specific haplotypes, and one female-specific haplotype in intron 3 of the LRRTM4 gene that were significantly associated with attempted suicide in bipolar disorder subjects. Males are 3.54 fold more likely to complete suicide, however females are 1.4 fold more likely to attempt suicide, indicating sex-specific differences in both suicide attempt and completion (https://www.cdc.gov/vitalsigns/suicide/index.html). Even though environmental factors may contribute to the disparity in suicide statistics between males and females, there is also biological evidence supporting sex-specific differences, including genetic biomarkers (Levey et al., 2016; Niculescu et al., 2015). Therefore, we chose to look at male and female subjects separately for these analyses because of both phenotypic differences in suicide attempt and completion statistics, and because of previous studies implicating sex-specific loci in psychiatric disorders. Our study found that the top male-specific haplotype overlaps with the two significant male-specific variants and the female-specific haplotype overlaps with the most significant female-specific variant from our GWAS study. Notably, our previous GWAS identified female-specific variants within this region, but there were also nominally significant male-specific signals that were identified.

LRRTM4 includes a large intron, intron 3, that contains common sex-specific variants and haplotypes that are associated with attempted suicide in bipolar disorder based on our GWAS and sequencing studies. This gene is found in the 2p11–12 linkage peak that was associated with attempted suicide in four previous studies (Hesselbrock et al., 2004; Butler et al., 2010; Willour et al., 2007; Zubenko et al., 2004). This new sequencing study resulted in a more detailed examination of the area of interest, allowing for the futher characterization of LD structure and allowing us to conduct dense haplotype analysis. Haplotypes themselves can even be the functional genomic region of interest (Stram, 2017). The vast majority of the variants contributing to the significant haplotypes are predicted brain eQTLs for LRRTM4. The female-specific haplotype included 80 variants, 75% of which were predicted to be brain eQTLs. The top male-specific haplotype contained 48 variants, 83% of which were also predicted brain eQTLs. A recent study by Brown, Sankararaman, and Pasaniuc (2018) demonstrated that haplotype-based eQTL mapping can be more effective at identifying complex regulatory regions than SNP-based methods (Brown et al., 2018). Ying, Li, Sham, and Li (2018) investigated this phenomenon in 20 different tissues (Ying et al., 2018). They found in several of the tissues examined that the haplotype-based eQTLs were enriched in regions previously associated with complex traits, and that this enrichment was tissue specific. They concluded that by using haplotype-based eQTL methods, they could better understand how genetic variants work together to regulate tissue-specific gene expression and the risk for complex disease. Consistent with this argument, our haplotype results may be more functionally relevant than our individual SNP findings, but additional research would be needed to demonstrate this.

LRRTM4 contains multiple hormone response elements (HREs) within the sequenced intronic region where the identified variants and haplotypes our group identified are located. Several of the variants contributing to the significant haplotypes fell either directly within or nearby (±10 bp) HREs (Figures S1–S8, Tables S11–S16). Past studies have shown that variants within HREs can cause changes in the binding and activation of different hormone receptors, potentially leading to changes in gene expression and increased risk for disease (Clinckemalie et al., 2013; Moghanibashi, Mohamadynejad, Rasekhi, Ghaderi, & Mohammadianpanah, 2012; Yu et al., 2011). Furthermore, it may be possible that even variants and/or haplotypes that affect binding affinity of regulatory proteins in close proximity to HREs may be functionally important for gene expression.

Our study had certain limitations. First, for rare single variant analyses, the sample set was only large enough to detect association of variants with a MAF of 0.05 and a genotypic relative risk ≥2.6 with 80% power using the variant tests we employed (calculated using Quanto v1.2.4; http://biostats.usc.edu/Quanto.html). For more common single variants, however, we had 80% power to identify a genotypic relative risk of ≥1.7 with MAF ≥0.30. For gene-based burden testing, we had 80% power to detect a cumulative damaging rare variant burden MAF of 0.05 and a genotypic relative risk ≥2.2. We also used only a subset of GWAS samples, which may have affected our ability to detect significance due to decreased power and potentially increased type I error (Christley, 2010). The significant SNP results from our study may be false positives and true statistical significance would not hold up in a larger dataset. Therefore, replication efforts in larger sample sets will be required in order to determine the actual significance of our results. While our GWAS sample set included all definitions of suicide attempt, our current sample set focused on a more stringent definition of suicide attempt (attempters with a definite or serious intent to die). This difference in definition could contribute to the observed differences for our variant rs7580390 in the current study (uncorrected p-value in overall sample set = .00042) versus the GWAS (uncorrected p-value in overall sample set = .0553). Another limitation to our study design is that we employed a targeted sequencing approach. We designed this study to include the sequencing of regions that were likely to affect gene function, including both exons and potential regulatory regions including the large intron of LRRTM4. However, because we did not include introns from any other LRRTM family genes in our study, we could have missed significant variants or haplotypes that may be associated with suicide. Additionally, we only sequenced ±10 kb from each gene, which focuses on close potential regulatory regions and does not include possible long-distance regulatory elements. Finally, individuals in our study classified as nonattempters at the time of interview could attempt suicide at a later point in time which biases our study toward early-onset attempters.

In conclusion, this study identified two male-specific variants, five male-specific haplotypes, and one female-specific haplotype that are all significantly associated with suicidal behavior in bipolar disorder. These loci are located within intron 3 of LRRTM4 and may have a functional impact on the brain through the regulation of LRRTM4 gene expression, potentially in a sex-specific manner. Many of these variants are also found in or near predicted HREs, and it is known that glutamatergic signaling is regulated by hormones. This evidence suggests that hormones may play a role in the regulation of LRRTM4 and could be important to understanding the genetic basis for sex differences in suicidal behaviors.