A Rare Functional Noncoding Variant at the GWAS-Implicated MIR137/MIR2682 Locus Might Confer Risk to Schizophrenia and Bipolar Disorder

Abstract

Schizophrenia (SZ) genome-wide association studies (GWASs) have identified common risk variants in >100 susceptibility loci; however, the contribution of rare variants at these loci remains largely unexplored. One of the strongly associated loci spans MIR137 (miR137) and MIR2682 (miR2682), two microRNA genes important for neuronal function. We sequenced ∼6.9 kb MIR137/MIR2682 and upstream regulatory sequences in 2,610 SZ cases and 2,611 controls of European ancestry. We identified 133 rare variants with minor allele frequency (MAF) <0.5%. The rare variant burden in promoters and enhancers, but not insulators, was associated with SZ (p = 0.021 for MAF < 0.5%, p = 0.003 for MAF < 0.1%). A rare enhancer SNP, 1:g.98515539A>T, presented exclusively in 11 SZ cases (nominal p = 4.8 × 10⁻⁴). We further identified its risk allele T in 2 of 2,434 additional SZ cases, 11 of 4,339 bipolar (BP) cases, and 3 of 3,572 SZ/BP study controls and 1,688 population controls; yielding combined p values of 0.0007, 0.0013, and 0.0001 for SZ, BP, and SZ/BP, respectively. The risk allele T of 1:g.98515539A>T reduced enhancer activity of its flanking sequence by >50% in human neuroblastoma cells, predicting lower expression of MIR137/MIR2682. Both empirical and computational analyses showed weaker transcription factor (YY1) binding by the risk allele. Chromatin conformation capture (3C) assay further indicated that 1:g.98515539A>T influenced MIR137/MIR2682, but not the nearby DPYD or LOC729987. Our results suggest that rare noncoding risk variants are associated with SZ and BP at MIR137/MIR2682 locus, with risk alleles decreasing MIR137/MIR2682 expression.

Main Text

MicroRNA (miRNA) dysfunction has been hypothesized to play an important role in neurodevelopmental disorders such as schizophrenia (SZ) (MIM 181500).^1–3 Recent SZ genome-wide association studies (GWASs) further strengthen an etiological role for miRNAs. Among >100 genome-wide significant (GWS) SZ risk loci, the MIR137/MIR2682 locus at 1p21.3 is one of the most strongly associated.^4–10 The GWS SZ risk variants are also associated with the impaired dorsolateral prefrontal cortex hyperactivation¹¹ and prefrontal-hippocampal functional connectivity.¹² Common GWS (p ≤ 5 × 10⁻⁸) variants cluster around MIR137 (MIM 614303) and MIR2682, with much weaker association extending to dihydropyrimidine dehydrogenase (DPYD [MIM 612779]) (Figure 1A). Large-scale exome sequencing^13,14 did not identify SZ-associated variants in coding regions of DPYD or within MIR137/MIR2682 that could explain the association, thus implying the importance of noncoding variants in conferring SZ risk at this locus. MIR137 is abundantly expressed in brain, enriched at neuronal synapses,¹⁵ and regulates neuronal differentiation, migration, and dendritogenesis.^16–20 Interestingly, ∼25% of SZ GWAS loci contain MIR137 targets (predicted by TargetScan),^4,7,9,10 including several empirically validated targets CACNA1C (MIM 114205), ZNF804A (MIM 612282), TCF4 (MIM 602272), CSMD1 (MIM 608397), and C10orf26 (MIM 611129),^21,22 suggesting a central hub role for MIR137 in a SZ susceptibility gene network. MIR137 has also been shown to target a large number of genes associated with autism spectrum disorders (ASD [MIM 209850]).²³ Although MIR2682 has no known function, it is predicted (TargetScan) to target ankyrin 3 (ANK3 [MIM 600465]), a gene previously found to be associated with BP (MIM 125480) in GWAS.^24–26 MIR137/MIR2682 thus represents a SZ risk locus with strong biology relevant to SZ. Rare deletions of genomic segments flanking MIR137/MIR2682 have been reported in individuals with intellectual disability (ID)¹⁵ and ASD.^27,28 Although we previously ruled out rare and large copy-number variants (CNVs) at this locus in our SZ GWAS sample,²⁹ it remained to be explored whether there were any rare SNPs or small indels of strong effect that could explain additional SZ risk and help to inform the functionality of common risk variants at the same GWAS loci.^30–33

Figure 1.

Genomic Features of the Sequenced MIR137/MIR2682 Locus

(A) Associations at the MIR137/MIR2682 SZ GWAS locus:⁹ regions selected for sequencing (light blue/blue, promoter/enhancer; red, insulator) and counts of the identified sequence variants in cases (red bar) and controls (blue bar), as well as the ENCODE chromatin modification marks in LCL (GM12878), neuronal cell lines (NH-A and SKNSHRA), and HeLa cells. DNaseI HS indicate regulatory regions. CTCF track indicates transcriptional insulators. Histone methylation mark tracks for H3K4m1 and H3K4m3 indicate enhancers and promoters, respectively. Different tracks were overlaid with physical positions on UCSC genome browser. GWAS association p values were downloaded from Ricopili and expressed as –log P.

(B) Sequencing chromatograms at the rare enhancer SNP 1:g.98515539A>T (T/A shown on DNA minus strand) and a consensus DNA sequence motif for YY1 binding (activation form).

(c) Cross-species sequence conservation around the SNP 1:g.98515539A>T and at the YY1 motif shown in UCSC genome browser (Note: DNA plus strand sequence is shown).

We first sequenced ∼6.9 kb of MIR137 and MIR2682 and their upstream regulatory sequences (Figure 1A and Table S1 available online) in 2,610 SZ cases and 2,611 controls from the Molecular Genetics of SZ (MGS) EA GWASs.^4,7 NorthShore University HealthSystem’s IRB approved the human subjects protocol, and proper informed consent was obtained. The selection of the region for sequencing was based on the DNaseI hypersensitive site (DHS) mapping data from ENCODE (Encyclopedia of DNA Elements^34,35) from neuronal cells (SK-N-SH and NH-A; Figure 1A) and in fetal brain. We further classified these putative regulatory sequences as ENCODE-annotated transcriptional promoters (H3K4me3), enhancers (H3K4me1), or insulators (CTCF-binding sites)³⁴ (Figure 1A and Figure S1). The PCR-amplified genomic DNA amplicons were sequenced on an ABI 3730 DNA Analyzer. The automatically (SeqScape 2.5; ABI) called SNPs and indels were manually verified, followed by extensive sequencing quality control metrics including genotype call rate (>90%), genotype concordance rate (>99.9%) between sequencing data and known GWAS genotypes,⁴ and absence of Hardy–Weinberg equilibrium (HWE) departures (p < 0.001 in controls).

We identified 143 SNPs and Indels (Table S2), of which 133 (∼93%) were rare (MAF < 0.5%). The variant density, proportion of singletons, and MAF distribution of the identified variants were all similar compared to whole exome or genome sequencing results (NHLBI-Exome Sequencing Project [ESP] and UK10K-TwinsUK)^36,37 for the same targeted region (Figure S2, Figure S3, and Table S3). The proportion of rare variants identified from our sequencing of putative regulatory sequences was similar to that of missense SNPs and frameshift or nonsense SNPs, but higher than that of intronic SNPs (Fisher’s exact test p = 0.0007; Figure S3B) uncovered by exome sequencing of 4,298 EA subjects (NHLBI-ESP).³⁶ The enrichment of rare variants in the sequenced noncoding regions relative to introns was consistent with the expected functionality of the sequenced region and possible purifying selection acting on deleterious variants.³⁸

We identified two common (MAF > 5%) SNPs (rs2660304 and rs2660302) in linkage disequilibrium (LD) (r² = 0.74 and 0.50, respectively, based on 1000 Genomes) with the reported SZ-associated rs1198588.^4–10 As expected, they showed nominally significant association to SZ in the sequenced MGS sample (Table S2), providing evidence of accuracy of our sequencing data. We found no association of SZ to a 15 bp Sequence Tandem Repeat (STR; rs58335419) 54 bp upstream from MIR137 (Table S1 and Figure S4), although the STR was recently found to reduce in vitro biogenesis of mature MIR137.³⁹ As in a previous report of no association of this functional MIR137 STR with SZ in a Japanese sample,⁴⁰ our nonreplication of the originally reported weak association (p = 0.049)³⁹ with SZ might be due to insufficient power of our MGS sample. Alternatively, because cellular RNA level is ultimately a result of multilayer of complex gene regulation predominately at transcriptional level,⁴¹ the in vitro posttranscriptional effect of the MIR137 STR³⁹ might be masked by other variant(s) with opposite functional effect and thus does not manifest in vivo phenotypic changes. A burden test aggregating all 133 rare variants did not produce evidence for association. As promoter and enhancer mutations likely reduce transcription while insulator mutations likely increase transcription, we reasoned that simply collapsing two sets of rare variants expected to have opposite directions might reduce the power. We thus performed burden tests separately for these two sets of rare variants. We found that the promoter and enhancer rare variants as a whole were significantly associated with SZ: p = 0.021 for variants with MAF < 0.5% in either cases or controls (to capture both risk and protective variants), and p = 0.003 for variants with MAF < 0.1% (in cases or controls); while insulator variants showed no case-control difference (Table 1). This highlights the importance of considering the direction of putative functional effects in association testing of rare variants in aggregate.

Table 1.

SZ Association of Rare Variants Grouped by Promoter/Enhancer and Insulator in 2,610 SZ Cases and 2,611 Controls from the MGS EA GWASs

	MAF < 0.5%				MAF < 0.1%
	Minor Allele Count				Minor Allele Count
Variant Location	SZ Cases	Controls	OR	p Value	SZ Cases	Controls	OR	p Value
Promoter/Enhancer	223	176	1.27	0.021	105	65	1.62	0.003
Insulator (CTCF)	62	81	0.77	0.132	26	32	0.81	0.435
All	285	257	1.11	0.246	131	97	1.35	0.029

ENCODE-annotated transcriptional promoters (H3K4me3), enhancers (H3K4me1), or insulators (CTCF-binding sites) were based on chromatin histone methylation patterns³⁴ in neuronal NH-A cells (ChIP-seq Signal > 10). MAF, minor allele frequency in MGS controls or cases. OR, odds ratio. p value is the nominal significance calculated by two-sided Fisher’s exact test. Using MAF < 1% as a cut-off (not tabulated) did not yield significant association. A total of 5,262 DNAs (2,634 SZ cases and 2,628 controls) from the MGS EA GWAS sample^4,7,9,29 were used for Sanger sequencing. 5,223 subjects (2,610 cases and 2,611 controls) remained for association analysis after sample quality control.

The association was primarily driven by a single SNP, 1:g.98515539A>T, that presented in 11 SZ cases versus 0 controls (p = 4.8 × 10⁻⁴, Fisher’s exact test) (Table 2). We repeated the DNA sequencing for 1:g.98515539A>T and confirmed all 11 heterozygous SZ cases. Although we had a higher proportion of blood DNAs in MGS SZ cases (25%) than in controls (1%), the rare variant 1:g.98515539A>T was not overrepresented in DNAs from bloods (27%) or from LCLs (Table S4). Furthermore, we confirmed the presence of 1:g.98515539A>T by repeated sequencing of DNAs re-extracted from available blood samples of two heterozygous subjects and from an induced pluripotent stem cell (iPSC) line derived at RUCDR (Rutgers University Cell and DNA Repository) from a heterozygous subject’s cryopreserved peripheral lymphocytes (Figure S5). These results suggested that the presence of 1:g.98515539A>T in 11 SZ cases was not due to technical artifacts resulted from sequencing or LCL derivation. Please note that, with 1:g.98515539A>T left out, the rare variants with MAF < 0.1% remained excess in SZ cases (p = 0.026; Table 1), and the association with GWAS-implicated common variants in this region did not change much in MGS sample (from p = 0.0005 to 0.0006). To estimate the empirical significance of the observed association with 1:g.98515539A>T, we performed 1 million random permutations for all rare SNPs in the region and determined the region-wise p value was 0.004. Interestingly, two rare insulator variants (1:g.98520090G>C and 1:g.98519714C>G) were more frequent in controls than in cases (p = 0.006 and 0.046, respectively, Fisher’s exact test); however, the associations did not survive multiple testing correction (Table 2), and the more strongly associated 1:g.98520090G>C was not functional (see functional studies below). We further genotyped 1:g.98515539A>T in three additional EA SZ samples: (1) the Irish Case-Control Study of SZ (ICCSS) sample (1,034 cases and 658 controls)^42,43 by a customized TaqMan assay (Life Technologies), (2) the University College London (UCL) sample (900 cases and 1,314 controls) by allele-specific PCR using KASPar reagents (LGC Genomics) on a LightCycler 480 (Roche), and (3) the Genomic Psychiatric Cohort (GPC; 500 cases and 200 controls)⁴⁴ by extracting the genotype of the 1:g.98515539A>T from available whole-genome sequencing data sets and found the rare allele in 2 SZ cases and 1 control (Table 3).

Table 2.

Single Variants Nominally Associated with SZ in 2,610 SZ Cases and 2,611 Controls from the MGS EA GWASs

Variant Name	Position (hg19)	Function	Alleles	MAF_case	MAF_ctrl	OR	p Value	Perm_P
1:g.98515539A>T	98,515,539	Enhancer	A:T	0.0021	0.0000	NA	0.00048	0.004
1:g.98520090G>C	98,520,090	Insulator (CTCF)	G:C	0.0002	0.0019	0.10	0.00627	NS
1:g.98519714C>G	98,519,714	Insulator (CTCF)	C:G	0.0023	0.0048	0.48	0.04627	NS
rs2660302	98,520,219	Promoter/enhancer	A:T	0.1573	0.1803	0.85	0.00807	NT
rs2660304	98,512,127	Promoter/enhancer	T:G	0.1695	0.1933	0.85	0.01314	NT

Functional regions were defined as ENCODE-annotated transcriptional promoters (H3K4me3), enhancers (H3K4me1), or insulators (CTCF-binding sites) based on chromatin histone methylation patterns³⁴ in neuronal NH-A cells (ChIP-seq Signal > 10). Alleles are coded on plus strand. Alleles listed as major:minor. OR, Odds ratio. p value is the nominal significance calculated by two-sided Fisher’s exact test. Perm_P is the region-wise significance after multiple testing correcting based on one million random permutations. NS, not significant; NT, not tested due to common SNPs.

Table 3.

Association of 1:g.98515539A>T with SZ and BP in the Combined Samples

	Cases		Controls
	Minor Allele T	Major Allele A	Minor Allele T	Major Allele A	Fisher p Value
MGS-SZ	11	5,177	–	5,192
ICCSS-SZ	1	2,067	–	1,316
UCL-SZ	1	1,799	1	2,627
GPC-SZ	–	1,000	–	400
NIMH-BP	2	2,090
UCL-BP	8	3,778
GPC-BP	1	2,799	1	2,799
TwinsUK			1	3,205
1000 Genome			–	170
Combined-SZ	13	10,043	3	15,709	0.0007
Combined-BP	11	8,667			0.0013
Combined-SZ/BP	24	18,710			0.0001

MGS^4,7,9 were Sanger sequenced. ICCSS^42,43 were genotyped by TaqMan. UCL were genotyped by allele-specific PCR using KASPar reagents. GPC genotypes were extracted from whole genome sequenced data. NIMH-BP were genotyped by TaqMan. Two public whole-genome sequence (WGS) data sets were used as population controls (TwinsUK and 1000 Genomes). Fisher’s Exact Test p values (single-sided) for the combined analysis were shown, and the combined control sample was used. UCL samples. The case sample included 1,917 BP and 932 SZ cases. All SZ or BP samples, except for UCL sample, have been previously described. UCL sample included 1,917 BP and 932 SZ cases, as well as 1,348 control subjects comprised of 868 screened subjects who had no first degree family or personal history of psychiatric illness and an additional 480 unscreened normal British subjects obtained from the European Collection of Cell Cultures (ECACC). BP cases had been given an NHS clinical diagnosis of BP by the International Classification of Disease 10 (ICD-10) and then needed to fulfill RDC for BP with clinical data collected by the lifetime version of the Schizophrenia and Affective Disorder Schedule (SADS-L).⁷⁶ The SZ cases were diagnosed in an analogous manner. National Health Service (NHS) multicenter research ethics approval was obtained for UCL sample.

Because SZ and BP share genetic architecture,⁴⁵ we also genotyped 1:g.98515539A>T in three EA BP samples: (1) NIMH-BP (1,046 cases)^46,47 by a customized TaqMan assay (Life Technologies), (2) UCL (1,893 cases) by allele-specific PCR using KASPar reagents (LGC Genomics), and (3) GPC (1,400 cases and 1,400 controls) by extracting the genotype of the 1:g.98515539A>T from available whole-genome sequencing data sets. We found 1:g.98515539A>T in 11 BP cases and 1 control (Table 3). Furthermore, we identified 1 heterozygous subject in whole-genome sequenced population controls (publically available; TwinsUK: 1,603 controls, 1000 Genomes: 85 controls) (Table S3). In the combined case (i.e., case-control) samples, the association of 1:g.98515539A>T with SZ, BP, and SZ and BP combined was 0.0007, 0.0013, and 0.0001, respectively (Table 3). We also examined 1:g.98515539A>T, in two family collections: (1) Clinical Neurogenetics (CNG) SZ families (307 subjects in 67 pedigrees)^48,49 by Sanger sequencing and (2) Irish Study of High-Density SZ Families (ISHDSF) (1,400 subjects in 274 pedigrees)^42,43 by a customized TaqMan assay (Life Technologies). We identified one SZ proband and an unaffected father in a CNG family to be heterozygous for 1:g.98515539A>T. The presence of the rare risk allele in an unaffected parent is consistent with incomplete penetrance, which is even commonly seen for SZ-associated rare CNVs of large effect⁵⁰ and for causal variants in most complex disorders.⁵¹

Next, we investigated whether allele T of 1:g.98515539A>T in the 11 SZ MGS EA cases originated from the same haplotypic background. We merged the genotypes of 1:g.98515539A>T with MGS GWAS genotypes⁴ and phased the MIR137 local region using BEAGLE⁵² for all subjects. The T alleles were on the same haplotype, suggesting that they were inherited from the same common ancestor. The shared common-SNP haplotype spans ∼77.2 kb (from rs1938567 to rs1198588) (Table S5) with a frequency of 0.787 in MGS controls, 0.792 in TwinsUK controls, and 0.790 in a publically available GWAS sample with 9,562 subjects of European ancestry (The Atherosclerosis Risk in Communities-ARIC Study data set downloaded from dbGAP). These results suggest that the rare risk allele is not just from an unusual ancestry that happens to carry this rare SNP on the haplotype, and our MGS controls are not an atypical EA subpopulation. To further assess any confounding effect of population stratification on our observed association with 1:g.98515539A>T, we analyzed the genotypic ancestry principal components (PCs)⁴ of the MGS cases carrying 1:g.98515539A>T and other MGS subjects (Figure S6). MGS cases and controls appeared to be matched very well, with northern European population as a predominate cluster. The MGS cases carrying 1:g.98515539A>T spread all over the main northern European cluster (Figure S6), indicating that these individuals with the rare variant were not preferentially from a small regional subpopulation. For the SZ subjects carrying 1:g.98515539A>T, we also calculated kinship coefficients using genome-wide SNPs with SNP frequencies estimated based on the whole MGS sample. Out of all possible pairs of individuals with the rare variant 1:g.98515539A>T, only one was greater than 0.02, nine were close to 0.01, and the rest of them were zero. This result demonstrated that the SZ cases carrying 1:g.98515539A>T are unlikely from a localized region, which is consistent with ancestry principal components analysis. From all of these analyses, we concluded that the observed disease association with 1:g.98515539A>T was unlikely due to population stratification.

SNP 1:g.98515539A>T is within an ENCODE-annotated neuronal cell (NH-A and SKNSH) specific enhancer and promoter ∼3.6 kb upstream of MIR137 (Figure 1A). It is 5 bp away from the binding motif (CCAT) of TF YY-1 (gene activating form; Figure 1B); the flanking sequence of the YY1 motif with the major allele of 1:g.98515539A>T is highly conserved in mammals (Figure 1C). We hypothesized that the risk allele of 1:g.98515539A>T reduced enhancer activity by interfering with YY1 binding. To test the hypothesis, we cloned the putative enhancer sequence (the transcribed minus strand) flanking 1:g.98515539A>T into a luciferase reporter gene vector (pGL3-promoter) (Figure 2A; Table S6). As expected from the ENCODE functional annotation, the cloned sequence exhibited robust enhancer activity as indicated by luciferase expression in a human neuroblastoma cell line (SH-SY5Y), but not in HeLa cells. The enhancer activity of the cloned sequence was reduced ∼50% by the risk allele A relative to the major allele T (we refer to the nucleotide on the minus strand hereinafter for the functional study section) of 1:g.98515539A>T (Figure 2A). However, 1:g.98520090G>C (nominally overrepresented in controls; p = 0.006), near an insulator (CTCF binding-site), did not display any effect on transcription in the reporter gene assay (Figure 2B) or on open chromatin (regulatory elements) in a formaldehyde-assisted isolation of regulatory elements (FAIRE) assay⁵³ in lymphoblastoid cell lines (LCLs; Figure S7).

Figure 2.

Reporter Gene Assay for Variant Functionality in Neuroblastoma (SH-SY5Y) and HeLa Cell Lines

Putative regulatory sequence flanking the rare enhancer (A) SNP (bp: 98,515,539; hg19) or the CTCF (B) SNP (bp: 98,520,090; hg19) was cloned in a configuration of minus strand upstream of the SV40 promoter of pGL3-promoter vector. The resultant reporter gene vectors carrying regulatory sequences with different alleles of a SNP were transiently transfected into neuroblastoma (SH-SY5Y) and HeLa cell lines, and the relative luciferase activity was measured for each reporter gene construct. Allele-specific effects on enhancer activity or insulator activity of the cloned sequence were expressed as the relative luciferase unit, with the reporter gene expression of pGL3-promoter (no insert) as a control for basal SV40 promoter activity. Note that in (B), the cloned sequence shows an enhancer activity in SH-SY5Y cells and repressor activity in HeLa cells, which is consistent with ENCODE functional annotations (CTCF and H3K4m1 peaks in NH-A, but only CTCF peak in HeLa). For normalizing the transfection efficiencies between different constructs, the firefly luciferase constructs were cotransfected with the Renilla luciferase pRL-CMV Vector (Promega). We used Lipofectamine 2000 Transfection Reagent (Invitrogen) for the transfection when the cultured cells are at 60% confluence for SH-SY5Y and 90% confluence for HeLa cells on 96-well plates. 48 hr after transfection, we measured the firefly and Renilla luciferase activities using Dual-Luciferase Reporter Assay System (Promega) according to the manufacturer’s protocol. Data were from 3 independent experiments and expressed as mean ± SD. Statistical significance was generated by Student’s t test.

We further carried out an electrophoresis mobility shift assay (EMSA) to examine whether the reduced enhancer activity by allele A of 1:g.98515539A>T correlated with altered DNA binding to TFs in SH-SY5Y cells. We found that the YY1 motif-flanking DNA probe carrying the risk allele A had a much weaker YY1 binding capacity than the probe with the major allele T (Figure 3A). Consistent with the EMSA result, computational analysis of local DNA breathing dynamics also predicted an effect of 1:g.98515539A>T on differential local DNA breathing associated with specific TF binding^54,55 (Figure 3B). Major allele T features a conformational DNA dynamics (probability for bubble formation p = 5.8 × 10⁻⁴) that favors strong YY1 binding, while the risk allele A nearly silences the DNA breathing in the vicinity of the YY1 binding sequence (Figure 3B). This effect could explain the weak YY1 binding associated with allele A in EMSA (Figure 3A). Allele T distinctively activates bubble formation with significantly higher probability that favors strong specific YY1 binding.⁵⁶

Figure 3.

The Rare Allele A of 1:g.98515539A>T Reduces TF YY1 Binding

(A) EMSA for YY1 binding. A double-stranded oligonucleotide (29 bp; on minus strand) flanking the YY1-binding site and 1:g.98515539A>T (allele T or A; AGAGGTGCTGTGAACACACAGCCATTTTC t/a TAGCAGCTTTTTGACTGTATGTTACCATA) was incubated with the nuclear extracts of neuroblastoma (SH-SY5Y) cells. Oligonucleotide probe (20 fmol) with allele T (lane 2) produced a much denser band of specific DNA-protein binding complex than the risk allele A (lane 6). The DNA-protein complex was recognized by antibody of YY1 (C-20; sc-281; rabbit polyclonal; Santa Cruz Biotechnology), showing up as a super-shift band (lanes 4 and 8), but weaker for allele A. The specificity of the YY1 binding is shown by the abolishment of the DNA-YY1 binding complex in the presence of excess (50 pmol) unlabeled probe (lanes 3 and 7). The EMSA was performed using the LightShift Chemiluminescent EMSA kit (Thermo Scientific). Binding reactions were separated by electrophoresis on 6% polyacrylamide gels on XCell SureLock Mini-Cell Electrophoresis System (Life Technologies).

(B) Computing models of the effect of 1:g.98515539A>T on DNA breathing dynamics (∼100 bp flanking the SNP site) that is necessary for binding of TFs. Langevin molecular dynamic (LMD) simulations, based on the Extended Peyrard-Bishop-Dauxois (EPBD) nonlinear model of DNA, assessed bubble-formation probability and average strand separation as the mechanistic parameters characterizing the TF binding activity of the sequence.^54,75 The sequences are simulated in 1,000 separate realizations. Sequence containing allele A in close proximity to YY1 binding site predicts weak DNA breathing activity (top), in contrast to the much stronger breathing potential of the sequence with allele T (bottom). The white horizontal lines mark the SNP sites. The YY1 binding sequence ending at the SNP site is shown below the SNP line. Vertical axis, bubble length in bp; color axis, bubble probability at specific nucleotide positions (horizontal axis) predicted for DNA openings with amplitude > 2Å at temperature of 310°K.

Although 1:g.98515539A>T at the MIR137/MIR2682 locus is proximal to the transcription start of the MIR137 host gene (MIR137HG), the enhancer sequence surrounding the SNP could still affect other adjacent genes (DPYD and LOC729987) (Figure 4A) through long-range regulation.^57,58 We therefore performed the 3C assay^58–60 in SH-SY5Y cells to examine whether the 1:g.98515539A>T site can physically interact with core promoters of DPYD and LOC729987 (Figure 4A). We identified specific physical interaction of the enhancer sequence flanking 1:g.98515539A>T with other putative regulatory sequences upstream of MIR137/MIR2682, but not with the core promoters of DPYD or LOC729987 (Figure 4B). This suggests the functional 1:g.98515539A>T might influence expression of MIR137/MIR2682, but not of DPYD or LOC729987.

Figure 4.

Chromatin Conformation Capture (3C) for Detecting the Physical Interaction between the Regulatory Sequences of MIR137/MIR2682 Locus in SH-SY5Y Cells

(A) Genomic locations of 3C probes (P1 to P6) on UCSC genome browser tracks along ∼300 kb region (chr1:98,378,000-98,688,000). Each probe is a plus-strand sequence ∼150 bp upstream of a HindIII cutting site. P3 is the bait probe adjacent to the enhancer sequence where 1:g.98515539A>T resides. P2, P4, and P5 target the other sequenced regulatory regions upstream MIR137/MIR2682. P1 and P5 target the core promoters of the flanking genes DPYD and LOC729987, respectively.

(B) PCR products from different pairs of 3C probes on an agarose gel (2%). The interacting genomic regions in SH-SY5Y cells were captured and enriched by sequentially cross-linking of chromatin, HindIII cutting of cross-linked chromatins, and religation of HindIII-digested DNAs. Only physically interacting genomic regions were enriched and produced specific amplification (corresponding to each pair of 3C probes) in PCR. The specific PCR products were verified by the expected size (Table S7) and by direct DNA sequencing. The HindIII-digested but unligated (i.e., no ligase) DNAs served as negative controls. PCR controls are an amplicon within two consecutive HindIII cutting sites and thus this shows specific amplification for both ligated and unligated chromatins.

Together, our reporter gene assay, EMSA, and the 3C experiment on the rare SNP 1:g.98515539A>T suggest a mechanism of reduced expression of MIR137/MIR2682 as contributing to SZ risk at this locus. This is consistent with a recent report showing an association of the SZ risk allele of the GWAS-implicated rs1625579 with decreased MIR137 expression in SZ postmortem brain tissue.⁶¹ The regulatory effect of SNP 1:g.98515539A>T is also consistent with the possible functional mechanism underlying the reported microdeletions involving MIR137 in ASD^27,28 and ID individuals.¹⁵ Although we have confirmed the transcriptional effect of the SZ-associated rare SNP 1:g.98515539A>T, we cannot rule out other functional variants at MIR137 locus.³⁹ Neurons differentiated from induced pluripotent stem cells (iPSCs) generated from SZ cases carrying the rare risk allele would be an appropriate experimental model to test the functional effect of 1:g.98515539A>T on MIR137/MIR2682 expression and on SZ-relevant cellular and physiological phenotypes.^62,63 We found a dramatic increase of MIR137/MIR2682 expression during the dopaminergic neuronal differentiation and maturation from iPSCs⁶⁴ (Figure S8). Such isogenic human iPSC-derived neurons differing only at the functional SNP site will also be an invaluable cellular model to study the downstream molecular target of MIR137 in a more disease-relevant physiological condition, complementing the current knowledge of genome-wide transcriptional effect of MIR137 obtained from overexpressing MIR137 in human neuronal stem cell line.^65,66

In summary, we have identified a functional rare noncoding risk variant for SZ at one of the most strongly associated SZ susceptibility loci from GWASs, highlighting the importance of rare noncoding variants in SZ genetics. Rare functional variants identified at GWAS-implicated loci not only explain additional genetic risk but also can provide unparalleled direct links to causal variants or mechanisms given their relatively larger effects compared to common risk variants at the same locus.^{30,31,67–70} The existence of rare noncoding SZ-risk variants at the SZ GWAS-implicated MIR137/MIR2682 locus is in line with recently reported polygenic burden of rare disruptive (nonsense, essential splicing site, and frameshift) variants in genes implicated by SZ GWAS and CNV studies.¹³ Our study also demonstrates an approach of analyzing rare noncoding variants based on a priori knowledge of directional functionality of a variant; a higher burden of rare noncoding variants in SZ cases was only observed in putative promoter and enhancer regions, but not in transcriptional insulators. Moreover, we have shown that a rare functional variant at MIR137/MIR2682 locus might confer risk for developing both SZ and BP, although common variants at this locus have been suggested to confer shared risk effect across major psychiatric disorders.⁷¹ Indeed, rare CNVs associated with a psychiatric disorder frequently show variable phenotypic expressivity, i.e., with shared risk to other psychiatric disorders or neurodevelopmental conditions.⁷² We acknowledge that although the association of 1:g.98515539A>T with SZ in the MGS sample remained significant after multiple-testing correction (p = 0.004), it did not reach genome-wide significance, and additional confirmation in independent samples will be necessary to definitively establish the association. It is however noteworthy that replicating the association with rare variants can be challenging, even in larger samples.^31,73,74 For SZ, there has been no report of a single rare variant showing genome-wide significant association, even in recent large-scale exome-sequencing studies.^13,14 Given that we have provided evidence that technical artifacts or population stratification are unlikely to explain our results, our study warrants further resequencing efforts with more comprehensive coverage of putative regulatory sequences at the MIR137/MIR2682 locus, confirmation of the observed association in larger independent samples, and a deeper understanding of the causal link between SZ risk alleles and disease phenotypes.

Web Resources

The URLs for data presented herein are as follows:

• 1000 Genomes, http://browser.1000genomes.org

• ANNOVAR, http://www.openbioinformatics.org/annovar/

• dbSNP, http://www.ncbi.nlm.nih.gov/projects/SNP/

• ENCODE, http://genome.ucsc.edu/ENCODE/

• GATK, http://www.broadinstitute.org/gatk/

• NHLBI Exome Sequencing Project (ESP) Exome Variant Server, http://evs.gs.washington.edu/EVS/

• Online Mendelian Inheritance in Man (OMIM), http://www.omim.org/

• Ricopili, www.broadinstitute.org/mpg/ricopili/

• UCSC Genome Browser, http://genome.ucsc.edu

• UK10K Consortium, http://www.uk10k.org/

Accession Numbers

The accession number for SNP 1:g.98515539A>T is ss1425684360. Previously unreported SNPs or indels identified in our study were submitted to dbSNP with accession numbers from ss1425684307 to ss1425684424.