Genome-wide association study identifies five new schizophrenia loci

Abstract

We examined the role of common genetic variation in schizophrenia in a genome-wide association study of substantial size: a stage 1 discovery sample of 21,856 individuals of European ancestry and a stage 2 replication sample of 29,839 independent subjects. The combined stage 1 and 2 analysis yielded genome-wide significant associations with schizophrenia for seven loci, five of which are new (1p21.3, 2q32.3, 8p23.2, 8q21.3 and 10q24.32-q24.33) and two of which have been previously implicated (6p21.32-p22.1 and 18q21.2). The strongest new finding (P = 1.6 × 10⁻¹¹) was with rs1625579 within an intron of a putative primary transcript for MIR137 (microRNA 137), a known regulator of neuronal development. Four other schizophrenia loci achieving genome-wide significance contain predicted targets of MIR137, suggesting MIR137-mediated dysregulation as a previously unknown etiologic mechanism in schizophrenia. In a joint analysis with a bipolar disorder sample (16,374 affected individuals and 14,044 controls), three loci reached genome-wide significance: CACNA1C (rs4765905, P = 7.0 × 10⁻⁹), ANK3 (rs10994359, P = 2.5 × 10⁻⁸) and the ITIH3-ITIH4 region (rs2239547, P = 7.8 × 10⁻⁹).

In stage 1, we conducted a mega-analysis combining genome-wide assocation study (GWAS) data from 17 separate studies (with a total of 9,394 cases and 12,462 controls; Table 1 and Supplementary Tables 1,2). We imputed allelic dosages for 1,252,901 autosomal SNPs (Table 1, Supplementary Table 3 and Supplementary Note) using HapMap3 as the reference panel¹. We tested for association using logistic regression of imputed dosages with sample identifiers and three principal components as covariates to minimize inflation in significance testing caused by population stratification. The quantile-quantile plot (Supplementary Fig. 1) deviated from the null distribution with a population stratification inflation factor of λ = 1.23. However, λ₁₀₀₀, a metric that standardizes the degree of inflation by sample size, was only 1.02, similar to that observed in other GWAS meta-analyses^2,3. This deviation persisted despite comprehensive quality control and inclusion of up to 20 principal components (Supplementary Fig. 1). Thus, we interpret this deviation as indicative of a large number of weakly associated SNPs consistent with polygenic inheritance⁴. We also examined 298 ancestry-informative markers (AIMs) that reflect European-ancestry population substructure⁵. Unadjusted analyses showed greater inflation in the test statistics than we saw for all markers (AIMs λ = 2.26 compared to all markers λ = 1.56). After inclusion of principal components, the distributions of the test statistics did not differ between AIMs (λ = 1.18) and all markers (λ = 1.23), a result inconsistent with population stratification explaining the residual deviation seen in Supplementary Figure 1. Moreover, the results of a meta-analysis using summary results generated using study specific principal components (Supplementary Note) were highly correlated with those from the mega-analysis (Pearson correlation = 0.94, with a similar λ = 1.20; Supplementary Fig. 2). Of the ten SNPs in Table 2, four increased and six decreased in significance, suggesting that the most extreme values did not result from systematic inflation artifacts. Therefore, our primary analysis used unadjusted P values (nevertheless, see Table 2 for stage 1 P values adjusted for λ (ref. 6).

Table 1.

Study design and samples

Collection	Country	Platform	Cases included by sex				Controls included by sex
			Male	Female	Unknown	Total	Male	Female	Unknown	Total
Cardiff UK	UK	Affymetrix 500K	320	152	0	472	1,442	1,492	0	2,934
CATIE	United States	Affymetrix 500K; Perlegen 164K	308	94	0	402	161	46	0	207
ISC-Aberdeen	UK	Affymetrix 5.0	536	184	0	720	447	251	0	698
ISC-Cardiff	Bulgaria	Affymetrix 6.0	270	257	0	527	291	318	0	609
ISC-Dublin	Ireland	Affymetrix 6.0	188	82	0	270	258	602	0	860
ISC-Edinburgh	UK	Affymetrix 6.0	267	101	0	368	146	138	0	284
ISC-London	UK	Affymetrix 5.0; Affymetrix 500K	369	149	0	518	207	284	0	491
ISC-Portugal	Portugal	Affymetrix 5.0	213	133	0	346	80	135	0	215
ISC-SW1	Sweden	Affymetrix 5.0	93	75	0	168	82	85	0	167
ISC-SW2	Sweden	Affymetrix 6.0	231	159	0	390	116	113	0	229
MGS	United States, Australia	Affymetrix 6.0	1,863	816	0	2,679	1,140	1,344	0	2,484
SGENE-Bonn	Germany	Illumina 550K	238	236	0	474	664	640	0	1,304
SGENE-Copenhagen	Denmark	Illumina Human 610-Quad	280	202	0	482	268	189	0	457
SGENE-Munich	Germany	Illumina 300K	279	155	0	434	167	184	0	351
SGENE-TOP3	Norway	Affymetrix 6.0	132	116	0	248	176	175	0	351
SGENE-UCLA	The Netherlands	Illumina 550K	529	175	0	704	310	321	–	631
Zucker Hillside	United States	Affymetrix 500K	128	64	0	192	92	98	0	190
Grand totals for the GWAs			6,244	3,150	0	9,394	6,047	6,415	–	12,462
Multicenter Pedigree	Europe, United States, Australia	Illumina Human 610-Quad	n.a.	n.a.	0	583	0	0	0	0
SGENE-Aarhus	Denmark	Illumina Human 610-Quad	477	399	0	876	477	397	0	874
SGENE-Aarhus	Denmark	Centaurus	114	102	1	217	176	317	0	493
SGENE-Belgium	Belgium	Centaurus; Illumina 370K	326	184	0	510	149	192	0	341
SGENE-Copenhagen	Denmark	Centaurus	264	198	0	462	499	375	0	874
SGENE-Iceland	Iceland	Illumina 300K	346	185	0	531	5,802	5,813	0	11,615
SGENE-England	UK	Illumina 300K	71	22	0	93	48	40	0	88
SGENE-Helsinki	Finland	Illumina 300K	112	70	0	59	122	75	0	147
SGENE-Kuusamo	Finland	Illumina 300K				123				50
SGENE-Hungary	Hungary	Centaurus	105	136	0	241	89	125	0	214
SGENE-Italy	Italy	Illumina 300K	48	36	0	84	50	39	0	89
SGENE-Munich	Germany	Illumina 300K	280	186	0	163	887	912	0	185
SGENE-Munich	Germany	Centaurus				303				1,614
SGENE-Russia	Russia	Centaurus	132	343	0	475	178	290	0	468
SGENE-Sweden	Sweden	Centaurus	158	94	0	252	178	109	0	287
SW3	Sweden	Affymetrix 6.0	327	212	0	539	457	448	0	905
SW4	Sweden	Affymetrix 6.0	656	407	0	1,063	605	568	0	1,173
UQ and ASRB	Australia	SequenomMassArray	347	190	21	558	487	455	15	957
ISGC and WTCCC2	Ireland	Affymetrix 6.0	968	342	0	1,310	245	778	0	1,023
Grand totals for the replication follow up			4,731	3,106	22	8,442	10,449	10,933	15	21,397

Stage 1 describes the 17 samples that provided full GWAS genotyping data, and stage 2 describes the 19 studies that provided results for the top SNPs identified in the combined analysis of stage 1 studies. Stage 2 replication SGENE-Belgium had four cases missing sex information. Stage 2 replication SGENE-Aarhus (focused genotyping sample) had one case missing sex information. Stage 2 replication-University of Queensland had 21 cases and 15 controls missing sex information. Sex information for the two stage 2 replication SGENE-Munich samples are combined. Sex information for the two stage 2 replication SGENE-Finnish (Helsinki and Kuusamo) samples are combined to enable that these two samples are located adjacent to each other in the table (rather than alphabetically). Multicenter Pedigree was a family sample, and so case sex counts are not applicable (n.a). SGENE, Schizophrenia Genetics Consortium; ISC, International Schizophrenia Consortium; TOP3, Thematic Organized Psychoses Research 3; UCLA, University of California at Los Angeles; SW1, Sweden 1; SW2, Sweden 2; WTCCC, Wellcome Trust case Control Consortium; for the Multicenter Pedigree study, the number of cases indicates the number of families; CATIE, Clinical Antipsychotic Trials of Intervention Effectiveness; MGS, Molecular Genetics of Schizophrenia; UQ, University of Queensland; ASRB, Australian Schizophrenia Research Bank; ISGC, Irish Schizophrenia Genomics Consortium.

Table 2.

Top genome-wide association results for schizophrenia

SNP	Chr.	Mb	Alleles	Frequency	P (GC-adjusted P )	OR (95% CI)	Consistency of direction	Gene	Distance (kb)
rs1625579	1p21.3a	98.3	TG	0.80	5.72 × 10−7 (6.52 × 10−6)	1.14 (1.08–1.19)	+−++++−+	MIR137	Intragenic
					2.65 × 10−6 (n.a.)	1.11 (1.07–1.16)
					1.59 × 10−11 (6.87 × 10−10)	1.12 (1.09–1.16)
rs17662626	2q32.3a	193.7	AG	0.91	3.09 × 10−6 (2.60 × 10−5)	1.22 (1.13–1.30)	+ −+ +++	PCGEM1	343
					1.70 × 10−3 (n.a.)	1.16 (1.06–1.27)
					4.65 × 10−8 (1.25 × 10−6)	1.20 (1.13–1.26)
rs2021722	6p21.3-p22.1	30.3	CT	0.78	4.30 × 10−11 (2.76 × 10−9)	1.18 (1.13–1.23)	+ ++ −++	TRIM26	Intragenic
					1.55 × 10−3 (n.a.)	1.10 (1.03–1.17)
					2.18 × 10−12 (2.88 × 10−10)	1.15 (1.11–1.19)
rs10503253	8p23.2a	4.2	AC	0.19	3.84 × 10−7 (4.71 × 10−6)	1.14 (1.09–1.19)	+ ++ +−+	CSMD1	Intragenic
					7.60 × 10−3 (n.a.)	1.08 (1.01–1.14)
					4.14 × 10−8 (8.98 × 10−7)	1.11 (1.07–1.15)
rs7004633	8q21.3a	89.8	GA	0.18	1.45 × 10−8 (3.22 × 10−7)	1.16 (1.11–1.21)	++−+++−+	MMP16	421
					0.011 (n.a.)	1.05 (1.01–1.10)
					2.75 × 10−8 (7.03 × 10−7)	1.10 (1.07–1.14)
rs7914558	10q24.32a	104.8	GA	0.59	1.58 × 10−7 (2.27 × 10−6)	1.11 (1.07–1.15)	+ ++ +++	CNNM2	Intragenic
					1.07 × 10−3 (n.a.)	1.08 (1.03–1.13)
					1.82 × 10−9 (3.11 × 10−8)	1.10 (1.07–1.13)
rs11191580	10q24.33a	104.9	TC	0.91	2.23 × 10−8 (4.58 × 10−7)	1.22 (1.15–1.29)	++++++++	NT5C2	Intragenic
					5.09 × 10−3 (n.a.)	1.09 (1.02–1.16)
					1.11 × 10−8 (3.72 × 10−7)	1.15 (1.10–1.20)
rs548181	11q24.2	125.0	GA	0.88	2.91 × 10−8 (5.69 × 10−7)	1.20 (1.13–1.26)	+++++−+	STT3A	1
					0.068 (n.a.)	1.04 (0.98–1.11)
					8.87 × 10−7 (1.74 × 10−5)	1.11 (1.07–1.16)
rs12966547	18q21.2	50.9	GA	0.58	1.00 × 10−6 (1.03 × 10−5)	1.10 (1.06–1.14)	+−++++++	CCDC68	126
					2.29 × 10−5 (n.a.)	1.08 (1.04–1.12)
					2.60 × 10−10 (5.99 × 10−9)	1.09 (1.06–1.12)
rs17512836	18q21.2	51.3	CT	0.02	2.35 × 10−8 (4.78 × 10−7)	1.40 (1.28–1.52)	−+++++++	TCF4	Intragenic
					0.085 (n.a.)	1.08 (0.96–1.20)
					1.05 × 10−6 (2.86 × 10−5)	1.23 (1.14–1.31)

The SNPs listed are those with a stage 1 P < 5 × 10⁻⁸ and/or a combined stage 1 and 2 P < 5 × 10⁻⁸. These ten independent (r² < 0.2) SNPs represent eight physically distinct genomic loci, as there are two SNPs listed for two loci (10q24.32-q24.33 and 18q21.2). For the MHC region, only one SNP is listed for clarity. The eight susceptibility loci represent three previously reported and five new loci (Supplementary Table 7). Stage 1 is the discovery GWAS mega-analysis. Stage 2 is the replication sample (single-tailed meta-analysis P values are weighted by 1/s.e.), and because the P values are single tailed, some 95% confidence intervals contain 1 (if 0.10 < P < 0.05). Combined values include stages 1 and 2 (two-tailed meta-analysis P values are weighted by 1/s.e.). For each SNP, P values and odds ratios are listed for stage 1 (top), stage 2 (middle) and combined stage 1 and 2 analysis (bottom) with the genomic control (GC)-adjusted values bracketed (n.a., not applicable for stage 2). Alleles are listed with the stage 1 risk allele first; the frequency (in stage 1 controls) and odds ratio (OR) refer to the stage 1 risk allele. Bolded P values indicate P < 5 × 10⁻⁸, except for in the stage 2 data, where bolded values indicate P < 0.05. The directions of association in eight replication samples are represented by + if the associations are in the same direction, − if they are in opposite directions and a blank space if the data are not available. Mb is the base position based on hg18. Cytogenetic bands are listed for each SNP, though because only one of multiple MHC SNPs are listed, a band range is given in that instance. The nearest gene (or microRNA) is listed, with the distance (kb) from the gene (or if the SNP is intragenic) noted. None of these SNPs showed a significant test for hetereogeneity among the samples. Chr., chromosome.

^aNew finding.

In stage 1 (Table 2, Supplementary Table 4 and Supplementary Figs. 3 and 4), 136 associations reached genome-wide significance (P < 5 × 10⁻⁸)⁷. The majority of these associations (N = 129) mapped to 5.5 Mb in the extended major histocompatibility complex (MHC, 6p21.32-p22.1), a region of high linkage disequilibrium (LD) previously implicated in schizophrenia in a subset of the samples used here^4,8,9. The other stage 1 regions included new regions (10q24.33 and 8q21.3) and previously reported regions (18q21.2 at TCF4 (encoding transcription factor 4) and 11q24.2 (ref. 8)). The signal at 11q24.2 is ~0.85 Mb from NRGN (encoding neurogranin) and is uncorrelated with the previously associated variant near this gene⁸.

In Table 2 and Supplementary Table 4, we denote regions of association by the most significant marker. Associated SNPs with r² ≥ 0.2 in HapMap3 (CEU+TSI populations) were not considered independent. However, we noticed instances where multiple SNPs within 250 kb of each other yielded evidence for association (P < 10⁻⁵) despite weak LD (r² < 0.2) between them. For regions with P < 10⁻⁶, we performed a conditional analysis using as covariates the dosages of the strongest associated SNP, principal components 1–4 and 6 and study indicator. We observed multiple statistically independent signals at the MHC. Although a number of SNPs within the MHC were potentially independent per HapMap r² values, only rs9272105 withstood formal conditional analysis, showing P = 1.8 × 10⁻⁶ conditional on association to the best SNP, rs2021722 (stage 1 P = 4.3 × 10⁻¹¹, inter-SNP distance = 2.4 Mb, r² = 0.01 in HapMap). Excluding the MHC region, we identified six regions with at least one SNP associated at P < 10⁻⁵ and a second SNP with a conditionally independent P < 10⁻³ (Supplementary Table 5). We performed 100 simulations after permuting case-control status randomly within each study. In contrast to the six regions in the real dataset, we never observed more than a single region with co-localized statistically independent signals in any simulated genome-wide scan, indicating our observation is highly unlikely to have occurred by chance.

Noteworthy co-localizing independent signals occurred at three regions (Supplementary Table 5): one region with a genome-wide significant association at 10q24.32-q24.33 (Table 2), a second region that nearly met this threshold at MAD1L1 (encoding mitotic arrest deficient-like 1; rs10226475, P = 5.06 × 10⁻⁸; Supplementary Table 4) and a third region at CACNA1C (encoding calcium channel, voltage-dependent, L type, α 1C subunit), the latter of which has previously been associated with bipolar disorder¹⁰ and other psychiatric phenotypes including schizophrenia¹¹. The conditionally independent signal at CACNA1C was more significant than any observation made in 100 permutations of the entire experiment (both conditional P < 10⁻⁵) and supports CACNA1C in schizophrenia after genome-wide correction (P < 0.01), even without considering these prior reports.

In stage 2, we evaluated in 29,839 independent subjects (8,442 cases and 21,397 controls) the most significant SNPs (N = 81) in each LD region where at least one SNP had surpassed P < 2 × 10⁻⁵ (Supplementary Table 6) in the mega-analysis. Of 22 SNPs from the MHC, 5 surpassed the genome-wide significant threshold in stages 1 and 2 combined (minimum P = 2.2 × 10⁻¹² at rs2021722; Supplementary Table 6). Excluding the MHC region, a sign test for consistency between stages 1 and 2 was highly significant (P < 10⁻⁶), with the same direction of effect as observed stage 1 also being observed in stage 2 for 49 of 59 SNPs. A Fisher’s combined test revealed the distribution of stage 2 P values was unlikely to have occurred by chance (P < 10⁻¹⁵). We also performed a transmission analysis using the family based Multicenter Pedigree replication sample in conjunction with a GWAS of 622 parent-offspring schizophrenia trios from Bulgaria¹², and the stage 1 associated allele was over-transmitted to cases for 44 of the 59 SNPs (one-sided P = 1.0 × 10⁻⁴). Thus, the stage 2 replication results are highly consistent with the stage 1 discovery results.

In the combined dataset (stages 1 and 2), five new (1p21.3, 2q32.3, 8p23.2, 8q21.3 and 10q24.32-q24.33) and two previously reported (6p21.32-p22.1 and 18q21.2) loci met genome-wide significance (Figs. 1,2, Table 2, Supplementary Tables 6,7 and Supplementary Fig. 4). After adjusting for λ (ref. 6), four loci (1p21.3, 6p21.32-p22.1, 10q24.32-q24.33 and 18q21.2) remained significant at P ≤ 5 × 10⁻⁸. For the primary analyses (unadjusted for λ), the strongest new association was at 1p21.3 (rs1625579; P = 1.6 × 10⁻¹¹), which is over 100 kb from any RefSeq protein-coding gene but is within intron 3 of AK094607, which contains the primary transcript for MIR137 (ref. 13). The next best locus, 10q24.32 (Supplementary Table 5 and Supplementary Fig. 5), has independent associations 130 kb apart at rs7914558 (P = 1.8 × 10⁻⁹) and rs11191580 (P = 1.1 × 10⁻⁸), implicating a 0.5-Mb region containing multiple genes (Supplementary Fig. 5). The third best locus, rs7004633 (P = 2.8 × 10⁻⁸) on 8q21.3, is 400 kb from the nearest gene (MMP16, encoding matrix metallopeptidase 16). The fourth best locus, rs10503253 (P = 4.4 × 10⁻⁸) at 8p23.2, is in an intron of CSMD1 (encoding CUB and Sushi multiple domains 1). Finally, rs17662626 (P = 4.7 × 10⁻⁸) at 2q32.3 is intergenic, mapping 300 kb from a non-coding RNA, PCGEM1 (prostate-specific transcript 1)¹⁴.

Figure 1.

Manhattan plot for stages 1 and 2. Standard −log₁₀ P plot of the study results. For the stage 1 results, 16 regions with one or more SNP achieving P < 10⁻⁶ are highlighted in color and labeled with the name of the nearest gene. SNPs selected for stage 2 replication are highlighted, with the resulting combined P value after replication (that is, after incorporation of stage 2 results) indicated by the large diamonds. Blue highlighting indicates SNPs that were less significantly associated after replication, and pink highlighting indicates SNPs that were more significantly associated after replication.

Figure 2.

Regional association plots for five new schizophrenia loci. Regional P value plots for each of the five new schizophrenia loci: 1p21.3, 2q32.3, 8p23.2, 8q21.3 and 10q24.32-q24.33. Each plot shows the most associated SNP (key SNP) and its genomic region from the first column of Table 2: stage 1 scan results for each SNP ± 200 kb to the key SNP are shown. On the x axis is the genomic position, and on the y axis is −log₁₀ P. Larger SNP symbols indicate higher LD (based on HapMap 3 data) to the key SNP than smaller SNP symbols. Color coding (from red to blue) denotes LD information; see also the legend within the plot.

MIR137 has been implicated in regulating adult neurogenesis^15,16 and neuronal maturation¹⁷, mechanisms through which variation at this locus could contribute to brain development abnormalities in schizophrenia. Of relevance, two independent schizophrenia imaging studies found MIR137 to be one of three microRNAs with targets significantly enriched for association¹⁸. In stage 1, SNPs in or near 301 high-confidence predicted MIR137 targets (with a TargetScan¹⁹ probability of conserved targeting ≥0.9) were enriched for association compared with genes matched for size and marker density: 17 predicted MIR137 targets (Supplementary Table 8) had at least one SNP with P < 10⁻⁴, which is more than twice as many as the control gene sets (P < 0.01). Excluding the MHC and MIR137, of the nine loci with genome-wide significant support either in stage 1 or in the combined set (six loci, 2q32.3, 8p23.2, 8q21.3, 10q24.32-q24.33, 11q24.2 and 18q21.2; Table 2 and Supplementary Tables 6,7) or in a joint analysis with bipolar disorder (three genes, CACNA1C, ANK3 and ITIH3-ITIH4, described below), four genes (TCF4, CACNA1C, CSMD1 and C10orf26) have predicted MIR137 target sites according to analyses using three different prediction programs (TargetScan¹⁹, PicTar²⁰ and miRanda²¹). In vitro overexpression and locked nucleic acid–mediated knockdown of MIR137 in neuronal cell line N2a leads to changes in expression levels of TCF4 protein, strongly supporting the prediction that TCF4 is a target of MIR137 (L.-H. Tsai, personal communication). Our observations suggest MIR137-mediated dysregulation as a new etiologic mechanism in schizophrenia.

The International Schizophrenia Consortium (ISC) reported evidence for a polygenic contribution to schizophrenia⁴. An independent family based study confirmed these results, greatly minimizing the possibility of population stratification artifact¹². We reevaluated the polygenic model, dividing stage 1 samples into independent training and testing sets (Supplementary Note). The training set had 15,429 subjects (over twice the size of the ISC training set), and the testing set consisted of 6,428 individuals independent of the ISC report. The proportion of variance (Nagelkerke’s r²) explained in the testing set increased from 3% in the ISC to around 6% here (Supplementary Table 9 and Supplementary Fig. 6). This estimate is much lower than the true total variation in liability that is tagged by all SNPs because SNP effects are estimated with error^3,4,22–25. The polygenic model appears to explain a substantial fraction of the heritability of schizophrenia⁴, as has been shown for other complex traits^3,26–28. Some of these additional risk loci are likely contained near the most highly significant results of our stage 1 analysis. Supporting this hypothesis, of the top loci that did not reach genome-wide significance in the combined stage 1 and 2 analysis, a sign test (P < 10⁻⁴) and a Fisher’s combined test (P < 10⁻⁵) both showed an excess of same-direction allelic association (41 of 51 non-MHC SNPs) in the discovery and replication datasets.

Clinical, epidemiological and genetic findings suggest shared risk factors between bipolar disorder and schizophrenia²⁹. In stage 1, three genes with strong support had prior genome-wide significant associations with bipolar disorder: CACNA1C, the region containing ITIH3-ITIH4 (encoding inter-α (globulin) inhibitors H3 and H4) and ANK3 (encoding ankyrin 3, node of Ranvier (ankyrin G))^10,11,30 (Supplementary Table 10). We performed a joint analysis with the Schizophrenia Psychiatric Genome-Wide Association Study (GWAS) Consortium (PGC) for bipolar disorder applying identical analytical methods. After removing duplicate subjects, we analyzed 16,374 cases with schizophrenia, schizoaffective disorder or bipolar disorder and 14,044 controls. Support for shared susceptibility was strengthened (Supplementary Table 11) at CACNA1C (rs4765905, P = 7.0 × 10⁻⁹), ANK3 (rs10994359, P = 2.5 × 10⁻⁸) and the ITIH3-ITIH4 region (rs2239547, P = 7.8 × 10⁻⁹), each of which reached genome-wide significance. A coding variant in ITIH4 (p.Pro698Thr; rs4687657) is in perfect LD with the most associated SNP. Although we included all subjects from an earlier report¹⁰, the increased support found with additional independent cases (N = 11,987) and controls (N = 7,835) provides further evidence for shared risk effects of schizophrenia and bipolar disorder.

The risk variants implicated here confer small risks (odds ratios ~1.10), but the polygenic analysis shows many more susceptibility variants with effects for which our sample is underpowered (Supplementary Table 12). At every stage where samples were added, we found an increase in the number of genome-wide significant loci and enhancement of signals at CACNA1C, ANK3 and ITIH3-ITIH4 when schizophrenia and bipolar disorder were jointly analyzed. Thus, gains in power offset any penalty for increased heterogeneity.

In summary, we report seven genome-wide significant schizophrenia associations (five of which are new) in a two-stage analysis of 51,695 individuals. We also report loci that confer susceptibility to both bipolar disorder and schizophrenia. The association near MIR137, associations in multiple predicted MIR137 targets and the known role of MIR137 in neuronal maturation and function together suggest an intriguing new insight into the pathogenesis of schizophrenia.

URLs

PLINK, http://pngu.mgh.harvard.edu/~purcell/plink/; Haploview, http://www.broadinstitute.org/scientific-community/science/programs/medical-and-population-genetics/haploview/haploview.

METHODS

Methods and any associated references are available in the online version of the paper at http://www.nature.com/naturegenetics/.

Stephan Ripke¹, Alan R Sanders^2,3, Kenneth S Kendler^4–6, Douglas F Levinson⁷, Pamela Sklar^1,8, Peter A Holmans^9,10, Dan-Yu Lin¹¹, Jubao Duan^2,3, Roel A Ophoff^12–15, Ole A Andreassen^16,17, Edward Scolnick¹⁸, Sven Cichon^19–21, David St. Clair²², Aiden Corvin²³, Hugh Gurling²⁴, Thomas Werge²⁵, Dan Rujescu²⁶, Douglas H R Blackwood²⁷, Carlos N Pato²⁸, Anil K Malhotra^29–31, Shaun Purcell¹⁸, Frank Dudbridge³², Benjamin M Neale¹⁸, Lizzy Rossin¹, Peter M Visscher³³, Danielle Posthuma^34,35, Douglas M Ruderfer¹, Ayman Fanous^5,36,37, Hreinn Stefansson³⁸, Stacy Steinberg³⁸, Bryan J Mowry^39,40, Vera Golimbet⁴¹, Marc De Hert⁴², Erik G Jönsson⁴³, István Bitter⁴⁴, Olli P H Pietiläinen^45,46, David A Collier⁴⁷, Sarah Tosato⁴⁸, Ingrid Agartz^16,49, Margot Albus⁵⁰, Madeline Alexander⁷, Richard L Amdur^36,37, Farooq Amin^51,52, Nicholas Bass²⁴, Sarah E Bergen¹, Donald W Black⁵³, Anders D Børglum^54,55, Matthew A Brown⁵⁶, Richard Bruggeman⁵⁷, Nancy G Buccola⁵⁸, William F Byerley^59,60, Wiepke Cahn⁶¹, Rita M Cantor^14,15, Vaughan J Carr⁶², Stanley V Catts⁶³, Khalid Choudhury²⁴, C Robert Cloninger⁶⁴, Paul Cormican²³, Nicholas Craddock^9,10, Patrick A Danoy⁵⁶, Susmita Datta²⁴, Lieuwe de Haan⁶⁵, Ditte Demontis⁵⁴, Dimitris Dikeos⁶⁶, Srdjan Djurovic^16,67, Peter Donnelly^68,69, Gary Donohoe²³, Linh Duong²⁵, Sarah Dwyer^9,10, Anders Fink-Jensen⁷⁰, Robert Freedman⁷¹, Nelson B Freimer¹⁴, Marion Friedl²⁶, Lyudmila Georgieva^9,10, Ina Giegling²⁶, Michael Gill²³, Birte Glenthøj⁷², Stephanie Godard⁷³, Marian Hamshere^9,10, Mark Hansen⁷⁴, Thomas Hansen²⁵, Annette M Hartmann²⁶, Frans A Henskens⁷⁵, David M Hougaard⁷⁶, Christina M Hultman⁷⁷, Andrés Ingason²⁵, Assen V Jablensky⁷⁸, Klaus D Jakobsen²⁵, Maurice Jay^79,132, Gesche Jürgens⁸⁰, René S Kahn⁶¹, Matthew C Keller⁸¹, Gunter Kenis⁸², Elaine Kenny²³, Yunjung Kim⁸³, George K Kirov^9,10, Heike Konnerth²⁶, Bettina Konte²⁶, Lydia Krabbendam⁸⁴, Robert Krasucki²⁴, Virginia K Lasseter^85,132, Claudine Laurent⁷⁹, Jacob Lawrence²⁴, Todd Lencz^29–31, F Bernard Lerer⁸⁶, Kung-Yee Liang⁸⁷, Paul Lichtenstein⁷⁷, Jeffrey A Lieberman⁸⁸, Don H Linszen⁶⁵, Jouko Lönnqvist⁸⁹, Carmel M Loughland⁹⁰, Alan W Maclean²⁷, Brion S Maher^4–6, Wolfgang Maier⁹¹, Jacques Mallet⁹², Pat Malloy²⁷, Manuel Mattheisen^19,21,93, Morten Mattingsdal^16,94, Kevin A McGhee²⁷, John J McGrath^39,40, Andrew McIntosh²⁷, Duncan E McLean⁹⁵, Andrew McQuillin²⁴, Ingrid Melle^16,17, Patricia T Michie^96,97, Vihra Milanova⁹⁸, Derek W Morris²³, Ole Mors⁵⁵, Preben B Mortensen⁹⁹, Valentina Moskvina^9,10, Pierandrea Muglia^100,101, Inez Myin-Germeys⁸⁴, Deborah A Nertney^39,40, Gerald Nestadt⁸⁵, Jimmi Nielsen¹⁰², Ivan Nikolov^9,10, Merete Nordentoft¹⁰³, Nadine Norton^9,10, Markus M Nöthen^19,21, Colm T O’Dushlaine²³, Ann Olincy⁷¹, Line Olsen²⁵, F Anthony O’Neill¹⁰⁴, Torben F Ørntoft^105,106, Michael J Owen^9,10, Christos Pantelis¹⁰⁷, George Papadimitriou⁶⁶, Michele T Pato²⁸, Leena Peltonen^{45,46,108,132}, Hannes Petursson¹⁰⁹, Ben Pickard¹¹⁰, Jonathan Pimm²⁴, Ann E Pulver⁸⁵, Vinay Puri²⁴, Digby Quested¹¹¹, Emma M Quinn²³, Henrik B Rasmussen²⁵, János M Réthelyi⁴⁴, Robert Ribble^4–6, Marcella Rietschel^91,112, Brien P Riley^4–6, Mirella Ruggeri⁴⁸, Ulrich Schall^97,113, Thomas G Schulze^112,114, Sibylle G Schwab^115–117, Rodney J Scott¹¹⁸, Jianxin Shi¹¹⁹, Engilbert Sigurdsson^109,120, Jeremy M Silverman^8,121, Chris C A Spencer⁶⁸, Kari Stefansson³⁸, Amy Strange⁶⁸, Eric Strengman^12,13, T Scott Stroup⁸⁸, Jaana Suvisaari⁸⁹, Lars Terenius⁴³, Srinivasa Thirumalai¹²², Johan H Thygesen²⁵, Sally Timm¹²³, Draga Toncheva¹²⁴, Edwin van den Oord¹²⁵, Jim van Os⁸⁴, Ruud van Winkel^42,82, Jan Veldink¹²⁶, Dermot Walsh¹²⁷, August G Wang¹²⁸, Durk Wiersma⁵⁷, Dieter B Wildenauer^115,129, Hywel J Williams^9,10, Nigel M Williams^9,10, Brandon Wormley^4–6, Stan Zammit^9,10, Patrick F Sullivan^{77,83,130,131}, Michael C O’Donovan^9,10, Mark J Daly¹ & Pablo V Gejman^2,3