Genetic correlations
We first applied stratified linkage disequilibrium (LD) score regression (S-LDSC)13 with the baseline-LD model14 to estimate the liability-scale SNP heritability for SCZ and OB. The liability-scale SNP heritability (without constrained intercept) was 36.9% (95% confidence interval [CI] =34.4~39.3%) for SCZ, 23.2% (95% CI, 22.1~24.3%) for OB. Then we used bivariate LDSC to estimate genetic correlations between SCZ and OB. The genetic correlation between SCZ and OB (rg= 0.12, p = 3.25E-14) was significant positive. In view of the fact that there is mild overlap between the samples, we conducted the LDSC with constraining intercept on the assumption of no population stratification. These estimates were slightly weaker, but remained significant (Table S1).
Local genetic correlations
The ρ-HESS (Heritability Estimation from Summary Statistics) method14 was adopted to determine local genetic correlations across the genome between SCZ and OB. After multiple correction, strong local correlations were found in 18 different regions, leading by 16q12.11 (chr16: 29036613..31382943) for SCZ with OB (P = 1.95E-19) (Figure 2; Table S2~S4).
Partitioned genetic correlation
SNPs associated with SCZ were enriched in 20 of 63 functional categories, leading by Human Promoter Villar (Enrichment = 13.1, P = 2.95E-05). SNPs associated with OB were enriched in 34 of 96 functional categories, leading by Conserved Primate (Enrichment = 10.8, P = 1.47E-18). There were 17 functional categories significantly associated with both SCZ and OB (Table S5&S6). To characterize genetic overlap at the level of functional categories, we calculated stratified-LDSC between SCZ and OB in those 17 functional groups (Figure 3, Table S7). SCZ was significantly correlated with OB at 8 out of 17 functional categories, with genetic correlation (rg) ranging from 0.12 (H3K27ac) to 0.15 (Conserved Mammal).
Identification of SNPs from cross-trait GWAS meta-analysis
Based on evidence for significant genetic correlations between SCZ and OB, we performed cross-trait meta-analyses to identify risk SNPs underlying the joint phenotypes SCZ-OB. We used two complementary approaches—MTAG (Multi-Trait Analysis of GWAS) 17 and CPASSOC (Cross Phenotype Association) 18. A total of 27 shared independent SNPs reached genome-wide significance (Table S8). After excluding SNPs that were significant in the single-trait GWAS of SCZ or OB, or were in LD (LD r2 ≥ 0.02) with any of previously reported significant SNPs, we identified 3 novel pleiotropic SNP (rs10777956, rs308697, rs79780963) associated with the joint phenotype SCZ-OB (Table S8). rs10777956 was mapped in ANKS1B, rs79780963 was mapped in NT5C2. rs308697?
Colocalization
Colocalization analysis was further performed to determine whether the genetic variants driving the association in two traits are the same or different. Most shared loci between SCZ and OB colocalized at the same candidate causal SNPs (PPH4 > 0.95) (20/27) (Table S9). Among the 3 novel pleiotropic loci, 2 loci (rs308697, rs79780963) showed evidence of colocalization (PPH4 > 0.95).
Mendelian randomization
We finally conducted a bi-direction two-sample MR using identified loci, which shows significant association with single phenotype GWAS of SCZ or OB, as IVs. All SNPs used in the Mendelian randomization analysis were strong instruments (F >10). As a result, evidence for the causal effect of SCZ on OB has been confirmed. However, due to the genetic commonality we found in the preceding part, the causality might be contributed by IVs with pleiotropy. After excluding IVs with potential directional horizontal pleiotropy, we re-conducted MR and found robust causality (OR 1.02, P <0.001). Egger intercept was not different from 0, which suggested that remained IVs were not pleiotropic. On the other hand, the causality of OB with SCZ was unstable. Before pleiotropy control, OB had a potential effect on increasing the risk of SCZ (OR 1.15, P= 0.015). Nevertheless, pleiotropy of IVs cannot be ruled out (Eggerintercept = 0.0085, P <0.001). We then performed MR-PRESSO and Radial-MR to minimize pleiotropy level and confirmed that OB had no significant effect on SCZ (OR 1.05, P =0.12) (Figure 4, Table S10).
Tissue-level SNP heritability enrichment
We adopted S-LDSC13 and Cell type specific analyses 28 to evaluate tissue-level enrichment of SNP heritability for SCZ and OB, using Genotype-Tissue Expression (GTEx) for different tissues. After adjusting for the baseline model, we identified tissues with shared significant SNP-heritability enrichment for SCZ and OB, which were all in brain (Figure 5, Table S11-12). In detail, SNPs associated with OB were specifically enriched in 6 different brain regions, leading by brain putamen and brain frontal cortex. Whereas, for SCZ, we observed the specific enrichment of SNPs in 13 different brain regions, leading by brain anterior cingulate cortex and brain frontal cortex (Table S11-12).
Identification of shared functional genes
We adopted SMR to recognize hypothetical functional genes underlying SCZ and OB, by jointly analysing GWAS summary data for SCZ, OB and eQTL summary data from GTEx (tissue shown SNP heritability enrichment in both SCZ and OB)30. In frontal cortex, we identified 90 and 150 non-MHC genome-wide significant genes associated with SCZ and OB, which passed the HEIDI-outlier test. Among these genes, 10 gene shared by SCZ and OB, leading by MICB (pSMR= 1.8E-06, pHEIDI=0.06; pSMR= 4.2E-04, pHEIDI=0.14). Additionally, 12,12,8,13, and 6 gene shared by SCZ and OB were detected in anterior cingulate, brain caudate, hypothalamus, putamen, and nucleus accumbens, respectively. ITIH4, DFNA5, GOS1, and FTSJ2 were found in most of the 6 shared enriched tissues (Figure 6, Table S13).