Supplementary Figure1 Results of Principal Component Analysis(A)(B) Results of GWAS case, control and the HapMap subjects of 4 populations (JPT, CHB, CEU and YRI) (A): Phase I <vector1/2>, (B): Phase II<vector1/2>

(C)(D) Results of the East Asian subject: GWAS case, control and JPT, CHB of the HapMap (C): Phase I<vector1/2>, (D): Phase II<vector1/2>, (E): Phase I <vector3/4>, (F): Phase II

<vector3/4>

Ctrl_Hondo: controls in Hondo (Japanese-Mainland) cluster, Ctrl_exceptHondo: controls in non-Hondo cluster (Ryukyu or Chinese: excluded from the association analysis), Case_Hondo:

case in Hondo (Japanese-Mainland) cluster, Case_exceptHondo: case in non-Hondo cluster (Ryukyu or Chinese: excluded from the association analysis)

(A) (B)

(C) (D)

(E) (F)

Supplementary Figure2 QQ plotsHorizontal and vertical axes show expected P values under a null distribution and the observed P values, respectively. The plots follow the line y = x under the null hypothesis of no

association.

(1) Combined GWAS of Japanese datasets (Phases I and II) (2) Phases I/II and PGC

Supplementary Figure 3: Regional association plot of the Combined GWAS of Japanese datasets (Phases I and II)

for the implicated loci by the previous studies

Blue lines indicate the recombination rate for the ASN ancestry in the 1000 Genome Project. The Y axis is -log 10(P-values) of the SNPs and the X axis is chromosomal position (hg19).

The linkage disequilibrium (r2) between the top and the remaining SNPs is indicated by color.

(A) ODZ4 (TENM4) (B) TRANK1

(C) MLL2~DHH (D) CACNA1C

(E) ANK3 (F) NCAN

(G) ADCY2 (H) 6q16.1

Supplementary Figure4: Correlation of the effect size for the top SNPs based on (A) meta-analysis of the Japanese

samples (Phases I and II) and the PGC-BD, (B) Combined GWAS of Japanese datasets only (Phases I and II) and (C)

PGC-BD dataset onlyOR: odds ratio, PhaseI/II: Meta-analysis of the Japanese samples, EUR_frequency: frequency of the CEU, JPN_frequency: frequency of our control samples

Overlap SNPs both in PhaseI/II and PGC-BD: excluding A/T or G/C SNPs

(A) Top SNPs (based on results of the meta-analysis of Phase I/II and the PGC-BD)

SNPs with P<5x10-5 SNPs with P<5x10-7

N of SNPs=663, r2=0.67 N of SNPs=70, r2=0.68 (B) Top SNPs (based on results of the

combined GWAS of Japanese

datasets only: Phase I/II)

SNPs with P<5x10-5 SNPs with P<5x10-7 SNPs with

P<5x10-7 (remove rs329674 as outlier)

N of SNPs=277, r2=0.055 N of SNPs=28, r2=0.12 N of

SNPs=27, r2=0.34

NOTE: most of the SNPs were located in FADS regions expect rs329674 which showed opposite direction

(C) Top SNPs (based on results of the PGC-BD dataset only)

SNPs with P<5x10-5 SNPs with P<5x10-7

N of SNPs=596, r2=0.17 N of SNPs=100, r2=0.0028

Supplementary Figure5: Quantiles plotsThe increasing risk profile score (RPS) was associated with increasing odds of detection of a phenotype. The threshold for selecting risk alleles was

the PT with the highest R2 values (for the Japanese subjects, Phase II/Phase I: discover/target pair, PT <0.5, Phase I/Phase II: discover/target pair, PT

<0.3 and the PGC/Phase I+II: discovery/target pair, PT <0.3). The RPS was converted to deciles (1 = lowest, 10 = highest RPS), and decile 1 was

defined as the reference to estimate the risk in each decline by comparing the case/control ratio. (A) Phase II/Phase I: discover/target pair, (B)

Phase I/Phase II: discover/target pair, and (C) PGC/Phase I+II: discovery/target pair.

(A) (B) (C)

Supplementary Figure6: “Leave-one-disease-out” analysis of RPS after excluding one “Non-psychiatric” controlY axis indicates the variance explained (Nagaelkerke’s R2). Target sets with bold are the main comparison (identical to the results in Figure 2). Target sets with underline are included in

Phase I