Summary Background The discovery of disease-associated loci through genome-wide association studies (GWAS) is the leading genetic approach to the identifi cation of novel biological pathways underlying diseases in humans. Until recently, GWAS in ischaemic stroke have been limited by small sample sizes and have yielded few loci associated with ischaemic stroke. We did a large-scale GWAS to identify additional susceptibility genes for stroke and its subtypes. Methods To identify genetic loci associated with ischaemic stroke, we did a two-stage GWAS. In the fi rst stage, we included 16 851 cases with state-of-the-art phenotyping data and 32 473 stroke-free controls. Cases were aged 16 to 104 years, recruited between 1989 and 2012, and subtypes of ischaemic stroke were recorded by centrally trained and certifi ed investigators who used the web-based protocol, Causative Classifi cation of Stroke (CCS). We constructed casecontrol strata by identifying samples that were genotyped on nearly identical arrays and were of similar genetic ancestral background. We cleaned and imputed data by use of dense imputation reference panels generated from whole-genome sequence data. We did genome-wide testing to identify stroke-associated loci within each stratum for each available phenotype, and we combined summary-level results using inverse variance-weighted fi xed-eff ects meta-analysis. In the second stage, we did in-silico lookups of 1372 single nucleotide polymorphisms identifi ed from the fi rst stage GWAS in 20 941 cases and 364 736 unique stroke-free controls. The ischaemic stroke subtypes of these cases had previously been established with the Trial of Org 10 172 in Acute Stroke Treatment (TOAST) classifi cation system, in accordance with local standards. Results from the two stages were then jointly analysed in a fi nal meta-analysis. Findings We identifi ed a novel locus (G allele at rs12122341) at 1p13.2 near TSPAN2 that was associated with large artery atherosclerosis-related stroke (fi rst stage odds ratio [OR] 1·21, 95% CI 1·13–1·30, p=4·50 × 10–⁸; joint OR 1·19, 1·12–1·26, p=1·30 × 10–⁹). Our results also supported robust associations with ischaemic stroke for four other loci that have been reported in previous studies, including PITX2 (fi rst stage OR 1·39, 1·29–1·49, p=3·26 × 10–¹⁹; joint OR 1·37, 1·30–1·45, p=2·79 × 10–³²) and ZFHX3 (fi rst stage OR 1·19, 1·11–1·27, p=2·93 × 10–⁷; joint OR 1·17, 1·11–1·23, p=2·29 × 10–¹⁰) for cardioembolic stroke, and HDAC9 (fi rst stage OR 1·29, 1·18–1·42, p=3·50 × 10–⁸; joint OR 1·24, 1·15–1·33, p=4·52 × 10–⁹) for large artery atherosclerosis stroke. The 12q24 locus near ALDH2, which has previously been associated with all ischaemic stroke but not with any specifi c subtype, exceeded genome-wide signifi cance in the meta-analysis of small artery stroke (fi rst stage OR 1·20, 1·12–1·28, p=6·82 × 10–⁸; joint OR 1·17, 1·11–1·23, p=2·92 × 10–⁹). Other loci associated with stroke in previous studies, including NINJ2, were not confi rmed. Interpretation Our results suggest that all ischaemic stroke-related loci previously implicated by GWAS are subtype specifi c. We identifi ed a novel gene associated with large artery atherosclerosis stroke susceptibility. Follow-up studies will be necessary to establish whether the locus near TSPAN2 can be a target for a novel therapeutic approach to stroke prevention. In view of the subtype-specifi city of the associations detected, the rich phenotyping data available in the Stroke Genetics Network (SiGN) are likely to be crucial for further genetic discoveries related to ischaemic stroke.