Articles Genetic basis of lacunar stroke: a pooled analysis of individual patient data and genome-wide association studies Matthew Traylor, Elodie Persyn, Liisa Tomppo, Sofia Klasson, Vida Abedi, Mark K Bakker, Nuria Torres, Linxin Li, Steven Bell, Loes Rutten-Jacobs, Daniel J Tozer, Christoph J Griessenauer, Yanfei Zhang, Annie Pedersen, Pankaj Sharma, Jordi Jimenez-Conde, Tatjana Rundek, Raji P Grewal, Arne Lindgren, James F Meschia, Veikko Salomaa, Aki Havulinna, Christina Kourkoulis, Katherine Crawford, Sandro Marini, Braxton D Mitchell, Steven J Kittner, Jonathan Rosand, Martin Dichgans, Christina Jern, Daniel Strbian, Israel Fernandez-Cadenas, Ramin Zand, Ynte Ruigrok, Natalia Rost, Robin Lemmens, Peter M Rothwell, Christopher D Anderson, Joanna Wardlaw, Cathryn M Lewis, Hugh S Markus, Helsinki Stroke Study, Dutch Parelsnoer Institute-Cerebrovascular Accident (CVA) Study Group, National Institute of Neurological Disorders and Stroke (NINDS) Stroke Genetics Network, UK DNA Lacunar Stroke Study Investigators, International Stroke Genetics Consortium Summary Background The genetic basis of lacunar stroke is poorly understood, with a single locus on 16q24 identified to date. Lancet Neurol 2021; 20: 351–61 We sought to identify novel associations and provide mechanistic insights into the disease. Published Online March 25, 2021 https://doi.org/10.1016/ Methods We did a pooled analysis of data from newly recruited patients with an MRI-confirmed diagnosis of lacunar S1474-4422(21)00031-4 stroke and existing genome-wide association studies (GWAS). Patients were recruited from hospitals in the UK as part of See Comment page 329 the UK DNA Lacunar Stroke studies 1 and 2 and from collaborators within the International Stroke Genetics Consortium. Clinical Pharmacology and Cases and controls were stratified by ancestry and two meta-analyses were done: a European ancestry analysis, and a The Barts Heart Centre transethnic analysis that included all ancestry groups. We also did a multi-trait analysis of GWAS, in a joint analysis with and NIHR Barts Biomedical a study of cerebral white matter hyperintensities (an aetiologically related radiological trait), to find additional genetic Research Centre, Barts Health NHS Trust, William Harvey associations. We did a transcriptome-wide association study (TWAS) to detect genes for which expression is associated Research Institute, Queen Mary with lacunar stroke; identified significantly enriched pathways using multi-marker analysis of genomic annotation; and University of London, evaluated cardiovascular risk factors causally associated with the disease using mendelian randomisation. London, UK (M Traylor PhD); Department of Medical and Molecular Genetics Findings Our meta-analysis comprised studies from Europe, the USA, and Australia, including 7338 cases and (E Persyn PhD, 254 798 con­ trols, of which 2987 cases (matched with 29 540 controls) were confirmed using MRI. Five loci Prof C M Lewis PhD) and Social, (ICA1L-WDR12-CARF-NBEAL1, ULK4, SPI1-SLC39A13-PSMC3-RAPSN, ZCCHC14, ZBTB14-EPB41L3) were found Genetic, and Developmental to be associa­ ted with lacunar stroke in the European or transethnic meta-analyses. A further seven loci Psychiatry Centre (Prof C M Lewis), King’s College (SLC25A44-PMF1-BGLAP, LOX-ZNF474-LOC100505841, FOXF2-FOXQ1, VTA1-GPR126, SH3PXD2A, HTRA1-ARMS2, London, London, UK; COL4A2) were found to be associated in the multi-trait analysis with cerebral white matter hyperintensities (n=42 310). Department of Neurology, Two of the identified loci contain genes (COL4A2 and HTRA1) that are involved in monogenic lacunar stroke. The Helsinki University Hospital, TWAS identified associations between the expression of six genes (SCL25A44, ULK4, CARF, FAM117B, ICA1L, Helsinki, Finland (L Tomppo MD, D Strbian MD); NBEAL1) and lacunar stroke. Pathway analyses implicated disruption of the extracellular matrix, phosphatidylinositol Department of Laboratory 5 phosphate binding, and roundabout binding (false discovery rate <0·05). Mendelian randomisation analyses identified Medicine, Institute of positive associations of elevated blood pressure, history of smoking, and type 2 diabetes with lacunar stroke. Biomedicine, Sahlgrenska Academy at University of Gothenburg, Gothenburg, Interpretation Lacunar stroke has a substantial heritable component, with 12 loci now identified that could represent Sweden (S Klasson MSc, future treatment targets. These loci provide insights into lacunar stroke pathogenesis, highlighting disruption of the A Pedersen MD, Prof C Jern MD); vascular extracellular matrix (COL4A2, LOX, SH3PXD2A, GPR126, HTRA1), pericyte differentiation (FOXF2, Department of Molecular and GPR126), TGF-β signalling (HTRA1), and myelination (ULK4, GPR126) in disease risk. Functional Genomics, Weis Center for Research (V Abedi PhD), Neuroscience Funding British Heart Foundation. Institute (C J Griessenauer MD, R Zand MD), and Genomic Copyright © 2021 The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY 4.0 license. Medicine Institute (Y Zhang PhD), Geisinger Health System, Danville, PA, USA; Introduction microbleeds, and brain atrophy.3 Few therapeutic inter­ Department of Neurology and Lacunar strokes are small subcortical infarcts that arise ventions have been shown to reduce small vessel disease. Neurosurgery, Brain Center from ischaemia in the territory of the deep perforating One obstacle to developing new therapeutic approaches Rudolf Magnus, University Medical Center Utrecht, arteries of the brain.1,2 They account for up to a quarter of has been a lack of understanding of the underlying Utrecht, Netherlands all ischaemic strokes and are usually due to cerebral small pathophysiology. One method that has been successfully (M K Bakker MSc, Y Ruigrok MD); vessel disease, which is also the most common pathol­ used to discover pathophysiological processes and uncover Stroke Pharmacogenomics and ogy underlying intracerebral haemorrhage and vascular potential treatment targets in other complex diseases is the Genetics, Sant Pau Institute of Research, Hospital de la Santa cognitive impairment. Radiologically, small vessel disease use of genetic data derived from genome-wide associ­ation Creu I Sant Pau, Barcelona, is also characterised by the presence of cerebral white studies (GWAS). Recent GWAS have identified 35 loci Spain (N Torres PhD, matter hyperintensities, enlarged perivascular spaces, associated with risk of ischaemic stroke and its major I Fernandez-Cadenas PhD); www.thelancet.com/neurology Vol 20 May 2021 351 Articles Centre for the Prevention of Stroke and Dementia, Nuffield Research in context Department of Clinical Neuroscience, University of Evidence before this study pathogenesis, including extracellular matrix dysfunction, Oxford, Oxford, UK (L Li DPhil, Using the terms “stroke”, “small vessel stroke”, “lacunar stroke”, myelination, and pericyte differentiation. The current findings Prof P M Rothwell FMedSci); “small vessel disease”, “white matter hyperintensities”, also show that individuals with increased genetic predisposition Clinical Neurosciences, “genetics”, and “GWAS”, we searched PubMed and GWAS to elevated blood pressure, smoking, and type 2 diabetes are at University of Cambridge, Cambridge, UK Catalog for relevant reports published between Jan 1, 2010, increased risk of lacunar stroke, suggesting a causal role for (S Bell PhD, D J Tozer PhD, and Jan 1, 2020. We considered only peer-reviewed reports in these factors. Prof H S Markus FMedSci); English. Previously, a single locus on chromosome 16q24 has Product Development Implications of all the available evidence been robustly associated specifically with lacunar stroke; Personalized Health Care, No treatments for prevention of lacunar stroke are available, F Hoffmann-La Roche, by contrast, more than 30 loci have been associated with broad aside from management of vascular risk factors, such as blood Basel, Switzerland stroke phenotypes. (L Rutten-Jacobs PhD); Institute pressure. This is due in part to poor understanding of the of Neurointervention, Added value of this study mechanisms underlying the disease. Our findings highlight Paracelsus Medical University, The present findings substantially expand the number of novel mechanisms underlying lacunar stroke pathogenesis, and Salzburg, Austria genetic associations with lacunar stroke, with five loci now therefore point to pathways that have potential to be targeted (C J Griessenauer); Institute of Cardiovascular Research, Royal associated directly and a further seven associated with lacunar by therapeutics. Improved treatment of elevated blood pressure Holloway University of stroke jointly with white matter hyperintensities. These loci and type 2 diabetes, as well as prevention of smoking in the London, London, UK highlight several key mechanisms in lacunar stroke population, is likely to reduce the burden of lacunar stroke. (Prof P Sharma MD); Neurovascular Research Group, Department of Neurology of subtypes;4–6 but although many loci have been identi­fied For the GWAS, lacunar stroke cases were recruited Hospital del Mar-IMIM (Institut with the other major stroke subtypes (cardioembolic and from a combina­ tion of acute stroke admissions and Hospital del Mar d’Investigacions Mediques), large artery stroke), only one locus (16q24) has been outpatient ser­vices from Europe, the USA, South America, Universitat Autonoma de robustly shown to be associated with lacunar stroke to and Australia. Study inclusion criteria are detailed Barcelona/DCEXS-Universitat date.6 This is surprising because lacunar stroke is the in the appendix (pp 99–114). Cases and controls were Pompeu Fabra, stroke subtype most likely to be caused by monogenic stratified by ancestry (European, south Asian, African, Barcelona, Spain (Prof J Jimenez-Conde MD); disease,7 and sporadic lacunar stroke has been strongly Hispanic) and analysed separately. We did two meta- Evelyn F McKnight Brain associated with a family history of stroke.8 Additionally, analyses: a European ancestry analysis, and a transethnic Institute, Department of studies of other MRI markers of cerebral small vessel analysis that included all ancestry groups. For each Neurology, University of Miami disease have shown a substantial genetic component; a contributing study, approval for inclusion in this analysis Miller School of Medicine, Miami, FL, USA recent GWAS identified 31 loci across three phenotypes.9 complied with local ethical standards and with local (Prof T Rundek MD); institutional review board or ethics committee over­sight. Neuroscience Institute, Saint Methods All participants provided written informed consent for Francis Medical Center, School Study design and phenotype definitions genetic studies. of Health and Medical Sciences, Seton Hall University, We did a GWAS of lacunar stroke to identify novel Lacunar stroke samples were divided into two strata: South Orange, NJ, USA associations and provide mechanistic insights into the MRI-confirmed group and the standard phenotyping (Prof R P Grewal MD); lacunar stroke. We recruited cases of lacunar stroke group. In the MRI-confirmed group, lacunar stroke was Department of Neurology, from hospitals across the UK as part of the UK DNA defined as a clinical lacunar syndrome17 with an anatom­ Skane University Hospital, Lund, Sweden Lacunar Stroke studies 1 and 2,10 and from collaborators ically compatible lesion on MRI (subcortical infarct, (Prof A Lindgren MD); within the International Stroke Genetics Consortium; ≤15 mm in diameter), either as a region of high intensity Department of Clinical Sciences and re-analysed data from previous studies.11–14 Because on diffusion-weighted imaging for acute infarcts or as Lund, Neurology, Lund University, Lund, Sweden MRI confirmation of lacunar stroke is more reliable a region of low intensity on fluid-attenuated inversion (Prof A Lindgren); Department than standard phenotyping,15,16 we focused on recruiting recovery or T1 imaging for non-acute infarcts,3 and the of Neurology, Mayo Clinic, MRI-confirmed cases. First, we did a GWAS of these absence of causes of stroke other than small vessel disease. Jacksonville, FL, USA data to identify novel genetic loci associated with lacunar MRIs were centrally reviewed according to a standard (Prof J F Meschia MD); Department of Public Health stroke. Second, we used a multi-trait approach to detect proforma to confirm the diagnosis of lacunar stroke and Solutions, Finnish Institute for additional genetic variation associated with lacunar identify any exclusion criteria. All patients underwent Health and Welfare, Helsinki, stroke in a joint analysis with cerebral white matter comprehensive stroke investigation, including brain MRI, Finland (Prof V Salomaa MD, hyperintensities from a large-scale GWAS.9 Third, we imaging of the carotid arteries, and electro­cardiogram. A Havulinna DSc); Institute for Molecular Medicine Finland did a transcriptome-wide association study (TWAS) to Echocardiography was done if appropriate. Exclusion (FIMM HiLIFE), Helsinki, identify transcribed genes for which expression is criteria were stenosis of more than 50% in the extracranial Finland (A Havulinna); Center associated with lacunar stroke, and used mendelian or intracranial cerebral vessels, or previous carotid end­ for Genomic Medicine randomisation to assess common cardio­vascular risk arterectomy; cardioembolic source of stroke, defined (C Kourkoulis BS, K Crawford BS, S Marini MD, Prof J Rosand MD, factors that contribute to the disease. We did separate according to criteria from the TOAST trial11 as high or C D Anderson MD), Henry and analyses in MRI-confirmed and standard phenotyping moderate probability; cortical infarct on MRI; subcortical Allison McCance Center for groups to assess whether MRI confirmation improves infarct of more than 15 mm in diameter, because these Brain Health (C Kourkoulis, power to detect genetic associations. can be caused by embolic mechanisms (striato­capsular 352 www.thelancet.com/neurology Vol 20 May 2021 Articles infarcts); and any other specific cause of stroke (eg, lupus anticoagulant, cerebral vasculitis, dissec­tion, and GWAS in TOAST-phenotyped GWAS in MRI-confirmed GWAS of white matter lacunar stroke group lacunar stroke group hyperintensity volumes monogenic cause of stroke). (4351 cases, 225 258 controls) (2987 cases, 29 540 controls) (n=42 310) In the standard phenotyping group, lacunar stroke was defined according to the TOAST criteria,11 based on a clinical lacunar syndrome and the absence of other causes of stroke, or non-lacunar infarction on CT. Meta-analysis of MRI-confirmed Multi-trait analysis of lacunar and TOAST groups stroke and white matter (European analysis: 6030 cases, hyperintensities Genotyping and imputation 248 929 controls; transethnic analysis: 7338 cases, Genotyping arrays, quality control filters, and imputation 254 798 controls) reference panels are listed in the appendix (pp 7–8). All studies inferred the genetic ancestry of samples by comparison with reference populations using principal components analysis. European samples in this study are Mendelian randomisation to Transcriptome-wide association identify causal cardiovascular study and colocalisation to defined as those that segregated with European ancestry risk factors identify implicated genes reference samples. The majority of studies were imputed to the Haplotype Reference Consortium build. If this Figure 1: Analysis pipeline was not possible due to logistical or ethical reasons, GWAS=genome-wide association study. imputation to 1000 Genomes Phase 3 (all ancestry groups) panels was used.14,18 We defined significant loci as those containing SNPs Prof J Rosand, C D Anderson), Statistical analysis with a p value of less than 5 × 10–⁸ and in linkage equilibrium Department of Neurology (Prof J Rosand, C D Anderson), All studies used logistic regression to assess the association (r²>0·1) with other lead SNPs. If multiple loci met these and J Philip Kistler Stroke of single nucleotide polymorphism (SNP) allele dosages criteria within 1 Mb we used conditional and joint multiple Research Center with lacunar stroke, including ancestry informative princi­ SNP analysis (GCTA-cojo) to evaluate whether these SNPs (Prof N Rost MD), pal components as covariates as appropriate. All studies remained genome-wide significant in a joint modelling Massachusetts General Hospital, Boston, MA, USA; included cases with geographically matched controls, as scenario.23 We used Nagelkerke R² values to calculate Program in Medical & confirmed by principal components analysis. Some studies the proportion of variance explained by genome-wide Population Genetics, Broad had a combination of cases based on TOAST diagnosis significant SNPs using the NagelkerkeR2 function in the Institute of Harvard and MIT, of lacunar stroke, which were re-analysed for MRI R fmsb library.24 To obtain an estimate of R², we subtracted Cambridge, MA, USA (C Kourkoulis, K Crawford, confirmation of lacunar stroke, and cases based on TOAST the R² estimate for the model that included only principal S Marini, Prof J Rosand, diagnosis of lacunar stroke only, for which MRI was either components from the model that also contained the C D Anderson); Division of not acquired or was inaccessible. In these circumstances, genome-wide significant SNPs. We used a genome-based Endocrinology, Diabetes and Nutrition, Department of we analysed the MRI-confirmed and TOAST only (stan­ restricted maximum likelihood (GREML) approach, imple­ Medicine (Prof B D Mitchell PhD) dard phenotyping) groups separately and divided the study mented in GCTA software,25,26 and LD score regression,22 to and Department of Neurology controls between the two groups to avoid any sample estimate the heritability of lacunar stroke (MRI-confirmed (Prof S J Kittner MD), University overlap. Any cases with subsequent MRI confirmation of and non-confirmed). of Maryland School of Medicine, Baltimore, MD, USA; lacunar stroke were omitted from the TOAST only group; We did multi-trait analysis of GWAS (MTAG),27 Geriatrics Research and all individuals were analysed only once. performing a joint analysis with a large study of cerebral Education Clinical Center, Meta-analysis was done based on a fixed-effects inverse- white matter hyperintensities on MRI (n=42 310),9 which Baltimore Veterans variance weighted procedure using the METAL tool.19 shares a common cause with lacunar stroke through Administration Medical Center, Baltimore, MD, USA Meta-analysis was done in the MRI-confirmed and standard cerebral small vessel disease, to uncover additional genetic (Prof B D Mitchell, phenotyping groups separately, and in all studies com­ variation associated with lacunar stroke. We considered Prof S J Kittner); Institute for bined. We used the principles described by Winkler and associations significant if they had a p value of less than Stroke and Dementia Research colleagues20 to scrutinise datasets used in the meta-analysis. 5 × 10–⁸ in MTAG, had a p value of less than 0·05 for (ISD), LMU Munich, Munich, Germany (Prof M Dichgans MD); For each study, we filtered out SNPs with imputa­ tion association with white matter hyperintensities and lacunar Munich Cluster for Systems INFO scores of less than 0·7 or minor allele frequency stroke in univariate analysis, and showed greater signi­ Neurology (SyNergy), Munich, (MAF) of less than 0·01. Additionally, we removed low ficance in MTAG than in univariate analyses for white Germany (Prof M Dichgans); frequency or poorly imputed SNPs in smal­ler studies by matter hyperintensities or lacunar stroke. To confirm our Clinical Neurosciences, University of Helsinki, Helsinki, removing variants for which INFO × MAF × number of associations, we used an alternative approach, Bayesian Finland (D Strbian); cases equalled less than 2.5 Genomic control correction multivariate analysis of summary statistics (using Neurovascular Research based on genomic inflation factor λ was used for each study R package BMASS).28 Laboratory and Neurovascular to adjust for any residual inflation.21 Linkage disequilibrium We used the TWAS approach to identify genes for which Unit, Institut de Recerca, Hospital Vall d’Hebron, (LD) score intercept values were used to assess whether genetically altered expression was associated with lacunar Universitat Autonoma de population structure had been sufficiently resolved at stroke. Analyses were done using FUSION,29 from gene Barcelona, Barcelona, Spain the meta-analysis level.22 After meta-analysis, we excluded expression models derived from the Genotype-Tissue (I Fernandez-Cadenas); SNPs that were not present in at least 50% of cases. Expression (GTEx) portal V7,30 CommonMind Consortium Experimental Neurology, www.thelancet.com/neurology Vol 20 May 2021 353 Articles (CMC),31 and Young Finns Study (YFS) datasets.32 The RA=risk allele. OA=other allele. RAF=risk allele frequency. MTAG=multi-trait analysis of genome-wide association study. SNP=single nucleotide polymorphism. *Positions based on reference genome hg19. †ICA1L, CARF, NBEAL1, ULK4, and SLC25A44 CMC gene expression tissues (labelled as CMC brain) 2·2 × 10–⁷‡ 1·9 × 10–¹⁰ 7·9 × 10–¹³ 6·4 × 10–¹³ 6·0 × 10–¹³ 8·2 × 10–¹⁰ 5·3 × 10–¹⁶ 1·6 × 10–⁹ 5·2 × 10–⁹ 9·2 × 10–⁹ 3·1 × 10–9 were collected from the dorso­lateral prefrontal cortex of 0·0005 p value MTAG individuals with schizophrenia or controls (TWAS n=452). In the YFS study (labelled as YFS whole blood), peripheral blood gene expression was analysed (TWAS n=1264). 0·070 (0·01) 2·8 × 10–¹⁰ 0·050 (0·01) 7·9 × 10 ¹¹ 0·049 (0·01) 2·0 × 10–¹² 0·037 (0·01) 5·8 × 10–⁶ 1·8 × 10 ⁵ 0·050 (0·01) 1·4 × 10–⁸ 0·037 (0·01) 2·3 × 10–⁷ 0·030 (0·01) 3·1 × 10–⁵ 7·1 × 10–⁵ – – 0·019 (0·01) 0·0078 p value Among the available GTEx tissues, we focused our TWAS –0·011 (0·01) 0·33 0·015 (0·01) 0·12 analysis on the aortic artery (TWAS n=267), coronary hyperintensities artery (TWAS n=152), tibial artery (TWAS n=388), and White matter 0·057 (0·01) 0·045 (0·01) (n=42 310) whole blood (TWAS n=369), based on the assumption that these tissues would be the most relevant for lacunar β (SE) stroke pathogenesis. Bonferroni correction for multiple were associated in TWAS analysis and confirmed by colocalization. ‡Because A→T and C→G SNPs are removed by MTAG, results are presented for the SNP in highest linkage disequilibrium (rs9842261). testing was applied, taking into account the total number (transethnic: 7338 cases, 6·0 × 10–¹⁰ 1·6 × 10 ⁶ 6·4 × 10–⁹ 2·4 × 10 ⁸ 1·4 × 10–⁵ 9·0 × 10–⁸ 5·2 × 10–⁸ 1·2 × 10–⁶ 1·1 × 10 ⁶ 0·00050 0·00014 – – – of tested genes across the tissues (cutoff for significance: 0·0024 p value p<1·5 × 10–⁶). TWAS results were further investi­gated with 225 258 controls) Lacunar stroke colocalisation analysis of expression quanti­tative trait loci 1·09 (0·02) 1·09 (0·02) 1·22 (0·04) 1·22 (0·04) 1·16 (0·03) 1·19 (0·03) 1·18 (0·03) 1·14 (0·02) 1·07 (0·02) 1·12 (0·02) 1·12 (0·03) 1·11 (0·02) and GWAS signals with the R package COLOC,33 to assess OR (SE) whether the observed expression quantitative trait loci and GWAS associations were consistent with a shared association. In a bioinformatics analysis for novel associations, Studies we used Phenoscanner to query whether genome-wide 12 14 14 13 12 13 11 11 12 12 11 12 6030 cases, 219 389 controls) signifi­ cant SNPs were associated with DNA methyla­ Lacunar stroke (European: 7·2 × 10–¹⁰ 3·6 × 10–⁵ 9·9 × 10 ⁷ 2·5 × 10–⁸ 2·5 × 10–⁵ 3·5 × 10–⁵ 3·2 × 10 ⁶ 4·3 × 10 ⁶ 3·7 × 10–⁹ 1·7 × 10–⁷ 3·7 × 10–⁶ 1·7 × 10–⁵ tion,34,35 metabolite levels, or protein levels from genome- – – – p value wide studies at genome-wide significance (p<5 × 10–⁸) in other GWAS. We scanned DrugBank and the Drug Gene 1·25 (0·04) 1·18 (0·03) 1·14 (0·02) 1·21 (0·04) 1·14 (0·03) 1·10 (0·02) 1·10 (0·02) 1·15 (0·03) 1·13 (0·02) 1·17 (0·03) 1·11 (0·02) 1·11 (0·03) Interaction Database to assess the therapeutic potential of OR (SE) targeting associated genes.36,37 To identify biological pathways significantly associated with risk of lacunar stroke, we used the multi-marker 0·89 0·64 0·60 0·29 analysis of genomic annotation (MAGMA) tool.38 We first 0·83 0·34 0·67 0·81 0·07 0·10 0·77 0·11 RAF used MAGMA to calculate the significance of each gene based on association results, and then used these gene- RA/OA level statistics to estimate enrichment of Gene Ontology G/A A/G G/A A/G Table: Genome-wide significant loci for lacunar stroke in univariate or multi-trait analysis T/A T/C C/T T/C C/T C/T T/C T/C pathways from the Molecular Signatures Database using a gene-set enrichment analysis approach.39 We investigated rs61000833 rs79043147 rs12445022 rs72934535 rs11838776 rs9958650 rs2984613 rs7766042 rs4621303 rs2293576 rs2303655 only Gene Ontology terms containing at least four and Identifier rs225744 fewer than 200 genes and considered pathways attaining a false discovery rate (FDR) of less than 0·05 as being significantly associated with lacunar stroke. Downstream Intergenic Intergenic Intergenic Intergenic Genomic We did mendelian randomisation analyses to evaluate context Intronic Intronic Intronic Intronic Intronic Intronic Exonic whether any lipid (LDL, HDL, or triglycerides),40 blood pressure (systolic blood pressure, diastolic blood pressure, pulse pressure),41 metabolic (type 2 diabetes, body-mass 203 968 973 41 839 370 47 434 986 87 575 332 5 389 832 156 197 380 121 518 378 1 366 718 142 562 417 105 447 838 124 233 181 111 040 681 Genome-wide significance in multi-trait analysis position* index),42,43 or lifestyle risk factors (ever smoking) have a Genome-wide significance for lacunar stroke causal impact on lacunar stroke based on genetics.44,45 Chromosome Base Instrumental variables were independent (LD r²<0·01) genome-wide significant (p<5 × 10–⁸) variants associated with each trait from previous analyses, and are listed in the appendix (pp 9–74). For blood pressure traits, we 2 3 11 16 18 1 5 6 6 10 10 13 included SNPs associated at genome-wide significance ICA1L-WDR12-CARF- with any of the three traits in all analyses. For body-mass SLC25A44-PMF1- ZBTB14-EPB41L3 index, we used the set of independent SNPs provided by SPI1-SLC39A13- LOC100505841 HTRA1-ARMS2 PSMC3-RAPSN FOXF2-FOXQ1 VTA1-GPR126 LOX-ZNF474- study authors.42 We calculated the ratio of the SNP risk SH3PXD2A ZCCHC14 NBEAL1† factor effect size by the corresponding effect size for COL4A2 BGLAP† ULK4† lacunar stroke and aggregated effects across all risk factor- associated SNPs using an inverse-variance weighted 354 www.thelancet.com/neurology Vol 20 May 2021 Articles procedure. As secondary analyses, we used median, A weighted median, and MR-Egger approaches to aggregate SPI1-SLC39A13- across SNPs. We used the MR-Egger intercept to assess PSMC3-RAPSN evidence of directional pleiotropy. In all analyses, we used ZBTB14-EPB41L3 the MendelianRandomization package in R.46 Results are -CARF-NBEAL1 ICA1L-WDR12 presented as odds ratios (ORs) per genetically predicted ZCCHC14 ULK4 increase in each risk factor (original scale). 8 Role of the funding source The funder had no role in study design, data collection, 6 –log10(p value) data analysis, data interpretation, or writing of the report. Results 4 We meta-analysed studies from Europe, the USA, and Australia, including previous GWAS and additional cases 2 and controls from the UK DNA Lacunar Stroke studies and the International Stroke Genetics Consortium, B comprising a total of 6030 cases and 248 929 controls of -CARF-NBEAL1 ICA1L-WDR12 European ancestry, and 7338 cases and 254 798 controls in LOX-ZNF474-LOC100505841 the transethnic analysis (figure 1). 2987 (40·7%) cases SPI1-SLC39A13- HTRA1-ARMS2 PSMC3-RAPSN SLC25A44-PMF1-BGLAP (matched with 29 540 controls) had confirmation by MRI. 16 SH3PXD2A Study cohorts, including genomic inflation λ values, are COL4A2 14 described in the appendix (pp 5–6, 100–11). Following FOXF2-FOXQ1 VTA1-GPR126 meta-analysis, the LD score intercept value in the 12 ZCCHC14 –log10(p value) European analysis was 1·046 (SE 0·008) and the λ1000 value 10 was 1·007, whereas in the transethnic analysis, the 8 λ1000 value was 1·005, indicating no substantial inflation. SNP heritability (h²) of MRI-confirmed lacunar stroke, 6 calculated using GREML methods in a European ancestry 4 subset of 1693 cases and 10 171 controls genotyped on the 2 same array, was 0·17–0·21 (SE 0·02), assuming stroke 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 20 22 prevalence of 1–3%, and that 20% of these are lacunar Chromosomes strokes. Using LD score regression, estimates of SNP Figure 2: Manhattan plot of genome-wide SNP associations by genomic position herita­bility were lower than GREML estimates, but were (A) Lacunar stroke transethnic analysis. (B) Lacunar stroke multi-trait analysis. SNP=single nucleotide higher in the MRI-confirmed population (h²=0·065, polymorphism. Dashed lines indicate the genome-wide significance threshold of 5 × 10–⁸. SE 0·017) than in the non-MRI-confirmed population (0·0081, 0·0025). The genetic correlation between and slightly lower when including all lacunar strokes Department of Neurosciences, MRI-confirmed and non MRI-confirmed groups using (0·37, 0·09, p=4·0 × 10–⁵). In the joint analysis with KU Leuven, Leuven, Belgium (R Lemmens MD); VIB Center for LD score regression was significant (rg=0·61, SE 0·21, cerebral white matter hyperintensities, variants in seven Brain & Disease Research, p=0·0033). additional loci reached genome-wide significance for Department of Neurology, Five loci were associated with lacunar stroke: three in lacunar stroke overall (table, figure 2). Four of these loci University Hospitals Leuven, European samples and three in the transethnic analysis, (SLC25A44-PMF1-BGLAP, LOX-ZNF474-LOC100505841, Leuven, Belgium (R Lemmens); Centre for Clinical Brain with one locus associated in both (table, figure 2). SH3PXD2A, and COL4A2) have previously been associ­ Sciences, UK Dementia Regional association plots and forest plots for these loci ated with white matter hyperintensities.9,48 Regional Research Institute and Row are provided in the appendix (pp 115–134). Four of the loci association plots and forest plots for the loci are provided Fogo Centre for Research into were novel and one had been identified previously in the appendix (pp 120–134). the Ageing Brain, University of Edinburgh, Edinburgh, UK (16q24).6 One other locus (ICA1L-WDR12-CARF-NBEAL1) None of the 12 loci reaching genome-wide significance (Prof J Wardlaw FRCR) was associated in gene-based analyses in MEGASTROKE, showed evidence of heterogeneity (p=0·05 to p=0·98; Correspondence to: and was associated in a recent multi-trait analysis of appendix pp 123–134). In two regions (SH3PXD2A Prof Hugh S Markus, intracerebral haemorrhage and lacunar stroke.5,47 and HTRA1-ARMS2) multiple apparently independent Stroke Research Group, We next applied MTAG to identify additional genetic (LD r²<0·1) SNPs reached genome-wide significance. Clinical Neurosciences, University of Cambridge, variation underlying lacunar stroke in a joint analysis However, in a joint modelling scenario that was done Cambridge CB2 0QQ, UK with an aetiologically related trait, cerebral white matter using GCTA-cojo, only a single SNP at each of these hsm32@medschl.cam.ac.uk hyperintensities (n=42 310). Genetic correlation between regions had genome-wide significance, showing that a See Online for appendix lacunar stroke and cerebral white matter hyperintensities, single variant remains the most parsimonious explanation For FUSION software see calculated using LD score regression,22 was substantial for of the association at this locus.23 We discarded two regions https://github.com/gusevlab/ the MRI-confirmed group (rg=0·46, SE 0·10, p=4·6 × 10–⁶) according to our protocol: one region on chromosome fusion_twas www.thelancet.com/neurology Vol 20 May 2021 355 Articles Tissue Direction COLOC.PP4 discarded from this analysis. The ZBTB14-EPB41L3 locus CMC brain Negative 0·00 that was associated with lacunar stroke was not associated GTEx aortic artery Positive 0·25 with white matter hyperintensities (p=0·33 and effect in GTEx coronary artery 0·50 GTEx tibial artery 0·75 the opposite direction). Similarly, for the ULK4 locus GTEx whole blood 1·00 associated with lacunar stroke, the lead SNP did not reach YFS whole blood significance for white matter hyperintensities (p=0·12), Lacunar stroke Multi-trait analysis of but it was in the consistent effect direction and thus could lacunar stroke and WMH reflect a lack of study power. The 12 loci showed stronger effects in the MRI- SLC25A44 1q22 confirmed group than in the standard phenotyping group (in European ancestry analysis, appendix p 86), although the difference was not statistically significant (one-tailed p=0·07), with a median proportional increase in OR CARF of 3·4%. The 12 loci explained 1·4% of the overall herita­ bility, and 6·5–8·1% of the lacunar stroke heritability FAM117B from GWAS arrays, as calculated in this study. 2q33.2 We did a TWAS to identify genes for which expression was associated with lacunar stroke (figure 3). Genetically ICA1L elevated expression of SLC25A44 was associated with lacunar stroke in multi-trait analysis in arterial tissues, NBEAL1 whereas genetically decreased expression of ULK4 was associated with lacunar stroke in arterial tissues, whole blood, and the brain. At the 2q33·2 locus, genetically Gene symbol 3p22.1 elevated expression of CARF, FAM117B, ICA1L, and ULK4 NBEAL1 were all associated with lacunar stroke. All associ­ ations were confirmed by colocalisation analysis between the gene expression and lacunar stroke associ­ ations AGBL2 (posterior probability >0·7). Five other associations were identified in the TWAS, but these were not confirmed by DDB2 colocalisation analysis (posterior probability <0·7, figure 3). 11p11.2 We used Phenoscanner to evaluate whether the 12 lead SNPs were associated with DNA methylation, metabolite FNBP4 levels, or protein levels from large-scale studies.49–51 11 of the 12 lead SNPs showed associations with DNA methyla­ MADD tion at genome-wide significance, which was more than expected by chance based on randomly selected SNPs across the genome (p<0·01). Ten of the SNPs were associ­ 12q24.13 ated in multiple independent studies (appendix pp 87–97). ERP29 Conversely, none of the 12 SNPs were associated with metabolite or protein levels. Tissue Querying databases that catalogue drug-gene relation­ Figure 3: Genes for which expression is associated with lacunar stroke in six tissues from transcriptome-wide ships showed that 11 of the genes listed in table are association analysis categorised as druggable, indicating that they have poten­ Evidence of colocalisation of gene expression and lacunar stroke signals is shown by triangle size, with larger tial for therapeutic development (appendix p 98). However, triangles indicating stronger evidence of colocalisation. CMC=CommonMind Consortium. COLOC.PP4=posterior probability of hypothesis 4 in colocalisation analysis (that there is a consistent association between lacunar stroke no existing drugs target any of the genes identified in and expression of the given gene). GTEx=Genotype-Tissue Expression portal. WMH=white matter hyperintensities. this study. YFS=Young Finns Study. A pathway analysis based on the multi-trait analy­ sis results using MAGMA identified five significantly 17q25 showed an association solely with white matter associ­ated Gene Ontology gene sets: phosphatidyl­ hyperintensities, with no association with lacunar stroke inositol 5 phosphate binding (p=2·2 × 10–⁶, FDR 0·020), (lead SNP p=0·39); a second region on chromosome 14 extra­cellular matrix structural constituent (p=6·2 × 10–⁶, (EVL-DEGS2), was not as significant in MTAG analysis 0·027), extracellu­ lar matrix constituent conferring elas­ (p=1·2 × 10–⁹) as in white matter hyperintensities alone ticity (p=8·9 × 10–⁶, 0·027), middle ear morphogenesis (p=1·2 × 10–¹²), so an independent contribution of lacunar (p=2·3 × 10–⁵, 0·049), and roundabout binding (p=2·7 × 10–⁵, stroke to the association could not be determined. Further 0·049). No pathways were significant when based solely on evidence is required to determine whether these two lacunar stroke results. Results for all pathways with FDR of regions are associated with lacunar stroke, so they were less than 0·5 are presented in the appendix (pp 75–76). 356 www.thelancet.com/neurology Vol 20 May 2021 Articles Mendelian randomisation analyses using an inverse p value OR (95% Cl) variance weighted approach found positive associations of Systolic blood pressure 3·1 × 10–6 1·04 (1·02–1·06) diastolic, systolic, and pulse pressure, type 2 diabetes, and Diastolic blood pressure 3·1 × 10–5 1·06 (1·03–1·09) ever smoking with lacunar stroke (figure 4). No significant Pulse pressure 0·0025 1·04 (1·01–1·06) finding showed any evidence of pleiotropy, as assessed Type 2 diabetes 0·00019 1·10 (1·05–1·16) using the MR-Egger intercept. There was some evidence Body-mass index 0·84 0·99 (0·92–1·07) of a negative association between HDL and lacunar LDL 0·34 1·05 (0·95–1·15) stroke, but this result did not reach Bonferroni-corrected HDL 0·031 0·88 (0·78–0·99) significance. There was no evidence of an association with Triglycerides 0·066 1·14 (0·99–1·32) body-mass index, low density lipoprotein or triglycerides. Ever smoking 0·0025 1·25 (1·08–1·44) Secondary analysis for all risk factors using median, 0·6 0·8 1 1·4 1·8 weighted median, and MR-Egger approaches are presented in the appendix (pp 135–143). Favours no lacunar stroke Favours lacunar stroke Figure 4: Associations between genetically proxied cardiovascular risk factors and lacunar stroke from Discussion mendelian randomisation analysis using the inverse variance weighted method Despite the public health importance of lacunar stroke as Estimates are presented as ORs per genetically proxied increase in each risk factor (original scale). OR=odds ratio. the cause of a quarter of all strokes, previous GWAS studies have only identified one genetic locus associated proliferation and promotes apoptosis (erythrocyte mem­ with the disease, in contrast with the 35 identified for brane protein band 4·1-like 3). ischaemic stroke and its major subtypes.5 We did a In multi-trait analysis we identified seven further loci, all GWAS of lacunar stroke, including the largest number of which are reported as being associated with lacunar of cases with MRI confirmation to date, identifying stroke at genome-wide significance for the first time. 11 novel loci in addition to replicating the one previously Two have not been reported as being associated with any reported locus. cerebrovascular disease previously. One locus lies in an The primary analysis identified four novel loci. intergenic region between the VTA1 (vacuolar protein One association on chromosome 11, encompassing sorting-associated protein VTA1 homolog) and GPR126 SPI1-SLC39A13-PSMC3-RAPSN, was identified in both (G protein-coupled receptor 126) genes. GPR126 is activated European and transethnic analyses. The lead SNP is a by type IV collagen and has an important role in synonymous variant in SLC39A13, which encodes for myelination.58 GPR126 also binds laminin-211,59 an extra­ solute carrier family 39 member 13, a transmembrane cellular matrix protein produced by astrocytes and present protein with roles in zinc transport. Mutations in this in the brain, with key roles in development and function gene cause a form of Ehlers-Danlos syndrome, a group of of the blood-brain barrier,60 in part through regulation of connective tissues disorders that affect the vasculature pericyte differentiation, a mechanism previously impli­ and can cause stroke;52 vascular abnormalities have been cated through the FOXF2 gene.61,62 Small vessel disease- reported in SLC39A13-knockout mice.53 We additionally related endothelial dysfunction has also been shown identified a locus for which the lead SNP resides in an to prevent oligodendrocyte precursor cell maturation, intron of ULK4 (encoding for serine/threonine-protein contributing to impaired myelination.63 One hypothesis kinase ULK4) on chromosome 3. The TWAS analysis is that the GPR126 variant might exacerbate this process, suggests ULK4 is the most likely implicated gene, with inhibiting repair from myelin damage. The second genetically decreased expression of ULK4 being associated previously unreported association lies in an intergenic with lacunar stroke. Variants in close LD with the lead region, the nearest gene to which is HTRA1 (encoding for SNP have been implicated in diastolic blood pressure in serine protease HTRA1), a gene in which rare homozy­gous large-scale GWAS.54 However, the direction of effect was variation leads to cerebral autosomal reces­ sive arterio­ opposite to that for lacunar stroke, suggesting this likely pathy with subcortical infarcts and leucoencephalo­pathy.64 reflects pleiotropy rather than a causal pathway. Variants HTRA1, through processing of LTBP-1, promotes TGF-β in close LD have also been associated with another signalling in the vascular extracellular matrix (ECM).65 The cardiovascular disease, acute aortic dissection.55 ULK4 presence of both rare and common risk variants in HTRA1 belongs to the family of serine-threonine protein kinases, points to it being a key molecule in lacunar stroke a group of phosphorylating kinases involved in diverse pathogenesis, and is a feature shared with another gene processes including cell proliferation and differ­entiation, identified in this study, COL4A2, in which rare variants also apoptosis, and embryonic development. Its deficiency cause monogenic forms of cerebral small vessel disease.7 leads to hypomyelination,56 and it has been associated Candidate gene studies have previously shown associations with neuropsychiatric traits.57 Finally, we report a novel not reaching genome-wide signi­fi­cance in COL4A2 with association on chromosome 18, located between ZBTB14, lacunar stroke and the same region has also previously encoding a zinc finger transcription factor (zinc finger been associated with intra­cerebral haemorrhage in multi- and BTB domain-containing protein 14), and EPB41L3, trait analysis, and coronary artery disease.47,66,67 Four other encoding for a membrane protein that inhibits cell loci identified (SH3PXD2A, LOX-ZNF474-LOC100505841, www.thelancet.com/neurology Vol 20 May 2021 357 Articles SLC25A44-PMF1-BGLAP, FOXF2-FOXQ1) were associated the direct effects of LDL-lowering medications, such as with broad stroke in MEGASTROKE (associations of all statins, on the incidence of lacunar stroke is likely to SNPs in MEGASTROKE are shown in the appendix be small. pp 11–18) or a pre­ vi­ ous meta-analysis,5,62 although, to Our study emphasises the benefit of accurate pheno­ our knowledge, this is the first study to confirm their typing using MRI. Using this approach, the heritability of association specifically with lacunar stroke. At the lacunar stroke using GREML was substantial, and larger SLC25A44-PMF1-BGLAP locus, the TWAS results point to than previous estimates based on TOAST subtyping.74 an association of genetically elevated expression of Using LD score, the heritability was larger in the MRI- SLC25A44 with lacunar stroke, which was validated in confirmed group, but estimates were considerably lower colocalisa­tion analysis. SLC25A44 has an important role than for GREML. The use of MRI subtyping also increased in catabolism of branched-chain amino acids in brown the strength of association for the lacunar stroke-associated adipose tissue by transporting them into mitochondria,68 variants, although this increase was not statistically and thus has potential as a mediating factor in the significant. These results suggest that further genetic risk relationship between metabolic disease and lacunar stroke. factor studies in lacunar stroke are likely to be more However, variants in close LD have also been associated successful if MRI subtyping is used. with mosaic Y chromosome loss,69 highlighting mosaicism Our study has limitations. The analysis was done in a as an alternative mechanism. Further functional studies predominantly European ancestry population. Large will be required to untangle these relationships with studies including diverse ancestries should be done to lacunar stroke. The strength of the association for all assess the generalisability of our findings to all ethnic associated variants was moderate to large in the context of groups. The MTAG approach relies on the relatively GWAS (OR of 1·10 to 1·25 in Europeans) and notably strong assumption that associated variants act on both larger than effects previously reported for variants associ­ traits, which might not always be the case for white matter ated with broad stroke phenotypes.5 This observation is hyperintensities and lacunar stroke, because they reflect consistent with the variants acting specifically on the downstream effects of a shared common ancestor, small lacunar stroke subtype rather than on stroke as a whole. vessel disease. To control for this, we considered only We also found that 11 of the 12 lead SNPs influence SNPs showing an association with both traits and showing DNA methylation at genome-wide significance, pointing greater significance in MTAG analysis than with white to epigenetic changes being one source through which matter hyperintensities or lacunar stroke alone, as being risk of lacunar stroke is conferred. Whether this genetically significant. However, independent replication will remain altered DNA methylation influences transcription of the gold standard for confirming these and all other nearby genes, and which genes are affected, should be the reported associations in this Article. Previous studies have focus of further study. A pathway analysis implicated suggested that a more conservative threshold of a p value several biological processes in lacunar stroke patho­ less than 1 × 10–⁸ should be considered in GWAS using physiology. Two pathways involved the ECM, the network larger imputation panels such as this study.75 If we were to of extracellular molecules that provide scaffolding and use this threshold, one locus (ZBTB14-EPB41L3) would no biochemical support to surrounding tissues. Disruption longer be significant. Additional caution should therefore of the vascular ECM has been hypothesised to be a be applied when interpreting this finding, particularly as it key component in pathogenesis of cerebral small vessel was not significant in MTAG analysis. To increase sample disease, particularly in monogenic forms, and several size and study power, we used publicly available controls of the genes implicated in this study (COL4A2, LOX, in analyses. As such, we could not determine whether SH3PXD2A, GPR126, HTRA1) have key roles in the these individuals had a history of lacunar stroke. Our ECM.70 Therefore, our findings support this hypothesis analyses did not adjust for age and sex, and there is an and suggest ECM dysfunction also has a key role in ongoing debate about the importance of including such sporadic cerebral small vessel disease. covariates in genetic studies.76 In analyses with substantial We did Mendelian randomisation to assess whether differences between case and control populations, subtle cardiovascular risk factors showed evidence of a causal biases could occur. association with lacunar stroke. We found support for In summary, these findings represent substantial genetically predicted elevated blood pressure (systolic, progress in identifying the genetic mechanisms under­ diastolic, and pulse pressure), type 2 diabetes, and lying lacunar stroke, a disease for which little is known smoking being associated with lacunar stroke. The results about the molecular causes. Our findings highlight are consistent with those from observational studies and diverse mechanisms contributing to the disease, implicat­ suggest that targeting these factors would reduce risk ing disruption of the vascular ECM (COL4A2, LOX, of lacunar stroke.71 There was evidence, not reaching SH3PXD2A, GPR126, HTRA1), pericyte differenti­ ation Bonferroni-corrected significance, for a protective effect (FOXF2, GPR126), TGF-β signalling (HTRA1), and of increased HDL on risk of lacunar stroke, and no myelina­tion (ULK4, GPR126) in disease risk. These results association with LDL, replicating findings in previous provide novel insights into the pathogenesis of lacunar studies.72,73 Overall these findings show that the impact of stroke, and highlight multiple candidates for functional 358 www.thelancet.com/neurology Vol 20 May 2021 Articles experiments to identify the specific mechanisms con­ (20190203), and the Swedish state under the agreement between the ferring risk of lacunar stroke that could be targeted Swedish government and the county councils (the ALF agreement, ALFGBG-720081). The genetic data from Geisinger was made therapeutically. available through the collaboration with Regeneron Genetic Centre. Contributors The Edinburgh Mild Stroke Study (MSS2) was funded by the Wellcome MT, CML, and HSM designed the experiment. MT did the meta-analysis Trust (WT088134/Z/09/A). MT was supported by The Barts Charity and and downstream analyses. HSM provided oversight. EP analysed the the National Institute of Health Research (NIHR) Barts Biomedical TWAS. LT, SK, VA, MKB, NT, SB, and YZ did the statistical analysis. Research Centre. RL is a senior clinical investigator of FWO Flanders. LL, LR-J, DJT, CJG, AP, PS, JJ-C, TR, RPG, AL, JFM, VS, AH, CK, KC, PMR and the Oxford Vascular Study are funded by the NIHR Oxford SM, BDM, SJK, JR, MD, CJ, DS, IF-C, RZ, YR, NR, RL, PMR, CDA, JW, Biomedical Research Centre and the Wellcome Trust. CDA is supported and HSM were responsible for recruitment and phenotyping of cohorts. by the US National Institutes of Health (R01NS103924 and MT and HSM verified the data. All authors contributed to, reviewed, K23NS086873). CML is supported by the NIHR Biomedical Research and approved the final draft of the paper. All authors had full access to Centre at South London and Maudsley NHS Foundation Trust and King’s all the data in the study and had final responsibility for the decision College London. HSM is supported by an NIHR Senior Investigator to submit for publication. award, and his work is supported by the Cambridge Universities NIHR Comprehensive Biomedical Research Centre. Declaration of interests AL reports grants from the Swedish Heart and Lung Foundation, Region References Skåne, Skåne University Hospital, Freemasons Lodge of Instruction Eos 1 Pantoni L. Cerebral small vessel disease: from pathogenesis and in Lund, Lund University, the Färs & Frosta Foundation (a Sparbanken clinical characteristics to therapeutic challenges. Lancet Neurol 2010; Skåne Foundation), and the Swedish Research Council, during the 9: 689–701. conduct of the study; and personal fees from AstraZeneca, Bristol Myers 2 Wardlaw JM, Smith C, Dichgans M. Small vessel disease: Squibb/Pfizer, Portola, and Bayer, outside the submitted work. mechanisms and clinical implications. Lancet Neurol 2019; VS reports honoraria for consultation from Novo Nordisk and Sanofi 18: 684–96. and grants from Bayer, outside the submitted work. AH reports grants 3 Wardlaw JM, Smith EE, Biessels GJ, et al. Neuroimaging standards from the Academy of Finland, outside the submitted work. JR reports for research into small vessel disease and its contribution to ageing and neurodegeneration. Lancet Neurol 2013; 12: 822–38. grants from the US National Institutes of Health and OneMind, during the conduct of the study, and personal fees from Boehringer Ingelheim, 4 Malik R, Rannikmäe K, Traylor M, et al. Genome-wide meta-analysis identifies 3 novel loci associated with stroke. Pfizer, and New Beta Innovation, outside the submitted work. Ann Neurol 2018; 84: 934–39. PMR reports personal fees from Bayer and Bristol Myers Squibb, 5 Malik R, Chauhan G, Traylor M, et al. Multiancestry genome-wide outside the submitted work. CDA reports grants from the US National association study of 520,000 subjects identifies 32 loci associated Institutes of Health, the American Heart Association, Massachusetts with stroke and stroke subtypes. Nat Genet 2018; 50: 524–37. General Hospital, and Bayer, and personal fees from ApoPharma, 6 Traylor M, Malik R, Nalls MA, et al. Genetic variation at 16q24.2 is outside the submitted work. JW reports grants from the Wellcome Trust, associated with small vessel stroke. Ann Neurol 2017; 81: 383–94. during the conduct of the study. HSM reports personal fees from BIBA 7 Tan R, Traylor M, Rutten-Jacobs L, Markus H. New insights into Medical, outside the submitted work. All other authors declare no mechanisms of small vessel disease stroke from genetics. competing interests. Clin Sci (Lond) 2017; 131: 515–31. Data sharing 8 Jerrard-Dunne P, Cloud G, Hassan A, Markus HS. Evaluating the GWAS summary statistics from these analyses are available at GWAS genetic component of ischemic stroke subtypes: a family history For the GWAS Catalog see Catalog and on the Cerebrovascular Disease Knowledge Portal. study. Stroke 2003; 34: 1364–69. https://www.ebi.ac.uk/gwas/ Individual-level data from the NINDS Stroke Genetics Network study are 9 Persyn E, Hanscombe KB, Howson JMM, Lewis CM, Traylor M, summary-statistics available to researchers through dbGAP. Markus HS. Genome-wide association study of MRI markers of cerebral small vessel disease in 42,310 participants. Nat Commun For the Cerebrovascular Disease Acknowledgments 2020; 11: 2175. Knowledge Portal see http:// The UK Household Longitudinal Study is led by the Institute for Social 10 Kilarski LL, Rutten-Jacobs LCA, Bevan S, et al. Prevalence of www.cerebrovascularportal.org and Economic Research at the University of Essex and funded by the CADASIL and Fabry disease in a cohort of MRI defined younger For the NINDS Stroke Genetics Economic and Social Research Council. The survey was done by NatCen onset lacunar stroke. PLoS One 2015; 10: e0136352. Network data on dbGAP see and the genome-wide scan data were analysed and deposited by the 11 Adams HP Jr, Bendixen BH, Kappelle LJ, et al. Classification of https://www.ncbi.nlm.nih.gov/ Wellcome Trust Sanger Institute. Information on how to access the data subtype of acute ischemic stroke. Definitions for use in a projects/gap/cgi-bin/study. can be found on the Understanding Society website https://www. multicenter clinical trial. TOAST. Trial of Org 10172 in acute stroke cgi?study_id=phs000615.v1.p1 understandingsociety.ac.uk/. This research made use of the UK Biobank treatment. Stroke 1993; 24: 35–41. Resource under application number 36509. Ethical approval for UK 12 Bellenguez C, Bevan S, Gschwendtner A, et al. Genome-wide Biobank was received from the research ethics committee (REC reference association study identifies a variant in HDAC9 associated with 11/NW/0382). We are grateful to deCODE genetics for providing data for large vessel ischemic stroke. Nat Genet 2012; 44: 328–33. this analysis. We acknowledge the contribution of Giorgio Boncoraglio 13 NINDS Stroke Genetics Network (SiGN), International Stroke (Fondazione IRCCS Instituto Neurologico Carlo Besta, Milan, Italy). Genetics Consortium (ISGC). Loci associated with ischaemic stroke and its subtypes (SiGN): a genome-wide association study. This work, including collection and genotyping of the UK Young Lacunar Lancet Neurol 2016; 15: 174–84. Stroke DNA Study 2 (DNA Lacunar 2), was supported by a British Heart 14 McCarthy S, Das S, Kretzschmar W, et al. A reference panel of Foundation Programme Grant (RG/16/4/32218). The NINDS Stroke 64,976 haplotypes for genotype imputation. Nat Genet 2016; Genetics Network study was funded by the US National Institute of 48: 1279–83. Neurological Disorders and Stroke, National Institutes of Health 15 Rajajee V, Kidwell C, Starkman S, et al. Diagnosis of lacunar (U01 NS069208 and R01 NS100178). Collection of the UK Young Lacunar infarcts within 6 hours of onset by clinical and CT criteria versus Stroke DNA Study 1 (DNA Lacunar) was primarily supported by the MRI. J Neuroimaging 2008; 18: 66–72. Wellcome Trust (WT072952), with additional support from the Stroke 16 Markus HS, Khan U, Birns J, et al. Differences in stroke subtypes Association (TSA 2010/01). Genotyping of the DNA Lacunar samples was between black and white patients with stroke: the South London supported by a Stroke Association Grant (TSA 2013/01). The principal Ethnicity and Stroke Study. Circulation 2007; 116: 2157–64. funding for the WTCCC2 stroke study was provided by the Wellcome 17 Bamford J, Sandercock P, Dennis M, Burn J, Warlow C. Trust, as part of the Wellcome Trust Case Control Consortium 2 project Classification and natural history of clinically identifiable subtypes (085475/B/08/Z and 085475/Z/08/Z and WT084724MA). Collection of cerebral infarction. Lancet 1991; 337: 1521–26. and genotyping of the Sahlgrenska Academy Study on Ischemic Stroke 18 Auton A, Brooks LD, Durbin RM, et al. A global reference for were primarily supported by the Swedish Research Council human genetic variation. Nature 2015; 526: 68–74. (grant #2018–02543), the Swedish Heart and Lung Foundation www.thelancet.com/neurology Vol 20 May 2021 359 Articles 19 Willer CJ, Li Y, Abecasis GR. METAL: fast and efficient meta-analysis 44 Liu M, Jiang Y, Wedow R, et al. Association studies of up to of genomewide association scans. Bioinformatics 2010; 26: 2190–91. 1.2 million individuals yield new insights into the genetic etiology 20 Winkler TW, Day FR, Croteau-Chonka DC, et al. Quality control and of tobacco and alcohol use. Nat Genet 2019; 51: 237–44. conduct of genome-wide association meta-analyses. Nat Protoc 2014; 45 Haycock PC, Burgess S, Wade KH, Bowden J, Relton C, 9: 1192–212. Davey Smith G. Best (but oft-forgotten) practices: the design, 21 Devlin B, Roeder K. Genomic control for association studies. analysis, and interpretation of Mendelian randomization studies. Biometrics 1999; 55: 997–1004. Am J Clin Nutr 2016; 103: 965–78. 22 Bulik-Sullivan BK, Loh PR, Finucane HK, et al. LD score regression 46 Yavorska OO, Burgess S. MendelianRandomization: an R package distinguishes confounding from polygenicity in genome-wide for performing Mendelian randomization analyses using association studies. Nat Genet 2015; 47: 291–95. summarized data. Int J Epidemiol 2017; 46: 1734–39. 23 Yang J, Ferreira T, Morris AP, et al. Conditional and joint 47 Chung J, Marini S, Pera J, et al. Genome-wide association study of multiple-SNP analysis of GWAS summary statistics identifies cerebral small vessel disease reveals established and novel loci. additional variants influencing complex traits. Nat Genet 2012; Brain 2019; 142: 3176–89. 44: 369–75, S1–3. 48 Verhaaren BF, Debette S, Bis JC, et al. Multiethnic genome-wide 24 Nagelkerke NJ. A note on a general definition of the coefficient of association study of cerebral white matter hyperintensities on MRI. determination. Biometrika 1991; 78: 691–92. Circ Cardiovasc Genet 2015; 8: 398–409. 25 Lee SH, DeCandia TR, Ripke S, et al. Estimating the proportion of 49 Bonder MJ, Luijk R, Zhernakova DV, et al. Disease variants alter variation in susceptibility to schizophrenia captured by common transcription factor levels and methylation of their binding sites. SNPs. Nat Genet 2012; 44: 247–50. Nat Genet 2017; 49: 131–38. 26 Yang J, Lee SH, Goddard ME, Visscher PM. GCTA: a tool for genome- 50 Gaunt TR, Shihab HA, Hemani G, et al. Systematic identification of wide complex trait analysis. Am J Hum Genet 2011; 88: 76–82. genetic influences on methylation across the human life course. 27 Turley P, Walters RK, Maghzian O, et al. Multi-trait analysis of Genome Biol 2016; 17: 61. genome-wide association summary statistics using MTAG. 51 Chen L, Ge B, Casale FP, et al. Genetic drivers of epigenetic and Nat Genet 2018; 50: 229–37. transcriptional variation in human immune cells. Cell 2016; 28 Turchin MC, Stephens M. Bayesian multivariate reanalysis of large 167: 1398–1414.e24. genetic studies identifies many new associations. PLoS Genet 2019; 52 Bin BH, Hojyo S, Hosaka T, et al. Molecular pathogenesis of 15: e1008431. spondylocheirodysplastic Ehlers-Danlos syndrome caused by 29 Gusev A, Ko A, Shi H, et al. Integrative approaches for large-scale mutant ZIP13 proteins. EMBO Mol Med 2014; 6: 1028–42. transcriptome-wide association studies. Nat Genet 2016; 48: 245–52. 53 Hirose T, Shimazaki T, Takahashi N, et al. Morphometric analysis of 30 Battle A, Brown CD, Engelhardt BE, Montgomery SB. Genetic thoracic aorta in Slc39a13/Zip13-KO mice. Cell Tissue Res 2019; effects on gene expression across human tissues. Nature 2017; 376: 137–41. 550: 204–13. 54 Ehret GB, Munroe PB, Rice KM, et al. Genetic variants in novel 31 Fromer M, Roussos P, Sieberts SK, et al. Gene expression elucidates pathways influence blood pressure and cardiovascular disease risk. functional impact of polygenic risk for schizophrenia. Nat Neurosci Nature 2011; 478: 103–09. 2016; 19: 1442–53. 55 Guo DC, Grove ML, Prakash SK, et al. Genetic variants in LRP1 and 32 Raitakari OT, Juonala M, Rönnemaa T, et al. Cohort profile: ULK4 are associated with acute aortic dissections. Am J Hum Genet the cardiovascular risk in Young Finns Study. Int J Epidemiol 2008; 2016; 99: 762–69. 37: 1220–26. 56 Liu M, Xu P, Guan Z, et al. Ulk4 deficiency leads to 33 Giambartolomei C, Vukcevic D, Schadt EE, et al. Bayesian test for hypomyelination in mice. Glia 2018; 66: 175–90. colocalisation between pairs of genetic association studies using 57 Liu M, Xu P, O’Brien T, Shen S. Multiple roles of Ulk4 in summary statistics. PLoS Genet 2014; 10: e1004383. neurogenesis and brain function. Neurogenesis (Austin) 2017; 34 Staley JR, Blackshaw J, Kamat MA, et al. PhenoScanner: a database 4: e1313646. of human genotype-phenotype associations. Bioinformatics 2016; 58 Paavola KJ, Sidik H, Zuchero JB, Eckart M, Talbot WS. Type IV 32: 3207–09. collagen is an activating ligand for the adhesion G protein-coupled 35 Kamat MA, Blackshaw JA, Young R, et al. PhenoScanner V2: receptor GPR126. Sci Signal 2014; 7: ra76. an expanded tool for searching human genotype-phenotype 59 Mehta P, Piao X. Adhesion G-protein coupled receptors and associations. Bioinformatics 2019; 35: 4851–53. extracellular matrix proteins: Roles in myelination and glial cell 36 Wishart DS, Feunang YD, Guo AC, et al. DrugBank 5.0: a major development. Dev Dyn 2017; 246: 275–84. update to the DrugBank database for 2018. Nucleic Acids Res 2018; 60 Menezes MJ, McClenahan FK, Leiton CV, Aranmolate A, Shan X, 46: D1074–82. Colognato H. The extracellular matrix protein laminin α2 regulates 37 Cotto KC, Wagner AH, Feng YY, et al. DGIdb 3.0: a redesign and the maturation and function of the blood-brain barrier. J Neurosci expansion of the drug-gene interaction database. Nucleic Acids Res 2014; 34: 15260–80. 2018; 46: D1068–73. 61 Yao Y, Chen ZL, Norris EH, Strickland S. Astrocytic laminin 38 de Leeuw CA, Mooij JM, Heskes T, Posthuma D. MAGMA: regulates pericyte differentiation and maintains blood brain barrier generalized gene-set analysis of GWAS data. PLoS Comput Biol integrity. Nat Commun 2014; 5: 3413. 2015; 11: e1004219. 62 Chauhan G, Arnold CR, Chu AY, et al. Identification of additional 39 Subramanian A, Tamayo P, Mootha VK, et al. Gene set enrichment risk loci for stroke and small vessel disease: a meta-analysis analysis: a knowledge-based approach for interpreting genome-wide of genome-wide association studies. Lancet Neurol 2016; expression profiles. Proc Natl Acad Sci USA 2005; 102: 15545–50. 15: 695–707. 40 Teslovich TM, Musunuru K, Smith AV, et al. Biological, clinical and 63 Rajani RM, Quick S, Ruigrok SR, et al. Reversal of endothelial population relevance of 95 loci for blood lipids. Nature 2010; dysfunction reduces white matter vulnerability in cerebral small 466: 707–13. vessel disease in rats. Sci Transl Med 2018; 10: eaam9507. 41 Hoffmann TJ, Ehret GB, Nandakumar P, et al. Genome-wide 64 Hara K, Shiga A, Fukutake T, et al. Association of HTRA1 mutations association analyses using electronic health records identify new and familial ischemic cerebral small-vessel disease. N Engl J Med loci influencing blood pressure variation. Nat Genet 2017; 49: 54–64. 2009; 360: 1729–39. 42 Yengo L, Sidorenko J, Kemper KE, et al. Meta-analysis of 65 Beaufort N, Scharrer E, Kremmer E, et al. Cerebral small vessel genome-wide association studies for height and body mass index in disease-related protease HtrA1 processes latent TGF-β binding ~700000 individuals of European ancestry. Hum Mol Genet 2018; protein 1 and facilitates TGF-β signaling. Proc Natl Acad Sci USA 27: 3641–49. 2014; 111: 16496–501. 43 Mahajan A, Taliun D, Thurner M, et al. Fine-mapping 66 Rannikmäe K, Sivakumaran V, Millar H, et al. COL4A2 is type 2 diabetes loci to single-variant resolution using high-density associated with lacunar ischemic stroke and deep ICH: meta- imputation and islet-specific epigenome maps. Nat Genet 2018; analyses among 21,500 cases and 40,600 controls. Neurology 2017; 50: 1505–13. 89: 1829–39. 360 www.thelancet.com/neurology Vol 20 May 2021 Articles 67 Nikpay M, Goel A, Won HH, et al. A comprehensive 1,000 73 Georgakis MK, Malik R, Anderson CD, Parhofer KG, Hopewell JC, Genomes-based genome-wide association meta-analysis of coronary Dichgans M. Genetic determinants of blood lipids and cerebral artery disease. Nat Genet 2015; 47: 1121–30. small vessel disease: role of high-density lipoprotein cholesterol. 68 Yoneshiro T, Wang Q, Tajima K, et al. BCAA catabolism in brown Brain 2020; 143: 597–610. fat controls energy homeostasis through SLC25A44. Nature 2019; 74 Bevan S, Traylor M, Adib-Samii P, et al. Genetic heritability of 572: 614–19. ischemic stroke and the contribution of previously reported candidate 69 Wright DJ, Day FR, Kerrison ND, et al. Genetic variants associated gene and genomewide associations. Stroke 2012; 43: 3161–67. with mosaic Y chromosome loss highlight cell cycle genes and 75 Wu Y, Zheng Z, Visscher PM, Yang J. Quantifying the mapping overlap with cancer susceptibility. Nat Genet 2017; 49: 674–79. precision of genome-wide association studies using whole-genome 70 Joutel A, Haddad I, Ratelade J, Nelson MT. Perturbations of the sequencing data. Genome Biol 2017; 18: 86. cerebrovascular matrisome: a convergent mechanism in small vessel 76 Pirinen M, Donnelly P, Spencer CC. Including known covariates disease of the brain? J Cereb Blood Flow Metab 2016; 36: 143–57. can reduce power to detect genetic effects in case-control studies. 71 Rutten-Jacobs LCA, Markus HS. Vascular risk factor profiles differ Nat Genet 2012; 44: 848–51. between magnetic resonance imaging-defined subtypes of younger-onset lacunar stroke. Stroke 2017; 48: 2405–11. 72 Hindy G, Engström G, Larsson SC, et al. Role of blood lipids in the development of ischemic stroke and its subtypes: a mendelian randomization study. Stroke 2018; 49: 820–27. www.thelancet.com/neurology Vol 20 May 2021 361