Genome-wide association studies of cerebral white matter lesion burden: the CHARGE Consortium

Abstract

Objective

White matter hyperintensities (WMH) detectable by magnetic resonance imaging (MRI)are part of the spectrum of vascular injury associated with aging of the brain and are thought to reflect ischemic damage to the small deep cerebral vessels. WMH are associated with an increased risk of cognitive and motor dysfunction, dementia, depression, and stroke. Despite a significant heritability, few genetic loci influencing WMH burden have been identified.

Methods

We performed a meta-analysis of genome-wide association studies (GWAS) for WMH burden in 9,361 stroke-free individuals of European descent from 7 community-based cohorts. Significant findings were tested for replication in 3,024 individuals from 2 additional cohorts.

Results

We identified 6 novel risk-associated single nucleotide polymorphisms (SNPs)in one locus on chromosome 17q25 encompassing 6 known genes including WBP2, TRIM65, TRIM47, MRPL38, FBF1, and ACOX1. The most significant association was for rs3744028 (P_discovery= 4.0×10⁻⁹; P_replication =1.3×10⁻⁷; P_combined =4.0×10⁻¹⁵). Other SNPs in this region also reaching genome-wide significance are rs9894383 (P=5.3×10⁻⁹), rs11869977 (P=5.7×10⁻⁹), rs936393 (P=6.8×10⁻⁹), rs3744017 (P=7.3×10⁻⁹), and rs1055129 (P=4.1×10⁻⁸). Variant alleles at these loci conferred a small increase in WMH burden (4–8% of the overall mean WMH burden in the sample).

Interpretation

This large GWAS of WMH burden in community-based cohorts of individuals of European descent identifies a novel locus on chromosome 17. Further characterization of this locus may provide novel insights into the pathogenesis of cerebral WMH.

INTRODUCTION

The burden of vascular injury to the aging brain is far greater than that of clinically-manifest neurological conditions, such as stroke and dementia, and the harmful effects of brain vascular disease begin well before clinical symptoms become apparent.¹ Abnormalities of the white matter of the brain are readily identified on magnetic resonance imaging (MRI) as areas of signal hyperintensity on T2-weighted and fluid-attenuated inversion-recovery sequences. White matter hyperintensities(WMH), also known as leukoaraiosis, are remarkably common in healthy individuals beginning in middle age, with prevalence ranging from 40–70% in the fifth decade of age.² They contribute a substantial public health burden since they are associated with an increased risk of cognitive deficits, motor function impairment, dementia, depression, and stroke. The pathophysiology of WMH is poorly understood but may reflect ischemic damage to the small deep vessels of the brain resulting from arteriolosclerosis and leading to myelin degeneration or rarefaction, proliferative astrogliosis, fibrinoid changes and fibrosis of the vessel wall.³ In addition to age, elevated blood pressure is a major risk factor for WMH. Genetic factors appear also to play a significant role, with a heritability ranging from 55% to 80%.^4–6 Yet, studies conducted to date offer few specific clues about these genetic factors.^7–9 Genome-wide association studies (GWAS) provide a powerful tool for identifying genes related to complex traits.¹⁰ We have conducted a meta-analysis of GWAS data for WMH burden in 9,361 individuals of European descent (mean age 69.5 years; 42.6% men) and belonging to 7 community-based cohort studies: the Aging Gene-Environment Susceptibility-Reykjavik Study (AGES-Reykjavik); the Atherosclerosis Risk in Communities (ARIC) Study; the Austrian Stroke Prevention Study (ASPS); the Cardiovascular Health Study (CHS); the Framingham Heart Study (FHS); and two cohorts from the Rotterdam Study (RS). The participating studies were prospective population-based cohort studies with available cerebral MRI and genome-wide genotyping data and collaborating in the Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE) Consortium.¹¹

SUBJECTS AND METHODS

Consortium organization and study sample

The CHARGE consortium is an investigator-initiated collaboration to facilitate GWAS meta-analyses among multiple large and well-phenotyped cohort studies.¹¹ All participating studies worked cooperatively to address issues related to phenotype harmonization, covariate selection and to develop analytic plans for within-study GWAS analyses and meta-analysis of results. Each study received institutional review board approval of its consent procedures, cohort’s examination and surveillance, DNA collection and use, and data access and distribution. All participants in this study gave written informed consent for study participation, MRI scanning, and use of DNA.

Details of cohort recruitment, risk factor assessment, phenotyping and genotyping are described in the Supplementary Material. Briefly, in all cohorts, participants were excluded if they lacked information on MRI, genotypes, or both, or if they suffered a stroke or transient ischemic attack prior to MRI. In addition, CHS did not genotype participants with clinical cardiovascular disease at baseline. The ASPS and Rotterdam Study did not perform MRI scans in participants with dementia, and FHS analyses excluded participants who had dementia at the time of MRI scanning. Table 1 shows the number and characteristics of participants from each cohort.

Table 1.

Characteristics of study participants in discovery and replication cohorts

Characteristics	AGES-Reykjavik	ARIC†	ASPS	CHS†	FHS	Rotterdam Study I	Rotterdam Study II	3C-Dijon	AGES II
Number with MRI and genotype data	2762	820	765	2184	2319	421	591	1449	1773
Excluded for TIA or stroke	295	22	0*	0*	87	41	24	32	166
Number in analyses	2467	798	765	2184	2200	380	567	1417	1607
Mean ± SD age at MRI	76.1 ± 5.4	63.1 ± 4.4	65.1 ± 8.0	71.8 ± 4.9	63.9 ± 11.3	72.7 ± 7.7	67.2 ± 5.3	72.3 ± 4.1	76.4 ± 5.6
Number of men (%)	1007 (40.8%)	333 (41.7%)	331 (43.3%)	851 (38.97%)	1010 (45.9)	176 (46.4)	283 (49.9)	554 (39.1%)	672(41.8%)
Mean ± SD log(WMH+1)	1.19 ± 0.85	0.83 ± 0.38	0.89 ± 0.88	0.70 ± 0.22	0.69 ± 0.61	1.98 ± 1.21	1.22 ± 0.87	1.46 ± 0.67	1.20 ± 0.84
Mean WMH burden ± SD	3.84 ± 4.47	1.47 ± 0.98	3.12 ± 6.75	2.19 ± 1.40	1.65 ± 3.51	13.09 ± 14.58	5.21 ± 7.36	5.50 ± 4.88	3.79 ± 4.25
Median (IQR) WMH burden	2.0 (0.6; 5.8)	1.0 (1.0; 2.0)	0.9 (0.2; 3.1)	2 (1.0; 3.0)	0.6 (0.3; 1.4)	7.8 (0.7; 15.0)	3.1 (1.3; 5.0)	4.0 (2.8–6.4)	2.0 (0.6; 5.7)
Cardiovascular risk factor at MRI
Mean ± SD Systolic BP	142 ± 20	119 ± 17	143 ± 23	134 ± 21	127 ± 19	145.7 ± 20	143.5 ± 18	148.6 ± 23.1	142.9 ± 20.6
Hypertension (%)	1968(79.8%)	215 (27.1%)	503 (65.8%)	1082(49.5%)	940 (43.6%)	111 (29%)	391 (69%)	1080 (76.2%)	543 (80.8%)
Diabetes Mellitus (%)	255 (10.4%)	80 (10.1%)	68 (8.9%)	221 (10.1%)	261 (12.2%)	15 (3.9%)	51 (9.0%)	115 (8.1%)	184 (11.6%)
Current smoker (%)	308 (12.4%)	145 (18.2%)	92 (12.0%)	226 (10.3%)	252 (11.7%)	71 (18.7%)	167 (29.5%)	79 (5.6%)	181 (11.3%)
Prevalent CVD at MRI, %	328 (13.3%)	53 (6.8%)	68 (8.9%)	116 (5.3%)	242 (11.1%)	29 (7.6%)	38 (6.7%)	58 (4.1%)	229 (14.4%)

Definition of baseline characteristics was uniform across all studies: Hypertension was defined as systolic blood pressure (BP) ≥140 mm Hg, diastolic blood pressure ≥90 mm Hg, or being on antihypertensive treatment; Diabetes mellitus was defined as a casual or 2 hour postprandial blood glucose ≥200mg/dL, a fasting blood glucose ≥126mg/dL, or use of insulin or oral hypoglycemic agents; CVD (cardiovascular disease) was defined as presence of congestive heart failure, coronary heart disease or intermittent claudication.

^†ARIC and CHS expressed WMH as a grade (0–9 scale); all other studies expressed WMH as a volume (expressed in mL);

^*By design: In ASPS and CHS, participants with prevalent TIA or stroke were not included.

SD: Standard deviation; IQR: Inter-quartile range

MRI scans

In each study, MRI scans were performed and interpreted in a standardized fashion, without reference to demographic or clinical information. The field strength of the scanners used ranged from 0.5T to 1.5 Tesla. T1-and T2-weighted scans in the axial plane were obtained for all participants. These were complemented by either scans obtained with fluid attenuation inversion recovery or proton density sequences to allow better separation of white matter hyperintensities and cerebrospinal fluid. In AGES-Reykjavik, ASPS, FHS, and RS, WMH volume was estimated on a quantitative scale using custom-written computer programs based on an automatic segmentation algorithm or a semi-automatic segmentation analysis involving operator-guided removal of non-brain elements. In ARIC and CHS, WMH volume was estimated on a semi-quantitative 10-point scale by visual comparison with eight templates that successively increased from barely detectable white matter lesions to extensive, confluent abnormalities. Study participant’s brain images were compared with the reference standards after aligning them to approximately the same apparent size. Hence, visual grades are inherently normalized for brain size.

Comparability between the volumetric and visual scales was carefully evaluated. The distribution of WMH scores and the standardized slopes of associations between WMH scores and age were examined in each cohort and shown to be similar across the cohorts. Direct comparison of volumetric segmentation to visual scoring was performed using WMH measurements obtained by both methods in 1134 ARIC participants and showed a very good overall agreement (R²>0.80)(Supplementary Material).¹² In addition, volumetric analysis of the visual scoring standard image sets was conducted and showed excellent agreement on the log scale (R²=0.98) (Shibata, personal communication).

Genotyping

The participating studies used different genotyping platforms. Thus, genotype data in each study was used to impute to the 2.5 million non-monomorphic, autosomal SNPs described in the HapMap’s CEU panel. Extensive quality control (QC) analyses were performed in each cohort(Supplementary Material). Briefly, participant-specific quality controls filters were applied based on missing call rate, cryptic relatedness, and number of Mendelian errors per individual. SNP-specific quality controls included filters for call rate, minor allele frequency, Hardy-Weinberg equilibrium, differential missingness by outcome or genotype, and imputation quality.

Statistical analyses and meta-analysis

Within each study, a linear regression model was used to evaluate the association of the natural log-transformed volume of WMH(log(WMH+1)) with the number minor allele (0 to 2) at each SNP. Analyses were adjusted only for age, sex, and total intra-cranial volume (except in ASPS, ARIC and CHS). ARIC and CHS also adjusted for study site, and FHS adjusted for familial structure. The most significant association analyses were further adjusted for systolic blood pressure or for hypertension. All studies screened for latent population substructure, which was negligible(Supplementary Material).

A fixed-effects meta-analysis of results from the seven discovery cohorts was conducted using a z-score based method, as justified in the Supplementary Material. This approach is appropriate given that WMH was measured on different scales (mlor unit-less grade) in the various cohorts. The genomic control parameter was calculated and used to remove any residual population stratification within cohort and in the combined meta-analysis.¹³ We used a genome-wide significance threshold of 5×10⁻⁸. This threshold was first proposed based on the testing of 1 million SNPs and uses the simple Bonferroni correction for multiple tests.¹⁴ It is widely used in GWAS studies and is similar to genome-wide significance thresholds estimated by more complex methodologies.¹⁵

Replication and extension

Replication was sought for SNPs reaching genome-wide significance. The first replication sample comprised 1607 white participants (mean age 76.4±5.6) from the AGES-Reykjavik study who had not undergone genome-wide genotyping and, thus, were not part of the discovery sample. The second comprised 1417 white elderly non-institutionalized participants in the 3C-Dijon study (mean age 72.3±4.1). Association analyses in the replication cohorts used the same statistical models as those in the discovery cohorts. Results of the association of the genome-wide significant SNPs with WMH burden in the two replication cohorts were combined by meta-analysis using an inverse-variance weighted method. Significance and extent of heterogeneity were evaluated using the Cochran Q and I² statistics.¹⁶

We also sought to explore the association of SNPs in chromosome 17q25 region with WMH burden in 807 African-Americans from the ARIC study (mean age 61.5±4.5)with MRI and genotype data. Details on these groups are provided in the Supplementary Material.

RESULTS

There was no evidence of genomic control factor inflation within each cohort analyses and in the meta-analysis (λ=1.02) (Supplementary Material). Figure 1 shows the genome-wide plot of P-values for the individual SNPs against their genomic position. Six SNPs reached the genome-wide significance threshold of P<5×10⁻⁸, and an additional 56SNPs showed associations with P<1×10⁻⁵. They belonged to 15 loci (r²<0.5 between loci) (Table 2). At a threshold of P<1×10⁻⁴, a total of 416 SNPs were associated with WMH burden (Supplementary Table 1).

Figure 1.

Genome-wide association results of WMH burden. Meta-analysis P-values are plotted against their genomic position. The horizontal line indicates the threshold for genome-wide significance.

Table 2.

Genome-wide association results for SNPs associated with WMH burden at P<1×10⁻⁵.

SNP ID	Chr	Position (build 36.3)	Risk allele	Allele freq.	SNP function	Nearest gene/transcript	Distance from gene (kb)	Direction of association in each cohort*	P	Number of additional SNPs associated at
										P<10−5	P<10−4
rs3744028	17	71400267	C	0.18	Intronic	TRIM65	0	+++++++	4.0×10−9	8	8
rs1055129	17	71384543	G	0.30	Intronic	TRIM47	0	+++++−+	4.1×10−8	13	18
rs7894407	10	105166169	T	0.63	Intronic	PDCD11	0	++−++++	6.1×10−7	4	13
rs1892525	1	68128202	G	0.69	Within transcribed sequence	RP11-518D3.1	0	+++++++	7.2×10−7	10	29
rs10814323	9	36021610	A	0.21	Intergenic	RECK	−5.3	+++++++	1.7×10−6	0	17
rs6992136	8	15786858	G	0.85	Intergenic	RPL32P19	−58.1	+−+++++	3.2×10−6	8	9
rs11731436	4	66437855	C	0.64	Intergenic	AC097110.1	−82.1	+++++++	3.3×10−6	0	22
rs1052053	1	154468797	A	0.62	Missense	PMF1	0	+++++−−	5.0×10−6	2	12
rs2167089	3	29009407	G	0.73	Intronic	AC098970.2	0	+++++++	6.0×10−6	0	0
rs10012573	4	110168035	A	0.94	Intronic	COL25A1	0	++++++−	6.0×10−6	0	0
rs11625623	14	51829096	G	0.23	Intergenic	PTGDR	15.9	+++++++	7.7×10−6	0	4
rs16901064	5	31422754	C	0.84	Intergenic	RNASEN	13.6	+++++++	7.8×10−6	2	2
rs6945846	7	114190152	C	0.20	Intergenic	FOXP2	71.8	+++++++	7.9×10−6	0	31
rs11629135	14	63982864	G	0.93	Intronic	MTHFD1	0	+++++++	8.6×10−6	0	10
rs9410016	9	137360523	G	0.41	Intergenic	C9orf62	−14.4	+++++++	9.7×10−6	0	5

^*in alphabetical order: AGES-Reykjavik, ARIC, ASPS, CHS, FHS, RS-I, and RS-II

The most significant association was identified for rs3744028 at position 71400267 (Genome Build 36) on chromosome 17q25. In all studies, the C allele (frequency = 0.18) was significantly associated with a greater WMH burden (P=4.0×10⁻⁹). This SNP belonged to a larger cluster of associated SNPs spanning a 108 kb region (Figure 2). The other SNPs in this region also reaching genome-wide significance are rs9894383 (P=5.3×10⁻⁹), rs11869977 (P=5.7×10⁻⁹), rs936393 (P=6.8×10⁻⁹), rs3744017 (P=7.3×10⁻⁹), and rs1055129 (P=4.1×10⁻⁸). The first four were in strong linkage disequilibrium (LD) with rs3744028 and with each other (r²>0.8). The fifth (rs1055129) was in moderate LD with rs3744028 (r²=0.5). rs3744028 is located in intron 2 of the tripartite motif-containing 65 gene (TRIM65); rs1055129 and rs3744017 are located in introns 1 and 5, respectively, of the TRIM47 gene; rs936393 is located in intron 2 of the WW domain binding protein 2 gene (WBP2); and rs9894383 and rs11869977 are located in the region between the TRIM47 and WBP2 genes. After adjustment for systolic blood pressure, the significance of the association was P=1.5×10⁻⁹ for rs3744028 and P=1.2×10⁻⁸ for rs1055129. Similarly, after adjustment for hypertension, the significance of the association was P=1.0×10⁻⁹ for rs3744028 and P=1.1×10⁻⁸ for rs1055129.

Figure 2.

Regional association plot centered around rs3744028. Meta-analysis P-values for SNPs located within a 200 kb region on chromosome 17q25 are plotted against their genomic position.

The next most significantly associated locus(although not reaching genome-wide significance)was identified on chromosome 10q24 in a region that partially overlaps with a region recently identified to influence systolic blood pressure in two genome-wide association studies of whites.^{17, 18} Four SNPs showed highly suggestive associations, with the most significant SNP, rs7894407 (P=6.1×10⁻⁷), located in intron 12 of the programmed cell death 11 gene (PDCD11).

In addition to the two above loci, three additional SNPs with P<1×10⁻⁵ were located within known coding genes, including the polyamine-modulated factor 1 gene (PMF1), the collagen type XXV alpha 1 gene (COL25A1), and the methylenetetrahydrofolate dehydrogenase 1 gene (MTHFD1)(Table 2).

Association of rs3744028 and rs1055129 with WMH burden was replicated in an independent sample of 1607 AGES-Reykjavik participants. The C allele of rs3744028 and the G allele of rs1055129 were significantly associated with a greater WMH burden (P=1.3×10⁻⁵ and 4.7×10⁻⁴, respectively). Adjustment for systolic blood pressure or hypertension did not substantially change these results (rs3744028, P=2.2×10⁻⁵ and 1.3×10⁻⁵, respectively; rs1055129, P=7.1×10⁻⁴ and 6.0×10⁻⁴, respectively).

Association of these 2 SNPs was also investigated in 1417 elderly white participants from the 3C-Dijon study. The C allele of rs3744028 was significantly associated with a greater WMH burden (P=0.002). However, no association of rs1055129 with WMH burden was observed (P=0.11). Adjustment for systolic blood pressure or hypertension did not substantially change these results (rs3744028, P=0.001 and 0.001, respectively; rs1055129, P=0.11 and 0.09, respectively).

A summary meta-analysis of the 2 SNP association results in the replication cohorts and in all cohorts combined are shown in Table 3. In the replication cohorts, significance of the association of rs3744028 and rs1055129 with WMH burden was 2.0×10⁻⁷ and 5.9×10⁻⁴, respectively. The associated risk alleles increased WMH burden by a quantity of 0.12 and 0.07, respectively, on the transformed log-scale, with WMH expressed in ml. This is equivalent to an increase of 8.3% and 4.8%, respectively, of the overall mean WMH burden(expressed in the same scale) in this sample. In the combined sample of 12,385 individuals of European descent, the significance of the association of rs3744028 and rs1055129 reached P=4.0×10⁻¹⁵ and 2.6×10⁻¹¹, respectively. The direction of effects was generally consistent across the studies and there was no evidence of between-study heterogeneity within the replication cohorts (Table 3).

Table 3.

Summary of meta-analyses results for the two top SNPs in discovery and replication cohorts.

Meta-analyses in the discovery cohorts and in the combined discovery and replication cohorts were conducted using the z-score method; Meta-analyses in the replication cohorts were conducted using both the z-score method and the inverse-variance weighted method.

		Meta-analysis in discovery cohorts			Meta-analysis in replication cohorts						Overall meta-analysis
SNP ID	Risk allele	Allele freq.	Direction of association*	P†	Allele freq.	Direction of association	Beta (SE) ‡	P	Phet¶	I2 (95% CI)	P†
rs3744028	C	0.18	+++++++	4.0×10−9	0.19	++	0.12 (0.02)	1.3×10−7 † 2.0×10−7 ‡	0.45	0% (0%–33%)	4.0×10−15
rs1055129	G	0.30	+++++−+	4.1×10−8	0.31	++	0.07 (0.02)	2.7×10−4 † 5.9×10−4 ‡	0.22	33% (0%–56%)	2.6×10−11

^*in cohort alphabetical order;

^†from z-score-based meta-analysis;

^‡from inverse-variance weighted meta-analysis;

^¶P value for Cochran’s Q-statistic for heterogeneity

CI: confidence interval

We also investigated the association of SNPs in the chromosome 17q25 region with WMH burden in 807 African-Americans from the ARIC cohort. Allele frequencies were 0.23, 0.13, for rs3744028 and rs1055129, respectively. None of the top SNPs identified in whites was associated with WMH burden in this sample of African-Americans (P=0.66, rs3744028; P=0.91, rs1055129). However, assuming the same effects sizes as those observed in the white replication cohorts, power to detect an association was 57% for rs3744028 and 13% for rs1055129 in this small sample of African Americans. Because the pattern of LD in these regions differs between populations of European and African descent (Supplementary Figure 3), we examined the association of 7 additional genotyped SNPs located within a 50kb distance of the top SNPs on chromosome 17, respectively. None of the 7 SNPs on chromosome 17q25 was associated with WMH burden(Supplementary Table 2). The lowest P-value was obtained for rs6501841 within the Fas-binding factor 1 (FBF1) gene (P=0.11).

A search for putative functional SNPs among our highly suggestive association results identified 4 non-synonymous coding variants and 2 SNPs located in splice sites (Supplementary Material). Five of the six putative functional SNPs were located on chromosome 17q25. rs2290771 (P=4×10⁻⁷) in a splice site of the WBP2 gene and rs3760128 (P=8×10⁻⁷) encoding a leucine to proline substitution in TRIM65 were in moderate to high LD with one another and with rs1055129 (r²=0.7). We also screened SNPs tagging the chromosome 17q25 locus against a set of previously reported expression-associated SNPs (eSNPs). Six SNPs, in moderate to strong LD with rs3744028 or rs1055129, were significantly associated with variation in exon-level expression of FBF1 in cortical brain tissue.¹⁹ Seven SNPs, including rs3744028, were associated with variation in mRNA expression levels of TRIM47 in lymphoblastoid cell lines.^{20, 21}

We also examined whether the 2 SNPs on chromosome 17q25 were associated with incident ischemic stroke, incident dementia, and Alzheimer’s disease (AD) using pre-computed results from genome-wide association meta-analyses of these endpoints in the CHARGE cohorts.^{22, 23} Neither rs3744028 nor rs1055129 was associated with risk of incident ischemic stroke, dementia, or AD (Supplementary Table 3).

A systematic review of published candidate gene association studies identified a limited number of polymorphisms most frequently investigated for their association with WMH.⁸ Data were available for 15 of the 20 candidate SNP spreviously reported associated with WMH. Two SNPs reached the nominal level of significance of P<0.05 (rs5498 (K469E) in the intercellular adhesion molecule 1 gene (P=0.006) and rs699 (M235T) in the angiotensinogen gene (P=0.01)) but these associations were not significant after correction for multiple tests (Supplementary Material and Supplementary Table 4).

DISCUSSION

In this meta-analysis of genome-wide association data from 9,361 individuals of European descent, we identified novel genetic associations with WMH burden. Six SNPs mapping to a locus on chromosome 17q25 reached genome-wide significance, of which five were in strong LD. We replicated associations of two of these SNPs – rs3744028 and rs1055129 – in an additional 3,024 individuals of European descent. In a meta-analysis combining data from all 12,385 individuals, both showed strong associations with variation in WMH burden (rs3744028, P=4.0×10⁻¹⁵; rs1055129, P=2.6×10⁻¹¹). These SNPs belong to a large cluster of associated SNPs spanning a ~100 kb region, which contains several genes: Two tripartite motif-containing genes, TRIM65 and TRIM47; the WW domain binding protein 2 gene (WBP2); the mitochondrial ribosomal protein L38 gene (MRPL38); the Fas-binding factor 1 gene (FBF1); the acyl-coenzyme A oxidase 1 gene (ACOX1), and the C-Elegans homolog UNC13D gene. TRIM65 and TRIM47 are members of a superfamily of ring-finger B-box coiled-coil (RBCC) proteins involved in a broad range of biological processes including innate immunity ²⁴, apoptosis, cell cycle regulation, vesicular trafficking, and neuroprotection.²⁵ TRIM47 was first identified as over-expressed in astrocytomas.²⁶ WBP2 encodes a protein that was shown to interact with the WW domain of the Yes kinase-associated protein (YAP) and may play an important role in transcription regulation.²⁷ FBF1 was identified in a yeast two-hybrid assay to interact with the Fas cell surface receptor, a member of the TNF-receptor superfamily and a key regulator of apoptosis. Histopathological studies have shown evidence of apoptosis in white matter lesions.²⁸ Moreover, apoptosis-related transcripts have been shown to be differentially expressed between white matter lesions and normal white matter of brains collected postmortem from donors.²⁹ ACOX1 is the first enzyme of the fatty acid beta-oxidation pathway. Similar to other defects in peroxisomal beta-oxidation, ACOX1 deficiency is a rare disorder characterized by severe white matter abnormalities. Mutations in the UNC13D gene have been implicated in familial hemophagocytic lymphohistiocytosis, a rare disorder characterized by an uncontrolled proliferation and infiltration of activated lymphocytes and macrophages and an overproduction of inflammatory cytokines. Abnormalities of the brain white matter have been documented in this disorder.³⁰ The two SNPs identified in this genome-wide association study have been associated with variation in FBF1 expression in the brain and TRIM47 in lymphoblastoid cell lines.^{21, 31} In addition, they were in moderate to high LD with several putative functional polymorphisms in TRIM65 and WBP2, raising the possibility that multiple genes may be at play in the relationship between sequence variation at the 17q25 locus and WMH burden.

This prospective meta-analysis of GWAS data for WMH volume included a large sample size of individuals of European descent with high-quality genotypic and MRI data. In this study, white matter disease burden was defined as a continuous variable rather than a dichotomized one (presence/absence). Indeed, white matter disease burden is continuously-distributed and graded in the population and analysis of the continuous trait generally yields greater power to detect genetic effects.

Despite these clear strengths, some limitations of the study must be acknowledged. WMH measures were obtained on different scales in the studies. Although there is very good agreement between the visual and volumetric scales known differences in sensitivity, specificity, and precision between the semi-quantitative and quantitative methods have likely reduced our power to detect genetic effects. Effect sizes of the two SNPs on chromosome 17q25 estimated from a meta-analysis of the replication results showed that each of the identified risk alleles was associated with a small increase in WMH burden (4–8% of the overall mean WMH burden in this sample). Our analyses considered burden of WMH over the whole brain and did not address possible regional differences in the distribution of WMH. In particular, periventricular and subcortical WMH are thought to have different pathophysiology, risk factors, and functional consequences.³² However, the significant genetic correlation between these two forms of white matter disease suggests that they are influenced at least in part by common genetic factors.³³ While this GWAS analysis conducted in whites of European descent has the advantage of minimizing potential population stratification issues, our findings may not necessarily be generalized to other ethnic groups. Indeed, we did not find any evidence of an association of SNPs at 17q25 with WMH burden in African-Americans. The lack of extension to African-Americans should be interpreted cautiously as our small sample lacked power to detect associations of similar effect size as those observed in whites. In addition, reduced genomic coverage of the chromosome 17q25 locus in African-American may also explain the lack of significant findings in this population. Finally, we can also not exclude the possibility that genetic factors influencing WMH burden may not fully overlap between African-Americans and Whites. Linkage studies performed in various ethnic groups have also failed to identify common chromosomal regions influencing WMH.^{9, 33} Additional GWAS for WMH burden in African-Americans or other ethnic groups are needed.

Fine-mapping of the WMH locus on chromosome 17q25 was not attempted in this study. Hence, information on the specific genes and functional variant(s) underlying the reported associations is lacking. Neither of the two WMH SNPs was associated with risk of incident ischemic stroke, dementia, or late-onset AD in previously conducted GWAS analyses of these outcomes in the CHARGE cohorts.^{22, 23} This is perhaps not surprising in light of the small effect size of these SNPs on WMH burden and the complex relationships between white matter injury and its clinical expression. Indeed, it is generally believed that endophenotypes may be closer, in the biological hierarchy, to the genetic underpinnings than clinical disease outcomes, and thus, the impact of genetic variation on these endophenotypes is expected to be greater. The effects of the identified SNPs on WMH burden are small and likely even smaller on the clinical endpoints. Hence, our power to detect such effects was limited and studies adequately addressing the association of the WMH SNPs with these outcomes will require very large cohorts or clinical samples. The trend toward a protective effect of the 2 SNPs on incident ischemic stroke should be interpreted with caution: First, the association analyses were underpowered and likely resulted in imprecise effect estimates. Second, the samples used in the meta-analyses of WMH and clinical endpoints GWAS data were not identical and difference in LD between samples could lead to inconsistent patterns of associations when non-causal variants are tested.³⁴ Finally, association of the WMH SNPs with clinical endpoints may be influenced by unrecognized (and unaccounted) factors such as other genetic loci or environmental variables, and such interactions may complicate interpretation of the association results.

In summary, this large GWAS of WMH burden in community-based cohorts of individuals of European descent has identified a novel locus on chromosome 17q25 and provides a first step towards characterization of the biological mechanisms that influence the pathophysiology associated with WMH burden, an important risk factor of stroke, cognitive impairment and dementia.