Genome-wide association analysis of susceptibility and clinical phenotype in multiple sclerosis

Abstract

Multiple sclerosis (MS), a chronic disorder of the central nervous system and common cause of neurological disability in young adults, is characterized by moderate but complex risk heritability. Here we report the results of a genome-wide association study performed in a 1000 prospective case series of well-characterized individuals with MS and group-matched controls using the Sentrix® HumanHap550 BeadChip platform from Illumina. After stringent quality control data filtering, we compared allele frequencies for 551 642 SNPs in 978 cases and 883 controls and assessed genotypic influences on susceptibility, age of onset, disease severity, as well as brain lesion load and normalized brain volume from magnetic resonance imaging exams. A multi-analytical strategy identified 242 susceptibility SNPs exceeding established thresholds of significance, including 65 within the MHC locus in chromosome 6p21.3. Independent replication confirms a role for GPC5, a heparan sulfate proteoglycan, in disease risk. Gene ontology-based analysis shows a functional dichotomy between genes involved in the susceptibility pathway and those affecting the clinical phenotype.

INTRODUCTION

Multiple sclerosis (MS) is a chronic and often disabling neurological disease associated with inflammation, loss of myelin and ensuing axonal damage in the central nervous system (CNS) (1). It is a relatively common disease, affecting approximately 1–2 per 1000 individuals; ancestry, latitude, sex, and age influence risk and incidence (2,3). Etiologically, MS is characterized by a polygenic heritable component that involves multifaceted interactions with environmental factors (4). Two genes, HLA-DRB1 and IL7R (CD127), have been unambiguously associated with disease susceptibility and/or protection following their identification as candidates by function (5–9). Additional disease genes were discovered through genome-wide association analysis (10,11), most notably the independently replicated IL2RA (CD25), CD58 (LFA3), CLEC16A (KIAA0350) and EVI5 (12–15), demonstrating the power of this approach to identify even the modest genetic effects that contribute to this complex neurological trait.

Clinical manifestations of MS are diverse and vary from benign to a rapidly evolving and ultimately incapacitating disease. Concordance in families for some early and late clinical features suggests that in addition to susceptibility, genomic variants influence disease severity or other aspects of the phenotype (16–19). It has been difficult to discern, however, whether phenotypic diversity reflects true etiological heterogeneity (20), modifying roles of specific genes (19), or some combination of the two. Unfortunately, published reports proposing specific genetic influences on the natural history of MS relied almost exclusively on small series of cases, retrospective clinical assessment and, in some cases, non-validated phenotypic endpoints. Furthermore, the confounding effects of drug treatment and population stratification have not been generally considered. It is also important to recognize that the aggregate contribution of individual germ line variants to the disease course may be quite modest. This is highlighted by observations that the clinical expression of MS can be different between monozygotic twin siblings. Nevertheless, the demonstration of even a modest genetic effect of a known gene on the course of MS could help elucidate fundamental mechanisms of disease expression and yield a major therapeutic opportunity.

To identify additional common variants associated with MS risk and phenotype, we conducted a genome-wide association study (GWAS) using the Sentrix® HumanHap550 BeadChip platform from Illumina in a 1000 prospective case series of well-characterized individuals with MS and matched controls. After stringent quality control, we compared allele frequencies for 551 642 SNPs in 978 cases and 883 controls and assessed genotypic influences on susceptibility, age of onset, disease severity, brain lesion load and normalized brain volume.

RESULTS

A total of 2068 individuals were enrolled within a 12 month period and genotyped using the Illumina Sentrix® HumanHap550 BeadChip platform. Following quality control filtering of SNPs and samples, the final dataset included 978 MS cases and 883 controls of European ancestry. Approximately, half of these were enrolled by the US site, with one-quarter from each of the two European sites. To assess bias introduced by population stratification, the entire dataset and each case–control group were scanned for the presence on non-random distribution of alleles using STRUCTURE (v.2.) (21) and principal component analysis (data not shown). This analysis resulted in the exclusion of 36 individuals (15 cases and 21 controls) enrolled at the US site and three cases from the EU sites. We then assessed the median distribution of test statistics with the genomic control parameter λ_GC (22): San Francisco λ_GC =1.03, Basel λ _GC =1.02, Amsterdam λ _GC =1.01; all sites combined, λ _GC =1.04. The results show that the variance inflation factors due to subpopulation structure within each cohort are less than 3%, and the pooled variance inflation factor is still less than 4%. In any case, origin of the sample was entered into the logistic regression model as a covariate to minimize the effects of the pooled variance inflation.

Table 1 lists demographic and clinical features of the study subjects. Overall, there was a high degree of demographic similarity among samples from the three sites, including frequency of HLA-DRB1*1501 carriers, but significant differences among sites were noted for some of the clinical variables used in the analysis of genotype–phenotype correlations. For all analyses, samples were pooled into a single dataset to increase statistical power (23).

Table 1.

Summary of demographic and clinical characteristics of study participants at the three sites at time of analysis

Variable	Total		San Francisco		Amsterdam		Basel
	Patients	Controls	Patients	Controls	Patients	Controls	Patients	Controls
Female, n (%)	653 (67)	584 (66)	334 (69)	287 (66)	137 (60)	148 (62)	182 (69)	149 (71)
Male, n (%)	325 (33)	299 (34)	153 (31)	147 (34)	92 (40)	90 (38)	80 (31)	62 (29)
Female:Male ratio	2.0:1	2.0:1	2.2:1	2.0:1	1.5:1	1.6:1	2.3:1	2.4:1
Age at onset of disease, mean (range)	34 (11–61)		34 (11–61)		35 (16–60)		32 (11–59)
Age at time of analysis, mean (range)	47 (21–69)	46 (21–74)	46 (21–69)	46 (23–69)	48 (25–67)	45 (24–68)	47 (21–69)	47 (21–74)
DRB1*1501+ individuals, n (%)	474 (48)	190 (22)	224 (46)	86 (20)	118 (52)	59 (25)	132 (50)	45 (21)
Disease duration (years) mean	11.5		10.7		11.4		12.9
Median (range)	9.0 (0–47)		8.0 (0–46)		10.0 (1–34)		11.0 (0–47)
Disease type, n (%)
Relapsing remitting	659 (67.9)		343 (70.4)		141 (61.8)		175 (68.4)
Secondary progressive	137 (14.1)		46 (9.5)		46 (20.2)		45 (17.5)
Primary progressive	72 (7.4)		18 (3.7)		29 (12.7)		25 (9.8)
Clinical isolated syndrome	100 (10.3)		79 (16.2)		10 (4.4)		11 (4.3)
Unknown	3 (0.3)		1 (0.2)		2 (0.9)		0 (0)
DMT, n (%)a	432 (54)		245 (63)		60 (32)		127 (58)
EDSS, n (%)b
<3	440 (55.4)		279 (71.9)		58 (31.0)		103 (47.0)
3 to <6	271 (34.1)		86 (22.2)		95 (50.8)		90 (41.1)
6–6.5	68 (8.6)		22 (5.7)		24 (12.8)		22 (10.1)
≥7	15 (1.9)		1 (0.3)		10 (5.4)		4 (1.8)
Mean (SD)	2.74 (1.81)		2.06 (1.64)		3.74 (1.72)		3.07 (1.67)
MSSS, nc	794		388		187		219
Median	2.97		2.23		4.82		3.17
Mean (SD)	3.45 (2.37)		2.70 (2.21)		4.81 (2.23)		3.61 (2.21)
T2 lesion load/mm3, nd	791		387		185		219
Median	2580		2164		3001		3276
Range	0–81317.2		0–64258.3		0–81317.2		0–34984.5
nBPV/cm3, ne	753		371		176		206
Mean (SD)	1530.8 (86.11)		1511.6 (81.91)		1554.5 (83.17)		1544.9 (88.55)
Range	1225–1767		1225–1704		1360–1750		1309–1767

^aDisease modifying treatment (DMT). Test of association with site: χ² = 50.64; P < 0.0001.

^bExpanded disability status scale (EDSS). Test of association with site: χ² = 102.17; P < 0.0001.

^cMultiple sclerosis severity scale (MSSS). Test of variation among sites (Kruskal–Wallis test): χ² = 107.82; P < 0.0001.

^dT2 lesion load/mm³. Test of variation among sites (Kruskal–Wallis test): χ² = 18.84; P < 0.0001.

^eNormalized brain parenchymal volume (nBPV). ANOVA of variation among sites: F = 19.77; P < 0.0001.

a–e analyses include only subjects with relapsing remitting and secondary progressive disease.

Figure 1 shows the genome-wide trend association results. As expected, the strongest association signal observed in this experiment was in the MHC region in chromosome 6p21. Q–Q plots of the association results are shown in Figure 2. P-values without adjustment for gender, DRB1*1501 and site of sample origin, and the corresponding P-values with adjustment, are plotted for all SNPs as a function of the null distribution, showing an excess in the tail of the distribution at P < 1 × 10⁻³ (without adjustment) and P < 1 × 10⁻⁴ (with adjustment). Following adjustment, the most extreme P-values are still smaller than expected, consistent with the presence of true genetic associations. P-values for SNPs in the HLA region are highlighted, showing that without adjustment, the SNPs from this region are concentrated among the low P-values, whereas with adjustment, they are distributed over the full range of P-values. Logistic regression analysis of these case–control data after adjusting for gender, site and DRB1*1501 identified 242 SNPs with P ≤ 1 × 10⁻⁴, including 65 within the MHC locus (Supplementary Material, Table S1.1).

Figure 1.

GWA trend test scan. Genome-wide association results for 551 642 SNPs in 978 MS cases and 883 controls using a linear trend test for additive allelic effects on disease penetrance (Cochran–Armitage trend test). As expected, the strongest association signal was located in the MHC region in chromosome 6p21.

Figure 2.

Q–Q plot distribution of the GWAS association results. P-values adjusted for HLA-DRB1*1501 and gender (A, trend test; B, genotypic test) are plotted for all SNPs as a function of the null distribution showing an excess in the tail of the distribution at P < 1 × 10⁻⁴. Red line is equal to expectation on H0. Grey/light green, P-values without adjustment for the HLA-DRB1 locus. Black/bright green, P-values with adjustment by including HLA-DRB1 as a covariate in the model. Green (light or bright), SNP loci in HLA region; black/grey, other SNPs.

The top peak SNPs located in the MHC class-II sub-region; rs3129934 (−Log10(P) 32.58) in the intron 3/4 of C6orf10, at the telomeric edge of the class II region, between BTNL2 and NOTCH4, rs9267992 (−Log10(P) 32.00) adjacent (telomeric) to LOC401252, rs9271366 (−Log10(P) 30.00) in the intergenic region between HLA-DRB1 and HLA-DQA1, and rs3129860 adjacent (telomeric) to HLA-DRA (Fig. 3A). It is likely that these signals reflect linkage disequilibrium (LD) with the well-established risk allele HLA-DRB1*1501, although following adjustment for gender and site, and upon conditional analysis for DRB1*1501, there is a residual effect in C6orf10 (−Log10(P) 4.00). The adjusted data also suggest the presence of independent association in the HLA-class I region (Fig. 3B), consistent with recent studies reporting multiple autonomous susceptibility signals in the locus (24,25). The top secondary signals are localized around TRIM 26, TRIM 15 and TRIM 10 members of the tripartite motif (TRIM) gene family. Although their functions are unknown, the presence of a RING domain suggests DNA-binding activity.

Figure 3.

Association and logistic regression analyses of the extended MHC region. Results of allelic tests of association (top panel) and results of subsequent logistic regression analyses conditioning upon HLA-DRB1*1501 (bottom panel).

Results for the non-HLA SNPs showing the strongest evidence of association (threshold of P ≤ 10⁻⁵) are shown in Table 2. These include 13 SNPs in eight chromosomes. Of special interest is the identification of Glypican5 (GPC5, 13q32), a heparan sulfate proteoglycan for which an IFNβ-associated pharmacogenomics role in MS was recently suggested (26). Also noteworthy are Tribbles Homolog 2 (TRB2) in 2p25.1-p24.3, a putative autoantigen in uveitis (27), and CUB and sushi multiple domains 1 (CSMD1) in 8p23.2, a complement regulatory protein highly expressed in the CNS (28). It is of interest to note that CMSD1 is located within a genomic region with a remarkable high mutation rate and as a consequence, is highly diverse between humans and non-human primates (29).

Table 2.

Top (non-HLA) markers associated with susceptibility, −Log10(P) ≥5.0

SNP	Chrom	Position	GeneSymbol	Alleles	Minor allele frequency	Odds ratio	Adj Geno Log P-valuea	Adj Trend Log P-valueb
rs397020	20	1153886	C20orf46	A/G	0.484	0.711	6.097	6.102
rs9523762	13	92129887	GPC5	A/G	0.346	1.361	5.155	5.921
rs1458175	12	40252128	PDZRN4	G/T	0.486	1.342	5.102	5.770
rs1529316	8	3815546	CSMD1	C/T	0.468	1.356	4.921	5.699
rs2049306	8	3819488	CSMD1	A/C	0.495	1.362	4.796	5.585
rs908821	3	142023416	SLC25A36	C/T	0.285	0.732	4.854	5.538
rs1755289	9	17928351	SH3GL2	C/T	0.389	0.739	5.041	5.523
rs16914086	9	98068032	TBC1D2	A/G	0.115	0.886	5.252	1.161
rs651477	2	119111921	EN1	A/G	0.263	1.378	4.699	5.143
rs7672826	4	182774844	MGC45800	A/G	0.341	1.366	4.886	5.102
rs1841770	3	149239384	ZIC1	G/T	0.470	0.746	4.310	5.076
rs1109670	2	9200636	DDEF2	A/C	0.256	1.381	4.000	5.046
rs7607490	2	12801718	TRIB2	A/G	0.115	1.558	4.000	5.013

^aGenotypic test of association (the logistic regression model, adjusted for covariates).

^bTrend test for association (the logistic regression model, adjusted for covariates).

Another gene of outstanding interest emerging from this scan is the aminophospholipid transporter (APLT), class I, type 8A, member 1 (ATP8A1), a member of a family of proteins driving uphill transport of ions across membranes. ATP8A1 also transports phosphatidylserine, a phospholipid component of myelin shown to be increased in the experimental autoimmune encepahlomyelitis brain, affecting the fluidity and integrity of the myelin sheath (30,31). In addition, phosphatidylserine on the surface of apoptotic cells serves as molecular addresses for the T-cell immunoglobulin mucin proteins for the effective clearance of apoptotic cells and prevention of autoimmunity (32).

Stratification by gender revealed additional genes of interest (Supplementary Material, Tables S1.2 and 1.3); 112 in the females only group, (39 intersecting with the all group and none with the males only group) and 60 in the males only group (nine intersecting with the all group). Some noteworthy examples include Tankyrase (TNKS) in females, a poly (ADP-ribose) polymerase associated with telomere length control and implicated in the regulation of EBV origin of plasmid replication (33), the prostaglandin receptor EP4 (PTGER4) in females, a gene recently associated with Crohn's disease (34), and the gamma-aminobutyric acid (GABA) A receptor beta 3 (GABRB3) in males, a gene coding for a subunit of a chloride channel that serves as the receptor for the gamma aminobutyric acid, a major inhibitory transmitter in the CNS and associated with Angelman syndrome, Prader–Willi syndrome and autism (35,36).

In order to overcome the potential bias for detection of false-positive associations in large genes with particularly dense SNP coverage, we applied the Šidák correction to the gene-wise minimum P-value. For each statistical and genetic model fitted to the case–control phenotype, values of P_min without correction and P_{min, corrected, adjusted} were obtained for 18 551 genes annotated in the Illumina array (counting different splice variants as different genes). The correlations among these gene-based measures of association are moderately strong (Supplementary Material, Table S2). The genes showing the strongest associations for each measure are, therefore, presented in Supplementary Material, Tables S3.1–3.4. The top MHC secondary signals (TRIM 26, TRIM 15 and TRIM 10) were also detected by the Šidák-corrected minimum P-value analysis of the data (Supplementary Material, Table S3). Outside chromosome 6p21, a group of genes located in chromosome 20p13 displayed robust association, including C20orf46, PSMF1 (Proteasome inhibitor subunit 1), SDCBP2 (Syndecan binding protein 2, or Syntenin) and SNPH (Syntaphylin or Synaptobrevin). The three top non-HLA hits identified in the screening, C20orf46, GPC5 and PCZRN4 (Table 2) survived this correction although GPC5 and PCZRN4 ranked lower than the genes mentioned above (Supplementary Material, Tables S3.1–3.4).

Given the lack of full-scale independent replication, these results were considered with caution. To add confidence to the associations reported here, we compared the results of the present analysis with those reported by the International Multiple Sclerosis Genetics Consortium (IMSGC) (11). SNPs showing support for association in both studies (thresholds P < 10⁻⁴) are shown in Table 3. R² values between the SNPs in the different studies are, however, very low. Nevertheless, the joint association of PARK2 (P = 0.0004), coding for an E3 ubiquitin ligase, is of interest because of a demonstrated role in Parkinson disease (37).

Table 3.

Top (non-HLA) susceptibility markers overlapping between the current and IMSGC studies

SNP	Chrom	Position	Gene	Alleles	Odds ratio	Adj Geno Log P-valuea	Adj Trend Log P-valuea	PTDT Log P-valueb	HapMap R2
rs11582200a	1	29810973	PTPRU	A/C	1.256	4.00	2.51	–	0.0001
rs4949238b	1	30091689	PTPRU	A/C	1.376	–	–	3.17
rs2144208a	6	162020735	PARK2	A/G	0.824	4.00	1.95	–	0.01
rs9458281b	6	161869691	PARK2	T/C	0.706	–	–	3.40
rs10283372a	8	74498045	RDH10	C/T	1.146	4.72	0.94	–	0.02
rs4738350 b	8	74478622	RDH10	A/G	0.469	–	–	3.40
rs2096735a	11	96368652	JRKL	C/T	1.151	4.00	1.27	–	0.0001
rs1941861b	11	97548590	JRKL	A/G	1.396	–	–	3.00
rs10140464a	14	48280389	RPS29	A/G	0.707	4.00	4.68	–	0.0001
rs10483556b	14	49029992	RPS29	G/A	0.740	–	–	2.89

^aTop SNP in current study.

^bTop SNP in IMSGC study (11).

Insofar genes reported by the Consortium and validated by others (11–15), a SNP within CD58 (rs1335532) ranked 16 094th (2.9 percentile) in the current study. The IMSGC SNP (rs12044852) was not represented in the Illumina platform, but according to HapMap data, both SNPs map to intron 10/11 and are in high LD (R² = 0.93) (Table 4). This indicates high concordance in the results of both studies and suggests that resequencing the flanking exons may identify a functional causative polymorphism in CD58. IL7R ranked 10 167th in this study (1.8 percentile). The exon 6 SNP rs6897932 is represented in both platforms and provides the strongest association in both studies, further supporting a functional role in disease susceptibility (8). The top IL2RA SNP in this study (rs11256497) ranked 7230th (1.3 percentile). Similar to the IMSGC top SNP (rs12722489), it is located in intron 1/2 but they are in relatively low LD (R² = 0.11). On the other hand, rs12722515 in this study, also in intron 1/2 is in high LD (R² = 0.93) and confers an OR = 1.21, but ranks 95 421th, suggesting allelic heterogeneity and that the causative variants may be located in the 5′ region of the gene. The best CLEC16A SNP in the current study was rs11074952 in intron 21/22 and ranked 1020 (0.18 percentile). This SNP is in weak LD with the IMSGC SNP rs28087 in intron 22/23 (R² = 0.24). In this study, the most tightly correlated SNP with the IMSGC is rs42369 (R² = 0.71), but shows no association ranking 113 789th, indicating the need for additional fine mapping of this intriguing gene. All the top-associated IMSGC SNPs are within the 3.0 percentile of markers in this study. On the other hand, the top EVI5 marker in this study (rs10747446, intron 10/11) is in high LD (R² = 0.92) with the IMSGC SNP (rs92833128, intron 10/11), but is not significantly associated and ranks 83 341th.

Table 4.

Performance of independently replicated IMSGC (non-HLA) Susceptibility markers in the current study

SNP	Chrom	Position	Gene	Alleles	Odds ratio	Adj Geno Log P-valuea	Adj Trend Log P-valuea	PTDT Log P-valueb	HapMap R2
rs10747446a,c	1	92820423	EVI5	A/G	1.00	0.99	0.12	–	0.92
rs10735781b	1	92833128	EVI5	C/G	1.29	–	–	3.65
rs1335532a,c	1	116812999	CD58	C/T	0.78	1.53	1.78	–	0.93
rs12044852b	1	116799821	CD58	T/G	0.68	–	–	3.00
rs6897932a,c	5	35910332	IL7R	C/T	1.21	1.48	2.00	–	–
rs6897932b	5	35910332	IL7R	T/C	0.81	–	–	2.23
rs11256497a	10	6127800	IL2RA	A/G	1.17	1.87	2.15	–	0.11
rs12722489b	10	6142018	IL2RA	T/C	0.74	–	–	2.88
rs12722525c	10	6121236	IL2RA	G/T	1.21	0.69	0.93	–	0.93
rs11074952a	16	11126107	CLEC16A	A/G	0.80	2.59	3.10	–	0.24
rs28087b	16	11160330	CLEC16A	G/A	1.16	–	–	1.56
rs42369a	16	11172113	CLEC16A	A/G	0.91	0.47	0.84	–	0.71

^aTop SNP in current study.

^bTop SNP in IMSGC study (11).

^cMost tightly correlated SNP in current study with top IMSGC SNP.

SNP rs7321717, located approximately 1240 kb from GPC5, showed a PTD P-value of 0.0004 in the IMSGC study. The metrics of LD between rs9523762 (this study) and rs7321717 (IMSGC) in GPC5 is 0.00014 (R²) and 0.01189 (D′). We followed the GPC5 data by genotyping 3 SNPs in an independent MS group consisting 974 affected individuals, recruited in the US using identical inclusion criteria. This dataset was compared to the control group used in the original screen generating robust evidence of replication when gender and HLA-DRB1*1501 status were included in the logistic regression model (Table 5).

Table 5.

Replication of GPC5

SNP	Allele, location	Wald Chi-square	Adjusted odds ratio (95% CI)a	Adj Trend Log P-value	Adj Geno Log P-value
rs553717	G/A, exon 3	1.01	A/G, OR = 1.12 (0.90–1.39)	0.50	0.22
rs9523762	G/A, intron 7-8	8.34	A/G, OR = 1.25 (1.07–1.45)	2.42	1.82
rs9516129	C/T, intron 7-8	5.77	T/C, OR = 1.23 (1.04–1.45)	1.79	1.66

^aAdditive model, covariates included in the logistic regression model: Gender, HLA-DRB1*1501(presence/absence).

Next, we performed a comprehensive statistical analysis to seek genome-wide associations with phenotypes related to the expression of MS. This analysis was restricted to individuals diagnosed as relapsing remitting or secondary progressive MS (N = 794) to limit the potential confounding influence of disease heterogeneity. Using regression analysis on the combined data from the three sites, and of data from each site separately with subsequent combination of the P-values by the Fisher's method, a number of allelic variants were modestly associated with the different phenotypic traits selected for this study (Supplementary Material, Tables S4–S7). Table 6 displays the top (P < 10⁻⁵) markers associated with the following phenotypic endpoints: age of onset, multiple sclerosis severity scale (MSSS), brain parenchymal volume (nBPV) and T2 lesion load.

Table 6.

Top markers associated with the MS phenotype, −Log10(P) ≥5.0

SNP	Chrom	Position	GeneSymbol	Alleles	Minor allele frequency	Adj Geno Log P-value	Adj Trend Log P-value
Age of onset
rs1386330	11	87459075	RAB38	C/T	0.13	5.82	6.66
rs7432623	3	176222035	NAALADL2	A/G	0.18	5.72	2.52
rs404694	16	78140299	MAF	T/G	0.27	5.61	0.73
rs17157903	7	103221987	RELN	C/T	0.14	4.06	5.49
rs2116078	8	73526543	KCNB2	T/G	0.48	5.00	5.48
rs1557351	18	52903312	WDR7	C/T	0.22	4.89	5.43
rs2842483	9	76135704	RFK	T/C	0.29	4.57	5.32
rs3804281	6	41853967	FRS3	C/T	0.46	5.31	0.72
rs12047808	1	176200971	C1orf125	A/G	0.13	4.82	5.24
rs5997184	22	25873695	CRYBA4	C/T	0.09	5.20	2.01
rs4704970	5	155433570	SGCD	A/G	0.20	3.59	5.13
rs1437898	2	133580362	FLJ34870	G/T	0.40	4.36	5.09
rs2803418	9	76138828	RFK	T/G	0.26	4.37	5.08
rs868824	7	109985872	IMMP2L	C/T	0.40	3.95	5.02
rs7914524	10	129994857	MKI67	C/T	0.17	5.01	3.19
MSSS
rs12553535	9	12902861	C9orf150	G/T	0.24	7.26	4.51
rs752092	15	99599457	CHSY1	C/T	0.37	5.84	1.69
rs12638253	3	158108793	FLJ16641	C/T	0.47	4.19	5.68
rs1478091	4	132148106	LOC132321	C/T	0.06	4.53	5.68
rs10936043	3	158179208	FLJ16641	A/G	0.46	4.16	5.66
rs8043243	15	99588976	CHSY1	C/T	0.42	5.53	2.34
rs2028455	18	46220003	C18orf24	A/G	0.31	5.47	3.44
rs299175	19	61005340	NALP11	C/T	0.46	4.28	5.37
rs10259085	7	7041671	C1GALT1	C/T	0.46	4.49	5.37
rs10518025	4	67893206	CENPC1	C/T	0.14	4.85	5.35
rs2035213	4	132145387	LOC132321	A/C	0.06	4.23	5.33
rs16925027	9	6968321	JMJD2C	A/G	0.24	5.33	0.66
rs6941421	6	15197130	JARID2	C/T	0.39	4.56	5.24
rs7191888	16	72138559	C16orf47	A/G	0.17	4.20	5.23
rs180358	11	116104602	MGC13125	G/A	0.23	4.51	5.22
rs10243024	7	115940554	MET	A/G	0.23	4.58	5.21
rs10516537	4	107985675	DKK2	A/C	0.15	5.10	3.77
rs337718	18	67925258	CBLN2	T/C	0.29	4.38	5.07
rs7253363	19	11543495	ACP5	G/T	0.05	4.66	5.05
rs12142240	1	46459321	LRRC41	C/T	0.29	4.52	5.00
rs7211577	17	14055005	COX10	A/G	0.49	4.96	5.00
Brain parenchymal volume
rs4866550	5	3361312	IRX1	C/T	0.32	6.06	2.09
rs10078091	5	25530762	CDH10	A/G	0.27	5.91	4.66
rs368380	20	14762090	C20orf133	C/T	0.33	5.73	1.47
rs4473631	4	174876499	MORF4	A/C	0.22	5.55	0.57
rs1869410	2	5207954	SOX11	C/T	0.28	5.40	3.70
rs261902	12	32367994	BICD1	T/C	0.16	4.42	5.36
rs11719646	3	56442731	CAST1	A/G	0.38	5.29	3.30
rs1354913	11	89794461	CHORDC1	A/G	0.18	5.23	0.16
rs13067869	3	175148379	NLGN1	G/T	0.07	5.20	4.83
rs9307252	4	102359544	PPP3CA	C/T	0.14	5.18	4.16
rs9480865	6	109023266	FOXO3A	C/T	0.16	4.52	5.19
rs9486902	6	108984745	FOXO3A	C/T	0.16	4.52	5.12
rs1927457	10	30048669	SVIL	C/T	0.31	4.69	5.11
rs716595	10	111996476	MXI1	A/G	0.08	3.81	5.10
rs11957313	5	169882972	KCNIP1	A/G	0.13	3.73	5.07
rs9319189	13	85516099	SLITRK6	A/G	0.35	5.01	2.54
rs10917727	1	160028398	CDCA1	C/T	0.47	5.00	4.28
T2 lesion load
rs12097667	1	94209	PLD5	A/C	0.15	6.38	0.73
rs1806468	7	212245	KIAA1706	C/T	0.11	5.65	4.13
rs146250	6	158286	GPR126	A/G	0.09	5.55	1.66
rs263153	6	279431	HIVEP2	C/A	0.09	5.55	1.74
rs6794496	3	422592	NPHP3	C/T	0.13	5.47	4.80
rs2602397	2	277843	CHRND	C/T	0.45	4.48	5.39
rs6899560	6	429349	FUT9	A/G	0.05	5.26	2.56
rs2039485	14	232964	NUBPL	C/T	0.22	4.63	5.22
rs305124	4	303690	HIP2	C/T	0.07	5.20	0.68
rs6917747	6	430506	IGF2R	A/G	0.15	4.72	5.17
rs11666377	19	74887	CPAMD8	C/T	0.14	4.42	5.15
rs12202350	6	98302	IGF2R	C/T	0.09	5.10	4.71
rs6512158	19	405242	CPAMD8	A/C	0.14	4.29	5.02

In order to explore the potential functionality of the associated SNPs across phenotypes, we conducted a gene ontology (GO) analysis of the most significant markers to assess enrichment in any of the pre-defined functional categories (Biological process, molecular function and cellular component) using a GO tree machine (Vanderbilt University) (Table 7). In this analysis, we included genes containing at least one SNP with a P-value <1 × 10⁻⁴ for each of the traits (Supplementary Material, Tables S1.1, 4–7). To avoid redundancy, only the most specific of any given set of nested significant GO categories is reported. Antigen processing and presentation were among the most significantly enriched GO categories for disease susceptibility, together with, surprisingly, CNS development. On the other hand, neural processes including axon guidance and glutamate signaling were over-represented in phenotypes directly related to CNS damage (i.e. T2 lesion load and brain volume, respectively). Cellular mechanisms were primarily represented in time-dependent processes such as MSSS and age of onset.

Table 7.

Gene ontology analysis

GO (Biological process)	n	Genes (observed)	Genes (expected)	Ratio (O/E)	P-value
Susceptibility
CNS development	6	MOG, PARK2, SH3GL2, ZIC1, CHST9, JRKL	0.52	11.5	0.00001
Organ morphogenesis	5	SPRY2, CITED2, ABLIM1, NPR1, ZIC1	0.98	5.1	0.003
Antigen presentation (MHC class I)	3	HLA-A, HLA-B, HLA-G	0.09	33.3	0.00009
Antigen presentation (MHC class II)	3	HLA-DPB1, HLA-DQA2, HLA-DQB1	0.08	35.7	0.00006
MSSS
Protein amino acid N-linked Glycosylation	2	FUT8, TM4SF4	0.15	13.3	0.009
Cellular respiration	2	ME3, COX10	0.15	13.3	0.009
Embryonic development	2	FUT8, KLF4	0.1	20.0	0.005
Age of onset
Cell adhesion	9	CDH12, DLG1, CNTN6, OPCML, PCDH10, TPBG, PPFIBP1,CASK PSCD1	3.3	2.8	0.005
Signal transduction	25	FRS3, RASSF8, PDZD8, CPE, DAPK1, DOCK1, EDNRB, DKK1, RASD2, RAB38, RASGRP3, CNTN6, GRIK1, HTR7, KDR, OR51B6, OR51M1, OR51I1, PDE4D, PDE6A, RGR, VIP, SPSB1, IRS2, PSCD1	16.1	1.6	0.009
CNS development	4	CNTN6, GRIK1, PBX1, PCP4	0.63	6.4	0.003
Excretion	3	VIP, NPHS2, KCNK5	0.22	13.6	0.001
Brain parenchymal volume
Glutamate signaling pathway	2	GRIN2A, HOMER2	0.07	28.6	0.002
T2 Lesion load
Calcium-mediated signaling	3	EGFR, PIP5K3, MCTP2	0.11	27.2	0.0002
G-protein signaling	3	DGKG, EDNRB, EGFR	0.29	10.3	0.003
Axon guidance	2	SLIT2, NRXN1	0.12	16.7	0.007
Hemopoiesis	3	JAG1, LRMP, BCL11A	0.44	6.8	0.009
Regulation of cell migration	2	JAG1, EGFR	0.13	15.4	0.008
Amino acid metabolism	5	EGFR, MSRA, SLC6A6, UBE1DC1, SLC7A5	1.06	4.7	0.004

DISCUSSION

We report the largest GWAS performed so far in a deeply phenotyped MS dataset. The recently reported IMSGC study screened 931 trios with 334 923 SNPs (11). Here we used a case–control design (n = 978 versus 883) with higher SNP density (551 642 SNPs). In contrast to the IMSGC study, which included a second replication stage of top susceptibility hits and candidate genes, the emphasis of this study was to isolate both, disease risk and modifiers. The present scan has identified a number of novel non-HLA candidate genes associated both with susceptibility and with clinical manifestations of the disease. As expected, each significant association has a very modest effect and reflects a small share of the genetic variance affecting MS risk and progression.

Among the most interesting association results is the glypican proteoglycan 5 (GPC5) gene. Glypicans are a class of heparan sulfate proteoglycans bound to the external surface of plasma membranes that play an important role in the control of cell division and growth regulation, but are also implicated in brain pattering, synapse formation, axon regeneration and guidance, and are found in dense networks in active MS plaques, where they may be involved in sequestering pro-inflammatory chemokines (38–42). GPC5 is expressed in neurons (43), and its interaction with growth factors, chemokines and extracellular matrix proteins may affect neuronal growth and repair (44). Recently, we reported GPC5 allelic differences associated with extreme responses to IFNβ in MS patients (26). It is conceivable that a single gene participates in both traits, conferring susceptibility and affecting the phenotype (in this case, a pharmacogenomic response). Alternatively, it is tantalizing to speculate that drug response differences in MS reflect, at least in part, singular etiologies characterized by different genetic backgrounds.

A second gene of interest emerging both from this study and the IMSGC GWAS is PARK2 or Parkin (6q25.2–27), a component of a multiprotein E3 ubiquitin ligase complex that mediates the targeting of substrate proteins for proteasomal degradation. A 4.5 kb transcript is expressed in many human tissues but is abundant in the brain. Mutations in this gene cause a loss of function which may compromise the poly-ubiquitination and proteasomal degradation of specific protein substrates, potentially leading to their deleterious accumulation in autosomal recessive juvenile-onset parkinsonism (37). Parkin mutations also affect mitochondrial function and apoptosis in neuronal cells, processes of importance in MS pathogenesis (45). Interestingly, variations within this gene were found to be associated with susceptibility to leprosy and other infectious diseases (46,47). An important caveat, however, is the very dense coverage of this gene in the Sentrix® HumanHap550 BeadChip (455 SNPs), increasing the possibility for potential type I errors. The signal in PARK2 did not survive the Šidák correction, suggesting the need for a conservative interpretation of its role in MS susceptibility.

The genes that we have associated with the MS phenotype are related to a broad range of cellular functions, but we did not find the expected high representation of genes mediating immunological effector functions. One exception is the association between NALp11 and the severity metric MSSS. NALPs are implicated in the activation of proinflammatory caspases via their involvement in multiprotein complexes called inflammasomes (48). It is striking that RELN (reelin), a gene also implicated in susceptibility to schizophrenia, autism, bipolar disorder, depression, and temporal lobe epilepsy (49–51), shows significant association with age of onset in MS (i.e. first clinical expression of disease). We speculate that the observed effect on age of onset is related to the role of reelin on neuronal survival (52) and layering of neurons in the cerebral cortex and cerebellum, affecting perhaps the threshold of neuronal plasticity required to avoid the clinical manifestation of the neurological damage (53). The associations between NLGN1 (neuroligin 1), HIP2 (huntingtin interacting protein 2) and CDH10 (cadherin 10) with T2 lesion load and brain atrophy, respectively, are consistent with roles for factors facilitating ‘homeostatic’ maintenance of brain function in disease.

Deconstructing the genetic events leading to MS will clarify its pathogenesis and improve opportunities to treat, prevent and cure this disease. This study increases the resolution of the MS genomic map by updating the roster of candidate disease susceptibility and progression loci. The GO summary analysis reflects the compartmentalization of the genetic involvement in the susceptibility versus the neurodegenerative phases of this disease.

MATERIALS AND METHODS

Study participants

The study cohort was ascertained through a prospective multi-center effort initiated in 2003. Three MS clinical centers were involved in patient enrollment and biological specimen collection using identical inclusion criteria, two in Europe (Vrije Universiteit Medical Center, Amsterdam; and University Hospital Basel) and one in the USA (University of California San Francisco). This study recruited patients of northern-European ancestry, preferentially with a relapsing onset of MS, although individuals with all clinical subtypes of the disease participated, including clinically isolated syndrome (CIS), relapsing remitting MS (RRMS), secondary progressive MS (SPMS), primary progressive MS (PPMS) and progressive relapsing MS (PRMS). CIS was defined as the first well-established neurological event lasting ≥48 h, involving optic nerve, spinal cord, brainstem or cerebellum. In CIS patients, the presence of two or more hyperintense lesions on a T2-weighted magnetic resonance imaging (MRI) sequence was also required for enrollment into the study. The diagnosis of RRMS was made utilizing the new International Panel criteria (54,55). SPMS was defined by 6 months of worsening neurological disability not explained by clinical relapse. PPMS was defined both by progressive clinical worsening for more than 12 months from symptom onset without any relapses, and abnormal cerebrospinal fluid as defined by the presence of ≥2 oligoclonal bands or an elevated IgG index. If acute relapses were superimposed on this steadily progressive course, patients were considered to have PRMS. Individuals were excluded if they had experienced a clinical relapse or received treatment with glucocorticosteroids within the previous month of enrollment. The concomitant use of disease modifying therapies for MS was permitted. For all subjects, the expanded disability status scale (56) and MSSS (57) scores were assessed, and brain MRI scans were performed within 2 weeks of entry into the study. In this study, age of onset was defined as the first episode of focal neurological dysfunction suggestive of CNS demyelinating disease (17). This information was obtained via individual recall and verified through review of medical records. The control group consisted of unrelated individuals, primarily spouses/partners, friends and other volunteers. Control subjects were of northern-European ancestry and matched as a group, proportionally with cases according to age (±5 years), gender. A familial history or current diagnosis of MS as well as a relation to another case or control subject were considered exclusionary for this group. Protocols were approved by the Committees on Human Research at all Institutions and informed consent was obtained from all participants prior to participation in the study.

Image acquisition

Brain MRI scans were performed on all cases upon entry into the study in 1.5 (Europe) and 3 (US) Tesla instruments using common sequences and protocols for data acquisition. MRIs were analyzed centrally at the project-MRI Evaluation Center (Basel) without the knowledge of disease subtype, duration or treatment history. Conventional spin echo, T1-weighted images were acquired for 5 min following administration of a single dose (0.1 mm/kg) of contrast agent (TE/TR = 8/467 ms, 256 × 256 × 44 matrix, 240 × 240 × 132 FOV, NEX = 1). Qualitative analysis for the presence of gadolinium enhancement was performed on post-contrast T1-weighted images. Brain lesions were identified and marked by consensus reading on simultaneously viewed T2 long and proton density-weighted images. Additional evaluations were done for new T2 lesions and volume of black holes, volume of T2 and T1 gadolinium enhanced lesions. The volumes were measured using an interactive digital analysis program (AMIRA).

The SIENAX algorithm (an adaptation of SIENA—Structural Image Evaluation using Normalization of Atrophy—for cross-sectional measurements) was used to estimate whole nBPV, normalized for subject head size (58,59). SIENAX extracts brain and skull images from the single structural acquisition. The brain image is registered to a standard space using the skull image to determine the registration scaling. Tissue segmentation, with partial volume estimation, is carried out to calculate total brain volume. SIENAX software are freely available as part of the FMRIB Software Library (www.fmrib.ox.ac.uk/fsl).

Genotyping, quality control and statistical analysis

Genotyping of the complete dataset was performed at the Illumina facilities using the Sentrix® HumanHap550 BeadChip. Sample success rate was 98.2% and the overall call rate was 99.85%. In addition to sample genotyping efficiency, a five-point quality control procedure was implemented before analysis, including assessment of marker allelic frequency, Hardy–Weinberg equilibrium, gender consistency, reproducibility and population genetic structure. Association analysis was carried out on 551 642 SNPs in 978 cases and 883 controls.

DRB1*1501 genotyping was performed using a validated gene-specific TaqMan assay designed to identify, specifically, the presence or absence of DRB1*1501 and/or *1503 alleles. An internal positive control (ß-globin) was included in each well to confirm that the reaction amplified successfully. PCR was carried out in a total volume of 10 µl, containing 20 ng DNA, 1x TaqMan Universal PCR Master Mix (Applied Biosystems), 0.6µM DRB1*1501/1503 specific primers (forward 5′;-ACG TTT CCT GTG GCA GCC TAA-3′, reverse 5′-TGC ACT GTG AAG CTC TCC ACA A-3′), 0.3 µm control primers (forward 5′-ACT GGG CAT GTG GAG ACA GAG AA-3′, reverse 5′-AGG TGA GCC AGG CCA TCA CTA AA-3′), 0.225 µm VIC-labeled DRB1*1501/1503 specific probe (5′-AAC AGC CAG AAG GAC ATC CTG GAG CA-3′) and 0.025 µm 6FAM-labeled control probe (5′-TCT ACC CTT GGA CCC AGA GGT TCT TTG AGT-3′). Amplification was carried out in an ABI PRISM 7900HT Sequence Detection System (Applied Biosystems) with an initial 95°C for 10 min, followed by 50 cycles of 95°C for 15 s and 62°C for 1 min. Results were confirmed with tagging SNP rs3135388 (60).

Two statistical methods were used to account for different models of genetic effect (1): the genotype-based case–control test for testing the dominant/additive/recessive allele effects on disease (61), and (2) a linear trend test for additive allelic effects on disease penetrance [Cochran–Armitage trend test (62)], with an asymptotic χ² distribution (1 d.f.). In order to address the problem of multiple testing, we implemented the false discovery rate (FDR) method (63). Logistic regression models were used to incorporate multiple covariates in the analysis of the transformed data, including gender, center of sample origin and DRB1*1501 status.

To correct for a potential bias towards detection of false-positive associations in large genes, the Šidák-corrected minimum P-value (64) for each gene, adjusted for LD, was obtained using the formula:

P_{min, corrected} = 1−(1−P_min)^M_eff, where P_min is the minimum P-value obtained from the SNPs tested within the gene under consideration, for association with a particular phenotype using a particular statistical and genetic model. M_eff is the effective number of SNPs tested within the gene, obtained by the method of Li and Ji (65), which is based on the eigenvalues of the correlation matrix (LD matrix) among the SNPs in the gene (66,67).

If the LD between every pair of SNPs tested within the gene is zero, then M_eff is equal to M, the actual number of SNPs tested.

For the genotype–phenotype correlation analysis, four different phenotypes were assessed: T2 lesion load, nBPV, MSSS and age of onset. Covariates included in the model were gender, center of sample origin, age onset, disease duration, treatment status and treatment duration. Transformation of the phenotypic data (cube-root transformation for T2 lesion, square-root transformation for MSSS and log transformation for age of onset) was performed prior to data analysis in order to meet the normality assumption for the regression model. We also performed analyses separately at each site, using the Fisher's method to combine the P-values over sites.

SUPPLEMENTARY MATERIAL

Supplementary Material is available at HMG Online.

FUNDING

Recruitment of study participants, clinical and MRI assessment and genome scan genotyping was funded by GlaxoSmithKline, of which several authors are employees.