A meta-analysis of multiple myeloma risk regions in African and European ancestry populations identifies putatively functional loci

Abstract

Background

Genome-wide association studies (GWAS) in European populations have identified genetic risk variants associated with multiple myeloma (MM).

Methods

We performed association testing of common variation in eight regions in 1,264 MM patients and 1,479 controls of European ancestry (EA) and 1,305 MM patients and 7,078 controls of African ancestry (AA) and conducted a meta-analysis to localize the signals, with epigenetic annotation used to predict functionality.

Results

We found that variants in 7p15.3, 17p11.2, 22q13.1 were statistically significantly (p<0.05) associated with MM risk in AAs and EAs and the variant in 3p22.1 was associated in EAs only. In a combined AA-EA meta-analysis, variation in five regions (2p23.3, 3p22.1, 7p15.3, 17p11.2, 22q13.1) was statistically signficantly associated with MM risk. In 3p22.1, the correlated variants clustered within the gene body of ULK4. Correlated variants in 7p15.3 clustered around an enhancer at the 3′ end of the CDCA7L transcription termination site. A missense variant at 17p11.2 (rs34562254, Pro251Leu, OR=1.32, p=2.93×10⁻⁷) in TNFRSF13B, encodes a lymphocyte-specific protein in the tumor necrosis factor receptor family that interacts with the NF-κB pathway. SNPs correlated with the index signal in 22q13.1 cluster around the promoter and enhancer regions of CBX7.

Conclusions

We found that reported MM susceptibility regions contain risk variants important across populations supporting the use of multiple racial/ethnic groups with different underlying genetic architecture to enhance the localization and identification of putatively functional alleles.

Impact

A subset of reported risk loci for multiple myeloma have consistent affects across populations and are likely to be functional.

INTRODUCTION

Multiple myeloma (MM), a neoplasm of malignant plasma cells arising in bone marrow, comprises 1.9% of all cancer deaths and 20% of all hematological cancer deaths (www.seer.ca.gov)(1). MM is uncommon, with an age-adjusted incidence rate of 7.9/100,000 in males and 5.1/100,000 in females in the U.S. in 2012 (www.seer.cancer.gov)(1). Clinical manifestations range from asymptomatic (smoldering) myeloma to active symptomatic disease (2). There is a 2- to 3- fold higher risk of disease in African Americans compared to individuals of European origin and a 2-fold increased risk in relatives of MM cases (3, 4), suggesting a heritable component to this disease.

A genome-wide association study (GWAS) of 1,675 cases and 5,903 controls from a Northern European population identified two genome-wide significant novel loci associated with MM risk at 3p22.1 (rs1052501) and 7p15.3 (rs4487645), as well as a suggestive association (p~10⁻⁷) at 2p23.3 (rs6746082) (5). In a second GWAS of 4,692 cases and 10,990 controls from the United Kingdom and Germany, four additional genome-wide significant risk loci were identified at 3q26.2 (rs10936599), 6p21.33 (rs2285803), 17p11.2 (rs4273077), and 22q13.1 (rs877529)(6). For these common risk variants, the per allele odds ratios (OR) and risk allele frequencies (RAF) ranged from 1.19 to 1.39, and 0.11 to 0.76, respectively. In a European study involving a large multiple myeloma consortium, three of these regions (2p23.3, 3p22.1 and 7p15.3) replicated at p<0.05 (7). In the most recent published GWAS, the 2q12.3 region was implicated in MM risk in a discovery set of 972 cases and 1,064 controls of European origin and was replicated in a similar set of 297 cases (8). This study also replicated six of the seven known regions for MM risk (8).

For common susceptibility alleles shared across populations, underlying genetic differences in linkage disequilibrium (LD) across racial/ethnic groups can be leveraged to more precisely localize markers of disease risk (9). In the present study, we examined MM susceptibility regions for individuals from North America of African (AA) and European (EA) ancestry and conducted GWAS plus imputation-based fine-mapping in an attempt to identify putative functional variants that better capture risk in these populations.

MATERIALS AND METHODS

Ethics Statement

All studies had approval from their respective Institutional Review Boards according to the Declaration of Helsinki Ethical Principles for Medical Research Involving Human Subjects in 1964. Signed informed consent was obtained from all participants at the time of blood/saliva collection. The participants in this study were recruited at multiple sites described below.

African Ancestry Study Participants

Study participants included 1,150 AA patients with MM enrolled in the phase 1 collection (through 11/11/2014) of the African American Mulitple Myeloma Study (AAMMS), from 11 clinical centers (Winship Cancer Institute and Grady Memorial Hospital at Emory University, MD Anderson Cancer Center at University of Texas, Robert H. Lurie Comprehensive Cancer Center at Northwestern University, Sidney Kimmel Comprehensive Cancer Center at Johns Hopkins University, Karmanos Cancer Institute at Wayne State University, University of Chicago Comprehensive Cancer Center, Siteman Cancer Center at Washington University, St. John Providence Health System, Norris Comprehensive Cancer Center at the University of Southern California (USC) and the Henry Ford Health System) and four National Cancer Institute (NCI) Surveillance, Epidemiology, and End Results (SEER) cancer registries (California, Detroit [excluding patients from Karmanos Cancer Center and Henry Ford Hospital], New Jersey, and Louisiana). USC is the data coordinating center that receives, processes and maintains all de-identified clinical and questionnaire data and biospecimens. English speaking AA patients diagnosed with active or smoldering MM at age 20 years or older were eligible for enrollment. Forty-three AA MM patients were included from the Multiethnic Cohort (MEC), a cohort of 215,251 men and women aged 45–75 years at recruitment from Hawaii and California (10). Incident cancer cases were identified through linkage with the Hawaii Tumor Registry and/or the Los Angeles County Cancer Surveillance Program; both NCI-funded SEER registries. An additional 28 AA MM patients from the University of California at San Francisco (UCSF) study were also included. That study enrolled 370 MM patients of all races treated for MM at UCSF between 1989 and 2010 (11). Additional details of the study, which also contributed patients to the EA GWAS meta-analysis, can be found in Supplementary Methods. Finally, 84 AA MM patients’ samples, collected from the Multiple Myeloma Research Consortium (MMRC) institutions and shipped to the MMRC Tissue Bank at the Mayo Clinic Scottsdale, were provided (2).

A comparison set of 7,078 multiple myeloma-free participants (4,447 males and 2,631 females) from the African Ancestry Prostate Cancer GWAS Consortium (AAPC, consisting of 13 independent studies) and from a breast cancer GWAS of AA women (AABC, consisting of nine independent studies) were used as controls (12, 13). Further details on the contributing studies are provided in the Supplementary Methods.

Genotyping and Imputation

DNA was extracted at the USC Genomics Core Laboratory from buffy coat or saliva samples from the 1,150 AAMMS and 43 MEC patients. For the 28 UCSF patients, DNA was extracted from white blood cells harvested after mobilization of stem cells with granulocyte colony-stimulating factor in preparation for autologous bone marrow transplant and shipped to USC for genotyping. For the 84 MMRF patients, DNA was extracted from ACK-lysed peripheral blood samples using a Puregene kit (Qiagen). All 1,305 samples were then genotyped using the Illumina HumanCore GWAS array at the USC Genomics Core Laboratory.

Controls were previously genotyped using the Illumina 1M-Duo (Illumina Inc., San Diego, California, USA). Quality control (QC) steps for the controls are described in detail elsewhere (12, 13). Among cases, 37,046 single nucleotide polymorphisms (SNP) and 11 samples with a call rate < 98% were removed. Cases were further excluded based on the following criteria: (i) unexpected replicates (n=14); (ii) first or second degree relatives (n=2); (iii) self-reported sex conflicting with sex estimated by X chromosome heterozygosity or XXY sex chromosome aneuploidy (n=6). A subset of controls (n=100) were genotyped on both arrays for QC purposes; any SNP that was discordant between the two platforms was removed (n=3,134). To minimize error due to platform differences, only SNPs genotyped in both cases and controls were included for imputation (n=188,835). Prior to merging the case and control genotype data, variant alleles were translated to the 1000 Genomes Project (1KGP) forward strand and base pair positions were mapped to GRCh37/hg19. Imputation to 1KGP (March 2012 release) was conducted for 500 Kb regions around the eight previously identified risk variants and SNPs with INFO>0.80 and minor allele frequency (MAF)>0.01 were included in the analysis. The number of genotyped and imputed SNPs by info score (<0.8 and >0.8) for each region is provided in Supplementary Table 1.

Statistical Analysis

Principal components (PC) were calculated with EIGENSTRAT v5.0 (14) using 19,070 common SNPs (MAF>0.05) with low pairwise LD (r²<0.20) selected from the 188,835 overlapping genotyped SNPs in cases and controls. Unconditional logistic regression was performed adjusting for age (at diagnosis for cases and at blood draw for controls), sex, and PC1-5, as these PCs captured the variability of the study sample (results were similar when adjusted for 10 PCs). The dosage effects of the risk allele assuming an additive genetic model were analyzed in a one degree-of-freedom likelihood ratio test implemented in SNPTEST v2.4.0 (15).

European Ancestry Study Participants

Study participants included 1,264 EA MM cases and 1,479 EA controls from four genotyping centers: USC, UCSF (11), Mayo Clinic (Mayo), and University of Utah (UU) (Supplementary Methods). The USC GWAS consisted of four case-control studies (Los Angeles SEER (16), Seattle/Detroit SEER (17), University of British Columbia, University of Alabama at Birmingham) and two cohort studies (the Multiethnic Cohort Study (MEC) (10) and the Melbourne Collaborative Cohort Study (18)). The Mayo Clinic study included cases and controls from Mayo Clinic and Washington University (19).

Genotyping and Imputation

Cases and controls were genotyped at each center and imputation was performed using IMPUTE2 (20) or Beagle (21) with 1KGP as the reference panel. A description of each of the EA studies, genotyping platforms and methods, as well as imputation and quality control procedures are provided in the Supplementary Methods.

Statistical Analysis

Each study analyzed their data separately using unconditional logistic regression, adjusting for age, sex, and PCs (Supplementary Methods) (14). Data for 500 Kb around each of the eight loci were extracted from each center. Summary statistics were meta-analyzed using a fixed effects model weighted by the inverse standard error in METAL (22).

Assigning Significance Levels

The goal of our statistical analysis was two-fold: 1) to enhance the localization of the regions found to be genome-wide significant in the previous studies in Europeans using combined AA-EA meta-analyses; and 2) to search for new associations in regions within −/+ 250 kb of these index SNPs. Accordingly, within each of the eight regions of interest, SNPs (both typed and imputed) were classified into two groups: Group A SNPs (r²≥0.50 with index estimated in 1KGP EUR populations) and Group B (r²<0.50). For Group A SNPs, we used region-wide significance as our type I error rate (alpha-level), but for Group B SNPs, we required a more stringent experiment-wide significance across all regions. We were less stringent in our choice of criteria for statistical significance for the Group A SNPs because of the prior knowledge of association of risk with the more strongly correlated Group A SNPs.

Alpha-levels for each region were separately derived for the two groups of SNPs using permutation testing. To achieve numerically stable results, 1,000 replicates randomly shuffling the case/control status of all samples while preserving the orginal case/control ratio were generated for Groups A and B SNPs within each region. For each replicate, we recorded the minimum p-value of all tested SNPs and regarded the 5^th percentile of the 1,000 minimum p-values as the permuation-based significance level for the Group A SNPs in that particular region. The minimum alpha-level for all Group A SNPs across the eight regions was 1.48×10⁻³. By contrast, the significance levels for Group B SNPs were found at the 0.625^th percentile (0.05/8 × 100%= 0.625), a Bonferroni correction accounting for a total of eight regions. The significance levels for both groups across the eight regions are presented in Supplementary Table 2.

Combined Analysis in AA and EA Individuals

Summary statistics from the AA analysis and EA meta-analysis were meta-analyzed using a fixed effects model weighted by the inverse standard error using METAL (22). Region-specific alpha-levels defined in the AA analysis were applied to the AA/EA combined meta-analysis, as they are the most conservative. All r² values presented in the results are calculated using European (EUR) and African (AFR) populations from 1KGP.

Genomic Annotation

In order to choose an efficient group of SNPs to move forward for functional annotation, we used the regions that replicated in AA population with the Group A criteria. We included SNPs that were correlated (r²≥0.50) with the most significant SNP in a 500 Kb region and within two orders of magnitude of the smallest p-value observed. To integrate chromatin biofeature annotations with our genotyping data in these regions, we used the R package FunciSNP (Bioconductor.org) (23). We selected publicly available datasets relevant to the development of the B-cell lineage, most closely representing MM pathogenesis. The following ENCODE datasets were employed to filter correlated SNPs that lie within putative enhancer regions with Gene Expression Omnibus (GEO) accession IDs: B-cells CD20+ RO01778 DGF Peaks (GSM1014525), B-cells CD20+ RO01778 DNase I HS Peaks (GSM1024765, GSM1024766), B-cells CD20+ RO01794 HS Peaks (GSM1008588), CD20+ (RO 01778) H3K4me3 Histone Mod ChIP-seq Peaks (GSM945229), CD20+ RO01794 H3K27ac Histone Mods by ChIP-seq Peaks (GSM1003459), CD20+ (RO 01794) H3K4me3 Histone Mod ChIP-seq Peaks (GSM945198), CD20+ CTCF Histone Mods by ChIP-seq Peaks (GSM1003474), CD20+ H2A.Z Histone Mods by ChIP-seq Peaks (GSM1003476), CD20+ H3K4me2 Histone Mods by ChIP-seq Peaks (GSM1003471). The combinations of these histone modifications were used to segment the genome in these ENCODE cell lines into active and poised promoter regions with or without DNase I hypersensitivity, active and poised enhancer regions with or without DNase I hypersensitivity, putative regulatory sites with open chromatin, and CTCF bound sites outside promoters and enhancers. SNPs that could be mapped to core regions (DNase hypersensitive sites) of putative non-coding regulatory regions (enhancers and promoters) were further subjected to analysis of transcription factor binding site disruptiveness using the R/Bioconductor package motifbreakR (24). To define other physical map features (transcription start sites, 5′ UTR, 3′UTR) we downloaded annotations from the February 2009 release of the human genome (GRCh37/hg19) available from the UCSC genome browser (25). Finally, we used the highly conserved set of predicted targets of microRNA targeting at mircode.org (miRcode 11, June 2012 release) (26), and conserved high-quality microRNA target species from microRNA.org (June 2010 release) (27).

RESULTS

Race-specific Replication of Known Risk Regions

Among AAs, we replicated three of the previously published risk variants at p<0.05 (7p15.3, p=8.30×10⁻⁵; 17p11.2, p=1.60×10⁻²; 22q13.1, p=1.47×10⁻², Table 1); four regions in total including 3p22.1. All previously reported risk variants were common among AAs (Table 1; Supplementary Figure 1). We had ≥90% power to detect the published effect size observed in AAs for six SNPs (rs4487645, rs4273077, and rs877529 were significant), and 73–80% power for the other two SNPs (Table 1). There were no statistically significant associations using Group B alpha-levels, although a marginally significant association was observed in the 6p21.33 region (rs190055148, p=1.37×10⁻⁶, r²=0.002 (1KGP EUR) and r²=0.06 (1KGP AFR) with the index marker rs2285803) (Supplementary Figure 2).

Table 1.

The association of genetic risk variants with multiple myeloma risk in persons of European and African ancestry, including previously reported index single nucleotide polymorphisms and the most statitistically significant risk variants from a combined meta-analysis of the two populations.

Index SNPsa/most significantly associated SNPsb									Association in European ancestry				Association in African ancestry				Combined Meta
SNP	BP	Risk/Ref	Freq	OR		P-value			Freq	OR	P-value	Power	Freq	OR	P-value	Power	OR	P-value	P-het	r2 w/indexc
2p23.3
rs6746082a	25659244	A/C	0.76	1.29		1.22×10−7			0.79	1.15	5.17×10−2	0.96	0.55	1.04	3.77×10−1	0.80	1.07	7.51×10−2	0.24
rs6761076b	25607758	T/C							0.81	1.23	7.23×10−3		0.68	1.09	8.33×10−2		1.14	3.19×10−3	0.22	0.05/0.51
2q12.3
rs12614346a	107642482	A/G	0.33	1.39		1.70×10−5			0.31	1.00	9.45×10−1	0.99	0.16	1.00	9.81×10−1	0.99	1.00	9.74×10−1	0.95
rs13416655b	107621925	C/T							0.50	1.01	8.03×10−1		0.39	1.10	4.90×10−2		1.06	9.48×10−2	0.29	0.13/0.52
3p22.1
rs1052501a	41925398	G/A	0.20	1.32		7.47×10−9			0.22	1.23	4.42×10−3	0.99	0.63	1.06	2.21×10−1	0.99	1.11	9.86×10−3	0.09
rs143531651b	41816589	G/C							0.17	1.25	4.91×10−3		0.11	1.27	1.37×10−3		1.26	2.02×10−5	0.91	0.02/0.79
3q26.2
rs10936599a	169492101	G/A	0.75	1.26		1.74×10−13			0.79	1.12	8.41×10−2	0.92	0.93	1.08	3.84×10−1	0.73	1.11	5.65×10−2	0.75
rs9811216b	169487501	T/C							0.74	1.11	1.10×10−1		0.70	1.09	8.46×10−2		1.10	1.91×10−2	0.84	0.16/0.94
6p21.33d
rs2285803a	31107258	A/G	0.28	1.19		1.18×10−10		0.29		1.11	1.27×10−1	0.84	0.26	1.06	2.21×10−1	0.95	− d
7p15.3
rs4487645a	21938240	C/A	0.65	1.38		3.33×10−15		0.70		1.23	7.47×10−4	0.99	0.89	1.37	8.30×10−5	0.99	1.28	4.00×10−7	0.28
rs12540021b	21945563	G/A						0.75		1.24	6.30×10−4		0.89	1.43	2.27×10−5		1.31	1.27×10−7	0.19	0.71/0.67
17p11.2
rs4273077a	16849139	G/A	0.11	1.26		1.41×10−7		0.12		1.37	2.46×10−4	0.83	0.14	1.17	1.60×10−2	0.97	1.24	3.66×10−5	0.14
rs34562254b	16842991	A/G						0.11		1.45	2.39×10−5		0.13	1.25	1.33×10−3		1.32	2.93×10−7	0.17	0.33/0.90
22q13.1
rs877529a	39542292	A/G	0.44		1.23		2.29×10−16	0.45		1.21	4.31×10−4	0.97	0.48	1.11	1.47×10−2	0.99	1.15	4.31×10−5	0.21
rs139425b	39559742	C/G						0.46		1.21	4.43×10−4		0.71	1.21	5.54×10−4		1.21	8.41×10−7	0.93	0.18/0.95

^aIndex SNP in each region, OR and p-values from the literature (5, 6, 8).

^bMost significant Group A SNP in each region from the combined AA and EA meta-analysis

^cr² from 1KGP (AFR/EUR reference)

^dCombined analyses were not performed in the HLA region

In EAs, we replicated four variants at p<0.05 (3p22.1, p=4.42×10⁻³; 7p15.3, p=7.47×10⁻⁴; 17p11.2, p=2.46×10⁻⁴; 22q13.1, p=4.31×10⁻⁴; Table 1). We had ≥90% power to detect the reported effect size for six SNPs (three of the six were significant at p<0.05) and 83–84% power for the other two SNPs (rs4273077 was significant) Table 1. No statistically significant Group B SNPs were observed. The previously reported locus 2q12.3 (8) was not associated with MM risk in either AA or EA subjects.

Race-specific results for all regions are provided in Supplementary Tables 3, 4 and Supplementary Figure 2.

Combined Analysis in AA and EA Individuals

In an attempt to better localize the region harboring a functional variant, summary statistics from the AA and EA studies were meta-analyzed for seven of the eight published risk regions. The HLA region on chromosome 6p21.33 was excluded from the meta-analysis because of extreme sensitivity to population stratification due to race-specific extended haplotypes and underlying LD patterns requiring greater SNP density than available here for interpretable results (28).

We found statistically significant associations for Group A SNPs that were in LD with the index SNP (r²≥0.50) in all regions except 2q12.3 and 3q26.2 (Table 1; Supplementary Figure 3); however there were no significant associations for Group B SNPs in any region. Five of the eight index SNPs and three of the most significant SNPs from the combined analysis were more common among individuals of African compared to those of European ancestry, with rs1052501 showing the largest difference (RAF in AAs 0.63, in EAs 0.22) (Table 1; Supplementary Figure 1). Below we describe the most significant associations and functional annotation in the four regions that replicated in the AA population with the Group A criteria.

3p22.1

Variant rs143531651 was the most significantly associated SNP (OR=1.26, p=2.02×10⁻⁵) and was correlated with the index SNP only in EA populations (AA RAF=0.11, r²=0.02, EA RAF=0.17, r²=0.79; Table 1). In this region, all the significant correlated variants cluster within the gene body of ULK4, which encodes the serine-threonine protein kinase. Among these are two missense variants of unknown significance, rs17215589 (OR=1.20, 1.04×10⁻³) and rs35263917 (OR=0.84, p=1.39×10⁻³). In addition, there are three SNPs, rs73830585 (OR=1.19, p=1.60×10⁻³), rs73071261 (OR=1.19, p=1.61×10⁻³) and rs55916855 (OR=0.83, p=7.35×10⁻⁴) located within DNase I hypersensitive sites in the active promoter of ULK4. Variants rs73830585 and rs55916855 disrupt EGR1 and INSM1 transcription factor binding sites, respectively (Figure 1, Supplementary Tables 5, 6).

Figure 1. Genomic annotation of the 3p22.1, 7p15.3, and 22q13.1 regions.

UCSC browser views showing wiggle tracks from ENCODE data for CD20+ B-cells from two cell lines, RO01778 and RO01794. The peak calls from these data were used to segment the genome into non-coding functional regions as detailed in the inset at bottom right. a) Region 3p22.1 detailing the 5′ end of the ULK4 gene, where high-confidence SNPs overlap the central regulatory core region of the active promoter. b) Overview of the 7p15.13; an enhancer with active histone marks within intron 79 of DNAH11 as described in the text. c) Overview of region 22q13.1 where several SNPs overlap with the promoter and downstream enhancers of CBX7. DHS: DNase I hypersensitive site, CTCF: CTCF bound region, ENH [active]: enhancer with H3K27 acetylation (K27Ac), ENH+DHS: DNase I hypersensitive region found within an active enhancer, PROM [active]: promoter with H3K27 acetylation, PROM+DHS: DNase I hypersensitive region found within an active promoter. Other abbreviations: DGF: digital DNase I footprinting, K4M1: H3K4 mono-methylation, K4M2: H3K4 dimethylation, H2A.Z: H2A.Z histone modification (not used for segmentations).

7p15.3

Variant rs12540021 (OR=1.31, p=1.27×10⁻⁷), located in intron 79 of DNAH11 and downstream of CDCA7L was the most significantly associated SNP in this region and was correlated with the index SNP in AAs and EAs (r²=0.71 and r²=0.67, respectively). The eight top correlated SNPs in this region are clustered around a solitary enhancer toward the 3′ end of the DNAH11 gene region, and 3′ of the CDCA7L transcription termination site. DNAH11 encodes for a ciliary outer dynein arm protein and CDCA7L encodes a cell-cycle gene that is expressed in malignant plasma cells (29). The index SNP in this region, rs4487645 (OR=1.28, p=4.00×10⁻⁷), is situated in the DNase I hypersensitive site in the center of the active enhancer, where transcription factors are most likely to be bound. The risk allele of rs4487645 (C), disrupts GATA1, GATA2 and GATA5 motifs. Thus, the correlated variants in 7p15.3 overlap putative regulatory features consistent with an active enhancer region (Figure 1; Supplementary Tables 5, 6).

17p11.2

rs34562254 (OR=1.32, p=2.93×10⁻⁷) was the most significantly associated SNP in this region in the combined analysis and in the race-specific analyses (Table 1; Supplementary Tables 3, 4). This variant occurs roughly equally in both populations (MAF_AA= 0.13; MAF_EA = 0.11) but is more highly correlated with the reported index SNP in EA (r²=0.90) compared to AA (r²=0.33, Table 1) individuals. This missense variant (Pro251Leu) is located in exon 5 of TNFRSF13B, a lymphocyte-specific tumor necrosis factor receptor that interacts with the NF-κB pathway and regulates B-cell development (30, 31). This variant is predicted to be possibly damaging in PolyPhen2 (32) with a score of 0.72 (sensitivity=0.86, specificity=0.92), while it is labeled as a tolerated mutation in SIFT (33). Variant rs34562254 is conserved across some species (the rhesus, dog, and elephant), but is not present in others (mouse or zebrafish).

22q13.1

Variant rs139425 (OR=1.21, p=8.41×10⁻⁷) was the most significantly associated SNP in this region and is strongly correlated with the reported index SNP in EAs but not AAs (r²=0.95 and r²=0.18, respectively). This SNP did not overlap any biofeatures of interest. The top 35 SNPs in this region cluster within 10kb in and around the promoter and proximal intronic enhancers of of the polycomb group gene CBX7, which are epigenetically marked active regions. CBX7 is a tumor supressor gene which is down-regulated in multiple cancers (34, 35). Seven correlated SNPs overlap with DNase I hypersensitive sites within the aformentioned promoter and enhancer regions (Supplementary Table 6): rs877529 and rs139398 are located within the downstream enhancers; rs877529 disrupts several high-confidence binding sites including ETS1, ETV4 and PAX6; rs1005300, rs6001455, rs5995688, rs12158877 and rs139405 are situated in the promoter region; and the reference allele of rs1005300 disrupts KLF1/KLF4 binding sites (Figure 1).

DISCUSSION

This is the first study to examine the eight published GWAS risk regions for MM in AA individuals. We statistically significantly replicated four of the EA reported regions in the AA-only analysis, suggesting that these risk regions are shared across populations. In an AA-EA meta-analysis, we identified SNPs in seven of the eight reported regions that were more significant than the index SNP; five were statistically significant using Group A criteria. The differential LD between AA and EA populations in these combined analyses allows for a finer resolution of the signal and suggests that these alternate SNPs may be better proxies of the functional alleles. The genomic annotation of these variants highlights potential functional impact within enhancer regions, promoter regions, and protein coding sequence for some of the variants.

We were able to utilize information from the differential LD in the two populations as well as the genomic annotation to identify the regions we believe to be the most promising for functional follow-up. Three regions have SNPs that are significantly associated with disease risk and functional annotation that is highly suggestive of regulatory function (3p22.1, 7p15.3, 22q13.1). Both the race-specific and combined analyses identified the missense variant rs34562254 (Pro251Leu) as the most significant SNP in the fourth region (17p11.2). This SNP is located in TNFRSF13B and falls centromeric to a common 17p deletion observed in MM cases (36). TNFRSF13B encodes a protein that is a lymphocyte-specific member of the tumor necrosis factor (TNF) receptor superfamily that interacts with the NF-κB pathway, critical for B-cell activation and survival and proliferation of MM neoplastic cells (37, 38), and the target of proteosome inhibitors used in standard MM therapy regimens (38).

In 7p15.3, we identified eight variants that were moved forward for functional annotation. A single SNP, rs4487645, was mapped to DNase I hypersensitive region in the core of a putative enhancer with active histone modifications. This SNP is predicted to disrupt three of a highly related family of transcription factor binding motifs with strong effects, including GATA1, GATA2, and GATA5 transcription factors (match threshold p<10⁻⁴) involved in T-cell and hematopoietic stem cell differentiation (Supplementary Table 5; Supplementary Methods). Weinhold et. al. recently generated expression quantitative trait loci (eQTL) data on malignant plasma cells in 848 MM patients and found that the strongest association was for rs4487645, which showed cis-regulation of CDCA7L (29). This same variant and its enhancer were annotated in our data as a potentially functional candidate in B-cells. Thus, our approach utilizing differential LD patterns to identify SNPs for functional annotation may identify truly functional disease correlates even when expression data are unavailable or lack sufficient statistical power.

This study includes the largest existing collection of AA MM cases and controls and is the first to examine previously reported risk regions in this diproportinately impacted group. One limitation is that AA cases and controls were genotyped on different arrays with only a small number of overlapping SNPs (n=188,835 SNPs genome-wide) which limited our ability to identify novel variants (Group B SNPs) and to examine the overlap in the HLA region. However, we performed rigorous QC on genotyped SNPs, which allowed us to impute cases and controls together, thereby providing more accurate imputed data. Nevertheless, there were not a large number of genotyped SNPs in each region which made imputation challenging. For example, in the 17p11.2 and 22q13.1 regions, over half of the imputed SNPs for the AA’s with a MAF>1% were excluded due to poor quality scores (INFO <0.8 in IMPUTE2, Supplementary Figure 2 Supplementary Table 1).

Another limitation of this study was the relatively small sample size of the race-specific analyses; however, power was greatly enhanced by combining the data across ancestry groups which leveraged the differential LD in these two populations in an attempt to more accurately approximate the true signal. For example, in the EA analysis, we had 28% power to detect an OR of 1.25 for an allele frequency of 10% while in the combined analysis, which more than doubled the number of cases and added more than 7000 additional controls, we had 89% power to detect this same effect size using the minimum alpha-level for Group A SNPs (1.48×10⁻³, Supplementary Table 2). Because MM is a rare disease (~6/100,000 average annual age-adjusted incidence rate) with a relatively poor 5-year survival rate (~46%), it is challenging to accrue large numbers of patients necessary for detecting associations with small to moderate magnitude of risk. Therefore, unlike similar studies of common solid tumor malignancies, it is often difficult to achieve adequate statistical power. However, we were able to improve power by including a a large number of controls from pre-existing GWAS in AA men and women.

Although we did not conduct a combined analysis of the HLA region due to its extreme sensitivity to population stratification and long-range LD, we did observe signals in this region for both AAs and EAs that differed by race, as expected. A possible independent signal (rs190055148, p=1.37×10⁻⁶, r²=0.06 with index in 1KGP AFR and r²=0.002 in EUR) was observed in AAs that will require confirmation in a larger sample.

In this study, we replicated associations in four of eight published risk regions in AAs and five in the AA-EA combined analysis, which suggests common shared functional variants across racial groups. We identified four regions that are promising for functional follow-up, including 17p11.2, where the most significant SNP in the combined analysis is a missense variant. Traditional large-scale discovery efforts in AA populations will be required to better understand the degree to which there is a genetic basis underlying the excess risk of MM in this group.