Common variation in KITLG and at 5q31.3 proximate to SPRY4 predispose to testicular germ cell cancer

Introductory paragraph

Evidence suggests that testicular germ cell tumors (TGCT) have a strong underlying genetic component. We performed a genome-wide scan among 277 TGCT cases and 919 controls. Seven markers at 12p22 within c-KIT ligand (KITLG) reached genome-wide significance (P < 5.0 × 10⁻⁸). In independent replication, TGCT risk was increased 3-fold per copy of the major allele at rs3782179 and rs4474514 (OR=3.08, 95% CI 2.29, 4.13; OR=3.07, 95% CI 2.29, 4.13, respectively). We also replicated associations with rs4324715 and rs6897876 at 5q31.3 near sprouty 4 (SPRY4; P < 5.0 × 10⁻⁶ in discovery). Risk of TGCT was increased nearly 40% per copy of the major allele (OR=1.37, 95% CI 1.14, 1.64; OR=1.39, 95% CI 1.16, 1.66, respectively). All of the genotypes were associated with both seminoma and non-seminoma TGCT subtypes. These results demonstrate that common genetic variants affect TGCT risk and implicate KITLG and SPRY4 as TGCT susceptibility genes.

In the United States, testicular germ cell tumors (TGCT) are the most common cancers in young men, with a peak incidence among those aged 25 to 34 years¹. The age-adjusted incidence in white men has doubled since 1975 and is now 6.6 per 100,000. The incidence in white non-Hispanic men is nearly five-fold higher than among black men¹. The reasons for the increasing incidence and racial disparity in TGCT rates are unknown.

While environmental exposures have been postulated to play a role in the increasing incidence of TGCT, there also is evidence for a substantial genetic contribution to TGCT susceptibility. Brothers of TGCT patients have an eight- to 12-fold increased risk of disease, with the risk to monozygotic and dizygotic twins 75- and 35-fold increased, respectively, and fathers of patients have a four-fold increased risk^2,3. Consistent with the high familial risks compared to most other cancer types and ethnic differences in TGCT risk, the proportion of TGCT susceptibility accounted for by genetic effects is estimated at 25%, and TGCT has the third highest heritability among all cancers⁴.

Results from linkage studies and candidate gene approaches, however, have produced limited insight into TGCT susceptibility factors. An initial report of linkage on Xq27 was not replicated nor have other loci been identified with significant effects, which suggests that multiple loci, potentially of weak to moderate effect, contribute to disease susceptibility^5,6. The gr/gr deletion on the Y chromosome, studied as a candidate region, increases TGCT risk two- to three-fold, but carriage frequency of this variant is low (2–3%) suggesting it likely accounts for only a small component of risk⁷. Thus, despite the multiple lines of evidence suggesting a genetic etiology of TGCT, no genetic risk factor has been identified that can explain an appreciable proportion of TGCT cases.

To identify genes associated with TGCT development, we performed a genome-wide association study. Cases were 277 white, non-Hispanic men with pathologically defined TGCT seen at the University of Pennsylvania Health System (UPHS) or Fox Chase Cancer Center (FCCC) in Philadelphia, PA. We genotyped DNA extracted from venous blood using the Affymetrix® Genome-Wide Human SNP Array 6.0. We compared the frequency of observed genotypes among TGCT cases to those available from 919 white, non-Hispanic males from the Philadelphia region genotyped on the same Affymetrix platform (Table 1). Supplemental Figure 1 shows the quantile-quantile plot of χ² values for observed versus expected allele frequencies based on Fisher's exact test for the 611,254 markers meeting quality control criteria, indicating little evidence of population stratification and evidence of excess disease associations⁸. The calculated genomic control inflation (λ) factor was 0.944, and hence we report unadjusted test statistics⁹.

Table 1.

Age, family history of TGCT, and tumor type in the discovery and replication samples

Status Total	Discovery			Replication
	Case n=277		Control n=919	Case n=371		Control n=860
	#	%	#	#	%	#	%
Age (median, [interquartile range])	31a [24,39]		57 [52,62]	34 [28, 38]		35 [30, 39]
Family history of TGCT
No	239	86.3	-	314	84.6	769	89.4
Yes	30b	10.8	-	11c	3.0	10c	1.2
Unknown	8	2.9	-	46	12.4	81	9.4
Personal history of cryptorchidism
No	245	88.5		330	88.9	844	98.1
Yes	26	9.4		38	10.2	16	1.9
Unknown	6	2.2		3	0.8	0	0
Tumor type
Seminoma	85	30.7	-	230	62.0	-	-
Non-seminoma	180	65.0	-	141	38.0	-	-
Unknown	12	4.3	-	0	0	-	-

^aAge of diagnosis missing for six TGCT cases.

^bSixteen cases were selected based on family history of TGCT; among non-selected (n=261) cases, the proportion reporting any family history of TGCT was 5.4%.

^cDenotes reported family history of TGCT among first degree relatives only.

Eight markers reached statistical significance at a genome-wide threshold of P < 5.0 × 10⁻⁸ (Fig. 1, Supplementary Table 1). Seven of these (rs995030, rs1352947, rs1472899, rs3782179, rs3782181, rs4474514, rs11104952) including the most significant association (P = 3.54 × 10⁻¹⁰) at rs4474514 occurred within the KITLG (c-KIT ligand) gene region on 12q22 (Supplementary Fig. 2). These markers were in strong linkage disequilibrium with each other; pairwise D' and r² measures were all > 0.99 (Fig. 2a). The eighth marker (rs3770112, P = 4.93 × 10⁻⁸) mapped to the integrin alpha 4 (ITGA4) gene on 2q31.3. Because no other markers in this genomic region (±10MB) reached statistical significance at P < 1.0 × 10⁻³, we suspected this association might have arisen purely by chance. We further investigated this possibility by imputing genotypes near rs3770112 based on publicly available HapMap genotypic data¹⁰. After imputation, the test of association at rs3770112 no longer surpassed the genome-wide threshold (P = 0.05); as well, all other markers in the region remained below the threshold for advancing to replication. The correlation between observed and imputed P values for the 23 markers that were in the same linkage disequilibrium block with rs3770112 was very high (r=0.96), and information content and maximum posterior call probability for rs3770112 were both > 0.998. Taken together, these results strongly suggested that the association observed in the discovery phase was a false positive (Supplementary Fig. 5). We selected two markers in KITLG (rs3782179, rs4474514) to bring forward into replication.

Figure 1. Genome wide association results plotted for 277 TGCT patients and 919 controls.

The threshold for genome wide significance was P < 5.0 × 10⁻⁸ (top red line) based on Fisher's exact test. Markers that reached significance at P < 5.0 × 10⁻⁶ (bottom green line) based on Fisher's exact test also were considered.

Figure 2. Regional association plots and linkage disequilibrium structure.

(a, b) The −log₁₀ of the P-value for the association of each discovery phase marker and TGCT status for segments of chromosomes (a) 12q22 [red line shows P < 5.0 × 10⁻] and (b) 5q31.3 [green line shows P < 5.0 χ 10⁻]. NCBI Build 36 was used for map locations. From each of these regions, two markers were taken into replication and are indicated in blue. Linkage disequilibrium structures for the GWAS data and based on r² are shown.

Sixteen additional markers reached statistical significance at the P < 5.0 × 10⁻⁶ level (Supplementary Table S1). Of these, three (rs12521013, rs4324715, rs6897876) mapped 2.4 kb downstream of the SPRY4 (sprouty homolog 4) coding region on 5q31.3 (Fig. 2b, Supplementary Fig. 3), and two (rs17031166, rs1549383) mapped to a gene free region on 2p14 that is 500kb centromeric of SPRED2 (sprouty-related, EVH1 domain containing 2) (Supplementary Fig. 4). As both SPRY4 and SPRED2 have been implicated in the KIT/KITLG signaling pathway^11,12, and as these two regions were the only ones that contained more than one marker surpassing threshold significance, we also chose two markers at each of these loci (SPRY4: rs4324715, rs6897876; 2p14: rs17031166, rs1549383) to bring forward for replication.

The replication set consisted of a population-based set of 371 TGCT cases and 860 controls, all white non-Hispanic, recruited from residents of the metropolitan Seattle-Puget Sound region, and parents of 204 of the cases. We observed associations with rs3782179 (P_trend= 5.88 × 10⁻¹⁵) and rs4474514 (P_trend= 5.88 × 10⁻¹⁵) in KITLG and with rs4324715 (P_trend= 6.77 × 10⁻⁴) and rs6897876 (P_trend= 3.67 × 10⁻⁴) proximal to SPRY4 (Table 2), but not with rs17031166 (P_trend= 0.90) or rs1549383 (P_trend= 0.88) near SPRED2. TGCT risk was increased three-fold per copy of the major A-allele in KITLG rs3782179 and rs4474514 (odds ratio (OR) = 3.08, 95% CI 2.29, 4.13; and OR=3.07, 95% CI 2.29, 4.13, respectively). Homozygous carriage of the major A-allele at these loci was associated with over a four-fold increased risk of TGCT (OR=4.56, 95% CI 1.78, 11.7; and OR=4.56, 95% CI 1.77, 11.7, respectively) compared with homozygous carriage of the minor G-allele. We noted weaker associations for the two markers close to SPRY4. TGCT risk was increased nearly 40% per copy of the major T-allele in rs4324715 (OR=1.37, 95% CI 1.14, 1.64) and major C-allele in rs6897876 (OR=1.39, 95% CI 1.16, 1.66); and risk was increased 65–80% with homozygous carriage of the major alleles (OR=1.81, 95% CI 1.26, 2.58; and OR=1.68, 95% CI 1.17, 2.42, respectively) compared with homozygous carriage of their corresponding minor alleles. In addition to the case-control analysis, we performed a case-parent triad analysis, which also showed that carriage of the risk allele for the markers in KITLG and proximal to SPRY4 are associated with TGCT. The per allele relative risks (RR) for rs3782179 and rs4474514 (KITLG) were 2.5 (95% CI 1.6, .9) and 2.6 (95% CI 1.6, 4.0), respectively, and for rs4324715 and rs6897876 (proximal to SPRY4) 1.5 (95% CI 1.2, 2.1) and 1.5 (95% CI 1.1, 2.0), respectively. These family-based estimates provide additional evidence that population stratification did not bias results in the replication phase.

Table 2.

Associations of TGCT with replicated SNP markers

Gene	Markera	Risk allele	Genotype Countb		Phase	Per allele	OR (95% CI) Heterozygotec	Homozygoted	P-trende
			Controls	Cases
KITLG	rs3782179 A/G	A	597/276/44	229/45/3	Discovery	2.36 (1.73, 3.21)	2.39 (0.71, 8.03)	5.63 (1.73, 18.3)	1.95 × 10−8
			515/285/38	309/49/5	Replication	3.08 (2.29, 4.13)	1.31 (0.49, 3.48)	4.56 (1.78, 11.7)	5.88 × 10−15
	rs4474514 A/G	A	599/276/44	229/45/2	Discovery	2.45(1.79,3.35)	3.59(0.84, 15.3)	8.41 (2.02, 35.0)	7.34 × 10−9
			517/285/38	310/49/5	Replication	3.07 (2.29, 4.13)	1.31 (0.49, 3.48)	4.56 (1.77, 11.7)	5.88 × 10−15
SPRY4	rs4324715 T/C	T	230/433/255	87/145/33	Discovery	1.59 (1.30, 1.93)	2.59 (1.72, 3.89)	2.92 (1.89, 4.53)	3.57 × 10−6
			191/437/197	119/171/68	Replication	1.37 (1.14, 1.64)	1.13 (0.82, 1.57)	1.81 (1.26, 2.58)	6.77 × 10−4
	rs6897876 C/T	C	282/429/207	114/135/28	Discovery	1.59 (1.31, 1.94)	2.33 (1.50, 3.61)	2.99 (1.90, 4.69)	2.96 × 10−6
			251/428/154	156/149/57	Replication	1.39 (1.16, 1.66)	0.94 (0.66, 1.34)	1.68 (1.17, 2.42)	3.67 × 10−4

^adbSNP rsnumber and major/minor alleles.

^bNumber of individuals genotyped as homozygous for the risk allele/heterozygous for the risk allele/homozygous for the non-risk allele. Nomenclature for major/minor alleles based on calls from Affymetrix® Genome-Wide Human SNP Array 6.0. MAF for discovery phase markers given in Supplemental Table 1.

^cOR for heterozygous carriage of risk allele compared to homozygous carriage of non-risk allele.

^dOR for homozygous carriage of risk allele compared to homozygous carriage of non-risk allele.

^eCochran-Armitage test for trend.

We did not observe an interaction between KITLG and SPRY4 marker genotypes. In the replication set, marker genotypes in KITLG and SPRY4 were associated with both seminoma and non-seminoma germ cell tumors without indication that genotype associations differed between the two subtypes (Table 3). In subgroup analyses among those without a family history of TGCT and among those without cryptorchidism, two strong and well-established risk factors for TGCT, the genotypic ORs associated with KITLG and SPRY4 markers were only negligibly attenuated (results not shown). These findings indicate that for the majority of cases, KITLG and SPRY4 do not exert their effect solely based on mechanisms involving these known risk factors. Because of limited numbers, it was not possible to examine the effect of KITLG and SPRY4 among those with positive family history or personal history of cryptorchidism.

Table 3.

Associations of KITLG and SPRY4 SNP markers with seminoma and non-seminoma TGCT^a

			Seminoma OR (95% CI)			Non-Seminoma OR (95% CI)
Gene	Marker	Risk allele	Per allele	Heterozygoteb	Homozygotec	Per allele	Heterozygoteb	Homozygotec
KITLG	rs4474514 A/G	A	2.97 (2.08, 4.23)	1.42 (0.42, 4.87)	4.70 (1.44, 15.4)	3.27 (2.06, 5.18)	1.13 (0.25, 5.1)	4.34 (1.03, 18.23)
KITLG	rs3782179 A/G	A	2.95 (2.07, 4.21)	1.42 (0.42, 4.87)	4.67 (1.43, 15.3)	3.30 (2.08, 5.24)	1.13 (0.25, 5.1)	4.39 (1.04, 18.45)
SPRY4	rs6897876 C/T	C	1.38 (1.11, 1.72)	1.11 (0.72, 1.73)	1.78 (1.14, 2.79)	1.39 (1.07, 1.81)	0.72 (0.43, 1.2)	1.55 (0.93, 2.56)
SPRY4	rs4324715 T/C	T	1.40 (1.13, 1.73)	1.39 (0.92, 2.09)	1.98 (1.27, 3.09)	1.32 (1.01, 1.71)	0.83 (0.52, 1.32)	1.60 (0.97, 2.62)

^aAnalyses reflect the replication set only

^bOR for heterozygous carriage of risk allele compared to homozygous carriage of non-risk allele.

^cOR for homozygous carriage of risk allele compared to homozygous carriage of non-risk allele

We have identified variation at 12q22 as a major risk locus for TGCT susceptibility. For rs3782179 and rs4474514, we observed a three-fold increased risk of disease per major allele and a 4.5-fold increased risk of disease for homozygous carriage of the major allele. The identified region contains KITLG, also known as stem cell factor, the ligand for the receptor tyrosine kinase, c-KIT. The KITLG/KIT signaling pathway plays an important role in gametogenesis, hematopoesis and melanogenesis¹³. In mouse models, Kitl (encoded at the steel [Sl] locus) is required for multiple aspects of primordial germ cell (PGC) development, including proliferation, migration and survival^14,15. Kitl plays a crucial role in the migration of PGCs from the hindgut and subsequent targeting to the genital ridges, and down regulation of Kitl in the midline triggers localized apoptosis of PGCs¹⁵. Based on similarity in cellular ultrastructure, patterns of imprinting, and gene expression, multiple human studies have suggested that TGCT arise from PGCs¹⁶. Delayed differentiation of PGCs has been associated with development of testicular germ cell carcinoma in situ among patients with intersex conditions and abnormalities of chromosomal number¹⁷. These data support a role for KITLG in TGCT susceptibility.

Furthermore, loss of the transmembrane form of Kitl, which leads to decreased PGC number, has been identified as a TGCT susceptibility locus in the 129/Sv mouse¹⁸. In humans, activating mutations of KIT are the most common somatic point mutations in TGCT, present in 25% of seminomas, although rarely identified in non-seminomas¹⁹. Thus, both germline variation and somatic mutations in the KITLG/KIT signaling pathway are associated with TGCT. In addition, KITLG/KIT signaling plays an important role in male fertility²⁰, and mutations in Kitl lead to decreased germ cell number. Our findings suggest that the reported epidemiological association between TGCT and male infertility²¹ may be due, in part, to a common genetic basis.

As KITLG plays a role in determining level of pigmentation²², we postulated that inherited variation at this locus could provide a genetic explanation for the observed differences in TGCT incidence in whites and blacks. KITLG has undergone strong positive selection in the European and East Asian populations, with an extended haplotype of 400kb²³. Data from HapMap Phase 3 show significant differences (P = 4.3 × 10⁻²⁰) in the frequency of the risk alleles of KITLG (rs3782179 and rs4474514) when comparing the CEU (major allele frequency = 0.80) and ASW (African ancestry in Southwest USA: major allele frequency = 0.25) populations¹⁰. This finding suggests that inherited variation in KITLG may explain, in part, the observed differences in TGCT incidence between whites and blacks.

We also observed an association between TGCT risk and variation at 5q31 just downstream of SPRY4. As with KITLG, the major allele was associated with increased risk. SPRY4 is one of a family of four genes (SPRY1–4) that have been implicated as negative regulators of the RAS-ERK-MAPK signaling pathway in response to growth factors²⁴. Expression analyses and tumor studies have shown that SPRY4 is the most significantly downregulated gene when KIT signaling is inhibited by imatinib mesylate in gastrointestinal stromal tumors, supporting a functional relationship between the two proteins¹².

In summary, our results demonstrate that common genetic variants at the 12p22 and 5q31 loci are associated with TGCT and strongly implicate KITLG as a susceptibility gene in the pathogenesis of TGCT. In addition, these observations may explain, in part, two important features of the disease: the increased incidence in whites and the epidemiological association with male infertility.

METHODS

Genome-wide association study

For the discovery phase, we initially selected 353 TGCT patients seen at UPHS (n=303) and FCCC (n=50); all cases were from the Philadelphia region. TGCT cases from UPHS were from an ongoing clinic-based case-control study of genetic susceptibility of TGCT for which study participants were asked to complete a self-administered questionnaire that elicited information on known and presumptive risk factors for TGCT. TGCT cases from FCCC were obtained from the Biosample Repository Facility, which collects and stores blood samples and obtains information on family history of cancer, risk factors and demographics from participating patients. We classified each TGCT patient according to the histological diagnosis of his tumor: seminoma or non-seminoma (including yolk sac, choriocarcinoma, embryonal, teratoma and mixed cell type TGCT) germ cell tumor. Only those with primary disease in the testis were included.

Male controls (n=932) were selected from PennCATH, a UPHS single-center, hospital-based study of angiographic coronary artery disease (CAD) in almost 4,000 subjects undergoing cardiac catheterization. This study investigates the association of biochemical and genetic factors for CAD and its risk factors²⁵; information on personal history of cancer was not collected. All controls were from the Philadelphia region, 90% were 46 years or older and had already passed the peak age of TGCT development. Based on available age-specific TGCT rates, we estimated that only four TGCT cases would be expected to have arisen in this control group¹. It is unlikely that this potential small misclassification of phenotype would have biased results appreciably.

Controls had been genotyped previously using the Affymetrix® Genome-Wide Human SNP Array 6.0 platform and had passed genotyping quality controls measures analogous to those used for TGCT cases (see below).

We used the Affymetrix® Genome-Wide Human SNP Array 6.0 to obtain genotypes for TGCT cases. We used the Birdseed algorithm to determine genotypes for the combined TGCT case and CAD control sample set²⁶. Among the 353 case samples, 18 subsequently were excluded for not meeting case eligibility (two Leydig cell tumors, one female germ cell tumor erroneously coded as TGCT, 15 non-TGCT samples) and 11 replicate samples with lower genotyping call rates were excluded. Of the 324 unique samples from TGCT cases, 19 (5.9%) were excluded because of a low (< 95%) genotyping call rate, eight (2.5%) because of lower than expected genotypic heterozygosity across called markers (F_ST ≥ 0.06), and 20 (6.2%) because of Asian or African ancestry as determined by multidimensional scaling (MDS)²⁷; no cases were excluded for cryptic relatedness (proportion of genotypes IBD for all cases was < 0.20). Among the 932 CAD controls, 13 were excluded because of female or ambiguous sex.

After excluding 224,705 (24.7%) markers with a minor allele frequency (in the total sample) < 0.05, 1,594 (0.2%) that deviated from Hardy-Weinberg equilibrium (HWE; P < 1 × 10⁻⁷), 71,978 (7.9%) with an individual genotype call rate < 0.95, and 233 (0.03%) invalid markers, 611,112 markers remained in the discovery phase.

To further investigate potential bias that could arise from our choice of control group, we compared the minor allele frequencies of markers brought into replication between those controls with verified coronary heart disease (n=700) and those without (n=219). We noted no statistically significant differences for the six markers, nor were there noted differences comparing these controls to population-based controls used in the replication phase (data not shown).

Replication study

To replicate findings of the discovery phase, we used 371 cases, 860 controls and 204 sets of mothers and fathers of cases from a population-based case-control study of TGCT in western Washington State. Methods for recruitment of TGCT cases and parents in this study previously have been published²⁸. Briefly, all cases had first, primary TGCT diagnosed between 1999 and 2007 and were residents of three urban counties of western Washington aged 18 to 44 years at diagnosis. Control subjects did not have a personal history of TGCT and were frequency-matched on age and ascertained from the general population of the three counties using random digit telephone dialing. Family history of TGCT among first-degree relatives and personal history of cryptorchidism was ascertained through self-administered questionnaires. Only cases and controls who self-identified as white, non-Hispanic were included in the replication study.

Genotyping was accomplished using pre-designed TaqMan SNP Genotyping Assays according to manufacturer's specifications. Genotyping was run in duplicate for 1034 marker pairs (an average of 172 sample pairs per each of the six markers in replication). In total, six (0.58%) calls were discordant; the Spearman correlation coefficient was > 0.99. Genotyping calls were made without knowledge of case or duplicate status. We also re-genotyped the majority (94–99%) of TGCT cases from the discovery phase for markers in Table 2. Concordance between genotype calls obtained from the Affymetrix® chip and TaqMan assays for these four makers was 100%.

For both the genome-wide scan and replication study, all participants provided written informed consent approved by their local Institutional Review Boards.

Statistical analysis

For the discovery phase, we used PLINK software to adjust for missing genotypes and calculate rates of heterozygosity²⁹. Population stratification was assessed using multidimensional scaling (MDS) methods and all markers were tested for HWE. PLINK also was used to determine genotypic associations among the 277 TGCT cases and 919 CAD controls. Statistical significance was assessed using Fisher's Exact test, and for top hits we determined ORs and 95% CIs for the per allele, heterozygous, and homozygous effects of the minor allele (Supplementary Table 1).

Imputation was conducted using a computationally efficient hidden Markov model based algorithm as implemented in software MACH³⁰. MACH combines our genotyped data with phased chromosomes from the HapMap CEU samples and then infers the unknown genotypes in the study sample probabilistically by searching for similar stretches of flanking haplotype in the HapMap CEU reference sample. We only analyzed markers that passed the following imputation QC criteria: R² > 0.3, and MAF > 0.05 in both cases and controls. To account for uncertainty involved in the imputation, we analyzed case-control associations for imputed SNP markers using software SNPTEST³¹.

For the replication phase, analyses were performed using SAS v9.1.3 (SAS Institute, Cary, NC). We used unconditional logistic regression to determine per allele associations and associations of homozygous and heterozygous carriage of risk alleles with case status (overall and among specified subgroups), and present unadjusted ORs because age was not a confounder in our data. We assessed trend across genotype categories by the Cochran-Armitage test for trend.

Models containing markers coded on an ordinal scale (additive model) and a cross-product term were made to test for marker-marker interaction. To estimate and compare the associations within TGCT subtypes, we used multinomial logit models to obtain simultaneously the OR and 95% CI for the association between markers and each level of outcome after adjusting for age.