Investigation of six testicular germ cell tumor susceptibility genes suggests a parent-of-origin

Transcription

1 Investigation of six testicular germ cell tumor susceptibility genes suggests a parent-of-origin effect in SPRY4 Robert Karlsson 1, {,, Kristine E. Andreassen 2,{, Wenche Kristiansen 3, Elin L. Aschim 3,RoyM. Bremnes 4,5, Olav Dahl 6,7, Sophie D. Fosså 8,9, Olbjørn Klepp 10, Carl W. Langberg 11, Arne Solberg 12, Steinar Tretli 2, Patrik K.E. Magnusson 1, Hans-Olov Adami 1,13, Trine B. Haugen 3, Tom Grotmol 2, { and Fredrik Wiklund 1, { 1 Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, Stockholm, Sweden, 2 Department of Research, Cancer Registry of Norway, Oslo, Norway, 3 Faculty of Health Sciences, Oslo and Akershus University College of Applied Sciences, Oslo, Norway, 4 Translational Cancer Research Group, Institute of Clinical Medicine, University of Tromsø, Tromsø, Norway, 5 Department of Oncology, University Hospital North Norway, Tromsø, Norway, 6 Section of Oncology, Institute of Medicine, University of Bergen, Bergen, Norway, 7 Department of Oncology, Haukeland University Hospital, Bergen, Norway, 8 Department of Clinical Cancer Research, The Norwegian Radium Hospital, Oslo University Hospital, Oslo, Norway, 9 Faculty of Medicine, University of Oslo, Oslo, Norway, 10 Department of Oncology, Ålesund Hospital, Helse Sunnmøre HF, Ålesund, Norway, 11 Cancer Centre, Ullevål University Hospital, Oslo, Norway, 12 Department of Oncology, St Olavs University Hospital, Trondheim, Norway and 13 Department of Epidemiology, Harvard School of Public Health, Boston, MA, USA Received October 21, 2012; Revised April 4, 2013; Accepted April 22, 2012 Human Molecular Genetics, 2013, Vol. 22, No doi: /hmg/ddt188 Advance Access published on May 2, 2013 Recent genome-wide association studies have identified single-nucleotide polymorphisms (SNPs) associated with testicular germ cell tumor (TGCT) risk in the genes ATF7IP, BAK1, DMRT1, KITLG, SPRY4 and TERT. In the present study, we validate these associations in a Scandinavian population, and explore effect modification by parental sex and differences in associations between the major histological subtypes seminoma and non-seminoma. A total of 118 SNPs in the six genes were genotyped in a population-based Swedish-Norwegian sample comprising 831 TGCT case parent triads, 474 dyads, 712 singletons and 3919 population controls. Seven hundred and thirty-four additional SNPs were imputed using reference haplotypes from the 1000 genomes project. SNP TGCT association was investigated using a likelihood-based association test for nuclear families and unrelated subjects implemented in the software package UNPHASED. Forward stepwise regression within each gene was applied to determine independent association signals. Effect modifications by parent-of-origin and effect differences between histological subtypes were explored. We observed strong association between SNPs in all six genes and TGCT (lowest P-value per gene: ATF7IP ; BAK ; DMRT ; KITLG ; SPRY ; TERT ). Stepwise regression indicated three independent signals for BAK1 and TERT, two for SPRY4 and one each for DMRT1, ATF7IP and KITLG. A significant parentof-origin effect was observed for rs in SPRY4 (maternal odds ratio , paternal odds ratio , interaction P ). No significant effect differences between seminomas and non-seminomas were found. In summary, we validated previously reported genetic associations with TGCT in a Scandinavian population, and observed suggestive evidence of a parent-of-origin effect in SPRY4. To whom correspondence should be addressed at: Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, Box 281, SE Stockholm, Sweden. Tel: ; Fax: ; robert.karlsson@ki.se The authors wish it to be known that, in their opinion, the first two authors should be regarded as joint First Authors. The authors wish it to be known that, in their opinion, the final two authors should be regarded as joint Last Authors. # The Author Published by Oxford University Press. All rights reserved. For Permissions, please journals.permissions@oup.com

2 3374 Human Molecular Genetics, 2013, Vol. 22, No. 16 Table 1. Characteristics of cases and controls Norway Sweden Total Included in final analysis TGCT cases Full triads Dyads (mothers/fathers) 192 (150/42) 248 (178/70) 440 (328/112) 474 (340/134) Singletons Histological subtype Seminoma Non-seminoma Age at diagnosis (range, mean) (33) (32) (32) (32) Unrelated controls (TwinGene) INTRODUCTION Although testicular germ cell tumor (TGCT) is a relatively rare form of cancer in the general population, it is the most common malignancy in young men of Northern and Western European ancestry (1,2). The origin of TGCT is believed to be carcinoma in situ cells, whose malignant transformation is initiated during the early development from primordial germ cells or gonocytes (3). Although the etiology of TGCT is largely unknown, genetic components and conditions during pregnancy are known to play arole(3). Brothers and sons of patients with TGCT are at increased risk of contracting the disease themselves (4), and family studies have shown a sizable genetic component of disease risk (5). Thus, susceptibility genes for TGCT are likely to exist. Confirming the existence of such genes, genome-wide association studies (GWAS) have identified several single-nucleotide polymorphisms (SNPs) in the genes activating transcription factor 7 interacting protein (ATF7IP), BCL-2 antagonist/killer 1 (BAK1), doublesex and mab-3 related transcription factor 1 (DMRT1), KIT ligand (KITLG), sprouty homolog 4 (SPRY4) and telomerase reverse transcriptase (TERT) as associated with the risk of developing TGCT (6 8). The aim of the present study was to replicate and further characterize the associations in six TGCT susceptibility genes reported in previous GWAS. Common genetic markers within the six genes were explored in a large population-based case parent and case control TGCT sample ascertained in Sweden and Norway. Overall association between markers and TGCT risk was assessed, and analyses stratified by histological subtype (seminoma or non-seminoma) were performed. Finally, taking advantage of our family-based design, we studied parent-of-origin effects. RESULTS After sample and genotype quality control filtering, a total of 831 complete case parent triads, 474 case parent dyads, 712 singleton cases and 3919 unrelated male controls remained for analysis (Table 1). For these individuals, genotypes for 852 SNPs distributed across the six susceptibility genes were available for assessment of association to TGCT risk. A total of 118 of these SNPs were directly genotyped, while the remaining 734 were imputed using the 1000 genomes reference haplotypes. The association between each SNP and TGCT risk was tested in the combined case parent, case control sample. All six genes investigated had one or more markers significantly associated with TGCT after false discovery rate (FDR) correction for multiple testing (FDR Q-value,, 0.01). Per gene, the lowest P-values obtained were in TERT (rs ), in SPRY4 (rs ), in BAK1 (rs ), in DMRT1 (rs ), in ATF7IP (rs ) and in KITLG (rs ). Figure 1 shows the complete association test results, along with a genomic feature display and local recombination rate estimates from the 1000 genomes project. Association results from the complete set of genotyped and imputed markers are presented in Supplementary Material, Table S1. A forward stepwise regression procedure was performed for directly genotyped markers in the combined case parent, case control sample to assess the number of independent association signals in each gene. With a nominal P-value inclusion threshold of P, at each iteration, three markers were included for TERT (rs , rs , rs ), two markers for SPRY4 (rs , rs ), three markers for BAK1 (rs513349, rs , rs ), one marker for DMRT1 (rs755383), one marker for ATF7IP (rs ) and one marker for KITLG (rs ). Single-marker (marginal) association results for these SNPs are presented in Table 2. Furthermore, Supplementary Material, Figs S2 S12 display regional pairwise linkage disequilibrium (LD) and (marginal) association plots for each of these selected SNPs, prepared using LocusZoom (9). DMRT1 was previously reported to contain two independent association signals, tagged by markers rs and rs (10). Pairwise LD between the two markers was estimated to R 2 ¼ 0.32 (D ¼ 0.80) in our study population. Adjusting for rs755383, rs entailed an association P-value of , an association that is arguably noteworthy, but did not pass the predefined inclusion criteria of our stepwise regression procedure. Effect modification by the sex of transmitting parent (parent-of-origin effect) and effect differences between seminomas and non-seminomas were assessed for the independently associated markers within each gene by applying a Wald test of marker covariate interaction terms in the regression model. Parent-of-origin effects were tested using only case parent triads and dyads. None of the markers investigated showed a significantly different association to one histological subtype compared with the other after multiple testing correction. However, the sex of the parent transmitting an allele was a significant effect modifier for marker rs in the gene SPRY4. The G allele of rs had an estimated allelic odds ratio of 1.72 (95% CI ) when maternally inherited, and

3 Human Molecular Genetics, 2013, Vol. 22, No Figure 1. Manhattan plot of 2log 10 (P)-values from combined case parent, case control association analysis. Markers plotted in light gray were imputed, dark gray markers were directly genotyped. Star-shaped markers were selected in gene-wise forward stepwise regression. Symbols below the scatterplot show the extents of transcribed regions and exons (wider line) of genes. The overlaid gray line shows recombination rate estimates from the 1000 genomes reference panel in centi- Morgan/mega base pair units. Genomic coordinates (horizontal axis) refer to the hg19 build of the reference human genome. Table 2. Single-marker association results for markers selected by gene-wise forward stepwise regression with an inclusion threshold of P, Gene CHR SNP BP A1 A2 MAF OR (95% CI) P-value Q-value Case parent OR (95% CI) TERT 5 rs C T ( ) ( ) TERT 5 rs A C ( ) ( ) TERT 5 rs A G ( ) ( ) SPRY4 5 rs A G ( ) ( ) SPRY4 5 rs A G ( ) ( ) BAK1 6 rs A G ( ) ( ) BAK1 6 rs C T ( ) ( ) BAK1 6 rs G T ( ) ( ) DMRT1 9 rs C T ( ) ( ) ATF7IP 12 rs A G ( ) ( ) KITLG 12 rs A G ( ) ( ) Stepwise regression was restricted to directly genotyped markers. CHR, chromosome; BP, base pair position (hg19 build of the reference human genome); A1, reference allele (allelic odds ratio ¼ 1); A2, non-reference allele; MAF, minor allele frequency in founders; OR, odds ratio; CI, confidence interval; Q, false discovery rate Q-value (95% CI ) when paternally inherited. The effect modification was statistically significant after Bonferroni correction for 22 tests (11 markers selected by stepwise regression tested for two potential interactions; unadjusted P ¼ ). Bonferroni correction was applied since markers were selected to be independent. Furthermore, the effect remained significant (P ¼ 0.006) when applying the parent-of-origin likelihood ratio test (11), which tests for parent-of-origin effects adjusted for maternal carriage effects. Out of the 833 complete triads, 437 matings were informative for marker rs (233 heterozygous fathers and homozygous mothers, 204 heterozygous mothers and homozygous fathers). To further explore the pattern of modification by parental sex, all directly genotyped and imputed markers in the SPRY4 region were tested for parent-of-origin effects. Several neighboring markers to rs near the promotor region of SPRY4 showed evidence of parent-of-origin effects, while little evidence of effect modification by parental sex was observed in coding regions, introns or the 3 UTR region of SPRY4 (Fig. 2). Finally, we estimated the parent-of-origin effect in

4 3376 Human Molecular Genetics, 2013, Vol. 22, No. 16 Figure2. Manhattan plotof 2log 10 (P)-valuesfrom per-marker tests for parent-of-origineffects in the SPRY4region. Markersplottedin light gray wereimputed,black markers were directly genotyped. Labeled SNPs were among the markers first selected by forward stepwise regression, and thus first tested for parent-of-origin effects. The SPRY4 gene symbol below scatterplot indicates extent of the transcribed region and exons (wider line) of SPRY4. Genomic coordinates (horizontal axis) refer to the hg19 build of the reference human genome. rs in the two distinct subsamples of seminoma and non-seminoma patients separately, in order to evaluate whether the effect was limited to one of the two major histological subtypes. Our power to detect such a difference was limited due to the roughly halved sample sizes, but point estimates of the effect were similar between the subtypes and 95% confidence intervals were overlapping [in seminoma: maternal OR ¼ 1.66 (95% CI ), paternal OR ¼ 1.19 (95% CI ); in non-seminoma: maternal OR ¼ 1.85 (95% CI ), paternal OR ¼ 0.91 (95% CI )]. DISCUSSION The etiology of TGCT is largely unknown, although genetic components and conditions during pregnancy are known to play a role (3). Recent GWAS in the UK and USA have shown that polymorphisms in genes regulating primordial germ cell development, telomerase activity and sex-determination are involved (6 8). In the present study, we have replicated evidence of association between genetic markers in the genes ATF7IP, BAK1, DMRT1, KITLG, SPRY4 and TERT and TGCT risk in a large population-based Swedish-Norwegian study. All loci contained markers significantly associated with TGCT risk in our combined case parent, case control sample and stepwise regression indicated more than one independent association in SPRY4, TERT and BAK1 in the investigated population. The three independently significant markers in BAK1 were, however, physically very close (inter-marker distances 804 and 138 base pairs), and their selection by the stepwise regression procedure may be a numerical artifact due to collinearity rather than truly independent signals. Two independent association signals have previously been reported in the DMRT1 gene, represented by markers rs and rs (10). We were only able to identify one independent signal (rs755383) in our study. However, LD between these two markers was quite strong in our population (D ¼ 0.80), which may explain the observed lack of independent association between rs and TGCT. Further assessment of this region is warranted to firmly establish the number of independent association signals. Assessment of effect modification of transmitted alleles by parental sex was performed in the associated markers selected by the stepwise regression procedure. A significant parent-oforigin effect was found for the marker rs in SPRY4, for which maternally transmitted risk alleles were associated with an allelic odds ratio of 1.7, while paternally transmitted alleles had no apparent effect on disease risk. Assessment of all available markers in the SPRY4 region for parent-of-origin effects indicated that the effect may be confined to variation near its promoter region (Fig. 2). The parent-of-origin effect remained significant in a test adjusting for potential maternal effects. The effect modification of rs by parental sex is a novel aspect of the risk of TGCT associated with SPRY4 genetic variation, and further studies in independent populations are warranted to validate this finding. Sprouty proteins are potent receptor tyrosine kinase inhibitors that antagonize growth factor signaling via the mitogenactivated protein kinase (MAPK) cascade (12). SPRY4 (encoded by SPRY4) acts as a tumor suppressor implicated in the development of several cancers such as in the lung (13) and prostate (14), in addition to its role as a risk factor for TGCT, demonstrated in several genetic association studies (6,8,15,16). The most well-known parent-of-origin effect is imprinting, an epigenetic phenomenon in which the expression of a gene is modified dependent on the sex of the transmitting parent (17). The simplest interpretation of the parent-of-origin effect observed in the present study is that the SPRY4 gene is silenced, by imprinting or some other epigenetic mechanism, in the fathers of TGCT cases. Variation in the maternally inherited allele associated with SPRY4 function or SPRY4 expression levels would then unilaterally affect the child s disease risk. According to Geneimprint ( (18), an online database of imprinted genes in human and selected model organisms, 60 imprinted genes have been described in human. SPRY4 is not among them (webpage accessed 10 September 2012), but because the number of imprinted genes described so far in the mouse is higher than in human, the extent of human imprinting is probably underestimated (17). Furthermore, tissue-specificity of imprinting also likely accounts for the underestimation of imprinted genes because studies assessing several tissues and developmental stages are lacking (19). The Encyclopedia of DNA Elements (ENCODE) consortium recently published a series of papers on functional elements in

5 Human Molecular Genetics, 2013, Vol. 22, No the human genome, including DNA methylation status of CpG islands, which constitute targets for gene regulation by methylation (20). Visualization of ENCODE data is possible via the UCSC Genome Browser (21). Using this resource, we assessed the presence of CpG islands and methylation status in several ENCODE cell lines in the SPRY4 region (Supplementary Material, Fig. S1). Interestingly, the 5 promotor region of SPRY4, in which we observed a parent-of-origin effect, contains a CpG island, and methylation of the region was observed in multiple cell types in the ENCODE database. Although evidence for methylation in the region is a necessary criterion for imprinting, it is not sufficient evidence, and further functional studies are warranted to elucidate the biological mechanism behind our finding. The rapid increase in TGCT incidence suggests an important role of environmental factors in the development of this cancer form. In rats, exposure to the anti-androgenic fungicide vinclozolin was shown to induce an epigenetic transgenerational phenotype, transmitted through imprinting in the male germ line, and involving adverse reproductive effects (e.g. male infertility) (22). Assuming such effects exist also in humans, environmental factors could then induce new paternally imprinted loci, with properties similar to the parent-of-origin effect found in this study for the marker rs in SPRY4. Recent work has shown how SNP variants within imprinted loci also associate with parent-of-origin susceptibility to cancer, such as for basal cell carcinoma and breast cancer (23). Although imprinting has been implicated in the etiology of TGCT (24,25), parent-of-origin effects have to our knowledge not been reported before in the development of this cancer form. Therefore, future studies of TGCT genetic susceptibility, such as GWAS, would benefit from assessing effect modification by parental sex if feasible. It is possible that part of the so-called missing heritability of this and other complex traits is in fact obscured by parent-of-origin effects not being properly taken into account (17). Assessment of effect differences for transmitted alleles by tumor histological subtype was performed in a similar manner as for parental sex. No significant effect differences were found between seminomas and non-seminomas in the associated markers selected by stepwise regression. Our results are in harmony with one of the previous GWAS (8), where marker genotypes in KITLG and SPRY4 were associated with both seminoma and non-seminoma TGCT without indication that genotype associations differed between the two subtypes. The remaining genes investigated in the present study do not seem to have been subject to previous assessment with respect to histological subtypes. Although there are differences in tumor biology between seminoma and non-seminoma TGCT, most often null or inconsistent results are reported in genetic association studies on histological subtype of TGCT (26 29). Major strengths of our study include the population-based design, the large sample size and the sizable case parent component. The case parent subsample enabled us to assess parent-of-origin effects, and allowed us to employ family-based tests for association, which are robust against potential bias from population stratification (30). Weaknesses include possible survival bias because cases were retrospectively ascertained. However, because 97% of cases eligible for inclusion in this study were alive at the time of sample collection, we argue that this bias is of minor importance. Another potential weakness is that all controls are of Swedish origin. However, we expect the Swedish and Norwegian populations to be genetically similar, and a comparison of the allelic effect estimates in the combined case parent, case control sample to those from using only the case parent portion revealed no major differences in TGCT risk effect estimates (Table 2). Since the case parent estimates are robust to population stratification issues, we conclude that the results of the present study are not greatly affected by population stratification bias. Although not critical for the results reported herein, we note that among the case parent dyads, the number of available mother son dyads is more than double the number of father son dyads. Effect modification by parental sex and effect differences between histological subtypes were only tested for the one marker per independent association region most strongly associated with disease risk. The rationale for this restriction was conservation of power by decreasing the number of tests performed, and the assumption that markers with a large marginal association are more likely to harbor a detectable effect modification. A limitation of this approach is that alleles associated with opposite effects depending on if they are maternally or paternally inherited would not be identified. In conclusion, we have validated previous TGCT GWAS findings in a Scandinavian sample, and observed a parent-of-origin effect for the marker rs in SPRY4. Occurrence of a CpG island in this region and reported observations of regionspecific methylation in several cell lines support this finding; however, additional association studies in independent populations or molecular evidence for imprinting in testicular germ cell tissue are necessary to firmly establish the existence of this parent-of-origin effect. TGCT GWAS to date have been fairly small compared with those seen for other diseases and traits, probably due to the relative rarity of the disease. They have nevertheless been successful in finding risk loci, partly due to the uncommonly large effect sizes for being a complex trait (ORs up to 2.5 for KITLG). Most likely there are more loci of smaller effect waiting to be found, and pooling of existing GWAS data from different research groups is warranted. MATERIALS AND METHODS Ascertainment of case and control samples Ascertainment of the case parent, case control sample used in this study has been described in detail previously (28). In brief, TGCT cases and their parents were recruited in Sweden and Norway using nationwide cancer registries. All men with a TGCT diagnosis (ICD-10, C62) between 1995 and 2006 in Sweden and between 1990 and 2008 in Norway were invited to participate in the study. At the time of enrollment, 3% of the men in the eligible study population were deceased. In total, 2141 (52%) of the invited men consented to participate in the study. Men consenting to the study were slightly older at the time of diagnosis compared with non-responders (32.4 versus 31.1 years). There was no difference in tumor stage between responders and non-responders (22.4 versus 22.3% with

6 3378 Human Molecular Genetics, 2013, Vol. 22, No. 16 metastatic disease). A total of 1171 of the included tumors were seminomas and 970 non-seminomas. After approval from participating TGCT cases, biological parents were also invited to the study. All participating individuals donated a saliva sample from which DNA was extracted. In total, 1004 complete case parent triads, 440 case parent dyads and 718 case singletons were included in the study. Average age at diagnosis among TGCT cases was 32 years (range: years). Additionally, 3922 unrelated male twins participating in the Swedish Twin- Gene project (31) were included as population controls. Study population composition before and after quality control is described in detail in Table 1. The study was approved by the Regional Committee for Medical Research Ethics in Southern Norway, the Norwegian Social Science Data Services and the Regional Research Ethics Committee in Stockholm, Sweden. The dedicated Research Biobank in Oslo was approved by the Ministry of Health and Care Services. DNA extraction Saliva samples were collected using the Oragenew DNA sample collection kit (DNA Genotek Inc., Kanata, ON, Canada). Extraction of DNA was performed according to the manufacturer s protocol using ethanol precipitation. Nano- Dropw ND-1000 spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA) was used to assess DNA yield. Samples from 79 Swedish and 35 Norwegian cases had low DNA yield (,25 mg) and were excluded from further analyses. Selection and genotyping of SNPs Six genes were selected for analysis based on previous GWAS findings: ATF7IP, BAK1, DMRT1, KITLG, SPRY4 and TERT. To select SNPs capturing common genetic variation (minor allele frequency above 5%) across these genes, an aggressive tagging approach using the Tagger software (32) with a minimal coefficient of determination equal to 0.95 was applied. SNP selection was performed based on the catalog of common genetic variants generated from the International HapMap Project (33) using phase two genotypes from the Centre d Etude du Polymorphisme Humain individuals. From this procedure, 118 SNPs were selected for genotyping. Genotyping was performed on a MassARRAY iplex system (Sequenom, Inc., San Diego, CA, USA) at the Centre for Integrative Genetics, Norwegian University of Life Sciences, Ås, Norway. Quality control of SNPs and samples After genotyping, a set of quality control thresholds were applied to the data. We first excluded samples with more than 20% of genotypes missing or a heterozygosity rate.3 SD from the sample mean (indicating possible sample contamination) (n ¼ 277). Parent offspring relations were validated using sample-wide pairwise measures of genotype identity-by-state (IBS). Based on visual inspection of a plot of the mean number of alleles shared IBS versus the standard deviation of the number of alleles shared IBS, a threshold of SD(IBS) was set to indicate possible pedigree errors. Removing the parents in case parent pairs exceeding this threshold (n ¼ 41) resolved 446 of 534 observed Mendelian errors in the data. The 88 remaining Mendelian errors were interpreted as mis-called genotypes and set to missing. Sample-wide IBS measures also uncovered 17 pairs of unplanned duplicate samples. Both samples in each pair were excluded since their identities could not be unambiguously determined (n ¼ 34). In case parent triads or dyads where the proband failed sample quality control, the parents were removed as well (n ¼ 111). Fifteen SNPs failed genotyping completely or were reported to have problematic clustering results from the genotyping facility, and were not included in further analyses. We further excluded SNPs with no call for.10% of the sample (n ¼ 8) or a minor allele frequency,1% in founders (n ¼ 0). After quality control, 831 triads, 474 dyads, 712 case singletons and 3919 control subjects remained for analysis (Table 1), with data available from 98 SNPs. Imputation To increase resolution, markers were imputed in the investigated regions using BEAGLE software version (34,35) and the full set of reference haplotypes from the 1000 genomes project June 2011 data release (1094 individuals) (36). All samples were pre-phased using BEAGLE before imputation. Family information was taken into account during phasing for triads and dyads, for improved phasing accuracy. Imputation was then performed on the phased haplotypes, imputing triads, dyads, singleton cases and TwinGene controls in separate batches. After imputation, marker and genotype quality control of imputed markers was performed. The minimum of the per marker imputation R 2 across all batches was calculated, and all markers with minimum R 2, 0.8 were dropped. Imputed genotype probabilities were converted to genotypes using the BEAGLE utility beagle2linkage. Genotype probabilities lower than 0.9 were called as missing. The PLINK (37) tool set was then used for further quality control of imputed markers. Imputed markers with any of the following characteristics were dropped from further analyses: missingness 5%, Hardy Weinberg exact test P-value 10 26, differential missingness between TwinGene and all other samples combined P-value or differential missingness between TGCT cases and controls P-value Single-marker association tests Marker TGCT association analysis was performed using a likelihood-based association test for nuclear families and unrelated subjects with missing data, implemented in the software package UNPHASED (30). For the primary analysis, we performed an unmatched test, comparing haplotypes transmitted to cases to non-transmitted and control haplotypes (-hhrr option in UNPHASED). In addition to the unmatched test, we tested for association in the case parent subsample, using only family-based control haplotypes. This restricted test is robust to confounding by population stratification. In both cases, allelic main effects were assumed, leading to a 1-degree of freedom likelihood ratio test of association.

7 Human Molecular Genetics, 2013, Vol. 22, No Adjustment for multiple testing was performed using an FDR approach. FDR is less conservative than the commonly applied Bonferroni correction, and especially suitable when tests are correlated. We estimated FDR using a semi-parametric approach based on a modified Grenander density estimator and truncated maximum-likelihood estimation (38), implemented in the R package fdrtool (39). We considered SNPs with FDR Q-values, 0.01 to be significantly associated with TGCT. Potential effect differences of alleles between tumor histological subtypes and depending on the sex of the transmitting parent (parent-of-origin effects) were tested by including an interaction term between the marker and covariate in the regression model, and performing Wald tests of the null hypothesis of no interaction. The parent-of-origin effect likelihood ratio test (11) was applied in the complete triads for markers showing a significant interaction with parental sex, in order to verify that observed parent-of-origin effects were not confounded by maternal carriage effects. Stepwise regression for identifying independent association signals In order to determine whether the genes investigated contained more than one independent association signal, a forward stepwise regression procedure was performed in the combined case parent, case control sample using UNPHASED (30) and in-house software. Due to problems with collinearity when analyzing the full set of highly correlated imputed markers, only the markers that were directly genotyped in the familybased sample and either directly genotyped or imputed with high quality in the TwinGene control sample were included in the stepwise regression (n ¼ 75). Using UNPHASED, for each gene, one marker at a time was tested for association to case/control status conditioning on the most strongly associated marker found in single-marker analysis. Association beyond that resulting from linkage disequilibrium with the conditioning marker was assessed through likelihood ratio tests. If one or more markers were significantly associated with disease at P, when conditioning on the first marker, the one with the lowest P-value was added to the set of conditioning markers. The same process was then applied iteratively by again testing single markers conditioning on the two markers already included and so on, until no more significant additions could be made. SUPPLEMENTARY MATERIAL Supplementary Material is available at HMG online. ACKNOWLEDGEMENTS The authors would like to thank all the study participants who contributed to this research. We furthermore thank Gerd Agerberg at the Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, Stockholm, Sweden and Gjøril B. Aas and Jan Ivar Martinsen at the Department of Etiologic Research, Cancer Registry of Norway, Oslo, Norway for organizing the collection of the case parent study samples and establishing an invitation database. We also thank Christine Ø. Buen, Hilde Herning, Manpret Kaur and Susanne S. Windju at the Faculty of Health Sciences, Oslo and Akershus University College of Applied Sciences, Oslo, Norway for helping with DNA isolation. Conflict of Interest statement. None declared. FUNDING This work was supported by the Norwegian Cancer Society (grant number ); the Nordic Cancer Union (grant number S-12/07) and the Swedish Cancer Society (grant numbers 2008/708, 2010/808, 2011/484). REFERENCES 1. Huyghe, E., Matsuda, T. and Thonneau, P. (2003) Increasing incidence of testicular cancer worldwide: a review. J. Urol., 170, Rosen, A., Jayram, G., Drazer, M. and Eggener, S.E. (2011) Global trends in testicular cancer incidence and mortality. Eur. Urol., 60, Rajpert-De Meyts, E. (2006) Developmental model for the pathogenesis of testicular carcinoma in situ: genetic and environmental aspects. Hum. Reprod. Update, 12, Hemminki,K. and Li,X. (2004) Familialrisk in testicularcancer as a clue to a heritable and environmental aetiology. Br. J. Cancer, 90, Czene, K., Lichtenstein, P. and Hemminki, K. (2002) Environmental and heritable causes of cancer among 9.6 million individuals in the Swedish family-cancer database. Int. J. Cancer, 99, Rapley, E.A., Turnbull, C., Al Olama, A.A., Dermitzakis, E.T., Linger, R., Huddart, R.A., Renwick, A., Hughes, D., Hines, S., Seal, S. et al. (2009) A genome-wide association study of testicular germ cell tumor. Nat. Genet., 41, Turnbull, C., Rapley, E.A., Seal, S., Pernet, D., Renwick, A., Hughes, D., Ricketts, M., Linger, R., Nsengimana, J., Deloukas, P. et al. (2010) Variants near DMRT1, TERT and ATF7IP are associated with testicular germ cell cancer. Nat. Genet., 42, Kanetsky, P.A., Mitra, N., Vardhanabhuti, S., Li, M., Vaughn, D.J., Letrero, R., Ciosek, S.L., Doody, D.R., Smith, L.M., Weaver, J. et al.(2009) Common variation in KITLG and at 5q31.3 predisposes to testicular germ cell cancer. Nat. Genet., 41, Pruim, R.J., Welch, R.P., Sanna, S., Teslovich, T.M., Chines, P.S., Gliedt, T.P., Boehnke, M., Abecasis, G.R. and Willer, C.J. (2010) LocusZoom: regional visualization of genome-wide association scan results. Bioinformatics, 26, Kanetsky, P.A., Mitra, N., Vardhanabhuti, S., Vaughn, D.J., Li, M., Ciosek, S.L., Letrero, R., D Andrea, K., Vaddi, M., Doody, D.R. et al. (2011) A second independent locus within DMRT1 is associated with testicular germ cell tumor susceptibility. Hum. Mol. Genet., 20, Weinberg, C.R. (1999) Methods for detection of parent-of-origin effects in genetic studies of case-parents triads. Am. J. Hum. Genet., 65, Mason, J.M., Morrison, D.J., Albert Basson, M. and Licht, J.D. (2006) Sprouty proteins: multifaceted negative-feedback regulators of receptor tyrosine kinase signaling. Trends Cell Biol., 16, Tennis, M., Van Scoyk, M., Freeman, S., Vandervest, K., Nemenoff, R. and Winn, R. (2010) Sprouty-4 inhibits transformed cell growth, migration and invasion, and EMT and is regulated by Wnt7A through PPARg in non-small cell lung cancer. Mol. Cancer Res., 8, Wang, J., Thompson, B., Ren, C., Ittmann, M. and Kwabi-Addo, B. (2006) Sprouty4, a suppressor of tumor cell motility, is downregulated by DNA methylation in human prostate cancer. Prostate, 66, Poynter, J.N., Hooten, A.J., Lindsay Frazier, A. and Ross, J.A. (2012) Associations between variants in KITLG, SPRY4, BAK1, and DMRT1 and pediatric germ cell tumors. Genes Chromosomes Cancer, 51, Lessel, D., Gamulin, M., Kulis, T., Toliat, M.R., Grgic, M., Friedrich, K., Žunec, R., Balija, M., Nürnberg, P., Kastelan, Z. et al. (2012) Replication of genetic susceptibility loci for testicular germ cell cancer in the Croatian population. Carcinogenesis, 33, Guilmatre, A. and Sharp, A. (2012) Parent of origin effects. Clin. Genet., 81,

8 3380 Human Molecular Genetics, 2013, Vol. 22, No Morison, I.M., Paton, C.J. and Cleverley, S.D. (2001) The imprinted gene and parent-of-origin effect database. Nucleic Acids Res., 29, Schneider, E., Pliushch, G., Hajj, N.E., Galetzka, D., Puhl, A., Schorsch, M., Frauenknecht, K., Riepert, T., Tresch, A., Müller, A.M. et al. (2010) Spatial, temporal and interindividual epigenetic variation of functionally important DNA methylation patterns. Nucleic Acids Res., 38, The ENCODE Project Consortium (2012) An integrated encyclopedia of DNA elements in the human genome. Nature, 489, Kent, W.J., Sugnet, C.W., Furey, T.S., Roskin, K.M., Pringle, T.H., Zahler, A.M. and Haussler, D. (2002) The human genome browser at UCSC. Genome. Res., 12, Anway, M.D. and Skinner, M.K. (2006) Epigenetic transgenerational actions of endocrine disruptors. Endocrinology, 147, s43 s Kong, A., Steinthorsdottir, V., Masson, G., Thorleifsson, G., Sulem, P., Besenbacher, S., Jonasdottir, A., Sigurdsson, A., Kristinsson, K.T., Jonasdottir, A. et al. (2009) Parental origin of sequence variants associated with complex diseases. Nature, 462, Furukawa, S., Haruta, M., Arai, Y., Honda, S., Ohshima, J., Sugawara, W., Kageyama, Y., Higashi, Y., Nishida, K., Tsunematsu, Y. et al. (2009) Yolk sac tumor but not seminoma or teratoma is associated with abnormal epigenetic reprogramming pathway and shows frequent hypermethylation of various tumor suppressor genes. Cancer Sci., 100, Kawakami, T., Zhang, C., Okada, Y. and Okamoto, K. (2006) Erasure of methylation imprint at the promoter and CTCF-binding site upstream of H19 in human testicular germ cell tumors of adolescents indicate their fetal germ cell origin. Oncogene, 25, Figueroa, J., Sakoda, L., Graubard, B., Chanock, S., Rubertone, M., Erickson, R. and McGlynn, K. (2008) Genetic variation in hormone metabolizing genes and risk of testicular germ cell tumors. Cancer Causes Control, 19, Kristiansen, W., Haugen, T.B., Witczak, O., Andersen, J.M., Fosså, S.D. and Aschim, E.L. (2011) CYP1A1, CYP3A5 and CYP3A7 polymorphisms and testicular cancer susceptibility. Int. J. Androl., 34, Kristiansen, W., Andreassen, K.E., Karlsson, R., Aschim, E.L., Bremnes, R.M., Dahl, O., Fosså, S.D., Klepp, O., Langberg, C.W., Solberg, A. et al. (2012) Gene variations in sex hormone pathways and the risk of testicular germ cell tumour: a case parent triad study in a Norwegian Swedish population. Hum. Reprod., 27, Davis-Dao, C.A., Siegmund, K.D., Vandenberg, D.J., Skinner, E.C., Coetzee, G.A., Thomas, D.C., Pike, M.C. and Cortessis, V.K. (2011) Heterogenous effect of androgen receptor CAG tract length on testicular germ cell tumor risk: shorter repeats associated with seminoma but not other histologic types. Carcinogenesis, 32, Dudbridge, F. (2008) Likelihood-based association analysis for nuclear families and unrelated subjects with missing genotype data. Hum. Hered., 66, Rahman, I., Bennet, A.M., Pedersen, N.L., De Faire, U., Svensson, P. and Magnusson, P.K.E. (2009) Genetic dominance influences blood biomarker levels in a sample of 12,000 Swedish elderly twins. Twin Res. Hum. Genet., 12, De Bakker, P.I.W., Yelensky, R., Pe er, I., Gabriel, S.B., Daly, M.J. and Altshuler, D. (2005) Efficiency and power in genetic association studies. Nat. Genet., 37, The International HapMap Consortium (2007) A second generation human haplotype map of over 3.1 million SNPs. Nature, 449, Browning, S.R. and Browning, B.L. (2007) Rapid and accurate haplotype phasingand missing-data inference for whole-genomeassociationstudies by use of localized haplotype clustering. Am. J. Hum. Genet., 81, Browning, B.L. and Browning, S.R. (2009) A unified approach to genotype imputation and haplotype-phase inference for large data sets of trios and unrelated individuals. Am. J. Hum. Genet., 84, The 1000 Genomes Project Consortium (2010) A map of human genome variation from population-scale sequencing. Nature, 467, Purcell, S., Neale, B., Todd-Brown, K., Thomas, L., Ferreira, M.A.R., Bender, D., Maller, J., Sklar, P., De Bakker, P.I.W., Daly, M.J. et al. (2007) PLINK: a tool set for whole-genome association and population-based linkage analyses. Am. J. Hum. Genet., 81, Strimmer, K. (2008) A unified approach to false discovery rate estimation. BMC Bioinformatics, 9, Strimmer, K. (2008) fdrtool: a versatile R package for estimating local and tail area-based false discovery rates. Bioinformatics, 24,