Abstract

   Genome‐wide association studies have identified several germline
   variants in gastric cancer. Meanwhile, sequencing studies have
   characterized extensive somatic alterations that arise during gastric
   carcinogenesis. However, the relationship between the germline variants
   and somatic alterations is still unclear in gastric cancer. A total of
   11 susceptibility loci and 276 driver genes of gastric cancer were
   determined based on previous studies and publicly available database.
   An enrichment analysis was made to detect whether driver genes were
   enriched in susceptibility regions. Besides, we performed a pathway
   enrichment analysis to find common‐enrich pathways of cancer driver
   genes and susceptibility genes. Finally, on the basis of the gastric
   cancer samples and data from TCGA STAD project, we evaluated the
   associations between susceptibility loci and somatic alterations.
   Enrichment analysis showed that gastric cancer susceptibility genes
   were more likely to be enriched in driver genes than in all the genes
   (P = .05). The susceptibility genes and driver genes were commonly
   enriched in 8 biological pathways. Gastric cancer susceptibility locus
   of rs2285947 was associated with truncation mutation within Signaling
   by PDGF pathway (OR = 0.26, 95%CI: 0.12‐0.55, P = 3.93 × 10^−4). The
   rs1679709 was connected with COSMIC Signature15 (P = .026). Moreover,
   rs1679709 was also associated with copy number values of RFC4 which is
   related to Signature15. These results provide evidence for the
   relationship between germline variants and somatic alterations, which
   facilitate understanding the interactive mechanism of germline
   variations with somatic alterations in gastric cancer development.

   Keywords: association studies, enrichment analysis, gastric cancer,
   germline variants, somatic alteration

1. INTRODUCTION

   Gastric cancer is the second most common cancer and the second leading
   cause of cancer death in China.[32]1 The environmental risk factors for
   gastric cancer include high‐salt diet, smoking, and infectious
   agents.[33]2 Besides, there are still numerous genetic factors which
   determine an individual's predisposition to gastric cancer.[34]3 Many
   genetic variations, most of which are single nucleotide polymorphisms
   (SNPs), have been detected over the past years by the genome‐wide
   association studies (GWAS) of gastric cancer.[35]4 These common
   susceptibility loci included 1q22 (MUC1), 3q13.32 (ZBTB20), 5p13.1
   (PRKAA1), 5q14.3 (lnc‐POLR3G‐4), 6p21.1 (UNC5CL), 8q24 (PSCA), and
   10q23 (PLCE1).[36]5, [37]6, [38]7, [39]8, [40]9, [41]10

   In parallel, a growing number of whole‐exome and whole‐genome
   sequencing studies have been conducted to define the landscape of
   somatic mutations in gastric cancer. These studies have identified many
   driver genes, whose mutations confer selective growth advantage to
   tumor.[42]11 Some of these driver genes are previously known cancer
   genes (eg, TP53, ARID1A, and CDH1), while the others are new‐found
   significantly mutated genes in gastric cancer (eg, MUC6, CTNNA2, GLI3,
   and RNF43).[43]4, [44]12, [45]13, [46]14, [47]15, [48]16, [49]17,
   [50]18, [51]19, [52]20, [53]21, [54]22 Moreover, the copy number
   changes and characteristic mutational signatures also play important
   roles in gastric cancer development.[55]4, [56]16, [57]17, [58]18,
   [59]19

   Recent studies have revealed the associations between germline
   mutations and somatic alterations in tumor development. According to
   these studies, a large fraction of cancer predisposition genes can
   contribute to oncogenesis when they have somatic mutation events in
   tumors.[60]23 The germline MC1R status may influence somatic mutation
   burden in melanoma, and the common germline risk variants are connected
   with total somatic mutation count in breast cancer.[61]24, [62]25 In
   addition, another study reported that oncoprotein EWSR1‐FLI1
   preferentially bound to the risk allele of susceptibility SNP
   rs79965208 in Ewing sarcoma.[63]26 However, the associations between
   the genetic susceptibility variants and somatic alterations in gastric
   cancer are still unknown.

   In this study, we set out to examine the association between the
   established gastric cancer susceptibility loci and the somatic
   alterations. First, we made enrichment analyses to examine whether
   driver genes are enriched in germline susceptibility regions, and
   whether cancer susceptibility genes are enriched in driver genes. Then
   we made a pathway enrichment analysis to explore whether driver genes
   and susceptibility genes are enriched in the same pathways. Finally,
   serial association analyses were conducted to investigate how the risk
   SNP genotypes affect the somatic alterations during gastric cancer
   development (Figure [64]1).

Figure 1.

   Figure 1
   [65]Open in a new tab

   The flowchart of the study design. Firstly, we collected gastric cancer
   risk SNPs and driver genes. Next we built the susceptibility regions
   and identified cancer susceptibility genes (CSGs), following by an
   enrichment analysis to investigate whether driver genes are more likely
   to locate within susceptibility regions, and whether susceptibility
   genes and driver genes are enriched in common biological pathways.
   Finally we made association analyses between gastric cancer risk SNPs
   genotypes and several somatic events. GC, gastric cancer; CPG, cancer
   predisposing gene; MMR, DNA mismatch repair pathway

2. MATERIALS AND METHODS

2.1. Risk SNPs and cancer susceptibility genes

   Gastric cancer risk SNPs were extracted from the original gastric
   cancer GWAS studies in GWAS Catalog.[66]5, [67]6, [68]7, [69]8, [70]9,
   [71]10, [72]27 Only 1 SNP was remained when multiple variants were in
   linkage disequilibrium (LD, r ^2 ≥ .8), and the minor‐allele frequency
   (MAF) should be greater or equal than .05. In addition, we included SNP
   rs1679709 which was reported in a new gastric cancer GWAS study in 2017
   and meet the above selection criteria.[73]28

   Gastric cancer susceptibility regions were defined as 200 kb upstream
   and downstream of a risk SNP. The protein coding genes located in these
   susceptibility regions were defined as cancer susceptibility genes
   (CSGs). Moreover, based on the Genotype‐Tissue Expression (GTEx) v6p
   database, the protein coding genes whose expression in stomach tissue
   was associated with risk SNPs or SNPs in high LD (r ^2 ≥ .8) with risk
   SNPs were also defined as CSGs.[74]29 Genomic coordinates and gene
   symbols of the protein coding genes were gained from GENCODE version
   19.[75]30

2.2. Driver genes

   Gastric cancer driver genes were obtained from 3 sources: (1) 16
   gastric cancer‐related driver genes from Catalogue of Somatic Mutations
   in Cancer (COSMIC) Cancer Gene Census(v78)[76]31; (2) 175 driver genes
   of gastric cancer from the Integrative Onco Genomics (IntOGen)
   database[77]32; (3) 108 significantly mutated genes (SMGs) and 26
   somatic copy number alteration genes from previously published Whole
   Genome Sequencing (WGS) or Whole Exome Sequencing (WES) articles.[78]4,
   [79]12, [80]13, [81]14, [82]15, [83]16, [84]17, [85]18, [86]19, [87]20,
   [88]21, [89]22

2.3. Region enrichment analysis and gene‐based enrichment analysis

   First of all, we calculated the proportion of driver genes in
   susceptibility regions. A random sampling of SNP sets was carried out
   to appraise the background distribution of driver genes, and a total of
   10 000 sets of randomly sampled SNPs, which were similar with risk SNP
   sets in several genomic features, were obtained using the SNPsnap
   online server. The SNPsnap Web server provide matched sets of SNPs
   based on allele frequency, number of SNPs in LD, distance to nearest
   gene and gene density that can be used to calibrate background
   expectations.[90]33 We calculated the number of random SNP sets with
   proportions of driver genes equal to or larger than that of risk SNPs.
   Then we divided the calculated number by 10 000 and defined it as the P
   values for enrichment of driver genes in susceptibility regions. To
   detect whether the length of flanking regions will affect the
   enrichment result, we defined 50kb, 100kb, 200kb and 500kb upstream and
   downstream of a risk SNP as the susceptibility regions, calculating P
   values for each susceptibility region separately.

   In the gene‐based enrichment analysis, we computed the enrichment ratio
   of CSGs in all genes (56 318 genes from GENCODE V19) and that in driver
   genes. The fold enrichment ratio between these 2 ratios was counted as
   well. The P value of fold enrichment ratio was counted using the
   “phyper” function in R 3.3.1.

2.4. Pathway enrichment analysis

   Reactome pathways were downloaded from the MSigDB website
   ([91]http://software.broadinstitute.org/gsea/index.jsp/). Each Reactome
   pathway included in our analysis should contain at least 1 gastric
   cancer driver gene or susceptibility gene. We used the “phyper”
   function as implemented in R to compute the P value for enrichment of
   CSGs or driver genes in a given pathway. The P value was adjusted to
   account for multiple hypotheses testing with Benjamini‐Hochberg
   correction and we defined a pathway to be significant if the false
   discovery rate (FDR) ≤ 0.1.

2.5. Genotype

   We used data from TCGA STAD project to perform association analyses.
   The germline genotypes were generated using the Affymetrix Genome‐Wide
   Human SNP Array 6.0. There were 442 cases left after the standard
   quality control process. Then we conducted genotype imputation using
   SHAPIT for prephasing and IMPUTE2 for imputation, based on the 1000
   Genomes Project Phase III integrated variant set release.[92]34, [93]35
   Risk SNPs included in the association analysis should meet the
   standards as follows: imputation info ≥0.5, minor allele frequency
   (MAF) ≥0.01, and Hardy‐Weinberg equilibrium P‐values ≥.001.

2.6. Association analysis on somatic mutations within driver genes and
somatic truncation mutations within key pathways

   The mutation information of gastric cancer was available online, based
   on the whole‐exome sequencing data supplied by TCGA‐STAD project and we
   used the mutation annotation file
   (TCGA.STAD.mutect.a88b4065‐34b4‐4858‐9c16‐55def79c38f2.DR‐6.0.public.ma
   f) from Genomic data commons (GDC) data portal
   ([94]https://portal.gdc.cancer.gov/projects/TCGA-STAD). The 276 driver
   genes described above were included into analysis, and for each
   patient, a driver gene was considered mutated if one or more DNA
   mutations mapped to this gene.

   Pathways where driver genes and CSGs were both significantly enriched
   in were defined as key pathways. As a result, a total of 8 Reactome
   pathways were included in our analysis. For each patient, a key pathway
   was considered to be mutated if one or more truncation mutations were
   detected in this pathway.

2.7. Association analysis on somatic copy number alterations

   Output files of SNP6 copy number analysis (GISTIC2) were obtained from
   the Broad Institute Genome Data Analysis Center (GDAC) Firehose portal
   ([95]http://gdac.broadinstitute.org/runs/analyses__latest/reports/cance
   r/STAD-TP/CopyNumber_Gistic2/nozzle.html). We extracted the CNV
   regional information from the “all_lesions.conf_99.txt” file, and it
   contains data about the significant regions of amplification and
   deletion as well as which samples are amplified or deleted in each of
   these regions. Besides, we extracted gene‐level copy number values of
   gastric cancer driver genes in each TCGA STAD sample from the
   “all_data_by_genes.txt” file. We take the absolute values of the copy
   number values and our analysis only focus on the copy number values of
   276 driver genes.

2.8. Association analysis on COSMIC signatures

   We calculated the weight of each mutational signature contributing to
   an individual TCGA STAD sample using the “deconstructSigs” package in R
   3.3.1 (package details in
   [96]https://github.com/raerose01/deconstructSigs) with the mutation
   annotation file described above
   (TCGA.STAD.mutect.a88b4065‐34b4‐4858‐9c16‐55def79c38f2.DR‐6.0.public.ma
   f). We included the 11 COSMIC signatures which had been reported in
   gastric cancer into our analysis (COSMIC signatures 1, 2, 5, 13, 15,
   17, 18, 20, 21, 26, and 28). For each patient, the weights of each
   mutational signature were used as somatic phenotype.

2.9. Collection of DNA mismatch repair genes

   DNA mismatch repair (MMR) genes were collected from 2 resources:
   identified MMR genes in a reported study and the genes in KEGG mismatch
   repair pathway.[97]21 As a result, a total of 26 MMR genes were
   included in our analysis (Table [98]S1).

2.10. Statistical analyses for association analysis

   Logistic regression was performed for binary phenotypes, and multiple
   linear regression was performed for quantitative traits in the
   association analysis progress. The additive model was utilized in our
   study, and we controlled for age, gender, clinical stage, and the first
   10 principal components. The clinical information (age, gender, and
   clinical stage) was obtained from GDC data portal, and the missing
   clinical variable values were imputed with the corresponding median
   values in our study. The association p values would be adjusted by
   Benjamini‐Hochberg correction method, and all the tests were two‐sided,
   a false discovery rate (FDR) of 0.1 was used as significance threshold.
   All the association analyses were conducted in R‐3.3.1
   ([99]http://www.R-project.org/).

3. RESULTS

   We identified 11 gastric cancer risk SNPs according to our definitions
   and quality control standards (Table [100]S2). Besides, we also
   identified 74 cancer susceptibility genes and collected 276 driver
   genes from COSMIC Cancer Gene Census, IntOGen database, and reported
   WGS or WES studies (Tables [101]S3 and [102]S4). With the enrichment
   analysis, we found a trend that the proportion of driver genes in
   susceptibility regions decreased with the increasing size of
   susceptibility regions, as from 50 kb to 500 kb upstream and downstream
   of risk SNPs. Inversely, the enrichment P‐values increased from 0.0956,
   0.1248, 0.175 to 0.3064 (Figures [103]2 and [104]3A, Table [105]S5).

Figure 2.

   Figure 2
   [106]Open in a new tab

   The enrichment analysis of Driver genes in SNP susceptibility regions.
   The figure showed the proportion results of gastric cancer driver genes
   in the 4 SNP susceptibility regions: A, The 50 kb‐amplification
   susceptibility regions. B, The 100 kb‐amplification susceptibility
   regions. C, The 200 kb‐amplification susceptibility regions. D, The
   500 kb‐amplification susceptibility regions

Figure 3.

   Figure 3
   [107]Open in a new tab

   The tendency of Driver genes in SNP susceptibility region and the
   gene‐based enrichment analysis. A, The trend of proportions of gastric
   cancer driver genes and −log10 transformed enrichment p‐values of
   driver genes in 50 kb, 100 kb, 200 kb, and 500 kb upstream and
   downstream of risk SNPs. B, A bar graph of the percentage of CSGs in
   the 276 driver genes compared with all genes (56 318) in GENCODE (V19)

   Gene‐based enrichment analysis showed that 2 CSGs were among the 276
   driver genes (0.72%), which represented a 5.51‐fold enrichment compared
   with the 56 318 annotated genes in GENCODE (P = .05, Figure [108]3B).
   According to pathway enrichment analysis result, CSGs and driver genes
   were commonly enriched in 8 biological pathways. These pathways
   included insulin receptor signaling cascade, PI3K cascade, PPARA
   activates gene expression, semaphoring interactions, signaling by
   insulin receptor, signaling by PDGF, PERK regulated gene expression,
   and other semaphoring interactions (Figure [109]4, Table [110]S6).

Figure 4.

   Figure 4
   [111]Open in a new tab

   Reactome pathways with FDR<0.1 in hyper‐representative test for gastric
   cancer driver genes and gastric cancer susceptibility genes. A,
   Enrichment ratios of significant pathways. B, −log10 transformed false
   discovery rates (FDRs) of significant pathways

   Association analysis identified a total of 130 associations (P < .05)
   between somatic mutations in driver genes and risk SNP genotypes, of
   which the strongest associations were rs9841504 at 3q13.32 with POLE
   (P = 7.36 × 10^−4, Table [112]S7) and rs2285947 at 7p15.3 with SOS1
   (P = 2.06 × 10^−3, Table [113]S7). However, no association remained
   significant after FDR adjustment (Table [114]S7). We wondered whether
   the 130 SNP‐gene pairs remain significant on gene expression level,
   followed with an expression quantitative trait loci (eQTL) analysis in
   stomach tissues based on GTEx. As a result, the eQTL analysis
   identified 11 associations between the gene expression levels and
   genotypes (P < .05, Figure [115]S1, Table [116]S8). However, no results
   remain significant after FDR adjustment.

   Analysis of truncation mutations in key pathways identified one
   significant result with FDR < 0.1 (Table [117]S9). The risk allele (A)
   of rs2285947 (7p15.3) was associated with Signaling by PDGF pathway
   (OR = 0.26, 95% CI: 0.12‐0.55, P = 3.93 × 10^−4) (Table [118]1). To
   further explore the association between rs2285947 and signaling by PDGF
   pathway, we performed a stratified analysis by age, sex, ethnicity, and
   tumor stage status. As shown in Table [119]S10, the association
   remained significant in both age groups, females and Caucasians.

Table 1.

   The associations between rs2285947 and truncation mutation within
   Signaling by PDGF pathway
   Genotypes Cases with mutation (n = 28)[120]^a Cases without mutation
   (n = 386)[121]^a OR (95% CI)[122]^b P value[123]^b
   N % N %
   GG 18 64.29 131 33.94 1
   GA 10 35.71 191 49.48 0.36 (0.15‐0.85) 1.94E−02
   AA 0 0 64 16.58 ‐ ‐
   GA/AA 10 35.71 255 66.06 0.24 (0.10‐0.58) 1.38E−03
   Additive model 0.26 (0.12‐0.55) 3.93E−04
   [124]Open in a new tab
   ^a

   Patients with or without somatic truncation mutations in pathway.
   ^b

   Odds ratios, 95% confidence intervals, and P values were calculated
   using logistic regression models adjusting for age, gender, clinical
   stage, and the first 10 principal components.

   We investigated the interactions between risk SNP genotypes and somatic
   copy number alterations (SCNAs). Results showed there were several
   “germline‐SCNAs” associations, like rs2494938 (6p21.1) with the ERBB3
   gene copy number values (P = 2.38 × 10^−3), rs13361707 (5p13.1) with
   the significant focal amplification region 12p12.1 (P = 9.03 × 10^−3),
   and rs1679709 (6p22.1) with the significant focal deletion region
   9q21.11 (P = 5.33 × 10^−3) (Tables [125]S11‐S13). However, we did not
   find any association with FDR < 0.1.

   There were 11 patterns of mutational signatures found in gastric
   cancer, and our analysis identified 5 pairs of associations with
   P‐values less than .05 between risk SNPs genotypes and the weight of
   mutational signature (Table [126]S14). Signature15 were associated with
   rs1679709 (6p22.1) (P = .026), as signature15 was DNA mismatch repair
   related signature according to COSMIC website. Thus, we further
   investigated whether rs1679709 genotypes were associated with
   expression of MMR genes. The results showed rs1679709 was concerned
   with 1 gene in gene‐level copy number values: RFC4 (P = 1.25 × 10^−2)
   (Figure [127]5, Table [128]S15).

Figure 5.

   Figure 5
   [129]Open in a new tab

   The association between genotypes of rs1679709 and gene copy number
   values of RFC4 gene. The box plot displays the first and third
   quartiles (top and bottom of the boxes), the median (band inside the
   boxes), and the lowest and highest point within 1.5 times the
   interquartile range of the lower and higher quartile

4. DISCUSSION

   Region enrichment analysis showed that there was no difference between
   the proportion of driver genes in cancer susceptibility regions and in
   the background regions. A recent study has reported that cancer
   susceptibility regions have gene mutation frequencies comparable to
   background mutation frequencies.[130]36 Besides, with the gene‐based
   enrichment analysis we observed that gastric cancer CSGs were more
   likely to be enriched in driver genes, although it was only marginal
   significant. It makes sense as somatic cancer driver mutations and
   germline cancer predisposing mutations are highly overlapped, whereas
   such mutual interrogation had been underestimated due to the artifact
   of different research approaches.[131]23 In addition, pathway
   enrichment analysis found CSGs and driver genes were commonly enriched
   in 8 Reactome pathways. The result implied that germline mutations and
   somatic mutations may work together in some particular biological
   pathways during gastric cancer development.

   Our interaction analysis of somatic mutations in driver genes and risk
   SNPs genotypes identified several significant SNP‐gene pairs. As a
   previous study reported, genetic background could influence the somatic
   evolution of a tumor by modifying the likelihood of acquiring mutations
   in specific cancer genes,[132]37 which could possibly explain why there
   were a number of SNP‐gene pairs found in our study.

   A pathway‐based analysis was performed later, and we detected that the
   risk allele of rs2285947 (A) was significantly associated with the
   occurrence of truncation mutations in Reactome pathway Signaling by
   PDGF. PDGF pathway has long been implicated in cancers and is known to
   be involved in many biological processes.[133]38 Rs2285947 is an
   intronic variant in DNAH11 and is able to regulate expression levels of
   DNAH11 in stomach tissues based on GTEx (P = 3.40 × 10^−5), as DNAH11
   is a microtubule‐dependent motor ATPase according to GeneCards website.
   Results showed that risk allele of rs2285947 (A) has an inverse
   association with DNAH11 gene expression (effect size = −0.28) and PDGF
   signaling pathway (Beta = −1.35) at the same time. One possible
   interpretation is that rs2285947 may affect the ATPase activity by
   regulating DNAH11 expression level, and it may influence the release of
   PDGF‐R, since E5 oncoprotein can form a ternary complex with PDGF‐R and
   the 16K subunit of the vacuolar V‐ATPase.[134]39 These PDGF receptors
   may dimerize and undergo autophosphorylation and then attract
   downstream effectors to transduct the signal into the cell.[135]40,
   [136]41, [137]42 According to the “two‐hit” model which explain the
   interaction of germline and somatic mutation, individuals with elevated
   genetic predisposition may require fewer stages to develop a tumor than
   those at lower genetic risk. Thus, it is possible that the observed
   inverse association between rs2285947 and PDGF signaling pathway is a
   result of the continuous process of cancer development, where both
   germline variants and somatic alterations contribute to the development
   of gastric cancer.

   In the past few years, there are 30 patterns of mutational signatures
   be found across the spectrum of human cancer types from many
   large‐scale analyses,[138]43, [139]44, [140]45, [141]46, [142]47
   including 11 types of signatures found in gastric cancer. We observed a
   significant association between rs1679709 and the weights of COSMIC
   Signature 15. Signature 15 is linked to defective DNA mismatch repair
   according to COSMIC website.[143]48 In the following analysis, we found
   gene copy number values of MMR gene RFC4 was also related to rs1679709
   genotypes. Therefore, rs1679709 might influence defective DNA mismatch
   repair associated mutational signatures in gastric cancer, via
   regulating the copy number changes in relevant genes like RFC4.

   Nowadays fundamental gaps remain in our knowledge of how normal cells
   evolve into cancer cells and what are the vital potential genetic
   backgrounds. In this research, we did identified several interactions
   between germline susceptibility loci and somatic mutations in gastric
   cancer. However, the exact mechanism how the germline alleles affect
   the progression of later somatic events is still unknown because of the
   limit of study sample scale and the lack of functional experiments. In
   addition, it is a limitation that there are only 89 Asians among 443
   cases in TCGA STAD, but most germline susceptibility loci were
   identified in Asian populations. We believe with more susceptibility
   loci discovered and larger samples of sequencing studies performed in
   the future, especially in Asian population, the networks of informative
   interactions between germline mutation and somatic mutation in gastric
   cancer will eventually be revealed, which may make contributions to the
   precision medicine.

CONFLICT OF INTEREST

   The authors have no conflict of interest.

Supporting information


   [144]Click here for additional data file.^ (377.8KB, pdf)


   [145]Click here for additional data file.^ (11.2KB, xlsx)


   [146]Click here for additional data file.^ (11.2KB, xlsx)


   [147]Click here for additional data file.^ (13.3KB, xlsx)


   [148]Click here for additional data file.^ (26.9KB, xlsx)


   [149]Click here for additional data file.^ (10.1KB, xlsx)


   [150]Click here for additional data file.^ (11KB, xlsx)


   [151]Click here for additional data file.^ (376.5KB, xlsx)


   [152]Click here for additional data file.^ (17.8KB, xlsx)


   [153]Click here for additional data file.^ (18KB, xlsx)


   [154]Click here for additional data file.^ (11KB, xlsx)


   [155]Click here for additional data file.^ (258.3KB, xlsx)


   [156]Click here for additional data file.^ (42.7KB, xlsx)


   [157]Click here for additional data file.^ (56.2KB, xlsx)


   [158]Click here for additional data file.^ (20.1KB, xlsx)


   [159]Click here for additional data file.^ (11.9KB, xlsx)

ACKNOWLEDGMENTS