Tumor somatic mutations also existing as germline polymorphisms may help to identify functional SNPs from genome-wide association studies

Abstract

We hypothesized that a joint analysis of cancer risk-associated single-nucleotide polymorphism (SNP) and somatic mutations in tumor samples can predict functional and potentially causal SNPs from GWASs. We used mutations reported in the Catalog of Somatic Mutations in Cancer (COSMIC). Confirmed somatic mutations were subdivided into two groups: (1) mutations reported as SNPs, which we call mutational/SNPs and (2) somatic mutations that are not reported as SNPs, which we call mutational/noSNPs. It is generally accepted that the number of times a somatic mutation is reported in COSMIC correlates with its selective advantage to tumors, with more frequently reported mutations being more functional and providing a stronger selective advantage to the tumor cell. We found that mutations reported ≥10 times in COSMIC—frequent mutational/SNPs (fmSNPs) are likely to be functional. We identified 12 cancer risk-associated SNPs reported in the Catalog of published GWASs at least 10 times as confirmed somatic mutations and therefore deemed to be functional. Additionally, we have identified 42 SNPs that are tightly linked (R² ≥ 0.8) to SNPs reported in the Catalog of published GWASs as cancer risk associated and that are also reported as fmSNPs. As a result, 54 candidate functional/potentially causal cancer risk associated SNPs were identified. We found that fmSNPs are more likely to be located in evolutionarily conserved regions compared with cancer risk associated SNPs that are not fmSNPs. We also found that fmSNPs also underwent positive selection, which can explain why they exist as population polymorphisms.

The study identified candidates for causal SNPs among GWAS-detected cancer risk SNPs and SNPs that are tightly linked to cancer GWAS-reported ones. The results of the analysis can be used to guide functional studies of cancer risk-associated SNPs.

Introduction

Two major cancer genetics research activities are as follows: (1) identification of cancer risk-associated single-nucleotide polymorphisms (SNPs) by genome-wide association studies (GWASs) and (2) identification of somatic mutations in tumor samples. These disciplines barely talk to each other even though cross talk between these two areas is likely to be beneficial. Here we jointly consider cancer risk-associated SNPs and somatic mutations detected in tumor samples. Our hypothesis was that considering SNPs and somatic mutations together will help to identify cancer risk-associated SNPs that are functional.

GWASs have identified a very large number of SNPs associated with cancer risk (1–3). It is generally accepted that the majority of GWAS-detected SNPs are not functional/causal SNPs, but are rather proxies linked to unknown causal variants (4,5). Detection of causal/functional variants among GWAS-detected SNPs is challenging. Several bioinformatics tools have been developed to predict functional/potentially causal SNPs. These tools use SNP characteristics including the level of evolutionary conservation of the site (6), projected effect of the SNP on protein structure (7) and other SNP features (8) for assessing the SNP’s functionality. To our best knowledge, these tools never used somatic mutation data to predict SNP functionality.

To identify tumor somatic mutations that also exist as SNPs, we have overlapped SNPs reported in the dbSNP database (9,10), with somatic mutations from the Catalog Of Somatic Mutations In Cancer (COSMIC) (11,12). Several millions of unique somatic mutations are reported in COSMIC. Mutations differ by the number of times they are reported in COSMIC, with the majority of them being singletons. To a great extent, cancer development is driven by an acquisition of driver mutations providing selective advantage (e.g. a higher proliferation rate or better survival) to the mutant clone. Clone-carrying driver mutations survive better and propagate faster (13). A higher propagation rate of cell clones with driver mutations results in an excess of driver mutations when compared with selectively neutral passenger mutations (14,15). Therefore, mutations providing a selective advantage (driver mutations) are detected more frequently in tumor samples compared with selectively neutral passenger mutations (15). Therefore, the number of times a somatic mutation is reported can be used as an indicator of functionality (16,17). If a frequent and therefore potentially functional somatic mutation also exists as a SNP, one can expect that it will be functional as an SNP also. In this study, we used somatic mutations detected in tumor samples jointly with SNP to identify candidate functional/potentially causal variants among GWAS-detected cancer risk-associated SNPs.

Materials and methods

Identification of somatic mutations that also exist as SNPs

Our first step was to identify somatic mutations that also exist as SNPs, regardless of whether SNPs were reported as cancer risk-associated or not. Hereafter we call them mutational/SNPs (mSNPs). To identify mSNPs we overlapped confirmed somatic mutations, detected by whole genome screens and reported in COSMIC (Build 88) with SNPs reported in dbSNP database. Confirmed somatic mutations are mutations established to be somatic rather than germline polymorphisms based on sequencing of paired normal tissue or in some cases comparison of the detected variant with the SNP database. Matching somatic mutations to SNPs was done based on the condition that the following three characteristics are the same for the somatic mutation and the SNP: (i) chromosome number, (ii) nucleotide position on the chromosome and (iii) the type of nucleotide substitution, e.g. C>T. We used human genome Build 38 for both SNPs and somatic mutations.

The number of times a somatic mutation is reported in COSMIC was used as a measure of its functionality. Before counting individual mutations, we excluded duplicates related to using different reference transcripts for annotation. Exactly the same mutation can be reported in COSMIC several times depending on the transcript used as a reference. To remove reference transcript-related duplicates, we identified mutations with the same chromosomal position, same type of nucleotide substitution and the same sample ID and removed all duplicates. A total of 1 719 388 annotation duplicates were detected and removed. After removing annotation duplicates, 2 955 675 mSNPs were identified. Those mSNPs were detected in 40 550 tumor samples across 42 cancer types.

Minor allele frequency and number of mutational counts

mSNPs are genetic variants that exist as both somatic mutations and germline polymorphisms. As a result of their dual nature, mSNPs have two key characteristics: the number of counts in COSMIC and minor allele frequency (MAF). Mutational counts were estimated for confirmed somatic mutations detected by whole genome sequencing. MAFs were estimated based on the data from Trans-Omics for Precision Medicine (TOPMed) project (18). We found that the majority of mSNPs are rare: >95% of them have MAFs < 0.001. mSNPs with MAF < 0.001 were excluded from the analysis because they may be false positives and also because their practical/clinical significance is questionable. After removing rare mSNPs, the total number of mSNPs was 599 630. Among these, we studied mSNPs reported in COSMIC at least 10 times as somatic mutations. The goal of implementing this threshold was to exclude mSNPs that are likely not functional. The justification for using 10 counts as a threshold is given below. Hereafter we refer to the mSNPs reported at least 10 times in COSMIC as frequent mutational/SNPs (fmSNPs). In total, 8533 fmSNPs were identified (Supplementary Table S1, available at Carcinogenesis Online). The table includes mutation ID, SNP ID, number of times the mutation is reported in COSMIC, MAF from TOPMed database and prediction of functionality by functional analysis through hidden Markov model (FATHMM) (19).

Selecting threshold for mSNPs that are likely to be functional

We found that the majority of COSMIC mutations are singletons. Singletons are likely to be selectively neutral and their presence in tumor samples reflects the randomness of the mutational process (20). On the other hand, somatic mutations frequently detected in tumor samples are likely to be functional: they are positively selected because tumor cells need them to proliferate and survive (21–23). Even though we know that singletons are likely to be selectively neutral and frequent somatic mutations are likely to be functional it is difficult, however, to decide where to put a threshold between neutral and functional mutations.

We used two approaches to decide on the boundary between functional mSNPs and mSNPs that are likely to be noise. In the first approach, we categorized mSNPs based on the number of times they are reported in COSMIC. In each category, we estimated the proportion of mSNPs predicted to be deleterious by FATHMM method (19). Our assumption was that the proportion of mSNPs predicted to be deleterious by FATHMM reflects a proportion of functional SNPs in the group.

The second approach was gene based. We first identified genes linked to mSNPs and then checked if those genes cluster in cancer pathways. For this analysis, all mSNPs were divided into five categories based on the number of counts in COSMIC: singletons, mSNPs reported 2–4 times, SNPs reported 5–9 times, mSNPs reported 9–19 times and mSNPs reported ≥20 times in COSMIC. The grouping strategy was chosen to ensure that the numbers of mSNPs across categories are comparable to each other. For each category of genes, we conducted the pathway enrichment analysis and counted the number of cancer-related pathways among the 20 top pathways. The pathways were defined by Kyoto Encyclopedia of Genes and Genomes (KEGG). KEGG pathways were designed using published data on proteins interactions as well as experimental evidence (24). Therefore, circular reasoning (selecting cancer-relevant genes based on how frequently they are mutated in cancer) is unlikely to be an issue in this analysis. However, it would be an issue, if we used the proportions of COSMIC-defined cancer census genes as a measure of cancer relevance.

Cancer risk-associated SNPs

Cancer risk-associated SNPs were retrieved from the Catalog of the published GWASs (25). The database was accessed 14 October 2019, and findings from 194 cancer GWASs were available. Table 1 shows the numbers of SNPs, genes and numbers of published GWASs for different cancer types. Cancer types are defined in the table exactly how they defined in the catalog. Supplementary Table S2 (available at Carcinogenesis Online) shows the complete list of cancer risk-associated SNPs used in this study. A total of 1013 unique SNPs were reported in the catalog as cancer risk associated at the GWA significance level—P ≤ 5 × 10⁻⁸. These SNPs are linked to 1011 genes. Because some SNPs are associated with risk in multiple cancers, the number of reports/lines in Supplementary Table S1 (available at Carcinogenesis Online)is larger than the number of unique SNPs.

Table 1.

Number of SNPs and linked genes associated with risk of different cancer types based on the data from the Catalog of the published GWASs

Cancer type	Number of SNPs	Number of unique genes	Number of studies
Prostate cancer	161	136	30
Breast cancer	134	147	37
Lung cancer	75	95	23
Colorectal cancer	74	84	30
Testicular germ cell tumor	59	68	8
Basal cell carcinoma	45	53	8
Squamous cell lung carcinoma	43	59	4
Breast cancer (estrogen-receptor negative)	43	45	3
Non-melanoma skin cancer	41	56	2
Lung cancer in ever smokers	33	47	18
Pancreatic cancer	32	36	9
Lung adenocarcinoma	30	37	8
Multiple myeloma	26	27	4
Glioma	22	15	6
Breast cancer (early onset)	18	13	2
Epithelial ovarian cancer	18	22	2
Thyroid cancer	18	17	5
Bladder cancer	15	22	7
Esophageal adenocarcinoma	14	22	2
Squamous cell carcinoma	14	17	3
Melanoma	14	17	9
Nasopharyngeal carcinoma	13	16	5
Endometrial cancer	12	14	4
Renal cell carcinoma	11	10	6
Cutaneous squamous cell carcinoma	11	15	1
Esophageal cancer	11	12	2
Ovarian cancer	10	13	5
Cervical cancer	9	12	5
Endometrial endometrioid carcinoma	6	10	1

fmSNPs linked to the cancer risk-associated SNPs

GWASs typically report a single most significant risk-associated SNP in the region. The reported most significant SNPs are not necessarily causal/functional variants. Functional and potentially causal variants can be linked to the reported SNPs but not reported because they may have happened to be less significant. Therefore, one needs to look for functional (potentially causal) SNPs among the SNPs linked to the reported most significant variant. We identified fmSNPs among SNPs linked to those reported in the catalog of published GWASs. As a first step, we identified SNPs located in the adjacent ±50 kb regions. We used a 50 kb region because it is about the size of an average linkage disequilibrium (LD) block in the human genome (26). For SNPs located in the human leukocyte antigen region we used ±100 kb adjacent region because LD blocks in the human leukocyte antigen region are larger (27). We obtained LDs between the GWAS-reported SNP and the SNPs from the adjacent region from the LDLink database. Pairwise LDs were assessed separately for five major ethnic groups: Africans, Mixed Americans, East Asians, Europeans and South Asians (28). SNPs with R² ≥ 0.8 in at least one group were considered to be proxy for the reported cancer risk-associated SNP. Among those proxies we have identified fmSNPs as candidate functional SNPs.

Estimates of selection pressure on mSNPs

Evolutionary conservation of the site is often used as a measure of functionality of the genetic variant (29). Genetic polymorphisms, e.g. SNP, located in a site with a signature of negative or positive selection are likely to be functional, whereas SNPs located in sites with no evidence of selection are likely to be neutral. We compared mSNPs and mutations that are not reported as SNPs (not-mSNP) by evolutionary conservation. We used PhyloP method to estimate strength and direction of natural selection on a given site (30). The PhyloP analyzes the distribution of nucleotide substitutions in an evolutionary tree of 44 vertebrate species. The method estimates the expected number of substitutions per site under the assumption of neutral evolution and compares them with the number of substitutions that have actually occurred in the site on the tree to generate likelihood score. Positive scores indicate slower-than-neutral evolution and negative ones—faster-than-neutral evolution of the site. We categorized not-mSNPs and mSNPs by the number of counts in COSMIC and estimated PhyloP scores for each count category.

In a separate analysis, we used Phylogenetic Analysis with Space/Time models (PHAST) to identify SNPs located in evolutionary conserved regions (31). PHAST uses multiple alignments of sequences from 100 vertebrate species to identify evolutionarily conserved regions. We estimated proportions of SNPs located in evolutionary conserved regions for mSNPs stratified by number of COSMIC counts and GWAS-detected associations with cancer risk.

Results

Majority of mSNPs are singletons

Figure 1 shows the distribution of mSNPs by the number of counts in COSMIC. One can see that the majority of mSNPs are singletons. There were >3 million singletons that comprise >45% of all mSNPs. The proportion of the mSNPs with two counts was 27%, and the proportion of the mSNPs with three counts was 10%. The proportions of SNPs with at least 10 COSMIC counts was <0.5%. The complete list of mSNPs categorized by the number of times they are reported in COSMIC with corresponding counts (number of cases) and their percentages can be found in Supplementary Table S1 (available at Carcinogenesis Online).

Figure 1.

The number of singletons and other number of counts categories among mutational SNPs. The absolute majority of mutational SNPs in COSMIC are singletons. mSNPs with at least 10 counts are shown as filled circles and with <10 counts as open circles.

Proportions of mSNPs predicted to be pathogenic by FATHMM in count categories

Figure 2 shows proportions of mutational SNPs classified as ‘pathogenic’ by FATHMM. mSNPs were categorized in 51 groups based on the number of counts in COSMIC with mSNPs with >50 counts combined in one group. One can see that the association between the number of counts and the proportion of mSNPs predicted to be pathogenic by FATHMM is not uniform across count categories. mSNPs with one to nine counts show positive linear association between the number of counts and the proportion of pathogenic mSNPs (orange dots in Figure 2). However, among mSNPs with at least 10 COSMIC counts there is no association between the number of counts and the proportion of mutations predicted to be pathogenic (blue dots in Figure 2).

Figure 2.

Proportions of the mSNPs predicted to be pathogenic by FATHMM. There is a positive linear association between the number of counts and the proportion of mutations predicted to be pathogenic in first nine categories (small open circles) (Pearson r = 0.96, n = 8, P = 0.00004). Categories with 10–50 counts (small filled circles) or more (a large filled circle) show no association between the number of counts and the proportion of pathogenic mutations (Pearson r = −0.06, n = 39, P = 0.72). Increasing variance after nine COSMIC counts is likely due to small sample sizes in those groups (see Figure 1).

Genes identified by linkage to the frequent mSNPs cluster in cancer-related pathways

For this analysis, we used mutation-linked genes provided by COSMIC annotation somatic mutations. We identified genes linked to the mSNPs from 5 count categories: singletons, mSNPs with 2–4 counts, mSNPs with 5–9 counts, mSNPs with 10–19 counts, and the mSNPs with >19 counts in COSMIC and then ran a pathway enrichment analysis. Table 2 reports 30 most significant pathways for each count category. Among 30 most significant pathways, we were interested to identify pathways directly related to cancer (shown in red in Table 2). No cancer pathways were identified in the singleton category or mSNPs with two to four counts in COSMIC. There was one cancer-related pathway in the group with five to nine counts. Four cancer-related pathways were identified in the group with 10–19 counts and 7 in the group with ≥20 counts. We also considered cancer-related pathways (shown in orange in Table 2) that do not mention cancer directly. The results of the analysis of the cancer-related pathways are consistent with the results of the analysis of pathways directly associated with cancer. Therefore, pathway enrichment analysis shows that genes linked to the mSNPs reported ≥10 times in COSMIC tend to cluster in cancer-related pathways.

Table 2.

Top pathways for the genes linked to mSNPs categorized by the number of counts in COSMIC

COSMIC singletons (7290)		2–4 COSMIC counts (8430)		5–9 COSMIC counts (5340)		10–19 COSMIC counts (2747)		>19 COSMIC counts (968)
KEGG pathway	P-value	KEGG pathway	P-value	KEGG pathway	P-value	KEGG pathway	P-value	KEGG pathway	P-value
Pancreatic secretion	0.00000058	Endocytosis	0.00011	Ascorbate and aldarate metabolism	0.00078	Adherens junction	0.00015	Drug metabolism—cytochrome P450	0.0000083
Endocytosis	0.000087	Pancreatic secretion	0.00033	Pancreatic secretion	0.00085	Fc gamma R-mediated phagocytosis	0.00035	Metabolism of xenobiotics by cytochrome P450	0.000023
Calcium signaling pathway	0.00057	Hematopoietic cell lineage	0.0027	Endocrine and other factor-regulated calcium reabsorption	0.0028	Pentose and glucuronate interconversions	0.00051	Pentose and glucuronate interconversions	0.000034
Gastric acid secretion	0.0015	Epstein-Barr virus infection	0.0036	Fc gamma R-mediated phagocytosis	0.0031	Pathways in cancer	0.0023	Ascorbate and aldarate metabolism	0.000048
Neuroactive ligand-receptor interaction	0.0022	Renin secretion	0.0068	Protein digestion and absorption	0.0041	Cell adhesion molecules	0.0026	Chemical carcinogenesis	0.000058
Endocrine and other factor-regulated calcium reabsorption	0.0037	Leukocyte transendothelial migration	0.0082	Pentose and glucuronate interconversions	0.0042	Endometrial cancer	0.0027	Endometrial cancer	0.00006
Leukocyte transendothelial migration	0.004	TNF signaling pathway	0.0086	Long-term depression	0.0044	Long-term depression	0.0052	Prostate cancer	0.00017
Carbohydrate digestion and absorption	0.0064	Synaptic vesicle cycle	0.0096	Cell adhesion molecules	0.0079	Serotonergic synapse	0.0055	Porphyrin and chlorophyl metabolism	0.00025
Thyroid hormone synthesis	0.0098	Proximal tubule bicarbonate reclamation	0.0099	Metabolism of xenobiotics by cytochrome P450	0.0087	Axon guidance	0.0056	Adherens junction	0.00027
Fc gamma R-mediated phagocytosis	0.01	Calcium signaling pathway	0.017	Inflammatory mediator regulation of TRP channels	0.0097	Proteoglycans in cancer	0.0068	Pathways in cancer	0.00032
Starch and sucrose metabolism	0.013	Endocrine and other factor-regulated calcium reabsorption	0.018	Endocytosis	0.016	Focal adhesion	0.007	Drug metabolism—other enzymes	0.00052
Protein digestion and absorption	0.015	Biosynthesis of antibiotics	0.02	Chemical carcinogenesis	0.017	ECM-receptor interaction	0.0072	Central carbon metabolism in cancer	0.0016
Glycerolipid metabolism	0.016	Gastric acid secretion	0.031	cAMP signaling pathway	0.019	Prostate cancer	0.0083	Retinol metabolism	0.0016
Synaptic vesicle cycle	0.018	Fc gamma R-mediated phagocytosis	0.036	Axon guidance	0.021	Endocrine and other factor-regulated calcium reabsorption	0.012	Proteoglycans in cancer	0.0017
TGF-beta signaling pathway	0.018	Circadian entrainment	0.041	Thyroid hormone signaling pathway	0.023	HIF-1 signaling pathway	0.012	Arrhythmogenic right ventricular cardiomyopathy	0.0023
cAMP signaling pathway	0.019	Cell adhesion molecules	0.044	Thyroid hormone synthesis	0.024	Morphine addiction	0.012	Thyroid hormone signaling pathway	0.0026
Cell adhesion molecules	0.02	Long-term depression	0.045	Insulin secretion	0.024	GnRH signaling pathway	0.012	Thyroid cancer	0.0034
Epstein-Barr virus infection	0.028	Glycerolipid metabolism	0.045	Bladder cancer	0.024	Ascorbate and aldarate metabolism	0.013	Colorectal cancer	0.0046
Phosphatidylinositol signaling system	0.028	Mineral absorption	0.049	Epstein-Barr virus infection	0.025	Arrhythmogenic right ventricular cardiomyopathy	0.016	Cell adhesion molecules	0.0071
Mucin type O-Glycan biosynthesis	0.033	Inflammatory mediator regulation of TRP channels	0.049	Histidine metabolism	0.026	Leukocyte transendothelial migration	0.017	Steroid hormone biosynthesis	0.01
Renin secretion	0.041	Adherens junction	0.053	TNF signaling pathway	0.027	Vibrio cholera infection	0.017	Melanoma	0.011
Aldosterone synthesis and secretion	0.044	Complement and coagulation cascades	0.054	Calcium signaling pathway	0.027	Biosynthesis of antibiotics	0.019	Calcium signaling pathway	0.012
Salivary secretion	0.046	Epithelial cell signaling in Helicobacter pylori infection	0.055	Purine metabolism	0.028	Fc epsilon RI signaling pathway	0.019	Bladder cancer	0.019
Signaling pathways regulating pluripotency of stem cells	0.047	p53 signaling pathway	0.055	Circadian entrainment	0.028	Cholinergic synapse	0.02	Hepatitis B	0.019
AMPK signaling pathway	0.049	Arrhythmogenic right ventricular cardiomyopathy	0.055	Salivary secretion	0.029	Type II diabetes mellitus	0.02	MAPK signaling pathway	0.024
Aldosterone-regulated sodium reabsorption	0.049	Protein digestion and absorption	0.055	Drug metabolism—cytochrome P450	0.029	Circadian entrainment	0.02	Oxytocin signaling pathway	0.025
Olfactory transduction	0.05	Salivary secretion	0.057	Adherens junction	0.029	Calcium signaling pathway	0.021	B cell receptor signaling pathway	0.027
Hippo signaling pathway	0.055	NF-kappa B signaling pathway	0.069	Inositol phosphate metabolism	0.039	Thyroid cancer	0.041	Cholinergic synapse	0.04

Number of genes used in pathway analysis is shown in parenthesis.

About 1% of cancer risk-associated SNPs are fmSNPs

Based on the results described in the two previous sections, mSNPs reported ≥10 times in COSMIC (fmSNPs) are considered to be functional. One can expect that fmSNPs are functional not only as mutations but also as germline polymorphisms SNPs. Therefore, we checked if fmSNPs are represented among GWAS-detected cancer risk-associated SNPs reported in the Catalog of published GWASs. A total of 12 fmSNPs were identified among cancer risk SNPs (Table 3). Taking into account that there are 1013 unique cancer risk-associated SNPs reported in the Catalog of published GWASs, ~1% of SNPs from there are potentially functional fmSNPs.

Table 3.

Frequent mutational SNPs reported in the Catalog of published GWASs as cancer risk associated

rs ID	COSMIC ID	position	Ref	Alt	Gene	SNP	Count	MAF	GWAS cancer type	PubMed ID	P-value	Odds ratio
rs10934853	87135561	3:128319530	C	A	EEFSEC	intronic	24	0.43	Prostate cancer	19767754	3.1E-10	1.12
rs10936599	89751629	3:169774313	C	T	MYNN	syn	12	0.22	Colorectal cancer	20972440	3E-8	1.04
rs1801516	88127386	11:108304735	G	A	ATM	nonsyn	15	0.09	Melanoma	21983787	3.3E-09	1.19
rs2274223	94964513	10:94306584	A	G	PLCE1	nonsyn	26	0.31	Esophageal cancer	21642993	4E-20	1.35; 1.34
rs2292884	88401131	2:237534583	A	G	MLPH	nonsyn	72	0.35	Prostate cancer	21743057	4E-8	1.14
rs3765524	94964492	10:94298541	C	T	PLCE1	nonsyn	24	0.32	Esophageal and gastric cancer	20729852	2E-9	1.35
rs3781264	94964531	10:94310618	A	G	PLCE1	intronic	18	0.25	Esophageal and gastric cancer	20729852	4E-9	1.36
rs5768709	103312035	22:48533757	A	G	FAM19A5	intronic	15	0.35	Pancreatic cancer	22158540	1E-10	1.25
rs6983267	151207513	8:127401060	G	T	CASC8; CCAT2	intronic; non-coding	16	0.37	Colorectal and prostate cancer	28960316; 26034056	2E-21; 3E-27	1.18; 1.25
rs8034191	106432881	15:78513681	T	C	HYKK	intronic	16	0.26	Lung cancer	19654303	3E-26	1.29
rs8100241	99852108	19:17282085	G	A	ANKLE1; USHBP1	nonsyn; intronic	11	0.46	Breast cancer	22976474	4E-8	1.14
rs9364554	86706928	6:160412632	C	T	SLC22A3	intronic	10	0.21	Prostate cancer	26034056	6E-12	1.14

SNPs linked to the reported cancer risk-associated SNPs

We have identified 54 fmSNPs linked to the SNPs reported to be cancer risk associated by the Catalog of published GWASs (Supplementary Table 3, available at Carcinogenesis Online). Together with 12 cancer risk SNPs identified earlier (Table 3), the total number of candidate functional SNPs is 66. Those candidate SNPs are mapped to 47 unique genes (see Supplementary Table 3, available at Carcinogenesis Online, for gene information).

Comparison of selective pressure on not-mSNPs and mSNPs

Positive PhyloP score for a given nucleotide position (site) indicates negative selection, meaning that the substitution rate for the site is lower compared with the substitution rate expected under neutral evolution. Negative PhyloP score is indicative of positive selection for a given site—the substitution rate for the site is higher compared with the substitution rate expected under neutral evolution. Figure 3 shows PhyloP scores for the somatic mutations categorized based on the number of COSMIC counts. mSNPs and not-mSNPs were analyzed separately. Mean PhyloP score for not-mSNPs are shown as blue dots and PhyloP scores for mSNPs are shown as orange dots. Overall PhyloP scores for not-mSNPs are higher compared with mSNPs, indicating that mutations that do not exist as polymorphisms are under stronger negative selection compared with the mutations that also exist as germline polymorphisms.

Figure 3.

PhyloP scores for somatic mutations stratified by the number of times they are reported in COSMIC. Somatic mutations not reported as SNPs (blue dots) and somatic mutations reported as SNPs (orange dots) were analyzed separately. Large dots show mean PhyloP scores for the somatic mutations reported >50 times in COSMIC. Dotted lines show polynomial regression and solid lines moving averages. (a) All count categories; (b) somatic mutations with 1–9 COSMIC counts; and (c) somatic mutations reported ten or more times in COSMIC.

Because the level of evolutionary conservation of the site reflects the strength of purifying selection one can expect that mutations frequently detected in COSMIC and, therefore, expected to be functional, will be preferentially located in evolutionary conserved sites. This is exactly what we observed for not-mSNPs (blue trend line in Figure 3). However for mSNPs (orange dots in Figure 3) the picture is more complicated. At the beginning—COSMIC counts from 1 to 9 we observed a positive correlation between number of counts and PhyloP score (Figure 3b). In this range mSNPS curve parallels the curve for not-mSNPs. Starting from the count 10, however, the curve for not-mSNPs continues to rise, whereas the curve for mSNPs becomes down-bound (Figure 3c). The downward trend for mSNPs indicates that functional somatic mutations that also exist as germline polymorphisms tend to be under positive selection.

fmSNPs are more likely to be located in evolutionary conserved regions than not-fmSNPs

We estimated the proportion of SNPs located in evolutionary conserved regions. We subdivided all SNPs in those reported to be cancer risk-associated by the Catalog of published GWASs and SNPs that are located in physical proximity to the cancer risk-associated SNP (±1000 nucleotides) but not reported as cancer risk-associated. We selected physically linked SNPs for comparison because they are expected to be similar in terms of nucleotide and gene content. Each SNP category was further stratified into fmSNPs and not-fmSNPs. SNPs not reported to be cancer risk associated and not reported as fmSNPs were used as a reference. The results of the analysis are shown in Figure 4. The proportion of SNPs located in evolutionary conserved regions in the reference group was 0.043 ± 0.001. Proportions of SNPs in evolutionary conserved regions were significantly higher for all other three categories. The highest proportion of SNPs in conserved regions (58.3 ± 7.1%) was observed among fmSNPs reported to be cancer risk associated.

Figure 4.

Proportions of SNPs located in evolutionary conserved regions among SNPs categorized based on being cancer risk associated and frequently reported as somatic mutation (fmSNP). Black cylinders indicate standard error of the proportion.

Discussion

Our analysis was based on two assumptions: (1) somatic mutations frequently detected in tumor samples are functional and (2) if a functional mutation exists as a SNP, it is also functional as SNP. Our data indicate that fmSNPs, that is, somatic mutations with at least 10 counts in COSMIC, are likely to be functional. A total of 8536 unique fmSNPs have been identified in the analysis. Twelve fmSNPs have been reported to be associated with cancer risk by GWASs. An additional 54 fmSNPs tightly linked to the cancer risk SNPs have been identified making the total number of potentially functional cancer risk-related SNPs equal to 66. Those SNPs represent only a small fraction of all identified fmSNPs, as there are 8536–66 = 8470 fmSNPs that are not reported as cancer risk-associated. We think that many of the remaining 8470 fmSNPs may be cancer relevant: they may be associated with cancer progression, survival and/or response to treatment. Below we provide our pilot analysis supporting this hypothesis. Our analysis identified 23 fmSNPs reported at least 100 times in COSMIC. These SNPs are linked to 17 genes (listed here according to the number of COSMIC counts): CACNA1C, CACNB2, SMIM4, FCRLA, IRF5, MDM4, PCDHGA11, YAP1, MADCAM1, CACNA1G, PDE9A, SPATA3, PCDHAC2, PCDHA10, HSD17B4, ERBB2 and MYBPC1. Ten of them have published evidence of an association with cancer progression, survival and/or response to treatment. For example, it has been demonstrated that somatic mutations in CACNA1C are associated with adverse prognosis of endometrial cancer (32). Loss of IRF5 expression in ductal carcinoma contributes to metastasis (33). It has been demonstrated that MDMF influences immune response in breast cancer (34). Dysregulation of PCDHGA11 is associated with progression of various cancers (35). Overexpression of YAP1 is associated with poor prognosis in breast cancer (36). MADCAM1 plays an important role in response to oxorubicin treatment (37). Inactivation of CACNA1G in colorectal cancer increases cell proliferation and suppresses apoptosis (38). It has been demonstrated that PDE9A suppression induces apoptosis of breast cancer cells (39). HSD17B4 has been shown to increase liver cancer progression (40). ERBB2 plays a critical role in the development and progression of various cancer types, especially breast cancer (41).

Our pilot analysis, therefore, demonstrates that fmSNPs are enriched by functional cancer-related SNPs. Frequent mutational SNPs can be used to identify causal variants among SNPs detected by GWASs as well as for targeted association analysis of cancer-related phenotypes. The hypothesis that fmSNPs are enriched by functional polymorphisms is further supported by the observation that fmSNPs more frequently located in evolutionary conserved regions than not-fmSNPs (Figure 4).

Comparative analysis of the frequency of fmSNPs and known driver mutations shows that fmSNPs are not as frequently detected in tumor samples as known driver mutations. The maximal count of mSNPs was 286 for mutation COSM3931613 or SNP ID rs201777030. Considering that these counts were detected among 40 550 tumor samples, the frequency of the most frequent mSNP is 0.7%. The median count of fmSNPs is 13, which transforms into the median frequency of fmSNPs of 0.03%. Known driver mutations are typically detected at the frequency of 2%, with some them being much more frequent (found in 50% of samples) (42). Therefore, the estimated frequencies of fmSNPs by two orders of magnitude are lower compared with the frequencies of known driver mutations. The difference in mutation frequencies between known drivers and fmSNPs may be related to the fact that mSNPs exist also as polymorphisms. It is known that genetic polymorphisms are mostly neutral or slightly deleterious (43). If a mutation is only slightly functional, genetic drift and other random factors may increase its population frequency because the pressure of negative selection is too weak to completely eliminate them from the population genetic pool.

Strong driver mutations do not exist as germline polymorphisms since they occur in the genes with important cellular functions (cell cycle, apoptosis) (44,45). To be able to exist as germline polymorphisms, genetic variants have to be only marginally functional or positively selected. We found that mSNPs reported at least 10 times as somatic mutations tend to be positively selected (Figure 3c). This observation suggests that genetic variants with relatively strong functional effects can exist as polymorphism only if they are positively selected. One can expect that positive selection will finally lead to fixation of the advantageous mutations but it will take some time during which the variant will exist as a polymorphism.

One of the advantages of using fmSNPs to identify cancer-related functional SNPs is that counts of fmSNPs are linkage independent. If, for example, we have several SNPs in the region that are in perfect LD and only one of them exists as a frequent somatic mutation it is an indication that this specific SNP is functional but not the others in LD with it. The drawback of fmSNP approach to identify functional SNPs relates to the fact that we are using the number of mutational counts as a proxy for functionality. Thus functionality in our analysis is narrowly defined: we are talking about functionality related to the selective advantage of tumor cells, e.g. a higher proliferation rate, or better survival. Other relevant functional variants that do not affect the behavior of tumor cells directly (e.g. influencing smoking behavior) cannot be identified using fmSNP approach.

The positive association between the number of times a somatic mutation is reported in COSMIC and the probability that the somatic mutation is functional is likely to be continuous: the more frequent the mutation, the more likely it is to be functional. Therefore, using a threshold for the number of times a mutation is reported in COSMIC to define a functional SNPs is a simplification. There is no doubt that there are functional SNPs among mutational SNPs that do not meet the frequency criterion to be considered fmSNPs. As the number of studies deposited in COSMIC increases (the repository is updated quarterly), the total number of mutations reported in COSMIC is expected to increase, which will require to modify (elevate) the threshold for functional significance. Therefore, the researchers wishing to use somatic mutations data as a guidance for the identification of potentially functional SNPs will need to consider the frequency of a given count category in the current COSMIC version.

In conclusion, by overlapping somatic mutations reported in COSMIC and SNPs reported in dbSNP, we have identified genetic variants of a dual nature: existing as somatic mutations and as population polymorphisms, termed mSNPs. We used the number of mutational counts in COSMIC as a proxy for cancer-relevant functionality: mutations reported more often were assumed to be more functional. We identified >8000 frequent mSNPs—those reported in COSMIC ≥10 times are considered to be likely functional polymorphisms based on the result of this study. Twelve of fmSNPs are reported as cancer risk associated by a GWAS. Additionally, we have identified 54 fmSNPs linked to the GWAS-detected SNPs. These SNPs are candidates for functional/potentially causal cancer risk-associated SNPs. fmSNPs can be used to identify causal SNPs associated with cancer risk, survival, progression and response to treatment.

Glossary

Abbreviations

COSMIC

Catalog of Somatic Mutations in Cancer

FATHMM

functional analysis through hidden Markov model

fmSNPs

frequent mutational/SNPs

GWASs

genome-wide association studies

LD

linkage disequilibrium

MAF

minor allele frequency

mSNPs

mutational/SNPs

SNPs

single-nucleotide polymorphisms

Funding

This work was supported in part by the National Institutes of Health U19 CA148127 and P01 CA206980-01A1. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Conflict of Interest Statement: None declared.