Genome-Wide Association Studies of Cancer

Abstract

Knowledge of the inherited risk for cancer is an important component of preventive oncology. In addition to well-established syndromes of cancer predisposition, much remains to be discovered about the genetic variation underlying susceptibility to common malignancies. Increased knowledge about the human genome and advances in genotyping technology have made possible genome-wide association studies (GWAS) of human diseases. These studies have identified many important regions of genetic variation associated with an increased risk for human traits and diseases including cancer. Understanding the principles, major findings, and limitations of GWAS is becoming increasingly important for oncologists as dissemination of genomic risk tests directly to consumers is already occurring through commercial companies. GWAS have contributed to our understanding of the genetic basis of cancer and will shed light on biologic pathways and possible new strategies for targeted prevention. To date, however, the clinical utility of GWAS-derived risk markers remains limited.

INTRODUCTION

Over the past two decades, research advances in cancer genetics have identified and characterized mutated cancer predisposition genes that account for a subset of cancers with a Mendelian pattern of inheritance.¹ Examples of cancer predisposition genes are BRCA1 and BRCA2 in hereditary breast and ovarian cancer, the mismatch repair genes in Lynch syndrome, the APC gene in familial adenomatous polyposis (FAP), and the TP53 gene in Li-Fraumeni syndrome. Individuals with mutations in these genes have a much higher risk of developing cancer than those in the general population. Guidelines for genetic testing for these genes and strategies for cancer surveillance and prevention have been developed and incorporated into oncologic practice.^2–5

In recent years, genome-wide association studies (GWAS) have identified genetic variants, or susceptibility loci, for a variety of human diseases, including cancer. As methods for sequencing entire personal genomes become less costly, the number of such genetic variants discovered will continue to increase. While the majority of susceptibility loci found by GWAS confer only a modest risk of disease (Fig 1), for-profit companies have made personal genomic profiles available directly to consumers. Although clinicians and scientists have articulated concerns regarding the premature commercial dissemination of personal genomics⁶ and need for continued research,⁷ oncologists and other cancer care professionals will be asked by patients to explain results of tests and advise on desired interventions or screening. A recent survey⁸ indicated that 42% of physicians were aware of direct-to-consumer genomic tests for disease risk, and of those physicians aware of direct-to-consumer tests, approximately 40% had been asked by patients to interpret the results of such testing. It is therefore important for oncologists to be aware of current data regarding the ranges of risk, the clinical validity (accuracy of a test in predicting the clinical outcome), and clinical utility (risks and benefits resulting from the use of a test) associated with genetic variants predisposing to cancer.^9,10

Fig 1.

Phenotypic effect size and frequency of occurrence. (*) Named genes only reflect the most likely candidate genes to be implicated by the marker single nucleotide polymorphisms (SNPs) identified from the genome-wide association studies. (†) The marker SNPs mapping to JAK2 in myeloproliferative neoplasms and KITLG in testicular germ cell tumors have odds ratios of approximately 3.0, with allele frequencies ranging from 20% to 40%.

GENETIC SUSCEPTIBILITY TO CANCER

Mutations of more than 50 genes have been associated with high-penetrance cancer susceptibility syndromes (risk of cancer increased approximately 5- to 50-fold). However, these syndromes account for only a small fraction of the familial risk of cancer (Fig 2).^11–16 Using breast cancer as an example, mutations in BRCA1 and BRCA2 appear to account for fewer than 20% of the familial risk of breast cancer, with other rare genes (eg, TP53, PTEN) accounting for no more than 5% of the risk.¹³ Other cancer types are similar in that high-penetrance cancer susceptibility genes only explain a small fraction of the familial risk of cancer.

Fig 2.

Familial risk of common cancers. Most common cancers, including breast, colorectal, and prostate cancer, exhibit familial aggregation, with the disease being more common in family members than in the general population. While familial aggregation may be secondary to genetic or environmental factors, evidence from monozygotic twins of patients suggests that genetic factors are mainly responsible.^11,12 As demonstrated in this figure, known high-penetrance cancer predisposition syndromes explain only a fraction of the familial risk of cancer.^13–16 While genome-wide association studies have identified numerous low-penetrance loci for each of these three common cancers, their contribution to the familial risk of cancer remains limited and a large portion of the familial risk of these cancers remains unexplained. GWAS, genome-wide association study; SNP, single nucleotide polymorphisms.

Studies of monozygotic twins and inbred populations have provided strong evidence that a large fraction of cancers are mediated by genetic susceptibility.^11,12 Surprisingly, in these studies, even seemingly environmental cancers, such as lung cancer, demonstrate familial clustering and are likely to be mediated by genetic susceptibility to shared exposures. A scientific debate over the past decade has centered on whether genetic susceptibility to common diseases is a result of the joint action of several common variants each with low relative risk of disease, or the result of genetic variants with low population frequency but moderate to high risk of disease.¹⁷

Examples of rare genetic mutations of moderate effect size emerged from studies of individual candidate genes. For example, genes in a common pathway (ie, DNA damage response), such as ATM^18–20, CHEK2^21–23, BRIP1, ²⁴ and PALB2,²⁵ have been associated with increased risk for breast cancer. However, based on estimates of the risk allele frequencies ranging from 0.1% to 0.5% (ie, 0.2 to one individual in 100 would carry at least one copy of the risk allele) and the modest 2.0-fold increase in relative risk associated with each of these genes, the contribution of mutations in these genes to the familial aggregation of breast cancer is limited. As shown in Figure 2, together with the known high-penetrance genes, all known breast cancer predisposition genes account for only approximately 25% of the familial risk of breast cancer in outbred populations.¹³

Recently, genetic association studies have been used to discover common genetic variants or risk alleles (minor allele frequency > 10%) with small to moderate (approximately two-fold) risks of cancer.²⁶ Association studies compare the frequency of a genetic variant in disease-affected patients and healthy controls. Recent knowledge gained from the Human Genome Project and the International Hap Map Project together with technical advances in high throughput genotyping technology has resulted in GWAS for a variety of complex diseases including cancer.

GWAS

Study Design

Roughly 99.9% of DNA sequence is identical across different individuals. Given the vast size (3.2 billion base pairs) of the human genome, even this small discrepancy results in millions of potential variations.²⁷ The most common variations are single-base pair changes called single-nucleotide polymorphisms (SNPs). GWAS compare allele frequencies between individuals with a disease (patients) to individuals without disease (controls). By determining which SNPs occur more (or less) frequently in individuals with disease, genomic regions associated with a disease state can be identified and a statistical estimate of the level of increased (or diminished) risk associated with each SNP can be made. GWAS take advantage of the fact that stretches of DNA tend to be inherited together and adjacent alleles sort together nonindependently from generation to generation. This nonrandom association of alleles at nearby loci (linkage disequilibrium) allows certain SNPs to serve as proxies, or tagSNPs, for other nearby SNPs. The use of such marker SNPs allows one to reduce the number of SNPs that need to be genotyped to characterize individual genomic variation to about 500,000.²⁸

The design of a typical GWAS for cancer susceptibility is shown in Figure 3. DNA from hundreds or thousands of patients and controls is analyzed (genotyped) using commercially available oligonucleotide microarray chips, which allow high throughput analysis of up to a million SNPs in one reaction. Even with per SNP genotyping costs of a fraction of 1 cent, the total cost for genotyping a large sample size using 500,000 or more SNPs can be prohibitive. Cost-effective multistage designs, such as the one shown in Figure 3, retain most of the power of the optimum design often at less than half the cost.^29,30 For study designs with three or more phases, the significant SNPs are included for replication testing in different patient-control sample sets.

Fig 3.

Genome-wide association study (GWAS) design. A typical cancer GWAS starts with the selection of a large number of individuals affected with a specific type of cancer and a suitable comparison group without cancer. DNA is isolated from each individual and genotyped using commercially available genome platforms (ie, chips) that assess for common genetic variations in the form of single nucleotide polymorphisms (SNPs) across the entire human genome. Data are reviewed to ensure appropriate genotyping quality. In phase I, stringent statistical methods are used to assess for associations between SNPs passing quality thresholds and cancer risk. In phase II, SNPs found to be significantly associated with cancer risk (approximately 5% to 10% of all SNPs) are tested in a larger set of independent patients and controls using arrays containing the putative SNPs of interest. Some studies proceed to further evaluation of putative SNPs in a third phase using larger and more heterogeneous populations to confirm the genotype associations.

Interpreting GWAS

Initiatives, such as the Strengthening the Reporting of Genetic Association Studies (STREGA),³¹ have provided guidelines for reporting of genetic association studies and helped standardize GWAS in order to avoid upwardly biased odds ratios (ORs).³² Interpretations of GWAS must take into account possible effects of disease and population heterogeneity, involvement of multiple genetic and environmental factors, and possible gene-gene and gene-environment interactions.

Because power is a function of sample size, minor allele frequency, and the presumed genetic effect size, the detection of modest genetic effects with ORs of 1.3 or lower and minor allele frequencies under 10%, may require more than 10,000 patients and 10,000 controls for adequate statistical power.³³ Although large study consortia may approach this size, such efforts are often limited by cost. In addition to ensuring adequate statistical power, patients and controls must be well-matched to avoid population stratification, wherein misleading associations may result from differences in race or ethnicity of those with and without the disease of interest.^34–37 Family-based designs can minimize population stratification bias, but also compromise the power of associations.³⁸ As population stratification can lead to significant type I error rates (ie, false positives), principal component analysis^39,40 and other methods^41,42 have been developed that estimate the population ancestry of patients and controls based on genotypes of a panel of SNPs that are not associated with the disease of interest. Misleading associations may also result from genetic heterogeneity, wherein what appears to be a single disease is in fact a result of genetically separate phenomenon. For example, given the known phenotypic variation in breast cancer, combining all breast cancers into one group may obscure potential genetic associations.

Given multiple comparisons, stringent statistical thresholds are necessary to avoid spurious false-positive associations. The most widely used adjustment for multiple comparisons is the Bonferroni correction, in which the threshold P value (usually 5 × 10⁻²) is divided by the number of tests performed (approximately 500,000 depending on array used) resulting in an acceptable P value on the order of lower than 1 × 10⁻⁷. Even with statistically significant associations, positive results have to be interpreted with caution owing to “winner's curse,” wherein an artificial inflation in measures of association can result from the discovered loci having to pass through stringent significance thresholds.^43,44 Regardless of study design and statistical considerations, GWAS are prone to publication bias, as strongly positive associations are more likely to be reported than negative studies.

Interpretation and Clinical Relevance of Results

Genetic associations from GWAS are reported either as relative risks (RR) for cohort studies or as ORs for case-control studies. To date, the majority of cancer GWAS have found risk alleles with only modest associations with cancer risk (OR, 1.1 to 1.5). For example, a risk allele in the fibroblast growth factor receptor 2 (FGR2) gene has been found to be associated with a 1.26-fold increased risk of breast cancer, comparable in magnitude to the modest increased risk of breast cancer conferred by delaying first pregnancy to age older than 35.⁴⁵ However, a 1.26 risk may be reported to a patient by stating “your risk is increased 26%,” which may seem high, but is far less than the 100-fold increases in breast cancer risk associated with, for example, BRCA mutations. In addition to being misleading, the use of ORs also generally relate to risks over a lifetime, and not over a specific period of time. The very low level of risk associated with most variants found by GWAS remains the greatest barrier to the clinical application of these markers in cancer prevention.

It may also be possible to combine a set of risk alleles into a genomic risk profile for risk estimation of a particular cancer. In a multiplicative model, one simply multiplies one risk factor by the other (eg, 1.2 × 1.5 = 1.8) to derive the combined risk. Such genomic risk panels can then be compared to models that incorporate typical clinical variables (eg, Gail model for breast cancer risk). To date, such comparisons have revealed a very modest, if any, added value to genomic risk profiles.^46–48 In addition, more research is needed regarding epistasis (gene-gene interactions) that may confound results of genomic risk profiles. It is also possible that as more variants in the same pathway are discovered, there will be supra-multiplicative interactions that may lead to much higher relative risks. To date such supra-multiplicative interactions have not been found.

It may be common for individuals receiving commercial testing for cancer risk to be told that a certain genetic variant they were found to have is associated with a very high proportion of a type of cancer. The reporting of this type of population attributable risk % (PAR%) of a SNP may be misleading.⁴⁹ Because of the high allele frequencies of most susceptibility loci identified, the PAR% can be quite high even with very modest elevations in relative risk. In fact, the joint PAR% of common risk variants in some cancers approaches (and may exceed) 100%. These same risk loci, however, may explain only a small fraction of the excess familial relative risk of cancer.¹⁴ For example, for the dozen or so SNPs associated with breast cancer risk, the joint PAR% is higher than 70% while the risk loci explain only approximately 8% of the familial relative risk of breast cancer. A large PAR% does not exclude the possibility that other risk alleles exist, that multiple risk factors may occur in the same individual, or that environmental factors may also contribute to disease development.

When clinicians interpret results of GWAS SNPs, it is also important to note that many of the low-risk variants derived by GWAS and used in some commercial panels have not been replicated; a systematic review of 260 meta-analyses of 160 polymorphism gene associations concluded that there is insufficient scientific evidence to conclude that genomic profiles are useful in measuring genetic risk for common diseases or in developing personalized recommendations for disease prevention.⁵⁰ Finally, since the frequency of risk alleles and the incidence of types of cancer vary within populations, these values may not be generalizable and prospective studies are necessary to directly measure absolute risk, and to judge the accuracy of risk estimates calculated using retrospective methods.

GWAS OF CANCER

In recent years, more than 50 cancer GWAS have been published incorporating at least 15 different malignancies. Nearly all the cancer susceptibility loci identified to date are associated with modest increases in disease risk, with ORs generally below 1.5. Exceptions to this are the risk variants identified in JAK2 in myeloproliferative neoplasms and in KITLG in testicular cancer which are each associated with nearly a three-fold increased risk of disease.^51–53 The contribution of the identified susceptibility loci in explaining the genetic basis (heritability) of cancer and the potential clinical implications of GWAS findings is illustrated below in three common malignancies.

Breast Cancer

Breast cancer has been at the forefront of cancer GWAS with at least 13 independent loci implicated in disease risk (Table 2).^45,54–81 Of identified susceptibility loci, the most strongly associated risk SNP, with an OR of 1.26, was in FGFR2. The protein encoded by FGFR2 is a member of the FGFR family and is overexpressed in 5% to 10% of breast tumors.^82,83 While the precise mechanism(s) of FGFR2 deregulation in breast cancer etiology remains unknown, fine mapping of the region suggests that the causative variants lie in intron 2 of FGFR2. The 10q26 locus mapping to FGFR2 was implicated in a number of breast cancer GWAS using different patient populations and appears to be strongest in estrogen receptor–positive breast cancers.^84–86 Some of the other susceptibility loci identified in gene-containing regions have not been implicated in cancer previously (eg, TOX3, LSP1, STXBP4) and are being evaluated for their potential role in carcinogenesis. Other SNPs lie in regions devoid of genes (ie, the 8q24 region) where research into their impact on near-by genes is under investigation. As most GWAS-based associations have correlated with estrogen receptor–positive breast cancers, efforts to identify risk SNPs predicting for estrogen receptor–negative or triple-negative breast cancers are ongoing.

Table 2.

Genome-Wide Association Studies for Breast, Prostate and Colorectal Cancers

Locus	Implicated Gene	SNP	Per Allele OR Ranges*	Reference
Breast
1p11.2	Pericentric	rs11249433	1.16	54
2q35	Intergenic	rs13387042	1.2-1.25	54,55
3p24.1	SLC4A7	rs4973768	1.11	56
5p12	Intergenic (MRPS30)	rs4415084	1.16-1.19	57
		rs10941679
5q11.2	MAP3K1, MIER3, C5orf35	rs889312	1.13	45
6q22.33	ECHDC1, RNF146	rs2180341	1.41	58
6q25.1	ESR1	rs2046210	1.29	59
8q24	Intergenic	rs13281615	1.08	45
10q26	FGFR2	rs2981582	1.20-1.29	45
		rs1219648		57,60
		rs1078806		58
11p15.5	LSP1	rs3817198	1.07	45
14q24.1	RAD51L1	rs999737	1.06	54
16q12	TNRC9 (TOX3), LOC643714	rs3803662	1.16-1.28	45,54,55
17q23	STXBP4	rs6504950	1.05	56
Prostate
2p15	EHBP1	rs721048	1.15	61
2p21	THADA	rs1465618	1.08	62
2q31	ITGA6	rs12621278	1.33	62
3p12	Intergenic	rs2660753	1.18	64
3q21	Intergenic	rs10934853	1.12	63
4q22	PDLIM5	rs17021918	1.11	62
		rs12500426	1.08
4q24	TET2	rs7679673	1.10	62
6q25	SLC22A3	rs9364554	1.17	64
7p15	Intergenic	rs12155172	1.05	62
7p15.2-15.1	JAZF1	rs10486567	1.12-1.35	65
7q21.3	LMTK2	rs6465657	1.12	64
8p21	NKX3-1	rs2928679	1.05	62
		rs1512268	1.18
8q24	Intergenic	HapC 14 SNPs	2.10	66
		rs16901979	1.79-1.80	63,66
		DG8S737	1.64	67
		rs1447295	1.36-1.60	63,66–69
		rs1016343	1.37	64
		rs6983267	1.26-1.42	64,65,69
		rs4242382	1.41-1.87	64,65
		rs1006908	1.15	70
		rs620861	1.17	71
		rs16902094	1.14	63
10q11.2	MSMB	rs10993994	1.16-1.25	64,65
10q26.13	CTBP2	rs4962416	1.17-1.20	65
11p15	IGF2, IGF2AS, INS, TH	rs7127900	1.22	62
11q13.2	Intergenic	rs10896449	1.10-1.28	65
		rs7931342	1.19	64
		rs11228565	1.23	63
17q12	TCF2 (HNF1B)	rs4430796	1.18-1.38	63,65,72
		rs7501939	1.41	64
17q24.3	Intergenic	rs1859962	1.20-1.26	64,72
19q13.2	PPP1R14A	rs8102476	1.12	63
19q13.41	KLK2, KLK3	rs2735839	1.20	64
22q13	TTLL1, BIK, MCAT, PACSIN2	rs5759167	1.20	62
22q13	TNRC6B	rs9623117	1.18	73
Xp11.23-p11.22	NUDT10, NUDT11,	rs5945619	1.19	64
	LOC340602, GSPT2, MAGED1	rs5945572	1.23	61
Colorectal
8q23.3	EIF3H	rs16892766	1.25	74
8q24.21	LOC727677, POU5F1P1	rs10505477	1.17	75
Intergenic	rs6983267	1.17-1.27	74,76
	rs7014346	1.19	78
10p14	Intergenic	rs10795668	1.11	74
11q23	Intergenic	rs3802842	1.12	78
14q22-q23	BMP4	rs4444235	1.11	79
15q13	Intergenic	rs4779584	1.23-1.26	74,80
	GREM1	rs10318	1.19	80
16q22.11	CDH1	rs9929218	1.10	79
18q21.1	SMAD7	rs4939827	1.16-1.20	78,81
19q13.11	RHPN2	rs10411210	1.15	79
20p12.3	Intergenic	rs961253	1.12	79

Abbreviations: SNP, single nucleotide polymorphism; OR, odds ratio.

^*ORs < 1 in the original publication have been converted to ORs > 1 for the alternate allele.

Table 1.

Definition of Terms

Term	Definition
High-penetrance cancer susceptibility syndrome	A cancer predisposition syndrome wherein a mutation in the implicated gene produces a phenotype in a high proportion of individuals who carry the mutation
SNP	A variation in DNA sequence where a single nucleotide is replaced by another; SNPs are thought to represent the most common form of genetic variation in the genome
Clinical validity	The accuracy with which a genetic test can identify or predict the presence or absence of a particular clinical condition taking into account the specificity, sensitivity as well as the penetrance of the genetic variation
Clinical utility	The degree to which the use of a test informs clinical decision making and leads to improved health outcomes
Candidate gene studies	A study that identifies genetic associations by assessing genetic variants that are suspected of being involved in the expression of a particular trait or disease
GWAS	A systematic hypothesis-free search for genetic variations, in the form of SNPs, across the genome to identify genetic associations with a disease or trait
Minor allele frequency	Frequency of the less common allele of a polymorphic locus
Risk allele	Any one of several variants of a gene that occupy the same position locus on a chromosome and is associated with risk of a particular trait or disease
Linkage disequilibrium	The non-random association of alleles of different polymorphisms in a population
Population stratification	Genetic differences between cases and controls not due to the trait or disease being studied but rather due to sampling of populations with different ethnicities or ancestries; population stratification can lead to erroneous genetic associations
Genetic heterogeneity	Multiple genetic mutations can result in the same disease phenotype
Bonferroni correction	A multiple-comparison correction used when several statistical tests are being performed simultaneously; in GWAS, using the Bonferroni correction helps to avoid spurious genetic associations
Winner's curse	Results of GWAS may be subject to varying degrees of upward bias in effect size estimates due to having to pass through stringent statistical thresholds
Epistasis	Gene-to-gene interactions where the effects of one gene are modified by one or several other genes
PAR%	The reduction in incidence of a particular disease that would be observed if the population was entirely unexposed, compared with its actual exposure; in genetic epidemiology, combining knowledge of risk allele frequency and genotypic relative risk, the attributable fraction of cases that would not occur if no one in the population had the risk allele can be determined
FRR	The ratio of disease risk in biological relatives of affected individuals compared with disease risk in the general population; in general, the higher the FRR, the stronger the genetic effect
Publication bias	A type of reporting bias wherein statistically significant or positive results are more likely to be published

Abbreviations: SNP, single nucleotide polymorphism; GWAS, genome-wide association studies; PAR, population attributable risk; FRR, familial relative risk.

Importantly, as compared to the high-penetrance breast cancer susceptibility genes, the magnitude of risk associated with each of the risk SNPs identified in breast cancer GWAS is modest, with ORs largely ranging from 1.1 to 1.4 (Fig 1). The contribution of these loci to the familial risk of breast cancer is no more than approximately 8%¹⁴ thereby still leaving the majority of the familial risk of breast cancer unexplained. Modeling studies have predicted that together the seven most common breast cancer–associated SNPs would add little in terms of improved discriminatory accuracy when compared to, or when used in conjunction with, a standard clinical breast cancer risk model (eg, the Gail model).^46,47 The addition of information on 10 breast cancer risk SNPs to the Gail model predicted the risk of breast cancer only slightly better than the clinical model alone.⁴⁸ Similarly, in a cohort of BRCA mutation–negative women with and without breast cancer, incorporation of a genotypic risk score had limited discriminatory accuracy; however, the potential for reclassification of a clinically relevant proportion of women for altered recommendations for chemoprevention or magnetic resonance imaging screening has not been excluded.⁸⁷ While these clinical studies demonstrate the potential use of SNPs in cancer risk prediction, to date, breast cancer risk stratification based on SNPs remains premature.

Colon Cancer

Although familial susceptibility accounts for as much as 35% of colorectal cancer (CRC)¹⁷, only approximately 6% of all CRCs occur in the setting of a known genetic predisposition syndrome. Hereditary nonpolyposis colorectal cancer (Lynch syndrome) and FAP account for the majority of cases, while rare inherited syndromes, such as Peutz-Jeghers syndrome, juvenile polyposis, attenuated FAP, and MYH-associated polyposis, explain only 1% of CRCs.

The seven GWAS in CRC have identified 10 susceptibility loci (Table 2). Multiple CRC GWAS identified the 8q24 locus, containing the rs6983267 SNP, with an associated approximately 1.2-fold increased risk of disease.^74–76,78 This SNP was also associated with adenoma risk with an OR of 1.16.⁸⁸ While the 8q24 region is devoid of known genes, two recent publications suggest that the rs6983267 SNP may be connected to enhanced Wnt signaling and subsequent MYC regulation, known pathways in carcinogenesis.^89,90

Nearly half of the susceptibility loci in CRC are in linkage disequilibrium or are nearby genes of the transforming growth factor beta (TGF-ß) signaling pathway previously implicated in carcinogenesis.^91,92 Increased TGF-ß1 expression has been linked to tumor progression and recurrence in CRC and germline mutations in components of the TGF-ß signaling pathway, namely SMAD4 and BMPR1A, are responsible for the high-penetrance juvenile polyposis syndrome. Genes implicated from CRC GWAS along the TGF-ß pathway include: SMAD7, RHPN2, BMP4, BMP2, and GREM1.

Overall, the 10 risk loci identified are associated with modest 1.1- to 1.25-fold increases in the relative risk of CRC and account for only approximately 6% of the excess familial risk of CRC.⁷⁹ There is currently no evidence that individual SNPs or panels of SNPs add to the discriminatory accuracy of current clinical risk criteria based on age, personal and family history of adenomas or CRC, and pre-existing inflammatory bowel disease. Nor is there convincing evidence that these SNPs correlate with survival, early-age at onset, site of tumor, or a histologically more aggressive subset of disease.^79,93 By way of comparison, the relative risk for CRC for an individual carrying the 8q24 risk SNP is approximately 1.2- versus a 1.8-fold increased risk for the first-degree relatives of individuals with an adenoma⁹⁴ and a 2.5-fold increased risk for individuals with a first-degree relative with CRC.⁹⁵ At this time, recommendations for CRC screening would not be altered from that of the general population based solely on the presence of a CRC-associated risk allele.

Prostate Cancer

There is strong evidence for genetic predisposition to prostate cancer from family studies, including a two- to three-fold increased risk of disease in first-degree relatives of affected men.^96,97 Germline mutations in genes such as BRCA2 have been found to be associated with prostate cancer risk, however, such mutations explain less than 10% of the familial risk of prostate cancer.¹⁵

GWAS have identified more than a dozen prostate cancer risk loci with ORs mostly ranging from 1.2 to 2.0 (Table 2). As in CRC, bladder, and breast cancer, a number of prostate cancer GWAS identified risk SNPs mapping to the 8q24 locus. In a multiethnic study, Haiman et al⁷⁷ identified seven independent risk variants in the 8q24 region and, interestingly, observed that the risk variants were most common in the African-American population, possibly suggesting a partial explanation for the higher incidence of prostate cancer in African-American men. Risk SNPs at or near genes (ie, MSMB, KLK3 and KLK2) with plausible roles in prostate carcinogenesis have also been found.^64,65

The joint contribution of identified loci to the familial risk of prostate cancer approaches 20%. However, the contribution of the risk SNPs in improving the discriminatory accuracy of clinical models (eg, those based on prostate-specific antigen [PSA]) is likely to be modest, although in at least in one instance a kallikrein-linked SNP is highly correlated with serum levels of proteins associated with prostate cancer risk.⁹⁸ Studies correlating individual risk SNPs to more aggressive prostate cancer subsets have largely been unrevealing^69,99–101 and the risk SNPs have not been associated with survival after diagnosis.^102,103 In addition, a family history of prostate cancer still confers a greater risk than the presence of any individual risk allele.^101–103

The impact of carrying multiple risk SNPs on prostate cancer risk was assessed by Zheng et al¹⁰² with results demonstrating that men who carried ≥ four of five possible risk SNPs had a 4.5-fold increased risk of disease. Importantly, there was no evidence that the risk SNPs were associated with disease aggressiveness, earlier age at diagnosis or presence or absence of family history. A subsequent analysis demonstrated that these five risk alleles do not improve prediction models for disease risk or disease-specific mortality once known risk factors (ie, age, PSA, family history) or prognostic factors (ie, Gleason score, diagnostic PSA, stage, age, primary treatment) are taken into account.¹⁰³ To date, the validity and clinical utility of using individual or multiple SNP panels as a screening test for prostate cancer have not been demonstrated. Additional studies seeking to identify genetic variants that predict for early-onset or more aggressive disease and may be used for risk stratification are in progress.

Other Malignancies

As presented in Table 3,^{51–53,104–132} dozens of other SNPs have been associated with an increased risk of cancers of the lung, skin, thyroid, ovaries, pancreas, and other sites. With the exception of SNPs for testicular germ cell tumors and myeloproliferative neoplasms, none of these SNPs has been found to have a risk greater than 2.0.

Table 3.

Genome-Wide Association Studies for Other Cancers

Locus (entrez gene)	Implicated Gene	SNP	Per Allele OR Ranges*	Reference
Lung
5pter-p15.33	CLPTM1L, TERT	rs401681	1.16	104
		rs402710	1.18	105
		rs2736100	1.14	105
			1.23 (adenocarcinoma)	106
5p15.33		rs4975616	1.16	107
6p21.33-p21.3	BAT3, MSH5	rs3117582	1.24	104,107
15q24-q25	CHRNA3, CHRNA5, CHRNB4,	rs8034191	1.30-1.38	108–110
	PSMA4, LOC123688
		rs1051730	1.31-1.35	105,108,110
		rs8042374	1.33, NR	104,107
15q25.1	CHRNA3	rs938682	1.33	107
		rs12914385	1.29	107
BCC and CM
1p36.13	PADI4, PADI6, RCC2, AHRGEF10L	rs7538876	1.28 BCC	111
1q42.11-q42.3	RHOU	rs801114	1.28 BCC	111
7q32	KLF14	rs157935	1.23 BCC	112
9p21	MTAP, CDKN2A	rs7023329	1.18 CM	113
9p21	CDKN2A, CDKN2B	rs2151280	1.19 BCC	112
9p23	TYRP1	rs1408799	1.15 CM	114
11q14-q21	TYR	rs1126809	1.21 CM	114
			1.14 BCC
		rs1393350	1.29 CM	113
12q12-q13	KRT5	rs11170164	1.35 BCC	112
16q24	MC1R	rs258322	1.67 CM	113
20q11.2-q12	ASIP	Hap rs1015362G and rs4911414T	1.45 CM	114
			1.35 BCC
20q11.22	CDC91L1 (PIGU)	rs910873 and rs1885120	1.75 CM	115
Urinary bladder
3q28	TP63	rs710521	1.19	116
8q24.21	MYC, BC042052	rs9642880	1.22	116
8q24.2	PSCA	rs2294008	1.15	117
Neuroblastoma
2q35	BARD1	rs3768716	1.68	118
		rs6435862	1.68
6p22.3	FLJ22536, FLJ44180	rs4712653	1.35	119
		rs9295536	1.32
		rs6939340	1.37
Glioma
5p15.33	TERT	rs2736100	1.27	120
8q24.21	CCDC26	rs4295627	1.36	120
9p21.3	CDKN2B	rs1412829	1.42	121
		rs4977756	1.24	120
11q23.3	PHLDB1	rs498872	1.18	120
20q13.33	RTEL1	rs6010620	1.51	121
			1.28	120
Acute lymphoblastic leukemia (childhood)
7p12.2	IKZF1	rs4132601	1.69	122
10q21.2	ARID5B	rs10994982	1.62	123
Maps to 10q11.22 by HGNC		rs10821936	1.91	123
	rs7089424	1.65	122
14q11.2	CEBPE	rs2239633	1.34	122
Chronic lymphocytic leukemia
2q13	ACOXL, BCL2L11	rs17483466	1.39	124
2q37.1	SP140	rs13397985	1.41	124
2q37.3	FARP2	rs757978	1.39	125
6p25-p23	IRF4	rs872071	1.54	124
8q24.21	Intergenic	rs2456449	1.26	125
11q24.1	GRAMD1B	rs735665	1.45	124
15q21.3	NEDD4, RFX7	rs7169431	1.36	125
15q23	Intergenic	rs7176508	1.37	124
16q24.1	IRF8	rs305061	1.22	125
19q13.2-q13.3	PRKD2, STRN4	rs11083846	1.35	124
Follicular lymphoma
6p21.33	STG, PSORS1	rs6457327	1.69	126
Thyroid (papillary and follicular)
9q22.33	Intergenic	rs965513	1.75	127
14q13.3	Intergenic	rs944289	1.37	127
Myeloproliferative neoplasms
9p24.1	JAK2	rs10974944	3.10	52
Testicular germ cell cancer
4q24	Intergenic	rs4699052	1.21	53
5q31.3	SPRY4	rs4324715	1.37	51
		rs4624820	1.37	53
		rs6897876	1.39	51
6p21.3	BAK1	rs210138	1.50	53
12q22	KITLG	rs995030	2.55	53
		rs3782179	3.08	51
		rs4474514	3.07	51
		rs1508595	2.69	53
Pancreatic
1q32.1	NR5A2	rs3790844	1.30	127a
5p15.33	TERT-CLPTM1L	rs401681	1.19	127a
9q34	ABO	rs505922	1.20	128
13q22.1	Intergenic	rs9543325	1.26	127a
Ovarian
9p22	BNC2, CNTLN, LOC648570	rs3814113	1.22	129
Gastric (diffuse)
8q24.3	PSCA	rs2976392	1.62 (Japan)	130
			1.90 (Korea)
Esophageal (squamous cell)
4q21-23	ADH1B	rs1229984	1.79	131
12q24	ALDH2	rs671	1.67	131
Single locus 5p15.33 Multiple cancers	TERT-CLPTM1L	rs401681		132
Prostate			1.07
Lung			1.15
BCC			1.25
Urinary bladder			1.12
Cervical			1.31

Abbreviations: OR, odds ratio; NR, not reported; BCC, basal cell carcinoma; CM, cutaneous melanoma.

^*ORs < 1 in the original publications have been converted to ORs > 1 for the alternate allele.

CHALLENGES AND FUTURE DIRECTIONS

Identification of Functional Gene Variants and Targeted Prevention

Perhaps one of the most important advantages of GWAS is the use of an agnostic approach wherein genes previously not implicated in cancer susceptibility may be identified. For example, genes such as FGFR2, overexpressed in 5% to 10% of breast tumors, and MSMB, a gene coding for PSP94, an immunoglobulin binding factor synthesized by prostate epithelial cells, seem particularly plausible breast and prostate candidate cancer susceptibility genes, respectively. Variants in genes previously linked to cancer pathogenesis, such as KITLG in testicular cancer, may also help to explain cancer predisposition. A challenge to date has been that the majority of risk SNPs implicated in cancer susceptibility have not been associated with functional changes in the genes residing near the loci of these SNPs. For some of the identified loci, fine-scale genetic mapping and deep sequencing of the implicated regions is ongoing and will hopefully lead to identification of the causal genetic change. For loci that map to genes or nearby genes, functional analyses will be crucial in helping to tailor targeted preventive approaches, such as the potential use of tyrosine kinase inhibitors to prevent breast cancer in women with FGFR2 pathway deregulation.

Incorporation into Prospective Trials

Before incorporating risk variants into individualized cancer risk assessment, the hypothesis-generating GWAS results should be validated in prospective studies in heterogeneous populations to demonstrate the efficacy of these variants in predicting an individual's risk of disease and associated disease outcome. The importance of prospective clinical studies cannot be underestimated for several reasons. First, prospective studies can take into account environmental factors, and can also identify associations with disease-specific mortality or more aggressive clinical phenotypes. Second, prospective studies allow better estimates of absolute risk, sensitivity, specificity as well as positive and negative predictive value that take into account the specific cancer incidences in the target population. These measures are more clinically useful than ORs from retrospective studies. Other measures may include derivation of the discriminatory accuracy of genomic tests using receiver operating characteristic curves that plot graphically the true-positive rate (sensitivity) against the false-positive rate (1 − specificity). Such measures have only recently been applied to GWAS-derived risk markers,^46–48,133 and analysis from initial receiver operating characteristic curves indicate that even with strong genetic associations, effective discrimination between patients and controls is not guaranteed.¹³³ In addition, using methods of calibration testing, panels of risk variants should be compared to the predictive power of already existing cancer risk models that incorporate a variety of clinical variables, such as the predictive models available for breast and prostate cancer and the more recently developed CRC risk model for those older than 50 years. Finally, because experience with communication of high-penetrance cancer risk information has shown that people have difficulty understanding probabilistic information^134,135 and tend to persist in inaccurate estimates of their risk even in the context of specialized counseling,^136–139 the appropriate delivery mode of genetic risk information to individuals must be determined.

Current Challenges

At present, the modest increased risks of cancer associated with the known genetic variants are, for the most part, not medically actionable. There is a lack of data to justify use of individual SNPs or panels of risk variants as independent risk predictors over known clinical variables. Increased population screening based on genomic risk profiles may be harmful and lead to increased cancer screening tests associated with significant false-positive rates (eg, breast magnetic resonance imaging, serum PSA), to invasive tests (eg, colonoscopy), to unnecessary subsequent medical interventions (eg, biopsy), and to overall increased medical costs. Results of genomic risk profiles may also provide false reassurance to individuals who may, in fact, be at high-risk for a particular cancer based on other clinical variables.

Despite the paucity of evidence supporting the use of GWAS results for clinically useful cancer risk prediction, personal genomic testing is currently being offered directly to consumers outside the context of the health care system. Such genome scans may include cancer-specific scans for breast, prostate, colorectal, thyroid, skin, lung and urinary bladder cancer risk. Patients who obtain such tests through nontraditional channels, may nonetheless expect that their physicians will assist them with the interpretations of such data.^140,141 Indeed, physicians, including oncologists, have already been asked to explain results of genomic profiles to their patients. Despite the paucity of evidence for clinical utility, 52% to 75% of physicians in an initial survey felt that genomic tests had the potential to impact clinical management.⁸ In order to inform both patients and physicians, there has been a call for increased regulation or guidance with respect to the provision of such tests. However, such intervention has been limited as many genetic and genomic tests fall outside of the purview of the US Food and Drug Administration and other regulatory agencies.⁹ Such regulations will clearly be required for pharmacogenomic applications of inherited genetic variants which predict response and toxicity of preventive and therapeutic pharmacologic interventions. Recognizing the scientific merits as well as the limitations of recent discoveries of genetic variants, a recent update by the American Society of Clinical Oncology emphasizes the importance of further research to demonstrate the validity and clinical utility of genomic profiles.¹⁴²

Future Perspectives

GWAS are powerful tools that have enhanced our understanding of cancer genetics and will inevitably lead to the identification of novel pathways of carcinogenesis. Research regarding other forms of genetic variation in the human genome, including genomic structural changes and sequence variation, is just beginning to emerge and will provide further insight into the genetic basis of complex diseases. As our knowledge of the human genome continues to rapidly expand, health care providers must also be aware of the evidence base required for genomic profiles of cancer risk to be effectively incorporated into the practice of preventive oncology.

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

Although all authors completed the disclosure declaration, the following author(s) indicated a financial or other interest that is relevant to the subject matter under consideration in this article. Certain relationships marked with a “U” are those for which no compensation was received; those relationships marked with a “C” were compensated. For a detailed description of the disclosure categories, or for more information about ASCO's conflict of interest policy, please refer to the Author Disclosure Declaration and the Disclosures of Potential Conflicts of Interest section in Information for Contributors.

Employment or Leadership Position: None Consultant or Advisory Role: Mark E. Robson, Pfizer (C); Noah D. Kauff, Wyeth (C) Stock Ownership: Bert Gold, GlaxoSmithKline, Quest Diagnostics Honoraria: Karen E. Hurley, Genzyme Research Funding: Mark E. Robson, Kudos Pharmaceuticals; Robert J. Klein, sanofi-aventis Expert Testimony: Noah D. Kauff, Wyeth (C) Other Remuneration: None

AUTHOR CONTRIBUTIONS

Conception and design: Zsofia K. Stadler, Peter Thom, Kenneth Offit

Financial support: Kenneth Offit

Administrative support: Kenneth Offit

Collection and assembly of data: Zsofia K. Stadler, Peter Thom, Vincent Devlin

Data analysis and interpretation: Zsofia K. Stadler, Peter Thom, Kenneth Offit

Manuscript writing: Zsofia K. Stadler, Peter Thom, Mark E. Robson, Jeffrey N. Weitzel, Noah D. Kauff, Karen E. Hurley, Vincent Devlin, Bert Gold, Robert J. Klein, Kenneth Offit

Final approval of manuscript: Zsofia K. Stadler, Peter Thom, Mark E. Robson, Jeffrey N. Weitzel, Noah D. Kauff, Karen E. Hurley, Vincent Devlin, Bert Gold, Robert J. Klein, Kenneth Offit