Combined effects of E2F1 and E2F2 polymorphisms on risk and early onset of squamous cell carcinoma of the head and neck

Abstract

Deregulated expression of most members of the E2F family has been detected in many human cancers. We examined the association of common single nucleotide polymorphisms (SNPs) of E2F1 and E2F2 with risk of squamous cell carcinoma of the head and neck (SCCHN) in 1,096 SCCHN patients and 1,090 cancer-free controls. We genotyped ten selected SNPs in E2F1 and E2F2, including those at the near 5′ UTR, miRNA binding sites at the near 3′ UTR and tagSNPs according to bioinfotmatics analysis. Although none of the selected SNPs alone was significantly associated with risk of SCCHN, there was a statistically significantly increased risk of SCCHN associated with the combined risk genotypes (i.e. rs3213182 AA, rs3213183 GG, rs3213180 GG, rs321318121 GG, rs2742976 GT+TT, rs6667575 GA+AA, rs3218203 CC, rs3218148 AA, rs3218211 CC, rs3218123 GT+TT). Compared with those with 0–4 risk genotypes, an increased risk was observed for those who carried 5–8 risk genotypes (adjusted OR = 1.04; 95% CI = 0.86–1.26) and 9–10 risk genotypes (adjusted OR = 1.62; 95% CI = 1.14–2.30) in a dose-response manner (P = 0.045). Furthermore, the joint effect was more pronounced among patients with oropharyngeal cancer, younger adults (≤57 years old), men, non-smokers, non-drinkers, and individuals with family history of cancer first-degree relatives. Additionally, we also observed that those with 5–10 risk genotypes had an earlier SCCHN onset than those with 0–4 risk genotypes, particularly for non-smokers and/or non-drinkers. We concluded that E2F1 and E2F2 genetic variants may jointly play important roles in head and neck carcinogenesis.

INTRODUCTION

Squamous cell carcinoma of the head and neck (SCCHN) mainly comprises malignant tumors at the oral cavity, oropharynx, hypopharynx and larynx. Approximately 52,140 new SCCHN cases occurred, accounting for above 3.2% of all newly diagnosed non-cutaneous malignancies, with 11,460 deaths in 2011 in the United States [1]. Tobacco and alcohol use, as well as infection of human papillomavirus, are the known risk factors for SCCHN [2–5]. However, only a small part of exposed individuals develops SCCHN, suggesting that genetic susceptibility factors also play critical roles in carcinogenesis of the head and neck. Increasing evidence has showed that inherited differences in efficiencies of carcinogen metabolism, DNA repair, cell cycle control, apoptosis, or a combination of these factors may modulate individual risk of developing SCCHN [6–11]. However, the roles of genetic variants in the E2F family members in the etiology of SCCHN have not been elucidated.

The E2F family are transcriptional factors that affect cell fate in general, and cancer development in particular [12], because they play pivotal roles in regulating both cell proliferation and antiproliferative processes [13]. Deregulated expression or activity of most members of the E2F family has been detected in many human cancers [14]. For example, E2F1-E2F8 are downstream effectors of the Rb tumor suppressor, which contain one or more conserved DNA binding domains (DBDs) that bind to the target promoters and regulate their expression [15,16]. E2F1, E2F2 and E2F3a, which interact only with pRB, have been classified as the ‘activator E2Fs’ that functionally activate gene expression, and E2F1 and E2F3 can influence the p53 activity by binding directly to p53 [13]. Furthermore, the E2F1 or E2F3 acting as oncogenes are also found to be amplified in many malignancies, and large chromosomal deletions of the regions of E2F1, E2F2 and E2F3 genes have been detected in some patients with neuroblastoma, thyroid cancer, and pancreatic cancer [17].

The E2F1 gene is located on chromosome 20q, spanning approximately 10.71 kb, and it contains 7 exons (National Center for Biotechnology Information reference sequences, NCBI RefSeq). The E2F2 gene is located on chromosome 1p, spanning approximately 24.79 kb, and it also contains 7 exons (National Center for Biotechnology Information reference sequences, NCBI RefSeq). Generally, E2F1−/− may increase susceptibility to tumorigenesis in different tissues [12,13], because E2F1 can induce apoptosis that is mediated by both p53-dependent and p53-independent pathways [14]. One study showed that the absence of E2F2 cooperates with Myc to accelerate tumorigenesis in the oral cavity [18].

It is well known that genetic variants in the coding, promoter, and miRNA binding regions can result in altered gene functions. However, some studies have shown inconsistent results of E2F1 and E2F2 variants in association with risk of cancer, including SCCHN [19,20]. Considering that E2F3, including E2F3a and E2F3b, is more complicated to study, because they are generated by alternative promoters [15]. Here, we tested our hypothese that E2F1 and E2F2 variants or their combination are associated with risk of SCCHN by performing a case-control study to examine the association of common (with minor allele frequency ≥5%) single nucleotide polymorphisms (SNPs) of E2F1 and E2F2 with SCCHN susceptibility.

MATERIALS AND METHODS

Study Subjects and Sample Collection

Study subjects were recruited from our ongoing SCCHN study as described previously [9]. Briefly, this study included 1,096 patients with newly diagnosed and histopathologically confirmed SCCHN and 1,090 self-reported cancer-free controls recruited between October 1999 and October 2009 at The University of Texas M. D. Anderson Cancer Center. Patients were excluded, if they had received prior surgery (other than diagnostic biopsies), chemotherapy or radiation therapy before recruitment, any blood transfusion during the preceding 6 months, any malignancies other than SCCHN, second SCCHN primary tumors, primary tumors of nasopharynx or sinonasal tract, or primary tumors outside the upper aerodigestive tract. Of the 1096 SCCHN patients included in the analysis, 323 (29.55%) had cancers of the oral cavity, 557 (50.8%) of the oropharynx, and 216 (19.7%) of the larynx (including 44 of the hypopharynx). Controls were enrolled from those genetically unrelated hospital visitors who were not seeking health care but accompanying other cancer patients to visit clinics. Controls were frequency-matched to the cases by age (±5 years), sex and ethnicity. Approximately 92% of eligible patients and 85% of controls contacted agreed to participate in this study.

Because of the expected differences in genotype frequencies between ethnic groups, only non-Hispanic whites were included in the analysis. All recruited subjects who were willing to participate in studies were interviewed by trained interviewers. Demographic (e.g., age, sex, and ethnicity) and risk factor (e.g., the history of tobacco smoking and alcohol consumption) information were collected using a standardized, structured questionnaire [21]. Each subject signed for an informed consent and donated a one-time 30-ml venous blood sample (after the diagnosis and before the initiation of treatment for the cases), and the research protocol was approved by the M. D. Anderson institutional review board.

Bioinformatics Analysis and Selection of Polymorphisms

We selected SNPs according to the following strategies: firstly, we identified tagging SNPs of E2F1 and E2F2 genes extending 2000 bps at each end of the gene by Hapmapview4.2 (http://hapmap.ncbi.nlm.nih.gov/cgi-perl/gbrowse/hapmap24_B36/#search). Secondly, we used the bioinformatics tool of SNP Function Prediction (FuncPred, http://snpinfo.niehs.nih.gov/snpfunc.htm) to identify their potential functional relevance as followed: (1) affecting transcription factor binding sites (TFBS) activity in the putative promoter region (here we defined as 2-kb upstream from the first exon) ( http://molsun1.cbrc.aist.go.jp/research/db/TFSEARCH.html), (2) affecting the microRNA (miRNA) binding site activity. We limited the SNPs to those with minor allele frequency ≥5% in the Hapmap CEU populations. We also calculated pairwise linkage disequilibrium (LD) of all SNPs in the same gene and chose those SNPs that were not in LD (r² < 0.8), and plotted LD maps of those SNPs in E2F1 and E2F2 genes with the program Haploview (Vision 4.2) (Figure 1)

Figure 1.

The top panel shows the E2F1 (Figure 1A) and E2F2 (Figure 1B) gene region. The lower panel shows the linkage disequilibrium plot of the E2F1 and E2F2 region, using the CEU population from the International HapMap project. Each square number represents the pairwise r² relationships between the relevant two SNPs.The color of each SNP spot reflects its D′ value, which changes from red to white as the D′ value decreases. The haplotype blocks were estimated with the program Haploview.The minor allele frequency of all above alleles are more than 0.05. Other selected functional alleles predicted in our study rs3213180 in E2F1 and rs3218123, rs3218211 in E2F2 were not included in the International HapMap data.

As a result, we identified two common tagSNPs for the E2F1 gene: rs3213182 (in the 3′ near gene) and rs2071054 (in intron 4). We found another SNP rs3213183 in the 3′ near gene in complete LD with rs2071054 (r² = 1). We also selected an additional SNP rs3213180 (in the 3′ UTR) that was predicted to affect the microRNA (miRNA) binding site activity. Thus, three potentially functional SNPs were identified: rs3213180 (E2F1 c.*914 C>G), rs3213182 (E2F1 3′ 256 T>G) and rs3213183 (E2F1 3′ 527 C>T) (Table 1).

Table 1.

Information of Selected SNPs of E2F1 and E2F2 Genes

Gene	Selected SNPs	Location	TagSNPs	Function Predictedb	Position	Allele	MAFc
E2F1	rs3213180	3′ UTR	NA	miRNA bing site	c.*914	C>G	0.056
	rs3213182	3′ near gene	Yes	NA	3′ 256	T>G	0.067
	rs3213183	3′ near gene	Yesa	NA	3′ 527	C>T	0.267
E2F2	rs6667575	5′ near gene	Yes	TFBS: G:GATA-2/A:GATA-1,2	−1808	G >A	0.200
	rs3218121	5′ near gene	Yesa	TFBS	−604	C >T	0.100
	rs2742976	5′ near gene	Yes	TFBS:G: STATx/T: STATx	−289	G >T	0.408
	rs3218123	5′ near gene	NA	TFBS:G:SRY /T: Elk-1	−143	G >T	0.140
	rs3218148	Intron 1	Yesa	NA	5926	C >T	0.492
	rs3218203	Intron 6	Yes	NA	20153	G >C	0.217
	rs3218211	3′ UTR	NA	miRNA bing site	c.*578	C >T	0.500

^acomplete LD with TagSNP: SNP rs3213183 in the 3′ near gene in complete LD (r²=1) with rs2071054; rs3218121 (in the 5′ near gene) is in complete LD (r²=1) with rs3218127 and that rs3218148 (in intron 1, MAF > 5%) is in complete LD (r²=1) with rs2038027 (in intron 3, MAF < 5%)

^bFunction Predicted were used the bioinformatics tool of SNP: FuncPred, http://snpinfo.niehs.nih.gov/snpfunc.htm) and http://molsun1.cbrc.aist.go.jp/research/db/TFSEARCH.html

^cMAF of rs3213180, rs3218123, and rs3218211 were cited from http://www.ncbi.nlm.nih.gov/SNP, others were form Hapmap-CEU data

For the E2F2 gene, five common tagSNPs were identified, including rs3218127 (in intron 1), rs2038027 (in intron 3), rs2742976 (in the 5′ near gene), rs3218203 (in intron 6), and rs6667575 (in the 5′ near gene). We found that rs3218121 (in the 5′ near gene) is in complete LD (r² = 1) with rs3218127 and that rs3218148 (in intron 1, minor allele frequency, MAF > 5%) is in complete LD (r² = 1) with rs2038027 (in intron 3, MAF < 5%); furthermore, we also identified two additional functional SNPs: rs3218211 (in the 3′UTR, miRNA bing site) and rs3218123 (TFBS) in the E2F2 gene. After we limited the SNPs to those with minor allele frequency ≥5% in the Hapmap CEU population, seven SNPs were selected for genotyping: rs6667575 (E2F2 −1808 G >A), rs3218121 (E2F2 −604 C >T), rs2742976 (E2F2 −289 G >T), rs3218123 (E2F2 −143 G >T), rs3218148 (E2F2 intron 1 C >T) and rs3218203 (E2F2 intron 6 G >C), and rs3218211 (E2F2 c.*578 C >T). Of these, E2F2 rs2742976, rs3218121, rs3218123, and rs6667575 may be at TFBS, according to bioinformatics analysis (Table 1).

Genotyping of E2F1 and E2F2 Polymorphisms

The DNA was extracted from peripheral blood leukocytes by the DNA Blood Mini Kit (Quagen, Valencia, CA), according to the protocol of the manufacturer. The quantification of DNA was determined by a Nanodrop analyzer (ND-1000 spectrophotometer [Nano Drop Technologies, Inc., Wilmington, DE]). All SNPs were genotyped using the TaqMan assay. The PCR amplification was run and the plate was read with the built-in Sequence Detection Software on an ABI-Prism 7900 instrument (Applied Biosystems, Foster City, CA). Each plate included four negative controls, duplicated positive controls and eight repeated samples. The conditions of amplification were as follows: 50°C for 2 min, 95°C for 10 min followed by 40 cycles of 95°C for 15 sec, and 60°C for 1 min.

Statistical Analysis

The differences in the distribution of categorical variables, including demographic characteristics, smoking and drinking status and E2F1and E2F2 genotype/allele frequencies between cases and controls, were tested by Chi-square or Fisher’s exact test (when an expected cell count was less than 5). The deviation from Hardy-Weinberg equilibrium in genotype distribution among controls was tested by Chi-square goodness-of-fit test, and the LD coefficient r² for each of the SNP pairs was also examined. The crude and adjusted odds ratios (ORs) with the 95% confidence intervals (CI) were estimated by unconditional univariate and multivariate logistic regression analyses to assess the association between E2F1 and E2F2 variant genotypes and risk of SCCHN, combined effects of the ten variants, which were further stratified by the primary tumor site, age, sex, smoking, drinking status, and family history of cancer in first-degree relatives, and which were evaluated by the P value in multivariate logistic regression models with adjustment for age, sex, smoking, drinking status, and family history of cancer in first-degree relatives. The P values of trend test was also estimated by unconditional logistic regression analyses, which were carried out to examine the homogeneity and trend of ORs, after controlling for possible confounding of age, sex, smoking status, alcohol use, and family history of cancer first-degree relatives. To adjust multiple tests, we performed permutations by simulating an additional 1000 same logistic models for each SNP and combined ten SNPs by randomly permuting case-control status of all subjects to generate a null distribution with correction for the overfitting caused by selection of a model including ten SNPs [22]. We used Kaplan-Meier curves to describe the cumulative hazard of combined genotypes for age at onset of SCCHN. Statistical significance was tested using Log rank test. All statistics tests were two-sided with a 0.05 significance level. Data were analyzed using SAS statistical software (SAS version 9.2; SAS Institute Inc., Cary, NC) and Stata software (version 8.2; StataCorp LP, College Station, TX)..

RESULTS

Characteristics of the Study Population

The frequency distributions of selected variables between SCCHN cases and controls are described in Table 2. There was no significant difference in the distributions of age between the cases and the controls (P = 0.510) with a mean age of 57.1 (±11.1 years) for the cases and 56.6 (±11.0 years) for the controls. Of all the subjects, 75.4% of the cases and 76.3 % of the controls were men (P = 0.598). The cases were more likely to be current smokers (37.9%) and current drinkers (50.9%) or former drinkers (22.0%), compared with the controls (14.5%, 40.5% and 16.1%, respectively) (P < 0.001 for all). Approximately 60.4% of the SCCHN cases had a family history of cancer in their first-degree relatives, compared with 62.2% of the controls (P = 0.388), which may reflect the nature of controls selected from visitors to a cancer hospital.

Table 2.

Frequency Distribution of Selected Variables among SCCHN Patients and Cancer-free Controls

Variable	Cases n (%)	Controls n (%)	P-valuea
All subjects	1096 (100)	1090 (100)
Age (mean ± standard deviation in years)	57.14±11.14	56.61±11.01	0.258
Age (years)
≤ 57	585 (53.4)	572 (52.5)	0.674
> 57	511 (46.6)	518 (47.5)
Sex
Male	826 (75.4)	832 (76.3)	0.598
Female	270 (24.6)	258 (23.7)
Smoking status
Current	415 (37.9)	158 (14.5)	<0.001
Former	377 (34.4)	397 (36.4)
Never	304 (27.7)	535 (49.1)
Drinking status
Current	558 (50.9)	441 (40.5)	<0.001
Former	241 (22.0)	176 (16.1)
Never	297 (27.1)	473 (43.4)
Family history of cancer in first-degree relatives			0.388
Yes	662 (60.4)	678 (62.2)
No	434 (39.6)	412 (37.8)
Tumor site
Oropharynx	557 (50.8)
Non-oropharynx b	539 (49.2)

^aThe differences between cases and controls were evaluated by two-sided χ² – test for discontinuous variables and Student’s t-tests for continuous variables.

^bIncluded patients with oral cavity (n = 323), patients with laryngeal cancer (n = 172) and patients with hypopharyngeal cancer (n = 44).

E2F1 and E2F2 Genotypes Distribution between Cases and Controls

The frequency distribution of genotypes and alleles of all SNPs are showed in Table 3. The observed genotype frequencies among the controls were all consistent with Hardy-Weinberg equilibrium (all P > 0.05). Although there was no altered risk associated with any of the selected SNPs alone, the frequencies of rs2742976 T, rs6667575 A, and rs3218123 T alleles (0.378, 0.281, and 0.134, respectively) among the cases were slightly higher than those among the controls (0.371, 0.274, and 0.133, respectively), whereas the frequencies of rs3213182 C, rs3213183 A, rs3213180 C, rs3218121 A, rs3218203 G, rs3218148 G, and rs3218211 T alleles (0.102, 0.293, 0.090, 0.760, 0.186, 0.448, and 0.472, respectively) among the cases were slightly lower than those among the controls (0.106, 0.301, 0.100, 0.770, 0.188, 0.449, and 0.485, respectively). As a result, we assumed in further combined analyses that rs2742976 T, rs6667575 A, rs3218123 T, rs3213182 C, rs3213183 A, rs3213180 C, rs3218121 A, rs3218203 G, rs3218148 G, and rs3218211 T alleles may be the putative risk alleles. To further adjust for multiple tests, we performed the permutations but did not find any directional change in the estimates for each of the SNPs (Supplement Table 1).

Table 3.

E2F1 and E2F2 Genotype and Allele Frequencies and their Association with Risk of SCCHN

Genotypes	Cases, n (%)	Controls, n (%)b	Adjusted OR (95% CI)c	Ptrend-testc
E2F1 rs3213182
AAa	880 (80.3)	867 (79.6)	1.00 (reference)
AC	208 (19.0)	214 (19.6)	0.96 (0.77–1.20)
CC	8 ( 0.7)	9 ( 0.8)	0.79 (0.29–2.18)	0.624
AC/CC	216 (19.7)	223 (20.5)	0.95 (0.77–1.19)
C allele	224 (10.2)	232(10.6)
E2F1 rs3213183
GGa	552 (50.4)	533 (48.9)	1.00 (reference)
AG	445 (40.6)	457 (41.9)	0.95 (0.79–1.14)
AA	99 (9.0)	100 (9.2)	0.93 (0.68–1.27)	0.522
AG/AA	544 (49.6)	557 (51.1)	0.95 (0.79–1.13)
A allele	643 (29.3)	657 (30.1)
E2F1 rs3213180
GGa	905 (82.6)	873 (80.1)	1.00 (reference)
GC	180 (16.4)	211 (19.4)	0.84 (0.67–1.06)
CC	11 ( 1.0)	6 ( 0.5)	1.63 (0.57–4.62)	0.311
GC/CC	191 (17.4)	217 (19.9)	0.86 (0.69–1.08)
C allele	202 (0.09)	223 (0.10)
E2F2 rs3218121
GGa	933(85.1)	926(84.9)	1.00 ( reference )
GA	158(14.4)	160(14.7)	0.95 (0.74–1.22)	0.776
AA	5 (0.5)	4 (0.4)	1.18 (0.28–4.90)
GA/AA	163 (14.9)	164 (15.1)	0.96 (0.75–1.23)
A allele	168 (7.6)	168 (7.7)
E2F2 rs2742976
GG	413 (37.7)	431 (39.5)	1.00 (reference)
GT	538 (49.1)	510 (46.8)	1.06 (0.87–1.28)
TT	145 (13.2)	149 (13.7)	1.03 (0.78–1.36)	0.721
GT/TTa	683 (62.3)	659 (60.5)	1.05 (0.88–1.26)
T allele	828 (37.8)	808 (37.1)
E2F2 rs6667575
GG	573 (52.3)	579 (53.1)	1.00 (reference)
GA	431 (39.3)	424 (38.9)	1.03 (0.85–1.24)
AA	92 (8.4)	87 (8.0)	1.13 (0.81–1.57)	0.514
GA/AAa	523 (47.7)	511 (46.9)	1.04 (0.88–1.25)
A allele	615 (28.1)	598 (27.4)
E2F2 rs3218203
CCa	731 (66.7)	719 (66.0)	1.00 (reference)
CG	323 (29.5)	333 (30.5)	0.94 (0.77–1.14)	0.855
GG	42 ( 3.8)	38 ( 3.5)	1.14 (0.71–1.81)
CG/GG	365 (33.3)	371 (34.0)	0.96 (0.80–1.15)
G allele	407 (18.6)	409 (18.8)
E2F2 rs3218148
AAa	347 (31.6)	326 (29.9)	1.00 (reference)
AG	517 (47.2)	550 (50.5)	0.87 (0.71–1.06)	0.850
GG	232 (21.2)	214 (19.6)	1.01 (0.78–1.29)
AG/GG	749 (68.3)	764 (70.1)	0.91 (0.75–1.10)
G allele	981 (44.8)	978 (44.9)
E2F2 rs3218211
CCa	314 (28.7)	286 (26.2)	1.00 (reference)
CT	530 (48.3)	551 (50.6)	0.86(0.70–1.06)	0.382
TT	252 (23.0)	253 (23.2)	0.90 (0.70–1.16)
CT/TT	782 (71.4)	805 (73.8)	0.87 (0.72–1.06)
T allele	1034 (47.2)	1056 (48.5)
E2F2 rs3218123
GG	828 (75.5)	831 (76.2)	1.00 (reference)
GT	243 (22.2)	228 (20.9)	1.00 (0.81–1.24)
TT	25 ( 2.3)	31 (2.9)	0.83 (0.47–1.42)	0.724
GT/TTa	268 (24.5)	259 (23.8)	0.98 (0.80–1.21)
T allele	293 (13.4)	290 (13.3)

^aAssumed risk genotypes.

^bThe observed genotype frequency among the control subjects were in agreement with Hardy-Weinberg equilibrium (p² +2pq + q² =1) (P = 0.541 for E2F1 rs3213180, P = 0.257 for E2F1 rs3213182, P = 0.494 for E2F1 rs3213183, P = 0.539 for E2F2 rs3218121, P = 0.144 for E2F2 rs2742976, P = 0.127 for E2F2 rs3218148, P = 0.400 for E2F2 rs3218203, P = 0.392 for E2F2 rs6667575, P = 0.328 for E2F2 rsrs3218211, P = 0.120 for E2F2 rs3218123)

^cAdjusted OR (95% CI) for age, sex, smoking status, alcohol use, and family history of cancer first-degree relatives in logistic regression models

Association and Stratification Analysis of Combined Effects of E2F1 and E2F2 Genotypes and SCCHN Risk

Considering possible combined effects from different variants or genotypes and potential locus-locus interactions of E2F1 and E2F2 SNPs on risk of SCCHN, we combined three E2F1 SNPs and seven E2F2 SNPs as the number of the putative risk genotypes (i.e., more frequent in the cases than in the controls: rs3213182 AA, rs3213183 GG, rs3213180 GG, rs321318121 GG, rs2742976 GT+TT, rs6667575 GA+AA, rs3218203 CC, rs3218148 AA, rs3218211 CC, and rs3218123 GT+TT). As shown in Table 4, when we divided the combined genotypes into three groups by the number of risk genotypes (i.e., 0–4 risk genotypes, 5–8 risk genotypes, and 9–10 risk genotypes), we found that a difference in the distribution of the combined variant genotypes between the cases and controls was statistically significant (P = 0.023) and that this difference was significantly associated with SCCHN risk in a dose-response manner (compared with those with 0–4 risk genotypes, adjusted OR = 1.04, 95% CI = 0.86–1.26 for those with 5–8 risk genotypes and adjusted OR = 1.62, 95% CI = 1.14–2.30 for those with 9–10 risk genotypes; P for trend = 0.045). We also observed the combined genotypes associated with risk of SCCHN in those carrying 9–10 risk genotypes, compared with those carrying 0–4 risk genotypes in subgroups of the stratified analysis; specifically, the combined effects were more pronounced among younger adults (adjusted OR = 1.74; 95% CI = 1.07–2.85), men (adjusted OR = 1.48; 95% CI = 1.00–2.20), never smoker (adjusted OR = 1.85; 95% CI = 1.07–3.17), never drinker (adjusted OR = 2.19; 95% CI = 1.22–3.95), individuals with family history of cancer first-degree relatives (adjusted OR = 1.64; 95% CI = 1.05–2.57), and oropharyngeal cancer (adjusted OR = 1.76; 95% CI = 1.17–2.64). To correct for potential overfitting caused by the combined ten SNPs in a logistic regression model, we randomly permuted case-control status of all subjects 1000 times, and we observed similar associations of the combined ten SNPs with SCCHN risk in the permuted (null) data similar to the original data (Supplement Table 2).

Table 4.

Association and Stratification Analysis of the Combined Genotypes of E2F1 and E2F2 Polymorphisms and the Risk of Squamous Cell Carcinoma of the Head and Neck

Variables	n, case/control	Number of combined risk genotypes of of E2F1 and E2F2 Polymorphismc					P trendd
		%(case/control)			Adjusted OR (95% CI)d	Adjusted OR (95% CI)d
		0–4	5–8	9–10	0–4 vs. 5–8	0–4 vs. 9–10
Totala	1096/1090	31.0/33.5	59.5/60.1	9.5/6.4	1.04 (0.86–1.26)	1.62 (1.14–2.30)	0.045
Tumor site
Oropharynx	557/1090	28.6/33.5	61.2/60.1	10.2/6.4	1.14 (0.90–1.44)	1.76 (1.17–2.64)	0.016
Non-oropharynxb	539/1090	33.6/33.5	57.7/60.1	8.7/6.4	0.92 (0.72–1.18)	1.43 (0.90–2.26)	0.556
Age
≤ median (57 years)	585/572	30.1/32.7	60.7/61.4	9.2/5.9	1.07 (0.82–1.39)	1.74 (1.07–2.85)	0.083
> median (57 years)	511/518	32.1/34.4	58.1/58.7	9.8/7.0	1.03 (0.78–1.35)	1.47 (0.89–2.42)	0.261
Sex
Male	826/832	31.2/32.6	59.1/60.7	9.7/6.7	1.00 (0.80–1.24)	1.48 (1.00–2.20)	0.191
Female	270/258	30.4/36.4	60.7/58.1	8.9/5.4	1.21 (0.82–1.79)	2.09 (0.98–4.48)	0.074
Smoking status
Never	304/535	30.9/33.3	58.2/60.4	10.9/6.4	1.03 (0.76–1.41)	1.85 (1.07–3.17)	0.109
Ever	792/555	31.1/33.7	60.0/59.8	9.0/6.5	1.04 (0.82–1.33)	1.41 (0.90–2.21)	0.238
Drinking status
Never	297/473	31.7/36.2	58.3/58.4	10.1/5.5	1.16 (0.84–1.59)	2.19 (1.22–3.95)	0.027
Ever	799/617	30.8/31.4	60.0/61.4	9.3/7.1	0.99 (0.78–1.25)	1.32 (0.86–2.04)	0.414
Family history of cancer in first-degree relatives
Yes	662/678	30.2/32.2	59.8/61.7	10.0/6.2	1.02 (0.80–1.30)	1.64 (1.05–2.57)	0.125
No	434/412	32.3/35.7	59.0/57.5	8.8/6.8	1.11 (0.82–1.50)	1.53 (0.87–2.66)	0.173

^aThe distribution of three groups of the combined genotypes between the cases and controls for total samples was statistically significant (P = 0.023).

^bincluded oral cavity, larynx and hypopharynx.

^cThe number represents the numbers of risk genotypes: the risk genotypes used for the calculation were rs3213182 AA, rs3213183 GG, rs3213180 GG, rs3218121 GG, rs2742976 GT+TT, rs6667575 GA+AA, rs3218203 CC, rs3218148 AA, rs3218211 CC, rs3218123 GT+TT genotypes.

^dAdjusted OR (95% CI) for age, sex, smoking status, alcohol use, family history of cancer first-degree relatives in logistic regression models.

Age at SCCHN Onset

In the case-only analysis of 1096 patients, the median age at SCCHN onset was 57 years for both those with 0–4 risk genotypes (a range of 22–89 years) and those with 5–10 risk genotypes (a range of 18–85 years). No significant difference in age at onset was observed for subgroups of sex, smokers, and drinkers, and tumor sites (data not shown). However, in never smoking patients, those with the combined 5–10 risk genotypes had a younger age at onset (a median age of 52 years) than those with the 0–4 risk genotypes (a median age of 54 years) (Log rank, P = 0.038) (Figure 2A). In never drinking patients, those with the combined 5–10 risk genotypes also had a younger age at onset (a median age 55 years) than those with the 0–4 risk genotypes (a median age of 56 years) (Log rank, P = 0.045) (Figure 2B). In both never smoking and never drinking patients, those with the combined 5–10 risk genotypes had a younger age at onset (a median age of 53 years) than those with the 0–4 risk genotypes (a median age 56 years) (Log rank, P = 0.034) (Figure 2C).

Figure 2.

Kaplan-Meier curves: cumulative risk of combined genotypes of rs3213182 AA, rs3213183 GG, rs3213180 GG, rs3218121 GG, rs2742976 GT+TT, rs6667575 GA+AA, rs3218203 CC, rs3218148 AA, rs3218211 CC, rs3218123 GT+TT for age at onset of SCCHN. Patients with 5–10 risk genotypes were associated with earlier onset of SCCHN, compared with those with 1–4 risk genotypes in never smokers (median age 52 yr vs. 54 yr, Long rank, P = 0.038, Figure 2A), in never drinkers (median age 55 yr vs. 56 yr, Long rank, P = 0.045, Figure 2B), and in never smokers and never drinkers (median age 53 yr vs. 56 yr, Long rank, P = 0.034, Figure 2C).

DISCUSSION

We investigated the associations of ten selected SNPs of E2F1 and E2F2 with risk of SCCHN in non-Hispanic whites in a hospital-based case-control study. We found that none of the ten SNPs in E2F1 and E2F2 alone had a significant main effect on risk of SCCHN in our study population. However, our results suggested a combined effect of the ten SNPs on risk of SCCHN in a genotype-dose response manner. Specifically, the combined effect was more pronounced among subgroups of men, non-smokers, non-drinkers, individuals with an early age at onset or with a family history of cancer in first-degree relatives, and patients with oropharyngeal cancer. Furthermore, we observed that those with 5–10 risk genotypes had an earlier age at SCCHN onset than those with the 0–4 risk genotypes in never smoking and/or drinking patients. Given the role of E2F1 and E2F2 genes in activating gene expression and influencing the p53 activity, it is biologically plausible that the E2F1 and E2F2 SNPs may modulate risk of SCCHN.

E2F1 and E2F2 as key transcription factors can interact with other genes in both the Rb-E2Fs pathway and other biological pathways that are involved in proliferation, cell cycle progression, and apoptosis [12,15,23]. For example, it has been shown that the retinoblastoma protein can restrict cells from entering the S phase by binding to and sequestering of the E2F activators in the model of mammalian cell cycle control [24]. Our finding of no main effect of each of E2F1 or E2F2 variants or variant genotypes on risk of SCCHN is consistent with previously published studies on other cancers. One case-control study showed that a SNP rs760607 (E2F2 −5368 A>G) in the promoter region of E2F2 had no main effect on risk of breast cancer [20]. However, conflicting results do exist for other cancers. For example, one study found that E2F2 rs760607 SNP had no association with ovarian cancer risk in the replication set, though an association with ovarian cancer risk was reported for the E2F2 rs760607 in a dominant model and for the E2F2 rs3218203 SNP in a recessive model in the discovery set [19]. However, no reported studies have investigated the association of E2F1 and E2F2 SNPs with SCCHN risk, nor any published studies have evaluated the combined effects of E2F1 and E2F2 variant genotypes on cancer risk. By analyzing ten SNPs simultaneously in our study, we found that the number of putative risk genotypes was associated with risk of SCCHN in genotype dose-response manner.

It is well known that genetic susceptibility contributes to the risk of developing cancer, which is also supported by some evidence in the present study, such as an early age onset, elevated risk in subgroups without apparent exposure to tobacco or alcohol, and among individuals with a family history of cancer in first-degree relatives, all indicating an important role of genetic background in the study population. Several selected SNPs (i.e. rs2742976 (E2F2 −289 G >T), rs6667575 (E2F2 −1808 G >A), and rs3218123 (E2F2 −143 G >T) are located in the E2F2 promoter region. Computer prediction analysis has shown that the −1808, −289 and −143 regions in the 5′ upstream gene sequence in E2F2 contain putative transcription factor binding sites for GATA-1,2, STATx, SRY, and Elk-1 (Table 1). It has been proposed that polymorphisms in the promoter region could destroy transcriptional regulation factors, leading to altered transcription [25]. Micro-RNAs (miRNAs) can also regulate the E2F activity, and miRNAs dysregulation has been implicated in malignancy. One study suggested that the E2F family proteins can target miRNAs by a negative feedback loop [23]. For instance, E2F1, E2F2 and E2F3 directly bind to the promoter of the miR-17~92 cluster activating its transcription, in turn, which negatively modulate translation of E2F1–3 mRNAs by binding sites in their 3′-untranslated regions [26,27]. Likewise, miR-330-3p exhibits its tumor suppressive activity through E2F1-mediated repression of Akt activation [28]. Both E2F1 rs3213180 and E2F2 rs3218211 are located in the 3′UTR and have miRNA binding sites as predicted, and E2F1 rs3213182 and rs3213183 are located in the 3′ near gene. Although we did not observe any their main effects on SCCHN risk, these variants could contribute in part to the observed joint effects on SCCHN risk. It is also likely that other ‘non-functional’ variants, such as tagSNPs in introns, may be markers for untyped functional SNPs and may be in LD with functional or disease causing variants in the genome [29].

There has been a trend for an increased risk in the number of young patients who developed SCCHN without recognized risk factors (i.e., tobacco and alcohol use) [30,31]. It has been reported that never-smoker patients with SCCHN tended to be younger than smokers with SCCHN [32], In the present study, we found that the combined risk genotypes of the ten SNPs were associated with an increased risk of SCCHN among younger adults, men, never smokers or never drinkers in our study population. Furthermore, we found that in 1096 SCCHN patients, those with the combined 5–10 risk genotypes had 1 years earlier at SCCHN onset than those with 0–4 risk genotypes in never drinkers, 2 year earlier in never smokers and 3 years earlier in never smokers and never drinkers. These indicated that SCCHN risk in never smoking and never drinking individuals was likely attributed to genetic factors compared with that in either never smokers or never drinkers, although at least 75% of SCCHN are attributable to a combination of cigarette smoking and alcohol drinking [33]. Therefore, our findings suggested that E2F1 and E2F2 genetic variants may play important roles in head and neck carcinogenesis.

Human papillomavirus (HPV) infection is a recognized risk factor for SCCHN [5]. A number of studies indicated that HPV-related SCCHN occurs predominantly at the oropharynx [34]. HPV16 seropositivity was associated with a 30-fold increased risk of oropharyngeal cancer among light drinkers or never smokers [35]. It has been found that HPV type 16 E7 can bind to E2F1 and activate E2F1-driven transcription through a retinoblastoma protein-independent manner [36]. Longworth et al. found that activating E2F2 transcription through its interaction with histone deacetylases (HDACs) can facilitate HPV31 E7 replication [37]. Indeed, we observed a positive association with risk of oropharyngeal cancer, which supports the likelihood of HPV infection-related SCCHN [34,38], and thus genetic variants in E2F1 and E2F2 may have modulated HPV-induced carcinogenesis in young individuals.

The strengths of our study included relatively large study sample size, our ethnically homogeneous study population, and unbiased genotyping data distributions in the control subjects as suggested by their agreement with Hardy-Weinberg equilibrium. However, there are still some limitations in our study. Firstly, small sample size in subgroup analyses may have attenuated statistical power, and therefore positive findings in stratification analysis need further confirmation in larger studies. Secondly, our study was hospital-based, which likely had recall bias and selection bias. However, the genotype frequencies of eight variants in our controls are similar to those for European populations published in the Hapmap website (http://hapmap.ncbi.nlm.nih.gov/cgi-perl/gbrowse/hapmap24_B36/) (data not shown), though the frequencies of rs3213180, rs3218123 and rs3218211 genotypes have not been reported for European populations in the Hapmap website. For example, the frequencies of the GG, GT, and TT genotypes of the rs2742976 among our 1090 Caucasians control subjects were 39.5%, 46.8%, and 13.7%, respectively, compared with those of Hapmap-CEU: 37.8%, 47.7%, and 14.4%, respectively. Hence, selection bias in our study, if any, was unlikely to be substantial.

In summary, our results suggested that any one of the E2F1 or E2F2 variants alone may not have a substantial effect on SCCHN risk, but a joint effect of the combined risk genotypes (i.e., rs3213182 AA, rs3213183 GG, rs3213180 GG, rs3218121 GG, rs2742976 GT+TT, rs6667575 GA+AA, rs3218203 CC, rs3218148 AA, rs3218211 CC, rs3218123 GT+TT) may contribute to risk of SCCHN, and this increased risk was more pronounced among the younger adults (≤ 57 years old), men, never smokers, never drinkers, individuals with a family history of cancer in the first-degree relatives, and patients with oropharyngeal cancer. In addition, those with the combined 5–10 risk genotypes had an earlier age at SCCHN onset than those with the 0–4 risk genotypes, particularly in never smoking and/or drinking patients. These results indicated the important role of the E2F1 or E2F2 variants in head and neck carcinogenesis, but our findings need to be validated by larger studies.

Abbreviations

CI

confidence interval

E2F1

E2F transcription factor 1

E2F2

E2F transcription factor 2

LD

linkage disequilibrium

MAF

minor allele frequency

OR

odds ratio

PCR

polymerase chain reaction

SCCHN

Squamous cell carcinoma of the head and neck

TFBS

transcription factor binding site

miRNA

Micro ribonucleic acid

5′ UTR

5′ untranslated regions

3′UTR

3′ untranslated regions