Cell-cycle checkpoint controls regulate cell-cycle progression and proliferation. Alterations in cell-cycle control mechanisms are linked to tumorigenesis.
Cell-cycle checkpoint controls regulate cell-cycle progression and proliferation. Alterations in cell-cycle control mechanisms are linked to tumorigenesis.
This case-control study included 147 cases and 147 controls. The authors used a pathway-based approach to assess the association between 10 potential functional single-nucleotide polymorphisms from 7 cell-cycle control genes and the risk of oral premalignant lesions (OPLs). They also used classification and regression tree analysis to examine high-order gene-gene and gene-smoking interactions.
Compared with the homozygous wild-type GG genotype of CCND1 P241P, individuals with the AG genotype exhibited an increased risk of OPL (odds ratio, 1.58; 95% confidence interval, 0.89-2.83), and carriers of the AA genotype had a significantly increased risk of OPL (odds ratio, 2.75; 95% confidence interval, 1.33-5.71), with risk increasing significantly with the increasing number of variant alleles (P = .006). The risk of OPL increased significantly as the number of unfavorable genotypes in the pathway increased (P = .002). The final decision tree in the classification and regression tree analysis contained 5 terminal nodes. Compared with the never smokers (the lowest risk group), the odds ratios for terminal nodes 2 through 5 ranged from 1.21 to 5.40.
The results illustrated the advantage of using a pathway-based approach for analyzing gene-gene and gene-smoking interactions. Specifically, the authors showed that genetic polymorphisms in cell-cycle control pathway genes may contribute to the risk of OPL. Cancer 2008. © 2008 American Cancer Society.
The chief clinical manifestations of oral premalignant lesions (OPLs) are leukoplakia, erythroplakia, lichen planus, and submucous fibrosis.1 The risk of developing oral cancer is high in individuals with OPLs. Lee et al followed 70 OPL patients with a medium follow-up of 7 years. The resulting incidence rate of cancer in the upper aerodigestive tract after treatment was 31.4%.2 The probability of showing evidence of severe epithelial dysplasia or malignancy at diagnosis was 90% for erythroplakia.3 A few studies suggest that the most common manifestation of OPL, leukoplakia, has a coincidence rate of up to 60% at the time of diagnosis of oral squamous cell carcinoma.1 The major risk factors for oral tumorigenesis are tobacco chewing, cigarette smoking, and alcohol consumption, but evidence also supports the role of genetic susceptibility.4–7
A study of the role of genetic polymorphisms in OPLs can be approached in different ways. A candidate gene study assesses the effects of a single genetic polymorphism at a time, whereas a pathway-based approach examines the joint effects of a panel of genetic polymorphisms in the same pathway. Because cancer is a multistep, multigenetic process, single-gene analyses usually provide limited power. Therefore, a pathway-based approach may provide more insight and enhance the prediction power.8–10
Chromosomal damage induced during carcinogenesis leads to the activation of cell-cycle checkpoint controls that regulate cell progression and proliferation. Under cell-cycle control, the affected cells experience either slow growth or apoptosis.11 Alterations in various cell-cycle control mechanisms are linked to tumorigenesis.12–14 Specifically, the alterations in the expression level of cell-cycle proteins were observed in both premalignant and malignant lesions arising in the oral cavity.15 The aim of this case-control analysis was to use a pathway-based approach to assess the association between 10 single-nucleotide polymorphisms (SNPs) from 7 cell-cycle checkpoint pathway genes (P53, P21, P27, CDK4, CDK6, CCND1, and STK15) and the risk of OPL. We also used classification and regression tree (CART) analysis to examine high-order gene-gene and gene-smoking interactions.
Adult patients (aged >17 years) with histologically confirmed OPL (leukoplakia and/or erythroplakia) were recruited between 1997 and 2006 from the patient population at the University of Texas M. D. Anderson Cancer Center in Houston, Texas. Patients who had acute intercurrent illnesses or infections, those who had undergone retinoid or carotenoid therapy within 3 months before study entry, and those with a history of cancer (except nonmelanoma skin cancer) who had received treatment within 2 years before enrollment were excluded from the study. Healthy individuals with no prior history of cancer (except nonmelanoma skin cancer) were selected from a database of control subjects recruited at Kelsey-Seybold clinics, the largest private multispecialty physician group in the Houston metropolitan area. Controls were matched to cases by age (±5 years), sex, and ethnicity. We included a total of 147 patients with OPL and 147 healthy controls in this study.
Epidemiologic data were collected including sociodemographic characteristics, recent and prior tobacco use, and alcohol consumption. Blood was drawn into heparinized tubes and delivered to the laboratory for DNA extraction and other molecular analyses. Each sample was given a study identification number before delivery to the laboratory. All laboratory workers were blinded to the case-control status of the samples. All study participants were provided written informed consent in accordance with the institutional review boards of M. D. Anderson Cancer Center and Kelsey-Seybold Clinic.
We conducted genotyping using the TaqMan method with a 7900HT Sequence Detection System (Applied Biosystems, Foster City, Calif), except for the p53 intron 3 polymorphism, which we genotyped using polymerase chain reaction (PCR)-restriction fragment length polymorphism.16 DNA samples (5 ng) were mixed with 1 × TaqMan buffer A, deoxynucleotide triphosphate (200 μM), MgCl2 (5 mM), AmpliTaq Gold (0.65 U), each primer (900 nM), and each probe (200 nM) to produce the amplification mixes (5 μL). The mixes were placed in the GeneAmp PCR system (Applied Biosystems) to perform the reaction in the following sequence of thermal conditions: 95°C for 10 minutes, followed by 50 cycles at 92°C for 30 seconds and 60°C for 1 minute. The final genotyping result was provided by the built-in software of the system.
The chi-square test was used to test for differences between the patients and the controls subjects in the distributions of sex, ethnicity, and smoking status. We used Pearson chi-square and Fisher exact tests to assess the association between each SNP in patients and control subjects. We applied Student t test (normal distribution variables) or the rank sum test (non-normal distribution variables) for continuous variables. Unconditional multivariate logistic regression was performed to compute odds ratios (ORs) as estimates of relative risk for overall and stratified analyses with 95% confidence intervals (CIs) while adjusting for possible confounding by age, sex, ethnicity, alcohol use, and smoking status (never, former, or current) when appropriate. An ever smoker was defined as an individual who had smoked at least 100 cigarettes in his/her lifetime. Ever smokers included former smokers, current smokers, and recent quitters (quit within the previous year). A former smoker was defined as an individual who had quit smoking at least 1 year before the interview.
In addition, we used a pathway-based approach to evaluate OPL risk as a function of the number of unfavorable genotypes in the cell-cycle control pathway. We determined the best fitting model for each SNP, ie, dominant risk model (comparing homozygous wild-type genotype with variant allele-carrying genotypes), recessive risk model (comparing homozygous wild-type and heterozygous genotypes with homozygous variant genotype), or additive model on the basis of P value. An SNP is included in our pathway-based analysis if the P value for the best fitting model is less than .5. Without prior knowledge of the genetic variant or OPL risk, we assumed that the minor variant was the risk allele. However, because our assumption may have been inaccurate, we used a more stringent measure to reassign the unfavorable genotype if the OR was ≤0.75 or the was OR ≤1.0 and the P value was ≤.05. If the genotype frequency for the variant homozygous genotype in both patients and control subjects was <5%, we combined it with that of the heterozygous genotype. All Pvalues were 2-sided, and associations were considered statistically significant at P < .05. We used Stata software (version 8.0; Stata Co., College Station, Tex) to complete these analyses.
For high-order gene-gene and gene-environment interactions, we used CART analysis (HelixTree Genetics Analysis Software, version 4.1.0; Golden Helix, Inc., Bozeman, Mont) to build a decision tree via recursive partitioning. Before growing a tree, we defined the measure for goodness of split using multiplicity-adjusted P values and controlled tree growth at .05. Starting with the root node that contained all the patient and control subject data, the CART process identified the split with the smallest multiplicity-adjusted P value, ie, the most optimal split, for the root node. This process continued for each subsequent node until there was no statistically significant split or there were <10 subjects in the terminal node.
The study population consisted of 147 patients with OPL and 147 healthy controls, frequency-matched by age, sex, and ethnicity (Table 1). Although the control subjects were slightly older than the patients (59.1 ± 11.0 years vs 57.5 ± 13.6 years), the difference was not statistically significant (P = .26). As shown in Table 1, there was a significant difference in smoking status between the 2 groups: 68.7% of patients and 44.9% of controls were ever smokers (P < .001). However, control subjects had significantly more alcohol use than cases (68.0% vs 53.7%, P = .012).
|Variable||Control (%)||Case (%)||P|
|Males||82 (55.8)||82 (55.8)|
|Females||65 (44.2)||65 (44.2)||1|
|Never||81 (55.1)||46 (31.3)|
|Ever||66 (44.9)||101 (68.7)||<.001|
|Former||55 (37.4)||58 (39.5)|
|Current & RQ||11 (7.5)||43 (29.3)||<.001|
|White||129 (87.8)||129 (87.8)|
|Hispanic||11 (7.5)||11 (7.5)|
|African American||7 (4.8)||7 (4.8)||1|
|Never||47 (32.0)||68 (46.3)|
|Ever||100 (68.0)||79 (53.7)||.012|
|Age, mean y (±SD)||59.1 (11.0)||57.5 (13.6)||.26|
There was no significant departure from Hardy-Weinberg equilibrium for the 10 SNPs in control subjects. Compared with the homozygous wild-type GG genotype of CCND1 P241P, those with the heterozygous genotype AG exhibited an increased risk of OPL (OR, 1.58; 95% CI, 0.89-2.83), and carriers of the variant homozygous genotype AA had a significantly increased risk of OPL (OR, 2.75; 95% CI, 1.33-5.71), with the risk increasing significantly with the increasing number of variant alleles (P = .006) (Table 2). In the stratified analysis (Tables 2 and 3), elevated risk conferred by the CCND1 variant allele was more evident among men (OR, 2.55; 95% CI, 1.08-6.04 for AG; OR, 9.89; 95% CI, 3.00-32.68 for AA; P for trend <.001), whereas no significant association was found among women. This trend of increasing risk for CCND1 was also seen in ever smokers (OR, 1.41; 95% CI, 0.64-3.14 for AG; OR, 2.96; 95% CI, 1.09-8.05 for AA; P for trend = .036) and younger individuals (aged ≤60 years) (OR, 2.08; 95% CI, 0.95-4.57 for AG; OR, 2.75; 95% CI, 1.02-7.42; P for trend = .03). Overall, there was no significant association between P53 R72P or CDK6 and OPL risk. However, never smokers with the homozygous variant CC genotype of P53 R72P exhibited a significantly decreased risk of OPL (OR, 0.16; 95% CI, 0.03-0.99). Younger individuals carrying the same genotype of this SNP showed a decreased risk of OPL (OR, 0.19; 95% CI, 0.03-1.29), although it is not statistically significant at P = .05. Younger patients with the heterozygous genotype CT of cyclin-dependent kinase 6 (CDK6) had a significantly reduced risk of OPL compared with the homozygous wild-type genotype CC (OR, 0.44; 95% CI, 0.19-0.98). There was a borderline significantly decreased risk of OPL for carriers of a variant allele of CDK6 among younger individuals (OR, 0.46; 95% CI, 0.21-1.00).
|Polymorphism||Subjects, N||Overall||Never Smoker||Ever Smoker|
|Controls||Cases||OR (95% CI)*||P||OR (95% CI)†||P||OR (95% CI)‡||P|
|GG||56 (38.1)||37 (25.7)||Ref.||Ref.||Ref.|
|AG||69 (46.9)||70 (48.6)||1.58 (0.89-2.83)||.12||1.81 (0.76-4.35)||.183||1.41 (0.64-3.14)||.394|
|AA||22 (15.0)||37 (25.7)||2.75 (1.33-5.71)||.006||2.66 (0.85-8.31)||.092||2.96 (1.09-8.05)||.033|
|Dominant||1.86 (1.08-3.21)||.025||2.00 (0.87-4.60)||.102||1.78 (0.84-3.75)||.13|
|Recessive||2.09 (1.11-3.95)||.023||1.85 (0.69-4.99)||.223||2.41 (1.00-5.76)||.049|
|P for trend||.006||.077||.036|
|P53 intron 3||147||111|
|No insert||103 (70.1)||76 (68.5)||Ref.||Ref.||Ref.|
|1 insert||41 (27.9)||33 (29.7)||1.17 (0.65-2.11)||.604||1.22 (0.49-3.05)||.669||1.11 (0.51-2.45)||.79|
|2 insert||3 (2.0)||2 (1.80)||1.15 (0.16-8.48)||.891||1.44 (0.18-11.83)||.732|
|Dominant||1.17 (0.66-2.08)||.599||1.25 (0.52-2.99)||.620|
|P53 intron 6||146||102|
|GG||104 (71.2)||73 (71.6)||Ref.||Ref.||Ref.|
|GA||40 (27.4)||27 (26.4)||1.04 (0.56-1.91)||.9061||1.06 (0.43-2.63)||.899||1.06 (0.45-2.45)||.9|
|AA||2 (1.4)||2 (2.0)||2.18 (0.27-17.68)||.465||3.21 (0.36-28.60)||.295|
|Dominant||1.08 (0.60-1.97)||.795||1.20 (0.50-2.84)||.685|
|GG||68 (49.6)||62 (56.4)||Ref.||Ref.||Ref.|
|GC||53 (38.7)||42 (38.2)||0.83 (0.47-1.46)||.508||0.87 (0.35-2.16)||.762||0.80 (0.38-1.68)||.555|
|CC||16 (11.7)||6 (5.4)||0.47 (0.15-1.44)||.185||0.16 (0.03-0.99)||.049||1.11 (0.21-5.90)||.898|
|Dominant||0.76 (0.44-1.30)||.314||0.62 (0.26-1.46)||.277||0.83 (0.40-1.71)||.611|
|Recessive||0.51 (0.17-1.52)||.224||0.17 (0.03-1.00)||.051||1.23 (0.24-6.30)||.805|
|P for trend||.192||.082||.737|
|AA||73 (51.1)||64 (47.1)||Ref.||Ref.||Ref.|
|AC||56 (39.2)||64 (47.1)||1.29 (0.76-2.19)||.343||2.01 (0.88-4.56)||.097||0.96 (0.47-1.97)||.921|
|CC||14 (9.8)||8 (5.9)||0.64 (0.24-1.72)||.381||0.95 (0.21-4.27)||.950||0.44 (0.11-1.68)||.229|
|Dominant||1.15 (0.70-1.91)||.576||1.77 (0.81-3.89)||.152||0.85 (0.43-1.69)||.65|
|Recessive||0.57 (0.22-1.49)||.256||0.68 (0.16-2.91)||.608||0.45 (0.12-1.64)||.224|
|P for trend||.959||.373||.381|
|CC||82 (56.6)||65 (52.9)||Ref.||Ref.||Ref.|
|CT||55 (37.9)||48 (39.0)||0.99 (0.57-1.71)||.967||0.56 (0.23-1.35)||.195||1.28 (0.61-2.71)||.51|
|TT||8 (5.5)||10 (8.1)||1.45 (0.48-4.36)||.51||1.02 (0.23-4.48)||.981||2.29 (0.38-13.85)||.366|
|Dominant||1.04 (0.62-1.77)||.875||0.63 (0.28-1.42)||.264||1.37 (0.67-2.81)||.394|
|Recessive||1.46 (0.49-4.28)||.496||1.27 (0.30-5.40)||.749||2.07 (0.35-12.20)||.42|
|P for trend||.69||.464||.311|
|CC||121 (84.6)||116 (84.7)||Ref.||Ref.||Ref.|
|CT||21 (14.7)||19 (13.9)||0.92 (0.44-1.91)||.819||1.30 (0.39-4.36)||.668||0.75 (0.28-1.99)||.566|
|TT||1 (0.7)||2 (1.5)||2.36 (0.16-35.96)||.535||3.69 (0.13-104.80)||.444|
|Dominant||0.97 (0.47-1.98)||.924||1.38 (0.42-4.52)||.597||0.81 (0.31-2.11)||.669|
|CC||81 (57.0)||86 (66.7)||Ref.||Ref.||Ref.|
|CT||55 (38.7)||35 (27.1)||0.64 (0.37-1.13)||.123||0.82 (0.35-1.96)||.659||0.48 (0.22-1.02)||.057|
|TT||6 (4.2)||8 (6.2)||1.62 (0.50-5.17)||.419||1.36 (0.27-6.75)||.707||1.90 (0.30-11.90)||.495|
|Dominant||0.73 (0.43-1.24)||.248||0.90 (0.40-2.02)||.790||0.55 (0.27-1.15)||.112|
|Recessive||1.90 (0.60-5.99)||.272||1.46 (0.30-7.03)||.639||2.60 (0.43-15.90)||.301|
|P for trend||.596||.980||.336|
|TT||88 (61.1)||78 (57.4)||Ref.||Ref.||Ref.|
|TA||53 (36.8)||49 (36.0)||1.08 (0.64-1.83)||.77||1.36 (0.61-3.02)||.455||0.77 (0.37-1.59)||.476|
|AA||3 (2.1)||9 (6.6)||2.49 (0.59-10.46)||.213||1.05 (0.08-14.51)||.971||4.72 (0.54-41.63)||.162|
|Dominant||1.17 (0.70-1.94)||.556||1.34 (0.61-2.92)||.469||0.93 (0.46-1.86)||.829|
|Recessive||2.41 (0.58-10.00)||.224||0.95 (0.07-13.01)||.968||5.24 (0.61-45.21)||.132|
|P for trend||.352||.518||.647|
|GG||104 (72.7)||104 (75.9)||Ref.||Ref.||Ref.|
|GA||36 (25.2)||30 (21.9)||0.81 (0.45-1.48)||.498||0.72 (0.27-1.87)||.494||0.85 (0.38-1.88)||.687|
|AA||3 (2.1)||3 (2.2)||1.17 (0.22-6.29)||.854||2.94 (0.28-31.37)||.372|
|Dominant||0.84 (0.47-1.50)||.553||0.62 (0.24-1.59)||.317||0.95 (0.44-2.04)||.894|
|Men||Women||Young Individual (Age ≤60 y)||Old Individual (Age >60 y)|
|Control||Case||OR (95% CI)*||Control||Case||OR (95% CI)*||Control||Case||OR (95% CI)†||Control||Case||OR (95% CI)†|
|AG||39||43||2.55 (1.08-6.04)||30||27||0.85 (0.36-2.04)||31||40||2.08 (0.95-4.57)||38||30||0.97 (0.38-2.49)|
|AA||10||23||9.89 (3.00-32.68)||12||14||0.84 (0.29-2.48)||12||20||2.75 (1.02-7.42)||10||17||2.67 (0.83-8.58)|
|Dominant||3.51 (1.56-7.93)||0.85 (0.38-1.92)||2.26 (1.09-4.71)||1.30 (0.54-3.14)|
|Recessive||5.37 (1.94-14.88)||0.92 (0.36-2.40)||1.83 (0.76-4.44)||2.72 (1.00-7.36)|
|P for trend||<.0001||.729||.030||.127|
|GC||30||20||0.65 (0.28-1.50)||23||22||1.15 (0.50-2.64)||28||26||1.06 (0.48-2.33)||25||16||0.59 (0.24-1.46)|
|CC||10||5||0.63 (0.17-2.42)||6||1||12||2||0.19 (0.03-1.29)||4||4||0.83 (0.13-5.42)|
|Dominant||0.65 (0.30-1.41)||0.92 (0.41-2.08)||0.85 (0.40-1.81)||0.62 (0.26-1.46)|
|Recessive||0.75 (0.21-2.76)||0.19 (0.03-1.22)||0.99 (0.16-6.29)|
|P for trend||.31||.434||.268||.368|
|CT||29||20||0.64 (0.29-1.43)||26||15||0.57 (0.24-1.38)||29||15||0.44 (0.19-0.98)||26||20||1.06 (0.45-2.49)|
|TT||5||3||0.72 (0.12-4.13)||1||5||7.75 (0.76-79.39)||3||3||0.68 (0.12-3.95)||3||5||3.17 (0.58-17.32)|
|Dominant||0.65 (0.30-1.41)||0.78 (0.34-1.78)||0.46 (0.21-1.00)||1.23 (0.54-2.78)|
|Recessive||0.84 (0.15-4.77)||9.50 (0.96-94.52)||0.89 (0.16-5.05)||3.09 (0.59-16.07)|
|P for trend||.33||.743||.095||.344|
In this pathway-based analysis, we excluded P53 intron3, P53 intron6, P21 3′UTR, and STK15 I57V because P values for the best fitting model were greater than .5. For the other SNPs, the unfavorable genotypes were AG/AA for CCND1 P241P, GG/GC for P53 R72P, AA/AC for CDK4 3′UTR, TT for P27 5′UTR, CC for CDK6 3′UTR, and AA for STK15 F31I. As shown in Table 4, we categorized the subjects into 3 risk groups according to the number of unfavorable genotypes: low-risk group (1-2 unfavorable genotypes), medium-risk group (3 unfavorable genotypes), and high-risk group (4-5 unfavorable genotypes). The OR for the medium-risk group was 5.01 (95% CI, 1.74-14.40) and for the high-risk group was 6.18 (95% CI, 2.14-17.86). The risk of OPL increased significantly as the number of unfavorable genotypes increased (P = .002).
|Risk (Unfavorable genotypes, n)||Controls, n||Patients, n||OR (95% CI)||P|
|Medium (3)||56||39||5.01 (1.74-14.40)||.0028|
|High (4-5)||39||43||6.18 (2.14-17.86)||.0008|
|P for trend||.002|
Ten genetic polymorphisms and smoking status were included in the CART analysis. The initial split of the root node was smoking status, showing the strongest association with OPL risk and supporting its critical role in the risk of OPL (Fig. 1). A total of 5 terminal nodes were constructed in the final tree. Never smokers, the lowest-risk group, were in terminal node 1. Table 5 shows the association between terminal node and OPL risk. The ORs for terminal nodes 2-5 ranged from 1.21 to 5.40 compared with terminal node 1, adjusting for age, sex, ethnicity, and alcohol use. The highest-risk group, terminal node 5, contained ever smokers carrying the common homozygous genotype of CDK6 3′UTR and the variant allele P27 5′UTR. The 5 terminal nodes were categorized into 3 risk groups on the basis of estimated OR: low (OR ≤ 1.21), medium (1.21 < OR ≤ 2), and high (OR > 2). The risk of OPL was significantly elevated in the 3 risk groups (P for trend <.001).
|Controls, n||Patients, n||OR||95% CI||P|
|P for trend||<.001|
|Low (OR <1.2)||107||62||Ref.|
|Medium (1.2 ≤ OR < 2)||24||25||1.90||0.98-3.68||.058|
|High (OR ≥2)||15||41||4.99||2.51-9.93||<.001|
|P for trend||<.001|
In this study, we used a pathway-based approach to increase our understanding of the complex interaction between genes in the cell-cycle control pathway, smoking status, and the overall risk of OPL.
Overall, in individual SNP analysis only the CCND1 P241P polymorphism was significantly associated with an increased risk of OPL for the variant homozygous genotype. Compared with the common homozygous genotype (GG), the rare homozygous genotype (AA) of CCND1 was associated with a 2.5-fold increased risk of OPL. The per-allele OPL risk for the rare allele A of CCND1 was 1.65 (P = .006). The results for CCND1 were consistent with those reported by Huang et al.4 The CCND1 gene, located at 11q13, encodes proteincyclin D1 and regulates the G1 phase of the cellcycle. The rare allele A of CCND1 has been associated with an increased risk of numerous malignancies, including cancers of the lungs, skin, head and neck, prostate, bladder, kidneys, esophagus, colorectum, endometrium, and larynx.4, 17–20 The alternative splicing of CCND1 mRNA into 2 transcripts, a and b, has been observed to be modulated by the AG alleles at nucleotide 870 within the splicer donor region of exon 4.21 Transcript b, which is preferentially spliced by allele A, exhibits a greater potential to become an oncogene in human malignancies. In the stratified analysis, an increased risk associated with the variant allele of CCND1 was observed in men, younger individuals, and ever smokers. The increased risk of the CCND1 variant allele with colorectal cancer among younger individuals and males was also observed by Huang et al20 in the Taiwanese population.
Because of the complex nature of cancer, single SNP analysis usually provides undetectable low or minor risk associated with cancer as shown by this study. Setting the threshold P value to .5, we will be able to include the minor, low risks in our pathway-based analysis while excluding SNPs without association with risk of OPL. The pathway-based analysis using selected genetic polymorphisms in the cell-cycle pathway showed a promising advantage over the single-candidate gene method. We found that individuals with 3 unfavorable genotypes exhibited a 5.01-fold increased risk of OPL, and those with more than 3 unfavorable genotypes showed a 6.18-fold increased risk, compared with individuals with 1 or 2 unfavorable genotypes. Each additional number of unfavorable genotypes was associated with a 1.79-fold (P for trend = .002) significantly increased risk.
Traditional logistic regression analysis cannot handle the sparseness caused by high-order interaction, ie, empty cells in the contingency table. Unlike logistic regression, CART analysis is an explorative, nonparametric approach that requires no assumption of a genetic model. Zhang and Bonney22 illustrated the applicability of tree-based analyses for genetic association studies. In our study, when compared with genetic polymorphisms, smoking was the most dominant factor in OPL risk. Among ever smokers, CART analysis identified CDK6-P27 and CDK6-CCND1 interactions. Ever smokers with a common homozygote CDK6 genotype and a variant allele-carrying genotype of P27 5′UTR had the highest percentage of OPL cases (75%). Individuals who smoked and carried at least 1 variant allele of CDK6 3′UTR and common homozygous and heterozygous genotypes of CCND1 P241P had the second-lowest percentage of OPL cases (38%). Both CDK6-P27 and CDK6-CCND1 interactions are biologically feasible. Cyclin-dependent kinase 6, CDK6, is located in chromosome region 7q21-q22 and starts its activity in the mid-G1 phase. It is crucial for progression from G1 phase and G1/S phase transition through interaction with CCND1. The complex mix, made up of CDK6/CDK4 with CCND1, regulates the tumor-suppressor protein pRB. Deregulated CDK6 expression has been associated with the dysfunction of cell-cycle control.23 P27 belongs to the KIP family of CDK inhibitors, which negatively regulate the interaction between CCND1 and CDK6/CDK4. Low expression of P27 has been linked to many tumor types and tumor progression.24 The polymorphism of P27 5′UTR is within a U-rich element and necessary for the transcription of P27 mRNA.25 Of note, the high-order gene-gene interactions identified above were found only in ever smokers, and no significant SNPs was identified for never smokers. Therefore, the risk of OPL may be regulated by potential higher-order gene-gene and gene-smoking interactions.
In conclusion, this study assessed a panel of genetic polymorphisms in the cell-cycle control pathway and the OPL risk. We used a pathway-based approach and a multifaceted approach (CART) to analyze high-order gene-gene and gene-smoking interactions. Our results revealed that the alteration of risk of OPL by SNPs in the cell-cycle control pathway can be attributed to smoking status. Therefore, this pathway-based approach has the advantage of identifying the complex relationship between genetic polymorphisms and cancer susceptibility involving multiple factors.