Cyclin D1 Pro²⁴¹Pro (CCND1-G870A) Polymorphism Is Associated with Increased Cancer Risk in Human Populations: A Meta-Analysis

The CCND1 gene encodes a key cell cycle regulatory protein, cyclin D1, which regulates transition from G₁ to the S phase during cell division. High activity of cyclin D1 leads to premature cell passage through the G₁-S transition, resulting in propagation of unrepaired DNA damage and accumulation of genetic errors, therefore leading to selective advantage for abnormal cell proliferation (1). Gene amplification and mRNA and protein overexpression of CCND1 have been characterized and shown in a variety of cancer types, thus providing strong evidence for the oncogenic role of CCND1 (2-4). CCND1 has been recognized as a promising biological marker in predicting tumor behavior (5).

A commonly occurring G-to-A polymorphism at nucleotide 870 (G870A) of CCND1 (CCND1-A870G) has been subject of many case-control association studies of various cancer types in different ethnic populations. CCND1-A870G, which corresponds to codon 241 (Pro²⁴¹Pro), is a silent variant and does not result in an amino acid alteration within the protein sequence. However, interestingly, CCND1-870A allele results in an alternatively spliced transcript of CCND1, called transcript b, which lacks PEST motif containing exon 5. PEST motif is critical for the degradation of cyclin D1; thus, transcript b (A allele) has shown to have a longer half-life than the transcript a (G allele) encoded protein. This highly suggests that individuals with more copies of the CCND1-870A are more likely to bypass the G₁-S checkpoint, thus contributing to cancer development (6).

Discrepancies and lack of an overall risk estimate from increasing number of case-control studies investigating the association of CCND1-A870G with common cancers prompted us to examine all related published literature. Given an altered function of the variant A870 allele and adhering to an established framework of meta-analysis (7), we sought to evaluate the association of CCND1-G870A polymorphism with cancer from combined evidence.

We searched for all studies reporting an association between CCND1-G870A (Pro²⁴¹Pro) polymorphism and cancer risk using PubMed. The following search terms and their combinations were used: “CCND1,” “cyclin D1,” “cancer,” and “polymorphism.” Additional studies were manually searched in the reference list of all studies identified via PubMed search. Here, we have focused on studies, with data on Asians and Caucasians, where sufficient numbers were available for statistical analysis. Data extracted from the selected studies included author, year of publication, country and/or dominant ancestry of the study populations, genotype data, as well as number of cases and controls.

The power of each study was evaluated as probability of detecting an association between CCND1 G870A polymorphism and cancer assuming an odds ratio (OR) of 1.5 (small effect size). The χ² test was used to determine departures of genotypic frequencies from the Hardy-Weinberg Equilibrium (HWE) in control subjects. Data were analyzed using the G*Power statistical program,⁶

Genotype specific risk was examined by considering the GG genotype as reference and calculating summary ORs for the AA and AG genotype. We also evaluated the odds of AA versus AG + GG and AA + AG versus GG, assuming recessive and dominant effects of the variant A allele, respectively. Raw data for genotype frequencies, without adjustment, were used for calculating study-specific estimates of the OR. Pooled ORs (summary estimates) were obtained using either the fixed-effects (Mantel-Haenszel) or random-effects (DerSimonian-Laird) models. The fixed-effects model was used in the absence of heterogeneity (8), whereas the random-effects model was used in its presence (9). Assuming genuine diversity in the results of various studies, the random-effects model incorporates between study variance. Heterogeneity between studies was estimated using the χ²-based Q test (10), the significance of which was set at P < 0.10 (11). Potential sources of heterogeneity were detected using the Galbraith plot (12). Heterogeneity was explored using subgroup analysis (10) with ethnicity (Asians and Caucasians) and cancer types. We have classified the cancer types in nine different categories including breast (13-19), colorectal (20-34), gynecologic [ovarian (35), cervical (36), and endometrial (37)], digestive tract [oral (38-40), esophageal (41-44), and stomach (45, 46)], blood-related [acute lymphoblastic leukemia (47), mantle cell lymphoma (48), and non-Hodgkin's lymphoma (49)], genitourinary [bladder (50-53), prostate (54, 55), and kidney (56)], lung (57-61), head and neck (62-65), and other [pituitary (66, 67), skin (68, 69), and liver (70, 71)] cancers.

A differential magnitude of effect in large versus small studies (publication bias) for genotype contrast was checked using a funnel plot. Publication bias may be absent if the plot resembles a symmetrical inverted funnel in which smaller, less precise, and more numerous studies have increasingly large variation in the estimates of their effect size (72). Funnel plot asymmetry was explored by plotting precision (1/SE) against effect size estimates (OR in log scale). Publication bias was statistically evaluated with Egger's regression asymmetry test, which detects whether the intercept deviates significantly from zero in a regression of the standardized effect estimates against their precision (73).

Using PubMed, we have identified 62 articles, which investigated the association of CCND1-G870A with cancers in Asians and Caucasian populations. Three articles (74-76) were excluded due to overlapping subjects. One article (28) examined the association in independent populations of Asians and Caucasians and thus was treated them as two separate studies. Four articles (21, 44, 45, 63) investigated two types of cancers, but only one cancer type, with the largest population, was selected from each article. The remaining studies investigated one cancer type. Tumor types containing less than three individual studies were combined into the other cancer group. Of the nine tumor types, colorectal cancer was the most studied, constituting over a quarter (26.7%) of all the studies (Table 1).

Overall, 60 studies provided genotype data from 18,411 cancer cases and 22,209 non–cancer controls. We have carried out a meta-analysis of CCND1-G870A polymorphism in overall, ethnic group, and cancer type under various genetic models. Significant association of this polymorphism with cancer risk has been consistently observed in many categories, including overall (Supplementary Fig. SA), ethnic groups, and 44.4% of the cancer types studied (Table 2A; Fig. 1). However, a high level of heterogeneity was also detected in the overall, ethnic group, and mainly in colorectal, digestive tract, head and neck, and lung cancer studies (Fig. 1). To identify which of the 60 studies that may be sources of heterogeneity, we used the Galbraith plot and accordingly identified 12 studies (20, 22, 23, 26, 37, 38, 41, 43, 44, 46, 57, 62) as the main contributors (Table 1; Fig. 2).

We excluded the 12 outlier studies and repeated the meta-analysis of the remaining 48 studies (16,399 cases and 19,818 controls) with reduced heterogeneity under various genetic models (Fig. 1). Accordingly, in the overall analysis, we have detected 1.1- to 1.2-fold increased risk associated with developing cancer (Table 2B). Of the 60 studies, 34 (56.7%) and 26 (42.6%) investigated Caucasian and Asian subjects, respectively. Reanalysis of 28 homogeneous Caucasian studies (10,313 cases and 11,625 controls) have shown 1.1- to 1.2-fold increased risk of developing cancer. Similarly, reanalysis 20 homogenous Asian studies (6,086 cases and 8,193 controls) have shown 1.1- to 1.3-fold increase risk of cancer.

We have also re–evaluated the risk associated with cancer types after removing the heterogeneous studies detected in colorectal, digestive tract, head and neck, lung, and gynecologic cancers. The remaining four cancer types (breast, blood-related, genitourinary, and other) were homogeneous across all comparisons. Table 2A shows that AA homozygotes were associated with significantly increased risk of developing cancer in four of the nine tumor types: breast (OR, 1.12; P = 0.04 in 4,718 cases and 5,183 controls), blood-related (OR, 1.62; P < 0.0001 in 1,336 cases and 1,131 controls), genitourinary (OR, 1.51; P = 0.0004 in 1,089 cases and 1,675 controls), and other cancers (OR, 1.31; P = 0.03 in 789 cases and 1,815 controls). Reanalysis following removal of outlier studies increased the risk to significance in two cancer types; colorectal (OR, 1.15; P = 0.02 in 3,584 cases and 5,500 controls) and head-neck (OR, 1.51; P = 0.02 in 642 cases and 541 controls; Table 2B).

Statistically significant association was also observed in genitourinary cancer (OR, 1.5; P < 0.0001) and blood-related cancer (OR, 1.5; P = 0.0002) under the recessive (AA versus AG + GG) model. Under the dominant (AA + AG versus GG) model, significant associations were also detected in four cancer types including blood-related (OR, 1.27; P = 0.006), colorectal (OR, 1.16; P = 0.004), breast (OR, 1.12; P = 0.02), and head and neck (OR, 1.30; P = 0.04) cancers (Table 2A and B).

ORs in the AA genotype were generally higher than those in the AG genotype in some of our analyses. To formally evaluate dosage effect within different categories, we applied the Cochran-Armitage tests for trend before and after removing the outlier studies (Table 2A and B). Significant dosage effects were observed in overall (P = 0.00001) and Asians (P = 0.00001); however, trend was not significant in Caucasians after removing the heterogeneous studies (20, 22, 26, 38, 41, 57). Among the cancer types, significant dosage effects were observed in blood-related (P = 0.0001), genitourinary (P = 0.0001), and other cancers (P = 0.001; Table 2A) before excluding the outliers and none were significant after outlier removal (Table 2B).

Table 1 shows the statistical power of all 60 studies ranging from 8.4% to 99.9%. Of the 60, 11 (18.0%) studies (15-19, 21, 24, 32, 33, 49, 59) have ≥80% power. These adequately powered studies include four cancer types (breast, blood-related, colorectal, and lung). The studies with the highest statistical powers were concentrated in breast cancer where 5 (15-19) of the 7 studies had powers ranging from 80.1% to 99.8%. Statistical power of the 12 studies considered to be sources of heterogeneity is <80%, indicating suboptimal sample sizes. Under the AA versus GG contrast, individual ORs of these studies ranged from extreme susceptibility (2.66-6.20) to very protective (0.36-0.74). In contrast to the other studies that reported susceptibility of the A allele, 3 (23, 38, 46) found that it is the G allele that showed increased risk for cancer. One study had a 6.0% population admixture (57) and another obtained their controls from a different population (26), suggesting selection bias. Genotype distribution of the control group in 5 studies 6.7% (18, 19, 24, 54, 57) deviated from the HWE, indicating possible potential biases in the selection of controls (Table 1).

Influence of each study on the pooled ORs was examined by repeating the meta-analysis omitting each study one at a time (77). This procedure did not change the pooled ORs supporting the robustness of our findings.

Funnel plots are shown in Fig. 3 for both AA versus GG and AA versus AG + GG comparisons. Arrangement of data points shows no evidence of asymmetry suggesting absence of publication bias. Formal evaluation using Egger's regression asymmetry tests generated intercept values of 0.65 for the homozygous model (t = 2.31, P = 0.71) and 28.6 for the recessive model (t = 5.67, P = 0.73).

Of the 60 studies, more than one-third (38.3%) investigated the modifier effects of environmental risk factors such as smoking, alcohol consumption, and diet. Fourteen studies discussed the influence of smoking and one included effects of alcohol consumption. The remaining 8 studies covered the combined effects of two or all of the above-mentioned risk factors.

Based on 60 published genetic association studies, including over 40,000 subjects, our meta-analysis provides evidence that the CCND1-G870A polymorphism is associated with risk of developing cancer. Compared with low cyclin D1 activity associated GG homozygotes in the population, AA homozygotes (high cyclin D1 activity) and AG heterozygotes (medium cyclin D1 activity) were associated with 1.1- to 1.2-fold increased risk of developing cancer in the overall comparisons. This risk in Caucasians was also 1.1- to 1.2-fold, whereas in Asians a slightly increased risk (1.1- to 1.3-fold) was detected. The most significant and highest risk effects were shown for AA homozygotes in overall (OR, ∼1.2-fold), Caucasians (OR, ∼1.2-fold), and Asians (OR, ∼1.3-fold). Risk for all cancer types was found to be increased, but statistical significance was obtained in only four types. Although risk associated with breast (1.1-fold), colorectal (∼1.2-fold), and other (∼1.3-fold) cancers were significant; P values were on the lower confidence levels. However, we have found that the increased risk associated with genitourinary (∼1.5-fold) and blood-related (∼1.6-fold) cancers was highly significant. Heterozygous AG individuals were found to be at increased risk in overall, ethnic groups, as well as breast and colorectal cancers.

Recessive or dominant effects of the A allele have been controversial with discrepant findings among published studies. Our findings showed that the dominant effects appear to prevail in the overall, ethnic group, and cancer type analyses including breast, colorectal, blood-related, and head and neck cancers. However, recessive effects were also detected in overall and Asians as well as genitourinary and blood-related cancers. Dominant effects of the CCND1 polymorphism were reported in lung cancer (78), whereas recessive effects have been shown in bladder (50, 53) and prostate (55) cancers. Both dominant (79) and recessive (25, 26) effects have been reported in colorectal cancer. Although our meta-analysis was done fairly with large sample size, the specific effects for each cancer type need further validation.

Pooled OR in blood-related cancers may well be attributed to one study (49) with high statistical power (99.4%) and weighted contribution of 84.3%. The other two (47, 48), with a total weighted contribution of 15.7%, have powers of 48.6% and 9.5%, respectively. By contrast, effect in genitourinary cancers may be attributed to four studies (52, 53, 55, 56) with statistical powers that ranged from 57.6 to 64.6 and a total weighted contribution of 84.9%. The remaining three studies with total weighted contribution of 15.1% have considerably less power ranging from 12.3% to 30.6%.

We have also shown that the gene dosage effects of the A allele places greater risk on individuals with the AA homozygous genotype when compared with the AG heterozygous individuals. This trend of increasing susceptibility with increasing presence of the A allele has been also reported in various cancer types (28, 56, 61, 65).

Between-study heterogeneity, selection bias, and false-positive discovery may be attributed to many factors including the selection of publications, differences in population characteristics and sample sizes, lack of comparable measures of phenotype, and deviations of allelic distributions from the HWE. We have addressed all these issues accordingly. (a) Heterogeneity among studies and pooled ORs. We have addressed heterogeneity in our studies using a random-effects framework, which evaluates the variation in the pooled ORs based on individual ORs of each study. Accordingly 12 of 60 studies have shown significant heterogeneity, which influenced our findings. Reanalyses of the categories, after removing these studies, has provided more homogeneous and more significant associations with cancer risk including greater precision among the tumor types. (b) Selection bias due to inappropriate matching of cases and controls. Lack of proper matching of controls to cancer subjects is unlikely in our study given that about 60% of the studies in our meta-analysis were age-matched, 20% were gender-matched, and the remaining 20% were matched by ethnicity. Furthermore, 18.3% of the studies were matched using the two or all of the above-mentioned matching criteria. (c) HWE. Control groups in 7 (11.5%) studies showed some evidence of departure from the HWE. Excluding these studies from the meta-analysis, however, had no real effect on the pooled ORs in overall, ethnic groups, and cancer type analyses. (d) Publication bias. Funnel plot analysis has shown the absence of publication bias under both homozygous and recessive genetic models. (e) Power and multiple corrections. False-positive findings are a possibility given the multiple comparisons performed and borderline significance of some associations. We have applied correction measures to minimize false-positive outcomes. More than 88% of the studies in this meta-analysis were underpowered when considered individually. Combining studies here, however, have achieved sufficient power to obtain reliable summary estimates for risk of CCND1-G870A polymorphism on cancer development.

Although magnitude of the associated risk was low, this polymorphism, however, affected a large portion (AA, 25.0%; AG, 50.0%) of the population. The risk effect shown is very small to be considered clinically useful at this stage. However, interaction of CCND1-A870G polymorphism with other genetic variants and environmental factors has been shown to be associated with further increase in cancer risk. For example, studies have shown an increased risk associated with interaction of CCND1-A870G and XPD in digestive tract cancers with a significant estimate of OR of 7.1 [95% confidence interval (95% CI), 4.0-12.5; ref. 57]. Recently, our own studies have also shown increased breast cancer risk associated by the interaction of CCND1-A870G and COMT-Met^108/158Val with OR of 2.2 (95% CI, 1.5-3.3) and 1.73 (95% CI, 1.1-2.8) in two independent Caucasian populations (17, 80).

We also identified several environmental risk factors from the studies that have suggested significant modifying risk effects of the CCND1 G870A polymorphism. ORs of 2.2 (95% CI, 1.2-4.1) and 2.7 (95% CI, 1.3-5.7) among smokers (29, 44) and 2.2 (95% CI, 1.2-4.2) and 2.07 (95% CI, 1.20-3.59) among drinkers (29, 32) have been reported. X-ray exposure has placed individuals with the AA genotype at significant high risk with an OR of 1.6 (95% CI, 1.2-2.1) when compared with GG individuals (59). In contrast, increased consumption of antioxidants has been found to have significant protective effects with an OR of 0.3 (95% CI, 0.2-0.7) in individuals with the AG genotype (13).

In this study, we showed that the CCND1 G870A polymorphism confers susceptibility to cancer development. Future investigations of the cyclin D1 G870A polymorphism warrant close attention to design and methodologic features. Well-designed epidemiologic studies would help illuminate the complex landscape of cell cycle and cancer risk. Nevertheless, our summary estimates support the hypothesis that the CCND1 870A variant may represent a low-penetrant cancer allele in the population.

No potential conflicts of interest were disclosed.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank Cheryl Crozier for research support, Sevtap Savas for insights on splice patterns of CCND1, Joseph Beyene for preliminary insights on methodology, and Kerime Arisan and Lincoln Lam for assistance in organizing the tables.