Associations of CYP1A1 gene polymorphisms and risk of breast cancer in Indian women: a meta-analysis

Abstract Reported associations of CYP1A1 polymorphisms with breast cancer have been inconsistent. In this meta-analysis examining breast cancer associations of three CYP1A1 polymorphisms (M1, M2 and M4) among Indian women may yield information that may be of clinical and epidemiological use for this particular demography. We searched MEDLINE using PubMed and Embase for association studies. From seven published case-control studies, we estimated overall associations and applied subgroup analysis to explore differential effects. All three polymorphisms exhibited overall increased risk, significant in M1 (OR 1.61–1.65, p = 0.04) and M4 (OR 2.02–3.92, p = 0.02–0.04). Differential effects were observed only in the M1 polymorphism where M1 effects were significant in South Indians (OR 2.20–4.34, p < 0.0001) but not the North population, who were at reduced risk (OR 0.64–0.77, p = 0.03–0.55). These populations were not materially different in regard to M2 and M4 as did the women stratified by menopausal status. In this meta-analysis, M1 and M4 effects may render Indian women susceptible, but may be limited by heterogeneity of the studies. Differential effects of the M1 polymorphism in breast cancer render South Indians susceptible compared to those in the North.


Introduction
Breast cancer is a leading cause of death among women in India [1]. The multifactorial nature of this disease is marked by the synergy of interaction among various environmental and genetic factors [2,3]. One genetic factor is the CYP1A1 protein, encoded by CYP1A1 gene, an interesting candidate that might influence susceptibility to breast cancer risk for reasons mentioned by Chen et al. [4]. CYP1A1 gene polymorphisms have been extensively studied, especially in relation to breast cancer susceptibility [4][5][6][7][8][9][10][11][12][13]. The CYP1A1 gene, located at 15q22-q24, comprises seven exons and six introns spanning 5810 base pairs [14]. Of the four common polymorphisms of the CYP1A1 gene identified, we examine three in the Indian population: M1 (rs4646903: T3801C, giving rise to a MspI restriction site in the 3'-noncoding region [15], M2 (rs1048943: A2455G), resulting in an amino acid change at codon 462 from isoleucine to valine within the heme-binding domain of exon 7 [16]; and M4 (rs1799814: C2453A), resulting in an amino acid substitution of threonine with asparagine at codon 461 [17]. Of the three polymorphisms, M4 is very rare.
Though the functional significance of variant CYP1A1 genotypes is unclear [4], studies of CYP1A1 in cultured human lymphocytes showed significantly elevated levels of inducible enzyme activity among M2 genotypes compared with the wild-type genotype [18]. M2 alleles appear to be associated with CYP1A1 inducibility at the level of transcription followed by 3-fold elevation in aryl hydrocarbon hydroxylase enzyme activity [19]. The M1 allele was also reported to be more readily inducible than the CYP1A1 wild-type allele [20,21]. A number of meta-analyses have appeared addressing associations of CYP1A1 polymorphisms with breast cancer. To our knowledge, no metaanalysis, including the four we chose to compare our study with, has focused on a population of a single country in Asia. Yet three of them examined CYP1A1 associations with breast cancer in Asian populations using subgroup analysis. Yao et al. [6] and Chen et al. [4] did this among Asians in general and east-Asians, respectively. Still these populations were a heterogeneous mix of different Asian nationalities. The Sergentanis et al. [5] examination of the Chinese population was more homogeneous. Still none of these meta-analyses have addressed CYP1A1 associations with breast cancer in the Indian population. Given that breast cancer is an overriding epidemiological issue in India [9] and that reports do not necessarily agree with each other, we conducted a meta-analysis of published case-control studies to examine the effect of CYP1A1 genetic polymorphisms on breast cancer risk among Indian women.

Selection of studies
Two strategies were used in searching MEDLINE using PubMed and Embase. In the first, search terms were broad such as "DNA polymorphism", "breast cancer" and "xenobiotic". In a separate second search, search terms were more specific: "CYP1A1" and "cytochrome P450 1A1". The broad and specific search terms were combined with "breast cancer" and "Indian" to find association studies as of 14 April 2015. Studies were eligible if they had genotypic data with a case-control design restricted to English. Figure 1 outlines our study selection process in a flowchart following PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines [22]. The two search strategies yielded a total of 31 citations, 23 of which were screened for not fulfilling our inclusion criteria. Excluding a duplicate article, full texts of the remaining seven were obtained which were judged as eligible; as such, they were included in the meta-analysis. We manually searched the reference sections of each of these studies for additional articles. Syamala et al. [7] provided sporadic and familial data thus were treated as separate studies. In all, this meta-analysis had a total number of eight studies from seven publications [7][8][9][10][11][12][13].

Data extraction
Two investigators independently extracted data and reached consensus on all the items. The following information was obtained from each publication: first author's name, published year, city and region in India, state of controls, matching criteria, sample source, genotype data, number of cases and controls. We also calculated frequencies of the variant allele as well as deviations of the controls from the Hardy-Weinberg equilibrium (HWE).

Quality assessment of the studies
Quality of the studies was evaluated with use of the Newcastle-Ottawa Scale (NOS) [23]. Each study was assessed based on three broad perspectives: selection, comparability, and exposure with a score ranging from 0 to 9. Study quality scores for high, medium and poor were ≥ 7, 4-6 and < 4, respectively.

Meta-analysis
Risks (odds ratios [ORs] and 95% confidence intervals) of breast cancer in three CYP1A1 polymorphisms (M1, M2, and M4) were estimated for each study. Pooled ORs were calculated for the following genetic models using variant (var) and wild-type (wt) genotypes: (i) homozygous (var-var, genotypes compared with the wt-wt), (ii) recessive (var-var vs. wt-var + wt-wt) (iii) dominant (var-var + wt-var vs. wt-wt) and (iv) co-dominant genetic model (var vs. wt). To compare effects on the same baseline, we used raw data for genotype frequencies to calculate study-specific estimates of the OR. Pooled ORs were obtained using either the fixed [24] (in absence of heterogeneity) or random [25] (in its presence) effects models. Heterogeneity between studies was estimated using the I 2 -based Q test [26] and quantified with the I 2 statistic which measures degree of inconsistency among studies [27]. Given the low power of the heterogeneity test [28], significance threshold was set at p = 0.10. Sources of heterogeneity were examined with subgroup analysis [27]. Sensitivity analysis, which involved omitting one study at a time and recalculating the pooled OR, was also used to test for robustness of the summary effects. Availability of analyzable data from the South and North Indian subpopulations from the included studies allowed this type of geographical subgrouping. Supported by precedent literature [29,30], the probability of differential risk associations between the South and North Indian subpopulations warranted testing for the presence of interactions; wherein the generated p values were subjected to the Bonferroni correction. Data were analyzed using Review Manager 5.3 (Copenhagen: Nordic Cochrane Centre, Cochrane Collaboration, 2014) and SigmaStat 3.5 (Systat Software, San Jose, CA). Significance was set at a p-value of  0.05 throughout except in heterogeneity estimation. Publication bias was not investigated because of the low sensitivity of the qualitative and quantitative tests when the number of studies is lower than ten [31]. Table 1 shows features of the included studies, published between 2005 and 2011. Five studies had Southern subjects [7,8,[11][12][13] compared to two from the North [9,10]. There were more hospital-based [7,[10][11][12] than population-based [8,9,13] studies. Two studies made no mention of matching [11,13]. Methodological quality of the studies indicates that they were moderate with most studies having a score of 6-7. Table S1 summarizes genotypic features of the component studies in the meta-analysis. In the M1 polymorphism association studies (1151 cases/1527 controls), populations of South and North India were examined in four [7,[11][12][13] and two [9,10] papers, respectively. These Indian geographical regions were likewise examined in four [8][9][10][11] and two [12,13] papers on the M2 polymorphism (1039 cases/1022 controls). The two studies [9,10] and a single one [11] on the M4 polymorphism (631 cases/611 controls) had North and South Indian subjects, respectively. Control subjects deviated from the HWE in the following: (i) two studies [9,12] in M1; and (ii) one study for M2 [8] and two for M4 [10,11]. For both M1 and M2, the mean frequency of the minor allele was greater from the Northern populations, though this was significant only for M1 (M1: North x ̅ = 0.32 ± 0.08 standard deviation (SD) versus South x ̅ = 0.18 ± 0.05 SD, p = 0.03; M2: North x ̅ = 0.18 ± 0.06 SD versus South x ̅ = 0.16 ± 0.11 SD, p = 0.81).   [9,10,12]. In the M4 polymorphism, premenopausal (168 cases/197 controls) and postmenopausal (118 cases/139 controls) subjects were all North Indians on account of two studies [9,10], one of which deviated from the HWE [10]. In the geography and menopausal status analyses, the probability of differential risk associations between North and South India as well as premenopausal and postmenopausal women warranted testing for presence of interactions.

Overall findings
In Table 2, all three polymorphisms showed overall increased risk across all comparisons. Significance was observed in the M1 (OR 1.61-3.43, p = 0.007-0.04) and M4 (OR 2.02-3.92, p = 0.02-0.04) polymorphisms. The significant homozygous and recessive M4 effects (OR 3.81-3.92, p = 0.02) were obtained in the absence of heterogeneity (pheterogeneity = 0.79-0.84, I 2 = 0%). All other effects in all the polymorphisms were heterogeneous, particularly high for M1 (pheterogeneity < 0.00001, I 2 = 85-96%). Figure 2 exemplifies the significant effects of M1 under conditions of high heterogeneity. Diamond denotes the pooled odds ratio (OR). Squares indicate the OR in each study, with square sizes directly proportional to the weight contribution (%) of each study. Horizontal lines represent 95% confidence intervals (CI). The Z test for overall effect indicates significance if p < 0.05. The chi-square test indicates heterogeneity (p < 0.10) warranting use of the random-effects model.

Subgroup analysis
3.3.1. Geography Table 3 shows subgroup associations stratified by geography in the M1 polymorphism. Mainly, the North Indians were protected (OR 0.64-0.77, p = 0.03-0.55), but not their Southern counterparts, who were at significant risk (OR 2.20-4.34, p < 0.0001). The tests of interaction highlight these contrasting effects, with significance generated in all (p = 0.00002-0.02) but the recessive model. Table 4 compares increased risk effects between the two Indian populations in the M2 polymorphism which did not materially differ given the non-significance of the p values of interaction (p > 1).

Menopausal subgroup
In all three polymorphisms, differential effects between premenopausal and postmenopausal women were not significant (Tests for Interaction: p > 1) where both subgroups generally exhibited increased risk effects (Tables 3-4). These effects were significant in the codominant model of M2 and M4 among postmenopausal women (Tables 4 and 5), suggesting this subgroup to be at particular risk.
Among premenopausal women, M2 effects indicated codominant null and protective dominant effects suggesting variability of the results (Table 4).

Sensitivity analysis
Direction of effects of the pooled ORs in M1 and M2 polymorphisms remained unchanged at increased risk with sensitivity treatment indicating that they were not materially altered suggesting robustness of our findings.

Discussion
With a combined sample size of 5981 (2821 cases and 3160 controls) in three CYP1A1 polymorphisms, this meta-analysis points to evidence of overall associations between the M1, M2 and M4 polymorphisms and breast cancer among Indian women. However, this association was limited by between study heterogeneity. Its sources were examined with subgroup analysis. Stratifying the studies according to geography revealed the North India subgroup to be more heterogeneous than their Southern counterparts.
Our meta-analysis differs from these three in that we examined the M1, M2 and M4 polymorphisms in the Indian population only enabling discernment of differences in effects between North and South Indians. The Indian population is the second largest in the world with diverse social, cultural, linguistic and biological features spread across a wide topography [32].
A fourth meta-analysis that focused on the M1 polymorphism addressed its effects in breast cancer among South Indians as a subgroup of four comparisons [33]. These effects indicated significant increased risk for up to 4-fold in all the genetic models they used, which was similar to our findings from five studies. The difference between that meta-analysis and ours was that our statistical comparison between the North and South Indian populations resulted in significant differential effects, protecting the former and placing the latter at risk.
The combination of population stratification leading to multiple comparisons and borderline significance of some associations open the possibility of false-positive findings [34]. Avoiding this problem warrants that the subpopulation be defined in terms of marker-allele frequencies and factors that influence disease such as geography. In this meta-analysis, we addressed population stratification in terms of differential allele frequencies between two distinct regions of India, which intensified significance.
Still, our study has been limited by the following: (i) given the multiplicity of comparisons for different genetic models, geography and menopausal status subgroups, and the unavoidable flexibility of choosing and defining the correlates, associations may have been detected by chance alone; (ii) low sample sizes of the component studies which translate to weak statistical power. But then, the point of meta-analysis is to combine studies resulting in increase of statistical power; (iii) that most of the studies were hospital-based may preclude extrapolation of our findings to the general population; (iv) heterogeneity of most of our findings, (v) deviation of some studies from the HWE which may have biased summary outputs and point to methodological weaknesses, such as biased selection of subjects, genotyping errors population stratification [35]. However, this was addressed with sensitivity analysis. (vi) Our study was based on unadjusted data, as the genotype information stratified for the main confounding variables was unavailable in the original papers. In addition, addressing of the confounding factors varied across the studies.
Yet, despite these weaknesses, the strengths of our meta-analysis include: (i) statistical homogeneity of the M4 homozygous and recessive effects imply combinability of the component studies; (ii) consistency of overall findings across the genetic models; (iii) because we confined our meta-analysis to a single country, we were able to delineate differential summary effects between populations of North and South India; (iv) controls were uniformly defined (cancer-free, healthy), minimizing non-differential misclassification bias.
The findings we report here highlight two points. The first is that individual epidemiological studies may lack statistical power or may have other interfering factors which prevent the unmasking of overall associations. Likewise, geographically local associations with specific alleles may exist which are due to linkage disequilibrium with other disease modifying alleles and are not reproduced in other populations. The second point is a caution with regard to drawing conclusions from individual studies. We report results that were not apparent from an examination of individual studies. We find that meta-analysis is a useful tool to examine broad trends in CYP1A1 associations with breast cancer in Indian populations, and may avoid possible misleading conclusions based on only single-population studies.
The pathways of carcinogen metabolism are complex, mediated by the activities of multiple genes in concert with CYP1A1 such as GSTM1 [12]. It is conceivable that breast cancer risk related to any one locus will be small because gene-gene as well as gene-environment interactions are likely to operate. Individual studies in our meta-analysis have addressed gene-gene [7,12] and geneenvironment [8,10,13] interaction and provided data which we excluded for concern with multiple testing issues.

Conclusion
To our knowledge, this is the first meta-analysis that details the association of CYP1A1 polymorphisms with breast cancer among Indian women. This focus on one country allowed us to examine differences with use of geography and menopausal subgroups. Because our meta-analysis summary effects are relevant only to this demography, we provide an epidemiological profile of Indian women's susceptibility to breast cancer on account of the CYP1A1 polymorphisms. Additional welldesigned studies based on sample sizes commensurate with detection of small genotypic risks should allow more definitive conclusions about the association of CYP1A1 polymorphisms and breast cancer among Indian women.