Polymorphisms of matrix metalloproteinases 1, 2, 3 and 9 and susceptibility to lung, breast and colorectal cancer in over 30,000 subjects

Abstract

A variety of susceptibility genes have been associated with cancer but definitive conclusions have been difficult to draw partly hampered by the small number of subjects in each study. We undertook a comprehensive genetic meta-analysis of all matrix metalloproteinase (MMP) genes investigated using an allelic-association case–control model in the 3 major cancers of lung, breast and colorectal cancer. Electronic databases were searched until and including July 2008 for any MMP genetic association study in lung, breast and colorectal cancer. Odds ratio (OR) and 95% confidence intervals (CI) were determined for each gene disease association using fixed and random effect models. Twenty-five studies addressing 5 polymorphisms in 4 genes were analyzed among 30,651 individuals (15,328 cases and 15,253 controls). The MMP-1 nt-1607 polymorphism was significantly associated with colorectal cancer in both the dominant (OR, 1.66; 95% CI, 1.14–2.42; p = 0.008) and recessive (OR, 1.59; 95% CI, 1.15–2.20; p = 0.005) models. MMP-2^1306C→T (OR, 0.53; 95% CI, 0.40–0.72; p < 0.0001) and 735^C→T (OR, 0.65; 95% CI, 0.53–0.79; p < 0.0001) were significantly associated with protection against lung cancer. No association was found with the MMP 1, 2, 3 or 9 polymorphisms with breast cancer, MMP-1, 3 or 9 with lung cancer or MMP-2, 3 or 9 with colorectal cancer. There may be a genetic influence in the development of colorectal and lung cancer. Subjects with the MMP-1 nt-1607 polymorphism have an increased association with colorectal cancer. Those homozygous for either the MMP-2/1306T or 735T allele may be at reduced risk of lung cancer, although the evidence base is small. © 2009 UICC

Matrix metalloproteinases (MMPs) are a family of zinc-dependent proteolytic enzymes1 involved in the degradation of various components of the extracellular (ECM) and basement membranes. MMPs can be split into 5 different groups depending on their substrate specificity and domain structure. These include collagenases (MMP-1, MMP-8 and MMP-13 cleaving collagen types 1, 2 and 3), gelatinases (MMP-2 and 9 which digest gelatin and collagen types 4, 5 and 9), stromelysins (MMP-3, MMP-10 digest a number of ECM molecules and MMP-11 digests serpins), matrilysins (MMP-7 and MMP-26 also digesting a number of ECM molecules) and membrane-type MMPs (MMP-15, 16, 17 and 24 and 25 activating MMP-2 and MMP-14 with collagenolytic activity on collagen types 1, 2 and 3).

MMPs have been found to play a role in tumor initiation and growth, stimulation of angiogenesis, activation of growth factors and receptors, apoptosis and metastasis of tumors.2–6 A number of studies have demonstrated the role of MMPs in breast cancer2 and colorectal cancer.7 Further studies have shown that inhibition of MMPs can reduce metastatic spread8–10 and influence lung cancer.11 Colorectal cancer patients with overexpression of MMP-7 have a 3-year survival rate of ∼76%, whereas those without overexpression have a 3-year survival rate at 36%12; while 5-year survival rates for lung cancer are lower in patients with MMP-7-positive tumors.13

A variety of single nucleotide polymorphisms (SNPs) in MMPs have been studied in the 3 main cancers with disparate results, partly because of insufficient size of any single study. To boost the power of these analyses and obtain more reliable odds ratios (OR), we sought to conduct a comprehensive genetic meta-analysis of all the MMP polymorphisms studied to-date in lung, breast and colonic cancer.

Material and methods

Data sources

Electronic searches using PubMed, Embase, Google Scholar and Yahoo were used to identify all published case–control studies evaluating any MMP gene in lung, breast or colorectal cancer in humans published until and including July 2008. Letters and abstracts were included in the searches. The Medical Subject Headings and text words used for the search included lung, breast, colorectal, colon cancer and matrix metalloproteinase in combination with polymorphism, gene, genotype or mutation. Search results were limited to human. All languages were searched. The references of all identified publications were hand-searched for additional studies and the PubMed option related articles was used to examine all relevant articles.

Study selection

Selection criteria included case–control studies, where cancer was analyzed as a dichotomous trait. Studies were selected if there was a confirmed histopathological diagnosis of lung, breast or colorectal cancer. All ethnic backgrounds were included in any language and were translated when necessary. Studies were excluded if (i) controls had a previous history of cancer and (ii) genotype frequency was not reported. For studies with more than 1 publication describing results among the same or overlapping groups of patients or controls, only the largest of the available published data set was included.

Statistical analyses

Data were analyzed using software for preparing and maintaining Cochrane reviews (Review Manager v4.2.8, Cochrane Collaboration, http://www.cc-ims.net/RevMan) and Comprehensive Meta Analysis v2.2.023 (Biostat, http://www.biostat.org). To determine the strength of genetic association, a pooled OR was calculated for each gene variant using fixed and random effects models, in addition to the calculation of 95% confidence intervals (CI). Fixed-effects summary ORs were calculated using the Mantel-Haenszel method,14–16 and the DerSimonian and Laird method was used to calculate random-effects summary ORs.17 The frequencies of at-risk genotypes were compared between cases and controls for each SNP analyzed. Tests for heterogeneity were performed for each meta-analysis with significance set at p ≤ 0.05.18 Publication bias was assessed using funnel plots, while the Egger regression asymmetry test was conducted for each SNP with 2 or more publications.19

The proportion of lung, breast or colorectal cancer cases in the population that could be attributed to a particular gene variant (population-attributable risk [PAR]) was estimated as follows:

• (1)

For this calculation, the fixed effects model was used and the prevalence of exposure was estimated as the genotype frequency among control subjects.

Results

From 1,209 papers identified in our primary search, 43 were selected as potentially relevant articles. Following examination of the complete texts, 13 did not meet our inclusion criteria leaving 30 candidate-gene case–control studies in which the presence or absence of lung, breast or colorectal cancer was analyzed in a dichotomous manner. In total, 16 polymorphisms in 6 genes were identified. Only genes with 2 or more publications on 1 SNP were included in our analyses, leaving a total of 25 publications addressing 5 polymorphisms in 4 genes (Fig. 1).

Figure 1. Inclusion algorithm.

From the 4 genes analyzed in detail (15,328 cases and 15,253 controls), the mean number of studies per candidate gene was 2.5 for breast cancer, 4.25 for colorectal cancer and 2.8 for lung cancer. Ten of the 13 meta-analyses had more than 500 cases, and 11 had a total participant size of greater than 1,000. The table summarizes the genotypic ORs for the polymorphisms and modes of inheritance evaluated in lung, colorectal and breast cancer (Table I).

Table I. Summary Table of MMPs and Lung, Colorectal and Breast Cancer

Gene (No. of studies)	Polymorphism	Genetic model	n Studies	n Cases	n Controls	pHet	OR (95% CI)	p- value (overall effect)
Lung cancer
MMP-1	nt-1607	Dominant	4	4,465	3,487	0.4	1.09 (0.99–1.21)	0.09
		Recessive	4	4,465	3,487	0.0003	1.12 (0.85–1.47)	0.42
MMP-2	1306C/T	Dominant	3	1,641	1,719	0.07	0.53 (0.40–0.72)	<0.0001
		Recessive	3	1,641	1,719	0.62	0.63 (0.37–1.06)	0.08
MMP-2	735C/T	Dominant	2	859	867	0.42	0.65 (0.53–0.79)	<0.0001
		Recessive	2	859	867	0.89	0.88 (0.56–1.39)	0.59
MMP-3	nt-1171	Dominant	3	2,430	2,023	0.81	1.03 (0.89–1.20)	0.66
		Recessive	3	2,430	2,023	0.5	0.97 (0.84–1.12)	0.71
MMP-9	1562C/T	Dominant	2	333	440	0.38	1.03 (0.73–1.45)	0.86
		Recessive	2	333	440	0.04	0.77 (0.01–60.82)	0.91
Colorectal cancer
MMP-1	nt-1607	Dominant	4	539	765	0.96	1.66 (1.14–2.42)	0.008
		Recessive	5	599	929	0.07	1.59 (1.15–2.20)	0.005
MMP-2	1306C/T	Dominant	3	379	460	0.01	0.97 (0.49–1.93)	0.81
		Recessive	3	379	460	0.66	1.37 (0.61–3.11)	0.44
MMP-3	nt-1171	Dominant	4	473	803	0.15	0.93 (0.50–1.71)	0.81
		Recessive	3	413	639	0.005	1.29 (0.66–2.51)	0.45
MMP-9	1562C/T	Dominant	5	700	963	0.27	0.9 (0.68–1.17)	0.43
		Recessive	4	575	836	0.35	0.6 (0.15–2.41)	0.47
Breast cancer
MMP-1	nt-1607	Dominant	2	231	260	0.47	1.16 (0.76–1.77)	0.49
		Recessive	2	231	260	0.21	1.08 (0.66–1.76)	0.75
MMP-2	1306C/T	Dominant	4	1,590	1,653	<0.0001	0.66 (0.40–1.07)	0.09
		Recessive	4	1,590	1,653	0.3	0.95 (0.54–1.66)	0.86
MMP-3	nt-1171	Dominant	2	586	603	0.11	0.94 (0.53–1.65)	0.83
		Recessive	2	586	603	0.08	0.71 (0.38–1.33)	0.29
MMP-9	1562C/T	Dominant	2	1,042	1,046	0.6	1.07 (0.88–1.30)	0.50
		Recessive	2	1,042	1,046	0.17	1.03 (0.16–6.75)	0.98

Lung cancer

MMP-1 nt-1602

Four studies were identified that investigated the MMP-1 nt-1602 polymorphism and its association with lung cancer.20–23 No significant association was observed in either the dominant (OR, 1.09; 95% CI, 0.99–1.21; p = 0.09) or recessive (OR, 1.12; 95% CI, 0.85–1.47; p = 0.42) model. Significant interstudy heterogeneity was observed in the recessive model (p_het = 0.0003). The funnel plot for the dominant model was symmetrical but the Egger's test showed significance (p = 0.034) suggesting a possibility of publication bias. With a recessive model, the funnel plot was not symmetrical and Egger's test was just significant (p = 0.045).

MMP-2 nt-1306C/T

Three studies investigated the association between the MMP-2 nt-1306C/T polymorphism and lung cancer.24–26 A significant association that showed protection against lung cancer was observed in the dominant CT/TT model when compared against the homozygote state (Fig. 2) (OR, 0.53; 95% CI, 0.40–0.72; p < 0.0001). However, there was no significant association in the recessive TT model (OR, 0.63; 95% CI, 0.37–1.06; p = 0.08). No significant interstudy heterogeneity was observed in either model. A funnel plot for the dominant model was not symmetrical and the Egger's test (p = 0.034) showed significance suggesting a possibility of publication bias. The funnel plot for the recessive model was not symmetrical while the Egger's test was nonsignificant.

Figure 2. Forest plot of association between MMP-2 1306T and lung cancer.

MMP-2^735C→T

Only 2 studies were identified for MMP-2^735C→T that evaluated its association with lung cancer.27, 28 A significant association was observed in the dominant CT/TT model (OR, 0.65; 95% CI, 0.53–0.79; p < 0.0001) that was protective in nature. There was no significant association observed in the recessive TT model (OR, 0.88; 95% CI, 0.56–1.39; p = 0.59).

MMP-3 nt-1171

Three studies were identified that investigated the association between the MMP-3 nt-1171 polymorphism and lung cancer.29–31 No significant association was observed in either the dominant (OR, 1.03; 95% CI, 0.89–1.2; p = 0.66) or the recessive model (OR, 0.97; 95% CI, 0.84–1.12; p = 0.71). No interstudy heterogeneity was observed in either model. Funnel plots were symmetrical and the Egger's test for both models showed no significance suggesting little evidence of publication bias.

MMP-9 1562C/T

Only 2 studies evaluated MMP-9 1562C/T polymorphism and its association with lung cancer.32, 33 No significant association was found in either the dominant CT/TT (OR, 1.03; 95% CI, 0.73–1.45; p = 0.86) or recessive TT (OR, 0.77; 95% CI, 0.01–60.82; p = 0.91) models.

Colorectal cancer

MMP-1 nt-1607

Five studies were identified that investigated the association between MMP-1 nt-1607 polymorphism with colorectal cancer,34–38 but the genotype data for Ghilardi et al. could not be interpreted for inclusion in a dominant model. In assessing the dominant 1G2G/2G2G model, there was a significant association with colorectal cancer when compared to homozygous 1G carriers (Fig. 3) (OR, 1.66; 95% CI, 1.14–2.42; p = 0.008). Similarly, the recessive 2G2G state showed significant association with colorectal cancer when compared against the homozygote 1G1G and heterozygote 1G2G state (Fig. 4) (OR, 1.59; 95% CI, 1.15–2.20; p = 0.005). No significant interstudy heterogeneity was observed in either the dominant or recessive models. Funnel plots were symmetrical and the Egger's test for both models showed no significance suggesting little evidence of publication bias.

Figure 3. Forest plot of association between MMP-1 nt-1607 2G and colorectal cancer.

Figure 4. Forest plot of association between MMP-1 nt-1607 2G2G and colorectal cancer.

MMP-2 nt-1306C/T

Three studies evaluated MMP-2 nt-1306C/T polymorphism and its association with colorectal cancer.37, 39, 40 No significant association was observed in the dominant CT/TT (OR, 0.97; 95% CI, 0.49–1.93; p = 0.81) or recessive TT homozygote (OR, 1.37; 95% CI, 0.61–3.11; p = 0.44) models.

MMP-3 nt-1171

Four studies investigated the association between the MMP-3 nt-1171 polymorphism and colorectal cancer.37, 41–43 No significant association was observed in either the dominant (OR, 0.93; 95% CI, 0.50–1.71; p = 0.81) or recessive (OR, 1.29; 95% CI, 0.66–2.51; p = 0.45) models. Significant interstudy heterogeneity was observed in the recessive model (p_het = 0.005). Funnel plots were symmetrical and the Egger's tests for both models were nonsignificant suggesting little evidence of publication bias.

MMP-9 ^1562C→T

Five studies evaluated MMP-9 ^1562C→T polymorphism and its association with colorectal cancer.37, 40, 44–46 Genotype data could not be interpreted for Xu et al. in a recessive model. Neither the dominant CT/TT (OR, 0.90; 95% CI, 0.68–1.17; p = 0.43) nor recessive TT model (OR, 0.60; 95% CI, 0.15–2.41; p = 0.47) showed significant association with colorectal cancer. No significant interstudy heterogeneity was observed. The funnel plot was symmetrical and Egger's test showed no significance suggesting little evidence of publication bias.

Breast cancer

MMP-1 nt-1602

Only 2 studies investigated the MMP-1 nt-1602 polymorphism and its association with breast cancer.47, 48 No significant association was observed in either the dominant 1G2G/2G2G (OR, 1.61; 95% CI, 0.76–1.77; p = 0.49) or recessive 2G2G (OR, 1.08; 95% CI, 0.66–1.76; p = 0.21) models.

MMP-2 nt-1306C/T

Four studies were identified investigating the MMP-2 nt-1306C/T polymorphism and its association with breast cancer.49–52 In assessing the dominant CT/TT model (Fig. 5), there was a (nonsignificant) negative (protective) association against breast cancer when compared against the C homozygote state (OR, 0.66; 95% CI, 0.40–1.07; p = 0.09). No association was observed in the recessive TT model (OR, 0.95; 95% CI, 0.54–1.66; p = 0.86) when compared against the heterozygote (CT) and homozygote state (TT). Funnel plots were symmetrical and the Egger's test showed no significance suggesting little evidence of publication bias.

Figure 5. Forest plot of association between MMP-2 1306T and breast cancer.

MMP-3 nt-1171

Only 2 studies were identified investigating an association between the MMP-3 nt-1171 polymorphism and breast cancer.53, 54 No significant association was observed in either the dominant (OR, 0.94; 95% CI, 0.53–1.65; p = 0.83) or recessive (OR, 0.71; 95% CI, 0.38–1.33; p = 0.29) models.

MMP-9 1562C/T

Two studies were also identified for MMP-9 1562C/T and breast cancer.55, 56 No significant association was observed in either the dominant CT/TT (OR, 1.07; 95% CI, 0.88–1.30; p = 0.50) or recessive TT (OR, 1.03; 95% CI, 0.16–6.75; p = 0.98) models.

Discussion

In this comprehensive genetic meta-analysis, 3 of the 5 polymorphisms analyzed in the 4 genes studied were significantly associated with the major cancers. The MMP-1 nt-1602 polymorphism was shown to increase the risk of colorectal cancer, whereas both the MMP-2 polymorphisms 1306C/T and 735C/T were protective in lung cancer, although the evidence base is small when compared against other complex disorders.57–59 Interestingly, for breast cancer the MMP-2 1306C/T polymorphism approached significance, showing a negative (protective) trend with breast cancer suggesting that the mechanism by which the MMP-2 1306C/T polymorphism confers protection may be the same in breast cancer as it is in lung cancer. The mean number of participants was ∼2,000 allowing more precise estimates to be made of effect sizes than can be estimated from any single study.

The attributable risk provided by the MMP-1 nt-1602 polymorphism (pooled OR ∼1.6) was broadly in agreement with previous studies in colorectal cancer37, 60–62 and similar to the effects of obesity63 and moderate alcohol consumption64 in the risk of developing colorectal cancer. The calculated PAR for the MMP-1 nt-1607 polymorphism in colorectal cancer is 36%. This value is higher than that reported for some other well-established risk factors such as family history (PAR, 8%) and low vegetable intake (PAR, 22%).65 Given the high incidence of colorectal cancer (36,000 new cases per year in the UK), if these estimates hold true they suggest that common MMP-1 nt-1607 variant may contribute to ∼12,960 cases of colorectal cancer in the UK alone each year.

Our data may help those involved in genome wide association studies (GWAS) where marker panels could be focused on these positive (susceptibility or protective) regions to determine replication. However, there remains an important role for hypothesis-driven studies even in the current GWAS era. Although whole genome searches can identify novel regions of interest, they remain a relatively blunt instrument even with the latest chips with a million SNP resolution. Any identified region has to be further resolved to determine its actual identity as opposed to the candidate gene approach where an a priori hypothesis is already established.

As with any meta-analysis, its interpretation must be made within the context of its limitations, including study selection, publication bias and variability in the methodological quality of the included studies. Although publication bias was quantified, it is an unlikely explanation for our findings as our funnel plots suggest but it cannot be completely excluded. Moreover, rigorous selection criteria (use of histopathological diagnosis of lung, breast or colorectal cancer) enriched the meta-analyses for phenotypic quality. The MMP-1 nt-1607 meta-analysis in colorectal cancer had more than 500 cases, MMP-2 nt-1306C/T more than 1,500 cases and the nt-735C/T polymorphism had more than 800 making these relatively sizable studies for risk estimates, albeit small by several other complex disorder genetic meta-analyses. More reliable assessment of attributable risk will only come from studies with much larger number of subjects. Using histopathologically confirmed diagnoses as a selection criterion may have helped to maintain comparable groups of cases; however, the selection of control groups varied considerably between studies. Statistical methods using marker genotype data may in the future permit the detection and control of confounding factors because of population stratification and selection bias in genetic association studies66 reducing the impact of variability within controls groups. However, the confounding effect of heterogeneity between different ancestral populations cannot be dismissed.

Our finding suggests a genetic basis for the development of colorectal cancer and a protection against lung cancer. Our results determine OR with greater reliability and more robust CI than has previously been assessed by any single study, but the evidence base remains small when compared against similar data from other complex disorders.59, 67, 68

Acknowledgements

P.S. is a Senior Fellow in the Department of Health (UK), and P.M. holds a Cancer Research UK studentship.