|Home | About | Journals | Submit | Contact Us | Français|
Lynch syndrome family members with inherited germline mutations in DNA mismatch repair (MMR) genes have a high risk of colorectal cancer (CRC) and cases typically have tumors that exhibit a high level of microsatellite instability (MSI). There is some evidence that smoking is a risk factor for CRCs with high MSI, but the association of smoking with CRC among those with Lynch syndrome is unknown.
A multicentered retrospective cohort of 752 carriers of pathogenic MMR gene mutations was analyzed, using a weighted Cox regression analysis, adjusting for sex, ascertainment source, the specific mutated gene, year of birth, and familial clustering.
Compared with never smokers, current smokers had a significantly increased CRC risk (adjusted hazard ratio [HR] = 1.62; 95% CI, 1.01 – 2.57) and former smokers who had quit smoking for 2 or more years were at decreased risk (HR = 0.53; 95% CI, 0.35 – 0.82). CRC risk did not vary according to age at starting. However, light smoking (<10 cigarettes per day) and shorter duration of smoking (<10 years) were associated with decreased CRC risk (HR = 0.51; 95% CI, 0.29 – 0.91 and HR = 0.52; 95% CI, 0.30 - 0.89 respectively). For former smokers, CRC risk decreased with years since quitting (P trend <0.01).
People with Lynch syndrome may be at increased risk of CRC if they smoke regularly. Although our data suggest that former smokers, short-term and light smokers are at decreased CRC risk, these findings need further confirmation, preferably using prospective designs.
People with Lynch syndrome (1), who have inherited mutations in DNA mismatch repair (MMR) genes, have an elevated cancer risk (2). The dominantly inherited germline MMR gene mutations are most frequently present in MLH1 and MSH2 but sometimes in MSH6 and PMS2 (3). Mutation carriers are at substantially increased risk of colorectal cancer (CRC); the syndrome accounts for about 2-3% of CRC diagnosed at any age in the US, and many more cases of earlier-onset CRC (4). There is also a predisposition to other tumors, particularly endometrial cancer (5,6). CRC occurs predominantly in the proximal colon, 90% of the tumors exhibit microsatellite instability (MSI), and progression from adenoma to carcinoma is rapid (2-3 years) (7), necessitating early and frequent screening.
For mutation carriers, the CRC risk to age 70 yrs has been recently estimated to be 30-50% and appears to be higher with mutations causing earlier onset disease (8,9). There has also been a shift in the cancer phenotype of Lynch syndrome from a predisposition toward stomach and endometrial cancers in the first few generations of the earliest recognized Lynch syndrome family (10) to colon and rectal cancers in its later generations (11), suggesting that the site of tumor development may be influenced by the environmental or lifestyle factors that have changed over the last century (12). It is therefore possible that there are environmental and/or other genetic factors that influence carcinogenesis and modify penetrance.
There is increasing evidence that smoking is a risk factor for CRC overall — it has been estimated that up to one in five CRCs in the United States may be attributable to tobacco use (13). Tobacco smoke contains a variety of genotoxic substances, including polycyclic aromatic hydrocarbons, nitrosamines, and heterocyclic and aromatic amines. Carcinogens in tobacco smoke might reach the colon both directly through the gut (14) and via the blood stream (15) and thus initiate damage to the colorectal mucosa.
People with Lynch syndrome might be particularly susceptible to tobacco carcinogens because of their inherited deficiency in DNA MMR. Among those with a family history of CRC, smoking is associated with risk of CRC (16) and particularly an increased risk of tumors with MSI (17). Evidence for an influence of smoking on CRC risk for people with Lynch syndrome is limited: studies published so far have either included only small numbers of known MMR mutation carriers (18) or lacked complete data on tobacco exposure (19). In this study, we analyzed the association between cigarette smoking and CRC risk among 752 individuals who were established carriers of pathogenic germline MMR gene mutations.
Subjects were from two sources: the Colon Cancer Family Registry (Colon CFR) and The University of Texas M. D. Anderson Cancer Center. The Colon CFR is an international consortium of six registries (Australasia, Hawaii, Mayo Clinic, Ontario, Seattle, and Southern California) formed as a family resource for collaborative research on CRC (20). Families with none, one, or more cases of CRC were ascertained through cancer family genetic clinics (clinic-based, n = 481) and cancer registries (population-based, n = 116). Subjects were also recruited at the M. D. Anderson Cancer Center (n = 155), a clinical referral site. Participant recruitment methods for likely Lynch syndrome families at M. D. Anderson have been described previously (21,22). Participants provided informed consent and the study had Institutional Review Board approval.
All subjects had a laboratory confirmed pathogenic mutation in a DNA MMR gene — MLH1, MSH2, or MSH6. A total of 868 MMR mutation carriers were identified. Two subjects were dropped from the analysis because of missing values for all smoking variables. Furthermore, we restricted the current analysis to 752 MMR mutation carriers born between 1920 and 1970 because smoking history from the very old, born before 1920 (n = 10) was likely to be subject to poor recall and the younger cohort, born after 1970 (n=104) had experienced a short duration of smoking exposure and insufficient years of follow-up to experience the event of interest (i.e. develop CRC).
All subjects from the Colon CFR had completed the Colon Epidemiology Questionnaire1, either by face-to-face interview or by mail; this asked about demographics, ethnicity, medical history, smoking, and other lifestyle factors. The questionnaire used at M. D. Anderson was designed to match the Colon CFR questionnaire.
The questionnaires defined regular smoking as having smoked at least 1 cigarette/day for 3 months or longer (CFR questionnaires) or >100 cigarettes overall (M. D. Anderson questionnaire), then asked about current and past smoking, average number of cigarettes smoked per day when smoking regularly, total duration of regular smoking in months and years, and the ages regular smoking started and stopped.
Confirmation of self-reported personal histories of CRC or other cancers and age at diagnosis were sought where possible through cancer registries, national death registries, pathologic review of slides, pathology reports, and hospital and other medical records.
Testing for mutations in MLH1, MSH2, and MSH6 for the Colon CFR subjects was done by denaturing high pressure liquid chromatography (dHPLC), followed by confirmatory sequencing. In addition, large genomic rearrangements were analyzed by multiplex ligation-dependent probe amplification (MLPA) analysis (20). Mutation testing for the M. D. Anderson participants was performed by a CLIA-certified laboratory (City of Hope, or Myriad Genetics) using direct Sanger sequencing and MLPA analysis.
The MMR mutation carriers were treated as a cohort from birth and Cox proportional hazards regression was used to calculate hazard ratios (HRs) and 95% confidence intervals (CIs) to estimate the risk of CRC associated with smoking. Because most of the subjects in this study were selected from multiple-case (high-risk) families, individuals with CRC (the cases) were, in general, oversampled. To adjust for this nonrandom sampling, we used a previously developed weighted-regression approach (23) that weights cases and unaffected carriers so that the overall cumulative cancer incidence in the cohort is equal to the population-based cancer incidence. Specifically, we obtained data on incidence of CRC for 5-year intervals from statistics compiled by the American Cancer Society Cancer Facts and Figures 20082, averaging over sex. We used relative risks of CRC for MMR-gene-mutation carriers for men and women under and over the age of 50 (8) to establish age-specific incidence data for the mutation carriers. The sex-averaged relative risks were 137.5 for mutation carriers up to age 50 and 4.1 over 50. To allow the relative risk to decline more quickly at younger ages, we allowed it to decay log-linearly from age 18 to 70. Multiplying the population-based incidence data by the relative risks for CRC in mutation carriers yielded age-specific incidence rates per year of 0.0034 (age 18-24), 0.0048 (age 25-29), 0.0051 (age 30-34), 0.0059 (age 35-39), 0.0059 (age 40-44), 0.0071 (age 45-49), 0.0084 (age 50-54), 0.0103 (age 55-59), 0.0111 (age 60-64), 0.0125 (age 65-69), and 0.0123 (age 70-79). These age-specific incidence rates were used to calculate sampling fractions to weight the proportion of affected mutation carriers and the unaffected carriers in each age stratum so that the proportion of affected carriers in each age group equaled the population proportions. Except for those 18-24 years, there were more affected mutation carriers (cases) in the sample than would be expected from a nonselected group of mutation carriers (see Supplementary Table 1). Therefore, for all intervals but the first, we downweighted information from the affected, compared to the unaffected, carriers.
A simulation study of this approach (23) applied to Cox regression showed that allowing for nonrandom sampling of subjects by using rates from an external referent population removed bias when the external rates were correctly specified and reduced bias even when the sampling fractions were not completely accurate. However, adjusting for nonrandom sampling via this method usually resulted in some increase in standard errors. This presumably is the result of the higher variance in the case group (as the effective sample size decreases with down weighting). Note that the test of the null hypothesis from the unweighted analysis is correct; i.e., to derive unbiased estimates, the weighted analysis is appropriate, but to test the null hypothesis, the p-value from the unweighted analysis was found to be valid and smaller in previous simulation studies (23).
Person-years of follow-up were determined starting at birth and ending at age at diagnosis of primary CRC (n=415) for those affected. Censoring age for the remaining subjects was age at diagnosis of another cancer (n=139) or age at questionnaire completion (n=198), whichever came first. It has been suggested that including time that a person is not at risk introduces “immortal person-time bias” (24) and since CRC is very rare before 25 years of age, we also analyzed the data counting person-years from age 25 onwards.
Smokers were categorized as never, former, and current. A never-smoker was defined as a person who had never smoked or smoked 1 cigarette a day for no more than 3 months (for CFR participants) or had smoked <100 cigarettes overall (for M. D. Anderson participants); a former smoker was one who had smoked at least 1 cigarette a day for 3 months or longer or had smoked >100 cigarettes but had quit 2 or more years before CRC diagnosis or censoring age; a current smoker smoked within 2 years of CRC diagnosis or censoring age. Other smoking variables examined were: age at smoking initiation, number of years smoked, years since quitting, and total pack-years. Time period for smoking assessment was up to the age at diagnosis or censoring. Smoking variables were also analyzed as time-dependent covariates. In this analysis, we analyzed the impact that smoking had on cancer risk for each year of an individual's life, allowing individuals to switch from current to former smoker at the age at which they indicated the change occurred.
HR estimates were adjusted for sex, data source (clinic or population), and the specific MMR gene mutated; these variables were included in the Cox model to adjust for potential confounding. The year of birth was also added to the regression model, to adjust for any residual birth-cohort effect for smoking. The proportional hazards assumption was verified by examining the Kaplan-Meier estimates of the survival function by smoking status and also by testing that the Schoenfeld residuals (25) (as implemented in STATA) were not significantly (p>0.05) correlated with age. Furthermore, we evaluated the association of each specific type of mutation (i.e. missense, nonsense, frameshift and splice-site MMR mutations) as well as the tumor sub-site (proximal, distal, rectal). To adjust for correlation within families, the Huber-White robust variance correction was applied (26,27). The available data on screening colonoscopy was examined to see if there was a difference by smoking status.
Cox analyses were stratified by age group (≤50 and >50 years), sex, data source, the MMR gene mutated, and birth cohort. All analyses were conducted using STATA version 8.0 software (Stata Corp., College Station, TX).
The cohort of 752 individuals, born 1920 – 1970, comprised 299 MLHI, 393 MSH2, and 60 MSH6 mutation carriers with a total of 426 diagnosed cases of CRC and 34,129 person-years of follow-up. Mean age was 45.4 years and mean age at diagnosis of CRC was 44.8 (range: 22 – 80) years. Overall, 385 mutation carriers (51.2%) had ever smoked cigarettes. Ever- and never-smokers did not differ in risk of CRC, but when the ever smokers were separated into former and current smokers, current smokers showed an increased risk (weighted HR = 1.62; 95% CI, 1.01 – 2.57) whereas former smokers showed a decreased risk (weighted HR = 0.53; 95% CI, 0.35 – 0.82). These estimates were adjusted for sex, source of ascertainment, mutated gene, and birth year. The weighted HR estimates for smokers were higher than the unweighted (1.62 vs. 1.49), but the 95% CIs were wider: 1.01 – 2.57 vs. 1.15 – 1.93 (Table 1). Although HRs did not vary by age at smoking initiation risk decreased inversely with time since quitting (P trend <0.01). The strongest inverse association was seen for those who quit more than 20 years ago (weighted HR = 0.26; 95% CI, 0.15 – 0.45). This group smoked a median of 10 cigarettes/day for a median of 9 years. Furthermore, the risk for those who smoked fewer than 10 cigarettes per day was lower than for never-smokers (HR = 0.51; 95% CI, 0.29 – 0.91) but an inverse association was not seen in those who smoked >10 cigarettes per day.
Although the duration smoked was not associated with CRC risk in the overall analyses, when analyzed as a time-dependent variable (Table 1), there was a statistically significant inverse association for those who smoked for <10 years (HR = 0.52; 95% CI, 0.3 – 0.89). This was further reflected in a significant inverse association for the lowest category of pack-years smoked (<10 pack years) (HR = 0.58; 95% CI, 0.37 – 0.9). However, among the smokers (with <10 years as referent), the risk increased with duration smoked (P trend 0.01).
There was no association between smoking and CRC when stratified by tumor sub-site: proximal, distal, rectal (results not shown). Similarly, there was no association by the specific types of MMR mutations (information about specific mutation type was available for 85% of the sample: 10% were missense, 25% nonsense, 26% splice-site, and 35% were frameshift mutations). In addition, analysis which started the Cox-Modeling at age 25 to allow for “immortal-time bias” had virtually no impact on the findings (data not shown).
History of screening colonoscopy was available from 595 people (79% of the participants). There was no difference between the never, former and current smokers in whether they had ever undergone a colonoscopy (90.4% of current smokers, 89.4% of former smokers and 88.3% of current smokers had ever had a colonoscopy). Furthermore, among the CRC unaffected, the proportion who had colonoscopy was comparable between the three smoking groups (of the 247 CRC unaffected with available data on colonoscopy, 84.4% never, 82.7% former and 84% current smokers had ever undergone colonoscopy).
Table 2 summarizes the estimates of associations between smoking and CRC, comparing current and former smokers to never smokers, by sex, ascertainment source, and gene mutated as well as age and birth cohort. The directions of the associations were consistent with the overall results in the stratified analyses, although the weighted HR estimates were not always statistically significant. Largely, the smoking associations did not vary by sex, data source or age category. However, the HR was higher for the 1960 or later cohort than the previous three cohorts among the current smokers and there was suggestion of a higher risk for both former and current smokers carrying the MSH6 mutations (although this was based on small numbers).
The influence of cigarette smoking on CRC risk for individuals with Lynch syndrome is not well understood. There is limited evidence suggestive of an increased risk for smokers but the studies have been small (18,19). Given the large sample size and extensive information on smoking habits in the present study, we were able more fully to examine the association between smoking and CRC. We found that among MMR gene mutation carriers, compared with never smokers there was a statistically significant increased risk of CRC for current smokers and a lower risk for former smokers. An inverse association with smoking was evident for those who smoked fewer than 10 cigarettes per day and, for former smokers, the younger they were when they quit the less likely they were to develop CRC.
In the general population, the relationship between smoking and CRC remains somewhat inconsistent (13,28). Smoking has, however, been found to be associated with a subset of CRC—namely, tumors exhibiting MSI. Slattery et al., in two separate case-control studies, examined the association between smoking and CRC risk for people with a family history and MSI-positive disease (16,17). They found an earlier age at onset of CRC for subjects with a known family history of cancer and an increased risk of MSI-positive CRC for smokers (OR = 1.6; 95% CI, 1.0 – 2.5 for men; and OR = 2.2; 95% CI, 1.4 – 3.5 for women). The association of smoking with MSI-positive tumors was strongest for current smokers (OR = 2.3; 95% CI, 1.5 – 3.3). However, because a family history and MSI-positive tumors are not necessarily indicative of Lynch syndrome, caution is required in generalizing these results to people specifically with Lynch syndrome. (Even though MSI is a characteristic of Lynch syndrome CRC tumors, about 10-15% of non-Lynch CRC tumors are MSI positive because of hypermethylation of the promoter region of MLH1). Using the same case-control data, Samowitz et al. (29) found an association of smoking with CpG island methylator (CIMP) phenotype and BRAF mutations suggesting that the mechanism for the observed association may be induction of CpG island methylation. Again, very few Lynch-syndrome-associated tumors have a CIMP-high status or BRAF mutations (30,31), so this epigenetic mechanism cannot explain an association between smoking and CRC in Lynch syndrome.
Two previous studies examined the association between smoking and CRC for people with Lynch syndrome. Watson et al. (19), studied 330 mutation carriers and found a 43% increased risk of CRC for smokers (HR = 1.43; P = 0.04). However, because 25% of the participants were deceased, the smoking history obtained by family report or medical records was limited; thus the study compared non-smokers with smokers but did not report on other smoking variables. Another recent study by Diergaarde et al. (18) involving 145 cases and 103 controls (not all of whom were confirmed MMR mutation carriers) found a 2.4-fold increased risk of CRC for current smokers (OR = 2.4; 95% CI, 1.1 – 5.3), consistent with our results but no association for former smokers.
Although increased risk of CRC for current smokers and declining risk after quitting are consistent with the known carcinogenicity of cigarette smoke, an inverse association with smoking among former smokers is difficult to explain. One can hypothesize that the former smokers in our study were a more heterogeneous group that included everyone from light smokers who had a history of smoking just 1 cigarette/day for 3 months to regular heavy former smokers. Because we also found that light smokers (1-10 cigarettes per day) and those that smoked for less than 10 years were at decreased risk, it is possible that occasional light smoking has an inverse association. However, the possibility that potential biases may give rise to this pattern cannot be ruled out. It is pertinent to point out that this is a study of people who have participated and given a blood sample. The unaffected mutation carriers are generally people in CRC families who have been motivated to participate; it is possible that they are more likely to be quitters than cases. Although the proportion of unaffected carriers who ever smoked was 49%, compared to 42% in the general population (32), the proportion of quitters was higher for the unaffected (31%) than the cases (27%). Hence, the inverse association in former smokers could be due to participation bias by health-aware unaffected people.
A further possibility is that smoking acts only on a subset of MMR-mutation carriers, causing an earlier onset of CRC. This may eliminate susceptibles from the smoking group, ensuring that, among those with long smoking histories, there will be a higher proportion of carriers who are resistant to tobacco: this group will be overrepresented among quitters, resulting in an inverse association with smoking.
Finally, drawing upon studies of smoking and breast cancer risk in BRCA1/2 mutation carriers: separate studies have reported a reduced risk for smokers (33,34), no association (35,36), and increased risk (37). In the largest and most recent of these studies (37), it was suggested that inclusion of prevalent cases interviewed or tested many years after diagnosis may have resulted in the inverse association with smoking seen in earlier studies. To avoid this bias, the authors suggest that the interval between cancer diagnosis and testing or interview/questionnaire completion should be minimal. Our data were collected over more than 10 years and therefore include prevalent cases of CRC: the median difference between diagnosis and questionnaire completion being 6 years. However, to overcome this source of bias, we limited the smoking exposure to that before age at diagnosis or censoring.
Lastly, the results of our stratified analysis showed that the HR for current smokers for the birth cohort of 1960 or later was higher than for the earlier three cohorts. However, this cannot be explained by change in smoking behavior over the years since in our data these were not heavier or longer term smokers than the other cohorts. It is possible that attributes of cigarettes may have changed over the years and although our results are based on a small subsample, it does raise the possibility that the more recent cigarettes may be more harmful.
Potential limitations of this study include an inability to account for the potential confounding effect of other known environmental and lifestyle exposures, such as physical activity; alcohol, aspirin and calcium use; fruit and vegetable intake; and body mass index because we lacked complete data for these variables. Strengths include a large sample size and detailed smoking history. Furthermore, risk estimates were corrected for possible selection bias by application of appropriate weights (23) and for familial correlation (26,27).
We evaluated the association between smoking and the risk of CRC using a cohort of verified MMR-mutation carriers. We found an increased risk of CRC for current smokers and a decreased risk for former, short-term and light smokers that requires further confirmation, preferably in prospective studies, along with an investigation of the underlying biology of the relationship. The positive association between current smoking and increased colorectal cancer risk provides additional evidence to recommend avoidance of smoking for people genetically predisposed to CRC to reduce their risk.
We thank the study coordinators and staff at the six CFR centers and at M. D. Anderson Cancer Center (Danny Osterwisch, Luis Orta, Nancy Viscofsky) for collecting and managing the data; John Langer from RTI for generating and providing the CFR datasets; Antonis Antoniou for providing an Excel spreadsheet tool and guidance for computing the age specific weights; and the study participants who have generously donated their time for this project.
The content of this manuscript does not necessarily reflect the views or policies of the National Cancer Institute or any of the collaborating centers in the CFRs, nor does mention of trade names, commercial products, or organizations imply endorsement by the US Government or the CFR.
Grant support: This work was supported by the National Cancer Institute, National Institutes of Health under RFA CA-95-011 and through cooperative agreements with members of the Colon Cancer family Registry and P.I.s. These include the Australasian Colorectal Cancer Family Registry (U01 CA097735); the Familial Colorectal Neoplasia Collaborative Group (U01 CA074799) [USC]; the Mayo Clinic Cooperative Family Registry for Colon Cancer Studies (U01 CA074800); the Ontario Registry for Studies of Familial Colorectal Cancer (U01 CA074783); the Seattle Colorectal Cancer Family Registry (U01 CA074794); the University of Hawaii Colorectal Cancer Family Registry (U01 CA074806); and the University of California, Irvine Informatics Center (U01 CA078296). In addition, grant support to the University of Texas M. D. Anderson Cancer Center was from National Institutes of Health Cancer Center Support Grant CA 16672, National Cancer Institute CA 070759 as well as R25 CA 57730 (Mala Pande).
Statement of Translational Relevance
Carcinogens in cigarette smoke have the potential to cause DNA damage and initiate carcinogenesis. People with Lynch syndrome may be particularly susceptible to the genotoxic effects of these carcinogens because of their inherited deficiency for DNA mismatch repair. Smoking is a preventable risk factor that is associated with many different types of cancers though not conclusively with colorectal cancer. We have shown that people with Lynch syndrome who smoke regularly are at increased risk for colorectal cancer and although the underlying mechanism of the relationship warrants further investigation, our results provide evidence for people with Lynch syndrome to avoid smoking to reduce their colorectal cancer risk. This has relevance in the practice of Oncology.