|
|
Background The identification of mutations in germ-line DNA mismatch-repair genes at the time of diagnosis of colorectal cancer is important in the management of the disease.
Methods Without preselection and regardless of family history, we recruited 870 patients under the age of 55 years soon after they received a diagnosis of colorectal cancer. We studied these patients for germ-line mutations in the DNA mismatch-repair genes MLH1, MSH2, and MSH6 and developed a two-stage model by multivariate logistic regression for the prediction of the presence of mutations in these genes. Stage 1 of the model incorporated only clinical variables; stage 2 comprised analysis of the tumor by immunohistochemical staining and tests for microsatellite instability. The model was validated in an independent population of patients. We analyzed 2938 patient-years of follow-up to determine whether genotype influenced survival.
Results There were 38 mutations among the 870 participants (4 percent): 15 mutations in MLH1, 16 in MSH2, and 7 in MSH6. Carrier frequencies in men (6 percent) and women (3 percent) differed significantly (P<0.04). The addition of immunohistochemical analysis in stage 2 of the model had a sensitivity of 62 percent and a positive predictive value of 80 percent. There were 35 mutations in the validation series of 155 patients (23 percent): 19 mutations in MLH1, 13 in MSH2, and 3 in MSH6. The performance of the model was robust among a wide range of cutoff probabilities and was superior to that of the Bethesda and Amsterdam criteria for hereditary nonpolyposis colorectal cancer. Survival among carriers was not significantly different from that among noncarriers.
Conclusions We devised and validated a method of identifying patients with colorectal cancer who are carriers of mutations in DNA repair genes. Survival was similar among carriers and noncarriers.
Practical and financial constraints often require initial testing (prescreening) of tumor DNA for microsatellite instability or immunohistochemical assessment of tumor sections for DNA mismatch-repair proteins, or both, before the genotyping of incident cases of cancer. Prescreening assumes that microsatellite instability is a consistent feature of tumors from carriers, but some tumors do not have microsatellite instability.8 Moreover, prescreening misses an appreciable number of mutations in DNA repair genes6,8 and has a relatively poor positive predictive value.6,8,13 In addition, analysis of patients with incident colorectal cancers for microsatellite instability and DNA mismatch-repair proteins14 without previous genetic counseling and explicit consent raises ethical concerns, because patients with tumors that do not have DNA mismatch-repair proteins and do have microsatellite instability are likely to carry heritable germ-line mutations.
There is inconclusive evidence that the prognosis of colorectal cancer among patients with the Lynch syndrome is better than that among patients with sporadic cases.15,16,17 Sporadic colorectal cancers with microsatellite instability have a better prognosis than do microsatellite-stable sporadic tumors,18,19,20 but this difference does not necessarily apply to tumors with germ-line defects in DNA mismatch-repair genes.
Mutational analysis of germ-line DNA mismatch-repair genes without previous testing of the tumor has not been undertaken in a prospective, population-based series.21 In this study, we looked for such mutations without considering the family history or the results of tumor testing in a prospective, population-based series of 870 cases of early-onset colorectal cancer, a group enriched for genetically determined disease.6,7,8,13,22,23 This approach allowed us to construct a clinically driven predictive model that aids in the prediction of carriers of germ-line mutations of DNA mismatch-repair genes.
Methods
Beginning January 2, 1999, we identified all patients in Scotland (population, 5.06 million) who had received a diagnosis of colorectal cancer before the age of 55 years. All such patients were identified soon after the diagnosis and regardless of family history. After the exclusion of 15 patients with dominant polyposis syndromes, 1259 patients agreed to undergo genetic counseling and gave informed written consent for germ-line and tumor analysis before the end of 2005. The study population comprised 870 consecutive patients who received a diagnosis of colorectal cancer between February 1999 and July 2003 and in whom mutational analysis and follow-up were complete. The study population also included 72 percent of 1206 patients with incident cases identified independently by the Scottish Cancer Registry. Of these patients, 80 did not respond to our request, 156 had psychiatric or physical illnesses that precluded their enrollment in the study, and only 100 patients declined to participate, which left us with 90 percent compliance overall. The study was approved by the research ethics committees and National Health Service (NHS) management of every participating hospital.
An invitation to participate in the study was extended to patients within a few weeks after diagnosis to minimize survival bias. Family history was obtained at the time of the interview, a blood sample was taken for DNA analysis, and tumor samples were acquired from NHS pathology departments. The date of disease onset was defined as the date of the first histologic diagnosis of colorectal adenocarcinoma. The cancer history of patients was obtained by personal interview and cross-reference of personal identifiers with the Scottish Cancer Registry. Tumors were staged according to Dukes' classification and tumor–node–metastasis (TNM) criteria in accordance with recommendations of the American Joint Committee on Cancer.
Mutational Analysis
Germ-line DNA obtained from blood leukocytes was analyzed for MLH1, MSH2, and MSH6 mutations. Denaturing high-performance liquid chromatography analysis (Transgenomic) was used for 11 exons of MSH2 and 16 exons of MLH1. Variants noted on chromatography were sequenced, as were MSH2 exons 1, 4, 5, 10, and 13 and MLH1 exons 8, 12, and 15 in every sample (ABI 3730 DNA Analyzer, Applied Biosystems). All 10 MSH6 exons were sequenced. Sequencing chromatographs were assessed visually and with the use of Sequencher software. Mutations were confirmed by reamplification of an independent sample of DNA and resequencing in both directions. MLH1 and MSH2 were assessed for deletions by multiplex ligation-dependent probe amplification (MLPA, MRC-Holland), with products separated on an ABI 3100 xl Genetic Analyzer (Applied Biosystems) and analyzed with the use of custom software (available on the Web site of the United Kingdom National Genetics Reference Laboratory at www.ngrl.org.uk/Manchester/Publications.htm#MLPA).
Analysis of Microsatellite Instability
Tumor DNA that had been purified from microdissected 10-µm tumor sections (QIAamp DNA minikit, Qiagen) was subjected to multiplex polymerase-chain-reaction assay with the use of BAT25, BAT26, D2S123, D5S346, and D17S250 markers24 and compared with control DNA from blood or normal tissue in the section. Products were analyzed with the use of ABI PRISM 3100 and Genescan software (Applied Biosystems). Tumors with more than one shifted marker were categorized as having a high degree of microsatellite instability, those with one unstable marker were categorized as having a low degree of microsatellite instability, and those with no instability were categorized as being microsatellite stable.
Immunohistochemical Analysis
Paraffin-embedded sections that had been stained with antibodies against MLH1, MSH2, and MSH6 were assessed independently by two microscopists. If discrepancies could not be resolved, a third pathologist reviewed all slides.
Statistical Analysis
Two-Stage Predictive Model
Univariate analysis of clinical variables by logistic regression identified independently significant predictors of mutational status. These predictors were included in multivariate logistic regression to construct a two-stage model for the prediction of whether patients carry germ-line mutations of DNA repair genes on the basis of clinical variables (see the Supplementary Appendix, available with the full text of this article at www.nejm.org). The least significant variables were removed one by one until all variables remained significant. All variables (except the age at onset) were included as categorical variables. The use of automatic stepwise backward and forward selection criteria, on the basis of the likelihood ratio and the Wald statistic, yielded similar results on logistic regression with SPSS software, so we selected the simplest model. The final multivariable analysis that estimates carrier probabilities for stage 1 of the model can be described by the following equation, which can be solved electronically at http://www1.hgu.mrc.ac.uk/Softdata/MMRpredict.php25: Pr/(1–Pr) = 1.39 x 0.89AGE x 2.57SEX x 4.45LOCATION x 9.53SYN/MET x 46.26CRCFH<50 x 7.04CRCFH
50 x 59.36ECFH.
In this equation, Pr is the carrier probability, and age refers to the age at diagnosis. Male sex is assigned a value of 1, and female sex a value of 0. A proximal tumor location is assigned a value of 1, and a distal location a value of 0. The presence of a synchronous (SYN) metachronous (MET) tumor is assigned a value of 1, and the absence of a SYN/MET tumor is assigned a value of 0. Colorectal-cancer family history (CRCFH) is assigned a value according to the age of the youngest relative with the disease, so CRCFH<50 and CRCFH50 are 1 and 0, respectively, if the youngest relative is less than 50 years of age and 0 and 1, respectively, if the youngest relative is 50 years of age or older; the assigned values are 0 and 0 if there are no affected relatives. The endometrial cancer family history (ECFH) is assigned a value of 1 if any first-degree relative has endometrial cancer.
Stage 1 of the model used exclusively clinical variables to represent the situation at diagnosis and identify subgroups enriched for carriers. These groups then proceeded to stage 2 (analysis of the tumor for microsatellite instability and immunohistochemical analysis). Combining stage 1 with stage 2 allowed for the assessment of the model's performance and usefulness in predicting the carrier state at various cutoff probabilities, as compared with the use of modified Amsterdam and Bethesda criteria, which were similarly combined with data on microsatellite instability and immunohistochemical analysis.
Replication Set
We assessed the validity of the model in an independent retrospective series of 155 Scottish patients under the age of 45 years (mean age, 34 years) who had received a diagnosis of colorectal cancer between February 1973 and June 1998. The mutational analysis was undertaken without consideration of the family history, the microsatellite instability of tumors, or immunohistochemical analysis.
Survival Analysis
Patients were followed by contacts with hospital teams, general practitioners, and systematic flagging with the Registrar General for Scotland. We also cross-referenced our data with records from the Scottish Health Service Information and Statistics Division. Data were censored on April 30, 2005, and no patients were lost to follow-up. Patients were followed until they died or the study ended. Kaplan–Meier analysis was used to compare survival according to genotype, first for all patients and then in an analysis that excluded patients who had undergone surveillance screening or had had previous cancers. Data were excluded for patients who had undergone previous surveillance (4 carriers) or had a previous potentially lethal cancer (26 noncarriers and 1 carrier); 1 carrier fulfilled both criteria.
Results
Pathogenic mutations were identified in 38 of 870 patients (4 percent) (Table 1). Carriers were younger than noncarriers (mean age, 42.7 and 48.2 years, respectively; P<0.001) (Table 2), although 37 percent of all carriers were between the ages of 45 and 54 years. The mean age was 38.5 years for carriers of MLH1 mutations, 43.8 years for those with MSH2 mutations, and 49.0 years for those with MSH6 mutations (P=0.005), suggesting gene-specific penetrance effects. Carrier frequencies in men (6 percent) and women (3 percent) were significantly different (P<0.04). Most carriers fulfilled Bethesda criteria (95 percent), but only 42 percent fulfilled Amsterdam criteria.
|
|
Microsatellite Instability and Immunohistochemical Analysis
The presence of a high degree of microsatellite instability had a sensitivity of 67 percent for germ-line mutations, as compared with a sensitivity of 27 percent for the presence of a low degree of microsatellite instability and a sensitivity of 93 percent for the presence of any degree of microsatellite instability. The positive predictive values were 45 percent, 24 percent, and 36 percent, respectively (Table 3). The presence of a high degree of microsatellite instability had a sensitivity of 83 percent for the detection of MLH1 mutations, of 75 percent for the detection of MSH2 mutations, and of 17 percent for the detection of MSH6 mutations. The presence of a low degree of microsatellite instability had sensitivities of 17 percent, 25 percent, and 50 percent, respectively, and the presence of any degree of microsatellite instability had sensitivities of 100 percent, 100 percent, and 67 percent, respectively. Table 3 lists the results of immunostaining for mismatch-repair proteins MSH2, MHL1, and MSH6, along with sensitivities, positive predictive values, and 95 percent confidence intervals. The absence of MSH6 protein predicted mutations in MSH2 or MSH6 (positive predictive value, 56 percent), as did the absence of MSH2 for mutations in MSH2 or MSH6 (positive predictive value, 52 percent), reflecting the biologic interaction between these proteins.
|
Tumor microsatellite instability and immunohistochemical data were assessed in stage 2 to refine the carrier prediction derived from stage 1. Clinical variables that were significant on univariate analysis (Table 4, and Table 3 of the Supplementary Appendix) were used to construct stage 1 of the model. The factors and P values in univariate analysis were age (P<0.001), sex (P=0.03), the location of the tumor (P<0.001), the presence of synchronous or metachronous tumors (P=0.001), having a first-degree relative with colorectal cancer (P<0.001), and having a first-degree relative with endometrial cancer (P=0.006). (The equation for the calculation of carrier probability is listed and described in the Methods section.25)
|
As compared with stage 1 of the model combined with microsatellite-instability analysis at stage 2 (at the optimal 0.05 probability cutoff), the Amsterdam criteria combined with microsatellite instability had a low sensitivity (39 percent, as compared with 65 percent in our model) but a positive predictive value of 100 percent. A combination of the Bethesda criteria and microsatellite instability was sensitive (88 percent) but had a positive predictive value of only 32 percent, as compared with 80 percent in our model. Although the inclusion of microsatellite-instability analysis enhanced the ability to predict the carrier state for all cutoff points in stage 1 of our model and for the Amsterdam and Bethesda criteria, to be clinically useful it would necessitate the genotyping of DNA from biopsy specimens within a very short time after diagnostic biopsy.
Replication Set
Mutation analysis of germ-line DNA from the replication series identified mutations in 35 of 155 samples (23 percent): 19 mutations in MLH1, 13 in MSH2, and 3 in MSH6. Clinical variables for all 155 patients were entered into the prediction algorithm,25 and carrier probabilities were generated for each subject. The discriminatory power of the model across cutoff values was similar in both the prospective and retrospective series (receiver-operating-characteristic [ROC] curves are shown in Figure 1 of the Supplementary Appendix). Furthermore, there was no significant difference (P=0.3) between the area under the ROC curve for these independent series (0.85 for the prospective series; 95 percent confidence interval, 0.77 to 0.93; and 0.82 for the replication series; 95 percent confidence interval, 0.72 to 0.91).
Survival Analysis
Follow-up for a maximum of six years three months generated a total of 2938 patient-years for survival analysis (134 carriers and 2804 noncarriers). There was no significant difference in survival between carriers and noncarriers (five-year survival rate of 74 percent for carriers and 63 percent for noncarriers, P=0.18) (Figure 1A). The censoring of data from patients who had previous cancers or had undergone previous screening or surveillance did not change the result significantly (P=0.24) (Figure 1B). Figure 1C and 1D show survival according to tumor stage; carriers with stage II tumors survived longer (five-year survival rate of 100 percent) than those with stage I tumors (five-year survival rate of 75 percent). There was no significant difference in survival between carrier groups according to the extent of tumor spread at diagnosis: the five-year survival rate among patients with localized disease was 95 percent for carriers and 87 percent for noncarriers (Figure 1E); for those with metastatic disease, the five-year survival rate was 42 percent for both carriers and noncarriers (Figure 1F).
|
This large prospective, population-based study provides robust estimates of the prevalence of mutations in DNA mismatch-repair genes in incident cases of colorectal cancer. We did not prejudice the analysis by using the family history or by first testing the tumor for microsatellite instability or expression of DNA mismatch-repair proteins. This strategy allowed us to identify predictors of mutational status using univariate analysis and to develop a two-stage, clinically driven predictive model using a multivariate analysis that estimates carrier probabilities. As was previously stated, the factors involved in the model have been fitted into the equation that appears in the Methods section. A clinician-friendly electronic version to estimate the likelihood that a given patient with colon cancer has a mutation is available at www1.hgu.mrc.ac.uk/Softdata/MMRpredict.php.25
The model was replicated in an independent series, even though the replication set consisted of younger patients and recruitment was retrospective. Among the 870 patients with cancer, the prevalence of the mutation was 4 percent, which causes some imprecision in the values in Table 3. For instance, testing for microsatellite instability combined with the use of Bethesda or Amsterdam criteria or stage 1 of our model (probability cutoff set at 0.05) gave sensitivities of 88 percent (95 percent confidence interval, 77 to 98 percent), 39 percent (95 percent confidence interval, 24 to 55 percent), and 65 percent (95 percent confidence interval, 50 to 80 percent), respectively, whereas positive predictive values were 32 percent (95 percent confidence interval, 23 to 41 percent), 100 percent (95 percent confidence interval, 82 to 100 percent), and 53 percent (95 percent confidence interval, 39 to 68 percent), respectively. All the confidence intervals are shown in Table 3 of the Supplementary Appendix.
Immunostaining of biopsy specimens obtained during colonoscopy (stage 2 of the model) is a feasible means of preoperative prediction of the carrier state. Of patients who underwent surgical resection, 80 percent underwent a preoperative endoscopic diagnostic biopsy. At the 0.05 cutoff, the model identifies a subgroup of 17 percent of patients for whom tumor immunostaining could inform a decision concerning the surgical procedure. The use of a combination of clinical measures and immunohistochemical staining of the tumor gives a positive predictive value of 80 percent and a sensitivity of 62 percent for mutation carriers. This information could be used in preoperative counseling about options for surgical prophylaxis, including total colectomy, rather than segmental resection, or combining colectomy with hysterectomy for postmenopausal women or those who do not want more children. Carrier identification could also be used to inform decisions about adjuvant therapy, since there may be differences between carriers and noncarriers in their responsiveness to chemotherapy.27,28 In all, combining the model with immunohistochemical data identifies only 1 in 29 of all patients (3.4 percent) as being likely to carry a mutation (positive predictive value, 80 percent) and so represents a highly efficient means of identifying patients for mutation testing.
By varying the cutoff values, the model allows resource-constrained health systems to optimize the efficiency of carrier detection by matching available financial and sample resources with the capacity for immunohistochemical analysis, microsatellite-instability analysis, and an evaluation of mutational factors.
The model requires cautious application, because its performance has been assessed in only 73 patients with mutations in the primary set of 870 patients and the replication set of 155 subjects. The prevalence of mutations reported by Aaltonen et al.6 and Hampel et al.8 was half that in our primary set (4 percent) and 1/10th that in the replication set (23 percent), which probably is a reflection of the age groups studied (mean ages, 68 years, 62.9 years, and 34 years, respectively), and reinforces our rationale for studying early-onset cases. The sex-specific difference in the frequency of mutations (a ratio of 1:17 for men and 1:37 for women) is noteworthy and consistent with sex-dependent penetrance effects10,12 and environmental factors, such as hormonal protective effects and sex-linked modifier genes. Such modifiers may explain sex-based differences in the population risk of colorectal cancer.
A few patients had the same mutations; some of these patients were related to each other, whereas others may have carried founder mutations. The proportion of MSH6 mutations (18.4 percent) is higher than that previously reported,29,30 although it is in accord with findings in families with the Lynch syndrome with predominantly extracolonic or late-onset colorectal cancers.31 A minority of tumors from MSH6 carriers had a high degree of microsatellite instability, as noted previously.29,30,32 The presence of other DNA mismatch-repair defects may increase the risk of cancer that does not have a microsatellite-instability phenotype — the MLH1 D132H variant, for example.33 The absence of microsatellite instability in these carriers emphasizes the importance of systematic approaches, without prescreening, to the identification of carriers among patients with incident cases.
Our study population was not categorized according to family history or patterns of referral, reducing the likelihood of bias. In contrast, previous reports23,34,35 studied selected populations referred to genetic centers. Amsterdam criteria seem to be insufficiently sensitive (42 percent) for routine use in incident cases, supporting our previous observations.7 Furthermore, 18 patients who fulfilled the Amsterdam criteria had no identifiable mutation. Some of these patients may have had genomic rearrangements of MSH6, intronic or promoter mutations, or PMS2 mutations, but use of the criteria for identifying hereditary nonpolyposis colorectal cancer fails to detect a clinically significant number of carriers. Although the Bethesda criteria are highly sensitive for the identification of carriers (95 percent), they require the analysis of microsatellite instability of 64 percent of tumors and three times as many analyses of germ-line genes as does our stage 1 model combined with immunostaining.
We found no significant difference in survival between genotypes, perhaps because of the relatively small number of carriers, but any possible trend is small and clinically irrelevant. This result concurs with results of population-based studies of family history36,37,38 but conflicts with results of retrospective analyses of families with the Lynch syndrome.15,16,17,39 Lead-time and selection bias probably explain these discrepancies. Thus, families with the Lynch syndrome with several surviving affected relatives are more likely to be included in retrospective studies. Effects on reproductive fitness also mean that fewer patients with alleles associated with a poor prognosis would be recruited. In a similar way, highly penetrant alleles impart excess rates of death with each incident tumor, as well as an increased likelihood of having multiple fatal synchronous or metachronous tumors. This contrasts with cases involving families with the Lynch syndrome who have a survival benefit as a result of surveillance.40 In any case, our finding emphasizes the importance of early cancer detection and prevention in patients with germ-line mutations in DNA mismatch-repair genes, whether or not such patients fulfill the criteria of having the Lynch syndrome.
Supported by a grant (C348/A3758) from the Cancer Research UK Programme, grants (K/OPR/2, /2/D333, and CZB/4/94) from the Scottish Executive Chief Scientist Office, and a grant (G0000657-53203) from the Medical Research Council.
No potential conflict of interest relevant to this article was reported.
We are indebted to Andrea Leitch, Diana Reinhardt, Naila Haq, Kathryn Drew, Antonella Maffe, and Alistair Thomson; to the nursing and office staff employed by the Colorectal Cancer Genetics Susceptibility study and the Scottish Colorectal Cancer study for their work in recruitment, especially Ruth Wilson, Nicola Cartwright, Maureen Edwards, Sheena MacDonald, Polly Somerville, Cathy Johnston, Jackie Kerrigan, Marie Manzi, Janet Chauhan, and Lisa Ferguson; to Jon Warner, Nicola Dunlop, and Austin Diamond at the Clinical Genetics Laboratory at the Western General Hospital, Edinburgh; to Stuart Bayliss, Lee Murphy, Ewan McDowall, Paul Fineron, Alistair Lessells, David Goudie, Zosia Miedzybrodzka, Neva Haites, Rosemarie Davidson, Ian Finlay, David Harrison, Frank Carey, Duncan Jodrell, Chris Twelves, David Brewster, and Roger Black; and to the Wellcome Trust Clinical Research Facility at the Western General Hospital in Edinburgh, the Scottish Cancer Registry, the Scottish Cancer Intelligence Unit of the Information and Statistics Division, and the Practitioner Services Division of the Scottish National Health Service for their excellent collaborative relationships; and to the surgeons, oncologists, pathologists, and colorectal cancer nursing teams in every Scottish hospital who made the work possible.
Source Information
From the Colon Cancer Genetics Group, School of Molecular and Clinical Medicine (R.A.B., A.T., S.M.F., H.C., M.G.D.), and the Public Health Sciences (H.C.), University of Edinburgh; and the Medical Research Council Human Genetics Unit (R.A.B., A.T., S.M.F., H.C., M.G.D.) and the Clinical Genetics Department (R.C., M.E.P.), Western General Hospital — all in Edinburgh; and the Research Institute in Healthcare Science, School of Applied Sciences, University of Wolverhampton, Wolverhampton, United Kingdom (I.D.N.).
Address reprint requests to Dr. Dunlop at the Medical Research Council Human Genetics Unit, Western General Hospital, University of Edinburgh, Crewe Road, Edinburgh EH4 2XU, United Kingdom, or at malcolm.dunlop{at}hgu.mrc.ac.uk.
References
| |||||||||||||||||||||||||||||||||||||||||||||||||||||
This article has been cited by other articles:
HOME | SUBSCRIBE | SEARCH | CURRENT ISSUE | PAST ISSUES | COLLECTIONS | PRIVACY | HELP | beta.nejm.org Comments and questions? Please contact us. The New England Journal of Medicine is owned, published, and copyrighted © 2008 Massachusetts Medical Society. All rights reserved. |
Previous
