Identification of a Quantitative MINT Locus Methylation Profile Predicting Local Regional Recurrence of Rectal Cancer

See commentary p. 2699

Rectal cancer is the second most common cancer of the digestive system in the United States (1). Worldwide, colorectal cancer has the second highest incidence (2), with rectal tumors constituting 33% of large bowel tumors (3). An important clinical feature of rectal cancer is its close anatomic relation to the small pelvis, which makes it prone to local recurrence after surgical removal even at early-stage disease. Locoregional recurrence occurs in approximately 10% of rectal cancer patients after total mesorectal excision (TME) surgery with curative intent (4, 5). However, locoregional recurrence is difficult to treat, with a very poor overall survival prognosis.

Fixation in the small pelvis makes malignancies of the rectum suitable for external beam radiation therapy. The Dutch multicenter TME clinical trial showed significant reduction of local recurrence by adding short-course (5 × 5 Gy) preoperative radiation therapy to TME (6). The multidisciplinary treatment of rectal cancer is a subject of many clinical trials randomizing patients to regimens that include various neoadjuvant therapies combined with radical surgery (7–10).

(Neo)adjuvant regimens aim at two clinical outcome parameters: improvement of local control and/or reduction of distant recurrence, the latter occurring in approximately 25% of patients after radical primary tumor resection (4). The improved local control shown by the TME clinical trial has not translated into an overall survival benefit in the trial analyses after six years of follow-up, and the data show that overall survival is determined predominantly by distant recurrence (4). The trial data further show that only 25% of nonirradiated tumors with distant spread also recur locally, whereas 60% of the locally recurrent tumors show distant disease spread (4, 6). The distant-spreading, nonlocally recurring tumor may therefore constitute a separate subclass of rectal cancer.

Allocation of patients to neoadjuvant therapies might lead to overtreatment; because 10% of rectal cancer patients will develop local recurrence, 90% of patients may be overtreated. Neoadjuvant therapies used in the treatment of rectal cancer have their specific morbidities, as has been shown for both radiotherapy (11) and combined chemoradiotherapy (12). It is therefore of great importance to define biomarkers that can categorize tumors into local and distant-spreading type preoperatively to target the multimodality treatment regimens towards a more systemic or local approach.

Molecular analysis of primary tumor tissues is an attractive form of preoperative diagnostics because rectal primary tumors are easily accessible for biopsy, which is routinely done in the preoperative work-up. In contrast to colorectal cancer, to date few biomarkers have been described to be predictive or of prognostic value specifically in rectal cancer. It is important especially in multivariate analyses to distinguish rectal cancer from other bowel adenocarcinomas or colon cancer. Specific factors need to be taken into account for rectal cancer as compared with colon cancer, such as circumferential margin involvement, type of surgical procedure, or distance of the tumor to the anal verge (13).

Epigenetic changes, such as changes in DNA methylation status, are regarded as early events contributing to carcinogenesis (14). Methylation of cytosine residues in DNA is one of the mechanisms regulating transcriptional activity (15). In cancer, aberrant DNA hypermethylation of specific regions as well as global hypomethylation is observed (16). In this study, we investigated epigenetic changes in rectal cancer and specifically methylation of methylated-in-tumor (MINT) loci. MINT loci are CpG dinucleotide-rich regions located in nonprotein-encoding DNA regions and have been reported to become methylated in a tumor-specific manner in (colo)rectal cancer (17, 18), gastric cancer (19), and recently malignant melanoma (20).

In a previous study, we quantitatively studied methylation of MINT loci in premalignant stages of rectal cancer and showed that the MINT methylation index increases during adenomatous transformation of normal epithelium (18). We also identified a quantitative, two-MINT methylation biomarker panel (MINT3 and MINT17) that is predictive of distant recurrence in early, node-negative rectal cancers patients (18). In the present study, we assessed the value of this panel to predict local recurrence of patients enrolled in the multicenter, randomized, clinical TME trial.

Primary paraffin-embedded archival tissue (PEAT) specimens were obtained from 325 nonirradiated patients enrolled in the TME multicenter clinical trial (4). Patients used in this study fulfilled the following criteria: nonirradiated, tumor-node-metastasis (TNM) stage I-III, with no evidence of disease after surgical resection. Using power calculations, a sample size of 250 patients was calculated to be sufficient to obtain statistical significance for predicting recurrence (with α = 0.05 and a power of 90%), as described previously (18). Allowing for 30% loss of patient samples due to availability and quality of PEAT and DNA, 75 additional PEAT specimens were collected. In 11 patient PEAT blocks, tumor tissue was no longer present on the section. DNA was isolated from 314 randomly selected patient specimens with sufficient tumor cell numbers. After processing and sodium bisulfite treatment, samples of 251 patients yielded sufficient amount of quality DNA for absolute quantitative assessment of methylated alleles (AQAMA). The selected group of patients analyzed in our previous study (18) did not differ from nonselected patients in the nonirradiated treatment arm or from the complete trial population. Trial eligibility criteria and follow-up protocols have been described previously (4, 21, 22). Nonirradiated patients were selected because the predictive value of the tested biomarkers for local recurrence probability should be tested in patients who did not receive adjuvant therapy, as this affects local recurrence.

For external validation studies, 43 additional TME trial patients that adhered to the same selection criteria were added. Further, 42 nonirradiated patients who participated in the transanal endoscopic microsurgery (TEM; ref. 23) were added and these have been previously described (24). DNA was extracted from adenomatous and cancer tissues. The adenomas were subdivided into cases consisting of only adenoma tissue in the resection (A; n = 21) and adenoma fractions of cases with a carcinoma focus infiltrating at least in the submucosa (A, C+; n = 8). The carcinomas were subdivided into groups: tumor fractions consisting of a mixture of adenoma and carcinoma tissue (C+A; n = 6) and only carcinomas (C; n = 7). We further collected normal rectal epithelial tissue from the tumor specimen resection margins as normal controls from 19 patients operated for rectal cancer at the Saint John's Health Center.

From the TME trial patient's primary tumor PEAT specimens, 7-μm tissue sections were cut and mounted on nonadhesive glass slides. Tumor-representative areas on H&E-stained sections were identified and marked by a surgical pathologist specializing in rectal cancer (JHJMvK). Per patient, two tissues sections were deparaffinized. DNA was isolated from microdissected tissue from the marked areas and modified by sodium bisulfite, as previously described (25). Salmon sperm DNA was added as a carrier (26). Before and after sodium bisulfite modification, double-stranded and single-stranded DNA were quantified using PicoGreen and OliGreen assays (Molecular Probes; Invitrogen), respectively. Sufficient input DNA for AQAMA was determined as described (17). To assess background signal, a salmon sperm DNA sample without tumor DNA was included in all assays in triplicate. To prevent any bias, PEAT blocks and isolated DNA samples were coded.

Primary tumor methylation levels of MINT loci 1, 2, 3, 12, 17, 25, and 31 were assessed by AQAMA in triplicate (17). Controls for specificity of AQAMA for both methylated and unmethylated sequences, as well as controls for nonspecific amplification, were included (17, 27). As a final outcome of the analysis, the methylation index (MI) of a sample was the mean value of three MI measurements that were calculated for each well as follows: [copy number_{methylated alleles}/(copy number_{methylated alleles} + copy number_{unmethylated alleles})]. Of these three measurements the SD was calculated. Values with a SD <0.1 were accepted and used in analysis. When the SD was >0.1 this was in most cases due to one or two of the three methylated or unmethylated measurements being undetermined. These cases were identified and the MI was then calculated as mean methylated copy number/(mean methylated copy number + mean unmethylated copy number). If one of three methylated or unmethylated measurements could not be determined this value was not incorporated in the calculation. If two or more of three measurements could not be determined the value was zero. Cases with six complete results and with SD > 0.1 were individually evaluated. When after omitting the most deviate MI value the SD was still >0.1 the sample was excluded from analysis.

Differences in recurrence probability, survival, and clinical and tumor pathologic factors were analyzed between TME trial patients assigned by unsupervised random forest (RF) clustering using MINT3 and MINT17 as described (18). Unsupervised RF clustering dissimilarity algorithms are based on individual decision tree predictors, and it automatically dichotomizes the expressions into clusters in a data-driven approach (28). Groups are therefore not established by employing cutoffs. The internal validation of RF clustering further eliminates the need for separate training and validation sets to test reproducibility of cluster formation. External validation of the data in an independent patient group is still required. The Gini-Index (GI; refs. 29, 30) of each input variable (MINT locus) was given. This index measures the inequality of two distributions and is defined as the ratio between the area spanned by the observed cumulative distribution and the area of a uniform hypothetical cumulative distribution for a nondiscriminating variable. A higher GI shows increasing inequality and therefore higher discrimination of that input variable.

Specimens yielding insufficient or low-quality DNA for PCR were excluded. To compare ordinal variables, Mann-Whitney and Kruskal-Wallis U tests were carried out. Differences in age were assessed using t tests. Cumulative incidences, accounting for death as a competing risk, were used to visualize survival differences (31), and significance was assessed by the log-rank test. For multivariate analysis the Cox proportional hazards model was used, with results presented as hazard ratios and 95% confidence intervals. Covariables entered in the model included T stage, N stage, circumferential margin status, distance of the tumor from the anal verge, and tumor differentiation. All clinical correlative analyses with identified clusters were done using SPSS statistical software (version 16.0.1, SPSS Inc.). A two-sided P value of ≤0.05 was considered statistically significant. Data on patients alive or free of recurrence were censored at the time of the last follow-up.

In our previous study, methylation levels at the seven MINT loci were measured in normal, adenomatous, and cancer rectal tissue (18). The results showed that in normal tissue, all MINT loci, except for MINT17, were mostly unmethylated. Significantly higher methylation of MINT2, MINT3, and MINT31 was detected in adenoma and cancer tissue compared with normal tissue. Further analysis of the quantitative data showed nonparametric distributions indicating the presence of subgroups for MINT1, MINT2, MINT12, MINT17, MINT25, and MINT31 in adenoma and cancer tissue. Based on these findings, we concluded that all seven MINT loci had potential to subclassify rectal cancer patient groups with corresponding methylation level patterns. Quantitative methylation data of the seven MINT loci were then used to do unsupervised RF clustering analysis. An important aspect of RF clustering is that it eliminates a validation-training approach because of its internal validation quality. External validation of the data is still required. The multidimensional scaling plot as an outcome of using all seven MINT loci in the RF clustering algorithm suggested the presence of two clusters (18). The GI of the RF analysis, indicating variable importance, appointed MINT3 and MINT17 as the two MINT loci that carried the most information to form the clusters (18). GI being a measure of inequality, the higher the GI, the more the clusters can be considered different based on that specific biomarker. MINT3 and MINT17 were shown to have the highest GI (20.2 and 20.7, respectively), compared with the five other MINT loci (range, 6.0-13.5; ref. 18). The multidimensional scaling plot of the RF clustering using only these two biomarkers showed four clearly separate groups (18). Methylation level differences at the seven studied MINT loci between the four clusters are given in Supplementary Table S1. For a more simplified representation, individual patient MI values of MINT3 and were rendered in XY-plots. In Fig. 1A the patients are labeled according to the outcome of the RF analysis using seven MINT loci that showed the presence of two clusters. In Fig. 1B the patients are labeled according to the outcome of the RF analysis using only MINT3 and MINT17 that showed the presence of four clusters. Figure 1B further shows that the groups can be divided by a cutoff of MI = 0.73 for MINT3 and MI = 0.14 for MINT17 with almost no misclassification. This is important because such a two-dimensional cutoff algorithm can be used for validation experiments (see below). The two-biomarker cluster allocation of Fig. 1B continued to be used for the clinical correlation studies.

Next, we were interested in comparing the probability of locoregional recurrence among the four clusters. As shown in Fig. 2A, the incidence of locoregional recurrence was lowest for patients in cluster 3. The quantitative methylation pattern of cluster 3 patients was that the tumors showed significantly higher methylation at MINT3 and lower methylation at MINT17. This pattern corresponds with the clinically relevant rectal cancer patient cluster at high risk for distant recurrence and decreased cancer-specific and overall survival that we previously identified (18). Clusters 1, 2, and 4 showed similar probability outcomes. The difference in local recurrence probability became more evident and reached statistical significance after combining clusters 1, 2, and 4 (P = 0.03; Fig. 2B). This result shows that the specific combination of increased methylation at MINT3 and decreased methylation at MINT17 is predictive of reduced local recurrence probability. From these data it was shown that quantitative methylation assessment of MINT3 and MINT17 identifies a patient group with an inversed risk incidence for distant recurrence and local recurrence.

Supplementary Table S1 shows the significance of MI differences between cluster 3 and the combined clusters 1, 2, and 4. It is shown that the 95% confidence intervals of MINT3 and MINT17 do not overlap, indicating that the rectal cancer patient groups' epigenetic classification is very distinct. The results also show that MINT1 and MINT2 can be helpful as discerning biomarkers for both patient groups.

Using univariate analysis, standard clinicopathologic parameters including sex, age, TNM stage, N status, tumor differentiation, location of distant recurrences, resection type, and circumferential margin status were compared between cluster 3 and the combined group of clusters 1, 2, and 4. These parameters did not significantly differ between the two patient groups identified based on methylation levels of MINT3 and MINT17 (Table 1). In addition, no significant difference was observed for these standard clinicopathologic factors between the patient groups when nodal status was taken into account (Supplementary Table S2).

To assess whether the observed prognostic value of the clusters was independent of standard prognostic variables, we carried out multivariate analyses. The Cox regression method was used to analyze standard prognostic factors of rectal cancer: T stage, N stage, circumferential margin status, distance of the tumor from the anal verge, and tumor differentiation (Table 2). Based on the epigenetic subclassification, the multivariate analysis showed patients of clusters 1, 2, or 4 to be at significant, >10-fold, increased risk of local recurrence compared with cluster 3 patients. Nodal involvement was further significantly associated with increased local recurrence incidence, and tumor involvement of the circumferential resection margin showed borderline significance.

Because we found the nonirradiated cluster 3 patients to be at reduced risk for local recurrence, we were interested to see how the recurrence probability rates of this group compared with those of preoperatively irradiated patients from the TME trial (Fig. 3). The irradiated patients were selected using the same clinical parameters that did not differ significantly from the nonirradiated selected patients (data not shown). The difference in the locoregional recurrence rate in cluster 3 patients was 3% versus 4.7% in patients who received 5 × 5 Gy before surgery (P = 0.43). Patients of high-risk cluster 1, 2, or 4 had significantly higher recurrence probability over time postoperatively compared with irradiated patients (P < 0.0001). These results indicate that patients with TME alone identified by the quantitative two-marker MINT profile have at least a comparable, locoregional recurrence rate compared with irradiated patients. This group will likely not be disadvantaged when preoperative radiotherapy before TME is not given and can be spared from unnecessary treatment morbidity.

To test the reproducibility of cluster allocation by the cutoffs established in Fig. 1B we measured MINT3 and MINT17 methylation levels in primary tumor tissues of 43 additional TME trial patients, and an independent group of 42 patients consisting of rectal adenomatous and cancer tissue from patients treated by TEM. Figure 4 shows the results, with 19% (8 of 43) of TME and 19% of TEM patients allocated to the prognostic cluster 3. The suggested cutoffs therefore show allocation of an independent group of rectal cancer patients, including patients with adenomas and very early disease stage, consistent in size and comparable with the size of the test group patients (27%). In the additional TME patient group there was only one event for local recurrence (which did occur in the high-risk cluster). The TEM group was treated differently and also contained premalignant stages and therefore was not comparable. Further external validation of these clinical findings in future studies is needed.

To show the MI of the two biomarkers in normal rectal mucosa we added those data of 19 specimens to Fig. 1B. The results of the normal rectal tissue analysis shows that these are well separable from a cluster 3 patient, which is important as these patients may be selected for treatment of their rectal tumor by surgery alone.

Although recurrence rates have decreased to about 10% after the introduction of TME surgery, locally recurrent cancer remains an important clinical problem (1). In a previous study (18), MINT methylation was shown to increase early during tumor progression, indicating that methylation of MINT loci is a factor acquired early during rectal tumorigenesis, and therefore can be used to subclassify early disease. Preoperative molecular profiling of the primary tumor could potentially be of great value in identifying patients at high risk of developing local recurrence. This study shows that based on absolute quantitative methylation levels of the MINT3 and MINT17 loci, rectal cancers with a high risk of local recurrence can be identified.

The MINT3 locus CpG island is localized on chromosome 1p34-35 just downstream of RBBP4 (retinoblastoma-binding protein 4; ref. 32), SYNC1 (encoding the syncoilin protein involved in cell cytoskeleton and extracellular matrix proteins; ref. 33), YARS (tyrosyl-tRNA synthetase, involved in angiogenesis; refs. 34, 35), and s100p-binding protein (s100p is overexpressed in many solid tumors; refs. 36, 37). MINT17 is localized on the long arm of chromosome 12, just upstream of the Harakiri (HRK) gene, a member of the BCL2 gene family, which encodes apoptosis regulatory proteins and its expression in gastrointestinal cancers is known to be regulated by promoter region methylation (38). A regulatory or functional relation between the MINT3 and MINT17 CpG islands and the above mentioned genes has not been established. MINT locus methylation may therefore be seen as a surrogate biomarker. In the literature, several studies have shown the clinical utility of MINT loci methylation as a predictive biomarker in colon and gastric cancer, melanoma, and renal cell carcinoma (18, 39–42). Our large-scale study adds novel findings that MINT locus methylation levels may have utility in predicting rectal cancer local recurrence. Importantly, in this study we propose an algorithm using only two MINT biomarkers with a simple cutoff. We showed validation of the algorithm to have excellent clustering capacity, even in early-stage disease, and the ability to separate cases well from normal rectal tissue.

This is the second study specific for rectal cancer from our group that shows the clinical utility of MINT3 and MINT17 methylation levels. Strong probability differences between the two patient groups could be shown for local recurrence, with cluster 3 patients having >10-fold increased risk of developing local recurrences than patients allocated to the other clusters. Local recurrence rates of cluster 3 patients were comparable with those of irradiated rectal cancer patients, and this shows that leaving out preoperative radiation can be done safely with the advantage of reducing treatment morbidity. This molecular stratification approach needs to be further investigated in a randomized multicenter trial to be validated.

Subdivision according to nodal status (data not shown) showed that patients in cluster 3 with a positive nodal status were at significantly increased risk of local recurrence. Node-negative patients in cluster 3, however, showed a significantly decreased cancer-specific and overall survival compared with the other clusters in our previous study (18). This is explained by the fact that these node-negative patients were found to be at significantly increased risk of distant recurrence in our previous study (18). Early metastasizing of rectal cancer, in the absence of evident nodal spread, may occur via the hematogenic route. This is supported by our group that circulating tumor cells can be detected in peripheral blood of early stage I/II colorectal cancer patients which has prognostic clinical utility (43). Our findings are in accordance with data from the TME trial showing that survival is determined predominantly by distant and not by locoregional recurrence (4). This suggests that nonlocally recurrent and distantly spreading rectal cancer constitutes a separate subclass of rectal cancers that can be identified by our two-biomarker MINT methylation profile.

The TME trial showed reduction of rectal cancer local recurrence rates by adding preoperative radiotherapy (7). A reduction of distant recurrence of rectal cancer, as well as improved disease-free and overall survival, has been shown in several randomized controlled trials, either alone or in combination with radiotherapy (44–46) using adjuvant chemotherapy. This is the first study to show that locoregional recurrence patterns of American Joint Committee on Cancer stage I, II, and III rectal cancers can be distinguished using preoperatively assessable, quantitative epigenetic subclassification of primary rectal tumor tissue. Our previous study showed the methylation status of the described MINT loci to have utility in predicting distant recurrence probability in stage I and II disease. Therefore, a new treatment stratification approach for rectal cancer can be suggested as follows: after preoperative assessment of primary tumor biopsy specimens of MINT3 and MINT17 methylation levels, about 30% of patients could be spared from preoperative radiation therapy, but might benefit from systemic treatment. The other 70% should receive preoperative radiotherapy, and if node-positive, postoperatively systemic treatment can be considered. The identification of patients who do not need preoperative radiotherapy would likely reduce long-term morbidity and improve quality of life (47). The results of our study should be further evaluated to improve planning of therapeutic regimens for rectal cancer patients in the preoperative phase.

No potential conflicts of interest were disclosed.

Grant Support: Ruth and Martin H. Weil Fund, the Leslie and Susan Gonda (Goldschmied) Foundation, and the Roy E. Coates Research Laboratories.

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