Article Text
Abstract
Objective Pathological staging used for the prediction of patient survival in colorectal cancer (CRC) provides only limited information.
Design Here, a genome-wide study of DNA methylation was conducted for two cohorts of patients with non-metastatic CRC (screening cohort (n=572) and validation cohort (n=274)). A variable screening for prognostic CpG sites was performed in the screening cohort using marginal testing based on a Cox model and subsequent adjustment of the p-values via independent hypothesis weighting using the methylation difference between 34 pairs of tumour and normal mucosa tissue as auxiliary covariate. From the 1000 CpG sites with the smallest adjusted p-value, 20 CpG sites with the smallest Brier score for overall survival (OS) were selected. Applying principal component analysis, we derived a prognostic methylation-based classifier for patients with non-metastatic CRC (ProMCol classifier).
Results This classifier was associated with OS in the screening (HR 0.51, 95% CI 0.41 to 0.63, p=6.2E−10) and the validation cohort (HR 0.61, 95% CI 0.45 to 0.82, p=0.001). The independent validation of the ProMCol classifier revealed a reduction of the prediction error for 3-year OS from 0.127, calculated only with standard clinical variables, to 0.120 combining the clinical variables with the classifier and for 4-year OS from 0.153 to 0.140. All results were confirmed for disease-specific survival.
Conclusion The ProMCol classifier could improve the prognostic accuracy for patients with non-metastatic CRC.
- cancer epidemiology
- colorectal cancer
- methylation
Statistics from Altmetric.com
Significance of this study
What is already known on this subject?
Harmful overtreatment and undertreatment of some patients with colorectal cancer (CRC) due to imprecise prognosis prediction based on the traditional tumour, node, metastases system highlights the need of additional prognostic markers.
DNA methylation is known to be involved in CRC formation and progression and recent studies indicate its prognostic impact.
Well-known examples in other cancers (hypermethylation at the genes MGMT in glioma and GSTP1 in prostate cancer) show the clinical relevance of DNA methylation as potential diagnostic, prognostic and predictive markers.
What are the new findings?
This study investigated the genome-wide DNA methylation status of two large independent cohorts of patients with non-metastatic CRC from a multicentric, population-based case–control study.
We successfully constructed a prognostic classifier, based on 20 CpGs, called the prognostic methylation-based classifier for patients with non-metastatic CRC (ProMCol classifier).
We clearly demonstrated that the usage of the ProMCol classifier improved prognosis prediction for overall survival and disease-free survival in both cohorts.
The validation of the ProMCol classifier in an independent cohort revealed a significant reduction of the prediction error.
How might it impact on clinical practice in the foreseeable future?
A combination of the novel ProMCol classifier and standard clinical variables can predict patients’ prognosis considerably more accurately.
An improved risk stratification of patients with non-metastatic CRC after surgery is essential in order to provide more effective and personalised adjuvant treatment strategies.
Introduction
Colorectal cancer (CRC) is the third most common cancer worldwide accounting for 1.36 million new cases annually.1 The 5-year survival rate of patients is highly dependent on the stage of the disease.2 At present, the most accurate means for the prediction of patient survival remains pathological staging according to the tumour, node, metastases (TNM) system, but it has been recognised that even patients within the same tumour stage display a strong heterogeneity for prognosis and treatment response.3 4 Especially for stage II patients, there is an ongoing debate if adjuvant chemotherapy should be recommended or not.4–6 The current classification system provides only limited information for the clinical prognostication highlighting the need for additional prognostic markers to avoid a potential undertreatment or overtreatment of patients.
In the research field of diagnostic, prognostic and predictive biomarkers for cancer, DNA methylation has gained increasing prominence.7–10 DNA methylation, predominantly defined as an addition of a methyl group at cytosine residues located adjacent to guanine bases (CpG dinucleotides), is one of the major epigenetic mechanisms, important in many physiological and pathophysiological processes.11 Since many years, the dysregulation of DNA methylation has been known to have a key role in cancer development and progression.12
Meanwhile, the accessibility of genome-wide methylation profiling revealed innovative prognostic methylation markers in different cancers13–17 but several of the studies on CRC prognosis were performed on a limited sample size. So far, among all methylation-based prognostic biomarker candidates for CRC, the CpG island methylator phenotype (CIMP) status has been the most promising indicator for prognostication,8 however, to date there is no consensus definition for CIMP leading to inconsistent study results.18 19 A better understanding of this disease and detection of novel markers for the prediction of patients’ prognosis may enable individual treatment options in order to improve patients’ outcomes and quality of life.
In a genome-wide approach investigating a large, multicentric study cohort, a prognostic methylation-based classifier for patients with non-metastatic CRC, the ProMCol classifier, was identified and successfully validated in an independent study cohort.
Materials and methods
Study cohort
All included patients were enrolled in the ongoing population-based case–control study DACHS (Darmkrebs: Chancen der Verhütung durch Screening), described in detail elsewhere.20 21 In total, 22 hospitals in the study area of the Rhine-Neckar-Odenwald region of south-western Germany were involved in recruitment. Extensive patient information was gained by standardised questionnaires, carried out by trained interviewers. Follow-up (FU) information concerning vital status, date and cause of death was obtained by the local population registries and health authorities 3 and 5 years after CRC diagnosis.22 The study was approved by the ethical committees of the Medical Faculty of the University of Heidelberg and of the Medical Chambers of Baden-Württemberg and Rhineland-Palatinate. Written informed consent was obtained from each participant. In the present study, only patients with non-metastatic disease recruited between 2003 and 2007 were included (for whom 5-year FU was complete at the time of this analysis). Genome-wide DNA methylation data are so far available for patients recruited until 2007 only. Eligible for the screening cohort (n=689) were all DACHS patients whose tumours were examined in the pathology institutes in Heilbronn, Ludwigshafen, Mannheim and Speyer, whereas for the validation cohort (n=408) only tumours examined in the pathology institute of Heidelberg were selected. In addition to the patients’ tumour tissues, the validation cohort contained corresponding normal mucosa tissue samples, enabling the analysis of tumour and normal mucosa tissue pairs for 34 patients. Patients who had received neoadjuvant therapy were excluded for the survival analysis. The median FU time for all patients in the screening and the validation cohort was approximately 5 (4.99) years. Table 1 gives an overview of the clinicopathological characteristics of the included patients.
DNA isolation
Tissue samples were collected from all de-central pathology institutes in the tissue bank of the National Center for Tumor Diseases (NCT, Heidelberg, Germany). All samples were provided in accordance with the regulations of the tissue bank and the approval of the ethics committee of Heidelberg University. For each sample, tumour DNA was isolated from four formalin-fixed and paraffin-embedded (FFPE) tissue slices à 5 µm. A haematoxylin-stained slice of every tumour block was evaluated by an experienced pathologist in order to mark the regions with high tumour cell content. After deparaffinisation, the DNA of the manually microdissected tumour tissue of the screening cohort was isolated following a semiautomated protocol using the Maxwell 16 MDx instrument (Promega, USA) in combination with the DNeasy Blood & Tissue Kit (Qiagen, Hilden, Germany) whereas the DNA of the validation cohort was extracted manually with the DNeasy Blood & Tissue Kit according to manufacturer’s recommendations. For both cohorts, the isolated DNA was eluted with 50 µl elution buffer.
Methylation profiling using the Infinium HumanMethylation450 BeadChip array (Illumina)
DNA quality and quantity was checked using the Quant-iT PicoGreen dsDNA Assay Kit (Thermo Fisher Scientific, San Diego, USA) and the integrated Infinium HD FFPE QC Assay Kit (Illumina, USA). Samples were randomised in 96 sample batches according to Optimal Sample Assignment Tool23 and replicates were included to check for batch effects. A minimum of 100 ng DNA, but if available, 250 ng DNA diluted in 45 µl RNase-free water was used per sample for the following experimental steps of the methylation profiling, all performed according to manufacturers’ recommendations. After bisulfite conversion using the Zymo EZ-96 DNA Methylation Kit (Zymo Research, USA), the DNA restoration step (Illumina) for FFPE samples was conducted, followed by the genome-wide methylation analysis using the Infinium HumanMethylation450 BeadChip (Illumina, USA) interrogating over 485 000 CpG loci. Samples of the screening cohort were analysed more than 1 year apart from samples of the validation cohort. The methylation score of each CpG site was defined as β-value, ranging from 0 to 1, with 0 representing a completely unmethylated status and 1 representing a fully methylated status.
Preprocessing and normalisation of methylation data
Preprocessing and statistical analyses were all performed using the computational environment R, V.3.3.1 (http://www.r-project.org/). Raw data files generated by the iScan array scanner were read and preprocessed using the ‘minfi’ package, included in the Bioconductor collection of R packages. The standard Illumina normalisation procedure (‘preprocessIllumina’) was used to correct for technical differences between the Infinium I and II assay designs. For all analyses, filtering criteria were applied according to Sturm et al.24 In the screening cohort, probes that failed in more than 10% samples, based on detection p-value using a significance level of 0.01, were excluded. To allow for an independent validation, no filtering was applied to the validation cohort. Instead, here exactly the same CpG sites were used as in the screening cohort.
Missing information in all clinical variables, except for microsatellite instability (MSI), were imputed with the R-package ‘mice’ using information from the other clinical variables as listed in table 1. For the imputation of MSI status, a random forest based on 100 CpG sites was applied. These 100 CpG sites were pre-elected using distance correlation sure independence screening.25
Statistical analyses
A paired Wilcoxon signed-rank test was used to find the differentially methylated CpG sites between the tumour and normal tissue pairs. The difference of tumour and normal tissues for every CpG site was estimated via the (pseudo-)median of the sample differences. For statistical inference, a filtering step for SD >0.05 was applied to the selected CpG sites. A variable screening for prognostic CpG sites for overall survival (OS) was performed for single CpG sites using marginal testing based on a Cox model adjusted for age as continuous covariate and gender, smoking behaviour, MSI status, tumour stage (II vs I, III vs I) and tumour location (colon proximal, colon distal or rectum) as categorical covariates. Subsequently, the p-values were adjusted via independent hypothesis weighting (IHW)26 with 20-fold and a significance level of 0.1. As auxiliary covariate for IHW, the estimate of the mean difference in methylation between tumour and normal mucosa tissue was used as ordinal covariate, manually partitioned into 16 intervals ([−1,–0.35), [−0.35,–0.3), [−0.3,–0.25),…, [0.25,0.3), [0.3,0.35), [0.35,1]). For the 1000 CpG sites with the smallest adjusted IHW p-value, we evaluated the apparent Brier score regarding 3-year survival in the screening cohort, using the R package ‘pec’. Finally, the 20 CpG sites with the smallest apparent Brier score in the screening cohort were selected for the construction of the methylation-based prognostic classifier. The first principal component of the β-values of these 20 CpG sites was calculated. The ProMCol classifier was obtained by multiplying the β-values of the 20 CpG sites with the respective weights of the first principal component (this is essentially the score of the first principal component omitting the centring of the β-values). For determining the number of CpG sites that we used for the classifier, we applied a 10-fold internal crossvalidation approach with three repetitions.
Two Cox models were fitted using the R package ‘survival’. The first model included only the clinical covariates (as specified above), the second one included both the clinical covariates and the prognostic classifier. The given HRs are based on a change from the lower to the upper quartile of the prognostic classifier. Both the Cox model based only on clinical covariates and the Cox model based on clinical covariates and the prognostic classifier (which were fitted using the screening cohort data) were independently validated in the validation cohort. The prediction error curve and the Brier score for the 3-years survival in the validation cohort were evaluated using the loss function approach described in Gerds and Schumacher27 using the R package ‘pec’. The area under the curves (AUCs) were calculated following the incident/dynamic approach described in the work of Song and Zhou28 using the R package ‘survAUC’.
Complementary, we performed a replication analysis, fitting new Cox models to the validation cohort. Kaplan-Meier curves were generated for the two subgroups ‘ProMCol high’ and ‘ProMCol low’ using the median value of the prognostic classifier as cut-off. Log-rank tests were used to assess the significance of the survival differences between the groups. Moreover, the Kaplan-Meier plots support the assumption of proportional hazards. In all analyses, a two-tailed significance level of 0.05 was used.
Results
Analysis of CpG-based methylation data from the genome-wide analysis
After preprocessing and filtering for SD >0.05, 269 306 probes were included in the analysis. The replicates (n=11) analysed in both cohorts showed correlations with Spearman correlation coefficient r ≥0.98. As the tumour and normal mucosa tissue pairs (n=34), available from the validation set, were used for the IHW procedure these tumour samples were excluded from the validation of our classifier later on. Of 688 patients in the screening cohort and 366 residual patients in the validation cohort, 52 patients had to be excluded because of neoadjuvant treatment. Highly distinct pathological and clinical characteristics of patients with non-metastatic and metastatic disease require a separate analysis of these two patients groups. As the statistical power was too limited for the analysis of the metastatic cases (94 patients in the screening and 52 in the validation cohort), impeding the identification of reliable prognostic biomarkers, we focused our analysis on the patients with non-metastatic CRC. Finally, 572 patients with non-metastatic disease were included for further analysis in the screening cohort and 249 patients in the validation cohort. An overview over the study design and the analysis steps is given in online supplementary figure 1.
Supplementary file 1
Identification of candidate CpG sites for the OS of patients with non-metastatic CRC
As a starting point, we determined differentially methylated CpG sites across the genome in the analysed 34 tumour and normal mucosa tissue pairs (see online supplementary figure 2). The difference in methylation between tumour and normal tissue served as covariate for adjustment of the p-values via IHW in the Cox model, used for the identification of prognostic CpG sites for OS. IHW improves the power of large-scale multiple testing by adjusting the p-values using weighted multiple-testing using data-driven weights. Online supplementary figure 3 illustrates the assigned weights according to the β-value difference of tumour and normal mucosa tissue. As a result of the IHW procedure, we obtained high weights for CpG sites with low methylation levels in tumour compared with normal mucosa tissue. In contrast, the IHW procedure assigned no weight to CpG sites with higher methylation levels in tumour tissue than in normal mucosa tissue. Consequently, the IHW procedure removed 114 405 CpG sites without a positive weight (weights=0) from the further model building process.
Construction of the ProMCol classifier in the screening cohort
After the adjustment of the p-values via IHW, only the 1000 CpG sites showing the smallest p-values in the survival analysis were used for the construction of the prognostic classifier using the screening cohort. The Brier score, a measure for the accuracy of probabilistic predictions, was calculated for the 3-years survival of the patients with non-metastatic disease. The 20 CpG sites with the best Brier score (depicted with their characteristics in table 2) were selected for the prognostic classifier.
The informative values of all 20 CpG sites were positively correlated (Pearson correlation coefficient r=0.19–0.68, median r=0.44) among each other. Moreover, they all were negatively correlated with the patients’ stage (Spearman correlation coefficient r =−0.25 to −0.04, median r=−0.17). Remarkably, 13 of the 20 CpG sites included in our classifier were determined as deoxyribonuclease I-hypersensitive sites (DHSs) and 14 were assigned to enhancer regions. The results of a principal component analysis (PCA) revealed that a linear combination based on the first principal component of all 20 CpG sites is sufficient for an appropriate presentation of their information value (see online supplementary figure 4). Using the weights of the first principal component, we derived a methylation-based prognostic classifier for patients with non-metastatic CRC, called ProMCol classifier (the individual PCA weights for the single CpG sites are given in table 2), based only on the screening cohort. The β-values of each CpG site were multiplied with the individual PCA weights and added up to the final score.
Depending on the methylation status of the 20 CpG sites, the ProMCol classifier can theoretically adapt values between 0.00 and 4.30, while 0.00 represents a completely unmethylated status in all 20 CpG sites and 4.30 a complete methylation in all CpG sites. In our screening cohort, we received ProMCol classifier values between 1.20 and 3.51. The correlation between the ProMCol classifier and the patients’ stage was r=(−0.20).
Associations of the ProMCol classifier with survival of patients with non-metastatic CRC
Using the ProMCol classifier, we fitted a Cox regression analysis in the screening cohort. In one model, only the standard clinical variables (gender, age, tumour stage, tumour location, smoking behaviour and MSI status) were included, in the other model the clinical variables in combination with the ProMCol classifier. Here, the ProMCol classifier was significantly associated with OS of patients with non-metastatic disease with HR 0.51, 95% CI 0.41 to 0.63 and p=6.2E−10 (results are shown in table 3A). Patients with a high methylation status showed a better prognosis for OS than patients with a low methylation status, meaning the higher the ProMCol classifier value, the better the patients’ prognosis. For an independent validation of the ProMCol classifier, the prediction error for the OS of patients with non-metastatic CRC was calculated in the validation cohort. This analysis revealed a clear advantage in prediction probability for adding the ProMCol classifier to the model (figure 1A). Whereas the prediction error, calculated by the model using only clinical variables, was 0.127 for the 3-year survival of the patients, a model combining the clinical variables with the ProMCol classifier improved the prediction with a smaller error value of 0.120. For the 4-year survival, the prediction error could be reduced from 0.153 with clinical variables to 0.140 using the combination of clinical variables and the ProMCol classifier. As an addition, we computed a time-dependent AUC (figure 1B), showing for the 3-year (4-year) OS of the patients an increase of AUC from 0.705 (0.704) calculated only with clinical variables to 0.750 (0.743) calculated with clinical variables and the ProMCol classifier.
Unadjusted Kaplan-Meier plots for OS are shown for all patients with non-metastatic disease (stage I–III) and specified for stage II and III patients dependent on the methylation status of the ProMCol classifier in both cohorts in figure 2. Patients with a high methylation status expressed by the ProMCol classifier value >2.86 (corresponding to the median of the classifier in the screening cohort) clearly showed a better prognosis for OS than patients with a low methylation status corresponding to a classifier value ≤2.86 (p=3.0E−5 in the screening cohort and p=8.5E−4 in the validation cohort). After 3 (5) years of FU, 92.8% (82.5%) of the patients with CRC in the screening cohort with a ProMCol classifier value >2.86 in contrast to 79.2% (68.2%) with a classifier value ≤2.86 were still alive. In the validation cohort, the absolute 3-year and 5-year survival rates of patients with CRC with a ProMCol classifier value >2.86 were 90.1% and 82.4%, respectively, compared with 75.4% and 66.1% for patient with a low ProMCol classifier value (≤2.86).
To ensure the specificity for CRC, we analysed the association of the ProMCol classifier with disease-specific survival (DSS) of patients with non-metastatic disease according to our procedure for OS. The multivariate Cox regression analysis revealed a significant association in the screening cohort with HR 0.50, 95% CI 0.38 to 0.67, p=1.40E−6 (table 3B). In the independent validation, the prediction error concerning DSS was reduced from 0.097 (calculated only with clinical variables) to 0.092 for the 3-year survival and from 0.124 to 0.112 for the 4-year survival (figure 1B). Corresponding to this, the AUC for DSS (figure 1D) was increased from 0.728 to 0.772 for the 3-year survival and from 0.728 to 0.768 for the 4-year survival.
As an addition, we computed a dynamic AUC (figure 1B), showing for the 3-year (4-year) OS of the patients an increase of AUC from 0.705 (0.704) calculated only with clinical variables to 0.750 (0.743) calculated with clinical variables and the ProMCol classifier.
In addition, we performed a replication analysis, fitting a new Cox model for the clinical variables and our ProMCol classifier for OS (table 4A) and DSS (table 4B) on the validation cohort. In this cohort, the ProMCol classifier showed a significant association with OS (HR 0.61, 95% CI 0.45 to 0.82 and p=0.001) and DSS (HR 0.55, 95% CI 0.38 to 0.79, p=0.001) of patients with non-metastatic disease. For non-DSS, no significant association (p=0.302) was found in the validation cohort.
Discussion
In this study, we aimed to identify novel DNA methylation markers with a prognostic relevance for patients with CRC based on a genome-wide analysis of DNA methylation in tumour tissues.
Differentially methylated CpG sites between the tumour and normal mucosa tissue were identified and used as starting point for the survival analysis, as was done before by Wei et al.17 After investigation of the prognostic information of single CpG sites, a prognostic classifier for patients with non-metastatic CRC, the ProMCol classifier, was constructed. The prognostic validity of this ProMCol classifier was proven in an independent validation cohort. A significant reduction of the prediction error for the patients’ prognosis regarding OS was achieved by using our classifier in combination with standard clinical factors (such as gender, age, tumour stage, tumour location, smoking behaviour and MSI status) instead of using these factors alone. The fact that the prognostic power of the ProMCol classifier in the validation cohort is even stronger for the prediction error of the DSS than for OS indicates the specificity of this marker set. Accordingly, no association was found for non-DSS in the validation cohort.
In clinical practice, patients’ prognosis is estimated mainly based on the pathological staging according to the TNM system. However, often the estimate obtained from this traditional clinical classification is not sensitive and precise enough. Harmful undertreatment of patients having a high risk of recurrence and unnecessary overtreatment associated with toxic side effects, decreased quality of life and high treatment costs are possible consequences. By the assessment of our ProMCol classifier, risk stratification and prognosis prediction could be improved for patients before treatment with adjuvant therapy.
Higher methylation at the 20 CpG sites included in the ProMCol classifier was associated with better survival outcome. The biological effect of hypomethylation at single CpG sites on CRC survival is unknown, but it appears to be in concordance with findings in other cancers (when compared with normal tissue) that global hypomethylation increases with tumour progression.29 Although, the loss of DNA methylation at CpG sites had been the first epigenetic abnormality discovered in cancer cells, hypomethylation in cancer had been overlooked in preference of hypermethylation for many years of research.12 But, as the example of gene activation of the metastasis-associated S100A4 gene in CRC cell lines induced by hypomethylation30 suggests, loss of methylation can lead to an activation of genes that play an important role in cancer.12
Thirteen of the 20 CpG sites of our ProMCol classifier have been assigned to DHSs where several regulatory elements, including enhancers, silencers, promoters and locus control regions are known to be located in.31 32 In addition, 14 of the 20 CpG sites are indicated to be located in enhancer regions, DNA elements containing specific sequence motifs that interact with DNA-binding proteins in order to increase DNA transcription.33 Only 2 CpG sites belong to intergenic regions, whereas 18 CpG sites are located within the gene body or untranslated regions of specific genes. However, the effect of epigenetic changes in these genes on CRC prognosis has not been investigated so far.
One of the included CpG sites (cg11056055) is located in the body of the proto-oncogene c-MET. Bound to its ligand, the hepatocyte growth factor, c-MET induces dimerisation and activation of the receptor, which is involved in processes of embryogenesis, cellular survival and cellular migration and invasion. Overexpression and gene amplification of c-MET in the progression of CRC was shown in several reports.34–36 Meta-analyses confirmed the prognostic value of high c-MET expression as marker for poor prognosis in patients with CRC.37 38 Our results, showing lower methylation levels at one specific CpG site to be associated with poorer prognosis, suggest the possibility of an epigenetic regulation of c-MET expression in CRC. The same applies to the CpG site (cg01131395) within the gene LRCH1 that is located on chromosome 13q, where frequent aberrations associated with the progression from colorectal adenoma to carcinoma have been identified before.39 40 Increased LRCH1 messenger RNA (mRNA) expression has been observed in carcinomas in comparison with adenomas.41 Assuming a direct link between epigenetic regulation and gene expression, this finding is in line with our observation of lower methylation at this specific CpG site being associated with poorer prognosis. In contrast, no direct association between hypomethylation and higher gene expression could be concluded in the CpG site cg23750514 located in the KCNQ1 gene. Contrary to our finding, that hypomethylation is associated with poor prognosis in patients with non-metastatic CRC, the loss of KCNQ1 mRNA and protein expression in stage II and III colon cancer has been proposed as a prognostic factor for poor disease-free survival.42 Additionally, another report showed that low expression of KCNQ1 protein was significantly associated with poor OS in patients with CRC with liver metastases after hepatic resection.43 Some other CpG sites (cg18195165, cg19184885 and cg22522598) included in the ProMCol classifier belong to the genes FBXL18,44 NPEPPS45 and BRD446 that are all known to be important in cell cycle progression and cell proliferation. Moreover, for NPEPPS47 and BRD448 49 an altered expression in connection to CRC was reported. The CpG site cg16336556 is located in the gene LTBP1 that was found to be implicated in transforming growth factor beta signalling pathway contributing to CRC formation processes.50 51 Somatic mutations in the gene GMDS, in which the classifier CpG site cg08804626 is located, were proposed to be the reason for the loss of fucosylation in CRC cells, leading to their escape from tumour immune surveillance and to cancer progression.52 For the genes of the residual CpG sites incorporated into the ProMCol classifier, so far no associations with CRC have been reported.
Even though the cohorts investigated here were of a remarkable size (screening cohort 689 and validation cohort 408 patients) and the samples were derived from a large population-based case–control study with complete information on clinical and molecular characteristics, the statistical power was too limited for the analysis of the metastatic cases and other interesting subgroups. However, for the patients with non-metastatic disease, we were able to replicate our results in an independent cohort and achieved a significant reduction of the prediction error for the patients’ prognosis regarding OS and DSS using the ProMCol classifier.
In summary, in this study we developed a prognostic classifier, the ProMCol classifier, that is strongly associated with OS and DSS of patients with non-metastatic CRC and that was able to significantly reduce the prediction error for patients’ prognosis. After validation in further studies the assessment of the ProMCol classifier in combination with standard clinical factors and genetic markers might contribute to a better estimation of patients’ prognosis and facilitate decisions regarding the therapeutic strategy for patients with CRC.
Acknowledgments
Regarding the DACHS study, we thank Ute Handte-Daub, Ansgar Brandhorst and Petra Bächer for their excellent technical assistance. We would also like to thank the following hospitals and cooperating institutions that recruited patients for this study: Chirurgische Universitätsklinik Heidelberg, Klinik am Gesundbrunnen Heilbronn, St Vincentiuskrankenhaus Speyer, St Josefskrankenhaus Heidelberg, Chirurgische Universitätsklinik Mannheim, Diakonissenkrankenhaus Speyer, Krankenhaus Salem Heidelberg, Kreiskrankenhaus Schwetzingen, St Marienkrankenhaus Ludwigshafen, Klinikum Ludwigshafen, Stadtklinik Frankenthal, Diakoniekrankenhaus Mannheim, Kreiskrankenhaus Sinsheim, Klinikum am Plattenwald Bad Friedrichshall, Kreiskrankenhaus Weinheim, Kreiskrankenhaus Eberbach, Kreiskrankenhaus Buchen, Kreiskrankenhaus Mosbach, Enddarmzentrum Mannheim, Kreiskrankenhaus Brackenheim, Cancer Registry of Rhineland-Palatinate, Mainz. We are also very grateful for the support of the pathologies in the provision of tumour samples: Institut für Pathologie, Universitätsklinik Heidelberg; Institut für Pathologie, Klinikum Heilbronn; Institut für Angewandte Pathologie, Speyer; Pathologisches Institut, Universitätsklinikum Mannheim; Institut für Pathologie, Klinikum Ludwigshafen; Institut für Pathologie, Klinikum Stuttgart; Institut für Pathologie, Klinikum Ludwigsburg. In addition, we like to thank the Genomics and Proteomics Core Facility for provision of the methylation analysis services, especially Matthias Schick, Dr Melanie Bewerunge-Hudler and Professor Dr Stefan Wiemann.
References
Footnotes
Contributors Substantial contributions to the conception and design (MG, DE, BB), development of methodology (MG, DE, BB), acquisition of data (MG, MH, PK, EH, JC-C, HB, BB), analysis and interpretation of data (MG, DE, BB), draft of the article (MG, DE, BB), critical revision of the article (all authors), final approval of the version (all authors).
Funding This project was supported by grants from the German Research Council (BR 1704/6-1, BR 1704/6-3, BR 1704/6-4, CH 117/1-1, HO 5117/2-1, HE 5998/2-1, KL 2354/3-1, RO 2270/8-1, BR 1704/17-1), the German Federal Ministry of Education and Research (01KH0404, 01ER0814, 01ER0815, 01ER1505A, 01ER1505B) and the Interdisciplinary Research Program of the National Center for Tumor Diseases (NCT), Germany.
Disclaimer The sponsors had no role in the design and conduct of the study, collection, management, analysis and interpretation of the data and preparation, review or approval of the manuscript.
Competing interests None declared.
Patient consent Detail has been removed from this case description/these case descriptions to ensure anonymity. The editors and reviewers have seen the detailed information available and are satisfied that the information backs up the case the authors are making.
Ethics approval Ethical committees of the Medical Faculty of the University of Heidelberg and of the Medical Chambers of Baden-Württemberg and Rhineland-Palatinate.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement The data that support the findings of this study are available on reasonable request from the corresponding author (MG). The data are not publicly available due to restrictions of informed consent.