An invasion front gene expression signature for higher-risk patient selection in stage IIA MSS colon cancer.

Stage II colon cancer (CC) encompasses a heterogeneous group of patients with diverse survival experiences: 87% to 58% 5-year relative survival rates for stages IIA and IIC, respectively. While stage IIA patients are usually spared the adjuvant chemotherapy, some of them relapse and may benefit from it; thus, their timely identification is crucial. Current gene expression signatures did not specifically target this group nor did they find their place in clinical practice. Since processes at invasion front have also been linked to tumor progression, we hypothesize that aside from bulk tumor features, focusing on the invasion front may provide additional clues for this stratification. A retrospective matched case-control collection of 39 stage IIA microsatellite-stable (MSS) untreated CCs was analyzed to identify prognostic gene expression-based signatures. The endpoint was defined as relapse within 5 years vs. no relapse for at least 6 years. From the same tumors, three different classifiers (bulk tumor, invasion front, and constrained baseline on bulk tumor) were developed and their performance estimated. The baseline classifier, while the weakest, was validated in two independent data sets. The best performing signature was based on invasion front profiles [area under the receiver operating curve (AUC) = 0.931 (0.815-1.0)] and contained genes associated with KRAS pathway activation, apical junction complex, and heme metabolism. Its combination with bulk tumor classifier further improved the accuracy of the predictions.

Molecular portraits of colorectal cancer morphological regions.

Heterogeneity of colorectal carcinoma (CRC) represents a major hurdle towards personalized medicine. Efforts based on whole tumor profiling demonstrated that the CRC molecular subtypes were associated with specific tumor morphological patterns representing tumor subregions. We hypothesize that whole-tumor molecular descriptors depend on the morphological heterogeneity with significant impact on current molecular predictors. We investigated intra-tumor heterogeneity by morphology-guided transcriptomics to better understand the links between gene expression and tumor morphology represented by six morphological patterns (morphotypes): complex tubular, desmoplastic, mucinous, papillary, serrated, and solid/trabecular. Whole-transcriptome profiling by microarrays of 202 tumor regions (morphotypes, tumor-adjacent normal tissue, supportive stroma, and matched whole tumors) from 111 stage II-IV CRCs identified morphotype-specific gene expression profiles and molecular programs and differences in their cellular buildup. The proportion of cell types (fibroblasts, epithelial and immune cells) and differentiation of epithelial cells were the main drivers of the observed disparities with activation of EMT and TNF-α signaling in contrast to MYC and E2F targets signaling, defining major gradients of changes at molecular level. Several gene expression-based (including single-cell) classifiers, prognostic and predictive signatures were examined to study their behavior across morphotypes. Most exhibited important morphotype-dependent variability within same tumor sections, with regional predictions often contradicting the whole-tumor classification. The results show that morphotype-based tumor sampling allows the detection of molecular features that would otherwise be distilled in whole tumor profile, while maintaining histopathology context for their interpretation. This represents a practical approach at improving the reproducibility of expression profiling and, by consequence, of gene-based classifiers.

IntOMICS: A Bayesian Framework for Reconstructing Regulatory Networks Using Multi-Omics Data.

Integration of multi-omics data can provide a more complex view of the biological system consisting of different interconnected molecular components. We present a new comprehensive R/Bioconductor-package, IntOMICS, which implements a Bayesian framework for multi-omics data integration. IntOMICS adopts a Markov Chain Monte Carlo sampling scheme to systematically analyze gene expression, copy number variation, DNA methylation, and biological prior knowledge to infer regulatory networks. The unique feature of IntOMICS is an empirical biological knowledge estimation from the available experimental data, which complements the missing biological prior knowledge. IntOMICS has the potential to be a powerful resource for exploratory systems biology.

Using empirical biological knowledge to infer regulatory networks from multi-omics data.

BACKGROUND: Integration of multi-omics data can provide a more complex view of the biological system consisting of different interconnected molecular components, the crucial aspect for developing novel personalised therapeutic strategies for complex diseases. Various tools have been developed to integrate multi-omics data. However, an efficient multi-omics framework for regulatory network inference at the genome level that incorporates prior knowledge is still to emerge. RESULTS: We present IntOMICS, an efficient integrative framework based on Bayesian networks. IntOMICS systematically analyses gene expression, DNA methylation, copy number variation and biological prior knowledge to infer regulatory networks. IntOMICS complements the missing biological prior knowledge by so-called empirical biological knowledge, estimated from the available experimental data. Regulatory networks derived from IntOMICS provide deeper insights into the complex flow of genetic information on top of the increasing accuracy trend compared to a published algorithm designed exclusively for gene expression data. The ability to capture relevant crosstalks between multi-omics modalities is verified using known associations in microsatellite stable/instable colon cancer samples. Additionally, IntOMICS performance is compared with two algorithms for multi-omics regulatory network inference that can also incorporate prior knowledge in the inference framework. IntOMICS is also applied to detect potential predictive biomarkers in microsatellite stable stage III colon cancer samples. CONCLUSIONS: We provide IntOMICS, a framework for multi-omics data integration using a novel approach to biological knowledge discovery. IntOMICS is a powerful resource for exploratory systems biology and can provide valuable insights into the complex mechanisms of biological processes that have a vital role in personalised medicine.

Colorectal Tumour Mucosa Microbiome Is Enriched in Oral Pathogens and Defines Three Subtypes That Correlate with Markers of Tumour Progression.

Long-term dysbiosis of the gut microbiome has a significant impact on colorectal cancer (CRC) progression and explains part of the observed heterogeneity of the disease. Even though the shifts in gut microbiome in the normal-adenoma-carcinoma sequence were described, the landscape of the microbiome within CRC and its associations with clinical variables remain under-explored. We performed 16S rRNA gene sequencing of paired tumour tissue, adjacent visually normal mucosa and stool swabs of 178 patients with stage 0-IV CRC to describe the tumour microbiome and its association with clinical variables. We identified new genera associated either with CRC tumour mucosa or CRC in general. The tumour mucosa was dominated by genera belonging to oral pathogens. Based on the tumour microbiome, we stratified CRC patients into three subtypes, significantly associated with prognostic factors such as tumour grade, sidedness and TNM staging, BRAF mutation and MSI status. We found that the CRC microbiome is strongly correlated with the grade, location and stage, but these associations are dependent on the microbial environment. Our study opens new research avenues in the microbiome CRC biomarker detection of disease progression while identifying its limitations, suggesting the need for combining several sampling sites (e.g., stool and tumour swabs).

An Explorative Analysis of ABCG2/TOP-1 mRNA Expression as a Biomarker Test for FOLFIRI Treatment in Stage III Colon Cancer Patients: Results from Retrospective Analyses of the PETACC-3 Trial.

Biomarker-guided treatment for patients with colon cancer is needed. We tested ABCG2 and topoisomerase 1 (TOP1) mRNA expression as predictive biomarkers for irinotecan benefit in the PETACC-3 patient cohort. The present study included 580 patients with mRNA expression data from Stage III colon cancer samples from the PETACC-3 study, which randomized the patients to Fluorouracil/leucovorin (5FUL) +/- irinotecan. The primary end-points were recurrence free survival (RFS) and overall survival (OS). Patients were divided into one group with high ABCG2 expression (above median) and low TOP-1 expression (below 75 percentile) ("resistant") (n = 216) and another group including all other combinations of these two genes ("sensitive") (n = 364). The rationale for the cut-offs were based on the distribution of expression levels in the PETACC-3 Stage II set of patients, where ABCG2 was unimodal and TOP1 was bimodal with a high expression level mode in the top quarter of the patients. Cox proportional hazards regression was used to estimate the hazard ratios and the association between variables and end-points and log-rank tests to assess the statistical significance of differences in survival between groups. Kaplan-Meier estimates of the survival functions were used for visualization and estimation of survival rates at specific time points. Significant differences were found for both RFS (Hazard ratio (HR): 0.63 (0.44-0.92); p = 0.016) and OS (HR: 0.60 (0.39-0.93); p = 0.02) between the two biomarker groups when the patients received FOLFIRI (5FUL+irinotecan). Considering only the Microsatellite Stable (MSS) and Microsatellite Instability-Low (MSI-L) patients (n = 470), the differences were even more pronounced. In contrast, no significant differences were observed between the groups when patients received 5FUL alone. This study shows that the combination of ABCG2 and TOP1 gene expression significantly divided the Stage III colon cancer patients into two groups regarding benefit from adjuvant treatment with FOLFIRI but not 5FUL.

Stool sampling and DNA isolation kits affect DNA quality and bacterial composition following 16S rRNA gene sequencing using MiSeq Illumina platform.

Many studies correlate changes in human gut microbiome with the onset of various diseases, mostly by 16S rRNA gene sequencing. Setting up the optimal sampling and DNA isolation procedures is crucial for robustness and reproducibility of the results. We performed a systematic comparison of several sampling and DNA isolation kits, quantified their effect on bacterial gDNA quality and the bacterial composition estimates at all taxonomic levels. Sixteen volunteers tested three sampling kits. All samples were consequently processed by two DNA isolation kits. We found that the choice of both stool sampling and DNA isolation kits have an effect on bacterial composition with respect to Gram-positivity, however the isolation kit had a stronger effect than the sampling kit. The proportion of bacteria affected by isolation and sampling kits was larger at higher taxa levels compared to lower taxa levels. The PowerLyzer PowerSoil DNA Isolation Kit outperformed the QIAamp DNA Stool Mini Kit mainly due to better lysis of Gram-positive bacteria while keeping the values of all the other assessed parameters within a reasonable range. The presented effects need to be taken into account when comparing results across multiple studies or computing ratios between Gram-positive and Gram-negative bacteria.

Characterization of molecular scores and gene expression signatures in primary breast cancer, local recurrences and brain metastases.

BACKGROUND: Breast cancer is a leading cause of cancer-related death in women worldwide. Despite extensive studies in all areas of basic, clinical and applied research, accurate prognosis remains elusive, thus leading to overtreatment of many patients. Diagnosis could be improved by introducing multigene molecular scores in standard clinical practice. Several tests that work with formalin-fixed tissue have become routine. Molecular scores usually include several genes representing processes, response to oestrogens, progestogens and human epidermal growth factor receptor 2 (Her2), respectively, which are combined additively in single values. These multi-gene scores have the advantage of being more robust and reproducible than single-gene scores. Their utility may be further enhanced by combining them with classical diagnostic parameters. Here, we present an exploratory study comparing the RISK and research versions of Oncotype DX recurrence score (RS), Prosigna Risk of Recurrence (ROR) and EndoPredict (EP) with respect to their prognostic potential for ipsilateral recurrence and/or distant relapse in brain, and we compared the scores to the intrinsic subtypes based on PAM50. METHODS: RNA was extracted from formalin-fixed, paraffin-embedded (FFPE) tissue cores of primary tumours, local recurrences and brain metastases. Gene expression was measured on a NanoString nCounter Analysis System. Intrinsic subtypes and molecular scores were computed according to published literature and RISK, RS, ROR and EP were compared against each other and to the intrinsic subtypes Luminal A (lumA), Luminal B (lumB), Her2-enriched (Her2↑), Basal-like (basal), and Normal-like (normal) of PAM50. Local recurrences and brain metastases were compared to their corresponding primary tumours. RESULTS: All four molecular scores were highly correlated. Highest correlations were observed among genes related to proliferation while lower correlations were found among oestrogen-related genes. The scores were significantly higher in primary tumours progressing to brain metastases as compared to recurrence-free primary tumours and primary tumours that relapsed as local recurrences. CONCLUSIONS: RISK and ROR-P are prognostic for primary tumours metastasizing to the brain. All four scores, RISK, RS, EP and ROR-P failed to discriminate between primary tumours that remained recurrence-free and primary tumours relapsing as local recurrences.

Cross-platform Data Analysis Reveals a Generic Gene Expression Signature for Microsatellite Instability in Colorectal Cancer.

The dysfunction of the DNA mismatch repair system results in microsatellite instability (MSI). MSI plays a central role in the development of multiple human cancers. In colon cancer, despite being associated with resistance to 5-fluorouracil treatment, MSI is a favourable prognostic marker. In gastric and endometrial cancers, its prognostic value is not so well established. Nevertheless, recognising the MSI tumours may be important for predicting the therapeutic effect of immune checkpoint inhibitors. Several gene expression signatures were trained on microarray data sets to understand the regulatory mechanisms underlying microsatellite instability in colorectal cancer. A wealth of expression data already exists in the form of microarray data sets. However, the RNA-seq has become a routine for transcriptome analysis. A new MSI gene expression signature presented here is the first to be valid across two different platforms, microarrays and RNA-seq. In the case of colon cancer, its estimated performance was (i) AUC = 0.94, 95% CI = (0.90 - 0.97) on RNA-seq and (ii) AUC = 0.95, 95% CI = (0.92 - 0.97) on microarray. The 25-gene expression signature was also validated in two independent microarray colon cancer data sets. Despite being derived from colorectal cancer, the signature maintained good performance on RNA-seq and microarray gastric cancer data sets (AUC = 0.90, 95% CI = (0.85 - 0.94) and AUC = 0.83, 95% CI = (0.69 - 0.97), respectively). Furthermore, this classifier retained high concordance even when classifying RNA-seq endometrial cancers (AUC = 0.71, 95% CI = (0.62 - 0.81). These results indicate that the new signature was able to remove the platform-specific differences while preserving the underlying biological differences between MSI/MSS phenotypes in colon cancer samples.

The Unfolded Protein Response: A Novel Therapeutic Target for Poor Prognostic BRAF Mutant Colorectal Cancer.

BRAFV600E mutations occur in ∼10% of colorectal cancer cases, are associated with poor survival, and have limited responses to BRAF/MEK inhibition with or without EGFR inhibition. There is an unmet need to understand the biology of poor prognostic BRAFMT colorectal cancer. We have used differential gene expression and pathway analyses of untreated stage II and stage III BRAFMT (discovery set: n = 31; validation set: n = 26) colorectal cancer, and an siRNA screen to characterize the biology underpinning the BRAFMT subgroup with poorest outcome. These analyses identified the unfolded protein response (UPR) as a novel and druggable pathway associated with the BRAFMT colorectal cancer subgroup with poorest outcome. We also found that oncogenic BRAF drives endoplasmic reticulum (ER) stress and UPR pathway activation through MEK/ERK. Furthermore, inhibition of GRP78, the master regulator of the UPR, using siRNA or small molecule inhibition, resulted in acute ER stress and apoptosis, in particular in BRAFMT colorectal cancer cells. In addition, dual targeting of protein degradation using combined Carfilzomib (proteasome inhibitor) and ACY-1215 (HDAC6-selective inhibitor) treatment resulted in marked accumulation of protein aggregates, acute ER stress, apoptosis, and therapeutic efficacy in BRAFMT in vitro and xenograft models. Mechanistically, we found that the apoptosis following combined Carfilzomib/ACY-1215 treatment is mediated through increased CHOP expression. Taken together, our findings indicate that oncogenic BRAF induces chronic ER stress and that inducers of acute ER stress could be a novel treatment strategy for poor prognostic BRAFMT colorectal cancer. Mol Cancer Ther; 17(6); 1280-90. ©2018 AACR.

A critical comparison of topology-based pathway analysis methods.

One of the aims of high-throughput gene/protein profiling experiments is the identification of biological processes altered between two or more conditions. Pathway analysis is an umbrella term for a multitude of computational approaches used for this purpose. While in the beginning pathway analysis relied on enrichment-based approaches, a newer generation of methods is now available, exploiting pathway topologies in addition to gene/protein expression levels. However, little effort has been invested in their critical assessment with respect to their performance in different experimental setups. Here, we assessed the performance of seven representative methods identifying differentially expressed pathways between two groups of interest based on gene expression data with prior knowledge of pathway topologies: SPIA, PRS, CePa, TAPPA, TopologyGSA, Clipper and DEGraph. We performed a number of controlled experiments that investigated their sensitivity to sample and pathway size, threshold-based filtering of differentially expressed genes, ability to detect target pathways, ability to exploit the topological information and the sensitivity to different pre-processing strategies. We also verified type I error rates and described the influence of overexpression of single genes, gene sets and topological motifs of various sizes on the detection of a pathway as differentially expressed. The results of our experiments demonstrate a wide variability of the tested methods. We provide a set of recommendations for an informed selection of the proper method for a given data analysis task.

Identification of "BRAF-Positive" Cases Based on Whole-Slide Image Analysis.

A key requirement for precision medicine is the accurate identification of patients that would respond to a specific treatment or those that represent a high-risk group, and a plethora of molecular biomarkers have been proposed for this purpose during the last decade. Their application in clinical settings, however, is not always straightforward due to relatively high costs of some tests, limited availability of the biological material and time, and procedural constraints. Hence, there is an increasing interest in constructing tissue-based surrogate biomarkers that could be applied with minimal overhead directly to histopathology images and which could be used for guiding the selection of eventual further molecular tests. In the context of colorectal cancer, we present a method for constructing a surrogate biomarker that is able to predict with high accuracy whether a sample belongs to the "BRAF-positive" group, a high-risk group comprising V600E BRAF mutants and BRAF-mutant-like tumors. Our model is trained to mimic the predictions of a 64-gene signature, the current definition of BRAF-positive group, thus effectively identifying histopathology image features that can be linked to a molecular score. Since the only required input is the routine histopathology image, the model can easily be integrated in the diagnostic workflow.

Image-based surrogate biomarkers for molecular subtypes of colorectal cancer.

MOTIVATION: Whole genome expression profiling of large cohorts of different types of cancer led to the identification of distinct molecular subcategories (subtypes) that may partially explain the observed inter-tumoral heterogeneity. This is also the case of colorectal cancer (CRC) where several such categorizations have been proposed. Despite recent developments, the problem of subtype definition and recognition remains open, one of the causes being the intrinsic heterogeneity of each tumor, which is difficult to estimate from gene expression profiles. However, one of the observations of these studies indicates that there may be links between the dominant tumor morphology characteristics and the molecular subtypes. Benefiting from a large collection of CRC samples, comprising both gene expression and histopathology images, we investigated the possibility of building image-based classifiers able to predict the molecular subtypes. We employed deep convolutional neural networks for extracting local descriptors which were then used for constructing a dictionary-based representation of each tumor sample. A set of support vector machine classifiers were trained to solve different binary decision problems, their combined outputs being used to predict one of the five molecular subtypes. RESULTS: A hierarchical decomposition of the multi-class problem was obtained with an overall accuracy of 0.84 (95%CI=0.79-0.88). The predictions from the image-based classifier showed significant prognostic value similar to their molecular counterparts. CONTACT: popovici@iba.muni.cz. AVAILABILITY AND IMPLEMENTATION: Source code used for the image analysis is freely available from https://github.com/higex/qpath . SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Towards the identification of tissue-based proxy biomarkers.

Accurate patient population stratification is a key requirement for a personalized medicine and more precise biomarkers are expected to be obtained by better exploiting the available data. We introduce a novel computational framework that exploits both the information from gene expression data and histopathology images for constructing a tissue-based biomarker, which can be used for identifying a high-risk patient population. Its utility is demonstrated in the context of colorectal cancer data and we show that the resulting biomarker can be used as a proxy for a prognostic gene expression signature. These results are important for both the computational discovery of new biomarkers and clinical practice, as they demonstrate a possible approach for multimodal biomedical data mining and since the new tissue-based biomarker could easily be implemented in the routine pathology practice.

Joint analysis of histopathology image features and gene expression in breast cancer.

BACKGROUND: Genomics and proteomics are nowadays the dominant techniques for novel biomarker discovery. However, histopathology images contain a wealth of information related to the tumor histology, morphology and tumor-host interactions that is not accessible through these techniques. Thus, integrating the histopathology images in the biomarker discovery workflow could potentially lead to the identification of new image-based biomarkers and the refinement or even replacement of the existing genomic and proteomic signatures. However, extracting meaningful and robust image features to be mined jointly with genomic (and clinical, etc.) data represents a real challenge due to the complexity of the images. RESULTS: We developed a framework for integrating the histopathology images in the biomarker discovery workflow based on the bag-of-features approach - a method that has the advantage of being assumption-free and data-driven. The images were reduced to a set of salient patterns and additional measurements of their spatial distribution, with the resulting features being directly used in a standard biomarker discovery application. We demonstrated this framework in a search for prognostic biomarkers in breast cancer which resulted in the identification of several prognostic image features and a promising multimodal (imaging and genomic) prognostic signature. The source code for the image analysis procedures is freely available. CONCLUSIONS: The framework proposed allows for a joint analysis of images and gene expression data. Its application to a set of breast cancer cases resulted in image-based and combined (image and genomic) prognostic scores for relapse-free survival.

Parish registers from Transylvania - sources of the history of medicine (late 18th - early 20th centuries).

This paper aims to analyse the typology of medical-historical information provided by parish registers from Transylvania - a category of primary sources used mainly by historical demographers. The approach is descriptive and prospective in character: it creates a typology of the medical information to be found in the sources, while highlighting possible research directions and approaching a series of methodological and interpretation issues. The parish registers contain references to medical actors (the midwife, the physician, the death inspector), to medical activities (vaccination), and to events regarding the history of medicine (multiple births, infant mortality, death-causing diseases and accidents, epidemics, etc.). Despite the fact that they provide mainly demographic data, some epidemiological and medicine-related information can prove interesting for researchers in the field of the history of medicine. Such information is suitable for serial analyses and in some cases even for collective biography studies of the medical staff (e.g., the birth assistants and midwives), thus cross-referencing in many respects with cultural and social history. However, probably for reasons related to the sources' accessibility, medical historians have not seemed very interested in these data, a situation which will hopefully change in the near future due to the newly compiled historical population databases. The conclusions reached in this paper point towards the variety of medical-historical information contained in parish registers, highlighting the need for reconsidering them as sources not only for historical demography, but also for medical history.

Test of four colon cancer risk-scores in formalin fixed paraffin embedded microarray gene expression data.

BACKGROUND: Prognosis prediction for resected primary colon cancer is based on the T-stage Node Metastasis (TNM) staging system. We investigated if four well-documented gene expression risk scores can improve patient stratification. METHODS: Microarray-based versions of risk-scores were applied to a large independent cohort of 688 stage II/III tumors from the PETACC-3 trial. Prognostic value for relapse-free survival (RFS), survival after relapse (SAR), and overall survival (OS) was assessed by regression analysis. To assess improvement over a reference, prognostic model was assessed with the area under curve (AUC) of receiver operating characteristic (ROC) curves. All statistical tests were two-sided, except the AUC increase. RESULTS: All four risk scores (RSs) showed a statistically significant association (single-test, P < .0167) with OS or RFS in univariate models, but with HRs below 1.38 per interquartile range. Three scores were predictors of shorter RFS, one of shorter SAR. Each RS could only marginally improve an RFS or OS model with the known factors T-stage, N-stage, and microsatellite instability (MSI) status (AUC gains < 0.025 units). The pairwise interscore discordance was never high (maximal Spearman correlation = 0.563) A combined score showed a trend to higher prognostic value and higher AUC increase for OS (HR = 1.74, 95% confidence interval [CI] = 1.44 to 2.10, P < .001, AUC from 0.6918 to 0.7321) and RFS (HR = 1.56, 95% CI = 1.33 to 1.84, P < .001, AUC from 0.6723 to 0.6945) than any single score. CONCLUSIONS: The four tested gene expression-based risk scores provide prognostic information but contribute only marginally to improving models based on established risk factors. A combination of the risk scores might provide more robust information. Predictors of RFS and SAR might need to be different.

Context-dependent interpretation of the prognostic value of BRAF and KRAS mutations in colorectal cancer.

BACKGROUND: The mutation status of the BRAF and KRAS genes has been proposed as prognostic biomarker in colorectal cancer. Of them, only the BRAF V600E mutation has been validated independently as prognostic for overall survival and survival after relapse, while the prognostic value of KRAS mutation is still unclear. We investigated the prognostic value of BRAF and KRAS mutations in various contexts defined by stratifications of the patient population. METHODS: We retrospectively analyzed a cohort of patients with stage II and III colorectal cancer from the PETACC-3 clinical trial (N = 1,423), by assessing the prognostic value of the BRAF and KRAS mutations in subpopulations defined by all possible combinations of the following clinico-pathological variables: T stage, N stage, tumor site, tumor grade and microsatellite instability status. In each such subpopulation, the prognostic value was assessed by log rank test for three endpoints: overall survival, relapse-free survival, and survival after relapse. The significance level was set to 0.01 for Bonferroni-adjusted p-values, and a second threshold for a trend towards statistical significance was set at 0.05 for unadjusted p-values. The significance of the interactions was tested by Wald test, with significance level of 0.05. RESULTS: In stage II-III colorectal cancer, BRAF mutation was confirmed a marker of poor survival only in subpopulations involving microsatellite stable and left-sided tumors, with higher effects than in the whole population. There was no evidence for prognostic value in microsatellite instable or right-sided tumor groups. We found that BRAF was also prognostic for relapse-free survival in some subpopulations. We found no evidence that KRAS mutations had prognostic value, although a trend was observed in some stratifications. We also show evidence of heterogeneity in survival of patients with BRAF V600E mutation. CONCLUSIONS: The BRAF mutation represents an additional risk factor only in some subpopulations of colorectal cancers, in others having limited prognostic value. However, in the subpopulations where it is prognostic, it represents a marker of much higher risk than previously considered. KRAS mutation status does not seem to represent a strong prognostic variable.

Gene expression patterns unveil a new level of molecular heterogeneity in colorectal cancer.

The recognition that colorectal cancer (CRC) is a heterogeneous disease in terms of clinical behaviour and response to therapy translates into an urgent need for robust molecular disease subclassifiers that can explain this heterogeneity beyond current parameters (MSI, KRAS, BRAF). Attempts to fill this gap are emerging. The Cancer Genome Atlas (TGCA) reported two main CRC groups, based on the incidence and spectrum of mutated genes, and another paper reported an EMT expression signature defined subgroup. We performed a prior free analysis of CRC heterogeneity on 1113 CRC gene expression profiles and confronted our findings to established molecular determinants and clinical, histopathological and survival data. Unsupervised clustering based on gene modules allowed us to distinguish at least five different gene expression CRC subtypes, which we call surface crypt-like, lower crypt-like, CIMP-H-like, mesenchymal and mixed. A gene set enrichment analysis combined with literature search of gene module members identified distinct biological motifs in different subtypes. The subtypes, which were not derived based on outcome, nonetheless showed differences in prognosis. Known gene copy number variations and mutations in key cancer-associated genes differed between subtypes, but the subtypes provided molecular information beyond that contained in these variables. Morphological features significantly differed between subtypes. The objective existence of the subtypes and their clinical and molecular characteristics were validated in an independent set of 720 CRC expression profiles. Our subtypes provide a novel perspective on the heterogeneity of CRC. The proposed subtypes should be further explored retrospectively on existing clinical trial datasets and, when sufficiently robust, be prospectively assessed for clinical relevance in terms of prognosis and treatment response predictive capacity. Original microarray data were uploaded to the ArrayExpress database (http://www.ebi.ac.uk/arrayexpress/) under Accession Nos E-MTAB-990 and E-MTAB-1026.