CORDIS - Résultats de la recherche de l’UE
CORDIS

Modelling and predicting sensitivity to targeted therapies in colorectal cancers

Final Report Summary - COLTHERES (Modelling and predicting sensitivity to targeted therapies in colorectal cancers)

Executive Summary:
Tumours are presently categorized and treated according to where they arose in the body. The revelation that cancer is a genetic disease and that accumulation of molecular alterations in the genome of somatic cells is the primary driver of tumour progression have revolutionized oncology. It is now manifest that cancers, which were thought to be indistinguishable based on light microscopy, are actually different diseases requiring distinctive approaches to therapy. These findings lead to the concepts of precision oncology or precision cancer medicine. COLTHERES was conceived to apply these revolutionary concepts to colorectal tumours. A central goal of the consortium has been to define molecular signatures of sensitivity and resistance to agents targeting oncogenic nodes in the EGFR signaling pathway in Colorectal Cancer (CRC).

COLTHERES has achieved and in some instances exceeded its planned goals. Members of the consortium have shown that activating KRAS mutations are the cause of resistance to EGFR-inhibitors in 30-40% of patients but other molecular alterations can cause resistance as well and cause resistance in addition 30-40% of patients. These alterations can be mutations in downstream key molecules, over expression of activating or competing molecules or loss of inhibitors etc. The COLTHERES team has worked on identification of these alterations and on the development of comprehensive gene expression signatures that can be used to identify more accurately responders to cetuximab treatment than by assessing KRAS mutations alone (Tian et al Gut 2012; Popovic et al., JCO 2013).

Since patients with KRAS or BRAF mutations will not respond to EGFR inhibitors, alternative therapies for these patients are urgently required. Inhibition of the BRAF (V600E) oncoprotein by the small-molecule drug PLX4032 (vemurafenib) is highly effective in the treatment of melanoma, however, colon cancer patients harboring the same BRAF (V600E) oncogenic lesion have poor prognosis and show a limited response. Members of the COLTHERES consortium found that BRAF(V600E) inhibition causes a rapid feedback activation of EGFR, which supports continued proliferation in the presence of BRAF(V600E) inhibition (Prahallad et al., Nature 2011). Our data suggest that BRAF (V600E) mutant colon cancers for which there are currently no targeted treatment options available might benefit from combination therapy consisting of BRAF and EGFR inhibitors.

Research by the COLTHERES consortium has also shown that mutations in genes like KRAS, NRAS and BRAF are causally associated with acquired resistance to targeted therapies for colorectal cancer (Misale et al., Nature 2012, Misale et al., Science Translational Medicine). Our results demonstrate that resistance mutations in KRAS and other genes are highly likely to be present in a subpopulation of tumour cells before treatment. Resistance to therapy remains a fundamental obstacle to successful therapies and COLTHERES partners discovered recently reported how epigenetic modification modulates based-resistance to chemotherapy in CRCs (Moutinho et al., JNCI 2013). Notably, research by COLTHERES has found that resistance-associated mutations can be detected in the blood of patients (liquid biopsies) several months before radiographic evidence of disease progression is observable. This finding may offer an opportunity to anticipate and counter resistance by using combination therapies before patients relapse.

Perhaps the best testimony to the success of COLTHERES is the rapid translation of scientific findings generated by the consortium into clinical trials, which are already recruiting patients as described at: http://www.coltheres.org.

Project Context and Objectives:
Precision cancer medicine based on the genetic milieu of individual colorectal tumours has long been postulated, but until recently this concept was not supported by clinical evidence. The observation that a subset of colorectal cancer (CRC) responds to anti-Epidermal Growth Factor Receptor (EGFR) antibody therapies, based on knowledge of which other mutant alleles are also present, has heralded the widespread realization that colorectal cancer medicine is now inevitably going to become more ‘personalised’. This has therefore stimulated research and development of clinically validated diagnostic tools and biomarkers for the prospective identification of responder patients. Remarkably, these studies have also increased knowledge into the molecular basis of colorectal cancer.

At the root of understanding targeted drug sensitivity or resistance, is the concept that each tumour has an independent genetic identity. Intrinsic to this concept, therefore, is that molecular therapies targeting one specific oncoprotein are likely to be useful only in a fraction of patients, i.e. those who display the genetic lesion that is both drugable and is centrally important to the tumours’ continued survival or growth. In addition to pre-existing, or de novo resistance mechanisms, it is now becoming apparent that secondary, or acquired mutations, during drug treatment, along with gene expression changes and/or epigenetic variations, also lead to resistance to targeted therapies in a remarkably short time. This makes it imperative to understand all the possible compensatory routes to acquired resistance, so that clinicians can be ready with rational combinations of targeted treatments to circumvent, reverse or even preclude resistance emergence. From the beginning COLTHERES had clear and focused objectives:
• To define routes of sensitivity and primary resistance to targeted agents mediated by oncogenic nodes in the EGFR signalling pathway in CRC;
• Identify biomarkers of secondary resistance that are likely to emerge for novel targeted therapies (such as RTK, BRAF, PI3K and MEK inhibitors)
• Identify genes mediating sensitivity and resistance to agents effectors of the EGFR signaling pathway.
• Define validated risk stratification criteria to be used in personalised patient-screening methodologies to predict individual therapy response and resistance profiles for colorectal cancer patients.

To achieve these objectives COLTHERES was organized in specific tasks. Task 1 was set to define a comprehensive algorithm of positive and negative predictors by integrated mutational, gene copy number, epigenetic, transcriptional analyses and pathway-specific (phospho)protein measurements of responsive and non-responsive tumours. This will be achieved applying state of the art ‘–omics’ technologies (described in the next sections) to fully characterize colorectal tumour samples and therapeutic response data that are already available through the clinical units of the consortium. Specifically COLTHERES consortium members have large retrospective series available. To overcome these limitations COLTHERES applied a new approach described in the figure below in which molecular lesions (biomarkers) and compounds targeting the corresponding oncogenic proteins are functionally assessed using novel in vitro and in vivo models


Task 2 was focused on developing cellular models that closely recapitulate the genetic milieu of individual colorectal tumours. The approach allows for the routine and precise mutation or correction of any endogenous gene within a human cell-line grown in culture, enabling the creation of gold-standard ‘isogenic’ in vitro disease models that: a) accurately recapitulate the genetics present in real cancer patients; and b) provide a matched normal genetic background for referencing the effects of targeted drugs and/or rationally determining mutational-based resistance mechanisms Using this approach the mutations found in the genes representing key nodes in the EGFR signalling pathway will have been inserted (knock-in). Among these are the hotspot mutations found in KRAS, PIK3CA, BRAF, PTEN beta catenin etc. This strategy ultimately lead to isogenic mutant vs wild type cellular models in the same context that they occur in real patients. A variety of models will be created covering the variety of known and newly discovered mutant oncogenes; in both isolation and in specific resistance imparting combinations. For example, there are 7 major variants of mutant KRAS, which at this time are assumed (probably erroneously) to be equal in imparting resistance to Cetuximab. COLTHERES will insert all these mutations and many others emanating from the research into novel isogenic cell-lines for further study.

These cellular tools have been explored by COLTHERES partners to identify the oncogenic events and the signalling pathways that contribute to primary resistance to targeted therapies in CRCs for subsequent assessment in clinical trials. Key proof-of-concept data that isogenic models could have predicted the now known clinical data that specific mutations can cause resistance to EGFR-therapy is provided in figure 7; and Di Nicoloantonio 2009). HD is now providing most of the world’s major pharma and biotech companies with these tools to rationalise targeted drug discovery; many of which will be tested in trials conducted by this consortium.

Prospective models of secondary or ‘acquired’ drug resistance are also sorely needed, as it is clear that secondary resistance is responsible for rapid treatment failure and its molecular bases are presently largely unknown. COLTHERES has explored the mechanisms of secondary resistance on two levels, each of them novel 1) Running two dedicated trials in patients; sampling the patients at repeated time points to study changes in treatment efficacy and acquired resistance (genotyping these samples is part of task 1); and 2) Using cell-based models of acquired resistance in vitro (task 3 cellular models) and in vivo (task 4 ’xenopatients’). Functional genetic screens have been performed to validate putative primary markers of resistance from task 2; and prospectively predict candidate biomarkers of secondary resistance to targeted therapies within task 3. To achieve this, the consortium includes the laboratory that initially developed the siRNA bar code screening technology to rapidly identify genes whose suppression cause resistance to cancer drugs. This approach was successfully used by COLTHERES members to identify a major pathway of resistance to HER2-targeted therapy in breast cancer. This strategy and complimentary gene-overexpression strategies will be used in CRC cell lines that are highly sensitive to EGFR and other targeted agents to find genes causing resistance to molecularly targeted therapies.

To further identify and validate candidate secondary resistance mechanisms, Task 4 exploited sub-cutaneous implantation of fresh tumour samples (liver metastases from CRCs) in immunocompromised mice (‘xenopatients’) followed by the establishment of continued ‘cancer lines’ in experimental animals. From individual patient-derived material a large cohort of tumour-bearing animals have already been generated by members of the consortium; these will represent ‘xeno-patients’ undergoing control treatment or systemic treatment with Cetuximab, or other targeted agents. The ‘xeno-patients’ approach was used to understand the mechanisms of secondary resistance (by chronic treatment of ‘responsive xeno-patients’ with targeted agents) and for the preclinical validation of combinatorial treatments aimed at targeting multiple signalling networks.

Next targeted functional-genomic and drug combination approaches were employed based on pathway/systems analysis of COLTHERES data to find rational strategies to reverse drug resistance. This approach is based on the notion that acquisition of resistance by mutation, up-regulation or down-regulation of parallel pathway targets, could be reversed directly (or indirectly downstream in specific pathway) by rational combinations of targeted agents already in existence. To exploit this feature, Task 5 will search for genes that when inhibited with targeted siRNAs or inhibitor compounds either reverse resistance to EGFR signalling inhibition in cellular models. These will then offer new targets for rational therapeutic strategies to be used for combinatorial pharmacological ‘attacks’ on core oncogenic signalling pathways in CRC. The results of the ‘omics analysis on clinical samples from CRC patients as well as the results of the functional analysis (in vitro and in vivo) generated were integrated at multiple levels through biostatistical analysis in task 5 which includes experts in statistical data analysis. These analyses will generate risk stratification algorithms for personalised screening methodologies and prediction of individual therapy response and resistance for CRC patients receiving drugs targeting the oncogenic nodes of the EGFR signalling pathway. The results were disseminated through peer reviewed publication by COLTHERES members. More importantly, the results have been directly translated into practical approaches for the benefit of patients though the design of novel biomarker hypothesis driven trials. These are presently ongoing and are based on the knowledge gathered on the mechanisms of sensitivity and resistance to agents targeting the EGFR oncogenic signaling on defined patient populations.
Project Results:
The COLTHERES tumour database
At the roots of COLTHERES was the idea of building a tumour sample database of FFPE CRCs samples and their related clinical data, in order to use these samples & data retrospectively, if needed, to define and validate signatures of sensitivity and resistance to agents targeting oncogenic nodes in the EGFR signalling pathway in CRC To successfully achieve this milestone and the assurance of the best clinical advisory and assessment in the case that clinical trials are developed, the COLTHERES consortium established three main objectives for the WP: i) to manage the ethical approval/regulatory issues; ii) to provide logistic and dedicated sample tracking support and iii) to build and to maintain databases including sample information, images and related clinical data. To make sure that the objectives were realistic and their degree of accomplishment could be measured, four tasks and three deliverables were defined. It is important to remark that all tasks have been completed and no major deviations need to be reported.

The first objective was to develop the inventory of available samples and data, with prior engagement of the clinical partners, who confirmed their ability to provide a unique set of CRC samples. Available samples were re-assessed by the institutions’ pathologists (a first assessment was carried out while drafting the proposal). The final database included tumour specifications (primary, non-relapsed from a single anatomical site), histological classification and clinical data (demographic, diagnostic, follow-up therapeutic and outcome data). To date, the database is being maintained a COLTHERES partner, the Data Coordination Centre at the Swiss Institute of Bioinformatics (SIB). The final number of samples and clinical data collected was lower than previously expected after careful revision of all the samples, as some were discarded for quality or quantity reasons. However, the database is subject to periodic updates from the clinical sites, so the number is expected to grow. The second task is closely related to the setting of the samples & clinical data. Under the title “Legal/ethical issues management”, the task’s objective was to produce and approve all necessary documents related to ethical issues by the participant institutions (i.e. patient informed consent or case report form), and to ensure that the samples and clinical data comply with all applicable local and international regulation. An ethical committee was set-up to deal with the ethical, legal and societal issues which might arise during the project.

The construction of a Virtual Tumour Bank by the SIB was crucial to centralise all the information on the samples in a web-based format containing the following information: basic sample data (i.e. local inventory code, type, etc.), tumour specifications data (primary, non-relapsed from a single anatomical site), histological classification and sample quality information. To successfully do so, data was confidentially collected from different clinical centres. Once collected, data was cleaned and consolidated in a homogeneous format. Finally, all the information was uploaded in a website only accessible to partners.
The final VTB is available for end users. The page includes the heading, menu bar and footer as specified in the template file, as well as the custom page contents in the central part of the page, as specified in the homepage source file.

Also, the option of a Java-based implementation of the VTB was explored. It was however discarded after development, security and maintenance costs considerations. The database includes all the clinical variables of interest and a specific protocol which was decided under consensus by all the clinical groups and institutions providing clinical data within COLTHERES. All samples not complying with the requirements after second or further checks and/or with incomplete clinical data were rejected. The clinical database was created as stated in the proposal. All patient clinical data are coded keeping patient’s identifiers to a minimum (the consortium members use sequential identification numbers to identify the samples).

During the final part of the project, when the scientific impact became remarkable because of high impact publications and conferences, various COLTHERES partners (AG, NKI, VHIO, SIB, KUL) were invited to join the colorectal cancer subtype consensus consortium (CRCSC) hosted by SAGE: (doi:10.7303/syn2623706).
SAGE is a non-profit research organization based in Seattle, US and collaborates with a worldwide network. SAGE is using the Synapse platform to store and manage large datasets (http://sagebase.org/synapse/).

In addition, SAGE provides bioinformatics analysis tools and support.

The CRCSC members have agreed to study the use of the Coltheres database in larger academic-based datasets like SAGE’s in the benefice of CRC research.

The goals of the CRCSC are perfectly aligned with those of COLTHERES:
(i) to compare and validate the major published colorectal subtypes;
(ii) to conduct an integrative analysis across the pooled data sets to establish a robust consensus of molecular subtypes;
(iii) to define the clinical and molecular hallmarks of these subtypes;
(iv) to investigate the clinical value of these subtypes in patient studies.

This alignment, together with the fact that and the SAGE consensus effort can achieve much larger numbers of samples than the COLTHERES consortium in their own, has led to the agreement of the COLTHERES partners to study the inclusion of the COLTHERES database within the SAGE database in the near future.

Development of positive predictors to therapies targeting the EGFR pathway
The presence or the absence of activating mutations in KRAS, BRAS, PIK3CA exon 20 and NRAS can help to predict the efficacy of cetuximab before treatment in mCRC, thus improving the cetuximab therapeutic index. Analysis of these mutations in randomized trials indeed confirms these initial findings, presentation of extended KRAS, BRAF and NRAS mutation analysis on the OPUS randomized trial of Folfox +/- cetuximab at ASCO GI 2014 by Coltheres member (Tejpar S et al, Effect of KRAS and NRAS mutations on treatment outcomes in patients with metastatic colorectal cancer (mCRC) treated first-line with cetuximab plus FOLFOX4: New results from the OPUS study., ASCO GI Jan 2014) as well as publications on non Coltheres series (Douillard et al NEJM 2013). A novel mechanism of primary resistance was identified by the Coltheres consortium. In a study led by UNITO, KRAS gene amplifications were found to predict primary resistance to anti –EGFR therapy. We performed a screening of 1,039 CRC samples to assess the prevalence of KRAS amplification in this tumour type and further evaluated the role of this genetic alteration on the sensitivity to anti EGFR therapies. We detected KRAS amplification in 7/1,039 (0.67%) and 1/102 evaluable CRC specimens and cell lines, respectively. KRAS amplification was mutually exclusive with KRAS mutations. Tumours or cell lines harbouring this genetic lesion are not responsive to anti-EGFR inhibitors. Although KRAS amplification is an infrequent event in CRC, it might be responsible for precluding response to anti-EGFR treatment in a small proportion of patients.

Construction of cellular models recapitulating the genetic milieu of CRCs
Coltheres generated isogenic cancer models carrying specific genetic alterations and expression changes present in target patient populations. These lines have been used as tools for performing functional genomics (gene over-expression and siRNA approaches) approaches to dissect and validate candidate resistance biomarkers and tools prospectively screen for other markers of primary or secondary drug resistance e.g. signalling pathway proteins, drug transporters & metabolisers. The aim was also to generate second-generation ‘isogenic’ cancer models carrying secondary resistance genes/alterations to screen for resistance reversing drug combinations in vitro and in vivo and to perform ‘omics on these disease models for subsequent correlation and validation with patients sample data. Over 50 isogenic cell lines were generated and made available to Coltheres partners.

Functional-genomic analysis of cellular models identify new markers of primary/secondary resistance
Within Coltheres a tumour-stromal cell co-culture assay system was developed. HD successfully established an imaged-based approach to analysing tumour-stromal cell co-cultures, where the tumour cells were labelled with GFP to enable their growth to be differentiated from that of the stromal cell component of the co-culture. As proof-of-concept HD co-cultured GFP-expressing LIM1215 CRC cells with MRC5 stromal fibroblasts and where able to demonstrate that the stromal cell population drove resistance to cetuximab in the tumour cell population. HD demonstrated that MRC5 fibroblasts secrete high levels of HGF and that addition of exogenous HGF to LIM1215 monocultures also rendered them less sensitive to cetuximab, which are consistent with the possibility that HGF secretion by the stromal cell population could be rescuing the LIM1215 cells from the effects of cetuximab inhibition in the co-culture. We extended the analysis of this co-culture system to other MAPK pathway-targeted agents and similarly found that the presence of the stromal cell population also reduced the sensitivity of the tumour cells to these agents.

As a proof-of-concept for the co-culture system HD set out to test the effect whether the presence of stromal cells have an effect on the sensitivity of LIM1215 CRC cells to EGFR inhibition with the monoclonal antibody drug cetuximab.

Reversing primary and secondary resistance with targeted siRNAs and compounds
HD and NKI analysed a large panel of colon cancer lines for Mek-inhibitor sensitivity, with two lines (SW480 and SW620) successfully selected as being resistant were used in the resistance reversing screens in the next period. To avoid duplication of effort with NKI, who made significant progress with these types of screens in 2D, and in recognition of the emerging importance of the tumour microenvironment in the response to KRAS pathway-targeted agents, HD decided to focus effort on the development of more complex 3D assays formats. These changes were officially ratified with the EU. A pair of DLD1 (CRC) cell lines isogenic for the presence (parental) or absence (KRAS G13D KO) of a KRAS G13D mutation was used to set up 3D soft agar assays and to study KRAS dependency in 2D versus 3D assay formats. These results were proof-of-concept to the importance of performing screens for MAPK pathway-target compounds under 3D in addition to 2D assay conditions. To extend the observations made with the DLD1 KRAS isogenic pair and to build a panel of cell lines that could be used as screening tools > 40 (mainly non-isogenic) cell lines were screened for ability to grow under 3D conditions; suitable conditions were identified for >20 cell lines. A subset of these cell lines was evaluated for MEK inhibitor sensitivity in 2D versus 3D (soft agar) to determine their dependency on the MEK/ERK pathway. This revealed that similar to the DLD1s all cell lines tested that carry a KRAS mutation were more sensitive to MEK inhibition in 3D further supporting the rationale for perform a screen in 3D.

Predicting sensitivity and resistance to targeted agents using phase 0 clinical trials of CRCs
Coltheres we exploited xenogafted tumors derived from bioptic samples of CRC patients who initially responded to anti EGFR therapies and then relapsed with the aim to unveil the molecular bases of acquired resistance and to test new therapeutic combinations according to the genomic profile of the resistant tumors characterized. Established xenopatients retained the histopathologic and genetic characteristics of the original samples. Preclinical study and validation of potential new therapeutic targets or pharmacological combinations can take a strong advantage from well-designed xenopatients platforms as demonstrated in previous works (Bertotti et al., Cancer Discovery 2011).

Within Coltheres we proceeded with the accrual of bioptic material of colorectal cancer (CRC) patients who relapsed upon cetuximab (or panitumumab) treatment and their implantation into severely immunocompromised animals (NOD/SCID mice). Samples were transplanted in mice within 3-5 hours after biopsy. Upon transplantation, tumour pieces were allowed to engraft within 35~40 days without anti-EGFR treatment of xenopatients. After the first engraftment, tumours were expanded in cohort of 6 mice each and after 3 weeks, treatment with cetuximab started. Notably, we found that xenopatients derived from relapsed metastatic CRC (mCRC) patients showed full resistance to cetuximab, therefore confirming the optimization and validation of the procedure. Using the approach described above, we generated a first example of mouse xeno-transplant (patient -derived xenograft, or PDX) from a lung metastasis of a CRC patient who responded and subsequently relapsed upon anti-EGFR therapy (cetuximab). This tumor carried a KRAS exon4 mutation A146T. After implantation and engraftment the xenografted tumor was serially transplanted until production of four cohorts. Mice were then randomized to vehicle alone, cetuximab monotherapy, pimasertib monotherapy and their combination. Notably, cetuximab alone or the MEK inhibitor pimasertib alone had limited effectiveness, while combinatorial (cetuximab-pimasertib) treatment prominently impaired tumor growth and induced moderate shrinkage.

We also established a second type of xeno-transplanted mouse from a mCRC patient horbouring a different genetic lesion as mechanism of acquired resistance to anti-EGFR treatment. Notably we demonstrated that MET amplification was driving acquired resistance in this particular case (Bardelli et al., Cancer Discovery 2013). Therefore, as for the choice of therapeutic regimens, we focused on small-molecule inhibitors of MET that were administered individually or in combination with cetuximab. We selected JNJ-38877605, the MET-specific tool compound (not in clinical use), and crizotinib, a dual MET/ALK inhibitor that has shown promising antitumor activity in MET-amplified esophagogastric adenocarcinomas (Lennerz et al., J Clin Oncol 2011). The xenopatients showed cetuximab resistance also after different transplantations and expansions in mice, therefore confirming the value of this preclinical system. All treatments potently delayed tumor growth compared to vehicle or cetuximab alone. In detail, the antitumor activity of crizotinib was not enhanced by the addition of cetuximab, and the most effective modality in producing durable disease stabilization proved to be the JNJ-38877605–cetuximab combination.

Integrative bioinformatics data analysis
Coltheres investigated how knowledge of molecular characteristics of clinical CRC tumor samples, in particular their gene expression profiles, could contribute to the identification of determinants of resistance to pathway-targeting chemotherapies. A specific objective was to establish a detailed description of heterogeneity of primary CRC, test for the existence of “intrinsic” disease subgroups and propose a system of molecular subgroups that could be used directly as biomarkers or could be used as stratification factors when studying potential biomarkers. A second specific objective was to find new candidate resistance biomarkers by correlating molecular features with relapse and response data. A third objective was to deduce markers of resistance by investigating molecular variation in model systems or in patients under drug pressure. Finally, there was the intention to see if signatures derived from models were helpful in interpreting signatures found in human tumor data. The investigations performed in COLTHERES include:
- Subtype discovery in gene expression from primary CRC tumors.
- Consolidation of different subtype systems into a consensus system.
- Integrative study of multiple data sources (mutation, copy number, methylation, microRNA, proteomics) to characterize the proposed subtypes.
- Analysis of association between gene expression and response to anti-EGFR therapy in gene expression profiles of clinical cohorts.
- Analysis of correlation btw molecular characteristics and resistance in mouse xenografts.
- Analysis of tumor evolution and resistance emergence in patients using liquid biopsies.
- Functional genetic screens revealing causative mechanisms of resistance.
Given the successes of the functional screening studies and of the analysis of xenopatients cohorts (WP7), whose relevance for the clinical applications was compelling, biochemical pathway analysis related to resistance has been driven by those empirical-experimental analyses.
The analysis of diversity among gene expression patterns of primary CRC suggests a continuum of variance with a few (3-7 depending on the desired level of resolution) very distinct characteristic types. The situation in CRC is similar to that seen in many other solid tumors, for example ovarian cancer or glioblastoma. A subdivision of CRC in 3 subtypes differing in levels of stroma-EMT and levels of immunological infiltrations is quite clear, while a finer subdivision is based on more subtle multigenic differences that correspond less well to well-defined biomolecular patterns.
After the publication of subtype systems by six groups, a common consensus was found that encompasses the description of four subtypes, which we consider sufficiently well and clearly defined, that their roles can now be investigated in any research direction concerning properties of primary colon tumors. The validity and extension of these systems to metastatic CRC is an open question. The consensus system with four consensus molecular subtypes (CMS) with distinguishing features:
1) CMS1 (“MSI Immune”), strong immune components, high prevalence of MSI and CIMP, few somatic copy number alterations (SCNA) but many point mutations (hypermutation);
2) CMS2 (“Canonical”), epithelial, most highly chromosomally unstable (frequent SCNA), marked WNT and MYC signaling activation;
3) CMS3 (“Metabolic”), epithelial, with evident metabolic dysregulation, some overlap with CMS1 in relation to MSI status and hypermutation;
4) CMS4 (“Mesenchymal”), prominent transforming growth factor β activation, stromal invasion, and angiogenesis.

Survival analysis shows that patients with CMS4 tumors have significantly worse relapse-free survival and overall survival, CMS2 patients longer survival after relapse and a larger proportion of long-term survivors. The CMS1 group has the worst survival after relapse.
In the analysis of association between gene expression data and response to anti-EGFR therapy it appeared that there was little new information that could be exploited for an efficient discrimination of the tumors that are sensitive from those that are resistant to the treatment, in addition to what was already known or to the information incorporated in the subtype system. This might be due to the fact that one key determinant of resistance for metastatic tumors are minor tumor subclones harboring resistance-conferring genetic aberrations and for the primary tumor patients the existence of such aberrations in tumor cells that are seeding metastatic sites or are quiescent somewhere in the body. These cells do not impact the gene expression pattern of the bulk tumor tissue used for profiling. Deep sequencing technologies might bring some improvement in the future for the detection of minor tumor subclones. These subclones are small but can grow out quickly under treatment when most of the cells in the tumor mass stop growing or undergo apoptosis.

COLTHERES has identified key emerging factors of resistance in the amplification of growth pathway-controlling genes such as HER2, KRAS and MET and in the accumulation of activating KRAS mutations. These observations also suggest therapeutic options that might help to overcome resistance and to prolong patient life. In Genotype-Response correlation studies in mouse xenografts HER2 amplifications correlated with resistance and with constitutive activation of the receptor. A multi-arm trial in xenopatients showed that combined inhibition of HER2 and EGFR induced a more long-lasting regression. In longitudinal analysis of DNA from serum, detection of activating KRAS mutations could be causally associated with acquired resistance directly in metastatic CRC patients. Analysis of metastases from patients who developed resistance to cetuximab or panitumumab showed the frequent acquisition of secondary KRAS mutations or, more rarely, KRAS amplifications. Functional studies showed that cells with expression of mutant KRAS remained usually sensitive to combinatorial inhibition of EGFR and of the mitogen-activated protein-kinase kinase (MEK). Another less frequent event associated with acquired resistance was amplification of the MET gene and this was also detectable in circulating tumour DNA before relapse was clinically evident. Also in patient-derived colorectal cancer xenografts, MET amplification correlated with resistance to EGFR blockade, which could be overcome by MET kinase inhibitors. These results highlight the role of MET in mediating primary and secondary resistance to anti-EGFR therapies in colorectal cancer and encourage the use of MET inhibitors in patients displaying resistance as a result of MET amplification. Similarly, functional genetic screens suggested that BRAF mutant colon cancers for which there are currently no effective targeted treatment options available, might benefit from combination therapy consisting of BRAF and EGFR inhibitors.
These findings may offer an opportunity to anticipate and counter resistance by using combination therapies before patients relapse. KRAS mutant alleles could be detected in the blood of patients several months before radiographic detection of disease progression, thus revealing also the potentially huge value of regular blood monitoring for the clinical management of colorectal cancer patients at high risk of metastasis. These findings also offer immediate opportunities to design various clinical studies aimed at determining the optimal personalized treatment.
Design of innovative hypothesis-driven clinical Studies
Approximately 40% of mCRC show a silencing of the MGMT gene. In particular, in a retrospective analysis on 244 CRC samples. COLTHERES Partners (Idibell) found that 71% of tumors with G>A mutation of KRAS showed asimultaneous epigenetic inactivation of MGMT, thus demonstrating a strong association between the MGMT promoter hypermethylation and the presence of KRAS mutations. Furthermore, the same Authors described that MGMT hypermethylation is present also in 35% of CRC with KRAS wild type.

It has also been showned that MGMT hypermethylation is associated with mutations G:C > A:T in KRAS but not in adenomatous polyposis gene (APC), suggesting that MGMT hypermethylation may succeeds APC mutations but precedes KRAS mutations in colorectal carcinogenesis. In conclusion, the loss of MGMT expression compromises DNA repair in tumor cells and play a significant role in the progression of solid tumors, particularly for CRC, and in sensitivity to anticancer therapy based on DNA damage. The mechanism of action of temozolomide is DNA methylation at O6-guanine site, leading to a base pair mismatch.

For this reason, COLTHERES partners (Idibell and Niguarda Hospital) hypothesized that MGMT inactivation through hypermethylation may confer sensitivity to alkylating agents such as temozolomide or to its analog dacarbazine in mCRC. From the above observations, temozolomide could represent a useful therapeutic alternative in patients relapsed/refractory to conventional second-line therapies.

Based on this observation COLTHERES partners launched a clinical trial entitled ‘Phase II study of temozolomide in metastatic colorectal cancer patients resistant to standard therapies and with O6-methylguanine-DNA methyltransferase (MGMT) promoter hypermethylation.

The pre-selection of patients based on MGMT hypermethylation could be applied to improve the efficacy of this drug in mCRC. The purpose of this study is to elucidate the value of patients selection for MGMT status and to consolidate the role of temozolomide as a single agent treatment in this mCRC setting.

Data coordination center and dissemination
The final goal of COLTHERES was to develop clinically relevant personalized algorithms to be used for the prediction of individual therapy response to molecularly targeted therapies in CRCs. Throughout the project, the consortium was committed to rapidly releasing data both to the scientific community and to the general public. Primary data has been made available to the scientific community by publication in field-specific scientific / peer-reviewed journals and a dedicated dynamic web site – including both public and private sections – has been set up early in the project to publicise the results generated by the consortium. Articles, conferences, press releases and other publications were posted on the home page. Users were able to circulate posts (using Twitter, posting on their Facebook page or emailing the post to others), follow COLTHERES on Twitter (http://twitter.com/#!/coltheres) and subscribe to the COLTHERES newsletter. Posts were categorised (articles, conferences, “Nature Paper”, press releases and publications) and archived for easy access.
The generation and publishing of a rational strategy or ‘cascade’ of in vitro and in vivo tests using defined disease models and markers to direct future drug research and development activities has been achieved. Scientific results have also been presented during international scientific conferences and special care has of course been paid to the use of unpublished data by partners. Results have attracted the interest of both specialized and lay media as well as that of pharmaceutical industries.
Data produced by the consortium has been stored in databases (data coordination center) and consistent datasets downloadable by the scientific community has been balanced with the protection of patient interests. The confidentiality of the data produced has been rigorous.
Dissemination was done via press releases, organisation of seminars, poster discussion / presentation and talks /lectures at conferences. The dissemination activities are listed in section 2 of this report.

Potential Impact:
The results of COLTHERES show convincingly that there are strategies to improve patient survival based on the molecular characterization of the disease and the selection of an appropriate combination of existing drugs. This could have a potentially quick positive clinical and therapeutic impact and strongly advocates for the rapid implementation of molecular tumour characterization in clinical practice. While the reported studies already suggest some therapeutic options (clinical trials ongoing), they call for further studies to use the new technologies to determine which is the optimal treatment depending on the molecular signature of the tumor and which is optimal use of the new profiling and monitoring strategies in patient management.
The functional and model system studies suggested immediate opportunities for testing the utility of rational combination therapies, so for example the combination of EGFR and BRAF inhibitors in BRAF mutated CRCs or of HER2 inhibitors and MEK inhibitors in KRAS mutated CRCs. Beyond these first propositions, even more importantly, they show a way how good combination therapies could be determined.

The CRC subtype system that was developed could help in this problem of mapping model systems back to the relevant patient group. A first application of the system is in prognostic prediction, as survival patterns differ significantly even if not hugely across subtypes. A second application is in the retrospective analysis of clinical trials to test, if they can yield relevant information on treatment benefit. The CMS1 group appears to be associated to a constitutive over-activation of the MAPK pathway and to be more resistant to current anti-growth factor receptor treatments like cetuximab. On the other hand the hypermutability goes along with the generation of a rich repertoire of cancer-specific antigens and immune reactivation strategies might be most promising in this CRC subtype. Tumors of the CMS4 group might rely on active angiogenesis to support their growth and anti-angiogenic therapies might be more successful in this subtype, maybe in combination with TGF-β pathway inhibition.
Globally, the most important impacts are the contribution to innovative paradigms for the characterization and treatment of cancer. Liquid biopsies allow patient monitoring to optimize interventions and can revolutionize clinical oncology. Model system and functional screening studies can suggest how to match best treatment options to molecular profiles and make these more widely actionable than they are today. In this regard, discoveries made within COLTHERES are presently being tested in ad hoc clinical trials (designed by COLTHERES partners).
List of Websites:
The COLTHERES public website is www.coltheres.org
P1 UNITO Alberto Bardelli (alberto.bardelli@ircc.it)
P2 IDIBELL Manel Esteller (mesteller@idibell.cat) and Anna Martinez-Cardus (amartinezc@idibell.cat)
P3 HD Charlotte Batley (c.batley@horizondiscovery.com)
P4 AG Inès Goossens-Beumer (ines.goossens@agendia.com)
P5 NKI Rene Bernards (r.bernards@nki.nl)
P6 VHIO Josep Tabernero (jtabernero@vhio.net)
P7 ONCG Salavatore Siena (Salvatore.Siena@OspedaleNiguarda.it)
P8 KUL Sabine Tejpar (sabine.tejpar@uz.kuleuven.ac.be)
P9 UNILIV Michael Clague (clague@liv.ac.uk)
P10 SIB Mauro Delorenzi (Mauro.Delorenzi@isb-sib.ch)
P11 ARTTIC Paul Crompton (pdc@arttic.be)