Aberrant Splicings Due to Microsatellite Instability in Colorectal Cancer : Physiopathological and Clinical Impact

WP1 - To identify exon/intron junctions that are specifically affected by aberrant splicing events in MSI CRC. Sequences will first be evaluated at CNG using Illumina's Pipeline CASAVA (Consensus Assessment of Sequence And Variation) software. This program converts intensity scores into base calls, quality scored alignments and additional formats for downstream analysis, thus rapidly transforming data into biologically relevant information. The filtered data will then be transferred to the CIT platform (Carte d'Identité des Tumeurs; http://cit.ligue-cancer.net, dir: A. de Reynies) to be analysed by our bio-informatics expert. The Cufflinks workflow will be used to quantify transcript expression levels in MSI tumours. This allows transcript assembly, discovery and differential expression measures at transcript-level resolution. Because the standard Cufflinks workflow does not support gene fusion discovery or quantification, several new features will be incorporated into it. Firstly, SoapFuse will be used to detect fusion break points and to predict fusion junction sequences. These will be integrated into the human reference genome and gene annotation from the Gencode project to provide a comprehensive, integrated annotation of gene features for mapping of splicing reads. The integration INCa - PRTK 2014 17/51 process will abide by several rules to minimise the potential to disturb the quantification of expression level in forthcoming analyses. With the help of our customised reference genome and annotation, TopHat, an aligner that supports splice junction and gene fusion mapping will be used for RNASeq mapping. Cufflinks will then be used to find new splice variants, including new exon skipping isoforms. These will be integrated into the annotation using Cuffmerge. Depending on the alignment results obtained with TopHat, several filters will be applied to remove low quality candidates for splice variation and gene fusion, as well as candidates that are incompatible with existing annotated transcripts. Where necessary, reads will be assembled by AbySS and then aligned onto the reference genome by BLAT to provide more information for refining the annotation. This will produce a transcriptome assembly containing high confidence gene fusion and exon skipping events. Identification of these events will then be performed in individual samples. Cuffdiff will be used to analyse the mapping result, also based on this transcriptome assembly, for calling differentially expressed genes and transcripts and for detecting differential splicing changes. Finally, CummeRbund will be used to interpret and visualise the results. The reliability of RNA-seq analysis will be verified by searching for aberrantly spliced transcripts already reported in MSI cancers (eg. MRE11 and HSP110) and for point mutations in the coding sequences of target genes for MSI (eg.TGFBR2, IGF2R, TCF7L2, AXIN2, PTEN, RIZ) and in other cancer-related genes that serve as internal positive controls (eg. KRAS, BRAF, TP53, PIK3CA). It is worth noting this project is part of several others developed jointly with CIT-Ligue and that are aimed at characterizing MSI CRC using Omics technologies. Importantly, this data is already available for a significant number of samples and could therefore be exploited if required. Data from the RNASeq cohort of patients will be comprehensively analysed to identify recurrent splicing aberrations (expected to be mostly exon skipping) that occur specifically in MSI colon tumours compared to MSS CRCs and matching normal colonic mucosa. Amongst these, the study will focus on splicing aberrations that are due to MSI and that affect exons with a flanking upstream intron containing a ≥ 15 bp LNCR that is located ≤ 6 bp from the intron-exon junction (splice acceptor site) (RNASeqMSI-exon pre-list). As stated above, about 2,000 human genes contain at least one intron with a LNCR very close to the AG splice acceptor site at the intron-exon junction. Approximately 100 human genes could be affected by recurrent and specific aberrant splicing events due to MSI in CRC (mostly exon skipping; deduced from experiments performed in a limited series of CRC cell lines and primary tumours using exon arrays; preliminary and unpublished results). WP2 - To investigate for functional links between MSI and aberrant splicing events. Following the RNASeq analysis, confirmation of the aberrant splicing events due to MSI will be required using another methodological approach in order to eliminate false positive events. This will be achieved with real-time quantitative INCa - PRTK 2014 18/51 RT-PCR using internal, specific probes (Applied biosystems). For each skipped exon in the RNASeqMSI-exon pre-list, a common pair of forward and reverse primers located in the flanking exons will be designed. Two internal probes will be designed, located either within the skipped exons or spanning the flanking exons at their junction in order to detect normal or aberrantly spliced mRNA in a competitive manner, respectively. It is highly sensitive and also avoids false positive signals due to contamination with genomic DNA. Candidate exons that will be retained are those that display aberrant and recurrent skipping in MSI CRC cell lines and primary tumors as compared to MSS CRC controls. In line with our working hypothesis, the study will then determine whether each confirmed splicing aberration is MSI-driven. This will be achieved as described earlier for T17 deletions in intron 8 of HSP110 that were identified specifically in MSI CRC and lead to exon 9 skipping. Briefly, allelic profiles of adjacent intronic LNCR (see above) will be analyzed using fluorescence-based genotyping in the panel of MSI and MSS CRC cell lines, as well as in the complete series of MSI primary tumors from the RNASeq patient cohort and paired normal mucosa (in order to assess polymorphic status). This will be performed using the same method developed earlier in our laboratory for analysis of the HSP110 T17 DNA repeat. Following migration of PCR products on an ABI 3100 Genetic Analyzer with GS400HD ROX size standards and POP-7 polymer (Applied Biosystems), GeneMapper V4.0 software (Applied Biosystems) will be used to analyze LNCR traces, with application of an AFLP (Amplified Fragment Length Polymorphism) method. Traces will be considered acceptable when the peak amplitudes are between 100 and 6,000 fluorescence units. An MSI Perl script has been developed to automatically compare LNCR traces in normal and tumour samples, thus allowing detection of aberrant LNCR peaks that fall outside of the polymorphic zone observed in the normal population. As with HSP110, it is expected (i) to detect somatic deletions/insertions in some of the candidate LNCRs using this approach, and (ii) to identify those whose somatic alterations due to MSI are significantly associated with exon skipping-related events at the RNA level (MSI exon final list). WP3 - To identify splicing events and/or LNCR mutations with clinical relevance in MSI CRC patients. The clinical relevance of candidate genes (MSI exon final list) will be assessed using multivariate survival regression models for RFS (Relapse-Free Survival). The number of candidate splicing events and/or LNCR mutations is approximately 100 (see WP1 and WP2 above) and INCa - PRTK 2014 19/51 other known clinical determinants such as stage, treatment and age at diagnosis will be considered in the multivariate models. False positives are one of the major pitfalls in identifying potentially relevant markers amongst dozens of candidates. As the number (p) of covariates to be considered will be of the same order as the number of individuals, the "high-dimensional setting" will be reached. Consequently, the usual algorithms for survival regression models (e.g. coxph in R) will fail to estimate the parameters and to identify events with clinical relevance. In our analysis, three main methodological issues require special attention, particularly at the algorithmic level. WP3 will be divided into two main steps, the first of which concerns comparisons and the development of statistical algorithms. Once tuned, the algorithms will be run to identify a prognostic biomarker(s) that involves splicing events and/or LNCR mutations with clinical relevance. High-dimensional regression models and lasso algorithms. In the first step only splicing mutations and clinical determinants in the survival regression models will be considered. In this case, variable selection and parameter estimation will be conducted with a lasso (or elastic net) algorithm (see Simon et al. for the Cox model, and Gaiffas et al. for Aalen model, both implemented in R). In previous publications it was demonstrated that cut-point values resulted in maximal survival differences between patient groups with large or small deletions in the HSP110 T17 LNCR, and with high or low expression of mutant HSP110 mRNA due to exon 9 skipping at the INCa - PRTK 2014 20/51 mRNA level. Since other splicing events could present with the same threshold effect, the cut-points for up to 100 candidate splicing events will be carefully determined. Even in classical statistics, cut-point determination is a known difficulty because of overdetection. As recently proposed in a related context, the "lasso with pre-screening algorithm" could be adapted to incorporate the cut-point determination into the main algorithm. Other points. Missing data will be handled by multiple imputations from nearest-neighbour or regression methods. The study will consider possible interactions with the use of chemotherapy. As a final step, estimations will be run on the training cohort (Saint-Antoine) to derive a prognostic biomarker that will include splicing events and/or LNCR mutations with clinical relevance. This biomarker will be validated on the test cohort (multi-centre). Bootstrap analysis will be conducted to ascertain the biomarker. WP4 - To initiate functional studies on a limited number of clinically relevant, cancer related genes whose splicing is highly perturbed in MSI cancer cells, and to develop biological tools to simplify screening in future clinical assays. As stated earlier, A preliminary study has been performed using exon arrays in a small series of MSI CRC cell lines and primary tumours (unpublished). This was conducted to evaluate feasibility, time, cost, adverse factors, effect size (statistical variability) and to improve upon the study design prior to performing the present full scale research project. The functions of the 100 detected candidate genes (some of which may overlap with those identified from RNAseq screening) were frequently related to a cancer-related processes such as macromolecular synthesis (30%), cell proliferative capacity or cell death (20%), drug resistance (10%) and others (WNT pathway, metastatic process, changes in chromosome structure). The present project expect to identify several robust MSI-driven splice mutants that are clinically relevant (see WP2 and WP3 above). These mutants could have either oncogenic or antioncogenic functions, given that some (such as HSP110) may be produced at high levels even though they have negative impacts on the tumour cell (see our functional hypothesis above concerning the expected detrimental influence of MSI at LNCR in CRC). In this context, in vitro functional studies will be designed to characterize the oncogenic impact of a small number (n=5) of putative clinically relevant mutants. These experiments will be based on transient silencing or overexpression using siRNA or plasmids and ad hoc biological read-out in CRC cellular models already available in our laboratory. Depending on the results obtained, further investigations could then be planned using stably transfected CRC models xenografted into nude mice. In addition, the study plan to validate biological tools (e.g. Antibodies) to optimize the future screening of patients using routine assays, similar to our work with HSP110 and the HSP110DE9 mutant. Tissue Microarrays (TMAs) will be constructed from routinely prepared, formalin-fixed and paraffin-embedded blocks collected INCa - PRTK 2014 21/51 retrospectively from the Pathology Department of Saint-Antoine hospital. Neoplastic tissue will be sampled, including the tumour invasive front (3 to 6 samples of 0.6 mm diameter tissue cores). When available, paired lymph node metastasis will also be sampled. Immunohistochemistry with antibodies generated specifically (subcontracting) to recognize wild-type or mutant candidate proteins will be performed on TMAs. Exon skipping events are frameshift in 2/3 of cases and thus generate truncated proteins that have immunogenic, aberrant C-terminal tails. Correlations between protein expression and clinico-pathological features will be evaluated, as well as their prognostic and predictive values. TMA slide images will be captured as high-resolution digital files and evaluation of each staining will be done by two pathologists.