413. Stratification of Metastatic Colorectal Cancer Patients Using NGS Sequencing, Neo-Epitope Detection and Immune Infiltrate AnalysisFang Yin Lo1, Nitin Mandloi4 Timothy Yeatman2, Kiran V Paul4, Ashwini Patil4, Steven Anderson3, Ravi Gupta4 and Anup Madan1

1Covance, Redmond, WA; 2Gibbs Cancer Center, Spartanburg, SC; 3Covance, Durham, NC; 4MedGenome Inc., Foster City, CA

Presented at AACR 2017

Introduction Colorectal cancer (CRC) is the third most common type of cancer in the United States. Although chemotherapy, radiation and targeted therapies can improve survival rates, recent studies have shown the potential benefit of immunotherapies to improve outcomes for patients with advanced CRC. Targeted therapies that use monoclonal antibodies (mAbs) to EGFR have been shown to benefit some CRC patients.1 Until recently, KRAS has been the only predictive biomarker for anti-EGFR therapy for metastatic CRC. However, 40% to 60% of patients with wild-type KRAS do not respond to anti-EGFR therapy. Therefore, to accurately predict patients’ response to treatments and improve clinical outcomes, additional prediction and treatment methods are imperative. One of the many efforts to improve prediction for CRC patient’s response to the anti-EGFR therapy is the development of gene expression based RAS signature scores for identification of RAS activated tumors independent of mutations in the KRAS gene.2,3 In addition to passive immunotherapy using mAb, there have been major advances in targeted active immunotherapy in other tumors, including checkpoint inhibitors and cancer peptide vaccines.4,5 In melanoma, there has been preliminary clinical findings indicate that combined targeted therapies and simultaneous active immunotherapies such as blockade of multiple immune checkpoints could promote therapeutic synergy and improve clinical outcomes for patients. In addition, chromosomal rearrangements have the potential to alter gene function in many different ways.

Recently there have been major advances in detecting these chromosomal rearrangements. Fusion genes such as BCR-ABL and EML4-ALK have become targets for therapy in cancer. There is considerable effort being placed on combinatorial ways of tumor stratification to improve responses for these cancers. Similarly, since no single treatment can apply to all CRC patients, we aim to stratify patients using a combination of the following methods:

▶ RAS signature score based on the expression profile of 18 genes, enabling measurements of mitogen-activated protein/extracellular signal-regulated kinase (MEK) pathway functional output independent of tumor genotype

▶ Expression profile of immune checkpoint inhibitor target genes, such as PD1 and PDL1

▶ DNA mutational profiles of genes such as KRAS, APC, BRAF and NRAS

Further, we investigate the potential immune reactivity in these CRC samples, and thereby the potential benefit of immunotherapy, by evaluating the tumor neo-epitope burden, and the quality of immune cell infiltration based on exome-seq and RNA-seq analysis.

Methods55 FFPE samples were selected from a cohort of 468 samples with matching FF samples. These 55 samples have about 1:1:1 ratio of high, medium and low RAS scores. Here we showed our capability to obtain RAS signature scores with concordant results using different platforms including whole transcriptome RNA-seq, Affymetrix® microarray, targeted RNA-seq and Nanostring®. We discovered that samples that have RAS activating mutations such as KRAS and BRAF have significant higher RAS scores (p<0.001). On the contrary, expression of PD-L1 was significantly lower in tumor samples harboring mutations of genes such as MET, PTEN, NRAS, FBXW7, and GNAS. Kruskal-Wallis test showed that the expression of PD-L1 was significantly lower in samples with higher RAS signature scores (p<0.05). Usingthe RNA-sequencing data, we were able to detect gene fusion events in these tumor samples. After filtering out low confidence results, a total of 730 gene fusion events were detected among the 55 tumor samples. While most of the gene fusion events were only detected once within the sample cohost, some were detected in multiple samples. For example, the fusion between KANSL1 and ARL17A was detected in 18 of the 55 samples. This is a relatively new discovery that had just started being mentioned in other cancer research institute reports.9 Other fusions that appeared multiple times include SAMD5 and SASH1. Interestingly, we discovered that significantly fewer fusion events were detected in samples with lower RAS signature scores than samples with higher RAS scores (p < 10-5).

Table 1. Summary of Mutation Identified from Exome seq

Variant Classification

No. of Unique Variants in All 53

Samples

No. of Variants Shared by >2

Samples

No. Shared Mutations that

Belong to Cancer Census Genes

Non-synonymous SNV

779 mutations 263 genes

42 mutations 23 genes

4 mutations 4 genes

Splice Site Mutants126 mutations

30 genes1 mutation 1 gene

1 mutation 1 gene

In/Del 0 0 0

Several driver gene mutations were identified in this study, KRAS, TP53, PIK3CA, APC and HER2. APC mutations could be germline in origin, although this could not be confirmed due to the lack of paired normal samples.

Figure 1. Colorectal cancer samples cohort selection strategy. The cohort was selected by filtering out colorectal cancer samples available as formalin-fixed, paraffin-embedded (FFPE) and flash frozen (FF). Samples were then filtered for known RAS score obtained from Affymetrix array. Known RAS scores are divided into 3 groups evenly: low (<33% percentile), medium (33%-66% percentile), high (>66% percentile).

Figure 2. Multi-platform comparison. Samples derived from the same 55 FFPE blocks were assayed across multiple platforms. The method design to combine RNA analysis (gene expression signature scores) with DNA analysis (i.e., mutation status) allows for comparison of RAS signature scores and overall gene expression from different platforms.

Figure 3. Flowchart for the analysis. 55 samples went through 5 different platforms for gene expression measurements – Affymetrix, whole transcriptome RNA-Seq by two different library preparation methods. Data went through quality control and normalization. For RAS score calculation, 18 genes were used based on previous study.5

Figure 4. The number of mutation versus KRAS status. KRAS mutant samples have significantly higher number of non-synonymous mutations than KRAS wild type samples.

Figure 5. RAS signature scores and the number of gene fusion events. (A) Distribution of gene fusion events of all samples. Only high and medium confidence gene fusion events based on results from JAFFA8 were considered. (B) Samples with lower RAS scores have significantly fewer gene fusion events detected than samples with higher RAS scores.

Figure 6. Scatter plot of enriched GO cluster representatives. Multidimensional scaling is applied to the list of significantly enriched GO terms in fusion genes found in the CRC samples.9

Figure 7. Neo-epitope prediction and prioritization.

Figure 8. Work flow for tumor microenvironment analysis. First a content analysis was done using expression of gene signatures associated with epithelial, stromal and immune cells. Next the immune cell compartment was further stratified into seven different immune cell types using signatures that are specific to each of the following immune cells: CD8 T-cells, CD4 T-cells, T-regulatory cells, NK cells, B-cells, macrophages and myeloid derived suppressor cells.

Figure 9. Tumor content analysis. RA: RNA Access RNAseq. RD: ribosomal depletion RNAseq. Stromal, Immune and Epithelial scores were generated using gene signatures for each of the three compartments. The expression of the genes contained in the signatures were integrated into a score using ssGSEA (single sample Gene Set Enrichment Analysis). Cellular content across different compartments is similar with TCGA data.

Figure 10. Immune phenotyping for CRC samples. RA: RNA Access RNAseq. RD: ribosomal depletion RNAseq. Myeloid-derived suppressor cells are the most prevalent cell type in colorectal cancer both in our data and TCGA data. The second most prevalent cell type is macrophages. Ribosomal depletion RNAseq data is more sensitive than RNA Access data for detecting immune cell types.

References1. Gallagher DJ, Kemeny N. Metastatic colorectal cancer: from improved survival to potential cure. Oncology. 2010; 78:237-248.

2. EGFR gene copy number as a prognostic marker in colorectal cancer patients treated with cetuximab or panitumumab: a systematic review and meta-analysis.

3. KRAS mutation status is predictive of response to cetuximab therapy in colorectal cancer.Lièvre A, Bachet JB, Le Corre D, Boige V, Landi B, Emile JF, Côté JF, Tomasic G, Penna C, Ducreux M, Rougier P, Penault-Llorca F, Laurent-Puig P. Cancer Res. 2006 Apr 15; 66(8):3992-5.

4. Loboda A et al. A gene expression signature of RAS pathway dependence predicts response to PI3K and RAS pathway inhibitors and expands the population of RAS pathway activated tumors.

5. BMC Medical Genomics 2010, 3:26Dry JR et al. Transcriptional Pathway Signatures Predict MEK Addiction and Response to Selumetinib (AZD6244). Cancer Res. 2010 Mar 15; 70(6):2264-73.

6. Nat Rev Cancer. 2012 Mar 22;12(4):252-64. doi: 10.1038/nrc3239.

7. Cancer J. 2011 Sep-Oct;17(5):343-50. doi: 10.1097/PPO.0b013e318233e5b2.

8. Genome Medicine 2015, 7:43

9. http://newswise.com/articles/rutgers-cancer-researchers-examine-gene-fusion-and-treatment-implications-for-breast-cancer

10. Supek F, Bošnjak M, Škunca N, Šmuc T. PLoS ONE 2011.