Supplementary Figure S3. Preferential targeting of Side Population (SP) CSCs in NCI-H69 cells by VS-5584.

Supplementary Figure S5. Representative Aldefluor FACS data of cells dissociated from MDA-MB-231 and MCF-7 tumors (Fig. 3B and 3D).

Supplementary Figure Legends. Legends for Supplementary Figures S1-S5

<p>Supplementary Figure S1. Aldefluor+ SUM159 breast cancer cells show enriched CSCs and tumor initiating-capability</p>

<p>Supplementary Materials and Methods. More detailed Materials and Methods</p>

Supplementary Figure S2. Side Population of SUM159 cells enriches for tumor-initiating capability.

Supplementary Figure S4. Representative images of tumorspheres formed by SUM159 and MCF7 cells.

3100 Background: Cancer vaccines against neoantigens have shown promise in treating advanced stage cancers [1,2]. Typically, detection of somatic mutations underlying neoantigen expression is done using WES, a cost and time-ineffective platform for routine clinical use. Here, we assess the feasibility of a large NGS panel for use in cancer vaccine development through discovery of somatic mutations and prediction of neoantigen load. Methods: Various patient tumor samples with a range of mutational loads [n = 30; mean: 84 mutations, range: 34-248] were profiled using CANCERPLEX, a 435-gene panel for identifying clinically-relevant genomic alterations in cancer [3]. For each mutation, all possible 9- and 10-mer peptides were enumerated. For mutations that result in novel open reading frames (neoORF), peptides overlapping the entire neoORF were considered. The 28 most frequent HLA-A and HLA-B alleles in the Caucasian, African, and Asian populations were used. Peptides were computationally evaluated for binding to HLA using the NetMHCpan-3.0 algorithm [4]. For each neoantigen, the rank relative to the collective mutant and wildtype 9- or 10-mers was determined. Neoantigens were classified as strong binders if rankmutant/rankwt ≥ 0.5. For each HLA allele, the neoantigen load was calculated as the number of strong binders per given allele. Results: The number of neoantigens was significantly correlated with mutational load. The neoantigen load was dependent on the HLA allele binding partner, with HLA alleles B07:02, A33:03, A30:01, and B42:01 predicted to have higher binding affinity. This allele dependent distribution suggests that certain HLA alleles are better binding partners for potential neoantigens or preferential for certain peptide. Conclusions: The high number of neoantigens identified suggests that CGP is a feasible alternative to WES for cancer vaccine development. CANCERPLEX accurately detects somatic mutations [3] thus enabling prediction of tumor neoantigens, an emerging biomarker in immune-oncology [5]. Future studies incorporating HLA transcription testing with CGP are warranted to further establish the platform for cancer vaccine development.

22 Background: Tumor genomic instability is positively correlated with immunotherapy response. It confers different tumor phenotypes, including high TMB (TMB-H) and high MSI (MSI-H). Recently the US Food and Drug Administration approved MSI-H phenotype as a biomarker for immunotherapy, highlighting its importance, but also bringing up the question of how TMB as another promising biomarker is going to add value in the field. Here, we characterized TMB and MSI profiles to better understand the potential TMB contribution and identify genomic markers for it. Methods: 734 solid tumor were collected, with 462 CRC, and 272 GA samples. Large panel Next-Generation Sequencing assay with the ability to determine TMB and MSI, was performed on each sample. Based on the TMB and MSI status, patients were grouped into four categories: THMH (TMB-H and MSI-H), THMS (TMB-H and MSI-Stable), TLMH (TMB-Low and MSI-H), and TLMS (TMB-Low and MSI-Stable). To identify genes that are related to the interplay of TMB and MSI, Random Forest and Lasso Regression models were applied to identify genes most predictive of the four categories. Results: TMB and MSI are highly correlated in our cohort of CRC and GA tumors. However, 5.8% CRC and 12.9% GA samples are under THMS (Table 1). In these samples, alternate DNA repair pathways are potentially dysregulated, including the nucleotide excision pathway (ATRX, APC), DNA double strand break repair (FANCF, SETD2), and the previously described proofreading pathway (POLD1). We hypothesize that these patients may also derive clinical benefit from immunotherapy. Conclusions: Immunotherapy benefits could be extended to more patients by jointly measuring MSI and TMB. The corresponding marker genes could also be extended beyond the commonly known POLE/POLD1 genes. Orthogonal validation by Whole Exome Sequencing data of the in silico mined marker genes is currently underway. [Table: see text]

e15632 Background: While commercially available hotspot testing detects over 99% of the activating mutations in KRAS located at codons 12, 13, 61, and 146, rare oncogenic variants beyond these regions can significantly impact treatment decisions in colorectal cancer (CRC). Herein we showcase the utility of comprehensive gene panel testing in identifying low frequency yet clinically relevant KRAS alterations to enable informed clinical treatment decisions. Methods: 575 CRC specimens were analyzed using CANCERPLEX over a three-year period (KEW, Inc; Cambridge, MA). CANCERPLEX is a comprehensive large gene panel comprising 435 cancer associated genes. A rare KRAS alteration was defined as a variant with a frequency > 0.02% in COSMIC. Variants with clinical actionability in CRC were defined as activating mutations with inferred therapeutic resistance to anti-EGFR antibody therapy. The analytical sensitivity and specificity for single nucleotide polymorphisms (SNPs) in CANCERPLEXwas 99.2% and > 99.9%, respectively. Results: KRAS Q22K, L19F, and G60D were identified as rare oncogenic variants in our analysis of 575 CRC cases. Q22K, L19F, and G60D has been previously reported in 0.009% (7/73000), 0.017% (13/73000), 0.003% (2/73000), respectively, in CRC cases (COSMIC). All three variants have been reported to have transforming potential in vitro and in vivo and trigger activation of downstream oncogenic pathways [PMID:11095964;26284123; 17150185; 20949621]. As activating mutations in KRAS predict lack of response to EGFR antibody therapies, cetuximab and panitumumab were reported as contraindications in all three cases. While not a rare variant, we detected activating mutations in codon 146 in 3% (17/575) of our cohort; strikingly, 5 of the KRAS A146 variants were not identified by prior less comprehensive modes of RAS testing. Conclusions: Low frequency, clinically relevant KRAS variants in CRC may be overlooked using commercially available RAS hotspot assays. In this study we demonstrate that a comprehensive gene panel may detect clinically consequential rare KRAS mutations that may define patients’ treatment and clinical course.

e18597 Background: Rapid advances in NGS technology have enabled high throughput generation of data allowing for more comprehensive approaches to precision medicine and improved cancer patient care. However, efficient and timely variant interpretation and reporting to guide physicians’ therapeutic decisions remains an unmet need. Manual interpretation of variants is onerous, unreproducible, and non-scalable. At the same time, full automation of highly-nuanced clinical interpretation without human intervention is still at a nascent stage and has high error rates. Herein, we describe development of a semi-automated platform that augments accuracy and efficiency of clinical interpretation and reporting by over 20-fold. Methods: To facilitate accurate and efficient variant interpretation and clinical reporting we developed and validated GENEKEEPER (KEW, Inc.; Cambridge, MA), a cloud-based knowledgebase and reporting tool. The underlying database architecture seamlessly integrates patient sequence data with numerous resources including, the internal knowledgebase, trial registries, professional guidelines, and other public databases. The internal knowledgebase includes over 40,000 variants from > 3,000 patient tumor specimens, annotated by PHD/MD expert analysts to be consistent with the AMP guidelines. Results: In an analysis of 20 sequenced tumors, curated using our platform, we noted a 10-fold increase in efficiency and 2-fold increase in accuracy as compared to manual curation. We also observed a near 100% congruency of reported clinical trials and therapies between different curators. The flexible architecture enabled facile implementation of AMP guidelines to enhance standardization and accuracy of variant interpretation. Overall the platform demonstrated a 20-fold increase in efficacy in report generation. Conclusions: Semi-automated curation platforms promote precision and efficacy in analyzing and reporting clinically relevant variants to guide informed treatment decisions. GENEKEEPER significantly reduces the laboriousness of manual curation and will empower clinical laboratories to exponentially scale-up NGS operations.

e15101 Background: Although the difference in right vs. left sidedness of colorectal cancer (CRC) in response to targeted therapy has been reported, we hypothesized that right-sided colorectal cancer (RCC) is more likely to have genetic alterations associated with resistance of anti-EGFR therapy. We tested this hypothesis using comprehensive genomic sequencing (CGS) on a set of samples from well-characterized CRC patients. Methods: Two-hundred-one primary colon cancer patients with either RCC or left-sided colorectal cancer (LCC) were analyzed. We investigated gene alterations using 415 gene panel, which includes the gene alterations associated with resistance of anti-EGFR therapy: TK receptors ( ERBB2, MET, EGFR, FGFR1, and PDGFRA), RAS pathway ( KRAS, NRAS, HRAS, BRAF, and MAPK2K1), and PI3K pathway ( PTEN and PIK3CA). We defined the patients who had no alterations in any of the genes as “all wild-type”, who are theoretically considered as responders of anti-EGFR therapy. Other patients with genetic alterations in resistance pathways were defined as “mutant-type”. Results: Fifty-six patients (28%) and 145 patients (72%) had RCC and LCC, respectively. Mutation of PIK3CA, BRAF, ACVR2A, MSH6, and CTNNB1 were significantly associated with RCC. Conversely, mutation of APC and TP53 were significantly associated with LCC. Regarding the gene alterations associated with anti-EGFR therapy, only 6 of 56 patients (11%) of RCC were “all wild-type”; in contrast to 41 of 145 patients (28%) of LCC that were “all wild-type” ( P = 0.009). In 45 Stage IV patients treated with anti-EGFR therapy, RCC showed significantly worse progression-free survival (PFS) than LCC ( P = 0.019), and “mutant-type” RCC showed worse PFS compared to the others ( P = 0.018). Conclusions: RCC is more likely to have the genetic alterations associated with resistance of anti-EGFR therapy compared to LCC. Primary tumor sidedness is a surrogate for the non-random distribution of molecular subtypes in CRC.

Objectives Anti-epidermal growth factor receptor (EGFR) therapy has been found to be more effective against left-sided colorectal cancer (LCRC) than right-sided colorectal cancer (RCRC). We hypothesized that RCRC is more likely to harbor genetic alterations associated with resistance to anti-EGFR therapy and tested this using comprehensive genomic sequencing. Materials and methods A total of 201 patients with either primary RCRC or LCRC were analyzed. We investigated tumors for genetic alterations using a 415-gene panel, which included alterations associated with resistance to anti-EGFR therapy: TK receptors (ERBB2, MET, EGFR, FGFR1, and PDGFRA), RAS pathway (KRAS, NRAS, HRAS, BRAF, and MAPK2K1), and PI3K pathway (PTEN and PIK3CA). Patients whose tumors had no alterations in these 12 genes, theoretically considered to respond to anti-EGFR therapy, were defined as “all wild-type”, while remaining patients were defined as “mutant-type”. Results Fifty-six patients (28%) and 145 patients (72%) had RCRC and LCRC, respectively. Regarding genetic alterations associated with anti-EGFR therapy, only 6 of 56 patients (11%) with RCRC were “all wild-type” compared with 41 of 145 patients (28%) with LCRC (P = 0.009). Among the 49 patients who received anti-EGFR therapy, RCRC showed significantly worse progression-free survival (PFS) than LCRC (P = 0.022), and “mutant-type” RCRC showed significantly worse PFS compared with “all wild-type” LCRC (P = 0.004). Conclusions RCRC is more likely to harbor genetic alterations associated with resistance to anti-EGFR therapy compared with LCRC. Furthermore, our data shows primary tumor sidedness is a surrogate for the non-random distribution of genetic alterations in CRC.

e23122 Background: Breast cancer is the leading cancer among women both in US and Japan. Triple negative breast cancer (TNBC) is the subset of breast cancers that are negative for estrogen receptor (ER), progesterone receptor, and HER2. While the selective ER modulator tamoxifen or anti-HER2 therapy for breast cancer patients with ER positive or HER2 protein overexpression are the most successful examples of targeted therapies, only limited therapies are available for patients with TNBCs, which are associated with a poor prognosis. Advance in next-generation sequencing enables us to identify actionable driver mutations that can be potentially treated by targeted therapies in each cancer patient. The aim of this study is to examine actionable driver mutations in TNBCs in Japan by comprehensive genomic sequencing (CGS) with 435 gene panel, and compare the driver events in Japan with TCGA database to validate the utility of CGS. Methods: We examined all exons of 435 known cancer genes in Japanese TNBC patients (N = 53) by CGS and evaluated for concordance among independent data obtained from The Cancer Genome Atlas-TNBC whole exome sequencing database (N = 123). Results: Oncogenic driver mutations were identified in 51 of 53 Japanese patients (96%) with TNBC and 36 of 53 patients (67%) harbored mutations in genes associated with FDA-approved targeted therapies, indicating the potential clinical utility of a large gene panel for evaluating patients with TNBC. Among 80 total genetic alterations, frequently mutated genes ( > 10% patients) were TP53, PIK3CA and PTEN. Overall, the mutation spectrum of the Japanese patients is similar to that of the TCGA population, except amplification of MYC. Conclusions: Use of a CGS panel of 435 genes can reliably identify all of the critical mutations in TNBC patients, which are similar as TCGA data. The information derived from CGS can be used to determine the optimal treatment for TNBC patients.

e15592 Background: Molecular heterogeneity represents a significant hurdle in realizing precision medicine for gastric cancer (GC). Large-scale whole exome sequencing projects have identified distinct molecular subtypes to help define the heterogeneity of GC. However, it remains unclear whether the targeted gene panel-based sequencing can provide optimum targeted therapies and clinical utility in GC. The aim of this study is to generate comprehensive genomic profiling data and classify Japanese GC into actionable clusters associated with targeted therapies. Methods: FFPE tumor tissues were obtained from surgical or biopsy specimens of 207 Japanese patients with GC. Extracted DNA was subjected to genomic sequencing for 435 cancer related genes including 69 druggable genes with FDA approved targeted therapies. Somatic mutations, copy number alterations (SCNA), microsatellite instability (MSI) and Epstein-Barr virus (EBV) infection were evaluated using sequencing data. Results: Genomic sequencing identified at least one alteration of 435 genes in 194 pts (94%), and that of 69 druggable genes in 141 pts (68%). The most frequently altered druggable gene was ERBB2 (14%), following BRCA2 (11%) and ATM (10%). We successfully classified 207 tumors into four molecular subtypes, similar to the previously report; EBV (4%), MSI (8%), chromosomal instability (58%) and genomically stable subtype (30%). Frequent alterations of druggable genes ( > 5%) were widely observed through these subtypes. To discover the novel classifications associated with targeted therapies, we classified 207 tumors using mutation rate and hierarchical clustering. We identified a hypermutated group (n = 32), and a remaining non-hypermutated group (n = 175) which were sub-divided into six clusters including five actionable ones; ERBB2 (n = 25), CDKN2A and CDKN2B (n = 10), KRAS (n = 10), BRCA2 (n = 9) and ATM cluster (n = 12). Interestingly, we experienced a case of unresectable GC with a remarkable response for anti-HER2 therapy in the ERBB2 cluster. Conclusions: Genomic sequencing using a 435-gene panel has the potential to provide the information of optimum targeted therapies for upcoming precision medicine in Japanese GC.

AIM:Develop and apply a comprehensive and accurate next-generation sequencing based assay to help clinicians to match oncology patients to therapies.MATERIALS & METHODS:The performance of the CANCERPLEX® assay was assessed using DNA from well-characterized routine clinical formalin-fixed paraffin-embedded (FFPE) specimens and cell lines.RESULTS:The maximum sensitivity of the assay is 99.5% and its accuracy is virtually 100% for detecting somatic alterations with an allele fraction of as low as 10%. Clinically actionable variants were identified in 93% of patients (930 of 1000) who underwent testing.CONCLUSION:The test's capacity to determine all of the critical genetic changes, tumor mutation burden, microsatellite instability status and viral associations has important ramifications on clinical decision support strategies, including identification of patients who are likely to benefit from immune checkpoint blockage therapies.

HER2-targeted therapy is considered effective for KRAS codon 12/13 wild-type, HER2-positive metastatic colorectal cancer (CRC). In general, HER2 status is determined by the use of immunohistochemistry (IHC) and fluorescence in situ hybridization (FISH). Comprehensive genomic sequencing (CGS) enables the detection of gene mutations and copy number alterations including KRAS mutation and HER2 amplification; however, little is known about the utility of CGS for detecting HER2-positive CRC. To assess its utility, we retrospectively investigated 201 patients with stage I-IV CRC. The HER2 status of the primary site was assessed using IHC and FISH, and HER2 amplification of the primary site was also assessed using CGS, and the findings of these approaches were compared in each patient. CGS successfully detected alterations in 415 genes including KRAS codon 12/13 mutation and HER2 amplification. Fifty-nine (29%) patients had a KRAS codon 12/13 mutation. Ten (5%) patients were diagnosed as HER2 positive because of HER2 IHC 3+, and the same 10 (5%) patients had HER2 amplification evaluated using CGS. The results of HER2 status and HER2 amplification were completely identical in all 201 patients (P < .001). Nine of the 10 HER2-positive patients were KRAS 12/13 wild-type and were considered possible candidates for HER2-targeted therapy. CGS has the same utility as IHC and FISH for detecting HER2-positive patients who are candidates for HER2-targeted therapy, and facilitates precision medicine and tailor-made treatment.

e15121 Background: The advent of whole genomic sequencing has greatly facilitated the identification and characterization of cancer-relevant genes. Given the potential genetic and environmental differences between ethnic populations, it is unclear if the underlying oncogenic drivers detected by genomic profiling will be consistent across populations. CancerPlex is a large panel genomic test of 400-plus genes associated with cancer. Here we demonstrate alignment of CancerPlex with TCGA database and validate the utility of the platform. Further, we test the hypothesis that the large-scale genomic sequencing platform captures the actionable genomic driver mutations in the Japanese population. Methods: A retrospective analysis of genomic data obtained from 201 archived formalin-fixed, paraffin-embedded samples from Japanese patients with colorectal cancer (CRC) were analyzed by a large next generation sequencing panel of 415 genes including full-gene sequencing for single nucleotide polymorphisms, short insertions and deletions, and copy number variations. Mutations and frequency of variants present in key oncogenic drivers for each patient were quantified and compared to publicly available whole genome sequencing data of 489 colorectal cases in the TCGA database. Testing was performed in a CLIA-certified laboratory (KEW Group Inc., Cambridge, MA). Results: Actionable mutations were identified in > 99% of the Japanese colorectal patients. Sequencing data obtained from CancerPlex identified 60% of Japanese cases with alterations in the RTK/RAS pathways, similar to that determined in CRC cases by TCGA. Of these, 26 Japanese cases had BRAF mutations, in which only 13 harbored the V600E mutation. Additionally, 83% of Japanese cases presented with alterations in the DNA double strand break repair pathway, comparable to data obtained from TCGA demonstrating alterations in 70% cases. Conclusions: Large panel genomic profiling of CRC patients captures the actionable genomic driver mutations in the Japanese population, validating its utility in stratifying patients for therapy. In addition, significant differences in the BRAF mutation profile were detected, supporting the use of full-gene sequencing vs hot-spot testing.

Background: Comprehensive genomic sequencing (CGS) has the potential to revolutionize precision medicine for cancer patients across the globe. However, to date large-scale genomic sequencing of cancer patients has been limited to Western populations. In order to understand possible ethnic and geographic differences and to explore the broader application of CGS to other populations, we sequenced a panel of 415 important cancer genes to characterize clinically actionable genomic driver events in 201 Japanese patients with colorectal cancer (CRC).Methods: Using next-generation sequencing methods, we examined all exons of 415 known cancer genes in Japanese CRC patients (n = 201) and evaluated for concordance among independent data obtained from US patients with CRC (n = 108) and from The Cancer Genome Atlas-CRC whole exome sequencing (WES) database (n = 224). Mutation data from non-hypermutated Japanese CRC patients were extracted and clustered by gene mutation patterns. Two different sets of genes from the 415-gene panel were used for clustering: 61 genes with frequent alteration in CRC and 26 genes that are clinically actionable in CRC.Results: The 415-gene panel is able to identify all of the critical mutations in tumor samples as well as WES, including identifying hypermutated tumors. Although the overall mutation spectrum of the Japanese patients is similar to that of the Western population, we found significant differences in the frequencies of mutations in ERBB2 and BRAF. We show that the 415-gene panel identifies a number of clinically actionable mutations in KRAS, NRAS, and BRAF that are not detected by hot-spot testing. We also discovered that 26% of cases have mutations in genes involved in DNA double strand break repair pathway. Unsupervised clustering revealed that a panel of 26 genes can be used to classify the patients into eight different categories, each of which can optimally be treated with a particular combination therapy.Conclusions: Use of a panel of 415 genes can reliably identify all of the critical mutations in CRC patients and this information of CGS can be used to determine the most optimal treatment for patients of all ethnicities.

Abstract Although durable responses to single agent immune checkpoint inhibitors have been reported, additional approaches are needed to improve upon this therapeutic benefit. Combinations of immunotherapy agents with tumor microenvironment modulators have the potential to overcome barriers that tumor cells develop to evade the immune system, and provide benefit to a greater proportion of patients. Focal Adhesion Kinase (FAK) and its family member, PYK2, are potentially valuable targets due to their roles in regulating key cellular populations in the tumor microenvironment. In addition to targeting cancer stem cells, the FAK/PYK2 dual inhibitors, VS-6063 and VS-4718, have been shown to inhibit monocyte-derived macrophages, reduce tumor-associated macrophages in xenograft models, and promote a CD8+ T cell-mediated anti-tumor response in squamous cell carcinoma models. We now report that the combination of VS-4718 with an anti-PD-1 mAb shows improved efficacy over anti-PD-1 mAb alone and extends survival of MC38 syngeneic tumor bearing animals. Analysis of MC38 tumors at day 12 of treatment revealed a significant increase in the CD8+ T cells/Treg ratios in tumors in the VS-4718 + anti-PD-1 combination group, providing a mechanistic understanding for the enhanced efficacy of this combination. To explore additional combination options, we tested the combination of VS-4718 with anti-4-1BB in the MC38 model. Consistent with what was observed with the anti-PD-1 combination, VS-4718 also enhanced the efficacy of an anti-4-1BB mAb. To further delineate direct effect of FAK inhibition on human T cells, in vitro T cell proliferation assays were conducted. VS-6063 and VS-4718 dose-dependently stimulated proliferation of CD8+ cytotoxic T cells isolated from healthy donors. This is in distinct contrast to other protein kinase inhibitors, such as the SRC inhibitor dasatinib which impaired the proliferation of CD8+ cytotoxic T cells. In addition, both VS-4718 and VS-6063 decreased CD8+ T cell exhaustion markers, and increased T cell-mediated tumor cell killing in vitro. These data provide a rationale for clinical trials in cancer patients to test whether a FAK/PYK2 inhibitor in combination with an immune checkpoint inhibitor could increase the breadth of responsive tumor types, increase the number of responders, and confer more durable anti-tumor responses. Citation Format: Yan Wang, Jennifer E. Ring, Kam Sprott, David T. Weaver, Jonathan A. Pachter. FAK/PYK2 inhibition enhances immune checkpoint inhibitor efficacy. [abstract]. In: Proceedings of the 107th Annual Meeting of the American Association for Cancer Research; 2016 Apr 16-20; New Orleans, LA. Philadelphia (PA): AACR; Cancer Res 2016;76(14 Suppl):Abstract nr 568.