Ashley H. Salazar

Molecular subtypes in head and neck cancer exhibit distinct patterns of chromosomal gain and loss of canonical cancer genes.

Head and neck squamous cell carcinoma (HNSCC) is a frequently fatal heterogeneous disease. Beyond the role of human papilloma virus (HPV), no validated molecular characterization of the disease has been established. Using an integrated genomic analysis and validation methodology we confirm four molecular classes of HNSCC (basal, mesenchymal, atypical, and classical) consistent with signatures established for squamous carcinoma of the lung, including deregulation of the KEAP1/NFE2L2 oxidative stress pathway, differential utilization of the lineage markers SOX2 and TP63, and preference for the oncogenes PIK3CA and EGFR. For potential clinical use the signatures are complimentary to classification by HPV infection status as well as the putative high risk marker CCND1 copy number gain. A molecular etiology for the subtypes is suggested by statistically significant chromosomal gains and losses and differential cell of origin expression patterns. Model systems representative of each of the four subtypes are also presented.

Combined Targeted DNA Sequencing in Non-Small Cell Lung Cancer (NSCLC) Using UNCseq and NGScopy, and RNA Sequencing Using UNCqeR for the Detection of Genetic Aberrations in NSCLC

The recent FDA approval of the MiSeqDx platform provides a unique opportunity to develop targeted next generation sequencing (NGS) panels for human disease, including cancer. We have developed a scalable, targeted panel-based assay termed UNCseq, which involves a NGS panel of over 200 cancer-associated genes and a standardized downstream bioinformatics pipeline for detection of single nucleotide variations (SNV) as well as small insertions and deletions (indel). In addition, we developed a novel algorithm, NGScopy, designed for samples with sparse sequencing coverage to detect large-scale copy number variations (CNV), similar to human SNP Array 6.0 as well as small-scale intragenic CNV. Overall, we applied this assay to 100 snap-frozen lung cancer specimens lacking same-patient germline DNA (07–0120 tissue cohort) and validated our results against Sanger sequencing, SNP Array, and our recently published integrated DNA-seq/RNA-seq assay, UNCqeR, where RNA-seq of same-patient tumor specimens confirmed SNV detected by DNA-seq, if RNA-seq coverage depth was adequate. In addition, we applied the UNCseq assay on an independent lung cancer tumor tissue collection with available same-patient germline DNA (11–1115 tissue cohort) and confirmed mutations using assays performed in a CLIA-certified laboratory. We conclude that UNCseq can identify SNV, indel, and CNV in tumor specimens lacking germline DNA in a cost-efficient fashion.

Enhancing Next‐Generation Sequencing‐Guided Cancer Care Through Cognitive Computing

Next‐generation sequencing (NGS) has emerged as an affordable and reproducible means to query tumors for somatic genetic anomalies. To help interpret somatic NGS data, many institutions have created a molecular tumor board to analyze the results of NGS and make recommendations. This article evaluates the utility of cognitive computing systems to analyze data for clinical decision‐making.

Abstract 4499: Integration of targeted RNA sequencing to targeted DNA sequencing for the characterization of clinically relevant variants in a population of thousands of patients treated on clinical trial

Background: Targeted sequencing has become common in the care of some patients with cancer, both for the detection of standard of care clinical mutations as well as in the care of patients with advanced disease who are looking for clinical trials or other non-standard therapies. Many patients have mutations detected, although many variants are of unknown significance, and many patients have no mutations at all. We added RNA targeted sequencing along with DNA targeted sequencing to add utility of targeted sequencing strategy in cancer genomics and assessed the performances of RNA sequencing analysis. Methods: Genomic sequencing of investigative biomarkers was prospectively offered to selected patients. DNA and RNA libraries were prepared separately from a retrieved archival FFPE tumor sample or a fresh frozen tumor sample from each patient. Relevant targets were enriched by custom designed Agilent SureSelect hybrid capture baits using standard protocols. Samples were sequenced on Illumina HiSeq 2000/2500 platforms. We compared somatic variants detected by DNA alone to those detected by DNA plus RNA using the UNCeqR algorithm and software. Results: From a population of 2200 patients consented as part of LCCC1108 (NCT01457196), a subset of 300 patients was selected at random for RNA profiling. Selected patients were 7 to 83 years old (median, 54) and the cases included more than 20 cancer types including uterus, breast, ovary, and thyroid etc. And stages of cancers were as follows; I, 34.6%, II, 16.8%, III, 25.2%, and IV, 23.4% respectively. Sequencing of RNA was successful in 90% of specimens using 2.5 ug of RNA. RNA sequencing could detect 97.5% of sequence variants called by DNA sequencing and detect 20% more variants than DNA sequencing. Also 97.9% of variants showed higher mutant allele fraction (MAF) in RNA sequencing than DNA sequencing. We further interrogated the impact of certain classes of mutations on transcript structure and abundance. Specifically, we observed that splice site mutations and indels were associated with detectable alterations in the full length transcripts of their respective genes as well as overall gene abundance, with nonsense and frameshift mutations associated with decreased relative expression of the transcript relative to controls. We also observed that gene amplification of EGFR and ERBB2 was reflected in the RNA sequencing as increased gene expression with statistical significance by Fishers’ exact test (P Conclusion: Using clinical samples, including relatively small quantities, we were able to confirm that nucleotide variants detected at DNA level leads to significant alteration at the transcription level and to have additional information potentially helpful for better management of cancer patients. Citation Format: Woochang Lee, Heejoon Jo, Xiaoying Yin, David A. Eberhard, Nirali M. Patel, Michele C. Hayward, Ashley H. Salazar, Joel S. Parker, William Y. Kim, Henry S. Earp, Norman E. Sharpless, David N. Hayes. Integration of targeted RNA sequencing to targeted DNA sequencing for the characterization of clinically relevant variants in a population of thousands of patients treated on clinical trial. [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 4499.

Abstract 5334: Integrated analysis of squamous tumors identifies novel targets and dissects gene regulation

Proceedings: AACR Annual Meeting 2014; April 5-9, 2014; San Diego, CA Background: Studies by The Cancer Genome Atlas (TCGA) and others have identified regions of somatic copy number alteration (SCNA) in head and neck squamous cell carcinoma (HNSCC) and lung squamous cell carcinoma (LUSC). Both tumor types exhibit frequent SCNAs in chromosomes 3q, 9p, and 11q. Although CDKN2A is the putative target of 9p deletions, numerous targets have been proposed in the 3q amplicon, including SOX2, PIK3CA, and TP63. Because expression levels are affected by underlying genomic events, we hypothesized that an integrated analysis of multiple genomic data types would provide increased ability to identify target genes in SCNA regions when compared to methods based on copy number alone while providing insight into mechanisms regulating expression. Findings may be subsequently validated by examining multiple tumor types or performing training/testing analyses. Techniques: Gene expression (GE), DNA copy number (CN), DNA methylation (ME), and microRNA expression (miR) data were obtained from the TCGA studies of HNSCC and LUSC. Linear models were constructed for each gene to investigate the effect of changes in CN, ME, and miR on GE. Analysis of model output provided an approach to identify target genes in SCNA regions and assess the effect of genomic alterations on expression. Results: Genome-wide GE, CN, ME, and miR data were available for n = 279 (HNSCC) and n = 159 (LUSC) tumor samples. Univariate modeling detected a strong overall association between GE and CN, as measured by the coefficient of determination (model R2). Although some genes found in SCNA studies produced large model R2— e.g. SOX2— a number of genes with large model R2 are relatively unknown. Notable examples include DVL3, and SENP2, which were implicated as driver genes in the 3q amplicon by a recent study of LUSC but not identified in the TCGA report. Remarkably, the model R2 for DVL3 and SENP2 was also high in HNSCC, and such a finding in a distinct tumor type provides validation and merits additional study. Output from models additionally including ME and miR as covariates contributes insight into the diversity of regulation of gene expression, but the strength of the association between GE and CN could mask other effects. Thus we constructed linear models in which the response variable was the residuals from the GE/CN model and the covariates were ME or miR. The TCGA study of glioblastoma noted frequent methylation of MGMT. When modeling MGMT we detected a highly significant association between ME and the GE/CN residuals in both HNSCC and LUSC. These results illustrate a pronounced effect of ME on GE conditional on CN. Conclusion: Linear modeling techniques provide a flexible and powerful basis for performing integrated analysis of genomic data. Our approach produces predicted results when analyzing known cancer genes, highlights lesser known genes for future study, provides insight into gene regulation, and draws attention to genes relevant in multiple tumor types. Citation Format: Vonn Walter, Ying Du, Xiaoying Yin, Wei Sun, Matthew D. Wilkerson, Michele C. Hayward, Ashley H. Salazar, Charles M. Perou, David N. Hayes. Integrated analysis of squamous tumors identifies novel targets and dissects gene regulation. [abstract]. In: Proceedings of the 105th Annual Meeting of the American Association for Cancer Research; 2014 Apr 5-9; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2014;74(19 Suppl):Abstract nr 5334. doi:10.1158/1538-7445.AM2014-5334

Abstract 5598: Correlating molecular and histopathologic tumor purity: An analysis of 816 patients

Proceedings: AACR Annual Meeting 2014; April 5-9, 2014; San Diego, CA Background: Clinical cancer samples frequently have significant contamination by stromal cells that increase the difficulty of mutation calling. Histopathologic light microscopy (LM) estimation of purity is commonly used to qualify samples for molecular testing. With large datasets generated by current technologies, more accurate estimations of purity may be facilitated by use of SNP chip (SNP) or sequencing (seq) data. We present a computational framework that considers a variety of technical factors, including percent (%) tumor, seq quality, and mutation type that impact the detection and evaluation of somatic mutations reported to patients. Methods: LCCC1108 (“The development of a tumor molecular analyses program and its uses to support treatment decisions”) is a registry study for cancer patients that performs targeted sequencing with a panel of >200 clinically actionable genes for clinical use. A subset of patients have additional analyses performed including DNA SNP and RNA seq. Any clinically actionable variant is orthogonally confirmed in a CLIA-certified lab. Results: As of 12/1/2013, 816 patients have enrolled. 350 cases have both molecular and LM estimates of tumor purity. Agreement between molecular estimates of % tumor and LM review was low (Pearsons correlation coefficient 0.16 for the most favorable comparison). No clear etiology could explain the low agreement, although examination of extreme cases in which LM approached 100% tumor and molecular estimates approached zero generally included tumors with low rates of molecular alterations (e.g. adenoid cystic carcinoma, chronic myelogenous leukemia). Additionally, independent assessment of % tumor by molecular analysis such as SNP vs. seq also had low agreement (0.13), although a clear pattern could be seen in the disagreement for certain cases. Methods that rely heavily on copy number(CN) alterations often under-predicted estimates of % tumor relative to methods that relied on the mutant allele fraction (MAF) in cases where there were few CN alterations but where point mutations predominated. Excluding these outliers, the agreement was high and molecular estimates of % tumor more reproducible. Most molecular tumor estimates performed more poorly or failed as the % stromal infiltration rose above 50%. Seq estimates of % tumor tended to be noisier for coverage <200x. % tumor as a function of the MAF of actionable mutations was useful in cases where a canonical mutation (e.g. p53) could be used to estimate % tumor as well as homozygous vs heterozygous mutations of actionable genes. Conclusion: In clinical practice, accurate estimates of % tumor spanning the clinical spectrum are critical for somatic mutation calling. Pure analytic approaches to estimates of % tumor may fail in tumors with low rates of somatic mutation and CN alteration. Ultimately, a combination of methods will be required to describe samples for optimal variant detection. Citation Format: Heejoon Jo, Nirali M. Patel, Karen E. Weck, David A. Eberhard, D. Neil Hayes, Joel S. Parker, Michele Hayward, Ashley Salazar, Juneko E. Grilley-Olson. Correlating molecular and histopathologic tumor purity: An analysis of 816 patients. [abstract]. In: Proceedings of the 105th Annual Meeting of the American Association for Cancer Research; 2014 Apr 5-9; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2014;74(19 Suppl):Abstract nr 5598. doi:10.1158/1538-7445.AM2014-5598

Abstract 2016: MicroRNA expression subtypes in head and neck squamous cell carcinoma.

Proceedings: AACR 104th Annual Meeting 2013; Apr 6-10, 2013; Washington, DC Background: Head and neck squamous cell carcinoma (HNSCC) is a heterogeneous disease whose underlying etiology is unknown. Gene expression (GE) subtypes were detected in HNSCC by Chung et al. (Cancer Cell 2004) and validated in a recent study by our group. However, to the best of our knowledge no corresponding investigation of microRNA (miRNA) expression subtypes in HNSCC has been performed, as previous studies compared miRNA expression patterns in tumor and normal samples. Techniques: After receiving informed consent, we obtained tumor samples from an incident surgical sample of HNSCC patients at the University of North Carolina Hospital. miRNA and GE assays were performed using Affymetrix GeneChip miRNA 1.0 and Agilent 44K Gene Expression platforms, respectively. miRNA expression subtypes were discovered using ConsensusClusterPlus, statistical significance of miRNA expression patterns was assessed with SigClust, SAM was used to identify differentially expressed miRNAs, and the miRBase and DAVID databases were used to identify target genes and enriched pathways. Results: After applying quality control procedures, miRNA expression data was available for 103 patients and 830 hsa-miRNAs, while GE data was available for 138 patients and 15,597 genes. Both miRNA and GE data was available for 75 patients. Clinical data was available for all patients. Three miRNA expression subtypes were detected, and all pairwise comparisons of the expression patterns in the subtypes were statistically significant after performing a Bonferroni correction. The association between GE and miRNA expression subtypes was not statistically significant. Differentially expressed miRNAs were identified when expression in each subtype was compared to the remaining subtypes. Examples include miR-15a, which was underexpressed in one subtype and has been shown act as a tumor suppressor miRNA, as well as both miR-155 and miR-21, which were overexpressed in a different subtype and have been shown to have oncogenic effects. Target genes corresponding to each collection of differentially expressed miRNAs were also identified, and results from the DAVID functional annotation database suggest that a number of cancer-related pathways are enriched for each set of target genes. Pearson correlation coefficients were computed for expression of each differentially expressed miRNA and the expression of its target genes. Several statistically significant negative miR-target gene correlations were found after adjusting for multiple comparisons, including miR-15b and SOX5, miR-520e and VEGFA, as well as miR-98 and FRAS1. Conclusion: miRNA expression subtypes of HNSCC provide a valuable method for categorizing tumors that complements known GE expression subtypes. Moreover, the subtypes identify known and novel miRs with tumor suppressor and oncogenic effects via their interaction with target genes. Citation Format: Vonn Walter, Neil Hayes, Ying Du, Matthew D. Wilkerson, Michele Hayward, Ashley Salazar, Xiaoying Yin. MicroRNA expression subtypes in head and neck squamous cell carcinoma. [abstract]. In: Proceedings of the 104th Annual Meeting of the American Association for Cancer Research; 2013 Apr 6-10; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2013;73(8 Suppl):Abstract nr 2016. doi:10.1158/1538-7445.AM2013-2016