Wang, Chen

Abstract Purpose: Gene expression–based molecular subtypes of high-grade serous tubo-ovarian cancer (HGSOC), demonstrated across multiple studies, may provide improved stratification for molecularly targeted trials. However, evaluation of clinical utility has been hindered by nonstandardized methods, which are not applicable in a clinical setting. We sought to generate a clinical grade minimal gene set assay for classification of individual tumor specimens into HGSOC subtypes and confirm previously reported subtype-associated features. Experimental Design: Adopting two independent approaches, we derived and internally validated algorithms for subtype prediction using published gene expression data from 1,650 tumors. We applied resulting models to NanoString data on 3,829 HGSOCs from the Ovarian Tumor Tissue Analysis consortium. We further developed, confirmed, and validated a reduced, minimal gene set predictor, with methods suitable for a single-patient setting. Results: Gene expression data were used to derive the predictor of high-grade serous ovarian carcinoma molecular subtype (PrOTYPE) assay. We established a de facto standard as a consensus of two parallel approaches. PrOTYPE subtypes are significantly associated with age, stage, residual disease, tumor-infiltrating lymphocytes, and outcome. The locked-down clinical grade PrOTYPE test includes a model with 55 genes that predicted gene expression subtype with >95% accuracy that was maintained in all analytic and biological validations. Conclusions: We validated the PrOTYPE assay following the Institute of Medicine guidelines for the development of omics-based tests. This fully defined and locked-down clinical grade assay will enable trial design with molecular subtype stratification and allow for objective assessment of the predictive value of HGSOC molecular subtypes in precision medicine applications. See related commentary by McMullen et al., p. 5271

PURPOSE The Fast Healthcare Interoperability Resources (FHIR) is emerging as a next-generation standards framework developed by HL7 for exchanging electronic health care data. The modeling capability of FHIR in standardizing cancer data has been gaining increasing attention by the cancer research informatics community. However, few studies have been conducted to examine the capability of FHIR in electronic data capture (EDC) applications for effective cancer clinical trials. The objective of this study was to design, develop, and evaluate an FHIR-based method that enables the automation of the case report forms (CRFs) population for cancer clinical trials using real-world electronic health records (EHRs). MATERIALS AND METHODS We developed an FHIR-based computational pipeline of EDC with a case study for modeling colorectal cancer trials. We first leveraged an existing FHIR-based cancer profile to represent EHR data of patients with colorectal cancer, and then we used the FHIR Questionnaire and QuestionnaireResponse resources to represent the CRFs and their data population. To test the accuracy of and overall quality of the computational pipeline, we used synoptic reports of 287 Mayo Clinic patients with colorectal cancer from 2013 to 2019 with standard measures of precision, recall, and F1 score. RESULTS Using the computational pipeline, a total of 1,037 synoptic reports were successfully converted as the instances of the FHIR-based cancer profile. The average accuracy for converting all data elements (excluding tumor perforation) of the cancer profile was 0.99, using 200 randomly selected records. The average F1 score for populating nine questions of the CRFs in a real-world colorectal cancer trial was 0.95, using 100 randomly selected records. CONCLUSION We demonstrated that it is feasible to populate CRFs with EHR data in an automated manner with satisfactory performance. The outcome of the study provides helpful insight into future directions in implementing FHIR-based EDC applications for modern cancer clinical trials.

Abstract Background: Traditionally, mutational burden and mutational signatures have been assessed by tumor-normal pair DNA sequencing. The requirement of having both normal and tumor samples is not always feasible from a clinical perspective, and led us to investigate the efficacy of using RNA sequencing of only the tumor sample to determine the mutational burden and signatures, and subsequently molecular cause of the cancer. The potential advantages include reducing the cost of testing, and simultaneously providing information on the gene expression profile and gene fusions present in the tumor.Results: In this study, we devised supervised and unsupervised learning methods to determine mutational signatures from tumor RNA-seq data. As applications, we applied the methods to a training set of 587 TCGA uterine cancer RNAseq samples, and examined in an independent testing set of xxx TCGA colorectal cancer RNAseq samples. Both diseases are known associated with microsatellite instable high (MSI-H) and driver defects in DNA polymerase ɛ (POLɛ). From RNAseq called variants, we found majority (>95%) are likely germline variants, leading to C>T enriched germline variants (dbSNP) widely applicable in tumor and normal RNAseq samples. We found significant associations between RNA-derived mutational burdens and MSI/POLɛ status, and insignificant relationship between RNAseq total coverage and derived mutational burdens. Additionally we found that over 80% of variants could be explained by using the COSMIC mutational signature-5, -6 and -10, which are implicated in natural aging, MSI-H, and POLɛ, respectively. For classifying tumor type, within UCEC we achieved a recall of 0.56 and 0.78, and specificity of 0.66 and 0.99 for MSI and POLɛ respectively. By applying learnt RNA signatures from UCEC to COAD, we were able to improve our classification of both MSI and POLɛ. Conclusions: Taken together, our work provides a novel method to detect RNAseq derived mutational signatures with effective procedures to remove likely germline variants. It can leads to accurate classification of underlying driving mechanisms of DNA damage deficiency.

Abstract Objective To design, develop, and evaluate a scalable clinical data normalization pipeline for standardizing unstructured electronic health record (EHR) data leveraging the HL7 Fast Healthcare Interoperability Resources (FHIR) specification. Methods We established an FHIR-based clinical data normalization pipeline known as NLP2FHIR that mainly comprises: (1) a module for a core natural language processing (NLP) engine with an FHIR-based type system; (2) a module for integrating structured data; and (3) a module for content normalization. We evaluated the FHIR modeling capability focusing on core clinical resources such as Condition, Procedure, MedicationStatement (including Medication), and FamilyMemberHistory using Mayo Clinic’s unstructured EHR data. We constructed a gold standard reusing annotation corpora from previous NLP projects. Results A total of 30 mapping rules, 62 normalization rules, and 11 NLP-specific FHIR extensions were created and implemented in the NLP2FHIR pipeline. The elements that need to integrate structured data from each clinical resource were identified. The performance of unstructured data modeling achieved F scores ranging from 0.69 to 0.99 for various FHIR element representations (0.69–0.99 for Condition; 0.75–0.84 for Procedure; 0.71–0.99 for MedicationStatement; and 0.75–0.95 for FamilyMemberHistory). Conclusion We demonstrated that the NLP2FHIR pipeline is feasible for modeling unstructured EHR data and integrating structured elements into the model. The outcomes of this work provide standards-based tools of clinical data normalization that is indispensable for enabling portable EHR-driven phenotyping and large-scale data analytics, as well as useful insights for future developments of the FHIR specifications with regard to handling unstructured clinical data.

Abstract While extensive, the genetic studies of DLBCL tumors to date have primarily focused on somatic mutations that are associated with development of lymphoma, i.e., driver mutations. Identification of key genetic events that impact and/or predict response to therapy and clinical outcome of lymphoma patients has not been accomplished. In an exploratory study, we used whole-exome sequencing to identify novel biomarkers that can predict which patients have an aggressive form of DLBCL, and therefore would likely benefit from a different and more aggressive treatment plan. Using whole-exome sequencing data from 54 newly diagnosed DLBCL patients, we evaluated the association of somatic coding single nucleotide (cSNV) and copy number (CNV) variants with aggressive DLBCL. All patients were treated with R-CHOP or immunochemotherapy, and disease aggressiveness was based on relapse, with patients classified as having aggressive disease (AD) (n=13, defined as treatment failure, relapse, or progression within 24 months of diagnosis) versus non-aggressive disease (n=41, defined as achieving at least 24 months of relapse-free survival). An in-house algorithm, patternCNV was used to detect somatic CNVs. Logistic regression was used to assess statistical significance. In the cSNV analysis, 24 genes were found to be significantly (p ≤ 0.05) associated with AD. Of these genes, CIITA (p=0.01) was previously identified as a mutational target in DLBCL and is a recurrent gene fusion partner in lymphoid cancers. Metacore analyses showed that seven of these genes, including CIITA, can be placed in the same regulatory network centered around CREB1. Interestingly, the CREB binding protein (CREBBP) is a known target of pathogenic mutations in DLBCL. In the CNV analyses, we identified 245 gene amplifications and 209 gene deletions associated with AD (p ≤ 0.05). Deleted genes were localized in three major chromosomal regions, 6p22.3-6p22.1 (26 genes), 6q13-6q16.3 (31 genes), and 6q21-6q24.2 (86 genes). In the CNV amplification analysis, the genes were localized in two major chromosomal regions, 3q12.3-3q29 (90 genes) and 19q13.12-19q13.43 (122 genes). We also quantified the association of each regional deletion/amplification with patient outcome. As expected, each deleted region was univariately associated with time to relapse: 6p22.3-6p22.1 (HR = 7.19, 95% CI: 2.66-19.46, p=1.33x10-5), 6q13-6q16.3 (HR = 4.44, 95% CI: 1.74-11.32, p=6.84x10-4), and 6q21-6q24.2 (HR = 5.42, 95% CI: 2.12-13.87, p=9.23x10-5) as well as the two amplification regions: 3q12.3-3q29 (HR = 4.60, 95% CI: 1.49-14.24, p=4.27x10-3) and 19q13.12 -19q13 (HR = 13.79, 95% CI: 4.39-43.34, p=2.81x10-8). Deletions at 6q21-6q24.2 have been previously reported in DLBCL and two tumor suppressor genes associated with lymphoma pathogenesis are localized in the region, TNFAIP3 (p=0.14) and PRDM1 (p=0.02), the latter of which was significantly associated with AD. However, the genes in the 6p21 locus that were most significant (p=0.002) were SLC22A16, GPR6, SOBP, MATTL24, DDO, and REV3L. SLC22A16, which also had cSNVs in two tumors, is an organic cation transporter that has been shown to transport chemotherapeutic drugs including doxorubicin. Successful drug response has been correlated with the level of activity and expression of this transporter in tumor cells. To determine if expression of SLC22A16 would have an impact on doxorubicin transport, and may therefore impact response to R-CHOP we overexpressed either an empty vector control or SLC22A16 in OCI-Ly7 DLBCL cells, which do not express endogenous SLC22A16 mRNA. We found that OCI-Ly7 cells expressing SLC22A16 have increased 14C-doxorubicin uptake and are more sensitive to doxorubicin-induced cell death when compared to the empty vector control cells. These data suggest that loss of SLC22A16 expression through gene deletion could impact the ability of DLBCL cells to respond to therapy and indicates an aggressive form of the disease. In summary, our data are the first to use whole exome sequencing to identify genetic variants associated with aggressive DLBCL, which will require replication in additional samples. Nevertheless, our data highlight the role for somatic CNVs in patient outcome and we biologically validate the potential impact of deletions in SLC22A16 at 6p21. Disclosures: No relevant conflicts of interest to declare.

Abstract Background: Archived formalin fixed paraffin embedded (FFPE) samples are valuable clinical resources to examine clinically relevant morphology features and also to study genetic changes. However, DNA quality and quantity of FFPE samples are sub-optimal, and resulting NGS-based genetics variant detections are prone to false positives. Evaluations of wet-lab and bioinformatics approaches are needed to optimize variant detection from FFPE samples. Results: As a pilot study, we designed within-subject triplicate samples of DNA derived from paired FFPE and fresh frozen breast tissues to highlight FFPE-specific artifacts. For FFPE samples, we tested two FFPE DNA extraction methods to determine impact of wet-lab procedures on variant calling: QIAGEN QIAamp DNA Mini Kit ("QA"), and QIAGEN GeneRead DNA FFPE Kit ("QGR"). We also used negative-control (NA12891) and positive control samples (Horizon Discovery Reference Standard FFPE). All DNA sample libraries were prepared for NGS according to the QIAseq Human Breast Cancer Targeted DNA Panel protocol and sequenced on the HiSeq 4000. Variant calling and filtering were performed using QIAGEN Gene Globe Data Portal. Detailed variant concordance comparisons and mutational signature analysis were performed to investigate effects of FFPE samples compared to paired fresh frozen samples, along with different DNA extraction methods. In this study, we found that five times or more variants were called with FFPE samples, compared to their paired fresh-frozen tissue samples even after applying molecular barcoding error-correction and default bioinformatics filtering recommended by the vendor. We also found that QGR as an optimized FFPE-DNA extraction approach leads to much fewer discordant variants between paired fresh frozen and FFPE samples. Approximately 92 % of the uniquely called FFPE variants were of low allelic frequency range (<5%), and collectively shared a “C>T|G>A” mutational signature known to be representative of FFPE artifacts resulting from cytosine deamination. Based on control samples and FFPE-frozen replicates, we derived an effective filtering strategy with associated empirical false-discovery estimates. Conclusions: Through this study, we demonstrated feasibility of calling and filtering genetic variants from FFPE tissue samples using a combined strategy with molecular barcodes, optimized DNA extraction, and bioinformatics methods incorporating genomics context such as mutational signature and variant allelic frequency.

Abstract Tumors arise and evolve by iterative steps of mutation, subclonal selection, and clonal expansion due to growth advantage of the fittest subclones and external mutation induction and selection pressure from radiation and drug therapies. Various studies have shown that greater clonal complexities in primary tumors are correlated with poor clinical outcome of tumor progression and/or drug resistance. Recently, we have reported that very few of the driver mutations, all of which were clonal, detected in primary diffuse large B-cell lymphoma (DLBCL) were associated with DLBCL outcome as measured by 24-month event-free survival (EFS24). In the current project, we identified and studied subclonal mutations in primary DLBCL tumors and assessed their associations with EFS24, as well as the role of activation-induced cytidine deaminase (AID) in genomic mutations occurred in DLBCL tumors. The detection of subclonal mutations is still a significant bioinformatics challenge. In addition, identification of clonal mutations in samples with lower tumor purities faces similar challenges due to the low concentration of reads supporting the mutant alleles. The current methods of somatic mutation calling are not sufficiently sensitive to identify the low concentration mutations in tumor DNA sequencing data. We implemented a bioinformatics workflow for low-concentration and subclonal mutation detection which is based on the positional read pile-up data and a back-fill approach (PUB). The pile-up files were first generated using the re-aligned and re-calibrated BAM files after read alignment using Burrows-Wheeler Aligner (BWA). The variant positions with alternative bases were annotated by attributes defined in the Variant Quality Score Recalibration (VQSR). PUB then used a boosting method and a generalized linear model (GLM) to train a model of 'good quality' variants using common variants from HapMap, and prioritized and called clonal and subclonal variants based on the trained model. The VQSR attributes related to alternative allele depth were less-weighed in order to call subclonal mutations. The somatic mutations were then identified by Fisher's Exact Test using sequencing depths of the reference and alternative alleles from paired tumor and germline sequencing data. The exome sequencing data of paired tumor and peripheral blood from 48 newly diagnosed DLBCL patients were analyzed using PUB. Thirty-six of the 48 patients achieved EFS24 with the other 12 patients experienced primary treatment failure. Most of the tumors studied had tumor purities between 40-70%. PUB identified substantially higher number of somatic mutations, both clonal and subclonal, compared to those detected using existing somatic callers. We observed that the prevalence of mutations in previously reported driver genes were higher using thresholds of mutation concentration ≥ 5% and Fisher's Exact Test p ≤ 0.05, including EZH2 (mutated in 26% of DLBCL tumors analyzed by PUB, compared to 12.7% as previously reported), MLL2 (72% vs. 31%), CD79B (22% vs. 14%), TNFRSF14 (36% vs. 20%), MEF2B (22% vs. 16.4%), CARD11 (46% vs. 21.8%), and MYD88 (18% vs. 11%). In addition, other genes involved tumorigenesis that have not been previously linked to DLBCL also harbored both clonal and subclonal mutations with substantial prevalence, including FGFR3 (12%), KIT (24%), and ATM (28%). Furthermore, association of genes displaying clonal and subclonal mutations with EFS24 identified potential biomarkers for DLBCL outcome. Among these, two genes EPGN and ASTE are involved in epithelial growth factor receptor signaling and had association p values of 0.0001 and 0.0016, respectively. The oncogene MAFB was also associated with EFS24 (p= 0.0016). We searched for AID induced mutations among all identified variant positions and concluded that there was no evidence of AID site enrichments compared to a simulated data set. In summary, we developed and applied a sensitive bioinformatics pipeline for the identification of both clonal and low concentration somatic mutations in primary DLBCL exome sequencing data which further revealed the clonal complexity of the primary DLBCL tumors. Several of the genes were identified as potential biomarkers for DLBCL outcome. Disclosures Maurer: Kite Pharma: Research Funding. Ansell:Bristol-Myers Squibb: Research Funding; Celldex: Research Funding. Link:Genentech: Consultancy, Research Funding; Kite Pharma: Research Funding. Cerhan:Kite Pharma: Research Funding.

Abstract Purpose: Predicting surgical outcome could improve individualizing treatment strategies for patients with advanced ovarian cancer. It has been suggested earlier that gene expression signatures (GES) might harbor the potential to predict surgical outcome. Experimental Design: Data derived from high-grade serous tumor tissue of FIGO stage IIIC/IV patients of AGO-OVAR11 trial were used to generate a transcriptome profiling. Previously identified molecular signatures were tested. A theoretical model was implemented to evaluate the impact of medically associated factors for residual disease (RD) on the performance of GES that predicts RD status. Results: A total of 266 patients met inclusion criteria, of those, 39.1% underwent complete resection. Previously reported GES did not predict RD in this cohort. Similarly, The Cancer Genome Atlas molecular subtypes, an independent de novo signature and the total gene expression dataset using all 21,000 genes were not able to predict RD status. Medical reasons for RD were identified as potential limiting factors that impact the ability to use GES to predict RD. In a center with high complete resection rates, a GES which would perfectly predict tumor biological RD would have a performance of only AUC 0.83, due to reasons other than tumor biology. Conclusions: Previously identified GES cannot be generalized. Medically associated factors for RD may be the main obstacle to predict surgical outcome in an all-comer population of patients with advanced ovarian cancer. If biomarkers derived from tumor tissue are used to predict outcome of patients with cancer, selection bias should be focused on to prevent overestimation of the power of such a biomarker. See related commentary by Handley and Sood, p. 9

Abstract Background Archived formalin fixed paraffin embedded (FFPE) samples are valuable clinical resources to examine clinically relevant morphology features and also to study genetic changes. However, DNA quality and quantity of FFPE samples are often sub-optimal, and resulting NGS-based genetics variant detections are prone to false positives. Evaluations of wet-lab and bioinformatics approaches are needed to optimize variant detection from FFPE samples. Results As a pilot study, we designed within-subject triplicate samples of DNA derived from paired FFPE and fresh frozen breast tissues to highlight FFPE-specific artifacts. For FFPE samples, we tested two FFPE DNA extraction methods to determine impact of wet-lab procedures on variant calling: QIAGEN QIAamp DNA Mini Kit ("QA"), and QIAGEN GeneRead DNA FFPE Kit ("QGR"). We also used negative-control (NA12891) and positive control samples (Horizon Discovery Reference Standard FFPE). All DNA sample libraries were prepared for NGS according to the QIAseq Human Breast Cancer Targeted DNA Panel protocol and sequenced on the HiSeq 4000. Variant calling and filtering were performed using QIAGEN Gene Globe Data Portal. Detailed variant concordance comparisons and mutational signature analysis were performed to investigate effects of FFPE samples compared to paired fresh frozen samples, along with different library preparations. In this study, we found that five times or more variants were called with FFPE samples, compared to their paired fresh-frozen tissue samples even after applying molecular barcoding error-correction and default bioinformatics filtering recommended by the vendor. We also found that QGR as an optimized FFPE-DNA extraction approach leads to much fewer discordant variants between paired fresh frozen and FFPE samples. Approximately 92% of the uniquely called FFPE variants were of low allelic frequency range (<5%), and collectively shared a “C>T|G>A” mutational signature known to be representative of FFPE artifacts resulting fromcytosine deamination. Based on control samples and FFPE-frozen replicates, we derived an effective filtering strategy with associated empirical false-discovery estimates. Conclusions Through this study, we demonstrated feasibility of calling and filtering genetic variants from FFPE tissue samples using a combined strategy with molecular barcodes, optimized DNA extraction, and bioinformatics methods incorporating genomics context such as mutational signature and variant allelic frequency.