VirMAP for cancer: Characterization of the intratumoral virome in virally associated cancers and a resource for investigators.

Abstract

2568 Background: A major barrier to understanding the role of viruses in cancer progression has been the lack of sensitivity of current tools to assemble viral genomes in an untargeted way. Here, we utilize our novel VirMAP algorithm to identify eukaryotic and prokaryotic viruses associated with cancers, using RNA sequencing data from The Cancer Genome Atlas (TCGA). Methods: VirMAP was run on 16 TCGA RNASeq datasets of primary tumor tissues, with a focus on cancers with known viral associations (including cervical, head and neck, liver, stomach, esophagus and ovarian). Previously published survival data were used for exploratory analysis. Overall survival (OS) was assessed using the Kaplan-Meier method, and p-values were calculated using log-rank test. Multivariate analysis (MVA) was done using Cox proportional hazards model to control for age, sex, stage, and histology. Results: Viral sequence recovery (among 537 billion sequencing reads from 6185 patients) was at least equal to previously published algorithms such as Kraken, but with higher taxonomic resolution; overall recovery was 0.0041% from 16 tumor types. Cancers with high viral loads such as cervical cancer demonstrated high recovery rates (0.012% of all reads mapped to viral taxa, followed by 0.004% for liver, 0.0018% head and neck, 0.0016% stomach, 0.0009% esophageal, and 0.0002% ovarian). There was a wide range of viral signature richness, with cervical cancer showing the highest viral read count predominated by 18 different HPV serotypes; in contrast, stomach cancer had lower abundance but contained over 300 unique viral species. Exploratory analysis revealed novel correlations between viral signatures and OS. In cervical cancer, presence of HPV45 predicted worse survival (median OS 837 days vs 4086, p=0.0043); MVA demonstrated hazard ratio (HR) of 5.7 [2.6-11.4, p<0.0001]. In stomach cancer, patients often had more than one viral taxon present (mean 3.7, range 0-18); presence of seven or more unique taxa predicted worse OS (median OS 403 days vs 1153, p=0.0001). MVA showed HR of 1.1 for every additional taxon (p=0.0004). Across all 16 tumor types, the detection of phage virus was associated with worse OS (1073 days vs. 2094 days for non-phage virus and 2131 days for no virus, p<0.0001). Conclusions: VirMAP performs deep characterization of the tumor-associated virome in both high and low abundance settings, with generation of partial genome reconstructions and strain level taxonomic classification including phages. Further work is ongoing to optimize viral read normalization based on genome length, as well as subset analysis with disease-specific modifiers such as histology, race, and exposures. Utilizing publicly available platforms such as TCGA, clinical and correlative published data can be leveraged to better understand the nuanced pleotropic effects of the tumor virome on cancer progression.