Metagenomic Assay for Identification of Microbial Pathogens in Tumor Tissues

Baldwin, Don A.; Feldman, Michael; Alwine, James C.; Robertson, Erle S.

The screen was performed on an initial set of eight oropharyn-geal squamous cell carcinoma (OSCC)/head and neck carcinoma samples from FFPE tissue specimens. Human p16 overexpression from the CDKN2A gene is correlated with oncogenic human papillomavirus (HPV) infection in OSCC, and p16 immunohistochemistry is used as a prognostic molecular biomarker in clinical pathology laboratories, with high sensitivity but poor specificity for HPV (27) . The OSCC samples included five p16-positive tumors and three p16-negative tumors as determined by pathology at the Hospital of the University of Pennsylvania. From our PathoChip screen, four of the five p16(Ï©) tumors produced high detection signals across the 68 PathoChip probes for HPV16, and the fifth tumor showed signals for a small subset of HPV16 probes. The remaining tumors were negative for PathoChip HPV16 detection ( Fig. 3) . Despite good hybridization to other p16(Ï©) samples, three of the HPV probes for tumor 2025 and two probes for tumor 2028 had no detectable signal This is suggestive of an HPV strain variation, which was similar to the results from polymorphic sites in the CMV positive-control experiment. Development of PathoChip analysis strategies using OSCC tumor screening data. Oncogenic viruses may undergo significant genomic rearrangements or deletions in host tumors. Furthermore, viral strains can be widely polymorphic, and detection of a new pathogen may rely on signal from a single probe. Several levels of data analysis are therefore needed to detect three main classes of "hits" that might be expected in a screening project (Fig. 4) . Accession signal (AccSig), the average of all probes for an accession adjusted for human DNA cross-hybridization, was calculated to screen for detection by a majority of probes in an accession's set. MAT (model-based analysis of tiling arrays) scores (28) from a sliding window of probes were calculated to detect local areas of high signal regardless of accession boundaries. t tests with multiple testing correction were employed at the individual probe level to identify probes with signal consistently higher than back- from the NCBI DNA sequence databases and concatenated to form a metagenome. Wherever possible, regions of target sequence unique to the accession (a and c) were used to select multiple 60-nt probes (1, 2, and 4 to 6) for microarray synthesis, and probes to target regions that share similar sequences in at least two viral accessions (b) were also identified. Probes to prokaryotic and eukaryotic pathogens may map to intergenic, gene, or rRNA sequences or a mixture of target types, depending on the availability of sequence data. (B) Parallel and iterative design processes were used to assemble the PathoChip probe collection that covers unique and conserved target regions, supplemented with high-resolution probe tiling for known cancer-associated microorganisms. (C) The PathoChip tumor screening protocol simultaneously assays DNA and RNA from small amounts of tissue recovered from formalin-fixed, paraffin-embedded (FFPE) tumor specimens. (1), low signal due to sequence polymorphisms (2, 3, and 4), and missing signal due to deletion in AD169 (5 and 6).