Metagenomic Assay for Identification of Microbial Pathogens in Tumor Tissues

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

ground across the population of tumors, and an outlier analysis was conducted for probes with high signal but only in one or a few tumors from the screening population. Data from a screening project of 100 OSCC tumors were used to evaluate these analysis methods. AccSig for HPV16 was consistent with p16 pathology reports ( Fig. 5A ; see also Table S1 in the supplemental material), with 80% of p16(Ï©) tumors producing an AccSig value of more than 100. Of the eight p16(Ï©) tumors with low or no HPV16 AccSig, four showed high signals for a subset of HPV16 probes or produced significant AccSig values for HPV26 or HPV92. The sliding window analysis recapitulated AccSig results and highlighted the differences between detection events for full or partial HPV16 genomes. In Fig. 5B , metagenome regions with a MAT score of more than 3,000 were compiled for each sample, and the individual probes within each region were ordered by map position in a plot of probe signals. This analysis detected a number of other organisms, including pathogenic oral bacteria, although the signals were lower than those of the HPV genomes. These preliminary candidates will be investigated using confirmatory PCR and sequencing methods. Analyses at the individual probe level also demonstrated utility for identifying candidates. A large majority of HPV16 probes passed a t test significance threshold for detection signals which were greater than background across the tumor population (Table 3) , as would be expected for a genome that is so common in OSCC. Many HPV16 probes also passed the outlier test, indicating that although the signals are consistently different from background, the population's range of intensities is wide and therefore also contains outliers. In contrast, fewer HPV18 and HPV26 probes were significant by t test, reflecting the much lower apparent occurrence of these genomes in this tumor population (Table 3) . However, the outlier analysis easily identified the relatively larger number of probes that produced HPV18 or HPV26 detections by AccSig or MAT score in a few positive samples. For these rarer candidates, some probes were significant by t test because they produced lower but consistent signals over background throughout the population, which may be an account of copy number of genomes present and is not surprising. This also illustrates the need to examine probe-level hybridization intensities, not just to analyze algorithm output scores, when considering candidates for follow-up validation, regardless of the method used for their initial identification.