Metagenomic Assay for Identification of Microbial Pathogens in Tumor Tissues

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

The ability of a highly multiplexed, metagenomic assay to detect small nonhuman genomes in an overwhelming background of human sequences will be affected by several factors, including nucleic acid extraction and recovery, target size and copy number, participation in amplification reactions if used, and specific probe performance. The last three factors likely contributed to the differences in assay performance, in which 10,000 copies of JC or BK polyomavirus (5-kb genomes in a 4-kb double-stranded circular DNA plasmid vector) were detected with probe intensity ranges of 61 to 4,889 (JC virus, 42 probes) and 4 to 442 (BK virus, 9 probes). In contrast, adenovirus type 5 (36-kb genome, double-stranded linear nonintegrated DNA) was detected over an intensity range of to three hypothetical tumor samples. All probes for Acc2 show high signal in tumor 1 (left), so this candidate should be detectable by comparing the accession's all-probe averages from test samples with those of control samples. A subset of Acc3 probes show high signal in tumor 2 (middle), perhaps due to strain sequence differences or partial deletion of the genome, reducing the all-probe accession average and making detection more difficult. In this case, a sliding window analysis of local probe signals is not biased by accession annotation and may be more sensitive for candidate identification. A single probe for Acc1 has high signal in tumor 3 (right), so a third tier of analysis based solely on individual probe performance is needed to detect organisms not specifically targeted by the PathoChip but sharing sequence homology with one or a few probes. 342 to 65,325 using 63 probes and 100 target genome copies; differences in genome size and conformation may affect participation in whole-genome amplification reactions. Furthermore, probes clearly have different hybridization affinities despite sharing the same bioinformatic design criteria. PathoChip assay sen-sitivity will therefore vary across accessions and between protocol options, but the inclusion of multiple probes per accession and integration of candidates from different levels of data analysis provide avenues for optimizing the chances of detecting a pathogen in the screening projects. The HPV16 genome, for example, is a Integrating multiple probe analyses which include accessionlevel, sliding window, and individual probe comparisons detected HPV16 in 34 of 48 OSCC tumors individually tested by PathoChip screening, a 71% occurrence rate somewhat higher than the estimated 63% rate previously reported (29, 30) but not unreasonable given the rapidly increasing prevalence of papillomaviruses in oropharyngeal cancers (31) . The results of the assay were highly concordant with the molecular pathology reports for p16 overexpression and in some cases suggested that an HPV strain other than HPV16 may be responsible. HPV16 detections by PathoChip assays were confirmed by PCR using primers that are independent of PathoChip probes and by recovery and sequencing of HPV16 regions located outside those targeted by capture probes on the HPV genome.