Abstract
Background
Microbiome analysis generally requires PCR-based or metagenomic shotgun sequencing, sophisticated programs, and large volumes of data. Alternative approaches based on widely available RNA-seq data are constrained because of sequence similarities between the transcriptomes of microbes/viruses and those of the host, compounded by the extreme abundance of host sequences in such libraries. Current approaches are also limited to specific microbial groups. There is a need for alternative methods of microbiome analysis that encompass the entire tree of life.
Results
We report a method to specifically retrieve non-human sequences in human tissue RNA-seq data. For cellular microbes we used a bioinformatic 'net', based on filtered 64-mer small subunit rRNA sequences across the Tree of Life (the 'electronic tree of life', eTOL), to comprehensively (98%) entrap all non-human rRNA sequences present in the target tissue. Using brain as a model, retrieval of matching reads, re-exclusion of human-related sequences, followed by contig building and species identification, is followed by reconfirmation of the abundance and identity of the corresponding species groups. We provide methods to automate this analysis. A variant approach is necessary for viruses. Again, because of significant matches between viral and human sequences, a 'stripping' approach is essential. In addition, contamination during workup is a potential problem, and we discuss strategies to circumvent this issue. To illustrate the versatility of the method, we report the use of the eTOL methodology to unambiguously identify exogenous microbial and viral sequences in human tissue RNA-seq data across the entire tree of life including Archaea, Bacteria, Chloroplastida, basal Eukaryota, Fungi, and Holozoa/Metazoa, and discuss the technical and bioinformatic challenges involved.
Conclusions
This generic methodology may find wider application in microbiome analysis including diagnostics.