Reply to Grigoriev et al., "Sequences of SARS-CoV-2 'Hybrids' with the Human Genome: Signs of Non-coding RNA?"

REPLY

High throughput sequencing reads from virally infected cells provide detailed information about both the infected host cells and invading viruses (1). For example, RNA-sequencing techniques from infected cells contains reads that unequivocally align to either the host or the viral transcriptomes, enabling quantification of host and viral gene expressions (2). Occasionally, there are reads with split characteristics, having one part (e.g., the 5′ end) unambiguously matching the host and another part (e.g., the 3′ end) clearly matching the viral genomes. The split characteristic with unambiguous matching on either part is the key here, typically requiring convincing stretches of sequence matches such as >30 bp that we used in our analysis (3). Such reads are termed host-virus chimeric reads (HVCRs). Indeed, HVCRs that surpass statistical reproducibility and signal-to-noise standards might carry novel insights into the biology of host-virus interactions (4, 5). Thus, it is important to unambiguously detect statistically rigorous and biologically relevant HVCRs. We and others have shown that detection of relevant HVCRs is complicated by unfaithful reverse transcriptase and polymerase enzymes that template-switch during typical high throughput sequencing library preparation protocols (6–9).

The conventional HVCRs with split characteristics that we and others used in our studies should not be confused with what we term “composite” host reads that contain short matches to the viral genome or, vice-versa, viral reads that contain short sequence matches to the host genome in the middle of the reads. Such “composite” viral reads seem to be the subject of the letter contributed by Grigoriev et al. Our work only evaluated the biological relevance of conventional HVCRs and showed that in the context of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection, they are most likely artifacts of library construction. Due to the short nature of sequence matches within “composite” reads (such as those identified by Grigoriev et al.), they are more prone to statistical anomalies and alignment errors and are likely to align by chance to at least some regions of the 3.2 billion base pairs encoded in the human genome. Thus, any analysis of “composite” events would need to include empirical or theoretical probabilities of such observations under rigorous control experiments to rule out template switching, alignment errors, or statistical anomalies.

Nevertheless, to avoid any misinterpretation, it is important to note that the observations of composite reads by Grigoriev et al. have no bearing on our original findings (3) and follow-up studies by others (10, 11) that HVCRs in SARS-CoV-2-infected cells do not support integration events and are infrequent and artifactual.