Petabase-scale Papillomavirus Discovery

Reviewer #1 (Public review):

Summary:

The manuscript titled "Petabase-scale papillomavirus discovery" capitalizes on Logan assemblages to identify novel and known PVs.

This is a brilliant use of Logan assemblages, and this paper highlights the use of this, plus also shows that one person's trash is another one's gold. Superb paper and I applaud the authors for starting with the PVs as easier to identify due to their set of genes coupled with the conserved L1 protein and associated typing for PVs (10% pairwise identity threshold for identification of new PV types).

Strengths:

This study highlights the hidden gems in public resources, especially if mined properly. Thanks to Logan assemblages, this is possible, and this manuscript highlights this with their data mining of papillomaviruses, identifying known and novel PVs in pangolins, lizards, fish, and white rhinos.

Weaknesses:

None identified

Reviewer #2 (Public review):

This study applies the Logan assemblage of the Sequence Read Archive to investigate papillomavirus diversity at a large scale. The authors combine sequence similarity searches, phylogenetic analyses, protein annotation, structural comparisons, and metadata integration to identify and characterize novel papillomavirus sequences. Beyond papillomavirus sequence discovery, the study develops a framework for associating viral sequences with host, geographic, and ecological metadata through the integration of both established and recently developed large-scale public sequence repositories. This framework may provide a foundation for similar investigations across other viral taxa.

An important aspect of the work is the systematic processing and integration of metadata across a large number of sequencing libraries. The approaches developed to aggregate, curate, and interpret host and environmental information may be broadly applicable to future large-scale studies of other viral families. At the same time, interpretations based on metadata-derived ecological and host-association patterns should be considered in the context of the inherent limitations of public sequencing repositories, including uneven taxonomic representation, heterogeneous sampling strategies, laboratory-derived samples, and variable sequencing protocols. The comments below primarily address methodological details, interpretation of ecological patterns, and opportunities to further strengthen the robustness of the analyses and conclusions.

Pages 136-137: The manuscript focuses primarily on L1-containing contigs. Could the authors provide a summary of libraries containing other PV hallmark genes (e.g., E1 or E2) but lacking full-length L1 sequences? This would help assess how much additional PV diversity may remain inaccessible under an L1-centered framework. Furthermore, the manuscript does not provide a detailed discussion of technical or biological explanations for libraries with detected PV sequences but lacking L1 sequences. Could such cases represent incomplete assemblies, low-abundance infections, highly divergent PVs, or endogenous papillomavirus-derived elements? Clarifying these possibilities would help readers interpret the biological significance of these detections.

Page 141: The rationale for performing the search in two sequential Logan releases is unclear. Does Logan v1.1 fully supersede v1.0? If so, why was the initial search performed on v1.0 rather than directly on v1.1? Providing a clearer description of the differences between the two database versions and explaining how these differences motivated the two-round search strategy would improve reproducibility.

The data could have been explored at the libraries' read level. The analysis is currently limited to presence/absence and diversity patterns derived from assembled PV contigs. However, the identified PV-positive libraries provide an opportunity to explore abundance-related metrics. Read counts, coverage estimates, or other measures of sequence representation could be used to characterize PV abundance within libraries, providing additional ecological context and helping distinguish low-level incidental detections from strongly represented infections. In addition, the curated PV sequence dataset generated in this study could serve as a reference for targeted read-mapping analyses. Aligning reads from a subset of libraries classified as PV-negative may help determine whether PV sequences are present below assembly detection thresholds. Such an analysis could provide valuable insights into the sensitivity of assembly-based virus discovery approaches and help establish practical coverage or read-count thresholds for detecting low-abundance papillomaviruses. These results could have important implications for future surveillance, clinical, and environmental studies aimed at PV detection.

Pages 155-156: Clustering was performed using sequence identity, while host, geographic, and ecological annotations were assigned from representative centroids. How frequently did clusters contain sequences associated with conflicting metadata (e.g., distinct hosts or geographic regions), and how were such cases handled?

Page 160: The authors state that the 70% query coverage threshold was selected based on an observed bimodal distribution. Could this analysis be shown explicitly (e.g., in a supplementary figure), and could the authors discuss how sensitive the number of novel PV calls could have been to alternative coverage thresholds?

Pages 167-168: The rationale for clustering highly divergent sequences at 60% nucleotide identity should be explained. Is this threshold associated with established genus-level classifications in this viral family, or was it chosen empirically?

Pages 167-168: For the 45 sequences lacking nucleotide-level matches, did the authors investigate amino-acid similarity to known PV L1 proteins? Such analyses would help determine whether these sequences represent deeply divergent PVs or potentially more distant viral lineages.

Pages 181-183: The biological interpretation of host-associated PV diversity may depend on library type. Could the authors summarize the proportion of samples originating from field collections, laboratory animals, cell culture systems, or experimental infections?

Pages 243-248: Could the observed geographic and ecological patterns be influenced by laboratory-derived samples? Distinguishing field-collected samples from laboratory, captive, or experimental material would strengthen the ecological interpretations.

Pages 254-261: The biome analyses focus on PV occurrence. An analysis of host composition across biomes would be highly informative and could help disentangle whether observed patterns reflect PV ecology or underlying host distributions.

Page 294: Statements regarding structural similarity appear to rely primarily on visual comparisons. Could the authors provide quantitative structural alignment metrics (e.g., RMSD, TM-score, DALI score, Foldseek score) to support these conclusions?

Page 321-324: Given the scale and curation of the dataset, the final case-study section is limited. Broader comparative analyses of gene content, ORF architecture, and composition related to host associations, phylogenetic relationships, and/or ecological variables could provide additional evolutionary insights beyond a small number of illustrative examples.

Page 333: Given the emphasis on the feasibility of petabase-scale sequence mining, the manuscript would benefit from a more detailed description of the computational resources required. The reported ~10-hour runtime is difficult to interpret without information regarding hardware specifications, CPU-hours, memory requirements, storage footprint, and cloud infrastructure (if used). Such information is important for evaluating the reproducibility and practical applicability of the approach.

Pages 408-409: Were metadata available regarding viral enrichment procedures, particle purification, or size-selection protocols? Such information could influence the interpretation of PV detection frequencies across library types.

Pages 415-416: The differentiation between viral and endogenous viral sequences is one of the biggest challenges in viral metagenomics and large-scale data mining for viral sequences. This issue is particularly relevant because the distinction between exogenous and endogenous viral sequences may directly affect estimates of novel PV diversity and inferred host associations. The manuscript acknowledges that papillomavirus sequences recovered from DNA-based libraries may derive from integrated viral DNA. However, there is no systematic analysis addressing the potential contribution of endogenous papillomavirus elements (EVEs) to the reported diversity estimates. Given the large number of host genome sequencing projects represented in the SRA, some detected PV-like sequences may correspond to integrated or fossil viral sequences rather than exogenous viruses. The authors could discuss this possibility more explicitly and provide analyses evaluating the prevalence of integration signatures, disrupted ORFs, host-genome flanking regions, or other indicators that would help distinguish endogenous viral elements from actively circulating PVs.

Pages 532-533: The study is described as "petabase-scale"; however, the analyses were performed on a pre-assembled and compressed representation of the SRA rather than directly on petabase-scale raw sequencing data. The authors may wish to clarify this distinction and explicitly acknowledge that the computational burden is substantially reduced by the Logan framework and, from this perspective of computational power applied, this study is not in the same context as Serratus and Logan.

Perspective comment: One of the strengths of this study is the generation of a highly curated papillomavirus protein dataset spanning a broad range of known and newly identified PV diversity. Given the increasing importance of structure-based homology detection in virology, the authors may wish to discuss the potential of this resource for future structure-guided discovery efforts. Recent studies have shown that protein structure prediction and comparison can reveal extremely distant evolutionary relationships that are undetectable at the sequence level. The curated PV dataset generated here could serve as a valuable reference for searching unannotated proteins from metagenomic "dark matter" datasets for structural homologs or convergent folds related to papillomavirus proteins. Such approaches may help identify highly divergent PV lineages or previously unrecognized viral proteins that retain structural similarity despite extensive sequence divergence.