Basic reporting

This paper provides an insightful analysis of a virome surveillance. It offers an original approach that is both rigorous and accessible. The findings are mostly well-supported, and it is likely to be highly cited. The paper is an elegant example of research that will be widely recognized.

This article presents an original approach to the virome of a remote ecosystem by using ants as a proxy. With just 209 ants, the authors were able to detect 22,406 virus-like contigs belonging to 56 families. Seventeen of the 29 ant colonies were identified thanks to the “accidental/non targeted” recovery of the COI gene. This approach is likely to lead to an increased level of detection in poorly studied areas. This will be beneficial for the global virome description. Notably, the authors highlighted the overrepresented families of Parvoviridae and Circoviridae. Sequences of 403 Parvovirus were analysed based on their SF3 proteins, with more than 200 amino acids available for comparison with publicly available data. This revealed an increased diversity, as well as an expanded geographical distribution and potential host range. Additionally, 45 complete genomes of novel cyclovirus were resequenced and compared with publicly available data, providing further insights into this family.

Experimental design

The work on the Parvoviridae and Circoviridae families is very thourough; however, due to the nature of the sequences, similar conclusions could not be made for other virus groups. The number of contigs and virus families identified in the study were based on contigs with lengths ≥200 nt, and retained viral BLASTx assignations of these contigs with e-values < 0.001 (M&M). It is not specified in the text that the BlastX was done on a complete non-redundant protein database (GenBank non-redundant database is indicated on the legend of fig 1). The amino acid identity recovered, as reported in Figure 1, was as low as <25%. Figure 1 is informative but can be misleading as a virus species can be represented multiple times, e.g. the two closely related points for the nepovirus can represent two different viruses or two contigs covering different parts of this segmented virus. In addition, the percentage of homology represented in figure 1 can be from very conserved genes (e.g. RdRP) or from putative genes with low homology even within well described families (the same virus could have multiple contigs with very varied homology to the closest sequence from the database). The legend of this figure should also be clarified as to whether the amino acids homology is per sequence alignment, or the homology given by BlastX, where only the matching region of the molecule is measured (in which case, this can be a fraction of a short 200 nt contigs (67 aa)).

In the manuscript, the authors have been cautious not to overstate their findings. It is evident that ants are a good proxy to access difficult regions, and the authors note that the ants are “not completely unbiased”. Judging by Figure 1, they are clearly biased towards animal, mostly invertebrate ssDNA viruses (as mentioned p14L12). Few plant viruses are detected and mycoviruses are not discussed at all. The fact that these viruses have to pass through additional steps in the trophic chain is discussed on page 19, but what can be said about viruses with low stability, concentration, or prevalence? The principle of VANA should yield nucleic acids protected by a capsid (in contradiction with the degradation observed). Are ants the best candidates for a plant metavirome? The authors should provide a more detailed discussion about this.

While the identification of the ants through the recovered reads matching the COI is a useful bonus, it is not definitive. The number of reads is small, and the VANA tool is not designed to recover non-encapsidated viruses. Additionally, if this experiment was to be repeated, it would be beneficial to have some morphological identification and/or a proper DNA barcoding on the ants (which would require collecting two samples for each species, one for the metagenome and one for the taxonomy).

Validity of the findings

It is clear that besides the Parvoviridae and Circoviridae families, the contigs extracted were mostly short and from different genomic regions. In some cases, this allowed for a taxonomic assignment, and presumably, in other cases, the contigs could only be used to make the Figure 1 (137 sequences were deposited on GenBank for the phylogenetic analyses out of the 22,406 contigs). But that is the nature of these metagenome studies. Therefore, I understand that the phylogeny used is there to illustrate that the virus contigs (or virus-like in some cases) fit into available taxonomies but it would be good to explain why neighbor joining method was chosen. 

Additional comments

There are a few additional small edits:

Page 3 L23: The sentence needs to be rewritten as it reads as if the viruses have medical or agricultural relevance to human (instead of the host).

Page 4 L14: Densely forested tropical regions do not represent major interfaces but rather provides interface as a consequence of human activities surrounding these forests. Additionally, densely forested tropical regions clustered together, represent fewer interfaces than if the forests were scattered across a larger territory. 

Page 4 L21: random/unbiased : all the tools relying on one animal will have preferred patterned, but those will be different to the human one. I like the way it is defined earlier “a less human-centric assessment of viral diversity at the ecosystem-scale”

Page 25 L4: Since endogenous paroviral elements are detected within invertebrate genomes, how many of the parvovirus contigs could be EPVs?

Figures with phylogenetic analyses (mostly 2 and 3): could you when the aligment is made on the protein or the nucleic acid and the size of the region aligned.