On 2013 Oct 23, Kenneth Daily commented:

The topic of this paper is very interesting - if there are long-term viruses that rapidly evolve along other more stable viruses. However, there are a few misstatements and omissions that would help strengthen this paper.

The authors write:

"Sequence reads from each sample were first assembled individually using MetaIDBA (16)."

The reference for this statement (Peng Y, 2012) refers to IDBA-UD, however.

Could the authors comment on which tool was used for metagenome assembly?

The authors write:

"The correlation coefficient between replicate samples from the same time point was at least 0.99, indicating a high degree of reproducibility (Fig. S1)."

In the supplemental methods, a section titled "Reproducibility of Contig Detection" gives the correlation coefficients from Figure S1.

1. The minimum correlation coefficient given is 0.9743. Hence, the above statement does not seem to be correct. I do not dispute that this is still a high degree of reproducibility across the two sample replicates.

2. The "day" given in this list does not correspond to the day in Figure S1; it seems to match the sample name given in the SRA submitted data (example: BLS020824_d13-1 and BLS020825_d13-2 would correspond with 13 in this table).

The methods in this paper are difficult to reproduce. No metadata about each of the samples (most importantly, the day on which they were taken) is present in the data supplied to SRA, BioSample, or BioProject, except for sample SRS413841.

Details on the versions and parameters for running the various tools are not included (for MetaIDBA, promer, Bowtie2, Glimmer, MetaPhlAn, PILER-CR, Blastn, and the "custom scripts"). Given the complexity and number of steps of the analysis and that most of the tools are openly available, some sort of pipeline of the procedures run on this sequencing data would assure reproducibility.

On 2013 Oct 23, Kenneth Daily commented:

The topic of this paper is very interesting - if there are long-term viruses that rapidly evolve along other more stable viruses. However, there are a few misstatements and omissions that would help strengthen this paper.

The authors write:

"Sequence reads from each sample were first assembled individually using MetaIDBA (16)."

The reference for this statement (Peng Y, 2012) refers to IDBA-UD, however.

Could the authors comment on which tool was used for metagenome assembly?

The authors write:

"The correlation coefficient between replicate samples from the same time point was at least 0.99, indicating a high degree of reproducibility (Fig. S1)."

In the supplemental methods, a section titled "Reproducibility of Contig Detection" gives the correlation coefficients from Figure S1.

1. The minimum correlation coefficient given is 0.9743. Hence, the above statement does not seem to be correct. I do not dispute that this is still a high degree of reproducibility across the two sample replicates.

2. The "day" given in this list does not correspond to the day in Figure S1; it seems to match the sample name given in the SRA submitted data (example: BLS020824_d13-1 and BLS020825_d13-2 would correspond with 13 in this table).

The methods in this paper are difficult to reproduce. No metadata about each of the samples (most importantly, the day on which they were taken) is present in the data supplied to SRA, BioSample, or BioProject, except for sample SRS413841.

Details on the versions and parameters for running the various tools are not included (for MetaIDBA, promer, Bowtie2, Glimmer, MetaPhlAn, PILER-CR, Blastn, and the "custom scripts"). Given the complexity and number of steps of the analysis and that most of the tools are openly available, some sort of pipeline of the procedures run on this sequencing data would assure reproducibility.