On 2016 Nov 08, P P Wolkow commented:

It is true that culture negative blood samples are difficult to work with and the interpretation of the obtained results is not easy. We believe that we have taken every care to avoid contamination and to analyze our data accordingly. However, we understand that you might not fully concur with our results.

Indeed NTC and blood samples contain the same taxa as viewed on Fig. 3a. However based on this sole statement, one cannot claim that these samples are similar. In fact they are different, what can be seen below based on following examples at the Order level. Bifidobacteriales in the healthy blood constituted 73.0% of reads vs. 12.8% in NTC samples (p = 2.76 x 10-7). In our opinion this level of significance confirms that the groups are different. Few other examples based on Figure 4 data: healthy vs NTC vs sepsis: Actinomycetales: 2.0% vs 7.7% vs 30.9%- p=0.04; Pseudomonadales: 6.7% vs 0.0% vs 4.4% - p=0.006; Sphingomonadales: 0.2% vs 11.4% vs 7.3%-p=3 x 10-7.

It is true that NTC samples cluster with clinical samples in PCoA analysis, however they cluster with septic samples but not with healthy ones. Clustering with the latter could potentially mean that the results for the healthy people are untrue due to sequencing a contamination only. Please note that only 3 out of 5 NTC samples passed the analytical threshold and only these are depicted on Fig. 2. In the two NTC samples filtered out from further analysis, the numbers of reads were very low, 480 and 73, respectively. Also, contamination of simultaneously processed NTC samples should result in similar abundance of phyla in these samples, which is not the case (Fig. 3).

PCR conditions are provided in Table 1. Negative control procedure was exactly the same for all samples analyzed and should have no impact on the results. All samples were prepared in one batch so we did not expect a batching effect.

An idea of retrospective contaminant read removal is in our opinion controversial. There are deep inter-individual differences between the NTC samples. Removal of the reads based on mean number of reads would lead to skewed results and negative read numbers in some samples.

We expected that results depicted on Fig. 5 would be self-explanatory. However, we admit that providing qPCR data would be superior. We are grateful for drawing our attention to this issue which will have implications for our future work.

On 2016 Oct 27, Susannah Salter commented:

The paper by Gosiewski et al (PMID27771780) draws strange conclusions from the data. Culture-negative blood samples are certainly difficult to work with as they are more susceptible to the influence of DNA contamination.

In the text attention is drawn to the sequenced controls (water) but some assertions are patently untrue: for example stating that Bifidobacterales were a noteworthy constituent of healthy blood when it also makes up >10% proportion of the negative controls (Fig 4), or stating that the control profiles are "completely different" to the blood samples despite appearing to contain most of the same taxa (Fig 4) and clustering with the clinical samples on PCoA (Fig 2).

The authors provide no detailed information about the number of PCR cycles, the negative control procedure, kit batching of samples, retrospective contaminant read removal etc, which would lend confidence that the described patterns are not artefacts of sample processing. qPCR would also help to clarify the background contaminant DNA levels and allow more robust conclusions to be drawn.

The supplemental figure has the most potentially interesting information but unfortunately it is not labelled or described in text. If the contaminant taxa are removed, there may be some nice signals hiding in there.

On 2016 Oct 27, Susannah Salter commented:

The paper by Gosiewski et al (PMID27771780) draws strange conclusions from the data. Culture-negative blood samples are certainly difficult to work with as they are more susceptible to the influence of DNA contamination.

In the text attention is drawn to the sequenced controls (water) but some assertions are patently untrue: for example stating that Bifidobacterales were a noteworthy constituent of healthy blood when it also makes up >10% proportion of the negative controls (Fig 4), or stating that the control profiles are "completely different" to the blood samples despite appearing to contain most of the same taxa (Fig 4) and clustering with the clinical samples on PCoA (Fig 2).

The authors provide no detailed information about the number of PCR cycles, the negative control procedure, kit batching of samples, retrospective contaminant read removal etc, which would lend confidence that the described patterns are not artefacts of sample processing. qPCR would also help to clarify the background contaminant DNA levels and allow more robust conclusions to be drawn.

The supplemental figure has the most potentially interesting information but unfortunately it is not labelled or described in text. If the contaminant taxa are removed, there may be some nice signals hiding in there.