12 major claims checked against the paper's own evidence: all adequately supported.
partialDiscussionReviewer 3
The biopsy site, not the tumor's tissue of origin, primarily determines whether a confident microbial signal is present in metastatic tumors.
The paper shows that in the O'Neill pan-cancer dataset, confident microbial signals are enriched in exposure-associated biopsy sites, but the analysis is limited by small numbers (only 10 samples with confident signals) and does not formally test the interaction between biopsy site and tumor origin.
Evidence: Results section 'Median(L)_adj shows that genuine microbial signals in tumor and normal tissues are confined to GI tissues and other exposure-associated biopsy sites' with figures.
“Analysis of the metastatic tumor in the public pan-cancer dataset() suggested that the biopsy site, not the tumor’s tissue of origin, primarily determines whether a confident microbial signal is present.”
supportedAbstract, Results paragraph 1-2Reviewer 1
DNA fragment length serves as a discriminator: genuine microbiome genomes have long fragments, contaminants are short.
The claim is directly supported by spike-in experiments in germ-free mouse tissues and cell lines, which show clear length differences.
Evidence: Figure 1 and results show that spike-in bacteria produce long reads comparable to host DNA, while non-spike-in reads are short.
“While genuine microbiome genomes have long genomic DNA fragments, contaminant DNA is typically short.”
supportedResults paragraph 3Reviewer 1
The Median Length-Adjusted (Med(L)adj) metric effectively separates genuine microbiota from contaminants.
The metric is validated on spike-in and negative-control datasets, showing clear separation of high and low values.
Evidence: Figures 2 and 3 show Med(L)adj values for spike-in bacteria are high, while non-spike-in taxa are low.
“Applying M e d i a n ( L ) a d j to spike-in bacteria in GFM tissues and to bacterial genera detected in human cell lines revealed a clear difference.”
supportedAbstract, Results paragraph 4-5Reviewer 1
Genuine microbial signals in human tissues are largely limited to tissues with natural microbial exposure (GI tract, cervix, vagina, skin).
The analysis of multiple datasets consistently shows that only tissues with natural exposure have high Med(L)adj values.
Evidence: Figures 3-5 show that CRC, gastric, and exposure-associated sites have high Med(L)adj genera, while others lack them.
“across multiple human tumor and normal tissues, we found genuine microbiome signals are largely limited to tissues with natural microbial exposure.”
supportedAbstract, Results paragraph 5-6Reviewer 1
No evidence of resident microbiome in kidney, brain, blood, and placenta.
The analysis of these tissues shows uniformly low Med(L)adj values and no long reads, even after pooling samples.
Evidence: Figures 4, 5, and 6 show that kidney, brain, lung, blood, and placenta lack high Med(L)adj genera.
“while other tissues (e.g. kidney, brain, blood, and placenta) showed no evidence of resident microbiome.”
supportedResults paragraph 7Reviewer 1
Long-read analysis confirms absence of confident microbiome signal in placenta and blood.
Reanalysis of public placenta PacBio data and newly generated blood PacBio data shows only short microbial reads and low Med(L)adj values.
Evidence: Figure 6 shows placenta and blood samples have low Med(L)adj values and no long reads.
“Long-read analysis re-confirms absence of confident microbiome signal in placenta and blood.”
supportedAbstract, Results (first two sections)Reviewer 2
Genomic DNA fragment length serves as a discriminator between genuine microbiome signals and contaminant DNA.
The paper provides direct evidence from spike-in experiments and negative controls showing that intact bacterial cells yield long DNA fragments while contaminant DNA is short.
Evidence: Figure 2 data from spike-in experiments; Figure 1 from germ-free mouse and cell line data.
“Here we show that genomic DNA fragment length serves as a discriminator: while genuine microbiome genomes have long genomic DNA fragments, contaminant DNA is typically short.”
supportedResults, second sectionReviewer 2
Median Length-Adjusted (Median(L)adj) effectively separates genuine microbiota from contaminants.
The metric is validated against spike-in controls and negative controls (GFM, cell lines) showing clear separation of high and low values.
Evidence: Figure 2C,D; Results section 'A read-length-based metric separates genuine microbiota from contaminants'.
“Applying Median(L)adj to spike-in bacteria in GFM tissues ... revealed a clear difference: spike-in bacterial reads consistently showed high values, whereas non–spike-in taxa ... had very low values.”
supportedResults, third sectionReviewer 2
Genuine microbial signals in human tissues are largely confined to sites with natural microbial exposure (GI tract, cervix, vagina, skin).
The paper shows that only tissues like CRC, gastric, cervix, skin, and vagina yield high Median(L)adj values, while kidney, brain, lung, and others do not.
Evidence: Figure 3, Figure 4; Results sections 'Median(L)adj shows that genuine microbial signals ... are confined to GI tissues and other exposure-associated biopsy sites' and analysis of O'Neill et al. 2024 dataset.
“applying the framework to human tumors and normal tissues, our study showed that in both primary and metastatic cancer, confident microbial signals were largely confined to GI tissues and other sites with natural exposure to microbes, such as skin and vagina.”
supportedAbstract, Results, DiscussionReviewer 2
No evidence for a pervasive intratumor microbiome in most primary or metastatic tumor types.
The paper demonstrates lack of high-integrity microbial signals in kidney, brain, lung, and most other cancers, even after pooling samples to increase depth.
Evidence: Figure 3, Figure 4, aggregate analyses of kidney and lung cancer samples.
“We found no solid evidence for widespread microbiomes across most other tumor types, in contrast to previous short-read–based reports in primary() and metastatic cancer().”
supportedResults, fourth sectionReviewers 2, 3
Residual long microbial reads in non-exposed tissues are dominated by well-known contaminants.
Analysis of reads >5 kb shows that in non-GI tissues, such reads map to common contaminants like Delftia and Bradyrhizobium, not to plausible tissue-resident taxa.
Evidence: Figure 5; Results section 'Residual long microbial reads are dominated by well-known contaminants'.
“the taxa supported by microbial reads >5 kb were far less abundant and were dominated by organisms commonly reported as laboratory or reagent-associated contaminants”
supportedResults, final sectionReviewer 2
The placenta and blood show no evidence of a resident microbiome under normal conditions.
Reanalysis of public placenta PacBio data and new blood PacBio data show only short, fragmented microbial reads, consistent with contamination.
Evidence: Figure 6; Results section 'Long-read analysis re-confirms absence of confident microbiome signal in placenta and blood'.
“our analysis detected only short microbial reads and no reliable long fragments (), providing no evidence for a bona fide placenta microbiome.”