On 2017 Apr 12, Yu-Chen Liu commented:

We sincerely appreciate your insightful feedbacks and constructive advices on our research. We strongly appreciate the advice on excluding all sequences that mapped on the mammal genome before the search of potential plant miRNAs. On the other hand, given the facts that the reads mapped on both plant and mammal, whether such reads were false positively mammal prompted cannot be assured before further experimental validation. On the prospect of potential candidate discovery attempts, reads mapped on both plant and mammal genomes should not be omitted indifferently. This, in my opinion, is a dilemma between the measures of avoiding false positive and increasing discovery rate. Maybe both measures should be taken in the future study.

Thank you again, for the great advices and dedication made in reviewing of this research.

On 2017 Apr 12, Kenneth Witwer commented:

Liu YC, 2017 reported mapping very low levels of mature plant miRNAs in a subset of public data gathered from 198 human plasma samples, concluding that this was evidence of "cross-kingdom RNAi"; however, both the authors and I observed that only one putatively foreign sequence, "MIR2910," mapped consistently and at levels above a reasonable noise threshold. No data were presented to support functional RNAi. I further noted that MIR2910 is a plant rRNA sequence, has been removed from miRBase, and also maps with 100% coverage and identity to human rRNA. In the comment below, Dr. Liu now links to unpublished predicted hairpin mapping data that were not included in the Liu YC, 2017 BMC Genomics conference article, which, like my comments, focused on mature putative xenomiRs. Dr. Liu states that mapping has been done not only to the putative MIR2910 mature sequence (as reported), but also to the predicted MIR2910 precursor hairpin sequence.

This is an interesting development, and I strongly and sincerely commend Liu et al for sharing their unpublished data in this forum. This is exactly what PubMed Commons is about: a place for scientists to engage in civil and constructive discourse.

However, examination of the new data reinforces my observation that the only consistently mapped "foreign" sequence in the Liu YC, 2017 study is a human rRNA sequence, not a plant miRNA, mature or otherwise. Beyond the 100% identity of the 21nt putative MIR2910 mature sequence with human rRNA, a 47nt stretch (80%) of the plant "pre-MIR2910" rRNA fragment aligns to human 18S rRNA with only one mismatch (lower-case), and indeed Liu et al allowed one mismatch:

Plant rRNA fragment:

UAGUUGGUGGAGCGAUUUGUCUGGUUAAUUCCGuUAACGAACGAGAC

Human rRNA fragment:

UAGUUGGUGGAGCGAUUUGUCUGGUUAAUUCCGaUAACGAACGAGAC

Dr. Liu provides the example of a plasma small RNA sequencing dataset, DRR023286, as the primary example of plant miRNA mapping, so let us examine this finding more closely. DRR023286 was by far the most deeply sequenced of the six plasma samples in the Ninomiya S, 2015 study (71.9 million reads), as re-analysed by Liu et al. Yet, like all other data examined in the Liu et al study, and despite the much deeper sequencing, DRR023286 yielded only the pseudo-MIR2910 as a clear "xenomiR" (mature or precursor). Of special note, previously reported dominant dietary plant xenomiRs such as MIR159, MIR168a, and the rRNA fragment "MIR2911" were not detected reliably, even with one mismatch.

The precursor coverage plots for DRR023286 (and less deeply sequenced datasets, for that matter), also according to the newly provided Liu et al data, show that any coverage is in the 5' 80% of the putative MIR2910 sequence: exactly the part of the sequence that matches human rRNA. The remaining 12 nucleotides at the 3' end of the purported MIR2910 precursor are conspicuously absent and never covered in their entirety. To give one example from the Liu et al data, in the deepest-sequenced DRR023286 dataset, even the single short read that includes just 11 of these 12 nucleotides has a mismatch. Furthermore, various combinations of these 3' sequences match perfectly to rRNA sequences in plant and beyond (protist, bacterial, etc.). Hence, the vanishingly small number of sequences that may appear to support a plant hairpin could just as convincingly be attributed to bacterial contamination...but we are already playing in the noise.

As noted, Liu et al allowed one mismatch to plant in their mapping and described no pre-filtering against human sequences. In the DRR023286 dataset, fully 90% of the putative mapping to plant MIR2910 included a mismatch (and thus human)...the other 10% were mostly sequences 100% identical to human.

In conclusion, the main points I raised previously have not been disputed by Liu et al:

1) numerous annotated plant miRNAs in certain plant "miRNA" databses appear to have been misannotated or the result of contamination, including some sequences reported by Liu et al that map to human and not plant;

2) the mature "MIR2910" sequence is a plant ribosomal sequence that also maps perfectly to human rRNA; and

3) the read counts for all but one putative plant xenomiR in the Liu et al study are under what one might consider a reasonable noise threshold for low-abundance RNA samples (plasma) with tens of millions of reads each.

Furthermore, the current interaction establishes a new point:

4) the plant rRNA sequence annotated by some as a "MIR2910" precursor maps almost entirely (and for all practical purposes entirely) to a human rRNA sequence.

Future, more rigorous searches for plant xenomiRs in mammalian tissues and fluids will require a pre-filtering step to exclude all sequences that map with one (or more) mismatches to all mammalian genomes/transcriptomes and preferably other possible contaminants, followed by a zero-mismatch requirement for foreign mapping.

On 2017 Apr 11, Yu-Chen Liu commented:

The authors appreciate the insightful feedbacks and agree with prospect that hypothesis derived from small RNA-seq data analysis deserve examination in skeptical views and further experimental validation. Regarding the skeptical view of Prof. Witwer on this issue, whether a specific sequence were indeed originate from plant can be validated through examining the 2’-O-methylation on their 3’ end (Chin, et al., 2016; Yu, et al., 2005). The threshold of potential copy per cell for plant miRNAs to affect human gene expression was also discussed in previous researches (Chin, et al., 2016; Zhang, et al., 2012).

Some apparent misunderstandings are needed to be clarified:

In the commentary of Prof. Witwer:

“A cross-check of the source files and articles shows that the plasma data evaluated by Liu et al were from 198 plasma samples, not 410 as reported. Ninomiya et al sequenced six human plasma samples, six PBMC samples, and 11 cultured cell lines 19. Yuan et al sequenced 192 human plasma libraries (prepared from polymer-precipitated plasma particles). Each library was sequenced once, and then a second time to increase total reads.”

Authors’ response:

First of all, the statement "410 samples" within the article was meant to the amount of runs of small RNA-seq run conducted in the referred researches. Whether multiple NGS runs conducted on same plasma sample should be count as individual experiment replicates is debatable. The analysis of each small RNA-seq run was conduct independently. The authors appreciate the kind comments for the potential confusion that can be made in this issue.

In the commentary of Prof. Witwer:

“Strikingly, the putative MIR2910 sequence is not only a fragment of plant rRNA; it has a 100% coverage, 100% identity match in the human 18S rRNA (see NR 003286.2 in GenBank; Table 3). These matches of putative plant RNAs with human sequences are difficult to reconcile with the statement of Liu et al that BLAST of putative plant miRNAs "resulted in zero alignment hit", suggesting that perhaps a mistake was made, and that the BLAST procedure was performed incorrectly.”

Authors’ response:

The precursor sequences of the plant miRNAs, including the stem loop sequences (precursor sequences) were utilized in the BLAST sequence alignment in this work. The precursor sequence of peu-MIR2910, “UAGUUGGUGGAGCGAUUUGUCUGGUUAAUUCCGUUAACGAACGAGACCUCAGCCUGCUA” was used. The alignment was not performed merely with the mature sequence, “UAGUUGGUGGAGCGAUUUGUC”. The stem loop sequences, as well as the alignment of the sequences against the plant genomes, was taken into consideration by using miRDeep2 (Friedländer, et al., 2012). As illustrated in the provided figures, sequencing reads were mapped to the precursor sequences of MIR2910 and MIR2916. As listed in the table below, a lot of sequencing reads can be aligned to other regions within the precursor sequences except the sequencing reads aligned to mature sequences. For instance, in small RNA-seq data of DRR023286, 5369 reads were mapped to peu-MIR2910, and 4010 reads were mapped to the other regions in the precursor sequences.  

miRNA | Run |Total reads | on Mature | on precursor

peu-MIR2910 | DRR023286 | 9370 | 5369 | 4010

peu-MIR2910 | SRR2105454 | 3013 | 1433 | 1580

peu-MIR2914 | DRR023286 | 1036 | 19 | 1017

peu-MIR2916 |SRR2105342 | 556 | 227 | 329

(Check the file MIR2910_in_DRR023286.pdf, MIR2910_in_SRR2105454.pdf, MIR2914_in_DRR023286 and MIR2916_in_SRR2105342.pdf)

The pictures are available in the URL:

https://www.dropbox.com/sh/9r7oiybju8g7wq2/AADw0zkuGSDsTI3Aa_4x6r8Ua?dl=0

As described in the article, all reported reads mapped onto the plant miRNA sequences were also mapped onto the five conserve plant genomes. Within the provided link a compressed folder file “miRNA_read.tar.gz” is available. Results of the analysis through miRDeep2, were summarized in these pdf files. Each figure file was named according to the summarized reads, sequence run and the mapped plant genome. For example, reads from the run SRR2105181 aligned onto both Zea mays genome and peu-MIR2910 precursor sequences are summarized in the figure file “SRR2105181_Zea_mays_peu-MIR2910.pdf”.

In the commentary of Prof. Witwer:

“Curiously, several sequences did not map to the species to which they were ascribed by the PMRD. Unfortunately, the PMRD could not be accessed directly during this study; however, other databases appear to provide access to its contents.”

Authors’ response:

All the stem loop sequences of plant miRNAs were acquired from the 2016 updated version of PMRD (Zhang, et al., 2010), which was not properly referred. The used data were provided in the previously mentioned URL.

In the commentary of Prof. Witwer:

“Counts were presented as reads per million mapped reads (rpm). In contrast, Liu et al appear to have reported total mapped reads in their data table. Yuan et al also set an expression cutoff of 32 rpm (log2 rpm of 5 or above). With an average 12.5 million reads per sample (the sum of the two runs per library), and, on average, about half of the sequences mapped, the 32 rpm cutoff would translate to around 200 total reads in the average sample as mapped by Liu et al.”

Authors’ response:

Regarding the concern of reads per million mapped reads (rpm) threshold, the author appreciate the kind remind of the need to normalize sequence reads count into the unit in reads per million mapped reads (rpm) for proper comparison between samples of different sequence depth. However the comparison was unfortunately not conducted in this work. Given the fact that the reads were mapped onto plant genome instead of human genome, the normalization would be rather pointless, considering the overall mapped putative plant reads only consist of ~3% of the overall reads. On the other hand, the general amount of cell free RNA present in plasma samples was meant to be generally lower than within cellar samples (Schwarzenbach, et al., 2011).

Reference

Chin, A.R., et al. Cross-kingdom inhibition of breast cancer growth by plant miR159. Cell research 2016;26(2):217-228.

Friedländer, M.R., et al. miRDeep2 accurately identifies known and hundreds of novel microRNA genes in seven animal clades. Nucleic acids research 2012;40(1):37-52.

Schwarzenbach, H., Hoon, D.S. and Pantel, K. Cell-free nucleic acids as biomarkers in cancer patients. Nature Reviews Cancer 2011;11(6):426-437.

Yu, B., et al. Methylation as a crucial step in plant microRNA biogenesis. Science 2005;307(5711):932-935.

Zhang, L., et al. Exogenous plant MIR168a specifically targets mammalian LDLRAP1: evidence of cross-kingdom regulation by microRNA. Cell research 2012;22(1):107-126.

Zhang, Z., et al. PMRD: plant microRNA database. Nucleic acids research 2010;38(suppl 1):D806-D813.

On 2017 Apr 07, Kenneth Witwer commented:

Caution is urged in interpreting this conference article, as described in more detail in my recent commentary. For careful analysis of this issue, using greater numbers of studies and datasets and coming to quite the opposite conclusions, see Kang W, 2017 and Zheng LL, 2017. Here, Liu et al examined sequencing data from two studies of a total of 198 plasma samples (not 410 as reported). Although no canonical plant miRNAs were mapped above a reasonable background threshold, one rRNA degradation fragment that was previously and erroneously classified as a plant miRNA, MIR2910, was reported at relatively low but consistent counts. However, this rRNA fragment is found in human 18S rRNA and is thus most simply explained as part of the human degradome. The other reportedly detected plant miRNAs were mostly found in a small minority of samples and in those were mapped at average read counts of less than one per million. These sequences may be amplification, sequencing, or mapping errors, since reads were mapped directly to plant (with one mismatch allowed) with no pre-filtering against mammalian genomes/transcriptomes. Several purported plant sequences, e.g., ptc-MIRf12412-akr and ptc-MIRf12524-akr, map perfectly to human sequences but do not appear to map to Populus or to other plants, suggesting that the plant miRNA database used by the authors and published in 2010 may include some human sequences. This is not a surprise, given pervasive low-level contamination in sequencing data, as reported by many authors.

Of course, even if some of the mapped sequences were genuine plant RNAs, they would be present in blood at greatly subhormonal levels unlikely to affect biological processes. No evidence of function is provided, apart from in silico predictions of human targets of the putative MIR2910 sequence, which, as noted above, is a human sequence. Thus, the titular claim of "evidences of cross-kingdom RNAi" is wholly unsupported. Overall, the results of this study corroborate the findings of Kang W, 2017 and previous studies: that dietary xenomiR detection is likely artifactual.

On 2017 Apr 07, Kenneth Witwer commented:

Caution is urged in interpreting this conference article, as described in more detail in my recent commentary. For careful analysis of this issue, using greater numbers of studies and datasets and coming to quite the opposite conclusions, see Kang W, 2017 and Zheng LL, 2017. Here, Liu et al examined sequencing data from two studies of a total of 198 plasma samples (not 410 as reported). Although no canonical plant miRNAs were mapped above a reasonable background threshold, one rRNA degradation fragment that was previously and erroneously classified as a plant miRNA, MIR2910, was reported at relatively low but consistent counts. However, this rRNA fragment is found in human 18S rRNA and is thus most simply explained as part of the human degradome. The other reportedly detected plant miRNAs were mostly found in a small minority of samples and in those were mapped at average read counts of less than one per million. These sequences may be amplification, sequencing, or mapping errors, since reads were mapped directly to plant (with one mismatch allowed) with no pre-filtering against mammalian genomes/transcriptomes. Several purported plant sequences, e.g., ptc-MIRf12412-akr and ptc-MIRf12524-akr, map perfectly to human sequences but do not appear to map to Populus or to other plants, suggesting that the plant miRNA database used by the authors and published in 2010 may include some human sequences. This is not a surprise, given pervasive low-level contamination in sequencing data, as reported by many authors.

Of course, even if some of the mapped sequences were genuine plant RNAs, they would be present in blood at greatly subhormonal levels unlikely to affect biological processes. No evidence of function is provided, apart from in silico predictions of human targets of the putative MIR2910 sequence, which, as noted above, is a human sequence. Thus, the titular claim of "evidences of cross-kingdom RNAi" is wholly unsupported. Overall, the results of this study corroborate the findings of Kang W, 2017 and previous studies: that dietary xenomiR detection is likely artifactual.

On 2017 Apr 11, Yu-Chen Liu commented:

The authors appreciate the insightful feedbacks and agree with prospect that hypothesis derived from small RNA-seq data analysis deserve examination in skeptical views and further experimental validation. Regarding the skeptical view of Prof. Witwer on this issue, whether a specific sequence were indeed originate from plant can be validated through examining the 2’-O-methylation on their 3’ end (Chin, et al., 2016; Yu, et al., 2005). The threshold of potential copy per cell for plant miRNAs to affect human gene expression was also discussed in previous researches (Chin, et al., 2016; Zhang, et al., 2012).

Some apparent misunderstandings are needed to be clarified:

In the commentary of Prof. Witwer:

“A cross-check of the source files and articles shows that the plasma data evaluated by Liu et al were from 198 plasma samples, not 410 as reported. Ninomiya et al sequenced six human plasma samples, six PBMC samples, and 11 cultured cell lines 19. Yuan et al sequenced 192 human plasma libraries (prepared from polymer-precipitated plasma particles). Each library was sequenced once, and then a second time to increase total reads.”

Authors’ response:

First of all, the statement "410 samples" within the article was meant to the amount of runs of small RNA-seq run conducted in the referred researches. Whether multiple NGS runs conducted on same plasma sample should be count as individual experiment replicates is debatable. The analysis of each small RNA-seq run was conduct independently. The authors appreciate the kind comments for the potential confusion that can be made in this issue.

In the commentary of Prof. Witwer:

“Strikingly, the putative MIR2910 sequence is not only a fragment of plant rRNA; it has a 100% coverage, 100% identity match in the human 18S rRNA (see NR 003286.2 in GenBank; Table 3). These matches of putative plant RNAs with human sequences are difficult to reconcile with the statement of Liu et al that BLAST of putative plant miRNAs "resulted in zero alignment hit", suggesting that perhaps a mistake was made, and that the BLAST procedure was performed incorrectly.”

Authors’ response:

The precursor sequences of the plant miRNAs, including the stem loop sequences (precursor sequences) were utilized in the BLAST sequence alignment in this work. The precursor sequence of peu-MIR2910, “UAGUUGGUGGAGCGAUUUGUCUGGUUAAUUCCGUUAACGAACGAGACCUCAGCCUGCUA” was used. The alignment was not performed merely with the mature sequence, “UAGUUGGUGGAGCGAUUUGUC”. The stem loop sequences, as well as the alignment of the sequences against the plant genomes, was taken into consideration by using miRDeep2 (Friedländer, et al., 2012). As illustrated in the provided figures, sequencing reads were mapped to the precursor sequences of MIR2910 and MIR2916. As listed in the table below, a lot of sequencing reads can be aligned to other regions within the precursor sequences except the sequencing reads aligned to mature sequences. For instance, in small RNA-seq data of DRR023286, 5369 reads were mapped to peu-MIR2910, and 4010 reads were mapped to the other regions in the precursor sequences.  

miRNA | Run |Total reads | on Mature | on precursor

peu-MIR2910 | DRR023286 | 9370 | 5369 | 4010

peu-MIR2910 | SRR2105454 | 3013 | 1433 | 1580

peu-MIR2914 | DRR023286 | 1036 | 19 | 1017

peu-MIR2916 |SRR2105342 | 556 | 227 | 329

(Check the file MIR2910_in_DRR023286.pdf, MIR2910_in_SRR2105454.pdf, MIR2914_in_DRR023286 and MIR2916_in_SRR2105342.pdf)

The pictures are available in the URL:

https://www.dropbox.com/sh/9r7oiybju8g7wq2/AADw0zkuGSDsTI3Aa_4x6r8Ua?dl=0

As described in the article, all reported reads mapped onto the plant miRNA sequences were also mapped onto the five conserve plant genomes. Within the provided link a compressed folder file “miRNA_read.tar.gz” is available. Results of the analysis through miRDeep2, were summarized in these pdf files. Each figure file was named according to the summarized reads, sequence run and the mapped plant genome. For example, reads from the run SRR2105181 aligned onto both Zea mays genome and peu-MIR2910 precursor sequences are summarized in the figure file “SRR2105181_Zea_mays_peu-MIR2910.pdf”.

In the commentary of Prof. Witwer:

“Curiously, several sequences did not map to the species to which they were ascribed by the PMRD. Unfortunately, the PMRD could not be accessed directly during this study; however, other databases appear to provide access to its contents.”

Authors’ response:

All the stem loop sequences of plant miRNAs were acquired from the 2016 updated version of PMRD (Zhang, et al., 2010), which was not properly referred. The used data were provided in the previously mentioned URL.

In the commentary of Prof. Witwer:

“Counts were presented as reads per million mapped reads (rpm). In contrast, Liu et al appear to have reported total mapped reads in their data table. Yuan et al also set an expression cutoff of 32 rpm (log2 rpm of 5 or above). With an average 12.5 million reads per sample (the sum of the two runs per library), and, on average, about half of the sequences mapped, the 32 rpm cutoff would translate to around 200 total reads in the average sample as mapped by Liu et al.”

Authors’ response:

Regarding the concern of reads per million mapped reads (rpm) threshold, the author appreciate the kind remind of the need to normalize sequence reads count into the unit in reads per million mapped reads (rpm) for proper comparison between samples of different sequence depth. However the comparison was unfortunately not conducted in this work. Given the fact that the reads were mapped onto plant genome instead of human genome, the normalization would be rather pointless, considering the overall mapped putative plant reads only consist of ~3% of the overall reads. On the other hand, the general amount of cell free RNA present in plasma samples was meant to be generally lower than within cellar samples (Schwarzenbach, et al., 2011).

Reference

Chin, A.R., et al. Cross-kingdom inhibition of breast cancer growth by plant miR159. Cell research 2016;26(2):217-228.

Friedländer, M.R., et al. miRDeep2 accurately identifies known and hundreds of novel microRNA genes in seven animal clades. Nucleic acids research 2012;40(1):37-52.

Schwarzenbach, H., Hoon, D.S. and Pantel, K. Cell-free nucleic acids as biomarkers in cancer patients. Nature Reviews Cancer 2011;11(6):426-437.

Yu, B., et al. Methylation as a crucial step in plant microRNA biogenesis. Science 2005;307(5711):932-935.

Zhang, L., et al. Exogenous plant MIR168a specifically targets mammalian LDLRAP1: evidence of cross-kingdom regulation by microRNA. Cell research 2012;22(1):107-126.

Zhang, Z., et al. PMRD: plant microRNA database. Nucleic acids research 2010;38(suppl 1):D806-D813.

On 2017 Apr 12, Kenneth Witwer commented:

Liu YC, 2017 reported mapping very low levels of mature plant miRNAs in a subset of public data gathered from 198 human plasma samples, concluding that this was evidence of "cross-kingdom RNAi"; however, both the authors and I observed that only one putatively foreign sequence, "MIR2910," mapped consistently and at levels above a reasonable noise threshold. No data were presented to support functional RNAi. I further noted that MIR2910 is a plant rRNA sequence, has been removed from miRBase, and also maps with 100% coverage and identity to human rRNA. In the comment below, Dr. Liu now links to unpublished predicted hairpin mapping data that were not included in the Liu YC, 2017 BMC Genomics conference article, which, like my comments, focused on mature putative xenomiRs. Dr. Liu states that mapping has been done not only to the putative MIR2910 mature sequence (as reported), but also to the predicted MIR2910 precursor hairpin sequence.

This is an interesting development, and I strongly and sincerely commend Liu et al for sharing their unpublished data in this forum. This is exactly what PubMed Commons is about: a place for scientists to engage in civil and constructive discourse.

However, examination of the new data reinforces my observation that the only consistently mapped "foreign" sequence in the Liu YC, 2017 study is a human rRNA sequence, not a plant miRNA, mature or otherwise. Beyond the 100% identity of the 21nt putative MIR2910 mature sequence with human rRNA, a 47nt stretch (80%) of the plant "pre-MIR2910" rRNA fragment aligns to human 18S rRNA with only one mismatch (lower-case), and indeed Liu et al allowed one mismatch:

Plant rRNA fragment:

UAGUUGGUGGAGCGAUUUGUCUGGUUAAUUCCGuUAACGAACGAGAC

Human rRNA fragment:

UAGUUGGUGGAGCGAUUUGUCUGGUUAAUUCCGaUAACGAACGAGAC

Dr. Liu provides the example of a plasma small RNA sequencing dataset, DRR023286, as the primary example of plant miRNA mapping, so let us examine this finding more closely. DRR023286 was by far the most deeply sequenced of the six plasma samples in the Ninomiya S, 2015 study (71.9 million reads), as re-analysed by Liu et al. Yet, like all other data examined in the Liu et al study, and despite the much deeper sequencing, DRR023286 yielded only the pseudo-MIR2910 as a clear "xenomiR" (mature or precursor). Of special note, previously reported dominant dietary plant xenomiRs such as MIR159, MIR168a, and the rRNA fragment "MIR2911" were not detected reliably, even with one mismatch.

The precursor coverage plots for DRR023286 (and less deeply sequenced datasets, for that matter), also according to the newly provided Liu et al data, show that any coverage is in the 5' 80% of the putative MIR2910 sequence: exactly the part of the sequence that matches human rRNA. The remaining 12 nucleotides at the 3' end of the purported MIR2910 precursor are conspicuously absent and never covered in their entirety. To give one example from the Liu et al data, in the deepest-sequenced DRR023286 dataset, even the single short read that includes just 11 of these 12 nucleotides has a mismatch. Furthermore, various combinations of these 3' sequences match perfectly to rRNA sequences in plant and beyond (protist, bacterial, etc.). Hence, the vanishingly small number of sequences that may appear to support a plant hairpin could just as convincingly be attributed to bacterial contamination...but we are already playing in the noise.

As noted, Liu et al allowed one mismatch to plant in their mapping and described no pre-filtering against human sequences. In the DRR023286 dataset, fully 90% of the putative mapping to plant MIR2910 included a mismatch (and thus human)...the other 10% were mostly sequences 100% identical to human.

In conclusion, the main points I raised previously have not been disputed by Liu et al:

1) numerous annotated plant miRNAs in certain plant "miRNA" databses appear to have been misannotated or the result of contamination, including some sequences reported by Liu et al that map to human and not plant;

2) the mature "MIR2910" sequence is a plant ribosomal sequence that also maps perfectly to human rRNA; and

3) the read counts for all but one putative plant xenomiR in the Liu et al study are under what one might consider a reasonable noise threshold for low-abundance RNA samples (plasma) with tens of millions of reads each.

Furthermore, the current interaction establishes a new point:

4) the plant rRNA sequence annotated by some as a "MIR2910" precursor maps almost entirely (and for all practical purposes entirely) to a human rRNA sequence.

Future, more rigorous searches for plant xenomiRs in mammalian tissues and fluids will require a pre-filtering step to exclude all sequences that map with one (or more) mismatches to all mammalian genomes/transcriptomes and preferably other possible contaminants, followed by a zero-mismatch requirement for foreign mapping.