Reply to Backes and Keller: Identification of novel tissue-specific and primate-specific human microRNAs

We thank Backes and Keller (1) for their comments on our recent publication (2).

With regard to the first example in their cluster analysis point, it is great to see that Backes and Keller (1) uncovered more clusters among our reported data, as this provides further support for the nonaccidental nature of these molecules. With regard to the second example, where they did not find a direct overlap, it is indeed because of variations in individual samples, as Backes and Keller already speculated; there are a handful of such instances that include 8 of the 13 clusters mentioned by Backes and Keller.

Regarding the inclusion of cluster 28 as a miRNA cluster, as we state in our report (2), we made use of miRBase Release 20. In fact, our analyses are based on the hsa.gff3 file (“General Feature Format”) that was made available on June 24, 2013 as part of miRBase Release 20 on that day: this .gff3 file lists no mature miRNA entry for the miR-486-2 precursor, making this case tantamount to our discovery of an unannotated miRNA-producing arm in a previously reported precursor. Release 21 of miRBase, which was not used in our study, contains an updated .gff3 entry for miR-486-2.

As to Backes and Keller’s (1) third point regarding the merging of overlapping novel precursors: as we state in our report (2), we performed miRNA discovery by applying miRDeep2 separately to each dataset. In the example shown by Backes and Keller (1), the two mature miRNAs and their corresponding precursors were identified separately and in different subsets of the analyzed datasets, and thus should not be examined as a single sequence. Additionally, this particular example by Backes and Keller argues for processing each dataset separately (which is precisely what we did) and against “dataset pooling.” Indeed, had we carried out discovery after pooling datasets we might have missed this signal and would have unnecessarily increased the chances of erroneous predictions. We note that our findings in this regard are concordant with figure 3 of our report (2), where we show that the newly discovered miRNAs exhibit much stronger tissue specificity than miRBase miRNAs. As described in our Materials and Methods (2), when precursors that were discovered independently in different datasets overlapped on the genome, we opted to report a single “island” for them (which avoids artificially inflating the reported miRNA counts): in these cases, the on-line data repository accompanying our article (hyperlinks are listed in column T of dataset S2 of ref. 2) shows each mature miRNA with its matching precursor separately to reflect their distinct transcriptomic history.