Abstract
microRNAs (miRNAs) are a class of endogenously expressed small non-coding RNAs that regulate target genes at the post-transcriptional level. One significant feature of miRNA is that their genomic locations are often clustered together in the genome. In a previous study (Wang et al., 2016), we proposed a “functional co-adaptation” model to explain how clustering helps new miRNAs survive and develop functions during long-term evolution. In a manuscript recently posted at bioRxiv (doi:10.1101/274811), Marco claimed that he re-analyzed our data and came to a different conclusion. However, we found his analyses were conducted in an inappropriate approach. He also claimed that the absence of substitution in highly conserved miRNAs does not support the “functional co-adaption” model based on the misunderstanding of our model. In summary, the analyses and claims of Marco, which are flawed, do not refute our model.
miRNAs are a class of endogenously expressed small noncoding RNAs (~22 nt in length) that down-regulate the expression of target genes at the post-transcriptional level. A salient feature is that many animal miRNAs are clustered into discrete genomic regions (Lagos-Quintana et al. 2001; Lau et al. 2001; Lai et al. 2003; Altuvia et al. 2005; J. Graham Ruby et al. 2007; Marco et al. 2013; Mohammed, Siepel, et al. 2014). The clustering patterns suggest that miRNAs in the same cluster might be co-transcribed (Baskerville and Bartel 2005; Saini et al. 2007; Ozsolak et al. 2008; Wang et al. 2009; Ryazansky et al. 2011) and be functionally related by targeting the same gene or different genes in the same biological pathway (Bartel, 2004; Grun et al., 2005; Kim and Nam, 2006; Yu et al., 2006). For example, the mir-17~92 cluster plays an important role in mammalian development and tumorigenesis (Lu et al., 2007; O’Donnell et al., 2005; Ventura et al., 2008; Xiao et al., 2008). Gene deletion experiments suggest members in the mir-17~92 cluster have essential and overlapping functions (Ventura et al., 2008). The mir-106b~93~25 and mir-222~221 clusters are upregulated and modulate G1/S phase transition in gastric cancer, and members of the two cluster have functional associations by targeting genes in the Cip/Kip family members of Cdk inhibitors (Kim et al., 2009). The brain specifically expressed mir-379~410 cluster is required for the activity-dependent development of hippocampal neurons, and multiple miRNAs from the cluster are necessary for the correct elaboration of the dendritic tree (Fiore et al., 2009). miRNAs in mir-23a~27a~24-2 cluster also have cooperative effects in various health and diseased conditions (Chhabra et al., 2010).
We recently proposed a “functional co-adaptation” model to systematically investigate the functional relatedness of clustered miRNAs (Wang et al., 2016). We provided several lines of evidence to support the “functional co-adaptation” model. First, we found the observed number of genes co-targeted by miRNAs in the same cluster but with different seeds are significantly higher than the number obtained by random permutations. Second, we found genes targeted by multiple miRNAs from the same clusters, in general, have lower expression levels than what. Third, we show that the miRNAs in the same cluster with different seeds tend to target genes in the same biological pathways. Fourth, we transfected four members of the mir-17~92 cluster into human 293FT cells individually and quantified the alteration of mRNA abundance with deep-sequencing, which verified the overlapping of target genes experimentally. Fifth, we also experimentally determined the target genes of miR-92a, the founding member of the mir-17~92 in Drosophila, and examined the relationship between the target genes of miR-92a in Drosophila and the target genes of the mir-17~92 cluster in humans. Our experimental results well supported the “functional co-adaptation” model. Finally, we also conducted evolutionary analysis to show that positive Darwinian selection drives the evolution of the newly formed miRNA clusters in both primates and Drosophila.
In a manuscript recently posted at bioRxiv (Marco, 2018; doi:10.1101/274811), Marco claimed that he re-analyzed our data and found “No evidence of functional co-adaptation between clustered microRNAs”. Macro claimed that the observed overlap of target genes by the clustered miRNAs are mostly caused by the similarity between two seed sequences in the miR-182/183/96 cluster. Marco argued that clustered miRNAs from different miRNA families do not share more targets than expected by chance after correcting for these factors. Marco also raised a series of critiques about the “function co-adaptation” model. As we already responded to Marco’s critique elsewhere, his critiques are based on the analysis that was conducted in an inappropriate manner, which can be summarized as follows:
First, Marco misunderstood the “functional co-adaption” model, which led him to make the argument that “microRNAs in a cluster are primarily under positive selection”. Although our population genetic analysis suggests Darwinian selection drives the evolution of the newly formed miRNA clusters in primates and in Drosophila, our model does not necessarily suggest all the clustered miRNAs are driven by positive selection (Wang et al., 2016). What we proposed is that, new miRNAs originate nearby a pre-existed miRNA would have higher chance to be maintained in the initial stage of cluster formation due to the tight genetic linkage. Then positive Darwinian selection might drive the newly emerged miRNAs to develop functions related to the pre-existed miRNAs in the same cluster or drive the evolution of all the new miRNAs in the same cluster to develop related functions during the long-term evolution. Once the cluster is fully established, the miRNAs in the same cluster will be maintained by purifying selection and become highly conserved after that (Wang et al., 2016). Thus, one could not expect to observe signature of ongoing positive selection in the well-established clusters such as the miR-17~92 or the miR-182/183/96 cluster which are ancient and conserved after the establishment, as Marco did. Marco’s observation that “both seed sequences (of miR-183-5p and miR-96-5p) have been conserved since their origin and, therefore, there is no evidence of substitutions happening in the seed of these microRNAs for the last 600 million years” could not refute our model. Also, he also used the deep conservation of the clustered miRNAs in other clusters (mir-106b~25 cluster, mir-23b~24 cluster, and mir-379~410 cluster of Fig. S1 in his manuscript) to argue against the “functional co-adaptation” model. The deep conservation of the seed sequences as Marco showed can only suggest these miRNAs are conserved due to extremely strong selective constraints during vertebrate evolution. These observations do not provide any evidence to defy the “functional co-adaptation” model since one cannot tell whether the miRNAs have changed since emergence as no outgroup sequence available.
Second, Marco did not properly conduct the permutation test. The major concern Marco raised is whether the observed number of genes targeted by at least two conserved miRNAs with different seeds from the same miRNA clusters is statistically higher than the number obtained under the assumption of randomness. We found 1,751 human genes were conserved targets of at least two distinct miRNAs (with different seeds) of the same miRNA cluster (Wang et al., 2016). The number obtained by Marco was 1,963, which is very similar between these two studies. However, Marco has conducted biased and flawed permutation test processes, which generated a pattern that the observed number slightly but still significantly higher than the expected number under the assumption of randomness (P = 0.0359). Since the difference between the observed and expected numbers are quite smaller obtained by Marco compared to what we obtained (Wang et al., 2016), Marco argued that the difference observed by Wang et al. (2016) was caused by the similarity of the targets between two seed sequences of the mir-183~182 cluster, and “the expected high number of common targets between pairs of microRNAs that have a large number of targets each”.
The discrepancy mainly lies in the permutation test procedures. In our previous study (Wang et al., 2016), to test whether miRNAs in the same clusters tend to regulate overlapping sets of genes, we obtained expression profiles of miRNAs and mRNAs from five tissues of human males as determined in previous study (Brawand et al., 2011; Meunier et al., 2013). Since the co-adaption of clustered miRNAs is the result of co-evolution between miRNA and target sites, in the permutation analysis, we first shuffled the co-expressed seed:target pairing (TargetScan PCT > 0.5), and then we tested how many genes were targeted by at least two miRNAs (with distinct seeds) in the same clusters. These permutation tests were performed for 1,000 replicates. By this way, the conservation level and length of 3’ UTR of target mRNAs, the number of miRNAs for each target gene, and the compositions of each miRNA cluster are fully controlled. Applying this procedure to the pooled dataset of miRNA-mRNA co-expression from different tissues, the result of Wang et al. was successfully reproduced (Fig. 1). When the co-expression data of each tissue was analyzed individually, the similar pattern was still observed (Fig. 1). Importantly, when the mir-182~183 cluster was excluded, we can still observe similar patterns (Fig. 2). Therefore, Marco’s argument that “the high overlap between targets in some clustered miRNAs is actually the random consequence of the similarity between their seed sequences, and is no associated to whether the miRNAs are clustered or not.” is invalid. Here, the condition to be tested is whether miRNAs with distinct seeds from the same cluster have more common target genes than expected under randomness, rather than to test whether miRNAs are clustered. Marco failed to reproduce our results because he only shuffled the location of the miRNAs and kept the seed: targeting pairing unaltered.
Furthermore, it is hard to understand why Marco argued that the targets shared between miR-183 and miR-96 in the mir-183~182 cluster should be excluded in the analysis. The seeds of miR-183-5p and miR-96-5p are very similar: AUGGCAC and UUGGCAC for the former and latter, respectively. However, BLAST2SEQ analysis between the precursor sequences of human mir-183 and mir-96 does not find significant similarity, suggesting these two miRNA precursors are unlikely to be duplicated miRNAs. Instead, the functional co-adaptation model might well explain the large number of target genes shared between these two miRNAs: During long time evolution, the adaptive changes in miRNA seed region or target sites on mRNAs drive the clustered miRNAs to regulate the same or functionally related genes. Therefore, this cluster serves as a strong evidence that convergent evolution has occurred between the seeds of miR-183 and miR-96 due to functional coadaptation.
Curiously, Macro did not report his re-analysis results of the miR-17~92 cluster over-expression data we generated (Wang et al., 2016). Many previous studies have demonstrated that the mir-17~92 cluster plays an important role in tumorigenesis, development of lungs and immune systems (Lu et al., 2007; O’Donnell et al., 2005; Ventura et al., 2008; Xiao et al., 2008), and deletion of the mir-17~92 cluster revealed that miRNAs in this cluster has essential and overlapping functions (Ventura et al., 2008). Furthermore, we found the conserved target genes shared between members of the mir-17~92 cluster is significantly higher than the simulated ones (Fig. 3). Importantly, we selected four distinct mature miRNAs in the miR-17~92 cluster (miR-17, miR-18a, miR-19a, and miR-92a) and transfected each miRNA mimic as well as the miRNA mimic Negative Control (NC) into human 293FT cells (Wang et al., 2016). With high-throughput mRNA-Seq, we found the predicted target genes (TargetScan PCT > 0.5) of each transfected miRNA are significantly more down-regulated than genes that do not have the target sites (Figure 4 of Wang et al. 2016). We identified 301, 55, 345 and 268 high-confidence target genes (TargetScan PCT > 0.5) for miR-17, 18a, 19a and 92a respectively that were down-regulated with log2(FoldChange) < -0.1 in the corresponding miRNA transfection experiments (totally 775 high-confidence genes after removing overlapping genes, Figure 4I of Wang et al. 2016). Among these 775 high-confidence target genes, 172 were targeted by at least two out of the four miRNAs, significantly higher than the number obtained by randomness (P < 0.001, see Figure 4I and Table S8 of Wang et al. 2016 for details). These results well support the “functional co-adaptation model” we proposed. If there is really “No evidence of functional co-adaptation between clustered microRNAs” as Macro argued, how can one explain these observed patterns?
Based on the observation that Drosophila new miRNAs often arose around the pre-existing ones to form clusters, Marco and colleagues proposed a “drift-draft” model which suggests that the evolution of miRNA clusters was influenced by tight genetic linkage and largely non-adaptive (Marco et al., 2013). Under such a model, the motifs of the pre-existing miRNAs would protect new miRNAs to be transcribed and processed properly since those motifs were already interacting with the miRNA processing machinery. Thus, the de novo formed new miRNAs are sheltered by the established ones in the same cluster because mutations that abolish the transcription or processing of the new miRNA will affect the pre-existing ones as well and are hence selected against. On the other hand, if a de novo formed miRNA is located in a discrete locus, it will have a higher probability to degenerate, either by mutations abolishing its transcription or by mutations impairing its processing. Although Marco argued the “drift-draft” and “functional co-adaptation” models are mutually exclusive, we did not think the “functional co-adaptation” we proposed is strictly “an alternative to the drift-draft model”.
Our previous results and others suggest that many newly-emerged miRNAs are evolutionarily transient, with a high birth-and-death rate (Berezikov et al. 2006; Rajagopalan et al. 2006; Lu, Shen, et al. 2008; Lu et al. 2010). Therefore, it is possible that the newly emerged miRNAs in the clusters would be sheltered by the pre-existing established miRNAs. However, the protection effect alone cannot explain why miRNAs in the same cluster have significantly higher numbers of overlapping target genes. Moreover, many de novo formed novel miRNAs will degenerate even after they are fixed in the populations if they are not maintained by functional constraints (Berezikov et al. 2006; Lu, Shen, et al. 2008). Thus developing functions related to the pre-existing miRNAs will help the novel miRNAs to survive and stabilize. The “functional co-adaptation” model we proposed well account for the evolution and function of de novo formed new miRNAs in the clusters (Figure 2D of Wang et al. 2016). Since miRNAs in the same clusters are usually co-transcribed temporally or spatially (see below for details), the newly formed miRNAs might gradually develop functions to target genes that are related to the pre-existing miRNAs in the same cluster; or multiple de novo formed new miRNAs in the same cluster interplay to regulate overlapping sets of target genes. Therefore, although miRNAs in the same cluster have independent origins, they might regulate overlapping sets of target genes through convergent evolution. After that the clustering patterns of miRNAs and the modular regulation of target genes will be stabilized by natural selection during long-term evolution. Of course, the evolutionary process of miRNAs is also companied by the co-evolution of the target sites. In a separate study, we also showed the target sites of miRNAs also experienced frequent births and deaths (Luo et al., 2018). But the evolution of the target sites alone would not cause the clustering pattern of miRNAs. A plausible scenario is that after a new miRNA originates in a cluster, the substitutions that change the sequences and expression of the new miRNAs, the interactions between miRNAs in the same cluster, and the co-evolution between miRNAs and the target sites, jointly affect the evolution of the clustering pattern of miRNAs.
Understanding the molecular mechanisms and evolutionary principles of the miRNA clustering would deepen our understanding of the regulatory roles of miRNAs in various biological processes or diseases. The “functional co-adaptation” model we propose is well supported by evolutionary and functional genomic data.