Sex-biased microRNA expression in mammals and birds reveals underlying regulatory mechanisms and a role in dosage compensation

Identification of miRNAs with sex-biased expression in mouse, opossum, and chicken

To identify miRNAs with sex-biased expression in mouse, gray short-tailed opossum, and chicken (red jungle fowl), we prepared and sequenced a total of 72 small RNA libraries corresponding to brain, heart, liver, and gonad (ovary and testis) samples from three male and three female adult individuals per species (Methods). To mitigate the extensive sequencing biases that have been identified for standard library preparation methods (Sorefan et al. 2012; Baran-Gale et al. 2015), we chose to prepare the libraries with the NEXTflex Small RNA-seq Kit (Methods), which improves the ligation step through the use of partially degenerate adapters (Baran-Gale et al. 2015). We found that this protocol allowed substantially broader coverage of expressed miRNAs compared to standard methods (Supplemental Fig. S1). After adapter trimming and filtering based on quality scores and read length, each library was represented by an average of 19.4 million high-quality reads with a size of 15–28 nt. Novel miRNA precursors were predicted for each library using miRDeep2 (Friedländer et al. 2012), with a cutoff score of 5, corresponding to a signal-to-noise ratio of at least 10:1 in all tissues except chicken testis (Methods). Only precursors that were identified in at least three of six tissue replicates (male or female) and that did not overlap with other noncoding RNAs (ncRNAs) were kept for further analysis (Methods). Genomic coordinates of all known and putative novel miRNA loci investigated in this study are provided in Supplemental Table S1. All novel miRNAs loci have furthermore been submitted to miRBase (Kozomara and Griffiths-Jones 2014) and their assigned miRBase IDs are listed in Supplemental Table S2. After an additional filtering step performed by miRBase, 40 mouse, 220 opossum, and 135 chicken miRNA precursors were added to the database.

Next, we estimated miRNA expression levels by remapping all reads to the respective genome using Bowtie (Langmead et al. 2009) and a sequence of decreasingly stringent settings (Methods) to allow for mismatches due to miRNA editing and 3′ modifications (Kawahara et al. 2007; Burroughs et al. 2010; Chiang et al. 2010; Warnefors et al. 2014). Because some miRNAs belong to multicopy families with high sequence similarity, it was not always possible to pinpoint the precise genomic locus from which a miRNA read was derived. We therefore grouped mature miRNAs into multimap groups (MMGs; Methods), which were thereafter treated as single expression units (Robert and Watson 2015). We chose a relatively conservative MMG cutoff of 5%, meaning that if a miRNA shared >5% of its total reads with another miRNA, these two miRNAs were assigned to the same MMG. Reads that mapped two MMGs (i.e., <5% cutoff) were divided equally between the MMGs. The 843 mouse, 893 opossum, and 564 chicken MMGs are listed in Supplemental Table S3. MMGs were considered expressed in a given tissue if they contributed at least 10 transcripts per million miRNA-mapped reads (TPM) in three of six replicates.

For each tissue and species, we then used DESeq2 (Love et al. 2014) to identify miRNAs with sex-biased expression at a global false discovery rate (FDR) of 5% (Methods). Differential expression between ovary and testis samples was analyzed and corrected separately due to the mixture of sex-related and tissue-related effects, which otherwise reduced the stringency of the significance test for the somatic samples (Methods). An overview of sex-biased miRNA expression per tissue is given in Figure 1, with complete details available in Supplemental Table S4. The most male-biased and most female-biased miRNA for each species and tissue are listed in Table 1. We did not identify any individual miRNAs that were sex-biased in a somatic tissue in more than one species (Discussion), although several were sex-biased in multiple tissues from a single species (Supplemental Table S4).

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Figure 1.

Sex-biased miRNA expression in mouse, opossum, and chicken. Male-to-female (M:F) ratios were calculated for three male and three female replicates per tissue and the significance assessed with DESeq2 (Love et al. 2014), with Benjamini-Hochberg correction for multiple tests (Benjamini and Hochberg 1995) performed separately for somatic and gonadal samples. For display purposes, the log₂-transformed M:F ratios were capped at −3 and 3, with more extreme ratios replaced by these values. Each data point corresponds to a multimap group (MMG) comprised of one or more mature miRNAs (see main text), and MMGs were considered sex-linked if they included at least one sex-linked member. Expression values correspond to the mean normalized counts provided by DESeq2. The most sex-biased miRNAs per tissue are listed in Table 1. The Z-linked and W-linked miRNA miR-122-5p is visible as the rightmost data point in the chicken liver panel. All data shown in this figure are included in Supplemental Table S4.

Independent transcription drives sex-biased miRNA expression in mouse tissues

We found evidence of sex-biased miRNA expression in all of the investigated somatic tissues (Fig. 1), with the largest proportion found in mouse heart (13%, 33 significant MMGs of 253 tested) and liver (8.6%; 19 of 222). These findings are consistent with the large numbers of protein-coding genes that are expressed in a sex-biased manner in various mouse tissues, notably with more pronounced differences in liver compared to brain (Yang et al. 2006). Although it is well established that many X-linked miRNAs are male-biased with high expression levels in the mouse and opossum testis (Song et al. 2009; Meunier et al. 2013), and this trend was visible also in our data (Fig. 1; Supplemental Table S4), we found that most sex-biased miRNAs in somatic tissues were autosomal (Fig. 1). The expression pattern of these miRNAs thus cannot be explained by escape from X inactivation or other effects immediately linked to the presence of sex chromosomes, but is instead likely achieved through direct or indirect regulation by hormones or other sex-biased factors.

We speculated that one regulatory route to sex-biased miRNA expression might simply be coexpression with a sex-biased protein-coding gene, especially since it is known that many miRNAs, including both those generated via canonical and noncanonical pathways, are located within the introns of protein-coding genes and transcribed from the same promoter as their host (Rodriguez et al. 2004; Baskerville and Bartel 2005; Berezikov et al. 2007; Meunier et al. 2013). In our data, there was considerable potential for coexpression, with 9 of 33 (27%) MMGs with significantly sex-biased expression in mouse heart having at least one intronic member, and the corresponding numbers for mouse liver were 7 sex-biased MMGs of 19 (37%). However, when we compared the sex-biased expression of intronic miRNAs with that of their host genes, based on expression data from Marin et al. (2017) (see also Methods), we did not find any significant correlation between miRNA and host gene male-to-female (M:F) ratios in heart, and only a weak correlation in liver, which was no longer significant after correction for multiple tests (Fig. 2A). These results indicate that miRNA sex bias is not in general a secondary effect caused by sex-biased regulation of protein-coding genes in somatic tissues, although a stronger correlation was seen for the gonads (Fig. 2A). The male-biased expression of miR-1948-5p/3p and miR-455-5p/3p in mouse liver nevertheless appeared to constitute an exception, since the host genes Ttc39c and Col27a1 were male-biased to a similar degree (Fig. 2A). However, we found that both miRNA loci were associated with mouse liver DNase I hypersensitivity (DHS) sites identified by Ling et al. (2010) (Methods), which is an indication that they possess independent regulatory elements and can be transcribed separately from the host. Although the possibility remains that some miRNAs are transcribed from both miRNA-specific and host-associated promoters (Ozsolak et al. 2008), these observations suggest that the sex bias we observe for mouse miRNAs occurs in parallel to the sex bias of protein-coding genes, but for the most part, is not a direct consequence thereof. Similar decoupling of intronic miRNA expression from that of host genes due to independent transcription and/or post-transcriptional processing has previously been observed across mammalian tissues (Ozsolak et al. 2008; Wen et al. 2015).

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Figure 2.

Transcriptional mechanisms underlying sex-biased miRNA expression. (A) Correlation between sex-biased expression of intronic miRNAs (y-axis) and their host genes (x-axis). Male-to-female (M:F) expression ratios were calculated with DESeq2. For each tissue, the Spearman correlation coefficient (rho) is given, together with its Benjamini-Hochberg corrected P-value (Benjamini and Hochberg 1995). (B) Overview of the two miRNA loci that were associated with sex-biased DHS regions (Ling et al. 2010) in mouse liver. In both cases, a sequence of 1 kb is depicted. The displayed miRNA precursor region corresponds to the 5p and 3p sequences with the intervening loop sequence. (C) Expression levels in mouse liver of the 5p and 3p mature miRNAs shown in B, for three female and three male replicates. Raw read counts were normalized with DESeq2. Statistical significance: (*) P < 0.05; (**) P < 0.01; (***) P < 0.001.

Although these analyses suggest that the majority of sex-biased miRNAs are transcribed from independent promoters, they do not address whether these promoters are themselves targets of sex-biased regulation. The DHS site that we were able to associate with the mir-1948 locus had previously been annotated as male-biased (Ling et al. 2010), which would be consistent with the male-biased expression of miR-1948-5p and -3p (Fig. 2A). This finding was nevertheless somewhat surprising, given that sex-biased regulation of protein-coding genes is thought to primarily take place at distally located regulatory elements rather than promoters, with 80% of all sex-biased DHS sites being located >100 kb away from sex-biased genes (Ling et al. 2010). To investigate the role of sex-biased promoters further, we intersected the coordinates of all 185 liver-expressed single-member MMGs (i.e., miRNAs for which most sequencing reads map to a single genomic locus) with genome-wide DHS annotations from Ling et al. (2010) (Methods). We could associate 95 miRNAs with 61 DHS sites within 5 kb upstream or overlapping the miRNA (P < 0.001 based on 1000 permutations of the DHS site coordinates), and were able to locate one additional miRNA-associated sex-biased DHS site, this time a female-biased site associated with the equally female-biased mir-802 locus (Fig. 2B). In spite of the limited sample size, the association between sex-biased DHS sites and sex-biased miRNA loci was significant (two of two sex-biased and six of 59 unbiased DHS sites were associated with sex-biased miRNAs, P = 0.015, Fisher's exact test), which suggests that, for a subset of miRNAs, sex-biased expression is likely associated with differential promoter accessibility in males and females.

Overall, our findings point to diverse transcriptional mechanisms behind sex-biased miRNA expression in the mouse, including potential piggybacking on sex-biased host genes in gonads and possibly liver, regulation at miRNA promoters, and presumably, regulation at distally located regulatory elements, such as enhancers. Conceivably, an additional level of regulation might be achieved through differential processing of miRNA transcripts in males and females (Bhat-Nakshatri et al. 2009; Nothnick et al. 2010). Given that 5p/3p pairs from the same miRNA locus frequently have highly similar M:F ratios (e.g., miR-1948-5p/3p, miR-455-5p/3p, and miR-802-5p/3p) (Fig. 2), we consider it more likely that such sex-biased regulation, if present, would affect the general efficiency of miRNA biogenesis rather than modulate the balance between 5p and 3p mature miRNAs through arm switching (Chiang et al. 2010).

Somatic miRNA sex bias mirrors gonadal expression in mouse but not chicken

In addition to the sex-biased miRNA expression found in somatic tissues, we consistently found pronounced sex differences in the gonads (Fig. 1). The magnitude of these differences is not surprising, given that ovary and testis transcriptomes are shaped by sex-biased as well as tissue-biased factors (Discussion). Previous studies have shown that the X Chromosome of mouse and opossum is enriched for recently expanded testis-expressed miRNA families, whereas no similar enrichment exists for the chicken Z Chromosome (Guo et al. 2009; Meunier et al. 2013). The mechanistic basis of this phenomenon has been a matter of intense study (Song et al. 2009; Royo et al. 2015; Sosa et al. 2015). As previously noted, the surplus of X-linked, testis-biased miRNAs in both mouse and opossum was clearly visible also in our data (Fig. 1; Supplemental Table S4). Even so, the majority of ovary-biased and testis-biased miRNAs in our data set were autosomal, and a large proportion of them could be detected in at least one somatic tissue, although testis-biased autosomal miRNAs were significantly less likely to be somatically expressed (Fig. 3A). The relative lack of shared miRNA expression between testis and somatic tissues fits well with known expression patterns of mRNAs and long noncoding RNAs, which are more frequently testis-specific than ovary-specific in all three species investigated here (Necsulea et al. 2014). However, it should be noted that ovary samples contain a larger proportion of somatic cells, and the power to detect ovary-specific miRNAs therefore might be reduced.

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Figure 3.

Somatic expression of ovary-biased and testis-biased miRNAs. (A) Proportion of autosomal MMGs with ovary-biased or testis-biased expression that are detected in at least one somatic tissue. (B) Somatic sex bias of ovary-biased and testis-biased miRNAs. If an MMG was detected in more than one somatic tissue, the most extreme M:F ratio was chosen. Statistical significance: (**) P < 0.01; (***) P < 0.001; (n.s.) not significant.

We were interested in whether miRNAs were consistently sex-biased between gonads and somatic tissues, as might be the case if they fulfilled sex-specific functions across a variety of conditions. To this end, we compared the degree of somatic sex bias for ovary-biased and testis-biased autosomal miRNAs (Fig. 3B), while choosing the tissue with the strongest sex bias if a miRNA was expressed in multiple somatic tissues (Methods). In mouse, we found that ovary-biased miRNAs tended to be female-biased in somatic tissues (median M:F ratio: −0.29), whereas testis-biased miRNAs tended to be male-biased (median M:F ratio: 0.13, P = 3.4 × 10⁻⁵, Mann-Whitney U test with Benjamini-Hochberg correction [Benjamini and Hochberg 1995]). A similar, although not significant trend was hinted at in the opossum data (P = 0.30), but was not observed in chicken (P = 0.93). These findings thus point to a more general coordination of sex-biased expression in mouse, which is largely absent in the other species, especially chicken, although we did notice that a novel, autosomal miRNA (gga-miR-2184a-5p) was the second most female-biased miRNA in chicken liver and was also significantly female-biased in heart and gonads (Supplemental Table S4). Although we were unable to address the root of the apparent discrepancy between mouse and chicken with the present data, it is tempting to speculate that hormonal regulation might facilitate joint control of miRNA expression across tissues. Indeed, many mammalian miRNAs are known to be regulated by gonadal hormones (Yang and Wang 2011; Klinge 2012). The majority of these studies were carried out in human cell lines, but there is considerable overlap with the miRNAs for which we observe significant sex-biased expression in our mouse data, including miR-17-3p and miR-19a-3p from the mir-17-92 cluster (Castellano et al. 2009), and miR-451a-5p (Bergamaschi and Katzenellenbogen 2012). In contrast to the seemingly widespread contributions of gonadal hormones to miRNA regulation in placental mammals, they play a reduced role in the establishment of male and female somatic tissues in chicken (Zhao et al. 2010), which would be consistent with the lack of coordinated sex bias in this species. Global studies of the hormonal impact on miRNA expression will however be required to determine whether the coordination of miRNA sex bias is driven by gonadal hormones and whether the effect is direct or indirect.

To search for functional commonalities of sex-biased miRNA regulation across mouse tissues, we analyzed Gene Ontology annotations of miRNA target genes using target predictions from TargetScan (Agarwal et al. 2015) and the GOrilla tool (Eden et al. 2009) (Methods). Only protein-coding genes with a minimum expression of 1 FPKM (fragments per kilobase of transcript per million mapped reads) in the tissue of interest were considered based on data from Marin et al. (2017). To avoid the frequent false positives associated with functional analyses of miRNA targets (Bleazard et al. 2015), we compared targets of sex-biased miRNAs only to targets of unbiased miRNAs instead of to all genes expressed in the tissue of interest (Methods). We further chose to focus on mouse heart and liver for this analysis, since the extent of sex-biased miRNA expression was too limited in brain and too extensive in gonads to permit a reasonable division of protein-coding genes into targets of male-biased, female-biased, and unbiased miRNAs, respectively. Although we did not observe any significant functional enrichment for targets of female-biased miRNAs, we did identify a number of Gene Ontology terms associated with male-biased miRNAs in both heart and liver (Supplemental Table S5). These terms were highly similar for the two tissues, with the top liver category “regulation of cell communication” being the eleventh most enriched term in heart, and the top heart category “positive regulation of cellular process” being the fourth most enriched term in liver. The potential functional parallels between targets of sex-biased miRNAs in heart and liver are noteworthy in light of the apparent coordination of sex-biased miRNA expression in the mouse and underline the value of conducting screens for sex-biased genes across multiple tissues.

Gene retention on the chicken W Chromosome allows nonbiased expression of miR-122-5p

Unlike mouse and opossum, chicken does not have a global mechanism of dosage compensation, meaning that Z-linked protein-coding genes generally tend to be more highly expressed in males compared to females (Ellegren et al. 2007; Itoh et al. 2007; Mank and Ellegren 2009; Julien et al. 2012). The most prominent example of a Z-linked, male-biased miRNA was miR-2954-3p (Table 1; see next section), but the association between Z linkage and male-biased expression did not hold for all miRNAs on the chicken Z Chromosome. A particularly striking exception was miR-122-5p, which was the most abundant miRNA in the chicken liver and almost completely unbiased between males and females (Fig. 1; Supplemental Table S4). The expression pattern of miR-122-5p in chicken is thus similar to miR-122-5p expression in mouse and opossum (Supplemental Table S4), although the mir-122 locus is autosomal in these species. Given that miR-122-5p is expressed at extreme levels (miR-122-5p contributed >60% of all miRNA-mapped reads per chicken liver library), it is difficult to imagine how sufficient dosage compensation could be achieved through regulatory means.

However when we remapped miRBase miRNAs to the most recent chicken genome release (galGal5) as part of our annotation procedure, we were able to locate a copy of the mir-122 locus on the updated W Chromosome, the presence of which likely grants equal miR-122-5p dosage in both sexes (a third mir-122 copy is present on contig chrZ_NT_463593v1_random, although this might be an assembly error). This miRNA therefore joins the highly select group of dosage-sensitive genes that were retained on the W Chromosome (Bellott et al. 2017). We did however detect significant female bias for miR-122-5p in the chicken heart, which could be due to sex-specific control or may indicate slight regulatory diversification of the Z-linked and W-linked copies in this tissue. Similar sex bias of miR-122-5p was previously observed in the zebra finch heart based on a comparison of one male and one female individual (Luo et al. 2012). In addition, remapping confirmed W linkage of the mir-7b locus (Ayers et al. 2013), which belongs to a family with both Z-linked and autosomal members. Given that the chicken W Chromosome is thought to host only 28 protein-coding genes (Bellott et al. 2017), miRNAs thus make up a sizable fraction of the total W-linked gene repertoire.

Beyond mir-122 and mir-7b, we did not detect any additional W-linked or Y-linked miRNAs in any species using remapping of known miRNAs (Methods) or de novo annotation with miRDeep2, although the latter did predict two novel miRNAs (gga-mir-12266-2 and gga-mir-12266-3) on W-labeled contigs. Because these methods would not pick up miRNAs that lack an equivalent on the X or Z Chromosome and were generated through an atypical mechanism (Yang et al. 2010), we additionally probed the data set for sex-specific reads derived from the Y or W Chromosomes (Methods). However, although we successfully identified the mir-7b locus in this manner, we did not recover reads from the putatively W-linked loci mentioned above, suggesting they do not represent bona fide W-linked miRNA genes (Supplemental Fig. S2). Although we did identify a number of male-specific reads in mouse and opossum, many male-specific reads were also present in chicken, indicating that most such reads might not be Y-derived, but instead a consequence of the transcriptional complexity of the testis (Soumillon et al. 2013). Indeed, when we mapped the male-specific reads from mouse to the Y Chromosome, not a single match was found (Supplemental Fig. S2). We therefore consider it unlikely that additional miRNAs, unique to the Y or W Chromosomes, are robustly expressed in the tissues studied here.

Male-biased expression of miR-2954-3p serves as a gene-specific dosage compensation mechanism in birds

Similar to previous reports (Zhao et al. 2010; Luo et al. 2012; Lin et al. 2014), we observed male-biased expression of the Z-linked miR-2954-3p (Fig. 4A) in all chicken tissues. Although the mir-2954 locus is located within an intron of the XPA gene, XPA is not male-biased to the same extreme degree (Methods; Marin et al. 2017). Notably, the male bias of miR-2954-3p is considerably more pronounced (up to almost eightfold in chicken liver) than the maximally twofold difference that is expected between males and females due to gene dosage alone (Lin et al. 2014). The high expression in chicken males is therefore likely a result of increased production and/or stability of miR-2954-3p, in combination with the dosage effect. The biological relevance of such extreme male bias is nevertheless not currently understood. Although previous work in zebra finch has elucidated the involvement of miR-2954-3p in song response (Gunaratne et al. 2011; Lin et al. 2014), these observations cannot explain why miR-2954-3p is male-biased in neural as well as non-neural tissues both in zebra finch and a non-songbird such as chicken (Fig. 4A). To understand the functional impact of miR-2954-3p, we therefore used TargetScan (Agarwal et al. 2015) to predict target genes in the chicken genome (Methods). Consistent with an earlier study (Luo et al. 2012), targets of miR-2954-3p were markedly enriched on the Z Chromosome (Fig. 4B), particularly when we considered only 8-mer sites, which are the strongest type of target site (Agarwal et al. 2015). For this site type, 47 of the 129 (36.4%) predicted miR-2954-3p targets were on the Z Chromosome, whereas the corresponding number for targets of all autosomal miRNAs was 327 of 7185 (4.6%), corresponding to an eightfold enrichment (P = 2.9 × 10⁻⁵⁸, χ²-test).

View larger version:

• In this window
• In a new window

• Download as PowerPoint Slide

Figure 4.

Role of miR-2954-3p in dosage compensation. (A) Expression of miR-2954-3p in chicken tissues. Raw read counts were normalized with DESeq2. (B) Proportion of Z-linked genes among the genes that were predicted to be targets of miR-2954-3p and autosomal miRNAs. Target prediction was performed with TargetScan and only 8-mer sites were considered. (C) Log₂-transformed M:F expression ratios for Z-linked target genes of miR-2954-3p or autosomal miRNAs. (D) Proportion of annotated ohnologs between Z-linked targets of miR-2954-3p and autosomal miRNAs. (E) Log₂-transformed expression ratios for protein-coding genes following miR-2954-3p knockdown compared to control. Genes that are down-regulated by miR-2954-3p are expected to have positive ratios. The first panel shows Z-linked genes in purple and autosomal genes in gray. The following two panels show Z-linked genes predicted to be more (light purple) or less (dark purple) dosage-sensitive based on chicken expression data and ohnolog annotations. Only genes that were 1-to-1 orthologs and located on the Z Chromosome in both chicken and zebra finch were included in the analyses. Statistical significance: (*) P < 0.05; (**) P < 0.01; (***) P < 0.001.

Although there is no chromosome-wide dosage compensation in chicken, Z-linked genes differ in their M:F expression ratios, with a tendency for dosage-sensitive genes to be more equally expressed between the sexes (Zimmer et al. 2016), but the regulatory means through which this gene-specific dosage compensation is achieved are not fully understood (Discussion; Melamed and Arnold 2007; Mank and Ellegren 2009; Itoh et al. 2010; Livernois et al. 2013; Wang et al. 2017). We reasoned that miR-2954-3p, with its strong male bias, broad expression, and preference for Z-linked genes might provide an alternative path to gene-specific dosage compensation. In order to evaluate whether miR-2954-3p regulation is indeed associated with more similar expression patterns in males and females, we calculated M:F ratios for all Z-linked protein-coding genes based on expression data from Marin et al. (2017) and compared target genes of miR-2954-3p to target genes of autosomal miRNAs (Methods). Consistent with the dosage compensation hypothesis, we observed significantly smaller M:F ratios for miR-2954-3p targets (Fig. 4C), indicating that the miRNA compensates for gene dosage differences, leading to more similar expression levels between males and females. As expected, we did not observe a significant difference between the two types of targets when we repeated the analysis for autosomal genes (Supplemental Fig. S3A).

If down-regulation by miR-2954-3p serves as a dosage compensation mechanism, we would furthermore expect an enrichment of dosage-sensitive genes among its targets. A recent study demonstrated that one class of dosage-sensitive genes known as ohnologs (gene duplicates retained after whole genome duplication) are preferentially dosage-compensated compared to other genes on the chicken Z Chromosome (Zimmer et al. 2016). Drawing on this observation, we counted the number of reported ohnologs (Singh et al. 2015) that were identified in our target prediction analysis. We found that 25.5% (12 of 47) of Z-linked miR-2954-3p targets were indeed ohnologs, whereas only 7.8% (15 of 192) of Z-linked genes targeted by autosomal miRNAs belonged to this category (P = 0.0015, χ²-test) (Fig. 4D). No significant difference was found among autosomal target genes (Supplemental Fig. 3B). As expected under the dosage compensation hypothesis, our data thus show that miR-2954-3p preferentially targets Z-linked dosage-sensitive genes.

The results described above make a compelling case for the involvement of miR-2954-3p in gene-specific dosage compensation, but it should be noted that the targeting properties of miR-2954-3p were so far determined based solely on computational predictions. To extend our results to targets with experimental support, we turned to a data set generated by Lin et al. (2014), who measured mRNA expression following knockdown of miR-2954-3p in a male zebra finch cell line. We first wished to verify the Z preference of miR-2954-3p and therefore analyzed the degree of up-regulation of Z-linked and autosomal mRNAs following miR-2954-3p knockdown. Consistent with our target predictions in chicken, we found that Z-linked genes in zebra finch were significantly more up-regulated than autosomal genes (Fig. 4E). The enrichment of Z-linked miR-2954-3p targets was further underlined when we considered only mRNAs with significant up- or down-regulation following miR-2954-3p knockdown (FDR 5%): Among the autosomal genes, only 45.2% (1296 of 2868) were up-regulated after knockdown, suggesting that many autosomal genes might not be direct targets of miR-2954-3p, whereas 62.0% (93 of 150) of Z-linked genes were up-regulated, as expected if some Z-linked genes are directly suppressed by miR-2954-3p (P = 8.0 × 10⁻⁵, χ²-test). Our findings thus demonstrate that miR-2954-3p disproportionally targets Z-linked genes in both chicken and zebra finch, which separated over 70 million years ago (Prum et al. 2015).

As miR-2954-3p-induced dosage compensation appears to be a conserved feature of bird genomes, we speculated that predictions of dosage sensitivity derived from chicken data should be reflected in the response to miR-2954-3p knockdown in zebra finch, in spite of the evolutionary distance between the two species. We therefore divided genes into those with above-median and below-median M:F ratios, based on expression in the chicken brain, and evaluated their response to miR-2954-3p knockdown in the zebra finch cell line, which was derived from a non-neural head tumor (Lin et al. 2014). We found that genes with low M:F ratios in chicken were more strongly up-regulated in zebra finch after knockdown (P = 2.2 × 10⁻⁴) (Fig. 4E), consistent with a conserved role of miR-2954-3p in gene-specific dosage compensation. We confirmed that the same effect was evident when we repeated the analysis based on zebra finch M:F ratios (P = 0.0027) (Supplemental Fig. S4), again based on gene expression in the brain (Itoh et al. 2010). Further to the conserved association with chicken M:F ratios, we found that ohnolog annotations from the chicken genome also correlated with the effects of miR-2954-3p in zebra finch, with ohnologs being more strongly up-regulated after miR-2954-3p knockdown (P = 0.0039) (Fig. 4E). Together, our findings thus demonstrate that the strongly male-biased, ubiquitously expressed miR-2954-3p preferentially targets and down-regulates dosage-sensitive genes on the bird Z Chromosome. To our knowledge, this is the first report of miRNA regulation as a means of dosage compensation.