An integrated expression atlas of miRNAs and their promoters in human and mouse

Abstract

MicroRNAs (miRNAs) are short non-coding RNAs with key roles in cellular regulation. As part of the fifth edition of the Functional Annotation of Mammalian Genome (FANTOM5) project, we created an integrated expression atlas of miRNAs and their promoters by deep-sequencing 492 short RNA (sRNA) libraries, with matching Cap Analysis Gene Expression (CAGE) data, from 396 human and 47 mouse RNA samples. Promoters were identified for 1,357 human and 804 mouse miRNAs and showed strong sequence conservation between species. We also found that primary and mature miRNA expression levels were correlated, allowing us to use the primary miRNA measurements as a proxy for mature miRNA levels in a total of 1,829 human and 1,029 mouse CAGE libraries. We thus provide a broad atlas of miRNA expression and promoters in primary mammalian cells, establishing a foundation for detailed analysis of miRNA expression patterns and transcriptional control regions.

Figure 1. Selection of robust miRNAs and Drosha CAGE peak analysis

(a) Number of miRBase high-confidence criteria (Table 2) satisfied by human pre-miRNAs annotated in miRBase. Pre-miRNAs with a statistically significant (P < 0.05) Drosha CAGE peak are shown in orange; mirtrons are shown in yellow. (b) Genomic locus of mir-223 in human with the total number of FANTOM5 (blue) and ENCODE (red) CAGE tags as a function of the genomic position of their 5′ end, showing a Drosha CAGE peak at the 3′ end of the pre-miRNA. FANTOM5 sRNA reads are shown at the bottom, colored by their read count as defined by the color key. The exact 5′ and 3′ ends of the pre-miRNA were determined from FANTOM4 full-length sequencing data. (c) Number of CAGE tags as a function of their starting position relative to the 3′ end of the pre-miRNA, averaged across human pre-miRNAs in the robust set (n = 795). The 3′ end of the pre-miRNA was selected as the 3′ end of the most prevalent sRNA on the 3′ arm of the pre-miRNA in the FANTOM5 sRNA data, with the position indicated as zero corresponding to the first nucleotide downstream of the 3′ end of the pre-miRNA.

Figure 2. Expression profile and cell ontology analysis of mature miRNAs

(a) Miru visualization of FANTOM5 primary cell samples based on their expression profile of robust mature miRNAs. (b) Number of most abundant miRNAs contributing at least 50% of the total miRNA expression in each human sRNA library in FANTOM5. (c) Reverse cumulative distribution of the maximum expression across the FANTOM5 samples of human miRNAs in the robust set, permissive set, and robust candidate set. (d) Examples of miRNAs enriched or depleted in specific primary cell samples. Expression of miR-122-5p, miR-142-5p, and miR-302a-5p was enriched in hepatocytes, leukocytes, and pluripotent stem cells, respectively; miR-100-5p and miR-29a-3p were broadly expressed, but depleted in leukocytes and pluripotent stem cells, respectively. cpm., counts per million.

Figure 3. Analysis of the curated miRNA promoters of miRNAs in the robust set

(a) (left panel) Sequence conservation of the human genome, evaluated as the average phastCons score, in the promoter region of non-coding pri-miRNAs (containing intergenic mature miRNAs; n = 132), coding pri-miRNAs (containing intronic mature miRNAs; n = 415), transcription factor (TF)-coding transcripts (n = 1,651), other protein-coding transcripts (n = 15,350), and long non-coding RNAs (lnc; n = 1,461). The sequence conservation of randomly selected genome regions is shown in gray. The shaded area corresponds to one standard deviation in the estimated mean phastCons score. (right panel) The average sequence conservation at promoter regions of miRNAs was higher than at the promoter regions of non-TF protein-coding genes (Mann-Whitney P = 2 × 10⁻¹⁶, two-sided) and of long non-coding RNAs (Mann-Whitney P = 1 × 10⁻³⁵, two-sided). Error bars correspond to one standard deviation in the estimated mean phastCons score. (b) Distance between the transcription start site (TSS) of the pri-miRNA and the 5′ end of the first pre-miRNA is highly correlated between human and mouse both for coding and non-coding pri-miRNAs, suggesting strong conservation of the genomic extent of pri-miRNAs. (c) Spearman correlation between the expression level of pri-miRNAs, as measured by CAGE, and mature miRNAs, as measured by sRNA sequencing, compared to a background distribution consisting of correlations between randomly paired pri-miRNAs and mature miRNAs. Correlations for polycistronic pri-miRNAs were averaged across the mature miRNAs. (d) Spearman correlation in expression level between mature miRNAs originating from the same pri-miRNA, compared to a background distribution consisting of correlations between mature miRNAs originating from different pri-miRNAs. (e) Cell type-dependent expression of miRNA paralogs. While mir-128-1 was broadly expressed across most primary cell samples in FANTOM5, its paralog mir-128-2 was enriched in brain samples. tpm, tags per million. (c-d) The box extends from the lower to the upper quartile, with the center line at the median; the whiskers indicate the full range of the data.