EVPsort: an atlas of small ncRNA profiling and sorting in extracellular vesicles and particles

Abstract

Extracellular vesicles and particles (EVPs) play a crucial role in mediating cell-to-cell communication by transporting various molecular cargos, with small non-coding RNAs (ncRNAs) holding particular significance. A thorough examination of ncRNA expression and sorting mechanisms within EVPs is imperative for advancing the clinical applications of EVPs. We have developed EVPsort, which not only provides an extensive overview of ncRNA profiling in 3,162 samples across various biofluids, cell lines, and disease contexts but also seamlessly integrates 19 external databases and tools. This integration encompasses information on associations between ncRNAs and RNA-binding proteins (RBPs), motifs, targets, pathways, diseases, and drugs. With its rich resources and powerful analysis tools, EVPsort extends its profiling capabilities to investigate ncRNA sorting, identify relevant RBPs and motifs, and assess functional implications. EVPsort stands as a pioneering database dedicated to comprehensively addressing both ncRNA expression and sorting within EVPs. It is freely accessible at https://bioinfo.vanderbilt.edu/evpsort/.

Graphical Abstract

Introduction

Extracellular vesicles and particles (EVPs) are present in nearly all biological fluids and constitute a diverse category of cell-secreted entities that transport a wide range of bioactive molecules [1]. These bioactive molecules include proteins, lipids, DNA, and RNA, which are delivered to recipient cells, exerting profound effects on normal development and physiology, immune response, cancer progression, and various diseases [2]. Dysregulated cargos within EVPs have been associated with cancer growth, stage, and metastasis [3].

Among the various molecules found in EVPs, small ncRNAs, particularly miRNAs, have garnered significant attention [4]. Notably, miRNAs maintain their regulatory functions even after being transferred to recipient cells [5]. miRNAs are highly stable within EVPs and have demonstrated correlations with cancer outcomes and progression, making them promising candidates as disease indicators and therapeutic molecules [6]. For instance, miR-23b-3p, miR-10b-5p, and miR-21-5p were found to be dysregulated in plasma EVPs and correlated with overall survival in lung cancer patients [7]. Another example includes miR-320 which has been identified as a potential prognostic biomarker in ovarian cancer [8].

An increasing number of studies have performed genome-wide profiling of small ncRNAs in EVPs, characterizing their composition in specific environmental or physiological contexts [9]. Meanwhile, several databases have been developed to aggregate datasets from various studies, providing valuable resources for investigating the distribution and function of small ncRNAs in EVPs across different conditions [10–15]. For example, the exRNA atlas [13] offers access to 10,356 extracellular RNA profiles across 77 studies. It is worth noting that among these samples, only 990 were derived from EVs, while the majority (9,366) were obtained from cell-free biofluids, which contain a mixture of EVs, nonvesicular particles, and other RNA carriers such as lipoproteins (Table 1). Another resource, liqDB [14] contains small RNA expression profiles from 1,607 manually annotated EVs across 19 different biofluids. For those interested in breast cancer biomarkers, exoBCD [12] serves as an integrated database, providing exosome-derived miRNA profiles from 21 breast cancer EV samples. EVAtlas [10] is a collection of 2,030 small RNA-seq datasets for human EVs across 24 conditions. Additionally, EVmiRNA comprises 462 samples of EVs from 17 sources [15], while miREV curates 428 samples of EVs [11] (Table 1).

Table 1.

Comparison between EVPsort and other EV databases

Database	Biotype	Source	Num. of datasets	Customized DE analysis	Sorting
EVPsort	small RNAs	EVPs	3162	√	√
exRNA atlas [13]	small RNAs	EVs, cell free RNA	990 EV, 9366 cell-free RNA	/	/
liqDB [14]	small RNAs	EVs	1607	√	/
ExoBCD [12]	miRNA	exosome	21	/	/
EVAtlas [10]	small RNAs	EVs	2030	/	/
miREV [11]	miRNA	EVs	428	/	/
EVmiRNA [15]	miRNA	EVs	462	/	/

While significant efforts have been dedicated to building comprehensive databases for characterizing small ncRNA profiling in EVs, there is a notable scarcity of resources focused on understanding the sorting and packaging of small ncRNAs into EVs. Emerging evidence suggests that small ncRNAs in EVPs are not randomly secreted but are selectively sorted [1]. However, the mechanisms governing this selective sorting remain largely unknown. To address this gap, we introduce EVPsort, a curated collection of 3,162 small RNA-seq datasets obtained from both EVPs and their matched donor cells. EVPsort serves as a unique resource for studying condition-dependent loading of ncRNAs into EVPs, as well as exploring potential mediators and downstream functional effects (Table 1). EVPsort offers a comprehensive view of expression profiles of various small ncRNA types, including miRNAs, rRNA fragments, tRNA fragments, Y RNAs, snRNAs, and snoRNAs, across diverse biofluids, cell lines, and disease states. Furthermore, EVPsort seamlessly integrates 19 external databases and tools, encompassing RNA-binding protein (RBP) binding and presence, miRNA-gene, miRNA-disease/drug associations, motif discovery, and RBP interactions. With access to these external resources and powerful analysis tools, EVPsort enables customized differential analysis of ncRNA profiles between groups of interest, identification of RBPs and motifs associated with sorting, and prediction of downstream functional effects (Figure 1A). EVPsort is freely available at https://bioinfo.vanderbilt.edu/evpsort/.

Figure 1. Overview of EVPsort.

(A) The schema of EVPsort. (B)The statistics of EVPsort.

Materials and Methods

Data sources

We compiled a diverse collection of human small RNA datasets by searching the Gene Expression Omnibus (GEO) database using specific terms related to extracellular vesicles and particles (EVPs) and their aliases. To ensure data quality and relevance, we exclusively retained datasets generated through high-throughput sequencing. Notably, we excluded studies related to the intracellular exosome complex, which primarily participates in RNA degradation.

Our search query in the GEO database included terms “exosome”, “ectosome”, “microvesicle”, “EV”, “extracellular RNA”, variations of “extracellular vesicle”, “exomere”, and “supermere” all limited to Homo sapiens. To further refine our dataset selection, we employed the filter “high throughput sequencing”. We also manually reviewed datasets from the Extracellular RNA Communication Consortium (ERCC) and scrutinized publications from popular EV extraction kit vendors’ websites.

Each dataset underwent meticulous manual examination to filter out studies that were not related to EVPs or did not involve small RNA-seq. Ultimately, we identified 3,162 samples from 89 studies for subsequent quality control and uniform data processing. Additionally, we manually curated detailed metadata for each study, encompassing information on EV sources (including biofluids, cell lines, and primary cell cultures), EV and particle types, EV extraction methods, library preparation methods, and sample conditions (e.g., diseases and treatments).

Quality control and uniform processing of small RNA-seq datasets

We conducted uniform processing and analysis of all raw sequencing data. One challenge we encountered was the variability in adapter patterns within raw reads due to the use of diverse commercial and customized small RNA library preparation kits across different studies. Some studies did not provide adapter sequences, and existing tools struggled to detect them accurately, complicating the analysis process. To address this issue, we developed FindAdapt, a tool designed to automatically and accurately detect adapter patterns from raw reads without requiring any prior information (https://github.com/chc-code/findadapt). FindAdapt identified adapter sequences along with any random bases at the 5’ and 3’ ends of the reads. The detected adapter pattern was then used as input for the standard small RNA-seq analysis pipeline, TIGER (https://github.com/shengqh/TIGER). In the TIGER pipeline, adapters were trimmed, trimmed reads were mapped to both host and non-host genomes, and the expression of various small RNA biotypes was quantified. Specifically, Cutadapt (v2.10) was used to trim 3’ adapters and random 5’ and 3’ bases if present. Reads shorter than 16 nucleotides were categorized as “too short” and discarded. Both raw reads and adapter-trimmed reads underwent quality control using FastQC (v0.11.9) (www.bioinformatics.babraham.ac.uk/projects/fastqc). Post-trimmed reads were first mapped to a customized reference composed of the host genome (GRCh37.p13) and known sequences of host mature transcripts derived from specific library databases, including miRbase for miRNAs, GtRNAdb2 for tRNAs, and SILVA for rRNA. Bowtie1 (v1.3.0) was used for mapping, allowing one mismatch. Mapped reads were assigned to various categories of small non-coding RNAs, including miRNA, tRNA fragments, rRNA fragments, snRNA fragments, snoRNA fragments, Y RNA, and lincRNA fragments. Unmapped reads longer than 19 nucleotides were subsequently aligned in parallel to non-host genomes, which included exogenous structural RNA databases and curated exogenous genome databases (bacteria, fungus, algae, and virus). This alignment allowed for zero mismatches.

small ncRNAs profiling and comparison

We normalized raw counts to the total number of post-trimmed reads. For each small RNA, we reported both the raw count and the normalized value reported as read per million (RPM). To visualize the data, we used bar plots to illustrate the number and the percentage of reads: after trimming; mapped to host small RNAs; mapped to host genome and non-host genome; and assigned to each type of ncRNAs. Principal component analysis (PCA) was employed to uncover expression variation and patterns across samples/conditions. Heatmap was also utilized to depict expression abundances across samples. Moreover, EVPsort offers users the flexibility to customize these plots by selecting conditions and specific small ncRNAs of interest.

We used DESeq2 to identify differentially expressed miRNAs between user-defined groups. The criteria for differentially expressed miRNAs included a false discovery rate (FDR) less than 0.05 and an absolute log₂ fold change of at least 1. We employed boxplots to illustrate the expression distribution of individual miRNAs across different categories such as biofluid types, disease conditions, and EV extraction methods.

RBP-miRNA interactions

We obtained validated RBP-miRNA interactions from ENCODE RIP-chip, eCLIP, and iCLIP experiments [16]. This involved downloading the peak file targeted by each individual RBP, and considering miRNAs located within a 10k bp distance from binding peaks to be bound by the RBP. In total, 65,301 interactions were obtained, involving 147 RBPs and 1,494 miRNAs. Additionally, 318,824 RBP-miRNA interactions were predicted using beRBP [17] with its generalized model, accounting sequences within 1k bp of distance from all miRNAs.

To identify RBPs associated with differentially expressed miRNAs, we applied Gene Set Enrichment Analysis (GSEA). This involved comparing a ranked list of differentially expressed miRNAs against a set of miRNAs targeted by RBPs. Specifically, we ranked differential miRNAs by their statistic values from DESeq2, and then compared the ranked list against validated and predicted miRNA bindings of each RBP, separately. We considered RBPs with a false discovery rate (FDR) of less than 0.25 to be significantly associated.

Motif identification

To identify potential regulatory elements within differentially expressed miRNAs, we performed motif enrichment analysis on miRNAs sequences, which were significantly upregulated (log₂FC > 1) or downregulated (log₂FC < −1) at a FDR < 0.05. We utilized the MEME algorithm from the MEME suite for motif discovery. The analysis was set to detect up to 20 motifs, with an E-value cut-off of 0.05. Only motifs present in at least four sequences were considered.

miRNA-gene and miRNA-pathway regulations

We sourced miRNA-gene interactions from DIANA-TarBase v7.0 [18], a repository containing 424,026 high-quality, manually curated, and experimentally validated interactions in human. To predict pathways regulated by one miRNA, we performed functional enrichment analysis on target genes of the miRNA against GO, KEGG, REACTOME, and Wikipathways using WebGestalt [19]. We considered functional gene sets/pathways with an FDR below 0.25 to be significantly enriched.

miRNA-disease and miRNA-drug associations

We enhanced EVPsort by integrating several valuable miRNA-related resources of drugs and diseases. We combined miRNASNP v3 [20], GWAS catalog [21], and COSMIC [22] to provide insights of a miRNA region about potential genotype-phenotype associations, germline and somatic mutations, and their potential impacts. To further bolster the understanding of miRNA dysregulation in the context of diseases, we incorporated the curated resources of miR2Disease [23] and dbDEMC [24]. We obtained miRNA-drug associations from SM2miR [25] and Pharmaco-miR [26]. This integration of diverse resources enriches the understanding of miRNA mutation/dysregulation and its implications in various diseases.

Database implementation

EVPsort was built using Django 4.1, a robust Model-View-Template (MVT) web development framework. Data management and accessibility were facilitated by the MySQL 5.7 relational database system. The hosting services were provided by the Apache 2.4 web server. To enhance the user experience, we utilized DataTables.js for rendering interactive tables, Plotly.js for generating dynamic data visualizations, and the STRING API for displaying the complex network of RNA Binding Protein interactions.

Results

Summary statistics of EVPsort

EVPsort serves as a comprehensive repository of small RNA-seq profiling data, encompassing 3,162 samples from 89 studies. Among them, 31 studies include both EVPs and matched donor cells. EVPsort quantifies expression of 2,638 miRNA, 531 rRNA, 610 tRNA, 860 Y RNA, 1,916 snRNA, 1,457 snoRNA across 10 biofluid types, 75 cell lines, and 66 conditions (Figure 1B). In addition to small RNA profiling, EVPsort connects small RNAs to their upstream mediators like motifs or RBPs, as well as downstream target genes and functional networks. These connections are made by 15 external databases and four tools (beRBP, MEME, GSEA, and WebGestalt) (Figure 1A). Specifically, for upstream RBP mediators, EVPsort obtains 65,301 validated RBP-miRNA binding data from ENCODE CLIP experiments [16] and predicts 318,824 RBP-miRNA interactions using beRBP [17]. EVPsort includes information on RBP presence in 286 exosomal studies from Vesiclepedia [27]. It also provides insight into 13,547 RBP protein-protein interactions through STRING [28]. In terms of downstream targets and effects, EVPsort includes 424,026 validated miRNA-gene interactions from DIANA-TarBase v7.0 [18] and conducts functional enrichment analysis against GO, KEGG, Wikipathways, and Reactome using WebGestalt [19]. Furthermore, EVPsort establishes links between miRNAs and drugs/diseases by incorporating data from miRNASNP [20], GWAS catalog [29], COSMIC [22], miR2Disease [23], dbDEMC [24], SM2miR [25], and Pharmaco-miR [26] (Figure 1B). EVPsort conducts MEME and GSEA to identify motifs and RBPs associated with differentially expressed miRNAs in a comparison between two conditions, respectively (Figure 1A).

View an miRNA of interest

EVPsort offers a wealth of information for studying an miRNA of interest, including its expression in EVPs and matched host cells, upstream RBPs, downstream target genes and pathways, and related diseases and drugs. EVPsort provides three ways to search an miRNA: by miRNA, by binding RBP, and by target gene (Figure S1A). When queried by an RBP, EVPsort returns the miRNAs bound by the specific RBP. Conversely, when queried by a gene, EVPsort provides a list of miRNAs targeting that gene.

Case study 1: hsa-mir-451a in EVPs

Using “hsa-mir-451a” as an example for how to use the database, EVPsort returns two miRNAs, hsa-mir-451a and hsa-mir-451b, when searching for “451” (Figure S1A). Selecting “hsa-mir-451a” brings up a detailed page with five tabs: Basic Info, Expression, Regulatory, Disease, and Drug (Figure S1B). In the ‘Basic Info’ tab, users can find key details about miR-451a, including its ID, miRBase accession number, coordinates, mature and primary sequences. It also lists 85 studies expressing miR-451a (Figure S1B). The ‘Expression’ tab displays the expression distribution of miR-451a across various experiments, including studies, EVP types, EVP extraction methods, conditions, cell lines, biofluids, and primary cell cultures. Notably, miR-451a is significantly more enriched in EVs than in donor cells in 22 of 25 comparable matched studies, suggesting preferential release into EVs rather than retention in cells (Figure S1C). The ‘Regulatory’ tab provides information on RBPs binding to miR-451a and its target genes. One well-known target is MYC, regulated by miR-451a [30]. Additionally, EVPsort lists pathways enriched in its target genes. Notably, the top significant pathways include mTOR and AMPK signaling pathways, which have been previously linked to miR-451a regulation [31] (Figure S1D). In the ‘Disease’ tab, users can explore variants in the miR-451a region and its differential expression in cancer datasets obtained from dbDEMC (Figure S1E). Unfortunately, the ‘Drug’ tab reports no drug information found for miR-451a.

Investigate/reanalyze a study

In addition to miRNA-centered exploration, EVPsort offers study-centered investigation. Users can search for a study of interest using keywords (Figure 2A) or browse all studies available in EVPsort. Each study page provides an in-depth view of the study, including details about experimental design, quality control, and ncRNA profiling. EVPsort also offers tools and modules for users to reanalyze the data and generate new hypotheses.

Figure 2. Explore a study on KRAS-dependent sorting.

(A) Search EVPsort using the keyword “KRAS”. (B) The ‘Study info’ tab of the study. (C) The ‘Data Summary’ tab of the study. (D) The ‘RNA profiling’ tab of the study.

Case study 2: KRAS-dependent sorting of miRNAs to EVPs

Searching EVPsort using the keyword “KRAS” retrieves the study GSE67004 [32]. This study generated small RNA-seq datasets for three isogenic colorectal cancer cell lines and their secreted exosomes: DKS-8 with wild-type KRAS, DLD1 with a heterozygous KRAS mutant G13D, and DKO-1 with a homozygous KRAS mutant G13D. This paired design, combined with different KRAS genotype status, allows for the study of KRAS-dependent secretion of small ncRNAs.

Selecting this study leads to a detailed study page with four tabs: Study Info, Data Summary, RNA profiling, and Comparison. The ‘Study Info’ tab displays the experimental design obtained from GEO, including the abstract, EV extraction method, and experimental metadata (Figure 2B). The ‘Data Summary’ tab provides information on data quality for each sample and condition, including the number and percentage of reads after trimming, reads mapped to the host genome and host small RNAs, as well as reads mapped to different small RNA biotypes (Figure 2C). This tab shows high-quality data with a low percentage of unmapped reads and notable differences in small RNA composition between cells and exosomes (Figure 2C). The ‘RNA Profiling’ tab contains expression profiles for various small ncRNA biotypes, such as miRNA, Y RNA, tRNA, snRNA, snoRNA, rRNA, mt_tRNA, and other small RNAs. For each biotype, EVPsort provides PCA plots, heatmaps, and a searchable, downloadable expression table. For example, the PCA plot in the miRNA tab illustrates distinct differences between miRNA sources (cells or exosomes) and KRAS status (Figure 2D). The most abundant miRNAs, such as miR-10a-5p and miR-21-5p, are also highlighted (Figure 2D).

The ‘Comparison’ tab provides a list of all samples and equips interactive tools for defining groups of interest, enabling users to perform differential analysis between these defined groups. To investigate KRAS-dependent sorting, users can define four groups by selecting corresponding samples: DKS8_cell, DKS8_EV, DKO1_cell, and DKO1_EV (Figure S2A). Two differential analyses are conducted: DKS8_EV vs. DKS8_cell and DKO1_EV vs. DKO1_cell, to identify small RNA sorting patterns in DKS-8 and DKO-1, respectively. In DKS-8, 358 miRNAs are differentially expressed (|log₂FC| > 1 & FDR < 0.05) in exosomes compared to cells (Figure S2B). Similarly, in DKO-1, 377 miRNAs show differential expression (|log₂FC| > 1 & FDR < 0.05) in exosomes compared to cells (Figure S2C). Notably, miR-30b/c/d/e expression is decreased in both DKS-8 and DKO-1 exosomes compared to matched cells (Figures S2B and S2C). Furthermore, a comparison between these two differential analyses (DKO1_EV vs. DKO1_cell - DKS8_EV vs. DKS8_cell) assesses KRAS-dependent sorting, specifically how the KRAS genotype status affects small RNA sorting from cells into exosomes (Figure S2A). In total, 97 miRNAs are identified as being sorted differently into exosomes between DKS-8 (WT) and DKO-1 (KRAS mutant) (FDR < 0.05) (Figure S2D). For instance, miR-100-5p abundance is down-regulated in DKS-8 but up-regulated in DKO-1 exosomes, suggesting that the KRAS mutant promotes the sorting of miR-100-5p into exosomes (log₂FC=4.2, FDR= 1e-06). In contrast, miR-146b-5p expression is increased in DKS-8 but decreased in DKO-1 exosomes, indicating that the KRAS mutant retains miR-146b-5p within cells (log₂FC=−3.7, FDR= 1e-13). These results from EVPsort align with the findings of the original study [32].

Case study 3: Identify RBPs in miRNA sorting into EVPs

In our previous study, we observed that the sorting of most miRNAs is condition-dependent [9]. RBPs play a crucial role in facilitating the transfer of RNA into extracellular vesicles through RBP-RNA complexes [33]. To identify the RBPs responsible for miRNA sorting, EVPsort integrates a map of RBP-miRNA binding data from ENCODE CLIP experiments [16] and employs computational prediction using beRBP [17]. As an example, using the study GSE106224, EVPsort identified two RBPs (DGCR8 and DROSHA) from ENCODE and 20 RBPs from beRBP that were enriched in the binding of upregulated miRNAs in Preterm Labor (PTL) EVs compared to plasma (FDR < 0.001) (Figure S3A). Among these RBPs, some have been previously reported to mediate the sorting of non-coding RNAs into EVs, including HNRPDL, HNRNPA1L2, IGF2BP2, and YB-1 [34]. Notably, four RBPs (RBMS2, RBMS3, RBMS1, and RBMS28) share the same RNA recognition motif (e.g., RRM, RBD, or RNP domain) and belong to the Pfam protein family PF00076. Additionally, the identified RBPs exhibit close interactions with each other, suggesting their cooperation in the sorting process (Figure S3B). In summary, EVPsort serves as a powerful tool for the identification of RBPs involved in the sorting of non-coding RNAs into EVs.

Discussion

EVPs mediate intercellular communication by exchanging molecular cargos. To unlock their biological significance and clinical potential, we present EVPsort, the largest database for small ncRNA profiling in EVPs, consisting of 3,162 datasets. In addition to miRNAs, EVPsort quantifies diverse small ncRNA types such as tRNA, rRNA, Y RNA, snRNA, snoRNA, and more. EVPsort offers a comprehensive view on miRNAs, including their percentage, abundance, and distribution, miRNA-gene/pathway regulations, and miRNA-disease/drug associations. With interactive tools, users can conduct customized differential analysis and assess complex interaction effects between two factors, simplifying the reanalysis of publicly available datasets.

While EVPs partially mirror the transcriptome of host cells, they exhibit distinct RNA profiles, indicating selective RNA loading [1, 9, 32]. EVPsort stands out by not only cataloging EVPs but also their host cells for comparative analysis, enabling the study of small ncRNA enrichment and shared patterns. Recent studies highlighted the main role of RBPs in RNA loading into EVPs [35]. EVPsort integrates validated and predicted RBP-miRNA interactions for identifying potential RBPs responsible for RNA sorting in certain conditions. Additionally, EVPsort employs MEME for motif discovery associated with ncRNA enrichment. EVPsort is the first database dedicated to both ncRNA profiling and sorting.

Comparing small ncRNA profiles across studies is challenging due to significant variability driven by technical factors (such as differences in EV and RNA isolation methods) and biological noise (such as inherent heterogeneity in EV carrier proportions) [13, 36, 37]. These factors are intertwined with the biological factors of interest, making bias removal and fair comparisons difficult. Additionally, downstream analytical biases are influenced by computational methods and assumptions, such as the assumption of uniform total input RNA content across experimental conditions. However, this assumption may not hold true, especially in diseases with differential RNA secretion [38]. To explain between-study variability, exRNA atlas uses computational deconvolution and defines six cargo types [13]. This might help tracing alterations in EV carrier compositions but cannot detect perturbations in a certain EV carrier between studies.

EVPsort offers a rich resource for miRNAs and facilitates the identification of RBPs responsible for RNA sorting, leveraging protein presence and interactions data from Vesiclepedia, and STRING. Nevertheless, information on other ncRNA types is limited to their abundance, primarily due to their unknown functions. As EVPs carry multiple ncRNAs and biomolecules with a given miRNA, the intricate language of EVP-mediated communication remains to be unraveled. Enhancing our understanding of this complexity is essential for developing EVP-based systems for disease diagnosis and RNA therapeutics delivery.

Supplementary Material

Figure S2. Customized analysis of KRAS-dependent sorting.

(A) Define groups of interest. (B) Differentially expressed miRNAs in EV compared to cells in DKS-8 (KRAS wildtype) cell line. (C) Differentially expressed miRNAs in EV compared to cells in DKO-1 (KRAS mutant) cell line. (D) miRNAs with KRAS-dependent sorting.

Figure S3. Identify RBPs involved in miRNA sorting.

(A) RBPs enriched in the binding of differentially expressed miRNAs in Preterm Labor (PTL) EVs compared to plasma. (B) RBPs interactions from STRING.

Figure S1. Explore miR-451a in EVPsort.

(A) Search EVPsort by miRNA containing “451”. (B) The ‘Basic Info’ tab of miR-451a. (C) The ‘Expression’ tab of miR-451a. (D) The ‘Regulatory’ tab of miR-451a. (E) The ‘Disease’ tab of miR-451a.