miRViz: a novel webserver application to visualize and interpret microRNA datasets
Pierre Giroux1,1, Ricky Bhajun2,1, Stéphane Segard2, Claire Picquenot2, Céline Charavay2, Lise Desquilles1, Guillaume Pinna 03, Christophe Ginestier ®4, Josiane Denis1, Nadia Cherradi 1 and Laurent Guyon ®1,2,*
1Univ. Grenoble Alpes, CEA, IRIG, Inserm, BCI, 38000 Grenoble, France, 2Univ. Grenoble Alpes, CEA, IRIG, Inserm, BGE, 38000 Grenoble, France, 3Institute for Integrative Biology of the Cell (I2BC), CEA, CNRS, University Paris-Sud, University Paris-Saclay, F-91191 Gif-sur-yvette, France and 4Aix-Marseille Université, CNRS, INSERM, Institut Paoli-Calmettes, CRCM, Epithelial Stem Cells and Cancer lab, F-13273 Marseille, France
Received February 10, 2020; Revised March 20, 2020; Editorial Decision April 03, 2020; Accepted April 06, 2020
ABSTRACT
MicroRNAs (miRNAs) are small non-coding RNAs that are involved in the regulation of major pathways in eukaryotic cells through their binding to and re- pression of multiple mRNAs. With high-throughput methodologies, various outcomes can be measured that produce long lists of miRNAs that are often dif- ficult to interpret. A common question is: after differ- ential expression or phenotypic screening of miRNA mimics, which miRNA should be chosen for further investigation? Here, we present miRViz (http://mirviz. prabi.fr/), a webserver application designed to visual- ize and interpret large miRNA datasets, with no need for programming skills. MiRViz has two main goals: (i) to help biologists to raise data-driven hypothe- ses and (ii) to share miRNA datasets in a straightfor- ward way through publishable quality data represen- tation, with emphasis on relevant groups of miRNAs. MiRViz can currently handle datasets from 11 eukary- otic species. We present real-case applications of miRViz, and provide both datasets and procedures to reproduce the corresponding figures. MiRViz of- fers rapid identification of miRNA families, as demon- strated here for the miRNA-320 family, which is signif- icantly exported in exosomes of colon cancer cells. We also visually highlight a group of miRNAs asso- ciated with pluripotency that is particularly active in control of a breast cancer stem-cell population in cul- ture.
INTRODUCTION
MicroRNAs (miRNAs) are non-coding RNAs of around 22 nucleotides that regulate protein-coding gene products at the post-transcriptional level, by directing the RNA- induced silencing complex (RISC) to its mRNA targets. Canonical binding of miRNAs corresponds to almost per- fect Watson-Crick pairing of the so-called ‘seed’ sequence with its mRNA targets, often in the 3’ UTR (1). The seed sequence comprises the six nucleotides at positions 2-7 in the 5’ region of the mature miRNA. Due to the small num- ber of nucleotides involved in this target recognition, miR- NAs lack specificity, and they often have dozens to hun- dreds of target mRNAs. Groups of miRNAs that share the same seed sequence bind to similar sets of mRNA targets, and are thus classified as miRNA families (1).
The popularization of high-throughput technologies has generated a vast quantity and diversity of large tables of miRNAs. For example, microarrays and sequencing tech- nologies measure miRNA expression levels under various experimental conditions, to provide data that are often con- verted into differential expression levels (2-4). Similarly, phenotypic high-content screening has led to large func- tional datasets (5). The common questions are then how to interpret these tables, and which miRNAs to select for further evaluation and validation. In the following, we de- fine as ‘hits’ those miRNAs with high scores that indicate their potential interest. These can be differentially expressed miRNAs, highly expressed miRNAs, or miRNAs with high scores or small P-values.
To help in the analysis of such long lists of miRNAs, we have designed and built a free-to-use webserver to visual- ize and interpret miRNA datasets, entitled miRViz (http:// mirviz.prabi.fr/). With miRViz, users can visualize their own and/or pre-loaded miRNA datasets on predefined miRNA networks, with various options to highlight or hide subsets
*To whom correspondence should be addressed. Tel: +33 438 780 453; Fax: +33 438 785 058; Email: laurent.guyon@cea.fr
*The authors wish it to be known that, in their opinion, the first two authors should be regarded as joint First Authors.
@ The Author(s) 2020. Published by Oxford University Press on behalf of Nucleic Acids Research.
of the data. These operations are accessible through the top menu on the left bandeau. The data-processing methods, such as data normalization, are not implemented in miRViz; these analyses should be conducted before using miRViz. For example, analysis for differential expression should be performed with one of the numerous available tools, such as DESeq2 (6), before visualization in miRViz.
MiRViz is designed to be as intuitive as possible. Ad- ditionally, the miRViz help file (‘Download help’, top- right) guides users, click by click, together with a down- loadable video tutorial. While a few studies, including our previous study, have proposed miRNA-network ap- proaches (7,8), to the best of our knowledge, none of them have proposed a dedicated tool for biologists to in- terpret miRNA datasets. There are tools to build miRNA- mRNA target networks, including miRNet (http://www. mirnet.ca/), although the users have to previously identify miRNAs of interest from their datasets (9). MiRViz and miRNet can thus be used in sequence. Bracken et al. re- viewed the websites for enrichment analysis (10). These websites are also complementary to miRViz, and can be used with the same datasets. For example, TAM (http: //www.lirmed.com/tam2/Home/) and miEAA (https://ccb- compute2.cs.uni-saarland.de/mieaa_tool/) perform ontol- ogy enrichment at the level of miRNAs, and miRPath (http://snf-515788.vm.okeanos.grnet.gr/) and miRPathDB (https://mpd.bioinf.uni-sb.de/) do so at the level of the mRNA targets (11-14).
This paper is organized as follows. In the following sec- tion, we present the different miRNA networks and the preloaded datasets. We then present the webserver func- tionalities, and provide practical examples to highlight the strengths of miRViz for experimental data analysis. Impor- tantly, in the Supplementary Materials we provide all of the necessary datasets and step-by-step procedures to repro- duce the figures shown here with miRViz.
MATERIALS AND METHODS
Implementation
MiRViz is composed of a JavaScript front-end server and a Java back-end server. While the best user experience is provided when running miRViz on wide screens with high resolution, it still remains of interest with lower resolution, where zooming out may sometimes be necessary to avoid overlapping components (for more details, see the Trou- bleshooting section of the downloadable mirviz-help.pdf). User datasets are not uploaded to the server but are safely loaded into the local browser. MiRViz is freely available without login requirements.
MicroRNA networks
MiRViz is built around predefined miRNA networks, in which each node is represented by a circle that corresponds to a unique mature miRNA. In its 2020 version, miRViz can be used to visualize miRNA data from 11 different species, including human (hsa), mouse (mmu), Caenorhabditis el- egans (cel) and Drosophila (dme). The architecture of the webserver was designed to easily add new networks and new species in the future.
In miRViz, we propose eight different predefined net- works. Each of the networks has its own rules for the con- nection of the miRNA nodes. The first network used by miRViz, ‘Seed2_7’, allows direct visualization of miRNA families by connecting pairs of miRNA nodes that share the same seed sequence. The two miRViz networks en- titled ‘Genomic_Distance’ connect neighboring miRNA genes on the genome. The ‘2k’ version links neighbor- ing miRNAs if they are closer than 2 kilobases (kb), to visually identify polycistronic miRNA clusters (15). The ‘50k’ version has a less stringent threshold of 50 kb, to account for large genomic reorganizations, as can be en- countered in tumors and as demonstrated in an exam- ple below. Three co-regulation networks connect miRNA nodes that share common mRNA targets (7). More pre- cisely, in ‘Diana50’, ‘TargetScan54’ and ‘DianaTarBase50’, the miRNA nodes are connected if they share more than 50%, 54% and 50% common mRNA targets, respectively, as predicted by Diana MicroT v3 (16) or Target-Scan v6.2 (17), or by experimental validation and gathering in Diana- TarBase v8 (18). We added two simplified networks with fewer nodes to ease visualization and interpretation: ‘Ge- nomic_Distance_50k_clusters_3+’ which only keeps the 820 miRNAs that are in clusters of size 3 or more, and ‘Tar- getScan54_degree_10+’ which keeps the same layout of ‘Tar- getScan54’ but removing the nodes connected to 9 or fewer nodes. In the Supplementary Materials, we provide the ex- act formulae and justifications of the rules that underlie the connections of the miRNA nodes in each of the networks.
Pre-loaded datasets
Three miRNA tables are pre-loaded in miRViz to ease comparisons with user datasets: ‘hsa_miRNAmine_cells’, ‘hsa_miRNAmine_tissues’ and ‘hsa_TissueAtlas’. For the two first datasets, the data were gathered from miRNAmine (3). For the third dataset, the data were gathered from Tis- sueAtlas (4). MiRNA expression was transformed into log2 scales, and then averaged across all of the experiments per- formed under the same conditions (the number of different experiments used to average is indicated in brackets). Ex- pression on a log2 scale spans from 0 to 20.
Experimental microRNA datasets
We have used various public miRNA expression datasets in the practical examples provided (from both microarrays and sequencing) to highlight miRViz strength (19,20). We have also used an in-house functional miRNA screening dataset (21). The datasets are provided as Supplementary Tables and the data pre-processings are detailed in the Sup- plementary Materials. Briefly, the first dataset relates to miRNA sequencing, from the public identifier SRA106214. It provides miRNA expression levels in colon cancer cell line LIM1863, and in their secreted small vesicles, together with differential expression between vesicles content and the parental cells (Supplementary Table S1). The second dataset provides the prognostic value for overall survival of patients with adrenocortical carcinoma, for each individual miRNA, expressed in the log10 of the P-value obtained after log-rank test (Supplementary Table S2). The third dataset
provides the differential expression between human embry- onic stem cell and differentiated cells from the public dataset GSE14473 (Supplementary Table S3). The fourth dataset provides the change of the percentage of breast cancer stem cells after transfection of individual miRNA mimics (Sup- plementary Table S4). The last dataset presents the en- richment of two selected gene ontologies among predicted mRNA targets for each individual miRNA (‘regulation of gene expression’ and ‘small-GTPase-mediated signal trans- duction’, Supplementary Table S5).
Statistical tests
MiRViz provides visualization of the aggregation of miR- NAs on various networks, in particular inside families (Seed2_7 network) or genomic clusters (Genomic_Distance networks). To assess how significant miRNA hits are aggre- gated in a given family or cluster, a ‘local’ statistical test can be performed. Let p denote the proportion of hits on the network (p = n hits / n nodes, with n hits the number of hits, and n nodes the number of nodes with miRNA measure- ment in the network), and n_cluster the number of nodes in the cluster/family of interest with miRNA measurement (e.g. expressed miRNAs for differential expression measure- ments). Under the null hypothesis H0, the number of hits in this cluster/family is given by the binomial distribution, and the P-value is the probability under H0 to get equal or more hits in the cluster/family of interest. To assess how sig- nificant miRNA hits are connected in a given network, we also proposed a ‘global’ statistical test, in which the number of hit pairs (i.e. an edge connecting two miRNA hit nodes) are first counted. Then, to evaluate the null hypothesis H0, the number of hit pairs is counted after a hit randomization procedure in which an miRNA is randomly designated a hit, while keeping constant the total hit number. The global P-value is estimated as the proportion of randomized tri- als for which the measured number of hit pairs is below the randomized one.
RESULTS AND DISCUSSION
MiRViz highlights the selective export of the miR-320 family into colon cancer cell exosomes
The first network used by miRViz, Seed2_7, allows direct visualization of miRNA families by connecting pairs of miRNA nodes that share the same seed sequence. The Seed2_7 networks can be visualized through miRViz for all of the 11 species proposed. As a demonstration of how miRViz can be used to interpret miRNA expression datasets, a few click tutorials are provided in the help file (top-right button on the website), which shows how to vi- sualize tissue-specific miRNAs. Here, we propose a com- plementary example using the publicly available miR-seq dataset that profiles the LIM1863 colon cancer cell line and three different sorts of extracellular vesicles isolated from the culture supernatants (19). Figure 1 shows the differentially expressed miRNA families in two types of immunoaffinity-isolated exosomes (i.e. A33 in Figure 1A; EpCAM in Figure 1B) and shed microvesicles (Figure 1C).
MiRNAs with lower and higher expression in vesicles com- pared to cells are colored in green and red, respectively. The nodes that correspond to the miRNAs with very low expres- sion under all of the conditions is set as semi-transparent. Interestingly, miRNAs tend to have similar differential ex- pression in each family, which is highlighted by miRViz showing uniform colors in each family. Indeed, when defin- ing an miRNA with high differential expression (log2 fold change > 1) as a hit, miRViz shows that hits significantly aggregate in miRNA families in all three sort of vesicles (global P-value < 10-5, Supplementary Materials section 6). We can thus hypothesize that there is an active export of selective family members through exosomes, in accor- dance with active export shown in other experimental cases (22,23). MiRViz can quickly identify these families, and pro- vide a way to share the result. The hsa-miR-378/422a fam- ily is exported specifically in immunoaffinity-isolated A33 exosomes (P-value = 0.008; paired Wilcoxon test; Figure 1A). The hsa-miR-320 family is significantly exported in both sorts of exosomes (P-value = 0.002; paired Wilcoxon test; Figure 1A, B). This export is clear only for exosomes (i.e. not for small vesicles), and stronger for the A33 exo- somes. We have confirmed these data by independent RT- qPCR measurements (Supplementary Figure S1). Finally, all of the five expressed members of the hsa-let-7-3p/miR- 98-3p family are significantly exported in all of the three types of extracellular vesicles (P-value = 6 × 10-5; paired Wilcoxon test; Figure 1). In addition to overall aggregation behavior of miRNA hits, one can ask how significant hits are aggregated in a given family. All the three families de- scribed above for which the differential expression was sig- nificant, showed also a significant segregation (local P-value < 0.002, Supplementary Materials section 6).
High expression of miRNAs of the Xq27.3 cluster is predictive of better prognosis in adrenocortical carcinomas
Both Genomic_Distance networks can be visualized with miRViz for all of the 11 species. To demonstrate the inter- est of the Genomic_Distance_50k network in the context of cancer, we reanalyzed the public data from Assie et al. (24), which contains both the miR Nome of tumor samples from patients diagnosed with adrenocortical carcinoma and their overall survival (OS) information. For each miRNA, the patients were separated into two groups of equal size, which depended on the miRNA quantification, and a P- value was calculated on the OS after log-rank tests. A hit is defined here as an expressed miRNA (median expression among patients > 10 reads) for which the P-value is <10-2. Figure 2A shows the Kaplan-Meier curves of two microR- NAs of interest, together with the P-value of the log-rank test, and the node colored according to the P-value. Fig- ure 2B shows the P-value in a log10 scale overlaid onto Ge- nomic_Distance_50k network, zoomed in on chromosomes 14 to X, with a green gradient for good prognosis miRNAs (miRNAs for which high expression is of good prognosis for the patient), and a red gradient for poor prognosis miR- NAs. Two large clusters show up in miRViz (Figure 2B, C):
- cluster 14q32.2 (spanning 197 kb) that is predictive of poor prognosis, i.e. patients who show high expression of the
A
miR-378i
miR-4638-3p
miR-378d
miR-378
miR-37Be
miR-378c
☒
miR-378f
☒
☒
miR-378h
miR-378b
miR-422a
☒
☒
☒
☒
☒
☒
☒ 4-7b-3p
miR-98-3p
☒
3
☒
☒
et,71-1-3p
niR-1185-1-3p
☒
☒
☒ ct-71-2-3p
☒
☒
☒
☒
☒
☒
☒
TIR-4789-5p
ct-7a-3p
☒
☒
☒
☒
☒
☒ ☒ ☒
miR-320c
☒
niR-320b
☒
niR-320d
TiR-320a
☒ ☒
☒
☒
☒
☒
miR-4429
☒
☒
☒
☒
☒
☒
☒
Differential expression (log2 scale)
-2
-0.5
0 0.5
2
☒
☒
☒ ☒
miR repressed in exosomes (stronger relative expression in cells)
miR overpressed in exosomes (smaller relative expression in cells)
☒
☒
☒
☒
Semi-transparent: miR with maximum expression (log2) < 5
B
miR-378i
miR-4638-3p
miR-378d
miR-378
miR-37Be
miR-378c
niR-3781
☒
miR-378h
niR-378b
miR-422a
☒
☒
☒
at-7b-3p
miR-98-3p
☒
-
TR-1
85-
3
☒
☒
et-71-1-3p
miR-1185-1-3p
☒
☒
☒
☒
01-71-2-3p
☒
☒
☒
☒
nIR-4789-5p
t-7a-3p
☒
☒
☒
☒
☒
☒ ☒
miR-320c
TiR-320b
☒
niR-320d
☒
☒
niR-320a
☒
☒ ☒
miR-4429
☒
☒
☒
Differential expression (log2 scale)
-2
-0.5
0 0.5
☒
2
☒
☒
☒
miR repressed in exosomes (stronger relative expression in cells)
miR overpressed in exosomes (smaller relative expression in cells)
☒
☒ ☒
☒
Semi-transparent: miR with maximum expression (log2) < 5
C
☒
☒
☒
☒
☒
☒
☒
☒ ☒
☒
☒
☒
☒
☒
☒
☒
☒
☒
☒
☒ ☒
☒
☒
☒
☒
☒
☒
☒
☒
☒
☒
☒
☒
☒
☒
☒
☒
miR-378i
miR-4638-3p
miR-378d
miR-378
miR-378e
miR-378c
niR-378f
AIR-378h
miR-378b
miR-422a
et-7b-3p
nIR-98-3p
3
etaf1-1-3p
IR-1185-1-3p
at-71-2-3p
IR-4789-5p
t-7a-3p
miR-320c
iR-320b
miR-320c
miR-320a
miR-4429
Differential expression (log2 scale)
-0.5
0
0.5
-2
2
miR repressed in microvesicles sMVs (stronger relative expression in cells)
miR overpressed in microvesicles sMVs (smaller relative expression in cells)
Semi-transparent: miR with maximum expression (log2) < 5
miRNAs of the cluster are associated with shorter OS (24 hits out of 48 expressed miRNAs, P-value = 6.1 x 10-7); - cluster X27q3 (spanning 95 kb) that is predictive of good prognosis, i.e. patients who show high expression of these miRNAs are associated with longer OS (9 hits out of 13 expressed miRNAs, P-value = 6.7 x 10-7).
While both clusters were described in the original publi- cation (24), miRViz proposes a rapid method to easily iden- tify such clusters and a way to visualize the data. Addi- tionally, in Figure 2C three clusters are highlighted in blue: two clusters of the hsa-miR-29 family, located in 1q32.2 and 7q32.3, and associated with good prognosis (three hits out of four in each cluster, P-value = 4.1 x 10-3); and the hsa-miR-450b-5p/503-5p/424-5p cluster, located in Xq26.3 and associated with adverse prognosis (two hits out of seven expressed miRNAs, P-value = 0.37). The lat- ter is not significantly enriched, as it is probable to find a few miRNA hits out of seven just by chance when there are so many hits (in this case 44 hits out of 241, most of the hits are in the 14q32 cluster). It is interesting to note that the mature miRNAs from the -3p strand of the miR- 29 families that are transcribed from both chromosomes
1 and 7 share the same seed, AGCACC, which suggests redundancy.
MiRViz visually identifies the miR-302/519 stem-cell family in the regulation of breast cancer stem cell equilibrium
As proof of purpose for the Diana50 network, Figure 3A shows the differential expression of miRNAs in hu- man embryonic stem cells that were cultured under two different conditions that favor either pluripotency or dif- ferentiation. Here, a ‘stem cell’ miRNA cluster clearly shows up in red, which highlights the overexpressed miR- NAs in the pluripotent stem cells. Most of these miR- NAs have already been hypothesized to cooperatively regulate pluripotency (25). The group comprises miR- 17/20/93/106/302/ .. /519/520 with shifted seed sequences (AAAGUG, AAGUGC, AGUGCU), and miR-411 with seed sequence AGUAGA.
Figure 3B represents a functional screening dataset where we measured the relative levels of breast cancer stem cells (bCSCs) in a human breast adenocarcinoma cell line (SUM159 cells) upon systematic and individual overexpres- sion of miRNAs (21). The green (resp. red) nodes represent miRNAs that upon overexpression lead to smaller (resp.
A
miR-514a-5p
1.0
Percentage of Overall Survival
0.8
0.6
0.4
0.2
0.0
low miR expr. < 1840 - 22 patients
high miR expr. > 1840 - 22 patients
p-val = 0.0078
0
50
100
150
Time (in Months)
miR-411-5p
1.0
Percentage of Overall Survival
0.8
0.6
0.4
0.2
0.0
low miR expr. < 30196 - 22 patients
high miR expr. > 30196 - 22 patients
p-val = 0.013
0
50
100
150
Time (in Months)
B
14
17
20
X
| P-value OS (Overall Survival) 0.001 0.1 1 0.1 0.001 | |
|---|---|
| miR tumour suppressor (high expression = good prognosis) | oncomiR (high expression = bad prognosis) |
| Transparent: miR with median expression (normalised count) < 10 | |
C
1
2
29c-5p
29b-2-5p
29a-5p
29b-1-5p
29c.3p
29b-3p
29a-3p
29b-3p
3
hsa-miR.29b-3p
sa-miR-29a-3p UAGCACCA
UAGCACCA
chr1-
chr7-
miRBase
miRBase
miR cluster (14q32.2)
770-5
411-50
134-5e
0 22
485-50
431-6
42
127-2p
270-0p
O 379-19 4116 291-12 240-5%
9
485-04
3
4
377-
€
sa-miR-411-5p
C 4300
0 5:0650
C 540hda
0 7255-5a
C 4700 68
0 3614
UAGUAGAC
O 3275-de
O 15tb
O 1247de O 124700
O 20db do
@ chilido
4115-do
chr14+
3173-20
C 4750
C 44070 4857
C 48310 4401
miRBase
3
450b.5p
450a
450a-
542-5p
503-5
424-5p
450₺
450a-1-3p
50a-
542-3p
503-
424
sa-miR-424-5p CAGCAGCA
chrX-
miRBase
miR cluster (Xq27.3)
513c.5p
513b-5p
513a-5p
513a-5p
506-5p
508-3p
514b-3p
509-3p
509-3p
509-3p
☐
510-3p
514a-3p
514a-3p
514a-3p
513c:3p
☐
513b.3p
513a-3p
513a.3p
506-3p
507
508-5p
514b-5p
509-5p
509-3-5p
509-5p
510.5p
514a.5p
5141-5p
514a.5p
sa-miR-5142.5p UACUCUGG
chrx- MIRBase
1
2 :***********::::::*******:::::**********::…***
2
P-value OS (Overall Survival)
0.001
0.1
1
0.1
1
0.001
miR tumour suppressor
oncomiR
(high expression = good prognosis)
(high expression = bad prognosis)
Semi-transparent: miR with median expression (normalised count) < 10
3
higher) bCSC proportions. To determine the efficiency of miRViz to compare different datasets and the possibility that it can raise biological questions of interest, we can compare Figure 3A and B. Here, the miRViz representa- tion highlights the redundant action of the ‘miRNA stem cell’ cluster on the balance of the bCSC phenotype. It is, however, surprising that miRNAs for which expression was correlated with pluripotency in normal cells (Figure 3A, in red) indeed lead to decreased proportions of bCSCs when overexpressed (Figure 3B, in green). This suggests that the fine-tuning of this specific group of miRNAs might have an important and yet unknown role in the maintenance of the ‘stem’ state of normal and cancer cells. Interestingly, miRViz identifies this group of miRNAs, and suggests target gene redundancy, which might explain why their individual knock-downs in separate experiments (data not shown) had little or no effects on the bCSC equilibrium. To assess these hypotheses, additional experiments are necessary. Results obtained with miRViz suggest that these miRNAs needs to be collectively studied, and that their simultaneous knock- down may help to restore the expression of the target genes responsible for the stem-cell features.
Diana50 and TargetScan54 structures are correlated with bi- ological functions
To show the link between network organization and miRNA functions, we performed gene ontology enrich- ment on the predicted mRNA targets for each individual miRNA. For a given ontology and miRNA, the small P- values (typically <10-5) suggest that the miRNA regulates the corresponding function under certain cellular condi- tions. Figure 4 and Supplementary Figure S2 show that miRNAs that are assumed to regulate a given ontology (i.e. pathway or function) are not randomly spread out in the networks. Supporting our previous study (7), the Di- ana50 and TargetScan54 networks are structured in two parts. The upper part of both networks contains miRNAs that are almost all predicted to regulate gene expression, together with the two more central subnetworks of let-7 and miR-17/93 (Figure 4A, Supplementary Figure S2A). The lower parts of both of these networks contain many miRNAs that are predicted to regulate signal transduction through small GTPases (Figures 4A, Supplementary Figure S2B). Altogether, these observations show that the Diana50 and TargetScan54 structures correlate with biological pre-
A
B
203-6p
3024-
Da-5p
03.5p
5196-3: 520h
3-5p
5166-3p.520
5206 302a-35
sted-3
5204-30 9020-30
SIBd-3
6202-30
411-5p
0-Sp
7-5p
Ob-50
373
30
-50
12-3p
b-5º 373.3
520€
512-3p
372-30
$20d-3
5189-3p
520€
372-30
026-3p
5200-
026-3p
5209-3
520c-Jp
170c-Jp
0
0
O-
0000
Log2 Fold Change on miR expression
-2
-0.5
0.5
2
Log2 Fold Change on bSCS proportion
-3
-1
1
3
Higher expression in differentiated cells
Higher expression in totipotent cells
Smaller proportion of bCSC after individual miR transfection
Higher proportion of bCSC after individual miR transfection
Semi-transparent: miRs not measured
Semi-transparent: miRs not transfected
dictions, and that the positions of the hits in the networks in- form the users of putative regulated pathways, e.g. miRNA hits in the upper part might be important regulators of gene expression.
Which network for which dataset, and further validations
An important question is: which network should be used for a given dataset? The simple answer is ‘all’. We suggest to use all the networks, starting with small ones (Seed2_7, Di- ana50 and Genomic_Distance_50k_clusters_3+ for human). We suggest then to visually identify enrichment of highly connected groups of miRNAs. In our experience, we often find enrichment in the Seed2_7 and/or Genomic_Distance networks. Yet, the most appropriate network to explore a specific dataset depends on the biological question to an- swer. When focusing on mRNA targets repressed by miR- NAs, the Seed2_7 (and in a second time the co-regulation networks: Diana50, DianaTarBase50 and TargetScan54) should be used. When trying to identify polycistrons, Ge- nomic_Distance_2k shows co-expressed miRNAs in a given neighborhood on the genome. The 50k version is more ded- icated for large genomic reorganizations, as found in cancer. To note, even if miRViz is particularly useful for raw expres- sion and differential expression datasets, any large miRNA dataset with numerical scores can benefit from miRViz. Fig- ures 2, 4 described above present practical examples with P-values and phenotypic scores.
Another important question is: what to do with miRViz results? First, the mapping itself is interesting, and miRViz provides an export function to show a given enrichment in a figure. Second, it guides the validation steps. When inves- tigating the phenotypic role of a given miRNA, a practi- cal experiment consists in knocking-down (or over express- ing) this miRNA and measure the phenotypic outcome. If many miRNAs of a given family are co-expressed, modu- lating one miRNA out of the whole family may lead to no or minimal phenotypic effect, as the other miRNAs from the family may still repress the mRNA targets. A suggestion would be to modulate the whole family, which is also true for groups of highly connected miRNAs in co-regulation networks, such as for the miR-302/519 stem-cell family de- tailed above.
CONCLUSION
We propose that the webserver application miRViz can be used to visualize numerical miRNA datasets. We have illustrated the results that can be obtained for miRViz through various examples, including miRNA expression and functional screening datasets. For miRNA profiling, the network-based visualization proposed here provides clear ways to present datasets that are complementary to volcano plots for expression data. In particular, the Seed2_7 network allows rapid identification of hit miRNA families, and quickly identifies miRNA redundancies.
A
B
-log10(p-value) 5
10
-log10(p-value) 5
10
miR is unlikely to regulate the ontology
miR is likely to regulate the ontology
miR is unlikely to regulate the ontology
miR is likely to regulate the ontology
SUPPLEMENTARY DATA
Supplementary Data are available at NAR Online.
ACKNOWLEDGEMENTS
We thank C. Cochet and J.J. Feige for helpful discus- sion, and all external testers of miRViz. We thank Christo- pher Berrie for scientific English editing. This research was funded by the Institut National de la Santé et de la Recherche Médicale (INSERM), the Commissariat à l’Energie Atomique et aux Energies Alternatives (CEA, DRF/IRIG), the University Grenoble Alpes (UGA).
Author contributions: L.G. and C.C. designed the webserver. L.G., P.G. and R.B. built the miR networks. S.S. conceived the architecture of the webserver. S.S. and C.P. coded the webserver. C.C. and L.G. carried out functional tests of the webserver. L.G. and L.D. analyzed the miR datasets. G.P., C.G., J.D. and N.C. performed the experiments. L.G. wrote the manuscript with input and approval from all authors.
FUNDING
Funding for open access charge: Budget of the laboratory (funding from INSERM institution).
Conflict of interest statement. None declared.
REFERENCES
1. Bartel,D.P. (2018) Metazoan microRNAs. Cell, 173, 20-51.
2. Gong,J., Wu, Y., Zhang,X., Liao, Y., Sibanda,V.L., Liu, W. and Guo,A.Y. (2014) Comprehensive analysis of human small RNA sequencing data provides insights into expression profiles and miRNA editing. RNA Biol., 11, 1375-1385.
3. Panwar,B., Omenn,G.S. and Guan, Y. (2017) MiRmine: a database of human miRNA expression profiles. Bioinformatics, 33, 1554-1560.
4. Ludwig,N., Leidinger,P., Becker,K., Backes,C., Fehlmann,T., Pallasch,C., Rheinheimer,S., Meder,B., Stähler,C., Meese,E. et al. (2016) Distribution of miRNA expression across human tissues. Nucleic Acids Res., 44, 3865-3877.
5. Izumiya,M., Tsuchiya,N., Okamoto,K. and Nakagama,H. (2011) Systematic exploration of cancer-associated microRNA through functional screening assays. Cancer Sci., 102, 1615-1621.
6. Love,M.I., Huber, W. and Anders,S. (2014) Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol., 15, 550.
7. Bhajun,R., Guyon,L., Pitaval,A., Sulpice,E., Combe,S., Obeid,P., Haguet, V., Ghorbel,I., Lajaunie,C. and Gidrol,X. (2015) A statistically inferred microRNA network identifies breast cancer target miR-940 as an actin cytoskeleton regulator. Sci. Rep., 5, 8336.
8. Jiang, Q., Hao, Y., Wang,G., Juan,L., Zhang,T., Teng,M., Liu, Y. and Wang, Y. (2010) Prioritization of disease microRNAs through a human phenome-microRNAome network. BMC Syst. Biol., 4, S2.
9. Fan, Y., Siklenka,K., Arora,S.K., Ribeiro,P., Kimmins,S. and Xia,J. (2016) miRNet - dissecting miRNA-target interactions and functional associations through network-based visual analysis. Nucleic Acids Res., 44, W135-W141.
10. Bracken,C.P., Scott,H.S. and Goodall,G.J. (2016) A network-biology perspective of microRNA function and dysfunction in cancer. Nat. Rev. Genet., 17, 719-732.
11. Li,J., Han,X., Wan, Y., Zhang,S., Zhao, Y., Fan,R., Cui,Q. and Zhou, Y. (2018) TAM 2.0: Tool for microRNA set analysis. Nucleic Acids Res., 46, W180-W185.
12. Backes,C., Khaleeq,Q.T., Meese,E. and Keller,A. (2016) MiEAA: MicroRNA enrichment analysis and annotation. Nucleic Acids Res., 44, W110-W116.
13. Vlachos,I.S., Zagganas,K., Paraskevopoulou,M.D., Georgakilas,G., Karagkouni,D., Vergoulis,T., Dalamagas,T. and Hatzigeorgiou,A.G. (2015) DIANA-miR Path v3.0: Deciphering microRNA function with experimental support. Nucleic Acids Res., 43, W460-W466.
14. Backes,C., Kehl,T., Stöckel,D., Fehlmann,T., Schneider,L., Meese,E., Lenhof,H .- P. and Keller,A. (2016) miRPathDB: a new dictionary on microRNAs and target pathways. Nucleic Acids Res., 45, gkw926.
15. Chang,T.C., Pertea,M., Lee,S., Salzberg,S.L. and Mendell,J.T. (2015) Genome-wide annotation of microRNA primary transcript structures reveals novel regulatory mechanisms. Genome Res., 25, 1401-1409.
16. Maragkakis,M., Reczko,M., Simossis, V. a., Alexiou,P., Papadopoulos,G.L., Dalamagas,T., Giannopoulos,G., Goumas,G., Koukis,E., Kourtis,K. et al. (2009) DIANA-microT web server: elucidating microRNA functions through target prediction. Nucleic Acids Res., 37, 273-276.
17. Lewis, B.P., Burge,C.B. and Bartel,D.P. (2005) Conserved seed pairing, often flanked by adenosines, indicates that thousands of human genes are microRNA targets. Cell, 120, 15-20.
18. Karagkouni,D., Paraskevopoulou,M.D., Chatzopoulos,S., Vlachos,I.S., Tastsoglou,S., Kanellos,I., Papadimitriou,D., Kavakiotis,I., Maniou,S., Skoufos,G. et al. (2018) DIANA-TarBase v8: a decade-long collection of experimentally supported miRNA-gene interactions. Nucleic Acids Res., 46, D239-D245.
19. Ji,H., Chen,M., Greening,D.W., He,W., Rai,A., Zhang,W. and Simpson,R.J. (2014) Deep sequencing of RNA from three different extracellular vesicle (EV) subtypes released from the human LIM1863 colon cancer cell line uncovers distinct mirna-enrichment signatures. PLoS One, 9, e110314.
20. Stadler,B., Ivanovska,I., Mehta,K., Song,S., Nelson,A., Tan, Y., Mathieu,J., Darby,C., Blau,C.A., Ware,C. et al. (2010) Characterization of microRNAs Involved in Embryonic Stem Cell States. Stem Cells Dev., 19, 935-950.
21. El Helou, R., Pinna,G., Cabaud,O., Wicinski,J., Bhajun,R., Guyon, L., Rioualen,C., Finetti,P., Gros,A., Mari,B. et al. (2017) miR-600 acts as a bimodal switch that regulates breast cancer stem cell fate through WNT signaling. Cell Rep., 18, 2256-2268.
22. Villarroya-Beltri,C., Gutiérrez-Vázquez,C., Sánchez-Cabo,F., Pérez-Hernández,D., Vázquez,J., Martin-Cofreces,N., Martinez-Herrera,D.J., Pascual-Montano,A., Mittelbrunn,M. and Sánchez-Madrid,F. (2013) Sumoylated hnRNPA2B1 controls the sorting of miRNAs into exosomes through binding to specific motifs. Nat. Commun., 4, 2980.
23. Santangelo,L., Giurato,G., Cicchini,C., Montaldo,C., Mancone,C., Tarallo,R., Battistelli,C., Alonzi,T., Weisz,A. and Tripodi,M. (2016) The RNA-binding protein SYNCRIP is a component of the hepatocyte exosomal machinery controlling microRNA sorting. Cell Rep., 17, 799-808.
24. Assié,G., Letouzé, E., Fassnacht,M., Jouinot,A., Luscap, W., Barreau,O., Omeiri,H., Rodriguez,S., Perlemoine,K., René-Corail,F. et al. (2014) Integrated genomic characterization of adrenocortical carcinoma. Nat. Genet., 46, 607-612.
25. Laurent,L.C., Chen,J., Ulitsky,I., Mueller,F .- J., Lu,C., Shamir,R., Fan,J .- B. and Loring,J.F. (2008) Comprehensive microRNA profiling reveals a unique human embryonic stem cell signature dominated by a single seed sequence. Stem Cells, 26, 1506-1516.