High-content screen in human pluripotent cells identifies miRNA-regulated pathways controlling pluripotency and differentiation

Background By post-transcriptionally regulating multiple target transcripts, microRNAs (miRNAs or miR) play important biological functions. H1 embryonic stem cells (hESCs) and NTera-2 embryonal carcinoma cells (ECCs) are two of the most widely used human pluripotent model cell lines, sharing several characteristics, including the expression of miRNAs associated to the pluripotent state or with differentiation. However, how each of these miRNAs functionally impacts the biological properties of these cells has not been systematically evaluated. Methods We investigated the effects of 31 miRNAs on NTera-2 and H1 hESCs, by transfecting miRNA mimics. Following 3–4 days of culture, cells were stained for the pluripotency marker OCT4 and the G2 cell-cycle marker Cyclin B1, and nuclei and cytoplasm were co-stained with Hoechst and Cell Mask Blue, respectively. By using automated quantitative fluorescence microscopy (i.e., high-content screening (HCS)), we obtained several morphological and marker intensity measurements, in both cell compartments, allowing the generation of a multiparametric miR-induced phenotypic profile describing changes related to proliferation, cell cycle, pluripotency, and differentiation. Results Despite the overall similarities between both cell types, some miRNAs elicited cell-specific effects, while some related miRNAs induced contrasting effects in the same cell. By identifying transcripts predicted to be commonly targeted by miRNAs inducing similar effects (profiles grouped by hierarchical clustering), we were able to uncover potentially modulated signaling pathways and biological processes, likely mediating the effects of the microRNAs on the distinct groups identified. Specifically, we show that miR-363 contributes to pluripotency maintenance, at least in part, by targeting NOTCH1 and PSEN1 and inhibiting Notch-induced differentiation, a mechanism that could be implicated in naïve and primed pluripotent states. Conclusions We present the first multiparametric high-content microRNA functional screening in human pluripotent cells. Integration of this type of data with similar data obtained from siRNA screenings (using the same HCS assay) could provide a large-scale functional approach to identify and validate microRNA-mediated regulatory mechanisms controlling pluripotency and differentiation. Electronic supplementary material The online version of this article (10.1186/s13287-019-1318-6) contains supplementary material, which is available to authorized users.

Distinct from TFs, microRNAs (miRs) are small (~22 nt) RNA molecules that act as major post-transcriptional regulators, by binding to partially complementary target mRNAs, blocking their translation and/or leading to their degradation [6]. Despite their small number (2,654 mature miRs, miRBase release 22), they regulate a large fraction of the transcriptome. Their biogenesis involves their transcription as primary miRs (pri-miRs), which assume secondary self-complementary hairpin structures, that are processed in the nucleus by the DROSHA/DGCR8 microprocessor complex giving rise to precursor hairpin miRs (pre-miRs) [7]. These pre-miRs are exported to the cytoplasm and further processed by Dicer to generate the mature miR duplex, composed by one 5′ and 3′ strands (-5p and -3p suffixes in miR names, respectively). This duplex is loaded into the RNA-induced silencing complex (RISC), which contains an Argonaute (AGO) protein, and the passenger strand is removed and degraded (historically referred as star "*" strand), while the guide strand drives the complex to the target mRNAs, inhibiting their translation and/or destabilizing and reducing their transcript levels [6,8].
Transcriptomic studies have identified miRs specifically expressed in pluripotent cells, and repressed upon differentiation, that could be involved in the maintenance of their undifferentiated properties, as well as miRs with opposite roles, induced upon differentiation [9,10]. For instance, Stadler et al. found that many miRs members of the miR-302 and miR-371-373 clusters and of the miR-17 family were highly expressed in pluripotent cells and downregulated in most cell lines during differentiation (including hESCs and NT2 cells); conversely, several miRs allocated in specific chromosomal clusters (such as miR-24/miR-27a/miR-23a) were induced during differentiation [10]. Interestingly, pluripotency-related miRs can improve iPSC reprogramming induced by TFs or even completely substitute them [11][12][13][14][15].
Despite the advances in functional genomics, with genome-wide methods able to profile the transcriptome, proteome, and miRNome of pluripotent cells, a true understanding of how these gene products impact cell biology depends on experimental approaches able to systematically identify and quantify molecular and phenotypic events affected by gain or loss-of-function. In this context, high-content screening (HCS) combines the automated acquisition of fluorescence microscopy images of cell-based assays (carried in multiwell plates) with quantitative digital image processing, allowing the simultaneous study of multiple characteristics in stem cells [16]. Specifically, by coupling immunofluorescence of selected markers to nuclear and cytoplasmic fluorescent stains, several morphometric and intensity measurements can be simultaneously quantified at the single-cell level [17,18], allowing the functional dissection of phenotypic consequences of miR [19]. Up to date, only three small focused screenings have evaluated miR roles in pluripotency, all using a mouse ESC line knockout for DGCR8 [18]. This cell line lacks most endogenous miRs [20][21][22] and was initially generated to assess the role of DGCR8 in miR processing and to study the global role of miRs in early embryonic development and ES cell differentiation. Although mESCs from Dgcr8 −/− embryos are morphologically normal and express ESC-specific markers, they show an extended population doubling time with accumulation in the G1 cell cycle phase and show large differentiation defects, failing to repress pluripotency markers [23]. How results from these screenings translate to cells containing Dgcr8, and human cells, remains to be shown.
With this in mind, our aim was to identify microRNAs and their predicted targeted pathways or processes, contributing to pluripotency or differentiation of human pluripotent cells. To this end, we developed a cell-based assay to investigate, by HCS, the functional effects of 31 miRs differentially expressed between pluripotent and differentiating cells [10,24], using two of the most studied human pluripotent cell lines in the literature, namely NTera-2 (NT2) embryonal carcinoma (EC) cells, a classical model to study pluripotency [25,26], and the H1 cell line, the first derived hESCs [1]. These cells share several characteristics, including the expression of pluripotency markers (e.g., OCT4, Nanog) [27][28][29], miR expression profiles [30], and similar responses upon differentiation induced by all-trans retinoic acid (atRA), including changes in miR expression profile [10], morphology, and expression of pluripotency markers [31]. Importantly, NT2 cells also closely resemble H1 hESCs in terms of cell cycle control [32]. Human and mouse ESCs are characterized by a high proliferation rate and a shortened G1 phase, limiting the differentiation associated with this phase [32][33][34]. Importantly, the Cdk1-Cyclin B complex seems the only complex with a cell cycle-dependent behavior in mESCs [35] and, in contrast to other cyclins (A, D, or E), Cyclin B1 (CCNB1) is the only cyclin required until implantation of the embryo [36]. Cyclin B levels increase during late-S across G2 and M phases and drastically drop upon re-entry into G1, when it abruptly translocates into the nucleus (shortly before nuclear envelope breakdown), being degraded during G1 [37]. Components of the Cyclin B1 pathway are centrally involved in the dissolution of the pluripotent state, strongly connecting G2 phase with pluripotency [38].
With that in mind, in addition to Oct4 and several morphometrical parameters, we evaluated Cyclin B1 as a surrogate marker of the undifferentiated status associated with the G2 phase of the cell cycle in pluripotent cells. By clustering miRs based on the induced multiparametric phenotypic profiles, and by applying enrichment analysis to the sets of predicted targets shared by clusters of miRs inducing similar phenotypes, we were able to uncover signaling pathways and biological processes potentially mediating their effects on pluripotency and differentiation. Specifically, we show evidence that miR-363-3p decreases the transcript levels of its predicted targets, the receptor NOTCH1 and PSEN1 (a Y-secretase component mediating Notch activation), counteracting the differentiation induced by the Notch ligand DLL1.

Study design (HCS assay and screening)
Detailed supplemental procedures can be found in an additional file (see Additional file 1). Given the lack of systematic studies evaluating the function of miRs in human pluripotent cells, we choose to functionally evaluate a selected set of 31 miRs previously shown to be differentially expressed between undifferentiated and differentiating pluripotent human cells (see Additional file 2: Table S1). These miRs were found to be induced (11 miRs) or repressed in nine hESC lines (including H1 and H9) and NT2 cells, upon transference from a condition favoring their undifferentiated state to a condition promoting a rapid shift toward multilineage differentiation [10]; most miRs were initially cloned and characterized in mESCs [39] and/or hESCs [40]. Additionally, part of the selected pluripotency-related miRs were expressed at higher levels in mESCs and miPSCs (as compared to mouse embryonic fibroblasts-MEFs), had promoters associated with Oct4/ Sox2/Nanog/Tcf3, and also were actively transcribed during miPSC reprogramming [24].
Based on the above, we designed a cell-based assay, compatible with HCS, to simultaneously evaluate morphological and molecular changes related to pluripotency, cell cycle status, and differentiation ( Fig. 1, top left panel). Cells were immunostained for OCT4 and Cyclin B1 and counterstained with nuclear (Hoechst) and cytoplasm (HCS CellMask Blue) fluorescent dyes. This allowed the quantitative measurement of several morphometrical as well as intensity features of the markers OCT4 and Cyclin B1 at the single-cell level, in the nuclei and cytoplasm compartments; stainings were optimized so that segmentation of the nucleus and cytoplasm could both be performed in the blue channel, respectively as high and medium fluorescence intensity objects. We used this assay to carry two functional miR screens, using H1 hESCs and NT2 EC cells (Fig. 1, top right panel). For each screening, two 96-well plates were used to accommodate triplicates of the 31 miR mimics evaluated plus negative controls (miR-Ctr, a miR without targets in the human transcriptome) and positive controls for differentiation (atRA and esiRNA-OCT4), as well as a lethal siRNA (siRNA-UBC, targeting the essential gene UbiquitinC) to control for transfection efficiency. Following reverse transfection and culture for 3-4 days, automated image acquisition (9 sites per well, × 10 objective) was performed using an ImageXpress micro XLS HCS system (Molecular Devices) and image and data analysis was performed using CellProfiler and KNIME open-source softwares.
Transfection efficiency was evaluated by comparing the total number of cells (nucleus counts) in wells with cells transfected with a lethal siRNA-UBC control, to those transfected with control miR-Ctr. To further evaluate if both plates of each screening were comparable, for each plate, we evaluated the reduction of nuclear OCT4 staining (as compared to cells transfected with miR-Ctr) in the wells with cells transfected with esiRNA against OCT4 or treated with 10 μM atRA.
Concentration of staining reagents (dyes or antibodies) was defined following titration, in order to guarantee a non-limiting excess of reagents, i.e., the reagent concentration tested, in which there is no quantifiable increase in staining intensity, as compared to the lower concentration tested. Of notice, automation of image acquisition and analysis with the exact same parameters, an inherent characteristic of the HCS approach, guaranties an unbiased and reproducible quantitation, thus allowing the results obtained from distinct wells to be compared [41,42]. In addition, in order to further validate the quantitative results obtained by the antibodies used in our HCS assay, we compared them to those obtained with distinct antibodies (a total of three anti-OCT4 and two anti-cyclin B1 antibodies). Moreover, western blotting was used to further validate the specificity of selected antibodies and, also, to compare how the quantitative results obtained by HCS relate to those obtained by densitometric analysis of western blot protein bands. A detailed description can be found in the "Antibody validation" section of the supplemental procedures (Additional file 1).

Image and data analysis
Image and data analysis was carried out using the softwares CellProfiler (cellprofiler.org) [43,44] and KNIME (knime.com) [45]. Once segmented, several morphometrical features, as well as intensity-related measurements, were obtained from the nucleus and cytoplasm of each cell. Moreover, in order to minimize experimental plateto-plate variation, we carried a normalization using the miR-Ctr wells (present in both plates) as references. The resulting normalized values are represented as percentage of control (POC), allowing a direct comparison of all treatment conditions in both plates of each screening [45]. Median values from each quantified parameter were combined in a multiparametric phenotypic profile representing the effect of each miR in the whole population. Details are provided in the supplemental experimental procedures (see Additional file 1).
Phenotypic clustering of miRs, identification of shared predicted targets, and pathway analysis In order to obtain a less redundant and more naturally interpretable set of biologically relevant phenotypic parameters, the following features were selected to compose multiparametric phenotypic profiles: cell count, solidity (a feature varying from 0, for complex shapes with reentrances, up to 1, for solid shapes), eccentricity (varying from 0 to 1, from round to increasingly elliptical shapes), nuclear and cellular areas, nuclear and Fig. 1 General outline of the study. First: A high-content screening (HCS) assay based on H1 hESCs and NTera-2 cells was developed, to explore pluripotency and differentiation. Several morphometrical features, as well as intensity-related measurements (from OCT4 and Cyclin B1 staining), were obtained from nuclei and cytoplasm compartments. Second: The HCS assay was used in a miR functional screen in 96-well plates to investigate the effects of 31 miRs. Following transfection and culture, images were acquired with an ImageXpress micro XLS HCS system (Molecular Devices). Third: Following image and data analysis (with CellProfiler and KNIME softwares, respectively), the multiparametric phenotypic profile describing the functional effect of each miR was submitted to hierarchical clustering (using Cluster 3.0 and Java Treeview), allowing the identification of miRs with similar functional effects (pro-pluripotency or pro-differentiation). Fourth: TargetScan was used to identify predicted targets shared by miRs in the same cluster (i.e., inducing similar phenotypes). Signaling pathways and biological processes enriched for these shared targets were identified using the DAVID tool cellular perimeter, and nuclear and cytoplasmic OCT4 and CCNB1fluorescence intensities. The phenotypic profiles obtained for all miR treatments were submitted to hierarchical clustering using the software Cluster 3.0, using centered correlation metrics and average linkage [46], and heatmaps and clusters were generated and visualized using Java Treeview [47]. Next, we used all predicted mRNA targets of each miR, downloaded from TargetScan Human 7.1 [48], to identify and select transcripts commonly targeted by miRs belonging to the same cluster. Finally, we used the DAVID (V6.8) [49,50] to identify enriched signaling pathways and biological processes. Details are provided in the supplemental experimental procedures (see Additional file 1).

Evaluation of miR effects on transcript levels of predicted target
NT2 cells were reverse transfected with selected mimics in 12-well plates and cultured for 2 days prior to RNA extraction. Total RNA was reverse transcribed using High Capacity cDNA Reverse Transcription Kit, and qPCR was performed using either TaqMan Universal PCR Master Mix or Power SYBR Green PCR Master Mix. Relative gene expression was obtained using the 2^-DDCT method [51] (see Additional file 1, detailed supplemental experimental procedures). TaqMan probes and SYBR Green PCR primers used are listed in Table S2 and Table S3, respectively (see Additional file 2).

Regulation of Notch pathway by microRNAs in pluripotent stem cells
To further dissect the potential repression of Notch signaling by miRs promoting pluripotency features, we obtained a list of genes related to the Notch pathway and identified all miRs targeting them. Four selected miRs with more than ten targets (miR-302c-3p, miR-101-3p, miR-363-3p, and miR-92a-3p) and two miRs with less than five targets (miR-222-3p and miR-371-3p) were individually transfected NT2 cells in 96-well plates and posteriorly co-cultured with OP9-ctrl or OP9-DL1 cells. Cells were fixed and stained for OCT4, nucleus, and cytoplasm and imaged as described before. The nuclear OCT4 intensity was measured using CellProfiler, and the results were compared to the cells transfected with pre-miR negative control and co-cultured with OP9-Ctr. Details are provided in the supplemental experimental procedures (see Additional file 1).

Statistical analysis
GraphPad Prism 6.0 software (La Jolla, CA, USA) was used for the statistical analysis. All experiments were performed in triplicates. For qPCR data, we used a onetailed non-parametric Mann-Whitney test. For HCS experiments, we used a non-parametric Kruskal-Wallis test followed with Dunn's multiple comparison test.

Screening quality controls
Transfection efficiency was similarly high in both screening plates for both cell lines (transfection efficiency > 85%), as indicated by the marked reduction in the total number of cells in wells transfected with the lethal siRNA-UBC control, as compared to those transfected with negative miR-Ctr molecules (see Additional file 3: Figures S1 and S2). Additionally, we also evaluated the reduction in OCT4 nuclear intensity following transfection with an esiRNA against OCT4 (as compared to controls wells), allowing us to address the quantitative inter-plate reproducibility of our experimental setting (a sum of transfection efficiency, anti-OCT4 staining, image segmentation, and quantitation of the resulting fluorescence intensity). The reduction in OCT4 nuclear intensity upon transfection with esiRNA-OCT4 (or upon atRA-induced differentiation), as compared to cells transfected with miR-Ctr, was also comparable between both plates in the two screenings (see Additional file 3: Figures S1 and S2). In order to further validate the quantitative results obtained by the antibodies used in our HCS assay, we compared them to those obtained with distinct antibodies using quantitative microscopy. As compared to untreated cells, the relative reduction in the levels of OCT4 and cyclin B1 upon atRA treatment (as quantified by HCS) was strikingly similar for all antibodies used, ranging from 14 to 16%, respectively (see Additional file 3: Figure S3A). Densitometric analysis of western blots carried with proteins extracted from cells cultured in parallel, under the same experimental conditions, was much larger, with a 31-34% reduction in OCT4 levels and 50-53% reduction in cyclin B1 levels (see Additional file 3: Figure S3B). Altogether, these results validate the quantitative results obtained by the antibodies used in our HCS assay and allowed us to confidently proceed with the analysis of the screening results.

Phenotypic screening in NT2 and H1 cells
For both cell lines, for each miR and each parameter quantified, we obtained 27 values derived from 9 images acquired in each of the 3 replica wells. The median value for OCT4 nuclear median intensity was used to generate Fig. 2a. The primary results quantified for all 27 sites (and all miRs) in NT2 and H1 cells can be seen (respectively) in (Additional file 3: Figure S4A and B), allowing the variability of our HCS assay to be visually accessed. Overall, both cell lines showed similar responses to the evaluated miRs. In general, miRs enriched in pluripotent cells, and whose expression was reportedly downregulated upon differentiation of H1 and/or NT2 cells [10], had a propluripotency effect in our experimental setting, increasing nuclear and cytoplasmic OCT4 and CCNB1 levels. On the other hand, miRs reported as upregulated upon differentiation of H1 and/or NT2 cells [10] had the opposite effect ( Fig. 2a). For information on the evaluated miR, regarding their families and the clusters they belong, see (Additional file 2: Table S1) and Fig. 2b.
Following hierarchical clustering of the multiparametric profiles, miRs eliciting similar phenotypic effects were grouped together (for details, see bottom right panel of Fig. 1 and supplemental experimental procedures in Additional file 1). Importantly, median intensity levels of OCT4 and CCNB1 in the nucleus and Effects of miRNAs on nuclear OCT4 levels in H1 hESCs and NT2 cells. a H1 hESCs and NT2 cells were transfected with miR mimics and the negative control miR-Ctr (PMC) and then submitted to quantitative fluorescence microscopy. OCT4 nuclear median intensity is represented as a percentage relative to the value observed in cells transfected with PMC (NT2 cell depicted as triangles and H1 hESCs as circles). Colored bars in the bottom of the figure indicate the observed change in miRNA expression in H1 and NT2 cells, following induction of differentiation [10]. b Graphical representation of selected primary transcripts of miR clusters, with the relative position of the pre-mirs and of the mature miRs. Mature miRs are represented as colored boxes in the 5′ (-5p miR) or 3′ (-3p miR) side of the corresponding pre-mir. Each miR family is represented with a distinct color. Families were defined, as represented in the TargetScan 7 database. For simplicity, only the first representative miR of the family was used in the figure legend. Additional information regarding families can be found in Table S1 (in Additional file 2). Boxes with dotted lines indicate miRs that were not evaluated in our study  Figures S4 and S5). This corroborates the assumption that (at least for these microRNAs and cell lines) changes in OCT4 and CCNB1 levels are able to indicate if a given microRNA acts as a pro-pluripotency or prodifferentiation factor. As can be seen in the clustering results obtained for NT2 cells (Fig. 3) and H1 cells (Fig. 4), the miRs were divided into two major clusters: a cluster containing miRs with pro-differentiation effects (i.e., reducing OCT4 and CCNB1 levels), identified as cluster "A" for both cell lines, and a cluster with pro-pluripotency miRs (i.e., increasing OCT4 and CCNB1 levels), identified as cluster "B" in NT2 cells, and as clusters "B" and "C" in H1 hESCs.
One of the most striking differences between the cell lines was the effect elicited by miR-19ab-3p (miR19-3p family) and miR-20ab-5p (also from family miR-17-5p/ 20-5p/93-5p/106-5p/519-3p/526-3p), deriving from the paralog clusters mir-17~92a-1 and mir-106a~363 (Fig. 2). Despite being repressed upon differentiation of both cell lines, these miRs markedly reduced OCT4 and CCNB1 levels in NT2 EC cells, but had the opposite effect in H1 hESCs. This result specifically highlights how miR-20-5p Fig. 3 Clustering of miR-induced multiparametric phenotypic profiles in NT2 cells. NT2 cells were transfected with miR mimics and the negative control miR-Ctr (PMC) and then submitted to quantitative fluorescence microscopy. Multiparametric phenotypic profiles specific for each miR were submitted to hierarchical clustering analysis (centered correlation and average linkage). The resulting dendograms above or to the left of the heatmap indicate, respectively, the similarities between the phenotypes induced by miRs (potentially reflecting shared targets and mechanisms of action) or the correlation between distinct phenotypic features. The measurements depicted in the heatmap were calculated as percentage of control (POC) (PMC in black). Highest POC values above or below the control are represented in bright yellow or blue, respectively and miR-106-5p have different effects, despite belonging to the same family. Another unexpected finding was the observed effect of miR-145-5p, miR-181d-5p, and miR-222-3p, which despite being induced upon differentiation of H1 hESCs [10] caused an increase in OCT4 and CCNB1 levels in both cell lines.
Although the two main miR groups, resulting from the hierarchical cluster analysis in both cell lines, successfully separated pro-pluripotency from pro-differentiation miRs, in H1 hESCs, miR-29a/b was grouped with propluripotency miRs in cluster B (instead of cluster A), as a result of its effects on cell morphology and, more importantly, as a result from a marked reduction in cell counts. Interestingly, this group included miR-18a/b-5p, miR-101-3p, and miR-363-3p (all with similar effects in NT2 cells), as well as miR-24-3p. Worth noticing, miR-92a-3p (from the same family as miR-363-3p) did not reduce cell counts in neither of the cell lines. Moreover, while both miRs increased cell size in H1 hESCs, in NT2 cells, they reduced. These findings demonstrate that the cell context largely affects the net phenotypic effect of miRs, even from the same family.

Identification of pathways targeted by clustered miRs
Given that miRs can have similar phenotypic effects by targeting common components of a given signaling pathway (or biological process) or, alternatively, by targeting distinct components of the same pathway or process [52,53], we identified the predicted mRNA targets of each miRNA, using TargetScan Human 7.1 [48] and selected transcripts commonly targeted by miRs belonging to the same cluster. Finally, we used the Database for Annotation, Visualization and Integrated Discovery (DAVID) [49,50] to identify signaling pathways and biological processes enriched for the set of shared targets (see Additional file 4, an Excel file with all results from the pathway analyses), which potentially correspond to the most relevant pathways (Fig. 1, bottom left panel).
We next compared the pathways identified for each of the three phenotypic clusters with pro-pluripotency characteristics in NT2 cells. From 102 potentially regulated pathways, 45.1% were exclusive for subcluster B.1, 7.8% for subcluster B.2a, and 19.6% for subcluster B.2b (top left Venn diagram in Fig. 5). On the other hand, from 148 pathways identified for the pro-pluripotency clusters in H1 hESCs, 10.1% were exclusive for subcluster B.1, 17.6% for subcluster C.1, and 22.3% for subcluster C.2 (top right Venn diagram in Fig. 5). Interestingly, some pathways were shared by all clusters, like FoxO, mTOR, and DICER pathways. A similar analysis was carried for the pathways identified for the pro-differentiation clusters in both cell lines (see Additional file 5, an Excel file with all pathway comparisons).
Next, we compared all the pathways identified for propluripotency and pro-differentiation clusters in both cell Fig. 4 Clustering of miR-induced multiparametric phenotypic profiles in H1 hESCs. H1 hESCs were transfected with miR mimics and the negative control miR-Ctr (PMC) and then submitted to quantitative fluorescence microscopy. Multiparametric phenotypic profiles specific for each miR were submitted to hierarchical clustering analysis (centered correlation and average linkage). The resulting dendograms above or to the left of the heatmap indicate, respectively, the similarities between the phenotypes induced by miRs (potentially reflecting shared targets and mechanisms of action) or the correlation between distinct phenotypic features. The measurements depicted in the heatmap were calculated as percentage of control (POC) (PMC in black). Highest POC values above or below the control are represented in bright yellow or blue, respectively lines (bottom Venn diagram in Fig. 5). Interestingly, among pathways exclusively enriched for targets of propluripotency-related miRs, we found one describing "Presenilin action in Notch and Wnt signaling." Importantly, this pathway was among the pathways exclusively enriched for miR-targets of B.1 clusters, in NT2 and H1 hESCs, which corresponded to pluripotency-related miRs with the distinguishing feature of restricting proliferation.

miR effects on predicted target transcript levels
Given that mammalian microRNAs predominantly act to decrease target mRNA levels [6], in order to evaluate the potential regulation of some predicted targets by selected miRs, we transfected NT2 cells with miR mimics or a control miR and then carried quantitative PCR. From 10 predicted targets of miR-29b-3p (from clusters A.3 and B.2 in NT2 and H1, respectively), a statistically significant decrease (P < 0.05), as compared to controls, was only observed for KLF4 and APC. However, a noticeable decrease was also observed for STAT3, IL2RA, FGFR1, and TGFB3 transcripts, whereas CDKN2B were significantly increased, unexpectedly (Fig. 6a)  It is possible to observe that some miRs caused statistically significant reductions in the levels of their target mRNA transcripts (Fig. 6b). For instance, miR-181d-5p reduced the levels of PTEN and MAPK1, miR-20a-5p of FGF2 and TGFBR2, miR-24-3p of TCF3 and GSK3B, miR-30a-5p of MAPK1, and miR-222-3p and miR-373-3p of TGFBR2.

miR-363-3p regulates Notch signaling pathway and promotes pluripotency features
The identification of the pathway describing "Presenilin action in Notch and Wnt signaling" as exclusively enriched with predicted targets from pluripotency-related miRs, characteristically restricting the proliferation of NT2 and H1 cells, attracted our attention, and we focused further studies on this pathway. In order to investigate the potential post-transcriptional regulation of the Notch pathway, we compiled a list of all Notch pathway components using the KEGG database and used TargetScan to find which components were predicted targets of the evaluated miRs (see Additional file 2: Table S4). Next, we selected distinct miRs to functionally test their ability to counteract the differentiation induced by Notch activation. To that end, we transfected NT2 cells with pre-miR negative control (PMC) and co-cultured them with murine OP9-DL1 stromal cells, expressing the Notch activating ligand Delta-Like 1 (DL1) in their surface, or with control OP9 stromal cells (OP9-Ctrl). Additionally, NT2 cells were transfected with selected miRs and co-cultured with OP9-DL1 cells.

Discussion
We present the first focused miR functional highcontent screen carried in human pluripotent cells. By identifying targets shared by miRs causing similar multiparametric phenotypic profiles, we were able to pinpoint post-transcriptionally regulated signaling pathways involved in pluripotency and differentiation. In particular, we show that the pluripotency-related microRNA miR-363-3p targets NOTCH1 and PSEN1 inhibit differentiation mediated by Notch signaling. Moreover, by comparing NT2 cells to H1 hESCs, we show how extremely contrasting effects can result from closely related miRs in the same cell or from the same miR in different cells.

Similarities and differences between hESCs and NTera-2 cells
Despite overall similarities, some miRs elicited contrasting effects depending on the cellular context. For example, OCT4 and CCNB1 levels were strongly repressed by miR-19a/b-3p and miR-20a/b-5p in NT2 cells, but were robustly induced in H1 cells. In turn, miR-92a-3p and miR-363-3p had opposing effects on cell size, decreasing it in NT2 cells, while increasing it in H1 hESCs. Conversely, some closely related miR (members of the same family) had opposing effects in the same cell line, despite identical seeds. For instance, miR-20a/b-5p and miR-106a-5p showed pro-differentiation and propluripotency effects in NT2 cells, respectively. In turn, miR-92-3p and miR-363-3p had opposite effects on proliferation and nucleus size in the same cellular context, in both cell lines.
Importantly, we found that miR-106a-5p, miR-302a/b/ c/d-3p, miR-372-3p, and miR-373-3p had the same effect in H1 hESCs and NT2 cells: increasing proliferation (i.e., cell counts), cell size (i.e., area and perimeter), and expression of OCT4 and CCNB1 in nucleus and cytoplasm (i.e., median intensity) and reducing nucleus size. Of notice, increases in cell size and reduction in nucleus size were unanticipated effects captured only as a result of our multiparametric high-content screen.
In mice, the miR-290~295 cluster is broadly expressed throughout the embryo, but diminishes a few days following implantation, remaining highly expressed in extraembryonic tissues, in line with its roles in implantation and placenta development [56,57]. In contrast, miR-302a-d cluster is expressed in the embryo proper only after implantation [58,59]. While deletion of miR-302a-d particularly affects neural development, deletion of miR-290 and miR-302a-d clusters results in early embryonic lethality, revealing redundant functions during early development [60].
Also using DGCR8 KO mESCs, Ma et al. evaluated the effects of 40 microRNAs expressed at higher levels in MEFs and EBs (as compared to mESC) and predicted to target pluripotency-associated transcription factors, including Oct4, Sox2, Nanog, Klf4, c-Myc, Lin28a, Sall4, Rex1, and Stella [21]. Many miRs evaluated in our study were also evaluated by Ma et al., in terms of their effects on Oct4 staining, though not in a quantitative way. Of all 40 miRs, 20 decreased Oct4 staining (including miR-24-3p, 27a-3p, 92a-3p, and 29a-3p), 12 mildly decreased Oct4 levels (including miR-23a-3p), and 8 miRs increased Oct4 intensity (including miR-18a-5p and 30a-5p). Of notice, we observed similar effects on Oct4 nuclear median intensity (in both human cell lines) only for miR-23a-3p, miR-24-3p, 27a-3p, and miR-29a-3p, which decreased Oct4 levels in our screenings. Opposite results were found for miR-92a-3p, which increased Oct4 levels in our screening, and for miR-18a-5p and miR-30a-5p, which repressed Oct4 levels. Nevertheless, our results regarding Fig. 7 miR-363-3p targets NOTCH1 and PSEN1 and inhibits Notch-induced differentiation in NT2 cells. a NT2 cells were transfected with miR mimic and negative control (PMC), and 72 h later, they were co-cultured with OP9-Ctrl or OP9-DL1 stromal cells for more 24 h. Nuclear OCT4 median intensity was then quantified by automated quantitative fluorescence microscopy. Kruskal-Wallis test followed with Dunn's multiple comparison test. b NT2 cells were transfected with miR-363-3p mimics and PMC and NOTCH1, and PSEN1 transcript levels were evaluated 48 h later by qPCR. Mean expression level of PMC transfected cells were used as reference for the calculation of the relative expression, using the 2^-DDCT method. *p < 0.05 (one-tailed non-parametric Mann-Whitney test) miR-92a-3p and miR-30a-5p were in line with their expected roles based on the expression behavior in differentiating hESCs [10]. On the other hand, when evaluating mESC colonyforming ability, 17 miRs decreased it (including miR-24-3p, miR-27a-3p, miR-29b-3p), 15 did not affect it (including miR-29a-3p, miR-27b-3p, miR-92a-3p, miR-145a-5p, miR-18a-5p, and miR-30a-5p), and 8 enhanced it (including miR-23a-3p and 23b-3p). Since colonyforming ability derives, in part, from the proliferation and survival of plated cells, we compared it to the miR effects on cell counts obtained by us. In H1 hESCs, we found that miR-24-3p and miR-29a/b-3p decreased cell counts (in line with lower colony numbers), while miR-27a-3p increased cell counts (in contrast to decreased colony numbers). In line with higher colony numbers, miR-23a-3p also increased cell counts in H1 hESCs and NT2 cells. In contrast to the decreased colony formation ability observed by Ma et al., we found that miR-24-3p, miR-27a-3p, and miR-29b-3p increased cell counts in NT2 cells. These contrasts between DGCR8 KO mESCs and H1 hESCs may derive from differences related to the absence of endogenous miRs, but also from differences regarding the pluripotency state of mouse and human ESCs (naïve vs primed, respectively). Importantly, according to Ma et al., miR-27a and miR-24 would exert their pro-differentiation effects, in part, by directly targeting the pluripotency-associated factors Oct4 and Foxo1 and the signal transducers gp130 and Smads [21]. Given our findings, it is likely that the above mechanism may not completely apply to hESCs.
As mentioned before, many miRs able to downregulate AP activity on Dgcr8 knockout mESCs cannot do so on wild-type cells [61]. Additionally, DGCR8 also acts in the maturation of snoRNAs and in the degradation of telomerase RNA [62]. Finally, in the absence of endogenous miRs, the RNAi machinery of DGCR8 KO cells becomes completely available to the exogenously introduced miR, likely resulting in an enhanced effect (including off-target effects of siRNAs). Altogether, our results further stress important differences that compromise a straightforward comparison and translation of the findings obtained with DGCR8 KO mESCs to those obtained with wild-type ESCs, further highlighting the importance of our study. For further discussions, the reader can refer to our review comparing all arrayed genetic screens (siRNAs and micro-RNAs) carried in the context of pluripotency, reprogramming, and differentiation, published up to date [18].

microRNAs in naïve and primed pluripotency
In a more recent work, Gu et al. used DGCR8 KO mESCs to evaluate the transition from mESCs in a naive state of pluripotency to epiblast-like cells (EpiLCs) in a primed state. The authors identified 50 upregulated miRs, including those of the miR-302 cluster, while other miRs with similar seeds, such as miR-290 cluster and miR-17-20-106 families, were not altered. By using qRT-PCR, Gu et al. found that miRs from the miR-302 cluster strikingly repressed the expression of naive markers (Rex1 and Klf2) and induced the post-implantation epiblast marker Fgf5; nevertheless, those from the miR-290 cluster also facilitated the exit from naive pluripotency [22]. The role of miR-302 cluster miRs in the transition to the primed state would be in line with its expression following implantation [58,59]. Moreover, two studies identified microRNAs expressed at higher levels in naive cells of mice or humans (including those from the miR-371-373 cluster in human, or miR-290~295 in mice), as well as those expressed at higher levels in primed cells (miR-302 family and miR-363) [63,64].
Based on these results and given that the miRs enriched in both pluripotency states share similar seeds, it is evident that miRs from the same or related families (but belonging to clusters transcriptionally regulated in an independent manner during development) can have pluripotencyrelated roles in both naive and primed cells. Moreover, they may likely contribute bi-directionally during naive-primed interconversion, depending on the experimental setting, as exemplified by miRs from the miR-302 and miR-290 clusters facilitating the exit from naive pluripotency [22]. However, given that DGCR8 KO mESC fails to completely silence the naive program and to establish the primed one, it remains to be shown if these findings translate to wild mESCs or hESCs.

Context-dependent effects of microRNAs on pluripotency and reprogramming
While the pro-pluripotency roles of miR-302 and miR-294 were expected, since they are enriched in pluripotent cells and are repressed during differentiation, the propluripotency role of miR-181 family members is more intriguing. While Stadler et al. [10] found these miRs to be induced during differentiation of hESCs (miR-181b and d) and NT2 cells (miR-181a and b), we found that miR-181d-5p considerably increased OCT4 nuclear levels and cell counts, also reducing cell and nucleus size, in both our screenings. Importantly, ESCC miR mimics from miR-302 and miR-290 clusters and from the miR-181 family all promote the initiation phase during OSK-mediated reprogramming of MEFs into iPSCs [69][70][71], in line with their pro-pluripotency roles. In mESCs, miR-181 family members are transcribed from three distinct clusters, differentially bound by Oct4, Sox2, Nanog, and TCF3 TFs, and expressed in pluripotent and differentiated cells. While miR-181a-1~181b-2 cluster is bound by all these TFs and expressed in neural precursors and MEFs, miR-181c~181d cluster is bound by all TFs except TCF3 and is expressed only in mESCs; finally, cluster miR-181a-2~181b-1 (in chr1) is expressed only in neural precursors and is not bound by any of these TFs [24]. Interestingly, the differential regulation of these clusters by TCF3 during reprogramming could be implicated in the stage-specific regulation of reprogramming by Tcf3, promoting it in the early phase, while inhibiting in the later stages [72].
MiRs from the miR-181 family are also induced during mESC differentiation, targeting Cbx7, the primary Polycomb ortholog of Polycomb repressive complexes 1 (PRC1) which reads repressive H3K27me3 histone marks left by PRC2, leading to derepression of bivalent genes encoding lineage-specific TFs and markers [2,[73][74][75]. In contrast, ectopic expression of Cbx7 in mESC enhances self-renewal and inhibits differentiation and X chromosome inactivation; in turn, upon its knockdown, mESC colonies display a flattened morphology [73], characteristics related to the naïve or primed states, respectively. Thus, similar to the miR-302 family, although miR-181 family possesses pluripotency-related properties, it may be associated with the transition to the primed state, by targeting Cbx7.
In addition to miR-181d-5p, miR-145-5p and miR-222-3p were also induced in pluripotent cells upon differentiation [10] and similarly found to considerably enhance OCT4 nuclear levels in both our screenings (increasing cell counts and reducing cell and nucleus sizes). For miR-145-5p, our results were particularly unexpected, as this miR was shown to target OCT4, SOX2, and KLF4, negatively regulating their translation during cell differentiation [76]. Moreover, the inhibition of this miR during reprogramming was shown to increase the efficiency of iPSC generation [77]. These conflicting findings may result from differences between the transient effect mediated by transfection of mimics, as compared to the sustained and strong expression mediated by lentiviral vectors. In fact, contrasting findings arising from different methods used were previously pointed out by our group in the study of the miR-29 family. Using transfection of synthetic miRs or inhibitors, our group and others showed that miR-29a hampers reprogramming by different mechanisms, including activation of WNT/beta-catenin signaling (by targeting GSK3B) and targeting of active DNA demethylation enzymes of the TET family [71,78,79]. Strikingly in contrast, expression of miR-29 throughout the process using a retroviral vector promotes reprogramming, an effect resulting from passive global demethylation following constitutive knockdown of the targets Dnmt3a and Dnmt3b [80]. However, transient reduction of DNMT3a/b by siRNAs does not affect reprogramming [81].
Many effects of miRs (such as ESCC miRs) on pluripotency and reprogramming have been attributed to their function in the regulation of key cell cycle regulators [20,61,[82][83][84]. In fact, while induction of cell proliferation increases reprogramming efficiency, cell-cycle arrest inhibits it [85]. Thus, many miRs physiologically associated with differentiation of ESCs may promote reprogramming and pluripotency in vitro, trough cell cycle-mediated effects.

miR-363-3p sustains pluripotency by repressing Notchinduced differentiation
By comparing all the pathways enriched for predicted targets of the pro-pluripotency clusters to those of the pro-differentiation clusters, we noticed that a pathway describing "Presenilin action in Notch and Wnt signaling" was exclusively associated with pluripotency-related clusters, more specifically with those that characteristically restricted proliferation in NT2 and H1 hESCs. This prompted us to explore the potential role of Notch signaling in pluripotency and more specifically to evaluate the potential role of selected miRs in the inhibition of Notch signaling.
In general, Notch signaling begins with the interaction of surface-bound ligands (such as Delta or Jagged) with transmembrane Notch receptors (NOTCH1 to 4), leading to its cleavage (by a gamma-secretase complex containing presenilin, PSEN1) and release of the Notch intracellular domain (NICD), which translocates to the nucleus and forms a complex with the protein RBP-Jk (CSL). This complex binds to DNA cis-regulatory elements along with coactivators proteins (including p300 and proteins of the MAML family), activating the transcription of its target genes [86]. As shown by us, Notch signaling (mediated by OP9-DL1 cells) induced differentiation of co-cultured NT2 cells. Moreover, miR-363-3p was able to counteract Notch-induced differentiation, also reducing the transcript levels of its predicted targets PSEN1 and NOTCH1, central players in Notch signaling, in line with the previously reported direct targeting of NOTCH1 by miR-363-3p [87,88].
Components of the NOTCH pathway are expressed both in hESCs and in NTERA2 cells [89]. Although initial studies concluded that Notch would positively regulate the proliferation of hESCs [90,91], later studies revealed that this conclusion was likely derived from the unspecific cytotoxic effect of the Notch inhibitor used in these studies, including DAPT [91] and L685458 [90] and, also, by the unspecific activation of Notch by Tripsin/EDTA passaging during culture [92]. In fact, although the Notch pathway is inducible in hESCs, it remains inactive and is not necessary for the propagation of undifferentiated cells, but instead is required for the maintenance of differentiating cells that accumulate in culture [92]. The pro-differentiation role of Notch in hESCs was corroborated in a study showing that Notch signaling inhibition by GSI-18 maintained cells undifferentiated, preventing spontaneous differentiation to all three germ layers, while allowing trophoblastic differentiation [93]. Interestingly, Oct4 physically interacts with RBP-Jk in mESC [94], and siRNA knockdown of OCT4 in H1 hESCs induces the transcription of Notch pathway components and targets, including NOTCH2, TLE2, DTX4, and HES1 [95], suggesting a connection between Notch and Oct4-regulated gene expression.
Interestingly, Annab et al. identified two coexisting subpopulations in hESC, one composed of rounder cells expressing higher levels of Nanog and KLF4 and a more heterogeneous population expressing high levels of Notch components, including NOTCH1, NOTCH2, JAG1, DLL1, HES1, and DTX1 [96]. Given that hESCs chemically reverted to the naive pluripotency state express higher levels of Nanog and KLF4 [97], this could indicate that the expression of Notch components could be associated with a primed population of hESCs. Strikingly, a recent study in search of surface markers capable to distinguish naive and primed states of hESC identified NOTCH receptors 1 and 2 as being specifically expressed in the primed state, while the LIF coreceptor (GP130/IL6ST) was detected exclusively in naive-state hESCs [98].
Importantly, a mechanism involving the oscillatory activation of Notch (mediated by the inhibition of Notch signaling by its transcriptional target HES1) would contribute to the heterogeneous differentiation responses displayed by ESCs. Specifically, cells expressing low or high levels of Hes1 (i.e., with active or inactive Notch signaling, respectively) differentiate preferentially into neural or mesodermal cells, respectively [99][100][101].
Altogether, this would indicate that expression of Notch components in ESCs would be associated with primed cells displaying a more prone and heterogeneous potential to differentiation, what would be linked to the oscillatory activity of the Notch-Hes1 axis. In turn, the absence of Notch components would be associated with a more homogeneous cell population of naïve cells expressing higher levels of pluripotency factors and with the potential to give rise to the trophectoderm. Up to date, the function of the Notch pathway in the context of the transition between naive and primed states has not been investigated in the literature. However, a recently deposited patent (WO2014174470), defining conditions (referred to as WIS-NHSM) for naive cell generation [102], suggests that Notch inhibitors could contribute to the consolidation of the naive pluripotency state of ESCs [103]. We are currently undertaking this investigation. Thus, although miR-363 is enriched in primed cells [63,64], inhibition of the Notch pathway may likely be a general mechanism for the maintenance of pluripotency in both naive and primed ESCs.

Conclusions
By using automated quantitative fluorescence microscopy (i.e., high-content screening) to evaluate the phenotypic effects of microRNA mimics transfected into pluripotent cell lines and by identifying transcripts commonly targeted by microRNAs inducing similar multiparametric phenotypic profiles (as revealed by hierarchical clustering and in-silico target prediction), we were able to identify signaling pathways and biological processes post-transcriptionally modulated by distinct microRNA groups. Integration of this type of data with similar data obtained from siRNA screenings (using the same HCS assay) could provide a large-scale functional approach to identify and validate microRNAmediated regulatory mechanisms controlling pluripotency and differentiation.

Additional files
Additional file 1: Supplemental experimental procedures. A detailed description of materials and methods used. (DOCX 52 kb) Additional file 2: Table S1. Synthetic miR mimic molecules used in this study and their miRBase accessions. Table S2. List of genes and corresponding TaqMan probe-IDs qPCR. Table S3. List of genes and corresponding SYBR Green primers and PCR conditions used for qPCR. Table S4. List of Notch pathway components and the corresponding targeting miRNAs. (DOCX 33 kb) Additional file 3: Figure S1. Plate controls of the miRNA screen in NT2 cells. Figure S2. Plate controls of the miRNA screen in H1 hESCs. Figure S3. Antibody validation. Figure S4. Effects of miRNAs on nuclear OCT4 levels in NT2 cells and H1 hESCs cells. Figure S5 TFs: Transcription factors IMSL contributed to the conception and design, collection and/or assembly of data, data analysis and interpretation, manuscript writing, and final approval of the manuscript. JLSS contributed to the collection and/or assembly of data, data analysis and interpretation, manuscript revision, and final approval of the manuscript. SBPL contributed to the collection and/or assembly of data, data analysis and interpretation, manuscript revision, and final approval of the manuscript. DS contributed to the design, manuscript revision, and final approval of the manuscript. HLOB contributed to the design, manuscript revision, and final approval of the manuscript. BS contributed to the data analysis and interpretation, manuscript revision, and final approval of the manuscript. ACC contributed to the data acquisition and analysis and final approval of the manuscript. CHT contributed to the data acquisition and analysis and final approval of the manuscript. VMF contributed to the design, data interpretation, and final approval of the manuscript. DTC contributed to the conception, manuscript revision, and final approval of the manuscript. MAZ contributed to the conception, manuscript revision, and final approval of the manuscript. MG contributed to the conception, manuscript revision, and final approval of the manuscript. MM contributed to the conception and design, manuscript revision, and final approval of the manuscript. RAP contributed to the conception and design, collection and/or assembly of data, data analysis and interpretation, manuscript writing, and final approval of the manuscript.

Availability of data and materials
Part of the data generated or analyzed during this study are included in this published article [and its supplementary information files]. The remaining datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.
Ethics approval and consent to participate Not applicable.

Consent for publication
Not applicable.

Competing interests
The authors declare that they have no competing interests.