High-Content Analysis of MicroRNAs Facilitates the Development of Combinatorial Therapies for Vascular Diseases


 In response to vascular injury vascular smooth muscle cells (VSMCs) alternate between a differentiated (contractile) and a dedifferentiated (synthetic) state or phenotype. Although parts of the signaling cascade regulating the phenotypic switch have been described, little is known on the role of miRNAs involved. To systematically address this issue, we have established a microscopy-based quantitative assay and identified 23 miRNAs that induced contractile phenotypes when over-expressed. These were then correlated to miRNAs identified from RNA-sequencing when comparing cells in the contractile and synthetic states. Using both approaches, six miRNAs (miR-132-3p, miR-138-5p, miR-141-3p, miR-145-5p, miR-150-5p, and miR-22-3p) were filtered as candidates that induce the phenotypic switch from synthetic to contractile. To identify potentially common regulatory mechanisms of these six miRNAs, their predicted targets were compared with five miRNAs sharing ZBTB20, ZNF704, and EIF4EBP2 as common potential targets and four miRNAs sharing 16 common potential targets. The interaction network consisting of these 19 targets and additional 18 hub targets were created to facilitate validation of miRNA-mRNA interactions by suggesting the most plausible pairs. Furthermore, the information on drug candidates was integrated into the network to predict novel combinatorial therapies that encompass the complexity of miRNAs-mediated regulation. This is the first study that combines phenotypic screening approach with RNA sequencing and bioinformatics to systematically identify miRNAs-mediated pathways and to identify potential drug candidates to positively influence the phenotypic switch of VSMCs.

formation, which is the prerequisite for atherosclerosis 5 . It is established by now, that 61 inhibiting VSMCs phenotypic switching may be beneficial in advanced stages of this 62 disease. 63 The cells in the quiescent/contractile phenotype show low levels of migration and 64 proliferation. Morphologically, the contractile VSMCs display a fusiform or spindle-like 65 shape, abundant myofilaments and a heterochromatic nucleus 6 . In contrast, the synthetic 66 VSMCs adopt a rhomboid shape without specific filamentous cytoplasm, but with the 67 extensive rough endoplasmic reticulum, Golgi complex, and a euchromatic nucleus 6,7 . The        As cells in the contractile state were elongated (Fig. 1A), we subsequently undertook a concentrations. During the learning phase, we taught the KNIME software how to 238 distinguish between the contractile and synthetic phenotype of HAoVSMCs and generated 239 an optimized workflow (Supplemental Fig.1). The larger the E value (E > 1 in the range) 240 is, the slender the cell appears. CSI values indicated how likely the cell morphology 241 resembled a closed circle. The larger the CSI (range 0 < CSI < 1) is, the closer the cell is 242 to a circle. Accordingly, E and CSI values were E > 3 and 0 < CSI < 0.4, for the contractile 243 phenotype, respectively; and 1 < E < 3 and 0.6 < CSI < 1, respectively for the synthetic 244 phenotype. In order to confirm that the KNIME could precisely recognize the different 245 phenotypes of HAoVSMCs, validation was performed by defining the cells manually 246 (~1200 cells). By this, we could test and adapt this software to a high number of cells and 247 could evaluate and predict the precision and feasibility of the entire workflow.

248
In order to verify that the cell morphology can be used as an effective high-throughput 249 screening method to detect phenotypic switch, four miRNAs (miR-22-3p, miR-145-5p, 250 miR-214-3p, and miR-663a, (miR-22, miR-145, miR-214, and miR-663a in the following)) 251 that are known to effectively induce a phenotypic switch were used 13-17 . Images of the 252 transfected cells were taken at intervals of 24h, 48h, and 72h and analyzed in KNIME.

253
After segmentation, the cells were grouped to contractile, synthetic and undecided 254 phenotypic groups (0.4 < CSI < 0.6). Finally, the ratio of contractile / synthetic (Ratio of 255 con / syn) was calculated (Ratio = number of contractile / number of synthetic). Majority 256 of our test miRNAs significantly promoted the conversion of cells into contractile 257 phenotype as compared with the control group (Fig. 2). The ratios of con / syn in 258 HAoVSMCs were increasing after transfection of miR-22, miR-145, miR-214, and miR-259 663a. To summarize, we established an accurate, reliable, fast, and easy to apply 260 screening method based on cell morphology, which could be upscaled for screening 261 multiple miRNAs. Until now, the morphological parameters, such as CSI, have not been 262 used in any high-throughput screening approach.  The oligoes for selected miRNAs over-expression were prepared (Dharmacon, CO, USA) 279 (Supplemental Table 1). 384-well plates were pre-coated with these oligoes 29,40-42 (three 280 wells per oligo), HAoVSMCs were seeded and incubated for 72h. Then, the plates were 281 13 fixed with 3% PFA, stained with Hoechst 33342 and DiIC12(3). Images were analyzed 282 using the KNIME software and E, CSI, and ratio of con / syn of cells in the treatment group 283 and control group were calculated (Supplemental Fig 2). The heatmap (Supplemental  Table 2). The overall information of these groups was shown in the heatmap of each 312 replicate, indicating closeness and the difference between the groups (Fig. 3A). When 313 averaged over all four replicates, 153 miRNAs were upregulated, and 143 miRNAs were 314 downregulated in the contractile cell group compared to the control group, as shown in 315 the volcano plot (Fig. 3B).

316
After comparing with the results of the microscopy-based screen and sequencing of 317 miRNAs, we found six overlapping miRNAs (Fig. 3C). All six molecules were hits in the 318 microscopy-based screen when overexpressed and shifted cells to contractile phenotype. 319 In agreement to that, they also were found at elevated levels in cells undergoing a switch 320 to contractile phenotype. Among the over-expressed miRNAs we could identify miR-22 321 and miR-145 that served as the robust positive controls for the microcopy-based analysis 322 (Fig. 2). Besides, we identified four other molecules that induced the contractile phenotype 323 when over-expressed (miR-132-3p, miR-138-5p, miR-141-3p, and miR-150-5p). The 324 quality of the data is underlined by the observation that all four miRNAs were quantified 325 as strong hits in microscopy-based switch assays (Supplemental Fig. 2).  145 were considered as overlapping hit for this analysis (Fig. 3C). All six miRNAs belong 340 to different miRNA seed families and we used miRWalk 34 to extract the putative target 341 genes for every molecule. In total, we obtained 3,497 putative target genes via miRWalk 342 database for our 6 miRNAs (Supplementary Table 3). Of which, miRNA-145 (n=738) and  Table 4), 374 with only little fraction (15 individual miRNA-mRNA interactions) with strong validation 375 proofs available so far. 376 We also analysed whether the shared targets among four and five miRNAs (19 transcripts 377 collectively) build a network; for this, we used the STRING program 58,59 . STRING 378 database aims to collect, score and integrate all publicly available sources of protein-379 protein interaction information, and to complement these with computational predictions.  Drug targeting of miRNA-mediated regulatory networks 407 As the switch phenotype from the contractile to synthetic underlines numerous vascular 408 pathologies, the pharmacological targeting of the key regulators could be a promising 409 option to increase the therapeutic success. Consequently, we utilized the Drug-Gene 410 Interaction Database (DGIdb) (http://www.dgidb.org/) to predict the candidate drugs for 411 the targets of six miRNAs identified in this study (Fig. 4). Initially, we concentrated to find 412 the drugs that influence the activity of the hub targets; and indeed we found that 14 out of 413 17 hub targets can be linked to approved or experimental drugs. Not surprisingly, one and 414 the same target can be influenced by several drugs. For instance, dasatinib and 415 gemcitabine could be potentially used as a cocktail to efficiently inhibit the activity of EGFR 416 (Fig. 5). 417 On the other hand, some of these drugs may be effective on several hub proteins 418 simultaneously. For example, Erlotinib could be effective on EGFR 65 and CBL 66 , both 419 regulated by miR-141, or resveratrol on TP53 67 and SIRT1 68 , both targeted by miR-22 420 (Fig. 5). This type of drug-target interactions resemble pleiotropic activity of miRNAs and 421 may be of particular interest if a given miRNA has a limited regulatory activity upon other 422 processes, but switch phenotype. Ideally, a cocktail of drugs could be designed to inhibit 423 key targets of the selected miRNA. Potentially, a lower dose of the individual drug can be 424 used in such a mixture in order to achieve the desired effect, thereby, reducing unspecific 425 and side effects. Knowledge on miRNAs-mRNA interactions within a particular context 426 may serve as recipes for disease tailored or even personalized drug multiplexing.           Cell shape segmentation by KNIME software and the ratio of con / syn after miRNAs transfection. A. Automated detection of contractile and synthetic phenotypes. Contractile phenotype is indicated by red arrows and the synthetic phenotype is indicated by yellow arrows. B. Quanti cation of the phenotypic switch of HAoVSMCs after miRNA transfection. Compared with the control group, the transfection groups (miR-22, miR-145, miR-214, and miR-663a) appear to have signi cantly higher ratios of con / syn at 48h and 72h. * p < 0.05, ** p < 0.01.

Figure 3
Differential expression of miRNAs in the contractile phenotype comparing with the control group. A.
Heatmap of each replicate, indicating closeness between these groups and the difference between them. Red color indicates high expression of miRNAs, and green color indicates low expression of miRNAs. N: normal serum, L: low serum. B. Volcano plot shows that the individual up-regulated and down-regulated miRNAs after averaging replicates of the group with the contractile phenotype and the control group. Red dots indicate the upregulated miRNAs, and green dots represent downregulated miRNAs. The thresholds are: upregulated miRNAs (Log2FC > 0.6, FC > 1.5, p < 0.05), downregulated miRNAs (Log2FC < -0.6, FC < 2/3, p < 0.05). C. Overlap between the hit miRNAs derived from microscopy-based screening and sequencing.

Figure 4
Predicted drugs affecting hub targets and the common targets among four and ve miRNAs. Subset of the combinatorial interactions among miRNAs-targets-drugs.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. Supplements20210508MKeese.pdf