MiRNome profiling identifies up-regulation of miR-10b in RDEB-cSCC
In order to identify differences in the miRNA expression profile between cultured primary keratinocytes (KC) and cSCC cells (Supplementary Information and Supplementary Table S1 in Additional File 1), we conducted an Affymetrix GeneChip™ miRNA 4.1 expression analysis. Out of 2578 analyzed mature miRNAs unique to human, 50 miRNAs were found to be significantly (p-value ≤0.05 and false discovery rate (FDR) ≤ 0.2) ≥ 2-fold up- or down-regulated in RDEB-cSCCs (n = 4) compared to RDEB-KC (n = 6), and 56 in HC-cSCCs (n = 3) compared to healthy control (HC)-KCs (n = 5, Supplementary Fig. S1A, Supplementary Tables S4 and S5 in Additional File 1). When analyzing miRTarbase v6.1 predicted targets of the deregulated (≥ 2-fold, p-value ≤0.05 and FDR ≤ 0.2) miRNome, various cancer-related Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, e.g. signaling pathways regulating pluripotency of stem cells, appeared significantly enriched (Supplementary Fig. S1B, Additional File 1).
To assess the ability of the miRNA expression profile to distinguish cSCCs from other experimental groups we performed principal component analysis (PCA). Annotation of the samples within the distinct clusters showed a clear separation of experimental groups (Fig. 1a). Even though the miR-10 family was found to be overall upregulated in cSCCs, in-depth analysis of major contributors driving the unsupervised cluster separation highlighted miR-10b, which was 2.2-fold (p < 0.05) upregulated in RDEB-cSCCs and 2.2-fold (p < 0.05) in HC-cSCCs, respectively, compared to their non-malignant controls (Fig. 1b, c). Despite the fact that miR-10a appeared to be the most deregulated miRNA in RDEB-cSCCs in microarray analysis, but not in subsequent qPCR, we assume that high microarray scores most likely derived from a certain hybridization error rate, as miRs-10a and -10b differ in only one nucleotide (Fig. 1e, Supplementary Fig. S1C-E, Additional File 1). In addition, consistently increased miR-10b (on average 9-fold), but not miR-10a levels in RDEB-cSCC were found in previously generated RNA sequencing (RNA-seq) data, where immortalized HC-KC lines were used as controls (Supplementary Fig. S1F,G, Additional File 1). Thus, we focused on miR-10b and dropped miR-10a from further experiments. Predicted miR-10b targets were further tested for gene set enrichment in cancer hallmarks (Molecular Signature Database v6.2), and showed significant association with metastatic processes like epithelial to mesenchymal transition (EMT) (Fig. 1d).
In summary, the results of a microarray based miRNA expression profiling revealed a deregulated miRNome able to distinguish cSCC from keratinocytes. In particular, miR-10b was significantly upregulated in cultured cSCC.
Expression of miR-10b in cSCC tissues
In order to localize and further substantiate upregulation of miR-10b expression in tissue, we optimized a combined IHC-ISH protocol using LNA probes.
When tissue specific expression was examined in FFPE-sections of archival skin and tumor biopsies, miR-10b was found to be upregulated in cSCC, particularly in RDEB-cSCCs. Expression was predominantly co-localized to keratin-positive cells (Fig. 1f, Supplementary Fig. S2A-C, Additional File 1), tumor vasculature and lymphocytes (Supplementary Fig. S2D in Additional File 1). Notably, both vasculature and hemopoietic cells have previously been shown to express miR-10b in a tumor context [15, 18, 27,28,29]. Further analysis of RDEB-tissues showed a strong expression of transcription factor TWIST1, a known upstream driver of miR-10b-mediated tumor malignancy. In addition, TWIST1 was also found to be upregulated in cultured SCC cells in IF and Western blot analysis (Supplementary Fig. S2E,F, Additional File 1) [14, 30]. Single cells in an RDEB-cSCC lymphnode metastasis (RDEB-SCC LM) expressed high levels of miR-10b. Less pronounced miR-10b abundance was observed in HC-cSCC (Fig. 1f).
To investigate whether miR-10b expression was correlated to differentiation, cells were incubated in the presence of calcium and serum to induce differentiation, and respective marker gene expression was analyzed in parallel to miR-10b expression levels. However, no correlation between differentiation and miR-10b expression was observed (data not shown).
Taken together, results of the IHC-ISH show high abundance of miR-10b in cSCC tumor biopsies.
MiR-10b confers anchorage-independent aggregation capabilities to keratinocytes and attenuates mobility
Metastasis requires the dissemination and successful establishment of clones with tumor-initiating potential at distant niches [31]. In order to investigate the biological role of miR-10b, we analyzed its expression levels in experimental cell lines using IHC-ISH, in parallel to 3D tumor spheroid formation assays. As the cytoplasm is considered as the site of miRNA maturation and action, fluorescence intensity, excluding the nuclear region, was assessed at a single cell resolution. MiR-10b was confirmed to be highly expressed in RDEB-cSCC cell lines using ISH probes specific for mature miR-10b, and in two out of three HC-cSCC cultures, as compared to primary HC-KCs (Fig. 2a, b, Supplementary Fig. S3,4 in Additional File 1).
Next, we performed aggregate formation assays to explore the capacity of experimental cell lines to form anchorage independent spheroids [32]. All three RDEB-cSCCs, and two out of three HC-cSCCs formed stable aggregates (Fig. 2c, d). RDEB-KCs, which were used as surrogate for functional experiments in all downstream experiments, stably overexpressing miR-10b (RDEB-KCmiR-10b), as well as HC-KCmiR-10b, phenocopied RDEB-cSCCs in the formation of stable spheroids (Fig. 2e). Expression and maturation of miR-10b, driven by the constitutive Pol-III U6 promoter, was confirmed by qPCR (Supplementary Fig. S5A, Additional File 1). In addition, we found that KCmiR-10b were significantly smaller than their parental cells (p-valueRDEB-KC = 0.002; p-valueHC-KC < 0.001; Supplementary Fig. S6A in Additional File 1). Like in RDEB-cSCC derived aggregates, viable cells were found predominantly in the outer spheroid layers of KCmiR-10b spheres (Supplementary Fig. S5I in Additional File 1). Parental RDEB- and HC-KCs formed fewer and rather loose aggregates, which did not withstand mild pipetting (Fig. 2e).
To further substantiate the link between miR-10b overexpression and enhanced spheroid formation capacities, we knocked-out the MIR10B gene locus in RDEB-cSCC cells (RDEB-SCC1MIR10B−/−) using the CRISPR/Cas9 technology, and performed minimal dilution to potentially obtain single clones, which was confirmed by Sanger sequencing and qPCR on mature miR-10b expression levels (Supplementary Fig. S5J,K). In spheroid formation assays we observed, that MIR10B knock-out reduced the stability of aggregates and resulted in an increased number of single cells and fragmented aggregates (Fig. 3a-c). While PCR-mediated confirmation of MIR10B knock-out showed only bands corresponding to successful deletion, we found that over time single cells that had escaped knock-out and subsequent clearance by minimal dilution returned to dominance, independent of a potential proliferative advantage (Fig. 3d, e). This was observed in several clones and over several cultivation passages, pointing towards a potential survival advantage of cells expressing miR-10b. When subjecting these mixed cultures again to 3D-sphere formation assays, their behavior resembled that of parental cells (Fig. 3a-c). Another striking difference between parental and MIR10B−/− cells was a reduced capacity to grow out of tumor spheroids upon transfer to culture dishes. Spheroids adhered to dishes, and circularly outgrowing cells became visible after 24 h in RDEB-SCC1 derived aggregates, and to a much lower extent in RDEB-SCC1MIR10B−/− cells. Again, this was reversed in mixed culture experiments. A similar outgrowth pattern to RDEB-SCC1 was also observed in two out of three HC-cSCC derived spheroid experiments (Fig. 3f).
As spheroid formation is a key attribute of cancer stem cells (CSCs), and points towards the presence of cancer stem cell-like properties, we analyzed accepted CSC markers (CD44 / CD24) [33]. RDEB-cSCC cultures showed an overall high expression of CD44, and varying levels of CD24, including a CD44high / CD24−/low cell population. In support of the hypothesis that miR-10b expression was associated with a CSC-like phenotype, overexpression of miR-10b in KCs resulted in an overall ~ 2-fold enrichment of CD44high / CD24−/low cells (Supplementary Fig. S6B,C in Additional File 1).
We next examined whether the above observed properties were attributed to mobility-, or adhesion-associated mechanisms [32]. Therefore, we conducted wound closure assays using RDEB-KCmiR-10b and RDEB-SCC1MIR10B−/− cells. We found that those cells expressing high levels of miR-10b (i.e. RDEB-KCmiR-10b versus parental, parental versus RDEB-SCC1MIR10B−/−) demonstrated significantly impaired mobilization (time point 6 h: p < 0.01) (Fig. 4a, b). In addition, when transiently transfecting SCC1MIR10B−/− with a miR-10b mimic, as well as when using the mixed population, gap closure was accelerated again (Fig. 4b, c). Notably, HC-cSCCs showed faster gap closure than RDEB-cSCC and primary HC-KCs, the latter two showing a similar migratory potential (Fig. 4d, e). When overexpressing miR-10b in HC-KCs, no impact on migration was observed (data not shown). The reason for the unexpected delay in gap closure in RDEB-cSCCs remains speculative, but might be due to prior exposure of RDEB-cells to pathological processes like chronic inflammation, or the difference in matrix composition, which is dramatically altered in RDEB patients due to C7 absence [5,6,7,8,9]. Still, our results show that RDEB-SCCs migrate more slowly that HC-cSCCs in general, and miR-10b has an impact on mobility. No impact on proliferation was observed in the presence or absence of miR-10b.
Taken together, our data suggest for the first time that CSC-like properties can be conferred by miR-10b, a heretofore unknown aspect of miR-10b driven malignancy in cSCCs, and indicates an impact on motility in RDEB.
DIAPH2 is deregulated in RDEB-cSCC and predicted to be a target of miR-10b
To analyze the impact of miR-10b on its targetome, we first screened the scientific literature in an automated text-mining approach for miR-10b in cancer, which highlighted the previously reported, direct miR-10b target transcription factor HOXD10 [14], (Supplementary Fig. S7A in Additional File 1). We observed significantly reduced levels of HOXD10 protein in three RDEB-cSCC cell lines compared to HC-KC, and also in an RDEB-cSCC tissue section, compared to HC- and RDEB-skin [34], (Supplementary Fig. S7B-D in Additional File 1).
To next identify novel downstream targets of miR-10b, data driven miR-10b target identification was implemented based on transcriptome data generated from miRNA microarray-matched RDEB-cSCC and RDEB-KC samples. Of the 576 differentially expressed genes identified (≥ 2-fold ↓↑) in RDEB-cSCCs, 114 were reported in merged repository data (n = 3923) of validated miR-10 targets (miRTarbase v6.1), as well as computationally predicted targets by seed sequence and evolutionary conservation (TargetScan v7.2). Dysregulated putative miR-10 interaction partners were further prioritized by strength of their inverse correlation of expression with miR-10 signal (Supplementary Fig. S7E in Additional File 1). To nominate a disease relevant miR-10b target, we analyzed publicly available survival data from metastatic stage IV head and neck squamous cell carcinoma (HNSCC) (n = 86 patients, The Cancer Genome Atlas / TCGA), as this cancer type was previously described to have high genetic similarities to RDEB-cSCC [8]. The top 20 candidate miR-10b targets with highest inverse correlation were then used to stratify HNSCC patients (Supplementary Fig. S8A and Supplementary Table S6 in Additional File 1). We nominated diaphanous related formin 2 (DIAPH2) for further evaluation based on significant differences in Kaplan Meier survival curves (p < 0.05, log-rank test, Fig. 5a), together with the fact that it was listed as a putative target of miR-10b. Its potential disease-relevance was substantiated by its recent association with colon carcinoma, and its potential role in actin-organization and microtubule stabilization [35,36,37]. When re-analyzing normalized RNA-seq data generated by Cho et al., retrieved from GEO-repository (GSE111582), both, DIAPH2 and HOXD10, showed lower expression in RDEB-cSCC versus RDEB-skin tissue [8], (Supplementary Fig. S8B,C in Additional File 1).
To confirm DIAPH2 as a direct target of miR-10b, a dual luciferase reporter assay was established by cloning the 3’UTRs of DIAPH2 and HOXD10 as a control, respectively, downstream of a firefly luciferase reporter gene. Constructs were then co-transfected with a miR-10b mimic into HC-KCs, which express only low levels of endogenous miR-10b. In the presence of miR-10b mimic, luciferase signal was significantly reduced (p-valueDIAPH2 = 0.015; p-valueHOXD10 < 0.01) compared to scrambled (SCR) control (Supplementary Fig. S7F,G in Additional File 1).
We next assessed DIAPH2 expression in cultured cSCCs and control KC. Both, at the mRNA and the protein level, DIAPH2 was downregulated in RDEB-cSCC cells, with no significant difference between HC- and RDEB-KC (Fig. 5b, d, Supplementary Fig. S7I in Additional File 1). In addition, DIAPH2 was also slightly downregulated in RDEB-KCmiR-10b, as shown by sqRT-PCR (Supplementary Fig. S5B in Additional File 1). In HC-cSCCs, DIAPH2 expression levels were not as consistently reduced as in RDEB-cSCC and differed significantly between cell lines (Fig. 5c, d). Notably, following transient transfection of HC-KCs with miR-10b mimic, DIAPH2 protein levels were found to be significantly reduced (~ 29%, p = 0.05) compared to SCR control (Supplementary Fig. S7H in Additional File 1), and it recurred in RDEB-SCC1MIR10B−/− as investigated by Western blot analysis and immunofluorescence microscopy (Fig. 5e). DIAPH2 downregulation was also evident in RDEB-cSCC tissue sections (Fig. 5f). Taken together, DIAPH2 appeared as target of miR-10b in RDEB-cSCCs, although inconsistent results in HC-cSCCs indicate further, miR-10b independent regulatory mechnisms.
To examine if a loss of DIAPH2 in keratinocytes phenocopies the observed migratory behavior in RDEB-KCmiR-10, we knocked out DIAPH2 in immortalized HC-KCs using the CRISPR/Cas9 technology (HC-KCDIAPH2−/−). Clones were generated by minimal dilution and knock-out was confirmed by Sanger sequencing and Western blot analysis (Supplementary Fig. S5C,D in Additional File 1). While proliferation was not altered, DIAPH2−/− cells showed significantly impaired motility over parental HC-KCs (Supplementary Fig. S5E,F). In addition, when analyzing HC-KCDIAPH2−/− in a 3D-sphere formation assay, a trend towards enhanced aggregation was observed, with a distribution of spheroid size similar to cSCCs (Supplementary Fig. S5G,H). Overall, the functional impact of DIAPH2 knock-out resembled our observations in RDEB-KCmiR-10 and RDEB-SCC1MIR10B−/−.
Our work taken as a whole has demonstrated that miR-10b is overexpressed in cSCC cells and tissues, and that miR-10b confers tumor-associated properties to KCs. In addition, deregulation of DIAPH2, a novel putative miR-10b downstream target, might represent a contributing factor in cSCC malignancy.