SAP domain-dependent Mkl1 signaling stimulates proliferation and cell migration by induction of a distinct gene set indicative of poor prognosis in breast cancer patients

We devised a screening method to identify genes co-regulated with tenascin-C by Mkl1 in a SAP domain-dependent manner without involvement of SRF. For this purpose, we used HC11 mammary epithelial cells that react strongly to the overexpression of Mkl1 with induction of tenascin-C expression [13]. We compared three HC11 strains that either overexpress the C-terminal red fluorescent protein (RFP)-tagged full length Mkl1 (HC11-FL), Mkl1-RFP with a mutated SRF-interaction site (HC11-mutB1) or Mkl1-RFP with a deletion of the SAP domain (HC11-ΔSAP). None of the three Mkl1 variants appear to be toxic to the cells, as we have not observed any changes in viability or cell morphology. HC11-FL cells were shown to overexpress Mkl1 7.1-fold above the endogenous Mkl1 present in parental HC11 cells [13], and were used as control cells in our study. All cell strains were FACS sorted to express similar levels of Mkl1-RFP proteins. These cells were used for transcript profiling and gene lists of interest were established as shown in Figure 1A, B. A scatter plot (Figure 1A) of all transcripts expressed in HC11-mutB1 versus HC11-FL control cells (y-axis) and all transcripts expressed in HC11-ΔSAP versus HC11-FL control cells (x-axis) shows that a large majority of transcripts does not differ significantly between the three cell strains (log fold change (FC) ≈ 0; black dots). Setting the threshold to a 2-fold reduction (logFC = -1; grey lines), three gene sets can be distinguished: 1) blue dots represent genes that are lower in HC11-mutB1 than in HC11-FL control cells, but are unaffected in HC11-ΔSAP cells, thus representing typical SRF/Mkl1 target genes; 2) green dots represent genes that are lower in HC11-ΔSAP than in HC11-FL control cells, but are unaffected in HC11-mutB1 cells (this gene set includes tenascin-C); and 3) red dots indicate genes with reduced expression in both HC11-mutB1 and HC11-ΔSAP cells compared to HC11-FL control cells. Thus, this approach enabled us to form three gene sets that were distinct from the large majority of transcripts and were dependent for expression on the B1 site of Mkl1, the SAP domain, or both. The three groups presented by a Venn diagram (Figure 1B) contain 141 probesets for transcripts that depended on the function of the B1 site but not the SAP domain for their induction, 113 probesets for transcripts that depended on both of these Mkl1 domains and a third group of 205 probesets for transcripts co-regulated with tenascin-C that did not require an interaction of Mkl1 with SRF but depended on the SAP domain for induction (complete probeset lists and annotations are found in Additional file 1: Table S1, Additional file 2: Table S2 and Additional file 3: Table S3). This analysis revealed that the SAP-dependent mechanism of tenascin-C regulation by Mkl1 is shared by a large cohort of genes. Below the Venn diagram, we indicated which cells were deficient in the respective transcripts. Thus, the typical SRF/Mkl1 target genes are reduced in HC11-mutB1 cells, while the SRF-independent/SAP-dependent genes are reduced in HC11-ΔSAP cells. The intermediate group that requires both Mkl1 activities is reduced in both the HC11-mutB1 and HC11-ΔSAP cells.

Functional analysis of the three gene lists using the IPA software revealed different molecular and cellular functions (Figure 1C) and different disease associations (Figure 1D) for the three types of gene signatures. Thus, the SRF-dependent/SAP-independent signature implicated a function of these genes in cellular movement and the linked diseases included connective tissue disorders, inflammatory disease and skeletal and muscle disorders, which are the main features known to be regulated by SRF/Mkl1 interaction [23‐25]. The SRF-dependent/SAP-dependent group of genes includes as major functions post-translational modification, protein degradation and protein synthesis, and the top disease association is cancer. Finally, the genes of the SRF-independent/SAP-dependent group were associated with extremely high significance with cell cycle and cancer (-logP ≥ 25 and ≥ 30, respectively), while the SRF/Mkl1 target genes were associated with the same two categories at low significance only (-logP ≥ 2 and ≥ 7, respectively). These data imply that SAP-dependent induction of transcription by Mkl1 may counteract the known differentiation-promoting effect of SRF/Mkl1-induced transcription. A list of SAP-dependent genes with published cancer-related functions, whose transcripts were downregulated more than 3-fold in HC11-ΔSAP compared to HC11-FL control cells, is presented in Table 1. To confirm that these transcripts are indeed differentially expressed in the different HC11 cell strains, qRT-PCR analysis was performed using cDNA from three different batches of the respective HC11 strains. Differences in gene expression between HC11-ΔSAP and control cells are presented in Table 1 and in more detail in Additional file 4: Figure S1. The qRT-PCR results agreed with the data obtained by transcript profiling. We also tested the SAP-dependent gene expression in the HC11 strains when grown in the presence of serum. It is interesting to note that in the presence of 3% FCS, these transcripts remained strongly reduced in HC11-ΔSAP compared to control cells (Table 1). Thus, the induction of these genes seems to depend mainly on whether the SAP domain is present in the transfected Mkl1 construct.

In addition, we monitored changes in the expression of some of the SRF-independent/SAP-dependent Mkl1 targets on a protein level. In agreement with the changes seen at the transcript level, we confirmed the reduction of tenascin-C, Wisp1 and Nox4 proteins in cells overexpressing the ΔSAP-Mkl1 construct compared to the HC11-FL control and HC11-mutB1 cells (Additional file 4: Figure S2). Using zymography, we found that Mmp2, a gene that was not affected by Mkl1 overexpression at the transcript level was highly expressed in all three cell strains, whereas Mmp3 and/or 12, which belonged to the SRF-dependent/SAP-dependent gene set, were almost completely lacking in HC11-mutB1 as well as HC11-ΔSAP cells, corresponding to the data obtained by transcript profiling.

Since we have previously shown that the SAP domain of Mkl1 interacts with the proximal promoter of tenascin-C to induce its transcription [13], we tested whether this was also the case for other transcripts of the same group. The promoters of the SRF-independent/SAP-dependent genes listed in Table 1 encompassing at least 500 bp upstream of the transcription start site (TSS) were fused to the secreted alkaline phosphatase (SEAP) reporter gene of pSEAP2-Basic. We tested the induction of each promoter-reporter construct by co-transfection with FL-Mkl1 (Figure 2A). This revealed that the majority of the new promoters tested (8 out of 12) were induced at least 2-fold by Mkl1 in comparison to co-transfection with an inactive Mkl1 devoid of the transactivation domain, indicating that these are indeed direct Mkl1 target genes. The promoter constructs that did not respond to Mkl1 overexpression may represent genes that are indirectly regulated by Mkl1, or the relevant promoter regions were not contained in the constructs tested. Next, we investigated whether the induction was SAP-dependent and SRF-independent by comparing the reporter activity after co-transfection with mutB1- versus ΔSAP-Mkl1 variants (Figure 2B). Indeed, the promoter-reporter constructs induced by FL-Mkl1 were also strongly induced by mutB1-Mkl1, but not by ΔSAP-Mkl1. In contrast, the promoter construct for Acta2, a gene from the SRF-dependent/SAP-independent gene set was strongly induced by ΔSAP-Mkl1 but not by mutB1-Mkl1, as expected for a typical SRF/Mkl1 target gene [18, 55, 56]. All promoters that revealed SAP-dependency were shortened to 200 bp upstream of the TSS to test whether this was sufficient to relay the Mkl1 response, as it has been seen previously for tenascin-C [13]. With the exception of Krt5 and Nox4, for which some activity was lost by shortening the promoters, the 200 bp proximal promoters of all other genes tested were induced equally well as the longer constructs (Figure 2B). Thus, we conclude that there are many genes that are regulated similarly as tenascin-C requiring the SAP domain of Mkl1 to induce transcription from their proximal promoter.

Next, we tested whether the differential gene expression seen in the different HC11 strains overexpressing either FL-, mutB1- or ΔSAP-Mkl1 constructs have functional consequences on their behavior. Since most of the SAP-dependent transcripts are proposed to have a function in cancer, we decided to analyze two main functions important for cancer progression: proliferation and cell migration. An approximately equal overexpression of the different Mkl1 protein variants in the HC11 cell lines was confirmed by Western blot analysis (Figure 3A). An HC11 cell strain stably transfected with an empty vector [13] expressing only endogenous Mkl1 (below the detection limit in Figure 3A) was also included in these studies. The proliferation rates of the HC11 strains were analyzed using a 5-bromo-2′-deoxyuridine (BrdU) incorporation assay. The incorporated BrdU was measured immediately after plating (0 h) as well as at 24, 48, 72 and 96 h. Compared to empty vector-, FL- or mutB1-transfected HC11 strains, there was a significant decrease in BrdU uptake into newly synthesized DNA in HC11-ΔSAP cells over the entire time period tested (Figure 3B). To investigate cell motility, we used a transfilter migration assay. Similarly to the effect on cellular proliferation, the expression of ΔSAP-Mkl1 significantly inhibited HC11 cell migration by 2.7-fold compared to endogenous or full length Mkl1 expression, and more than 3.5-fold compared to mutB1-Mkl1 expression (Figure 3C).

Thus, overexpression of FL-Mkl1 protein in HC11 cells did not affect their behavior. However, overexpression of ΔSAP-Mkl1 led to a significant reduction in the proliferative and migratory ability of HC11 epithelial cells, either through a dominant-negative effect of ΔSAP-Mkl1 on SRF-mediated action and/or a positive impact of the SAP-dependent Mkl1 target genes on these functions important for cancer progression.

We have previously found that the SAP-dependent induction of tenascin-C was triggered by applying mechanical strain to fibroblasts. Mammary epithelial cells are also exposed to mechanical strains, both during normal development, pregnancy and lactation, as well as under pathological conditions such as in cancer. Therefore, we tested whether tenascin-C and other members of the SAP-dependent Mkl1-induced gene set are mechanoresponsive in HC11 cells. We tested two paradigms: 1) static strain that was shown to induce c-fos, a very prominent mechanoresponsive gene in HC11 cells [57] that we used as a control, and 2) cyclic strain. While we were able to confirm induction of c-fos by applying static strain at 20% for 1 h, there was no induction of tenascin-C under these conditions compared to cells at rest (Figure 4A). However, using 15% cyclic strain at a frequency of 0.3 Hz for 1 h, we found that not only the control gene c-fos but 11 out of 16 SAP-dependent genes, including tenascin-C were significantly upregulated above the expression levels obtained in resting cells (Figure 4). Even though significant, the induction of tenascin-C was minimal (Figure 4A) compared to 18-fold upregulation for Adamts16 or 10-fold upregulation for Lox (Figure 4B), both of which are enzymes involved in extracellular matrix (ECM) remodeling and cancer progression [31, 58]. Being mechanoresponsive, the SAP-dependent Mkl1 target genes might be activated in stiff tumor tissue, which further confirms their relation with cancer.

In order to investigate whether the SRF-independent/SAP-dependent genes were prognostic of accelerated cancer progression in human patients, we used the bioinformatics tool Gene expression-based Outcome for Breast cancer Online (GOBO) that allowed us to investigate a breast tumor data set containing 1881 samples analyzed by Affymetrix Human Genome U133A arrays. GOBO is designed to assess gene expression levels and association with outcome of single genes or gene sets in multiple subgroups of this breast cancer data set [59]. Here, we analyzed two sets of genes, namely the SRF/Mkl1-induced gene set (SRF-dependent/SAP-independent) and the SAP-dependent gene set (SRF-independent/SAP-dependent) containing tenascin-C. The analysis was performed across tumor samples stratified according to PAM50 subtypes [60], estrogen receptor (ER)-status and histological grade. In contrast to the SRF/Mkl1 target genes that were predominantly associated with tumors classified as normal-like and with lower histological grades (1 and 2) (Figure 5A), elevated expression of SAP-dependent genes was associated with extremely high significance (P < 0.00001) with typical high-proliferative poor outcome classes in breast cancer, such as basal-like, HER2-enriched, luminal B, ER-negative and histological grade 3 tumors (Figure 5B). Next, a functional correlation analysis to find a possible interconnection between the SAP-dependent Mkl1 target genes was performed using the GOBO tool (Additional file 4: Figure S3). This analysis explores the correlation of expression of individual genes in our gene sets with eight different co-expressed gene modules emulating breast cancer-specific as well as general tumor biological processes [61]. Interestingly, whereas the gene set of SRF/Mkl1 targets did not show a significant correlation with any of these modules, the genes in the SAP-dependent gene set were correlated with a very high significance (P < 0.00001) with two proliferation modules – mitotic checkpoint and mitotic progression. Both modules contain genes related to central mitotic processes involved in either the regulation of the M-phase and the mitotic checkpoint or in carrying out the M-phase. Finally, the association of our gene sets with outcome using distant metastasis free survival (DMFS) as an endpoint and 10-year censoring was analyzed. The survival analysis was performed in all tumors for which DMFS follow-up is available (1379 cases), as well as in 21 groups that were stratified based on gene expression subtypes (PAM50 classifier), ER-status, lymph node (LN)-status, histological grade, and treatment status. Samples in the whole cancer data set (1881 patients) were stratified into three quantiles, low, intermediate and high, based on SRF-dependent/SAP-independent or SRF-independent/SAP-dependent gene expression. Interestingly, high expression of SRF/Mkl1-induced genes was associated with a better clinical outcome for all tumors, as well as for LN-negative and untreated tumors compared to low and intermediate expression of these genes (Figure 6A). In contrast, both high and intermediate expression of the SAP-dependent genes was associated with bad clinical outcome in all tumors, and particularly in LN-negative, systemically untreated, ER-positive, Grade 1 and 2 tumors (Figure 6B). Similar results were obtained for the typical breast cancer gene CCNB1 by Ringnér et al. [59]. The Kaplan-Meier survival analyses were supported by the corresponding multivariate analyses (Figure 7A, B). The hazard ratio for the variate Grade shows statistical significance, proving that the influence of high SAP-dependent gene expression on patient survival is independent of tumor grade. Among all tumors for which DMFS data are available, a hazard ratio of 0.44 (95% CI = 0.28-0.68; P = 0.0003) for the low SRF-independent/SAP-dependent tercile was detected compared to the high SRF-independent/SAP-dependent tercile (Figure 7B, all tumors). This indicates that patients with tumors expressing high levels of the SAP-dependent genes are more than twice as likely to develop metastatic disease. Similar hazard ratios, in the range of 0.28-0.44 for the low tercile compared to the high tercile were also detected among subgroups of untreated, LN-negative, ER-positive, Grade 1 and 2 tumors (Figure 7B). Thus, the association of high SRF-independent/SAP-dependent gene expression with reduced DMFS among patients not receiving adjuvant therapy, as well as among LN-negative, ER-positive, Grade 1 and 2 patients indicates that increased expression of the SAP-dependent Mkl1 target genes plays a significant role in the natural metastatic progression of non-aggressive towards highly aggressive breast cancer in human patients.

Full length Mkl1 (FL-Mkl1) and the two Mkl1 mutants, mutB1-Mkl1 comprising alanine substitutions of four amino acids (underlined) in the B1 domain of Mkl1 (KKAKELKPKVKKLKYHQYIPPDQKQD) [78] and ∆SAP-Mkl1 with a deletion of the SAP domain [15], were constructed based on transcript variant 1 (GenBank accession number NM_153049) as previously described [13]. All Mkl1 variants were expressed as C-terminal RFP-tagged fusions. An empty vector expressing RFP alone was previously described [13].

HC11 mammary epithelial cells, kindly provided by Dr. N. Hynes (Basel, Switzerland), were grown in RPMI-1640 medium supplemented with 10% FCS, 5 μg/ml insulin (Sigma, Buchs, Switzerland) and 10 ng/ml epidermal growth factor (EGF; Invitrogen, Zug, Switzerland). In most of the experiments, the HC11 cells were starved in 0.03% FCS/RPMI without EGF. To obtain HC11 cells stably expressing FL-Mkl1-RFP (HC11-FL), mutB1-Mkl1-RFP (HC11-mutB1), ∆SAP-Mkl1-RFP (HC11-∆SAP) or RFP alone (HC11-empty vector), cells were transfected using FuGENE® 6 (Roche, Basel, Switzerland) and selected with Geneticin (1 mg/ml; Roche) for 14 days before fluorescence-activated cell sorting (FACS) of RFP-positive cells on a Vantage SE (Becton Dickinson, Basel, Switzerland). Cell viability of the four HC11 cell strains was assessed by the CellTiter-Blue viability assay (Promega, Duebendorf, Switzerland).

Proliferation rates of the HC11 cell strains were determined using BrdU incorporation assay (Roche). After 24 h of starvation, cells were plated in triplicate on Black 96-well microtiter plates (PerkinElmer, Schwerzenbach, Switzerland) at 5 × 10³ cells/well in 3% FCS/RPMI and allowed to proliferate for 0, 24, 48, 72 and 96 h before labeling with BrdU for 2 h. BrdU incorporation into newly synthesized DNA was determined according to the manufacturer’s protocol using a Luminometer Mithras LB940 (Berthold Technologies, Regensdorf, Switzerland). Experimental values were normalized to the values of HC11-∆SAP cells at the time point 0. Data represent means ± SD from three independent experiments.

Cell migration was assayed using transwell polycarbonate membrane inserts (6.5 mm; Corning, Amsterdam, The Netherlands) with 8 μm pores as described [79]. After 24 h of starvation, 5 × 10⁴ cells were plated in the top insert chamber with 100 μl serum-free RPMI. The lower chamber was filled with 600 μl 10% FCS/RPMI. Cells were allowed to migrate across the filter for 22 h at 37°C before fixation and crystal violet-staining. Images of duplicate inserts were acquired on a Nikon Eclipse E600 using 10× magnification and a color CCD camera. Migration was quantified by measuring the area covered by migrated cells using the Fiji distribution of ImageJ [80]. Data represent means ± SD from three independent experiments.

2 × 10⁵ HC11 cells/well were seeded in BioFlex® 6-well culture plates (Flexcell International, Hillsborough, NC, USA) coated with either growth factor reduced-Matrigel (BD Biosciences, Basel, Switzerland) or fibronectin [11]. Cultures were starved for 24 h before applying either equibiaxial cyclic strain (15%, 0.3 Hz) or static strain (20%) at 37°C for 1 h using Flexcell FX-4000 (Flexcell International). Cells cultured under the same conditions and not exposed to strain were used as a resting control. After mechanical stimulation, cells were lysed and total RNA was isolated using the RNeasy Mini Kit (Qiagen, Basel, Switzerland).

HC11 cell strains stably expressing Mkl1 variants were starved for 48 h before total RNA was extracted, converted into labeled cDNA and hybridized to Affymetrix GeneChip Mouse Gene 1.0 ST arrays. RMA-normalized expression values were calculated with the Affy package from Bioconductor 2.4 [81], and differentially expressed genes were identified using moderated t-statistics calculated with the empirical Bayes method as implemented in the Bioconductor limma package [82]. To be considered as differentially expressed between HC11-FL and HC11-mutB1 or HC11-∆SAP cells, genes had to pass the filters: adjusted P-value ≤ 0.01 (with Benjamin-Hochberg false discovery correction), a minimum absolute linear fold change difference of 2.0 and a minimum average expression value of 4.0 (log2). Microarray data files are available from the Gene Expression Omnibus (GEO; http://​www.​ncbi.​nlm.​nih.​gov/​geo/​), accession number GSE44907. Using the above parameters, gene lists of the two contrasts (mutB1/FL and ∆SAP/FL) were compared resulting in the formation of three gene groups: SRF-dependent/SAP-independent, SRF-dependent/SAP-dependent and SRF-independent/SAP-dependent. The three gene sets were analyzed using the bioinformatics softwares: 1) IPA (Ingenuity® Systems; http://​www.​ingenuity.​com); and 2) GOBO (http://​co.​bmc.​lu.​se/​gobo) [59]. In order to use the latter tool, Affymetrix GeneChip Mouse Gene 1.0 ST IDs were mapped to Affymetrix Human Genome U133A IDs using Biomart for Ensembl build 66. The module “Gene Set Analysis Tumors” was used to investigate the expression pattern and to perform survival and functional correlation analyses for the SRF-dependent/SAP-independent and SRF-independent/SAP-dependent gene sets across 1881 breast cancers characterized by Affymetrix Human Genome U133A arrays.

Total RNA was isolated from HC11 cell strains after 24 h of incubation either in 0.03 or 3% FCS/RPMI. RNA was reverse transcribed and relative tenascin-C and c-fos mRNA levels were detected as described [12, 13]. Relative mRNA levels for the genes listed in Table 1, normalized to Gapdh, were measured using Platinum® SYBR® Green qPCR SuperMix-UDG with ROX (Invitrogen) and the primers listed in Additional file 4: Table S4. Real-time PCR was performed in a StepOnePlus Real-Time PCR System (Applied Biosystems, Rotkreuz, Switzerland) using a standard cycling profile. All samples were run in duplicate. Data were analyzed by the ∆Ct method [83] and presented as fold changes in mRNA expression levels between HC11-FL and HC11-∆SAP cells. RNA from stretched cells was analyzed by qRT-PCR using the efficiency ∆∆Ct method [84] that included a further normalization to the resting control. Data represent means ± SD from three independent experiments.

After 24 h of starvation, whole-cell extracts from the three HC11 strains were prepared in RIPA buffer and immunoblotting was performed as described [12, 13]. The following primary antibodies were used: mAb65F13 anti-Mkl1 [12], MTn12 anti-Tnc [85], anti-Wisp1/CCN4 (clone 214203, R&D Systems), anti-Nox4 (NB110-58851, Novus Biologicals), anti-Vcl (clone hVIN-1, Sigma) and anti-Gapdh (ab9485, Abcam).

After reaching 90% confluency, HC11 strains were starved for 48 h before conditioned medium was collected, concentrated and analyzed by zymography as described [86].

The tenascin-C promoter used in this study was described as TNC 247 bp [13]. Promoters of Acta2 [87] and all SRF-independent/SAP-dependent genes described in Table 1 were PCR-amplified from genomic DNA and corresponded to the sequences listed in Additional file 4: Table S5. Each promoter contained ≥ 500 bp 5′ of the TSS and was cloned into the pSEAP2-Basic (Clontech, Saint-Germain-en-Laye, France). For some promoters also 200 bp proximal promoter sequences were cloned as described above. All clones were verified by DNA sequencing.

HC11 cells in 6-well plates were cotransfected with 1 μg of the SEAP reporter vectors, 1 μg of pcDNA3 vectors encoding Mkl1 variants [13], and 200 ng of the secreted luciferase MetLuc vector (Clontech) used to normalize for transfection efficiency. Cells were cultured in 0.03% FCS/RPMI for 24 h before enzymatic activity measurements were performed as described [13]. Experimental values represent averages of three independent experiments, each performed in duplicate.