Normal stroma suppresses cancer cell proliferation via mechanosensitive regulation of JMJD1a-mediated transcription

ARTICLE Received 8 Sep 2015 |Accepted 15 Jun 2016 |Published 4 Aug 2016 Normal stroma suppresses cancer cell proliferation via mechanosensitive regulation of JMJD1a-mediated transcription Riina Kaukonen1, Anja Mai1, Maria Georgiadou1, Markku Saari1, Nicola De Franceschi1, Timo Betz2, Harri Sihto3, Sami Ventela ¨1,4, Laura Elo1,5, Eija Jokitalo6, Jukka Westermarck1, Pirkko-Liisa Kellokumpu-Lehtinen7, Heikki Joensuu3,8, Reidar Grenman4& Johanna Ivaska1,9 Tissue homeostasis is dependent on the controlled localization of specific cell types and the correct composition of the extracellular stroma. While the role of the cancer stroma in tumour progression has been well characterized, the specific contribution of the matrix itself is unknown. Furthermore, the mechanisms enabling normal—not cancer—stroma to provide tumour-suppressive signals and act as an antitumorigenic barrier are poorly understood. Here we show that extracellular matrix (ECM) generated by normal fibroblasts (NFs) is softer than the CAF matrix, and its physical and structural features regulate cancer cell proliferation. We find that normal ECM triggers downregulation and nuclear exit of the histone demethylase JMJD1a resulting in the epigenetic growth restriction of carcinoma cells. Interestingly, JMJD1a positively regulates transcription of many target genes, including YAP/TAZ (WWTR1), and therefore gene expression in a stiffness-dependent manner. Thus, normal stromal restricts cancer cell proliferation through JMJD1a-dependent modulation of gene expression. DOI: 10.1038/ncomms12237 OPEN 1Centre for Biotechnology, University of Turku, 20520 Turku, Finland. 2Institute of Cell Biology, Center for Molecular Biology of Inflammation, 48149 Muenster, Germany. 3Laboratory of Molecular Oncology, Translational Cancer Biology Program, University of Helsinki, 00290 Helsinki, Finland. 4Department of Otorhinolaryngology, Head and Neck Surgery, Turku University and Turku University Hospital, 20521 Turku, Finland. 5Department of Mathematics and Statistics, University of Turku, 20520, Turku, Finland. 6Institute of Biotechnology, Electron Microscopy Unit University of Helsinki, 00014 Helsinki, Finland. 7Department of Oncology, Tampere University Hospital, 36280 Tampere, Finland. 8Department of Oncology, Helsinki University Central Hospital, 00029 Helsinki, Finland. 9Department of Biochemistry and Food Chemistry, University of Turku, 20520 Turku, Finland. Correspondence and requests for materials should be addressed to J.I. (email: johanna.ivaska@utu.fi). NATURE COMMUNICATIONS | 7:12237 | DOI: 10.1038/ncomms12237 | www.nature.com/naturecommunications 1 In normal tissue, different cell types are spatially confined. For example, the normal epithelium is separated from the underlying stromal extracellular matrix (ECM) by the basement membrane, and thus they are not in a direct contact with each other. The stromal ECM is produced by the fibroblasts and serves as an important regulator of tissue homeostasis1. In contrast, invasive carcinomas contain a complex mixture of tumour cells and stromal components where the cancer cells interact with the altered ECM and are also embedded in it. Changes in the matrix stiffness as well as transition between twodimensional and three-dimensional matrix contacts influence cell proliferation profoundly2–5. Thus, cancer progression involves not only genetic alterations of cancer cells but also changes in the tumour microenvironment6. The role of the stroma as a potent antitumorigenic barrier has already been elucidated over two decades ago7,8. The exposure of carcinoma cells to normal basement membrane-like components has the capacity to revert breast cancer cells to a near-normal phenotype9. Furthermore, in coculture systems, normal fibroblasts (NFs) can inhibit the growth of certain cancer cells10,11, while a population of transformed cancer-associated fibroblasts (CAFs) can reverse the growth-inhibiting effect of NFs12. Thus, cross-talk between epithelial cells and fibroblasts is a critical feature of cancer progression. However, the specific contribution of the matrix itself has not been addressed in detail. Tumour stromata are characterized by increased tissue stiffness and altered matrix architecture that favour proliferation, metastasis and drug resistance via aberrant mechanosignalling13–15. Stiffness also activates the main mediators of mechanotransduction: transcription factors YAP (Yes-associated protein) and TAZ (transcriptional coactivator with PDZ-binding motif)16. Epigenetic regulation is also implicated in cancer progression as it profoundly regulates the transcription profile and phenotype of cells. Many lines of evidence suggest that demethylation of repressive histone methylation marks, such as histone H3 lysine 9 (H3K9), from the gene promoters by histone demethylases is a prerequisite for transcriptional activation17–19. JMJD1A (KDM3A) demethylates monomethyl and dimethyl histone H3K9 in vitro and in vivo and has been implicated as a positive regulator of transcription of several growth-promoting genes17–19. While the mechanisms whereby cancer stroma and CAFs contribute to tumour progression are being actively investigated, much less is known about how the normal stroma exerts tumoursuppressive signals to control tissue homeostasis. Furthermore, the role of epigenetic regulators in the ability of the cells to respond to the stiffness of the tumour microenvironment is not known. To investigate this, we compared the ability of matrices generated by NFs or CAFs from the same patient as well as matrices from immortalized NFs to influence cancer cell proliferation and gene expression. We find that the matrix generated by NFs, but not CAF matrix, profoundly inhibits cancer cell proliferation through mechanosensitive downregulation of the histone demethylase enzyme JMJD1a. Results Normal matrix inhibits cancer cell proliferation. Fibroblasts produce in vitro a robust cell-derived matrix (CDM) that recapitulates many features of the architecture and composition of in vivo ECM20. To investigate the potential effect of matrix on cancer cell proliferation, we generated matrices from telomerase-immortalized NFs (TIFFs) (Fig. 1a) and tested their effectiveness on two highly proliferative and widely studied cancer cell lines, namely cervical cancer HeLa and breast cancer MDA-MB-231 cells. Remarkably, both of these cell lines were significantly growth-inhibited (Fig. 1b) by the normal matrix compared with standard growth conditions on plastic. The growth-restrictive properties of CDM were only observed with the intact CDM as matrix proteins such as fibronectin or collagen I, or solubilized and re-plated CDM did not inhibit MDA-MB-231 cell proliferation (Supplementary Fig. 1a,b). Soluble factors were not implicated either because culturing the MDA-MB-231 cells in conditioned TIFF medium did not influence proliferation (Supplementary Fig. 1c). Thus, only the architecturally intact CDM possessed growth-inhibitory properties. When investigating matrix-induced effects in more detail we, rather unexpectedly, observed that growth inhibition induced by TIFF CDM was maintained in cancer cells following detachment from the matrix by trypsinization and replating on plastic (Fig. 1c). Even though the matrix-exposed cells were returned to plastic in full serum-containing medium, both MDA-MB-231 and HeLa cells continued to proliferate significantly slower than the same cancer cell lines cultured continuously on plastic. Thus, exposure of cancer cells to CDM from NFs is not only growth-inhibitory but has the potential to revert malignant cancer cell proliferation in a sustained manner. We observed that cancer cells grown on normal TIFF-derived CDM had significantly altered cell morphology compared with cells on plastic (Fig. 1d and Supplementary Fig. 1d). This was interesting as mechanical cues and environmental stiffness are known to affect the cytoskeleton and nuclear functions including chromatin condensation and global epigenetic status of a cell21–25, and therefore changes in cell morphology and gene expression could explain the matrix-dependent reversion in the cancer cell phenotype. Matrix induces gene expression changes. We hypothesized that exposure to CDM could induce changes in epigenetic modifiers, hence suppressing cancer cell growth in a sustained manner. To investigate this possibility, we performed Illumina whole-genome transcription analysis in MDA-MB-231 and HeLa cells harvested directly from TIFF CDMs after 6 days (CDM), following detachment from CDMs and replating on plastic for 5 days (CDM to plastic) or in cells grown continuously on plastic (Fig. 1e). As CDM induced sustained growth inhibition in both cell lines, we focused our attention on common transcriptional alterations of epigenetic enzymes. A single well-characterized histone demethylase JMJD1a was significantly downregulated in both cell lines, suggesting that this gene might be linked to the sustained phenotypic alterations triggered by the CDM in both cancer cell lines. Among the other significantly altered genes were signalling proteins SORBS2 and PDE7B, which were downregulated in both cell lines and in both conditions (CDM and CDM to plastic; Fig. 1f and Supplementary Data 1). In addition, expression of 27 genes on CDM and 150 genes following CDM detachment were altered significantly in both cell lines. Several matrix-modifying proteins (MMP3, PLAUR1 and COL1A2) were upregulated on the matrix in both cell lines, while many genes involved in regulating cholesterol synthesis (IDI1,ACAT1 and HMGCS1) were downregulated on the matrix. Following matrix detachment, several hypoxia-related genes were downregulated (JMJD1A (KDM3A),ALDOC,DDIT4,GDF15, ANG and MUC1) even though the rather thin CDM layer is unlikely to restrict oxygen diffusion in the in vitro cultures26.In addition, MYC-target genes (FAPB5 and NME1) were upregulated following CDM detachment and replating on plastic. Since the mechanism of sustained growth inhibition was of primary interest, we focused specifically on the epigenetic modifier enzyme JMJD1a that was downregulated following matrix detachment in both cell lines (Fig. 1f). ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12237 2NATURE COMMUNICATIONS | 7:12237 | DOI: 10.1038/ncomms12237 | www.nature.com/naturecommunications Normal matrix restrains JMJD1a expression. JMJD1a is a H3K9-specific demethylase that has been linked to several biological processes including growth, development and reprogramming17,27–29. Loss of JMJD1a decreases gene expression and increases the inhibitory H3K9me2 modifications in the promoter regions of multiple genes19. We validated that JMJD1a was downregulated in cancer cells following detachment from CDM both on the protein (Fig. 2a,b) and on mRNA levels (Fig. 2c). Interestingly, JMJD1A was downregulated already in cells grown on CDM at the protein level (Fig. 2d,e), while the mRNA levels of JMJD1a were not changed (Fig. 2f). This correlated with reduced stability of the JMJD1a protein on CDM when compared with plastic (Fig. 2g), indicating that CDM-induced repression of JMJD1a first on the protein level and subsequently on mRNA level correlates with the CDM-induced growth inhibition of cancer cells when compared with cells grown on plastic (Fig. 1b,c). Silencing of JMJD1a recapitulated the growth inhibition in plastic-cultured cancer cells and induced a flat quiescent-looking morphology (Fig. 2h and Supplementary Fig. 1e–f). Conversely, proliferation was enhanced in cells with high JMJD1a-GFP levels when compared with high green fluorescent protein (GFP)-expressing cells on plastic (Fig. 2i and Supplementary Fig. 1g). Importantly, forced expression of JMJD1a was sufficient to rescue the CDM-induced growth inhibition on TIFF-derived CDM (Fig. 2j). Furthermore, the growth of MDA-MB-231 (breast cancer) cells in an in vivo chicken embryo chorioallantoic membrane (CAM) assay as well as orthotopic tumour growth in mice were significantly reduced Fibronectin TIFF CDM d eCDM CDM to plastic MMP3 COL1A2 ACAT1 IDI1 HMGCS1 PLAUR1 JMJD1a DDIT4 ANG ALDOC MUC1 GDF15 FABP5 NME1 FBXO32 SORBS2 PDE7B 15027 2 Number of common genes: CDM CDM and CDM to pl. CDM to pl. f CDM CDM to plastic Trypsinization Plastic Plastic 6 Days 5 Days a TIFF CDM Plastic DAPI Actin β1-integrin bMDA-MB-231 1 2 3 4 0 20 40 60 80 100 Time (days) Proliferation (% of max) HeLa P< 0.0001 CDM=35.2±0.6 Plastic=26.5±0.6 Doubling time: 1 2 3 4 0 20 40 60 80 100 P= 0.0001 CDM=29.81±0.8 Plastic=23.6±1.0 Time (days) Doubling time: Collagen I c CDM to pl.=41.5±2.2 Plastic=35.9±0.86 Doubling time: P= 0.038 Time (days) 0 20 40 60 80 100 12 453 MDA-MB-231 Proliferation (% of max) CDM to pl.=42.9±4.3 Plastic=30.6±2.0 Doubling time: P= 0.014 0 20 40 60 80 100 12 453 Time (days) HeLa Figure 1 | Fibroblast-derived CDM induces sustained growth inhibition of cancer cells. (a) Collagen I and fibronectin staining of CDM generated by NFs (TIFFs). Scale bar, 20 mm. (b) Proliferation of MDA-MB-231 and HeLa cells plated on TIFF CDM or on plastic in full medium for the indicated times. n(HeLa) ¼10, n(MDA-MB-231) ¼8. (c) Proliferation of MDA-MB-231 and HeLa cells after detachment from TIFF matrices (6 days on matrix before detachment) and replating on plastic in full medium for the indicated times n(HeLa) ¼7, n(MDA-MB-231) ¼10. (d) Representative images of MDA-MB-231 cell morphology on CDM and plastic. Shown are maximum intensity projections of confocal images. Scale bar, 10 mm. (e) Schematic representation of the experimental set-up. Red arrows indicate the time points of sample collection for Illumina gene expression analysis. (f) Common gene expression changes in MDA-MB-231 and HeLa cells on CDM and 5 days after CDM detachment (CDM to plastic (pl.)). The numbers of commonly regulated genes (upor downregulated) in both cell lines are indicated in the table. Upregulated genes are marked with red and downregulated genes with blue. All data are mean±s.e.m. Unpaired t-test was used for statistical analyses. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12237 ARTICLE NATURE COMMUNICATIONS | 7:12237 | DOI: 10.1038/ncomms12237 | www.nature.com/naturecommunications 3 ed a JMJD1a Actin MDA-MB-231 CDM to plastic Plastic HeLa CDM to plastic Plastic JMJD1a Tubulin CDM Plastic JMJD1a GAPDH MDA-MB-231 CDM Plastic JMJD1a Tubulin HeLa f MDA-MB-231 JMJD1a protein levels CDM to plastic Plastic MDA-MB-231 1.5 0 0.5 1.0 n=7 P= 0.007 JMJD1a mRNA levels CDM to plastic Plastic n=5 P= 0.0072 1.5 0 0.5 1.0 bc Time (h) HeLa CDM to plastic Plastic 1.5 0 0.5 1.0 n=3 P= 0.0365 JMJD1a mRNA levels h JMJD1a protein levels CDM Plastic 1.5 0 0.5 1.0 n=5 P= 0.0075 MDA-MB-231 JMJD1a mRNA levels 2.0 1.5 0 0.5 1.0 MDA-MB-231 n=3 P=ns CDM Plastic iProliferation on CDMProliferation on plastic 0.0 0.1 0.2 0.3 0.4 siControl siJMJD1a P= 0.0014 MDA-MB-231 j k 0 10 20 30 siControl siJMJD1a Tumour area (mm 2 ) Tumour area (mm 2 ) P< 0.001 MDA-MB-231 l Proliferation (% of max.) JMJD1a Tubulin siControl siJMJD1a 0 h 1 h 3 h 5 h 0 h 1 h 3 h 5 h 0.0 0.5 1.0 1.5 Plastic CDM P= 0.031 P=ns JMJD1a protein levels Time of CHX g 0 20 40 60 80 0 50 100 Low JMJD1a=19.8±2.1 High JMJD1a=16.6±0.6 Doubling time P= 0.01 siControl=24.6±1.4 Doubling time P= 0.017 GFP=20.6±0.8 siJMJD1a=54.3±7.5 0 20 40 60 80 100 0 20 40 60 80 100 Proliferation (% of max.) Time (h) MDA-MB-231 siControl siJMJD1a Day 0 JMJD1a Tubulin Day 8 020 40 60 80 0 20 40 60 80 100 GFP JMJD1a Time (h) P< 0.001 Proliferation (% of max.) 50 Mr (K) 150 37 50 Mr (K) 150 Mr (K) 150 37 Mr (K) 150 50 150 50 Mr (K) 150 50 Figure 2 | Normal ECM restrains JMJD1a expression, cell proliferation and tumour growth. (a,b) Western blot (a) and quantification (b) of JMJD1a protein levels normalized to loading control. (c) JMJD1a mRNA expression (qRT–PCR) relative to GAPDH mRNA in MDA-MB-231 and HeLa cells after matrix detachment (6 days on TIFF CDM and 5 days on plastic; CDM to plastic) and on plastic. (d,e) Western blot (d) and quantification of JMJD1a protein levels (e) normalized to loading control. (f) JMJD1a mRNA expression (qRT–PCR) relative to GAPDH mRNA (f) in the indicated cells plated on either TIFF CDM or on plastic for 4 days. (g) Western blot quantification showing JMJD1a stability on CDM and plastic 24 h after plating. Time of the cycloheximide (CHX) treatment is indicated and Pvalues are calculated between 0 and 5 h. Paired t-test was used for statistical analysis, n¼3. (h) Proliferation of MDA-MB-231 cells upon JMJD1a silencing on plastic, n¼3. (i) Proliferation of JMJD1a-GFP or GFP-overexpressing MDA-MB-231 cells on plastic. Cells were sorted by FACS (JMJD1a: high and low; GFP: high), n¼3. (j) Proliferation of GFP control and JMJD1a-GFP-overexpressing MDA-MB-231 cells on TIFFderived CDM. n(GFP) ¼11 CDMs and n(JMJD1a-GFP) ¼12 CDMs. Two-way analysis of variance (ANOVA) was used to calculate the Pvalue. Data are mean±s.e.m. (k) Control or JMJD1a siRNA-transfected MDA-MB-231 cells (1 106) were implanted on CAM membranes inside a plastic ring to analyse tumour growth in vivo for 3 days. Shown are quantified tumour areas from three individual experiments n(siControl) ¼25, n(siJMJD1a) ¼23 eggs. (l) Orthotopic tumour growth assay. Control or JMJD1a siRNA-transfected MDA-MB-231 cells (2 106) were injected into the fat pad of nude mice (n¼19) and tumour growth was measured 8 days after injection. Western blot in showing the silencing efficacy of JMJD1a siRNA on the day of the injection (Day 0) and at the end of the experiment (Day 8). Shown are mean±s.d. and (g–i) mean±s.e.m. Paired t-test was used for statistical analyses in b–hand non-paired t-test in j,k. ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12237 4NATURE COMMUNICATIONS | 7:12237 | DOI: 10.1038/ncomms12237 | www.nature.com/naturecommunications upon silencing of JMJD1a (Fig. 2k,l). Thus, JMJD1a is downregulated following CDM exposure of cells and is a potent regulator of cancer cell proliferation in vitro and in vivo. NFs and CAFs generate architecturally distinct CDM. To test the ability of stromal ECM to influence cancer cells in a more clinically relevant model, we isolated CAFs from the tumour stroma of three head and neck squamous cell carcinoma (HNSCC) patients (Supplementary Fig. 2a) and NFs from an unaffected area of the same patients. CAFs were identified by the high expression of CAF marker smooth muscle actin-a(a-SMA), which was low or absent in the NFs and the TIFFs (Fig. 3a and Supplementary Fig. 2b). RNA-sequencing demonstrated that all three CAFs and NFs clustered together (Supplementary Fig. 2c,d), which indicated that CAFs derived from different patients were more similar to other CAFs than to their corresponding NFs and all the NFs resembled each other (Supplementary Fig. 2c,d). However, all NF and CAF cell lines expressed genes, which are typically highly expressed by fibroblasts, such as Vimentin, Fibronectin 1 and several different collagens validating that they are fibroblasts (Supplementary Fig. 2e). CAFs had elevated levels of YAP and TAZ (Fig. 3b–d) compared with NFs, and they were also more able to contract collagen gels (Supplementary Fig. 2f), in line with a previous report on the role of YAP and contractility in the CAF phenotype30. In addition, we observed a significant upregulation of b1-integrin, which has also been connected to contractility and mechanosignalling (Supplementary Fig. 2g,h). We analysed the CDM produced by the different fibroblasts using immunofluorescence and scanning electron microscopy (SEM). Immunofluorescence staining revealed that similarly to TIFF CDM (Fig. 1a) the matrix bundles in NF CDM were more sparse compared with the corresponding CAF matrix, which displayed a denser and more uniform collagen and fibronectin staining (Fig. 3e and Supplementary Fig. 3a,b). Furthermore, based on SEM analysis, the NF and TIFF CDMs had more uniform and aligned structures compared with CAF-derived matrices (Fig. 3f). According to the RNA-sequencing data, the mRNA expression of different types of collagen or fibronectin was not changed, suggesting that matrix assembly and/or turnover, rather than production, results in distinct CDM architecture between NF and CAF CDM. However, even though we could not find significant differences on mRNA levels of matrix-related genes or in cell adhesion to solubilized and re-plated NF and CAF CDM (Supplementary Fig. 3c), the NF and CAF CDMs may differ in their protein composition as suggested earlier by others31. These data show that CDMs generated by NFs and CAFs differ and that CAF-derived matrices from different patients share similar features that are distinct from matrices made by NFs from the same individual. CAF matrix lacks growth-inhibitory properties. To test the ability of the patient-derived stromal ECM to influence the proliferation of cancer cells, we cultured MDA-MB-231 and HeLa cells on CDMs derived from either NFs or CAFs. Interestingly, NF CDM was significantly growth-inhibitory compared with CDM generated by CAFs from the same patient (Fig. 3g), and the same was observed when comparing TIFF and CAF CDM (Supplementary Fig. 3d). Importantly, also the growth of the patient-derived primary squamous cell carcinoma (SCC) cells was inhibited by the NF CDM matrix compared with the CAF CDM (Supplementary Fig. 3e), demonstrating that the growthrestrictive ability of NF CDM is widely applicable to different carcinomas. Similarly to the TIFF CDM, the growth restriction was specifically due to the matrix and not soluble factors as coculture of SCC or MDA-MB-231 cells with NFs or CAFs separated by a filter or conditioned medium from NFs or CAFs had no effect on proliferation (Supplementary Fig. 3f–h). In line with results obtained in breast cancer cells, JMJD1a silencing was sufficient to inhibit the proliferation of patient-derived SCC cells (Supplementary Fig. 3i). Concurrent with the ability of NF-derived CDM to downregulate JMJD1a levels, cancer cells on plastic and CAF CDM expressed high levels of JMJD1a, whereas JMJD1a was downregulated on normal CDMs (TIFF and NF; Fig. 3h), further validating the ability of normal CDM to restrict proliferation by JMJD1a downregulation. JMJD1a levels correlate with activated stroma within tumours. The data above demonstrate that NF CDM downregulates and CAF CDM supports levels of JMJD1a. We found histologic features in human breast cancer and HNSCC tumours that are compatible with our experimental model, suggesting that the CAF matrix supports JMJD1a expression. We observed that JMJD1a expression coincides with the presence of a-SMA-positive stromal cells, which are characteristic for the tumour microenvironment32,33. Staining of 10 normal breast tissue samples, 28 primary breast cancer sections and 7 lymph node metastasis revealed that in normal breast tissues JMJD1a expression was low or absent and a-SMA was restricted to the basal mammary epithelial cells (Fig. 4a). In contrast, 27/28 of the breast cancers were JMJD1a-positive (low: 57%; intermediate: 29%; high: 11%) and 27/28 of the tumours had a-SMA-positive stroma (Fig. 4a). In addition, all metastases were positively stained for JMJD1a and a-SMA. In HNSCC patient samples, 10/14 of the cancers were JMJD1aand a-SMA-positive (JMJD1a low: 50%; JMJD1a high: 21%). All HNSCC samples that were highly JMJD1a-positive also exhibited intense a-SMA expression in the stroma (Fig. 4b). These analyses suggest that JMJD1a is expressed in breast and HNSCC carcinomas and correlates with the presence of a-SMA-positive stroma in patients. Since increased matrix stiffness has been linked to cancerassociated stromal alterations and cell proliferation, we measured the stiffness of the patient-derived fibroblast-generated CDMs and TIFF CDM using atomic force microscopy (AFM). High indentation forces of up to 30 nN were applied. Pairwise comparison of matrices generated from fibroblasts from the same patient demonstrated that CAF-derived matrices were significantly stiffer than the normal matrix (Fig. 4c). In addition, stiffness of the TIFF CDM was similar to the NF CDMs, in line with the similar growth-inhibitory properties of the matrices (Fig. 4d). These stiffness values for NF-generated CDMs and the higher range of stiffness in the CAF CDM are highly consistent with earlier measurements on CAF-contracted collagen gels30 and differences in tissue stiffness observed in normal breast tissue and cancer34. JMJD1a levels and localization are regulated by stiffness.The requirement for intact ECM and the fact that normal CDM was less stiff compared with the CAF CDM suggested that matrix stiffness, in addition to matrix architecture and possibly composition, could be involved in the ability of the normal matrix to inhibit cancer cell proliferation. To test this in a controlled manner, cancer cells were grown on collagen I-coated hydrogels of varying stiffness. MDA-MB-231, HeLa and patient-derived SCC cells proliferated significantly more on stiffer supports (Supplementary Fig. 4a,b), suggesting that the lower stiffness of NF CDM is likely to contribute to its growth-restrictive properties. JMJD1a has thus far been reported to localize to the nucleus19,27,35 in line with its function as a histone demethylase. However, nuclear fractionation analyses revealed that JMJD1a localizes both to the nucleus and to the cytoplasm in NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12237 ARTICLE NATURE COMMUNICATIONS | 7:12237 | DOI: 10.1038/ncomms12237 | www.nature.com/naturecommunications 5 MDA-MB-231 cells (Supplementary Fig. 4c), suggesting that it might be shuttling between the two compartments. YAP/TAZ transcription factors are well-established mechanosensitive regulators of cell proliferation, such that stiff matrix and cell spreading support YAP/TAZ protein stability and nuclear localization36. To study the potential mechanosensitivity of JMJD1a, we investigated JMJD1a localization and levels under conditions known to regulate YAP/TAZ36. YAP/TAZ protein YAP protein levels (RQ) Patient 3 NF CAF P=ns n=3 0.5 1.0 1.5 2.0 0 Patient 2 NF CAF P= 0.008 n=5 Patient 1 NF CAF 3.0 2.0 1.0 0 P= 0.0211 n=4 3.0 2.0 1.0 0 ×10,000 NF CAF Patient 2 P= 0.003 n=6 6.0 4.0 2.0 0 TAZ protein levels (RQ) Patient 1 NF CAF 4.0 2.0 0 1.0 3.0 P= 0.02 n=4 Patient 3 5 10 20 0 NF CAF 15 P= 0.043 n=3 α-SMA Tubulin NF CAF NF CAF Patient 2Patient 1 Patient 3 NF CAF CAF CDM NF CDM TIFF CDM Patient 1Patient 3 FibronectinCollagen NF CDM CAF CDM NF CDM CAF CDM 0 2 4 6 0 20 40 60 80 100 Time (days) Proliferation (% of max.) HeLa Doubling time: NF CDM: 116.3±23.8 CAF CDM: 61.4±6.8 P= 0.044 13 5 0 2 4 0 20 40 60 80 100 Time (days) Doubling time: NF CDM: 35.8±1.6 CAF CDM: 29.3±1.0 P= 0.002 MDA-MB-231 Proliferation (% of max.) 135 Mr (K) 50 50 ab cd e f g h Patient 1 Patient 2 Tubulin YAP TAZ NF CAF NF CAF Actin YAP TAZ Patient 3 Tubulin YAP TAZ NF CAF Mr (K) 75 50 50 50 37 NF CDM CAF CDM Plastic JMJD1a Actin 0.1 1.05 1.0 TIFF CDM CAF CDM Plastic JMJD1a Actin 0.65 1.02 1.0 Mr (K) Mr (K) 150 150 50 37 50 37 Figure 3 | Patient-derived CAF and NF CDMs are architecturally and functionally distinct. (a,b) Representative western blots showing SMA-a(a) and YAP/TAZ expression (b) in NFs and CAFs isolated from three SCC patients. (c,d) Quantification of YAP (c) and TAZ (d) expression in NFs and CAFs normalized to loading control. Data are mean±s.d. (e) Collagen I (red) and fibronectin (blue) staining of patient #1 and #3 NF and CAF CDM. Scale bar, 20 mm. (f) Representative SEM images of TIFF, Patient #1 NF and CAF CDM. Scale bar, 5 mm. (g) Proliferation of MDA-MB-231 and HeLa cells on Patient #1 NF and CAF CDM. n(MDA-MB-231) ¼10–11 and n(HeLa) ¼19 from three independent experiments. Data are mean±s.e.m. and Pvalues are calculated from the doubling times. (h) JMJD1a expression in MDA-MB-231 cells cultured on TIFF, NF and CAF CDMs and on plastic. Quantification shows relative JMJD1a amount normalized to loading control. (c,d) Non-parametric Mann–Whitney test was used for statistical analyses in c,d and non-paired t-test in g. ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12237 6NATURE COMMUNICATIONS | 7:12237 | DOI: 10.1038/ncomms12237 | www.nature.com/naturecommunications levels are known to be downregulated on soft supports. Interestingly, we found that also JMJD1a stability and therefore the protein levels were reduced when cells were cultured on soft hydrogels (Fig. 5a,b and Supplementary Fig. 4d). In addition to matrix stiffness, RhoA-signalling, actomyosin contractility and cell spreading are known regulators of YAP/TAZ localization and protein stability16. To test whether these ques regulate JMJD1a as well, we plated MDA-MB-231 cells on micropatterns with equal total adhesive surface distributed over a variable spreading area. In cells spreading on 800-mm2 fibronectin-coated micropatterns (large), both JMJD1a and YAP/TAZ were predominantly nuclear compared with cells spreading on 400-mm2(small) fibronectin-coated micropatterns, where both proteins were predominantly cytoplasmic (Fig. 5c,d). Accordingly, we observed that both JMJD1a and YAP/TAZ became cytosolic on lower stiffness hydrogels (0.5 kPa) as soon as a Normal breast Breast cancer Metastasis Breast cancer Metastasis None Low (+1) Medium (+2) High (+3) α-SMA Normal breast Breast carcinoma JMJD1a Negative stroma Low Medium α-SMA None Low Medium c Low Low Adjacent tissue slices Medium Medium ×10 ×20 Adjacent tissue slices Breast carcinoma S LB S S TT SS α-SMA JMJD1a HNSCC b None Low High JMJD1a and α-SMA JMJD1a ×10 ×20 d 0 100 200 300 400 Stiffness (Pa) NF matrix CAF matrix Patient 1 matrices 0 100 200 300 400 NF matrix CAF matrix Patient 3 matrices Stiffness (Pa) 0 100 200 300 400 Stiffness (Pa) TIFF matrix P< 0.0001 P< 0.0001 Figure 4 | JMJD1a levels correlate with a-SMA-positive stroma in human tumours. (a,b) JMJD1a and a-SMA staining from sections of the same normal breast tissue and primary breast carcinomas (a) or HNSCC (b) tissue. For breast carcinomas, examples of the immunostaining of different expression levels and two different magnifications of the same tissue are shown. Scale bar, 200 mm. S, stroma; T, tumour. Inset in ahighlights the a-SMA localization in basal cells in the normal mammary gland (B, basal; L, luminal). Quantification of the incidence of JMJD1a or a-SMA positivity in the analysed samples is shown. (c,d) Stiffness of the patient-derived NF or CAF (c) and TIFF (d) CDMs is expressed by the Young’s modulus, which was measured using AFM indentation. Each grey dot represents individual measurement and black line indicates the mean. Non-paired t-test was used for statistical analysis. Non-paired t-test was used for statistical analysis. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12237 ARTICLE NATURE COMMUNICATIONS | 7:12237 | DOI: 10.1038/ncomms12237 | www.nature.com/naturecommunications 7 the cells had fully adhered (3h) and were increasingly nuclear on stiffer supports both in MDA-MB-231 and patient SCC cells (Fig. 5e–g and Supplementary Fig. 5a–d), whereas another nuclear protein Son was not mechanosensitive on hydrogels (Supplementary Fig. 5e). Importantly, both JMJD1a and YAP/TAZ were predominantly cytoplasmic in MDA-MB-231 cells grown on soft, normal CDM compared with their nuclear localization on plastic or collagenand fibronectin-coated plastic (Fig. 5h,i and Supplementary Fig. 5f), and very similar regulation was observed also in the patient-derived SCC cells (Supplementary Fig. 5g). Correspondingly, JMJD1a was prominently cytoplasmic on NF CDM compared with its nuclear localization on CAF CDMs (Fig. 5j), indicating that CDM-induced regulation of JMJD1 localization and levels is similar to the previously established regulation of YAP/TAZ (Supplementary Fig. 5h). However, unlike YAP/TAZ, JMJD1a localization was not dependent on an intact actin cytoskeleton or Rho-signalling as g Nuclear JMJD1a/total CDM Plastic Collagen + FN 1.0 0.5 1.5 0 P< 0.0001 P< 0.0001 ns i a 0 1.0 0.5 1.5 2.0 P= 0.01 P= 0.029 0.5 50 PL Relative JMJD1a levels Stiffness (kPa) 0.5 kPa 50 kPa NNCC Lamin A/C GAPDH JMJD1a YAP TAZ 0.5 kPa4 kPaPlastic JMJD1a YAP/TAZ DAPI Merge f CDMPlastic Merge ROIActin DAPIJMJD1a Cyto Nuc+ Cyto Nuc Cell number 0.5 kPa 4 kPa 50 kPa Plastic 0 50 100 150 Cell number JMJD1a 5 15 25 5 15 25 YAP/TAZ LargeSmall JMJD1aJMJD1aYAP/TAZ YAP/TAZ DIC DIC Pattern Pattern 0.5 50 PL JMJD1a Tubulin kPa kl b e Cyto Nuc+ Cyto Nuc c Rho ROCK ECM Actomyosin Nucleus JMJD1a YAP/TAZ SRC ? ECM YAP/TAZ Nucleus JMJD1a SRC JMJD1a STIFF SOFT CA-Src expression on 0.5 kPa JMJD1a YAP/TAZ JMJD1a DAPI YAP/TAZ DAPI JMJD1a YAP/TAZ JMJD1a DAPI YAP/TAZ DAPI CA-Src expression on plastic Large Small NF CDM CAF CDM JMJD1a DAPI j JMJD1a d Cell number h Large Small Cyto Nuc+ Cyto Nuc 50 Mr (K) 150 75 37 75 50 150 ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12237 8NATURE COMMUNICATIONS | 7:12237 | DOI: 10.1038/ncomms12237 | www.nature.com/naturecommunications Blebbistatin, Cytochalasin D or ROCK inhibitor Y27632 treatment did not change JMJD1a localization, while YAP/TAZ became predominantly cytoplasmic with all these treatments (Supplementary Fig. 6a). Integrin signalling is known to be important for mechanosignalling37–40, albeit the specific requirement for integrin–ECM interaction in YAP/TAZ regulation remains controversial41–43. In order to investigate whether integrin b1 activity could regulate JMJD1a, we plated cells on integrin b1 inactivating (4B4) or activating monoclonal antibodies (12G10) and analysed JMJD1a localization in serumfree conditions. We found that locking integrins into active or inactive conformation was not sufficient to alter JMJD1a localization on soft or stiff (Supplementary Fig. 6b). Furthermore, silencing of integrin b1 from MDA-MB-231 cells had no effect on the nuclear localization of JMJD1a, suggesting that JMJD1a nuclear localization is unlikely to be dependent on a specific integrin heterodimer (Supplementary Fig. 6c). Next, we tested whether known integrin downstream effectors, Focal adhesion kinase or Rous sarcoma oncogene cellular homolog (SRC)-kinase, could regulate JMJD1a localization. We found that pharmacological inhibition of Focal adhesion kinase and SRC alone or in combination did not alter JMJD1a localization on plastic but SRC inhibition induced YAP/TAZ translocation to the cytoplasm (Supplementary Fig. 6d). However, expression of constitutively active SRC (CA-SRC) increased tyrosine phosphorylation of GFPJMJD1a (Supplementary Fig. 6e) and was sufficient to induce nuclear JMJD1a on soft 0.5kPa hydrogels (Fig. 5k), indicating that SRC activation is sufficient to support JMJD1a nuclear localization on soft. Interestingly, forced expression of CA-SRC did not alter YAP/TAZ localization on soft (Fig. 5k), demonstrating that stiffness-mediated regulatory pathways of YAP/TAZ and JMJD1a localization are distinct (Fig. 5i). JMJD1a regulates YAP/TAZ expression. JMJD1a and YAP/TAZ levels were significantly reduced on TIFF CDM (Fig. 6a) in line with their cytoplasmic translocation (Supplementary Fig. 5f), and this correlated with reduced transcription of the well-known YAP/TAZ target genes Connective tissue growth factor (CTGF) and Thrombospondin 1 (THBS1) on TIFF CDM compared with plastic (Fig. 6b). Furthermore, we observed that levels of YAP/TAZ and JMJD1a in individual MDA-MB-231 cells correlated significantly (Fig. 6c). Thus, we were interested to investigate the potential link between JMJD1a and YAP/TAZ. Chromatin immunoprecipitation (ChIP) assays revealed that JMJD1a is recruited to TAZ promoter (Fig. 6d, Supplementary Fig. 7a) and, in line with the demethylase activity of JMJD1a, transient JMJD1a silencing increased H3K9me2 methylation on the TAZ promoter (Fig. 6e and Supplementary Fig. 7a). Furthermore, JMJD1a silencing with two independent short interfering RNAs (siRNAs) reduced YAP/TAZ protein and mRNA levels in MDA-MB-231 and patient-derived SCC cells (Fig. 6f–h and Supplementary Fig. 7b–e) as well as expression of YAP/TAZ target genes in MDA-MB-231 (Fig. 6i and Supplementary Fig. 7a). Conversely to JMJD1a silencing, overexpression of JMJD1a-GFP (which localizes to the nucleus similarly to endogenous JMJD1a on plastic, Supplementary Fig. 7f) increased YAP/TAZ levels (Fig. 6j,k) as well as THBS1 and CTGF gene expression (Fig. 6l). Importantly, overexpression of JMJD1a was sufficient to increase YAP/TAZ levels even on TIFF-derived CDMs (Fig. 6m) and 4 kPa hydrogels (Fig. 6n), where YAP/TAZ protein-level stability is compromised because of increased cytoplasmic localization and degradation36. However, overexpression of wild-type or active YAP mutant (YAP-5SA (ref. 44)) alone was not sufficient to rescue proliferation on MDA-MB-231 cells grown on TIFF-derived CDMs or in JMJD1a-silenced cells, suggesting that additional JMJD1a target genes contribute to the CDM-induced growth inhibition (Supplementary Fig. 7g,h).We also found that silencing of JMJD1a had no effect on either YAP/TAZ nuclear localization (Supplementary Fig. 8a) or phosphorylation of LATS1/2 (Supplementary Fig. 8b), which is a negative regulator of YAP/TAZ protein stability. Furthermore, we could not detect increased YAP (S-127) phosphorylation, which is associated with reduced stability of YAP, upon JMJD1a silencing (Supplementary Fig. 8c) and protein stability of YAP/TAZ was not reduced upon JMJD1a silencing, suggesting that JMJD1a regulates YAP/TAZ on the transcriptional level (Supplementary Fig. 8d). This further demonstrates that JMJD1a is a previously undescribed transcriptional activator of YAP/TAZ expression, and that forced expression of JMJD1a can support YAP/TAZ levels even on soft substrates because of its ability to increase the transcription of YAP/TAZ. JMJD1a and YAP/TAZ expression correlates in human cancer. We found that JMJD1a and YAP/TAZ expression correlated also in human carcinomas. In a large cohort of primary breast tumours45, JMJD1a and YAP/TAZ (the YAP antibody recognizes both transcription factors) were strongly associated with several commonly assessed clinicopathological prognostic factors (Supplementary Tables 1–4 and Supplementary Note 1). In all, 689 (94.3%) and 262 (35.8%) out of the 731 tumours available for JMJD1a staining had positive cytoplasmic and nuclear staining, respectively, and 645 (86.5%) and 514 (68.9%) out of the 746 cancers available for YAP/TAZ staining had positive cytoplasmic and nuclear YAP/TAZ expression. Fully in line with their correlated expressions in vitro, we found that JMJD1a and YAP/TAZ levels significantly correlated in Figure 5 | Mechanosensitive regulation of JMJD1a on soft and stiff ECM and CDM. (a,b) Representative western blot (a) and quantification (b) showing JMJD1a expression in cells plated on 0.5 and 50 kPa hydrogels and on plastic (PL). Tubulin was used as loading control, n¼4 (mean±s.d.). (c) JMJD1a (red) and YAP/TAZ (green) and DAPI (blue) staining in MDA-MB-231 cells on large (800 mm2) and small (400 mm2) spreading area micropatterns (adhesive area is the same). Cell morphology is shown as DIC. (d) Quantification of cytoplasmic and nuclear JMJD1a and YAP/TAZ localization on small and large micropatterns. n(cells) ¼20 per pattern size. (e,f) Immunofluorescence staining showing (e) and quantifying (f) YAP/TAZ and JMJD1a localization on collagen I and fibronectin-coated hydrogels of varying stiffness (0.5, 4 and 50 kPa) and on plastic. Scale bar, 10 mm. (g) Representative western blot showing JMJD1a and YAP/TAZ nuclear (N) and cytoplasmic (CP) localization in cells plated on 0.5 and 50 kPa hydrogels. Lamin A/C and GAPDH were used as fractionation controls. (h,i) Immunofluorescence staining quantification (i) and representative images (h) of JMJD1a localization in cells plated on TIFF CDM, plastic or collagen and fibronectin ligands (2.5 mgml1collagen and 2.5 mgml1fibronectin). Nuclear localization of JMJD1a was quantified with the CellProfiler software. n(cells): CDM ¼74 cells; plastic ¼33 cells and collagen þFN ¼61. Scale bar, 10 mm. (j) Representative immunofluorescence images showing JMJD1a localization in MDA-MB-231 cells growing on NF and CAF CDMs for 3 days. Scale bar, 10 mm. (k) YAP/TAZ and JMJD1a localization in CA-SRC-expressing MDA-MB-231 cells growing on soft 0.5 kPa hydrogels or on plastic. Representative images from three independent experiments. Scale bar, 10 mm. (l) Model of distinct mechanotransductional regulation of YAP/TAZ and JMJD1a on soft and stiff. Red arrows indicate the pathway, which we and others have shown to regulate YAP/TAZ nuclear localization. Blue arrow indicates the SRC kinase-mediated and stiffness-dependent regulation of JMJD1a and YAP/TAZ. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms12237 ARTICLE NATURE COMMUNICATIONS | 7:12237 | DOI: 10.1038/ncomms12237 | www.nature.com/naturecommunications 9