All animal experiments were reviewed and approved by the Animal Experiments Committee of the Groningen University Medical Center (#AVD10500015129) and performed in accordance with the recommendations for animal experiments issued by the European Commission directive 2010\63. Pulmonary arterial hypertension (PAH) induction in Wistar rats (7–9 weeks old; Harlan, Horst, Netherlands) was performed by a single intraperitoneal injection of monocrotaline (60 mg/kg) followed by placement of an aortocaval (AC) shunt one week later as described previously (26, 27). 4 weeks post AC-shunt placement, rats were euthanized by continuous inhalation of 3%–5% isoflurane in air for the duration of the procedure. Hemodynamic examination was performed under continuous inhalation of 3%–5% isoflurane in air for the duration of the procedure, prior to euthanisation using the closed chest technique as described previously (28). A fluid-filled pressure catheter was inserted into the right internal jugular vein and guided to the pulmonary artery while monitoring pressure waveform using a bedside monitor. Right ventricular systolic and diastolic pressure, pulmonary arterial pressure, and pulmonary Wedge pressure were measured. Mean pulmonary arterial pressure (mPAP) was calculated as follows: mPAP(mmHg)=(2/3dPAP+1/3sPAP), in which dPAP and sPAP are diastolic and systolic pulmonary artery pressure, respectively. Total pulmonary vascular resistance was estimated as: PVR(mmHg⋅ml⋅min−1)=(mPAP−PCWP)CO, in which mPAP and CO are pulmonary capillary wedge pressure and cardiac output, respectively. Rats were euthanized under 3%–5% isoflurane anaesthesia by exsanguination. Hearts were harvested and processed for further analyses. The hearts were weighed and dissected into the left atrium, left ventricle, right atrium, right ventricle, and septum, which were all weighted individually. The right ventricle was cut in half and one half was fixed in 3.6% formalin and embedded in paraffin, the other half was snap frozen using liquid nitrogen.

Human umbilical vein endothelial cells (HUVEC; Lonza, Walkersville, MD) were cultured on 1% gelatin-coated culture plastics in endothelial cell medium consisting of RPMI 1640 (Lonza, Verviers, Belgium) supplemented with 20% Fetal Bovine Serum (FBS; Sigma, St. Louis, MO, USA), 50 µg/ml Endothelial Cell Growth Factors [ECGF; own isolate according to Burgess et al. (29)], 2 mm L-glutamine (Lonza, Basel, Switzerland), 1% penicillin/streptomycin (Gibco, Waltham, MA, USA) and 5 U/ml heparin (Leo Pharma, The Netherlands). Confluent HUVEC were dissociated with Trypsin EDTA in PBS (Gibco). HUVEC were used between passage 3 and 7. EndMT was initiated by culturing HUVEC in RPMI 1640 supplemented with 20% FBS, 1% penicillin/streptomycin, 2 mm L-glutamine, and 5 U/ml heparin and 10 ng/ml TGFβ1 (Peprotech, NJ, USA) for 72 h. Human Embryonic Kidney (HEK293T) cells were cultured in DMEM (Lonza, Basel, Switzerland) supplemented with, 10% FBS, 2 mm L-glutamine, and 1% penicillin/streptomycin.

HEK293T cells were transfected in 75 cm² plates in a reaction volume of 10 ml with 0.38 µg/ml pLKO.1-shEZH2 or 0.38 µg/ml pLKO.1-shSCR, 0.09 µg/ml pVSVG (envelope plasmid) and 0.38 µg/ml pCMV-R8.91 (gag-pol 2nd generation packaging plasmid) using Endofectin™ Lenti transfection reagent (GeneCopoeia, Rockville, MD, USA). 24 h post-transfection, HEK293T cells were placed on endothelial cell medium. 48- and 72-hours post-transfection, viral supernatants were collected, centrifuged at 500xg, and filtered through 0.45 µm filters. Viral supernatants were supplemented with 6 mg/ml polybrene (Sigma) and added to 70% confluent HUVEC (P3 up to P7) for two consecutive rounds of 24 h exposure. Transduced HUVEC were passaged once, and transduced cells were selected with 4 µg/ml puromycin O/N after which they were used for downstream stimulation or analysis.

HUVEC were pre-incubated with reduced serum medium OptiMEM (Gibco). Cells were transfected with 50 µM hsa-miR-29c-3p miRCURY LNA miRNA inhibitor (Qiagen, Hilden, Germany) using Lipofectamine 2000 reagent (Invitrogen, Waltham, MA, USA) according to manufacturer's protocol. Cells were incubated with the small interfering RNA (siRNA)/Lipofectamine mix for 4 h at 37°C and 5% CO₂ after which transfection medium was replaced by endothelial cell medium. The next day, HUVEC were used for downstream stimulation or analysis.

RNA was isolated using in TriZOL reagent (Ambion, Austin, TX, USA) in accordance with manufacturer's protocol. RNA concentration and purity were determined using a Nanodrop 1,000 spectrophotometer (Thermo Scientific, Waltham, MA, USA). cDNA synthesis was performed using RevertAid^TM First Strand cDNA Synthesis Kit (Thermo Scientific), according to manufacturer's protocol. 5 ng of total cDNA for gene transcript analysis was amplified. For analysis of microRNA transcripts, the ABI Taqman microRNA reverse transcription kit (Thermo Scientific) was used according to manufacturer's instructions using microRNA specific stem-loop primers (Table 1). 10 ng of total cDNA for microRNA expression analysis was amplified. For all transcript analyses, cDNA was amplified in duplicate on the Viia7 Real-Time PCR System (Thermo Fisher) in a reaction with 0.6 µM primers (gene transcript analysis, Table 2) or 0.5 µM primers (microRNA transcript analysis, Table 2) using SYBR Green chemistry (Roche Diagnostics GmbH, Mannheim, Germany). Data was analysed using the Viia7 software (Thermo Scientific). Cycle threshold (Ct) values for individual reactions were determined and normalized against GAPDH or ACTB (gene transcript analysis) or RNU6 (microRNA transcript analysis). Data is shown as fold change compared to control and was calculated using the 2^−ΔΔCt method.

Cells were trypsinized, pelleted and the chromatin was crosslinked with 1% formaldehyde (Sigma) for 8 min. Crosslinking was stopped by the addition of 125 mm glycine (104201, Merck, Kenilworth, NJ, USA). Cell pellets were lysed on ice with SDS lysis buffer (1% SDS, 10 mm EDTA, 50 mm Tris-HCl pH 8.0) supplemented with freshly added 100 mm proteinase inhibitor cocktail (P8340, Sigma) for 15 min. The chromatin was fragmented using the Biorupter (Diagenode, Seraing, Belgium) by applying 5 cycles (30″ ON/OFF) and cleared by centrifugation. Chromatin was diluted 10 times with RIPA buffer (0.1% SDS, 0.1% Na-deoxycholate, 1% Triton-X-100, 1 mm EDTA, 0.5 mm EGTA, 10 mm Tris-HCl, 140 mm NaCl) supplemented with freshly added 100 mm proteinase inhibitor cocktail (Sigma). 40 µl Protein A Dynabeads (#10002D, Invitrogen) were coated with 4 µg H3K27Me3 (07-449, Merck Millipore, Burlington, MA, USA) or IgG control (ab46540, Abcam) and added to the chromatin followed by incubation at 4°C O/N while rotating. The beads were collected using a magnet and washed 3 times with ice-cold PBS, the remaining complexes were eluted with elution buffer (100 mm NaHCO₃, 1% SDS) for 15 min. at RT. Eluted samples were supplemented with 5M NaCl and RNAse (EN0531, Thermo Scientific) and incubated for 4 h at 62°C to reverse crosslinking. 2 µl proteinase K (03115828001, Roche) was added and incubated for 1 h at 62°C. DNA fragments were purified with a QIAquick PCR purification kit (Qiagen) according to manufacturer's protocol and quantified by qRT-PCR using primers for miR-29c-3p and miR-98 (Table 2). Data is shown as fold enrichment over control IgG values.

Cells were harvested in RIPA buffer (Thermo Scientific) supplemented with 1% v/v protease inhibitor cocktail (#P8340, Sigma) and 1% v/v phosphatase inhibitor cocktail (#1861277, Sigma). Samples were sonicated and centrifuged, and protein concentration was measured with a DC protein assay (BioRad, Hercules, CA, USA). Equal amounts of protein were separated by electrophoresis on polyacrylamide gels followed by protein blotting onto nitrocellulose membranes using the semi-dry Transblot Turbo system (BioRad). Membranes were blocked in 5% Elk milk powder (FrieslandCampina, Amersfoort, the Netherlands) in TBS-T at RT for 1 h, and incubated with antibodies against GAPDH (1:2,000, CS-5274S, Cell Signaling, Danvers, MA, USA), EZH2 (1:1000, CS-5246S, Cell Signaling), VE-cadherin (1:1,000, CS-2500, Cell Signaling), eNOS (1:1,000, bs-610296, BD Bioscience, Franklin Lakes, NJ, USA), SM22α (1:1,000, ab14106, Abcam, Cambridge, England) or αSMA (1:1,000, ab124964 Abcam) at 4°C O/N. Membranes were washed twice in TBS-Tween (0.1%), once in AP detection buffer (100 mm NaCl, 100 mm Tris-Base, 50 mm MgCl₂, pH 9.5) and developed using AP detection buffer supplemented with 5-bromo-4-chloro-3′-indolyphosphatae p-toluidine salt (BCIP, 165 µg/ml) and nitro blue tetrazolium chloride (NBT, 330 µg/ml). Densitometry analysis was performed using Totallab 120 (Nonlinear Dynamics, Newcastle upon Tyne, England). Data is shown as fold change compared to control.

Cells were trypsinized and counted. Cytospins of the samples were prepared using the Cytospin 4 centrifuge (Thermo Scientific) using a cell suspension of 25,000 or 50,000 cells per 100 µl endothelial cell medium. Cytospins were made by spinning 5 min. at 550 rpm with low acceleration followed by air drying for 1 h at RT and stored at −20°C. Prior to staining the slides were thawed under a fan for 10 min. after which they were fixed with 2% paraformaldehyde in PBS for 30 min. at RT. For analysis of intracellular proteins, cells were permeabilized with 0.5% Triton-X-100 solution for 15 min at RT. Blocking of non-specific antibody binding was performed using 5% donkey serum in PBS for 10 min. at RT. Samples were incubated with antibodies against H3K27Me3 (1:200, 07-449, Merck Millipore) in PBS supplemented with 5% donkey serum at 4°C O/N. Samples were washed 3 times with 0.05% Tween-20 in PBS and incubated with Alexa Fluor® 594-conjugated antibodies (1:500, #A21207, Invitrogen) against rabbit IgG or Alexa Fluor® 647 conjugated antibodies (1:500, #A31571, Invitrogen) against mouse IgG in DAPI/PBS with 2% human serum for 1 h at RT. Samples were washed 3 times with 0.05% Tween-20 in PBS and spots were fixed with Citifluor covered with a coverslip and stored at 4°C until analysis. Samples were imaged on the EVOS FL (Life Technologies), and analysis was performed using ImageJ software (National Institutes of Health, Bethesda, MD, USA).

Perivascular fibrosis was scored after picrosirius red staining (PSR) and counterstaining with Weigert's haematoxylin (both Sigma-Aldrich, St. Louis, MO) following manufacturer's instruction. Samples were imaged on a NanoZoomer digital slide scanner (Hamamatsu Photonics, Shizuoka, Japan. For the quantification of perivascular fibrosis, 12–15 arteries per animal were scored positive (PSR⁺) or negative (PSR⁻) for PSR staining and the percentage of PSR⁺ arteries per animal was calculated (PSR⁺ arteries/total arteries × 100). Perivascular fibrosis was quantified using Aperio ImageScope (Leica Biosystems, Nussloch, Germany). For CD31/SM22α and CD31/H3K27Me3 staining on formalin-fixed, paraffin-embedded sections, prior to immunohistochemistry, deparaffinization was performed by placing sections in xylene I (15 min), xylene II (10 min), 100% ethanol (10 min), 96% ethanol (3 min), 70% ethanol (3 min) and demi water (3 min). Heat-induced antigen retrieval was performed with 0.1 M Tris/HCl (pH 9.0) at 80°C for 16 h. Blocking of non-specific antibody binding was performed using 1% BSA and 5% donkey serum in PBS for 30 min. at RT. Sections were incubated with antibodies against H3K27Me3 (1:200, 07-449, Merck Millipore) or SM22α (1:200, ab14106, Abcam) for 1 h at RT. Sections were washed 3 times followed by incubation with Alexa Fluor® 555-conjugated antibodies (1:500 #A31572, Invitrogen) against rabbit IgG 30 min at RT. Sections were washed 3 times, followed by incubation of the second primary antibody: CD31 (1:200, #AF3628, R&D Systems, Minneapolis, MN, USA) for 1 h at RT. After 3 washing steps, sections were incubated with the second secondary antibody: Alexa Fluor® 647 conjugated antibodies (1:500 #A21447, Invitrogen) against goat IgG in DAPI /PBS for 30 min at RT. Sections were washed 3 times and fixed with Citifluor covered with a coverslip and stored at 4°C until analysis. Sections were imaged on the TissueFAXS (TissueGnostics, Vienna, Austria) in combination with Zeiss AxioObserver Z1 microscope. Image analysis was performed using TissueQuest fluorescence (TissueGnostics) software.

Collagen production by endothelial cells was performed as described elsewhere (30). In short, HUVEC (control, shSCR and shEZH2) were cultured in 96-wells plates at a density of 50,000 cells/cm² and cultured for 96 h with fresh culture medium every 24 h. Where appropriate, HUVEC were stimulated with 10 ng/ml TGFβ1 for 96 h. Cells were gently lysed in extraction buffer (0.5% v/v Triton X-100, 20 mm NH₄OH in PBS) and the remaining matrix rinsed twice in PBS and fixed using 4% paraformaldehyde at RT for 20 min. Samples were blocked in 5% BSA and subsequently incubated with antibodies against collagen type 1 (1:1000, #ab34710, Abcam plc, UK) and collagen type 3 (1:1,000, #ab7778, Abcam plc, UK) at RT for 1 h. After three washes, samples were visualized using AF555-conjugated secondary antibodies (1:2000, #A31572, ThermoFisher) and recorded on a Varioscan spectrofluorometer (Thermo Scientific, Waltham, MA, USA) and plotted against standards of collagen. All samples were normalized against DNA content as measured in the cell lysates using FluoReporter Blue (#F2962, ThermoFisher Scientific) according to manufacturer's instructions. DNA standards were prepared in extraction buffer.

Statistical analysis was performed using GraphPad Prism software version 8 or 9 (San Diego, CA, USA) with an unpaired t-test when comparing the means of two groups or a 1-way ANOVA when comparing the means of more than 2 groups. Outliers within a group were detected with a ROUT outlier test (Q = 1%) and significant outliers were excluded from the analysis. Graphs and tables depict the mean with standard deviation. The number of independent experiments or animals used are indicated in the figure legend. P-values <0.05 were considered statistically significant.

To investigate the correlation between H3K27Me3 abundance and EndMT, we depleted its producing methyltransferase EZH2 in HUVEC using shRNA. shRNA targeted against EZH2 (shEZH2) decreased its expression at the gene and protein level by 3.3-fold and 13.9-fold, respectively, when compared to HUVEC that were transduced with a scrambled control sequence (shSCR, Supplementary Figure S1). Concurrently, the amount of H3K27Me3⁺ cells and the mean fluorescent intensity (MFY) was decreased in EZH2 deficient HUVECs (1.4-fold and 2.7-fold, respectively) when compared to shSCR (+TGFβ) controls (Figure 1).

To induce EndMT, HUVEC were treated with TGFβ1 as described before (7, 31, 32). Stimulation of HUVEC with TGFβ1 decreased the expression of NOS3 (2.0-fold, p = 0.0049) and increased the expression of PECAM1 (1.6-fold, p = 0.011) on gene expression levels as compared to untreated control cells, while expression of CDH5 and TEK remained unchanged (Figures 2A–D). On the protein level, VE-cadherin expression increased (1.5-fold, p = 0.0319), and the expression of eNOS remained unchanged (Figures 2E,F). The reduction in H3K27Me3 abundance did not affect endothelial gene expression, except for TEK, which increased (2.8-fold) compared to the shSCR controls (Figure 2B). Notably, on protein level, the decrease in eNOS expression was aggravated in shEZH2 HUVEC when compared to shSCR controls (Figures 2F,G).

Consistent with the induction of EndMT by TGFβ1, the expression of mesenchymal genes was induced in TGFβ1 stimulated HUVEC and aggravated by the reduction of H3K27Me3 (Figures 3A–D). For example, the expression of TAGLN and CNN1 increased 127.5- and 53.4-fold upon TGFβ1 stimulation (p = 0.0788 and p = 0.3414), which increased to 295.9- and 114.3-fold in shEZH2 HUVEC (Figures 3A,B). Similar, on protein level, SM22α and αSMA increased after the reduction in H3K27Me3 abundance compared to shSCR controls (Figures 3E–G).

EndMT is associated with increased extracellular matrix production induced by TGFβ1. Indeed, TGFβ1-stimulated HUVEC increased gene expression of ECM components COL1A1 and COL3A1 (26.1-fold, p = 0.0005 and 66.6-fold, p = 0.0002, respectively) and protein expression of collagen I and collagen III (4.2-fold, p ≤ 0.0001 and 5.2-fold, p ≤ 0.0001) compared to unstimulated control cells (Figure 4).

Counterintuitively, and in contrast to the aggravated induction of mesenchymal gene expression, lowering the H3K27Me3 abundance in TGFβ1-stimulated HUVEC blunts the expression of these ECM genes on gene (6.8-fold and 19.2-fold for COL1A1 and COL3A1, respectively, Figures 4A,B) and protein levels (3.9-fold and 10.6-fold for collagen I and collagen III, respectively, Figures 4C,D) compared to shSCR controls. In combination, these data suggest that the reduction of EZH2 and its related histone mark H3K27Me3 does not halt the EndMT phenotype but mitigates the fibrogenesis that is associated with EndMT.

As H3K27Me3 is a transcriptional repressor, the possibility of a direct interaction between H3K27Me3 and collagen is unlikely. Rather, the co-occurring decreases suggests another layer of regulation. It is widely accepted that the expression of collagens is co-regulated by microRNAs (miRNAs) (33–36) and changes in H3K27Me3 abundance associates with changes in miRNA expression levels (37, 38). For example, studies have shown that members of the miR-29 family were significantly reduced upon TGFβ1 expression, which associates with an increase in collagen expression (33, 34). Additionally, several studies show that pharmacological or siRNA-based knockdown models of EZH2 led to an upregulation of miR-29b due to reduced H3K27Me3 abundance (37, 38). Besides the miR-29 family, a literature search showed that other miRNAs that could be involved in collagen regulation and might be affected by EZH2 are the let-7 family including let7b and let-7d and miR-98. Therefore, we hypothesized that H3K27Me3 abundance affects the expression of collagen-targeting miRNAs during EndMT. Stimulation of HUVEC with TGFβ1 decreased the expression of miR-29a (6.8-fold, p = 0.0014), miR-29b (2.9-fold, p = 0.0152), miR-29c (2.0-fold, p = 0.1633) and miR-98 (3.3-fold, p = 0.0099), whereas the expression levels of the let-7 family remained unchanged (Figures 5A–F). The decrease of miR-29c and miR-98 was abolished by the reduction in H3K27Me3 abundance, whereas the expression levels of the let-7 family and the other miR-29 family members remained unchanged (Figure 5). Notably, miR-98 expression level returned to baseline, while miR-29c expression increased to above baseline by the reduction in H3K27Me3 abundance (Figures 5C,F), suggesting that their expression is (in part) epigenetically regulated by H3K27Me3. To validate that H3K27Me3 abundance is a determinant of miR-29c and miR-98 transcription, we performed chromatin immunoprecipitation of H3K27Me3. Detectable levels of DNA encoding for miR-29c, and miR-98 were present in all H3K27Me3 ChIP samples. After TGFβ1 stimulation, the abundance of DNA encoding for miR-29c increased in HUVEC, whereas no differences in the abundance of DNA encoding for miR-98 were observed. These data imply that miR-29c expression is regulated in part by H3K27Me3 abundance (Figures 5G,H).

MiR-29c reportedly modulates collagen expression in ECs (33). Indeed, the depletion of miR-29c in HUVEC with low H3K27Me3 abundance restored collagen expression without affecting mesenchymal gene expression, except for ACTA2 which increased 2.0-fold as compared to H3K27Me3-deficient HUVEC (Supplementary Figure S2). Specifically, the stimulation of shSCR HUVEC with TGFβ1 increased the expression of COL1A1 and COL3A1 on gene (5.4-fold, p = 0.0011 and 6.0-fold, p = 0.0017, respectively) and protein levels (3.7-fold, p ≤ 0.0001 and 9.4-fold, p ≤ 0.0001 fold respectively) (data not shown). This increase was blunted in H3K27Me3-deficient HUVEC (Figure 6), whereas the simultaneous blunting of miR-29c expression partly restored COL1A1 expression again by 2.0-fold (gene expression), although this failed to reach significance and collagen I expression by 1.6-fold and collagen III expression by 2.5-fold compared to HUVEC only deficient in H3K27Me3 (Figures 6A–D). These data imply that collagen I and collagen III expression is regulated by miR-29c during TGFβ1 induced EndMT.

To translate our findings to CVD pathology, we investigated if increased H3K27Me3 abundance affects the expression of the collagen-targeting miR-29c in vivo using an EndMT-prone model of pulmonary arterial hypertension (PAH). PAH is a disease characterized by increased right ventricular fibrosis consisting predominantly of interstitial and perivascular fibrosis (10, 39, 40). In this model, the development of PAH is observed as an increased mean pulmonary arterial pressure (mPAP) in PAH animals (31.17 ± 7.31, p = 0.0130), compared to sham control animals (21.60 ± 2.51) and an increased pulmonary vascular resistance (PVR, 0.303 ± 0.156 PAH and 0.083 ± 0.017 sham, p = 0.0159). As a consequence of increased mPAP and PVR, the right ventricle (RV) adapts to the increased afterload, eventually resulting in fibrosis (remodelling) and RV failure as indicated by increased RV weight (418.3 ± 99.89 PAH and 195.0 ± 22.61 sham, p = 0.0022 respectively) and decreased RV output (110.0 ± 44.22 PAH and 176.3 ± 32.26 sham, p = 0.0111) (Table 3). As the onset of cardiac fibrosis during PAH is predominantly perivascular (10, 39, 40), we investigated EndMT in the perivascular areas of the right ventricle. Indeed, in PAH rats an increase in the occurrence of perivascular fibrosis is observed as compared to its occurrence in sham control animals (61% and 34%, respectively, Table 3, Figures 7A,B). Interestingly, in the perivascular areas, H3K27Me3 abundance increased 1.6-fold in ECs of PAH rats compared to sham-treated rats (Figures 7C,D) which associated with an increased percentage of CD31⁺ SM22α⁺ cells (14.31% PAH and 3.94% sham, p = 0.0002, Figures 7C–E), indicative of EndMT. Additionally, increased H3K27Me3 abundance associated with a decreased expression of miR-29c in PAH rats compared to sham-treated rats (3.9-fold, Figure 7F).

These data imply that an increased abundance of H3K27Me3 is associated with a decreased expression of miR-29c and an increase in perivascular fibrosis in vivo.