Selective inhibition of stromal mechanosensing suppresses cardiac fibrosis

Abstract

Matrix-derived biophysical cues are known to regulate the activation of fibroblasts and their subsequent transdifferentiation into myofibroblasts^1-6, but whether modulation of these signals can suppress fibrosis in intact tissues remains unclear, particularly in the cardiovascular system^7-10. Here, we demonstrate across multiple scales that inhibition of matrix mechanosensing in persistently activated cardiac fibroblasts potentiates—in concert with soluble regulators of the TGFβ pathway—a robust transcriptomic, morphological, and metabolic shift towards quiescence. By conducting a meta-analysis of public human and mouse single-cell sequencing datasets, we identify the focal adhesion-associated tyrosine kinase SRC as a fibroblast-enriched mechanosensor which can be targeted selectively in stromal cells to mimic the effects of matrix softening in vivo. Pharmacological inhibition of SRC by saracatinib, coupled with TGFβ suppression, induces synergistic repression of key pro-fibrotic gene programs in fibroblasts, characterized by a marked inhibition of the MRTF-SRF pathway which is not seen with either drug alone. Importantly, the dual treatment alleviates contractile dysfunction in fibrotic engineered heart tissues and in a mouse model of heart failure. Our findings point to joint inhibition of SRC-mediated stromal mechanosensing and TGFβ signalling as a potential mechanotherapeutic strategy for treating cardiovascular fibrosis.

Fibrosis is an integral component of a vast range of cardiac pathological conditions, from inherited cardiomyopathies to ischemic heart disease^11-13. Fibrosis can occur in many distinct forms (e.g., interstitial, perivascular, or replacement fibrosis)¹⁴ depending on the molecular or physiological stressor, but one common denominator of most, if not all fibrotic hearts, is the mechanical stiffening that results from excessive deposition of extracellular matrix (ECM)^15,16. While this stiffening is a direct consequence of fibrosis, it is intriguingly also a potent activation signal for cardiac fibroblasts (CFs) that triggers their transdifferentiation into myofibroblasts (MyoFbs)^1,3,5,17, the key drivers of fibrotic remodelling. This inherent mechanosensitivity is driven by several key pathways including TGFβ^18,19 and YAP/TAZ^20,21, effectively creating a vicious cycle in which the stiffened microenvironment promotes further CF mechano-activation and fibrogenesis², even after the initial pro-fibrotic (inflammatory) signals subside²². Whether breaking this feedback loop can potentiate the deactivation of MyoFbs in vivo remains to be seen^7-9 and may hold the key to developing effective therapies against fibrosis across a broad range of diseases^4,10.

Suppression or reversal of fibroblast activation and related cell types by mechanical ‘unloading’ has been demonstrated previously in vitro^6,23-25. Culture on soft substrates⁶, for example, was shown to reduce the fraction of αSMA (ACTA2)-expressing cells in the early transient stages of aortic valvular fibroblast activation. The effects of soft matrix, however, taper off after long durations of stimulation (e.g., prolonged activation on rigid substrates), suggesting a certain window exists beyond which activated fibroblasts may become either more resistant to mechanically-induced reversal or more prone to re-activation^6,23,26. Such findings are consistent with the notion of a ‘mechanical memory’²⁷, indicating that additional perturbations (e.g., by soluble drugs⁸) might be required to achieve effective fibroblast state modulation. Importantly, mechanical regulation of fibroblast states has yet to be demonstrated convincingly in vivo, which is compounded by the lack of (myo)fibroblast-specific druggable targets, thus presenting major hurdles to clinical translation.

Here, we demonstrate that a combination of soluble (biochemical) and insoluble (mechanical) signals synergistically regulates cardiac stromal cell states to suppress fibrosis at the tissue level. Using phototunable biomaterials and CFs derived from fluorescent reporter induced pluripotent stem cells (iPSC-CFs), we show that dynamic matrix softening coupled with inhibition of the TGFβ pathway induces a phenotypic reversion of persistently activated CFs towards quiescence. Importantly, by conducting a meta-analysis of public single-cell RNA sequencing (scRNA-seq) datasets, we identify a mechanosensor that is activated and uniquely enriched in expression in stromal cells of the fibrotic heart that can be targeted to mimic the effects of matrix softening in vivo. We demonstrate the efficacy of our chemo-mechanical perturbations in engineered tissues and in mouse hearts, providing as proof-of-concept a new therapeutic strategy against fibrosis.

A mechanical feedback loop in fibrosis

We first sought to assess whether human iPSC-CFs, like primary CFs³, are inherently sensitive to microenvironment mechanics (Fig. 1A) by culturing them on: (i) hydrogels of varying stiffness (elastic modulus, E, in units of kPa) (Fig. 1B), (ii) rigid tissue culture plastic (~GPa) with or without soluble inhibitors of mechanotransduction (Fig. 1C), and (iii) rigid plastic at varying cell densities (Extended Data Figs. 1A-C). After ~10 passages, iPSC-CFs cultured on rigid substrates corresponding to the stiffness range of fibrotic hearts (>20 kPa) spontaneously transdifferentiated into αSMA+ MyoFbs with prominent stress fibres (Figs. 1D and Extended Data Figs. 1D-F). This spontaneous activation was suppressed on soft gels mimicking the compliance of embryonic hearts (0.1-3 kPa)^28,29, or with ‘mechano-suppressor’ drugs, without significantly altering cell proliferation (Extended Data Fig. 1E). At all stiffness ranges, TGFβ stimulation resulted in significantly enhanced transdifferentiation³⁰, whereas SB431542 nearly abolished it, indicating an expected dominant response to modulation of TGFβ signalling³¹. Importantly, immunostaining for YAP, a major mechanosensitive transcription factor (TF), revealed a broad correlation between its nuclear localization and the fraction of αSMA+ cells (Figs. 1E,F), both with an inflection point occurring at ~12 kPa. Such correlations were also evident in CFs cultured on plastic at varying cell densities: spontaneous transdifferentiation and nuclear YAP1 were both suppressed at high cell densities, consistent with reduced cell contractility and cytoskeletal tension³² (Extended Data Figs. 1A-C). Interestingly, nuclear localization of YAP was more responsive to mechanical signals (stiffness and density) than to TGFβ stimulation or inhibition.

Figure 1. CF sensitivity to matrix mechanics underlies a pro-fibrotic feedback loop.

(A) iPSC-CF differentiation protocol and experimental design. Representative immunofluorescence images of αSMA and YAP1 in iPSC-CFs cultured for ~10 passages on (B) hydrogels of varying stiffness, or on (C) rigid plastic with or without various ‘mechanosuppressor’ drugs. Scale bar = 100 (top) and 10 μm (bottom row). “N/C” = nuclear/cytoplasmic ratio. (D-E) % αSMA+ cells (threshold: >5X baseline, n = 3 biological replicates/group) and YAP1 N/C versus stiffness. For 2, 8, 16, 64 kPa, and plastic: DMSO (n = 24, 22, 18, 13, 11), blebbistatin (n = 14, 15, 16, 13, 11), Y27632 (n = 14, 19, 15, 24, 17), verteporfin (n = 14, 12, 15, 14, 15), SB431542 (n = 17, 22, 13, 21, 20), and TGFβ (n = 14, 13, 14, 13, 17 cells, respectively). Ordinary one-way ANOVA with Dunnett’s test (*’s) versus DMSO on plastic, and two-sided Student’s t-test (#’s). (F) % αSMA+ cells versus YAP1 N/C (equivalent n-numbers as D-E). (G) Left: UMAP of iPSC-CFs stimulated by TGFβ for 4d on plastic. Right: iPSC-CF data mapped to a reference human heart dataset³³ of control and DCM CFs (only “pathogenic variant-negative” samples were analysed to minimize confounding effects). (H) Feature plots of iPSC-CFs (left) and in vivo human CFs (right) displaying module scores for major fibrosis-associated genes. (I) Gene Ontology analysis for the largest MyoFb cluster, “MyoFb1”. Fisher’s exact test (Benjamini-Hochberg adjustment; one-sided). (J) Module scores for FSP1, TCF21, and PDGFRA. (K) Trajectories inferred by Slingshot, with “qCF1” as the cluster of origin. (L) Violin plots of Modules 1 & 2 (H & J), in qCF1 versus MyoFb1 (top) and Control versus DCM CFs (bottom). (M) Heatmap of differentially expressed fibrosis-related genes. (N) Cartoon illustrating the dual treatment strategy. Data are mean ± SEM.

To confirm that these cells were reliable models of bona fide cardiac MyoFbs, we performed scRNA-seq on the TGFβ-stimulated iPSC-CFs (on rigid plastic) and mapped the data to: (i) a reference human foetal tissue dataset of heart, liver, lung, kidney, and skin fibroblasts (Extended Data Figs. 2A-F), and (ii) a reference adult human heart dataset of healthy and dilated cardiomyopathy (DCM) patient samples³³ (Figs. 1G-L). The iPSC-CFs, when compared to fibroblasts of various foetal tissues, were most similar to primary cardiac fibroblasts, with high expression of TCF21 and PDGFRA (Extended Data Fig. 2A-C). More importantly, mapping the data to an adult human heart dataset showed several populations that clustered with healthy control CFs, and others that clustered with the diseased DCM CFs (Fig. 1G). The latter group (the most prominent of which we termed “MyoFb1”) expressed markedly high levels of major fibrosis-associated genes including COL1A1, POSTN, FN1, LOXL1, FAP, and MEOX1, similar to the DCM fibroblasts (Figs. 1H,L). By contrast, cells that mapped to the healthy control CFs expressed relatively higher levels of genes expressed abundantly in ‘ground-state’ CFs, including TCF21, FSP1 (S100A4), and PDGFRA (Figs. 1J,L). Based on their expression profiles, these non-activated CFs were deemed to be quiescent and were named “qCF1”. Setting qCF1 as the cluster of origin for Slingshot-based lineage trajectory inference yielded ‘outward’-pointing trajectories (Fig. 1K), with the quiescent CFs seemingly giving rise to several activated CF populations (protomyofibroblast-like intermediates) and MyoFb-like cells, characterized by progressive upregulation of pro-fibrotic genes (Fig. 1M and Extended Data Figs. 2G,H). These genes included downstream targets of YAP (e.g., ANKRD1), and the top enriched Gene Ontology terms for the most prominent MyoFb1 cluster included those related to ECM, cell adhesion/migration, and mechanotransduction (Figs. 1I and Extended Data Fig. 2H). Taken together with the mechanobiological perturbation experiments (Figs. 1A-F and Extended Data Figs. 1A-F), our findings pointed to a characteristic positive feedback loop in which fibrotic stiffening promotes further CF activation and, in turn, even more fibrogenesis (Fig. 1N). We therefore sought to assess whether breaking this mechanical feedback loop, in combination with inhibition of soluble signals that initiate the response (e.g., TGFβ), could suppress fibrosis synergistically.

Chemo-mechanical control of MyoFb states

To test whether matrix mechanics can regulate MyoFb states, we generated iPSC-CFs from a fluorescent reporter iPSC line with TAGLN (SM22α)-CFP, one of the major genes known to be upregulated in MyoFbs³⁴. The reporter iPSC-CFs were pre-stimulated with TGFβ for 4 days, and the resulting CFP+ MyoFb-like cells were seeded onto a photoresponsive hydrogel that undergoes rapid softening from ~28 kPa (fibrotic range) to ~8 kPa (normal range) upon exposure to 365 nm light³⁵ (Fig. 2A). Time-lapse imaging was performed to observe cell responses to dynamic matrix softening, either with or without a TGFβ inhibitor (5 μM SB431542, “TGFβi”) (Figs. 2B,C and Extended Data Figs. 3A,B). Surprisingly, softening in the presence of TGFβi resulted in a gradual loss of CFP signal (~24 hrs to reach 50% of baseline) with rapid changes in cell morphology (e.g., spindle-like elongation and reduction in cell size). However, such changes were less apparent with matrix softening alone (>38 hrs to 50% baseline) and were absent with TGFβi alone. Further analysis by immunofluorescence showed that soft matrix plus TGFβi induced a dramatic reduction in αSMA expression, loss of αSMA-decorated stress fibres, and a significant decrease in nuclear YAP1 that were not seen with TGFβi alone or soft matrix alone (Figs. 2D,E).

Figure 2. Dynamic matrix softening potentiates the reversal of MyoFbs when coupled with acute TGFβ inhibition.

(A) Schematic of the dynamically softening hydrogel system. R = Me; oNB = ortho-nitrobenzyl linker. (B) Representative time-lapse images of SM22α-CFP reporter MyoFbs subjected to dynamic softening, with or without a TGFβ inhibitor (“TGFβi”; 5μM SB431542). Scale bar = 100 μm. (C) Normalized 2D cell projected area, aspect ratio, and CFP intensity upon treatment: Stiff control, Stiff+TGFβi, Softened, or Softened+TGFβi (n = 5, 6, 6, 5 cells, respectively; additional cells analysed in Extended Data Fig. 3A,B). (D) Representative immunofluorescence images of YAP1, αSMA, and f-actin after 48 hrs. Scale bar = 50 μm. “N/C” = nuclear/cytoplasmic ratio. (E) Quantitation of: (i) cell area (from left to right: n = 50, 184, 388, 145, 212 cells), (ii) aspect ratio (n = 16, 187, 125, 60, 85 cells), (iii) % αSMA+ cells (n = 24, 49, 41, 22, 30 cells), and (iv) YAP1 N/C ratio (n = 24, 11, 20, 22, 22 cells)). Data from two separate experiments were normalized and pooled. (F) Top: module score for major MyoFb contractility genes and YAP/TAZ targets. Bottom: integrated UMAP plot with all treatment groups and cluster labels. (G) Separate UMAP plots (top) and corresponding cell density plots (bottom) for “Stiff (Ctrl)” versus “Stiff→Soft+TGFβi”. Dotted outlines from the density plot highlight the change in the size of the qCF1 cluster. (H) Bar graph illustrating the broad population re-distribution from MyoFb toward CF clusters with Stiff→Soft+TGFβi. Actively proliferating clusters CF6-8 were excluded from the graph. Actual cell numbers are shown for each bar. (I) SM22α expression in the MyoFb1 and qCF1 clusters upon Stiff→Soft+TGFβi treatment. P-values were calculated based on ordinary one-way ANOVA with Dunnett’s test (*’s) versus Undiff CF, and two-sided Student’s t-test (#’s). Data are mean ± SEM.

Analysis by scRNA-seq confirmed these trends at the transcriptomic level (Fig. 2F) and further revealed that soft matrix plus TGFβi induces a ~40% population re-distribution towards the ‘quiescent’ qCF1 cluster in the centre of the UMAP (Figs. 2G,H and Extended Data Figs. 3C). With the combination treatment, all identified MyoFb clusters shrank in terms of the % of the total population, whereas the CF clusters correspondingly increased in size, most notably the qCF1 cluster which increased ~2-fold in cell number. Such transitions occurred with limited changes in cell cycle distribution (Extended Data Figs. 3D-G) and were further supported by RNA velocity, which revealed cell trajectories pointing ‘inward’ toward the qCF1 cluster in the “Stiff→Soft+TGFβi” group, but not in the “Stiff (Ctrl)” sample (Extended Data Fig. 3H). Importantly, the expression of SM22α (TAGLN) in the MyoFb clusters upon “Stiff→Soft+TGFβi” were reduced to levels almost identical to those of control qCF1 cells (Fig. 2I), consistent with our time-lapse data (Figs. 2A-C).

To investigate which TFs are responsible for driving the population shifts observed with the chemo-mechanical perturbations, we performed TF activity fingerprinting by DoRothEA (Fig. 3A and Extended Data Fig. 4A). The top five variable TFs that were predicted to be: (i) most active in the MyoFb1 cluster relative to the qCF1 cluster, and (ii) simultaneously most sharply deactivated upon “Stiff→Soft+TGFβi” relative to the “Stiff (Ctrl)”—were members of the TEAD (co-regulators of YAP/TAZ) and SMAD families (TGFβ effectors), consistent with our dual targeting approach (Figs. 3A,B and Extended Data Fig. 4B). 24 TFs were uniquely activated or deactivated in the combination treatment versus soft matrix alone or TGFβi alone (Fig. 3C). Among them, the most strongly deactivated TF was serum response factor (SRF), a master regulator of smooth muscle cell and MyoFb differentiation states. Consistent with TF regulon analysis, SRF co-factors (e.g., myocardin-related transcription factor A; MRTF-A) as well as major downstream targets, including SM22α, were broadly downregulated in the combination group, with inconsistent effects seen in either soft matrix alone, or TGFβi only groups (Fig. 3D).

Figure 3. Combination treatment suppresses major fibrosis gene programs with synergistic inhibition of MRTF-A/SRF.

(A) Heatmap of the top 25 TFs whose activity levels were most variable between control versus “Stiff→Soft+TGFβi” in the qCF1 and MyoFb1 clusters, based on regulon analysis. The top-ranked TFs are highlighted in grey. (B) Violin plot of representative genes downstream of YAP/TAZ-TEAD (top) and TGFβ-SMAD2/3/4 (bottom). (C) Venn diagram of unique and shared TFs with variable activity versus Stiff. Right: The top 10 TFs uniquely activated or de-activated in the combination group. The #1 ranked TF, SRF, is highlighted (arrowhead). (D) Dot plot of SRF co-factors and targets. (E) Immunoblots of αSMA, collagen-I, periostin, and phospho-YAP1 (repeated n = 4, 3, 4, 3 times, respectively, with similar results). Images are from three blots from the same samples, processed under identical conditions. (F) Principal component analysis of the anti-MRTF-A CoIP-MS data. (G) Venn diagram of differential MRTF-A interactors identified. Enriched and depleted interactors are shown in black and red font, respectively. (H) Volcano plot of “MyoFb (Stiff Ctrl)” versus “Stiff→Soft+TGFβi”. The top-ranked enriched protein in the MyoFb (Stiff Ctrl) group, SORBS2, is highlighted (blue). (I) Left: Representative images and quantitation of MRTF-A N/C (top) and normalized total nuclear intensity of SORBS2 (bottom) upon siCTRL and siSORBS2 treatment (n = 15 and 12 cells, respectively). Scale bar = 50 μm. (J-L) Representative FRAP time-course images (J) and corresponding quantitation of: (K) normalized MRTF-A-tdTomato intensity over time, and (L) % recovery after 4 min. Coloured circles indicate bleached regions: blue = cytoplasmic (n = 14, 19); red = nuclear (n = 8, 16 cells for siCTRL and siSORBS2, respectively). Scale bar = 10 μm. P-values were calculated based on ordinary one-way ANOVA with Dunnett’s test (*’s) versus Stiff, and two-sided Student’s t-test (#’s). Data are mean ± SEM.

Synergistic effects were also evident at the protein level. Immunoblots and immunofluorescence staining revealed reduced levels of fibrosis-associated proteins including αSMA, collagen-I, and periostin (Fig. 3E and Extended Data Fig. 4C), as well as an increase in YAP phosphorylation at S127, consistent with reduced YAP-TEAD activity inferred by DoRothEA (Figs. 3A,B and Extended Data Fig. 4B). At the functional level, gel contraction assays showed that MyoFbs ‘primed’ on soft gels with TGFβi for 4d were far less contractile than those primed on soft gels without TGFβi, or those treated with TGFβi only (Extended Data Fig. 4D). Furthermore, the combination treatment induced broad repression of ECM genes and crosslinkers, and simultaneous upregulation of ECM degrading enzymes, which were not seen with either soft matrix alone or TGFβi alone (Extended Data Fig. 4E).

To better understand how the combination treatment synergistically inhibits the MRTF-SRF pathway in activated MyoFbs (Figs. 3C,D), we examined MRTF-A localization in response to (i) TGFβ perturbations in a soft matrix background, and (ii) culture on stiff matrix in the presence of TGFβi (Extended Data Figs. 4F-H). Either TGFβ alone or stiff matrix alone was sufficient to induce nuclear localization of MRTF-A, suggesting a redundancy in the upstream signals that promote its activation. Co-immunoprecipitation mass spectrometry (CoIP-MS) analysis supported these findings, showing that the enriched MRTF-A interactors in activated MyoFbs were mostly nuclear proteins, whereas those in undifferentiated CFs or Soft+TGFβi-treated MyoFbs were more commonly found in the cytoplasm (Fig. 3F and Extended Data Fig. 4I). Importantly, CoIP-MS identified a previously unreported interaction between MRTF-A and SORBS2 (ArgBP2), which was highly enriched in the MyoFb control but uniquely depleted in the Soft+TGFβi-treated cells (Figs. 3G,H). Immunofluorescence confirmed the co-localization of SORBS2 and MRTF-A in the nucleus and cytoplasm under the various drug treatments (Extended Data Fig. 4J,K). Interestingly, knockdown of SORBS2 by siRNA resulted in more cytoplasmic MRTF-A (Fig. 3I) and simultaneously increased the mobility of MRTF-A specifically in the nucleus (Figs. 3J-L). Taken together, the results suggested that Soft+TGFβi uniquely disrupts SORBS2 interactions with MRTF-A in the MyoFb nucleus, thereby increasing its mobility and promoting its exit to the cytoplasm.

SRC is a stroma-enriched mechanosensor

We next asked whether we could coerce MyoFbs to ‘perceive’ a stiff fibrotic environment as soft in vivo, and whether we could achieve this in a cell type-selective manner. Rather than attempting to directly alter the mechanics of the myocardium (e.g., ECM degradation in situ), we sought to identify cellular mechanosensors that could be targeted pharmacologically to mimic the effects of matrix softening. To achieve this, we performed meta-analyses of six independent scRNA-seq datasets, including those of healthy human adult^36,37 and foetal hearts, diseased human hearts (dilated cardiomyopathy; DCM)³³, and adult mouse hearts^38,39. Our analysis revealed that expression of SRC, a focal adhesion-associated signal transducer upstream of YAP/TAZ⁴⁰ (Fig. 4A), is highly enriched in stromal populations (most notably in CFs and pericytes) relative to other cell types of the heart (Figs. 4B,C). Notably, this enrichment pattern was unique to SRC compared to a broad panel of well-established mechanosensors (Extended Data Figs. 5A-G). Baseline SRC expression did not seem to increase in fibrotic DCM (Fig. 4C), but its activation state—as assessed by phosphorylation at Y416—increased dramatically (~8-fold in 6 weeks) in parallel with progressive fibrosis in a transverse aortic constriction (TAC) model of heart failure (Figs. D-F). The association between CF mechano-activation and fibrosis was further supported by a published proteomics dataset of human hypertrophic cardiomyopathy patient tissue samples⁴¹, which revealed a significant upregulation of fibrotic ECM, a decrease in YAP phosphorylation at S127 (i.e., increased nuclear YAP), and elevated expression of YAP/TAZ target genes (Fig. 4G).

Figure 4. SRC is a stroma-enriched mechanosensor that can be targeted pharmacologically to mimic matrix softening in vivo.

(A) Schematic illustration of SRC-mediated mechano-signalling. (B) Left: UMAP plot of the human Heart Cell Atlas³⁶. Right: Nebulosa-generated UMAP feature plot showing enriched SRC expression in stromal cells (CFs and pericytes). Right inset: pie chart of % of SRC+ cells. (C) Dot plot of SRC expression in common cardiac cell types identified in scRNA-seq datasets of: (i) normal adult human hearts³⁶, (ii) human foetal heart dataset from this study, (iii) normal and dilated cardiomyopathy patient hearts³³, and (iv) adult mouse hearts^38,39 (right). (D) Masson’s trichrome-stained sections showing progressive fibrosis in a TAC model of pressure-overload induced hypertrophy and heart failure: Sham (n = 7), TAC at 4 (n = 5) and 6 weeks (n = 6 animals). (E) Immunoblots of SRC phosphorylation at Y416 and αSMA expression in TAC hearts after 4 and 6 weeks. Blots are representative of three independently performed experiments with similar results. Images are from two blots from the same samples, processed in parallel. (F) Analysis of a public proteomics dataset⁴¹ of human hypertrophic cardiomyopathy tissue samples (from septal myectomy) showing fibrotic ECM, phosphorylated YAP, downstream YAP target genes, and ‘housekeeping’ proteins. (G) Strategy for virtual docking screen of ~10,000 compounds to identify small molecule inhibitors of SRC. (H) Front and side views of the docking (“starting ensemble”) and after MD simulation (“end ensemble”) for the top 5 candidate compounds after filtering. (I) Total interaction energy (top) and root mean square deviation (RMSD; bottom) for the top 5 candidate compounds. P-values were calculated based on ordinary one-way ANOVA with Dunnett’s test (*’s) versus Sham, and two-sided Student’s t-test (#’s). Data are mean ± SEM.

Given SRC’s unique expression patterns across cardiac cell types and its activation in fibrotic hearts, we next sought to identify small molecule inhibitors of SRC by performing a virtual docking screen of >10,000 compounds (Fig. 4H). Molecular dynamics simulations performed on the top 5 compounds revealed saracatinib (AZD-0530) to be the most promising candidate, capable of accessing SRC’s binding pocket with the strongest and most stable interactions (Figs. 4I,J). We therefore investigated whether saracatinib’s inhibition of SRC-driven mechanosensing could recapitulate the effects of matrix softening.

Saracatinib mimics matrix softening

To predict the effects of saracatinib on CFs, we used Connectivity Map (CMap)⁴², which quantifies the similarity between genetic (e.g., knockdown or overexpression) and pharmacological perturbations based on a comprehensive database. Running CMap for saracatinib revealed top similarity scores for knockdown of focal adhesion-associated genes like PXN, as well as knockdown of YAP1, consistent with saracatinib’s predicted inhibition of SRC-mediated YAP/TAZ mechanotransduction (Fig. 4A and Extended Data Fig. 6A). To confirm this experimentally, we performed immunofluorescence of MyoFbs treated with saracatinib (“SAR”) either with or without two different inhibitors of the TGFβ pathway SB431542 (“SB”) and pirfenidone (“PFD”) (Extended Data Figs. 6B-G). MyoFbs pre-stimulated for short (2 days) and long (7 days) durations on rigid plastic were examined in parallel to test whether the drugs would be effective even after prolonged durations of activation. At both 2d and 7d time points, the combination treatments (“SAR+SB” and “SAR+PFD”) resulted in significant decreases in the percentage of αSMA+ cells, whereas single drug treatments had limited effects (Extended Data Fig. 6D). Saracatinib, whether alone or in combination with TGFβi, led to morphological remodelling similar to those seen with matrix softening (Figs. 2C,E, and Extended Data Figs. 6E, H-J). Regardless of TGFβi, saracatinib also reduced nuclear YAP (Extended Data Fig. 6F) and increased phosphorylated YAP (Extended Data Figs. 6K,L), in line with CMap predictions. Importantly, nuclear localization of MRTF-A also simultaneously decreased only in the combination groups (Extended Data Fig. 6G), consistent with our scRNA-seq (Figs. 3C,D) and soft hydrogel experiments (Figs. 3G-L and Extended Data Figs. 4F-H). These trends were further supported by CoIP-MS on MRTF-A interactors: SAR+TGFβi, much like Soft+TGFβi, led to reduced fractions of nuclear MRTF-A interacting proteins, as well as markedly decreased SORBS2 protein levels (Extended Data Figs. 6L-P). In addition to these changes, SAR+TGFβi also reversed both mitochondrial respiration and glycolysis to levels comparable to those of unstimulated quiescent CFs (Extended Data Fig. 7A). Such effects of SAR on MyoFb states and SRC activity were seen without any negative consequences on the viability or function of CMs and ECs, presumably due to their low baseline expression of SRC (Extended Data Figs. 7B-H).

Dual mechanotherapy suppresses fibrosis

We next sought to assess the efficacy of the drug combination in a more physiologically relevant context. The drugs were first tested in engineered heart tissue (EHTs) that had undergone rapid fibrotic stiffening and compaction after TGFβ stimulation (Extended Data Fig. 8A). The combination treatment, after both short (2d) and long durations (7d) of stimulation, resulted in partial recovery in contractile function, de-compaction (‘relaxation’) of the shortened EHTs, and a decrease in tissue stiffness (elastic modulus) (Extended Data Figs. 8B-G). These changes, however, were much more modest with the single-drug treatments, consistent with our 2D experiments.

To assess the anti-fibrotic effects of the combination therapy in vivo, the drug combination was examined in two mouse models with two very distinct forms of cardiac fibrosis: myocardial infarction (MI) and TAC-induced pressure overload. The SAR+PFD combination treatment offered little therapeutic effect in acute MI hearts with high levels of inflammation and cell death, regardless of the timepoint at which the drug treatments were initiated (Extended Data Fig. 9). By contrast, the ‘two-hit’ approach showed dramatic anti-fibrotic and functional benefits in the more chronic TAC model (Fig. 5), even with delayed drug treatments beginning at 4 weeks post-surgery, at which point fibrosis had already progressed to some degree with marked activation of many of the same TFs identified in our in vitro CF dataset (Figs. 5A-B). Progression of both interstitial and perivascular fibrosis in the TAC hearts (~8-fold increase by week 9) was significantly suppressed with SAR+PFD (Figs. 5C,G; orange), whereas each drug alone had much more modest effects. Echocardiography showed mirroring trends in contractile function: the TAC-induced decrease in systolic function was partially rescued by SAR+PFD, but not with the single-drug treatments (Figs. 5D,E and Extended Data Figs. 9J-M). Interestingly, withdrawal of SAR+PFD after 2 weeks resulted in slight decreases in the recovered contractile function and fibrosis, suggesting the therapeutic effects may not be permanent.

Figure 5. Combination ‘mechanotherapy’ suppresses fibrosis and improves contractile function in failing mouse hearts.

(A) Heatmap of the top 20 most variable TFs in a published snRNA-seq dataset of TAC (5 weeks) mouse hearts³⁹, based on DoRothEA. Transcriptional regulators/effectors of the YAP/TAZ (teal), SRF (purple) and TGFβ (pink) pathways are highlighted. (B) Treatment timeline and experimental design. (C) Masson’s trichrome staining (top; scale bar = 200 μm) and M-mode echocardiography scans (bottom) at week 9 post-TAC. Images are representative of experiments repeated three times independently, with similar results. (D-G) Time-course of left ventricular ejection fraction (LVEF (%)) (D); for days 0, 18, 24, and 42, data from three separate experiments were averaged), and endpoint measurements of LVEF (E), left ventricular posterior wall thickness in diastole (LPWd) (F), and normalized fibrosis area (%) (G): Sham, TAC, S+P, S+P withdrawn, S, and P (n = 10, 12, 12, 6, 7, 7 animals, respectively). (I) Top: UMAP plot of TAC and drug-treated hearts (n = 3 hearts pooled per group). Bottom: Module score for key fibrosis genes in CF populations subsetted out from the original data. (J) Individual UMAPs for each experimental group (top) and corresponding density plots (bottom). Arrows indicate a MyoFb-like subcluster (blue) that grows in number after TAC but shrinks with S+P. (K) Cell composition in the five treatment groups. (L) Violin plot of Postn expression. (M) Dot plot showing broad expression changes in major ECM genes and targets of TGFβ/SMAD, MRTF-A/SRF, and YAP/TAZ induced by S+P. (N) Violin plots displaying module scores for the MRTF-A/SRF target gene set (panel M), in human fibroblasts³³ (left) and in TAC mouse CFs. Bars indicate mean. (O) Graphical summary. P-values were calculated based on ordinary one-way ANOVA with Dunnett’s test (*’s) versus Sham, and two-sided Student’s t-test (#’s). Data are mean ± SEM.

Single-nucleus RNA sequencing (snRNA-seq) on the mouse TAC hearts under the different drug treatments (Figs. 5I-N) identified a MyoFb-like subcluster that abundantly expresses fibrosis genes, including Col1a1, Postn, Tnc, Loxl2, Thbs1, and Meox1 (Fig. 5I). This population grew ~3-fold in the TAC hearts, but shrank most sharply with the ‘two-hit’ SAR+PFD treatment (Figs. 5J,K). Consistent with our in vitro studies, SAR+PFD led to simultaneous repression of key matrix genes and downstream targets of the TGFβ/SMAD, MRTF/SRF, and YAP/TAZ pathways (Fig. 5M). Importantly, the increased expression of a broad MRTF/SRF gene set in both human DCM and mouse TAC hearts was markedly downregulated by SAR+PFD (Fig. 5N), whereas the single drug treatments showed much more modest effects. Similar trends were observed with a YAP/TAZ target gene set in CFs, genes commonly associated with aerobic respiration and glycolysis, and in Sorbs2 expression (Extended Data Figs. 10A-D). However, the various drug treatments did not significantly affect the composition of immune cells, their expression of pro-inflammatory cytokines, or immune-CF signals inferred by enriched ligand-receptor pairs (Extended Data Figs. 10E-I). In line with our snRNA-seq analysis, immunoblots revealed progressive activation of SRC as well as increases in POSTN and αSMA in the TAC hearts, all of which were significantly suppressed with the combination treatment (Extended Data Figs. 10J,K). The findings were consistent with our histology, echocardiography, and in vitro scRNA-seq data, supporting the recurring observation that marked attenuation of fibrosis and contractile dysfunction requires inhibition of both the TGFβ pathway and SRC-driven stromal mechanosensing.

Discussion

Matrix-derived biomechanical cues have long been known to regulate fibroblast activation states in vitro⁴, suggesting that modulation of these signals could hold the key to suppressing or perhaps even reversing fibrosis. However, the development of effective therapies based on these findings has been hampered by the lack of fibroblast-specific druggable targets and delivery methods. Here, we demonstrate that selective inhibition of mechanotransduction in stromal cells, combined with inhibition of the TGFβ pathway, triggers a transcriptomic, metabolic, and morphological shift towards quiescence (Fig. 5O). We show across multiple scales—from human iPSC-derived cells and tissues to mouse hearts—that the effects of matrix softening can be approximated in vivo by targeting SRC, a stroma-enriched mechanosensor. Taken together, our studies provide the first proof-of-concept dual ‘mechanotherapy’¹⁰ against myocardial fibrosis.

Our data suggest that the synergy between matrix mechano-signalling and the TGFβ pathway is achieved at least in part by acute deactivation of the MRTFA-SRF pathway (Figs. 3C-D, 5N). Our studies suggest that this synergy could be explained by inhibition of nuclear localization of MRTF-A (Extended Data Fig. 6G), which was observed only in the combination treatments due to redundancy in the upstream signals that promote MRTF-A activity (Extended Data Fig. 4F-H). Importantly, we identified a previously unreported interaction between MRTF-A and SORBS2 in the MyoFb nucleus, the disruption of which by the ‘two-hit’ treatment led to cytoplasmic localization of MRTF-A (Figs. 3G-L, and Extended Data Figs. 4H-K, 6L-P). The precise chemistry and dynamics of direct or indirect binding between MRTF-A and SORBS2, as well as how exactly the drug combination disrupts this interaction, require further investigation. In addition, the involvement of (i) factors that stabilize MRTF/SRF transcript (e.g., by MBNL1)⁴³, (ii) metabolic regulation of the MRTF-SRF pathway (and vice versa), (iii) direct regulation of MRTF-A expression by YAP²⁰, and/or (iv) the role of epigenetic factors (e.g., mechanically sensitive non-coding elements, chromatin architecture^44,45) cannot be ruled out and should be examined in future studies.

The apparent synergy between mechanical and chemical signals showed promising anti-fibrotic effects and functional recovery in mouse TAC hearts (Fig. 5), but was much less effective against MI. This discrepancy points to potential limitations of our combination strategy in treating more acute forms of fibrosis with particularly high levels of cell death and inflammation (e.g., IL-1β^46,47). Our dual therapy thus appears better suited for treating diseases in which fibrotic remodelling occurs over longer timescales. In TAC mice, over long durations of 4–5 weeks, the combination treatment induced a re-distribution of CF populations toward the non-activated state (Figs. 5J,K), much like what we had observed in vitro (Figs. 2G,H). However, whether true reversal of mature MyoFbs was achieved remains unclear, without the use of multicolour lineage-tracing systems that rely on several highly specific cell state markers. Unfortunately, such marker genes are very limited for the many stages of CF activation and transdifferentiation. New approaches to characterize diverse stromal populations without having to rely on such non-specific genes are thus needed to better understand fibroblast dynamics and state transitions in response to anti-fibrotic therapies including our ‘two-hit’ strategy.

Another notable observation from our studies was that the combination treatment did not appear to result in any adverse effects on cardiac structure or function, presumably because non-fibroblast cell types express very low levels of the target protein SRC (Fig. 4B,C), unlike its downstream mechanotransducer YAP⁴⁸ or other well-studied mechanosensors such as focal adhesion kinase (Extended Data Fig. 5A). While additional studies are required to thoroughly assess safety and potential off-target effects, our findings provide a proof-of-concept therapeutic strategy which, in principle, selectively targets stromal cells. This approach of targeting stroma-enriched mechanosensors could potentially be expanded to other organ systems, thus providing a conceptual framework for the development of new anti-fibrotic therapies.

Materials and Methods

Differentiation of iPSC-derived cardiac fibroblasts

Existing induced pluripotent stem cells (iPSCs) lines from the Stanford Cardiovascular Institute Biobank were used in accordance with IRB guidelines (Stem Cell Research Oversight (SCRO) Protocol #308). The study did not involve new patient/donor recruitment. All iPSCs used in the study were thoroughly authenticated (see Reporting Summary) and tested for mycoplasma contamination prior to differentiation. iPSCs were differentiated into CFs as previously described^51,52. Briefly, iPSCs on Matrigel-coated (CB-40234, BD) plates were treated with 6 μM CHIR99021 (S2924, Selleck Chemicals) for 48 hrs, followed by 24 hrs of recovery in RPMI+B27 minus insulin medium. Cells were then treated with 5 μM IWR1 (I0161, Sigma) for 48 hrs to generate cardiac progenitor cells (CPCs). After 24 hrs of recovery in RPMI+B27 minus insulin, the CPCs were re-plated on day 6 at a density of 20,000 cells/cm² in Advanced DMEM/F12 medium (12634028, Gibco) supplemented with GlutaMAX (Invitrogen), 100 μg/ml L-ascorbic acid (A8960, Sigma), 2 μM retinoid acid (R2625, Sigma), 5 μM Y27632 (S1049, Selleck Chemicals) and 1% FBS. On day 7, the medium was replaced with Advanced DMEM/F12 + GlutaMAX medium containing 5 μM CHIR99021 and 2 μM of retinoid acid, followed by 2 days in Advanced DMEM/F12 + GlutaMAX medium to generate epicardial cells (EPCs). On day 11, EPCs were re-plated in Advanced DMEM/GlutaMAX medium containing 2 μM SB431542 (S1067, Selleck Chemicals) at a split ratio of 1:3, and the medium was refreshed every other day thereafter. To differentiate iPSC-EPCs into CFs, EPCs were re-plated onto collagen-I coated plates and treated with 10 μM FGF2 (100–18B, PeproTech) and 10 μM SB431542 in Fibroblast Growth Medium 3 (C-23025, PromoCell) for a minimum of 10 days. Differentiated iPSC-CFs were maintained on either collagen-I coated plastic or collagen-I coated static PDMS hydrogels of varying stiffness (5190, Sigma) for subsequent experiments. 5 ng/ml of TGFβ1 (240-B-010, R&D) was used to activate iPSC-CFs.

Generation and functional characterization of iPSC-derived cardiomyocytes

iPSC-CMs were differentiated as previously described^53,54. Briefly, CPCs were generated using the same protocol described above (days 0-5). After 5 μM IWR-1 treatment (days 3-5), the cells were cultured in RPMI+B27 minus insulin medium for another 2 days, then in fresh RPMI medium with B27 plus insulin (17504044, Gibco). Beating cells were observed on days 9-11. iPSC-CMs were re-plated and cultured in glucose-free RPMI medium (11879020, Gibco) for 48 hrs to achieve >90% purity. The contractility of the iPSC-CMs was measured using the SONY SI8000 Live Cell Imaging System (Sony Biotechnology, San Jose, CA), which enables the quantification of cellular motion in a noninvasive label-free manner. 6-well plates seeded with beating iPSC-CMs were first placed in the CO₂ incubator at 37°C. A minimum of four regions of interest per well were selected for analysis. Focus and light conditions of the phase contrast microscope were automated through the SI8000 software, and a high-speed charge-coupled video camera was used to capture cell activities at high frame rates of up to 150 fps. After image acquisition, the direction and magnitude of pixel displacements were calculated using a motion detection algorithm developed by Sony.

Generation and functional characterization of iPSC-derived endothelial cells

iPSCs were cultured on Matrigel-coated (CB-40234, BD) plates until reaching 80% confluence. The medium was switched to RPMI-B27 without insulin (Life Technologies) with 6 μM CHIR99021 for 2 days and then changed to 2 μM CHIR99021 for another 2 days. During differentiation, from days 4–12, the medium was changed to EGM2 (Lonza) supplemented with vascular endothelial growth factor (VEGF) (50 ng/mL) (PeproTech), bone morphogenetic protein 4 (BMP4) (20 ng/mL), and fibroblast growth factor 2 (FGF2) (20 ng/mL) (PeproTech). On day 12, cells were dissociated using TrypLE for 5 min and sorted using CD144-conjugated magnetic microbeads (Miltenyi Biotec) according to the manufacturer’s instructions. CD144-positive cells were seeded on 0.2% gelatin-coated plates and maintained in EGM2 medium supplemented with 10 μM SB431542. (Selleck Chemicals). After passage 2, iPSC-ECs were cultured in EGM2. The iPSC-ECs were analysed at passage three post-differentiation. EC tube formation assay was performed by first treating iPSC-ECs with test compounds (e.g., verteporfin and saracatinib). After 48 hrs, cells were trypsinised and re-plated at a density of 1x10⁵ cells on Matrigel basement matrix, and tube formation was analysed after 12 hrs. Cumulative tube length was quantified from microscopic images taken from 3 random fields for each condition. Quantification was performed with Wimasis image analysis software.

Dynamically softening biomaterials

Dynamically softening hydrogels were synthesized as previously described³⁵. Briefly, the two multi-arm PEG precursors (4-arm PEG-sulfo-BCN and 8-arm PEG macromers with azide groups connected via ortho-nitrobenzyl (oNB) linkages) were prepared and separately dissolved in PBS and sterile filtered (0.22 μm). Laminin and fibronectin were chemically modified with linear PEG-azide as previously described³⁵. The azide-functionalized laminin and fibronectin dissolved in PBS were added to the PEG-sulfo-BCN solution to achieve a final concentration in the gels of 100 μg/mL each. The PEG-sulfo-BCN/laminin-PEG-azide/fibronectin-PEG-azide solution was then mixed with the azide-bearing PEG macromers, and the resulting solution (200 μL) was pipetted onto azide-functionalized glass bottom 24-well plates (Cellvis). In parallel, 12 mm circular glass coverslips were passivated, washed with ethanol, and dried. The plates were then centrifuged at 2000 g for 15 min at room temperature to ensure that the hydrogel mixture would uniformly coat the well surface as it crosslinked. To ensure complete crosslinking, the plates were subsequently incubated for an additional 30 min at 37°C.

After rinsing with PBS, the hydrogels were seeded with the CFP reporter iPSC-MyoFbs at a density of 25k cells/well in FluoroBrite DMEM supplemented with 10% FBS, 1x L-glutamine, and 1x penicillin/streptomycin, and were allowed to adhere overnight. For dynamic hydrogel softening, stiff control gels (no softening) and oNB-containing gels were exposed to 365 nm light (~450 mW/cm2) for 4 min. Time-lapse images were collected every hour for 2 days on a Zeiss AxioObserver inverted microscope with a motorized scanning stage and environmental stage incubator (acquisition was paused for exposure to 365 nm light to initiate softening). At each image collection time point, a 7x7 tiled region near the centre of the gels was collected at 20X magnification. A custom GPU-accelerated image processing pipeline was used for image stitching, allowing for accurate flatfield correction, high-dimensional image alignment/registration, deconvolution, and background subtraction. The stitched images were cropped to be uniform in dimension using ImageJ, and masks of the objects in the individual channels were obtained. Only cells positive for CFP prior to the softening experiments (exposure to light) were analysed.

Single-cell and single-nucleus RNA sequencing

For single-cell RNA sequencing (scRNA-seq), cells were harvested and prepared following the instructions provided by 10X Genomics (Pleasanton, CA). In brief, cultured cells were dissociated using accutase (STEMCELL Technologies), washed with 1X DPBS (Gibco), strainer filtered, and re-suspended in 0.04% BSA. For single-nucleus RNA sequencing (snRNA-seq) of mouse hearts, snap-frozen apical tissue slices were cut into smaller pieces (< 1 mm³) on a pre-cooled surface on ice using a clean cold razor, and gently homogenized over two rounds using a Dounce homogenizer. The isolated nuclei were then strainer filtered (40 and 70 μm Flowmi) to remove debris and re-suspended in Diluted Nuclei Buffer (10X Genomics). For both scRNA-seq and scRNA-seq, the resulting single-cell/nucleus suspension was then loaded onto the 10X Chromium Controller for GEM generation, barcoding, and library construction following the instructions provided by the manufacturer. Quality control for the resulting libraries was performed with Bioanalyzer (Agilent Bioanalyzer 2100) prior to sequencing on Illumina HiSeq. Raw data was processed using Cell Ranger and subsequent analyses were performed using Seurat v5.1.0. Briefly, FASTQ files containing sequenced reads were mapped to the human reference genome (GRCh38) or mouse reference genome (mm10). Once the Seurat objects were generated for each condition, cells with fewer than 200 or more than 6,000 expressed genes, and a high percentage of mitochondrial genes (>20%) were removed. CFs with abnormally high expression of endothelial genes (<5%) were filtered. Each Seurat object was normalized separately, integrated into a single object, and scaled. Variable genes with an average expression >0.0125 and <3, and dispersion >0.5, were used for down-stream analysis. The top 30 principal components (PCs) were used for dimensionality reduction and clustering by t-SNE or UMAP. DEGs for each cluster were identified based on Wilcoxon rank sum test under the following thresholds: p-value <0.01, log fold-change >0.25, and > 25% cells expressing the gene. Cell cycle phase scores for each cell was calculated based on known canonical G2M and S phase markers. Cell trajectory inference and pseudotime analyses were performed using open source R packages Slingshot⁵⁵ (v1.6.1) and RNA velocity (velocyto.R, v0.6)⁵⁶. Cell density UMAP plots were generated using Nebulosa⁵⁷ (v1.14.0), and Gene Ontology enrichment and gene network analyses (“cnetplot”) were performed using clusterProfiler⁵⁸ (v4.12.6). Cell-cell communication inference was performed using CellChat⁵⁹ (v2.1.2). Transcription factor activity fingerprinting (regulon analysis) was performed using the single-cell adapted version of DoRothEA⁶⁰ (v.1.16.0). TFs of interest (e.g., Smads 2 and 3) were included during the analysis of DoRothEA viper scores. To determine the top TFs with the most variable activities, the TFs were ranked manually based on the largest positive or negative fold-change in the computed viper scores from: (i) a control group versus a given experimental group (e.g., “Stiff” versus “Soft+TGFβi”), and/or (ii) one cell cluster relative to another within the same treatment group (e.g., MyoFb1 versus qCF1 in the “Stiff (Ctrl)” group). All open source R packages were run under default settings provided in the corresponding vignettes.

Immunofluorescence imaging

After various treatments with drugs, gels, siSORBS (s16087, Thermo Scientific), or siCTRL, (#4390843, Thermo Scientific), cultured cells were washed with pre-warmed PBS and subsequently fixed with 4% paraformaldehyde (Fisher) for 15 min. Fixed cells were washed X3 with PBS and then permeabilized with 0.5% Triton-X (Fisher) in PBS for 10 min. Permeabilized cells were then blocked with 5% BSA in PBS for ≥1.5 hrs. Samples were incubated overnight with the following primary antibodies in 0.5% BSA solution, with gentle agitation at 4°C: αSMA (1:200, #ab7817, Abcam), YAP1 (1:500, #NB110-58358, Novus Biologicals), Ki67 (1:2000, #ab15580, Abcam), MRTF-A/MKL1 (1:150, #NBP2-45862, Novus Biologicals), and SORBS/ArgBP2 (1:250, #83189-3-RR, Proteintech). Samples were washed X3 in 0.1% BSA in PBS and incubated with the corresponding secondary antibodies at 1:500 dilution for 1.5 hrs at RT (Alexa Fluor 488, 546 and 647 nm; Invitrogen) along with DAPI nuclear staining (R37606, Thermo) and/or phalloidin (A22287, Thermo) when applicable. Images of adherent cells were taken with an ECHO microscope at 20X and 40X magnification. All images in each experiment were taken under the same imaging conditions and were analysed using ImageJ/FIJI (NIH).

Immunoblotting

Lysis buffer was first prepared on ice by adding 1% protease and phosphatase inhibitor cocktail (78441, Thermo) and 1% β-mercaptoethanol into 1X NuPage LDS buffer (Invitrogen) diluted in RIPA buffer. For cultured CFs and MyoFbs, cells were trypsinised, counted, and centrifuged to create a pellet. After 1X wash (resuspension and pipette mixing) with cold DPBS and a second round of centrifugation, the supernatant was aspirated, and lysis buffer was added to the pellet. Lysates were incubated on ice for 30 min, then at 4°C for 1 hr on a rocker. For snap-frozen tissues, frozen samples were transferred to a dish placed on dry ice and were quickly excised into submillimeter pieces, then immediately suspended in ice-cold lysis buffer. After pipette mixing, lysates were incubated on ice for 30 min, then at 4°C for 1 hr on a rocker. Lysed samples were then heated to 80°C for 10 min and centrifuged at maximum speed for 30 min at 4°C. SDS-PAGE gels were loaded with 5 - 15 μL of lysate per lane (NuPage 4%-12% bis-Tris; Invitrogen). Gel electrophoresis was run for 10 min at 100 V and 1 hr at 160 V. Separated samples were then transferred to a polyvinylidene fluoride membrane using a Bio-Rad Trans-Blot Turbo Transfer System (1704272 and 690BR3244, Bio-Rad). The membrane was blocked with 5% non-fat dry milk in TTBS buffer (Tris-buffered saline, Fisher; with 0.1% Tween-20), washed X3 in TTBS, and incubated with primary antibodies diluted in TBS to final concentrations generally on the order of ∼1 μg/ml at 4°C overnight: αSMA (1:1000, #ab7817, Abcam), Col-I (1:1000, NBP1-30054, Novus Biologicals), POSTN (1:1000, #ab14041, Abcam; 1:500 for all blots except for Extended Data Fig. 6K: #sc-398631, Santa Cruz Biotechnology), YAP1 (1:1000, #NB110-58358, Novus Biologicals for all blots except for Fig. 3E: 1:5000, #66900-1-Ig, Proteintech), phospho-YAP1 (1:1000, #4911, Cell Signaling), SRC (1:5000, #60315-1-Ig, Proteintech), phospho-SRC (1:1000, #2101, Cell Signaling), SORBS/ArgBP2 (1:5000, #83189-3-RR, Proteintech), and GAPDH (1:5000, #MA5-15738-HRP, Thermo). After washing X3 with TTBS, the membrane was incubated with 1:2000 diluted secondary Ab: anti-mouse HRP-conjugated IgG (1:2000, ##7076, Cell Signalling) or anti-rabbit HRP-conjugated IgG (1:2000, ##7074, Cell Signalling), at RT for ≥ 1.5 hrs. The membrane was washed X3 again with TTBS, developed with Clarity Western ECL Substrate (Bio-Rad) for ~5 min at RT, and imaged using the ChemiDoc MP Imaging System (Bio-Rad). For immunoblot source data, see Supplementary Fig. 1.

Affinity purification and peptide preparation

Immunoprecipitation of MRTF-A from primary cardiac fibroblasts was conducted using the Pierce^™ Classic Magnetic IP/Co-IP Kit as specified by the manufacturer’s instructions. Briefly, CFs were sonicated at 4°C for 1 min in lysis buffer containing 1X protease and phosphatase inhibitors to ensure nuclear rupture. Lysates were collected after 10 min centrifugation at 10,000 x g, and the total protein was quantified using a BCA assay. Input protein was collected from each sample, incubated in sample buffer for 10 min at RT and stored for subsequent Western Blot analysis, or stored as an unenriched protein lysate for suspension-trapping and subsequent LC-MS. Protein lysate (500 ug protein/condition in 500 ul reaction volume) was incubated with 5 ul MRTF-A antibody (1:100, #77098, Cell Signaling) or a species-matched control immunoglobulin (normal rabbit IgG, #2729, Cell Signaling) at 4°C overnight with mild agitation. Magnetic beads were pre-swollen in lysis buffer and then used to immunocapture antibodies from lysates for one hour at RT. Antibody-bead conjugates were collected from each sample using the DynaMag^™-2 Magnet. Pellets were gently washed three times in lysis buffer, three times in PBS, and three times in wash buffer (50 mM Tris, 15 0mM NaCl, and 1X protease/phosphatase inhibitor). For on-bead peptide preparation, all wash buffer was removed from the bead-protein conjugate and replaced with 100 mM triethyl ammonium bicarbonate (TEAB), which was mixed head-over-head on an automatic rotor for 10 min at RT twice. The samples were reduced with 10 mM dithiothreitol (DTT) for 5min at 55 °C, then incubated at room temperature for 25 mins on a rotor and alkylated with 20 mM iodoacetamide for 30 minutes at room temperature in the dark. Samples were then acidified to a pH ~1 with 2.6 μl of 27% phosphoric acid, followed by digestion on-bead overnight at 37°C with 600 ng of mass spectrometry grade Trypsin/LysC mix (Promega, USA). The digested peptides were eluted with two 35 μl increments of 0.2% formic acid in water and two more 40 μl increments of 50% acetonitrile with 0.2% formic acid in water. The four eluted fractions were consolidated and further desalted using MonoSpin C18 Solid-Phase Extraction (SPE) columns (GL Sciences) to remove residual iron (III) oxide. Finally, the samples were dried via SpeedVac (ThermoFisher Scientific) and reconstituted with 0.015% n-Dodecyl β-D-maltoside (DDM; Sigma USA) in 0.1% formic acid to a concentration of 3 ng/μl determined by a fluorometric peptide quantitative assay.

Liquid chromatography mass spectrometry

Samples were analyzed on the timsTOF Ultra (Bruker Daltonics, Germany) coupled to a nanoElute 2 (Bruker Daltonics, Germany) with 1 ng sample input. Samples were eluted off a PepSep Ultra C18 column (25 cm length x 75 um ID x 1.5 um particle size, P/N 1893484; Bruker Daltonics, Germany) at 50 °C connected to a fused silica 10 um emitter (Bruker Daltonics, Germany) inside a nanoelectrospray Captive Spray source (Bruker Daltonics, Germany) with a 90 min active gradient (2-24% Solvent B in 72-mins, 24-35% Solvent B in 18 mins, 35-95% Solvent B in 0.10 mins, wash at 95% for 4.90 mins; Solvent B: 0.1% formic acid in acetonitrile, Solvent A: 0.1% formic acid). The TIMS 1/K0 mobility range was 0.75-1.30 Vs/cm², with a ramp time of 120 ms and 100% duty cycle. The source was set to 1600 V, 3.0 l/min dry gas, and 200°C dry temp. diaPASEF windows were created with py_DIA ⁶¹ with a mass range of 350-1250 Da, and a mobility range of 0.60-1.45 1/K0 giving an estimated cycle time of 2.25 s. The collision energy ramped from 20 to 59 eV at 0.60 to 1.60 V·s/cm².

Mass Spectrometry Data Processing and Analysis

The “.d” files acquired in data-independent mode were analyzed using Spectronaut v 19.1 (Biognosys AG, Schlieren, Switzerland) with the directDIA+ feature against the UniProt Homo sapiens protein database. Proteolysis with Trypsin was assumed to be semi-specific, allowing for N-ragged cleavage with up to two missed cleavage sites, with a PSM false discovery rate (FDR) of 1%. Data filtering was based on the Q Value, and global normalization was utilized. Cysteine modified with iodoacetamide was set as a fixed modification. Variable modifications included oxidation of methionine and tryptophan, deamidation of asparagine and glutamine, and phosphorylation of serine, threonine, and tyrosine.

For Venn diagram analyses of enriched or depleted protein-protein interactors, non-specific binding was first filtered out using the IgG control by removing all proteins that were detected in >1 out of the 3 IgG replicates. Uniquely detected proteins were then identified for each experimental group (i.e., proteins with non-zero intensity values in only one group). Proteins with average intensity values at least 1.5x greater in one group than in all other groups were categorized as uniquely enriched. Conversely, those with average intensity values less than 0.5x that of all other groups were categorized as depleted.

Generation of MRTF-A-tdTomato construct and cell transfection

The construct containing MRTF-A coding sequence was synthesized by Twist Biosciences. The gene sequence was codon optimized for synthesis. Restriction enzyme sites were added to each sequence end (NheI on the 3’ and BshTI to the 5’ end) for excision. The MRTF-A sequence was ligated into the linearized tdTomato-N1 plasmid (Addgene #54642). Insertion was confirmed via Sanger sequencing. For transfection, primary CFs (Lonza, #CC-2904)) were plated in 8-well glass-bottom dishes (Ibidi, Cat. #80826) using Fibroblast Growth Medium (FGM3, PromoCell, Cat. No. C-23025) supplemented with TGF-β (PeproTech, Cat. No. 100-21C). After two days of treatment, the CFs were transfected with MRTF-A-tdTomato construct and MRTF-A-targeting siRNA using JetPrime transfection reagent, following the manufacturer’s protocol. Live-cell imaging was performed 24 hours post-transfection using a Zeiss LSM980 microscope equipped with an environmental chamber set to 37°C and 5% CO₂.

Fluorescence recovery after photobleaching (FRAP)

Transfected cells of interest were imaged using a 63X oil objective with 561 nm laser excitation and a detection range of 570–680 nm, capturing a 4-minute time series. After acquiring two initial frames, a selected region of interest (ROI) was bleached using 100% power from the 561 nm laser for 80 iterations. Image analysis was performed in ImageJ, measuring fluorescence changes over time in the bleached ROI and a similarly sized unbleached area. The fluorescence from the unbleached region was used to normalize the fluorescence recovery in the bleached ROI, correcting for photobleaching effects.

In silico molecular screen and molecular dynamics simulations

The 3D protein structure of SRC (PDB ID: 2src)⁶² was retrieved from the protein data bank (PDB). A ligand library was prepared from the Broad Institute Drug Repurposing Hub⁶³. In silico high-throughput molecular screen was performed using Autodock Vina 1.2.3⁶⁴. The top 5% of compounds were assessed for absorption, distribution, metabolism and excretion properties using SwissADME⁶⁵. The top 5 candidates were then selected for molecular dynamics (MD) simulations using GROMACS (v2022.2) using the first docking pose as the initial starting conformation. The energy was minimized using the steepest descent algorithm with a Verlet cutoff in with all atoms being parameterized according to the CHARMM36 (July 2021) force field using the CGenFF program. The protein-drug complexes were solvated explicitly in a periodic box of TIP3P water molecules and the charge was neutralized with sodium chloride. Two rounds of equilibration were run under both isothermal-isochoric and isothermal-isobaric conditions and MD simulations were run for all atoms over 100 ns. The resulting trajectories underwent centering, rotational, as well as transitional fitting.

Seahorse XF assay

Seahorse Mito Stress assay (Agilent, USA) was performed according to the manufacturer’s instructions. Briefly, 20,000 cells were plated onto Matrigel-coated 96-well Seahorse plates. Following drug treatment, the oxygen consumption rates (OCR) and extracellular acidification rates (ECAR) were measured. One hour before the assay, the culture medium was exchanged with Agilent Seahorse XF RPMI Basal media supplemented with 2 mM glutamine, 10 nM glucose, and 1 mM sodium pyruvate. Electron transport chain uncouplers were diluted in the same media and sequentially injected during the measurements at the following final concentrations; oligomycin (2.5 μM), FCCP (1 μM), rotenone (1 μM), and antimycin A (1 μM).

Engineered heart tissue fabrication and analysis

Engineered heart tissues (EHTs) were assembled as previously described⁶⁶. Briefly, tissue casting molds were prepared by placing Teflon spacers (C0002, EHT Technologies) in 24-well culture plates and adding 1.5 mL of a 2% agarose solution in DPBS (14190235, Gibco) to each well. After allowing the agarose solution to solidify, Teflon spacers were removed, and the EHT silicon racks (C0001, EHT Technologies) were placed in the agarose cavity. RPMI 1640 medium (11875119, Thermo Scientific) with B-27 supplement (17504044, Thermo Scientific) and 33 μg/mL of aprotinin (A1153, Sigma-Aldrich) were used as the EHT culture medium. An ECM solution was prepared by mixing 7 mg/mL bovine fibrinogen (F8630, Sigma-Aldrich) and 5% growth factor reduced Matrigel (356231, Corning) with the EHT culture medium. Cell seeding solution was made with a concentration of 2 × 10⁷ cells/mL, consisting of 93% of iPSC-CMs and 7% of iPSC-CFs in the ECM solution. To generate each EHT, 100 μL of the cell seeding solution was mixed with 3 μL of 100 U/mL thrombin (T6634, Sigma-Aldrich) in DPBS and added to the agarose cavity containing the silicon rack. 0.5 mL of the EHT culture medium was added to each well on top of the seeded EHTs and incubated at 37°C for 1 hour. The resulting solidified EHTs were carefully transferred to new 24-well plates containing 2 mL/well of the EHT culture medium.

Transverse aortic constriction (TAC) and myocardial infarction (MI) models

The laboratory animal care program at Stanford University is accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care International (AAALAC International). All experiments were conducted in accordance with NIH and Stanford University policies. All animals were housed in individually ventilated cages (4 animals per cage) located in the same room in the animal facility managed by the Veterinary Service Center (VSC) of Stanford University School of Medicine, under standard housing conditions (ambient temperature ~20–24°C, ~50% humidity, 12h light / 12h dark cycles). TAC and MI surgeries were performed on 8-10-week-old male C56BL/6N mice (n = 180 total). For TAC, anaesthesia was given and maintained throughout the surgery by 1-3% isoflurane 1L O₂/min, with pre-emptive Bupivacaine 1-2 mg/kg (local infiltration). After anaesthesia, the thoracic cavity was opened, and the aorta under the upper left sterna border was ligated with a 27G needle using 6.0 silk thread. The needle was removed after ligation of the aorta, and the thoracic cavity was closed 6.0 prolene suture. For MI, the permanent left anterior descending (LAD) coronary artery occlusion model was used. Pre-emptive analgesia was given by administration of 1-3% isoflurane 1L O₂/min and local infiltration with bupivacaine (1–2 mg/kg). LAD artery ligation was performed after aseptic thoracotomy by suturing the proximal LAD using a 5-0 Ethilon suture. Successful infarction was confirmed through blanching of the anterior myocardium. The chest, muscle layer, and skin were closed using 2–4 interrupted cross-stitches with 4-0 Vicryl absorbable sutures. For both MI and TAC models, controls consisted of sham-operated mice, which underwent the same surgical procedure but without ligation. All mice were allowed to fully recover on a heating pad after the surgeries. Systolic function was assessed by echocardiography 4 days after surgery for MI and 2 weeks after surgery for TAC. For drug administration, mice were selected randomly and orally administered saracatinib (20 mg/kg), and/or pirfenidone (200 mg/kg) in a solvent mixture of ethanol and corn oil, once daily. The sham group received the same volume of corn oil solvent. Echocardiography was performed periodically to monitor changes in cardiac function and/or structure.

Histological staining

Mouse hearts were harvested post-anesthesia and immediately snap-frozen in isopentane chilled in a bath of liquid nitrogen. Snap-frozen hearts were then embedded in Optimal Cutting Temperature (O.C.T.) compound and sliced into 20 μm sections. Tissue sections were subsequently stained with hematoxylin and eosin (H&E) and Masson’s trichrome (Sigma-Aldrich) following standard protocols. Quantification of these histochemical images was performed using ImageJ/FIJI (NIH). Myocardial fibrosis grades (fibrotic area fraction) were analysed from a minimum of ten randomly selected regions per trichrome-stained tissue sample.

Analysis of cardiac function using echocardiography

Echocardiography was performed using the Vevo2100 Ultrasound System (Visual Sonics) equipped with an MS400 transducer, following standard procedures. Hair removal cream (Nair) was used to remove any remaining fur around the left chest area of the mice prior to echocardiography. Mice were lightly anesthetized using 2% isofluorane mixed with 100% O₂ throughout the duration of scanning. For analysis of cardiac systolic function, M-mode images of the parasternal short-axis view at the papillary level were used to measure left ventricular ejection fraction (LVEF), fractional shortening (FS), left ventricle posterior wall thickness; end-diastolic (LVPWd), and left ventricular internal dimension in diastole (LVIDd). Measurements were averaged over at least three cardiac cycles for all functional parameters for analysis.

Statistics and reproducibility

Data are represented as mean values ± SEM. Numbers next to, or above, individual data points/bars indicate P-values which were obtained by two-sided unpaired Student’s t-test for comparisons between two experimental groups (marked also by #’s: #p<0.05, ##p<0.01, ###p<0.001, ####p<0.0001), and by ordinary one-way ANOVA plus Dunnett’s multiple comparison test (two-sided) for experiments with more than three groups (marked also by *’s: *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001). A value of p < 0.05 was considered statistically significant. All statistical analyses were performed using GraphPad Prism v6.0, except Figure 1I and Extended Data Figures 9E and I, for which built-in statistics were directly taken from the R packages used (ClusterProfiler v.4.12.6 and CellChat v2.1.2). All experiments were performed with at least three biological replicates per group and were repeated at least twice independently with similar results, unless otherwise noted in the legend (e.g., scRNA-seq and CoIP-MS). Micrographs, histological images, and echocardiography scans in main Figs. 1B, C; 2D; 4D, and 5C are representative of experiments repeated three times independently, and those in Figs. 2B, 3I, and J twice independently, with similar results. Sample sizes were not predetermined based on statistical analysis; instead, they were estimated based on prior experience, availability of materials and reagents, the specific experimental methods used, and feasibility considerations to ensure statistically significant results. The EHT experiments were repeated four separate times, and data from shared experimental conditions across these four experiments were normalized and pooled. Animal experiments were repeated three times, and data from shared experimental conditions across these three experiments were normalized and pooled. All EHTs and animals were randomly assigned to experimental groups, and all experiments and analyses of data were performed blinded to the investigator.

Extended Data

Extended Data Figure 1. iPSC-CFs undergo spontaneous activation on stiff environments or at low cell densities unless mechanosensing is inhibited.

(A) Representative immunofluorescence images of αSMA (ACTA2) and YAP1 staining in iPSC-CFs cultured for ~10 passages on rigid tissue culture plastic (~GPa) at varying cell densities (initial seeding at 2.5 – 40k cells/well). Images are representative of experiments repeated twice independently with similar results. Top row scale bar = 100 μm; bottom row scale bar = 10 μm. (B) % αSMA+ cells (left) and YAP1 N/C ratio (right) plotted versus cell seeding density: 2.5k, 5k, 10k, 20k, 40k cells/well: % αSMA+ cells (n = 13, 15, 15, 15, 20 randomly selected fields of view, respectively, from n = 3 biological replicates/group) and YAP1 N/C (n = 20, 18, 21, 26, 27 cells, respectively, from n = 3 biological replicates/group). (C) % αSMA+ cells plotted versus YAP1 N/C ratio (n-numbers equivalent to those of panel B). (D) αSMA immunoblot of iPSC-CFs undergoing spontaneous CF-to-MyoFb transition on hydrogels of varying stiffness for ~10 passages. Blots are representative of n = 2 independently performed experiments with similar results. (E) Reported stiffness ranges for embryonic, healthy adult, myocardial infarction (MI), heart failure with preserved ejection fraction (HFpEF), and transverse aortic constriction (TAC) hearts^15,49. Normalized 2D projected cell area (top) and % of Ki67+ cells (bottom) versus stiffness: 2, 8, 16, 64 kPa and rigid plastic. Normalized cell area (n = 21, 18, 22, 13, 11 cells, respectively, from n = 3 biological replicates/group) and Ki67+ (n = 9, 15, 7, 5, 5, 8 randomly selected fields of view, respectively, from n = 3 biological replicates/group). (F) Half-maximal effective stiffness (“ES50”, analogous to EC50 for a drug dose-response, in units of kPa) for each group, obtained from curve-fitting (data in main Fig. 1D, four-parameter dose-response stimulation with variable slope on GraphPad Prism v6.0). ES50 is defined as the gel stiffness at which the half-maximal response is produced, which in this case is the % of αSMA+ cells. ES50 for each group is plotted as individual bars, and the asymptotic standard error obtained from the best-fit curves (“Std. Error” values) are presented as error bars. P-values were calculated based on ordinary one-way ANOVA with Dunnett’s test (*’s) versus 2.5k/well (panel B), versus 2 kPa (panel E), or versus DMSO on plastic (panel E); and the two-sided Student’s t-test (#’s). Data are mean ± SEM (with the exception of panel F, for which individual bars ± error bars represent ES50 ± Std. Error).

Extended Data Figure 2. iPSC-CFs resemble primary human foetal CFs at the transcriptomic level.

(A) UMAP plot of quiescent iPSC-CFs mapped onto a reference dataset of primary fibroblasts (FBs) from various human foetal tissues (heart, lung, skin, kidney, and liver). (B) Feature plots showing expression of TCF21 and PDGFRA in iPSC-CFs and primary CFs. (C) Correlation plot of iPSC-CFs, FBs from various foetal tissues (lung, skin, kidney, liver), and major cardiac cell types (CFs, SMCs, CMs, and ECs). (D-F) Representative feature plots for: (D) pan-fibroblast genes, (E) CM marker genes, and (F) EC marker genes. (G) The 25 most variable genes along pseudotime inferred by Slingshot-based trajectory analysis of TGFβ-activated iPSC-CFs. Notable ECM genes (purple) and SMC/MyoFb-associated genes (orange) are highlighted. (H) Gene network plot (“cnetplot” generated by clusterProfiler) for the MyoFb1 cluster.

Extended Data Figure 3. Combination treatment drives morphological remodelling and population re-distribution at the single-cell transcriptomic level without altering cell cycle.

(A) Additional representative time-lapse images of SM22α (TAGLN)-CFP reporter iPSC-derived MyoFbs subjected to light-induced matrix softening over 38 hrs, either with or without TGFβi. Scale bar = 100 μm. (B) Normalized 2D cell projected area and cell aspect ratio (length/width; L/W) before and 38 hrs after matrix softening. Stiff control (n = 10), TGFβi (n = 12), Softened (n = 11), Softened+ TGFβi (n = 10 cells). Data from the two separate experiments were normalized and pooled. (C) Individual UMAP plots (top row) and corresponding cell density plots (bottom row) for all four treatment groups. (D) UMAP plot generated based on the cell cycle phase of each cell, scored using canonical markers. (E) Cell cycle distribution of the total population in each of the four treatment conditions. (F) Cell cycle distribution by cluster in the four treatment conditions. (G) Representative immunofluorescence images of Ki67 and DAPI-stained MyoFbs. Bottom: quantitation of % Ki67+ cells in the four treatment conditions: Stiff control (n = 11), TGFβi (n = 5), Soft (n = 6), Soft+TGFβi (n = 5 randomly selected fields of view, from n = 3 biological replicates/group). (H) (i) RNA velocity plots of Stiff control (purple) and Soft+TGFβi (orange), and (ii) ‘ΔRNA velocity’ generated by subtracting individual velocity vectors ($\overrightarrow{\text{Soft}+\text{TGF}{\beta }_1}\text{-}\overrightarrow{{\text{St}}_1\text{ff}}$) on the coordinate grid. Images in panels A and G are each representative of experiments repeated twice independently with similar results. P-values were calculated based on ordinary one-way ANOVA with Dunnett’s test (*’s) versus Stiff baseline control; and the two-sided Student’s t-test (#’s). Data are mean ± SEM.

Extended Data Figure 4. Combination treatment downregulates pro-fibrotic pathways and synergistically inhibits MRTF-A retention in the MyoFb nucleus.

(A) Heatmap of the top 25 TFs whose activity levels were most variable across the treatment conditions, for all identified clusters, based on activity fingerprinting by DoRothEA. (B) Violin plots of key downstream target genes of YAP/TAZ-TEAD. (C) Representative αSMA immunofluorescence images (left) and corresponding histogram of anti-αSMA intensity in single MyoFbs. Inset: bar graph of % αSMA+ cells (threshold set at 5-fold increase relative to baseline). Stiff (n = 49), TGFβi (n = 41), Stiff→Soft (n = 22), Stiff→Soft+ TGFβi (n = 30 randomly selected fields of view). Images are representative of experiments repeated n = 3 times independently with similar results. Scale bar = 100 μm. (D) Gel contraction assay of undifferentiated iPSC-CFs and MyoFbs after culturing (‘priming’) for 4 days on soft or stiff substrates, with or without TGFβi treatment. Primed cells were trypsinized and mixed with a commercially available gel solution for the contraction assay. 2D gel areas were measured after 12h (n = 3 biological replicates per group). (E) Dot plot showing the expression of major fibrosis-associated ECM proteins, ECM cross-linkers, and various MMP isoforms (collagenases and gelatinases) in the four groups after 2 days of treatment. (F) Schematic of experimental design illustrating TGFβ perturbations in a soft matrix background (orange), or stiff matrix culture in the presence of TGFβi. (G-H) Representative immunofluorescence images and quantitation of MRTF-A N/C ratio in the five treatment groups: Soft (n = 11), Stiff (n = 12), Stiff+LatB (latrunculin-B; n = 12), Soft+TGFβ (n = 11), Soft+TGFβ+SB (n = 11 cells). (I) Subcellular localization of enriched MRTF-A interactors in the CoIP-MS data. (J-K) Representative immunofluorescence images and quantitation of MRTF-A and SORBS2 co-localization in the nucleus and cytoplasm under the different treatment conditions: Undiff CF (n = 8), MyoFb (n = 37), TGFβi (n = 18), Soft (n = 12), Soft+TGFβi (n = 11 cells). Individual cells are shown as opaque data points in the background in panel K. Images in panels G and J are representative of experiments repeated twice independently with similar results. P-values were calculated based on ordinary one-way ANOVA with Dunnett’s test (*’s) versus Soft control (panel H); and the two-sided Student’s t-test (#’s) versus MyoFb (panel K). Data are mean ± SEM.

Extended Data Figure 5. SRC is unique among major cellular mechanosensors in its enrichment in cardiac stromal cells.

(A) Nebulosa feature plots for a broad range of major mechanosensors, including mechano-responsive TFs (e.g., YAP/TAZ), focal adhesion components (e.g., FAK (PTK2)), mechanically-gated ion channels (e.g., PIEZO2), integrins (e.g., ITGA1), and nuclear envelope proteins (e.g., LMNA) in adult mouse heart. SRC’s unique enrichment in CFs and mural cell populations is highlighted with arrowheads (magenta). (B-F) Representative UMAP plots and Nebulosa feature plots illustrating SRC expression in scRNA-seq datasets of: (B) human adult heart³⁶, (C) FBs isolated from various human fetal tissues, (D) mouse model of myocardial infarction (MI)³⁸, (E) additional mouse data set for TAC hearts³⁹, and (F) TGFβ-simulated iPSC-CFs. (G) Violin plot of SRC expression across all clusters in TGFβ-simulated iPSC-CFs.

Extended Data Figure 6. Saracatinib recapitulates the effects of matrix softening in vitro.

(A) Connectivity Map (CMap⁴²) analysis showing the top 20 genes for which overexpression (OE) or knockdown (KD) results in similar perturbation signatures as saracatinib treatment. (B-C) Schematic of the experiment timeline (B) and representative immunofluorescence images (C) of cells treated with various drug combinations: “SAR” = saracatinib; “SB” = SB431542; “PFD” = pirfenidone; Scale bar = 150 μm; inset = 50 μm. (D) % αSMA-positive cells (defined as ≥ 5X versus baseline control) over the course of the experiment. For days 2, 4, and 9: DMSO (n = 4, 11, 13), SAR (n = 2, 8, 10), SB (n = 3, 7, 8), PFD (n = 2, 8, 6), SAR+SB (n = 3, 6, 8), SAR+PFD (n = 3, 11, 14 randomly selected frames of view, respectively, from n = 3 biological replicates/group). (E-G) Cell aspect ratio (elongation) versus normalized 2D cell area (E), and nuclear/cytoplasmic ratio of (F) YAP1 and (G) MRTF-A at day 9. DMSO (n = 58), SAR (n = 44), SB (n = 41), PFD (n = 38), SAR+SB (n = 31), SAR+PFD (n = 44 cells). (H) Representative images of f-actin and DNA in cells treated with TGFβi, either alone or with three different SRC inhibitors (“DAS” = dasatinib; “BOS” = bosutinib). Scale bar = 150 μm. (I) % cells with actomyosin stress fibres in the different treatment groups: CF, MyoFb, TGFβi, TGFβi+DAS, TGFβi+SAR, TGFβi+BOS (n = 5, 8, 6, 5, 5, 5 randomly selected fields of view, respectively, from n = 3 biological replicates/group). (J) Scatter plot of cell aspect ratio (length/width; L/W) versus 2D area. CF (n = 35), MyoFb (n = 42), TGFβi (n = 69), TGFβi+DAS (n = 25), TGFβi+SAR (n = 35), TGFβi+BOS (n = 77 cells). (K) Immunoblots and corresponding densitometry quantitation of pYAP1, αSMA, collagen-I, and POSTN. Blots are representative of n = 2 independently performed experiments with similar results (for collagen-I and POSTN, data from two additional experiments were normalized and pooled). Images shown are from two blots derived from the same samples from the same experiment, processed in parallel. For groups with n = 2, the error bars indicate the range between the two data points. (L) Subcellular localization of enriched MRTF-A interactors in the CoIP-MS data. (M) PCA analysis of CoIP-MS data from Fig. 3F, now including the SAR and SAR+TGFβi groups. (N) Heatmap of the anti-MRTF-A CoIP-MS data, showing all seven experimental groups (including those from Fig. 3F-H). The top 10 differentially enriched MRTF-A protein interactors in the MyoFb group (purple) are shown on the right. (O) Anti-SORBS2 immunoblot of the CoIP-MS input. The blot was not repeated due to sample availability after running the CoIP-MS. (P) Representative images (left) and quantitation (right) of MRTF-A and SORBS2 co-localization in the nucleus and cytoplasm across the different treatment groups (including those from Extended Data Figs. 4J,K. Undiff CF (n = 9), MyoFb (n = 37), TGFβi (n = 19), Soft (n = 12), Soft+TGFβi (n = 11), SAR (n = 11), SAR+TGFβi (n = 9 cells). Images in panels C, H, and P are representative of experiments repeated twice independently with similar results. P-values were calculated based on ordinary one-way ANOVA with Dunnett’s test (*’s) versus DMSO (panels D-G) or Undiff CF control (panel I); and the two-sided Student’s t-test (#’s) versus MyoFb (panels J, P). Data are mean ± SEM.

Extended Data Figure 7. Saracatinib potentiates reversal of metabolic states in MyoFbs without affecting the viability and function of cardiomyocytes and endothelial cells.

(A) Seahorse XF Assay measurements of: (i) oxygen consumption rate (OCR), (ii) extracellular acidification rate (ECAR), and (iii) OCR versus ECAR after 2 days of drug treatments (n = 4 replicates/condition). (B-i) PrestoBlue viability assay for varying concentrations of a direct YAP/TAZ inhibitor verteporfin (‘VP’, top) and the SRC inhibitor saracatinib (‘SAR’, bottom) in iPSC-CMs, iPSC-ECs, iPSC-CFs, and iPSC-MyoFbs (n = 3 replicates per group). (B-ii) Half-maximal inhibitory concentration (IC50) values for VP and SAR, for each cell type. IC50 values were obtained from curve-fitting (data from panel B-ii, four-parameter dose-response inhibition with variable slope on GraphPad Prism v6.0). IC50 for each group is plotted as individual bars, and the asymptotic standard error obtained from the best-fit curves (“Std. Error” values) are presented as error bars. (C) Representative images (left) and quantitation (right) of EC tube formation assay after treatment with VP or SAR (1 μM each). n = 6 randomly selected fields of view, from n = 3 biological replicates/group. Images are representative of experiments repeated twice independently with similar results. (D) Heatmap (top) and line graph (bottom) illustrating the sharp decrease in SRC expression relative to cardiac genes (TTN, MYH7, and TNNT2) throughout the course of iPSC-CM differentiation (bulk RNA-seq, n = 3 technical replicates per time point). (E-G) Beating area, contraction velocity, and relaxation velocity measurements in VP- and SAR-treated iPSC-CMs at differentiation day 90 (D90): DMSO (n = 19), VP (n = 22), SAR (n = 24 randomly selected fields of view, from n = 3 biological replicates/group). (H) Western blot of phosphorylated SRC (Y416) and total SRC in MyoFbs under TGFβ drug perturbations. SAR was included as a positive control. Blots are representative of n = 3 experiments repeated with similar results. Images are from two blots from the same samples from the same experiment, processed in parallel. P-values were calculated based on ordinary one-way ANOVA with Dunnett’s test (*’s) versus DMSO control; and the two-sided Student’s t-test (#’s). Data are mean ± SEM (with the exception of panel B-ii, for which individual bars ± error bars represent IC50 ± Std. Error).

Extended Data Figure 8. Combination treatment alleviates contractile dysfunction in fibrotic engineered heart tissues.

(A) Schematic of EHT formation and fibrotic compaction of EHTs upon TGFβ stimulation. Right inset: snapshots of relaxed (diastole; cyan) and contracted (systole; magenta) states. (B) Representative images of EHTs for each condition in their relaxed states (diastole). The distance between the centre of the two silicon posts were used to measure the diastolic tissue length, L_d (yellow dotted line). Images are representative of experiments repeated four times with similar results. Scale bar = 1 mm. (C-F) Fractional shortening (FS%) and diastolic tissue length (L_d) measured over time for each drug treatment group (C,E) and FS% and L_d after 2 days of treatment with the different drugs, normalized to DMSO control (D,F). Ctrl (n = 13), TGFβ (n = 13), S (n = 10), P (n = 11), S+P (n = 12 EHTs). (G) Elastic modulus (stiffness) of EHTs 4 days after the drug treatments: Ctrl (n = 5), TGFβ (n = 7), S (n = 4), P (n = 3), S+P (n = 7). For panels C and E (between days 0 and 6) and for panels D and G, data from four separate experiments were normalized and pooled. P-values were calculated based on ordinary one-way ANOVA with Dunnett’s test (*’s) versus control EHTs; and the two-sided Student’s t-test (#’s). Data are mean ± SEM.

Extended Data Figure 9. Combination treatment in MI and TAC models of cardiac fibrosis.

(A) Schematic summary of the MI experiment. SAR+PFD (‘S+P’) dual treatment was started at three different time points: Sham (n = 6), MI (n = 14), and S+P starting on day 1 (n = 8), day 4 (n = 10), and day 7 (n = 4 animals) post-MI surgery. (B) Survival curves for the five treatment groups. (C) Representative images of Masson’s trichrome-stained tissue sections and (D) M-mode echocardiography scans for each group at day 21 post-MI. Images are representative of at least four animals per group, from one MI cohort. (E-I) Time-course of left ventricular ejection fraction, LVEF (%, (E)), and endpoint echocardiography measurements of LVEF (%) (F), fractional shortening (FS%, (G)), left ventricular posterior wall thickness in diastole (LPWd, (H)), and left ventricular internal diameter (LVIDd, (I)). At endpoint, Sham (n = 6), MI (n = 9), ‘S+P (d4-)’ (n = 8), and ‘S+P (d7-)’ (n = 4 animals). (J) Schematic summary of the TAC experiment, copied from Fig. 5B. (K-M) Time-course measurements of FS (K, left; for days 18, 24, and 42, data from three separate experiments were normalized and pooled), and week 9 measurements of: FS (K, right), left ventricular internal diameter (LVIDd, (L)), and heart weight / body weight (mg/g) ratio (M). Sham, TAC, S+P, S+P withdrawn, S, and P (n = 10, 12, 12, 6, 7, 7 animals, respectively). P-values were calculated based on ordinary one-way ANOVA with Dunnett’s test (*’s) versus Sham; and the two-sided Student’s t-test (#’s) versus MI (panels F-I) or versus TAC (panels K-M). Data are mean ± SEM.

Extended Data Figure 10. Combination treatment induces transcriptional and protein-level changes in TAC hearts without significantly affecting inflammation or immune signalling.

(A) Left: Western blot of phosphorylated YAP (S127) and total YAP in whole heart lysates from the different drug treatment groups: Sham, Sham S+P, TAC 4 wk, TAC 6 wk, S+P, S, P (n = 5, 5, 10, 9, 9, 8, 9 hearts per lysate, respectively). Right: snRNA-seq violin plot displaying module scores for major YAP/TAZ target genes (Ccn2 (CTGF), Ankrd1, Amotl2, Birc5, and Igfbp2) specifically in CFs. (B) Violin plots showing module scores for common genes associated with oxygen consumption rate (OCR, left) and extracellular acidification rate (ECAR, right). (C) Normalized mass spectrometry intensity values for SORBS2 (ArgBP2) in two published datasets of mouse TAC hearts^41,50: Kuzmanov et al and Rudebusch et al. (n = 5 and 4 tissue samples, respectively). (D) Violin plot of Sorbs2 expression in the five groups. Horizontal bars indicate mean. Cells with >0 Sorbs2 expression are shown. (E) Gene Ontology enrichment analysis for SAR+PFD versus TAC hearts. Fisher’s exact test (with Benjamini-Hochberg adjustment; one-sided). (F) Dot plot of major pro-inflammatory cytokine genes in the heart. (G-I) CellChat-based inference of cell-cell communication mechanisms, showing changes in modes of signalling from immune cells to CF populations (immune→CF) in the SAR+PFD group relative to TAC (H,I). A permutation test (randomization; one-sided) was used to compute communication probability strength of ligand-receptor pairs. (J-K) Immunoblots and corresponding quantitation of SRC phosphorylation and key fibrosis-associated proteins αSMA and POSTN for each group: Sham, Sham S+P, TAC (4 wk), TAC (6 wk), S+P, S, P (n = 5, 5, 10, 9, 9, 8, 9 hearts per lysate, respectively). Immunoblots in panels A and K are representative of n = 3 experiments performed independently with similar results. Images are from two blots from the same samples from the same experiment, processed under identical conditions. For panel K, data from an additional experiment was normalized and pooled. P-values were calculated based on ordinary one-way ANOVA with Dunnett’s test (*’s) versus Sham; and the two-sided Student’s t-test (#’s). Data are mean ± SEM.

Data Availability

Raw and processed single-cell/nucleus sequencing data have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession number GSE291370 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE291370). Proteomics data generated from this study have been uploaded to ProteomeXchange/MassIVE under accession numbers PXD061188/MSV000097212 (https://massive.ucsd.edu/ProteoSAFe/dataset.jsp?task=ee0071c049c54dd1bfa1bd63674dde91). Source data files for other types of data are provided with this paper. Public datasets re-analysed in this study are cited in the References and are also available in: NCBI GEO under accession number GSE120064³⁹ (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE120064), the European Genome-phenome Archive (EGA) under accession number EGAS00001006374³³ (https://ega-archive.org/studies/EGAS00001006374), the Human Cell Atlas (HCA) Data Coordination Platform (DCP) under accession number ERP123138³⁶ (https://www.ebi.ac.uk/ena/browser/view/ERP123138), the Broad Institute’s Single Cell Portal under study ID SCP498³⁷ (https://singlecell.broadinstitute.org/single_cell/study/SCP498/transcriptional-and-cellular-diversity-of-the-human-heart), the ArrayExpress database at EMBL-EBI under accession codes E-MTAB-7376³⁸ (https://www.ebi.ac.uk/biostudies/arrayexpress/studies/E-MTAB-7376) and E-MTAB-7365³⁸ (https://www.ebi.ac.uk/biostudies/arrayexpress/studies/E-MTAB-7365), ProteomeXchange under accession number PXD016492⁴¹ (https://proteomecentral.proteomexchange.org/cgi/GetDataset?ID=PXD016492), ProteomeXchange under accession number PXD007171 (https://proteomecentral.proteomexchange.org/cgi/GetDataset?ID=PXD007171-1&test=no), and the NCBI National Library of Medicine (NLM; genome assemblies GRCh38, GCF_000001405.26, https://www.ncbi.nlm.nih.gov/datasets/genome/GCF_000001405.26/ for human; and GRCm38, GCF_000001635.20, https://www.ncbi.nlm.nih.gov/datasets/genome/GCF_000001635.20/ for mouse).

Code Availability

No custom code or mathematical algorithms were developed for the study. All scRNA-seq/snRNA-seq data processing and analyses were performed using standard pipelines with Seurat (v.5.1.0) and open source R packages cited in the Methods and Reporting Summary, under default settings provided in the corresponding vignettes.