Increasing KLF15 in heart muscle cells may help prevent harmful changes and scarring using a gene-activating tool

To further evaluate the transcriptional dynamics of KLF15 in human cardiovascular diseases, we analyzed publicly available datasets derived from single-nucleus profiling of human dilated (DCM) and hypertrophic cardiomyopathy (HCM) as well as non-failing (NF) donors³⁰ and detected a significant reduction of KLF15 in cardiomyocytes of DCM and HCM patients compared to NF controls (Fig. 1h). Additionally, we applied a statistical approach to identify single-nucleotide variants (SNVs) that affect transcription factor binding.³¹ Analysis of cardiovascular genome-wide association studies (GWAS) showed that regulatory (r)SNVs associated KLF15 with its loss of function in four cardiovascular diseases (Fig. 1i). Collectively, these findings suggest that altered KLF15 transcriptional activity is a key driver of the phenotypic and pathological remodeling of cardiomyocytes in an evolutionarily conserved manner.

We hypothesized that normalizing the perturbed transcriptional activity of KLF15, as a key TF, will maintain homeostasis upon stress. Consequently, we tested CRISPRa (using dCas9VPR) as a safer, physiologically relevant strategy to normalize TF activity. Unlike classical open reading frame (ORF) overexpression (OE), which induced aberrant cytosolic KLF15 accumulation in vitro, CRISPRa preserved its native nuclear localization (Supplementary Fig. 4a), as previously observed.²¹ We used a mouse model with constitutive dCas9VPR expression restricted to cardiomyocytes (Myh6-dCas9VPR-tdTomato).²¹ We first validated robust expression of dCas9VPR, as indicated by the tdTomato reporter and the delivery of gRNAs shown by the expression of EGFP encoded by the AAV construct as well as by immunobloting in the adult heart (Supplementary Fig. 4b, c). To activate Klf15 expression, we administered validated gRNAs complementary to the 5’ transcriptional start site (TSS) upstream region of the Klf15 gene body via intravenous injection of recombinant AAV 2/9 (AAV9) at a dose of 1E + 12 viral genomes (vg) per mouse. This resulted in a significant induction of Klf15 expression at 2 and 8 weeks after AAV9 injection in ventricular tissue compared to saline-injected controls (Supplementary Fig. 4d). CRISPRa transgenic mice showed normal cardiac function and Klf15 expression (Supplementary Fig. 4e, f) as well as a similar response to pressure overload compared to wild-type littermates (Supplementary Fig. 4g–j). Next, we performed Klf15 re-expression experiments in CRISPRa adult mice, which were injected with either AAV9 Klf15 gRNAs (CRISPRa-Klf15), non-targeting (NT) control gRNAs (CRISPRa-NT), or saline and subjected to TAC or sham surgery eight weeks post-injection to ensure robust AAV transduction (Fig. 1j). Klf15 KO⁸ were used to generate a double CRISPRa/Klf15^−/− mouse line, which served as a negative control. AAV9 transduction efficiency and dCas9VPR-tdTomato expression were similar in sham and TAC groups, as demonstrated by reporter expression. CRISPRa-Klf15 mice showed smaller hearts after TAC as compared to respective controls 16 weeks post-surgery. CRISPRa/Klf15 KO mice subjected to TAC resulted in an exacerbated cardiac dilation as expected,^10,11 which was not rescued by injection of AAV9 Klf15 gRNAs (Fig. 1k and Supplementary Fig 5a, b). This indicated that the rescue mediated by CRISPRa depends specifically on KLF15. Echocardiography analysis confirmed comparable heart function among the groups prior to TAC, except for CRISPRa/Klf15 KO mice, which showed reduced systolic function prior to surgery. Successful and uniform pressure overload over all TAC groups was confirmed (Supplementary Fig. 5d, e). At 16 weeks post-TAC, RT-qPCR analysis revealed reduced Klf15 expression in all TAC groups, except for the CRISPR-Klf15 group, which showed normalized Klf15 expression (Fig. 1l and saline shown in Supplementary Fig 5c). After TAC, CRISPRa-Klf15 showed reduced cardiomyocyte cross-sectional area (CSA), reduced functional decline as shown by fractional shortening (FS), better survival, and reduced ventricular internal dimension at systole (LVIDs) as compared to all control TAC conditions (Saline and CRISPRa-NT) (Fig. 1m–p and Supplementary Fig 5f). Enhanced transcript levels of Klf15 positively correlated with FS (R² = 0.69) and negatively correlated with cardiac mass (R² = 0.77) in TAC CRISPRa-Klf15 mice (Supplementary Fig. 5g). These results establish CRISPRa-mediated Klf15 reactivation as a viable strategy to counteract pathological cardiac tissue remodeling and to restore functional homeostasis in the stressed heart.

Consistent with the Klf15 levels, cardiomyocyte expression of Aldh2, a direct metabolic target gene of KLF15, was decreased in TAC hearts compared to sham control but normalized in TAC CRISPRa-Klf15 hearts. This was validated by RT-qPCR (Supplementary Fig. 7c). Expression of Acta2 (Actin Alpha 2, Smooth Muscle), a marker of fetal cardiomyocytes and a stress-induced reprogramming, was reduced in TAC CRISPRa-Klf15 cardiomyocytes compared to TAC CRISPRa-NT cardiomyocytes (Fig. 2g and Supplementary Fig. 7c). Further analysis of KLF15-dependent transcriptional networks revealed an increase in the expression of developmental and hypertrophic genes that are negatively regulated by KLF15, including Mef2 (Myocyte enhancer factor 2), Gata4 (GATA binding protein 4), Wnt signaling, and myocardin-mediated serum response factor (SRF) target genes Nppa and Myh7 along with downregulation of genes involved in BCAA and lipid metabolism, which are positively regulated by KLF15^12–14 in TAC CRISPRa-NT hearts particularly in stressed cardiomyocytes (CM0). Expression of these transcripts was largely normalized in CRISPRa-Klf15 cardiomyocytes, particularly in CM2 (Fig. 2h). The regulation of α-SMA (Actin Alpha 2, Smooth Muscle), the protein product of the Acta2 gene, was further corroborated by immunofluorescence, which showed co-localization with cTnT (Cardiac Troponin T)-expressing cells indicating de-differentiation of cardiomyocytes (Fig. 2i and Supplementary Fig. 7d). Further cardiomyocyte de-differentiation hallmarks, including increased expression of Tnni1 (slow skeletal muscle troponin I) and decreased expression of Tnni3 (cardiac-specific troponin I) were observed in TAC CRISPRa-NT hearts but not in TAC CRISPRa-Klf15 hearts (Supplementary Fig. 7e).

Next, we performed bulk proteomic analysis. Although this approach lacks sc resolution and may obscure cell-specific effects, it offers valuable validation of our findings beyond the transcript level. This data showed that levels of proteins depending on KLF15 transcription, including hypertrophic factors α-SMA, MYH7, and ANP (atrial natriuretic peptide, the product of Nppa) were increased in TAC CRISPRa-NT hearts but were attenuated in TAC CRISPRa-Klf15 hearts. Metabolic and calcium handling proteins, including ALDH2, were more abundant in TAC CRISPRa-Klf15 compared to TAC CRISPRa-NT. Additionally, the levels of pro-fibrotic proteins were also attenuated in TAC CRISPRa-Klf15 compared to TAC CRISPRa-NT (Fig. 2j). Accordingly, GO processes analysis of more abundant proteins in TAC CRISPRa-Klf15 compared to TAC CRISPRa-NT revealed enrichment in gene transcription, while less abundant proteins were categorized to glycosylation and stress signaling, including Wnt and TGF-β signaling (Fig. 2k and Supplementary Fig. 7f). Thus, CRISPRa-enhanced expression of Klf15 normalized its downstream homeostatic transcriptional network in a subset of cardiomyocytes, which is sufficient to prevent transcriptional fetal reprogramming and maintain cardiac performance.

These tissues were subjected to next-generation sequencing (NGS). Principal component analysis and hierarchical clustering revealed three distinct gene expression profiles according to the experimental groups. Activation of KLF15 altered the gene expression profile in stressed tissues towards that of control conditions (Fig. 4e and Supplementary Fig. 11a, b). Specifically, CRISPRa-NT tissues upon stress showed reduced expression of KLF15, metabolic targets, and structural genes, along with the upregulation of developmental and hypertrophic markers. Moreover, TGF-β dependent target genes, including ACTA2, POSTN, CCL2 (C-C Motif Chemokine Ligand 2), CCN2 (connective tissue growth factor), and TAGLN (Transgelin) were upregulated along with reduced expression of antifibrotic genes BMP7 (bone morphogenetic protein 7) and AZGP1 in CRISPRa-NT upon stress. These changes were reversed in CRISPRa-KLF15 stress EHM (Fig. 4f). NanoString analysis validated increased expression of genes that can support cardiac function, including SLC8A1 (Solute Carrier Family 8 Member A1), GLUT4 (Glucose Transporter Type 4), and CKM (creatine Kinase, Muscle), as well as decreased expression of ACTA2 and BMP4 in CRISPRa-KLF15 EHM upon stress (Fig. 4g and Supplementary Fig. 11c).⁴¹ GO analysis of upregulated genes in CRISPRa-NT stress tissues compared to CRISPRa-NT control tissues was related to myocardial stress, including cell cycle activation and HIF-1 signaling, while the downregulated genes classified to channel activities involved in membrane potential. In CRISPRa-KLF15 stress tissues, upregulated genes clustered to mitochondrial respiration, whereas downregulated genes categorized to stress processes such as the TGF-β signaling pathway compared to CRISPRa-NT stress tissues (Fig. 4h). Additionally, control EHM with CRISPRa-KLF15 showed decreased TGF-β signaling pathway activity along with increased AZGP1 expression compared to NT control independent from the stress condition (Fig. 4i, j, and Supplementary Fig. 11d). Altogether, we demonstrated for the first time a reduction in KLF15 expression in a human model subjected to mechanical load. This was accompanied by cardiomyocyte reprogramming, a pro-fibrotic response, and functional decline, all of which were reversed by CRISPRa-mediated induction of KLF15.

We further aimed to investigate how KLF15 is regulated in hiPSC-cardiomyocytes under stress conditions. Increased secretion of TGF-β by various cell types is a characteristic feature of hypertrophic remodeling induced by pressure overload and has been shown to downregulate Klf15 expression in animal models.¹² Among the three TGF-β isoforms, TGF-β1 is the most extensively implicated in pathological cardiac remodeling.⁴² Therefore, we treated hiPSC-cardiomyocytes with TGF-β1, which resulted in a dose-dependent downregulation of KLF15 along with increased expression of the dedifferentiation marker α-SMA, specifically in cTnT-positive cells (Fig. 4k–m). Upregulation of reprogramming hallmark genes was observed, including ACTA2 and the cell cycle marker CCND1 (Cyclin D1), along with the reduced expression of mature sarcomeric ACTN2 (alpha-actinin-2) at the transcript and protein level (Fig. 4n, o). We also tested the effect of TGF-β2 in hiPSC-cardiomyocytes, which resulted in downregulation of KLF15 only at high concentrations. In contrast to the effect triggered by TGF-β1, TGF-β2 treatment resulted in ACTA2 and BMP7 downregulation, while CCND1 levels remained unchanged (Supplementary Fig. 12a, b). Next, we investigated the TGF-β1-mediated cascade regulating KLF15. Consequently, we blocked the TGF-β1 canonical activation of SMAD2/3 using the inhibitor SB431542, which prevented the TGF-β1-mediated reduction of KLF15, as well as the upregulation of ACTA2, CCND1, and α-SMA. Blocking of p38 activity (non-canonical TGF-β1 signaling) with the inhibitor SB203580 did not affect TGF-β1-mediated reduction of KLF15 or activation of ACTA2, CCND1, and α-SMA (Fig. 4p, q and Supplementary Fig. 12c). Of note, p38 inhibition resulted in accumulation rather than reduction of its phosphorylation. This is explained by the SB203580 mode of action, which does not block p38 MAPK phosphorylation itself but prevents its kinase activity, thereby preventing the phosphorylation of downstream substrates such as HSP27, which was demonstrated in our analysis.⁴³ Thus, downregulation of KLF15 is a cell-specific downstream effector of the canonical TGF-β1 pathway triggering fetal reprogramming in human cardiomyocytes.

To gain insights into the functional consequences of KLF15 deficiency, we analyzed excitation-contraction (EC) coupling in KLF15^+/− and KLF15^−/− hiPSC-cardiomyocytes. Compared to controls, a complete loss of KLF15^−/− reduced the amplitudes of systolic calcium transients and slowed the rates of calcium increase (time to peak) and decay (relaxation time 50%) compared to wild-type cells (Figs. 5d and S13a). Since EC coupling is tightly coupled to mitochondrial function through mechano-energetic coupling,⁴⁴ we also determined mitochondrial and metabolic function of KLF15-deficient cardiomyocytes. In fact, complete loss of KLF15^−/− decreased basal and maximal oxygen consumption rate (OCR) as well as ATP-linked respiration, indicating compromised mitochondrial function, while the presence of one KLF15 allele (KLF15^+/−) mitigated this impairment (Fig. 5e). However, additional rescue attempts via CRISPRa-KLF15 yielded no statistically significant alterations (Supplementary Fig. 13b, c). Accordingly, mitochondrial structure was impaired in KLF15^+/− and KLF15^−/− cardiomyocytes, resembling a more immature state characterized by decreased mitochondrial branches and junctions, indicative of a disrupted network, while the mitochondrial counts remained unchanged compared to KLF15^+/+ cells. Conversely, CRISPRa-KLF15 significantly increases mitochondrial branching in KLF15^+/− but not in KLF15^−/− cardiomyocytes (Fig. 5f and Supplementary Fig. 13d, e). These data characterized a pathological mitochondrial phenotype upon genetic loss of KLF15, which was reversed by its transcriptional activation in heterozygous cells. Major sarcomeric structural alterations were not observed (Supplementary Fig. 13f).

To investigate the impact of KLF15-deficiency in a pathological context, we tested the effect of CRISPRa-KLF15-mediated induction after TGF-β1 treatment (Fig. 5g). Treatment with TGF-β1 increased systolic calcium levels and accelerated the kinetics of calcium transients, indicating a hyperdynamic condition as can be observed in various forms of cardiac hypertrophy. Accordingly, maximal mitochondrial respiration was also increased in response to TGF-β1. These TGF-β1-induced alterations in EC coupling and mitochondrial energetics were all prevented by the CRISPRa-KLF15-mediated induction (Fig. 5h, i and Supplementary Fig. 13g). These data suggest that initial TGF-β-mediated reduction in KLF15 levels triggers a compensatory response, however, persistent suppression of KLF15 may contribute to the progression of heart failure.

We further evaluated the paracrine effects of the cells under different doses of KLF15 expression. In line with the data in Fig. 3, CRISPRa-KLF15 significantly induced the expression of AZGP1 in pathological conditions upon TGF-β1 treatment (Supplementary Fig. 14a) as well as in KLF15^+/+ and KLF15^+/- but not in KLF15^−/− hiPSC-cardiomyocytes (Fig. 5j). We next investigated whether KLF15^+/− and KLF15^−/− hiPSC-cardiomyocytes produce a pathological secretome that can affect fibroblasts, which may be rescued by induction of KLF15 by CRISPRa. Conditioned medium from KLF15^+/− and KLF15^−/− hiPSC-cardiomyocytes with and without CRISPRa-KLF15 or AZGP1 OE as a positive control was used to treat hiPSC-fibroblasts. Loss of KLF15 expression in cardiomyocytes yielded conditioned media that induced α-SMA elevation in fibroblasts. In contrast, supernatant from KLF15^+/− hiPSC-cardiomyocytes with CRISPRa-KLF15 or AZGP1 OE significantly ameliorated α-SMA activation in fibroblast, whereas this was not the case with medium from CRISPRa-KLF15 in KLF15^−/− hiPSC-cardiomyocytes (Fig. 5k and Supplementary Fig. 14b). Interestingly, BMP7 expression was also increased in KLF15^+/− hiPSC-cardiomyocytes with CRISPRa-KLF15 but not in KLF15^+/+ cardiomyocytes (Supplementary Fig. 14c). To evaluate the effects of these two potential antifibrotic factors induced upon stress in CRISPRa-KLF15 cardiomyocytes, we employed human recombinant AZGP1 and BMP7 proteins on hiPSC-fibroblasts treated with TGF-β1. AZGP1 exhibited a similar effect to SMAD2/3 inhibition, resulting in amelioration of the fibrotic response as evidenced by reduced levels of POSTN, α-SMA, and CTGF (Connective Tissue Growth Factor) following TGF-β1 treatment (Fig. 5l). AZGP1 abolished SMAD2/3 phosphorylation induced by TGF-β1, while BMP7 only reduced its phosphorylation. AZGP1 treatment did not significantly affect the phosphorylation levels of p38, AKT, and ERK1/2 (Fig. 5m). To determine if AZGP1 mediates the effects of KLF15 in cardiomyocytes, we analyzed hiPSC-cardiomyocytes with AZGP1 OE. Our analysis revealed no significant changes in the expression levels of genes associated with reprogramming or fibrosis, including ACTA2, KLF15, ACTN2, and BMP7. Additionally, we observed no alterations in α-SMA levels (Supplementary Fig. 14d, e). In addition, unlike in fibroblasts, AZGP1 OE in hiPSC-cardiomyocytes was unable to modulate upregulated ACTA2 expression upon TGF-β1 treatment (Supplementary Fig. 14f). These data indicate that AZGP1 does not contribute to cardiomyocyte reprogramming but instead primarily affects the fibroblast response. Finally, direct CRISPRa-KLF15 induction in hiPSC-fibroblasts did not alter the expression of AZGP1 and did not ameliorate the fibrotic response induced by TGF-β1. (Supplementary Fig. 14g). Altogether, these findings indicate that KLF15 regulates the fibrotic response through cardiomyocytes, where loss of KLF15 induces a pathological secretome that activates fibroblasts. This phenotype can be rescued by CRISPRa-mediated KLF15 restoration, with AZGP1 acting as a major downstream contributor.

To elucidate the KLF15-mediated cell-autonomous regulatory landscape establishing homeostasis in cardiomyocytes, lentiviral transduction of FLAG- and HA-tagged KLF15 was performed, followed by ChIP-seq and mass spectrometry (MS) analysis (Fig. 5n). ChIP-seq was validated by identification of KLF motifs in the enriched sequences (Supplementary Fig. 15a). Recruitment of KLF15 was identified at promoters of metabolic, mitochondrial, and developmental genes, indicating a direct regulation of KLF15 in these processes. Specifically, KLF15 enrichment was identified in SLC25A42 (Solute Carrier Family 25 Member 42), POLRMT (RNA Polymerase Mitochondrial), and POLG (DNA Polymerase Gamma, Catalytic Subunit), important for energy supply (Fig. 5o, p). Validating this finding, EHM generated from CRISPRa-KLF15 hiPSC-cardiomyocytes exhibited upregulated expression of these genes compared to EHM with CRISPRa-NT controls, as determined by NGS. These data validated KLF15’s activating role on these direct targets. (Fig. 5q). Moreover, protein pulldown followed by MS identified CSRP3 (Muscle LIM protein) as the most significantly enriched protein in the KLF15 pulldown (Fig. 5r and Supplementary Fig. 15b–e). CSRP3 is associated with hypertrophic cardiomyopathy and regulation of hypertrophic signaling, including NPPA, ACTA2, and structural proteins.^45,46 Consistently, EHM generated with CRISPRa-KLF15 hiPSC-cardiomyocytes showed regulation of these genes as compared to CRISPRa-NT control EHM (Fig. 5s). The genes encoding these proteins showed no significant enrichment of KLF15 in their promoter regions (Supplementary Fig. 15f). These data indicate that KLF15 exerts direct and indirect effects on metabolic, sarcomeric, and mitochondrial gene regulation.

Finally, we tested the efficiency of the TNNT2-Sauri-dCas9VPR construct and KLF15 activation in precision-cut human myocardium⁵³ as a model for physiological cell context and transducibility of human heart tissue. We first determined the ability of AAV9 particles to transduce the human myocardium slices using the EGFP reporter by culturing them for 11 days (Fig. 6g, h and Supplementary Fig. 16c). We used the biomimetic culture of human myocardium from patients with cardiomyopathies to transduce AAV9-TNNT2-Sauri-dCas9VPR containing the validated KLF15 gRNA or corresponding NT control. Hearts from three patients were prepared to derive slices for KLF15 or NT gRNA delivery. Tissue biomechanics were characterized for 11 days of culture. Expression of Sauri-dCas9VPR as well as KLF15 was confirmed by RT-qPCR (Fig. 6i). Cultured slices showed increased contractility over the time of culture in slices treated with AAV9 KLF15 gRNA compared to NT control (Fig. 6j). Cardiac specificity of the TNNT2 promoter was validated in vivo (Supplementary Fig. 16d). These results validated a method for delivery of a CRISPRa platform into human cardiomyocytes that does not exceed the payload of AAVs and that efficiently modulated KLF15 expression, an approach compatible with clinical translation.

All mouse experiments in this study were approved by the Lower Saxony State Office for Consumer Protection and Food Safety (LAVES), animal research project AZ-G 15-1840 and AZ-G 33.9-42502-04-20/3434. Approval for the use of human material was received from the local ethics committee of the Justus Liebig University Giessen (AZ 184/24). Cardiac tissue specimens obtained from patients with cardiomyopathies were used to prepare myocardial slices. All procedures conducted in this study adhered to the principles outlined in the Declaration of Helsinki. Informed consent was signed by all tissue donors.

Mice (Myh6-dCas9VPR)²¹ were crossed with Klf15/Klf15 knockout (KO) mice¹⁰ to obtain double transgenic Myh6-dCas9VPR/Klf15 KO mice. Wild-type C57BL/6N animals were obtained from Charles River Laboratories. Animals were fed ad libitum and housed in a temperature- and humidity-controlled environment in a 12 h day/night cycle. Both male and female mice were included in the study. Four- to six-month-old mice were injected via the tail vein with 1 × 10¹² vg/mouse rAAV2/9-Klf15 (KLF15 gRNAs A, B, C) or rAAV2/9-NT (NT gRNAs 1, 2, 3) diluted in 80 µL of 0.9% NaCl. Injected adult mice underwent transverse aortic constriction surgery (TAC) using a 26 G blunt cannula or sham surgery (control) as previously described.⁶⁷ To determine the level of pressure overload by aortic ligation, a high-frequency Doppler probe was used to measure the ratio between blood flow velocities in the right and left carotid arteries. TAC mice with a blood flow gradient < 60% were excluded. Echocardiography was performed 4 and 8 weeks post-surgery, and hearts were harvested 16 weeks post-surgery. Mice were anesthetized with 2% isoflurane inhalation, and ventricular measurements were performed with a VisualSonics Vevo 2100 Imaging System equipped with an MS400, 30 MHz MicroScan transducer. The observer was unaware of the genotypes and treatments. Images of excised organs were taken with a Stereo Lumar.V12 stereomicroscope (Carl Zeiss). Surgeries and echocardiography were performed and analyzed in a blinded manner with the support of the CRC1002 service unit.

Plasmids, the design of gRNA for the validation of human KLF15 gRNA, and the TRISPR vector used for the delivery of gRNAs were described previously.²¹ A list of gRNAs is provided in Table 1. The CMV promoter was exchanged with human TNNT2 promoter in pAAV-CMV-SauriCas9, pAAV-CMV-SauriCas9-KKH-puro, and pAAV-CMV-SaSauriCas9 (a gift from Yongming Wang, Addgene plasmid #135964, #135966, #135967) using XbaI and BshTI sites (both Thermo Fisher Scientific).⁴⁸ SauriCas9 was exchanged with SlugCas9 derived from pAAV-CMV-SlugCas9 (a gift from Yongming Wang, Addgene plasmid #163793)⁴⁷ using In-Fusion cloning (Takara) and BshTI and BamHI restriction sites (Thermo Fisher Scientific). Mutations in the RUVC and HNH domains were introduced by homology comparison and mutagenesis PCRs. The transactivation domain, HA-tag, and mini-polyA were introduced by a gene block (Eurofins Genomics) using the BamHI restriction site (Thermo Fisher Scientific). Sanger sequencing (MicroSynth Seqlab) and ITR integrity checks were performed using XmaI and KpnI sites (Thermo Fisher Scientific). Nuclease deficiency was checked by transfecting CMV-SauriCas9 or CMV-Sauri-dCas9 plasmids co-expressing the following EGFP targeting gRNAs: EGFP gRNA1: 5’-GTCAAGGAGGACGGCAACATC-3’, EGFP gRNA2: 5’-GCGACCAGGATGGGCACCACC-3’ into HEK293T cells transduced with an EGFP lentiviral expression plasmid. DNA was isolated, and PCRs were performed to check for deletion events using the following primer pair: EGFP fwd: 5’-GAGCAAGGGCGAGGAGCT-3’; EGFP rev: 5’-TTGTACAGCTCGTCCATGCC-3’.

Cardiac cells were isolated from whole mouse hearts using a modified Langendorff perfusion protocol. Briefly, hearts were perfused with perfusion buffer containing 150 mM NaCl, 5 mM HEPES, 5.4 mM KCl, 10 mM glucose, 2 mM Na-pyruvate, 1.2 mM MgCl₂, 10 mM taurine, and 12.38 mM 2,3-butanedione monoxime for 2 min at 37 °C followed by perfusion with perfusion buffer supplemented with 25 mM CaCl₂ and 210 µg/mL Liberase DH (Roche) for 10 min at 37 °C. Reactions were stopped by adding perfusion buffer supplemented with 0.5% w/v bovine serum albumin (BSA) and 200 µM CaCl₂. Hearts were minced and homogenized into a single cell suspension. For sc imaging, glass coverslips were coated with Matrigel (Corning) for 1 hour at 37 °C and cells were seeded on prepared coverslips. Cells attached within 30 min at room temperature. Live-cell imaging was performed on an Axio Imager M2 with ZEN software (Carl Zeiss).

Samples were prepared and processed for sc transcriptomics at the Next Generation Sequencing Facility, Institute of Human Genetics, University Medical Center Göttingen. Briefly, cells were distributed on 5184 nanowell chips ICELL8 250 v Chip (ICELL8 System, Takara Bio). Single live cells were identified using Hoechst 33342 and propidium iodide staining (NucBlue Cell Stain Reagent, Thermo Fisher Scientific) and the CellSelect Software (Takara Bio). Complementary DNA synthesis was performed by oligo-dT priming in a one-step RT-PCR reaction. P5 indexing primers, Terra Polymerase, and Reaction Buffer were added for library preparation. Transposase enzyme and reaction buffer (Tn5 mixture) were added to each well. P7 indexing primers were dispensed to wells. Final libraries were amplified and pooled as they were extracted from the chip. Pooled libraries were purified, and size was selected with Agencourt AMPureXP magnetic beads (Beckman Coulter) to obtain an average library size of 500 bp. A typical yield for a library comprised of ~1300 cells was 15 nM. Libraries were sequenced with a HiSeq4000 (Illumina) to obtain, on average, ~500K reads per cell (single-end, 50 bp). Raw sequencing files (bcl-files) were converted into a single fastq file using Illumina bcl2fastq software (v2.20.0.422) for each platform. Each fastq file was demultiplexed and analyzed using the Cogent NGS analysis pipeline (CogentAP) from Takara Bio (v1.0). In brief, the “cogent demux” wrapper function was used to allocate the reads to the cells based on the cell barcodes provided in the well-list files. Subsequently, the “cogent analyze” wrapper function was used to perform read trimming with cutadapt (version 3.2), genome alignment to Mus musculus gene annotation version 102 from ENSEMBL was done using STAR (version 2.7.7a), read counting for exonic, genomic, and mitochondrial regions was done using featureCounts (version 2.0.1), and a gene matrix was generated with the number of reads expressed for each cell in each gene.

Raw gene matrices underwent quality control (QC) filtering for cells and genes using the following parameters: (a) for cells, only those with at least 10,000 reads associated with at least 300 different genes, and (b) for genes, only those containing at least 100 reads mapped to them from at least three different cells. Mitochondrial and ribosomal content were calculated for each cell to ensure data quality. They were then filtered based on total counts and mitochondrial content based on five median absolute deviations. Only those genes with at least 200 reads mapped from a minimum of 3 different cells were considered, and mitochondrial, ribosomal, and hemoglobin genes were systematically removed. The Seurat package (version 4.1.2) was utilized in this analysis. Following QC, scRNA-seq data were normalized with NormalizeData. Variable features were identified using FindVariableFeatures, and data were scaled while excluding mitochondrial gene expression. Principal component analysis (PCA) was performed with RunPCA, and the optimal number of principal components was determined using an ElbowPlot. The normalized data were then used to identify cell clusters through FindNeighbors and FindClusters. Different clustering resolutions (0.1–0.5) were explored to find the most suitable resolution. Uniform Manifold Approximation and Projection (UMAP) was applied with RunUMAP for visualizing cell clusters. The Seurat package’s integration method was employed to correct for batch effects. The scRNA-seq data were split by sample using SplitObject, and each subset was transformed using SCTransform with the generalized linear model approach and a Poisson distribution. Integration features were selected with SelectIntegrationFeatures, and data were prepared for integration with PrepSCTIntegration. Integration anchors were identified using FindIntegrationAnchors, and data integration was performed with IntegrateData. The integrated dataset underwent additional dimensionality reduction, clustering, and visualization using PCA, UMAP, FindNeighbors, FindClusters, and DimPlot. Differential gene expression analysis was conducted on the integrated dataset to identify markers associated with different biological conditions, utilizing PrepSCTFindMarkers and FindMarkers. The packages dplyr, ggplot2, tidyverse, EnhancedVocano, and dittoseq were used for data visualization and manipulation.^68–70 Gene ontology analysis was performed by ShinyGO 0.82 and Enrichr.^71,72 For BITFAM TF inference, cardiomyocyte samples were extracted from the two selected datasets,^28,29, and their count matrices were subjected to basic BITFAM analysis to obtain a list containing matrices with the inferred TF activity and gene weights. The inferred activity matrix was extracted using the BITFAM_activities tool, and a data frame with the barcoded cells, the inferred activity, and the conditions was constructed for the Random Forest classification. The results were used to plot the TF importance. The BITFAM_weights tool was used to extract the downstream weights for each TF.

KLF15 loss of function in CRISPRa hiPSCs (RUCDRi002-A-15) was generated by ribonucleoprotein-based CRISPR/Cas9 targeting exon 2 of the KLF15 gene, as described previously.³⁸ Transgenic hiPSCs containing the CRISPRa system (RUCDRi002-A-15) were previously described.^37,38 Differentiation protocols for cardiomyocytes, fibroblasts, and EHM were previously described.^39,73 Human induced pluripotent stem cells were cultured in culture dishes coated with Matrigel (Corning) in Stem MACS iPS Brew XF, human (Miltenyi Biotec).

For hiPSC-cardiomyocyte differentiation, hiPSCs were grown until 70% confluency and washed twice with basal medium (RPMI 1640 + GlutaMAX, 1 mM sodium pyruvate (Gibco), 200 μM L-ascorbic acid, 100 U/mL penicillin, and 100 μg/mL streptomycin (Gibco)). For mesoderm induction, basal medium was supplemented with 2% v/v B27 no insulin (Thermo Fisher Scientific), 1 μM CHIR99021 (Merck Chemicals GmbH), 5 ng/mL BMP4 (Bio-Techne), 5 ng/mL FGF-2 (Peprotech), and 9 ng/mL Activin-A (R&D Systems), and cells were cultured for 3 days. Cells were washed twice with basal medium and cardiomyocyte differentiation was initiated for 9 days in basal medium supplemented with 2% v/v B27 supplement (Thermo Fisher Scientific) and 5 μM IWP-4 (ReproCELL Europe Ltd.). Cardiomyocytes were metabolically selected in RPMI 1640, no glucose, 2.2 mM Na-lactate, 100 μM β-mercaptoethanol (Gibco), 100 U/mL penicillin, and 100 μg/mL streptomycin (Gibco) for 5 days. Cells were detached using StemPro Accutase Cell Dissociation Reagent (Gibco) for 30 min and Versene (Gibco) for 5 min at room temperature and reseeded on Matrigel-coated flasks in base medium supplemented with 5 µM ROCK inhibitor Y-27632 (Stemgent). For the TGFβ treatments, hiPSC-cardiomyocytes were differentiated and treated with different concentrations of TGFβ1 or TGFβ2 (Peprotech) (final concentration: 0.001, 0.01, 0.1, 10, 100, and 1000 pM) for 5 days, and cells were harvested for analysis. For TGFβ inhibitor treatment, cells were pretreated for 1 h with respective inhibitors: SB431542 (SMAD2/3 inhibitor, final concentration: 10 µM, Peprotech) and SB203580 (p38 inhibitor, final concentration: 10 µM, Thermo Fisher Scientific) before adding TGFβ1 (final concentration: 10 pM, 1 h or 5 days). The medium was changed daily.

To differentiate fibroblasts, hiPSCs were plated on Matrigel-precoated flasks 4 days before starting the differentiation and maintained in Stem MACS iPS Brew XF, human (Miltenyi Biotec) with daily medium changes. On the day of differentiation, medium was replaced with RPMI 1640 + GlutaMax (Gibco) supplemented with 2% v/v B27 no insulin (Thermo Fisher Scientific), 1 µmol/L CHIR99021 (Stemgent), 5 ng/mL bFGF (Peprotech), 5 ng/mL BMP4 (R&D Systems), and 9 ng/mL Activin-A (R&D Systems), and refreshed after 24 h for mesoderm induction. After 2 days, growth factor composition was switched to the addition of 5 µmol/L IWP4 (Stemgent). 24 h later, cells were passaged on laminin (final concentration 0.9 µg/cm², BioLamina) coated flasks using TrypLE Express (Gibco) and replated in RPMI 1640 + GlutaMax, supplemented with 2% v/v B27 supplement (Thermo Fisher Scientific), 1 µmol/L CHIR00921, 50 ng/mL BMP4, and 4 µM retinoic acid (Sigma-Aldrich) to induce formation of epicardial cells. On day 10, fibroblast differentiation was induced by the addition of 25 ng/mL VEGFA (Peprotech) and 50 ng/mL bFGF in KnockOut DMEM (Gibco) supplemented with 10% v/v KnockOut serum (Gibco) and 2 mmol/L L-glutamine (Gibco). Cells were again passaged on day 14 and maintained in culture on vitronectin (final concentration: 0.9 µg /cm², Invitrogen) coated flasks for up to seven passages with weekly passaging intervals in the fibroblast culture medium (10 ng/mL bFGF, 5 ng/mL EGF (Peprotech), 20 ng/mL IGF-1 (Peprotech), 0.5 ng/mL VEGFA, 10% v/v fetal bovine serum, and 500 μM L-ascorbic acid (Sigma) in DMEM, low glucose, GlutaMAX™ Supplement (Gibco)). Medium was changed every 2 days.

For hiPSC-cardiomyocytes supernatant experiments for assessing the effect of AZGP1 in hiPSC-fibroblasts, supernatant from cardiomyocytes was collected daily for 5 days. To assess the concentration of AZGP1 in the supernatant, the human ZAG/AZGP1 ELISA Kit (Proteintech) was used by following the manufacturer’s instructions. The hiPSC-fibroblasts were seeded in culture vessels, and a medium containing 50% of fibroblast culture medium and 50% of supernatant from hiPSC-cardiomyocytes was added. Cells were collected 5 days after seeding for analysis. The medium was changed daily.

For human recombinant AZGP1 and BMP7 protein treatment, hiPSC-fibroblasts were cultured in the fibroblast culture medium and the cells were treated with TGFβ1 (final concentration: 10 or 100 pM, Peprotech) and human recombinant AZGP1 (final concentration: 1 µg/mL, R&D Systems) or BMP7 (final concentration: 400 ng/mL, Peprotech) for 2, 24 h or 5 days with the additional Zinc²⁺ supplement (final concentration: 1.5 nM, MERCK) in the medium. The cells were cultured in the fibroblast culture medium without FBS for 16 h before TGFβ1 and recombinant treatment.

For long-term cultivation of human myocardium, the MyoDish-1 tissue culture system (InVitroSys) was employed, following the protocol adapted from Fischer et al.⁵³ Briefly, tissue specimens obtained from patients with cardiomyopathies were cleared of connective tissue and intramuscular fat and subsequently trimmed to a 1 × 1 cm² cross-sectional area. Tissue blocks were embedded in 4% w/v low-melt agarose (Roth), mounted on the cooled (4 °C) stage of a vibratome (VT1200S, Leica Biosystems), and cut in the epicardium-tangential plane (2 mm vibration amplitude, 0.06 mm/s feed rate). Resulting 300 µm-thick slices were affixed (Histoacryl, B. Braun) to small plastic triangles with aligned longitudinal fiber orientation. Subsequently, slices were transferred to biomimetic culture chambers (InVitroSys) containing 2.4 mL pre-warmed (37 °C) culture medium (Medium 199, Gibco) supplemented with 100 U/mL penicillin/100 µg/mL streptomycin, 1% insulin/transferrin/selenite 100-x, 20 nM dexamethasone, 50 µM 2-mercaptoethanol) and placed in a standard incubator with 37 °C and 5% CO₂. Electrical pacing (0.5 Hz, 50 mA, 3 ms) was initiated via the MyoDish software. After 24 h of cultivation, the preload was set to 1 mN and remained constant throughout the subsequent culture period. To verify the effectiveness of AAV9 particles in transducing human myocardium, cultivated slices were incubated for two days with the reporter gene EGFP (pAAV-TNNT2-EGFP) at a concentration of 1 × 10¹¹ vg/slice. After 8 additional days of culturing, expression of EGFP in the human tissue slices was examined using confocal microscopy. After confirmation of AAV9 transduction in human tissue, cultured slices were transduced with 1 × 10¹¹ vg/slice TNNT2-Sauri-dCas9VPR containing the validated KLF15 gRNA or the corresponding NT gRNA controls. The contractile performance of transduced tissue slices was recorded for 11 days via the MyoDish software and analyzed with the “LabChart Reader” software (AD Instruments). Upon completion of the culture period, the tissue sections were extracted from the chambers and stored at −80 °C for subsequent immunoblotting and qPCR analysis.

For lentiviral production, 3 × 10⁶ HEK293T cells (Takara) were transfected with 1.23 µg pMD2, 2.27 µg psPAX2 (both kindly provided by Didier Trono, Addgene plasmid #12259 and #12260) and 3.5 µg pGIPZ-KLF15-gRNA (KLF15 gRNAs A, D, E)/pGIPZ-NT-gRNA (NT gRNAs 1, 2, 3)/LentiORF AZGP1 (horizon)/pGIPZ AZGP1 shRNA (horizon)/pGIPZ-FLAG-HA-KLF15/pGIPZ-FLAG-HA-EGFP transfer vectors using Turbofect (Thermo Fisher Scientific). Medium was exchanged the following day and subsequently collected for 3 days. The lentiviral particle-containing supernatant was concentrated using Amicon Ultra-15, 100 kDa MWCO (Sigma Aldrich), and the lentiviral particle concentration was estimated using Lenti-X GoSticks Plus (Takara). For hiPSC-cardiomyocyte or hiPSC-fibroblasts transduction, cells were seeded in the respective culture vessel. The cells were left to recover for three days (hiPSC-cardiomyocyte) or one day (hiPSC-fibroblasts) before transduction. At day 0, the culture medium was changed, and the virus was added to the culture medium. The following day, the media were changed. Transduction efficiency was assessed by fluorescence microscopy. Images were analyzed with ImageJ.

AAV9 preparation was carried out essentially as previously described.^21,74 For the generation of recombinant AAV9, shuttle vectors for these recombinants were co-transfected with AAV9 helper, pDG-9 (a gift from Dr. Roger Hajjar), into HEK293T cells (AAVPro 293T Cells, Takara 632273) to produce virus. To prepare the recombinant AAV9, HEK293T cells were plated at a density of 8 × 10⁶ per T-175 flask and maintained in DMEM/F12 (Gibco) containing 10% v/v FBS and 100 U/mL penicillin/100 µg/mL streptomycin at 37 °C and 5% CO₂. For each virus preparation, 48 flasks were used. Twenty-four hours after plating, cultures were transfected using Polyethylenimine “Max” (MW 40,000, Polysciences) as follows: for each T175 flask, 15 µg of helper plasmid and 5 µg of recombinant plasmid were mixed with 1 mL of DMEM/F12 (no antibiotics) and 160 µL of polyethylenimine (0.517 mg/mL), vortexed for 30 s, and incubated for 15 min at room temperature. This was then mixed with 18 mL DMEM/F12 containing 2% v/v FBS and 100 U/mL penicillin/100 µg/mL streptomycin, which was then used to replace the media on the cultures. The cultures were then rocked intermittently for 15 min before incubation. Three days after transfection, the cells collected from six T-175 flasks were centrifuged at 500 g for 10 min and then resuspended in 10 mL of lysis buffer (150 mM NaCl, 50 mM Tris-–HCl). The resuspended cells were then subjected to three rounds of freeze–thaw, followed by treatment with benzonase (Novagen) and 1 mM MgCl₂ at 37 °C for 30 min. The cell debris was collected by centrifugation at 3400×g for 20 min. The supernatant obtained from six T-175 flasks containing the AAV9 was then purified on an iodixanol gradient composed of the following four phases: 7.3 mL of 15%, 4.9 mL of 25%, 4 mL of 40%, and 4 mL of 60% iodixanol (Optiprep, SigmaAldrich) overlayed with 10 mL of cell supernatant. The gradients were centrifuged in a 70Ti rotor (Beckman Coulter) at 69,000 rpm for 1 h using OptiSeal Polyallomer Tubes (Beckman Coulter). Virus was collected by inserting a needle 2 mm below the 40–60% interface and collecting 4 or 5 fractions (~4 mL) of this interface and most of the 40% layer. The fractions were analyzed for viral content and purity by examining 10 µL of each fraction on a 12% SDS–PAGE gel (BioRad), followed by staining with InstantBlue (Expedeon) to visualize the viral capsid proteins VP1, VP2, and VP3. The virus was then collected from the fractions of several gradients, and the buffer was exchanged with lactated Ringer’s using an ultrafiltration device, Vivaspin 20, 100 kDa MWCO (GE Healthcare). The final viral preparation was then fractionated on a 12% SDS-PAGE gel, stained with InstantBlue, and then compared with a similarly stained gel of a virus with a known titer. Quantification of AAV genomic titers was done by RT-qPCR using gene-specific primers or polyA-specific primers on purified AAV vectors after density gradient ultracentrifugation.

GraphPad Prism 11 was used for statistical analyses unless otherwise specified. Differences with P < 0.05 were considered statistically significant. Datasets were checked for normal distribution using the Shapiro-Wilk test. Two-group comparisons wereperformed with Student’s t-test. For multiple group analyses, one-way ANOVA and Tukey’s post hoc test were used. If more than one independent variable was investigated, two-way ANOVA and Tukey’s post hoc test were used unless otherwise specified.