Fibroblast-specific TGF-β signaling mediates cardiac dysfunction, fibrosis, and hypertrophy in obese diabetic mice

Abstract

Aims

Transforming growth factor (TGF)-β is up-regulated in the diabetic myocardium and may mediate fibroblast activation. We aimed at examining the role of TGF-β-induced fibroblast activation in the pathogenesis of diabetic cardiomyopathy.

Methods and results

We generated lean and obese db/db mice with fibroblast-specific loss of TbR2, the Type 2 receptor-mediating signaling through all three TGF-β isoforms, and mice with fibroblast-specific Smad3 disruption. Systolic and diastolic function, myocardial fibrosis, and hypertrophy were assessed. Transcriptomic studies and in vitro experiments were used to dissect mechanisms of fibroblast activation. Fibroblast-specific TbR2 loss attenuated systolic and diastolic dysfunction in db/db mice. The protective effects of fibroblast TbR2 loss in db/db mice were associated with attenuated fibrosis and reduced cardiomyocyte hypertrophy, suggesting that in addition to their role in fibrous tissue deposition, TGF-β-stimulated fibroblasts may also exert paracrine actions on cardiomyocytes. Fibroblast-specific Smad3 loss phenocopied the protective effects of fibroblast TbR2 loss in db/db mice. Db/db fibroblasts had increased expression of genes associated with oxidative response (such as Fmo2, encoding flavin-containing monooxygenase 2), matricellular genes (such as Thbs4 and Fbln2), and Lox (encoding lysyl oxidase). Ingenuity pathway analysis (IPA) predicted that neurohumoral mediators, cytokines, and growth factors (such as AGT, TGFB1, and TNF) may serve as important upstream regulators of the transcriptomic profile of diabetic mouse fibroblasts. IPA of scRNA-seq data identified TGFB1, p53, MYC, PDGF-BB, EGFR, and WNT3A/CTNNB1 as important upstream regulators underlying fibroblast activation in db/db hearts. Comparison of the transcriptome of fibroblasts from db/db mice with fibroblast-specific Smad3 loss and db/db Smad3 fl/fl controls identified Thbs4 [encoding thrombospondin-4 (TSP-4), a marker of activated fibroblasts] as a candidate diabetes-induced fibrogenic mediator. However, in vitro experiments showed no significant activating effects of matricellular or intracellular TSP-4 on cardiac fibroblasts.

Conclusion

Fibroblast-specific TGF-β/Smad3 signaling mediates ventricular fibrosis, hypertrophy, and dysfunction in Type 2 diabetes.

Graphical Abstract

Graphical Abstract.

Time of primary review: 27 days

1. Introduction

The increased risk of heart failure in patients with diabetes¹ is not explained only by the high incidence of macrovascular complications that may increase the severity of hypertension, or cause early-onset atherosclerotic coronary disease,² but also reflects the development of a distinct cardiomyopathic condition, termed diabetic cardiomyopathy.^3–9 Myocardial interstitial and perivascular fibrosis is a prominent characteristic of diabetic cardiomyopathy and may contribute to the pathogenesis of diastolic dysfunction, thus increasing the prevalence of heart failure with preserved ejection fraction (HFpEF) in subjects with diabetes.^10–13 Moreover, fibrotic cardiac remodeling may be implicated in the increased incidence of both atrial and ventricular arrhythmias in patients with diabetes.¹⁴ Despite its potential significance, the molecular basis of diabetic cardiac fibrosis remains poorly understood.

Transforming growth factor (TGF)-β, a central effector of the fibrotic response, potently activates fibroblasts by stimulating collagen synthesis, by promoting fibroblast to myofibroblast conversion and by driving acquisition of a matrix-preserving fibroblast phenotype, associated with anti-protease expression and secretion of collagen crosslinking enzymes.¹⁵ Fibrogenic actions of TGF-β involve a series of intracellular effectors, the Smads^16–18 or Smad-independent pathways.¹⁹ Diabetes has been consistently associated with activation of TGF-β cascades in several different tissues, including the kidney, liver, and heart.^20–24 Several distinct mechanisms may stimulate TGF-β signaling in diabetic subjects. First, the diabetes-associated activation of neurohumoral angiotensin (AGT) II signaling may up-regulate a synthesis of TGF-β isoforms in peripheral tissues.¹⁵ Secondly, a high-glucose environment has been suggested to activate TGF-β signaling, not only by inducing TGF-β isoform transcription, secretion, and activation by various cell types,^25–27 but also by increasing expression and cell surface localization of Type 1 and Type 2 TGF-β receptors.²⁸ Thirdly, activation of oxidative pathways and up-regulation of matricellular proteins in diabetic subjects may promote the release of bioactive TGF-β from latent stores.²⁹ Despite extensive evidence documenting activation of the TGF-β system in diabetic tissues,^22,30 whether fibroblast-specific activation of TGF-β signaling mediates diabetes-associated fibrotic remodeling has not been investigated.

In the current study, we investigated for the first time the role of fibroblast-specific TGF-β signaling in diabetic hearts and we dissected the downstream signaling pathways responsible for TGF-β actions. We generated lean and Type 2 diabetic db/db mice with fibroblast-specific loss of TGFBR2/TbR2, the only Type 2 receptor-mediating signaling through all the three TGF-β isoforms. Fibroblast-specific loss of TGF-β signaling attenuated fibrotic remodeling, cardiomyocyte hypertrophy, and dysfunction in db/db mice. Fibroblast-specific Smad3 loss phenocopied the protective effects of fibroblast TbR2 loss in db/db mice, suggesting that in diabetic hearts, fibroblast-driven, TGF-β-mediated fibrosis and dysfunction are dependent on Smad3. Comparison of the transcriptional profile of cardiac fibroblasts from Type 2 diabetic fibroblast-specific Smad3 KO (FSmad3KO) mice and corresponding controls identified Thbs4 (encoding the matricellular protein TSP-4), a previously identified marker of activated fibroblast clusters^31,32 as a candidate gene that may mediate diabetes-associated fibroblast activation. However, TSP-4 stimulation and overexpression did not significantly modulate fibroblast activity.

2. Methods (detailed methodology is provided in the online supplement)

2.1. Generation of lean and Type 2 diabetic fibroblast-specific TbR2 and Smad3 KO mice

All animal experiments were performed according to the animal experimental guidelines issued by the Animal Care and Use Committee at Albert Einstein College of Medicine and conform to the Guide for the Care and Use of Laboratory Animals published by the National Institutes of Health. Both male and female mice were studied. Tamoxifen-inducible Col1a2CreER (#029567, The Jackson Laboratory, Bar Harbor, ME, USA) mice were bred in a db/+ C57BL/6J background (#000697, Jackson) and then were crossed with TbR2fl/fl (#012603, Jackson), in which the exon 4 is flanked by loxP sites. As a result, after tamoxifen injection, four genetically different animal groups were obtained: TbR2 fl/fl lean (n = 8; 4 males and 4 females), TbR2 fl/fl db/db mice (n = 13, 5 males and 8 females), lean fibroblast-specific TbR2 knockouts (FTbR2KO, n = 9, 3 males and 6 females), and db/db FTbR2KO (n = 7; 4 males and 3 females) (see Supplementary material online, Figure I). Similar breeding protocols were used to generate Smad3 fl/fl lean (n = 7; 4 males and 3 females), Smad3fl/fl dbdb (n = 17, 6 males and 11 females), lean FSmad3KO (n = 9; 5 males and 4 females), and db/db FSmad3KO (n = 7, 4 males and 3 females) (see Supplementary material online, Figure II). Intraperitoneal injections of tamoxifen (Sigma-T5648, CAS#10540-29-1) to abrogate TbR2 or Smad3 in Col1a2^CreER; TbR2 fl/fl or Col1a2^CreER; Smad3 fl/fl mice in lean or diabetic background were administered twice at 8 and 16 weeks of age at a dosage of 100 mg/kg (in order to ensure continuous targeting of fibroblasts that may originate from other cellular sources), daily over a course of five consecutive days. Following echocardiographic measurements at the age of 6 months, the animals were sacrificed for subsequent histological and transcriptomic experiments. Mice were euthanized using 2% inhaled isoflurane followed by cervical dislocation. The timeline of echocardiographic, histological, and cell biological studies in lean and obese fibroblast-specific KO and control animals is illustrated in Supplementary material online, Figure III.

2.2. 2D Echocardiography and strain echocardiography by speckle tracking

Echocardiographic studies were performed in all mice at the age of 6 months using the Vevo 2100 and 3100 systems (VisualSonics Inc., Toronto, ON, Canada). Speckle-tracking echocardiography was performed to assess systolic and diastolic function. The echocardiographic offline analysis was performed by a sonographer blinded to the study group.

2.3. Histological and quantitative analyses of collagen content

For histopathological analysis, murine hearts from 6-month-old mice were fixed in zinc formalin (Z-fix; Anatech, Battle Creek, MI, USA) and embedded in paraffin. Based on the protocols used in our laboratory, 10 sequential 5 μm thick sections were cut from base to apex at 300 μm intervals. For the assessment of collagen content and cardiomyocyte size, slides from mid-ventricular heart sections were used (five fields per animal). Collagen fibres were identified by picrosirius red staining using protocols established in our laboratory.³³ In diabetic hearts, fibrotic expansion of the interstitium is associated with concomitant cardiomyocyte hypertrophy that may outpace the increase in the collagen-positive area, making traditional assessment of collagen content as the percentage of the myocardial area occupied by collagen problematic. For this reason, we used two different methods for the assessment of collagen content. First, we quantitatively assessed the thickness of perimysial collagen fibres in lean and db/db hearts. Secondly, we quantitatively assessed the volume of collagen in each heart based on measurements of heart weight (HW) and of the collagen-stained area in myocardial sections (see Supplementary material online, Figure IV).

2.4. Wheat germ agglutinin lectin histochemistry and quantification of cardiomyocyte size and endomysial connective tissue thickness

Cardiomyocyte size was assessed using sections stained with Alexa Fluor 594-conjugated wheat germ agglutinin (WGA) lectin (1:100, #W11262, Invitrogen, Waltham, MA, USA), and the nuclei were counterstained with 4, 6-diamino-2-phenylindole (#17985-50, Electron Microscopy Sciences, Hatfield, PA, USA). The average peri-cardiomyocyte endomysial matrix area was also calculated (see Supplementary material online, Figure V).

2.5. FACS sorting of fibroblasts from lean and db/db hearts with functional Smad3 and Smad3 knockout

To study the effects of fibroblast-specific Smad3 loss on fibroblast gene expression profile, we sorted CD31⁻/CD45⁻/PDGFRα⁺ cardiac fibroblasts from lean and db/db Smad3 fl/fl and FSmad3KO mice at 6 months of age. Single-cell suspensions of interstitial cells for flow cytometry were prepared using a previously described protocol.³⁴ Fibroblasts were identified as non-endothelial cells (CD31⁻), non-hematopoietic (CD45⁻) cells expressing PDGFRα and were sorted with the FACSAria Sorter (BD Biosciences). FlowJo software was used for data analysis. RNA extracted from the sorted fibroblasts was used for RNA-Seq.

2.6. Cardiac fibroblast isolation and culture

Fibroblasts were isolated from 12-week-old mouse (C57/BL6J) hearts with protocols used in our laboratory.³⁴

2.7. Stimulation of cultured cardiac fibroblasts with TSP-4

Mouse cardiac fibroblasts at passage 2 were cultured in the presence or absence of recombinant mouse TSP-4 (10 ng/mL, 7860-TH-050, R&D Systems, Minneapolis, MN, USA) for 6 h. At the end of the experiment, RNA extraction was performed.

2.8. TSP-4 overexpression in cultured cardiac fibroblasts

Transformed Escherichia coli bacteria in stab culture were used as a source of plasmids: (i) plasmid containing murine THBS4 [pAAV.CMV.SV40.THBS4-HA.SV40(polyA), #155193, Addgene, Watertown, MA, USA] and (ii) empty plasmid serving as a control (pAAV-CA, #69616, Addgene). After transfection, medium with plasmid was discarded and replaced by standard medium without antibiotics for 12 h. Thereafter, cells were serum deprived for 24 (for RNA) or 48 h (for proteins) and collected for subsequent experiments. Efficiency of TSP-4 overexpression was confirmed using western blotting.

2.9. Library preparation for transcriptome sequencing

RNA isolation was performed in sorted cardiac fibroblasts from 6-month-old mice (four groups: Smad3fl/fl lean, FSmad3KO lean, Smad3fl/fl dbdb, and FSmad3KO dbdb) and in cultured cardiac fibroblasts stimulated with recombinant TSP-4 (two groups: TSP-4 stimulation vs. control) or treated with plasmid to overexpress TSP-4 (two groups: TSP-4 OE vs. control) using a TRIzol-based method (Invitrogen). RNA samples were sent to Novogene (Sacramento, CA, USA) to construct cDNA libraries using the NEBNext® Ultra™ RNA Library Prep Kit for Illumina® (NEB, Ipswich, MA, USA). In brief, the process of library construction consisted of (i) mRNA purification and enrichment from total RNA using oligo(dt)-attached magnetic beads, (ii) fragmentation of purified mRNA using divalent cations exposed to elevated temperatures in NEBNext first-strand synthesis reaction buffer, (iii) double-stranded cDNA synthesis, using RNase H–reverse transcriptase (first strand) and DNA polymerase I, dNTP, and RNase H (second strand); (iv) terminal cDNA ends reparation by exonucleases/polymerases and polyadenylation of the 3′ ends of the DNA fragments, (v) sequencing adaptors ligation, and (vi) cDNA fragments size selection (150–200 bp in length) which underwent polymerase chain reaction (PCR). PCR was performed using Phusion High-Fidelity DNA Polymerase, universal PCR primers, and Index (X) Primer. PCR products were purified (AMPure XP System, Beckman Coulter, Inc, Brea, CA, USA), and the library concentration (> 2 nM) and quality were assessed using a Bioanalyzer 2100 system (Agilent, Santa Clara, CA, USA).

2.10. Quality analysis, mapping, and assembly

The library preparations were sequenced on Illumina Novaseq 6000 devices, generating 150 bp paired-end reads. Adapter, poly-N, and low-quality reads from the raw data were excluded to purify the data analysis.

2.11. Gene expression, differential expression, enrichment, and co-expression analysis

HTSeq software v.0.6.1³⁵ was used to count the number of reads mapped to each gene. Read count of fragments per kilobase of transcript sequence per millions base pairs sequenced was used to calculate gene expression level, which considered the effects of both sequencing depth and gene length.³⁶ Read count obtained from the gene expression analysis was used for differential expression analysis. Cluster differential expression analysis for every gene in different cardiac fibroblast conditions was performed using the DESeq2 R software package (v.1.10.1).³⁷ A list of genes was ranked by differential gene expression as log2(fold change) between each comparison group. Positive values were up-regulated genes, whereas negative values were down-regulated genes. Top biological functions and canonical pathways associated with the differentially expressed mRNAs dataset were identified with ingenuity pathway analysis (IPA) (Qiagen, Germantown, MD). By comparing the imported RNA-Seq data generated in the Agilent platform with Ingenuity® Knowledge Base, a list of relevant networks, upstream regulators, and algorithmically generated mechanistic networks based on their connectivity was obtained. A score [P-score = − log10 (P-value)] according to the fit of the set of supplied genes and a list of biological functions stored in the Ingenuity Knowledge Base were generated. All RNA-Seq processed data have been deposited in NCBI’s Gene Expression Omnibus (GEO) and are accessible through GEO SuperSeries accession number GSE241180.

2.12. Analysis of scRNA-seq data from db/db and db/+ mice

In order to correlate the transcriptional signature of db/db fibroblasts from our bulk RNA-Seq to the transcriptome of db/db fibroblasts from scRNA-seq studies, we analysed published scRNA-seq data of cardiac interstitial cells from 4-month-old db/db and db/+ mice.³⁸ Raw sequencing reads in FASTQ format were processed using Cell Ranger v5.0.0 (10 × Genomics) to align the reads to the reference mouse genome and quantify transcriptional expression for individual cells. A subsequent analysis of scRNA-seq data was performed in R v4.1 using the Seurat v5 package. Pre-processing, including filtering and normalization, and scRNA-seq analysis workflow including dimensionality reduction, cell clustering, annotation, and differential expression testing were conducted following the methods originally described by Cohen et al.³⁸ Differentially expressed genes (DEGs) from our RNA-Seq analysis of CD31⁻/CD45⁻/PDGFRα⁺ cardiac fibroblasts were correlated with DEGs of fibroblasts as identified in the scRNA-seq analysis. IPA was performed to identify signaling pathways and upstream regulators that are enriched in db/db cardiac fibroblasts.

2.13. Protein extraction and western blotting

Protein from cultured cardiac fibroblasts from TbR2fl/fl and FTbR2KO lean animals and from cultured cardiac fibroblasts treated by a plasmid with either THBS4 DNA or an empty plasmid was extracted in RIPA lysis buffer, including 0.5 M ethylenediaminetetraacetic acid, 1 × protease, and phosphatase inhibitors. The following antibodies were used for western blotting: anti-phospho-TbR2 (phospho S225/S250, 1:500, #ab111564, Abcam), anti-TSP-4 (1:200, #AF2390-SP, R&D Systems), and anti-GAPDH (1:4000, #PA1-987, Invitrogen).

2.14. Statistical analysis

Sample size estimates were based on our own experience with comparable experimental studies to achieve 90% power at a significance level of 0.05. For all analyses, normal distribution was tested using the Shapiro–Wilk normality test. For comparisons of two groups, an unpaired two-tailed Student’s t-test using (when appropriate) Welch’s correction for unequal variances was performed. The Mann–Whitney U test was used for comparisons between two groups that did not show Gaussian distribution. For comparisons of multiple groups, one-way ANOVA was performed followed by Sidak’s multiple comparison test for normal distributions or the Kruskal–Wallis test followed by Dunn’s multiple comparison test for non-Gaussian distributions. Data are expressed as means ± S.E.M. Statistical significance was set at 0.05. Statistical analysis was performed with GraphPad Prism 9.4.

3. Results

3.1. Fibroblast-specific loss of TGF-β does not affect weight gain in db/db mice

Col1a2^CreER; Tgfbr2 fl/fl mice (FTbR2KO) and corresponding Tgfbr2 fl/fl controls were injected with tamoxifen at 2 and 4 months of age. To document fibroblast-specific disruption of TGF-β signaling, fibroblasts were harvested from FTbR2KO and Tgfbr2 fl/fl hearts 4 days after the last tamoxifen injection, and protein was isolated. Western blotting for p-TbR2 showed abrogated constitutive Type 2 TGF-β receptor signaling in FTbR2 KO cardiac fibroblasts (Figure 1A and B). To generate Type 2 diabetic mice with fibroblast-specific disruption of TGF-β signaling, FTbR2 KO mice and Tgfbr2 fl/fl controls mice were bred in a db/db C57Bl/6J background. Fibroblast-specific Tgfbr2 loss did not affect body weight (BW) and tibial length (TL) in both lean and db/db obese diabetic mice at 6 months of age (Figure 1C and D).

Figure 1.

Fibroblast-specific loss of TbR2 attenuates diabetes-associated cardiac hypertrophy without affecting BW. (A) Western blotting for phospho-TbR2 (phospho S225/S250) documents the loss of TbR2 in mice with fibroblast-specific TbR2 loss (FTbR2KO mice). (B) Quantitative analysis of p-TbR2 levels (**P < 0.01, n = 4/group, Student’s t-test with Welch’s correction). (C) Fibroblast-specific TbR2 loss does not affect BW in lean and db/db mice. (D) TL was comparable between groups. (E, F) Fibroblast-specific TbR2 loss decreased HW and HW/TL in db/db mice without any effects in lean mice. (G, H) In contrast, there were no significant effects on LW and LW/TL between groups. For body, heart, and lung weights, TbR2fl/fl lean: n = 8, FTbR2KO lean: n = 9, TbR2fl/fl db/db: n = 13, FTbR2KO db/db: n = 7. For TL, HW/TL, and LW/TL, TbR2fl/fl lean: n = 6, FTbR2KO lean: n = 7, TbR2fl/fl db/db: n = 13, FTbR2KO dbdb: n = 6. **P < 0.01; ^^^^P < 0.0001 lean vs. corresponding db/db mice. Statistical analysis was performed using Kruskal–Wallis test followed by Dunn’s multiple comparison test (D–G) and one-way ANOVA test followed by Sidak’s multiple comparison test (C, H).

3.2. Fibroblast-specific TGF-β signaling disruption significantly reduces HW in obese diabetic db/db mice

Fibroblast-specific loss of TGFBR2 significantly reduced HW and HW/TL in obese diabetic mice, but not in lean animals (Figure 1E and F). No significant differences in lung weight (LW) and in LW/TL were noted between groups (Figure 1G and H). Moreover, serum glucose levels were not significantly affected by fibroblast-specific TGFBR2 loss (see Supplementary material online, Figure VI).

3.3. Fibroblast-specific TGF-β signaling preserves systolic function in lean animals but mediates a modest depression in systolic function in obese diabetic mice

Echocardiographic analysis showed that fibroblast-specific TGFBR2 loss does not have significant effects on left ventricular end-diastolic volume (LVEDV) and left ventricular end-systolic volume (LVESV) in lean or obese diabetic mice (Figure 2A and B). Loss of TGFBR2 in fibroblasts reduced end-diastolic left ventricular anterior wall thickness in obese diabetic mice (Figure 2C), without significant effects on LV mass (Figure 2D). Lean and db/db mice had comparable ejection fraction that was not significantly affected by fibroblast-specific TGFBR2 loss (Figure 2E). Although db/db mice have preserved ejection fraction, they exhibit subtle evidence of systolic dysfunction that can be detected using speckle-tracking echocardiography (Figure 2F–L). db/db mice had a significant reduction in global longitudinal strain (GLS, Figure 2J), peak systolic longitudinal endocardial strain (PSLEndoS, Figure 2K), and peak systolic longitudinal epicardial strain (PSLEpiS, Figure 2L). Fibroblast-specific loss of TGFBR2 reduced GLS, and PSLEndoS in lean animals, suggesting a modest deterioration of systolic myocardial function. In contrast, GLS, PSLEndoS, and PSLEpiS were increased by fibroblast-specific loss of TGFBR2 in obese db/db mice, suggesting attenuated systolic dysfunction. Considering the increased activation of TGF-β signaling cascades in db/db hearts,²² the findings may reflect the requirement for tight regulation of TGF−β signaling for preservation of homeostatic ventricular function. The basal level of fibroblast TGF-β signaling in normal lean mouse hearts may be required for preservation of systolic function, whereas excessive fibroblast TGF-β signaling in diabetic mice promotes systolic dysfunction.

Figure 2.

Fibroblast-specific disruption of TGF-β signaling attenuates systolic dysfunction in db/db mice. (A–E) 2D echocardiography showed that fibroblast-specific TbR2 loss did not affect LVEDV (A) and LVESV (B) in 6-month-old lean and db/db mice. db/db mice lacking TGF-β signaling in fibroblasts (db/db; FTbR2KO) had decreased LVAWTd (C). However, LV mass (D) and LVEF (E) were not significantly affected by fibroblast-specific TbR2 loss. No effects of fibroblast-specific TbR2 loss were noted in lean animals. (F–I) Representative images of systolic longitudinal endocardial strain (%) using speckle-tracking echocardiography. The six curves on the images represent six different regions of the LV: posterior basal, posterior mid-ventricular, posterior apical, anterior basal, anterior mid-ventricular, and anterior apical. The black curve represents an average strain calculated for each animal. An arrow shows the average PSLEndoS value. (J–L) Quantitative analysis shows that, despite exhibiting a normal ejection fraction, db/db mice have systolic dysfunction, evidenced by reduced absolute GLS (J) and decreased PSLEndoS (K) and peak systolic longitudinal epicardial (PSLepi) strain (L), in comparison with lean animals. Fibroblast-specific TbR2 loss attenuates systolic dysfunction in db/db mice, increasing GLS, PSLendo, and PSLepi. In contrast, in lean mice fibroblast-specific TbR2 loss is associated with decreased GLS, and PSLendo, without significantly affecting PSLepi strain, suggesting subtle systolic dysfunction. Taken together, the findings suggest that fibroblast TGF-β signaling mediates systolic dysfunction in obese diabetic mice. For (A–C, E): TbR2fl/fl lean, n = 8; FTbR2KO lean, n = 9; TbR2fl/fl db/db, n = 13; FTbR2KO db/db, n = 7. For (D) TbR2fl/fl lean, n = 7; FTbR2KO lean, n = 9; TbR2fl/fl db/db, n = 13; FTbR2KO db/db, n = 7. For (J–L) TbR2fl/fl lean, n = 7; FTbR2KO lean, n = 9; TbR2fl/fl db/db, n = 7; FTbR2KO db/db, n = 6. *P < 0.05, **P < 0.01, ***P < 0.001. Statistical analysis was performed using one-way ANOVA test followed by Sidak’s multiple comparison test (A–C, E, L) and the Kruskal–Wallis test followed by Dunn’s multiple comparison test (D, J, K).

3.4. Fibroblast-specific TGF-β signaling preserves diastolic function in lean mice but mediates diastolic dysfunction in obese diabetic mice

Speckle-tracking echocardiography was used to examine the effects of fibroblast-specific TGF-β signaling on diastolic function (Figure 3A–F) and demonstrated significant effects of fibroblast TGF-β signaling in the pathogenesis of diastolic dysfunction in obese diabetic db/db mice. db/db mouse hearts exhibited marked reductions in peak diastolic longitudinal endocardial and epicardial strain rates (PDLEndoSR and PDLEpiSR, Figure 3E and F). In lean mouse hearts, fibroblast-specific TGFBR2 loss significantly reduced PDLEndoSR and PDLEpiSR. In db/db mice, TGFBR2 disruption in fibroblasts increased PDLEndoSR; however, the effects on PDLEpiSR did not reach statistical significance (Figure 3E and F). The findings may be consistent with a homeostatic role of fibroblast-specific TGF-β signaling in lean mouse hearts that preserves diastolic function. In contrast, in diabetic hearts, accentuation of TGF-β signaling in cardiac fibroblasts mediates diastolic dysfunction. To exclude effects of Cre on cardiac function, we performed echocardiographic imaging and speckle-tracking echocardiography in Col1a2CreER mice and Cre-negative controls at 6 months of age, after tamoxifen injections at 2 and 4 months of age. No significant effects of fibroblast-specific Cre expression on cardiac systolic and diastolic function were noted (see Supplementary material online, Figure VII).

Figure 3.

Fibroblast-specific disruption of TGF-β signaling attenuates diastolic dysfunction, interstitial fibrosis and cardiomyocyte hypertrophy in db/db mice, while worsening diastolic cardiac function in lean animals. (A–D) Representative images of diastolic longitudinal endocardial strain rate (1/s), using speckle-tracking echocardiography. The six curves on the images represent six different regions of the LV: posterior basal, posterior mid-ventricular, posterior apical, anterior basal, anterior mid-ventricular, and anterior apical. The black curve represents an average strain rate calculated for each animal. An arrow shows the average PDLEndoSR value. (E, F) Quantitative analysis shows that db/db mice have diastolic dysfunction, evidenced by reduced PDLendoSR and PDLepiSR in comparison with lean animals. Fibroblast-specific TbR2 loss attenuates diastolic dysfunction in db/db mice, increasing PDLEndoSR. In contrast, the effect on PDLepi did not reach statistical significance. In contrast, in lean mice fibroblast-specific TbR2 loss is associated with decreased PDLendoSR and PDLepiSR, suggesting impaired diastolic function. (G–I) Representative images with picrosirius red-stained sections from lean TbR2 fl/fl controls, lean FTbR2KO, db/db; TbR2 fl/fl controls, and db/db; FTbR2 KO mouse hearts and their quantitative analysis showed that db/db hearts exhibited thickening of perimysial collagen fibres, in comparison with lean controls and the fibroblast-specific loss of TGF-β signaling reduced perimysial collagen thickness (H) and collagen volume (normalized to TL) in diabetic mice (I). (J–L) Representative images of WGA lectin-stained sections from lean TbR2 fl/fl, db/db; TbR2 fl/fl, lean FTbR2KO, and db/db; FTbR2KO mice with respective quantitative analysis showed that db/db mice had increased cardiomyocyte size; the increase was attenuated in db/db mice with fibroblast-specific disruption of TGF-β signaling. WGA staining also labels the endomysium which was thickened in db/db mice. Fibroblast-specific TbR2 loss attenuated endomysial thickness in db/db mice. For (E, F) TbR2fl/fl lean, n = 7; FTbR2KO lean, n = 9; TbR2fl/fl db/db, n = 7; FTbR2KO db/db, n = 6. For (H, K, L) TbR2fl/fl lean, n = 8; FTbR2KO lean, n = 9; TbR2fl/fl db/db, n = 13; FTbR2KO db/db, n = 7. For (I) TbR2fl/fl lean, n = 6; FTbR2KO lean, n = 7; TbR2fl/fl db/db, n = 13; FTbR2KO db/db, n = 6. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. For (G) scale bar = 100μm; for (J) scale bar = 30μm. Statistical analysis was performed using one-way ANOVA test followed by Sidak’s multiple comparison test (E, F, I, K, L), and the Kruskal–Wallis test followed by Dunn’s multiple comparison test (H).

Taken together (Figures 2 and 3), the findings highlight the importance of tightly regulated TGF-β-mediated fibroblast activity in the myocardium. In normal lean animals, basal levels of fibroblast-specific TGF-β signaling preserve function and geometry, whereas in obese diabetic mice the increase in fibroblast TGF-β signaling promotes dysfunction and adverse remodeling.

3.5. Fibroblast-specific TGF-β signaling promotes interstitial fibrosis and contributes to hypertrophic remodeling of the diabetic heart

When compared with lean animals, db/db mice exhibit cardiomyocyte hypertrophy accompanied with increased collagen deposition.³³ Hypertrophic and fibrotic changes may play an important role in the pathogenesis of diabetes-associated diastolic dysfunction. Accordingly, we examined whether the protective effects of fibroblast-specific TGFBR2 loss on diabetes-associated cardiac dysfunction are due to attenuation of fibrosis and/or cardiomyocyte hypertrophy. Picrosirius red staining was used to label collagen fibres in lean and db/db mouse hearts (Figure 3G–I). Quantitative analysis showed that when compared with lean littermates, db/db mice have a marked increase in perimysial collagen thickness (Figure 3H). Fibroblast-specific TGFBR2 loss did not affect perimysial collagen thickness in lean mouse hearts but abrogated the diabetes-associated increase in perimysial collagen thickness (Figure 3H) and collagen content (Figure 3I).

WGA lectin staining showed a marked increase in cardiomyocyte size in db/db mouse hearts in comparison with lean littermates (Figure 3J and K). Fibroblast-specific loss of TGF-β signaling did not affect cardiomyocyte size in lean hearts, but attenuated diabetes-associated cardiomyocyte hypertrophy (Figure 3K). Moreover, WGA lectin staining intensely labels the endomysial interstitium, allowing quantitative assessment of the thickness of the endomysial matrix (Figure 3J and L). When compared with lean WT mice, db/db animals had increased endomysial thickness that was attenuated upon fibroblast-specific TGFBR2 loss (Figure 3L) Thus, fibroblast TGF-β signaling mediates fibrotic and hypertrophic remodeling in diabetic hearts.

3.6. Fibroblast-specific activation of the Smad3 cascade mediates dysfunction and hypertrophy in db/db mice

The fibrogenic effects of TGF-β may involve both Smad-dependent and Smad-independent pathways that exert distinct effects on fibroblast phenotype.³⁹ Extensive evidence suggests a central role for Smad3 in activation of cardiac fibroblasts.¹⁶ In db/db mice, global partial loss of Smad3 in Smad3^+/− haploinsufficient mice attenuated cardiac fibrosis.²² To examine whether diabetes-associated myocardial fibrosis involves activation of Smad3 signaling in fibroblasts, we generated lean and obese FSmad3KO mice using the Col1a2^CreER driver (see Supplementary material online, Figure II). Fibroblast-specific Smad3 loss in lean FSmad3KO mice has been previously documented by our group.⁴⁰

Fibroblast-specific Smad3 loss had no significant effects on the marked increase in BW noted in db/db mice and did not affect TL (see Supplementary material online, Figure VIII). Obese diabetic db/db mice had a significant increase in HW and in the HW/TL ratio in comparison with lean animals. Fibroblast-specific Smad3 loss was associated with a significant reduction in HW/TL; in contrast, effects on non-normalized HW did not reach statistical significance. LW, LW/TL, and plasma glucose levels were not affected by fibroblast-specific Smad3 loss (see Supplementary material online, Figure VIII).

Fibroblast-specific Smad3 loss did not affect LVEDV and LVESV in lean and obese diabetic db/db mice (Figure 4A and B). Left ventricular anterior wall thickness in diastole (LVAWTd) and LV mass were significantly increased in db/db mice, when compared with lean controls (Figure 4C and D), indicating cardiac hypertrophy. Fibroblast-specific Smad3 loss significantly reduced LVAWTd in db/db mice without any significant effects in lean animals (Figure 4C). The effects of fibroblast Smad3 loss on LV mass did not reach statistical significance (Figure 4D). Left ventricular ejection fraction (LVEF) was preserved in all groups (Figure 4E). Speckle-tracking echocardiography (Figure 4F–I) showed that despite the preservation of ejection fraction, db/db mice have systolic dysfunction, exhibiting a significant reduction in GLS (Figure 4J) and in PSLEndoS(Figure 4K). Effects on PSLEpiS did not reach statistical significance (Figure 4L). In lean mice, fibroblast-specific loss of Smad3 reduced GLS, PSLEndoS, and PSLEpiS, consistent with a modest deterioration of systolic myocardial function. In contrast, obese diabetic db/db FSmad3KO mice exhibited increased GLS and PSLEndoS in comparison with floxed controls, suggesting that fibroblast Smad3 loss attenuates diabetes-associated systolic dysfunction. Thus, in normal lean mouse hearts, basal fibroblast Smad3 signaling preserves systolic function, whereas in diabetic hearts, the accentuation of Smad3 mediates systolic dysfunction. Thus, the consequences of fibroblast Smad3 loss on systolic function (Figure 4) phenocopy the effects of fibroblast TGFBR2 loss (Figure 2), consistent with the notion that TGF-β actions in diabetic cardiac fibroblasts involve the Smad3 signaling pathway.

Figure 4.

Fibroblast-specific Smad3 loss phenocopies the effects of fibroblast-specific TbR2 disruption on left ventricular systolic function. (A–E) 2D echocardiography showed that fibroblast-specific Smad3 loss did not have significant effects on LVEDV (A) and LVESV (B) in 6-month-old lean and db/db mice. Diabetic mice lacking Smad3 in fibroblasts (db/db; Smad3KO) had decreased LVAWTd (C), but comparable LV mass (D) with db/db; Smad3 fl/fl controls. LVEF (E) was comparable between groups. No effects of fibroblast-specific Smad3 loss were noted in lean animals. (F–I) Representative images of systolic longitudinal endocardial strain (%) using speckle-tracking echocardiography. The six curves on the images represent six different regions of the LV: posterior basal, posterior mid-ventricular, posterior apical, anterior basal, anterior mid-ventricular, and anterior apical. The black curve represents an average strain calculated for each animal. An arrow shows the average PSLEndoS value. (J–L) Quantitative analysis shows that, despite exhibiting a normal ejection fraction, db/db mice have systolic dysfunction, evidenced by reduced absolute GLS and decreased PSLEndoS (K) in comparison with lean animals. Fibroblast-specific Smad3 loss attenuated systolic dysfunction in db/db mice, significantly increasing GLS, and PSLendo. Effects on PSLepi did not reach statistical significance (L). In contrast, in lean mice fibroblast-specific Smad3 loss is associated with decreased GLS, PSLendo, and PSLepi, suggesting subtle systolic dysfunction. Taken together with the data shown in Figure 2, the findings suggest that the effects of TGF-β signaling in lean and diabetic hearts predominantly involve Smad3 signaling. For (A–E) Smad3fl/fl lean, n = 7; FSmad3KO lean, n = 8; Smad3fl/fl db/db, n = 13; FSmad3KO db/db, n = 7. For (J–L) Smad3fl/fl lean, n = 7; FSmad3KO lean, n = 7; Smad3fl/fl db/db, n = 10; FSmad3KO db/db, n = 7. *P < 0.05, **P < 0.01, ^^P < 0.01 vs. corresponding lean. Statistical analysis was performed using one-way ANOVA followed by Sidak’s multiple comparison test (A–C, E, J–L) and Kruskal–Wallis test followed by Dunn’s multiple comparison test (D).

Speckle-tracking echocardiography was used to examine the effects of fibroblast-specific Smad3 signaling on diastolic function (Figure 5A–D) and demonstrated significant effects of fibroblast Smad3 signaling in the pathogenesis of diastolic dysfunction in obese diabetic db/db mice. db/db mouse hearts exhibited marked reductions in PDLEndoSR and PDLEpiSR, consistent with diabetes-associated diastolic dysfunction (Figure 5E and F). In lean mouse hearts, fibroblast-specific Smad3 loss significantly reduced PDLEndoSR, without affecting PDLEpiSR. In contrast, in db/db mice, Smad3 disruption in fibroblasts increased PDLEndoSR (Figure 5E). The findings suggest that fibroblast-specific Smad3 signaling plays a homeostatic role in lean mouse hearts, preserving diastolic function, whereas in diabetic hearts, Smad3 signaling in cardiac fibroblasts mediates diastolic dysfunction. The effects of fibroblast-specific Smad3 loss (Figure 5) phenocopy the effects of TGFBR2 deletion (Figure 3), suggesting that in diabetic hearts the effects of TGF-β on diastolic function are predominantly mediated via Smad3.

Figure 5.

Fibroblast-specific Smad3 loss phenocopies the effects of fibroblast-specific TbR2 disruption on left ventricular diastolic function, collagen deposition, and cardiomyocyte size. (A–D) Representative images of diastolic longitudinal endocardial strain rate (1/s), using speckle-tracking echocardiography. The six curves on the images represent six different regions of the LV: posterior basal, posterior mid-ventricular, posterior apical, anterior basal, anterior mid-ventricular, and anterior apical. The black curve represents an average strain rate calculated for each animal. An arrow shows the average PDLendoSR value. (E, F) Quantitative analysis shows that db/db mice have diastolic dysfunction, evidenced by reduced PDLendoSR (E) and PDLepiSR (F) in comparison with lean animals. Fibroblast-specific Smad3 loss attenuates diastolic dysfunction in db/db mice, increasing PDLendoSR. In contrast, in lean mice, fibroblast-specific Smad3 loss is associated with decreased PDLendoSR, suggesting impaired diastolic function. (G–I) Picrosirius red staining (scale bar = 80μm) with respective quantitative analyses shows that fibroblast-specific Smad3 loss in db/db mice attenuates perimysial collagen fibre thickening (G, H) but does not significantly affect total collagen volume (I). Fibroblast Smad3 loss had no significant effects on collagen content or thickness in lean mice. (J–L) WGA staining shows that cardiomyocyte size and endomysial thickness are increased in db/db mice (J) in comparison with lean controls (scale bar = 30μm). Fibroblast-specific Smad3 loss in db/db animals attenuated cardiomyocyte hypertrophy and decreased endomysial thickness (J–L). No significant effects of fibroblast Smad3 loss were noted in lean animals. Taken together with the observations shown in Figure 3, the findings suggest that the effects of fibroblast-specific TGF-β signaling on diabetes-associated diastolic dysfunction, fibrosis, and cardiomyocyte hypertrophy are mediated via Smad3. For (E, F) Smad3fl/fl lean, n = 7; FSmad3KO lean, n = 7; Smad3fl/fl db/db, n = 10; FSmad3KO db/db, n = 7. For (H, L) Smad3fl/fl lean, n = 7; FSmad3KO lean, n = 9, Smad3fl/fl db/db, n = 11; FSmad3KO db/db, n = 7. For (K) Smad3fl/fl lean, n = 7; FSmad3KO lean, n = 9; Smad3fl/fl db/db, n = 11; FSmad3KO db/db, n = 6. For (I) Smad3fl/fl lean, n = 7; FSmad3KO lean, n = 8; Smad3fl/fl db/db, n = 10; FSmad3KO db/db, n = 7. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Statistical analysis was performed using one-way ANOVA followed by Sidak’s multiple comparison test.

3.7. Fibroblast Smad3 signaling mediates hypertrophic and fibrotic remodeling of the diabetic heart

Picrosirius red staining was used to investigate the role of fibroblast-specific Smad3 signaling in diabetic cardiac fibrosis (Figure 5G). Quantitative analysis showed that when compared with lean littermates, db/db mice have a marked increase in perimysial collagen thickness (Figure 5H). Fibroblast-specific Smad3 loss did not affect perimysial collagen thickness in lean mouse hearts but attenuated the diabetes-associated increase in perimysial collagen thickness (Figure 5H). However, the effects of fibroblast-specific Smad3 loss on collagen volume did not reach statistical significance (Figure 5I). WGA lectin histochemistry was used to assess cardiomyocyte size (Figure 5J). Fibroblast-specific loss of Smad3 signaling did not affect cardiomyocyte size in lean hearts, but attenuated diabetes-associated cardiomyocyte hypertrophy (Figure 5K). Moreover, fibroblast-specific Smad3 loss attenuated the diabetes-associated increase in the thickness of the WGA-labelled endomysial matrix network (Figure 5J and L). Thus, the findings suggest that both hypertrophic and fibrogenic effects of fibroblast-specific TGF-β signaling in diabetic hearts involve activation of the Smad3 pathway.

Although our study was not statistically powered to identify sex-specific effects of fibroblast TGF-β and Smad3 signaling, we also performed separate analysis of echocardiographic and histological endpoints in male and female mice (see Supplementary material online, Figures IX–XII). Fibroblast-specific TGFBR2 loss attenuated cardiomyocyte hypertrophy in both male (see Supplementary material online, Figure IX) and female (see Supplementary material online, Figure X) db/db mice. Fibroblast-specific TGFBR2 loss reduced perimysial collagen thickness and collagen volume/TL in female mice; however, in male animals, the effects on collagen thickness and content did not reach statistical significance. Moreover, fibroblast-specific Smad3 loss attenuated cardiomyocyte size, perimysial collagen thickening, and endomysial thickness in male db/db mice (see Supplementary material online, Figure XI); however, in female animals, only the effect on cardiomyocyte size reached statistical significance (see Supplementary material online, Figure XII).

3.8. Db/db cardiac fibroblasts exhibit induction of genes associated with lipid metabolism and extracellular matrix (ECM) organization

To understand the molecular basis for the TGF-β/Smad3-mediated activation of fibroblasts in diabetic hearts, we harvested cardiac fibroblasts from lean and obese FSmad3KO mice and from corresponding floxed controls. RNA-Seq was used to characterize the transcriptomic profile of diabetic mouse fibroblasts in the presence or absence of Smad3. A total of 234 genes were up-regulated, and 190 genes were down-regulated in db/db Smad3 fl/fl mouse cardiac fibroblasts, when compared with lean Smad3 fl/fl controls (see Supplementary material online, Figure XIII). The top-ranked gene up-regulated in diabetic mouse fibroblasts was Fmo2 (log2FC = 1.99, P = 6.39E − 12), encoding flavin-containing monooxygenase 2, a protein involved in the oxidative response and found to serve as a negative regulator of fibrosis in infarcted hearts.⁴¹ The list of the top 50 genes that were differentially regulated in db/db cardiac fibroblasts (see Supplementary material online, Figure XIV) included the extracellular matrix (ECM) genes Fbn2 (encoding fibrillin-2; logFC = 1.64, P = 0.0001), Lox (encoding the matrix crosslinking enzyme lysyl oxidase; logFC = 1.63, P = 0.0004), Spock2 (encoding the proteoglycan testican-2; logFC = 0.88, P = 0.001), and the matricellular genes Spon1 (encoding spondin-1; logFC = 0.93, P = 0.0018) and Thbs4 (encoding thrombospondin-4, a marker of activated fibroblasts; logFC = 1.38, P = 0.0019). All these matrix genes were significantly up-regulated in db/db cardiac fibroblasts. IPA identified pathways differentially modulated in db/db mouse cardiac fibroblasts, including cyclic AMP response element binding protein (CREB) signaling, calcium signaling, and pathways involved in cholesterol biosynthesis (Figure 6A). Downstream effects analysis identified ‘fibrosis’ as a differentially regulated process with a z-score of 1.46 (suggesting predicted activation) and a P-value of 2.8E − 05. Gene ontology (GO) enrichment analysis showed that enriched terms included ‘extracellular matrix’, ‘proteinaceous extracellular matrix’, and ‘steroid metabolic process’ (Figure 6B). Moreover, upstream regulator analysis identified molecular signals that may be involved in the modulation of the transcriptional profile in db/db cardiac fibroblasts. Neurohumoral mediators, cytokines, and growth factors (such as AGT, TGFB1, and TNF), kinases (including MAP2K5 and MAPK7), and nuclear receptors (such as PPARG) were predicted to be potentially important upstream regulators of the transcriptomic profile of diabetic mouse fibroblasts (Figure 6C).

Figure 6.

(A) Differentially regulated pathways in db/db vs. lean mouse cardiac fibroblasts. IPA was used to analyse our RNA-Seq data. Pathways are ranked by P-value. A positive z-score (orange) indicates predicted activation in db/db mouse cardiac fibroblasts, whereas a negative z-score (blue) suggests inhibition. (B) GO enrichment analysis showed that enriched terms related predominantly to steroid biosynthetic and extracellular matrix responses, including ‘extracellular matrix’, ‘proteinaceous extracellular matrix’ and ‘steroid metabolic process’. (C) Upstream regulator analysis identifies molecular signals that may be involved in mediating differential gene expression in db/db vs. lean mouse cardiac fibroblasts. IPA software was used to analyse RNA-Seq data. Upstream regulators are ranked by P-value. A negative z-score (red) indicates predicted inhibition, whereas a positive z-score suggests predicted activation. Neurohumoral mediators (such as AGT), cytokines and growth factors (such as TGFB1 and TNF), kinases (including MAP2K5 and MAPK7), and nuclear receptors (such as PPARG) were predicted to be potentially important upstream regulators of the transcriptomic profile of diabetic mouse fibroblasts. Overlap P-values are calculated using Fisher’s exact test.

Next, we analysed published scRNA-seq data comparing gene expression levels between db/db and lean mouse cardiac fibroblasts harvested at 4 months of age.³⁸ Twenty-nine genes were found to be consistently up-regulated in db/db cardiac fibroblasts in both scRNA-seq and our own bulk RNA-Seq data (Figure 7). The list of consistently up-regulated genes included Fmo2, Fbn2, Spock2, the fibrogenic lncRNA Neat1,⁴²  Eln, Adamtsl2 (encoding a matrix glycoprotein implicated in negative regulation of TGF-β signaling⁴³), and Frzb, encoding the Wnt-binding Frizzled-related protein (Figure 7A). Moreover, 23 genes were found to be consistently down-regulated in db/db cardiac fibroblasts in both scRNA-seq and bulk RNA-Seq data (Figure 7B). IPA identified upstream regulators that may underlie differential gene expression in db/db mouse cardiac fibroblasts (Figure 7C–E). Several fibrogenic mediators were ranked in the top 20 upstream regulators by P-value (Figure 7C) or by z-score of activation (Figure 7D) and inhibition (Figure 7E). These mediators included TGFB1, p53,⁴⁴ MYC, PDGF-BB, EGFR, and WNT3A/CTNNB1 (Figure 7C and D).

Figure 7.

Differentially expressed genes in cardiac fibroblasts harvested from db/db and lean mouse hearts. Published scRNA-Seq data,³⁸ examining gene expression in cells harvested from 4-month-old db/db and corresponding lean control mice, and our own RNA-Seq data examining the transcriptional profile of CD31⁻/CD45⁻/PDGFRα⁺ fibroblasts harvested from 6-month-old db/db and lean control hearts were analysed. (A) 29 genes were found to be consistently up-regulated in db/db cardiac fibroblasts in both scRNA-seq and bulk RNA-Seq data, including Fmo2, Fbn2, Spock2, the fibrogenic lncRNA Neat1, Eln, Adamtsl2, and Frzb. (B) 23 genes were consistently down-regulated in db/db cardiac fibroblasts in both scRNA-Seq and bulk RNA-Seq data. P-values for bulk RNA-Seq data were calculated using the negative binomial distribution model. Seurat’s FindMarkers function with the MAST test was used for differential expression analysis of scRNA-Seq data. The analysis focused on genes with a detectable expression in at least 10% of db/db or lean cells. (C–E) IPA of the scRNA-Seq data identified upstream regulators that may underlie the changes in transcriptional profile noted in db/db mouse cardiac fibroblasts. The top 20 upstream regulators are ranked by P-value (C), positive z-score (D, indicating predicted activation in db/db fibroblasts) and negative z-score (E, predicting inhibition in db/db fibroblasts). IPA identified upstream regulators that may underlie the differential gene expression in db/db mouse cardiac fibroblasts (Figure 7C and D). The list includes several fibrogenic mediators, such as TGFB1, p53, MYC, PDGF-BB, and WNT3A/CTNNB1. Overlap P-values for upstream regulator analysis are calculated using Fisher’s exact test.

Considering the effects of fibroblast-specific TGF-β/Smad3 signaling on cardiomyocyte hypertrophy, we also analysed our RNA-Seq data to identify fibroblast-derived mediators that may regulate cardiomyocyte hypertrophy in diabetic hearts. Several pro-hypertrophic mediators, including Spp1 (encoding osteopontin), Prg4, Itgbl1, Agt (encoding the AGT II precursor, angiotensinogen), and Areg (encoding amphiregulin) were up-regulated in db/db cardiac fibroblasts and, upon secretion, may stimulate cardiomyocyte hypertrophy in a paracrine manner (see Supplementary material online, Table S1). Moreover, db/db cardiac fibroblasts exhibited reduced expression levels of several mediators with anti-hypertrophic properties, including Bmp7, Igfbp6, C1qtnf1, and Fgf18 (see Supplementary material online, Table S2).

3.9. Smad3 mediates expression of the fibrogenic marker Thbs4 in diabetic cardiac fibroblasts

Next, we examined the role of Smad3 in the changes in gene expression noted in diabetic cardiac fibroblasts by comparing the transcriptional profile of cardiac fibroblasts harvested from db/db FSmad3KO and db/db floxed animals. Fibroblast-specific Smad3 loss was associated with up-regulation of 198 and down-regulation of 292 genes (see Supplementary material online, Figure XVA). Thbs4 (encoding TSP-4, a matricellular protein associated with an activated fibroblast phenotype) was the top-ranked down-regulated protein-encoding gene in fibroblasts lacking Smad3 (see Supplementary material online, Figure XVB). Several other fibrosis and ECM-associated genes were also downmodulated in the absence of Smad3, including Col9a1, Col8a1, Col11a2, Col6a5, and Plod2 (encoding a lysyl hydroxylase with an important role in collagen deposition⁴⁵), and the matricellular genes Tnc (encoding tenascin-C), Postn, and Cilp (see Supplementary material online, Figure XVC).

3.10. Effects of stimulation with recombinant TSP-4 on the transcriptional profile of mouse cardiac fibroblasts

Our findings showed that Thbs4 is up-regulated in db/db cardiac fibroblasts and is the top-ranked down-regulated gene upon Smad3 loss (see Supplementary material online, Figure XV). Moreover, two independent studies using scRNA-seq to explore the heterogeneity of fibroblasts in failing hearts demonstrated that Thbs4 characterizes a cluster of activated fibroblasts that may mediate cardiac fibrosis in experimental models of heart failure.^31,32 However, whether Thbs4 expression is an essential driver of the fibrogenic properties of activated fibroblast clusters is not known. Thbs4 has been suggested to act both as a matricellular protein^46,47 and as an intracellular mediator that regulates endoplasmic reticulum stress.⁴⁸ Accordingly, to examine the effects of Thbs4 in fibrogenic activation of fibroblasts, we performed stimulation experiments with recombinant TSP-4 (to investigate outside-in matricellular actions) and TSP-4 overexpression studies (to explore intracellular effects). TSP-4 overexpression was documented using western blotting (see Supplementary material online, Figure XVI). The effects of TSP-4 on the transcriptomic profile of the fibroblasts were studied using RNA-seq. RNA-Seq demonstrated a marked up-regulation of Thbs4 mRNA in TSP-4-overexpressing fibroblasts (see Supplementary material online, Figure XVIB).

Stimulation with recombinant TSP-4 significantly increased expression of 115 genes and down-regulated expression of 175 genes (see Supplementary material online, Figure XVIIA). TSP-4 stimulation did not significantly affect levels of genes encoding the structural collagens and fibronectin (Col1a1, Col3a1, and Fn1) and did not modulate several other genes known to be up-regulated in db/db fibroblasts, such as Timp1, Cmyc, Eng, Thbs1, Lox, and Ccl11.⁴⁹ TSP-4 stimulation induced a modest up-regulation of Col11a1 (a marker of invasive cancer-associated fibroblasts),⁵⁰  Dact3, a gene encoding a β-catenin regulatory protein that has been suggested to inhibit renal fibrosis,⁵¹ and up-regulated expression of the Wnt receptor Fzd2.Conversely, TSP-4 stimulation down-regulated several fibrosis-associated genes, including Serpine1 (encoding fibrogenic PAI-1), the early immediate gene Egr1, involved in TGF-β−induced fibrogenesis,⁵² the fibrogenic growth factors Ereg (encoding epiregulin)⁵³ and HB-EGF, the transmembrane fibroblast activator CD44,⁵⁴ and the matricellular gene Cilp⁵⁵ (see Supplementary material online, Figure XVIIIA).

3.11. Effects of TSP-4 overexpression on the transcriptional profile of mouse cardiac fibroblasts

TSP-4 overexpression in cardiac fibroblasts was associated with up-regulation of 96 genes and down-regulation of 121 genes (see Supplementary material online, Figure XVIIB). There was very limited overlap in the gene expression patterns of TSP-4-overexpressing and TSP-4-stimulated cells, suggesting activation of distinct cascades upon outside-in vs. intracellular signaling. TSP-4 overexpression had no significant effects on structural collagen gene expression, Fn1, or on expression of genes up-regulated in db/db cardiac fibroblasts (see Supplementary material online, Figure XVIIIB). However, TSP-4 overexpression induced expression of Col8a2, the ADAM family member Adam23, and the fibrogenic and pro-hypertrophic growth factor Igf1. In addition, TSP-4 down-regulated expression of the early response gene Egr3 (involved in fibrogenic activation),⁵⁶ and the transcription factors Fos and Klf8 (see Supplementary material online, Figure XVIIIB). Taken together, these experiments showed no significant fibrogenic effects of TSP-4 and suggested that TSP-4 does not play a significant role in mediating the fibrogenic actions of Smad3 signaling in diabetic fibroblasts.

4. Discussion

Although diabetes is associated with expansion of the cardiac interstitium and increased deposition of ECM proteins^57,11,58,59 direct evidence documenting a role for fibroblasts in the pathogenesis of diabetic cardiomyopathy is lacking. Our study uses a fibroblast-specific loss-of-function approach in vivo to demonstrate for the first time a central role for fibroblast activation in the pathogenesis of diabetic cardiomyopathy. Using mice with fibroblast-specific abrogation of TGF-β responses, we demonstrate that fibroblast TGF-β signaling plays a central role in the development of interstitial fibrosis, cardiomyocyte hypertrophy and dysfunction in obese, Type 2 diabetic db/db mice. Fibroblast-specific Smad3 loss phenocopied the effects of TGF-β signaling disruption, suggesting that the fibrogenic and pro-hypertrophic actions of TGF-β in diabetic fibroblasts involve activation of the Smad3 cascade. Transcriptomic analysis showed that diabetic mouse cardiac fibroblasts exhibit induction of matrix crosslinking genes, and alterations in expression of genes involved in the oxidative response and in cholesterol biosynthesis. Moreover, Thbs4, the gene encoding the matricellular protein TSP-4 which has been identified as a marker of fibrogenic fibroblast clusters in models of cardiac fibrosis,^31,32 was identified as a candidate TGF-β-induced gene that may mediate activation of a diabetic fibroblast phenotype. However, TSP-4 stimulation and overexpression experiments did not stimulate a fibrogenic profile in cardiac fibroblasts, suggesting that Thbs4 may be simply a marker, and not a causative mediator, of fibroblast activation. Our studies establish for the first time the central involvement of fibroblast-specific activation of TGF-β signaling in diabetic cardiomyopathy.

4.1. Fibrosis as a central pathologic abnormality in diabetic cardiomyopathy

Several clinical studies have suggested that diabetic cardiomyopathy is associated with diffuse myocardial fibrosis that may contribute to dysfunction. In a histological study, patients with diabetes had increased fibrosis that was accentuated in subjects with poor glycaemic control.¹² Moreover, in patients with diabetes, myocardial interstitial fibrosis assessed through cardiac magnetic resonance imaging was associated with evidence of impaired systolic and diastolic function.⁶⁰

In experimental models of diabetes-associated heart disease, activated fibroblasts do not undergo conversion into α-SMA ⁺ myofibroblasts, but acquire a matrix-preserving phenotype, characterized by accentuated expression of anti-proteases [such as tissue inhibitor of metalloproteinases (TIMP-1)], reduced activation of MMP-2,⁶¹ and increased levels of matrix crosslinking enzymes (such as lysyl oxidase).^49,33 Diabetes-associated activation of matrix crosslinking fibroblasts was also suggested by scRNA-seq data in a model of streptozotocin (STZ)-induced diabetes, showing expansion of Lox-expressing fibroblasts.⁶² Moreover, another scRNA-seq study showed that db/db mouse cardiac fibroblasts have increased expression of genes associated with migration and with fibrogenic mitogen-activated protein kinase (MAPK) signaling.³⁸ However, despite abundant associative evidence suggesting that fibroblasts may contribute to diabetes-associated fibrosis and dysfunction, evidence documenting their role through fibroblast-specific interventions in vivo is lacking. A study using fibroblast-specific targeting in a model of high fat diet + STZ suggested a protective fibroblast-mediated pathway, involving activation of the small Rho GTPase Rnd3 that may act, at least in part, by suppressing TGF-β signaling.⁶³ However, the molecular signals that activate fibroblasts in diabetic hearts are not known.

4.2. The TGF-β system in diabetic tissues

TGF-βs serve as fibrogenic mediators, potently activating a matrix-synthetic and matrix-preserving fibroblast phenotype through induction of structural and matricellular ECM proteins, anti-proteases (such as PAI-1 and TIMP1), and collagen crosslinking lysyl oxidases.¹⁵ Activation of TGF-β cascades plays a major role in fibroblast activation and fibrotic remodeling in animal models of heart failure.¹⁷ Moreover, scRNA-seq studies suggested that stimulation of TGF-β pathways may underlie fibroblast activation in the myocardium of heart failure patients.⁶⁴ Myocardial activation of TGF-β signaling cascades has been consistently reported in models of Type 1 and Type 2 diabetes.^65,22,29 Several mechanisms may contribute to diabetes-induced activation of TGF-β signaling cascades, including up-regulation of TGF-β isoforms by high glucose and AGT II, increased expression of TGF-β receptors on the cell surface, TGF-β−activating effects of oxidative stress and of matricellular proteins, such as thrombospondin-1, and diabetes-associated down-regulation of endogenous TGF-β/Smad inhibitors, leading to accentuated fibrogenic Smad2/3 responses.⁶⁶

Our findings demonstrate that fibroblast-specific activation of TGF-β signaling plays an important role in the pathogenesis of diabetic cardiomyopathy, mediating both systolic and diastolic dysfunction (Figures 2 and 3). The effects of fibroblast TGF-β signaling on diastolic dysfunction may reflect not only the predictable increase in deposition of ECM proteins, but also paracrine effects on cardiomyocytes. Fibroblasts have been demonstrated to regulate cardiomyocyte size by secreting pro-hypertrophic growth factors, matricellular proteins,⁶⁷ and exosomes containing miRNAs.⁶⁸ A study of the transcriptional profile of db/db mouse cardiac fibroblasts identified several pro-hypertrophic mediators that may be secreted by fibroblasts to stimulate cardiomyocyte hypertrophy in diabetic hearts, including Spp1/osteopontin,⁶⁹ Prg4/proteoglycan 4,⁷⁰ Itgbl1,⁷¹ Areg/amphiregulin, the AGT II precursor angiotensinogen,⁷² and the pro-hypertrophic long non-coding RNA NEAT1.⁷³

4.3. Fibroblast-mediated diabetic cardiomyopathy involves Smad signaling cascades

The activating effects of TGF-β on fibroblasts have been suggested to involve both Smad-dependent and non-Smad pathways. Our study demonstrates that fibroblast-specific Smad3 loss phenocopied the effects of fibroblast TbR2 deletion (Figures 4 and 5), suggesting that the fibrogenic actions of TGF-β in fibroblasts predominantly reflect Smad3-dependent signaling. Our RNA-Seq analysis identified several matricellular genes which are induced in diabetic fibroblasts, including Thbs4, Fbn2, Spon1, and Spock2. Thbs4 was also the top-ranked protein-encoding gene down-regulated in db/db fibroblasts lacking Smad3. TSP-4, the protein encoded by Thbs4, has been suggested to serve as a major regulator of extracellular matrix production and organization⁷⁴ that plays a central role in cardiac remodeling and fibrosis.^46,75 Moreover, scRNA-seq studies in models of cardiac injury identified a fibrogenic fibroblast cluster, characterized by a high expression of Thbs4.^31,32 In contrast, our in vitro studies showed no significant direct effects of matricellular or intracellular TSP-4 on cardiac fibroblasts, suggesting that Thbs4 is simply a marker of activated fibroblasts and does not play a causative role in mediating diabetes-associated fibroblast activation.

4.4. The TGF-β/Smad3 cascade as a therapeutic target in HFpEF associated with diabetes and obesity

HFpEF patients with obese-inflammatory phenotypes typically exhibit elevated markers of fibrosis⁷⁶ and may benefit from anti-fibrotic therapies. Our study suggests that fibroblast-specific targeting of the TGF-β cascade may protect from dysfunction, attenuating both ECM deposition and cardiomyocyte changes. However, our findings also raise a word caution regarding the consequences of TGF-β targeting, suggesting that basal TGF-β signaling in cardiac fibroblasts may be required to preserve ventricular function. In lean animals, fibroblast-specific loss of either TbR2 or Smad3 was associated with dysfunction (evidenced by the highly sensitive speckle-tracking echocardiography assessment). Thus, our findings support the notion that cardiac homeostasis requires low-level basal TGF-β/Smad3 signaling and highlight the importance of tight regulation of the TGF-β cascade. Both complete loss of TGF-β signaling (in lean fibroblast-specific KO mice) and excessive TGF-β signaling (in obese Type 2 diabetic mice) have adverse functional consequences. Therapeutic implementation of anti-TGF-β approaches in HFpEF patients will require identification of subpopulations with excessive fibrosis driven by overactive TGF-β cascades.

4.5. Limitations

Our study has several limitations. First, identification of candidate fibroblast-derived mediators that may regulate cardiomyocyte hypertrophy in diabetic hearts is based on transcriptomic studies. Whether these mediators are secreted in the diabetic myocardium and play a role in cardiomyocyte hypertrophy and/or dysfunction is not known. Secondly, both clinical and experimental studies have suggested sex-specific differences in the pathogenesis and severity of diabetes-associated heart disease.⁷⁷ Although we studied both male and female animals, our experiments were not powered to study sex-specific differences. Thirdly, our speckle-tracking data show subtle perturbations in myocardial strains and strain rates in lean fibroblast-specific TbR2 and Smad3 KO mice. However, these functional defects were not accompanied by changes in cardiac geometry, cardiomyocyte morphology, or collagen deposition at 6 months of age. Experiments at later timepoints are needed to establish any homeostatic actions of fibroblast TGF-β signaling in the heart.

4.6. Conclusions

Our study provides the first direct evidence supporting a fibroblast-mediated mechanism of dysfunction and adverse remodeling in the diabetic heart, highlighting the importance of interstitial cells in the pathogenesis of diabetic cardiomyopathy. Diabetes-associated activation of the TGF-β/Smad3 cascade in fibroblasts may promote dysfunction not only by accentuating ECM deposition and crosslinking, but also by modulating cardiomyocyte phenotype and function. Thus, the TGF-β system may be a promising therapeutic target in patients with diabetes-associated heart failure.

Supplementary material

Supplementary material is available at Cardiovascular Research online.

Funding

This work was supported by the NIH grants R01 HL76246, R01 HL85440, and R01 HL149407, US Department of Defense grants PR181464 and PR211352 (N.G.F.), an American Heart Association (AHA) Career Development Grant (C.H.), and a post-doctoral grant from the Deutsche Forschungsgemeinschaft (I.T., TU 632/1-1).

Data availability

The data supporting the findings of the study are available from the corresponding author upon reasonable request. All RNA-seq data are available through the NCBI’s GEO and can be accessed through GEO SuperSeries accession number GSE241180 (token srongeuinjgpbup).