Diabetes Induces Cardiac Fibroblast Activation, Promoting a Matrix‐Preserving Nonmyofibroblast Phenotype, Without Stimulating Pericyte to Fibroblast Conversion

Abstract

Background

Interstitial and perivascular fibrosis may contribute to diabetes‐associated heart failure. Pericytes can convert to fibroblasts under conditions of stress and have been implicated in the pathogenesis of fibrotic diseases. We hypothesized that in diabetic hearts, pericytes may convert to fibroblasts, contributing to fibrosis and to the development of diastolic dysfunction.

Methods and Results

Using pericyte:fibroblast dual reporter (NG2^Dsred [neuron‐glial antigen 2 red fluorescent protein variant]; PDGFRα^EGFP [platelet‐derived growth factor receptor alpha enhanced green fluorescent protein]) mice in a type 2 diabetic db/db background, we found that diabetes does not significantly affect pericyte density but reduces the myocardial pericyte:fibroblast ratio. Lineage tracing using the inducible NG2^CreER driver, along with reliable labeling of fibroblasts with the PDGFRα reporter system, showed no significant pericyte to fibroblast conversion in lean and db/db hearts. In addition, db/db mouse cardiac fibroblasts did not undergo myofibroblast conversion and had no significant induction of structural collagens but exhibited a matrix‐preserving phenotype, associated with increased expression of antiproteases, matricellular genes, matrix cross‐linking enzymes, and the fibrogenic transcription factor cMyc. In contrast, db/db mouse cardiac pericytes had increased expression of Timp3, without any changes in expression of other fibrosis‐associated genes. The matrix‐preserving phenotype of diabetic fibroblasts was associated with induction of genes encoding oxidative (Ptgs2/cycloxygenase‐2, and Fmo2) and antioxidant proteins (Hmox1, Sod1). In vitro, high glucose partially recapitulated the in vivo changes in diabetic fibroblasts.

Conclusions

Diabetic fibrosis is not mediated through pericyte to fibroblast conversion but involves acquisition of a matrix‐preserving fibroblast program, which is independent of myofibroblast conversion and is only partially explained by the effects of the hyperglycemic environment.

Nonstandard Abbreviations and Acronyms

CSF1R

colony stimulating factor 1 receptor

ECM

extracellular matrix

EGFP

enhanced green fluorescent protein

MMP

matrix metalloproteinase

NG2

neuron‐glial antigen 2

PDGFR

platelet‐derived growth factor receptor

STAT

signal transducer and activator of transcription

TGF

transforming growth factor

TIMP

tissue inhibitor of metalloproteinases

α‐SMA

α‐smooth muscle actin

Clinical Perspective.

What Is New?

• Diabetes‐associated myocardial fibrosis is not mediated through pericyte to fibroblast conversion but involves acquisition of a matrix‐preserving phenotype by resident cardiac fibroblasts, in the absence of myofibroblast transdifferentiation.

• Increased expression of antiproteases, matricellular genes, cross‐linking proteins, and genes associated with the oxidative response is noted in diabetic cardiac fibroblasts and is only in part explained by an increase in glucose levels.

What Are the Clinical Implications?

• Targeting the matrix‐preserving and matrix‐cross‐linking properties of diabetic fibroblasts may hold promise in the treatment of heart failure with preserved ejection fraction in subjects with diabetes.

Diabetes, obesity, and metabolic dysfunction are associated with an increased risk of heart failure. ¹ , ² Moreover, the presence of diabetes worsens prognosis in patients with heart failure, in both populations with heart failure with reduced ejection fraction and heart failure with preserved ejection fraction. ³ , ⁴ , ⁵ Diabetes‐associated heart failure is only in part caused by accelerated atherosclerosis, or by more severe hypertension, but may also be because of a primary myocardial condition, termed “diabetic cardiomyopathy.” ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ The myocardial perturbations associated with diabetic cardiomyopathy involve not only cardiomyocytes but also immune cells, vascular cells, and fibroblasts. ¹¹ , ¹² , ¹³ Fibrotic remodeling of the interstitium and perivascular fibrotic changes have been described both in patients with diabetes ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ and in animal models of diabetes‐associated myocardial disease. ¹⁹ , ²⁰ , ²¹ , ²² Deposition of large amounts of extracellular matrix (ECM) in the myocardium may increase ventricular stiffness, contributing to diastolic dysfunction and to the development of heart failure with preserved ejection fraction.

Fibroblasts are the main effector cells in cardiac fibrosis. In the infarcted and pressure‐overloaded heart, increased deposition of ECM proteins is driven by expansion of the cardiac fibroblast population and by conversion into matrix‐synthetic myofibroblasts. ²³ , ²⁴ The origin of activated fibroblasts and myofibroblasts in the infarcted and remodeling myocardium remains hotly debated. Several studies have suggested that endothelial cells ²⁵ , ²⁶ hematopoietic progenitors, ²⁷ and macrophages ²⁸ may convert to fibroblasts, thus directly contributing to fibrotic cardiac remodeling. In contrast, other investigations used robust lineage tracing strategies to demonstrate that the vast majority of activated myofibroblasts in infarcted and failing hearts are derived from resident fibroblast‐like interstitial cells. ²⁹ , ³⁰ , ³¹ , ³² Although extensive experimental evidence suggests an important contribution of pericytes in renal ³³ , ³⁴ , ³⁵ , ³⁶ and pulmonary ³⁷ , ³⁸ fibrosis, mediated at least in part through fibroblast conversion, the role of pericytes in myocardial fibrosis remains enigmatic. Lineage tracing studies in mice demonstrated that a population of perivascular cells expressing the zinc finger transcription factor Gli‐1 may expand following cardiac pressure overload, accounting for a large fraction of activated myofibroblasts. ³⁹ However, whether these cells represent pericytes or periadventitial fibroblasts is unclear, and the role of mature pericytes in cardiac fibrosis remains enigmatic.

Diabetes is associated with marked perturbations in pericyte phenotype and function. ⁴⁰ , ⁴¹ Pericyte loss and/or dysfunction may underlie the microvascular complications of diabetes and has been implicated in the pathogenesis of diabetic retinopathy and nephropathy. ⁴² In vitro, cultured pericytes exhibit remarkable plasticity and are capable of transdifferentiating into fibrogenic cells in response to microenvironmental cues. ⁴³ In diabetic subjects, high glucose may stimulate oxidative responses in pericytes, leading to activation of a fibrogenic program, and may also disrupt the architecture of the microvascular basement membrane, ⁴⁴ perturbing interactions between endothelial cells and pericytes. Although the role of pericytes in diabetic fibrosis has been suggested, ⁴⁵ studies examining whether pericytes contribute to fibrotic remodeling in diabetic tissues have not been performed.

We hypothesized that, in diabetic hearts, pericytes may convert to fibroblasts and directly contribute to expansion of fibroblasts and the development of fibrotic changes. In order to study the fate and function of diabetic cardiac pericytes, we developed type 2 diabetic db/db pericyte:fibroblast dual reporter mice, and genetic tools for lineage tracing of cardiac pericytes accompanied by reliable identification of fibroblasts using the PDGFRα^EGFP (platelet‐derived growth factor receptor alpha enhanced green fluorescent protein) reporter system. ⁴⁶ We found no significant pericyte to fibroblast conversion in the diabetic myocardium. Moreover, transcriptomic analysis showed that pericytes in diabetic hearts do not upregulate synthesis of genes associated with ECM deposition or remodeling. On the other hand, diabetic mouse cardiac fibroblasts do not exhibit myofibroblast conversion and do not synthesize increased amounts of structural collagens but acquire matrix‐preserving and matrix cross‐linking properties, associated with alterations in genes involved in oxidative stress. Our findings suggest that fibroblasts and not pericytes are responsible for diabetes‐associated cardiac fibrosis and that the fibrotic changes are not dependent on myofibroblast conversion but predominantly involve acquisition of a matrix‐preserving phenotype.

Methods

The data that support the findings of this study are available from the corresponding author upon reasonable request.

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.

Generation of Lean and Diabetic Pericyte/Fibroblast Dual Reporter Mice

Pericyte reporter NG2^Dsred/+ (neuron‐glial antigen 2 red fluorescent protein variant; Jackson #008241), ⁴⁷ fibroblast reporter PDGFRα^EGFP/+ (Jackson #007669), ⁴⁶ , ⁴⁸ and Lepr^dm/+ mice on a C57BL/6J background (db/+) (Jackson #000697) were purchased from Jackson Laboratories (Bar Harbor, ME). Obese type 2 diabetic and lean fibroblast; pericyte dual reporter mice (NG2^Dsred; PDGFRα^EGFP; db/db, or NG2^Dsred; PDGFRα^EGFP; db/+ and NG2^Dsred; PDGFRα^EGFP; +/+) were generated by breeding NG2^Dsred/+; db/+ mice with PDGFRα^EGFP/+; db/+ mice. Six‐month‐old male and female mice were euthanized using 2% inhaled isoflurane followed by cervical dislocation and cardiac tissue was used for histological processing and for fluorescence‐activated cell sorting of pericytes and fibroblasts.

Generation of Diabetic Macrophage Reporter Mice

Lean and diabetic macrophage reporter mice were generated by breeding db/+; CSF1R^EGFP/+ (colony stimulating factor 1 receptor) mice with db/+ animals. Six‐month‐old male and female db/db; CSF1R^EGFP animals, and lean db/+:CSF1R^EGFP controls were euthanized for histological studies.

Lineage Tracing of Pericytes in Lean and Diabetic Animals

Figure S1 illustrates the breeding strategy for generation of the mice used for tracing of mural cells and simultaneous identification of fibroblasts. Tamoxifen‐inducible NG2CreER(NG2^iCre, Jackson #008538) mice were bred in a db/+ background (NG2^iCre/+; db/+), then were crossed with ROSA26^tdTomato (Jackson #007914) also in the db/+ background ROSA26^tdTomato; (db/+), to generate NG2^iCre/+; ROSA26^tdTomato; db/+ mice. These animals were then crossed with PDGFRα^EGFP; db/+ fibroblast reporter mice to enable the simultaneous identification of mural cell‐derived progeny, as well as fibroblasts in the diabetic background, as previously described by our group. ⁴⁶ For lineage tracing studies, 10‐ to 12‐week‐old NG2^iCre/+; R26^tdTomato; PDGFRα^EGFP; db/db or NG2^iCre/+; R26^tdTomato; PDGFRα^EGFP; db/+ (or +/+) mice received intraperitoneal injections of tamoxifen (Sigma‐T5648, CAS#10540–29‐1), at a dosage of 100 mg/kg, administered over 5 consecutive days once every 24 hours. Mice were euthanized for histology when they reached 6 months of age.

Immunohistochemistry

For histopathological analysis, mice were euthanized and hearts were fixed in Z‐fix (Anatech, Battle Creek, MI) and embedded in paraffin. Sequential 5‐μm sections were cut from base to apex at 250‐μm intervals. Following citrate buffer‐mediated antigen retrieval, sections were allowed to cool for 60 minutes and were then blocked with TBST (Tris‐buffered saline+0.1% Triton X), containing 10% donkey serum. The following antibodies were used for immunofluorescence studies: mouse anti‐α‐SMA (smooth muscle actin) antibody (dilution 1:100, Sigma F3777, St. Louis, MO) to identify vascular smooth muscle cells as well as any nonvessel associated myofibroblasts, goat anti‐tdTomato antibody (dilution 1:300, Biorbyt orb334992, Cambridge, UK) to identify NG2+ pericytes, and rabbit anti‐GFP antibody (dilution 1:100, D5.1–2956, Cell Signaling Technology, Danvers, MA) to identify PDGFRα+ resident fibroblasts. Lectin histochemistry with biotinylated (GSL‐1) Griffonia simplicifolia lectin 1 (Vectorlabs B‐1105‐2, Burlingame, CA) was performed to identify endothelial cells. Incubation with primary antibodies was performed overnight at 4 °C. Following washes in TBS, secondary antibodies were used for 1 hour at room temperature. Sections were washed and then incubated with Trueblack (Biotium 23 007, Fremont, CA) for 20 seconds to quench autofluorescence and then sealed with Fluoro‐Gel II mounting medium containing DAPI (EMS #17985–50, Hatfield, PA). Slides were then scanned using Zen 3.0 Pro software and the Zeiss Imager M2 microscope (Carl Zeiss Microscopy, White Plains, NY). Appropriate negative controls using isotype‐matched immunoglobulin G were performed to confirm the specificity of the immunofluorescence experiments.

Assessment of Myofibroblast Conversion

In order to examine whether fibroblasts in diabetic hearts exhibit myofibroblast conversion, we used myocardial sections from lean and db/db fibroblast:pericyte reporter mice for dual immunofluorescence, combining GFP staining (goat anti‐GFP antibody, dilution 1:100, ab5450, Abcam, Waltham, MA) to identify all PDGFRα+ fibroblasts and α‐SMA staining (Alexa Fluor® 594‐conjugated mouse anti‐α‐SMA antibody, dilution 1:100, ab202368, Abcam, Waltham, MA) to label myofibroblasts. Additional experiments used dual immunofluorescence, combining GFP staining (rabbit anti‐GFP antibody, dilution 1:100, D5.1–2956, Cell Signaling Technology, Danvers, MA) and periostin staining (goat antiperiostin antibody, dilution 1:25, AF2955, R&D Systems, Minneapolis, MN) to identify cells with early characteristics of myofibroblasts (in the absence of α‐SMA immunoreactivity). Sections from infarcted hearts (7 days after coronary occlusion) were used as a positive control. ⁴⁹

Macrophage Labeling and Quantification

To quantify myeloid cells in diabetic and wild type (WT) control dual reporter mice, we used immunofluorescent staining for lysozymeM, as previously described. ⁵⁰ Sections were dewaxed and subjected to antigen retrieval in a steamer for 30 minutes in citric buffer (pH 6.0) followed by cooling at room temperature for 1 hour. Sections were then blocked with serum for 1 hour at room temperature and were incubated overnight at 4 °C with the rabbit antilysozyme antibody (1:100; Novus Biologicals, NBP2‐61118). To examine the effects of diabetes on macrophage density, lean and db/db CSF1R^EGFP macrophage reporter mice were used for dual fluorescence with goat anti‐GFP antibody to label CSF1R+ macrophages (1:200; Abcam Ab6673) and the rat anti‐Mac2 antibody (1:200; Cedarlane, CL8942AP) to identify cells expressing the macrophage activation marker galectin‐3. The density of myocardial CSF1R+ and Mac2+ cells was quantitatively assessed.

Quantitative Analysis of Cell Density

Using default algorithms of the Intellesis Trainable Segmentation module of Zen 3.0 Pro software (Carl Zeiss Microscopy), an artificial intelligence‐based model was trained on images representing ventricles to identify and count pericyte, microvessel as well as fibroblast profiles. Using the Image Analysis module, specific settings were incorporated in the trained model to count the segmented objects. For quantitative analysis, 5–10 fields were scanned from 2–3 different sections from ventricular tissue under a magnification of 200×. The data were presented as the number of positive profiles per mm².

Picrosirius Red Staining

Collagen fibers were identified by picrosirius red staining using protocols established in our laboratory. ²² Perimysial collagen thickness was assessed by measuring the average thickness of each perimysial strand, by averaging measurements at 3 different points (which included the thickest and thinnest dimensions and a third random point). To assess the perimysial collagen thickness for each mouse, measurements for at least 10 perimysial strands were averaged.

Wheat Germ Agglutinin Lectin Histochemistry and Quantification of Cardiomyocyte Size

Cardiomyocyte size was assessed using sections stained with Alexa Fluor 488‐conjugated wheat germ agglutinin (WGA) lectin (W11261, Invitrogen). For each mouse, the area of 50 cardiomyocytes from subendocardial or subepicardial regions cut in cross‐section was measured. Cardiomyocytes were identified from 5 different random subendocardial or subepicardial fields, scanned from 2 different sections for each mouse. The mean cardiomyocyte cross‐sectional area was calculated for each mouse.

Fluorescence‐Activated Cell Sorting of Cardiac Pericytes and Fibroblasts

Single cell suspension for flow cytometry was prepared using a previously described protocol. ⁴⁶ Briefly, atria were removed from the myocardial tissue, and the ventricles were finely minced, suspended in digestion buffer cocktail of collagenase IV (2 mg/mL, Worthington Biochem) and dispase II (1.2 U/mL, Stemcell Technologies) in Dulbecco's PBS. Cells in single cell suspensions were blocked with antimouse CD16/CD32 (1:250, BD Biosciences) for 30 minutes at 4 °C. To identify endothelial cells, the cell suspension was incubated for 1 hour at 4 °C with anti‐CD31‐BV605 (1:100, BioLegend). Cell suspension was washed and labeled with calcein violet 450 (1.25 μmol/L, eBioscience, Invitrogen) to identify metabolically active cells and 7‐aminoactinomycin D (1:500, Invitrogen) to identify cells with compromised cell membranes. Nonendothelial cells (CD31‐) were gated to identify NG2^RFP+ pericytes and PDGFRα^EGFP+ fibroblasts, which were sorted with the FACSAria Sorter (BD Biosciences). FlowJo software was used for data analysis.

Stimulation of Isolated Cardiac Fibroblasts

Cardiac fibroblasts were isolated from C57BL/6J animals (n=4, each “n” consisted of 1 male and 1 female heart) using enzymatic digestion as previously described ⁵¹ and were cultured in Dulbecco's Modified Eagle's Medium (DMEM; Corning) containing 1 g/L glucose with 10% fetal calf serum. Cells were serum starved at passage 2 for 3 hours and subsequently stimulated for 24 hours with increasing concentrations of glucose‐containing media (DMEM with 1 g/L glucose, Corning; DMEM/F12 with 3.151 g/L glucose; DMEM with 4.5 g/L glucose, Sigma‐Aldrich) to obtain 3 different glucose concentrations: 5.6 mmol/L, 17 mmol/L, and 25 mmol/L. Cells were then harvested for subsequent TRIzol‐based RNA extraction.

RNA Isolation and Polymerase Chain Reaction Array

RNA isolation was performed in cultured cardiac fibroblasts using traditional TRIzol reagent method and in sorted cardiac fibroblasts or pericytes using Dynabeads mRNA Direct kit (61 012, Invitrogen). Following isolation, RNA was converted to cDNA using Qiagen RT2 First Strand kit (330 404, Qiagen). Quantitative polymerase chain reaction (PCR) was done using the following mouse Qiagen RT2 Profiler PCR arrays according to manufacturer's protocol: Extracellular matrix PCR array (PAMM‐013Z), oxidative stress and antioxidant defense array (PAMM‐065Z), and fibrosis array (PAMM‐120Z). The following conditions were used: 95 °C for 10 minutes, 40 cycles at 95 °C for 15 seconds and 60 °C for 1 minute on the CFX384™ Real‐Time PCR Detection System (Bio‐Rad). The data obtained were analyzed using the ΔCt method.

Statistical Analysis

For comparisons of 2 groups, an unpaired 2‐tailed Student's t test using (when appropriate) Welch's correction for unequal variances was performed. The Mann–Whitney test was used for comparisons between 2 groups that did not show Gaussian distribution. For comparisons of multiple groups, 1‐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±SE for normally distributed data sets and as median with interquartile range for non‐Gaussian distributions. Statistical significance was set at 0.05. Statistical analysis was performed with Graphpad Prism 9.4.

Results

Diabetic db/db Mice Exhibit Reduced Microvascular Density, Accompanied by a Trend Toward Reduced Pericyte Density

We have previously demonstrated that NG2^Dsred reporter mice identify pericytes in the mouse myocardium. ⁴⁶ In order to examine the effects of diabetes on the myocardial pericyte population, we generated type 2 diabetic pericyte reporter mice (db/db; NG2^Dsred). Dual fluorescence for Dsred and GSL identified abundant periendothelial NG2+ pericytes in lean and db/db hearts (Figure 1A). Dual fluorescence showed that the majority of NG2+ cells do not express α‐SMA, thus exhibiting characteristics of microvascular pericytes (Figure 1B, Figure S2). α‐SMA+ vascular smooth muscle cells in arterioles and intramyocardial coronaries had variable expression of NG2 (Figure S2). Quantitative analysis showed that at 6 months of age, db/db hearts have a significant reduction in microvascular density (Figure 1C) and exhibit a trend toward reduced pericyte density, when compared with lean mouse hearts (P=0.14, n=10–12, Figure 1D). The reduced microvascular density in db/db hearts is more consistently observed in the subendocardial segments but does not reach statistical significance in the subepicardial myocardium (Figure 1E through 1H).

Figure 1. Diabetic hearts exhibit a reduction in microvascular density, associated with a trend toward decreased pericyte content.

A, Dual fluorescence in myocardial sections of lean and db/db NG2^Dsred pericyte reporter mice identifies NG2+ pericytes (red) and endothelial cells stained with GSI (Griffonia simplicifolia lectin, green). NG2+ pericytes are closely associated with endothelial cells. Scalebar=80 μm. B, Dual fluorescence in myocardial sections of lean and db/db NG2^Dsred pericyte reporter mice for NG2 (red) and α‐SMA (green). NG2 labels both α‐SMA+ vascular smooth muscle cells (long arrows) and α‐SMA‐negative microvascular pericytes (short arrows). Scalebar=100 μm. C, Quantitative analysis shows that, when compared with lean mice, db/db mice have a significantly reduced cardiac microvascular density (P=0.02). D, db/db mice also had a trend toward reduced density of pericyte profiles (P=0.14). E through H, Microvascular density was significantly reduced in the subendocardial segments of the left ventricular myocardium (LV endo) (E). Subepicardial segments (LV epi) (G) had a trend toward reduced microvascular density. In contrast, pericyte density in subendocardial (F) and subepicardial segments (H) was not significantly different between lean and db/db mice (n=10‐12/group). Dsred indicates red fluorescent protein variant; NG2, neuron‐glial antigen 2; and α‐SMA, alpha smooth muscle actin.

Fibroblasts and Pericytes in the Diabetic Heart Remain Distinct Populations

Pericyte to fibroblast conversion has been suggested to contribute to tissue fibrosis in models of renal, ³³ , ³⁴ pulmonary ³⁸ and myocardial fibrosis. ³⁹ To examine whether pericytes in diabetic hearts acquire a fibroblast‐like phenotype, we generated lean and diabetic pericyte:fibroblast (NG2^Dsred; PDGFRα^EGFP) dual reporter mice. Dual fluorescence showed that in both lean and in diabetic hearts, PDGFRα+ fibroblasts and NG2+ pericytes are distinct populations (Figure 2A through 2F). Quantitative analysis showed no significant differences in the density of PDGFRα+ left ventricular fibroblasts between lean and obese diabetic db/db hearts (Figure 2G). Moreover, the density of double positive NG2+/PDGFRα+ cells (that could represent pericytes undergoing fibroblast conversion) was very low in both lean and obese diabetic mouse hearts (Figure 2G). The percentage of PDGFRα+ myocardial fibroblasts expressing the pericyte marker NG2 was <1% in lean mice and did not significantly change in db/db animals (lean: 0.90%±0.18 versus db/db: 1.03±0.12; P=0.34, n=7–11/group).

Figure 2. In both lean and db/db mice, NG2+ pericytes and PDGFRα+fibroblasts are distinct nonoverlapping populations of interstitial cells.

A through F, Staining of myocardial sections from lean (A through C) and db/db (D through F) NG2^Dsred;PDGFRα^EGFP pericyte:fibroblast dual reporter mice. Representative subendocardial (A and D), midmyocardial (B and E), and subepicardial (C and F) fields are shown. Dual fluorescence shows no significant overlap between NG2+ pericyte profiles (red) and the nuclei of PDGFRα+ fibroblasts (green). Scalebar=50 μm. G, Quantitative analysis shows no significant difference in the density of myocardial PDGFRα+ fibroblasts between lean WT and db/db mice. Moreover, the density of PDGFRα+ fibroblasts expressing the pericyte marker NG2 is very low in both WT and db/db hearts. H, The ratio of PDGFRα+ fibroblasts to NG2+ pericytes is significantly increased in db/db hearts (P=0.03, n=7–11/group). Dsred indicates red fluorescent protein variant; EGFP, enhanced green fluorescent protein; NG2, neuron‐glial antigen 2; PDGFRα, platelet‐derived growth factor receptor alpha; and WT, wild type.

Diabetic Hearts Exhibit an Increased Fibroblast: Pericyte Ratio

Quantitative analysis showed that the ratio of PDGFRα+ fibroblasts to NG2+ pericytes is significantly increased in db/db hearts (Figure 2H), suggesting that diabetes is associated with a perturbation in the cellular composition of the cardiac interstitium, favoring fibroblasts in relation to pericytes. The trend toward reduced pericyte density and the increased fibroblast to pericyte ratio may reflect diabetes‐associated conversion of pericytes to fibroblasts, resulting in loss of expression of pericyte markers, such as NG2. Thus, any pericyte‐derived fibroblasts may not be identified as pericytes using the reporter system. In order to examine whether pericytes convert to fibroblasts in diabetic hearts, we performed lineage tracing experiments.

Pericytes Do Not Undergo Fibroblast Conversion in Diabetic Hearts

To trace the fate of pericytes in diabetic hearts, we generated obese diabetic db/db; NG2^CreER; ROSA26^tdTomato mice. Because lack of reliable fibroblast markers is a major challenge in documenting acquisition of a fibroblast phenotype in vivo, we bred these mice with PDGFRα^EGFP fibroblast reporter animals for reliable fibroblast identification. NG2+ pericytes were labeled in lean NG2^CreER; ROSA26^tdTomato; PDGFRα^EGFP mice and in obese diabetic db/db; NG2^CreER; ROSA26^tdTomato; PDGFRα^EGFP animals after tamoxifen injection. There was a trend toward reduced density of NG2‐derived cells in diabetic hearts that did not reach statistical significance (Figure 3). In both lean and diabetic mice, the density of NG2‐derived PDGFRα+ fibroblasts was very low (Figure 3). The percentage of PDGFRα+ fibroblasts derived from NG2+ cells in lean WT mice was ≈1.3% and did not significantly increase in db/db animals (lean: 1.34%+0.34 versus db/db: 1.45+0.77; P=0.998, n=3–8/group).

Figure 3. Fate mapping of adult pericytes shows no significant pericyte to fibroblast conversion in lean and diabetic hearts.

Mice used for tracing of NG2 lineage cells were bred with PDGFRα^EGFP fibroblast reporter mice for reliable identification of fibroblasts. Dual fluorescent staining shows that PDGFRα+ fibroblasts (green nuclei) are not derived from NG2‐derived cells (red) in lean (A through F) and db/db (G through L) mouse hearts. D through F and J through L represent high‐magnification images of the insets in A through C and G through I, respectively. Few PDGFRα+ fibroblasts derived from NG2+ cells were noted (arrows) in both db/db and lean WT groups. Quantitative analysis shows that db/db mice and lean mice had comparable density of PDGFRα+ fibroblasts (M). Moreover, the number of PDGFRα+ fibroblasts that were derived from NG2+ cells was very low in both WT lean and db/db mouse hearts. N, Quantitative analysis of the total density of NG2‐derived profiles showed that db/db mice had a trend toward reduced density of NG2‐derived cells that did not reach statistical significance (P=0.19, n=3–8/group). Scalebar=40 μm for A through C and G through I; scalebar=20 μm for D through F and J through L. EGFP indicates enhanced green fluorescent protein; NG2, neuron‐glial antigen 2; PDGFRα, platelet‐derived growth factor receptor alpha; and WT, wild type.

Diabetes Is Associated With Activation of a Matrix‐Preserving and Matrix‐Cross‐Linking Phenotype in Cardiac Fibroblasts.

Our lineage tracing experiments do not support a significant role for pericyte to fibroblast conversion in diabetes‐associated myocardial fibrosis. However, even in the absence of pericyte to fibroblast transdifferentiation, pericytes may contribute to diabetic fibrosis by secreting fibrogenic mediators, or matrix‐remodeling enzymes. To test this hypothesis, we sorted myocardial pericytes and fibroblasts from lean and diabetic dual pericyte:fibroblast reporter (NG2^Dsred; PDGFRα^EGFP) mice, and we assessed expression of fibrosis‐associated genes using a PCR array. Flow cytometry confirmed the absence of a significant population of NG2+/PDGFRα+ cells in lean and db/db hearts (lean: 0.28%+0.04 of the CD31‐interstitial cells; diabetic: 0.14%+0.05 of the CD31‐interstitial cells, P=NS, Figure 4A). PDGFRα+ fibroblasts and NG2+ pericytes isolated from lean WT hearts have distinct transcriptomic profiles (Figure 4, Figure S3). Fibroblasts express much higher levels of genes encoding structural matrix proteins (Col1a2, Col3a1), certain matricellular proteins (including Thbs2 and Ccn2/Ctgf), matrix remodeling genes (Mmp2, Timp2, Timp3), Tgfbr2, Itgb5, and Dcn (which encodes for the antifibrotic protein decorin). In contrast, pericytes express higher levels of Acta2 (encoding α‐SMA), Itga1, and Cav1. (Figure 4).

Figure 4. In normal adult hearts, NG2+ pericytes and PDGFRα+ fibroblasts are nonoverlapping populations with distinct transcriptional profiles.

A, Cell suspensions were prepared from lean and db/db dual reporter NG2^Dsred; PDGFRα^EGFP mice. Viable (7‐AAD‐) and nonendothelial (CD31‐) cells were identified and subgated to select the NG2+ and PDGFRα+ populations, which were sorted into cell lysis buffer for RNA isolation. FACS analysis confirms the immunofluorescence data (Figure 2), demonstrating that the NG2+/PDGFRα+ population is small. B, The volcano plot summarizes the extracellular matrix array quantitative polymerase chain reaction data, comparing the profile of NG2+ pericytes and PDGFRα+ fibroblasts from lean WT mice. C through P, PDGFRα+ fibroblasts have higher levels of Col1a2 (C), Col3a1 (D), Ccn2 (E), Thbs2 (F), Mmp2 (G), Serpinh1 (H), Timp2 (I), Timp3 (J), Tgfbr2 (K), Dcn (L), and Itgb5 (M). On the other hand, NG2+ cells express higher levels of Itga1 (N), Acta2 (O), and Cav1 (P) (n=5/group). 7‐AAD indicates 7‐aminoactinomycin D; Dsred, red fluorescent protein variant; EGFP, enhanced green fluorescent protein; FACS, fluorescence‐activated cell sorting; NG2, neuron‐glial antigen 2; PDGFRα, platelet‐derived growth factor receptor alpha; RFP, red fluorescent protein; and WT, wild type.

Diabetes was associated with significant changes in the gene expression profile of fibroblasts (Figure 5, Figure S4). db/db and WT cardiac fibroblasts have comparably low expression of Acta2 (Figure 5). To examine whether this finding reflects the absence of myofibroblast conversion in diabetic hearts, we performed dual immunofluorescence in lean and type 2 diabetic PDGFRα^EGFP reporter mice. Dual fluorescence for α‐SMA and GFP (to identify fibroblasts) showed that α‐SMA immunoreactivity in 6‐month‐old lean and db/db mouse hearts was restricted to vascular smooth muscle cells (Figure S5A through S5L). Sections from infarcted hearts were used as a positive control, showing abundant α‐SMA+/PDGFRα+ infarct myofibroblasts (Figure S5M through S5O). Moreover, dual fluorescence for the early myofibroblast marker periostin and GFP showed no significant periostin immunoreactivity in fibroblasts residing in lean and db/db mouse hearts (Figure S6). Moreover, db/db fibroblasts do not have increased expression of the structural collagen genes Col1a2 and Col3a1. However, db/db fibroblasts exhibit markedly accentuated synthesis of Timp1 (encoding the protease inhibitor TIMP1 [tissue inhibitor of metalloproteinases 1]), Cmyc (encoding the fibrogenic c‐Myc oncogene), Eng (encoding the TGF‐β [transforming growth factor‐β] accessory receptor endoglin), and Thbs1 (encoding the fibrogenic TGF‐β activating matricellular protein TSP‐1 [thrombospondin‐1]). Levels of SerpinE1, which encodes the fibrogenic protease inhibitor PAI‐1 (plasminogen activator inhibitor 1), Lox, which encodes for the cross‐linking enzyme lysyl‐oxidase, and Ccl11 are also increased in db/db fibroblasts (Figure 5).

Figure 5. Diabetic fibroblasts acquire a matrix‐preserving and matrix cross‐linking phenotype, in the absence of myofibroblast conversion or induction of structural collagen synthesis.

A, The volcano plot summarizes the matrix quantitative polymerase chain reaction array data, comparing the transcriptional profile of sorted PDGFRα+ fibroblasts from lean and db/db reporter mice. B through K, Lean WT and db/db cardiac fibroblasts had no significant differences in expression of Acta2 (encoding the myofibroblast marker α‐smooth muscle actin) (B), Col1a2 (C), and Col3a1 (D). In contrast, db/db mouse cardiac fibroblasts had higher expression of Timp1 (encoding TIMP1) (E), cMyc (F), Eng (G), Thbs1 (encoding the matrix‐preserving matricellular protein thrombospondin‐1) (H), Serpine1 (encoding the antiprotease PAI‐1) (I), Lox (encoding the matrix‐cross‐linking enzyme lysyl oxidase) (J) and Ccl11 (K) (n=5/group). PAI‐1 indicates plasminogen activator inhibitor 1; PDGFRα, platelet‐derived growth factor receptor alpha; TIMP1, tissue inhibitor of metalloproteinases 1; and WT, wild type.

To examine whether acquisition of a matrix‐preserving phenotype by db/db mouse cardiac fibroblasts is associated with expansion of the matrix network, we performed picrosirius red staining to label the collagen fibers (Figure S7A through S7D). Quantitative analysis showed that db/db mice had increased thickness of the myocardial perimysial collagen fibers, when compared with lean mouse hearts (Figure S7E). WGA lectin histochemistry was performed to assess cardiomyocyte size and to label the endomysial matrix. Cardiomyocyte size was significantly higher in db/db mouse hearts (Figure S8). Moreover, type 2 diabetic db/db mice exhibited expansion of the WGA‐stained cardiac interstitium (Figure S8A through S8F). Thus, the findings suggest that diabetic‐myocardial fibrosis is not associated with direct induction of collagen synthesis by cardiac fibroblasts but may involve activation of matrix‐preserving and matrix‐cross‐linking pathways and upregulation of fibrogenic matricellular proteins.

Diabetes Is Associated With Increased Expression of Timp3 and Cebpb in Pericytes But Does Not Affect Synthesis of Other Fibrosis‐Related Genes

In contrast, pericytes sorted from WT and db/db hearts had comparable expression levels of collagens, MMPs (matrix metalloproteinases), and matricellular genes (Figure S9, Figure 6). Out of the 84 fibrosis‐associated genes assessed, only Timp3 (which encodes the protease inhibitor TIMP3) was found to be significantly upregulated in db/db pericytes, when compared with pericytes harvested from lean WT hearts. Moreover, db/db pericytes exhibit trends toward increased expression of Cebpb (encoding the transcription factor C/Ebpβ) and Itgb1 and reduced synthesis of Itgb5 and Tgfbr2 that did not reach statistical significance. Thus, diabetes does not promote a matrix‐synthetic phenotype and has modest effects on expression of fibrosis‐associated genes by cardiac pericytes.

Figure 6. Diabetic pericytes do not exhibit activation of a fibrogenic program.

A, The volcano plot summarizes the matrix quantitative polymerase chain reaction array data comparing the transcriptional profile of sorted NG2+ pericytes from lean and db/db reporter mice. db/db pericytes do not exhibit upregulation of structural collagens, matricellular genes and proteases. B through F, Out of the 84 genes assessed, only Timp3 (B) is significantly upregulated in db/db cardiac pericytes. db/db pericytes also exhibit trends toward increased Cebpb (C) and Itgb1 (D) levels, and decreased Itgb5 and Tgfbr2 levels, which did not reach statistical significance (D through F) (n=5/group). NG2 indicates neuron‐glial antigen 2; and WT, wild type.

Diabetic Fibroblasts Exhibit Marked Induction of Genes Associated With Oxidative Stress

Oxidative mechanisms have been extensively implicated in diabetic fibrosis. To explain the marked differences between fibroblasts and pericytes in expression of fibrosis‐associated genes, we hypothesized that fibroblasts and pericytes may exhibit distinct patterns of activation of oxidative mediators in response to diabetes. We used a PCR array to compare expression profiles of genes associated with oxidative stress between fibroblasts and pericytes harvested from lean and db/db mice (Figure 7, Figures S10 through S13). In comparison with pericytes, fibroblasts harvested from lean mice had significantly higher levels of Vim, Sod1, Scd1, Ptgs2, Psmb5, Gstp1, Gss, Gpx7, Fth1, Fmo2, Ctsb, and ApoE. In contrast, pericytes had significantly higher Ift172 expression levels. Ncf1 and Mpo expression was significantly higher in pericytes than in fibroblasts; however, levels of expression were low in both cell types (Figures S10, S11).

Figure 7. Effects of diabetes on oxidative gene expression in pericytes and fibroblasts.

A and B, Polymerase chain reaction array was performed to examine the effects of diabetes on expression of oxidative response‐related genes in cardiac fibroblasts and pericytes. The volcano plots show comparisons of oxidative gene expression between db/db and lean mouse cardiac fibroblasts (A) and between db/db and lean mouse cardiac pericytes (B). Diabetes had more pronounced effects on expression of genes involved in regulation of the oxidative response in fibroblasts than in pericytes. C through Q, db/db cardiac fibroblasts had marked induction of Ptgs2, which encodes cyclooxygenase (COX)‐2 (C), Fmo2, which encodes the oxidative protein FMO2 (flavin‐containing monooxygenase 2) (D), Sod1 (E), and Hmox1 (F), which encode the antioxidant proteins SOD1 (superoxide dismutase 1) and HO‐1 (heme oxygenase 1), respectively. Diabetes was also associated with increased fibroblast synthesis of Psmb5 (G), Sqstm1 (H), and Scd1 (I) and with trends toward increased expression of Txnip (J), Gpx7 (K), and Fth1 (L). None of these genes were modulated by diabetes in cardiac pericytes. Only 4 genes were modulated in diabetic vs control cardiac pericytes. Txnrd3 (M) and Gss (N) were significantly increased in db/db pericytes, whereas Ift172 (O) and Atr (P) were downregulated. Moreover, Cyba levels (Q) trended toward a lower expression in diabetic pericytes (P=0.1) (n=3–4/group). NG2 indicates neuron‐glial antigen 2; and PDGFRα, platelet‐derived growth factor receptor alpha.

In response to diabetes, fibroblasts exhibit more pronounced changes in synthesis of genes associated with oxidative stress than pericytes. (Figure 7, Figures S10, S11) In comparison with fibroblasts harvested from lean mouse hearts, db/db cardiac fibroblasts have marked induction of Ptgs2, which encodes COX‐2 (cyclooxygenase‐2); Fmo2, which encodes the oxidative protein FMO2 (flavin‐containing monooxygenase 2); and Sod1 and Hmox1, which encode the antioxidant proteins superoxide dismutase 1 and heme oxygenase 1, respectively. Diabetes is also associated with increased fibroblast synthesis of Psmb5, which encodes one of the essential subunits of the 20 S proteasome complex, Sqstm1 (which encodes the autophagy‐associated protein sequestosome 1), and Scd1, which encodes stearoyl‐coenzyme A desaturase 1, an enzyme involved in fatty acid metabolism. Moreover, db/db cardiac fibroblasts have trends toward increased expression of Txnip (encoding thioredoxin‐interacting protein) (P=0.053), Gpx7 (which encodes glutathione peroxidase 7) (P=0.11), and Fth1 (P‐0.126). None of these genes are modulated by diabetes in cardiac pericytes.

Out of the 84 genes assessed with the PCR array, only 4 genes were modulated in diabetic cardiac pericytes, in comparison with lean mouse pericytes (Figure 7, Figure S13). Txnrd3, which encodes thioredoxin reductase 3, and Gss, which encodes glutathione synthetase, are significantly increased in db/db mouse cardiac pericytes. In contrast, pericyte expression of Ift172, a gene encoding the ciliar protein intraflaggelar transport 172, and Atr, which encodes Ataxia telangiectasia and Rad3 related protein, are downregulated in diabetic animals. Two additional genes exhibit statistical trends that do not reach statistical significance: Cyba (encoding cytochrome B‐245) trends toward a lower expression (P=0.1) in diabetic pericytes, whereas Gclc (which encodes γ‐glutamylcysteine synthetase, an enzyme involved in glutathione synthesis) trends toward higher expression (P=0.06). None of these genes is differentially regulated in diabetic fibroblasts. Taken together, the findings suggest that diabetes exerts distinct actions on the oxidative molecular machinery in fibroblasts and pericytes and that when compared with fibroblasts, pericytes may exhibit attenuated oxidative responses to diabetic metabolic perturbations.

At 6 Months of Age, Lean and db/db Mice Have Comparable Cardiac Macrophage Density

Next, we examined whether activation of oxidative pathways and Ccl11 upregulation in db/db mouse cardiac fibroblasts are associated with increased myocardial infiltration with leukocytes. Immunofluorescent staining for the myeloid cell marker lysozyme M showed no significant increase in the density of myeloid cells in 6‐month‐old db/db hearts, when compared with age‐ and sex‐matched lean littermates (Figure S14A through S14E). To examine whether db/db mice exhibit infiltration with macrophages, we generated type 2 diabetic db/db; CSF1R^EGFP macrophage reporter mice. Lean and db/db reporter animals had comparable density of CSF1R+ macrophages (Figure S14F through S14O). Furthermore, we used immunofluorescence for Mac2, an antibody against galectin‐3, which is expressed predominantly by activated macrophages to assess macrophage activation. The majority of CSF1R+ macrophages in both lean and db/db hearts were Mac2‐negative. db/db mice and lean controls had comparable numbers of cardiac Mac2+ cells (Figure S14F through S14O).

High Glucose Induces Expression of Timp1, Cmyc, Mmp3, and Grem1 in Isolated Cardiac Fibroblasts, Without Affecting Collagen and Acta2 Gene Expression

Diabetes‐associated fibrosis may be due to the actions of high glucose concentrations or related to effects independent of hyperglycemia. To examine whether the matrix‐preserving phenotype in diabetic fibroblasts involves effects of high glucose, we studied the effects of normal (5.6 mmol/L), high (17 mmol/L) and very high (25 mmol/L) glucose concentrations on expression of fibrosis‐associated genes in mouse cardiac fibroblasts (Figure 8, Figure S15). Consistent with the in vivo findings, high glucose concentrations did not significantly modulate Acta2, Col1a2, and Col3a1 expression (Figure 8A through 8C). Stimulation with very high glucose increased Timp1 and Cmyc expression, recapitulating the in vivo findings. In contrast to the in vivo observations, high glucose does not affect Thbs1 and SerpinE1 levels and modestly but significantly reduces Lox expression. Moreover, high glucose has effects on several genes, which are not modulated in db/db fibroblasts, increasing expression of Mmp3, Mmp13, and Grem1 and reducing Serpinh1 and Tgfb2 synthesis.

Figure 8. Exposure to high glucose does not induce myofibroblast conversion and matrix synthesis but stimulates expression of Timp1, Cmyc, and Mmp3.

A through L, Expression of fibrosis‐associated genes was assessed using a polymerase chain reaction array in isolated mouse cardiac fibroblasts exposed to increasing concentrations of glucose‐containing media. There was a trend toward reduced expression of Acta2 (the gene encoding a‐smooth muscle actin) upon stimulation with high concentrations of glucose (A). Moreover, glucose stimulation did not affect expression of the structural collagens Col1a2 (B) and Col3a1 (C). Consistent with the in vivo findings in diabetic fibroblasts, high glucose upregulated synthesis of matrix‐preserving Timp1 (D) and fibrogenic Cmyc (E). In contrast to the in vivo findings, glucose did not affect expression of Thbs1 (F) or Serpine1 (G) and modestly downmodulated Lox synthesis (H). Moreover, high glucose induced Mmp3 expression (I) and suppressed synthesis of Serpineh1 (J) and Tgfb2 (K). Expression of the bone morphogenetic protein inhibitor Grem1 was upregulated upon exposure to high glucose (L) (n=4/group).

Discussion

Our study examines for the first time the potential involvement of pericytes in any model of diabetic fibrosis. Using NG2^Dsred; PDGFRα^EGFP fibroblast:pericyte dual reporter mice, and a well‐validated lineage tracing system ⁴⁶ combining the inducible mural cell‐specific inducible NG2^CreER driver with the PDGFRα^EGFP reporter for reliable identification of fibroblasts, we found that diabetes is associated with an increase in the cardiac fibroblast to pericyte ratio, without stimulating pericyte to fibroblast conversion. In both lean and obese diabetic mouse hearts, cardiac pericytes and fibroblasts are nonoverlapping populations of interstitial cells with distinct transcriptomic profiles. Diabetes has modest effects on gene expression in pericytes and does not affect levels of genes encoding matrix‐degrading proteases and structural and matricellular matrix proteins. On the other hand, diabetic mouse cardiac fibroblasts do not convert into myofibroblasts and do not have increased collagen synthesis but acquire a matrix‐preserving and matrix cross‐linking phenotype, associated with accentuated expression of Timp1, SerpinE1, and Lox. Activation of diabetic cardiac fibroblasts is associated with upregulation of genes involved in regulation of the oxidative response. The changes in gene expression profile noted in diabetic fibroblasts are partially recapitulated by the effects of high glucose on cardiac fibroblasts. Thus, our observations suggest that diabetic myocardial fibrosis does not involve fibrogenic conversion, or activation of pericytes but may be driven by stimulation of matrix‐preserving properties of resident cardiac fibroblasts in the absence of increased collagen synthesis or myofibroblast conversion.

Pericytes in Tissue Fibrosis

A large body of evidence derived from both animal model studies and investigations in human patients suggests that pericytes actively participate in tissue fibrosis. In models of renal fibrosis, pericytes were found to convert to matrix‐synthetic myofibroblasts. ³⁴ In the injured lung, nestin+/NG2+ pericytes were identified as an important source of collagen, ⁵² and in a model of bleomycin‐induced pulmonary fibrosis, pericytes derived from FoxD1+ progenitors were found to undergo myofibroblast transdifferentiation, contributing to the development of fibrotic lesions. Moreover, single cell RNA‐sequencing studies in human patients with kidney fibrosis or Dupuytren's disease provided additional support to the role of pericytes as an important source of myofibroblasts. ³⁶ , ⁵³ In vitro studies have implicated several pathways in pericyte to fibroblast conversion, including TGF‐β signaling cascades, ⁵⁴ proinflammatory MyD88 (myeloid differentiation primary response 88)/IRAK (interleukin‐1 receptor‐associated kinase)4, ⁵⁵ STAT3 (signal transducer and activator of transcription 3) activation, ⁵⁶ and stimulation of the TWEAK (tumor necrosis factor‐like weak inducer of apoptosis)‐Fn14 axis. ⁵⁷ Taken together, these published observations suggest a remarkable plasticity of tissue pericytes, which respond to a wide range of cytokines and growth factors by activating a fibrogenic profile and even by converting to fibroblasts and myofibroblasts.

Pericytes in Diabetic Myocardial Fibrosis

Pericytes are major cellular targets in diabetes and have been implicated in the pathogenesis of complications, such as diabetic retinopathy and nephropathy. ⁴⁰ The metabolic perturbations associated with diabetes trigger activation of pathways reported to mediate pericyte to fibroblast transdifferentiation, such as TGF‐β and proinflammatory signaling cascades. ²⁰ , ²¹ , ⁵⁸ , ⁵⁹ , ⁶⁰ Thus, it is plausible to hypothesize that diabetes‐associated myocardial fibrosis may involve, at least in part, pericyte to fibroblast conversion. However, our data do not support this notion. Robust lineage tracing protocols, combined with the use of a reporter system for reliable identification of fibroblasts showed no significant pericyte to fibroblast conversion in the diabetic heart (Figure 3). Moreover, diabetes had very limited effects on expression of fibrosis‐associated genes in cardiac pericytes. In both lean and obese diabetic mice, pericytes and fibroblasts were identified as nonoverlapping interstitial cell populations with distinct transcriptional profiles (Figures 2, 4). Myocardial NG2+ pericytes harvested from obese diabetic db/db mice exhibit no significant increase in expression of fibrogenic mediators, structural and matricellular ECM genes, and proteases involved in ECM remodeling. Out of the 84 genes studied using a PCR array, only 2 genes were differentially expressed in diabetic mouse cardiac pericytes when compared with pericytes from normal mouse hearts (Figure 6). Timp3, the gene encoding TIMP3, a TGF‐β‐inducible antiprotease that restrains MMP activation ⁶¹ protecting from ventricular dilation, was upregulated in db/db pericytes. Global loss‐of‐function experiments showed that TIMP3 inhibits angiotensin II‐induced myocardial fibrosis ⁶² ; however, the role of pericyte‐derived TIMP3 in regulation of fibrotic remodeling in the diabetic heart is unknown. Expression of Cebpb (which encodes CCAAT/enhancer‐binding protein β /CEBPB) was also upregulated in diabetic mouse cardiac pericytes, although the increase in levels was not statistically significant (P=0.056). In fibroblasts, CEBPB has been implicated in cytokine‐induced MMP1 synthesis ⁶³ and has been suggested to mediate interferon‐γ‐induced inhibition of matrix secretion. ⁶⁴ Considering the comparable levels of expression of genes involved in matrix synthesis and remodeling between control and diabetic cardiac pericytes, the significance of the increased Cebpb levels in db/db pericytes is unclear.

Diabetic Fibrosis Is Associated With a Matrix‐Preserving/Matrix‐Cross‐Linking Fibroblast Phenotype, in the Absence of Myofibroblast Conversion and Collagen Upregulation

In infarcted ⁶⁵ , ⁶⁶ and in pressure‐overloaded hearts ⁶⁷ cardiac fibrosis is typically associated with the emergence of myofibroblasts, activated fibroblast‐like cells that express contractile proteins and secrete large amounts of ECM proteins. In contrast, our findings show that diabetic cardiac fibrosis reflects activation of a matrix‐preserving, matrix‐cross‐linking fibroblast phenotype in the absence of myofibroblast conversion or increased collagen synthesis. db/db mouse cardiac fibroblasts have low expression of Acta2 that is comparable to the levels noted in normal adult mouse cardiac fibroblasts (Figure 4) and do not express the myofibroblast markers α‐SMA and periostin (Figures S5 and S6). Moreover, diabetic mouse cardiac fibroblasts show no significant upregulation of genes encoding structural collagens (Figure 5). In contrast, db/db fibroblasts have increased expression of the genes encoding the matrix‐preserving antiproteases TIMP1 and PAI‐1, and the collagen cross‐linking enzyme lysyl‐oxidase. Levels of TSP1, a matricellular protein that potently activates TGF‐β ⁶⁸ and inhibits MMP activity, ⁶⁹ , ⁷⁰ are also increased in db/db fibroblasts (Figure 5). Thus, increased interstitial matrix deposition in type 2 diabetic db/db mice likely reflects the induction of antiproteases, such as TIMP1, PAI‐1, and TSP1, which attenuate MMP activity, and the upregulation of cross‐linking enzymes (such as lysyl oxidase) that reduce the susceptibility of the interstitial ECM to enzymatic digestion. Associative clinical observations support the notion that induction of matrix‐preserving and matrix‐cross‐linking signals in the myocardium may increase ventricular stiffness, contributing to diastolic dysfunction. In patients with type 2 diabetes and hypertension, high circulating levels of TIMP1 are associated with diastolic dysfunction. ⁷¹ Moreover, in patients with heart failure with preserved ejection fraction, myocardial matrix cross‐linking and LOX (lysyl oxidase) levels correlate with the severity of diastolic dysfunction, assessed through tissue Doppler imaging. ⁷²

The Mechanisms of Fibroblast Activation in the Diabetic Heart

Which signals selectively activate matrix‐preserving mediators in diabetic fibroblasts? In experimental studies, exposure of fibroblasts to high glucose has been suggested to act as a potent fibrogenic stimulus, ⁷³ through several different mechanisms. First, high glucose concentrations generate reactive oxygen species, triggering fibroblast activation. Our findings demonstrated that diabetes has pronounced effects on expression of genes involved in regulation of the oxidative response in fibroblasts but not in pericytes (Figure 7). When compared with fibroblasts harvested from normal mouse hearts, db/db mouse cardiac fibroblasts exhibit marked induction of the genes encoding the oxidative proteins COX2 and FMO2 and the antioxidant proteins SOD‐1 (superoxide dismutase 1) and HO‐1 (heme oxygenase 1), suggesting activation of both oxidative and antioxidant pathways. Second, hyperglycemia activates fibrogenic neurohumoral cascades, including the renin‐angiotensin‐aldosterone system. ⁷⁴ Third, exposure to high glucose induces proinflammatory and fibrogenic cytokines, such as interleukin‐1 ⁷⁵ and TGF‐β, ⁷⁶ , ⁷⁷ and may also potentiate the profibrotic effects of TGF‐β by inducing a rapid externalization of TGF‐β receptors at the cell surface. ⁷⁸ To examine whether fibroblast activation in diabetic hearts reflects the consequences of hyperglycemia, we stimulated isolated cardiac fibroblasts with high glucose concentrations. Exposure to high glucose only partially recapitulated the effects of diabetes on cardiac fibroblasts (Figure 8). Consistent with the in vivo findings, high glucose did not increase expression of myofibroblast markers, such as Acta2, but significantly upregulated matrix‐preserving Timp1 and induced the fibrogenic transcription factor Cmyc. However, in contrast to the in vivo findings, high glucose not only did not increase Lox and Serpinh1 expression but also modestly downregulated these genes. These findings suggest that diabetes‐associated cardiac fibrosis may involve both hyperglycemia‐mediated actions and effects independent of exposure to high glucose.

Limitations

Our study has several limitations. First, although NG2 labels a large fraction of cardiac pericytes, there is no optimal marker for identification and tracing of pericytes in mouse tissues. Thus, we cannot exclude the presence of an NG2‐negative mural cell subpopulation in the mouse heart with distinct functional properties. Second, the absence of reliable and specific markers makes identification of tissue fibroblasts challenging. To document pericyte to fibroblast conversion, we used a highly specific strategy by breeding mice used for pericyte lineage tracing with PDGFRα^EGFP fibroblast reporter animals. Reduced expression of PDGFRα has been reported in activated fibroblasts undergoing myofibroblast conversion. ⁷⁹ Although there are no myofibroblasts in diabetic hearts, we cannot exclude diabetes‐associated emergence of fibroblast populations with low PDGFRα expression. Third, we studied young mice at 6 months of age. Pericyte contribution to fibrosis may be more prominent in older animals. Fourth, to investigate the molecular profile of fibroblasts and pericytes, we used PCR arrays, thus limiting our analysis to a few hundred genes, relevant to fibrotic and oxidative responses. Although our findings do not support a central role of pericytes in diabetes‐associated cardiac fibrosis, changes in mural cell phenotype may be implicated in the pathogenesis of other important cellular responses in diabetic hearts, such as regulation of angiogenesis, microvascular function, or cardiac hypertrophy. Diabetes‐associated loss of pericyte coverage may contribute to microvascular rarefaction and may increase microvascular permeability. Moreover, cardiac mural cell dysfunction may mediate the perturbations in regulation of blood flow, typically noted in diabetic hearts. ⁸⁰ Finally, because of the use of multiple statistical hypothesis tests in analysis of gene expression data, some differences may be statistically significant by chance alone. For this reason, we have interpreted our data taking into account patterns of expression involving multiple genes rather than isolated comparisons.

Conclusions

We demonstrated that in the diabetic heart: (1) pericytes do not undergo fibroblast conversion and do not serve as an important source of ECM proteins and fibrogenic mediators, and (2) fibroblasts do not undergo myofibroblast conversion and do not increase their collagen‐synthetic capacity but acquire a matrix‐preserving phenotype, associated with increased expression of MMP inhibitors and matrix‐cross‐linking enzymes. The findings highlight the diversity of cellular responses involved in cardiac fibrosis. In contrast to the prominent role of myofibroblast conversion in infarcted and pressure‐overloaded hearts, metabolic dysregulation in diabetes and obesity increases collagen content by inhibiting matrix degradation, thus leading to net accumulation of collagen in interstitial and perivascular areas and promoting diastolic dysfunction. The distinct cellular mechanisms of fibrogenic activation may have important implications in design of antifibrotic therapies for patients with heart failure and prominent fibrosis.

Sources of Funding

Dr Frangogiannis' laboratory is supported by National Institutes of Health grants R01 HL76246, R01 HL85440, and R01 HL149407 and by US Department of Defense grants PR151029, PR181464, and PR211352. Dr Hanna is supported by a postdoctoral award by the American Heart Association (831084). Dr Tuleta is supported by a postdoctoral grant from the Deutsche Forschungsgemeinschaft (TU 632/1–1).

Disclosures

None.

Supporting information

Figures S1–S15