A Reservoir of Fibroblasts Promotes Recovery from Limb Ischemia

Abstract

Background:

The angiogenic response to ischemia restores perfusion so as to preserve tissue. A role for mesenchymal-to-endothelial transition in the angiogenic response is controversial. This study is to determine if resident fibroblasts contribute to angiogenesis.

Methods:

We utilized the murine model of hindlimb ischemia, and in vivo matrigel plug assay together with lineage tracing studies and single cell RNA-sequencing (scRNA-seq) to examine the transcriptional and functional changes in fibroblasts in response to ischemia.

Results:

Lineage tracing using Fsp1-Cre: R26R-EYFP mice revealed the emergence within the ischemic hindlimb of a small subset of YFP⁺ CD144⁺ CD11b⁻ fibroblasts (E* cells) that expressed endothelial cell (EC) genes. Subcutaneous administration of matrigel in Fsp1-Cre: R26R-EYFP mice generated a plug that became vascularized within 5 days. Isolation of YFP⁺ CD11b⁻ cells from the plug revealed a small subset of YFP⁺ CD144⁺ CD11b⁻ E* cells which expressed EC genes. Pharmacological or genetic suppression of innate immune signaling reduced vascularity of the matrigel plug and abrogated the generation of these E* cells. These studies were repeated using human fibroblasts, with FACS analysis revealing that a small percentage of human fibroblasts that were induced to express EC markers in matrigel plug assay. Pharmacological suppression or genetic knockout of inflammatory signaling abolished the generation of E* cells, impaired perfusion recovery and increased tissue injury after femoral artery ligation. To further characterize these E* cells, scRNA-seq studies were performed, and revealed eight discrete clusters of cells expressing characteristic fibroblast genes, of which two clusters (C5 and C8) also expressed some EC genes. Ischemia of the hindlimb induced expansion of clusters C5 and C8. The C8 cells did not express CD144, nor did they form networks in Matrigel, but did generate angiogenic cytokines. The C5 fibroblasts most resembled E* cells in their expression of CD144 and their ability to form EC-like networks in Matrigel.

Conclusions:

Together, these studies indicate the presence of subsets of tissue fibroblasts which seem poised to contribute to the angiogenic response. The expansion of these subsets with ischemia is dependent upon activation of innate immune signaling, and contributes to recovery of perfusion and preservation of ischemic tissue.

Introduction

Angiogenesis, the formation of new blood vessels from pre-existing vessels^{1, 2}, is involved in growth and development, as well as in tissue regeneration and the response to ischemia. In response to angiogenic factors released by ischemic tissue, endothelial cells (EC) sprout from existing capillaries, migrate, proliferate and form lumens to generate a functional microvasculature². Other processes which contribute to the formation of new vessels include intussusception of existing capillaries^{1, 3}, the incorporation of circulating endothelial progenitors^{4, 5}, and cytokines released from circulating angiogenic cells^{6, 7}.

We and others have shown that fibroblasts may be reprogrammed into endothelial cells in vitro^8–10. Notably, this mesenchymal-to-endothelial transition (MEndoT) in vitro requires the activation of cell-autonomous innate immune signaling¹¹. The activation of innate immune signaling increases DNA accessibility to facilitate the action of lineage determination factors to change the cell phenotype^11–13. Whether MEndoT contributes to the recovery from ischemia in vivo is controversial. One group has presented evidence that resident fibroblasts may transdifferentiate into ECs, to increase capillary density and to enhance perfusion in the murine myocardial infarction (MI) model¹⁴, which is mediated partially through p53¹⁴. However, a subsequent paper supported by extensive lineage tracing found no evidence of MEndoT in the murine MI model¹⁵.

The possible role of MEndoT in peripheral arterial disease has not been studied. As there are significant differences in the pathobiology of peripheral versus coronary artery disease¹⁶, we sought to determine if MEndoT contributed to perfusion recovery in the peripheral circulation. Accordingly, we employed a hindlimb ischemia model and an in vivo matrigel assay, as well as lineage tracing and single cell transcriptional profiling, to detect evidence of MEndoT. Surprisingly, we discovered subpopulations of fibroblasts in the mouse hindlimb that express some EC genes at baseline. These subpopulations of fibroblasts expands during ischemia, an effect that requires innate immune activation. The expansion of this subpopulation appears to contribute to vascularity, recovery of perfusion, and preservation of ischemic tissue.

Methods

All of our data, analytical methods, and study materials will be made available to other researchers for purposes of reproducing the results or replicating the procedure on request. A detailed Materials and Methods section can be found in the Data Supplement.

Animal care and use

All animal studies were approved by the Institutional Animal Care and Use Committee of Houston Methodist Research Institute. All procedures were in accordance with institutional guidelines.

Murine hindlimb ischemia

Hindlimb ischemia was induced in 9-month old male fibroblast-specific protein 1 (Fsp1)-Cre: R26R-EYFP mice as previously reported with slight modification^{11, 17}.

In vivo matrigel plug assay

Mice were injected subcutaneously with vascular endothelial growth factor (VEGF)-reduced matrigel (300ul) plus heparin (30 U/ml) with: 1.) Polyinosinic:polycytidylic acid, PIC (30ng/ml); or 2.) angiogenic agents VEGF (50ng/ml), bone morphogenetic protein 4 (BMP4, 20ng/ml), basic fibroblast growth factor (bFGF, 20ng/ml), 8-Br-cAMP (0.1mM) or 3.) PIC plus angiogenic agents. Matrigel plugs were removed after five days for histology and FACS study.

FACS analysis

Single cell suspension was stained and run on BD LSRII machine. Data were analyzed by FlowJo software. Cell sorting was performed by BD FACS Aria.

Single cell RNA-sequencing (scRNA-seq)

YFP⁺CD11b⁻ cells were FACS sorted from hindlimb muscle isolated from mice without ischemic surgery or 28 days after ischemic surgery. 5000 Cells were captured using 10X chromium system following manufacturer’s protocol¹⁸. Each library was sequenced on one lane on an illumina Hi-seq 4000 system.

Data analysis

Results were expressed as the mean±SEM. Statistical comparisons between two groups were performed via Student t test. Statistical comparisons among three or four groups were performed via one-way ANOVA followed by Sidak’s multiple comparisons test. When measurements at different days were taken on the same mice, the data were analyzed using two-way mixed-effects ANOVA followed by Sidak’s multiple comparisons test, to account for the correlation of serial measurements within the same animal. *P<0.05 was considered significant.

Results

A subset of Fsp1⁺ cells undergoes transdifferentiation to endothelial lineage in vitro

To determine if MEndoT contributes to the perfusion recovery in hindlimb ischemia, we used a lineage tracing strategy. We labelled fibroblasts by crossing two transgenic strains: mice with a Cre recombinase gene driven by the promoter sequence of Fsp1¹⁹; and mice bearing the lineage reporter R26R-EYFP. The R26R-EYFP mice have a loxP-flanked STOP sequence followed by EYFP gene inserted into the ROSA26 locus. When these two strains are crossed, the F1 generation expresses Cre recombinase in Fsp1 expressing cells, deleting the STOP sequence and permitting EYFP expression. Thus, all cells that were at one time Fsp1⁺ will express EYFP and exhibit yellow fluorescence. The Fsp1-Cre model has been used to track the reprogramming of cardiac fibroblasts to myocytes or ECs in vivo^{14, 20}. Fsp1 may also express in hematopoietic cells. In our analyses we used FACS to exclude CD11b⁺ immune cells as described below.

We next performed studies to determine if YFP⁺ fibroblasts might have the capacity to transdifferentiate into endothelial lineage. We isolated YFP⁺ tail tip fibroblasts (TTFs) from the Fsp1-Cre: R26R-EYFP mice. These cells maintained the typical spindle-like morphology of cultured fibroblasts in vitro (Fig. ⅠA in the Supplement). When we exposed the TTF to our previously reported transdifferentiation formulation (see below, a combination of TLR3 agonist and angiogenic agents), we were able to generate a small subpopulation of CD144⁺ cells as we have previously observed (Fig. ⅠC in the Supplement)¹². These CD144⁺ cells formed networks in matrigel and incorporated acLDL, consistent with an endothelial phenotype (Fig. ⅠD in the Supplement). We have previously reported that the CD144⁺ cells obtained using this protocol also generate nitric oxide, and have a transcriptional profile by RNA-seq that closely resembles genuine ECs¹¹. For convenience, we term these fibroblast-derived CD144⁺ cells “endothelioid” or E* cells.

E* cells emerge in vivo in a model of transdifferentiation

The studies above, and our previous work, indicates that cell-autonomous innate immune signaling plays a significant role in transdifferentiation in vitro¹¹. We thus determined if such transdifferentiation occurred in vivo using a matrigel plug assay. Specifically, to matrigel we added components of our transdifferentiation formulation^{11, 12}, which has two essential components: 1. An activator of innate immune signaling, the toll-like receptor 3 (TLR3) agonist PIC (30ng/ml) which we have shown to increase DNA accessibility and phenotypic fluidity of the cell¹³; and 2. Angiogenic agents (AA) to promote endothelial gene activation, including VEGF (50ng/ml); BMP4 (20ng/ml); bFGF (20ng/ml); and 8-Br-cAMP 0.1mM. Experimental groups included matrigel alone; matrigel with PIC; matrigel with AA; or matrigel with PIC plus AA. These mixtures were injected subcutaneously into Fsp1-Cre: R26R-EYFP mice. Matrigel plugs were removed after five days for FACS, gene expression and histology examination. Matrigel plugs that incorporated PIC plus AA showed the most abundant neovascularization and blood content (Fig. 1A), which was confirmed by HE staining and vessel quantification (Fig. 1B&C). As shown in FACS analysis in Fig. 1D, YFP⁺CD144⁺ E* cells and YFP⁻CD144⁺ resident ECs increased markedly in matrigel plugs containing PIC plus AA. We observed the greatest number of E* cells in the PIC plus AA group, with around 15% of the YFP⁺ cells also expressing the EC marker CD144 (Fig. 1E&F). Notably, E* cells were seen in all groups, albeit at a lower level, with ~ 4% yield of YFP⁺CD144⁺ E* cells in the other groups (Fig. 1E&F). Mouse CD31 and CD144 gene expression was markedly upregulated in the matrigel plugs containing PIC plus AA, compared to the other groups (Fig. 1G&H). These studies indicated that insertion of the matrigel plug resulted in its neovascularization, precipitated by the invasion of resident ECs (YFP⁻CD144⁺), as well as resident fibroblasts (YFP⁺CD144⁻), some of which had features of E* cells (YFP⁺CD144⁺).

Fig. 1. Activation of TLR3 signaling facilitated fibroblast transdifferentiation in vivo.

VEGF-reduced matrigel (300ul) plus heparin (30U/ml) were mixed with either PIC (30ng/ml), angiogenic agents (AA) (VEGF 50ng/ml, BMP4 20ng/ml, bFGF 20ng/ml, 8-Br-cAMP 0.1mM) or PIC plus AA and injected into Fsp1-Cre: R26R-EYFP mice. Matrigel plugs were removed after five days. A. Gross morphology of matrigel plugs. B. HE staining of matrigel plugs. Scale, 100μm. C. Number of vessels per mm2 quantified from HE staining. D. Matrigel plugs were digested to get single cell suspensions. The FACS plots shown are cells gated on CD11b⁻ cells. E. The percentage of CD144⁺ cells in gated YFP⁺CD11b⁻ cells. F. Quantification of percentage in CD144⁺ cells in gated YFP⁺CD11b⁻ cells. G. Mouse CD31 gene expression in matrigel plugs. H. Mouse CD144 gene expression in matrigel plugs. One-way ANOVA was used for data analysis. All data are presented as mean ± S.E.M.. *, P< 0.05 compared to PIC group; #, P< 0.05 compared to AA group.

These data are consistent with the transdifferentiation of a subset of Fsp1⁺ cells into E* cells during angiogenesis in vivo. Alternatively, these data are consistent with the migration and/or expansion of tissue-resident E* cells into the matrigel plug. In either case, innate immune activation facilitates the emergence of these cells in the matrigel plug.

To further test the hypothesis that NFκB mediates the emergence of E* cells in vivo and contributes to angiogenesis, we examined the effect of NFκB inhibitor Bay117082 using the matrigel plug assay in Fsp1⁺ lineage tracing mice. We used the same experimental groups as in Fig. 1, with the addition of Bay117082 or vehicle. The inhibitor Bay117082 nearly abrogated angiogenesis and blood content in the matrigel plug assay (Fig. 2A) confirmed by HE staining (Fig. 2B&C). Bay117082 treatment also decreased the intra-plug YFP⁺ population and the number of YFP⁺CD144⁺ E* cells (Fig. 2D, E&F), suggesting that NFκB mediates YFP⁺ cell migration and infiltration, as well as the emergence of E* cells. Thus, inhibition of NFκB activity blocked Fsp1⁺ cell migration and emergence of E* cells, in association with decreased angiogenesis.

Fig. 2. NFκB mediated angiogenic transdifferentiation in vivo.

PIC (30ng/ml) and angiogenic agents (AA) (VEGF 50ng/ml, BMP4 20ng/ml, bFGF 20ng/ml, 8-Br-cAMP 0.1mM) were added to VEGF-reduced matrigel (300ul) plus heparin (30U/ml) with vehicle (PIC+AA) or with Bay117082 (Bay11, 200μM). Matrigel was injected into Fsp1-Cre: R26R-EYFP mice and matrigel plugs were removed after 5 days. Matrigel plugs were digested, and single cell suspension prepared for FACS analysis. A. Gross morphology of matrigel plugs. B. HE staining of matrigel plugs. Scale, 100μm. C. Number of vessels per mm² quantified from HE staining. D. The FACS plots shown were derived from cells gated on CD11b⁻ F4/80⁻ cells. E. Quantification of percentage of YFP⁺ cells in gated CD144⁺ CD11b⁻ F4/80⁻ cells. F. Quantification of percentage of YFP⁺ cells in gated CD11b⁻ F4/80⁻ cells. *, P< 0.05 compared to Veh group. All data are presented as mean ± S.E.M.. Veh, vehicle.

E* cells emerge with limb ischemia

To determine if E* cells may emerge in a more clinically relevant model, Fsp1-Cre: R26R-EYFP mice were subjected to femoral artery ligation as previously described^{11, 21}. Hindlimb muscles were removed from the operated (ischemic) and contralateral limb at 4, 7 and 21 days after operation, and digested for FACS analysis. FACS analysis (with gating to exclude CD11b⁺ immune cells) revealed a progressive, localized accumulation of a population of YFP⁺ cells that expressed the vascular endothelial marker CD144 resembling the E* cells derived with the prior models above. These E* cells began to appear at 7 days after injury and increased until 21 days (Fig. 3A). The total YFP⁺ cell number increased substantially from 7 to 21 days (Fig. 3B). The E* cells accounted for 4–6% of the total CD144⁺ cells at day 21 (Fig. 3C &D). By this time (21 days), perfusion of the ischemic hindlimb had recovered to about 80% of the contralateral limb (Fig. 3E).

Fig. 3. A subset of tissue fibroblasts contributes to vascular regeneration and perfusion during limb ischemia.

Fsp1-Cre:R26R-EYFP mice were subjected to hindlimb ischemia. Limb muscles were dissected from mice before and 4, 7 and 21 days after femoral artery ligation, and digested for FACS analysis. A. The FACS plots shown are cells gated to exclude CD11b⁺F4/80⁺ cells. B. Quantification of percentage of YFP⁺ cells in CD11b⁻F4/80⁻ cells. C. The FACS plots shown are cells gated on CD144⁺CD11b⁻F4/80⁻ cells. D. Quantification of percentage of YFP⁺ cells in CD144⁺CD11b⁻F4/80⁻ cells. E. Blood perfusion by Laser Doppler immediately and 21 days after femoral artery ligation. F. CD31 gene expression. G. eNOS gene expression. H. vWF gene expression. I. Fsp1 gene expression. J. IHC staining of CD144 and YFP in hindlimb tissue at day 0 and day 21. Scale, 100μm. Brown arrow points to CD144 staining and red arrow points to YFP staining. One-way ANOVA was used for data analysis. All data are presented as mean ± S.E.M.. *, P< 0.05 compared to day 0 in B&D; FB group in F-I). FB, fibroblasts; E* cells, the YFP⁺CD144⁺CD11b⁻ population.

To characterize the E* cells that were detected in the ischemic limb, we harvested ischemic limbs, disaggregated the tissue, and FACS sorted cells at 21 days. We then examined the gene expression of YFP⁺CD144⁻ fibroblasts, YFP⁻CD144⁺ ECs, and YFP⁺CD144⁺ E* cells using RT-PCR. Compared with YFP⁺CD144⁻ fibroblasts, YFP⁺CD144⁺ E* cells had significantly upregulated level of endothelial specific genes such as CD31, eNOS and vWF (Fig. 3F, G&H). Furthermore, fibroblast specific genes such as Fsp1 (Fig. 3I) were markedly downregulated in E* cells, as compared to YFP⁺CD144⁻ cells. However, expression of endothelial genes in E* cells were not as high as resident endothelial lineage cells, suggesting that these cells are not fully mature ECs. Dual IHC staining of CD144 and YFP in hindlimb tissue showed that, under physiological conditions, YFP⁺ cells primarily located in the interfascicle regions of the skeletal muscle. These YFP⁺ cells did not co-localize with CD144 signal nor within vascular structures (Fig. 3J, left panel). After ischemic recovery, some YFP⁺CD144⁺ E* cells were observed in vascular structures (Fig. 3J, right panel).

Dexamethasone inhibits the emergence of E* cells and recovery from hindlimb ischemia

To determine if innate immune activation is also playing a role in the emergence of E* cells in the ischemic hindlimb, we used the synthetic glucocorticoid dexamethasone, which has potent anti-inflammatory activities through NFκB inhibition²². We hypothesized that dexamethasone would inhibit the emergence of Fsp1⁺CD144⁺ E* cells and thereby inhibit angiogenesis. To test this hypothesis, we subjected the Fsp1-Cre:R26R-EYFP mice to femoral artery ligation. In one group, dexamethasone was injected i.m. at 0.6mg/kg once after surgery and once at 7 days after injury. Blood perfusion was monitored before and after injury, and 4, 7, 14 and 21 days after injury. The recovery of blood perfusion over time was reduced in the dexamethasone treated animals (Fig. 4A). Moreover, the dexamethasone injected mice exhibited more severe tissue loss, as assessed in a blinded manner using the established clinical scoring system (Fig. 4B&C). We dissociated the tissue and studied the cellular composition by FACS. Notably, FACS analysis revealed that dexamethasone markedly reduced the expansion of the Fsp1⁺ population and reduced the generation of YFP⁺CD144⁺ E* cells in the ischemic limbs from control mice (Fig. 4D).

Fig. 4. Dexamethasone inhibits transdifferentiation and recovery from hindlimb ischemia.

Fsp1-Cre:R26R-EYFP mice were subjected to hindlimb ischemia. Dexamethasone was injected i.m. at 0.6mg/kg once after surgery and at 7 days after surgery. Blood perfusion was monitored before, immediately after, and 4, 7, 14 and 21 days after surgery. Limb muscle was dissected from mice 28 days after surgery and digested for FACS analysis. A. Mean perfusion ratio of ischemic to unoperated limb at different time points after injury. Two-way ANOVA was used for this data analysis. B. Representative image of foot in vehicle and dexamethasone treated mice. C. Hindlimb ischemia score at day 21. D. Representative FACS plots shown are cells gated on CD11b⁻ F4/80⁻ cells. All data are presented as mean ± S.E.M.. *, P< 0.05 compared to Veh group. Dex, dexamethasone; Veh, vehicle.

NFκB mediates the emergence of E* cells in vitro and in vivo

Our previous work indicated that NFκB signaling is required for transdifferentiation in vitro¹². NFκB activation can be achieved by two pathways: the canonical pathway (p50-RelA), which is usually activated in response to TLR agonists and cytokines; and the non-canonical NFκB pathway (p52-RelB), important in B cells. To understand which NFκB plays the key role in transdifferentiation, we assessed the yield of induced ECs from Nfκb p50-deficient or p52-deficient TTF subjected to our standard transdifferentiation protocol (PIC + angiogenic agents, as described above). Although both knockouts impaired transdifferentiation, the Nfκb p50^−/− mouse TTFs manifested a more severe impairment in transdifferentiation efficiency in vitro (Fig. Ⅱ in the Supplement).

To further examine the role of NFκB signaling in the contribution of fibroblasts to angiogenesis, Fsp1-Cre:Rela^flox/flox mice were generated to knockout Rela in Fsp1+ fibroblasts. We hypothesized that, by blocking NFκB signaling in fibroblasts, angiogenesis would be inhibited.

Matrigel with PIC and angiogenic agents (AA) (VEGF 50ng/ml, BMP4 20ng/ml, bFGF 20ng/ml and 8-Br-cAMP 0.1mM) was injected into Fsp1-Cre:Rela^flox/+ and Fsp1-Cre:Rela^flox/flox mice and matrigel plugs were removed after 5 days. By comparison to the matrigel plugs from control (Fsp1-Cre:Rela^flox/+) mice, those retrieved from Fsp1-Cre:Rela^flox/flox mice manifested markedly less neovascularization and blood content (Fig. 5A), which was confirmed by H&E staining (Fig. 5B). To determine the effect of knocking down NFκB activity on angiogenesis in response to ischemia, we induced hindlimb ischemia in Fsp1-Cre:Rela^flox/+ (control) and Fsp1-Cre:Rela^flox/flox mice. Blood perfusion was monitored before, immediately after, and up to 21 days after surgery in Fsp1-Cre:Rela^flox/+ and Fsp1-Cre:Rela^flox/flox mice. The Fsp1-Cre:Rela^flox/flox mice showed decreased blood flow recovery (Fig. 5C) at day 14 and day 21 , and more severe clinical symptoms (Fig. 5D) compared with control Fsp1-Cre:Rela^flox/+ mice. These data reveal that NFκB signaling in Fsp1+ fibroblasts is critical for angiogenesis. However, the angiogenic effect of Fsp1+ myeloid cells can not be excluded.

Fig. 5. Fibroblast specific knockout of Rela inhibits vascular regeneration.

PIC (30ng/ml) and angiogenic agents (AA) (VEGF 50ng/ml, BMP4 20ng/ml, bFGF 20ng/ml, 8-Br-cAMP 0.1mM) were added to VEGF-reduced matrigel (300ul) plus heparin (30U/ml). Matrigel was injected into Fsp1-Cre:Rela^flox/+ and Fsp1-Cre:Rela^flox/flox mice and matrigel plugs were removed after 5 days. A. Gross morphology of matrigel plugs. B. HE staining of matrigel plugs. Scale, 100μm. C. Fsp1-Cre:Rela^flox/+ and Fsp1-Cre:Rela^flox/flox mice were subjected to hindlimb ischemia. Blood perfusion was monitored before, immediately after, and 4, 7, 14 and 21 days after surgery. Mean perfusion ratio of ischemic to unoperated limb at different time points after injury. Two-way ANOVA was used for this data analysis. D. Hindlimb ischemia score at day 21. All data are presented as mean ± S.E.M.. *, P< 0.05 compared to Fsp1-Cre:Rela^flox/+ group.

Persistence of endothelial lineage

Although we find that E* cells emerge during angiogenesis in the matrigel plug assay, it is not clear whether the E* cells persist in the long term. To determine the persistence of E* cells, we injected matrigel mixed with PIC plus AA into Fsp1⁺ lineage tracing mice and removed the matrigel plugs after 5, 14 and 28 days. It is reported that the neovascularization in matrigel plugs does not persist^{23, 24}. Similarly, in our study, matrigel plugs removed after 5 days manifested a dense vascular network. However, vascularity of the matrigel plugs was reduced by 14 and 28 days as shown by gross morphology (Fig. ⅢA in the Supplement) and by gene expression of endothelial specific markers CD31 and CD144 (Fig. ⅢB&C in the Supplement). Interestingly, the percentage of YFP⁺CD144⁺ E* cells represented about 5% of all CD144⁺ cells at day 5, but this percentage increased up to 30% at day 14 and remained stable through day 28 (Fig. ⅢD&E in the Supplement). Thus, it seems that the E* cells are more resistant to the vascular regression that affects the native ECs in the matrigel plug assay.

Human fibroblasts acquire features of E* cells in vivo

We wished to determine if human fibroblasts harbored the capacity to acquire markers of endothelial lineage in vivo. Accordingly, matrigel was mixed with or without HNDF. These HNDFs showed typical spindle-shaped fibroblast morphology in culture and did not express endothelial markers CD31 nor CD144 (Fig. ⅣA & B in the Supplement). We then injected the matrigel mixture in immunodeficient NOD-SCID mice. After five days, the matrigel plugs were removed for FACS analysis, gene expression and histology examination.

Angiogenesis was enhanced in matrigel plugs containing HNDFs (Fig. ⅣC in the Supplement). To verify the presence of human cells manifesting features of endothelial lineage, we used a human CD31 antibody (clone JC70A) that does not cross react with mouse CD31 (Fig. ⅤA&B in the Supplement). FACS analysis of the digested matrigel plugs revealed a distinct hCD31⁺ population (Fig. ⅣD&E in the Supplement). In addition, human CD31 gene expression in the matrigel was significantly upregulated compared with its basal level expression in HNDFs (Fig. ⅣF in the Supplement). Taken together, these data suggest that a subset of human fibroblasts can acquire features of ECs in vivo.

This subset of human fibroblasts may be the functional equivalent of murine E* cells. To determine the role of proliferation in the emergence of these putative E* cells, we repeated the above Matrigel studies together with BrdU incorporation. BrdU was injected intraperitoneally into the mice at 12-hour intervals 24 hours before matrigel plug collection. We once again observed the emergence of hCD31⁺ cells from human fibroblasts in the Matrigel plugs. In these cells there was minimal incorporation of BrdU (Fig. ⅣG in the Supplement), suggesting that these cells are not highly proliferative.

scRNA-seq identifies Fsp1⁺CD11b⁻ subpopulations mediating angiogenic response during hindlimb ischemic recovery

To further characterize the subsets of limb fibroblasts we performed scRNA-seq on all YFP⁺CD11b⁻ cells that were FACS sorted from hindlimb muscles isolated either before or 28 days after femoral artery ligation. We are particularly interested in 28 days since the blood flow recovery plateaus at this time point based on our previous observations. A total of 2668 and 2653 cells were captured in day 0 and day 28 respectively (Fig. ⅥA in the Supplement). Interestingly, the median number of genes detected per cell was nearly double in the cells recovered at day 28, as compared to cells recovered at day 0 (2849 vs 1592) (Fig. ⅥA in the Supplement), suggesting that during ischemic recovery, the Fsp1⁺CD11b⁻ cells have a more active transcriptional profile.

We identified eight clusters of YFP⁺CD11b⁻ cells (Fig. 6A) based on their transcriptional similarity using principal components analysis (Fig. Ⅶ in the Supplement) with t-SNE (t-distributed stochastic neighbor embedding, a non-linear dimensional reduction method). 20 principal components (Fig. Ⅷ in the Supplement) were used to distinguish these eight distinct clusters. All YFP⁺CD11b⁻ cells had high gene expression level of fibroblast markers such as Thy1²⁵, S100a4 (Fsp1^{14, 26}), Vim (Vimentin)²⁷, Lamc1²⁸, Nav1²⁸ and Igfbp6²⁸ (Fig. Ⅸ in the Supplement). Cells within all 8 clusters have high expression of Tcf4/Tcf7l2, a key transcription factor for skeletal fibroblasts²⁹. In addition, several key extracellular matrix genes, such as Fn1 (Fibronectin) and Col4a4 (Collagen IV) are highly expressed in all clusters. Collagen IV is the major form of collagen secreted by skeletal fibroblasts ³⁰. In a few clusters, collagen VI³¹ and/or collagen I genes were also detected.

Fig. 6. scRNA-seq revealed dynamic changes of cell subpopulations during ischemic recovery.

A. YFP⁺CD11b⁻ cells were sorted from hind limb muscles collected from day 0 before femoral artery ligation and 28 days after surgery for scRNA-seq. t-SNE plot of combined cells revealed 8 cell clusters based on transcriptional similarity. Cells are colored either by cluster ID (left panel) or sample ID (right panel). B. Percentage distribution of cell numbers in all 8 cell clusters of day 0 and day 28 sample respectively. C. Heatmap showing expression profile of top 10 marker genes for each cell cluster. D. Hierarchical clustering of cell clusters based on average expression of top cell marker genes. Pairwise distance between cell clusters is measured by Spearman Rank Correlation Coefficient.

These cells also had very low gene expression level of pericyte marker Pdgfrb and Cspg4 (NG2)³², low gene expression level of the smooth muscle cell/myofibroblast marker Acta2, as well as undetectable levels of the satellite cell markers Pax7, Pax3 and Myf5³³ (Fig. Ⅸ in the Supplement). Consistent with the negative selection of CD11b, we also observed low expression level of monocyte/macrophage identity genes such as Itgam (CD11b), Itgax (CD11c), Adgre1 (F4/80), or Fcgr1 (Fig. Ⅸ in the Supplement). Altogether, the transcriptional profiles supported the notion that each of these eight cell subsets were of a fibroblast phenotype.

Several fibroblast specific gene promoters have been used to trace cardiac fibroblasts, including periostin²⁶, α2 (type I) collagen¹⁴, Tcf21³⁴, PDGFRα³⁵ and Fsp1^{14, 26} while SMC actin has been used to trace myofibroblasts. These different lineage tracing methods may detect different subpopulations of tissue fibroblasts and/or may have tissue-specificity³⁶. Previous studies^{37, 38} have maintained that the Fsp-1 Cre is not specific for cardiac fibroblasts. A recent study suggests that Fsp1⁺ fibroblast population is angiogenic while the SMC actin⁺ fibroblast populations is not³⁶, suggesting that these lineage tracing systems may detect fibroblast subpopulations that do not completely overlap³⁶. In this regard, our scRNA-seq data show that Tcf21 is not detected in any of the 8 clusters of hindlimb fibroblasts, while Pdgfra was only detected in cluster 1, 6 and 7, periostin (Postn) only in cluster 1 and 2, collagen I genes (Col1a1) only in cluster 1, 2, 3 and 6, and Acta2 only in 1, 2 and 8 (Fig. Ⅸ in the Supplement). Thus, there is virtue in using a more inclusive promoter when lineage tracing is combined with single cell analysis, so as to avoid excluding a subpopulation in the cluster analyses.

The size of each cluster, as a percentage of the total YFP⁺ cells at day 0 and day 28 is shown in Fig. 6B. Interestingly, cluster 1 and cluster 2 contributed significantly less to total cell count at day 28. By contrast, cluster 5–8 increased as a percentage of the total cell count from day 0 to day 28. Cluster 8 sharply increased and accounted for around 60% of the total YFP⁺ population at day 28 (Fig. 6B). Fig. 6C shows the heat map of highly differentially expressed genes with all cells from both days 0 and 28 clustered based on the similarity of the transcriptional profile. Clusters 5, 7 and 8 had the highest transcriptional similarity as assessed by a hierarchical clustering analysis (Fig. 6D).

To identify subpopulations that might contribute to the angiogenic process, we visualized the expression of early and mature endothelial markers by ViolinPlot (Fig. 7A) among the eight clusters of cells. Early endothelial markers such as Kdr, Tie1 and Ecscr were expressed mainly in cluster 5–8. Mature endothelial markers such as Nos3 (eNOS) and vWF were expressed almost exclusively in cluster 5–8. Cluster 5 and cluster 8 were most highly enriched in early and mature endothelial markers (Fig. 7A). Pathway enrichment analysis showed that cluster 5 (Fig. 7B), cluster 8 (Fig. 7C) and cluster 7 (Fig. Ⅹ in the Supplement) were all enriched in endothelial related pathways such as angiogenesis and vasculature development. These data may indicate that clusters 5 and 8 are most poised to support angiogenesis.

Fig. 7. Identification of EC-like cell clusters.

A. Expression of EC marker genes across cell clusters showed by violin plot. B&C. Pathway enrichment analysis showed cluster 5 and 8 were enriched with endothelial related pathways.

To examine the potential developmental relationship between the 8 different fibroblast cell clusters, we used Slingshot³⁹ to infer the most probable differentiation trajectory based on pseudo-time analysis. Applying Slingshot to cell projections on the 2-dimentional UMAP space (Fig. ⅪA in the Supplement), we identified three single-branch trajectories comprising of 1–2–3–4–8–5, 1–2–3–4–8–7 and 1–2–3–4–8–6 (Fig. Ⅺ B-E in the Supplement). In addition, the finding that day 28 sample are more enriched in cells with high pesudotime values compared to day 0 samples (Fig. ⅪF in the Supplement) is consistent with our earlier observation that both transcriptional complexity and cellular diversity are increased during ischemic recovery (Fig. ⅥA-B, ⅦB in the Supplement).

Characterization of the function of fibroblast subsets

As discussed above, of all the resident YFP⁺ fibroblasts in the hindlimb, it was the C5 and C8 clusters that expressed the most early and mature endothelial genes. Accordingly, we focused on characterizing their angiogenic potential in vitro. The C5 cells have the highest gene expression of CD144 (Fig. 7A), and are thus most representative of the E* cells defined above. Thus, we used CD144 expression to isolate these cells. Because IFITM3 and Galectin-3 (but not CD144) were preferentially expressed in C8 cells, we used these surface markers in our sorting strategy to select C8 cells (Fig. Ⅻ in the Supplement).

Subsequently, we subjected Fsp1-Cre:R26R-EYFP mice to hindlimb ischemia and FACS sorted YFP⁺CD11b⁻ cells (FB, fibroblast) at day 0, YFP⁺CD144⁺CD11b⁻ cells (C5) (Fig. 8A) and YFP⁺ IFITM3⁺Galectin-3⁺CD144⁻CD11b⁻ cells (C8) (Fig. 8B) at day 28. In 2D culture on plastic plates, all three populations showed the typical spindle shape morphology of fibroblasts (Fig. 8D, left panel). Intriguingly, when cultured on matrigel, the C5 subset (but not the C8, nor the YFP⁺ cells derived from the skeletal muscle at day 0) could form networks similar to those typically generated by authentic ECs. Although C8 cells did not form networks in matrigel, the C8 cells could secrete angiogenic cytokines such as Pdgfa, Pdgfec and Sdf-1 to a greater extent that the general population of fibroblasts derived from the hindlimb at day 0 (Fig. 8C).

Fig. 8. Characterization of the function of fibroblast subsets in vascular formation.

Fsp1-Cre:R26R-EYFP mice were subjected to hindlimb ischemia. Limb muscles were dissected and FACS sorted for the following populations: YFP⁺CD11b⁻ cells (FB, fibroblast) at day 0, YFP⁺ CD144⁺CD11b⁻ cells (C5) and YFP⁺IFITM3⁺Galectin-3⁺CD144⁻CD11b⁻ cells (C8) at day 28. A. Quantification of percentage of CD144⁺ cells in CD11b⁻YFP⁺ cells. B. Quantification of percentage of IFITM3⁺Galectin-3⁺ cells in CD11b⁻CD144⁻YFP⁺ cells. C. Angiogenic cytokine level detected in FB and in cluster 8 cells using mouse angiogenesis proteome profiler antibody arrays. Only significantly upregulated cytokines were presented. D. Images of isolated cell population under bright field or under green fluorescence channel. Left panels are isolated cell population on plastic dish. Right panels are isolated cell population on matrigel. All data are presented as mean ± S.E.M.. *, P< 0.05 compared to FB group. FB, fibroblast.

Discussion

The seminal findings of this work include evidence that there are distinct subpopulations of fibroblasts in the murine hindlimb that can be defined using functional assays, proteomic markers and transcriptional profile. Furthermore, we show that a subpopulation of fibroblasts expressing CD144⁺ (E* cells) increases with activation of innate immunity and/or ischemia. Genetic knockout or pharmacological suppression of NFκB abrogates the generation of E* cells. In the hindlimb ischemia model, suppression of inflammatory signaling abrogates the generation of fibroblast-derived E* cells, impairs perfusion recovery, and exacerbates tissue loss after femoral artery ligation. scRNA-seq confirmed the existence of 8 distinct clusters of fibroblasts. Clusters 5 and 8 express genes encoding angiogenic pathways, and both subsets expand dramatically with ischemia. Cluster 5 most resembles E* cells in the expression of CD144 and in the ability to form networks in Matrigel, whereas cluster 8 generates angiogenic cytokines. These observations reveal subsets of fibroblasts that support angiogenesis. Whether these cells are derived from expansion or transdifferentiation of resident fibroblasts, their emergence in ischemic tissue is dependent upon inflammatory signaling, an observation consistent with our prior work^{11–13, 40–44} showing that inflammatory signaling promotes DNA accessibility and phenotypic plasticity.

Transdifferentiation of fibroblasts to ECs was reported to occur in a murine MI model¹⁴. However, a subsequent study which used more extensive lineage tracing techniques raised questions about this finding¹⁵. Since the myocardium has limited angiogenic capacity after an ischemic event⁴⁵, we wondered if such transdifferentiation might be detectable in models where the angiogenic response to ischemia is more robust, i.e. the matrigel plug assay and the hindlimb ischemia model. Our initial lineage tracing data in the ischemic hindlimb model detected low levels of YFP⁺CD144⁺ fibroblasts (E* cells), suggesting that transdifferentiation might occur. Indeed, in YFP⁺CD144⁺ (E*) cells isolated from the ischemic hindlimb, the fibroblast marker Fsp1 was downregulated, whereas endothelial genes CD31, eNOS and vWF were upregulated, although not to the levels of mature ECs.

However, our subsequent scRNA-seq studies suggested that the story is more complex. These studies indicated that as many as eight subsets of fibroblasts reside in the murine limb. These subsets of fibroblasts share common fibroblast genes, and by transcriptional profiling are not immune cells, ECs, pericytes or satellite cells. The fibroblast nature of these 8 subsets is confirmed by their high expression of genes known to be preferentially expressed in fibroblasts (e.g. Tcf4/Tcf7l2, a key transcription factor for skeletal fibroblasts²⁹, and key extracellular matrix genes, such as Fn1 (Fibronectin) and Collagen IV gene (Col4a4) which encodes the major form of collagen secreted by skeletal fibroblasts³⁰. Although these sets of fibroblasts are highly related, they can be separated into Clusters by their transcriptional profile.

To investigate the potential developmental trajectories of these eight subpopulations, we performed unsupervised pseudo-time analysis and the data suggest that cells share a common 1–2–3–4–8 trajectory and then branch into cluster 5, 6 and 7. Although further experimental evidence will be definitely required to prove that this trajectory occurs in development, and/or is recapitulated with ischemia, the attenuation of cluster 1/2 and the expansion of cluster 5/8 during ischemia could be explained by such a developmental trajectory.

Two of the Clusters (C5 and 8) express some genes characteristic of endothelial progenitors or mature ECs. Intriguingly, there is a dramatic shift in the composition of the fibroblast population during ischemia, with a marked increase in the size of Clusters 5 and 8 as a proportion of the total number of tissue fibroblasts in the ischemic hindlimb. We isolated total tissue fibroblasts at Day 0 (control fibroblasts), and compared them to cells from the Day 28 C5 and C8 clusters. In two-dimensional culture, cells from all three groups adhered to plastic and exhibited the spindle-shaped morphology typical of fibroblasts. However, when grown in Matrigel, only C5 cells formed networks characteristic of authentic ECs. However, C8 cells had the capacity to generate angiogenic cytokines, whereas control fibroblasts did not. Thus C5 and C8 cells represent distinct subsets of fibroblasts that have the ability to support an angiogenic response to ischemia.

Cluster 5 cells may be more likely to directly participate in angiogenesis as they express early and mature endothelial genes involved in angiogenesis and cell migration, and they express the endothelial adhesion molecule CD144. When we isolated these cells from the ischemic hindlimb, they had the morphological appearance of fibroblasts, but were able to form networks in Matrigel, much as authentic ECs. These “E* cells” may emerge from selective proliferation of pre-existing Cluster 5 fibroblasts, and/or from a phenotypic switch of the other fibroblast subsets. Whether these E* cells incorporate into the vasculature to form functional vessels remains to be shown definitively, although this hypothesis is most consistent with our observations. At this time, we are not prepared to claim that transdifferentiation of fibroblasts to endothelial cells occurs in vivo. This hypothesis requires further study, which will require the development of new lineage tracing models specific for the subsets we identified by scRNA-seq.

Intriguingly, the scRNA-seq data reveal a marked increase in the diversity of transcripts expressed by YFP⁺ cells isolated from the ischemic limb, suggesting that there may be an increase in DNA accessibility with ischemia. An increase in DNA accessibility and an increased diversity of gene expression would be consistent with our prior work revealing the role of inflammatory signaling in DNA accessibility and nuclear reprogramming. We have previously described that cell autonomous innate immune signaling is required for nuclear reprogramming to pluripotency^{10, 19} and transdifferentiation of fibroblasts to ECs in vitro¹¹. Signaling through NFκB and IRF3 cause global changes in epigenetic modifiers⁴⁰. In addition, NFκB activates inducible nitric oxide synthase (iNOS), which in turn S-nitrosylates epigenetic elements, such as Ring1A of the polycomb repressive complex 1, causing the dissociation of this repressive complex from the chromatin¹². Finally, transdifferentiation in vitro requires a metabolic shift⁴¹. The effect of innate immune signaling to increase epigenetic plasticity and nuclear reprogramming is termed transflammation^{11, 40, 42}.

In the current paper, we find that Nfκb deficient fibroblasts have reduced capability to transdifferentiate in vitro. Furthermore, the NFκB antagonist Bay117082 nearly abrogated the emergence of E* cells and reduced angiogenesis in the matrigel plug assay. Dexamethasone, a glucocorticoid compound that blocks NFκB²², inhibited the emergence of YFP⁺CD144⁺ E* cells, impaired the recovery of perfusion, and exacerbated tissue loss, in the limb ischemia model. Finally, the fibroblast specific knockout of Rela also abrogated angiogenesis in the matrigel plug assay. These transgenic mice also exhibited an impaired recovery of perfusion, and exacerbated tissue loss, in the limb ischemia model. These studies are consistent with a role for innate immune signaling in the generation and/or expansion of subsets of fibroblasts that support angiogenesis in vivo, which appear to contribute to the physiological response to ischemia.

A greater understanding of the different subsets of fibroblasts could have therapeutic benefit. For example, methods to promote the expansion of Clusters 5 and 8 could enhance the restoration of perfusion in ischemic tissues. Fibroblasts respond to injury by proliferating, secreting inflammatory cytokines, generating extracellular matrix, and contributing to scar formation⁴⁶. In addition, our studies and others indicate that they may be a source of regenerative cells after injury. An understanding of the role of tissue fibroblasts in regeneration has been facilitated by lineage tracing studies using mice transgenic for Cre-recombinase being driven by fibroblast genes such as periostin²⁶, α2 (type I) collagen¹⁴, Tcf21³⁴, PDGFRα³⁵ and Fsp1^{14, 26}. Using the Fsp1-Cre lineage tracing technique, a quantitative FACS method and in vivo angiogenesis model, we have shown that a subset of fibroblasts contribute to the angiogenic response to ischemia. Fibroblast derived CD144⁺ cells account for about 5–20% of the total CD144⁺ pool (which is primarily composed of authentic endothelial cells) in a matrigel plug assay and account for up to 4–6% in a hindlimb ischemia model. Notably, fibroblast derived CD144⁺ cells appear to maintain their endothelial markers for a sustained period and persist at least as well as native ECs in a matrigel plug model.

In conclusion, our data indicate the existence of subsets of fibroblasts that contribute to the angiogenic response to ischemia. The generation and/or expansion of these angiogenic subsets of fibroblasts are dependent upon innate immune activation. Therapeutic modulation of these fibroblast subsets may represent a novel therapeutic strategy for enhancing vascularity and repair of ischemic tissues.

Clinical Perspective.

1) What is new?

• There are eight different subsets of fibroblasts in the murine limb

• Two of these subsets contribute to recovery of perfusion in limb ischemia

• The expansion of these angiogenic subsets during ischemia requires inflammatory signaling

2) What are the clinical implications?

• New technologies such as single cell RNA sequencing are helping us to discover many novel cell subtypes

• The recovery from ischemia may require the activity of cells that we had previously consigned to a fibrotic role

• Therapeutic modulation of these subsets of fibroblasts might enhance angiogenesis and reduce fibrosis in the setting of ischemia; and their inhibition might be useful in cancer.

Non-standard Abbreviations and Acronyms

EC

Endothelial cells

MEndoT

Mesenchymal-to-endothelial transition

MI

Myocardial infarction

Fsp1

Fibroblast-specific protein 1

VEGF

Vascular endothelial growth factor

PIC

Polyinosinic:polycytidylic acid

BMP4

Bone morphogenetic protein 4

bFGF

Basic fibroblast growth factor

scRNA-seq

Single cell RNA-sequencing

TTF

Tail tip fibroblast

TLR3

Toll-like receptor 3

AA

Angiogenic agents

HNDF

Human neonatal dermal fibroblast