Introduction

Fibroblasts are mesenchymal cells that form the extracellular matrix (ECM), which supports an organ-specific shape and provides the environment for parenchymal cells1,2. The ECM consists of collagen fibers, elastic fibers, and other fibrous components, as well as proteoglycans and glycosaminoglycans that fill the space among the fibers; it provides nutrients, stores cytokines and growth factors, forms their concentration gradients, and regulates cell behavior.

Fibroblasts can be defined as resting fibroblasts, activated/proliferating fibroblasts, and myofibroblasts, depending on conditions. Under physiological conditions, a few quiescent resting-type fibroblasts are present in abundant stroma. Once activated under the inflammation, the resting fibroblasts proliferate and differentiate into active contractile cells, termed myofibroblasts. Transforming growth factor-β (TGFβ) drives the differentiation of fibroblasts toward myofibroblasts and facilitates their biosynthesis of collagens and proteoglycans3,4. Myofibroblasts further differentiate into “mature” myofibroblasts, which synthesize collagens more actively. Then, collagen fibers are correctly aligned as myofibroblasts decrease. An imbalance between tissue degradation and repair causes increased volumes of the ECM, leading to fibrosis5. Therefore, fibroblasts are directly involved in these pathogenesis and disease progressions.

Fibroblasts are the key players in the development of organ fibrosis, the uncontrolled growth of collagen fibers, leading to organ dysfunction. Cardiac fibrosis after myocardial infarction impairs cardiac function significantly, pulmonary fibrosis reduces gas exchange efficiency, and liver fibrosis disturbs the natural function of the liver. Although the initial stages of fibrosis likely share the same process as tissue repair, persistent activation of fibroblasts by repeated or prolonged stimulation leads to fibrosis.

Fibrosis has organ specificity, defined by the heterogeneity of fibroblasts. For example, systemic sclerosis gradually develops in the skin, esophagus, lungs, and heart but seldom in the liver6. Keloid is a continuously growing local fibrotic mass, and such a mass is not observed in other organs7. This heterogeneity is known to be established during the embryonic period8,9,10.

Fibroblasts have various origins at different developmental stages2. Some fibroblasts are transformed via the epithelial-mesenchymal transition (EMT), in which epithelial cells lose their unique molecular markers, such as E-cadherin or zona occludens-1, and express fibroblast marker proteins such as fibroblast-specific proteins11. Other fibroblasts are of hematopoietic stem cell origin. The multipotent progenitor cells can differentiate into various vascular and mesodermal cells, such as pericytes, adventitial cells associated with the vasculature, and mesenchymal stem cells of the bone marrow. They can contribute to the adult interstitial fibroblast population2,12,13. The heterogeneity of fibroblast origin, evidenced by the diversity of transcriptome, signal transduction pathways, expression of the ECM molecules, cytokines, and growth factors, determines the stromal property of individual organs and characterizes diseases accompanied by stromal reactions14,15,16. Although fibroblast heterogeneity is believed to underlie disease characteristics, their ontogeny-based processes of heterogeneity formation remain unknown.

Currently, techniques for differentiating induced pluripotent stem cells (iPSCs) into various parenchymal cells, including cardiomyocytes, neurons, and hepatocytes, have been established, and protocols for differentiation into accompanying fibroblasts have been reported. Here, we prepared iPSC-derived fibroblasts in differentiation lineages of the heart, skin, and liver and analyzed their expression profiles using single-cell RNA-sequencing (scRNA-seq). Our results indicate characteristic subpopulations of fibroblasts of each organ type, although they share common features. We classified these subpopulations into active ACTA2+ and resting ACTA2−-type fibroblasts and identified TBX20 and TRPS1 as common factors in the ACTA2+ subpopulations. Our results provide information on organ-specific fibroblast subpopulations and a basis for specific fibrotic conditions in different organs.

Results

Differentiation of hiPSCs to fibroblasts

Normal human iPSCs were differentiated into fibroblast-like skin-(DFB), heart-(CFB), and liver-type (HFB) cells as models for organ-specific fibroblasts (Supplementary Fig. 1). The differentiated cells exhibited spindle-shaped, fibroblast-like morphology (Fig. 1a). Quantitative RT-PCR (qRT-PCR) analysis showed significantly decreased expression of stem cell marker genes, such as NANOG and OCT3/4, and increased expression of fibroblast marker genes, such as vimentin (VIM) and COL1A1 in differentiated cells compared with undifferentiated iPSCs (Fig. 1b). To investigate the heterogeneity of fibroblasts, we performed transcriptome analysis using these cells as samples for each organ-specific fibroblast model.

Fig. 1: Differentiation of fibroblasts from human iPSC.
figure 1

a Phase-contrast views of iPSC and differentiated fibroblasts. DFB dermal fibroblasts, CFB cardiac fibroblasts, HFB hepatic fibroblasts. Scale bar indicates 200 µm. b Expression levels of stem markers, NANOG and OCT3/4, and fibroblast markers, COL1A1 and VIM, in iPSC and differentiated fibroblasts. Data represent the mean ± SD of three independent experiments. One-way ANOVA followed by Dunnett’s multiple comparison test was used to compare the differentiated group with iPSC. **p < 0.01. c Heatmap analysis of integrated scRNA-seq data from differentiated fibroblasts. Selected fibroblasts and other lineage marker genes are shown. d Violin plots indicate the expression levels of COLA1A, VIM, and NANOG in differentiated fibroblasts.

Full size image

Overview of transcriptome analysis

Single-cell RNA-seq (scRNA-seq) data were obtained from three fibroblast models and then integrated using Seurat, resulting in a total of 15,804 cells (DFB, 5515 cells; CFB, 4490 cells; HFB, 5799 cells). First, we compared the expression of fibroblast marker genes among the three fibroblast models and found that most marker genes, such as COL1A1 and PDGFRA, were expressed in all types (Fig. 1c, d, and Supplementary Fig. 2). In contrast, the expression of other cell marker genes, such as ALB (hepatocytes), GLI1 (neurons), MYH11 (smooth muscle cells), MSLN (enterocytes), PECAM1 (endothelial cells), PF4 (megakaryocytes), and PTPRC (immune cells), was substantially low. The expression of RGS5 (pericytes/smooth muscle cells) was observed at a high level in CFB, and EPICAM (epithelial cells) was partially observed at low levels in the three fibroblast models. The expression patterns of fibroblast marker genes differed slightly among the models (Fig. 1c and Supplementary Fig. 2). The expression levels of BGN, POSTN, and TNC were markedly lower in DFB (Fig. 1c and Supplementary Fig. 2). ACTA2, a myofibroblast marker, was expressed in all models; however, its expression was higher in CFB (Supplementary Fig. 2). The gene expression levels were also confirmed by qRT-PCR (Supplementary Fig. 2b). These differential gene expression patterns demonstrate that the three fibroblast models had different characteristics.

To ensure data reliability, we compared scRNA-seq data from these iPSC-derived fibroblasts with data from three primary human fibroblasts in public databases, PCS-201-012 (adult skin fibroblasts, 6775 cells, GSM5104820)17, BJ (neonatal skin fibroblasts, 6024 cells, GSM6894025)18, and CCD-18Co (neonatal colon fibroblasts, 2343 cells, GSM6894025)19, as well as a dataset from fifteen human organs including heart, skin, and liver (total 84,363 cells, GSE159929)20 (Supplementary Fig. 3). Cluster analysis and identification of cell types by marker gene expression patterns revealed that the iPSC-derived fibroblasts were different from organ fibroblasts and other cell types, such as endothelial cells and smooth muscle cells (Supplementary Fig. 3a and Supplementary Data 1)20. The iPSC-derived fibroblasts and three primary fibroblasts had very similar cluster compositions, consisting mainly of FB-a (cluster 0), FB-c (cluster 9), and a mitotic cell population (cluster 7) (Supplementary Fig. 3b and Supplementary Data 2). Organ fibroblasts, on the other hand, were dominated by FB-b (cluster 1). However, FB-a and FB-c were detected as major populations in blood, lymph nodes, marrow, small intestine, and spleen, whereas minor populations in the bladder, common bile duct, esophagus, heart, muscle, rectum, skin, stomach and trachea. Taken together, the overall gene expression levels of cultured fibroblasts may be different from in vivo organ fibroblasts.

Identification of subpopulations

To investigate the heterogeneity of fibroblasts, we further analyzed scRNA-seq data from the iPSC-derived fibroblasts. Clustering of the integrated scRNA-seq data to identify fibroblast subpopulations resulted in 15 clusters (Fig. 2a). The proportion of each cluster differed among types (Fig. 2b and Supplementary Table 1). When we examined the cluster characteristics of each cell type, CFB had a high percentage of Clusters 4, 8, and 14, with 39.3% of CFB. In Cluster 14, DFB and HFB comprised less than 0.5% of the cells in each cluster, and CFB comprised almost all the cells in Cluster 14. HFB had a high percentage of Clusters 1, 2, and 7, accounting for 33.6% of HFB. Clusters 9 and 13 were relatively high in HFB (these five clusters together accounted for 50.1% of the total). DFB had a high percentage of Clusters 0, 3, 6, and 12, but since Cluster 3 was assumed to be proliferating cells, as described below, the characteristic clusters of DFB were Clusters 0, 6, and 12, which accounted for 32.5% of DFB. These results suggested that fibroblasts differentiated from iPSCs using different methods have their own major subpopulations. Except for Cluster 14, all clusters consisted of three types, although their proportions differed among the clusters.

Fig. 2: Analysis of scRNA-seq data of the differentiated fibroblasts.
figure 2

a The Uniform minifold approximation and projection (UMAP) plot analysis of absolute data and separate data of scRNA-seq. b Bar graph showing the proportion of clusters in each sample. The Numbers inside the bars indicate the cluster numbers. c Bar graph showing the percentage of cells in each cell cycle phase. d Trajectory analysis using the Monocle 3. e Pseudotime analysis using the Monocle 3.

Full size image

The expression patterns of the characteristic genes in each cluster are shown in a heatmap (Fig. 3a, Table 1, and Supplementary Data 3). Fibroblast marker genes such as COL1A1 were expressed in all clusters (Fig. 3b and Supplementary Fig. 4), confirming that all clusters were fibroblast populations. Regarding the other cell markers, the expression of RGS5 was higher in Cluster 14 than in the other clusters, whereas the expression of MKI67 was higher in Clusters 3, 10, and 11, indicating that these were proliferating cell populations (Fig. 3b and Supplementary Fig. 4). Fibroblast marker gene expression was relatively higher in Clusters 4 and 8. The expression patterns of several marker genes tended to be slightly different in each cluster, suggesting that they were subpopulations. For instance, LUM expression was higher in Clusters 2, 7, and 13 (Supplementary Fig. 4). The expression patterns of collagen genes were similar among the clusters, and the overall expression levels were high in Clusters 4, 8, and 13 (Supplementary Fig. 5). The expression levels of some collagen genes were slightly different among the clusters. For instance, COL14A1 was high in Clusters 7 and 13, and COL8A1 and COL9A3 were high in Clusters 8 and 14 (Supplementary Fig. 5).

Fig. 3: Features within each fibroblast subpopulations.
figure 3

a Heatmap showing the top five differentially expressed genes (DEGs) within each cluster. The complete data are presented in Supplementary Data 3. b Violin plots indicate the expression levels of COLA1A, ACTA2, RGS5, and MKI67 in the clusters. c Gene ontology enrichment analysis of biological processes in each cluster. The complete data are presented in Supplementary Data 4.

Full size image
Table 1 List of top ten differentially expressed genes (DEGs)
Full size table

Characteristics of subpopulations

Mitotic populations

Evaluation of the expression patterns of cell cycle markers demonstrated that Clusters 3, 10, and 11 were mitotic cell populations (Fig. 2c and Supplementary Table 1). Cluster 10 consisted exclusively of cells in the G2/M phase (99.9%) and had the highest expression of MKI67 (Supplementary Fig. 4)). Cluster 3 consisted of 80.1% cells in the G2/M phase and 7.8% cells in the S phase, while Cluster 11 consisted of 37.9% G2/M and 62% S phase cells. Cluster 6 had a relatively higher percentage of G2/M and S phase cells (7.6% and 8.7%). The other clusters were mainly composed of G1 phase cells (>92%). Gene ontology (GO) analysis also revealed that Clusters 3, 10, and 11 had characteristics of proliferating cells, such as DNA synthesis and nuclear division (Fig. 3c and Supplementary Data 4).

Heart-type group

The clustering results showed that Clusters 4, 8, and 14 were characteristic of CFB (these three clusters accounted for 39.3% of CFB, 12.3% of DFB, and 6.3% of HFB). Therefore, we categorized these clusters as a heart-type group. Analysis of differentially expressed genes (DEGs) showed that SERPINE2, INHBA, CCDC80, FSTL1, and SPARC were highly expressed in this group (Table 1). Cluster 4 was characterized by high expressions of COL1A2, NEAT1, LOX, FN1, and SPARC (Table 1 and Supplementary Data 3). Cluster 8 was by those of INHBA, RPSS35, CCDC80, SERPINE2, and FSTL1. Cluster 14 was by those of NPPB, MYL7, EDN1, ANKRD1, and TAGLN. Furthermore, Cluster 14 showed high RGS5 expression, as described above (Supplementary Fig. 4). Trajectory analysis revealed that Cluster 14 was a distinct cell lineage from Clusters 4 and 8 (Fig. 2d, e), suggesting that Cluster 14 is an RGS5+ subpopulation distinct from Clusters 4 and 8. GO analysis showed that Clusters 4, 8, and 14 were associated with ECM and cytoskeletal organization. In addition, Cluster 4 was associated with lipid and alcohol metabolism (Fig. 3c and Supplementary Data 4).

Skin-type group

The clustering results showed that Clusters 0, 6, and 12 were characteristic of DFB (these three clusters accounted for 32.5% of DFB, 22.4% of CFB, and 14.3% of HFB). We categorized these clusters as a skin-type group. DEG analysis revealed that KRT18, KRT8, KRT19, MT2A, and ANKRRD1 were highly expressed in this group (Table 1). Cluster 0 was characterized by high expressions of KRT8, SERPINE1, KRT18, MT2A, and MT1E. Cluster 6 was by those of KRT19, HMGA1, ACAT2, MT2A, and FDPS. Cluster 12 was by those of ANKRD1, CCDC146, MEST, OGFRL1, and TAGLN. GO analysis showed that Clusters 0 and 6 were associated with lipid and alcohol metabolism (Fig. 3c and Supplementary Data 4). Cluster 6 was with nucleotide synthesis, suggesting that this cluster was in the pre-mitotic phase. Cluster 12 was associated with ECM and cytoskeletal organization.

Liver-type group and Cluster 5

The clustering results showed that Clusters 1, 2, and 7 were characteristic of HFB (these three clusters accounted for 33.6% of HFB, 18.1% of DFB, and 15.7% of CFB). Cluster 9 was the fourth most cell-rich cluster of HFB (10.0%). Cluster 13, although with fewer cells, was characteristic of HFB (HFB: 6.4%, DFB: 3.5%, CFB: 2.5%). These five clusters accounted for half of the HFB, and we categorized them as a liver-type group.

The DEG analysis showed that CXCL14, IGFBP3, BAMBI, CTHRC1, and FTL were highly expressed in this group (Table 1). Cluster 1 was characterized by high expressions of CDKN1A, S100A6, FTL, OST4, and MYL9. Cluster 2 was by those of S100A6, TFPI2, LUM, MDK, and PCOLCE. Cluster 7 was by those of CXCL14, HOPX, IGFBP3, LUM, and H19. Cluster 9 was by those of MYLK, PTX3, THBS1, ACTA2, and CSRP2. Cluster 13 was by those of CXCL14, H19, COL1A2, DLK1, and COL3A1. When the top 10 DEGs were compared among these five clusters, Clusters 7 and 13 shared five genes (CXL14, IGFBP3, H19, APCDD1, and BAMBI), and Clusters 1 and 2 did three genes (S100A6, FTL, and BAMBI) (Table 1), indicating that there were three subgroups; Clusters 1 and 2, 7 and 13, and 9. GO analysis showed that Clusters 1 and 7 were functionally different from the other three clusters (Fig. 3c and Supplementary Data 4). Clusters 9 and 13 were associated with ECM and cytoskeleton organization, and Cluster 2 was with nucleotide synthesis.

Cluster 5 contained a similar ratio of CFB, DFB, and HFB and could not be categorized into any group. It was characterized by high expressions of GNG11, HNRNPA1, PTMA, PRAC1, and TFPI2. This cluster contained a slightly higher percentage of proliferating cells (4.8% S-phase and 3.2% G2/M phase), and the results of trajectory analysis suggested that this cluster was between the mitotic group and Cluster 7 (Fig. 2d, e).

Regulators of subpopulations

To predict the upstream factors that define individual organ-specific fibroblast models, we analyzed the transcription factor (TF) network using the SCENIC program and identified 248 differentially regulated regulons (Supplementary Fig. 6 and Supplementary Data 5). As shown in a heatmap of the Regulon Specific Score (RSS), each organ-specific fibroblast model had specific TFs under regulation21 (Supplementary Fig. 6a). TFs such as PRRX2, HOXB13, MSX2, HOXD10, and HOXA13 for CFB, ZFP64, MEIS1, IRF9, BATF3, and RELB for DFB, and IRX2, HOXB5, POLR3G, KLF9, and KLF2 for HFB were high-scoring regulators (Supplementary Fig. 6c, d). When the top 50 factors (top 20%) were compared, 12 factors were common between DFB and CFB, and one factor was common between DFB and HFB (Supplementary Fig. 6e). In contrast, no factors were common between CFB and HFB. The main regulatory factors for each model consist of different repertoires of factors.

Next, a TF network analysis was performed on the clusters (Supplementary Fig. 7 and Supplementary Data 6). In hierarchical clustering, three major classes were identified: Clusters 4, 8, 12, and 14; Clusters 1, 2, 7, 9, and 13; Clusters 0, 3, 5, 6, 10, and 11 (Fig. 4a). The first class was the heart-type group (Clusters 4, 8, and 14), except for skin-type Cluster 12. The second class consisted exclusively of the liver-type group. The third class was a mixture of the skin-type (Clusters 0 and 6) and mitotic groups.

Fig. 4: Transcription factor activities in subpopulations.
figure 4

a Heatmap showing normalized regulon specific score (RSS) within each cluster. The complete data are presented in Supplementary Data 6. b Venn diagrams showing the top 25 (top 10%) regulons in the heart-, skin-, and liver-type populations.

Full size image

Among the top 10% of regulons (25/248 factors), seven common TFs, such as HOXB13, SOX6, RARG, HOXD10, TBX20, TRPS1, and NR2C1, were detected in the heart-type group (Fig. 4b). Eight other factors were common to Clusters 4 and 8, suggesting that these two clusters represented a similar subpopulation. The skin-type group had no common factors with any of the top 10% of regulons. Clusters 0 and 6 had nine factors in common, including DDIT3, MYC, and CEBPG, while Clusters 0 and 12 or Clusters 6 and 12 had three or one factor in common, respectively. This suggests that Clusters 0 and 6 are similar. The liver-type group had no common factors. Clusters 1, 2, 9, and 13 had one common factor, HOXB2. Clusters 1, 7, 9, and 13 had two factors, ZBTB33 and HOXB5. Clusters 1 and 13 had 14/25 factors in common, and Clusters 7 and 13 had 11/25 factors in common, but Clusters 1 and 7 had 7/25 factors in common. Cluster 13 may be independently differentiated from Clusters 1 and 7.

Regulators for myofibroblastic phenotype

Myofibroblasts express high levels of α-smooth muscle actin (ACTA2). High ACTA2 populations (ACTA2+) were identified as Clusters 8 and 14 in the heart-type group, Cluster 12 in the skin-type group, and Cluster 9 in the liver-type group (Fig. 3b). All of these clusters exhibited biological functions such as ECM and cytoskeleton organization by GO analysis (Fig. 3c), suggesting that ACTA2+ cells were myofibroblastic-like. Interestingly, Cluster 9 was rarely observed in DFB, whereas Cluster 12 was rarely in HFB (Supplementary Table 1).

A survey of ACTA2+ regulators indicated that TBX20 and TRPS1 were common regulatory factors in the top 10% of regulators (Fig. 5a). These two factors were also found in the top 10% regulators of Cluster 4, whose ACTA2 expression was substantially higher but lower than Clusters 8 and 14 in the heart-type group (Fig. 3b). These two factors, therefore, must be associated with the myofibroblast-like properties. The expression level of TRPS1 was observed in the ACTA2+ group, whereas TBX20 was not (Fig. 5b). Factors common to any of the ACTA2+ were ranked low in the ACTA2− group, except for Cluster 4 (Fig. 5c). HOXB5, ZBTB33, TCF21, and SOX6 were found as factors common to both ACTA2+ and ACTA2− in liver-type and heart-type groups. Conversely, FOXP4, RARG, FOXQ1, and NFKB2 were as common in heart- and skin-type groups (Fig. 5c). These results suggest that skin-type Cluster 12 and liver-type Cluster 9 were of a distinct ACTA2+ subtype.

Fig. 5: Transcription factor activities in ACTA2+ subpopulations.
figure 5

a Venn diagrams showing the top 25 (top 10%) regulons in the ACTA2+ population. b Violin plots indicate the expression levels of TBX20 and TRPS1 in the subpopulations. c Heatmap showing normalized regulon specific score (RSS) within each cluster.

Full size image

Discussion

Understanding the functional differences in fibroblast subpopulations is important for developing therapeutic and preventive methods for diseases associated with fibroblast functions. Despite the research progress, we have not understood the basis of fibroblast heterogeneity among organs. To reveal organ differences in human fibroblasts, we generated three organ-type fibroblasts from iPSC and performed a comparative analysis of their properties. We have found that (1) iPSC-derived fibroblasts using different methods have characteristic organ-type subpopulations that vary in their ratios; (2) These organ-type subpopulations consist of myofibroblast-like ACTA2+ and resting-type fibroblasts; (3) ACTA2+ subpopulations have distinct characteristics; (4) TBX20 and TRPS1 are common factors in the ACTA2+ subpopulation. Our study provides the basis for ontogeny-based processes of fibroblast heterogeneity formation.

Our scRNA-seq analysis revealed 15 clusters from three different organ-specific fibroblast models (Supplementary Fig. 8). These clusters can be categorized into five groups; mitotic-type (Clusters 3, 10, and 11), heart-type (Clusters 4, 8, and 14), skin-type (Clusters 0, 6, and 12), liver-type (Clusters 1, 2, 7, 9, 13), and undetermined (Cluster 5). CFB is composed of 9.4% mitotic-type, 39.3% heart-type, 22.4% skin-type, and 24.2% liver-type. DFB is composed of 24.0% mitotic-type, 12.2% heart-type, 32.5% skin-type, and 23.5% liver-type. HFB is composed of 19.9% mitotic-type, 6.3% heart-type, 14.3% skin-type, and 50.0% liver-type. The percentage of the mitotic-type is likely depending on the experimental conditions. We observed more clusters comprising the liver-type than the other types. This is probably the reason for the high ratio of liver-type in three organ-specific fibroblast models.

The heart-type group contains two subgroups. One is ACTA2+ Clusters 8 and 14, and the other is Cluster 4. This cluster appears in transition to Cluster 8, as it exhibits ACTA2 and the ECM organizing functionalities. The expression patterns of fibroblast marker genes and DEGs are similar between Clusters 4 and 8. Cluster 14 is different from these two clusters and is defined by high expression of RGS5, known as a general pericyte marker gene. Although the origin and function of RGS5-expressing fibroblasts have not been determined, overexpression of RGS5 suppresses cardiac hypertrophy and fibrosis in mice and inhibits collagen synthesis in cardiac fibroblasts22. Therefore, Cluster 14 is unlikely to participate in cardiac fibrosis. Recent studies have shown that RGS5 is expressed in skin wound fibroblasts and cancer-associated fibroblast (CAF) subpopulations23,24,25. Cluster 14 may share their properties.

The skin-type group shows high expressions of KRT8, KRT18, and KRT19, known as epithelial markers. Cluster 6 appears to be a pre-mitotic type, and Cluster 12 is the active ACTA2+ type. Trajectory analysis indicates that Cluster 0 is located in between. In the process of wound healing, dermal fibroblasts proliferate and differentiate into myofibroblasts expressing ECM molecules. Cluster 0 is the most abundant cell population and is likely in transition to both proliferative and ECM synthetic states.

The liver-type contains three subgroups: Clusters 1 and 2, Clusters 7 and 13, and ACTA2+ Cluster 9. Clusters 1 and 2 show moderately high expression levels of ACTA2 but lower expression levels of ECM-organizing genes. In contrast, Cluster 7 and 13 show low expression levels of ACTA2 but slightly higher expression levels of ECM-organizing genes, especially in Cluster 13. The top DEG of Clusters 1 and 2 are S100A6, which is upregulated in myofibroblasts but not in resting hepatic stellate cells (HSCs) in mice26. S100A6 activates HSCs and promotes liver fibrosis in mice27. Overall expression levels of S100A6 in the liver-type are higher than in the other groups. Those of S100A6 are quite low in ACTA2+ Cluster 8. These observations suggest that Cluster 8 is unlikely to participate in liver fibrosis. Clusters 7 and 13 have the common top DEG, CXCL14. CXCL14-expressing fibroblasts are associated with prostate cancer and Crohn’s disease28,29.

ACTA2+ myofibroblastic cells are found in each organ-type group and are likely to have different characteristics. Heart-type Cluster 14 has the marked expression of RGS5, as mentioned above. In contrast, Heart-type Cluster 8 has high expression of INHBA (inhibin subunit beta A), which promotes EMT and fibrosis30,31. Cluster 8 also has a high expression of COL11A1. Recently, subpopulations with high expression of INHBA and COL11A1 have been identified in CAFs and may be associated with poor prognosis in colorectal cancer32. Liver-type Cluster 9 has high expression of PTX3 (pentraxin3), which inhibits inflammation and fibrosis33,34,35. Skin-type Cluster 12 has a high expression of ANKRD1, which facilitates wound healing in skin tissue36. Since PTX3 and ANK have opposite functions to each other, the two ACTA2+ clusters appear to have opposite properties.

TBX20 and TRPS1 are found as common regulators in the ACTA2+ group and Cluster 4. TBX20 is identified as the common regulator in the heart-type group, in agreement with the previous report that TBX20 is a regulator specific to the cardiac system37. The expression level of TRPS1 is higher than that of TBX20 in the ACTA2+ group, suggesting that TRPS1 is an important regulator for myofibroblastic characteristics. A recent multi-omics study has identified TRPS1 as a regulator of regeneration in skin wound healing rather than scar formation38,39. This suggests that our identified ACTA2+ group is the regeneration-associated fibroblasts rather than myofibroblasts.

The major role of fibroblasts is to form and maintain the ECM. While type I collagen is the major component of collagens, type III collagen is abundant in blood vessels, and type V collagen is often observed in the reticular fibers40. The present analysis has found no significant differences in gene expression ratios among the three groups. Fibrillins, components of elastic fibers, have a direct impact on the synthesis of elastic fibers. FBN1 expression was increased in the CFB, supporting the fact that the lesions of Marfan syndrome caused by FBN1 are the heart valves and aorta41,42. When the ECM is degraded and remodeled, transiently expressed VCAN promotes fibroblast proliferation and differentiation into myofibroblasts by enhancing TGFβ-signaling43. VCAN expression was the highest in the CFB, followed by the DFB and HFB. Our result suggests that the CFB contains more fibrosynthetic cells.

Accumulating studies have illustrated phenotypic diversity within an organ and a tissue, including lung44, heart45,46, skin47, and periodontal tissue48. Our results, demonstrating the differences in the expression of ECM molecules, various signaling molecules, and regulon patterns, may provide the basis for organ-specific characteristics in fibrosis, tumor invasion, and other ECM-related pathological conditions.

Materials and methods

Cells and culture conditions

The feeder-free human iPSC line, 1383D2, was provided by the RIKEN BRC through the National BioResource Project of the MEXT, Japan49. The cells were maintained on laminin-511 (iMatrix511 silk; Matrixome, Suita, Osaka, Japan) coated dish with StemFit AK02N (Ajinomoto, Tokyo, Japan) in a humidified atmosphere of 5% CO2 at 37 °C. For subculture, the cells were lifted by Accutase (Innovative Cell Technologies Inc, San Diego, CA) and seeded in the fresh AK02N with 10 µM Y-27632 (Selleck Biotech, Tokyo, Japan). On the next day, the medium was replaced with AK02N without Y-27632.

Differentiation of human iPSC-derived dermal fibroblasts

Human skin-type fibroblasts were differentiated by a protocol established by Hewitt et al. 50 and Kim et al. 51. The hiPSC were seeded onto iMatrix511-coated plate in AK02N medium with 10 µM Y-27632. Next days, the cells were cultured in fibroblast differentiation medium 1 (FDM1; 3:1 Dulbecco’s modified eagle’s medium (DMEM; FujiFilm Wako Pure Chemical Co., Osaka, Japan): F12 (FujiFilm Wako Pure Chemical Co.), 5% fetal bovine serum (FBS; Merck KGaA, Darmstadt, Germany), 0.18 mM adenine (FujiFilm Wako), 10 ng/ml human epidermal growth factor (EGF; FujiFilm Wako Pure Chemical Co.), and 5 µg/ml human insulin (FujiFilm Wako Pure Chemical Co.) for 4 d. Then the cells were cultured in FDM1 containing 0.5 nM human bone morphogenetic protein 4 (BMP4; Reprocell Japan, Yokohama, Japan) for 3 d. At day 7, the cells were passed onto Geltrex-coated plate (ThermoFisher Scientific K.K., Tokyo, Japan) and cultured in FDM2 (1:1 DMEM: F12, 5% FBS and 1% non-essential amino acids (ThermoFisher Scientific K.K.)) for 7 d. At day 14, the cells were passaged onto a non-coated culture plate in EDM1 for 7 d. At day 21, the cells were passaged onto type-I collagen (Col-I; Nitta gelatin Inc., Osaka, Japan) coated dishes in EDM1 for a few days. The differentiated DFB were cultured in FibroGRO medium (Merck KGaA) on Col-I-coated dish.

Differentiation of human iPSC-derived cardiac fibroblasts

Epicardiac fibroblasts as the CFB were differentiated by a protocol established by Bao et al.52. The hiPSC were seeded onto an iMatrix511-coated plate in AK02 medium with 10 µM Y-27632. The next day, the cells were treated with 6 µM CHIR99021 (Selleck Biotech) in RPMI/B27 (RPMI (FujiFilmWako Pure Chemical Co.) containing with B27 supplement minus insulin (ThermoFisher Scientific K.K.)) for 24 h and then treated with RPMI/B27 for 2 d. At day 3, the cells were treated with 5 µM IWP-2 (Nacalai Tesque Inc., Kyoto Japan) in 1:1 conditioned media to fresh RPMI/B27. After 2 d (day 5), the cells were treated with RPMI/B27. On day 6, cardiac progenitors were detached by Accutase and seeded onto a Geltrex-coated plate in LaSR medium (Advanced DMEM/F12 (ThermoFisher Scientific K.K.) containing 1.3% GlutaMax (ThermoFisher Scientific K.K.) and 0.1 mg/ml ascorbic acid). Then, the cells were treated with 3 µM CHIR99021 in LaSR medium for 2 d and were cultured in LaSR for 3 d. At day 12, the cells (epicardial cells) were detached by Accutase and seeded onto a Geltrex-coated plate in LaSR medium with 0.5 µM A83-01. To differentiate epicardiac fibroblasts, the cells were cultured in LaSR with 5 ng/ml human basic fibroblast growth factor (bFGF; Reprocell Japan)) and 0.5 µM A83-01 (Merck KGaA) for 10 d. The differentiated CFB were cultured in FibroGRO medium on a Col-I-coated dish.

Differentiation of human iPSC-derived hepatic fibroblasts

Hepatic stellate cells as the HFB were differentiated by a protocol established by Miyoshi et al. 53. The hiPSC were seeded onto an iMatrix511-coated plate in AK02 medium with 10 µM Y-27632. The next day, the cells were incubated with Advanced-DMEM/F12 containing 2% B27 supplement, 1% N2 supplement (FujiFilm Wako Pure Chemical Co.), 10 µM CHIR99021, 10 ng/ml bFGF, 30 ng/ml BMP-4, and 10 ng/ml human activin A (API Co., Ltd, Gifu, Japan) for 3 d. Next, the cells were incubated with Advanced-DMEM/F12 containing 2% B27 supplement, 1% N2 supplement, 100 ng/ml bFGF, and 50 ng/ml BMP-4 for 3 d. The differentiated HSCs were cultured in FibroGRO medium on a Col-I-coated dish.

Quantitative real-time PCR

Total RNAs were extracted by RNeasy plus mini kit (QIAGEN K.K, Tokyo Japan) and cDNAs were synthesized with ReverTraAce qPCR RT master mix with gDNA Remover (Toyobo Co. Ltd., Osaka, Japan). Quantitative real-time PCR (qRT-PCR) was performed on a QuantStudio 3 Real-time PCR system (ThermoFisher Scientific K.K.) using iTaq Universal SYBR Green Supermix (Bio-Rad Laboratories Inc., Hercules, CA). The PCR primer sets are represented in Supplementary Table 2. The reaction was performed for 40 cycles of 95 °C for 15 s and 60 °C for 60 s followed by 95 °C for 10 min. The expression level was analyzed by the comparative Ct method and normalized to that of glyceraldehyde-3-phosphate dehydrogenase (GAPDH).

Analysis of Single cell RNA sequencing data

Rhelixa, Inc. (Tokyo, Japan) performed the single cell RNA library preparation by 10× chromium, sequencing by Illumina NovaSeq 6000, and data analysis by Cell Ranger v6.1.2. Further analysis of the raw data was performed with Seurat package v4.4.054. For quality control, we first filtered the cells with a threshold of over 2500 “nFeature_RNA”, 5000 to 200,000 “nCount_RNA”, and less than 10 “percent.mt”. The data of total 5515 cells of DFB, 4490 cells of CFB, and 5799 cells of HFB were normalized using the Log Normalization method and integrated with 2000 variable integration features. After merging the Seurat Objects, the data was scaled with the ScaleData function and identified the principal component analysis (PCA, npcs = 30). A clustering and dimensionality reduction of UMAP was performed using the FindNeighbours, FindClusters, and RunUMAP functions, respectively (dims = 1:15 and resolution = 1.2). Cell cycle scoring was performed with the CellCycleScoring function. The differentially expressed genes (DEGs) of each cluster were identified using the FindAllMarkers function (min.pct = 0.25 and logfc.threshold = 0.25). The trajectory analysis was performed using Monocle 3 package 1.3.155. Gene ontology analysis was performed using clusterProfiler package 4.6.256. Single cell regulatory network inference and clustering (SCENIC) package 1.3.1 were used to infer gene regulatory networks and the key transcription factors of subpopulations57. Venn diagrams were drawn by using Venn package 1.11. Heatmaps were created with ComplexHeatmap package 2.14.0.

Statistics and reproducibility

Statistical analyses were performed using IBM SPSS statistics 28 (IBM Corp, Armonk, NY). One-way ANOVA followed by Dunnett’s or Tukey’s multiple comparison test was used, and p-values < 0.05 were considered significant.