All cells were isolated from three proliferative hemangiomas from three patients less than 6 months old. IH samples were placed in 50-mL sterile tubes containing phosphate-buffered saline immediately after excision and stored at 4°C. Cell suspensions were obtained by incubating minced specimens in endothelial cell growth medium 2 (EGM-2, PromoCell) supplemented with 2 mg/ml collagenase (Sigma Aldrich), 0.02 mg/ml DNase (Sigma Aldrich), 10% fetal bovine serum (Thermo Fisher Scientific) and 1% penicillin–streptomycin. HemSCs and HemECs were separated by magnetic activated cell sorting (MACS) using a CD133 MicroBead Kit (Miltenyi Biotec) and CD31 MicroBead Kit (Miltenyi Biotec), respectively, from a single-cell suspension as previously described (7, 20). Next, HemMCs were separated by CD146-positive selection from CD133^-CD31^- cells using a CD146 MicroBead Kit (Miltenyi Biotec). After separation, HemMCs were cultured in StemPro™ MSC serum-free medium (Thermo Fisher Scientific), and HemSCs were cultured in EGM-2 supplemented with Growth Medium 2 Supplement Mix (PromoCell) and 10% fetal bovine serum. For transcriptional analysis, HemMCs were transferred to complete EGM-2 for 3 days to minimize the influence of the medium. HUVECs were purchased from Promocell and cultured in EGM-2 supplemented with Growth Medium 2 Supplement Mix (PromoCell). All kinds of cells at passages 3–7 were used in the following experiments, including in vivo studies.

For mesenchymal differentiation, cells were cultured in adipogenic medium, osteogenic medium and chondrogenic medium (STEMCELL Technologies), and human adipose-derived stem cells (ADSCs), which were isolated and expanded following previously reported protocols (21), were used as a positive control in this study. Endothelial cell differentiation and neurogenic differentiation were induced according to a previous report (7).

To fluorescently label cells, the vector LV051-PHBLV-U6-MCS-EF1-mcherry-T2A-PURO containing the mCherry-encoding sequence was purchased from HanBio Technology. When the culture reached approximately 60% confluence, the HemMCs were infected with Lentivirus carrying the aforementioned vector (multiplicity of infection=10) for 24 hours. Next, the cells were cultured in the complete medium for 3 days. To obtain stable fluorescence-labeled cells, the cells were then screened using puromycin (at the concentration of 2.0 ug/mL for 2 days) and examined by fluorescence microscope (Nikon ECLIPSE microscope).

When cell confluence reached 80%-90%, the HemMCs were dissociated using TrypLE™ Express Enzyme (Gibco) and suspended in Cell Staining Buffer (Biolegend). In this study, each test contained 5×10⁵ cells in 100 µL staining buffer. To reduce nonspecific Fc receptor (FcR)-mediated binding, 5 µL FcR blocking solution (Biolegend) was applied per test (10 min incubation at room temperature). The fluorescence dye-conjugated primary antibodies (listed in Supplementary Table 2 ) were applied at recommended concentration indicated in the manufacturer’s instruction. After 30 min incubation at 4°C in the dark, the cells were washed using staining buffer (2 mL per test) for 2 times and then suspended in 200 µL staining buffer. The samples were immediately analyzed by the LSRFortessa flow cytometer (BD Biosciences) or CytoFLEX LX flow cytometer (Beckman Coulter) and FlowJo software (BD Biosciences).

Total RNA was isolated from HemMCs and HemSCs using TRIzol and the RNA Nano 6000 Assay Kit with the Bioanalyzer 2100 system. After cDNA library preparation, clustering of the index-coded samples was performed on a cBot Cluster Generation System using TruSeq PE Cluster Kit v3-cBot-HS (Illumina). The library preparations were sequenced on the Illumina NovaSeq platform, and 150 bp paired-end reads were generated (Novogene Experimental Department). Hisat2 v2.0.5 was used for mapping. Feature Counts v1.5.0-p3 was used to count the read numbers mapped to each gene. Then, the FPKM of each gene was calculated based on the length of the gene and read count mapped to this gene. Differential expression analysis between two conditions/groups (three biological replicates per condition) was performed using the DESeq2 R package (1.20.0). The resulting P values were adjusted using Benjamini and Hochberg’s approach for controlling the false discovery rate. For differential expression analysis of two groups, genes with an adjusted P value <0.05 according to DESeq2 were considered differentially expressed. The data generated in this study has been deposited in NCBI’s Gene Expression Omnibus and is accessible through GEO Series accession number GSE216867.

The murine model of IH was established according to an adapted protocol of Boscolo et al (10). We used female BALB/c nude mice in this study. A total of 1×10⁶ cells (per implant), which were HemMCs or HUVECs alone or the mixture of them at a 1:1 ratio, were suspended in 100 µL medium, mixed thoroughly with 100 µL Matrigel (Corning). Then, the cell/Matrigel mixture (200 µL per implant) was injected subcutaneously onto the back of mice aged 3 weeks (n = 6/group). At 1, 2, 4 and 8 weeks after the injection, the animals were sacrificed, and Matrigel plugs were harvested. The implants were fixed overnight in 4% paraformaldehyde at 4°C, and then paraffin-embedded and sectioned.

The sections of eighteen paraffin-embedded IH samples were obtained from the Pathology Department of Shanghai Ninth People’s Hospital. These samples were from the patients who were assessed for the necessity of operation (2) and collected via surgical treatment in the last ten years at Shanghai Ninth People’s Hospital. Most of the lesions were located in the head or face, and it was difficult to achieve a satisfactory outcome after drug treatment or natural involution. Detailed patient information is listed in Supplementary Table S1 . The IH sample or Matrigel xenograft explant sections were deparaffinized by xylene and then hydrated in a series of graded alcohol solutions. Antigen retrieval was performed, using the PH 9.0 antigen retrieval buffer (GeneTech), by heating in a water bath at 60°C for 20 min. The sections were permeabilized in 0.1% Triton X-100 (dissolved in phosphate buffered saline (PBS)) for 5 min after they cooled down to room temperature. Next, the sections were blocked in 5% donkey serum (dissolved in PBS) at room temperature for 1 h and then incubated overnight with primary antibodies (details in Supplementary Table 2 ) at 4°C. The antibodies were applied at recommended concentration indicated in the manufacturer’s instructions. Next, the sections were washed with PBS-Tween 20 (PBST) for 3 times. For immunohistochemistry, the detection of the antigens and counterstaining of the nuclei was performed using a REAL EnVision Detection Kit (DAKO) according to the manufacturer’s instructions. Images of the processed sections were taken with a Nikon ECLIPSE microscope. For Immunofluorescence, the sections were incubated with fluorescein-conjugated secondary antibodies (details in Supplementary Table 2 ) for 1 h at room temperature and then mounted in DAPI containing mounting medium (SouthernBiotech). Immunofluorescence (IF) images were taken with a Leica TCS SP2 Acousto-Optical Beam Splitter confocal system.Immunostaining with each antibody was carried out on at least 3 samples from different patients, and representative images are presented.

Statistical significance was determined using GraphPad Prism 6 (GraphPad Software, La Jolla, CA, USA). The data were analyzed by ANOVA or Student’s t test, and differences were deemed significant at P<0.05.

IH specimens were collected from the Department of Plastic and Reconstructive Surgery of Shanghai Ninth People’s Hospital, Shanghai Jiao Tong University of Medicine. A diagnosis of IH was confirmed by experienced clinicians and pathologists. Written informed consent for use of the specimens was obtained before surgery. All experiments were performed according to protocols reviewed and approved by the Ethics Committee of Shanghai Ninth People’s Hospital, Shanghai Jiao Tong University of Medicine.

CD146 expression was measured in specimens from patients with all stages of IH. CD146 expression was mainly located in perivascular cells surrounding CD31⁺ endothelial cells in hemangioma specimens of different phases ( Figure 1A ). In addition, CD146 is expressed in some of the CD31⁺ HemECs in proliferative IH ( Figure 1A ). Many CD146⁺ cells were costained with Ki67⁺ ( Figure 1B ). The number of CD146⁺Ki67⁺ cells was significantly higher in proliferating IH tissues than in involuting IH tissues ( Figure 1C ). The expression of CD146 mostly overlapped with that of PDGFRβ in proliferative hemangioma ( Supplementary Figure 1A ). However, PDGFRβ was also positively expressed in CD31^-CD146^- cells embracing CD146⁺ HemMCs, which were probably telocytes (19) or fibroblasts. PDGFRβ was not expressed in HemMCs in involuting hemangioma ( Supplementary Figure 1A ). Moreover, glucose transporter 1 (GLUT1)-positive HemECs were surrounded by CD146⁺ cells in both the proliferative and involuting phases ( Supplementary Figure 1B ). These results indicated that CD146 was more suitable as a marker to distinguish mural cells of IHs in different phases.

HemMCs and HemECs were isolated from proliferating hemangioma tissues through MACS and cultured. To assess whether both HemMCs and HemECs survive and proliferate in a xenograft model, a mixture of Matrigel and the two kinds of cells, i.e., HemMCs and HemECs, were injected into immunodeficient nude mice. At 1 week, both CD31⁺ cells and CD146⁺ cells were found in the implants by immunostaining ( Figure 2 ). At 2 weeks, the density of both became higher ( Figure 2 ). Staining of a proliferative IH section was used as a positive control.

To further analyze the phenotypes of HemMCs, flow cytometry was performed to assess the expression of stem cell surface markers. Similar to HemSCs, HemMCs expressed the mesenchymal stem cell surface markers CD73, CD90 and CD105 but did not express CD14, CD34 or CD45 ( Figure 3A ). After culture, the percentage of CD146⁺ cells in HemMCs was maintained at 87.9%, while the percentage of CD146⁺ cells in HemSCs was 8.48%, both counted by flow cytometry (FCM). These results indicated that CD146 was suitable as a marker to distinguish cultured HemMCs. Most HemMCs (97.5%) expressed CD140b (PDGFRβ), which was consistent with our previous results indicating that CD146⁺ cells were localized to the perivasculature ( Figure 3A ). Then, to determine the mesenchymal stem cell differentiation potential of these cells, we cultured HemMCs in conditioned medium and compared them with HemSCs. ADSCs, which are known to have the potential for adipogenesis, osteogenesis and chondrogenesis, were cultured in the same conditioned medium as the positive control. Similar to HemSCs and ADSCs, HemMCs displayed the potential to differentiate into adipocytes, osteocytes, and chondrocytes in vitro ( Figure 3B ). HemMCs expressed the neural markers glial fibrillar acidic protein (GFAP) and β-tubulin III after conditioned induction ( Figures 3C, D ). Regarding endothelial differentiation, unlike HemSCs, HemMCs failed to express the EC markers CD31 and vascular endothelial cadherin (VE-cadherin) in the same endothelial differentiation medium.

To further evaluate differentiation potential in vivo, HemMCs were mixed with Matrigel and injected subcutaneously into nude mice. Two weeks or four weeks after injection, the HemMC/Matrigel implants were harvested, sectioned, and stained. Similar to the involuting infantile hemangiomas shown, hematoxylin and eosin (H&E) staining revealed that a portion of the explants formed adipose tissue at 2 weeks. At 4 weeks, many of the explants had turned into fat tissue ( Figure 4A ). IHC staining for human leukocyte antigen-ABC (HLA-ABC) or human nuclei antigen (HNA) indicated that these adipocytes were of a human source ( Figure 4A ). Moreover, mCherry-labeled HemMC/Matrigel explants were stained for perilipin-A, and the results showed that these cells were perilipin-A⁺ adipocytes derived from human cells ( Figure 4B ).

To identify the molecular signature of CD146⁺ HemMCs and their different role in IH from HemSCs, we performed RNA sequencing (RNA-seq) and compared the transcriptome between in vitro cultured HemMCs and HemSCs. A total of 216 differentially expressed genes, including 131 upregulated genes and 85 downregulated genes, were found between HemMCs and HemSCs ( Figure 5A ). Three biological replicates from each group were subjected to clustered analysis. Gene Ontology (GO) enrichment analysis showed that the upregulated genes in HemMCs were involved in the extracellular matrix and angiogenesis ( Figure 5B ), which implied that HemMCs exhibited a pro-angiogenesis-associated transcriptome.

To confirm the proangiogenic function of HemMCs, HUVECs were seeded on Matrigel to induce tube formation, cultured independently or cocultured with HemMCs or HemSCs. The results showed that the tube length and number of branch points of HUVEC tubes were increased in the HemMC coculture group compared with the HUVEC alone group or HemSC coculture group ( Figures 5C, D ). In addition, significantly enriched GO terms, including growth factor activity (GO: 0008083) and enhancer binding (GO: 0035326), may partly account for the proangiogenic function of HemMCs ( Supplementary Figure 2A, B ).

To assess the proangiogenic function of HemMCs in vivo, a mixture of Matrigel and two kinds of cells, i.e., HemMCs and HUVECs, were injected into immunodeficient nude mice. At 2 weeks, a considerable number of newly formed small vessels with blood cells were observed in the implants, and HLA-ABC and HNA immunostaining confirmed that these vessels were of human origin ( Figure 6A ). Immunofluorescence staining indicated that the vessels consisted of VE-Cahderin⁺GLUT1⁺ endothelial cells surrounded by mCherry-labeled HemMCs ( Figure 6B ). At 4 weeks, the number of vessels was significantly decreased ( Figure 6C ), while the number of mCherry⁺Perilipin-A⁺ adipocytes in the implants was increased ( Figure 6D ).