Human hair follicle organ dissection

All male participants with AGA signed informed consent and were notified about the study before participation. The study was approved by Huashan Hospital of Fudan University (ethical approval number 2019M-008). The clinical information of the study participants is shown in Table S1 (see Supporting Information). Samples from patients 1–10 were used for RNA sequencing and those from patients 1–20 were used for the follow-up functional studies. Anagen HFs in the vertex and occipital areas from male patients with AGA undergoing hair transplants were collected by follicular unit extraction, and haematoxylin and eosin staining was performed to confirm the hair cycle stage (Figure S1; see Supporting Information). Anagen follicles were used in this study and were used in RNA sequencing analysis and ex vivo HF studies.

RNA sequencing and bioinformatic analysis

The parameters, reagents and methods for sequencing and bioinformatic analysis can be found in Appendix S1 (see Supporting Information).^20-28

Human hair follicle culture and treatment

The isolated human HFs were cultured in Williams E medium (Gibco BRL, Grand Island, NY, USA). HFs that grew to lengths of 0·3–0·5 mm and macroscopically remained in anagen VI after 24 h were selected for further experiments.²⁹ For transfection experiments, HFs were transfected with 0·04 nmol L⁻¹ EGLN1, EGLN3, SFRP2 or SERPINF1 (formerly PEDF) small interfering (si)RNA mixed with 2 μmol L⁻¹ Lipofectamine RNAiMAX transfection reagent (Thermo Fisher Scientific, Waltham, MA, USA). The transfection effect was evaluated at 2 days post-transfection by immunofluorescence. For recombinant protein stimulation experiments, HFs were incubated with 100 ng mL⁻¹ recombinant pigment epithelium-derived factor (PEDF) protein (HY-P7054; MedChemExpress, Monmouth Junction, NJ, USA) or 1 μ$\mathrm{g}{\mathrm{mL}}^{-}$¹ recombinant secreted Frizzled-related protein 2 (SFRP2) protein (6838-FR-025; R&D Systems Inc., Minneapolis, MN, USA). The length of the hair shaft and the hair cycle were measured every 2 days.

Immunofluorescence staining

HF samples were embedded in paraffin. Next, histological sections were incubated overnight at 4 °C with primary antibodies for either EGLN1 (1 : 300, sc-271835; Santa Cruz Biotechnology, Santa Cruz, CA, USA), EGLN3 (1 : 250, ab184714; Abcam, Cambridge, UK), PEDF (1 : 500, ab180711; Abcam) or SFPR2 (1 : 500, ab92667; Abcam). The sections were then incubated with the corresponding secondary antibody for 1 h at room temperature. After removing the secondary antibody, the nucleus was counterstained with 4-amino-6-diamino-2-phenylindole (DAPI; Beyotime Biotech, Nanjing, China). The stained sections were observed under a microscope (Olympus, Tokyo, Japan), and representative images were captured.

Real-time polymerase chain reaction

DPCs were isolated from HF bulbs and cultured in Dulbecco’s modified Eagle’s medium as previously reported.³⁰ Total RNA was extracted and reverse transcribed from HFs or DPCs using TRIzol (Invitrogen, Carlsbad, CA, USA) and the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Foster City, CA, USA). Real-time polymerase chain reaction (PCR) was performed with SYBR Premix Ex Taq (Takara Bio, Shiga, Japan) and analysed with an ABI Prism 7900 Detector System (Applied Biosystems). The housekeeping gene GAPDH was used as the endogenous control.

Real-time cellular analysis assay

Real-time cellular analysis (Agilent, Santa Clara, CA, USA) was applied to monitor cell viability according to the manufacturer’s instructions. An xCELLigence system (Roche Applied Science, Penzberg, Germany) was used to record the cell growth curves automatically. Cell index was used to observe cell state (e.g. cell attachment, cell number). Cells at a density of 9000 cells per well were seeded in a 16-well microplate. After incubation for 6 h, the medium was refreshed with new treatment medium.

Western blot

Cell lysates were subjected to sodium dodecyl sulfate polyacrylamide gel electrophoresis and proteins were transferred onto polyvinylidene fluoride (PVDF) membranes. According to the standard protocol, the PVDF membranes were blocked in Tris-buffered saline with Tween (TBST, pH 7·4) and 5% bovine serum albumin for 2 h and incubated with their respective primary antibodies for 8 h at 4 °C, followed by the secondary antibody for 2 h at room temperature. After washing with TBST three times, the proteins were visualized using the ECL chemiluminescence system (Santa Cruz Biotechnology). ImageQuant TL software (Cytiva, Marlborough, MA, USA) was used to quantify the intensity of bands.

Deferoxamine treatment

Deferoxamine was purchased from Sigma-Aldrich (St Louis, MO, USA). The working concentration for treatment of DPCs was 300 ng mL⁻¹. RNA and protein were extracted from DPCs after 24-h and 48-h incubation periods with or without deferoxamine, respectively.

Statistical analysis

All data were analysed using R (v3.6.0; R Foundation, Vienna, Austria). The paired-sample t-test and unpaired-sample t-test were used for paired samples and unpaired samples, respectively. P-values < 0·05 following false discovery rate adjustment for multiple testing correction were considered to be statistically significant.

Whole-transcriptome signatures of hair follicles from individuals with androgenetic alopecia

To explore the pathogenesis of AGA, we performed RNA sequencing of protein-coding genes, lncRNAs and miRNAs in paired HFs of the frontal vertex and occipital scalp from 10 male patients with AGA (Figure S1). The whole-genome transcriptomic profiles of these two groups were pooled together and presented as a circle plot based on their P-values and the chromosome location of each gene (Figure 1a). Differential expression analysis identified 506 mRNAs, 127 lncRNAs and 55 miRNAs that were differentially expressed. Hierarchical clustering indicated that most mRNAs and miRNAs were upregulated in the vertex group compared with the occipital group (Figure 1b, c), whereas lncRNAs were mainly downregulated (Figure S2a; see Supporting Information). For the differentially expressed mRNAs, 308 genes were upregulated, and 198 downregulated genes were identified in the vertex group (Figure 1d). In miRNAs, we identified 35 upregulated miRNAs and 20 downregulated miRNAs (Figure 1e). We also detected 53 upregulated lncRNAs and 74 downregulated lncRNAs (Figure S2b).

These results showed significantly differential expression patterns between vertex and occipital HFs in AGA. The details of the differentially expressed genes, miRNAs and lncRNAs are listed in Tables S2–4 (see Supporting Information). In addition, we identified the correlation pairs of differentially expressed RNAs through correlation analysis and constructed regulatory networks using these correlation pairs. The lncRNA–mRNA coexpression networks and miRNA–mRNA and miRNA–lncRNA regulatory networks were constructed based on their coefficients using Cytoscape²⁸ (Figure S3a–c; see Supporting Information). Moreover, a competitive endogenous RNA network was constructed to identify the key regulators using miRNA–mRNA and miRNA–lncRNA pairs (Figure S3d). The results of the coexpression analysis are presented in Tables S5–7 (see Supporting Information).

Pathway enrichment analysis: hypoxia-inducible factor-1 and Wnt signalling pathways

We further performed KEGG pathway enrichment analysis to identify the altered signalling pathways in AGA. We found that the upregulated mRNAs in AGA vertex HFs were significantly enriched in the HIF-1 signalling pathway (adjusted P = 9·8 × 10⁻⁷). The downregulated mRNAs were mainly enriched in the Wnt signalling pathway (adjusted P = 7·9 × 10⁻⁵) and Hippo signalling pathway (adjusted P = 8·1 × 10⁻⁵) (Figure 2a). The expression of key HIF-1 signalling genes (EGLN1, EGLN3, SERPINE1, HMOX1, TIMP1, and IL6R) and Wnt signalling-related genes (SERPINF1, SFRP2, DKK1, LGR4, LGR5, WNT3A, WNT5A and WNT16) is shown in Figure 2(b–d). Furthermore, Gene Ontology enrichment analysis revealed that adaptive immune response, moulting cycle, hair cycle, extracellular matrix, cytokine receptor activity, DNA-binding transcription activator activity, and hormone activity pathways were dysfunctional in balding HFs (Tables S8–10; see Supporting Information).

We also performed miRNA pathway enrichment analysis using miRPathDB.^{20, 23} We found that the HIF-1 and Wnt signalling pathways (data not shown) were significantly enriched with differentially expressed miRNAs using KEGG and WIKI pathway analysis, respectively (Figure 2e). In particular, the target genes of miR-451a, miR-98-5p and miR-196b-5p were significantly enriched in the HIF-1 signalling pathway (Figure 2f). In addition, the focal adhesion, cell cycle, DNA damage response and transforming growth factor-β signalling pathways were highly involved. These miRNA-enriched pathways were coordinated with mRNA-enriched pathways, which emphasizes the importance of the pathways that we identified. Therefore, the HIF-1 and Wnt signalling pathways were selected for further functional studies.

Expression validation of differentially expressed genes involved in the hypoxia-inducible factor-1 and Wnt signalling pathways

To validate the expression levels of differentially expressed genes, a total of 20 AGA samples, including 10 independent pairs of HFs, and the initial 10 pairs of HFs for RNA sequencing were collected (Table S1). Haematoxylin and eosin staining showed that the obtained HFs were anagen HFs, and revealed a concomitant decrease in the maximum bulb diameter of vertex HFs compared with occipital HFs (Figure 3a). Real-time PCR was performed to assess the expression of HIF-1 pathway-related genes (EGLN1, EGLN3 and HMOX1) and Wnt signalling pathway-related genes (SERPINF1, SFRP2 and LGR5). The results showed that the mRNA levels of EGLN1, EGLN3, SERPINF1 and SFRP2 were significantly higher in vertex HFs compared with occipital HFs (Figure 3b; and Figure S4a; see Supporting Information).

Moreover, immunofluorescence staining showed increased mean (SD) protein expression of EGLN1 [1·04 (0·24) vs. 1·55 (0·35)], EGLN3 [1·02 (0·06) vs. 2·65 (0·38)], SERPINF1 [1·15 (0·25) vs. 1·75 (0·18)] and SFRP2 [1·16 (0·21) vs. 2·54 (0·53)] in vertex HFs (Figure 3c). These proteins were ubiquitously expressed in almost every cell type, and EGLN1 was mainly expressed in hair matrix cells (Figure 3c). Collectively, we discovered that HIF-1 and Wnt signalling pathway-related genes (EGLN1, EGLN3, SERPINF1 and SFRP2) were upregulated in AGA vertex HFs and might play important roles in AGA.

The effects of hypoxia-inducible factor-1 pathway-related EGLN1 and EGLN3 on dermal papilla cells and hair follicles

DPC proliferation activity, HF growth and the hair cycle play important roles in the process of HF miniaturization and AGA. To assess the effects of HIF-1 pathway-related genes (EGLN1 and EGLN3) in AGA, we treated DPCs and cultured HFs with EGLN1 or EGLN3 siRNA. Real-time PCR analysis confirmed the decreased expression of EGLN1 or EGLN3 after corresponding siRNA transfection into cultured DPCs (Figure 4a). Cell proliferation activity was examined by the xCELLigence system as reported,³¹ and we found that downregulation of EGLN1 or EGLN3 resulted in a significant increase in DPC proliferation (Figure 4b). Furthermore, the decreased expression of EGLN1 and EGLN3 in treated HFs was confirmed by immunofluorescence (Figure 4c). As shown in Figure 4d, EGLN1 or EGLN3 knockdown increased the length of hair shafts compared with the negative control. Moreover, a greater percentage of anagen HFs and a lower percentage of catagen HFs remained in the EGLN1 or EGLN3 siRNA-treated group compared with the negative control group after 6 days of culture (Figure 4e). The results revealed that EGLN1 or EGLN3 downregulation promoted DPC proliferation and HF growth, prolonged the anagen stage, and postponed HF catagen transition, suggesting the important role of EGLN1 and EGLN3 in AGA.

The effects of Wnt pathway-related SFRP2 and SERPINF1 on dermal papilla cells and hair follicles

The Wnt signalling pathway plays a key role in AGA, but the effect of SFRP2 or SERPINF1 has not been studied in AGA. Here, we show that SFRP2 or SERPINF1 siRNA treatment reduced the transcriptomic levels of targeted genes (Figure 5a) and significantly promoted DPC proliferation (Figure 5b). Consistently, addition of recombinant SFRP2 or SERPINF1 protein inhibited DPC proliferation (Figure 5c). In HFs, SFRP2 or SERPINF1 siRNA transfection decreased target protein expression as revealed by immunofluorescence (Figure 5d). In addition, SERPINF1 downregulation promoted HF growth, while the SERPINF1 recombinant protein significantly inhibited HF growth (Figure 5e). However, SFRP2 siRNA or recombinant protein treatment showed no effect on HF growth (Figure 5f). Compared with the negative control group, SERPINF1 downregulation resulted in a greater percentage of anagen HFs remaining and a lower percentage of catagen HFs, and SERPINF1 recombinant protein facilitated catagen transition and shortened the anagen stage (Figure 5g, h). Although SFRP2 is an inhibitor of the Wnt pathway, SFRP2 did not significantly affect the HF hair cycle (Figure 5g, i). These findings suggest that as SERPINF1 is an inhibitor of the Wnt pathway, its upregulation plays a role in HF pathological changes and the development of AGA.