2.1 Identification of the SOX2-positive human RSC subpopulation

To explore whether in situ RSCs are present in the adult human eye, scRNA-seq technology was used to perform transcriptome analysis of five areas of the adult human eye including pars plicata, pars plana, RPE, iris, and NR (Figure 1). The pars plicata sample with 7671 cells could be divided into 12 clusters (Figure 2A). Interestingly, cluster 4 in the pars plicata, which included 384 cells, specifically expressed the neural stem cell marker gene SOX2, and other clusters expressed the differentiated ciliary epithelium marker gene PALMD (Figure 2B,C). The pars plana sample with 7392 cells included 12 clusters, and all clusters expressed the differentiated ciliary epithelium marker gene PALMD (Figure 2D,E). However, there was no expression of SOX2 in the pars plana (Figure 2F). The tSNE plots of the RPE sample (7876 cells) and iris sample (841 cells) were divided into 14 clusters and four clusters, respectively (Figure 2G,J). The RPE marker gene RPE65 was specifically expressed in cluster 2 in the RPE sample, and the iris marker gene ITGB1 was specifically expressed in all clusters in the iris sample (Figure 2H,K). However, there was no expression of SOX2 in either sample (Figure 2I,L). The NR sample with 12,742 cells consisted of 12 clusters (Figure 2M). SOX2 and PALMD were both slightly expressed in cluster 4 (Figure 2N,O). We hypothesized that the NR tissue was mixed with a small number of pars plicata cells because of their adjacent anatomical relationship. In conclusion, we found that SOX2 was specifically expressed in cluster 4 in the pars plicata sample, which may represent the subpopulation of RSCs.

2.2 Functional analyses of the identified RSC cluster

To further establish the molecular differences among the 12 clusters of the pars plicata, we compared the gene expression in each cluster against all other clusters in the pars plicata. The top 5 differentially expressed gene (DEG) markers that distinguished each cluster from the others are shown in the heatmap (Figure 3A). Dynamic metabolic patterns are important characteristics of stem cells.^22-25 We found that most pathways of cluster 4 were related to the regulation of metabolism by KEGG analyses, suggesting the important characteristics of stem cells for cluster 4 (Figure 3B). GO analyses also indicated that the DEGs of the identified RSC cluster 4 were related to visual system development processes (Figure 3B). To further identify the functional roles of cluster 4, we examined the expression of an expanded set of genes related to cell properties in eye development (FOXE3,²⁶ TFAP2A,²⁷ PROX1,²⁸ and HES5²⁹), proliferation(CCND1,^{30, 31} STC2,³² ANXA1,³³ and ANXA2³⁴) and differentiation (ELF3,³⁵ ALDH1A1,³⁶ CRYAB,³⁷ and ENO1³⁸) (Figure 3C–E), and the results showed that these aforementioned genes were highly expressed in cluster 4 and further demonstrated the stem cell features of the identified RSC cluster.

2.3 AQP1 and TSPAN12 double-positive cells may represent RSCs

The sequencing data showed that the cluster 4 population contained 384 cells, of which SOX2-positive cells accounted for 76.04% (3.80% of total pars plicata cells). As a transcription factor, SOX2 localizes to the nucleus and has been shown to be necessary for the long-term self-renewal of neural stem cells.³⁹ For further stemness validation of SOX2-positive RSCs without destroying their cell structure, we screened SOX2-positive RSCs in cluster 4 by the specific surface markers AQP1 and TSPAN12 according to scRNA-seq data (Figure 4A–C). AQP1 and TSPAN12 were highly and especially expressed in cluster 4. AQP1 was previously reported to be upregulated in undifferentiated human embryonic stem cells (hESCs) and human induced pluripotent stem cells (hiPSCs), as well as hESC- and hiPSC-derived RPE cells.⁴⁰ TSPAN12 could accelerate mitotic progression to regulate proliferation, one of the stemness characteristics, by controlling the cell cycle.⁴¹ The specific surface markers AQP1 and TSPAN12 accounted for 70.31% of cluster 4 (3.52% of pars plicata cells), which is consistent with the percentage of the SOX2-positive subpopulation. Then, SOX2-positive RSCs were isolated by flow cytometry using the specific surface marker genes AQP1 and TSPAN12. The isolated cells accounted for approximately 1% of the whole pars plicata (Figure S1). The number of isolated AQP1 and TSPAN12 double-positive cells was in a reasonable range according to the sequencing results. The flow chart of the stemness verification experiments of double-positive cells after flow sorting is shown in Figure 4D.

2.4 SOX2-positive cells in the pars plicata have proliferation potential

Since proliferation is one of the characteristics of stemness, we tested the proliferation ability of isolated AQP1 and TSPAN12 double-positive cells. First, we sorted AQP1 and TSPAN12 double-negative cells in the pars plicata and performed SOX2 immunofluorescence staining as a negative control. We found that AQP1 and TSPAN12 double-negative cells did not express the neural stem cell marker SOX2 (Figure 5A). At the same time, we performed SOX2 in situ immunofluorescence staining on the eyeballs of the donors, and the experimental results showed that there were indeed scattered SOX2-positive cells in the pars plicata (Figure 5B). Then we conducted a suspension culture of the isolated AQP1 and TSPAN12 double-positive cells and double-negative cells for 7 days and found that the double-positive cells were positive for SOX2 and could be suspended into spheres, while the double-negative cells could not (Figure 5C,D), suggesting that AQP1⁺TSPAN12⁺ cells had proliferation potential. Asymmetric division is another characteristic of adult stem cells. We performed SOX2 immunofluorescence staining on the suspension spheres and determined the SOX2-positive rate. The results showed that the SOX2-positive rate in the suspension spheres ranged from 37% to 45% (Table 1), which is consistent with the characteristic asymmetric division of adult stem cells. Then, we carried out immunofluorescence staining of Nestin (a retinal progenitor marker), β-Tubulin III (an early neuron marker), and Pax2 (a mature glial marker) in suspension spheres. We observed that the isolated AQP1⁺TSPAN12⁺ cells were positive for Nestin and β-Tubulin III but negative for Pax2 (Figure 5E). These results showed that double-positive cells had a tendency to gradually differentiate into neurons from progenitor cells rather than incorporate them into the isolated retinas. We carried out a monolayer subculture of the identified RSCs after flow sorting. The results showed that the double-positive cells could be subcultured and proliferate, as shown in Figure 5F, suggesting that the double-positive cells had proliferation potential.

2.5 SOX2-positive pars plicata cells have differentiation potential

To further confirm the stemness of SOX2-positive RSCs from pars plicata of human eyes, we tested the cell differentiation potential. SOX2-positive RSCs from pars plicata were cultured in the same differentiation media for the differentiation experiments according to previous studies,^{9, 10, 42-44} and the differentiated cells were stained with immunofluorescence for different cell types of the RPE and the NR. After culturing for 3 weeks, we observed that the isolated cells were positive for markers of various retinal cells with different morphologies, including RPE65 (a marker of the RPE, Figure 6A), Nrl/Rho (markers of photoreceptors, Figure 6B,C), IBA1 (a marker of microglia, Figure 6D), GFAP (a marker of astrocytes, Figure 6E), calbindin (a marker of horizontal cells, Figure 6F), NeuN (a marker of neurons, Figure 6G), PKC-α (a marker of bipolar cells, Figure 6H) and NF-M (a marker of ganglion cells, Figure 6I). The differentiation ratio of SOX2-positive RSCs after being cultured in differentiation media for three weeks was determined, and the results showed that the majority of the differentiated cells were photoreceptor cells (as indicated by the presence of Nrl and rhodopsin staining) and a minority of the differentiated cells were RPE cells (as indicated by the presence of RPE65 staining) (Table 2 and Figure S2). These results indicate the differentiation abilities of the isolated SOX2-positive RSCs.

4.1 Samples and preparation of single-cell suspensions

Eyeballs from six donors were divided into five tissue regions within 32 h including the iris, pars plicata, pars plana, RPE, and neuroepithelial layer (Donors 1–6, Table S1). The samples were finely minced with scalpels and dissociated in a buffer containing papain (20 U/ml, LS003119; Worthington) supplemented with DNase-I (2000 U/ml, 10104159001; Roche) in Earle's Balanced Salt Solution at 37°C for 30 min. The cell suspensions were resuspended in Earle's Balanced Salt Solution and passed through a 30-μm cell strainer. Then, the viability of the samples was evaluated with AO/PI prior to loading on the single-cell platform. The samples displayed 90% viability on average.

4.2 Single-cell sequencing

Single-cell cDNA libraries of sample pools were generated using the 10X Genomics platform following the manufacturer's instructions (Document Number CG000206). Briefly, single-cell suspensions were loaded into 10X Chromium according to the kit instructions, followed by cDNA amplification and library construction according to standard procedures. Qubit was used for library quantification before pooling. The final library pool was sequenced on an Illumina Novaseq6000 instrument using 150 base-pair paired-end reads.

4.3 scRNA-seq data analysis

The Cellranger (version 3.1.0) pipeline with the default and recommended parameters was used for preprocessing and sequence alignment of data reads.⁶⁵ The human reference genome used for the alignment was GRCh38. Next, Gene-Barcode matrices were generated for each individual sample by counting unique molecular identifiers (UMIs) and filtering noncell-associated barcodes. Finally, the gene-barcode matrix was generated by the pipeline, and it contained the barcoded cells and gene expression counts.

After Cellranger processing, the count matrix, barcode information, and gene information of each sample were imported into R software (version 3.6.1), and each sample was analyzed separately according to the standard process of Seurat (version 3.1.5).⁶⁶ First, we performed preliminary filtering of the imported data, and the filtered data (features detected in at least three cells, cells in which at least 200 features were detected) were used for subsequent analysis. Next, to avoid the existence of low-quality cells interfering with subsequent results, we removed outlier cells. Doublet-cell prediction was performed on these cells by the R package DoubletFinder (version 2.0.3),⁶⁷ and doublet cells that could affect the clustering were removed. Therefore, we obtained 7671 cells in the plicata sample, 7392 cells in the plana, 7876 cells in the RPE, 12742 cells in the NR, and 841 cells in the iris. Second, the cell cycle of each sample was analyzed, and it did not affect the grouping of the samples. After the above analyses, we normalized the data by setting the parameter scale factor = 10,000, and the normalization method was LogNormalize. We also obtained the highly variable features by setting the parameter nfeatures = 2000 with the method “vst”. Similarly, we scaled and centered features in the dataset with the function ‘ScaleData’, and we regressed out confounding factors: number of UMIs, number of features, percentage of mitochondrial RNA, and percentage of hemoglobin RNA. Third, we performed a principal component analysis of the processed dataset to achieve dimensionality reduction, and the selection of these PCs was based on elbow and jackstraw plots. Clusters were determined by a clustering algorithm based on shared nearest neighbor (SNN). Additionally, the resolution parameter for clustering was 0.3. Finally, we chose t-distributed stochastic neighbor embedding (t-SNE) to visualize our clustering results. To obtain the differentially expressed genes of each cluster, we conducted a differential analysis for each cluster with the function ‘Find Markers’ and the parameter ‘Wilcox’.

4.4 Pars plicata section immunostaining

First, all pars plicata were isolated from the eyeballs of donors 7–12 (Table S1). A part of the isolated pars plicata was cut by the cross-section method and cryosectioned. The remaining pars plicata were dissociated and used for fluorescence-activated cell sorting (FACS) and subsequent experiments. The pars plicata cryosections were subjected to immunofluorescence labeling with the primary antibody mouse monoclonal SOX2 (ab79351, Abcam; 1:200 dilution) at 4°C overnight. Then, the sections were washed carefully and incubated with a Cy3-labeled goat anti-mouse IgG (H+L) secondary antibody (Beyotime, A0521) for 1 h. Images were captured by confocal microscopy (LSM 800; Zeiss, Germany).

4.5 Fluorescence-activated cell sorting

The pars plicata was dissociated from the eyeballs of donors 7–12 as previously described and passed through a 30 mm filter. The filtered cell suspensions were resuspended in FACS medium (PBS supplemented with 0.1% BSA), and the number of cells was counted. The cells were transferred into FACS tubes, aliquots of 20,000 cells were removed as unstained and single-stained controls, and the remainder of the cell suspension was incubated with dialyzed mouse anti-human TSPAN12-Alexa Flour 647 (FAB8910R; R&D Systems, dilution 1:20) and mouse anti-human AQP1-FITC (orb15121; Biorbyt, dilution 1:50). The cells were incubated on ice in the dark for 30 min. Cells were washed twice with FACS medium (spin at 400 × g for 5 min). Unstained and single-stained controls were used for proper gating. FACS was then performed on a flow cytometer (S3e; Bio-Rad, USA). The TSPAN12⁺AQP1⁺ cells were collected in FACS tubes containing 3 ml precooled DMEM.

4.6 Stem cell cultures

The stem cell cultures were prepared from donors 7–12 (Table S1). After FACS, the isolated cells were counted and resuspended in a serum-free medium containing 10 ng/ml fibroblast growth factor 2 (F, F3685; Sigma‒Aldrich), 2 μg/ml heparin (H, B9806; Sigma‒Aldrich) and 20 ng/ml EGF (E, SRP3027, Sigma‒Aldrich). The number of cells was counted after the indicated number of days in culture. To test the clonality of the isolated cells, the cells were plated at a density of one cell per well in individual 96-well plates in the presence of E+F+H. After 7 days in culture, the wells were inspected for the presence or absence of spheres.

4.7 Differentiation and immunostaining

To test the differentiation abilities of the isolated cells from donors 7–12 (Table S1), individual clonally derived spheres were plated in 24-well plates on a substrate of Matrigel (Corning) in the presence of 10% fetal bovine serum, 10 ng/ml fibroblast growth factor, 2 μg/ml heparin and 20 ng/ml EGF. The medium was changed every 3–4 days. Three weeks later, the plates were fixed with 4% paraformaldehyde for 20 min and permeabilized with 0.3% Triton X-100 for 10 min. Then, the sections were blocked with goat serum albumin for 60 min at 37°C. After washing with PBS three times, the sections were incubated with primary antibodies at 4°C overnight. Then, the retinas were washed carefully and incubated with secondary antibodies for 1 h. The following primary antibodies were used: mouse monoclonal SOX2 (ab79351; Abcam, 1:200 dilution), rabbit monoclonal Nestin (ab105389; Abcam, 1:200 dilution), mouse monoclonal β-Tubulin III (ab78078; Abcam, 1:200 dilution), mouse monoclonal Pax2 (H00005076-M01; NOVUS, 1:500 dilution), rabbit monoclonal GFAP (80788; Cell Signaling Technology, 1:200 dilution), rabbit monoclonal IBA1 (019-19741; WAKO, 1:100 dilution), rabbit monoclonal rhodopsin (27182; Cell Signaling Technology, 1:200 dilution), mouse monoclonal Nrl (sc-374277; Santa Cruz Biotechnology, 1:50 dilution), mouse monoclonal RPE65 (sc-390787; Santa Cruz Biotechnology, 1:50 dilution), mouse monoclonal Calbindin (ab82812; Abcam, 1:200 dilution), mouse monoclonal NF-M (sc-32273; Santa Cruz Biotechnology, 1:50 dilution), mouse monoclonal PKC-alpha (ab32376; Abcam, 1:200 dilution) and mouse monoclonal NeuN (94403; Cell Signaling Technology, 1:500 dilution). The following secondary antibodies were used: Alexa Fluor 488-labeled goat anti-rabbit IgG (H+L) (Beyotime, A0423), Alexa Fluor 488-labeled goat anti-mouse IgG (H+L) (Beyotime, A0428), Cy3-labeled goat anti-mouse IgG (H+L) (Beyotime, A0521), and Cy3-labeled goat anti-rabbit IgG (H+L) (Beyotime, A0516).

4.8 Statistical analysis

All data are presented as the mean ± SD and were analyzed with SPSS 20.0 software (IBM, Chicago, IL, USA). Data are presented as the means ± SDs. p-Values were calculated using Student's t-test. Differences were considered statistically significant at p < 0.05.