2.1 Tissue processing

Eleven donated lenses of 10 donors, of which D2 provided two lenses, without a history of ocular disease were obtained from the Eye Bank Center of Wenzhou Medical University (Table 1). Donated eyes were obtained post-mortem within <6 h from death and dissected within 1 h of enucleation. Four lenses were used for scRNA-seq and the additional six lenses were used for immunofluorescence and tissue clearance (Table 1).

For samples D1–3, the entire lens capsule (less the posterior 4 mm capsulorhexis) containing all lens epithelia and early fibre cells was used. A circular capsulorhexis of the central 4 mm of the posterior capsule was performed, and the remaining lens capsule with superficial lens cells was sheared radically from the posterior margin to the equator using surgical scissors. After removing fibre cells from the nuclear portion of the lens, the fibre cells in the cortical portion of the lens were carefully removed. D4 was treated similarly to D1–D3, except the most central anterior cells (6 mm capsulorhexis) were removed. We collected the capsule without the central 6 mm anterior zone by a circular central anterior capsulorhexis, and the remaining capsule with superficial lens cells was collected by the same procedure described above.

This study was approved by the Laboratory Human Ethics Committee of Wenzhou Medical University (2020-132-K-117-02) and was carried out in compliance with the Declaration of Helsinki.

2.2 scRNA-seq

Lens superficial cells were digested in 600 μL digestion solution consisting of 400 μL pancreatin (0.25%), 100 μL collagenase A (2 mg/mL) and 100 μL dispase II (10 mg/mL) for 8 min at 37°C and gently pipetted 20 times to assist cell dissociation. Then, 600 μL DMEM containing 10% FBS was added to the digestion solution for neutralization. The capsule was removed after neutralization, and the sample was passed through a 40-μm filter. The obtained single-cell suspension was centrifuged at 200g for 5 min, the supernatant was discarded, and the cells were resuspended in 1150 μL phosphate-buffered saline (PBS). The single-cell suspension (300–600 live cells/μL determined by Count Star) was immediately used for scRNA-seq with 10x Genomics.

2.3 Bioinformatics analysis of scRNA-seq data

Reads were aligned and unique molecular identifier counts were obtained using CellRanger pipeline (10x Genomics). For further processing, integration and downstream analysis, R (version 4.1.1) and Seurat (version 4.1.0) were used.¹² Cells with gene numbers of <500 and >7500, and potential stress signals of >20% mitochondrial reads were excluded. The sequencing depth was normalized for each cell in the dataset, and Seurat-based canonical correlation analysis was performed on D1–3 sample datasets to remove inter-sample batch effects. The dataset was subsequently subjected to linear dimensionality reduction by principal component analysis (PCA). PCA dimensions were evaluated and selected by the results of elbow plots. Data were clustered in R using a Louvain graph algorithm. The cluster resolution was selected by the results of the clustree (version 0.4.4) program.¹³ Cells were projected onto two-dimensional coordinates using the umap algorithm. Marker genes for lens epithelium and fibre cells were used to identify cell types. Epithelial clusters were extracted for reclustering, and differentially expressed genes between clusters were identified using the FindAllMarkers function. Gene ontology (GO) functional enrichment analysis was performed using the clusterProfiler package (version 4.2.2). Integrated analysis of intact capsule samples (D1–3) with ring retrieved capsule samples (D4) was performed using Seurat (version 4.1.0). To infer the differentiation trajectory and genes whose expression was significantly associated with the pseudotime, we used the destiny package (version 3.8.1).

2.4 Immunofluorescence

Three donated lenses were immediately fixed in a fixation solution (40% formaldehyde:absolute ethanol:double distilled water:glacial acetic acid = 1:4:4:1) for 48 h at 4°C, embedded in paraffin, and sectioned. Before staining, the tissue sections were dewaxed in xylene, rehydrated using a graded series of ethanol solutions and washed in distilled water. Then, the sections were placed in 100× citrate antigen retrieval solution in a 98°C water bath for 20 min. After cooling slowly to room temperature, the sections were washed with PBS three times, each time for 5 min, and blocked in 5% sheep serum and 1% bovine serum albumin in PBS for 1 h, and then incubated with primary antibodies against human CD24 (1:200, Cat. # ab202073, Abcam, Cambridge, MA, USA), ADAMTSL4 (1:100, Cat. # A4785, ABclonal, Wuhan, Hubei Province, China), and C8orf4 (1:100, Cat. # ab229680, Abcam) at 4°C overnight. An Alexa Fluor 488-conjugated secondary antibody (1:800, Cat. # ab150077, Abcam) was applied and DAPI (Cat. # 62248, Thermo Fisher Scientific, Waltham, MA, USA) was used for counterstaining. Images were obtained under a Zeiss LSM710 confocal microscope.

2.5 iDISCO tissue clearing, 3D imaging and Imaris analysis

The iDISCO process was based on the protocol of Renier et al.¹⁴ A fresh lens was fixed with paraformaldehyde for 48 h before the clearing process. Then, a gradient of methanol (20%, 40%, 60%, 80% and 100% methanol/PBS for 1 h each at room temperature) for dehydration and incubation in 66% dichloromethane/33% methanol with shaking were applied. Because of a colour change caused by fixation in paraformaldehyde, the lens was bleached with H₂O₂ (5% H₂O₂/95% methanol, shaking at 4°C overnight), which was followed by rehydration in gradient methanol (80%, 60%, 40% and 20% methanol/PBS for 1 h each at room temperature). Immunostaining employed traditional immunofluorescence methods, including permeabilization and blocking, primary and secondary antibody incubation (CD24, Cat. # EPR19925, Abcam; donkey anti-rabbit secondary antibody, Alexa Fluor™ 594, Cat. # A-21207, Thermo Fisher Scientific; DAPI). Then, dehydration in methanol and incubation in dichloromethane were repeated as described above. Dibenzyl ether was used to calibrate reflection index for approximately 48 h.

A Lightsheet microscope (Nuohai Life Science Co., Ltd) was used for 3D imaging with Imaris software (Imaris 9.7, Oxford Instruments PLC, UK) for quantitative analysis. Surface algorithms were used for remodelling and thickness calculation.

3.1 Different cell types exist in the superficial tissue of lens

Lens epithelial and superficial fibre cells from four donor lenses of various ages were used for single-cell RNA sequencing. After filtering low-quality cells, the transcriptome profiles of 21,711 cells were further analysed (D1: 10373 cells; D2: 5135 cells; D3: 6203 cells) (Figure 1A).

Data of 21,711 single cells were embedded in a uniform manifold approximation and projection (UMAP). Using unbiased low-resolution clustering, they were divided into three major cell clusters using established lens markers and annotated as epithelial and fibre cells (Figure 1C–E).¹⁵ The unknown cluster was relatively isolated from the other two clusters in the UMAP plot (Figure 1C). After integration to remove inter-sample batch effects, the different samples did not overlap in this cluster (Figure S1A). Moreover, the expression of lens markers was relatively weak compared with the other two clusters (Figure 1D). Additionally, the gene expression profiles were similar to that of the lens epithelial cluster (Data S1). Therefore, this cluster may have contained different subpopulations reflecting the heterogeneity of different ages, considering that the biological replicates in this study were insufficient for age difference analysis and the unknown cluster was not analysed further.

3.2 Identification of two subpopulations of LECs and one LFC cluster with novel specific markers by scRNA-seq

Data of 18,596 sequenced cells belonging to fibre and epithelial cell clusters were extracted for reclustering analysis (D1: 9970 cells; D2: 3686 cells; D3: 4940 cells). As shown in Figure 2A, we identified three clusters including one LFC cluster and two LEC clusters, in which the proportion of cells in the three samples was similar (Figure S1B). Expression of LEC markers was not identical in the two epithelial cell clusters. C8orf4 and ADAMTSL4 genes were specifically expressed in the two epithelial cell clusters, respectively (Figure 2B,C,E), and CD24 was specifically expressed in the LFC clusters (Figure 2B,C,E). The expression similarities suggested that C8orf4⁺ and ADAMTSL4⁺ cells were LECs, but differed in expression of specific markers. After performing differential analysis of gene expression among the three clusters (Figure 2D), we found that the three clusters were three cell types with different gene expression profiles.

3.3 Different cell types have different localization in lens superficial tissue

Differential gene expression analysis of the three clusters showed that C8orf4 and ADAMTSL4 were expressed at high levels in the two subpopulations that expressed classical LEC markers, whereas CD24 was highly expressed in the cluster expressing classical LFC markers.

To confirm our findings, marker expression in the three clusters was assessed by immunofluorescence (Figure 3A). C8orf4 was mainly expressed in the anterior central region of the lens epithelium, and ADAMTSL4 was mainly expressed in the equator and pre-equatorial region of the lens epithelium (Figure 3A). Interestingly, CD24 was expressed only in superficial fibre cells directly adjacent to the lens epithelium, including the lens anterior and posterior regions (Figures 3A,F and 4).

Functional enrichment analysis based on differentially expressed genes showed that C8orf4⁺ cells were related to the response to nutritional and oxygen levels (Figure 3B), while ADAMTSL4⁺ cells were related to eye development (Figure 3C). Moreover, CD24⁺ cells showed downregulation of genes related to epithelial cell proliferation and upregulation of genes related to eye development, including some LFC markers (Figure 3D).

Considering the different localization of the two clusters of LECs, the equator region of the capsule with superficial lens cells from one donor lens was used for scRNA-seq (D4: 4651 cells). The size of the central zone of the lens epithelium lacks a unified definition and standard, and specific dimensions may vary on the basis of factors such as researchers, measurement methods and tools employed. In general, the diameter of the central zone of the lens epithelium is within the range of 5 mm or less.^{16, 17} We believe that the capsule without the central 6 mm anterior zone removed all the central lens epithelium. Then, the data were compared with those from whole capsule samples. The difference in the expression profile of LEC markers as well as C8orf4 and ADAMTSL4 between whole capsule samples and the equator capsule sample reconfirmed the distinct localization of the LEC subpopulations (Figure 3E).

Lens transparency confirmed the location of the CD24⁺ fibre cell cluster. A 3D view and axial section of the lens also showed an obvious difference between 6 months (6M) and 17 and 49 years (17Y and 49Y, respectively) (Figure 4B,C). CD24⁺ fibre cells appeared to be two rings in 6M, whereas only one ring was observed in 17Y and 49Y lens. From the anterior capsule to the equator and posterior capsule, the thickness of nuclei (DAPI⁺) and CD24, together with their spatial localization was recorded. CD24 was expressed beneath the capsular epithelium (Figure 4C–E). Detail imaging of CD24 in the 6M lens is shown in Figure S2A, which obviously appeared as two rings. The outer ring gradually thinned from the front to the back, while the inner ring gradually thickened just over the equator (Figure S2B). CD24 appeared as a single ring in 17Y and 49Y lenses similarly to the outer ring in the 6M lens (Figure S2C). The unique double ring of CD24 in the 6M lens indicated a role of CD24 in lens development. In particular, the unique double ring staining in the 6-month lens required further sample confirmation.

3.4 Characteristics of the differentiation of human lens superficial cells

LECs differentiate into LFCs throughout life. However, the relationship between cell differentiation of the two epithelial cell subtypes and superficial fibre cells found in this study was unclear. We focused on these cells to investigate the differentiation status of lens superficial tissue. The differentiation states and transitions between them were computationally reconstructed using diffusion maps.¹⁸ This method embeds the data in a low-dimensional space where distances between cells represent progression through a gradual but stochastic process such as differentiation. In the diffusion maps, the data showed a structure with gradual transitions between two different clusters and cells originating from a common origin (Figure 5A,B).

We found expression of progenitor cell markers ALDH1A3 and HES5,¹⁹ and ADAMTSL4⁺ cells gradually decreased as cells progressed away from their common origin (Figure 5C), which were derived from ADAMTSL4⁺ cells. We further noted that the left arm of the differentiation trajectory terminated at CD24⁺ cells and showed increasing expression of MIP, CRYBA1 and CD24 (Figure 5C), which was consistent with the superficial fibre cell phenotype. On the right arm of the differentiation trajectory, cells derived from ADAMTSL4⁺ cell transitioned to C8orf4⁺ cells, during which expression of GJA1, IGFBP5 and C8orf4 was increased, suggesting that this branch represented differentiation towards central epithelial cells (Figure 5C). In this branch, the expression levels of PAX6 and SOX2, which are considered to be markers of lens progenitor cells,²⁰ remained relatively high (Figure 5C).

The diffusion map recapitulated superficial cells during the lens differentiation process, and therefore we computationally inferred with the two branches and ordered the cells in accordance with their progression through pseudotime (Figure 6A). This allowed us to identify genes with expression changes during the differentiation process. We found 71 genes that showed pseudotime-dependent expression with the same directionality along both differentiation trajectories (Data S2). These included genes associated with known progenitor characteristics, such as HES4, ID3,^{21, 22} and FOXE3, which are associated with LEC differentiation,^{23, 24} and genes with no previously reported associated with lens differentiation, such as TINAGL1, PBX1 and SLC16A1 (Figure 5B). Additionally, we identified 905 genes with branch-specific expression patterns. We then clustered the gene expression trend on each of the two branches and identified sets of genes that changed over pseudotime (Figure 6E,F, Data S3). On the superficial fibre cell branch, we found two clusters of genes with increased or decreased expression during differentiation. The increased cluster was enriched for genes involved in eye development and the LFC differentiation process (e.g., PROX1, BFSP2 and TDRD7, Figure 6E, Figure S3A). The decreased cluster was enriched for genes involved in energy metabolism (e.g., COX7A2L, PHGDH and DGUOK, Figure 6E, Figure S3B). Genes involved in RNA splicing showed a transient phase of upregulation during the superficial fibre cell differentiation process (e.g., SRRM1, SFPQ and SRSF11, Figure 6E, Figure S2C). During the differentiation process towards central epithelial cells, we identified two broad clusters of genes that gradually increased or decreased their expression. Genes with increasing expression were involved in the response to nutrient and oxygen levels and the epithelial cell proliferation process (e.g., CAMK2N1, GLUL and IGFBP5, Figure 6F, Figure S3D), while genes with decreasing expression were involved in negative regulation of cell differentiation and stem cell population maintenance (e.g., HES5, CDK5RAP2 and DIXDC1, Figure 6F, Figure S3E).

Considering that the diffusion map suggested a common origin subpopulation of the two differentiation trajectories for ADAMTSL4⁺ cells, we extracted ADAMTSL4⁺ cells for reclustering (Figure 7A) and performed differential analysis of gene expression at a resolution of 0.2 (Figure 7B). We found that cluster 5 (104, 0.06%) showed a distinct stemness signature as indicated by expression of the cell proliferation marker MKI67, a gene commonly used for stem cell identification, and high expression of STMN1 relative to other clusters. Previous studies have shown a population of high STMN1-expressing MIKI67⁺ cells in the isthmus that were characteristic of actively cycling stem cells.²⁵ Additionally, PTTG1,^{26, 27} CENPF,²⁸ TOP2A,^{29, 30} and BIRC5³¹ have been effectively studied as biomarkers for various types of stem cells (Figure 7C). Next, we performed GO functional enrichment analysis of the significantly differentially expressed genes in cluster 5, which showed that cluster 5 cells were closely related to the cell division process (Figure 6D). Taken together, cluster 5 may be lens epithelial stem/progenitor cells and exhibit the characteristics of STMN1^highMKI67⁺PTTG1⁺CENPF⁺TOP2A⁺BIRC5⁺.