Global Transcriptional and Epigenetic Reconfiguration during Chemical Reprogramming of Human Retinal Pigment Epithelial Cells into Photoreceptor-like Cells

Retinal degenerative diseases are frequently caused by the loss of retinal neural cells such as photoreceptors. Cell replacement is regarded as one of the most promising therapies. Multiple types of stem and somatic cells have been tested for photoreceptor conversion. However, current induction efficiencies are still low and the molecular mechanisms underlying reprogramming remain to be clarified. In this work, by combining treatment with small molecules, we directly reprogrammed human fetal retinal pigment epithelial (RPE) cells into chemically induced photoreceptor-like cells (CiPCs) in vitro. Bulk and single-cell RNA sequencing, as well as methylation sequencing, were performed to understand the transcriptional and epigenetic changes during CiPCs conversion. A multi-omics analysis showed that the direct reprogramming process partly resembled events of early retina development. We also found that the efficiency of CiPCs conversion from RPE is much better than that from human dermal fibroblasts (HDF). The small molecules effectively induced RPE cells into CiPCs via suppression of the epithelial-to-mesenchymal transition (EMT). Among the signaling pathways involved in CiPCs conversion, glutamate receptor activation is prominent. In summary, RPE cells can be efficiently reprogrammed into photoreceptor-like cells through defined pharmacological modulations, providing a useful cell source for photoreceptor generation in cell replacement therapy for retinal degenerative diseases.


Introduction
Photoreceptors are responsible for converting light into electrical signals for further processing and integration, which is essential for the visual system. Retinal degenerative diseases, such as age-related macular degeneration (AMD) and retinitis pigmentosa, are typically characterized by the degeneration of retinal pigment epithelial (RPE) cells and photoreceptors, causing irreversible vision loss [1][2][3]. Cell therapy has been extensively investigated to repair or replace damaged RPE cells or photoreceptors [4][5][6]. In particular, cell replacement therapy based on directed differentiation of embryonic stem cells (ESCs) [7] and induced pluripotent stem cells (iPSCs) [8] into photoreceptor cells holds tremendous promise for the treatment of retinal diseases [9]. However, the potential risk of generating progenitor cell tumors presents a challenge for their use in clinical therapy [10].

Treatment with ASO
To test if PTBP1 ASO could improve conversion efficiency, we used the mouse PTBP1 ASO synthesis and transfection protocols previously published [22]. PTBP1 ASO was obtained by Integrated DNA Technologies (Northridge, CA, USA). The sequence of the target region in human PTBP1 for ASO synthesis is 5 -GGGTGAAGATCCTGTTCAATA-3 . The backbone of the ASO contains phosphorothioate modification. Fluorescein was attached to the 3 end of the ASO.
Cells were seeded on D0 and cultured in the growth medium to reach confluency at around 70-80% 24 h later. On D1, in addition to a medium change of PIM with 5F, 75 pmol of PTBP1 ASO was transfected with Lipofectamine RNAiMAX (Thermo Fisher Scientific, Waltham, MA, USA). Forty-eight hours later, the medium was replaced by fresh PIM containing 5F, then the following process was the same as that of treatment with 5F alone after D3.

Time-Lapse Imaging
For live-cell time-lapse imaging, synchronized RPE cells were plated on glass-bottom dishes coated with 0.1% gelatin. The next day, the culture medium was replaced with PIM plus 5F, then image recording was conducted on the first day of 5F treatment in a temperature-and CO 2 -controlled chamber. At least 6 positions per well were acquired every 30 min with a Zeiss Axio Observer Z1 inverted fluorescence microscope.

RT-qPCR
Total RNA was extracted using the PureLink™ RNA Mini Kit (Thermo Fisher Scientific, Waltham, MA, USA), and 1 µg RNA was reverse transcribed to cDNA with a High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Waltham, MA, USA). The total RNA concentration was determined by the NanoDrop Spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA). Real-time qPCR was carried out with PowerUp™ SYBR™ Green Master Mix (Applied Biosystems, Waltham, MA, USA) on a CFX96 real-time PCR detection system (Bio-Rad, Hercules, CA, USA). Fold changes were calculated using the 2 −∆∆Ct method. In order to improve the data accuracy, we adopted GAPDH and ACTB as reference genes and normalized the data by taking the geometric mean. Primers are listed in Supplementary Table S4.

Trypan Blue Exclusion Assay
Cell viability was estimated using the trypan blue exclusion assay. Viable cells have intact membranes to exclude trypan blue, while dead cells lose membrane integrity and can thus be stained by this dye. Cells with 5F/5FA induction at different time points were dissociated by incubation with 0.25% trypsin-EDTA (Sigma-Aldrich, St. Louis, MO, USA) at 37 • C for 5 min, centrifuged and then resuspended in the culture medium. A trypan blue solution was added to the cell suspensions in a ratio of 1:1. Total cells and dead cells (stained in blue) were counted using a hemocytometer. The percentage of living cells was calculated.

Western Blots
RPE cells, reprogramming intermediates, and CiPCs were harvested and dissociated in a lysis buffer. After protein extraction, samples were loaded onto 12% gels and separated via SDS-PAGE (sodium dodecyl sulfate polyacrylamide gel electrophoresis). Then the proteins were transferred to polyvinylidene difluoride (PVDF) membranes (Bio-Rad, Hercules, CA, USA), blocked in 5% skim milk (Sigma-Aldrich, St. Louis, MO, USA) at room temperature for 1 h, and incubated with primary antibodies at 4 • C overnight. The membrane was transferred into secondary antibodies the next day and incubated for 1 h at room temperature. The specific signals were detected with ECL Western Blotting Substrate (Thermo Fisher Scientific, Waltham, MA, USA) and visualized with a chemiluminescence detection system (Bio-Rad, Hercules, CA, USA). Signal intensity was analyzed with ImageJ (v2.1.0). The antibodies used are provided in Supplementary Table S4.

Establishment of Mouse Model and Cell Transplantation
All animal work was conducted in accordance with the guidelines of the University of California, Los Angeles Animal Research Committee (ARC). 8-10-week-old C57B6L/J mice were maintained under a 12 h light/12 h dark illumination cycle with normal food and water. Anesthesia was induced via intraperitoneal injection of ketamine (87.5 mg/kg) and xylazine (12.5 mg/kg) before each surgical procedure. To establish retinal degeneration, we employed the widely used sodium iodate (NaIO 3 ) mouse model [23]. Unlike the genetic retinal degeneration models such as rd1 mouse with RPE unaffected [24], NaIO 3 leads to RPE and subsequent photoreceptor degeneration, which better recapitulates the advanced state in dry AMD patients. The procedures for NaIO 3 model establishment and cell transplantation were adjusted from published articles [17,25,26]. Briefly, 35 mg/kg of sterile 1% NaIO 3 (Sigma-Aldrich, St. Louis, MO, USA) in saline was administered through a tail vein injection. One month after NaIO 3 treatment, 5F/5FA treated RPE cells at D10 were individualized by incubation at 37 • C for 5 min with 0.25% trypsin-EDTA and resuspended into single-cell suspension with balanced salt solution (BSS; Alcon, Fort Worth, TX, USA). The cells were counted on a hemocytometer to reach a dilution of 1 × 10 5 cells/µL and kept on ice until subretinal transplantation. Briefly, 1% tropicamide ophthalmic solution was used to dilate the pupil. A tiny bleb was visible below the retina on the superior nasal location by injecting 1.5 µL of 1 × 10 5 /µL cell suspension or BSS through the sclera approximately 1 mm behind the limbus under a surgical microscope using a 33-gauge blunt-end microliter syringe (Hamilton, Reno, NV, USA) as previously described [27]. The contralateral eye received the same treatment. The animals were monitored for revival from anesthesia and ocular inflammation.
Selected mouse eye frozen sections were washed three times in PBS before being blocked in 5% BSA with 0.5% Triton-X 100 for two hours at room temperature. Then, the process was followed by incubation with primary antibodies at 4 • C overnight. After three subsequent PBS washes, secondary antibodies were incubated for 1 h at room temperature. Sections were mounted with DAPI Fluoromount-G (Southern Biotech, Birmingham, AL, USA). Images were taken on an Eclipse 80i Nikon microscope and analyzed using ImageJ (v2.1.0). Primary and secondary antibodies used in this article are provided in Supplementary Table S4. 2.10. RNA Sequencing (RNA-seq) and Data Processing Total RNA was extracted using a PureLink™ RNA Mini Kit (Thermo Fisher Scientific, Waltham, MA, USA). Then mRNAs were used for library construction by using the NEBNext ® Ultra™ II RNA Library Prep Kit (New England Biolabs, Ipswich, MA, USA) following the manufacturer's recommendations. Libraries were quantified using the 2200 TapeStation (Agilent, Santa Clara, CA, USA) and sequenced using the Illumina HiSeq 3000 or Novaseq S2.
All the sequenced raw reads were filtered and trimmed with Trim Galore (v0.4.1) before subsequent analysis. For bulk RNA-seq data processing, the clean reads were mapped to the human genome (GRC38) using Hisat2 [28]. Then, the number of reads uniquely mapped to each gene was determined using FeatureCounts from subread2 [29]. The differentially expressed genes (DEGs) were identified by DESeq2 [30] with p values less than 0.01. Principal component analysis (PCA) was performed using all the genes after removing the batch effect by DESeq2.

Enzymatic Methylation Sequencing (EM-seq) and Data Processing
Genomic DNA of different cells was extracted using DNeasy Blood and Tissue Kits (Qiagen, Valencia, CA, USA). Genomic DNA was spiked with 0.5% lambda DNA and fragmented to 250-350 bp using Bioruptor (Diagenode, Denville, NJ, USA). EM-seq libraries were prepared from 300 ng of fragmented DNA using the NEBNext Enzymatic Methyl-seq Kit (New England Biolabs, MA, USA).
For EM-seq data processing, reads were filtered and trimmed with Trim Galore (v0.4.1) followed by mapping to the human genome (GRC38) using Bismark (v0.17.0) [34]. Mapped reads were further deduplicated and filtered for non-conversion. Estimation of methylation levels was determined in the CpG context with Bismark.

10× scRNA-Seq and Data Processing
For scRNA-seq, the preparation of single cells was performed according to protocols from 10× Genomics. Cells were trypsinized into single cell suspensions and resuspended in an appropriate buffer with viability of over 90%. Then cell suspensions were introduced into 10× Chromium for single-cell 3 transcriptome profiling.
The resulting FASTQ files were processed using CellRanger v3, using GRC38 as the reference genome. The read count matrix generated by CellRanger was then analyzed using Seurat v3 [35]. Cells were further filtered by the number of genes detected (with at least 500 genes but no more than 6000 detected) and the percentage of reads mapped to the mitochondrial genome out of total reads (less than 10%). The top 2000 genes were identified by variable feature selection based on a variance stabilizing transformation ("VST"). Then 50 principal components (PCs) were utilized to calculate the k-nearest neighbors (KNN) graph based on the Euclidean distance in PCA space. Clusters were then visualized using a Uniform Manifold Approximation and Projection (UMAP) plot to annotate the cell types by gene markers. Genes with a log2 (fold change of expression) of at least 0.25 and FDR < 0.01 were selected as DEGs. Cell cycle phase assignments were performed using the Seurat package. Monocle2 was used for pseudo-time analysis [36].
To comprehensively evaluate the conversion rate of a related cell fate, the AddMod-uleScore function of the Seurat v3 R package was used to calculate the Rod score and Cone score, respectively. For Rod score calculation, the "RCVRN", "PDE6G", "VSX2", "CRX", "OTX2", "RHO", "NRL", "PAX6", "ASCL1", "RXRG", "RORB", and "THRB" were used. For Cone score calculation, "GNAT2", "OPN1SW", and "OPN1LW" were used. For each sample, this calculates the average expression of genes in the module, subtracted by the average expression of a randomly selected set of control genes with similar expressions across the samples. As with input to the function, we used the normalized expression as described above, and in each case, we used 100 random control genes.

Statistical Analysis
All data were presented in the form of mean ± standard deviation (SD). For comparisons between multiple groups, an analysis of variance (ANOVA) followed by a Tukey's or Bonferroni post hoc test was used. Comparisons between two groups used the Student's t-test. For comparison between groups with unequal variances, a non-parametric Mann-Whitney test was used to compare two groups. Replicates were obtained by measuring distinct donor samples (3 biological replicates), and statistical analyses were based on at least 3 experimental replicates. Details of the number of biological replicates and p values are provided in the figure legends. For all analyses, a significance level α = 0.05 was set with a 95% confidence level, and differences were considered significant at a p value ≤ 0.05.

Direct Conversion of RPE Cells into CiPCs with Small Molecules
It is very instructive that fibroblasts could be directly and chemically reprogrammed into functional CiPCs by five small molecule factors (5F), but the efficiency was rather low [14]. We first tested the effect of cell source on the conversion process, and both HDF and RPE cells were used for 5F-mediated CiPCs induction (Figure 1a). For HDF, no obvious morphological changes were observed at the early stage (Day (D) 3) of induction ( Figure 1b). After 10 days (D10) (Figure 1b), HDF were reprogrammed into CiPCs (H-CiPCs), which was consistent with the previous study. For RPE cells, both bright-field images and live-cell time-lapse imaging showed that they gained dramatic morphological changes as early as the first day of reprogramming (Figure 1b,c, Supplementary Video S1) and exhibited an intermediate state (D3). On D10, most of the RPE-derived CiPCs (R-CiPCs) appeared as single cells with round soma, among which were scattered cells with tubular structures and small colonies (Figure 1b). Extended cultures showed that cells were able to be maintained in photoreceptor differentiation medium (PDM) to D22 (Supplementary Figure S1a). These results reflect a faster response of RPE cells to 5F compared to HDF.
We also tested the small molecules individually and found that they failed to generate a good number of RCVRN-positive cells (Supplementary Figure S2a). Additionally, different seeding densities were also tested to determine the proper starting density. As shown in Supplementary Figure S2b, 3 × 10 4 cells/cm 2 was the most suitable seeding density for RPE cell reprogramming.

Suppression of PTBP1 Partly Enhances CiPCs Induction
PTBP1 is an RNA-binding protein and is reported to inhibit neuronal differentiation by suppressing the splicing of a subset of neural targets [37]. Recently, studies have shown remarkable efficiency in neuronal conversion by downregulating PTBP1, although this observation is still being debated [22,[38][39][40]. Antisense oligonucleotides (ASO), a promising therapeutic agent, could bind sequences specifically to the target RNA and modulate protein expression [41]. To test whether repression of PTBP1 could improve the reprogramming efficiency, we combined PTBP1 ASO treatment (added from D1 to D3) with 5F (5FA) for CiPCs induction. We found that the morphological changes due to 5FA treatment were similar to those of 5F-induced H-CiPCs and R-CiPCs (Figure 2a,b). Compared to 5F, 5FA leads to downregulation of PTBP1 (Figure 2c,d) and elevated expression of rod photoreceptor-specific genes such as RHO and RCVRN at D10 (Figure 2e,f). Correspondingly, immunostaining analysis indicated that RHO and RCVRN positive cells in 5FA condition were significantly increased in both H-CiPCs (10.1% ± 1.2%, p < 0.01 and 13.7% ± 4.5%, p < 0.01) and R-CiPCs (38.3% ± 8.2%, p < 0.001 and 42.1% ± 6.0%, p < 0.001) ( been widely used to test various retinal cell transplantation treatments [17,26] (Supplementary Figure S3a). As is shown in Supplementary Figure S3b,c, after 4 weeks of NaIO3 injection, the thickness of the retina's outer nuclear layer decreased significantly. We collected the R-CiPCs after 10 days of induction for subretinal transplantation. Upon immunofluorescence staining analysis at 4 weeks post-transplantation, we observed that the transplanted cells survived well and maintained the characteristics of CiPCs in the subretinal space of the injected eyes in close apposition with the outermost layer of the host retina (inner nuclear layer and occasional patches of the remaining outer nuclear layer). We confirmed that these cells were of human origin by staining with human nuclear antigen antibody (HNA), and around 23% of them were RCVRN positive (Figure 2i,j, Supplementary Figure S3d,e). Collectively, these results suggest RPE cells could be a promising cell source for photoreceptor cell generation for retinal cell replacement therapy.  Meanwhile, we also tested if suppression of PTBP1 alone is sufficient to induce reprogramming. As shown in Figure 2a Our results indicate that PTBP1 suppression itself is not enough to reprogram HDF and RPE cells into photoreceptor-like cells in vitro. Collectively, these results reflect that PTBP1 ASO further promoted, in part, the conversion rate based on 5F treatment.

CiPCs Sustained the Photoreceptor-Like Features In Vivo
To explore the practical implications of R-CiPCs from a clinical perspective, we transplanted the cells into the subretinal space in a NaIO 3 -induced RPE and photoreceptor degeneration mouse model, which has a similar pathological process to dry AMD and has been widely used to test various retinal cell transplantation treatments [17,26] (Supplementary Figure S3a). As is shown in Supplementary Figure S3b,c, after 4 weeks of NaIO 3 injection, the thickness of the retina's outer nuclear layer decreased signifi-cantly. We collected the R-CiPCs after 10 days of induction for subretinal transplantation. Upon immunofluorescence staining analysis at 4 weeks post-transplantation, we observed that the transplanted cells survived well and maintained the characteristics of CiPCs in the subretinal space of the injected eyes in close apposition with the outermost layer of the host retina (inner nuclear layer and occasional patches of the remaining outer nuclear layer). We confirmed that these cells were of human origin by staining with human nuclear antigen antibody (HNA), and around 23% of them were RCVRN positive (Figure 2i,j, Supplementary Figure S3d,e). Collectively, these results suggest RPE cells could be a promising cell source for photoreceptor cell generation for retinal cell replacement therapy.

Small Molecules Effectively Reshaped the Transcriptional Profile of RPE Cells
To comprehensively investigate the molecular changes during reprogramming, RNAseq was performed on the cells at the initial (D0, HDF, and RPE cells), early (D3, with 5F and 5FA induction), and late stages of reprogramming (D10, with ASO, 5F, and 5FA induction). Principal component analysis (PCA) distinguished each type of the CiPCs well based on their cell origins, and both followed a reprogramming path similar to fetal retina development (Figure 3a, Supplementary Figure S4a,b), revealing the effective reshaping of the transcriptional profile. In particular, the photoreceptor genes showed higher expression levels in R-CiPCs than in H-CiPCs (Figure 3b).
Then we compared the transcriptional profile of R-CiPCs at different stages to explore the mechanism of R-CiPCs induction. Comparing the early intermediates at D3 to RPE control, 1068 DEGs were detected (Figure 3c, Supplementary Table S1). GO analysis showed that the upregulated genes were enriched in neuronal genesis, neurotransmitter uptake, and photoreceptor cell differentiation, while the downregulated genes were enriched in regulation of epithelial cell migration, epithelium development, and cell−cell junction assembly (Figure 3d).
Notably, the neuronal genes such as SOX8 and IGFN1 showed increased expression but the epithelium-related genes such as EGF and FGF2 decreased (Supplementary Figure S4c,d).
In addition, photoreceptor-specific transcriptional factors, such as ASCL1, RXRG, THRB, and RORB, were upregulated in RI (Supplementary Figure S4e). These results reflected an emerging lineage reprogramming from epithelial cells to neuronal cells in the intermediate state.
At D10, the transcriptome profile revealed consistent upregulation of the rod-specific genes in R-CiPCs compared to RPE cells (Supplementary Figure S4f, Supplementary Table S2), but only a slight increase in cone-related gene expression (Supplementary Figure S4g). As shown in Figure 3e,f, the upregulated genes were enriched in photoreceptor cell differentiation while the downregulated genes were enriched in epithelium development. Consistently, GSEA revealed that both RI and R-CiPCs expressed enriched genes involved in photoreceptor differentiation and function, while the control RPE cells showed enrichment in cell substrate junction, which confirmed continuous conversion from RPE cells to photoreceptors (Figure 3g,h, Supplementary Figure S4h).
In addition, we found that RPE cells were prone to exhibiting epithelial-mesenchymal transition (EMT) with increased passage number, as previously reported (Supplementary Figure S5a). The RPE cells undergoing EMT became less differentiated and lost their lineage-specific features. Correspondingly, we noticed that EMT-like morphological changes can be effectively eliminated through 5F treatment. Consistent with our observations, RNA-seq results demonstrated that the expression of EMT-related genes was significantly downregulated in R-CiPCs compared with RPE cells (Supplementary Figure S5b), which was confirmed by RT-qPCR and GSEA (Supplementary Figure S5d,j). Further analysis identified a decrease in TGF-β pathway genes in R-CiPCs (Supplementary Figure S5c,e). Because TGF-β has been reported to induce EMT [42], 5F could suppress EMT through the inhibition of the TGF-β pathway. In addition, we found that RPE cells were prone to exhibiting epithelial-mesenchymal transition (EMT) with increased passage number, as previously reported (Supplementary Figure S5a). The RPE cells undergoing EMT became less differentiated and lost their lineage-specific features. Correspondingly, we noticed that EMT-like morphological changes can be effectively eliminated through 5F treatment. Consistent with our observations, RNA-seq results demonstrated that the expression of EMT-related genes was significantly downregulated in R-CiPCs compared with RPE cells (Supplementary Figure  S5b), which was confirmed by RT-qPCR and GSEA (Supplementary Figure S5d,j). Further analysis identified a decrease in TGF-β pathway genes in R-CiPCs (Supplementary Figure ASCL1, a powerful pro-neural transcription factor, has been reported to stimulate the regeneration of retinal neurons from Müller glia [43]. A previous study has reported that NF-κB induced Ascl1 during the induction of CiPCs [14]. Consistently, we found that the upregulated genes in the CiPCs were enriched in the positive regulation of the NF-kB pathway, accompanied by upregulation of ASCL1 (Supplementary Figure S5f,g).
Glutamate is an essential neurotransmitter released by photoreceptors for retinal synaptic circuitry [44]. Activation of the glutamate receptor pathway has been reported to induce ASCL1 expression [45]. Our results demonstrated that the glutamate receptor pathway-related genes were gradually upregulated in R-CiPCs compared to RPE cells, accompanied by upregulation of ASCL1 (Supplementary Figure S5g,h). This result was also confirmed by RT-qPCR of GRM4 and GLUD2 and GSEA (Supplementary Figure S5i,k). Thus, our data implied the potential involvement of the glutamate receptor pathway in the photoreceptor lineage transition, which needs to be further elucidated.

Global DNA Methylation Remodeling during Direct Reprogramming of RPE Cells to CiPCs
To investigate DNA methylation changes due to reprogramming, base-resolution methylomes were generated from control RPE cells as well as the reprogrammed R-CiPCs (Supplementary Table S3). Remarkably, both 5F and 5FA induced genome-wide demethylation (Figure 4a). VPA was reported to induce DNA demethylation through the action of ten-eleven translocation (TET) family enzymes in an active DNA demethylation pathway in HeLa cells [46]. Thus, the global demethylation may result from the increased expression of TET family genes due to VPA treatment (Figure 4b). Of note, 4195 genes showed dramatic demethylation in their promoters (Figure 4c). GO analysis indicated these genes were functionally enriched in the regulation of glutamate secretion and neurogenesis (Figure 4d,e,h). Notably, we found that the majority (437 out of 632) of the DEGs with demethylated promoters exhibited a significant increase in gene expression (Figure 4g), especially for the genes related to neurogenesis (Figure 4i).
In addition to the global demethylation, we noticed 166 genes showed significantly increased DNA methylation in their promoters (Figure 4c). These genes were functionally enriched in epithelium development (Figure 4f,j), and their hypermethylation was accompanied by attenuated gene expression (Supplementary Figure S4d). The hypermethylation of epithelium lineage genes could be attributed to the increased expression of DNA methyltransferase 3 (DNMT3) family genes (Figure 4k). These results demonstrated the removal of epithelium lineage features, and the permission of photoreceptor features due to DNA methylation remodeling during reprogramming.

Single-Cell Analysis of Cell Populations during CiPCs Reprogramming
Since direct reprogramming is an unsynchronized process, traditional methods such as bulk RNA-seq of inhomogeneous cell populations cannot fully dissect the detailed changes in the reprogramming pathway. Therefore, scRNA-seq provides a greater opportunity to capture the transcriptional state and delineate the trajectory of direct reprogramming at a single-cell resolution. We conducted 10× genomics scRNA-seq at different time

Single-Cell Analysis of Cell Populations during CiPCs Reprogramming
Since direct reprogramming is an unsynchronized process, traditional methods such as bulk RNA-seq of inhomogeneous cell populations cannot fully dissect the detailed changes in the reprogramming pathway. Therefore, scRNA-seq provides a greater opportunity to capture the transcriptional state and delineate the trajectory of direct reprogramming at a single-cell resolution. We conducted 10× genomics scRNA-seq at different time points during RPE cell reprogramming to CiPCs. Sequencing data were obtained from 23,661 individual cells at early and late reprogramming phases (D3 and D10) (Figure 5a), with 4378 median genes and 82,255 mean confidently mapped reads per cell. We projected all cells on a UMAP plot and identified five transcriptionally distinct clusters: RPE cells, retinal progenitor cells (RPCs), reprogramming intermediate 1 (RI1), reprogramming intermediate 2 (RI2), and CiPCs, which reveal a dynamic transcriptomic transition from the parental RPE cells (Figure 5b). We specified the cell types by their unique expression of specific known marker genes (Figure 5c). COL4A1 and MKI67 are well known to be highly expressed in RPE cells and RPCs, respectively, whereas ANXA1, HMGA1, and NEAT1 are widely expressed in neural cell types [47][48][49][50]. In addition to RPE cells, a fraction of RPCs were also present in the control group, which is consistent with the previous study [51]. The major components of the 5F treated cells at D3 were RPE cells, RPCs, and RI1, while 5FA treated cells at D3 were mainly composed of RPE cells, RPCs, RI1, and RI2. Additionally, CiPCs made up most of the 5FA group at D10. Compared with the control group, the proportion of RPE cells exhibited a gradual decline in the 5F D3, 5FA D3, and 5FA D10 groups, reflecting the progressively increasing conversion rate (Figure 5d). Detailed analysis showed that around 46% of the CiPCs at D10 were rod-like cells, while only 1% committed to the cone photoreceptor fate (Figure 5e), which is consistent with the immunostaining result (Supplementary Figure S2g). Similar to the bulk RNA-seq result, glutamate receptor pathway genes were significantly upregulated in RI and CiPCs (Supplementary Figure S6a). To understand the molecular features of RI1, RI2, and CiPCs, we compared the transcriptome of these cell subtypes to RPE cells. The quantities of DEGs identified in RI1, RI2, and CiPCs were 211, 1098, and 859, respectively. Although these differences were highly significant, the cell conversion process was ongoing and modest. GO analysis showed that the DEGs in these subtypes were all enriched in neurogenesis, axon genesis, and neuron differentiation (Figure 5g-i). Thus, it is confirmed that the pharmacologic reprogramming induced RPE cells into a neuronal lineage, which can go further and be more mature.
Then we determined the cell cycle state for each subtype via the Cell-Cycle Scoring analysis based on the scRNA-seq data. A decreased proportion of cells in a particular cell cycle phase suggests a more rapid transition in that phase [52]. Multiple studies have established that the G1 phase length is short in self-renewing cells, and can prolong with cell differentiation [53,54]. As shown in Figure 5f, there were no RPCs in the G1 phase, revealing its highly proliferative feature. Compared to RPE cells, the RI1 and RI2 groups exhibited shorter G1 phases, which indicated their reentry into the cell cycle and proliferation during the induction process. With further induction, CiPCs showed extended G1 phases similar to RPE cells, implying they turned into a relatively stable condition.

Trajectory and Pseudo-Time Analysis Identify Dynamic Pathways during CiPCs Reprogramming
To further delineate the cell transition pathways, we used the Monocle method to determine a pseudo-temporal order between cell types [36]. Monocle uses an algorithm

Trajectory and Pseudo-Time Analysis Identify Dynamic Pathways during CiPCs Reprogramming
To further delineate the cell transition pathways, we used the Monocle method to determine a pseudo-temporal order between cell types [36]. Monocle uses an algorithm to learn the sequence of gene expression changes that each cell must go through as part of a dynamic biological process. The pseudo-time trajectory is constructed by placing each cell at its proper position based on the gene expression changes. The pseudo-time trajectory axis indicated that the RI state was between RPE cells and CiPCs (Figure 6a-d).
As the pseudo-time values increased, cells transited through the RI state and eventually progressed towards CiPCs (Supplementary Figure S6b-e).
Then, we identified 5626 genes differentially expressed along the pseudo-time (FDR < 0.01), including the typical photoreceptor marker genes (Figure 6e). GO analysis on the top 1000 pseudo-time-related DEGs revealed that these genes were enriched in glutamate homeostasis and nervous system development, which was also consistent with the bulk RNA-seq result (Figure 6f). Among them, RPE lineage-related genes such as TYR, SERPINF1, and TTR were downregulated over the pseudo-time progression (Figure 6g, Supplementary Figure S6f), and the rod photoreceptor-lineage related genes such as RCVRN, CRX, NR2E3, PDE6G, and NRL were gradually upregulated over pseudotime (Figure 6h, Supplementary Figure S6f).
top 1000 pseudo-time-related DEGs revealed that these genes were enriched in glutamate homeostasis and nervous system development, which was also consistent with the bulk RNA-seq result (Figure 6f). Among them, RPE lineage-related genes such as TYR, SER-PINF1, and TTR were downregulated over the pseudo-time progression (Figure 6g, Supplementary Figure S6f), and the rod photoreceptor-lineage related genes such as RCVRN, CRX, NR2E3, PDE6G, and NRL were gradually upregulated over pseudo-time (Figure 6h,  Supplementary Figure S6f).

RPE-Derived CiPCs Resemble Photoreceptors in Human Fetal Retina
To investigate the similarity of CiPCs derived from fetal RPE cells and human developing photoreceptor cells, we further compared our scRNA-seq datasets with the results

RPE-Derived CiPCs Resemble Photoreceptors in Human Fetal Retina
To investigate the similarity of CiPCs derived from fetal RPE cells and human developing photoreceptor cells, we further compared our scRNA-seq datasets with the results from human fetal retina (fRetina) at fetal day (FD) 125, when all major retinal cell types, including photoreceptors, are present [55]. Through integrating the datasets and projecting them onto a single UMAP plot, nine distinct clusters were obtained (Figure 7a,b). Among them, seven clusters were previously well-defined in fRetina, including retinal progenitor cells (RPCs),

photoreceptors (PR), amacrine cells (AC), retinal ganglion cells (RGC), horizontal cells (HC), ON bipolar cells (ON BC), and OFF bipolar cells (OFF BC). The intermediates (RI1 and RI2)
and RPE cells could not be distinguished well, while CiPCs could be subsequently divided into two groups: CiPCs-precursors (CiPCs-pre) and CiPCs-photoreceptors (CiPCs-PR). Compared with RPE cells and RI, CiPCs-PR were closer to the photoreceptor subtypes in fRetina with higher expression levels of photoreceptor marker genes (Figure 7c,d). Furthermore, this combination enabled us to identify the proportion of photoreceptor-like cells in the D10 samples (Figure 7e). Cell proportion analysis showed that photoreceptors (CiPCs-PR) constituted about 28.3% of all the cells (724 out of 2552) in the D10 sample.
CiPCs could be subsequently divided into two groups: CiPCs-precursors (CiPCs-pre) and CiPCs-photoreceptors (CiPCs-PR). Compared with RPE cells and RI, CiPCs-PR were closer to the photoreceptor subtypes in fRetina with higher expression levels of photoreceptor marker genes (Figure 7c,d). Furthermore, this combination enabled us to identify the proportion of photoreceptor-like cells in the D10 samples (Figure 7e). Cell proportion analysis showed that photoreceptors (CiPCs-PR) constituted about 28.3% of all the cells (724 out of 2552) in the D10 sample.

Unique DNA Methylation Profiles in RPE Cells May Contribute to the Better Efficiency in Photoreceptor Induction Than That in Fibroblasts
To investigate the role of DNA methylation in determining the efficiency of reprogramming, methylome comparison was also conducted on H-CiPCs and R-CiPCs. Similar to R-CiPCs, H-CiPCs exhibited a low methylation level in the promoter of photoreceptor genes, reflecting the activation of these genes (Figure 8a-c). Genome-wide analysis of all of these genes showed that the DNA methylation pattern in fRetina was closer to R-CiPCs in most regions (Figure 8d). Among all the 27,535 gene promoters analyzed, 18,986 from R-CiPCs exhibited a closer methylation pattern to that of fRetina (Figure 8e). We further filtered the top 1000 promoters according to the methylation difference between H-CiPCs and R-CiPCs. Notably, the genes corresponding to these promoters were enriched in retinal development (Figure 8f). Further investigation of the methylation profiles of HDF and RPE cells showed that most of these DNA methylation differences already existed before 5F treatment (Figure 8b,c,f,g). As a result, the retained epigenetic memory of their cellular origin could contribute to the varying reprogramming efficiency for H-CiPCs and R-CiPCs.
in most regions (Figure 8d). Among all the 27,535 gene promoters analyzed, 18,986 from R-CiPCs exhibited a closer methylation pattern to that of fRetina (Figure 8e). We further filtered the top 1000 promoters according to the methylation difference between H-CiPCs and R-CiPCs. Notably, the genes corresponding to these promoters were enriched in retinal development (Figure 8f). Further investigation of the methylation profiles of HDF and RPE cells showed that most of these DNA methylation differences already existed before 5F treatment (Figure 8b,c,f,g). As a result, the retained epigenetic memory of their cellular origin could contribute to the varying reprogramming efficiency for H-CiPCs and R-CiPCs.

Discussion
Chemical approaches lead to major advancements in cell reprogramming and regenerative medicine. For example, small molecule treatment has been reported to reprogram human somatic cells to pluripotent stem cells, reflecting the powerful potential for small molecules in cell reprogramming [56]. Here, we demonstrate that small molecules could efficiently convert RPE cells into postmitotic photoreceptor-like cells, providing another example of the usefulness of chemical induction in cell lineage conversion. DNA methylation has been frequently attributed to an essential role in lineage transition [57]. Epigenetic regulators, such as the TET gene family and DNMT3 gene family, have been reported to jointly regulate promoter epigenetic landscapes. Consistently, this study identifies increased expression of all the genes in the TET and DNMT3 families, revealing their critical role in reshaping the DNA methylation landscape of CiPCs. Additionally, ASCL1, a known transcription factor, has been reported to induce promoter methylation of fibroblast-specific genes during reprogramming [58]. In general, the DNA methylation reconfiguration during chemical reprogramming is the collective outcome of epigenetic factors and transcription regulators.
The epigenetic memory of the original cells has strong effects on reprogramming efficiency. By stabilizing the original state of genes of the previous donor cell type, it impedes the conversion process and causes incomplete reprogramming to a different cell type [59]. Both RPE and photoreceptor cells are developed from the optic vesicle in the embryonic stage, suggesting that they may share similar epigenetic states. Our results demonstrated that RPE cells not only responded faster to the 5F treatment but also exhibited higher efficiency compared to HDF. These results suggest that similar epigenetic states between the original cells and induced cells could facilitate cell lineage transition during CiPCs reprogramming.
The RNA-binding protein PTBP1 has been considered to be a promising target for direct reprogramming of non-neuronal cells into functional neurons [38]. Recent in vivo studies showed that downregulation of PTBP1 could reprogram Müller glia into retinal ganglion cells with high efficiency [39]. However, these claims have been challenged for technical reasons and lack solid proof to demonstrate lineage switching [60]. Our results show that PTBP1 ASO coupled with 5F could improve the CiPCs induction efficiency, but repression of PTBP1 alone is not sufficient to reprogram RPE cells or HDF into photoreceptors, suggesting the PTBP1 pathway only plays an auxiliary role in CiPCs induction.
Recent research has also revealed that ESCs and iPSCs may be good sources for generating CiPCs. However, potential contamination of undifferentiated pluripotent stem cells or incomplete differentiation may result in tumorigenicity [61]. RPE cells are differentiated cells with no risk of developing tumors. Furthermore, exogenous RPE cells can be easily transplanted into the subretinal space between the RPE and outer segments of photoreceptors. This raises the possibility of future in situ reprogramming of subretinal transplanted RPE cells into photoreceptors for better integration into the retina structure in vivo.
RPE cells are involved in the transportation of nutrients, recycling of proteins, elimination of photoreceptor debris, and also secretion of some growth factors [62]. All of these functions enable RPE cells to play a critical role in regulating the microenvironment in the eye. In fact, photoreceptor loss in patients is often accompanied by dysfunction or death of RPE cells [63]. Therefore, it is also possible to transplant CiPCs and RPE cells simultaneously when both types of cells are produced in vitro. In this way, we not only treat the loss of photoreceptors but also reconstruct the subretinal microenvironment with both RPE cells and CiPCs.
Reprogramming efficiency is majorly determined by the effect of the small molecules as well as the intrinsic properties of the source cells. RPE cells in amphibians can regenerate the entire retina under an injury state, and several studies have shown that mammalian RPE cells can be reprogrammed into retinal progenitor cells and retinal neurons in vitro [64]. Its innate plasticity further assures the effectiveness of the treatment with small molecules. Since in vivo reprogramming holds tremendous promise in regenerative medicine, it would be of great interest to determine whether endogenous RPE cells can be converted into photoreceptor-like cells within the subretinal space in future studies.

Conclusions
Our study establishes an improved approach to generating CiPCs from direct reprogramming of somatic cells. RPE cells, as the source cell, exhibited superior conversion efficiency compared to HDF via five small molecule factors (5F). When the chemical treat-Cells 2022, 11, 3146 20 of 24 ment was applied, further downregulation of PTBP1 could promote conversion efficiency. Through multi-omics sequencing, we found that 5F treatment could induce global transcriptional and epigenetic remodeling for the epithelium-to-retinal neuron transition. More detailed analysis revealed that 5F could suppress EMT through inhibition of the TGF-β pathway but enhance CiPCs conversion through activation of the glutamate receptor pathway. Additionally, the preferable performance of RPE cells for CiPCs induction may benefit from their retinal lineage epigenetic modifications. Collectively, our study has significant implications for chemical approaches in the future clinical use of retinal cell therapy.   [14,18,46,[65][66][67][68][69][70][71][72][73][74][75][76][77][78] are cited in the Supplementary Materials (Supplementary Table S5).
Data Availability Statement: All raw RNA-seq, EM-seq, and scRNA-seq data reported in this paper have been deposited in the National Center for Biotechnology Information Gene Expression Omnibus (GEO) database, at the accession number GSE192882. The RNA-seq and WGBS data of human fetal retina were obtained from GSE87064 [79]. The single-cell data from human fetal retina was obtained from GSE142526 [55]. The WGBS data of human fibroblasts were obtained from GSE8634 [80]. There are no restrictions on data availability, and all data will be made available upon request directed to the corresponding authors.