We first aimed to assess the ability of sci-Plex to assign cells to specific wells in an expected manner by observing the in vitro kinetics and dynamic gene expression program of Ascl1-dependent MG reprogramming. Retinas were isolated from P11 mice expressing a constitutive rtTA driving a tet-inducible Ascl1-GFP. The retinas were dissociated and grown as previously described to enrich for MG (4). We have previously shown that induction of Ascl1, by addition of doxycycline to the medium, induces neurogenesis in the MG, and that the new neurons generated from the reprogrammed MG most closely resemble bipolar neurons (4) (Fig. 1A). Taking advantage of the ease by which condition number can be scaled with sci-Plex, we induced Ascl1 in replicate for 1-5 days at varying windows of time, allowing us to temporally track the conversion of MG to neurons (Fig. 1A-B). Upon collection, nuclei were isolated in their individual treatment wells, the nuclei were barcoded with well-specific polyadenylated oligonucleotides “hashes”, and then pooled for sci-RNA-seq2 (13, 14). We recovered 12,850 cells across two experiments and 30 wells (Timecourse: 7,004, Pulse: 5,846) with a median UMI count of 3,172 and 2,398 after all filtering steps, respectively (Fig. S1 A-K). The hash recovery rate was 68% for both experiments (Fig. S1C-D, G-H). The experiments were integrated and visualized by UMAP using Monocle3 (14, 18). The clusters were annotated according to cell type using the expression of known marker genes (Fig. S1L-N, Table S1) (14, 18). Most cells fell into one of four clusters: (1) MG; (2) Progenitor-like cells (ProL); (3) Transition; and (4) neuronal precursors (NeuPre). In addition, we identified other clusters of non-MG derived cells, including microglia, astrocytes, and fibroblast-like cells (FibroL), potentially derived from vascular cells (Fig. 1C).

Having annotated the cells, we next ordered all the cells originating from reprogrammed MG into a pseudotime trajectory to better understand the sequence of gene regulatory events on the path from MG to neuronal identity (Fig. 1D). We then assessed how the timing and duration of Ascl1 overexpression affected reprogramming progression when Ascl1 was added continuously (Timecourse) or pulsed (Pulse). In the Timecourse experiment, Ascl1 expression was induced for either 0,1,2,3,4 or 5 days and then the cells were collected (Fig. 1B). During the timecourse, we saw that Ascl1 stimulated a rapid transition away from MG (i.e. the start of the pseudotime trajectory) that persisted across all timepoints (Fig. 1E). As expected, increased duration of Ascl1 overexpression led to cells further along the pseudotime trajectory, consistent with the need for Ascl1 expression to drive cells from the ProL state towards the neuronal state.

In the Pulse experiment, we initiated Ascl1 expression in all samples at the same time, but then removed doxycycline after either 1, 2, 3, 4, or 5 days to shutdown Ascl1 overexpression. The cells were all collected 5 days after the start of Ascl1 expression regardless of Ascl1 expression duration (Fig. 1B, Pulse). When Ascl1 was maintained for the entire 5 day period (Fig. 1D-E), the distribution of cells across pseudotime was similar to that observed in the Timecourse experiment, with few MG remaining. However, the pseudotime profiles were remarkably different when Ascl1 overexpression was removed for any period of time. When Ascl1 was expressed in the cells for 2 or fewer days, the majority of cells were identified as MG (Fig. 1E-F). After 3 or more days of Ascl1 expression, transition and neurogenic precursor (NeuPre) cells began to accumulate, though a substantial number of MG were also present even after 4 days of Ascl1 exposure and one day without (Fig. 1F). These results suggest that even after 4 days of Ascl1 expression, removal of Ascl1 allows a subset of the cells to return to a glial state, while another subset of cells that have made it far enough along the reprogramming trajectory are no longer dependent on Ascl1 to remain committed to their neuronal fate.

To further understand the molecular features of the cells as they lose dependence on Ascl1, we analyzed changes in gene expression around the pseudotime stage associated with transition to the neuronal fate. The loss of dependence on Ascl1 appeared around a pseudotime value of 15 (Fig. 1E). To find the genes dynamically expressed during this time, we performed differential gene expression analysis across pseudotime using cells with pseudotime values from 10-20 (Dataset S1). We were particularly interested in identifying transcriptional regulators and signaling molecules. In the top 250 most differentially expressed genes, three major regulatory patterns (Cluster 1: decreasing, Cluster 2: increasing, Cluster 3: a transient increase) were observed (Fig. 1G-H). The downregulation of Id1 and Id3 in Cluster 3 was consistent with known biology: these factors (1) form heterodimers with bHLH TFs making them incapable of binding their target sites (19) and (2) have been previously implicated in maintaining retinal progenitors and Muller glia in a progenitor-state (2, 4–8). Genes in Cluster 2, which increase during the progenitor-to-neuron transition, are also associated with the normal development of neurons (Elavl3, Map2) and with processes such as neuronal maturation and synaptogenesis (Nrxn3, Syt1). There were also four TFs that increased in their expression right around the loss of Ascl1 dependence: Sox4, Neurog2, Myt1, and Rorb. All of these genes are expressed in the retina during the early stages of neurogenesis (20–25). Of note, Myt1 is a direct target of Ascl1 and although it has been studied more thoroughly in the brain, it is associated with repression of progenitor gene expression programs (20, 21). Another factor on the list, Rorb is involved in the specification of photoreceptor, amacrine, and horizontal but not bipolar cell fates (26, 27). Our results thus show the ability of sci-Plex to be used in MG cultures and we have identified potential regulators of critical steps in the MG to neuron reprogramming paradigm.

The above results supported the potential of sci-Plex to study neurogenic reprogramming, so we next embarked on a larger-scale screen for factors that can improve the efficiency of neurogenesis from MG. We screened through 92 small molecules from the TocriScreen Stem Cell Library (Dataset S2). The compounds target a range of biological functions with a bias towards epigenetic modulators and modulators of common signaling pathways (Fig. 2A). We followed the same treatment regime as for the timecourse experiment in which MG were isolated from P11 mice that express a constitutive rtTA and harbor a tet-inducibe Ascl1-GFP. The MG were selectively cultured before induction of Ascl1 with doxycycline. Small molecules from the TocrisScreen Stem Cell Library were added at the same time as the doxycycline and remained in culture until the cells were collected on Day 5 (Fig. 2B). The compounds were applied across three doses (0.1, 1, and 10 uM) with each compound being tested on one of two different collection days. We included three wells per collection day in which only doxycycline was added (Only Ascl1) and three wells per collection day in which doxycycline was not added (No Ascl1). To maximize the range of compounds tested, we opted to perform the sci-Plex experiment once, and then follow up with compounds that elicited strong effects through subsequent in vivo validation experiments. In sum, we screened 92 small molecules and two control conditions from a total of 265 individual wells after filtering out conditions with low cell recovery.

We used sci-RNA-seq2 to profile a total of 27,152 cells (Collection 1: 6,282, Collection 2: 20,870) (Fig. S2A-G). The median UMI count was 3,551 for Collection 1 and 2,166 for Collection 2 (Fig. S2H-I). A hash could be called for 65-74% of the cells (Fig. S2F). We performed dimensionality reduction and clustering with Monocle3 and used marker genes to annotate cell types by cluster. All cell types not of retinal/MG origin were discarded (Fig. S2J-L), and the dimensionality reduction and cell type annotation was repeated (Fig. 2C, Table S1).

To identify the molecules most likely to promote reprogramming, we pooled the clusters corresponding to neurogenic precursors (NeuPre: Gap43, Snap25, Dcx) and bipolar neurons (BP: Snap25, Vsx2, Cabp5, Otx2, Grik1) together into a common Neuron annotation and for each drug, calculated the abundance of Neurons relative to Ascl1(Fig. 2D and Dataset S2). Many of the molecules did not have clear effects on the reprogramming process. However, we found a several molecules that increased neurogenesis and others that reduced neurogenesis. The molecules that most markedly increased the percentage of Neurons, included DBZ (28), DMH-1 (29), RepSox (30, 31), SB 431542 (32, 33), and WH-4-023 (34–36). These compounds target one of three molecular signaling pathways: Notch, BMP/TGF-beta, and LCK/SRC. We also calculated the number of cells recovered per well, and noticed that for a few treatments, far more total cells, regardless of cell type identity, were recovered than average. This was most prominent for metformin, sodium 4-phenylbutyrate, and valproic acid (Fig. 2E). Both sodium 4-phenylbutyrate and valproic acid are HDAC inhibitors which is of interest because TSA, another HDAC inhibitor, is a significant stimulus for Ascl1-dependent reprogramming in vivo (2). Unexpectedly, metformin, a compound that is approved for the treatment of diabetes with other potential clinical uses(37), also increased cell recovery in our assay. Because metformin has not been evaluated in the context of retinal reprogramming, we decided to pursue it and the 5 compounds that increased neurogenesis with further experiments in vivo. Thus, the use of sci-Plex aided us in narrowing down 92 compounds to 6 for follow up investigations.

In addition to detecting shifts along the reprogramming trajectory, we also detected what appeared to be previously undetected cell states (Fig. 2C, Unknown clusters). To characterize these, we performed Biological Process GO term enrichment on genes with high regional specificity for each cluster (Dataset S3). Cluster Unknown 1 was enriched with genes related to ER stress, Unknown 2 had markers of Immune reactive cells, and Unknown 3 had non-retinal neuronal characteristics (Fig. S2M-O, Dataset S3). The Unknown 4 cells had 77 genes with high specificity to the cluster, but only two enriched gene sets “regulation of ion transmembrane transport” and “action potential” (Dataset S3). With such a small number of enriched gene sets, we were unable to make further conclusions about the identity of that cell cluster and have left them annotated as Unknown.

We next set out to understand how each treatment altered the distribution of cells across these novel clusters. We generated a heatmap in which treatments with similar cell type compositions were clustered together (Fig. S2P). In doing this, we found that for a small number of treatments, the majority of cells come from either the ER stressed cluster of cells or the Unknown cluster of cells (Fig. S2Q-R). Interestingly, the treatments that led to the Unknown cluster include two histone deacetylase (HDAC) inhibitors (CI 994, SAHA) (38, 39). It’s possible that these cells are deregulated without having yet experienced a change in fate towards a more specific cell type. The treatments that lead to increased cell counts in the ER stressed cluster include two bromodomain (acetylated lysine binders) inhibitors (I-BET 151 dihydrochloride, CPI 203) (40, 41). While these observations don’t directly help us identify better treatments for reprogramming MG, they do help us find compounds that affect the molecular state of MG cells exposed to Ascl1 overexpression (2).

We next aimed to assess whether the positive hits that increased neurogenesis from the in vitro sci-Plex screen could be validated in vivo. Before testing compounds from the in vitro screen in vivo, we sought to validate that in vivo MG express the signaling pathways targeted by our small molecule compounds during neurogenic reprogramming. We analyzed scRNA-seq libraries made from our in vivo regeneration experiments where we used mice engineered to express Ascl1 specifically in the MG (Fig. 3A), then induced retinal injury with an intravitreal injection of NMDA, and a co-injection of TSA (Fig 3B). The mice were sacrificed and GFP expressing cells were sorted and subjected to 10x scRNA-seq either 5, 9, or 21 days after injury (the day 5 and 9 libraries were originally published in (42)). Dimensionality reduction was performed to generate a UMAP from which clusters were defined and cell types were annotated using Marker genes (Fig. 3C-D, Fig S3A-H, Table S1). A small number of rod and cone cells are present even in the earliest time point and most likely are contaminating cells from the FACs sorting. As expected, with increasing time after injury, we observe greater numbers of MG-derived bipolar cells. This is most evident in the time between 9 and 21 days. However, even between 5 and 9 days there is a small increase in bipolar cells. Additionally, when pseudotime analysis was performed, a shift in the MG cluster towards a more mature state became apparent even by 9 days (Fig. 3E). Lastly, using the pseudotime analysis across all time points, we looked at how the expression of key target genes for each of the drugs are expressed throughout the reprogramming process (Fig. 3F). In all cases, at least some of the target genes are expressed early and decrease over time, making them ideal candidates for small molecule intervention.

Using the same mice as above, we aimed to validate five hits in vivo (the γ-secretase inhibitor DBZ, Lck and Src inhibitor WH-4-023, LKB1/AMPK activator metformin, the BMP inhibitor DMH-1, and the TGF-BR1 inhibitor SB43152) for their ability to increase neurogenic reprogramming of MG in vivo. We originally also identified RepSox as a hit, but because RepSox and SB43152 have the same target, we decided to only pursue SB43152. Ascl1 was induced in the MG with tamoxifen, retinal injury was performed by NMDA, and TSA was given in conjunction with one of the small molecule hits from the screen; animals were sacrificed two weeks later, and their retinas were assayed for lineage traced MG-derived cells (GFP+) and the retinal neuronal maker (Otx2+) (Fig 3G). Out of the compounds tested, two significantly boosted the number of Otx2+ neuronal cells (Fig 3H-J).

The first identified compound was DBZ, a γ-secretase/Notch inhibitor (Fig. 3I). This is consistent with reports in the chick retina where the γ-secretase inhibitor DAPT increased neuronal differentiation from proliferating MG (43–45). Inhibition of Notch is also well known to promote neural differentiation from retinal progenitors in mouse and human (46, 47). While this result was not unexpected, DBZ represents a new compound found to promote neurogenesis in the mammalian retina. The second positive hit, metformin, showed the biggest effect on MG-neurogenesis (Fig. 3H-I). In some retinas, metformin increased the rate of neurogenesis to 75% of the Ascl1-expressing MG, a higher rate than any of our previous pharmaceutical treatments. The MG derived neurons expressed Otx2, and were primarily located in the INL; this is similar to what we observe in the ANT treatment condition, and so it does not appear that metformin altered the fates of the MG-derived neurons; In addition, the neurons in the metformin treated retinas do not express markers of ganglion/amacrine cells (Figure S3K). These in vivo experiments thus confirm that the in vitro sci-Plex screening is an effective method to identify new compounds that increase the regeneration potential of mammalian MG.

Mice were housed and treated under University of Washington Institutional Animal Care and Use Committee approved (UW-IACUC). Animals were maintained in groups of 2-5 per cage. For the in vivo experiments, Glast-CreER:LNL-tTA;TetO-mAscl1-ires-GFP mice are from mixed backgrounds of C57BL/6 and B6SJF1. The tetO-mAscl1-ires-GFP mice were a gift from M. Nakafuku (U. Cincinnati). Both sexes were used in all experiments and the mice used for injections were between the ages of P40-P60. To induce Ascl1 in MG in vivo, tamoxifen (Cayman Chemical, #13258-1G) was given intraperitoneally at 1.5mg per 100μl for four consecutive days. Intravitreal injections were performed on isoflurane anesthetized mice using a 32-gauge Hamilton syringe. All intravitreal injections were done at 2μl, and contained either NMDA (Sigma, #M3262-100MG) + drug or TSA (Tocris #1406/1) +drug. NMDA injections were done in 1μl volume at a concentration of 100mM. TSA injections were given in 1μl volume at a concentration of 1μg/μl. Drugs used were all from the Tocriscreen Stem Cell Library (Tocris, 7340) and were administered in 1μl of DMSO at 10mM. The mice were either randomized by animal tag number for injections (L1-4) or the vehicle (DMSO) was given to the left eye and the drug was given to the right eye (M). Blinding was performed for experiments

Muller glia cultures were derived from the retinas of Postnatal Day 11 (P11) rtTa:tetO-Ascl1-ires-GFP mice of both sexes. After retina harvest, the retinas were incubated for 7 min at 37°C in a solution of papain and deoxyribonuclease (DNase) (Worthington, #LK003172). Incubation was followed by trituration of the mixture to dissociate the cells, then to stop the reaction, an equal volume of ovomucoid (Worthington, #LK003182) was added. To isolate the cells, the solution was spun at 4°C at 300g for 10 minutes. Pelleted cells were resuspended in growth medium consisting of Neurobasal medium (Gibco, #10888-022), 10% tet-approved fetal bovine serum (tet-FBS) (Takara/Clontech, #631367), N2 (Gibco, #17502-048), 1 mM l-glutamine (Gibco, #25030-, 1% penicillin-streptomycin (Gibco, #15140-122), and mouse epidermal growth factor (100 ng/ml) (R&D Systems, #2028-EG-200). Cells were plated in a six-well dish at a density of about 2 retinas per 10cm2 well, then incubated at 37°C. The entire volume of medium was changed every 48 hours until confluent or after 7 days. At confluence or after 7 days, cells were removed from the plate using TrypLE (Gibco, #12605-028), spun at 4°C at 300g for 10 minutes, then resuspended in a small volume of growth medium. The entire volume of cell suspension was added to a freezing medium containing 40% tet-FBS, and 10% dimethyl sulfoxide (DMSO), with the remaining 50% being the growth medium from the resuspension. The cells were then frozen for at least 24 hours at -80C before being moved to storage in liquid nitrogen for at least one day. Cells were thawed at room temperature, then spun at 300 RCF for 10 minutes. After thawing, cells were seeded in 500 uL of 10% tet-FBS growth medium in 24 well plates or in 250ul of the same medium in 48 well plates. Cells were grown for 24-48 hours in the growth medium until they appeared healthy.

To harvest cells for sci-Plex nuclei processing, the cells were rinsed either in 24- or 48-well plates with 200ul HBSS (Gibco, #14025-134), and then 200ul of Accutase (Sigma, #A6964-100ML). For experiments performed in 24-well plates, the accutase was incubated on the cells an additional 1 minute at room temperature before being removed. The cells were incubated with any residual accutase at 37°C for 5 minutes. Cells were then resuspended in 200 uL of 1% tet-FBS-reduced growth medium with a wide-bore P200 pipette tip to isolate single cells. The entire volume of the single cell suspension was transferred to a v-bottom 96-well plate on ice. The hashing of glia used a protocol adapted from the original sci-Plex publication(13). Briefly, dissociated cells were centrifuged at 600 x g for 5 minutes in 96-well v-bottom plates. Note that all centrifugation steps, including this one, were performed in a chilled swinging bucket centrifuge (4°C). Media was aspirated, and 200 uL 1 x dPBS (no calcium, no magnesium) was added to each well. The plate was centrifuged at 600x g for 5 minutes and the dPBS was pipetted off. To each well containing cells, 50 uL of CLB+hash solution (45 uL of Cold Lysis Buffer – 10mM Tris/HCl pH 7.4, 10mM NaCl, 3mM MgCl2, 0.1% IGEPAL, 1% (v/v, Sigma-Aldrich, I8896), SuperaseIn RNase Inhibitor (20 U/μL, Ambion, AM2694), 1% (v/v) BSA (20 mg/ml, NEB, B9000S) + 5μL of hash oligo (10 uM, IDT)) was added and then mixed by pipetting up and down for 5-10 strokes. The plate was incubated on ice for 3 minutes, after which, 200 uL of fixation buffer (5% paraformaldehyde (EMS, cat. no. 50-980-493), 1.25x dPBS) was mixed with the nuclei by pipetting up and down for 10 strokes. The cells were incubated with the fixative for 15 minutes on ice during which the hashes were affixed to the nuclei. All nuclei from all wells were then collected into a single 15 or 50 mL conical tube, depending on the size of the experiment, and centrifuged for 5 minutes at 800xg. Pellets were resuspended in 1mL of NSB (Nuclei Buffer + SuperaseIn + BSA – 10mM Tris/HCl pH 7.4, 10mM NaCl, 3mM MgCl2, 1% (v/v) BSA, 1% (v/v) SuperaseIn RNase Inhibitor), and centrifuged again for 5 minutes at 800 x g. One final wash with 1 mL of cold NSB was performed. Upon resuspension, nuclei were transferred to a 1.5 mL LoBind microcentrifuge tube (Eppendorf, Z666491) and counted using a hemocytometer. The 1.5 mL tube was placed within a 15 mL conical and centrifuged in a swinging bucket centrifuge for 5 minutes at 800 x g. The nuclei were resuspended in 500 uL of NSB and flash frozen in liquid nitrogen before being stored at -80C.

For immunohistochemistry, whole eye globes were fixed for 30 minutes in 4% PFA after lens removal. Eyes were then soaked in 30% sucrose overnight and processed for sectioning the following day. For sectioning, retinas were frozen in O.C.T and cryosectioned at 18μm thickness. Slides were washed in PBS two times for 10 minutes and then primary antibodies were applied overnight in 10% normal horse serum (Vector Labs, #S-2000-20), 0.5% Triton-X-10 (Sigma, #T8787-100ML) and PBS. The following day sections were rinsed two times for 10 minutes in PBS and secondary antibodies were applied at a concentration of 1:1000 in PBS for 1 hour. Slides were washed again in PBS and then coverslipped using Fluromount-G (Southern Biotechnology, #0100-01)

A Zeiss LSM880 confocal microscope was used to take images from the in vivo experiments. For quantifications, four images per retina at 20x objective were taken and both eyes were combined for a biological n. Statistical significance between treatments was determined using an unpaired t-test.

Dissociated and GFP+ FACs sorted cells prepared 21 days after NMDA injury were subjected to the standard 10x workflow for Single Cell 3’ v3.1 Dual Index Gene Expression kit (Dual Index Kit TT Set A 96 rxns, 10x Genomics, 1000215).

After hashing, the nuclei were subjected to the original 2-level sci-RNA-seq protocol(14) with minor modifications. Briefly, the hashed nuclei were thawed on ice, subjected to centrifugation at 800 x g for 5 minutes, the liquid was aspirated away, and then the nuclei were incubated on ice for 3 minutes in 500 uL of NSB + 0.2% Triton X-100 (ThermoFisher, A16046.AP). Permeabilized nuclei were centrifuged for 5 minutes at 800 x g, resuspended in 400 uL NSB, and counted using a hemocytometer. The desired number of nuclei were subjected to one final centrifugation step at 800 x g for 5 minutes. They were then resuspended to a concentration of ∼800-3000 nuclei per uL across 3-4 x twin.tec™ 96 Well LoBind PCR Plates (Eppendorf, 0030129512) in NSB. The RT reaction was carried out using an increasing temperature gradient (4°C 2 min, 10°C 2 min, 20°C 2 min, 30°C 2 min, 40°C 2 min, 50°C 2 min, 55°C 15 min), but was otherwise the same as in Cao et. al.

All nuclei were pooled. DAPI (4′,6-diamidino-2-phenylindole, 3 uM final) (Invitrogen, D1306) was used to stain the DNA content of cells so that doublets and debris could be removed by sorting on the DAPI height vs DAPI area, and FSC vs SSC respectively using a FACSAria III cell sorter (BD Biosciences). Into each well of a twin.tec™ 96 Well LoBind PCR Plate, ∼180 nuclei were sorted. This was performed across 4-5 plates depending on the experiment. Each of the wells that the nuclei were sorted into contained 5 uL of EB buffer (Qiagen, 19086), 0.5 uL of 5 x mRNA Second Strand Synthesis buffer (New England Biolabs, E6111L), and 0.25 uL of mRNA Second Strand Synthesis enzyme (New England Biolabs, E6111L). The plates were incubated at 16°C for 3 hours. Tagmentation was performed at 55°C for 5 minutes in 5.75 uL of Tagmentation mix (1.1 uL N7 loaded custom TDE1 enzyme (MacroLab, UC Berkeley) per 632.5 uL 2xTD buffer (20 mM Tris-HCl pH 7.6, 10 mM MgCl₂, 20% v/v Dimethyl Formamide (Sigma-Aldrich, 227056-100ML)). The reaction was stopped with 12 uL of DNA binding buffer (Zymo Research, D4004-1-L). Ampure-based bead purification, PCR, and final library purification were performed as published previously (14). The library was visualized using a D1000 Screen Tape (Agilent Technologies, 5067-5583) and quantified by Qubit using Broad Range DNA reagents (Invitrogen, Q32853).

In-depth protocols can be found on protocols.io under the title. Single cell RNA sequencing library preparation (2-level sci-RNA-seq)