Episodic live imaging of cone photoreceptor maturation in GNAT2-EGFP retinal organoids

Cone photoreceptors are critical for color vision and are impaired in retinal diseases such as cone-rod dystrophy, retinitis pigmentosa, Leber congenital amaurosis and retinoblastoma (Mustafi et al., 2009; Xu et al., 2014). Modeling of human cone development and disease in animals is challenged by human-specific cone development features (Singh et al., 2018). Human pluripotent stem cell (hPSC)-derived retinal organoids (ROs) closely recapitulate human retinal development, with similar developmental timeline, cellular composition and laminated structures (Aparicio et al., 2017; Bell et al., 2020). Combined with CRISPR gene editing, ROs provide opportunities to build cone disease models for mechanistic studies and therapeutic screening and are a potential source for retinal cell or tissue transplantation (Gasparini et al., 2019).

hPSC-derived ROs with cell type-specific fluorescent reporters can be used to monitor a cell's normal and disease-related behaviors. Photoreceptor reporter lines have been generated by introducing a CRX-GFP cassette (Kaewkhaw et al., 2015), a RCVRN-eGFP cassette (Guan et al., 2022) or a mouse Crx-mCherry cassette (Gagliardi et al., 2018) into the AAVS1 locus. A rod reporter line was made by replacing the NRL coding sequence with EGFP (Phillips et al., 2018); cone reporter lines were produced by inserting a mouse cone-arrestin (mCar)-GFP cassette (Gasparini et al., 2022) or inserting T2A-mCherry at the C terminus of GNGT2 (Nazlamova et al., 2022); and a retinal ganglion cell (RGC) reporter line was produced by inserting a P2A-tdTomato-P2A-Thy1.2 cassette at the C-terminus of BRN3B (POU4F2) (Sluch et al., 2017). Additional lines reporting multiple cell types include a SIX6-GFP/BRN3B-tdTomato double reporter line that separately labels all retinal cells and RGCs, and a VSX2-Cerulean/BRN3B-EGFP/RCVRN-mCherry triple reporter line that differentially labels retinal progenitor cells, RGCs and photoreceptors (Lam et al., 2020; Wahlin et al., 2021). These lines have been used to investigate retinal morphogenesis, improve organoid differentiation, and purify specific retinal cell types for transcriptome profiling and cell transplantation.

To build a platform with which to study cone development and diseases, we sought to generate a human cone reporter iPSC line in which EGFP is specifically expressed in both immature and mature cones with minimal disruption of normal cone development. As a candidate, we identified guanine nucleotide-binding protein subunit alpha transducin 2 (GNAT2), which encodes the cone-specific α-subunit of transducin, a G-protein that couples visual pigment opsin to the cone phototransduction cascade (Morris et al., 1997; Morris and Fong, 1993). Prior studies demonstrated that Gnat2^−/− mice exhibit complete loss of cone phototransduction without changes in rod phototransduction or in cone or rod morphology (Ronning et al., 2018). In human retina, GNAT2 expression is limited to cones, controlled by a cone-specific promoter, and initially induced following the early cone expression of RXRG and THRB but prior to expression of mature cone markers ARR3, OPN1SW or OPN1LW (Hoshino et al., 2017; Morris et al., 1997; Welby et al., 2017) (Fig. S1A; data from Hoshino et al., 2017). However, unlike RXRG or THRB, GNAT2 does not downregulate in adult cones (Hoshino et al., 2017; Welby et al., 2017). A similar onset order was observed in CRX-GFP⁺ photoreceptors in human ROs, with GNAT2 first detected at day (d)37 (Kaewkhaw et al., 2015). These features suggested that the endogenous GNAT2 promoter linked to a fluorescent protein could serve as an ideal cone-specific reporter.

In principle, cell type-specific fluorescent reporter lines enable live imaging to characterize developmental and disease processes. Episodic imaging (i.e. repeated imaging at defined intervals) of specific RO regions may enable long-term monitoring of individual retinal cells, yet embedding beyond the initial 6 weeks was shown to impair photoreceptor development (Decembrini et al., 2020; Rashidi et al., 2022). Others have immobilized and imaged mature ROs for up to 3 weeks (Achberger et al., 2019), but longer-term imaging has not been demonstrated.

In this study, we produced GNAT2-EGFP cone reporter iPSC lines in which cones are robustly, specifically and innocuously labeled with EGFP. We also established RO hydrogel immobilization and episodic live imaging methods that enable long-term assessment of individual EGFP⁺ cone morphological changes, inner segment development and mitochondria localization. This EGFP-GNAT2 cone reporter line, combined with the immobilization and imaging techniques, provides a useful tool to study cone development and disease.

To assess the suitability of a GNAT2 cone reporter, we compared GNAT2 expression to that of other potential cone markers in human fetal, adult and RO single-cell RNA-sequencing (scRNA-seq) datasets. In a combined human fetal retina, adult retina and early-stage RO dataset produced via 3′ end-counting, GNAT2 was mainly detected in cones from adult retina (Fig. S1B-E; data from Lu et al., 2020). In contrast, our deep, full-length scRNA-seq analysis of fetal retinal progenitor cells and photoreceptors showed robust and specific GNAT2 expression in the majority of fetal cones from post-conception week 13 to 19 (Fig. S1F,G; data from Shayler et al., 2023 preprint), consistent with the onset timing in bulk RNA-seq (Fig. S1A). This sensitivity and specificity compared favorably with that of other potential cone markers, such as THRB, RXRG, GNGT2, ARR3, OPN1SW or OPN1LW, which were either expressed in both cones and rods (THRB, RXRG, GNGT2; although the THRB TRβ2 isoform was cone specific) or expressed in only a subset of cones (THRB only in L/M cones, ARR3 mainly in more mature cones, and OPN1SW and OPN1LW mainly in S and L/M cones, respectively) (Fig. S1H-O; data from Shayler et al., 2023 preprint). These patterns were preserved in human ROs, in which GNAT2 was first detected at d37 in CRX-GFP positive cells (Kaewkhaw et al., 2015). These analyses supported the potential utility of a GNAT2-driven cone reporter.

To generate a GNAT2 cone reporter line, we used CRISPR/Cas9-mediated homologous recombination to insert an EGFP-P2A cassette at the N-terminus of GNAT2 in the human WTC11 iPSC-derived line WTC11-mTagRFPT-LMNB1 (Allen Institute for Cell Science) (Fig. 1A,B). The N-terminal position of the EGFP-P2A cassette is predicted to enable GNAT2 translation with a single proline residue added to the N-terminus (Fig. 1B). WTC11-mTagRFPT-LMNB1 cells and derivatives express an mTagRFPT-LMNB1 fusion protein that labels nuclear envelopes and enables live imaging of nuclei together with other fluorescent protein markers.

Briefly, a homology donor plasmid was constructed by inserting the EGFP-P2A coding sequence between left and right homology arms (LHA and RHA, respectively) containing human GNAT2 genomic sequences 882 bp upstream and 854 bp downstream of the translation start codon (Fig. 1C). The sgRNA spanned the intended insertion site, eliminating the need to introduce a silent mutation on the homology donor plasmid (Fig. 1A). No antibiotic resistance marker was included in the donor vector to enable scarless editing. Following electroporation of the donor plasmid and a plasmid co-expressing GNAT2 sgRNA and Cas9-T2A-Puro (PX459) (Ran et al., 2013), cells were selected with puromycin, single-cell cloned, and screened by PCR using location-specific and insert-flanking primer pairs (Fig. 1A). PCR with location-specific primers flanking the LHA showed integration of the EGFP-P2A cassette with correct orientation in five of 48 clones tested (Fig. 1D), while insert-flanking primers distinguished two mono-allelic from three bi-allelic knock-in clones (Fig. 1E and data not shown). The two mono-allelic clones carried mutations on the non-knock-in alleles, whereas the three bi-allelic clones carried no unintended mutations at the knock-in junctions. Partial sequencing of EGFP-P2A-GNAT2 (hereafter referred to as GNAT2-EGFP) clone (C)-41 revealed no unintended mutations at any of the top five predicted gRNA off-target sites (Fig. S2A,B, Table S1), and karyotyping revealed no chromosomal abnormalities (Fig. S2C). All subsequent experiments were carried out using this clone.

We next evaluated the ability of the GNAT2-EGFP iPSCs to make ROs with cone-specific EGFP expression. We used a modified version of the protocol by Kuwahara et al. (2015) to improve RO consistency. Initial culture medium was supplemented with small-molecule inhibitors of WNT signaling (IWR1) and TGF-β superfamily signaling (SB431542 and LDN193189) for 6 days, followed by addition of BMP-4 to induce anterior neural ectoderm and eye field specification (Aparicio et al. 2023). As the parental WTC11-mTagRFPT-LMNB1 had poor early survival, the starting cell number was increased from 12,000 to 48,000, the retinal pigment epithelium induction reversal was optimally timed from d23 to d28, and long-term maintenance with retinoic acid was begun at d72 (Fig. 2A), which promotes photoreceptor maturation and long-term survival (Kelley et al., 1994; Zhong et al., 2014). During the first 30 days, ROs increased in size and adopted a structure consistent with that of nascent neural retina (Fig. 2B). Subsequently, ROs continued to grow, and a brush border likely representing photoreceptor inner and/or outer segments was evident by ∼d140 and remained visible until the latest analysis on d245 (Fig. 2B). In mature ROs, EGFP⁺ cells formed uneven patches occupying the outermost layer (Fig. 2B).

To evaluate the specificity of GNAT2-EGFP expression, we immunostained d105 RO sections from two independent differentiations with cone markers ARR3 and RXRγ and assessed their colocalization with EGFP. At this age, most EGFP⁺ cells had elongated cell bodies occupying the outermost layer (Fig. 2C,D). Among 487 cells examined for ARR3 and EGFP, 475 (97.5%) co-expressed both, ten expressed ARR3 only, and two expressed EGFP only (Fig. 2C,E). Among 561 cells examined for RXRγ and EGFP, 520 (92.7%) co-expressed both, 41 expressed RXRγ only, and none were EGFP⁺ and RXRγ⁻ (Fig. 2D,F, unpaired two-tailed Student's t-test, P=0.019, for % EGFP⁺,ARR3⁺ versus EGFP⁺,RXRγ⁺). Similarly, at d86, all EGFP⁺ cells expressed RXRγ and photoreceptor marker CRX (Fig. S3A-D). Moreover, 35.2% of RXRγ⁺ cells were EGFP⁻ at d86 (Fig. S3C) compared to 7.3% at d105 (Fig. 2F, unpaired two-tailed Student's t-test, P=0.02). These results are consistent with the sequential onset of RXRγ, GNAT2-EGFP and ARR3 expression during cone maturation (Hoshino et al., 2017; Welby et al., 2017) (Fig. S1A).

Additionally, immunostaining of GNAT2-EGFP ROs for retinal progenitor cell markers PAX6 and CHX10 (VSX2), RGC marker BRN3B (POU4F2), bipolar cell marker CHX10, amacrine cell marker AP2α (TFAP2A), rod marker NRL, Müller glia marker CRALBP (RLBP1) and horizontal cell marker calbindin, at various ages, showed the presence of all retinal cell types and no co-expression of EGFP (Fig. S3E-L). Although most retinal cells resided in their expected RO layers, some misplaced EGFP⁺,RXRγ⁺ and EGFP⁺,ARR3⁺ cones resided deep within the ROs, as reported by others (Capowski et al., 2019; Wahlin et al., 2017). Taken together, these results confirm the specific and robust labeling of cones in GNAT2-EGFP ROs.

To evaluate the effects of the EGFP-P2A tag on GNAT2 expression, western blot analysis of ROs derived from two independent differentiations of GNAT2-EGFP C-41 and parental WTC-mTagRFPT-LMNB1 iPSCs was performed. The results showed that GNAT2-EGFP ROs had a slight, but not statistically significant, decrease in GNAT2 expression and no change in GNAT2 size (Fig. S3M,N).

We next assessed the feasibility of high-resolution live imaging of GNAT2-EGFP ROs to monitor the development of cone cells over time. We performed episodic live confocal imaging on GNAT2-EGFP ROs and captured z-stack images at different maturation stages, with cones represented by cytoplasmic+nuclear EGFP and nuclear membranes represented by mTagRFPT-LMNB1 (Fig. 3A,B). The cytoplasmic+nuclear EGFP distribution is consistent with P2A-mediated separation of EGFP from GNAT2 and the ability of EGFP to diffuse across the nuclear envelope (Seibel et al., 2007; Wei et al., 2003). From d62 to d111, EGFP⁺ cell bodies elongated and gradually populated the outermost organoid layer. By d147, they developed more mature cone morphology, with the appearance of inner segments and pedicles similar to those in older ROs. By d195, cones retained similar morphology, and cell bodies were often displaced away from the outermost layer as previously described (Gasparini et al., 2022).

Unexpectedly, the live-imaged mTagRFPT-LMNB1 nuclear envelope signal was much weaker in EGFP⁺ cones than in EGFP⁻ cells (Fig. 3B; Fig. S4A). Similarly, immunostaining of d160 GNAT2-EGFP ROs and d105 ROs differentiated from the parental WTC-mTagRFPT-LMNB1 iPSCs showed decreased LMNB1 in EGFP⁺ and RXRγ⁺ cone nuclei (Fig. S4B,C). The decreased cone LMNB1 expression is unlikely to relate to the heterozygous mTagRFPT knock-in allele as the wild-type LMNB1 allele remains. These analyses confirm that live imaging can be used to display the distribution of EGFP⁺ cones and mTagRFPT⁺ nuclei.

Live confocal imaging also enabled assessment of organelle development. At d245, imaging with MitoView dye, which accumulates in mitochondria and allows live detection, showed that mitochondria coalesced at the mature cones' inner segment ellipsoid bodies (Fig. 3C,D), as confirmed by immunostaining with the mitochondrial marker (TOMM20) (Fig. 3E).

To evaluate the maturation of individual cone cells, ROs were embedded in HyStem-C^TM hydrogel on Millicell^TM cell culture inserts, and the same organoid regions were repeatedly imaged during long-term culture (Fig. 4A). HyStem-C^TM is based on hyaluronic acid and collagen crosslinked polymers and was chosen because its rigidity can be tuned, and because hyaluronic acid and collagen are major components of retinal extra cellular matrix and vitreous humor (Achberger et al., 2019; Hemshekhar et al., 2016; Tram and Swindle-Reilly, 2018) and deemed likely to be biocompatible. Pilot experiments showed that 0.25-1% HyStem-C^TM supported long-term immobilization, whereas organoids dislodged from 2% and 4% hydrogel. ROs embedded in 1% hydrogel at d121 or older could be cultured in this manner for at least 120 more days. The embedded ROs displayed continuous morphological changes with the appearance of photoreceptor inner segment protrusions, suggesting that embedding does not restrict cone growth and maturation (Fig. 4B,C). Indeed, in organoids embedded on d139, mature cones remained stable and individual cells were identifiable until at least d267 (18.3 weeks) (Fig. 4D).

Episodic live confocal imaging of immobilized ROs allowed us to capture developmental features of the same cells over time. As an example, we used EGFP intensity and cell shape-based segmentation on the acquired three-dimensional (3D) z-stack images to define the volumetric change in 24 cone inner segments in three ROs immobilized at d125 and imaged between d126 and d153 (Fig. 4E). The individual cells were within a defined region in which spatial relationships were maintained. We observed the inner segments enlarged from a mean of 193 µm³ to 523 µm³ (P<0.0001) (Fig. 4F), with an average increment of 12.2 µm³ per day. However, the initial inner segment size and rate of individual cone inner segment growth varied depending on the organoid and/or organoid region, ranging from a maximum rate of 27.14 µm³ per day for one cone from RO #3 to a slight decline of 0.3 µm³ per day for another from RO #2 (Fig. 4G-I). Taken together, episodic live imaging of hydrogel-embedded GNAT2-EGFP ROs enabled long-term evaluation of cone development, such as inner segment morphogenesis and formation of mitochondria-rich ellipsoid bodies, demonstrating the versatility of this reporter system.

In this study, we generated a cone-specific GNAT2-EGFP iPSC reporter line and demonstrated its utility for tracking individual cone development in long-term live-embedded ROs. By tagging GNAT2 with scarless CRISPR insertion and placing the EGFP-P2A at the N-terminus, we aimed to create a reporter line that faithfully recapitulates cone development with minimal effect on GNAT2 expression or protein sequence. The GNAT2-EGFP iPSC line generates all retinal cell types and robustly labels GNAT2⁺ cones throughout RO differentiation, with EGFP detected in cone precursor cell bodies at d62 and subsequently in maturing cone axon terminals, nuclei and inner segments. Indeed, EGFP was detected as early as d34, when RO cone precursors are initially born (Kaewkhaw et al., 2015). However, as cone-specific genes might transiently express before cell fate is determined, lineage tracing is needed to determine whether the earliest GNAT2-EGFP expression is strictly limited to the cone photoreceptor lineage.

Live imaging revealed that maturing cone precursors develop inner segments and extend pedicles towards the outer plexiform layer between ∼d120 and ∼d150, coinciding with a time of rapid maturation and high glycolytic activity in the photoreceptor layer (Browne et al., 2017). This timeline is slightly delayed compared to that of foveal cones in the human fetal retina, which elongate and develop inner segments and synaptic pedicles between fetal week 14 and 18 (d98-d126) (Hendrickson and Zhang, 2019), but is similar to that of the later-born peripheral cones that RO cone transcriptomes best resemble (O'Hara-Wright and Gonzalez-Cordero, 2020).

Recently, two other cone reporter human iPSC lines have been described – one generated by piggyBac-mediated insertion of GFP under the control of mCar promoter (Gasparini et al., 2022) and the other generated by inserting a T2A-mCherry cassette into the GNGT2 locus (Nazlamova et al., 2022). Cone-arrestin is first expressed at a later stage of cone maturation than GNAT2, limiting its ability to label immature cones (Hoshino et al., 2017; Welby et al., 2017) (Fig. S1A). Moreover, in mature ROs, only ∼80% of cells labeled with the mCar-GFP reporter were ARR3⁺ whereas >95% were recoverin⁺ (Gasparini et al., 2022), potentially reflecting some non-cone expression. Our CRISPR-based editing of GNAT2 is similar to the Nazlamova et al. (2022) editing of GNGT2, but used an antibiotic-free scarless knock-in approach so the inserted sequence contains only EGFP-P2A. Additionally, 3′ end counting and our full-length scRNA-seq analyses indicate that GNAT2 RNA is highly cone specific, whereas GNGT2 RNA is expressed in both cones and rods and has no discernable difference in cone- versus rod-specific splicing or exon usage (Fig. S1E,H,N,O) (Lu et al., 2020; Shayler et al., 2023 preprint). Nevertheless, transcript expression need not directly correlate with protein expression, as evidenced by the cone-specific GNGT2 protein in developing and adult human retinae (Ong et al., 1997) and the cone-specific mCherry protein in GNGT2-T2A-mCherry ROs (Nazlamova et al., 2022), suggesting that post-transcriptional mechanisms enforce cone-specific translation of GNGT2 and the linked mCherry in GNGT2-T2A-mCherry ROs, despite similar cone and rod GNGT2 RNA levels.

Furthermore, we established a protocol with which to track individual cone development. Live embedding in HyStem-C^TM hydrogel enabled long-term immobilized RO cultures and monitoring of cone development with episodic live confocal imaging (i.e. repeated imaging at defined intervals). Immobilizing the ROs in thin layers of hydrogel allows nutrient diffusion from all sides and allows RO to grow and mature with minimal restriction (Fig. 4A). Cells were tracked for at least 120 days starting at >d120. The hydrogel maintained structural integrity and transparency during the entire period, allowing repeated imaging of the same RO regions. However, live embedding might be more disruptive in younger <d80 ROs owing to the ongoing rapid organoid growth and morphological changes. The biocompatibility of HyStem-C^TM hydrogel was evident from the fairly consistent growth of cone inner segments and may relate to its derivation from hyaluronic acid and collagen, which are major components of vitreous humor and retinal extracellular matrix (Achberger et al., 2019; Hemshekhar et al., 2016; Tram and Swindle-Reilly, 2018). Combining this GNAT2-EGFP cone reporter with further CRISPR editing and live imaging provides a powerful tool to study cone development and disease.

WTC-mTagRFPT-LMNB1 human iPSC line was obtained from Allen Institute for Cell Science (https://www.allencell.org/cell-catalog.html), cell line ID AICS-0034 cl.62. This line was produced as described in Roberts et al. (2017). The WTC-mTagRFPT-LMNB1 iPSCs and the subsequent GNAT2-EGFP iPSCs were cultured in feeder-free conditions in mTeSR Plus Medium (Stem Cell Technologies, 100-0276) on Matrigel (Corning, 354277)-coated 35 mm dishes. Cells were seeded at 25,000 cells per dish, fed daily and passaged every 5 days at ∼70% confluency. When passaged, cells were washed with DPBS (Corning, 21-031-CV), and colonies were gently lifted by 2 min incubation in ReLeSR (Stem Cell Technologies, 05872) at 37°C. After neutralization with mTeSR Plus Medium, cells were centrifuged for 3 min at 300 g and seeded at desired densities. WTC-mTagRFPT-LMNB1 iPSCs and GNAT2-EGFP iPSCs were authenticated and confirmed to be mycoplasma negative.

The 882 bp LHA and 854 bp RHA were PCR amplified from WTC-mTagRFPT-LMNB1 iPSC genomic DNA with CloneAmp HiFi PCR Premix (Takara, 639298). The LHA, EGFP-P2A and RHA were cloned into pUC118 backbone using In-Fusion Snap Assembly (Takara, 638949). The sgRNA targeting the GNAT2 ATG^Met start codon (AAGACGGCAAATATGGGAAG) was identified using the CRISPick online sgRNA design tool (https://portals.broadinstitute.org/gppx/crispick/public) and cloned into the PX459 sgRNA Cas9-T2A-Puro expression plasmid (Addgene #62988) (Ran et al., 2013) according to the accompanying Zhang Laboratory Target Sequence Cloning Protocol (https://media.addgene.org/data/plasmids/62/62988/62988-attachment_KsK1asO9w4owD8K6wp8.pdf). The resulting plasmids were sequenced to confirm correct assembly. A full list of cloning primers can be found in Table S2.

WTC-mTagRFPT-LMNB1 human iPSCs were dissociated into single cells with 5 min incubation in Accutase (Life Technologies, A1110501) at 37°C. Then, 200,000 cells were electroporated with 500 ng PX459 and 1000 ng homology donor plasmid in 10 µl electroporation buffer R using a Neon Transfection System (Invitrogen) with one pulse of 1400 V and a pulse width of 20 ms. After electroporation, cells were placed in mTeSR Plus Medium supplemented with CloneR (Stem Cell Technologies, 05888) for 36 h and then selected in 250 ng/ml puromycin for 36 h. After a recovery phase of 4 days, colonies were dissociated into single cells with Accutase and seeded into 96-well plates at an average density of 0.5 cells per well to ensure that colonies are derived from single cells. Colonies were expanded and genotyped using two primer pairs: an LHA flanking primer pair that produces a 990 bp band from clones with correct integration of the insert, and an insert flanking primer pair that produces a single 1131 bp band from bi-allelic knock-in clones, the 1131 bp band and a 351 bp from mono-allelic knock-in clones, or a single band at 351 bp from wild-type clones. The 1131 bp knock-in band and 351 bp wild-type allele band were gel purified (Qiagen, 28604) and sequenced to check for potential mutations introduced during editing. To check for potential off-target mutations, we PCR amplified ∼1 kb regions spanning the top five program predicted off-target sites (IDT CRISPR-Cas9 gRNA checker) from the edited clone-41 and the unedited WTC-mTagRFPT-LMNB1 iPSCs and aligned the sequences. All PCRs utilized CloneAmp HiFi PCR Premix according to the manufacturer's instructions. The predicted off-target sequences and amplifying primers are provided in Tables S1 and S2. The GNAT2-EGFP iPSC line was karyotyped by the Department of Pathology and Laboratory Medicine, Children's Hospital Los Angeles.

Retinal organoids were generated following steps modified from a previous protocol (Kuwahara, et al., 2015; Aparicio et al. 2023). Briefly, iPSCs were dissociated on d0 with Accutase and resuspended in Aggrewell media (Stem Cell Technologies, 05893) supplemented with 20 µM Y-27632 (Cayman Chemical, 10005583) and ‘ISL’ cocktail, which consists of 3 µM IWR1 (Cayman Chemical, 13659), 10 µM SB431542 (Cayman Chemical, 13031), and 0.1 µM LDN193189 (Sigma-Aldrich, SML0559). Cells were plated in round-bottom 96-well plates at 48,000 cells per well in 200 µl medium to allow aggregate formation. From d1 to d5, medium was changed to gfCDM [45% Hams F12 (Thermo Fisher Scientific, 11765047), 45% IMDM (Thermo Fisher Scientific, 12440046), 10% KSR (Life Technologies, 10828028), 1× Chemically Defined Lipid (Gibco, 11905-031), 0.5× Glutamax (Life Technologies, 35050061), 450 µM monothioglycerol (Sigma-Aldrich, M6145) and 1× Penicillin-Streptomycin (Corning, 30-002-CI)] supplemented with ISL. On d6, gfCDM medium was supplemented with 0.75 nM BMP-4 (R&D Systems, 314-BP-050), followed by 1/2 medium change on d9 and d12, 3/4 medium change on d15, and full medium change on d19 and d21 with fresh gfCDM only without BMP-4. From d23 to d27, medium was changed to RPE induction medium that consisted of Dulbecco's modified Eagle medium (DMEM)/F12 (Thermo Fisher Scientific, 21331020), 1× N2 supplement (Thermo Fisher Scientific, 17502048), 1× Glutamax, 1× Penicillin-Streptomycin, 3 µM CHIR99021 (Cayman Chemical, 13122) and 5 µM SU5402 (Cayman Chemical, 13182). Starting at d28, medium was changed to RDM3S-KZ, which consisted of DMEM/F12, 10% fetal bovine serum (FBS; Omega Scientific, FB-01), 1× Glutamax, 1× N2 supplement, 1× Penicillin-Streptomycin and 0.5× fungizone (Omega Scientific, FG-70). Taurine (Sigma-Aldrich, T8691; 0.1 mM) was added from d30 onward. On d30, ROs were transferred to a 48-well cell culture plate pre-coated with HEMA (Sigma-Aldrich, P3932-25G). From d37 to d42, medium was transitioned to RO maintenance medium (MM) adapted from Zhong et al., (2014): 2/3 RDM3S-KZ+1/3 MM on d37, 1/3 RDM3S-KZ+2/3 MM on d40, and all MM on d42. MM consisted of an equal volume of DMEM (VWR, 54000-305) and DMEM/F12, supplemented with 1× B27 supplement (Thermo Fisher Scientific, 12587010), 1× NEAA, 1× Penicillin-Streptomycin, 1× fungizone, 10% FBS, 0.1 mM taurine and 1× Glutamax. ROs were subsequently cultured in MM with 1 µM retinoic acid (Sigma-Aldrich, R2625) from d72 to d100 and 0.5 µM retinoic acid from d100 onward. Differences from the protocol of Aparicio et al. (2023) included the following: (1) cultures were initiated with 48,000 cells; (2) only BMP-4 was added on d6, no IWR1; (3) induction reversal was initiated on d23 for 5 days; and (4) RA was first added on d72.

Samples comprised of four to six ROs were fixed in 4% paraformaldehyde for 12 min, washed with DPBS three times, incubated in 30% sucrose solution overnight at 4°C, embedded in OCT compound and cryo-sectioned into 20 µm sections. For immunostaining, slides were washed with TBS, blocked for 1 h at room temperature (2.5% horse serum, 2.5% donkey serum, 2.5% human serum, 1% bovine serum albumin, 0.1% Triton X-100 and 0.05% Tween 20 in 1× TBS), incubated in primary antibodies at 4°C overnight, followed by TBS wash, secondary antibody and 4′,6-diamidino-2-phenylindole (DAPI) incubation at room temperature for 1 h, washing and mounting with Mowiol with anti-fade. Samples were imaged on Zeiss LSM710 or Leica Stellaris 5 confocal microscopes and processed using Fiji ImageJ's Cell Counter plugin. ROs from three independent differentiation experiments were collected at different ages, and multiple sections from at least two ROs from each differentiation were analyzed. Primary antibodies are listed in Table S3.

Samples comprised of eight to ten ROs at d70 were dissociated with 10 U/ml Papain (Worthington Biochemical, LK003176) and 100 U/ml DNaseI (Qiagen, 79254) in Earle's balanced salt solution at 37°C for 30 min, with periodic trituration using a P1000 pipette to aid dissociation. Dissociated ROs were washed with PBS, lysed with RIPA buffer (Cell Signaling Technology, 9806S) plus a protease inhibitor cocktail (Sigma-Aldrich, 5892970001) for 1 h on ice, and centrifuged at 15,000 g at 4°C for 10 min. Supernatants were collected, and proteins were quantified via bicinchoninic acid (BCA) assay. Lysates containing 50 µg total protein were mixed with 4× sample buffer (Invitrogen, NP0007), boiled at 95°C for 5 min, separated on SDS-polyacrylamide gel (Invitrogen, NP0335BOX) and transferred to a PVDF membrane (Sigma-Aldrich, GE10600023). After blocking with 5% milk and 0.05% Tween 20 in 1× TBS, the membrane was sequentially incubated with anti-GNAT2 primary antibody (Thermo Fisher Scientific, PA5-22340) and horseradish peroxidase (HRP)-conjugated anti-rabbit IgG (Cell Signaling Technology, 7074P2). Protein bands were detected using SuperSignal™ Ultimate Sensitivity Chemiluminescent Substrate (Thermo Fisher Scientific, A38555). After imaging, the membrane was stripped with stripping buffer (Thermo Fisher Scientific, 46430) for 15 min, re-blocked with milk and incubated with anti-β-actin primary (Cell Signaling Technology, 4967S) and HRP-conjugated secondary antibodies (for antibody dilutions, see Table S3). Fiji ImageJ was used to quantify protein bands, and GraphPad Prism was used to perform unpaired two-tailed Student’s t-test and to generate the graph.

ROs were subjected to live episodic imaging using a Zeiss LSM780 NLO inverted confocal microscope, with fitted temperature control and a CO₂ chamber that kept ROs at 37°C and 5% CO₂ during imaging. A Zeiss C-Achroplan 32×/0.85 W Korr M27 lens (working distance 1.1 mm) was used to accommodate the long distance between the lens and ROs embedded in cell culture inserts. Non-immobilized ROs were submerged in RO medium and imaged in a Lab-Tek eight-well chambered coverglass (Thermo Fisher Scientific, 12-565-338). For mitochondria live imaging, ROs were incubated with 1× SPY555 vital DNA dye (Spirochrome, SC201) overnight and MitoView 650 (Biotium, 70075) at 200 nM for 30 min prior to imaging. For RO immobilization, individual ROs were live embedded in 100 µl 1% HyStem-C^TM hydrogel (Advanced Biomatrix, GS312) on a Millicell^TM 12 mm cell culture insert with 0.4 µm hydrophilic PTFE membrane (Sigma-Aldrich, PICM01250). The cell culture insert was then submerged in RO medium in a 24-well plate, and medium was changed following the same protocol as for non-immobilized ROs. For live confocal imaging, the rim of the cell culture insert was marked at the 12 and 3 o'clock positions to indicate orientation, and inserts were placed in Cellvis 24-well coverglass-bottom plates. Ubiquitous autofluorescent debris on the PTFE membrane was used as points of reference for the regions of interest. EGFP was excited at 488 nm and collected between 491 nm and 553 nm, mTagRFPT and/or SPY555 were excited at 561 nm and collected between 571 nm and 642 nm, and MitoView 650 was excited at 633 nm and collected between 642 nm and 735 nm. Collected 3D z-stack images were processed using Imaris (Version 8.4.2, Oxford Instruments), and individual cone cells were segmented using Imaris Surface function. The inner segment was manually segmented from the cell body at the thinnest connecting point, and the volume was recorded in Imaris. GraphPad Prism was used for statistical analysis with repeated measures one-way ANOVA and to generate graphs.

The authors thank Narine Harutyunyan and Andrew Salas of the Stem Cell Analytics Core Facility and G. Esteban Fernandez of the Cellular Imaging Core Facility of the Saban Research Institute of Children's Hospital Los Angeles for assistance. The authors also thank the Aaron Nagiel laboratory for sharing antibodies and Sumitha Prameela Bharathan for advice on immunofluorescence. Microscopy and image processing were performed at the Translational Biomedical Imaging Laboratory supported by University of Southern California (USC) and the Saban Research Institute.