Three-dimensional retinal organoids from mouse pluripotent stem cells mimic in vivo development with enhanced stratification and rod photoreceptor differentiation

Generation and maintenance of ESCs and iPSCs from Nrl-GFP mice

ESC and iPSC clones were derived based on published protocols [31,32]. In brief, ESC clones were isolated from the inner cell mass (ICM) of 3.5-day blastocysts of Nrl-GFP transgenic mice on feeders of mouse embryonic fibroblast (MEF) in the presence of leukemia inhibitory factor (LIF; Millipore, Darmstadt, Germany) and 3i [33] and maintained under the same conditions at 37 °C in a humidified environment of 5% CO₂. iPSC clones were generated by infection of the E14.5 MEFs from Nrl-GFP mice of Dox-inducible lentiviral vectors carrying Oct3/4, Sox2, Klf4, and c-Myc genes individually and then cultured on MEF feeder cells for 3 weeks. Embryonic stem cell–like colonies were picked and transferred onto fresh feeder cells, followed by evaluation of quality by morphology, proliferation rate, and Nanog expression. Nrl-GFP ESCs and iPSCs were maintained on feeder cells in knockout Dulbecco's Modified Eagle Medium (DMEM; Life Technologies, Carlsbad, CA, Catalog number: 11995073) supplemented with 1X Eagle′s minimum essential medium (MEM) non-essential amino acids (Sigma; St. Louis, MO, catalog number: M7145), 1X GlutaMAX (Life Technologies), 1X penicillin-streptomycin (PS; Life Technologies), 1×β-mercaptoethanol (Life Technologies), 2,000 U/ml Leukemia Inhibitory Factor (LIF; Millipore, Darmstadt, Germany), and 15% embryonic stem cell–qualified fetal bovine serum (FBS; Life Technologies). Cells were passaged using TrypLE Express (Life Technologies).

Differentiation of Nrl-GFP ESCs and iPSCs into 3D neural retinas

ESCs and iPSCs (passages 15–25) were differentiated using the SFEBq protocol [18] and after modifications in the culture conditions as follows: (i) The hypoxic condition was applied from D0 to D10; (ii) 36 uncut retinal organoids were cultured in a 100 mm Poly(2-hydroxyethyl methacrylate) (poly-HEMA)-coated (Sigma) dish after D14; (iii) β-mercaptoethanol was freshly added to the media every time; (iv) retinoic acid (RA; Sigma) was added to the culture from D14 to D26; (v) retinal organoids were supplied with DMEM/Nutrient Mixture F-12 (F12) with GlutaMAX (Life Technologies, catalog number: 10565042) supplemented with 2% FBS (Life Technologies), 1X penicillin-streptomycin (PS, Life Technologies), 1X MEM non-essential amino acid (NEAA, Sigma), 1X N2 supplement (Life Technologies), and 1X B27 supplement (Life Technologies); and (vi) 1X β-mercaptoethanol (Life Technologies) was added to all media every time. Then, 72 retinal organoids were transferred at D7 to a 100 mm poly-HEMA-coated Petri dish with 10 ml retinal maturation medium (DMEM/F12 with GlutaMAX with 1X PS). The maturation medium of the retinal cultures was changed at D10, and 1 mM taurine and 500 nM RA were added. At D14, half of the media was changed and the culture density reduced to 36 organoids per 100 mm dish. The cultures were maintained until D26 with half of the media changed every 3 days. At D26, the full media were replaced with DMEM/F12-containing GlutaMAX that was supplemented with 2% FBS, 1X PS, 1X NEAA, 1X N2, and 1X B27. Half of the media was then changed every 2 days. The cultures were incubated at 5% O₂ from D0 to D10 and with 20% O₂ from D10 onward.

Immunohistochemistry

ESCs and iPSCs were plated into four-well Nunc® Lab-Tek® II Chamber Slides (Sigma) at a density of 56×10³ cells per well. After 48 h, the cells were washed once with DPBS (Dulbecco's phosphate-buffered saline (DPBS, 2.6 mM KCl, 1.47 mM KH₂PO₄, 137.93 mM NaCl, 8.06 mM Na₂HPO₄-7H₂O, pH 7.0-7.3; Life Technologies, catalog number: 14190250) and fixed for 30 min using 4% paraformaldehyde (PFA; Electron Microscopy Science, Hatfield, PA). The cells were washed three times with DPBS and incubated for 1 h at room temperature in blocking solution containing DPBS with 10% FBS (Atlanta Biologicals, Flowery Branch, GA) and 0.2% Triton X-100 (Sigma). The cells were then incubated with anti-Nanog antibody (Mouse, 1:200 in blocking solution, ReproCELL, Kanagawa, Japan) overnight at 4 °C, followed by three washes of 5 min each with DPBS. Then 4',6-diamidino-2-phenylindole (DAPI; Life Technologies) and Alexa Fluor 568 anti-mouse antibody (1:250 in blocking solution, Life Technologies) were added, and the cells were incubated at room temperature for 1 h. After washing three times with DPBS (5 min each with gentle shaking), the cells were rinsed once with Ultrapure water (Life Technologies) to remove the sodium ions before the mounting solution was added. Fluorescence images were captured using the EVOS^® XL system (Life Technologies).

For immunohistochemistry (IHC) of the differentiating neural retina, the organoids were fixed in 4% PFA (Electron Microscopy Science) for 1 h and cryoprotected in 15% sucrose for at least 2 h, followed by 30% sucrose overnight, before embedding in optimum cutting temperature (OCT) compound (Sakura Finetek, Torrance, CA). The OCT blocks were sectioned at 10 μm thickness and incubated at room temperature for at least 1 h, before immunostaining or storage at −80 °C. The immunostaining procedures have been recently described [20]. Primary antibodies were used at the following dilutions: laminin (rabbit, 1:50; Sigma), Zonula occludens-1 (ZO-1, rabbit, 1:100; Life Technologies), brain-specific homeobox/POU domain protein 3A (Brn3a, rabbit, 1:200; Santa Cruz, Dallas, TX), calbindin (Calb, rabbit, 1:1,000; Calbiochem, San Diego, CA), S-opsin (Opnsw, goat, 1:200; Santa Cruz), Prospero homeobox 1 (Prox1, rabbit, 1:200; Abcam, Cambridge, MA), Ceh-10 homeodomain-containing homolog (Chx10, sheep, 1:200; Abcam), protein kinase C alpha (PKCα, rabbit, 1:1,000; Sigma), GFP (chick, 1:500; Abcam), rhodopsin (Rho, mouse, 1:1,000; a gift from Dr. Robert Molday, University of British Columbia, Canada), glutamine synthetase (GS, rabbit, 1:200; Abcam), ADP-ribosylation factor-like protein 13B (Arl13b, rabbit, 1:1,000; Proteintech, Rosemont, IL), gamma-tubulin (γ-tubulin, mouse, 1:500, Sigma), and synaptophysin (mouse, 1:200, Abcam). Sections were incubated with the primary antibody at 4 °C overnight. Species-specific secondary antibodies, conjugated with Alexa Fluor 488 or 568, were diluted in antibody dilution buffer (1:250, Life Technologies) and incubated at room temperature for 1.5 h. DAPI (Life Technologies) was used for nuclear staining. For comparison of intensity, images in the same section were captured using identical gain and voltage. Fluorescence images were captured using an LSM 880 confocal microscope (Zeiss, Jena, Germany).

FACS analysis

ESCs and iPSCs were dissociated using TrypLE Express (Life Technologies). The cells were centrifuged at 180 ×g for 5 min and transferred to a T25 flask (Corning, Corning, NY) in 4 ml maintenance media to remove the feeder cells, which adhered to the flask in 20 min. Suspended cells were then fixed and stained with the FIX & PERM® Cell Fixation & Permeabilization Kit (Life Technologies) as suggested by the manufacturer. The cells were stained with anti-Nanog antibody overnight at 4 °C, followed by secondary antibody for 2 h at room temperature. The cells were then resuspended in DPBS containing 1 mM EDTA, filtered by 40 μm cell strainer (BD Bioscience, San Jose, CA), and kept on ice until fluorescence-activated cell sorting (FACS) analysis was performed using a MACSQuant Analyzer (Miltenyi Biotec, San Diego, CA).

For sorting the GFP+ cells, the retinal organoids were dissociated in 0.25% trypsin-EDTA (Life Technologies). The cells were centrifuged at 400 ×g for 5 min, resuspended in DPBS containing 1 mM EDTA, and filtered with a 40 μm cell strainer (BD Bioscience). The samples were kept on ice until sorting with FACSAriaII (BD Bioscience). Cell viability was evaluated by integrity (gated by DAPI) and size (gated by forward scatter, FSC). Gating of GFP+ cells was based on wild-type retinal organoids without GFP at the same differentiation day.

RNA extraction and RT–PCR

Total RNA was purified from tissue homogenates or flow-sorted GFP+ photoreceptors using TriPure isolation reagent (Roche, Basel, Switzerland). RNA quantity was assessed with a NanoDrop 1000 spectrophotometer (Thermo Scientific, Rockville, MD). Reverse transcription (RT) was performed using a SuperScript III Reverse Transcription-PCR kit (Life Technologies) and complementary DNA (cDNA) was generated using oligo(dT) primer. The resulting cDNA was amplified using gene-specific primers (Table 1). PCR products were separated by 1% agarose gel (Lonza, Walkersville, MD) and detected under ultraviolet (UV) illumination (PerkinElmer, Waltham, MA).

Transcriptome analysis

RNA sequence (RNA-seq) profiles of high-quality total RNA (RNA integrity number ≥7.5) from the GFP+ photoreceptors were obtained using the Illumina platform. These paired-end RNA-seq data were compared to reported single-end mouse rod transcriptome [30] as follows. After the Illumina adapters were trimmed, defined in the TruSeq3-PE-2.fa file, by Trimmomatic (version 0.36) [34] from paired-end data, each sample was quantified against Ensembl release 84 [35] cDNA FASTA file by Kallisto (version 0.42.5) [36] via a custom pipeline in Perl (Perl version 5.16.3). The quantifications were then normalized and summarized at the gene level with tximport [37] with the option “countFromAbundance=lengthScaledTPM.” The statistical environment R (version 3.3.1) installed on a 64-bit GNU Red Hat Linux system (Centos 7.2) was used to further analyze the data and to generate heatmaps. The count values from tximport were converted to count per million (CPM) via edgeR [38]. Logarithmically transformed (base 2) CPM values were used for the PCA and clustering. All GFP+ data were averaged for each time point to avoid replicates. The heatmap function has been used previously [20]. Photoreceptor genes data were clustered with the affinity propagation algorithm [39] using the negative distance matrix option as the similarity measure. PCA analysis was performed with custom plotting, using prcomp function in R without scaling and centering the data. To highlight the time matching, the first principal component was plotted against the 11^th component on the x- and y-axes, respectively.

Statistical analysis

All data were expressed as mean ± standard error of the mean (SEM), unless specified. An unpaired two-tailed Student t test was used to compare the mean between two groups. Results with a p value of less than 0.05 were considered statistically significant. At least three independent biologic replicates were performed (n = number of experiments).

Characterization of ESCs and iPSCs from Nrl-GFP mice

Nrl-GFP ESCs and iPSCs were evaluated by the morphology, proliferation rate, and expression of Nanog, which is a critical transcription factor for pluripotency [40]. There was no significant difference between the ESCs and iPSCs in terms of morphology (Figure 1A) and proliferation rate (Figure 1B), both of which were consistent with other publications [41,42]. Nanog expression was detected in the nuclei of all ESC and iPSC colonies (Figure 1C). Flow analysis showed that all ESCs and iPSCs were pluripotent and exhibited Nanog expression (Figure 1D).

Efficient generation of optic vesicles and optic cups from mouse ESCs and iPSCs

The SFEBq protocol was modified to efficiently obtain a more mature neural retina by changing the oxygen level and medium components during differentiation (Figure 2A). Oxygen tension has been demonstrated to impact the viability and cell fate commitment of neurons during early neuronal development in mice and humans [43,44]. Thus, the hypoxic condition was applied from D0 to D10 in the modified protocol. Under normoxia (20% O₂) and hypoxia (5% O₂) conditions, mouse ESCs and iPSCs could be differentiated into embryoid bodies with a neuroepithelium at D4, an optic vesicle at D7, and an optic cup at D10, without significant morphological differences (Figure 2B). To assess the efficiency, differentiating organoids with optic vesicle and optic cup formation were quantified at D7 and D10, respectively. Organoids in hypoxic condition demonstrated significantly higher differentiation into optic vesicles and optic cups (Figure 2C). For ESCs and iPSCs, more than 70% of the organoids formed optic vesicles by D7 (n = 10, range from 50% to 95%), and more than 50% formed optic cups on D10 (n = 10, range from 30% to 85%), compared with 40% at D7 (n = 10, range from 30% to 50%) and less than 20% (n = 10, range from 8% to 30%) at D10 with the original SFEBq method. To avoid bias in counting, 60 organoids at D10 were randomly selected and harvested for RNA. RT–PCR analysis was performed to measure the expression of early eye field transcription factors. As predicted, the embryoid bodies under hypoxic conditions revealed higher expression of Pax6 and Rax, suggesting commitment to the retinal cell fate by a greater number of cells (Figure 2D).

Polarization of the neural retina and genesis of early-born neurons

To investigate whether the neural retinas generated by the modified protocol mimic the in vivo mouse retina in terms of polarity, lamination, and the presence of all cell types, immunostaining was performed using various retinal markers. Apical-basal polarity of the neural retina is critical for the proper functioning of photoreceptors [45]. The neural retinas generated using the modified protocol revealed polarity, with the outer side basal and the inner side apical (Figure 3A).

Ganglion cells and horizontal cells are the two earliest-born cell types in 3D retinas, as evident by the expression of Brn3a and calbindin (Calb), respectively (Figure 3B). As early as D14, ganglion cells could be detected at the basal side of the neural retina, with a substantial increase in the number at D18. However, the number of ganglion cells decreased at D21, which may be due to the lack of axon formation or other secretory factors from the brain in culture [46,47]. Horizontal cells started to be born at D14, and few cells were detected at the middle part of the neural retina. At D18, the horizontal cells migrated to the basal side, and their number greatly expanded at D21, with limited connections among the cells.

The birth of ganglion cells and horizontal cells is followed by the generation of cone photoreceptors and amacrine cells (Figure 4). At D18, cones and amacrine cells were detected with minimal expression of S-opsin (Opnsw) and Prox1, respectively, at the middle part of the neuroblastic layer (NBL). At D26, cone cells were found at the apical side of the NBL, while amacrine cells were mainly located at the basal side. With the development of amacrine cells, the inner plexiform layer (IPL) started to form at this stage. Further maturation of the cone photoreceptors and amacrine cells was indicated by the significant increase in expression of Opnsw and Prox1, respectively. At D32, the cone photoreceptors had typical morphology and were located in the outer nuclear layer (ONL), which is at the apical side of the neural retina. Notably, the outer plexiform layer (OPL) and the IPL were detected at this stage. Most amacrine cells were located at the inner nuclear layer (INL), with some misplaced at the ganglion cell layer (GCL).

The Nrl-GFP ESC and iPSC lines were also differentiated using the original SFEBq protocol for comparison (Appendix 1). At D26, the inner retina neurons were not fully developed, and only the NBL and the GCL could be observed. The morphology of the neural retina in these experiments was similar to that reported by others [23,24,48].

Development of bipolar neurons

During in vivo retinal development, formation of the OPL and the IPL is followed by the development of bipolar cells to connect the two plexiform layers [49]. Chx10 is a marker for proliferative cells or bipolar cells, depending on the differentiation stage. At an early stage of differentiation (D18), proliferative cells were found throughout the NBL, as indicated by Chx10 staining. As the cells started to commit bipolar cell fate, a small portion of Chx10+ cells were located at the basal side of the NBL at D26, while most remained in the NBL, which were proliferative retinal progenitor cells. With further maturation of the neural retina, all cells exited the cell cycle at D32, and Chx10+ cells were detected only at the ONL, indicating the proper location of bipolar cells. PKCα is the marker for ON-bipolar cells, and its expression indicates the maturation of bipolar cells. A substantial expression of PKCα was detected only in the late stage of differentiation, connecting the OPL and the IPL (Figure 5).

Differentiation of rod photoreceptors and Müller glia

Nrl is the earliest marker of post-mitotic rods; Nrl expression is detected at rod birth, followed by Rho expression as the differentiation proceeds [26,29]. RT–PCR analysis of the neural retina showed that Nrl expression began at D15, with a continuous increase during differentiation (Figure 6A). A minimal amount of rhodopsin (Rho) expression was detected at D18 and increased with time as the rod photoreceptors matured (Figure 6A). As GFP is under the control of the Nrl promoter in ESCs and iPSCs, differentiating cells committed to rod cell fate would express GFP in organoids. Flow analysis of the dissociated organoids revealed a progressive increase in GFP+ cells and in the intensity of GFP (Figure 6B). Notably, the GFP+ rod cells constituted more than 70% of the total population at D35, closely mimicking the mature mouse retina in vivo [26] (Figure 6C). At D18, there were few GFP+ cells with weak signal intensity, but more GFP+ cells with higher intensity were detected by D21 (Figure 6B,D). After D25, GFP seemed polarized to the apical side of the neural retina, with D35 having the highest GFP+ cell number and signal intensity (Figure 6D). Thus, the retinal organoids generated by the HIPRO protocol demonstrated longer survival and more GFP+ rods compared to the original SFEBq protocol (Appendix 1).

Most rod photoreceptors and Müller glia are born postnatally in the mouse retina [50]. At D18, rod photoreceptors were evident by the minimal expression of Reep6, a marker for rod photoreceptors [51], at the apical side of the neural retina. Expression of Reep6 increased at D25, indicating the maturation of the rods. In addition, GS, a marker for Müller glia, was mainly located at the basal side, while Reep6 was polarized to the apical side. From D32 and onward, the Müller glia matured further and spanned the entire neural retina, and the outer limiting membrane (OLM) was evident at the apical side of NR, with Reep6 protruding beyond the OLM. With the HIPRO protocol, the organoid cultures were maintained until D35, with an intact structure of the neural retina (Figure 7).

Genesis of photoreceptor cilium and synaptic morphology

Photoreceptor outer segments originate as modified primary cilium [52,53], and limited ciliogenesis has been one of the shortcomings of 3D retinal organoid cultures [18,19,21,54]. Using the HIPRO protocol, we facilitated cilia growth in retinal organoids and detected elongation of the ciliary axoneme at the apical side of the neural retina (Figure 8A). At D25, basal bodies, indicated by immunostaining of γ-tubulin, were polarized to the apical side of the neural retina, suggesting the start of ciliogenesis. At this stage, immunostaining of ADP-ribosylation factor-like protein 13B (Arl13b), a marker of ciliary axoneme, was barely detectable. As the photoreceptors continued to mature, more cells with elongated cilium were observed at D32, as indicated by Arl13b. Colabeling with Rho and Arl13b demonstrated their partial colocalization, with Rho immunostaining extending beyond Arl13b, suggesting the initiation of outer segment biogenesis (Figure 8B).

In addition, we observed the formation of plexiform layers, as indicated by synaptophysin immunostaining. At D26, only the IPL was detected with a trace amount of synaptic vesicles. As the development proceeded, the OPL was detected with significant labeling with synaptophysin. Increased synaptic staining was also detected in the IPL at D32, suggesting further maturation of the INL cell types (Figure 8C).

Gene profiling

After the transcriptome data from the GFP+ rods derived from the D25, D28, and D35 organoids were normalized with the published RNA-seq of the developing mouse rods [30], we applied PCA to the whole data set (Figure 9A). The principal component 1 (PC1) captured more than 30% of all variations in the data and perfectly captured the variance between time points. The PCA plot showed that the D25 and D28 data coprojected on PC1 with the P2 and P4 time points of the mouse rods, whereas D35 falls between P6 and P10 but closer to the P6 developmental stage. A heatmap showing the expression profiles of the photoreceptor genes validated the PCA results and further demonstrated that the D35 rods in the organoid culture were more mature than in the P6 retina (Figure 9B). Notably, the developing rod photoreceptors in the retinal organoids also expressed cone genes, consistent with the proposed concept of the potential origin of mammalian rods from S-cones [30].