The study adhered to the tenets of the declaration of Helsinki and was approved by the institutional review board of LV Prasad Eye Institute, Hyderabad. Donor retinas from cadavers as well as from patients due to conditions such as staphyloma and open globe injury were used to establish a primary mixed retinal cell culture system. The cadaveric donor eyes were collected (within 24 hours of death) in sterile moist glass bottles from Ramayamma International Eye Bank, LV Prasad Eye Institute, and washed with sterile phosphate buffered saline (PBS) containing 2X concentrations of penicillin and streptomycin. The retinal tissues were removed using a pair of sterile forceps from the donor eye by making a posterior cut and washed gently with 1X PBS. The enucleated eyeballs from patients were collected after obtaining written informed consent and immediately transported to the lab on ice. The retina was removed from these eyeballs similar to the cadaveric eyes. The retinal tissues collected from either of the sources were washed gently with 1X PBS to remove the RPE and choroid pigments. The tissue was then chopped into small pieces using a sterile surgical blade. The chopped tissues were again washed with 1X PBS and treated with 1X trypsin EDTA (0.25%) for a period of 15–20 min at 37°C. Trypsin activity was arrested by adding complete DMEM (DMEM + 10% FBS + 1% penicillin–streptomycin) and centrifuged at 1,000 rpm for 3 min. The dissociated pieces were collected and resuspended in 2 ml PBS and gently triturated with a P1000 pipette tip to further obtain a suspension of cells. The suspensions of the cells were then passed through a 70-micron-size cell strainer to remove undigested pieces of tissues, if any. The cells were collected after the centrifugation and resuspended in DMEM containing 10% serum and 1% antibiotics. The cells were seeded in a sterile tissue culture grade T-75 mm flask and kept undisturbed for 7 days under standard cell culture conditions followed by changing medium every 3 days. 1ng/mL of granulocyte macrophage cytokine stimulating factor GM-CSF [REC. HUMAN GM-CSF 10 UG BIOSOURCE (TM), PHC2015], was added to the culture until the first medium change.

Immunofluorescence was done to characterize the cells in MRC. Briefly, the cells (approximately 5,000 cells/ml) were seeded on a sterile glass coverslip and allowed to attain 70–80% confluency. The cells were fixed with 4% formaldehyde in PBS for 10 min at room temperature. The cells were washed with 1X PBS and permeabilized with 0.5% Triton X-100 in PBS for 10 min. This was followed by incubation with blocking buffer consisting of 2% BSA in PBS for 1 h at room temperature. The primary antibodies were diluted with blocking buffer and added to the cells for overnight incubation at 4°C. The primary antibodies were used for identification of cells in the MRCs including mouse anti–ionized calcium-binding adaptor molecule 1 (for microglia, Abcam, Catalog No. ab178680), rabbit anti–glial fibrillary acidic protein (for astrocytes, Catalog No. Dako, Z0344), rabbit anti nestin (for neuronal progenitor cells, Millipore, Catalog No. ABD 69), rabbit beta-III tubulin (β-III tubulin; for neurons, Abcam, Catalog No. ab18207), and rabbit anti glutamine synthetase (GS; for Müller glia, Abcam, Catalog No. ab176562). The cells were washed thrice with 1X PBS followed by incubation for 45 min at room temperature with secondary antibodies (diluted in blocking buffer) Alexa Fluor 488 conjugated anti rabbit (Life Tech, Catalog No. A11008), Alexa Fluor 594 conjugated anti rabbit (Life Tech, Catalog No. A11012), and Alexa Fluor 594 conjugated anti mouse (Life Tech, Catalog No. A11005). The cells were washed thrice with 1X PBS, mounted with SlowFade Gold Antifade containing DAPI (Life Technologies, Ref. S36939), and scanned using an EVOS fluorescent microscope.

The cells from earlier passages (P1 and P2) were used for the experiment. The cell viability was estimated using an Alamar blue dye–based assay using different concentrations of cobalt chloride (CoCl₂; Sigma Aldrich, Catalog No. C-8661-25G). Briefly, 2,000 cells were seeded on a 96-well plate and allowed to attain 70–80% confluency. Prior to the exposure of stress, the complete DMEM was replaced with serum free medium and incubated for 6 h. The cells were then treated with 100 and 150 μM CoCl₂ for inducing hypoxia for a period of 24 h in serum free medium. Alamar blue reagent (10 μl) (Life Technologies, Catalog No. DAL1025) was added onto the cells containing 100 μl of medium and kept for incubation at standard cell culture conditions for 3 h. DMEM with Alamar blue served as blank. The absorbance of the medium was measured; the blank values were subtracted from cells’ absorbance value; and the percentage of viability and significance were calculated.

Once the optimization for hypoxia drug concentration was achieved, 15,000 cells were seeded on a glass coverslip and allowed to grow for 70–80% confluency. CoCl₂ (150 μM) was used for the treatment for 24 h in serum-deprived medium. The cells deprived of serum but not exposed to stress for the same duration were used as control.

Gene expression by PCR was done for characterizing the cultured retinal cells as well as to measure the expression of genes under hypoxia. In brief, the total RNA was extracted from the retinal cells before and after the stress induction by TRIzol method. Total RNA was reverse transcribed into cDNA using a Verso cDNA synthesis Kit (ThermoFisher Scientific, Catalog No. AB1453B) according to the manufacturer’s protocol. The primer sequences used for conventional PCRs are given in the Supplementary Table S1. In order to quantify the average mRNA expression in the entire MRC population, we performed quantitative real-time (qRT) PCR on an Applied Biosystems 7900 HT system for a total reaction volume of 20 μl. Reaction mixture (10 μ) included iTaq^TM Universal SYBR^® Green Supermix (BIO-RAD, Catalog No. 172-5121), 200- nM of primer, and cDNA. The relative measure of the concentration of the target gene (Ct) was calculated by using software SDS 2.4. Analysis of gene expression changes was done using the 2^–ΔΔCt method. Statistical analyses were performed using 2^–ΔΔCt ± SEM in three technical and biological replicates. The housekeeping gene β-actin was used as a normalizing control. The primer sequences used for qRT PCRs are given in the Supplementary Table S2.

In order to assess the glial hyperreactivity under hypoxia, we plan to compare the protein expression in GFAP-positive and IBA1-positive cells in control and stressed conditions. In order to achieve this, we performed immunocytochemistry and quantitative protein imaging. Large-scale imaging was performed using lasers with excitation at 405, 488, and 594 nm for DAPI, Alexa 488, and Alexa 594 with a Leica SP8 laser scanning confocal microscope with a 40X dry objective. In order to quantify the protein level in a large section of MRC, we acquired a panorama using the mosaicking technique for a field of view (1.8 × 1.8 mm) containing 10 × 10 square sections (each section of dimension 180 × 180 μm) with approximately 20% overlap. A hundred sections were stitched to obtain the spatial protein profiling for a large section by Leica LAS X Software. In order to show the representative images of GFAP and IBA1 expression under no stress and hypoxia, 3D imaging was performed through acquiring Z-stack images along the z-axis (total Z height = ∼12 μm, Z-stack thickness between each slice = 0.5 μm) with a 63× oil immersed objective. The fluorescence intensity corresponding to various regions of interest was acquired using Leica LAS X software. In order to quantify the glial reactivity in MRC, we further created a 3D surface plot of GFAP and IBA1 expression from the panorama images using ImageJ software.

In order to perform cytosolic calcium imaging, cells were loaded with 2 μM Fluo-4 (Molecular Probes, Life Technologies, Grand Island, NY, United States) for 30 min in Hank’s Balanced Salt Solution (HBSS) (Invitrogen, Life Technologies, Grand Island, NY, United States). The cells were then washed thrice with HBSS followed by fluorescence imaging (excitation: 488 nm) at 37°C. Time-lapse movies were acquired every 10 s, for 10 min, and raw data were analyzed with MATLAB (The MathWorks, Natick, MA, United States). An image segmentation algorithm, based on principal component analysis, was optimized for automated segmentation of cells present in MRC. The maximum amplitude of Fluo-4 intensity and Ca²⁺ spike count were computed for control and hypoxia and were represented through box plots. In order to perform the correction for the photobleaching effect, we used second-order polynomial fitting and estimation of coefficients.

The time course of Ca²⁺ transients obtained for 600 s was analyzed using MATLAB. The analysis was performed for all the segmented cells obtained from the live imaging video for respective conditions. In order to quantify the activity level in MRC, we obtained the raster plot via peak identification from the time course of Fluo-4 intensity. Although the cell size has not been accounted for to make any size-based correction, the Fluo-4 intensity was normalized with respect to basal-level Fluo-4 intensity for each cell (Swain et al., 2018). We performed k-means clustering based on Ca²⁺ spike count and Ca²⁺_max (maximum calcium amplitude) (number of clusters k = 4 for control condition). The automation in classification of cells yielded different types of cells, black and cyan cells with lower activity, green cells with moderate activity and red cells with highest activity. Here, we report the cells with higher amplitude and spike count as the hyperactive cells and the cells with moderate amplitude and spike count as moderately active cells. Furthermore, we used the boundaries obtained from the control condition as the reference to classify the calcium transients under the hypoxic condition. The relative percentages of four subpopulations were represented using stack bar plots. The data set corresponding to control and hypoxia was tested for normality using the Jarque–Bera test. As the majority of the data set were not normally distributed, we have used the Kruskal–Wallis test to study the effects of hypoxia on a mixed retinal population. Statistical tests were performed at the significance level of 0.05. In box plots, the results were presented in terms of median, interquartile range, and whiskers 10–90%. We also performed the Kruskal–Wallis test to check whether the percentages of cells corresponding to different clusters (having high, moderate, low, and no activity) are significantly different in the stressed condition compared to the no-stress condition.

Data sampling: In order to select data from multiple videos in an unbiased manner, 60% of the cells were randomly chosen from the MRC population (population size = 160, sampling repeated five times, size of each random sample = 90 cells) for clustering (Swain et al., 2018). All bar graphs were plotted to present mean ± SEM. The schematic representation of data analysis is given in Supplementary Figure S9.

The cells were heterogeneous in nature, which was evident from their morphology. The dissociated retinal cell cultures started to adhere after 3–4 days and became confluent within 3–4 weeks in culture. Most importantly, the cells were having both neuronal and glial morphology with a network of processes (Supplementary Figures S1, S2). Immunofluorescent staining and PCR characterization were done for glial as well as neural populations of the cultured cells. Immunofluorescence of the cells in MRC clearly showed positive staining for neuronal progenitor marker nestin (Figure 1A), Müller glial marker GS (Figure 1B), GFAP for astrocytes (Figure 1C), microglia marker IBA1 (Figure 1D), and β-III tubulin (Figure 1E) for the neuronal population. Likewise, PCR-based characterization of the cells also confirmed the expression of genes specific to glial cells, neural progenitor cells, and mature neurons in the dissociated retinal culture (Figure 2).

Further, to ascertain the robustness of this culture system and reproducibility of the cell types obtained, we have calculated the percentage of each cell type with respect to the total number of DAPI stained cells in each culture obtained from different cadaver retina samples. Figure 3 shows the stack bar representation of subpopulation percentages for each cell type in MRC corresponding to samples from four cadaver donor retinas. The result shows that the percentages of each cell type for the samples derived from different eyes are not significantly different (p > 0.05) (Figure 3).

The cells were actively dividing until passage four, and the earlier passage of the cells (P1 and P2) was used for the experiment. Prior to the experiment, a PCR-based characterization was done for the cell-specific genes to ensure all major glial cell types and mature neurons in the culture (Supplementary Figure S3). The serum-deprived cells were exposed to different concentrations of CoCl₂ for a time period of 24 h, and cell viability was measured by the Alamar blue method. We have used a concentration range from 100 to 250 μM of CoCl₂. The cell viability of controls was always maintained as 100%. We have found a concentration-dependent reduction of cell viability under hypoxic treatment. The results showed a significant reduction in cell viability when the cells were treated with 150 μM (62.33 ± 1.71, p < 0.05) (Figure 4).

In order to evaluate the changes in intracellular calcium level under hypoxia, we first characterized the basal-level cytosolic Ca²⁺ spiking in MRC. Figures 5A,B show the time-lapse imaging of cytosolic Ca²⁺ in MRC under the no-stress condition. (A movie file shows these details, Supplementary Movie S1). Next, we performed the time-lapse imaging of intracellular Ca²⁺ for hypoxia (Figures 5C,D; an additional movie file shows these details, Supplementary Movie S2). The spatial mapping of single-cell Fluo-4 intensity in the MRC population showed the Ca²⁺ spiking in a single cell under various conditions. The heat map representation of time-lapse Ca²⁺ responses provided prominent visualization of Ca²⁺ spiking (Figures 5B,D, and Supplementary Figure S4). The time-lapse videos were further processed by an image segmentation algorithm to acquire data, and the schematic diagram of the data acquisition process is shown in Supplementary Figure S5. The Ca²⁺ spiking pattern under no stress and hypoxia (Figures 6A,B) indicated that the intracellular Ca²⁺ oscillates at variable frequencies for different cells in the MRC population. Note that each of the cells in the whole population did not show Ca²⁺ spiking, indicating that there were some cells having relatively less activity.

In order to observe the Ca²⁺ spiking pattern in a large MRC population, we have plotted the raster plot for 160 cells (Figures 6C,D). The raster plot showed that there is an increase in Ca²⁺ spike count in the case of hypoxia compared to the no-stress condition. This was further validated using box plot representation (Figures 6E,F) showing the comparison of Ca²⁺ spike count and Ca²⁺_max between the no-stress condition and hypoxia. The box plot representation clearly indicated that hypoxia induces a significant increase in Ca²⁺ spike count in the MRC population (p < 0.05).

In order to obtain a subpopulation profiling of the calcium spiking pattern present in the MRC population, we implemented the k-means clustering (Supplementary Figure S6) under the no-stress condition (Figure 7A). Since Ca²⁺ spike count and Ca²⁺_max can be used to characterize the neuronal activity, we chose these two features to perform the clustering of calcium spiking over time. The result showed that the cells can be grouped into various categories, hyperactive cells (high spiking, high amplitude >6 spikes in 10 min), cells with moderate activity (moderate amplitude, moderate spiking, 1–6 spikes per 10 min, Ca²⁺_max > 0.5), and cells with lower activity (low amplitude, moderate spiking, 0–6 spikes per 10 min, Ca²⁺_max < 0.5). Using these boundary constraints corresponding to two features for each subpopulation of MRC under the no-stress condition, we performed the classification of the Ca²⁺ spiking under hypoxia (Figure 7B). Further, we plotted the stack bars representing the relative percentages of each category (Figure 7C). The percentage of hyperactive cells and cells with moderate activity were found to be higher in hypoxia compared to the no-stress condition (p < 0.05) (Figure 7D). Moreover, the percentage of low active cells was significantly lower in the case of hypoxia compared to control (p < 0.05).

Further clustering of Ca²⁺ spiking of MRC under normal conditions obtained from four donor retinas was performed. Figure 8A shows the stack bars representing subpopulation profiles of Ca²⁺ spiking corresponding to cells from each donor retina. This analysis showed that the percentages of each subtype are not significantly different across patients (p > 0.05) (Figure 8B).

Cells exposed to 150 μM CoCl₂ in serum-deprived medium to induce hypoxia and control cells were harvested after 24 h, and RNA was extracted. Real-time PCR was performed for three sets of heterogeneous cell cultures derived from three different retina sources. The expression of representative genes from different pathways known to be involved in DR pathogenesis including hypoxia signaling (HIF1-α, NERF2, and OXR1) inflammation (IL-1β, IL-8, and C3), angiogenesis (CXCR4 and VEGF), and apoptosis (BAX and Caspase 3) was measured. (Out of 11 genes measured, the expression of six genes was found to be significantly upregulated in hypoxia; Figure 9). These include genes such as HIF1-α, which was found to be significantly upregulated by 2.28 ± 0.37-fold under hypoxia (p < 0.05). Likewise, the genes involved in oxidative stress response such as OXR1 and NERF2 were upregulated under hypoxia (OXR1: 2.56 ± 0.53, p < 0.05; NERF2: 1.7 ± 0.4, p < 0.05). Further, the angiogenic genes such as VEGF and CXCR4 were upregulated 3.48 ± 0.8-fold, p < 0.05, and 6.89 ± 1.02-fold, p < 0.05, respectively. The expression of IL-1β was found to be 15.3 ± 2.5, p < 0.05, in hypoxia treated cells, even though the expression of other inflammatory genes such as C3, IL-8 was found to be higher in hypoxic treatment, while this increase was not found to be significant (C3: 1.53 ± 0.05, p > 0.05; IL-8: 1.7 ± 0.4, p > 0.05). Likewise, the apoptotic markers Caspase 3 and BAX showed an increased expression under hypoxia (Caspase 3: 2.26 ± 0.63, p > 0.05; BAX: 1.41 ± 0.3, p > 0.05).

Next, we hypothesized that the expression of cell type–specific protein is increased in the MRC population when subjected to hypoxia. Since significant spatial heterogeneity was observed for various proteins in MRC, the protein expression was quantified through large-scale imaging using a confocal microscope. To examine the hypoxic injury on microglia and astrocytes, we analyzed IBA1 and GFAP expression from the panorama images. Figures 10A,C show the representative 3D images of IBA1- and GFAP-positive cells chosen from MRC under hypoxic injury. In order to assess the glial reactivity, we constructed a 3D surface plot corresponding to spatial profiling of IBA1 and GFAP under control and stress conditions (Figures 10B,D).

The spatial pattern shows the differential expression of IBA1 and GFAP under stress compared to the control condition, indicating the activation of microglia and gliosis, respectively, under injury. Next, we performed a quantitative analysis of protein expression in a large number of cells under each condition through box plots (Figures 10E,F). The result suggests that hypoxia induces a significant increase in IBA1 expression (2-fold) and GFAP (1.7-fold) (p < 0.05). In addition to this, we also evaluated the protein expression of GS under the hypoxic condition, however, there was no significant difference (p > 0.05, Kruskal–Wallis test) between the control and hypoxic conditions (Supplementary Figure S7). A double staining of GFAP and GS was also performed in retinal cells under control and hypoxic conditions. This identified a clear categorization of GFAP- and GS-positive cells in the culture (Supplementary Figure S8), and upon treatment, the GFAP level was found to be elevated in the cells, and there was no change in the expression of GS.