2.1.

Animals

The Institutional Animal Care and Use Committee at the University of Colorado Anschutz Medical Campus approved all procedures and protocols. Pregnant Sprague-Dawley rats were purchased from Charles River Laboratories (Wilmington, Massachusetts). Animals were fed ad libitum and exposed to day–night cycles alternatively every 12 h. Litters were delivered in room air (RA) and randomly assigned to HO chambers (${\mathrm{FI}}_{\mathrm{O}2}=0.70$) or RA (${\mathrm{FI}}_{\mathrm{O}2}=0.21$) at day 1 of life ($n=10/\text{group}$). HO chambers were maintained at Denver’s altitude (1600 m; barometric pressure, 630 mm Hg; inspired oxygen tension, 122 mm Hg). HO exposure was continuous with brief interruptions for animal care ($<10\min /\text{day}$). Oxygen concentration was regulated using a ProOx device (Reming Bioinstruments). Eye tissue from pups in HO and normoxia conditions was harvested at P7 and P14. Half of the eyes were used for retinal flat mounts, and the other half were used for 3-D quantification methods. Animals were killed with intraperitoneal injections of pentobarbital sodium ($0.3\mathrm{mg}/\mathrm{g}$ body wt., Fort Dodge Animal Health, Fort Dodge, Indiana).

2.2.

Retina and Entire Eye Isolation from Sprague-Dawley Rats

Eyes harvested at P7 and P14 were enucleated using curved forceps and immediately immersed in 2% (weight/volume) paraformaldehyde for fixation (90 min). Eyes were transferred to phosphate-buffered saline (PBS), where the cornea and lens were removed under a dissecting microscope. Using fine-tipped forceps, sclera and choroid vasculature were separated from retinas and discarded [Fig. 1(a)]. Retinas were examined for tears or folding and included in the study based on overall integrity, measured by the absence of folding, tearing, and other structural abnormalities. Following dissection, intact retinas were transferred to 2-ml tubes (2 to $3\text{retinas}/\text{tube}$) of PBS with Triton X-100 (PBS-T, 0.1%) for staining.

Fig. 1

Flat mount and intact retina sample preparation: (a) Gross eye anatomy and (b) work flow of sample preparation. The cornea and lens are removed first, followed by the sclera and choroid. The retina is either cut four times to produce the clover-like structure used in retinal flat mounts (top row) or kept intact then cleared for use with DSLM imaging (bottom row).

2.3.

Two-Dimensional Retinal Immunofluorescence Sample Preparation

Tubes containing retinas in PBS-T were placed on an orbital rocker at room temperature for 1 to 4 h. PBS-T was drained and replaced with a PBS-T primary label [Biotinylated Griffonia lectin (IB4) Vector #B-1205] solution at 1:100. Tubes were kept on a rocker for 48 to 72 h at room temperature and then washed four times for 1 h with PBS-T. Retinas were immersed in PBS-T secondary label (Alexa-488 conjugated streptavidin, Jackson Immunoresearch #016-540-084) and kept wrapped in foil on a rocker for 24 h. Tubes were drained and washed with PBS-T four times for 1 h.

2.4.

Two-Dimensional Retinal Flat Mount Preparation

Precleaned microscope slides were prepared by adhering two secure seal spacers to each slide face, allowing two flat mounts per slide. A sample spoon was used to transfer intact stained retinas from 2-ml tubes to microscope slides under a dissecting microscope. To prepare flat mounts, retinas were oriented photoreceptor side up in a drop of PBS on the slides. Using fine-tipped scissors, four even interval incisions were made from the peripheral retina toward the optic nerve. A fine-tipped pipette was used to remove PBS from the retina, flattening the retina and revealing a clover-like structure [Fig. 1(b)]. No. 1 cover slips were placed on the secure seal spacers and adhered with Hydromount mounting media (National Diagnostics). Slides were stored in light-tight box at 4°C until imaging.

2.5.

Passive Optical Clearing of Isolated Retina and Intact Eyes

To render tissue optically transparent, retinas harvested from pups were incubated in 4% acrylamide and 0.25% photoinitiator, 2,2′-azobis[2-(2-imidazolin-2-yl) propane] dihydrochloride (VA-044, Wako Chemicals) monomer solution overnight at 4°C. Infused samples were then degassed in nitrogen for 5 min and incubated at 37°C for 2 h to establish hydrogel polymerization. In addition, the polymerized acrylamide hydrogel was cross-linked to endogenous proteins, such as those expressed on the surface of vascular endothelial cells.^29,30 Retinas were removed from hydrogel solution and washed in PBS. Retinas were then incubated in 4% sodium dodecyl sulfate (SDS) solution for 5 days at 37°C with gentle shaking. Retinas were then washed 5 to 8 times over the course of 8 h in PBS to remove excess SDS from sample tissue. Biotinylated GS I4B lectin (Vector Labs Catalog #B-1205) was used to label endothelial cells in retinal vasculature. Retinas were then washed in PBS to remove excess GS I4B lectin and incubated in streptavidin for 2 days. The secondary labeling solution contained the same amount of serum and Triton-X 100 with a 1:250 dilution of Alexa-488 conjugated streptavidin (Jackson ImmunoResearch Labs, catalog #016-540-084). Eyes were then washed in PBS to remove excess secondary label and incubated in refractive index matching solution (RIMS) index matching media until cleared. RIMS imaging media contained 40 g of Histodenz (Sigma D2158) in 30 ml of 0.02 M phosphate buffer with 0.1% Triton-X 100.

2.6.

Passive Optical Clearing-Treated Sample Mounting for Minimal Deformation

Eyes were rendered optically transparent through a 24-h incubation in RIMS at room temperature. Each eye was then stitched with a small suture at the optic nerve. This stitch was adhered to a $200\text{-}\mu \mathrm{l}$ pipette tip using adhesive (Loctite). Once eyes were mounted to pipette tips, they were resubmerged in fresh RIMS.

2.7.

Two-Dimensional Fluorescent Imaging of Retinal Flat Mount Samples

Entire retinal flat mounts were imaged by epifluorescence microscopy (Nikon Eclipse Ti-E) with an excitation wavelength of 488 nm. To obtain pictures of complete retinal flat mounts, the focal plane containing the primary plexus was imaged in nine segments using the $4\times$ air objective and Cy3 filter cube with an in-plane pixel size $1.5\mu \mathrm{m}$. Segments were assembled into a complete picture with a 10% overlap using NIS-Elements Ar image stitching software (Nikon). Additionally, $20\times$ images of each retina (in-plane pixel size $0.3\mu \mathrm{m}$) were taken at the midline between the retinal periphery and the optic nerve for quantification using Angiotool. One $20\times$ image was taken per flat mount petal, per retina.

Deconvolution epifluorescence imaging of retinal flat mounts was performed on a home-built deconvolution epifluorescence microscope. The microscope consists of a four-color LED excitation unit (Lumencor Spectra-X), microscope base (Olympus IX71), quad-band excitation and emission filter set (Semrock LED-DA/FI/TR-Cy5-4X-A), $40\times$ numerical aperture (NA) 1.3 oil microscope objective (Olympus UPLFLN $40\times \mathrm{O}$), automated XY stage (Mad City Labs Microdrive), automated objective piezo (Mad City Labs F-200), and a sCMOS camera (Hamamatsu C11440-22CU). The entire retinal flat mount was imaged using multiple areas with 30% overlap. Each image stack had an in-plane pixel size of $0.1625\mu \mathrm{m}$ and an axial step size of $0.5\mu \mathrm{m}$ over $100\mu \mathrm{m}$. Raw images were deconvolved on-the-fly using custom GPU software.⁴⁰ After deconvolution, individual image tiles were reassembled using FIJI.^41,42

2.8.

Light-Sheet Imaging of Passive Optical Clearing-Treated Retinas

Stitched retinas were mounted on a home-built cleared tissue-specific DSLM (Fig. 2).⁴³ The excitation arm consisted of a multicolor laser source (Coherent OBIS 488 nm, 532-nm diode laser, and Coherent OBIS 640; Thorlabs RGB46HA), fiber collimator (Thorlabs TC06FC-633), pair of galvanometer mirrors (Thorlabs GVSM002), scan lens (Thorlabs CLS-SL), tube lens (Nikon MXA22018), electrotunable lens (ETL-1 – Optotune EL-10-30), and $5\times$ NA 0.14 33.5-mm WD excitation objective (Edmund Optics 59-876). The excitation arm was used to create the sheet and displace the beam along the optical axis of the detection arm. The galvanometer mirrors and camera triggering were controlled by an Arduino Uno unit (Sparkfun DEV-11021) with three PowerShield DAC boards (Visgence, Inc.) while the laser intensity was controlled through analog input to the Coherent OBIS controller by an Arduino Uno unit (Sparkfun DEV-11021) with three PowerShield DAC boards (Visgence, Inc.).

Fig. 2

Light-sheet optical layout. Excluding sample motion and rotation ($XYZ\theta$), there are 4 deg of freedom within the system. The thinnest area of the exciting Gaussian light sheet was laterally translated using ETL-1 (black lines with arrows). ETL-1 was placed at a telecentric position to the back aperture of the excitation objective by physically translating the optic along the optical axis between the fiber coupler and the galvanometer mirrors until this position was determined. The physical location of the light sheet was altered in-plane or axially (blue line with arrows) using the 2-D scanning unit. The unit consisted of the scanning mirrors, scan lens ($f_{\text{scan}}=70\mathrm{mm}$), and tube lens ($f_{\text{tube}}=200\mathrm{mm}$). The position of the detection plane was altered using a $4f$ relay system containing ETL-2.

A 3-D printed imaging chamber, with glued #1 quartz coverslips (SPI Supplies 01019T-AB), was mounted to a manual rotation stage (Thorlabs PR01) and centered at the focal length of both the excitation and detection objectives (DOs), so the full range of excitation beam scanning was available ($\sim 6\mathrm{mm}$). An automated micrometer-driven $XYZ$ stage controlled using OpenStage hardware and software was utilized for sample positioning and tiling.⁴⁴ This stage positioned a sample rotation stage/sample mount consisting of an Arduino Uno controlled stepper motor (Sparkfun DEV-11021, Sparkfun ROB-09238) and rigid mounted $200\text{-}\mu \mathrm{l}$ pipette tip within the imaging chamber.

The detection arm accommodated a $4\times$ NA 0.13 17.2-mm WD (Nikon MRH00041) or $10\times$ NA 0.28 33.5-mm WD (Edmund Optics 59-877) DO using a nonrotating translatable objective mount (Thorlabs SM1NR1). A tube lens (Nikon MXA22018), adjustable periscope mirror (Thorlabs BB1-E02), and electrotunable lens (ETL-2 – Optotune EL-16-40-TC) were placed at the first focal point of a $4f$ system (Thorlabs AC254-300-A). An adjustable periscope mirror (Thorlabs BB1-E02), automated filter wheel (Thorlabs FW102C), and sCMOS camera (Hamamatsu C11440-22CU) completed the detection optics. ETL-2 was placed in a $4f$ configuration to allow the focus to be swept without changing telecentricity. To generate images the scanning mirror corresponding to the focal plane (fast-axis) was swept once per acquisition. The second scanning mirror positioned the light sheet along the detection axis. ETL-2 positioned the focal plane of camera at the axial location of the light sheet. ETL-1 optimized the alignment of each laser source, enabling fine tuning of the excitation arm focus within the sample. These values were user determined at multiple image planes throughout the sample. The full table of axial mirror, ETL-1, and ETL-2 values were then built using linear interpolation. This technique allowed the sample to stay immobile during volumetric imaging.

For all images in this work, fluorescence was collected with either the $4\times$ NA 0.13 or $10\times$ NA 0.28 DO. This yielded in-plane pixel sizes of 1.625 or $0.650\mu \mathrm{m}$ and axial step sizes of 3.5 or $1\mu \mathrm{m}$, respectively. A manually adjustable pinhole was utilized to limit the NA of the excitation objective to adjust the Rayleigh length to best match the imaging area. This yielded a Gaussian width of $\sim 20\mu \mathrm{m}$ for the $4\times$ DO and $10\mu \mathrm{m}$ for the $10x$ DO. Volumetric images of varying total sizes at orientations parallel or perpendicular to the optic nerve were acquired for each retina. The orientation that best visualized the vasculature was chosen for image analysis.

2.9.

Two-Dimensional Fluorescent Image Quantification

All 20× images were imported into Angiotool for automated network tracing.¹⁶ Analysis parameters were visually modified for accurate skeletonization. Controls were set to a low threshold of 15, high threshold of 255, and vessel diameter of 15 and to fill holes at 2000. Output metrics included vessel percentage area, total number of junctions, junction density, total vessel length, average vessel length, total number of endpoints, and mean lacunarity. Statistical analysis was performed in Graphpad PRISM using an ANOVA test.

Additionally, the binary image produced by Angiotool was analyzed using 2-D Sholl analysis in FIJI.^39,41 The number of intersections as a function of radius was normalized by the area of the circle. The resulting Sholl curves were then integrated using trapezoidal integration with unit spacing in MATLAB^®. Statistical analysis was performed in MATLAB^® using a Kruskal–Wallis test.

2.10.

Two-Dimensional Fluorescent Optically Sectioned Image Quantification

Optically sectioned images of retina flat mounts were analyzed using FIJI and Vaa3D.^32,36 Images were filtered using a multiscale adaptive enhancement filter.³⁰ Semiautomated network tracing was carried out on filtered images using APP-2.0.³¹ The resulting network trace was used to mask the retinal vasculature into a binary image. The binary image was analyzed using 2-D Sholl analysis for each image plane in FIJI.^39,41 The number of intersections as a function of radius was normalized by the area of the circle for each focal plane. The resulting Sholl curves were then integrated using trapezoidal integration with unit spacing in MATLAB^®. Statistical analysis was performed in MATLAB^® using a Kruskal–Wallis test.

2.11.

Three-Dimensional Fluorescent Image Network Quantification Pipeline

Volumetric images of intact retinas were analyzed in Vaa3D.³⁶ Raw images were filtered using a multiscale adaptive enhancement filter.³⁴ Semiautomated network tracing was carried out on filtered images using APP-2.0.³⁵ The resulting network trace was used to mask the retinal vasculature into a binary image. The binary image was analyzed using 3-D Sholl analysis in FIJI.^39,41 The number of intersections as a function of radius was normalized by the surface area of the sphere. The resulting Sholl curves were then integrated using trapezoidal integration with unit spacing in MATLAB^®. Statistical analysis was performed in MATLAB^® using a Kruskal–Wallis test.

3.1.

Superficial Vascular Plexus was Disturbed at P7 in Continuous Hyperoxia Exposure Model

P7 RA control retinas displayed dense and consistent vascular distributions with narrow avascular zones at the retinal periphery [Fig. 3(a)]. In contrast, P7 HO flat mounts showed simplified vessel branching and decreased vessel concentration throughout the primary plexus, with large avascular zones at the retinal periphery [Fig. 3(b)].

Fig. 3

Representative tiled and segmented epifluorescence images of P7 retinal flat mounts. (a) Control retinal flat mounts show normal vascular development and (b) HO retinal flat mounts show simplified and less expansive vasculature (shaded red area and red asterisk). Pixel size of $1.5\mu \mathrm{m}$ and total image size of $7.5\mathrm{mm}\times 7.5\mathrm{mm}$. Raw images segmented using automatically determined global threshold and Otsu algorithm in FIJI.

To better visualize the 3-D structure of the developing rat retinal vasculature, we imaged intact retinas using DSLM (Fig. 4, Videos 1 and 2). This imaging method offered two advantages over traditional flat mount preparations. The native structure of the eye was better preserved because the retina was not physically altered, and volumetric imaging proceeded rapidly, at $\sim 100$ to 500 ms per image plane. For Figs. 4(a) and 4(b), the image acquisition required roughly 5 min. For the multiple overlapping images that comprise Figs. 4(c) and 4(d), image acquisition required roughly 30 min. In comparison, acquisition of a similar volume in a retinal flat mount using a wide-field deconvolution microscope or a confocal microscope required nearly 10 h.

Fig. 4

Three-dimensional renderings of DSLM imaging of intact retinas. (a) Low-magnification 3-D rendering of P14 rat retinas. In-plane pixel size of $1.625\mu \mathrm{m}$ and axial step of $3.5\mu \mathrm{m}$. Total imaging volume $3.3\mathrm{mm}\times 3.3\mathrm{mm}\times 5\mathrm{mm}$. (Video 1, MP4, 2.5 MB [URL: http://dx.doi.org/10.1117/1.JBO.22.7.076011.1]) (b) Cross-sectional image of (a). In-plane pixel size of $1.625\mu \mathrm{m}$ and axial step of $3.5\mu \mathrm{m}$. Total imaging volume $3.3\mathrm{mm}\times 3.3\mathrm{mm}\times 5\mathrm{mm}$. (c) High-resolution 3-D image of intact P7 retina. In-plane pixel size of $0.65\mu \mathrm{m}$ and axial step of $1\mu \mathrm{m}$. Total imaging volume $5\mathrm{mm}\times 3\mathrm{mm}\times 3\mathrm{mm}$. (Video 2, MP4, 3.7 MB [URL: http://dx.doi.org/10.1117/1.JBO.22.7.076011.2]) (d) Zoomed area showing intact primary plexus (left), vertical sprouts (middle), secondary plexus (right), and choroid vasculature (far right). In-plane pixel size of $0.65\mu \mathrm{m}$ and axial step of $1\mu \mathrm{m}$. Total imaging volume $1.3\mathrm{mm}\times 1.3\mathrm{mm}\times 3.5\mathrm{mm}$.

Individual image planes contained within volumetric images showed the presence of the choroid layer of vasculature, as well as the primary plexus and the secondary plexus found at the deepest portion of the retina. Connections were observed between the two plexuses in control retinas, indicating the presence of vertical sprouts [Fig. 5(a)]. In the HO retinas, the strongest fluorescent signal was found at the site of the primary plexus. This appeared as dense fluorescence, suggesting an abnormal vascular distribution [Fig. 5(b)]. Additionally, the choroids of the HO retinas were consistently deformed compared to those of the control retinas [Fig. 5(b)].

Fig. 5

Representative individual image planes from volumetric DSLM images of P7 intact retinas. (a) P7 control retinas show normal development of the (1) primary plexus, (2) vertical sprouts, (3) secondary vascular plexus, and (4) smooth choroid. (b) P7 HO retinas show a chaotic (1) primary plexus, lack of vertical sprouts and secondary plexus, and (4) distorted choroid. Individual image planes isolated from full axial image stacks with a pixel size of $0.65\mu \mathrm{m}$.

3.2.

Secondary Vascular Plexus Remained Disorganized in P14 Retinas

Vascular development at P14 was more robust compared with P7 in both control and HO retinas. Relative to control P14 retinas [Fig. 6(a)], P14 HO retinas demonstrated reduced vascular complexity surrounding the optic nerve and larger avascular zones at the retinal periphery [Fig. 6(b)]. P14 DSLM images also showed more robust vasculature compared with P7 images. Control P14 retinas demonstrated more vertical sprouts and more organized primary and secondary plexuses [Fig. 7(a)]. HO P14 retinas demonstrated a strong fluorescent signal in the primary plexus, but they appeared disorganized compared to control retinas. [Fig. 7(b)].

Fig. 6

Representative tiled and segmented epifluorescence images of P14 retinal flat mounts. (a) P14 control retinal flat mounts show normal vascular development and (b) P14 HO retinal flat mounts show some decrease in primary plexus development, particularly surrounding the optic nerve (shaded red area and red asterisks). Pixel size of $1.5\mu \mathrm{m}$ and total image size of $7.5\mathrm{mm}\times 7.5\mathrm{mm}$. Raw images segmented using automatically determined global threshold and Otsu algorithm in FIJI.

Fig. 7

Representative individual image planes from volumetric DSLM images of intact P14 retinas. (a) P14 control retinas show a fully developed (1) primary and (3) secondary vascular plexus, as well as (2) vertical sprouts with a (4) smooth choroid. (b) P14 HO retinas show clear staining for a developed (1) primary plexus. However, there is diffuse staining, indicating disorganized (2) vertical sprouts, lack of a secondary plexus, and (4) distorted choroid. Individual image planes isolated from full axial image stacks with a pixel size of $0.65\mu \mathrm{m}$.

To verify these findings, we performed optical sectioning of retinal flat mounts utilizing a high NA objective, epifluorescence, and computational deconvolution. Figure 8 shows an axial montage of the vascular network isolated from a large-scale epifluorescence image ($\sim 900$ image areas per retina) of control and HO retinas at P7. Vertical sprouts and secondary plexuses were missing from HO retinas [red arrows note sprouts in Fig. 8(a)].

Fig. 8

Representative cross-sections from deconvolved epifluorescence images of P7 retinal flat mounts. (a) P7 control retinas show the beginning of a highly organized network sprouting from the primary plexus. Red arrows mark vertical sprouts that are found in multiple axial planes in P7 control retinas. (b) P7 HO retinas show little evidence of vertical sprouts. No continuous vertical sprouts are found in P7 HO retinas in this area. This result holds throughout the entire retina.

3.3.

Quantification of Vascular Complexity Using Traditional Network Tracing

We used Angiotool software to quantify the differences in the primary plexus between control and continuous HO exposed retinas at the two time points used in this study. Angiotool measures morphometric and spatial parameters of vascularity through network tracing [Fig. 9(a)]. We utilized the total vessel length, junction density, and mean lacunarity tools for assessment of the $20\times$ images (Figs. 2 and 5).¹⁶ We observed a significant decrease in total vessel length and junction density between control and HO P7 retinas [Figs. 9(b) and 9(c)]. This lack of vascular complexity appeared to resolve by P14, with no statistical difference between the two study groups at that time point. Inversely, we saw an increase in avascular space measured by mean lacunarity at P7 in HO relative to controls [Fig. 9(d)]. Again, there was no statistical difference between mean lacunarity in the two groups by P14. Angiotool was designed to measure morphometric and spatial parameters of vascularity in two dimensions through skeletonization and feature extraction. Because the algorithm does not consider axial network connections and requires high signal-to-noise for feature identification, it was not applied to 3-D and volumetric images.

Fig. 9

Angiotool analysis applied to retina vascular networks. (a) Example of Angiotool network tracing. (b) Total vessel length decreased in HO eyes at P7, resolves by P14. (c) Junction density decreased in HO eyes at P7, resolves by P14. (d) Mean lacunarity increased in HO eyes at P7, resolves by P14. * denotes $p<0.05$ using ANOVA test. All data are $\text{mean}\pm \text{standard deviation}$.

3.4.

Quantification of Vascular Complexity Using Sholl Analysis

We conducted Sholl analysis to quantify the variations in vascular distribution between control and HO retinas. Previously used to quantify characteristics of dendritic processes branching off neuronal cell bodies, Sholl analysis examines the number of branches that intersect concentric circles (retinal flat mounts) or spheres (intact retinas) of increasing radii around the cell body or central focal point of the image area (Fig. 10).^38,39

Fig. 10

Sholl analysis applied to retinal vascular networks. Schematic of Sholl analysis applied to retinal flat mount. Starting from the center of the retina, equally spaced concentric circles (shaded red) are superimposed on a segmented image of the retinal flat mount. Along each circle, the number of intersections with the vascular network is counted (inset: yellow star along red lines). This provides a measure of the vascular network complexity as a function of radial distance from the center.

Using this method, we observed an initial decrease in vascular branching in HO P7 retinas compared with those of the control [Figs. 11(a) and 11(c)]. This was consistent with our observations using DSLM and intact retinas in which there was a reduction or complete absence of the secondary plexus in HO retinas. Three-dimensional Sholl analysis applied to intact P14 retinas captured this observation while 2-D Sholl analysis applied to P14 retinal flat mounts failed to account for the lack of secondary plexus [Figs. 11(b) and 11(d)]. We utilized the “integrated” Sholl metric by taking the integral of the area under the Sholl analysis curve to calculate an approximate value of the total number of vascular network intersections (Fig. 12).⁴ This simple metric captures changes in 3-D vascular structure of intact retinas, providing a methodology to rapidly quantify the full vascular development of the retina.

Fig. 11

Sholl analysis results. Mean (solid) and standard deviation (shaded) of normalized Sholl curves for control (black) and HO (red). (a) P7 retinal flat mounts, (b) P14 retinal flat mounts, (c) intact P7 retinas, and (d) intact P14 retinas ($n=4$ for each group). There is a clear difference in vascular complexity at P7 for both flat mount and intact retinas. However, at P14, there is no clear difference in retinal flat mounts and a less pronounced difference in intact retinas.

Fig. 12

Box and whisker plots of integrated normalized Sholl metric. The Sholl metric is calculated for (a) retinal flat mounts and (b) intact retinas by integrating the area under each normalized curve. The metric is not comparable between flat mounts and intact retinas due to different normalizations (area versus volume). At P7, RA and HO flat mounts and intact retina vascular networks are statistically different ($n=4$, Kruskal–Wallis test, $p<0.01$). At P14, RA and HO flat mounts are not statistically different ($n=4$, Kruskal–Wallis test) while RA and HO intact retinas are ($n=4$, Kruskal–Wallis test, $p<0.01$).