Introduction

Retinopathy of Prematurity (ROP), which was first described in the 1940s, is a common blinding disease in premature babies with long term disability and diminished quality of life, creating a big burden to the aging society when they get old [1]. It is regarded as a vascular disease characterized first by vaso-obliteration and then pathological angiogenesis, excessive growth of blood vessels on the surface of the retina in response to the combined effects of immature retinal vasculature, and high levels of supplemental oxygen for neonatal clinical care [2–4]. In addition to the most well-known pathological retinal angiogenesis of this disease [2], reduced visual acuity, and neuronal dysfunction have also been observed in infants and children with a history of ROP [5, 6]. Alterations of electroretinographic (ERG) responses, including lower saturated amplitude, decreased sensitivity, and prolonged deactivation of rod photoreceptors have been observed, and these deficits appear to persist long even after ROP has been resolved [7–9]. Moreover, attenuated post-receptor responses are also present, although these deficits can be improved in children with a history of mild ROP, they stay in severe ROP [7, 9–11]. In rat or mouse models of oxygen-induced retinopathy (OIR) to mimic human ROP, altered ERG responses for rods and post-receptors, disrupted rod outer segment, morphological or neurochemical changes in horizontal, bipolar, or amacrine cells have been reported [3–5, 7, 12–14]. However, current treatments for ROP such as laser therapy, vitrectomy or VEGF inhibitor Bevacizumab, mainly focus on the pathological retinal angiogenesis [15–17]. Therefore, there is a great need to also investigate the long-term impact on retinal neurons of the potential therapeutic targets, which may demonstrate beneficial effects on attenuating the pathological retinal angiogenesis.

Oxidative stress has been highly implicated to contribute to the pathogenesis of ROP. Previous clinical studies have demonstrated increased oxidative stress in preterm infants [18]. Antioxidant supplement vitamin E has been shown to reduce the development and progression of ROP [19]. These observations suggest that targeting oxidative stress would be an efficient therapeutic approach in ROP treatment. Aldose reductase (AR) is the first enzyme in the polyol pathway for reducing glucose to sorbitol, using NAD(P)H as a cofactor [20]. Increased AR activity contributes to oxidative stress [21–26]. In the mouse retina, AR deficiency or inhibition of AR activity has been shown to reduce ischemia-induced oxidative stress [20, 27]. In our previous studies, genetic deletion or pharmacological inhibition of AR reduced the retinal vascular changes including central vaso-obliteration and neovascularization with reduced induction of oxidative stress in the mouse model of OIR [27]. In the current studies, we aimed to further evaluate the effects of genetic deletion of AR on retinal neuronal changes at both functional and cellular levels. We hypothesize that AR deficiency helps protect the retinal neurons in OIR.

Methods

Animals

All experimental and animal handling procedures were in accordance with the ARVO Statement for the Use of Animals in Ophthalmic and Vision Research. The use of animals was conducted according to the requirements of the Cap. 340 Animals (Control of Experiments) Ordinance and Regulations, and all relevant legislation and Codes of Practice in Hong Kong, and was approved by the Faulty Committee on the Use of Live Animals in Teaching and Research in The University of Hong Kong (CULATR 1691–08 and 2423–11).

Animal model of OIR

OIR was induced in wild-type (WT) and AR-deficient (AR−/−) mice [28]. Neonatal mice and their nursing dams were exposed to a 75 % oxygen (PRO-OX110 chamber controller; Biospherix Ltd, NY, USA) at postnatal day seven (P7), returned to room air on P12, and kept until P17 [29]. The hyperoxia from P7 to P12 induces retinal blood vessel regression, and the relative hypoxia-induced neovascularization reaches maximum on P17 [30]. During the exposure to high oxygen, soda lime was placed inside the chamber to serve as CO2 quencher. To reduce the runty phenotype, the litter size was limited to seven or eight pups for each mother [30]. AR−/− mice were observed to breed well, and the weight of pups on P7 (~4 g) and P17 (~6 g) was identical to that of WT mouse pups with limited litter size in OIR.

Samples were collected on P17 and P30. Both eyes were enucleated and fixed in 4 % paraformaldehyde (PFA) in phosphate buffered saline (PBS; 0.01 M; pH 7.4) overnight at 4 °C and then dehydrated in a graded series of ethanol and chloroform. Subsequently, eyeballs were infiltrated with paraffin for 3 hours (new paraffin changed every hour) and embedded. Sagittal sections (6 μm) were cut through the cornea parallel to the optic nerve using a microtome (Microm HM 315R, Heidelberg, Germany).

Assessment of retinal function by electroretinogram

Scotopic flash electroretinogram (ERG) was recorded from mice on P30 to assess retinal function. Before ERG measurement, mice were dark-adapted overnight and anesthetized with a mixture of Ketamine (50 mg/kg; Alfasan International, Woerden, The Netherlands) and Xylazine (0.5 mg/kg; Alfasan International, Woerden, The Netherlands). The eyes were further treated with 0.5 % proparacaine hydrochloride (Alcon, Alcon-Couvreur, Belgium) for topical analgesia and pupils were dilated with 1% Mydricyl (Alcon, Alcon-Couvreur, Belgium). The ERG signals were recorded by placing a gold active electrode on the cornea as well as a reference and a ground electrode in the forehead and tail region, respectively. All operations were performed under dim red light, and mice were kept warm on a heating pad during the ERG recording. Rod and cone responses were elicited by the single white flashes of intensity at 3 cd · s/m2 generated from the Color Dome Ganzfeld System (Diagnosys, Lowell, MA). Each flash lasted for 4 ms and the flash interval was 10 s. All ERG signals were amplified automatically by the system. For a-wave and b-wave, responses were recorded with a band-pass filtered from 0.3 to 300 Hz; for oscillatory potentials (OPs), responses were recorded with a band-pass filtered from 100 to 300 Hz. The sum of only OP2 to OP4 amplitudes was analyzed due to the potential of OP1 contaminated by the a-wave [31]. Fifteen readings from two eyes for each mouse were averaged and analyzed using the Epsion V5 software incorporated with the system.

Measurement of retinal layer thickness

Three sagittal sections with an optic nerve head for each eye were selected and stained with hemotoxylin and eosin (H&E). Two images from the same section were captured for central and mid-peripheral retinal areas with one from each side of the optic nerve, using an upright microscope with 400X magnification (Eclipse 80i, Nikon, Japan). The central retinal area was defined as ~100 μm away from the optic nerve; the mid-peripheral retinal area was defined as the middle area between optic nerve and the end of the retina on each side. Outer nuclear layer (ONL), inner nuclear layer (INL) and inner plexiform layer (IPL) thicknesses were then measured using Spot Advanced software (SPOTTM Imaging Solutions, Sterling Heights, USA).

Immunohistochemistry (IHC) for various retinal neurons and glial cells

Deparaffinized retinal sagittal sections were subjected to antigen retrieval by incubation with proteinase K (20 μg/ml in PBS) for 4 min at room temperature. Sections were blocked with goat or donkey serum (Vector Laboratories Inc.) and incubated with primary antibodies rabbit anti-PKCα (rod bipolar cell marker, 1:500; sc-208, Santa Cruz Biotechnology), goat anti-calretinin (amacrine cell marker, 1:1000; sc-11644, Santa Cruz Biotechnology), mouse anti-Tuj1 (retinal ganglion cell marker, 1:1000; MRB-435P, Convance), and rabbit anti-glial fibrillary acidic protein (GFAP 1:500, Dako, Denmark), respectively, overnight at 4 °C. For rabbit anti-calbindin (horizontal cell marker, 1:500; AB1778, Chemicon) staining, sections were incubated for 72 h at 4 °C. Signals were visualized by reacting with corresponding secondary antibodies Alexa Fluor® 488 donkey anti-goat IgG (1:500, A-11055, Molecular Probes, Invitrogen), or Alexa Fluor® 568 goat anti-rabbit IgG (1:500; A-11011, Molecular Probes, Invitrogen), or Alexa Fluor® 568 goat anti-mouse IgG (1:500, A-21134, Molecular Probes, Invitrogen), respectively, for 60 min at room temperature. The sections were washed with PBS and counterstained with DAPI (0.1 μg/ml in PBS, D9564, Sigma). The immunoreactivity was examined using an upright microscope with 400X magnification (Eclipse 80i, Nikon, Japan). For horizontal and rod bipolar cells, a scoring system was used according to previous studies [27, 32, 33]. Briefly, the slides for the same immunostaining at each time point were examined at the same time under the microscope with the same light exposure intensity. The scoring was done within 2 h. To minimize the potential bias, all the slides were re-coded and the examiners were blinded while performing the scoring. Based on the intensity and location of the staining, score one was assigned for the weakest immunoreactivity and discrete alignment of cell bodies, and score five represented the strongest immunoreactivity and evidently located cell bodies. For calretinin-positive amacrine cells in INL and Tuj1-positive RGCs in a ganglion cell layer (GCL), the number of cell bodies was counted and compared between the two genotypes. Representative images were taken from the central retinal areas of retinal sections with 400X magnification (Eclipse 80i, Nikon, Japan) using the same exposure of light intensity and time.

Statistical analysis

All experiments and analyses were performed in a double-blinded manner. The immunohistochemical investigations for all retinal sections using the same antibody were performed in one single experiment to eliminate inter-experiment errors and variation. One-way ANOVA and Bonferroni’s multiple comparisons test, Kruskal-Wallis test, and Dunn’s multiple comparison test were used for comparison of results as specified in the figure legends (Prism v5.0, GraphPad Software Inc., San Diego, CA). Statistically significant difference was set at p < 0.05. Data was presented as mean ± SEM.

Results

Preservation of retinal neuronal function in AR−/− mice

Scotopic ERG was performed to investigate the effect of OIR on neuroretinal function in WT and AR−/− mice on P30. Representative ERG waveforms recorded from room air and OIR retinae in both genotypes were demonstrated in Fig. 1b. Under room air, the ERG waveforms were very similar between the WT and AR−/− mice, and no significant difference was observed in the a-wave, b-wave amplitudes, and the sum of OPs (from OP2 to OP4) amplitudes between WT and AR−/− mice (Fig. 1c). In OIR, a-wave, b-wave, and Ops amplitudes were decreased significantly in WT but not AR−/− mice (Fig. 1c). AR−/− mice appeared to display better preserved ERG waveform than that in WT OIR mice (Fig. 1c). Delayed implicit time (time-to-peak) for a-wave was observed in both genotypes (Fig. 1d). These observations suggested that inner retinal functions were disturbed in OIR and partially preserved with AR deficiency.

Fig. 1
figure 1

ERG responses of WT and AR−/− mice on P30. The normal ERG responses with parameters that are usually measured for electrodiagnosis (a). Representative ERG waveforms were exhibited at low light intensity (3 cd · s/m2, dotted line) for both genotypes (b). No significant difference was observed in a-wave amplitudes between WT OIR and AR−/− OIR mice. Significantly reduced amplitudes of a-wave, b-wave, and the sum of OPs were found in WT OIR mice but not is AR−/− OIR mice (c). The delayed implicit time of a-wave was observed in both genotypes (D). ***p < 0.001, **p < 0.01, *p < 0.05, One-way ANOVA and Bonferroni’s multiple comparisons test. n = 5–9 for each group

Full size image

To further understand if AR deficiency protects the retinal functions in mouse OIR through preserving retinal neuronal structures, investigation on cellular levels of retinal neurons was conducted. Retinal thickness was measured and immunohistochemistry for the inner retinal cells was analyzed and compared.

Retinal layer thickness on P17 and P30

Hyperoxia exposure from P7 to P12 induces retinal blood vessel regression and the relative hypoxia-induced neovascularization reaches the maximum on P17 [30]. Therefore, besides P30, the relative hypoxia-induced retinal neuronal changes were also investigated on P17 corresponding to the retinal vascular changes in OIR.

The ONL, INL, and IPL thicknesses in central (~100 μm away from the optic nerve) and the mid-peripheral retinal areas (the middle area between the optic nerve and the end of the retina on either side of the optic nerve) were measured on H&E-stained retinal sections on P17 and P30 (Fig. 2a–b). The central retinal areas are vaso-obliterated while the mid-peripheral areas are the boundary between vaso-obliterated and peripheral areas where neovascularization occurred in the mouse OIR retina. The results were analyzed and compared between WT and AR−/− retinae. Generally, comparable retinal thicknesses at different layers were observed in both WT and AR−/− retinae at room air conditions (Fig. 2c). After OIR, there was no significant change in ONL thickness, but markedly reduced INL and IPL thickness in the central retinal areas were observed versus room-air groups on P17, 30 in both WT and AR−/− retinae (Fig. 2c). More importantly, the reductions in INL and IPL thickness in the mid-peripheral areas observed in WT mice on P17 after OIR were not observed with AR deficiency. In addition, thicker INL and IPL with AR deficiency were also detected on P30 (Fig. 2c).

Fig. 2
figure 2

Retinal thickness in the OIR model and WT mice. Measurement of ONL, INL, and IPL thicknesses in the central and mid-peripheral areas on H&E-stained retinal sections (a–b). In the central and mid-peripheral retinal area, comparable ONL, INL, and IPL thickness was found in both genotypes on P17 and P30 under room-air and OIR conditions (c). Significantly reduced INL and IPL thickness was observed in WT OIR mice (c). Significantly thicker IPL in mid-peripheral areas on P30 were observed in AR−/− OIR retinae (c). No significant change was observed in ONL thickness (c). ***p < 0.001, **p < 0.01, *p < 0.05, One-way ANOVA and Bonferroni's multiple comparisons test. n = 6–13 for each group. C central, MP mid-peripheral, GCL ganglion cell layer, IPL inner plexiform layer, INL inner nuclear layer, OPL outer plexiform layer, ONL outer nuclear layer. Scale bar: 100 μm (a); 25 μm (b)

Full size image

Inner retinal neurons on P17 and P30

Horizontal cells

Calbindin has been used as a marker for horizontal cells [34, 35]. The immunoreactivity of calbindin is present in cell bodies, dendrites, and axons of horizontal cells along the outer margin of INL [34]. In addition, amacrine cells and RGCs also appeared to be positive for calbindin staining [34]. Here, we focused on the calbindin-positive horizontal cells only.

On P17, evidently located calbindin-positive cell bodies along the outer margin of INL and faintly stained processes between horizontal cells were observed in both genotypes under room air, no significant difference was shown between the two genotypes based on the IHC scoring system (Fig. 3a–b). In OIR, the intensity of calbindin immunoreactivity along outer INL was decreased especially in the processes between horizontal cells in WT but not AR−/− OIR retinae (Fig. 3c–d), although IHC scoring system did not show a significant difference (Fig. 3i). On P30, strong staining of calbindin was exhibited in horizontal cell bodies along the outer margin of INL and processes between the cell bodies in both genotypes under room air (Fig. 3e–f), indicating matured horizontal cell interactions. In OIR, the intensity of calbindin immunoreactivity was decreased in WT retinae but not in AR−/− retinae, and the calbindin staining intensity was significantly stronger in AR−/− retinae than that in WT retinae (Fig. 3 g–i). In addition, in AR−/− OIR retinae, cell processes appeared to be more strongly calbindin-stained between horizontal cells when compared with WT OIR retinae (Fig. 3 h).

Fig. 3
figure 3

Evaluation of retinal horizontal cells in the OIR model and WT mice. Immunohistochemical staining of horizontal cells with an antibody against calbindin on P17 and P30 (red, a–h). Calbindin-positive horizontal cell bodies were present along the outer margin of INL in WT and AR−/− RA retinae (a–b, e–f). After OIR, reduced calbindin immunoreactivity was observed in WT but not AR−/− OIR retinae (c–d, g–h). A trend of calbindin immunoreactivity reduction was also exhibited in WT OIR retinae on P30 (i). Significantly more intense calbindin immunoreactivity was observed in AR−/− OIR retinae versus WT OIR retinae on P30 (i). The sections were counterstained with DAPI for nuclei identification (blue). Arrows, calbindin-positive horizontal cell processes. *p < 0.05, Kruskal-Wallis test, and Dunn’s multiple comparison test. n = 5–9 for each group. Scale bar, 25 μm

Full size image

Rod bipolar cells

PKCα was used for the rod bipolar cells examination. PKCα immunoreactivity was present not only in the rod bipolar cell bodies along the outer margin of INL but also in their processes extended into IPL. In addition, a few amacrine cells along the inner margin of INL were also stained with PKCα [34–36]. Here, we focused on PKCα-positive rod bipolar cells.

On P17, normally, PKCα immunoreactivity was present in rod bipolar cell bodies in the outer margin of INL and processes close to GCL (Fig. 4a–d). Strong PKCα immunoreactivity was observed in rod bipolar cells for WT and AR−/− room-air retinae (Fig. 4a and b). However, in OIR, a sharp decrease in PKCα immunoreactivity especially in the processes was observed in both genotypes (Fig. 4c and d). The IHC scoring system demonstrated significantly decreased PKCα immunoreactivity in both genotypes in OIR versus room-air controls (Fig. 4i). On P30, PKCα immunoreactivity was apparently exhibited in rod bipolar cell bodies along the outer margin of INL and processes close to GCL in both genotypes under room air (Fig. 4e and f). In OIR, PKCα immunoreactivity, particularly in the processes, was much reduced in WT retinae but not in AR−/− retinae (Fig. 4g and h). The IHC scoring system also showed a significant decrease in WT OIR retinae (Fig. 4i).

Fig. 4
figure 4

Evaluation of retinal rod bipolar cells in the OIR model and WT mice. Immunohistochemical staining of rod bipolar cells with an antibody against PKCα on P17 and P30 (red, a–h). PKCα immunoreactivity was present in rod bipolar cell bodies along the outer margin of INL and processes close to GCL in both WT and AR−/− RA retinae (a–b, e–f). After OIR, on P17, less PKCα immunoreactivity in rod bipolar cell processes adjacent to GCL was found (c–d) and significantly reduced PKCα immunoreactivity after OIR was observed in both genotypes (i). On P30, significantly reduced PKCα immunoreactivity, particularly in the processes, was observed in WT versus AR−/− OIR retinae (g–h). The sections were counterstained with DAPI for nuclei identification (blue). **p < 0.01, *p < 0.05, Kruskal-Wallis test, and Dunn’s multiple comparison test. n = 5–10 for each group. Scale bar, 25 μm

Full size image

Amacrine cells

Calretinin was used as a marker for amacrine cells. Calretinin immunoreactivity was observed in the amacrine cell bodies along the inner margin of INL and the three strata of the axon terminals in IPL, as well as some displaced amacrine cells in GCL [34, 35]. In addition, some RGCs were also stained with calretinin [34]. Here, we focused on calretinin-positive amacrine cells in INL as well as three strata in IPL.

On P17, calretinin immunoreactivity was strong in cell bodies in INL and three clear strata in IPL in both genotypes under room air (Fig. 5a and b). In OIR, markedly reduced intensity of calretinin immunoreactivity, seriously distorted three strata in IPL, and decreased cell bodies in INL were observed in WT retinae but not in AR−/− retinae (Fig. 5c and d). The number of calretinin-positive cell bodies along the inner margin of INL was quantified and a significant reduction was shown in WT OIR retinae but not AR−/− OIR retinae (Fig. 5i). On P30, calretinin immunoreactivity was similar to that on P17 in room air controls (Fig. 5e and f). In OIR, the intensity of calretinin immunoreactivity was similar in both genotypes (Fig. 5g and h). A significantly reduced number of cell bodies was observed in WT not AR−/− OIR retinae (Fig. 5i).

Fig. 5
figure 5

Evaluation of retinal amacrine cells in the OIR model and WT mice. Immunohistochemical staining of amacrine cells with an antibody against calretinin on P17 and P30 (green, a–h). Calretinin was expressed in amacrine cell bodies in the innermost layer of INL as well as in three clear and straight strata inside IPL in both WT and AR- /- RA retinae (a–b, e–f). After OIR, on P17, three distorted strata in IPL were observed especially in WT OIR retinae (c–d) and significantly reduced amacrine cells were shown in WT but not AR−/− OIR retinae (i). On P30, significantly reduced amacrine cells in INL were only observed in WT OIR retinae (g–h, i). The sections were counterstained with DAPI for nuclei identification (blue). *p < 0.05, One-way ANOVA, and Bonferroni’s multiple comparison test. n = 5–12 for each group. Scale bar, 25 μm

Full size image

RGCs

Tuj1 was highly reactive to neuron specific Class III β-tubulin and was, therefore, used as a marker for RGCs [37, 38]. Yet, Tuj1 immunoreactivity was observed not only in cell bodies along GCL but also those in INL (Fig. 6a). In addition, Tuj1 immunoreactivity was also present in the nerve fiber layer (Fig. 6a). Therefore, Tuj1-positive cell bodies along GCL were regarded as RGCs only when they were also DAPI-positive in merged images (Fig. 6a–c). The number of the Tuj1- and DAPI- positive cells in GCL was quantified and compared between the two genotypes with or without oxygen exposure. Under room-air conditions, a comparable number of RGCs was found in WT and AR-deficient retinae on P17 and P30. No significant difference was observed on P17 and P30 between the two genotypes (Fig. 6d).

Fig. 6
figure 6

Evaluation of retinal RGCs in the OIR model and WT mice. Immunohistochemical staining of RGCs with Tuj1 (red) and DAPI (blue, a–c). Only cells positive for both Tuj1 and DAPI were regarded as RGCs (arrows, a–c). No significant changes in the number of RGCs were observed between WT and AR−/− retinae under room air and after oxygen exposure on P17 and P30 (D). *p < 0.05, One-way ANOVA and Bonferroni’s multiple comparisons test. n = 5–9 for each group. Scale bar, 25 μm

Full size image

Müller cell gliosis on P17 and P30

GFAP is a known hallmark of reactive gliosis in astrocytes and Müller cells [39]. Normally, GFAP staining was localized in astrocytes around the blood vessels along the inner limiting membrane (ILM) on P17and P30 (Fig. 7a and b, e and f). On P17, more intense GFAP immunoreactivity in astrocytes was observed in ILM in WT OIR retinae than that in AR−/− OIR retinae (Fig. 7c and d). Strong GFAP immunoreactivity was also observed in Müller cell processes across IPL (Fig. 7c). In AR−/− OIR retinae, increased GFAP immunoreactivity was also found in ILM (Fig. 7d). Furthermore, less GFAP immunoreactivity in Müller cell processes was observed when compared with WT OIR retinae (Fig. 7d). IHC scoring demonstrated significant GFAP induction in WT but not AR−/− OIR retinae (Fig. 7i). On P30, after OIR, GFAP immunoreactivity was found to be increased in ILM in both genotypes (Fig. 7g and h). In addition, strong GFAP staining was exhibited in some cell processes across IPL in WT OIR retinae (Fig. 7g). However, in AR−/− OIR retinae, only faint GFAP immunoreactivity was shown in Müller cell processes across IPL (Fig. 7h). IHC scoring also revealed a significant increase in GFAP immunoreactivity in WT but not AR−/− OIR retinae (Fig. 7).

Fig. 7
figure 7

Evaluation of glial response in the OIR model and WT mice. Immunohistochemical staining of astrocytes and Müller cells with GFAP antibody on P17 and P30 (red a–h). GFAP immunoreactivity was present in astrocytes along ILM in RA controls for both genotypes (a–b, e–f). Increased GFAP intensity along ILM was observed in OIR groups versus RA controls for both genotypes (c–d, g–h). On P17 and P30, intense GFAP immunoreactivity was present in the Müller cell process (arrows) across IPL in WT but not AR−/− OIR retinae (c–d, g–h). The IHC scoring system also demonstrated significantly increased GFAP induction in WT but not AR−/− OIR retinae (I). The sections were counterstained with DAPI for nuclei identification (blue). ***p < 0.001, Kruskal-Wallis test, and Dunn’s multiple comparison test. n = 5–10 for each group. Scale bar, 25 μm

Full size image

Discussion

ROP is well characterized by abnormal vasculature including blood vessel regression and pathological angiogenesis [2]. In addition, ROP also results in retinal neuronal dysfunction [5, 6]. Our results not only compensated current limited studies of hypoxia-induced retinal neuronal changes at the cellular level in the mouse model of OIR, but also demonstrated a potential protective role of AR deficiency in hypoxic retinal neurons in OIR.

Retinal dysfunction has been reported in infants and children with a history of ROP as well as in animal models of OIR [5–14]. ERG was performed to identify retinal function in mice. In ERG recordings, an a-wave is directly produced by photoreceptor cells and a b-wave mainly reflects bipolar and Müller cell functions [40–42]. The OPs are regarded as primarily amacrine cell responses to the light, although ganglion cells are also involved in the generation of Ops [41–43]. It has been demonstrated that a-wave, b-wave, and OPs were all decreased in the animal model of OIR [41, 42, 44]. Our results also exhibited decreased a-wave, b-wave, and OPs amplitudes in WT OIR retinae versus WT room-air retinae. Interestingly, AR deficiency attenuated the reduced ERG responses (b-wave and OPs) observed in WT OIR retinae, suggesting that AR deficiency may help improve the inner retinal function in OIR.

Altered retinal neuronal function suggests related retinal structural changes [5, 7–9, 13, 14, 45]. In the current study, hypoxia-induced phenotypic changes in horizontal cells, amacrine cells, and rod bipolar cells were observed on P17 and P30 in the mouse model of OIR. Previously, morphological changes of horizontal cells, amacrine cells, and rod bipolar cells on P17 have been shown in the mouse model of OIR [3, 12]. Our observations demonstrated that the changes of retinal layer thickness and retinal neural cells observed on P17 persist until P30 when the retinal vasculature resolved, suggesting that the protection of retinal neurons should also be considered while targeting the retinal neovascularization in OIR.

With AR deficiency, attenuated changes in horizontal cells and amacrine cells on P17 and P30 as well as rod bipolar cells on P30 were observed when compared with WT OIR retinae. However, no significant change was observed in RGCs on all time points. In fact, previous reports have shown the majority of cell apoptosis on P14 was in INL but not in GCL in the mouse model of OIR [3]. It has been demonstrated that exposure of neonatal rats during the period undergoing significant maturation exhibited more severe damage in the model of OIR [46]. In the neonatal mouse retina, retinal neuronal maturation proceeded from the inner to outer retina with RGCs maturing first [47]. Our observed minimal effects of hypoxia on RGCs, significant effects on horizontal cells, amacrine cells, and rod bipolar cells, agreed well with this maturation process.

Glial responses have been reported in the animal models of OIR [29, 48]. Normally, retinal glia including astrocytes and Müller cells provide the support and nutrients for retinal neurons, maintain the blood retinal barrier (BRB), and function as a template to guide the developing retinal vasculature and neurons in normal conditions [4, 49]. In pathological circumstances, retinal glia became activated and resulted in “reactive gliosis” with GFAP up-regulation [29, 39, 50]. According to previous findings, Müller cell gliosis may contribute to neovascularization and neuronal cell death [48, 51]. Our results showed persistent GFAP induction on P17 and P30 in WT OIR retinae. With AR deficiency, Müller cell gliosis was found to be weaker on P17 and P30 when compared with WT OIR retinae. Therefore, reduced Müller cell gliosis with AR deficiency might contribute to the preserved inner retinal neuronal protection on a respective time point in the mouse model of OIR.

In addition, oxidative stress has also been highly implicated in the pathogenesis of OIR [52, 53]. Animal studies for ROP using antioxidants such as NADPH oxidase inhibitor [52], vitamin E [54], inducible nitric oxide synthase, or cyclooxygenase-2 inhibitors [55] have been found to reduce the progression of OIR. Previously, we have found induced oxidative stress in OIR and AR deficiency could reduce oxidative stress induction in the OIR retinae, especially in the inner nuclear layer and ganglion cell layer [27]. Reduced oxidative stress in the protected inner retinal neurons was observed with AR deficiency, suggesting a direct effect of AR on retinal inner neurons via oxidative stress modulation. Although oxidative stress were shown in GCL, no significant change in the number of RGCs was found, possibly due to the early maturation of RGCs with clearly evident RGC classes observed around P12, and matured RGCs may have more resistance to stressed environment [46, 56].

On P17 and P30, significant reductions in inner retinal thickness were observed in WT OIR retinae in both central and mid-peripheral areas, corresponding to vaso-obliteration and neovascularization in mouse OIR retinae [27, 29, 30, 57–61], respectively. AR deficiency appears to attenuate these reductions in INL and IPL thicknesses especially in the mid-peripheral neovascular areas. Previously, we have shown that genetic deletion of AR reduced neovascularization [27]. As the retinal vasculature supports the retinal neurons with nutrients and oxygen, there is, therefore, a possibility that the attenuated pathological neovascularization with AR deficiency may contribute to the functional and morphological preservation of inner retinal neurons. Further confirmation of the definitive and sequential roles of protected retinal vasculature and preserved retinal neurons with AR deficiency in OIR can be achieved by neuron-specific adeno-associated viral down-regulation of AR expression in retinal neurons.

On P17, our previous observations demonstrated reduced vaso-obliterated areas and neovascular areas in AR−/− OIR retinae [27], suggesting AR deficiency helps protect the normal vascular development as well as inhibit the pathological vessel proliferation. With the known persistent retinal neuronal dysfunction in ROP infants and animal models, further investigation on retinal neuronal changes and stress-responded gliosis were conducted in the current study. Induced gliosis as well as morphological and functional changes of retinal neurons were observed in OIR. AR deficiency demonstrated attenuated gliosis and neuronal abnormalities. Overall, these findings indicated that AR could be considered as a potentially therapeutic target in ROP benefiting both retinal vasculature and neurons, possibly through attenuating glial activation.