Animal model and ethics

All animal procedures were performed in compliance with the European Community directive guidelines (2010/63/EU) for the use of experimental animals, transposed into Portuguese law in 2013 (Decreto-lei 113/2013). The procedures were approved by the Animal Welfare Committee of the Coimbra Institute for Clinical and Biomedical Research (iCBR), Faculty of Medicine, University of Coimbra (ORBEA 02/2021), and received further approval from the Direção Geral de Alimentação e Veterinária (DGAV – approval no. 0421/000/000/2021). Additionally, the animal experimentation was conducted in accordance with the Association for Research in Vision and Ophthalmology (ARVO) Statement for the Use of Animals in Ophthalmic and Vision Research.

Male and female (10-week-old) Wistar Han rats (Charles River Laboratories, Lyon, France) were housed in certified facilities with a temperature- and humidity-controlled environment, maintained under a 12 h:12 h light–dark cycle, and with ad libitum access to food and water. The animals were randomly divided into the control and diabetic (T2D) groups. Type 2 diabetes was induced by a HFD containing approximately 40% total fat (Special Diet HF231—PF4486; Mucedola srl, Settimo Milanese, Italy) for 12 weeks, in combination with a single low-dose STZ injection (35 mg/kg in 10 mM sodium citrate buffer, pH 4.5; IP; after 6 h fasting) in the fourth week after starting the diet (Fig. 1a). Fasting glycaemia (Fig. 1b) and body weight (Fig. 1c) were monitored at the beginning of the study, i.e., just before starting the diet (week 0), and after 4, 5, 6, 8, and 12 weeks on HFD. Glycated haemoglobin (HbA1c) was measured at weeks 4 and 12 (Fig. 1d), and the oral glucose tolerance test (OGTT; 1.8 mg/kg glucose) was performed at weeks 0 and 12 (the results for week 12 are presented in Fig. 1e).

Fig. 1

Impact of diabetes on body weight and metabolic parameters. (a) Protocol of type 2 diabetes induction. Animals were fed a high-fat diet (HFD) for 12 weeks, combined with a single low-dose STZ injection (35 mg/kg, IP) in the fourth week after starting the diet. (b) Fasting glycaemia (6 h fasting), (c) body weight, and (d) HbA1c were monitored in diabetic Wistar Han rats (T2D group) and age-matched controls (Control group). (e) The OGTT (1.8 g/kg glucose) was performed after 6 h of fasting (only the results obtained after 12 weeks on HFD are presented). Results are presented as mean ± SEM. Statistical analysis was performed using the Mann–Whitney test. *P ≤ 0.05, **P < 0.01, ***P < 0.001, versus the age-matched Control group. STZ, streptozotocin; IP, intraperitoneal; HbA1c, glycated haemoglobin; T2D, type 2 diabetes; OGTT, oral glucose tolerance test; SEM, standard error of the mean

OCT data acquisition

Animals were anesthetised with an IP injection of ketamine (30 mg/kg; Nimatek, Dechra, Northwich, UK) and medetomidine (1 mg/kg; Sedator, Dechra, Northwich, UK) in 0.1 mL saline solution, followed by corneal anaesthesia with oxybuprocaine (4 mg/mL; Anestocil, Edol, Lisbon, Portugal), and full pupil dilation using a topical drop of tropicamide (10 mg/mL; Tropicil Top, Edol, Lisbon, Portugal). To keep the corneal surface moisturised, carmellose sodium (4 mg/0.4 mL; Celluvisc, Allergan, Dublin, Ireland) lubricating eye drops were regularly applied. The retinas from the right eye of control (N = 44) and diabetic (N = 45) animals were imaged using a Micron IV OCT System (Phoenix Technology Group, Pleasanton, CA, USA) at the baseline (week 0) and after 4, 8, and 12 weeks on HFD. Retinal OCT volumes were acquired consisting of 512 B-scans each (composed of 512 A-scans of 1024 pixels in length), imaged at a predefined retinal region (located above the optic disc, maintaining a horizontal alignment with its centre). The B-scans were saved as non-compressed TIFF image files. All OCT data acquisitions were performed by the same operator. In the molecular and cellular studies, multiple parameters were assessed using retinal samples from the same animals.

Automatic retinal segmentation and retinal thickness measurement

The retinal layer segmentation was achieved using a fully convolutional neural network, following a ResNet architecture, as previously described [32, 36, 37]. A total of six distinct structures (layers and layer aggregates) were segmented: the nerve fiber layer and ganglion cell layer complex (NFL-GCL), the inner plexiform layer (IPL), the inner nuclear layer (INL), the outer plexiform layer (OPL), the outer nuclear layer (ONL), and the inner/outer photoreceptor segments (IS/OS). Volumetric segmentation was determined by combining the 512 segmented B-scans. An expert (author S.O.) evaluated the quality of the segmentation in a masked fashion for both experimental groups. After obtaining the volumetric segmented OCT data, retinal thickness maps were computed for each of the six retinal layers, and the thickness of each retinal layer/layer conjugate, as well as the thickness of the entire retina, were calculated as the distance between the respective segmented boundaries, as reported in our previous work [32].

Texture analysis

For each layer, projection mean-value fundus (MVF) images [25] were computed from volume OCT data, as previously described [32]. MVF images consist of 2-dimensional (2D) images in which each pixel’s value is the average of the values of each A-scan between two retinal layer interfaces defining the layer. These allow the projection of the 3-dimensional (3D) OCT data onto a 2D plane, which mimics a fundus photograph capturing only the layer of interest. Texture-related features were extracted from the MVF images using the GLCM statistical method, as described previously [32]. The key concept of GLCM lies in analysing the probability of finding a transition from a grey-level A to a grey-level B along an established direction and distance in the image. By computing all possible different transitions at a pre-defined pixel distance and direction, GLCM provides texture information of the image being analysed. Twenty features per layer were computed, namely: the inverse difference moment/energy, contrast/inertia, correlation, angular second moment/uniformity/homogeneity, sum average, sum of squares, sum variance, sum entropy, difference variance, difference entropy, information measure of correlation I (IMCI), IMCII, and entropy, as described by Haralick et al. [38]; autocorrelation and maximum probability, as described by Haralick et al. [39]; cluster prominence and cluster shade, as described by Conners et al. [40]; INN and IDN, as described by Clausi et al. [41]; and dissimilarity, as described by Soh et al. [42].

ERG

Animals were anesthetised as described above, and electroretinograms were recorded at baseline (week 0) and after 4, 8, and 12 weeks on HFD, following the protocol from our previous study [32]. In brief, after a dark adaptation overnight, retinal ganglion cell (RGC) function was registered by stimulating both eyes with a 0.000095 cd·s/m2 light stimulus, following the scotopic threshold response (STR) protocol. The scotopic luminance response was also recorded under scotopic conditions by stimulation with light flashes ranging from 0.0095 to 9.49 cd·s/m2, inducing rod activation (a-wave), followed by the subsequent activation of the downstream retinal cells [oscillatory potentials (OPs) and b-wave]. Then, after light adaptation to a 25 cd/m2 white background for 15 min, the photopic luminance response was recorded in response to a 0.0095 to 9.49 cd·s/m2 light stimulus. To obtain a more isolated response from cones, the photopic flicker protocol was performed by stimulation with white light flashes (0.95, 3.00, and 9.49 cd·s/m2) delivered 10 times at 6.33 Hz. ERGs were recorded using a RETIport System (Roland Consult GmbH, Brandenburg, Germany), and the light stimulation was performed using a Ganzfeld stimulator (Roland Consult GmbH, Brandenburg, Germany), with the data being extracted from the RETIport software (Roland Consult GmbH, Brandenburg, Germany).

Following OCT and ERG recordings, atipamezole (1 mg/kg; Revazol, Dechra, Northwich, UK) was injected intraperitoneally to reverse medetomidine sedation.

ImmunofluorescenceRetinal wholemounts

Animals were anesthetised (2.5% isoflurane in O2; IsoFLO, Ecuphar, Oostkamp, Belgium) and euthanised by cervical dislocation. The retinas were then dissected, and the retinal wholemounts were prepared following the procedure previously reported by our group [32]. Briefly, after fixation with 4% paraformaldehyde (PFA) for 10 min, the retinal wholemounts were blocked for 1 h with 5% bovine serum albumin (BSA) and incubated for 72 h with antibodies against claudin-5, occludin, and zonula occludens-1 (ZO-1; Table 1). After washing, the retinas were incubated with the corresponding secondary antibodies (Table 1) for 24 h, followed by incubation with 1:1000 DAPI (Thermo Fisher Scientific, Waltham, MA, USA), and then mounted with the vitreous side up for visualisation under the LSM 710 Meta confocal laser scanning microscope (Zeiss, Jena, Germany) with a 20 × objective (Plan Achromat 20 × /0.8 M27).

Table 1 List of antibodies used for immunofluorescence of retinal wholemountsRetinal cryosections

Retinal sections (14-μm thick) were prepared as described previously [32]. Briefly, these were fixed in cold acetone for 10 min, permeabilized for 30 min, and blocked with 10% normal goat serum and 1% BSA before overnight incubation with primary antibodies against Iba1, OX-6/MHC II, glial fibrillary acidic protein (GFAP), vimentin, and nitrotyrosine (at 4 °C; Table 2). Sections were then washed and incubated with the secondary antibodies (Table 2) and 1:5000 DAPI for 1 h at room temperature. For image analysis, the images were captured with a fluorescence microscope (Axio Observer.Z1, Zeiss, Jena, Germany) using a 20 × objective (Plan Achromat 20 × /0.8 M27). Representative Z-stack images were acquired with a 20 × objective (Plan Achromat 20 × /0.8 M27) on a confocal microscope (Zeiss LSM 710, Zeiss, Jena, Germany).

Table 2 List of antibodies used for immunofluorescence of retinal sectionsEvans blue assay

Blood-retinal barrier permeability was qualitatively assessed using the Evans blue assay, following our group’s previously published protocol [32]. Animals were anesthetised with an IP injection of ketamine (75 mg/kg; Nimatek, Dechra, Northwich, UK) and medetomidine (1 mg/kg; Sedator, Dechra, Northwich, UK). After filtration (pore size, 0.2 μm), Evans blue (100 mg/kg in PBS; Sigma-Aldrich, St. Louis, MO, USA) was injected via the tail vein, and the animals were kept on a warm pad for 2 h. The eyes were enucleated and fixed with 2% PFA for 2 h, after which the retinas were dissected and mounted with the vitreous side up for visualisation under the LSM 710 Meta confocal laser scanning microscope (Zeiss, Jena, Germany) with a 20 × objective (Plan Achromat 20 × /0.8 M27).

Western blotting

Retinal lysates were prepared, and Western blot was performed as described previously [32]. In brief, equal amounts of protein (30 µg) were loaded onto 6%–12% polyacrylamide gels, separated through sodium dodecyl sulphate–polyacrylamide gel electrophoresis (SDS-PAGE), and transferred to polyvinylidene fluoride membranes (Millipore, Billerica, MA, USA). After blocking for 2 h at room temperature, membranes were incubated overnight with claudin-5, occludin, ZO-1, interleukin-1 beta (IL-1β), and tumour necrosis factor (TNF) primary antibodies (Table 3), followed by 2 h incubation with the corresponding secondary antibodies (Table 3). Immunoblots were imaged on the ImageQuant™ LAS 500 (GE Healthcare, Chicago, IL, USA), and band intensity was quantified using ImageQuant 5.0 software (Molecular Dynamics, Sunnyvale, CA, USA).

Table 3 List of antibodies used for Western blottingEnzyme-linked immunosorbent assay

For tissue preparation, one retina from each eye was homogenised in lysis buffer [20 mM imidazole HCl, 100 mM KCl, 1 mM MgCl2, 1% (vol./vol.) Triton X-100, 1 mM EGTA, 1 mM EDTA, 10 mM NaF, and 1 mM Na3VO4 (activated), supplemented with complete mini protease inhibitor cocktail tablets]. Then, following the manufacturer’s instructions, the protein levels of IL-1β and TNF were assessed in retinal homogenates using the rat IL-1β and TNF enzyme-linked immunosorbent assay (ELISA) kits (#RAB0478 and #RAB0480; Sigma-Aldrich, St. Louis, MO, USA), respectively.

Statistical analysis

ANCOVA tests were performed to assess differences between groups and each individual time point of the study (after 4, 8, and 12 weeks on HFD) after adjusting for the baseline (week 0), which was used as a quantitative covariate. Additionally, the results from retinal thickness, metabolic parameters, Western blot, and immunofluorescence were analysed using the Student’s t-test, or the Mann–Whitney test, if data were not normally distributed, which was assessed using the Shapiro–Wilk test. SPSS software (version 27; IBM Corp., Armonk, NY, USA) was used to perform the statistical analysis, where we considered a statistical significance of 5%. Results are presented as mean ± standard error of the mean (SEM), except for the graphs of the retinal texture, in which the results are expressed by box plots showing the median (horizontal line), 25th and 75th percentiles (box), and minimum and maximum (error bars). The results of the retinal thickness were calculated as the percentage of the baseline and are presented as mean ± SEM.