We characterized the expression of three ABC transporters in the retina: P-gp (Mdr1a/Abcb1a), Bcrp (Abcg2), and Mrp2 (Abcc2). We previously reported that P-gp mRNA is expressed in the retinas of mice that received chronic morphine (20 mg/kg for 6 d, twice daily) or vehicle (saline) via intraperitoneal (i.p.) injections (Bergum et al., 2022a). Using the RNA extracted from those same saline-treated retinas (n = 5), we performed qRT-PCR to measure Bcrp and Mrp2 expression relative to the reference genes β-actin and Tbp (Figure 1). We confirmed that Bcrp (relative gene expression [RGE] mean ± s.d.: 1.01 ± 0.12) and Mrp2 (1.02 ± 0.2) mRNA are detectable in the whole mouse retina at similar levels (t (4) = −0.11, p = 0.92; Figure 1).

We then performed immunohistochemistry (IHC) with fluorescence confocal microscopy to evaluate P-gp, Bcrp and Mrp2 expression at the iBRB of male and female mice (Figure 2). The iBRB was labeled using an antibody against occludin, a marker for the tight junctions of the endothelial cells which make up the iBRB (Morcos et al., 2001). The iBRB extends from the inner capillary bed in the ganglion cell layer (GCL), through the inner plexiform layer (IPL) and inner nuclear layer (INL), to the outer capillary bed in the outer plexiform layer (OPL) (Figure 2). The literature suggests that both endogenous opioid peptides and their receptors are predominantly expressed in the inner retina (Wamsley et al., 1981; Altschuler et al., 1982; Gallagher et al., 2010; Cleymaet et al., 2019; Berezin et al., 2022). Previous work has also suggested that P-gp and Bcrp are not expressed at the outer BRB (Chapy et al., 2016). Therefore, the outer BRB, which is established by the retinal pigment epithelium, was not analyzed in this study.

Immunolabeling of occludin and P-gp was performed in both retinal cryosections (n = 4, Figures 2A–D) and whole-mount preparations (n = 6, Supplementary Figures S1A–C) from male and female mice. The anti-occludin antibody produced robust labeling of vascular endothelium in both preparations. In cryosections, we found extensive P-gp immunolabeling exclusively co-localizing with that of occludin in the iBRB (Figures 2A–D). However, in the whole-mount retinas, we commonly observed regions of the occludin-positive vasculature with strong and some with weaker P-gp co-labeling (Supplementary Figures S1A–C). Across all the images obtained from both preparations (4–6/mouse, n = 51), the average Pearson’s correlation coefficient (r) was 0.32 ± 0.16 (mean ± s.d.). Using Costes’ automatic threshold and randomization procedures, we found this to represent true colocalization in all cases (p = 100%) (Costes et al., 2004). Colocalization parameters (r, Mander’s coefficients M₁ and M₂) were then averaged across all images obtained from a single mouse. Consistent with the observed P-gp immunolabeling patterns, the average r was significantly lower in whole-mount retinas (0.24 ± 0.12) than cryosections (0.42 ± 0.16) (p = 0.04, t-test, t (6.4) = 2.57; Supplementary Figure S1D). There was no significant difference in the Mander’s colocalization coefficient M₁ (indicating the extent of occludin-labeled area overlapping with P-gp) (Manders et al., 1993) between whole-mounts (0.20 ± 0.29) compared to the cryosections (0.41 ± 0.31) (p = 0.1, t-test, t (7.4) = 1.9; Supplementary Figure S1E). Similarly, there was no significant difference in the M₂ coefficient (indicating the extent to which P-gp overlaps with occludin) between whole-mount (0.16 ± 0.21) and cryosection preparations (0.30 ± 0.31) (p = 0.196, t-test, t (5.6) = 1.5; Supplementary Figure S1F). These results are consistent with the qualitative differences in P-gp immunostaining described above, and support the idea that while the level of P-gp expression may vary throughout the vasculature, P-gp expression in the mouse retina is confined to blood vessels. Importantly, we found no significant qualitative or quantitative (r, M₁, or M₂) differences between male and female mice (n = 7 and 3, respectively) (p > 0.05, t-test; Supplementary Figures S2A–C).

Bcrp immunolabeling in retinal cryosections from male and female mice (n = 7) also revealed robust expression of Bcrp in occludin-labeled iBRB blood vessels (Figures 2E–H). However, Bcrp labelling was not restricted to iBRB, but found in other structures in the inner retina and photoreceptor layer as well. Outside blood vessels, the anti-Bcrp antibody appeared to label neuronal projections, as well as a diffuse pattern in the OPL and GCL (Figures 2E–H). Across all images (6/mouse, n = 42), the average r was 0.46 ± 0.1. We found true colocalization in all images (p = 100%) (Costes et al., 2004). Colocalization parameters (r, Mander’s coefficients M₁ and M₂) were then averaged across all images obtained from a single mouse. Consistent with diffuse Bcrp staining in the inner retina, the average M₁ coefficient (indicating the extent of occludin-labeled area overlapping with Bcrp) was high (0.74 ± 0.13) compared to our results for P-gp. Conversely, the average M₂ coefficient (indicating the extent to which Bcrp overlaps with occludin) was more similar to the results for P-gp at 0.27 ± 0.09. As with P-gp, we found no significant qualitative or quantitative (r, M₁, or M₂) differences between male and female mice (n = 4 and 3, respectively) (p > 0.05, t-test; Supplementary Figures S2D–F). Despite the presence of Bcrp in some retinal cells outside the iBRB, its robust expression throughout the vasculature suggests it could be a significant transporter at the mouse iBRB.

Mrp2 immunostaining in retinal cryosections (n = 8) from male and female mice did not result in robust colocalization with occludin-labeled blood vessels (Figures 2I–L). Rather, it primarily labeled structures at the proximal INL, OPL and GCL, in addition to diffuse labeling of the IPL (Figures 2I–L). There were some imaged regions where Mrp2 appeared to overlap with occludin-labeled blood vessels, which we isolated for colocalization analysis (13–22/mouse, n = 145) (Figures 3A–H). We found true colocalization (p > 95%) in about a third of the 145 regions analyzed (Costes et al., 2004). Colocalization parameters (r, Mander’s coefficients M₁ and M₂) were then averaged across all negatively or positively colocalizing regions obtained from a single mouse. After accounting for paired samples (i.e., one mouse has both positive and negative regions), the average r in regions with negative colocalization (n = 90, −0.006 ± 0.03, Figures 3E–H) was significantly lower than in images with positive colocalization (n = 55, 0.10 ± 0.07, Figures 3A–D) (p = 0.0012, paired t-test, t (7) = -5.25; Figure 3I). Accordingly, the M₁ coefficient (indicating the amount of occludin overlapping Mrp2) was significantly lower in regions with no colocalization (0.04 ± 0.07) than in positive regions (0.23 ± 0.16) (p = 0.04, paired t-test, t (7) = −2.52; Figure 3J). The M₂ coefficient (indicating the amount of Mrp2 overlapping occludin) was not significantly different between images without colocalization (0.187 ± 0.166) than with colocalization (0.24 ± 0.14) (p = 0.18, paired t-test, t (7) = −1.5; Figure 3K), likely due to diffuse Mrp2 labeling. As with P-gp and Bcrp, we found no significant qualitative or quantitative (r, M₁, or M₂) differences in positively colocalizing regions between male and female mice (n = 5 and 3, respectively) (p > 0.05, t-test; Supplementary Figures S2G–I). Despite Mrp2’s occasional presence at the iBRB, its expression primarily by other cell types suggests that its principal function in the retina is not the extrusion of substrates from vascular endothelial cells. Thus, we chose to focus only on P-gp and Bcrp in our subsequent analyses.

We previously investigated the pharmacokinetics of morphine in the serum, brain and retina of male mice over 11 h after systemic morphine administration (20 mg/kg, i.p.) (Bergum et al., 2022a). Morphine content peaked in the serum, brain and retina 1 h following injection (Bergum et al., 2022a). Here, we investigated whether morphine deposition at this time point differs between the sexes and/or between females in different estrous stages. Male mice (n = 9), as well as females with a low estrogen:progesterone ratio (Low E/P; n = 8) and High E/P females (n = 20) received a single injection of morphine (20 mg/kg, i.p.) and were sacrificed 1 h later. The mean weight in the male group was 26.9 ± 1.82 g, 21.4 ± 1.17 g in the High E/P female group, and 20.8 ± 0.81 g in the Low E/P female group. Raw morphine content (ng/mL) in the plasma, brain, and retina was quantified by LC-MS/MS. The raw concentrations detected in the male mice in this study were comparable to our previously published data (Bergum et al., 2022a), with an average concentration in the brain of 47.7 ± 19.3 ng/mL (mean ± SD), 96.6 ± 12.7 ng/mL in the retina, and 346 ± 63.2 ng/mL in the plasma detected here. The data were log_e (ln) transformed to satisfy the assumptions of ANOVA. We found that tissue type, but not sex/estrus group (F = 0.68, p = 0.51), was a significant predictor of morphine content (F = 378, p < 2e-16). Raw morphine content was significantly higher in the plasma than retina in all groups: males (t (68) = 8.8, p < 0.0001), Low E/P females (t (68) = 8.2, p < 0.0001) and High E/P females (t (68) = 13.5. p < 0.0001) (Figure 4). Furthermore, raw morphine content was significantly higher in the retina than in the brain in all groups: males (t (68) = 5.3, p < 0.0001), Low E/P females (t (68) = 5.2, p < 0.0001) and High E/P females (t (68) = 9.6, p < 0.0001) (Figure 4). We found no significant sex or estrous stage differences (p > 0.05) in morphine deposition in the plasma, retina, or brain (Figure 4).

To account for differences in the tissue weight of each harvested brain/retina sample, we normalized morphine content in the retina and brain to the tissue weight (average of 4.76 ± 0.74 mg and 17.4 ± 8.77 mg, respectively). The data were again ln transformed for statistical analysis. As before, only tissue type was a significant factor (F = 1421, p < 2e-16, ANOVA). Morphine content in the retina was again significantly higher than in the brain in male mice (t (34) = 19.657, p < 0.0001), Low E/P females (t (34) = 19.218, p < 0.0001), and High E/P females (t (34) = 30.495, p < 0.0001) (Figure 5). No significant sex differences were found in morphine content in the retina or the brain (p > 0.05) (Figure 5).

We previously reported that P-gp expression in the retina and brain was not significantly different between male mice that received chronic morphine or vehicle (saline) (Bergum et al., 2022a), consistent with reports that elevated P-gp expression after morphine exposure can be attributed to a withdrawal response rather than the morphine treatment itself (Yousif et al., 2012; Chaves et al., 2016). Here, we tested whether Bcrp expression is influenced by chronic morphine exposure (20 mg/kg for 6 days, twice daily) (n = 7) compared to vehicle (saline, n = 9). All animals were sacrificed 1 h following the last injection. Expression of Bcrp, relative to the reference genes β-actin and Tbp, was calculated using the saline-treated retinas as the control group. The data were log₂ transformed to meet the assumptions of ANOVA. Tissue type, but not treatment, was a significant predictor of Bcrp expression (F = 96, p = 1.18e-7).

Similar to P-gp (Bergum et al., 2022a), Bcrp expression was lower in the retina than the brain in both the saline-treated (t (14) = −6.24, p < 0.0001) and morphine-treated groups (t (14) = −7.58, p < 0.0001) (Figure 6). Moreover, we found no significant differences between morphine- and saline-treated mice in terms of Bcrp expression in the retina (t (24.3) = −0.1, p = 0.92) and brain (t (24.3) = −1.8, p = 0.08) (Figure 6). Thus, neither P-gp nor Bcrp expression is expected to be altered by morphine administration.

We next investigated whether P-gp and Bcrp expression differs between the sexes and/or between female estrous stages in the context of acute morphine treatment (1 h after injection). Expression of P-gp and Bcrp, relative to the reference genes β-actin and Tbp, in the brains and retinas of acute morphine-treated males (n = 8), High E/P females (n = 11), and Low E/P females (n = 5) was quantified by qRT-PCR using the male retinas as the control group. The relative gene expression (RGE) values were transformed on the log₂ scale to satisfy the assumptions of ANOVA. We found that Gene (F = 253, p < 2e-16), Tissue (F = 509, p < 2e-16) and Group (F = 4.3, p = 0.027) were all significant predictors of RGE values, as were the interactions of Gene and Tissue (F = 170, p < 2e-16) and of Tissue and Group (F = 4.8, p = 0.011).

As we showed in chronically treated male mice (Bergum et al., 2022a), P-gp expression was significantly lower in the retina than the brain in all groups: males (t (63) = −14.219, p < 0.0001), Low E/P females (t (63) = −12.66, p < 0.0001), and High E/P females (t (63) = −18.422, p < 0.0001) (Figure 7). Similarly, Bcrp expression was significantly lower in the retina than the brain in all groups: males (t (63) = −2.287), p = 0.0255), Low E/P females (t (63) = −2.714, p = 0.0086), and High E/P females (t (63) = −7.695, p < 0.0001) (Figure 7).

In the brain, P-gp expression was significantly lower in males than in both Low E/P females (t (83) = -2.633, p = 0.027) and High E/P females (t (83) = -2.909, p = 0.0128) (Figure 8). The two female groups were not significantly different from one another (t (83) = 0.28, p = 0.9585) (Figure 8). As for Bcrp in the brain, we found that expression in males was significantly lower than High E/P females (t (83) = -2.63, p = 0.0272), but not significantly different than Low E/P females (t (83) = −1.5, p = 0.2830) (Figure 8). The two female groups were not significantly different from one another (t (83) = −0.65, p = 0.7926) (Figure 8). Together, these results are consistent with sex hormones influencing opioid transporter expression in some tissues, such as the brain (Tanaka et al., 2005; Suzuki et al., 2006; Bebawy and Chetty, 2009; Shchul’kin et al., 2017; Dalla et al., 2022).

Conversely, there was no effect of sex or estrous stage on retinal P-gp expression, as expression in males was not significantly different than Low E/P females (t (83) = −1.11, p = .51) or High E/P females (t (83) = −1.36, p = 0.37), nor were Low E/P and High E/P females significantly different from one another (t (83) = 0.005, p = 1.0) (Figure 8). Furthermore, we found no effect of sex or estrous stage on retinal Bcrp expression; expression in males was not significantly different than Low E/P females (t (83) = −0.56, p = 0.84) or High E/P females (t (83) = 1.82, p = 0.17), nor were Low E/P and High E/P females significantly different from one another (t (83) = 2.155, p = 0.085) (Figure 8). Thus, circulating sex hormones do not appear to significantly impact opioid transporter expression at the iBRB.

Our experimental design allowed for statistical comparison of P-gp and Bcrp expression in the retina and the brain. In the brain, we found a staggering three-fold difference in P-gp expression compared to Bcrp expression in all groups: males (t (63) = 11.9, p < 0.0001, ANOVA), Low E/P females (t (63) = 10.46, p < 0.0001) and High E/P females (t (63) = 14.306, p < 0.0001) (Figure 9). Interestingly, in the retina, the expression of the two transporters was more similar. While P-gp expression was greater than Bcrp expression in High E/P females (t (63) = 3.58, p = 0.0007), we found no such significant differences in males (t (63) = 0, p = 1.0) or Low E/P females (t (63) = .516, p = 0.61) (Figure 9).

We then examined the relationship between transporter expression in the retina or brain and the level of morphine deposition in that tissue, combining our male and female groups (n = 24) given the lack of sex/estrous effects on either parameter in the retina. Relative gene expression values and normalized morphine content (ng/mg) were left untransformed, and Spearman’s non-parametric rank correlation rho (ρ) was calculated for each gene in each tissue. In the brain, we found a weak correlation between transporter expression and morphine deposition. The correlation coefficient, ρ, between P-gp and morphine content was −0.0026 (p = 0.99), while for Bcrp it was −0.19 (p = 0.37). Conversely, we found stronger correlations between transporter expression and morphine content in the retina. For Bcrp, ρ was 0.29 (p = 0.17), while for P-gp we found a statistically significant correlation between expression and morphine deposition (ρ = 0.43, p = 0.036) (Figure 10).

There is some evidence that chronic morphine exposure may impair the integrity of the BBB, which could affect morphine pharmacokinetics in the CNS (Leibrand et al., 2019). If our morphine treatment paradigm affects the integrity of blood-CNS barriers, it may increase the permeability of the barriers to xenobiotics such as morphine. To test this, we used fluorescence imaging of fluorescein isothiocyanate (FITC) extravasation to assess the integrity/permeability of the BBB and BRB (Figure 11). Six male mice were perfused with FITC, and an additional five male mice were perfused with FITC conjugated to 10 kDa dextran (FITC-dextran) (Supplementary Figure S3). In each group, mice received either chronic morphine treatment, vehicle (saline) or were left untreated. Treated mice were sacrificed 1 h after the last injection when morphine concentrations in the brain and retina are at their peak. Barrier leakage was assessed in the retina and the paraventricular nucleus of the hypothalamus (PVN) of each mouse. Notably, the PVN was chosen due to the dense vascularization found in this hypothalamic nucleus (Frahm and Tobet, 2015). Retinal arterioles in the superficial vascular layer, as well as capillaries in the superficial, intermediate, and deep layers of the retinal vasculature were imaged (Kornfield and Newman, 2014). Leakage indices were similar across these layers, thus capillaries in the intermediate layer were chosen as representative images (Figure 11; Supplementary Figure S3). Images of FITC extravasation of the retinal arterioles can be found in Supplementary Figure S4.

We obtained similar leakage indices (as defined in the Methods) in mice perfused with FITC and those perfused with FITC-dextran (Supplementary Figure S3). The average leakage index in each tissue and treatment combination was compared by ANOVA. We found that neither tissue (F (16) = 0.24, p = 0.63), treatment (F (16) = 0.14, p = 0.87), nor the interaction of tissue and treatment (F (16) = 0.95, p = 0.41) were significant predictors of the leakage index. Accordingly, there were no significant pairwise differences in the PVN or retina of saline- and morphine-treated mice (Figure 11), nor were these different from untreated mice (Supplementary Figure S3). There were also no significant differences between leakage indices in the PVN and retina of mice in the saline-treated group (t (8) = 0.16, p = 0.88), the morphine-treated group (t (8) = 0.47, p = 0.65), nor in the untreated group (t (8) = -1.3, p = 0.22). Thus, neither chronic morphine nor vehicle treatment significantly altered the permeability of the BBB or BRB.

Adult male and female mice (8–26 weeks) were kept on a 12-h light:12-h dark cycle with lights on at 7:00 a.m. (Zeitgeber time ZT 0), fed standard chow and water ad libitum. All animal protocols were approved by the Colorado State University Institutional Animal Care and Use Committee in accordance with the NIH Guide for the Care and Use of Laboratory Animals.

Mice with Cre recombinase expressed upstream of the melanopsin (Opn4) promoter [Tg (Opn4-cre)SA9Gsat/Mmucd, #036544-UCD; Opn4-cre] were purchased from the Mutant Mouse Resource and Research Center (MMRRC) at the University of California at Davis. Opn4-cre mice were backcrossed into a 100% C57BL/6J background prior to purchase and maintained as hemizygotes (Opn4-cre +/−).

McKO mice, in which only ipRGCs lack functional MORs, were generated as described previously (Weibel et al., 2013; Cleymaet et al., 2021; Bergum et al., 2022b; Berezin et al., 2022). Oprm1 ^fl/fl breeders (Jackson Labs strain #030074) were generously provided by Dr. Brigitte Kieffer (Douglas Research Center, McGill University). Oprm1 ^fl/fl mice have exons 2 and 3 of the μ-opioid receptor (Oprm1) gene flanked by loxP sites and were maintained on a 50% C57BL/6J-50% 129Sv background for 5 generations before receipt (Weibel et al., 2013). Oprm1 ^fl/fl and Opn4-cre mice were crossed to produce 50% Oprm1 ^fl/fl mice and 50% McKO mice with the floxed Oprm1 gene on both alleles and Opn4-cre on one allele. This Mu^fl/McKO line was maintained on a background containing at least 75% C57BL/6J and 25% 129Sv. Male and female McKO mice, and their littermate controls, were used for the mass spectrometry and qRT-PCR experiments in the context of morphine treatment. Mice of this mixed background have been used in previously published studies on retinal morphine pharmacokinetics, and the pharmacokinetics of morphine in the retina and the brain were not found to be significantly different between McKO mice and their wild-type (C57BL/6J) counterparts (Bergum et al., 2022b; Bergum et al., 2022b).

Mice carrying a bacterial artificial chromosome (BAC) in which the melanopsin (Opn4) promoter drives expression of enhanced green fluorescent protein (EGFP) [Tg (Opn4-EGFP)ND100Gsat/Mmucd] were obtained from the Gene Expression Nervous System Atlas (GENSAT) project. These mice are hereafter referred to as Opn4:EGFP mice and were maintained on a C57BL/6J background. Untreated male and female Opn4:EGFP and Opn4-cre mice, and their littermate controls, were used for immunohistochemical and quantitative reverse transcription PCR (qRT-PCR) experiments, as both strains are on the commonly used C57BL/6J background.

Opn4-cre mice were crossed to Ai14 (Jackson Labs strain #007914) mice to express tdTomato following Opn4-driven Cre-dependent recombination (Opn4-cre-tdTomato). Opn4-cre-tdTomato mice were on a 100% C57BL/6J background. Male Opn4-cre-tdTomato mice, and their littermate controls, were used in the fluorescence extravasation studies. Positive Opn4-cre-tdTomato expression in the ganglion cell layer was used to guide the imaging of distinct retinal vasculature layers and did not affect FITC extravasation.

Procedures were as previously described (Berezin et al., 2022). Mice were anesthetized with Fluriso (isoflurane, VetOne) and sacrificed by cervical dislocation. After enucleation, the eyes were dissected in 0.1 M phosphate buffered saline (PBS; pH 7.4) and fixed in freshly prepared 4% paraformaldehyde (PFA) in PBS. For cryosectioning, eyecups were immersed then in 30% sucrose at 4°C for several days prior to being embedded in Optimal Cutting Temperature (O.C.T.) Compound (Tissue-Tek). The tissue was sectioned at 20 μm on a ThermoFisher CryoStar NX50 cryostat, directly mounted on SuperFrost Plus slides (Fisher Scientific), and stored at −20°C until use. For whole-mount preparations, immunohistochemistry was performed immediately following fixation. Both tissue types were subject to antigen retrieval. As previously described (Gallagher et al., 2012), the tissue was incubated in boiling 10 mM sodium citrate in PBS for 15 min followed by incubation in 5% sodium borohydride in PBS for 45 min. The tissue was then incubated in a blocking solution (5% serum and 0.5% Triton X-100 in PBS) for 3 h on a shaker table at RT. Retinas were incubated in primary antibodies diluted in the blocking solution for 24–36 h on a shaker table at RT (rabbit anti-P-glycoprotein 1:200, Abcam [EPR10364-57] #ab170904; rat anti-Bcrp 1:50, Enzo [BXP-53] #ALX-801–036; rabbit anti-Mrp2 1:100, Bioss #bs-1092R; mouse anti-occludin 1:250, Invitrogen [OC-3F10] #33–1500). Sections were then washed in PBS and incubated, either at RT for 2 h or overnight at 4°C, in the appropriate secondary antibodies. After the final washes in PBS, retinas were stained with DAPI (Advanced Cell Diagnostics, Newark, CA) and mounted in Vectashield Plus Antifade Mounting Medium (Vector Laboratories, Burlingame, CA).

Procedures were performed as previously described (Bergum et al., 2022a; Berezin et al., 2022), with minor modifications described below.

Male mice received i.p. injections of chronic morphine or vehicle (saline) as described in Section 4.2.1. One hour following the last injection, mice were deeply anesthetized with Fluriso (isoflurane, VetOne) and kept anesthetized by continuous isoflurane delivery via a nose-cone. Mice were transcardially perfused with 10 mL of FITC (fluorescein isothiocyanate, 1 mg/mL, ThermoScientific #46424) or FITC-Dextran (fluorescein isothiocyanate dextran 10 kDa, 0.5 mg/mL Fisher Scientific #F0918100MG) dissolved in PBS (0.1 mg/mL) at a rate of 4 mL/min using a syringe pump (New Era Pump Systems NE-300). Mice were then perfused with 15 mL of 4% PFA in PBS. The eyes were enucleated and post-fixed in 4% PFA at RT for 25 min (20 min puncture-fix, retinas dissected, 5 min fixation). Retinas were washed in PBS and then mounted in Vectashield Plus Antifade Mounting Medium (Vector Laboratories, Burlingame, CA). Whole brains were removed and post-fixed in 4% PFA at 4°C for at least 24 h and then cut into 100 μm coronal sections on a Leica vibratome (VT1200S).

To evaluate the permeability of the blood-brain barrier, we assessed the densely vascularized paraventricular nucleus of the hypothalamus (PVN). Z-stacks of PVN vasculature (≥20 μm) were taken in 1 μm increments on either side of the third ventricle. To evaluate the permeability of the blood-retina barrier, we obtained one to two representative Z-stack images of the central arterioles and venules and one to two representative Z-stack images of the capillary vasculature of the peripheral retina (all in 1 μm increments). Retinal capillaries were imaged from the superficial plexus in the ganglion cell layer through the apparent end of the retinal vasculature (≥20 μm). All images were ∼320 μm² and obtained with a ×20 air objective on a Zeiss LSM 900 confocal microscope (Carl Zeiss, Oberkochen, Germany).

Images were denoised using rolling ball background subtraction (25 pixels) followed by the Despeckle and Remove Outlier (3 pixel radius, threshold = 10) functions in Fiji (Schindelin et al., 2012). Maximum intensity Z-stack projections were created from each image (retinal capillary images were split into two stacks: one from the superficial plexus to the intermediate plexus, and the second from the intermediate plexus to the deep plexus terminating in the outer plexiform layer). Each maximum projection image is composed of 5–10 single optical sections taken at 1 μm increments, and thus represents approximately 5–10 μm of tissue vasculature. Images were thresholded using the Li method in Fiji and converted to binary masks. A composite image of each mask and the denoised image were visually inspected to ensure accurate representation of the imaged vasculature. Barrier leakage was quantified as previously described (Frahm and Tobet, 2015). The mask containing vessel “objects” was redirected to the denoised image and the average intensity within each vessel was measured. Each vessel “object” was expanded by 2 μm and the region outside each vessel was saved; the average intensity in each outside region was measured. Leakage index was defined as the average intensity outside the vessel divided by the average intensity inside the vessel. The final reported leakage index is the average across all vessels in all images of the PVN or retina of each mouse.

In order to determine whether FITC type should be included as a factor in our statistical model, we compared two mixed models of the data by ANOVA: 1) with tissue and treatment as dependent variables, and 2) with tissue, treatment, and FITC type as dependent variables. Both models had leakage index as the independent variable and had samples identified by mouse ID number. We found that the more complex model (2) was not significantly better than model (1) without FITC type (p = 0.223): the simpler model (1) had a lower Akaike Information Criterion (AIC; −65.8 compared to −62.1) and Bayes Information Criterion (BIC; −57.1 compared to-46.8).

All data processing, visualizations and statistical tests were performed using R (v4.2.1; R Core Team, 2022) in RStudio (v2022.7.2.576; R Studio Team, 2022). Data visualizations were made using the ggplot2 package in R (Wickham, 2016). For all tests, p < 0.05 was considered significant. T-tests and ANOVAs with post hoc Tukey adjustments were used when data were normally distributed and (if applicable) had homogeneous variance across groups. Whether the data was transformed (e.g., log₂ or ln) to satisfy these assumptions is noted in the text. Linear mixed-effects models were fit using the lme4 package in R (Bates et al., 2015), and pairwise comparisons were calculated using the emmeans package in R (Lenth, 2022).