Non‐invasive measurement of retinal permeability in a diabetic rat model

The gold standard for measuring blood‐retinal barrier permeability is the Evans blue assay. However, this technique has limitations in vivo, including non‐specific tissue binding and toxicity. This study describes a non‐toxic, high‐throughput, and cost‐effective alternative technique that minimizes animal usage.


| INTRODUC TI ON
Selective permeable barriers are vital for organ function and homeostasis. Such barriers include the EC barrier of blood vessels, the BBB, 1 the BRB, 2 the blood-spinal cord barrier, 3 and the blood-placental barrier. 4 Selective transport of micromolecular substances across these barriers occurs under physiological conditions but they are relatively impermeable to macromolecules such as albumin. 5 Evans blue (EB) dye has a high affinity for albumin 6 and has a poor ability to cross selective permeable barriers under normal circumstances and remains predominantly within the blood circulation. 5 The breakdown of the barrier, for example, via trauma or cytokine release, causes leak of dyes such as this across blood vessel walls including the BBB/BRB. When diseases such as diabetic retinopathy, 7 sepsis, 8 capillary leak syndrome, 9 or cancer 10 result in disruption of these barriers, there is an increase in vascular permeability and dyes such as EB may extravasate from the circulation into surrounding tissues. Accumulation of EB dye can be extracted from stained tissues and quantified by spectrophotometry. 11,12 Evans blue is extensively used to estimate changes in vascular permeability in a number of models including in stroke, 13 cerebral ischemia, 11 skin 14 (often referred to as a Miles assay, 15 endothelial damage caused by trauma, 16 and the breast cancer brain metastasis model. 10 Evans blue has also been used to measure breakdown in the blood-spinal cord barrier 17,18 and the BRB. 7,19 The EB assay is rapid and sensitive and has become a widely used method for estimating barrier integrity and vascular permeability. However, this methodology has a number of significant assumptions associated with its use which can lead to significant over-simplification of the research findings if unaccounted for. These include (a) the contribution of convective flux of albumin (and dye bound albumin) to the total flux varying from minimal to highly significantly dependent on hemodynamic parameters rather than barrier properties, (b) a substantive percent of free dye present in the animal following administration, resulting in flux being highly blood flow-dependent rather than barrier-dependent, 1,20 (c) lack of specific binding to albumin, (d) injection of dye dissolved in physiological solutions affects the structure of the dye, 21 (e) EB binds albumin when it is in the tissues, (f) change in tissue clearance can account for the changes, (g) problems with spectroscopic methods that have been used to estimate the amount of dye, and (h) in vivo toxicity. 20,22 These issues have been highlighted many times and are extensively discussed elsewhere.
This study describes a non-toxic and high-throughput alternative technique to the EB assay in the retina. In combination with the Micron IV retinal imaging microscope system (Phoenix Research Systems), we have used FFA and a novel post imaging analysis methodology described below. This technique can be carried out in less than 15 minutes, is minimally invasive, and can be repeated in animals, resulting in a reduction of the number of animals required when compared to the EB assay, which takes 2-3 hours to carry out per animal and requires multiple cohorts of animals per time point. In addition, we have estimated that our novel methodology is significantly more efficient than the EB assay to perform.

| Gender selection
Male Brown Norway rats, which are more susceptible to STZinduced diabetes (250-300 g, Envigo), were included in this study.

| STZ dose and route of administration
A total of 40 male Brown Norway rats were weighed and given a single i.p. injection of STZ (Sigma-Aldrich). An additional 14 control rats were administered 300 μL saline i.p.

| Fast or non-fasted prior to STZ dosing
Fasting can be used to minimize competition between STZ and glucose for low-affinity GLUT2 transporters on β-cells. In the period following STZ dosing, the animals enter a hypoglycemic state and begin to show signs of sudden weight loss, polyuria, and dehydration. In our initial pilot study, animals were fasted overnight, prior to dosing in the morning, but their weights dropped dramatically during that period and this had a subsequent effect on weight loss in the proceeding study, exceeding a 20% weight loss moderate severity threshold. Therefore, in the subsequent study we did not fast prior to dosing. We ran the risk of the animals not becoming diabetic (~10% of cases) but this allowed the animals to feed, retain, or increase their weights prior to dosing and avoid the subsequent decline in health and weight otherwise observed. Animals with weights exceeding 300 g tolerated the 50 mg/kg STZ dose and maintained weights throughout the study or until the point of insulin supplementation.

| Sucrose enrichment
In addition to water, a 15% (w/v) sucrose solution was made available in a separate drinking bottle to alleviate the initial hypoglycemic spike following STZ induction. The volume of sucrose intake was monitored over a 72-hour period.

| Insulin supplementation
Current blood glucose control in diabetic rodent models focuses on maintaining the diabetic animal in a state of moderate hyperglycemia, with normal weight gain in the absence of severe ketonuria.
This state can be achieved by once-daily injections of titrated insulin doses or by implantation of a continuous release insulin pellet.
To reduce animal stress by repeated injections, we subcutaneously implanted one third of a single insulin pellet (LinShin). The 7-mmlong implant has a diameter of 2 mm and is designed to facilitate handling and insertion. Upon implantation, gradual erosion of the implant starts at once and the effect of released insulin on the blood glucose level can be detected in <1 hour. Unlike injectable insulin, the implant releases a set basal dose of insulin every hour. Therefore, an animal given an optimal implant dose should show no glucosuria and ketonuria, which are both difficult to prevent by daily insulin injections, due to the action of the insulin lasting only part of the 24-hour cycle.

| Blood glucose measurement
On days 0, 4, and prior to sacrifice (day 28), blood glucose levels were tested using a sample of blood taken from the tail vein and an Accu-Chek blood glucose monitor. Rats with blood glucose levels of >15 mmol/L and above were deemed hyperglycemic. Streptozotocininjected rats that did not become hyperglycemic on day 4 were reinjected with STZ the following morning and subsequently evaluated for diabetes, as outlined above.

| Optimization of anesthetic knockdown
Due to the severity of the diabetic model and in combination with injectable anesthetics, gaseous anesthesia (halothane) was trialed.
However, this had constraints. Firstly, the small animal gaseous mask sat just underneath the eye making it very difficult to align the retinal microscope with the eye without dislodging the gaseous mask and risking the animal becoming lucid. Secondly, gaseous anesthesia was not sufficient to prevent the eye from responding to the light source, blinking and rolling during imaging, without giving a high percentage of halothane. Injectable anesthetic removed the cumbersome equipment required for gaseous knockdown, the control of multiple flow meters, and ultimately stabilized the animal to prevent the eye from moving during FFA. We therefore used an injectable anesthetic regimen combining ketamine hydrochloride, 37.5 mg/kg (Zoetis) and medetomidine hydrochloride, 0.25 mg/kg (Produlab Pharma BV) i.p.

| Optimization of Na-Fl dose and volume
Na-Fl (MW 376.27) was prepared in sterile water, and 0.2 μm filtered, and 0.1, 10, 100 mg/mL dilutions were prepared and stored at room temperature and away from direct light until required.
Intraperitoneal injections were administered in a non-injected lower quadrant of the abdomen while the anaesthetized rat was on the imaging cradle and the retina of the right eye had been previously located and aligned using the Micron IV. This was achieved by raising the right back leg, maintaining retinal alignment by minimal animal movement. The retinal imaging software was set to record prior to injection and continued to record until the dye had reached a saturation level in the retina (~3 minutes).

| Optimization of Na-Fl route of administration
Two routes of administration, intraperitoneal and intravenous, were tested with all three doses of Na-Fl. The tail was pre-warmed in a beaker of warm water (30-35°C) to dilate the blood vessels, and an intravenous injection was administered via the lateral tail vein.

F I G U R E 1 In vivo study overview and timeline. Male Brown
Norway rats were given a single dose of STZ (50 mg/kg) i.p. Nondiabetic controls were given an equivalent volume of sterile PBS i.p. On day 4, a tail vein blood sample was obtained and blood glucose levels were measured. Diabetic animals were implanted with an insulin pellet, and FFA was carried out on the left eye and in the same retinal position on days 0, 7, 14, 21, and 28. On day 28, animals were humanely killed by sacrifice and both eyes enucleated

| FFA analysis
Angiograms were imported as avi files into Fiji software, 23 and mean fluorescence intensity was measured in a major retinal vessel and nearby tissue (which includes unresolved capillaries) every 200 frames up to 2400 frames. An initial time course was plotted, and only the region where there was detectable tissue fluorescence but no major vessel saturation was used for analysis.

| Immunofluorescence
On day 28, animals were sacrificed and both eyes enucleated and immediately fixed in 4% PFA for 30 minutes. Retinae were excised from the scleral/choroidal cup and stained with an endothelial cell-specific marker, isolectin B 4 (lectin from Bandeiraea simplicifolia conjugated to biotin) followed by Alexa Fluor 488-streptavidin.
Whole-mounted retinae were imaged using confocal microscopy (Leica TCS SPE) to generate z stacks, and allow for analyses of vascular structural parameters across the superficial and deep plexuses.
Length per volume and mean radius were measured using Fiji software 23 by manually drawing along all vessels using the freehand line tool. The distance between the upper and lower plexus was taken as the vascular retinal thickness.

| Statistical analysis
Unless otherwise stated, all data are shown as mean ± SEM. All data, and graphs were formulated with Microsoft Excel (Microsoft Office Software), GraphPad Prism v7/8 (GraphPad Software Inc), Fiji, and Imaris. Vascular spacing statistical analysis was calculated using a one-way ANOVA with Bonferroni's correction, and 28-day permeability analysis was calculated using a two-way ANOVA with Bonferroni's correction. All results were considered statistically significant at *P < .05, **P < .01, ***P < .001, ****P < .0001.

| RE SULTS
To quantify the effective solute permeability (P s ) for the vascular wall, the parameter derivation in Appendix 1 was used, which is directly comparable to previous derivations from single capillaries. 24 To enable this calculation to take place, the fluorescence intensity was measured in one of the major retinal blood vessels and in a box outside the major retinal vessel containing no visibly distinct vessels (ie, no vessels larger than approximately 20µm in diameter, Figure 2A). The permeability to fluorescein (P fluorescein ) was estimated as described in Appendix 1 using FFA ( Figure 2B). The calculations make the assumptions that the surface area available for exchange is the same in the diabetic animal as the control.
To determine these parameters tissue was stained to quantify the retinal vasculature. The retina has two clear plexuses ( Figure 3A), one at the surface and one within the retinal layer. The vascular length ( Figure 3B) and radius ( Figure 3C) were calculated. There was no significant difference between the mean radius, the distance between the two plexuses ( Figure 3D), nor the length of vessel per confocal volume.
We therefore calculated the permeability using cohort parameters and assuming that the surface area was unchanged throughout for each cohort. Figure 4A shows that while permeability remained constant in non-diabetic animals, there was a significant and progressive increase in permeability in diabetic animals.
To determine whether permeability could be linked, the data for all three diabetic and six controls over 28 days are shown ( Figure 4B).
These results show that the increase in permeability is progressive and that paired (linked) analysis can be used to determine a progressive increase in permeability with diabetic duration, allowing for a substantial reduction in number of animals per group as well as a reduction in overall numbers.   of the tracer will be an important practical factor without creating a F I G U R E 4 Retinal Na-Fl permeability P t=0 Fluoresein significantly increases over 28 d in diabetic Brown Norway rats and remains unchanged in non-diabetic controls. (A) Permeability measurements from 11 diabetic and 14 non-diabetic rats. Bars show mean ± SEM. Paired one-way ANOVA, with Sidak post hoc test. (B) Paired results from three diabetic and six non-diabetic rats followed for four time points. Statistical analysis using a two-way ANOVA, with Bonferroni's post hoc test. # P < .05 ## P < .01, #### P < .0001. **P < .01 and ****P < .0001. # Significance between time points; *Significance between non-diabetic vs diabetic per time point sharp bolus (ie, more invasive carotid injection). 10 kDa (eg, a chemokine) to 70 kDa (eg, albumin) molecules would be feasible sizes to test, if a stable version bright enough to observe orders of magnitude lower fluxes was achievable. Their flux fold change would be measurable yet likely markedly different to Na-Fl itself. The relative comparison may elude to which layers within the wall are disrupted in pathologies. 35 This would provide a better understanding of the flow dynamics and size selectivity but would still not be able to determine vascular hydrostatic or osmotic pressures. The change of ratio between the mid-sized and Na-Fl molecules would allow a more detailed understanding of the specific vascular wall layer that is disrupted in DR pathology. S is the exchange area of the capillaries. ΔC = Difference in concentration between lumen C V and tissue C T but if we assume that initially C T = 0 and throughout C T ≪ C V :

| PERME AB ILIT Y QUANTIFIC ATION
If L Tot = total length and r = mean radius of the vessels within the tissue imaging volume V 1 (see Figure 2A), the exchange area (S) within V 1 becomes: V 1 has a total number of molecules (N V 1 ) being the sum of the molecules in the tissue (N T ) and capillaries (N c ) within it: Hence: When substituted in Equation A1 gives: As all imaging volume V 2 is encompassed by the major vessel, the mean number of molecules in the unit volume is C V .
where <·> denotes the mean over the respective volume. Further, the number of molecules can be assumed to be directly proportional to fluorescent intensity (I) measured throughout (ie, I ∝ N); I therefore substitutes for N to give: If we assume the capillaries have the same solute concentration as the major vessel then: Subsequently: Substituting Equation A12 into Equation A10 gives: