Prolonged systemic hyperglycemia does not cause pericyte loss and permeability at the mouse blood-brain barrier

Diabetes mellitus is associated with cognitive impairment and various central nervous system pathologies such as stroke, vascular dementia, or Alzheimer’s disease. The exact pathophysiology of these conditions is poorly understood. Recent reports suggest that hyperglycemia causes cerebral microcirculation pathology and blood-brain barrier (BBB) dysfunction and leakage. The majority of these reports, however, are based on methods including in vitro BBB modeling or streptozotocin-induced diabetes in rodents, opening questions regarding the translation of the in vitro findings to the in vivo situation, and possible direct effects of streptozotocin on the brain vasculature. Here we used a genetic mouse model of hyperglycemia (Ins2AKITA) to address whether prolonged systemic hyperglycemia induces BBB dysfunction and leakage. We applied a variety of methodologies to carefully evaluate BBB function and cellular integrity in vivo, including the quantification and visualization of specific tracers and evaluation of transcriptional and morphological changes in the BBB and its supporting cellular components. These experiments did neither reveal altered BBB permeability nor morphological changes of the brain vasculature in hyperglycemic mice. We conclude that prolonged hyperglycemia does not lead to BBB dysfunction, and thus the cognitive impairment observed in diabetes may have other causes.

streptozotocin and similar drugs shows extensive inter-individual variability and depends on administration method, dosage, genetic background and other factors 10 . Finally, visualization and quantification of BBB permeability is still challenging, and different protocols may lead to different results. Therefore, whether hyperglycemia causes BBB dysfunction remains an open question. The usage of genetic models of DM, such as the Ins2 AKITA mouse, constitutes a novel approach for studying hyperglycemia complications in vivo. The Ins2 AKITA mouse has a point mutation in the Ins2 gene resulting in a conformational change in the protein that leads to its accumulation in pancreatic beta cells causing cell death 11 . This toxicity is completely specific to the Ins2-expressing beta cells, thus mimicking human type 1 DM. Heterozygous male Ins2 AKITA mice display a non-obese phenotype and develop consistent hyperglycemia, hypoinsulinemia, polydipsia, and polyuria around the age of 4 weeks 11 . Thus, the Ins2 AKITA mouse constitutes a clear advance over chemically-induced DM, as the primary insult is beta cell-specific. Additionally, this mouse model is stable and reproducible and hence allows longitudinal studies of hyperglycemia complications.
In our study, we sought to determine whether prolonged hyperglycemia causes BBB dysfunction and leakage in the mouse brain by using several state-of-the-art methodologies to characterize BBB dysfunction. We conducted our studies in Ins2 AKITA heterozygous males and littermate wild-type controls. The Ins2 AKITA mice yielded consistent, prolonged hyperglycemia throughout the study. BBB leakage was studied in long-term hyperglycemic animals by injecting different exogenous tracers intravenously 12 . BBB integrity was studied by quantifying and visualizing these injected tracers in the CNS 12,13 . Complementarily, we analyzed whether hyperglycemia leads to transcriptional or morphological changes in the mouse brain microvasculature. In sum, these methods failed to demonstrate increased BBB permeability in Ins2 AKITA mice.
Collectively, our results -along with the study by Corem and Ben-Zvi (submitted) -lead us to conclude that persistent systemic hyperglycemia per se does not cause BBB permeability in the mouse brain. In the light of novel studies in human DM patients showing an association with BBB leakage and dementia 14 , our study suggests that factors other than hyperglycemia contribute to BBB dysfunction.

Material and Methods
Animals. Male heterozygous Ins2 AKITA mice 11 (referred here as Ins2 AKITA ) and littermate wild-type controls (referred here as WT) were used for this study. Mice were purchased from Jackson Labs (Bar Harbor, ME) on a C57Bl/6-J background and bred in house at the Scheele Animal House, Karolinska Institute. Animals were housed in Allentown XJ type II long cages (Allentown, NJ), with 12/12 h light-dark cycle and access to water and standard chow ad libitum. Cage bedding was changed 2-3 times per week due to polyuria. Glycemia (Bayer Contour, Solna, Sweden; range 0.6-33.3 mmol/l) and body weight were monitored weekly from 6 weeks of age. Mice were sacrificed between 26 to 38 weeks of age, and samples collected. All animals were sacrificed between 10-11 AM. The procedures were carried out in accordance with institutional and national Swedish policies following approval from the Animal Ethical Board of Northern Stockholm.
Quantification of pericyte coverage. Pericyte coverage was quantified as previously reported 15 . Briefly, 3D confocal images were projected with maximum intensity projection from 10 μm thick z-stacks, contrast enhanced, de-speckled, smoothened, and a threshold was set to generate a binary image. These images were then subjected to 'dilate' and 'close' commands prior to running the AnalyzeSkeleton plug-in in Fiji (https://fiji.sc/). Total skeletal length was calculated for both PECAM1 stained images (all vessels) and ANPEP stained images (pericyte-covered vessels), and the ratio of the vessel lengths was used as a measurement of coverage. One to three fields per mouse were acquired with 20× objective from cerebral cortex of seven WT and six Ins2 AKITA mice. All image processing was done automatically with a custom macro designed in Fiji.
Quantification of microglia processes. Three fields per mouse were acquired with 40× objective from cerebral cortex of six WT and seven Ins2 AKITA mice. Ten μm thick confocal stacks of AIF1 stained brain sections were analyzed with Imaris ×64 8.3.1 software (Bitplane, Belfast, UK). Microglia filament length in μm was measured in 3D-mode by filament module.

Intravenous injection and detection of leakage tracers.
To address BBB permeability, we injected intravenously either Evans Blue dye (EB, 2% in PBS, Sigma Aldrich, Saint Louis, MO) or lysine-fixable cadaverine conjugated to Alexa Fluor 488 or 555 (1 KDa, 5 mg/ml in 0.9% NaCl, Invitrogen). Circulation time was overnight for EB and 2 h for cadaverine. As a positive control for EB leakage we used pdgfb ret/ret mice 12 . Animals were subsequently perfused transcardially with PBS for 5′ and brains removed. Successful cadaverine injections were verified by examining kidneys of injected animals under a fluorescence stereomicroscope (Leica Microsystems). Quantification of extravasated EB or cadaverine in the brain parenchyma was performed as published 12,16 .
Microvasculature isolation for quantitative PCR and RNA sequencing. Purification of microvasculature was performed as described 17 with modifications. Briefly, after collagenase A treatment, brain tissue was first gently pressed through a 100 μm and then through a 40 μm cell strainer. The tissue homogenate was incubated with rat anti-PECAM1 antibody-coated magnetic beads (Dynabeads, Invitrogen, cat. #11035) at 4 °C for 1 h. Microvascular fragments adherent to magnetic beads were washed 6 times with HBSS containing 1% BSA followed by 2 times with HBSS alone. Microvasculature fragments were lysed in 350 μl RLT buffer (Qiagen, Hilden, Germany).
Quantitative PCR. Total RNA was isolated from whole cerebral cortex or purified brain microvasculature using RNeasy Mini kit (Qiagen). RNA was DNase-treated (Invitrogen) and samples were cleaned using RNeasy Mini-elute Cleanup Kit (Qiagen). cDNA was synthesized from 1 ug of total RNA using Superscript III (Invitrogen) and mRNA expression quantified by Taqman assays (Applied Biosystems) for mouse Pdgfrb (Mm00435546_ RNA sequencing data analysis. The RNA-seq libraries were prepared as previously described 18 20 . Gene expression levels in brain microvasculature between the Ins2 AKITA and WT mice were compared, and the genes with multiple-test corrected p < 0.05 were identified as significantly differentially expressed genes. The sequence data is deposited in NCBI GEO, accession number GSE113919. Statistical analysis. Data are presented as mean ± SEM. For two group comparisons two-tailed, unpaired student's t test was performed (Prism v5.01, GraphPad, La Jolla, CA). A p-value of ≤0.05 was considered statistically significant.

Results
The effect of hyperglycemia on blood-brain barrier permeability. We first checked blood glucose levels in WT and Ins2 AKITA mice. As reported elsewhere 21-23 , Ins2 AKITA animals had elevated glycemia at 6 weeks of age as compared to WT mice (Fig. 1a). By 11 weeks, Ins2 AKITA animals were consistently hyperglycemic (>20 mmol/l), whereas WT mice remained normoglycemic (<10 mmol/l) throughout the experiment (Fig. 1a). Due to hypoinsulinemia, body weight in Ins2 AKITA mice reached a plateau by 7 weeks of age (≈20 g), while WT animals continued to gain weight normally (Fig. 1b).
We then sought to address the effect of prolonged hyperglycemia on BBB permeability. We used two well-established methods to quantify vascular permeability in brain tissue 12 . Firstly, we injected the fluorescent tracer cadaverine-Alexa Fluor 488 or 555 (1 kDa) intravenously and perfused the animals after the dye had been circulating for 2 h. We could not detect a significant difference in tracer accumulation in brains between Ins2 AKITA and WT mice, either by microscopical visualization (Fig. 2a,b) or by fluorescence quantification (Fig. 2c). Additional analyses of BBB permeability in older mice (38 weeks of age) also failed to detect increased cadaverine extravasation into the brain of hyperglycemic mice (Fig. 2d). Secondly, we investigated whether larger molecular weight tracer and longer circulation time could reveal a compromised BBB function upon hyperglycemia. Thus, we intravenously injected the azo dye Evans Blue, which binds to albumin in the circulation, and allowed it to circulate overnight. As shown in Fig. 2e, quantification of Evans Blue accumulation in the brain was not significantly different between Ins2 AKITA and WT mice. As a positive control for BBB permeability, we quantified Evan's blue extravasation in Pdgfb ret/ret mice, a model of pericyte deficiency that reproducibly shows tracer leakage into the brain 12 (Fig. 2e). The results obtained by these analyses lead us to conclude that long-term hyperglycemia does not cause any significant vascular leakage in the mouse brain. Study of pericyte coverage of Ins2 AKITA and WT brain capillaries. Pericyte loss is one of the earliest events in diabetic retinopathy 24,25 , therefore we sought to analyze pericyte coverage of brain capillaries in hyperglycemic and WT mice. First, we analyzed pericyte coverage by immunohistochemistry (Fig. 3a,b). We stained endothelial cells by PECAM1 and pericytes by ANPEP and DES. We also analyzed the abundance of pericyte-derived extracellular matrix proteins VTN and LAMA2 12,18 . These immunostainings did not reveal any vascular abnormalities or aberrant pericyte coverage of brain capillaries from both Ins2 AKITA and WT mice (Fig. 3a,b). Furthermore, none of the four pericyte markers analyzed displayed abnormal expression as judged by their pattern of immunoreactivity (Fig. 3a,b). Unexpectedly, pericyte coverage, quantified as the percentage of the PECAM1 + endothelium covered by ANPEP + pericytes, was significantly increased in Ins2 AKITA as compared to WT (Fig. 3c). We then obtained mRNA from isolated microvascular fragments, containing endothelial cells, pericytes, and astrocyte end-feet 17 . In agreement with our immunohistochemistry results, we did not detect differences in mRNA expression of pericyte-specific genes (Pdgfrb and Rgs5) by quantitative, real-time PCR (Fig. 3d,e).
Collectively, these results suggest that hyperglycemia does not cause decreases in pericyte coverage or changes in pericyte-specific gene or protein expression in the mouse brain.
Brain vascular transcriptome analysis. To take a more comprehensive effort in characterizing the brain vasculature upon hyperglycemia, we extracted mRNA from brain microvasculature fragments isolated from Ins2 AKITA and WT mice and analyzed their transcriptome by RNA sequencing (Table 1). Twenty-three genes were significantly changed in Ins2 AKITA as compared to WT. One gene, Pin1, was found to be upregulated in Ins2 AKITA mice, whereas the remaining 22 were downregulated. We checked the cellular origin of the 23 dysregulated genes according to the recently described molecular atlas of the mouse brain vasculature 13 . Only 4 genes were clearly associated with endothelial cell expression (Table 1), while the remaining 19 had a broad expression pattern spanning at least 2 different cell types. We did not find differentially regulated genes related to the tight junctions at the BBB, nor known genes mediating endothelium-pericyte interactions. Of relevance for the hyperglycemic status of the Ins2 AKITA mice, Slc2a1 (Glut1) was 1.8-fold downregulated in Ins2 AKITA (Table 1). Immunohistochemistry analysis of SLC2A1 did not reveal an aberrant pattern of expression, however, a slightly weaker immunofluorescent staining was noticed in Ins2 AKITA as compared to WT (Fig. 4).

Analysis of glial cells.
Hyperglycemia has been reported to cause early morphological and gene expression changes in glial cells, particularly in astrocytes and microglia 26 . We analyzed astrocytes by GFAP immunostaining. Additionally, we measured Gfap mRNA expression in the cerebral cortex homogenates and from isolated brain microvasculature fragments. GFAP immunohistochemistry did not reveal any differences between Ins2 AKITA and WT brains (Fig. 5a). Gfap mRNA expression was found to be unchanged in Ins2 AKITA mice when compared to WT (Fig. 5b,c). We conclude that hyperglycemia does not cause major changes in astrocytic GFAP expression in the mouse brain.
We then studied microglial cells by immunohistochemistry on brain sections from Ins2 AKITA and WT mice. AIF1 + immunoreactivity was readily detected (Fig. 6a), with no differences in AIF1 + cell numbers between Ins2 AKITA and WT mice (Fig. 6b). However, the morphology of the AIF1 + cells was remarkably different. In WT, the cell processes were long and highly ramified (Fig. 6a). In contrast, Ins2 AKITA AIF1 + cells had an apparent larger cell body and significantly shorter and less ramified processes (Fig. 6a,c). These morphological changes are consistent with the description of "activated" retinal microglia caused by hyperglycemia 23,26 .

Discussion
The question whether hyperglycemia causes BBB permeability has been addressed in numerous reports 6 . Unfortunately, however, differences in experimental design and divergent and sometimes contradictory results between studies preclude consistent conclusions. This is not entirely surprising, however. The extremely delicate BBB architecture in vivo, including important features such as blood flow and its metabolic regulation (neurovascular coupling), makes it impossible to extrapolate in vitro results to the in vivo scenario. Previous animal studies are also problematic, since they applied chemically induced hyperglycemia 7,8 . Although popular, this technique has problems with specificity 9,27-29 and variability as it depends on administration route, dosage, and genetic background 10 . The advent of genetic mouse models, such as the Ins2 AKITA mouse 11 , offers the opportunity of studying the effect of hyperglycemia without many of the above-mentioned confounding factors. To date, a few reports have addressed the role of hyperglycemia on the blood-retinal barrier (BRB) in Ins2 AKITA mouse. Two studies found early signs of retinopathy including pericyte loss and increased BRB permeability 21,30 . Subsequently, it was shown that the presence of the rd8 mutation in the Crb1 gene -causing retinal degeneration 31 -might account for the In the present study, we aimed at evaluating hyperglycemia-induced BBB permeability by employing highly specific tools. Firstly, a range of well-established techniques for the study of BBB and neurovascular unit (NVU) integrity 12,13 , and secondly, a robust mouse model of hyperglycemia, the Ins2 AKITA , were used. Our data clearly show that prolonged systemic hyperglycemia does not elicit BBB dysfunction and leakage in the mouse brain. Moreover, different cellular and molecular components at the NVU appear intact in Ins2 AKITA mice, such as endothelial cells, pericytes, astrocytes, and basement membrane. Based on our results and the Corem and Ben-Zvi (submitted), we propose that hyperglycemia-induced BBB leakage is likely negligible.
Transcriptional analysis of microvascular fragments comparing hyperglycemic animals and WT controls revealed Slc2a1 to be 1.8-fold downregulated in brain vessel-fragments from Ins2 AKITA animals ( Table 1). Analysis of glucose transport in the brains of human patients has been controversial, with methods employed differing from surrogate tissues for SLC2A1 expression in the brain 32 to transient hyperglycemia conditions in healthy human beings 33 . Others have shown a decrease in glucose uptake in experimental non-obese DM animals 34 . Our study shows that Slc2a1 is downregulated at the transcriptional level specifically in the hyperglycemic mouse brain vasculature. Of note, allelic disorders associated with SLC2A1 mutations, glucose-deficiency syndrome, epilepsy, and dystonia, are associated with cognitive disability 35 . Thus, downregulation of SLC2A1 on brain blood vessels due to hyperglycemia could lead to metabolic changes in neural tissue leading to cognitive dysfunction especially after repeated hypoglycemic episodes.
Interestingly, the only gene that was found to be upregulated in Ins2 AKITA mice was Pin1, a phospho-serine/ threonine-proline isomerase that has been implicated in cellular proliferation. Pin1 has previously been reported to be upregulated in an in vitro model of hyperglycemia and in the aorta of streptozotocin-induced diabetic mice, resulting in mitochondrial oxidative stress, vascular relaxation dysfunction and vascular inflammation 36 . Thus, our results imply that even if hyperglycemia-induced upregulation of Pin1 might cause similar stresses to the brain vasculature, these do not induce significant BBB disruption. Our analysis of microglia showed that these cells change their morphology in Ins2 AKITA mice into a more amoeboid-like shape with shorter and thicker processes compared to the highly ramified process morphology seen in WT brains. Our findings are consistent with an "activated" microglial response. How the microglia is activated is unknown, but it has been suggested that hyperglycemia can potentially activate retinal microglia through  Table 1. RNA sequencing of microvascular fragments. Differentially expressed genes in Ins2 AKITA (n = 2) when compared to WT (n = 3) littermate controls. In the mouse brain, Gata2, Sema3G, Slc2a1, and Gm694 displayed restricted endothelial expression. The rest of the genes had a broad expression pattern (FPKM, fragments per kilobase of exon per million reads mapped). a variety of molecular mechanisms 37 . Furthermore, it is hypothesized that permeability at the BRB contributes to the microglia activation 37 . Our results reveal the same morphological changes in the hyperglycemic mouse brain than these already reported in retinas 23 . Because the integrity of the BBB and NVU are preserved in the Ins2 AKITA mice, other microglial activating factors than BBB permeability should be considered.  The association between DM and different brain pathologies is an underexplored but emerging research area. BBB dysfunction is commonly found in conditions such as Alzheimer's disease, and it is therefore tempting to speculate that hyperglycemia-induced BBB permeability is the underlying mechanism of brain pathology in human DM patients 14 . However, even the most prevalent microvascular DM complications, such as retinopathy or nephropathy, do not affect all DM patients 38 . Beyond hyperglycemia, other DM features, such as hyperglycemia/hypoglycemia cycles, dyslipidemia, or genetic susceptibility should be considered as possible causal factors for these pathologies. Our results presented herein show that hyperglycemia has minor, if any, effect on BBB permeability, and therefore fails to support increased BBB leakage as an underlying cause of DM-related brain pathologies.