DPP-4 Inhibitor and Sulfonylurea Differentially Reverse Type 2 Diabetes–Induced Blood–Brain Barrier Leakage and Normalize Capillary Pericyte Coverage

Diabetes affects 536.6 million people worldwide and reached a prevalence of 10.5% in 2021 (1). Diabetes causes damage to multiple organs, including the brain, thus constituting a major health problem.

Similar to the retinal and renal diabetic complications, where vascular damage leads to blood vessel leakage, in the brain, diabetes is associated with progressive blood–brain barrier (BBB) disruption (2–4) and dysregulated angiogenesis (5).

BBB leakage leads to an increased influx of neurotoxic molecules, inflammatory cells, and pathogens into the brain. The consequence is loss of the homeostasis of the neurovascular microenvironment, leading to chronic inflammation and other pathological immune responses, all of which can impair neuronal function (6). Indeed, type 2 diabetes (T2D) is now widely acknowledged to be a risk factor for neurodegenerative diseases like Parkinson disease (PD) (7), a neurodegenerative disorder characterized by the progressive loss of dopaminergic neurons and fibers and reduced dopamine in the striatum (8). One of the hypotheses is that T2D and PD share pathological microvascular alterations and neuroinflammation in the brain whereby T2D might further aggravate the microvascular pathology in PD (9).

The BBB is composed of several cell types, including endothelial cells, pericytes, and astrocytic endfeet (6). Pericytes are perivascular cells lining the entire microvasculature; their highest density is in the brain and retina (10). They maintain BBB integrity, have an essential role in angiogenesis, and contribute to vessel formation, vessel remodeling, and stabilization (11). The effects of T2D on pericytes are extensively studied in the retina, where pericyte dysfunction contributes to diabetic retinopathy characterized by neovascularization, leaky vessels, and edema (12). However, the impact of T2D on brain pericytes is much less investigated (13). It has been shown recently that T2D causes pericyte loss due to increased oxidative stress (14) and, in the brain, impairs vascular coverage, which is associated with increased vascular leakage and features of angiogenesis (15).

The link between metabolic dysfunction and neurodegeneration is further strengthened by studies demonstrating a beneficial effect of antidiabetic medications in PD and PD models that extends beyond metabolic effects (reviewed by Foltynie and Athauda [16]). Knowledge about the effect of antidiabetics on already established microvascular changes in the brain of T2D is, however, sparse. Given the detrimental effects that T2D-induced microvascular pathology and loss of BBB integrity may have on the neuronal environment and neuronal function, therapies that can reverse those changes are of clinical significance.

Here, we investigated whether the dipeptidyl peptidase-4 inhibitor (DPP-4i) linagliptin (Lina), or a sulfonylurea, glimepiride (Gli), both antidiabetic medications, can reverse any microvascular alterations induced by T2D in the brain. These two drugs were selected on the basis of their availability for clinical use and our previous findings demonstrating that they restored the impaired nigrostriatal dopaminergic neuronal function in aged T2D mice and patients with PD (17,18).

All procedures were in accordance with the ethical standards of the Karolinska Institute and Pronexus AB, where the studies were conducted. In this study, we used the material from a previous study with another objective (17). This is in line with the guideline principles at Karolinska Institute, with the aim to improve the ethical use of animals in testing according to the 3R principle (reduction, refinement, and replacement) (19). Two-month-old C57/BL6j male mice (n = 25) (Charles River Laboratories) were entered into the study and randomly assigned to the different experimental groups. Mice were housed under controlled conditions, in a 12-h light/dark cycle with free access to food and water.

To explore the effects of T2D and antidiabetics on microvascular changes, mice were fed either a standard diet (n = 7) or a high-fat diet (HFD; n = 18) for 12 months, starting from the age of 2 months. The HFD matched a formula containing 54% calories from fat (ssniff E15126–34); the standard diet was a normal chow diet. Mice fed the HFD were subdivided after 9 months into three groups. One group continued to receive the HFD only (n = 6), whereas the others received either Lina (in food, 5–7 mg/kg body weight per day; n = 7) or Gli (in food, 2–4 mg/kg body weight per day; n = 5) as described in our previous study (17).

Body weight and blood glucose levels were measured after overnight fasting in new cages. For measurement of plasma DPP-4 activity and total active GLP-1 levels, the fed-state blood was collected during killing, between 10 a.m. and 3 p.m., and measured by enzyme immunoassay and by ELISA, respectively (Meso Scale Discovery, Gaithersburg, MD). DPP-4 activity was decreased, and active GLP-1 was increased. (Complete data can be viewed in the Lietzau et al. article [17].)

Mice were sacrificed 12 months after initiation of the respective diet. Mice were anesthetized and transcardially perfused with saline followed by 4% paraformaldehyde. Brains were postfixed in 4% paraformaldehyde overnight and placed in a 20% sucrose solution. Brains were cut in 40-μm-thick sections and stored at 20°C.

Brain sections were blocked with 5% serum with 0.25% triton-X100 PBS (PBS-TX) for 1 h at room temperature (RT) and incubated with primary antibodies (Supplementary Table 1) diluted with 3% serum in PBS-TX overnight at RT. Sections were then incubated for 1 h at RT with the respective fluorophore-tagged secondary antibodies (Supplementary Table 2). For CD31 staining, the signal was amplified using biotinylated secondary antibody followed by fluorophore-conjugated streptavidin. Sections were mounted on gelatinized slides with mounting medium (polyvinyl alcohol/DABCO) and covered with a cover slip.

Confocal images were obtained from two striatal sections (according to anterior–posterior axis +0.62 to +1.18 relative to bregma) (20), using a Leica DMi8 confocal microscope. Two images per section were obtained at ×20 magnification from the dorsolateral and medial striatum (image size: 775 μm × 775 μm; z-stack size = 10 μm; step size = 1 μm). The same acquisition settings were applied for each image.

For three-dimensional modeling, z-stack images of CD31, CD13, and Col IV were reconstructed with Imaris software, using surface tool wizard (V.8.0, Bitplane, Zurich, Switzerland). All the images were processed with the same parameters.

Quantification of CD31⁺ vessel, CD13⁺ pericyte, and collagen IV (Col IV) density was performed on the maximum projected images, using the area fraction measurement tool of ImageJ (National Institutes of Health [NIH]). The density was expressed as the percentage of CD31⁺, CD13⁺, or Col IV⁺ area of the image area. Col IV density was normalized to the CD31⁺ vessel density. Pericyte coverage of the vessels was calculated by dividing the CD13⁺ pericyte density by the CD31⁺ vessel density on the same image and expressed as a percentage of the vessel density.

For vessel branch counts, vessel intersections were counted manually on the maximum projection images; z-stack images were reviewed to avoid counting vessel overlay. For pericyte counts, pericytes were identified by their perivascular localization and the colocalization of the cell body with a nucleus. Pericytes were counted manually by using a multipoint tool in ImageJ.

Activated pericytes were identified by NG2 (21). NG2⁺ pericytes were recognized by colocalization with CD13⁺ pericyte using the colocalization plugin tool of ImageJ; the nanolocalized NG2 was excluded from the analysis because it likely denotes glial cells. NG2⁺ pericyte density was calculated as percentage of NG2⁺ pericyte area of total image area using ImageJ. NG2⁺ pericytes were counted manually after identifying NG2⁺/DAPI pericytes. Then, NG2⁺ pericyte density and count were presented as a percentage of CD13⁺ pericytes.

Extravascular fibrinogen, IgG, and albumin were quantified to evaluate vascular leakage. CD31⁺ vessels were outlined to exclude intravascular plasma proteins. Then, by applying the image threshold, the area covered by each of extravascular fibrinogen, IgG, and albumin was quantified using ImageJ and reported as the percentage of the total image area.

CD11b was used to label microglia. Microglial activation was quantified on the basis of morphological changes: an activated microglia has a larger cell body and shorter processes (22). Modified from the previous method (23), ImageJ was used to measure microglial process density and cell body density. The processes were delineated and measured using the Tubeness plugin of ImageJ. The ratio of branch density to cell body density was calculated as an index of microglial activation, whereby the lowest value represents the most activated microglia.

The perivascular distribution of CD11b⁺ microglia was illustrated showing microglial cells within 20 μm of the vessel border by using the Distance Map plugin of ImageJ. The communication between the vessels and perivascular microglia was quantified by measuring the density of interaction points between CD11b⁺ cells and CD31⁺ vessels, using the colocalization plugin tool of ImageJ. The data are expressed as the ratio of the interaction point density to the CD31⁺ vessel density per image.

All data were analyzed using GraphPad Prism version 8. Data were examined for normal distribution using the Shapiro–Wilk normality test. Parametric unpaired Student t test was used for the comparison between the standard diet and HFD groups. One-way ANOVA was used for comparison among the HFD, HFD plus Lina, and HFD plus Gli groups followed by the Dunnett or Tukey comparison test. The data are presented as mean ± SD. The comparisons between the groups were considered significant when P < 0.05. Simple linear regression analysis was performed to evaluate correlations using a 95% CI graph. The slope considered significantly different when P < 0.05.

The data supporting the conclusions of this article will be made available upon contacting the corresponding author.

Being fed an HFD for 12 months caused glucose intolerance and obesity in all mice. Both Lina and Gli similarly reduced fasting glucose levels but had no effect on body weight. For complete data, see the article of Lietzau et al. (17). In Lina-treated mice, plasma DPP-4 activity and GLP-1 levels were significantly decreased or increased, respectively, as expected on the basis of the drug mechanism of action (see Supplementary Figs. 2 and 3 in the article of Lietzau et al. [17]).

An HFD significantly increased vascular density and the number of vascular branches compared with the standard diet group, indicating angiogenesis (t test for vessel density, P < 0.01; for vessel branches, P < 0.05) (Fig. 1A–C). Even though CD13⁺ pericyte density and number did not differ between the HFD and standard diet groups (t test for pericyte density and count, NS) (Fig. 1D and E), the pericyte coverage of the brain capillaries was significantly reduced in the HFD group compared with the standard diet group (t test, P < 0.01) (Fig. 1F), suggesting that an HFD induced angiogenesis but with a pericyte mismatch.

Interestingly, a 3-month treatment with Lina in HFD-fed mice significantly reduced the CD31⁺ vessel density compared with an HFD only, but the number of vessel branches was not significantly changed (for Lina, one-way ANOVA: CD31⁺ vessels, P < 0.05; for number of branches, NS) (Fig. 1A–C). In contrast, Gli further increased angiogenesis, as indicated by a significant increase in CD31⁺ vessel density and the number of vessel branches compared with the HFD group (for Gli, one-way ANOVA: CD31⁺ vessels, P < 0.05; number of branches, P < 0.01) (Fig. 1A–C). These findings indicate that Lina and Gli decreased or increased HFD-induced angiogenesis, respectively.

Notably, Gli, but not Lina, also significantly changed the number and density of CD13⁺ pericytes (twofold increase) compared with the HFD group (one-way ANOVA: for Lina, CD13⁺ density and count, NS; for Gli, CD13⁺ density, P < 0.01, and CD13⁺ count, P < 0.05) (Fig. 1A, D, and E). Importantly, even though their effect on vascular density was different, both Lina and Gli significantly increased pericyte coverage compared with the HFD group (one-way ANOVA: for Lina, CD13⁺ coverage, P < 0.05; for Gli, CD13⁺ coverage, P < 0.05) (Fig. 1A and F).

To test BBB integrity, we examined the presence of large plasma proteins in the brain parenchyma. As expected, HFD-fed mice had significantly increased extravasation of albumin and fibrinogen, and a tendency to increase vascular leakage of IgG, compared with the standard diet group (by t test: albumin, P < 0.05; IgG, NS; fibrinogen, P < 0.05) (Fig. 2A–F), confirming BBB leakage in diabetic animals.

Strikingly, Lina and Gli significantly reversed the extravasation of albumin, IgG (fivefold decrease), and fibrinogen (10-fold decrease) (one-way ANOVA: for Lina, albumin, P < 0.05, IgG, P < 0.05, and fibrinogen, P < 0.05; for Gli, albumin, P < 0.05, IgG, P < 0.05, and fibrinogen, P < 0.05) (Fig. 2A–F). As expected, fibrinogen extravasation was detected mainly around the vessels with low pericyte coverage in HFD-fed mice but was absent when pericytes sufficiently covered the blood vessels in the Lina- and Gli-treated groups (Fig. 2G).

To further understand the pericyte response, we analyzed pericyte activation by measuring the proportion of NG2⁺ pericytes, a marker expressed on activated pericytes (21), normalized to CD13⁺ pericytes. An HFD resulted in increased number and density of NG2⁺ pericytes compared with the standard diet group (t test: NG2⁺ number, P < 0001; NG2⁺ density, P < 0.01) (Fig. 3A–C), indicating pericyte activation.

The antidiabetics did not significantly change number or density of NG2⁺ pericytes when compared with the HFD group, although there was a tendency of Gli to increase both (one-way ANOVA: for Lina, NG2⁺ count and density, NS; for Gli, NG2⁺ count and density, NS) (Fig. 3A–C). However, when compared with the Lina group, the Gli group had a significantly higher NG2⁺ cell count and density (one-way ANOVA: NG2⁺ count, P < 0.05; NG2⁺ density, P < 0.05).

To investigate if another factor might affect BBB integrity and angiogenesis (24), microglial activation and perivascular location were examined. Extending previous observations on the number of microglial cells (17), HFD did not significantly change the activation score of microglia compared with the standard diet group (t test, n.s.) (Fig. 4A, B, and F). However, Gli, but not Lina, significantly reduced microglial activation, which resulted in a higher process to cell body ratio when compared with the HFD group (one-way ANOVA: for Lina, NS; for Gli, P < 0.01) (Fig. 4A, B, and F).

Next, perivascular microglia were evaluated by measuring the interaction points between CD11b⁺ microglia and CD31⁺ vessels (Fig. 4E). An HFD did not change the number of microglia interaction points with the CD31⁺ vessels compared with the standard diet group (t test, NS) (Fig. 4C and D). However, Gli, but not Lina, increased the number of the interaction points compared with the untreated HFD group (one-way ANOVA: for Lina, NS; for Gli, P < 0.05) (Fig. 4C and D). These data indicate that Gli can reduce the microglial activation and increase the microglial interaction with blood vessels. Further analysis showed that the microglial interaction with the vessels was positively correlated with CD13⁺ pericyte density (Fig. 4G).

We also investigated the effect of T2D on the expression of Col IV, a marker of the basement membrane of the capillaries (25). Values were normalized to the CD31⁺ vessel density and expressed as proportion of the vessel density. We found that HFD significantly increased Col IV density compared with the standard diet group (t test, P < 0.05) (Fig. 5A and B). Gli, but not Lina, further increased Col IV density compared with the HFD group (one-way ANOVA: for Lina, NS; for Gli, P < 0.05) (Fig. 5A and B).

T2D has a deleterious impact on the vasculature in different organs, resulting in complications like retinopathy and nephropathy (see the Forbes and Cooper review [26]). However, much less literature is available on T2D’s effects on the brain microvasculature. Here, we explored the impact of T2D and the effect of two clinically used antidiabetics, Lina and Gli, on microvascular changes in the striatum. We confirmed that HFD-induced T2D causes BBB leakage, angiogenesis, reduction in pericyte coverage of the vessels, and increased Col IV expression.

Both antidiabetics significantly reduced hyperglycemia and BBB leakage and normalized pericyte coverage. However, they had a differential angiogenic effect. Whereas Lina restored the T2D-induced angiogenesis and normalized the pericyte coverage, Gli further enhanced the angiogenic process and increased pericyte density, allowing adequate vascular coverage and increased Col IV expression. Interestingly, Gli was also associated with greater microglial interaction with the vessels and a reduction in microglial activation.

The 12-month HFD used in this study represents a clinically relevant long-term T2D model. T2D induced pathological angiogenesis, indicated by the increase in vascular density and vessel branches, and a reduction in pericyte coverage of the vessels. These findings extend previous findings from short-term T2D models and confirm the current knowledge of pathological angiogenesis induced by T2D in the brain (5,15,27).

The vascular changes resulted in BBB leakage in the untreated HFD group, a finding that is consistent with previous reports from patients with diabetes and animal models of diabetes (2,4). The molecular basis for BBB leakage of large plasma proteins is not fully understood, but amplification of endothelial caveolae leading to transcytosis of plasma proteins (28) and/or reduction in the expression of tight junctional (TJ) proteins (29), allowing paracellular permeabilization, has been suggested. Interestingly, the extravasation of, for example, fibrinogen localized in vessel areas devoid of pericytes underlines the importance of sufficient pericyte coverage for BBB maintenance (2,13,30). This reduction in pericyte coverage may cause dysregulation of the transcellular barrier (31) and reduction of TJ protein expression under diabetic conditions (2,13).

Both Lina and Gli reversed an already established T2D-induced BBB leakage. However, they had differential effects on vascular pathology, underlining the fact that common metabolic effects mediated by these drugs, such as the regulation of glycemia, are only a partial effect of those antidiabetic drugs on the brain.

Lina counteracted the angiogenic effect of T2D and restored pericyte coverage. Studies of the effect of antidiabetics on brain vessels are sparse. In a previous study using the diabetic Goto-Kakizaki rat model, researchers examined the effect of Lina on diabetes-induced angiogenesis in the brain (27). However, neither BBB integrity nor pericyte coverage were investigated. Consistent with our findings, Lina attenuated neovascularization induced by diabetes in the cortex and striatum (27). Interestingly, Lina did not lower the glucose level in the diabetic rat model, suggesting Lina effects are, at least in part, glucose independent. Findings of an in vitro study suggest Lina might reduce angiogenesis via modulation of the endothelin-1 system (27). The exact mechanism mediating the vascular actions of Lina in vivo, however, remains elusive. Lina inhibits DPP-4 activity and thereby increases the level of, among other peptides, endogenous GLP-1 (32). DPP-4 is expressed in brain endothelial cells (33) and may have a role in the regulation of endothelial cell proliferation (34).

Lina reversed the BBB leakage, likely as a result of counteracting the T2D-induced angiogenesis and restoring pericyte coverage. Endothelial cells require close pericyte contact and sufficient pericyte coverage to form a TJ and thereby maintain an intact BBB (30,35). Previous data in diabetic models showed that DPP-4is can reduce pericyte apoptosis (36) and increase TJ protein levels (37) in retina. DPP-4is also reduced vascular leakage and recovered TJ protein levels in transitory cerebral ischemia (38). It is conceivable that in our study, Lina improved BBB integrity by protecting pericytes, maintaining proper vascular pericyte coverage, and attenuating immature neovascularization.

Unlike Lina, Gli further enhanced the angiogenesis in the HFD group, paralleled by a corresponding increase in pericyte density. This resulted in mature vessels with sufficient pericyte coverage, which was reflected by a significant reduction in BBB leakage.

Gli acts via SUR1 and SUR2 (39) and belongs to a group of drugs able to cross the BBB (40). Interestingly, the SUR2 gene (Abcc9) is specifically expressed in brain pericyte (41). Thus, it is conceivable that the actions of Gli on the vasculature are mediated directly via SUR2 on brain pericytes. The observed increase in pericyte numbers in Gli-treated animals might be due to increased proliferation, reduced migration, or reduced apoptosis of pericytes. Gli has been suggested to increase proliferation of endothelial smooth muscle cells (42), but the molecular mechanisms underlying its effect on brain pericytes still need to be explored.

The strong angiogenic response by Gli may be mediated by changes in pericyte activity. Pericytes control the angiogenic process via several different signaling pathways [reviewed by Sweeney et al. (31)] and contribute to endothelial cell proliferation and sprouting through extracellular matrix modulation (43). Pericyte dysfunction during angiogenesis causes a pathological angiogenic pattern (44).

In our study, Gli increased the expression of NG2 in pericytes. NG2 has been described as an angiogenic marker of activated pericytes (45) and has a role in neovascularization and proliferation of pericytes and endothelial cells (21). Although the pathological activation of pericytes usually results in immature vessels with corresponding reduction in BBB integrity, no vascular leakage was detected in the Gli group, suggesting that pericytes were sufficiently recruited to the newly formed vessels, ensuring proper pericyte coverage.

Another interesting finding is the increase in the density of perivascular microglia after Gli treatment. Perivascular microglia have been suggested to have a dual effect on BBB integrity, depending on their activation state. Resting microglia, corresponding to a ramified phenotype, maintain vascular integrity and provide TJ proteins. However, vessels are phagocytosed, and BBB integrity is impaired by the activated microglial phenotype that is characterized by shorter processes and a larger soma area (46). Interestingly, Gli increased the ramification of microglia, indicating a resting status. This is consistent with previous data showing that sulfonylurea drugs can reduce microglial activation via inhibiting the NLRP3 inflammasome and NF-κB activation and reducing the proinflammatory response (47). Another study confirmed that Gli inhibits NLRP3 activation in microglia and brain endothelial cells in the cerebral hemorrhagic model, resulting increased TJ proteins and less vascular leakage (48). It is conceivable that Gli supports BBB integrity by attracting microglia toward the vessels, a process that may be mediated by pericytes; we found a significant correlation between pericyte density and microglial interaction with the vessels in this study. This finding is also consistent with our previous data (15) showing that perivascular microglial density correlated positively with pericyte density. However, more studies are needed to confirm this hypothesis of an interaction between microglia and pericytes.

We also investigated Col IV because it is important for vascular integrity and basement membrane function (25). Col IV is, among other cells, synthesized by pericytes (25). Gli further increased the expression of Col IV, which possibly resulted from the increase in the number and density of pericytes. In addition, it has been shown that NG2-positive pericytes, which were increased by Gli, are responsible for Col IV deposition in the basement membrane (49). This increment in Col IV production may contribute to consolidating the BBB integrity. Previous studies showed that modification of the expression of Col IV leads to vulnerability in the BBB, indicating its importance in supporting vascular integrity (50).

The BBB guards the brain’s microenvironment, and disruption of its integrity leads to a disturbance in the brain homeostatic balance, which ultimately has negative consequences for neuronal function (6). Recently, we showed that diabetes worsens the motor deficit in a PD model. This was associated with pathological vascular changes characterized by a reduction in pericytes and increased vascular interaction of activated microglia and changes in vascular density (15). Furthermore, diabetes worsened dopaminergic neuronal function and reduced dopamine release in the striatum of diabetic mice. However, Lina and Gli recovered the dopamine release in these mice (17). It is conceivable that the normalization of the vascular pathology, especially the reduction in BBB leakage, contributes to a normalized neuronal environment that may participate, at least partially, in the restorative effect on dopamine release, as observed by these antidiabetic drugs in T2D mice (17).

In conclusion, we confirmed that T2D induced vulnerability of the BBB and pathological alterations in brain microvessels. Importantly, we demonstrated that Lina or Gli can counteract several of these vascular pathologies. Our findings may have important implications for the use of these drugs not only to balance T2D-induced vascular alterations but possibly also other conditions with aberrant vessel formation, which will be interesting to explore. Importantly, these drugs may deter or slow the progression of neurodegenerative diseases via reversal of the T2D-induced pathological BBB leakage, which will maintain a healthy microenvironment around the neurons and could be a possible contributor to the regenerative effect of antidiabetic drugs described in neurodegeneration.

Funding. This work was mainly supported by the Swedish Parkinson Foundation and the Åhlens Foundation and by an award from Boehringer Ingelheim Pharma GmbH & Co. KG (to C.P.), Financial support was also provided by Hjärnfonden, Diabetesfonden, Ulla Hamberg Angeby och Lennart Angebys Stiftelse, the Swedish Stroke Foundation, the Swedish Heart-Lung Foundation, and by the ALF regional agreement on medical training and clinical research.

Duality of Interest. This study received funding from Boehringer Ingelheim Pharma GmbH & Co. KG. T.K. is employed at Boehringer Ingelheim Pharma GmbH & Co. KG.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Author Contributions. O.F.E., G.P., V.D., and C.P. designed the experiments, had full access to the data in the study, and take responsibility for the integrity of the data and the accuracy of the data analysis. O.F.E. performed the immunohistochemistry, microscopy, and image processing and analysis, and analyzed the data. D.K., E.V., and G.L. helped with the sectioning and edited the manuscript. T.N. and T.K. provided resources for the study, contributed to discussion, and edited the manuscript. O.F.E. and G.P. wrote the article. G.P. and C.P. secured funding for this project. All authors contributed to the article and approved the submitted version.