Disruption of cortical cell type composition and function underlies diabetes-associated cognitive decline

Aims/hypothesis Type 2 diabetes is associated with increased risk of cognitive decline although the pathogenic basis for this remains obscure. Deciphering diabetes-linked molecular mechanisms in cells of the cerebral cortex could uncover novel therapeutic targets. Methods Single-cell transcriptomic sequencing (scRNA-seq) was conducted on the cerebral cortex in a mouse model of type 2 diabetes (db/db mice) and in non-diabetic control mice in order to identify gene expression changes in distinct cell subpopulations and alterations in cell type composition. Immunohistochemistry and metabolic assessment were used to validate the findings from scRNA-seq and to investigate whether these cell-specific dysfunctions impact the neurovascular unit (NVU). Furthermore, the behavioural and cognitive alterations related to these dysfunctions in db/db mice were assessed via Morris water maze and novel object discrimination tests. Finally, results were validated in post-mortem sections and protein isolates from individuals with type 2 diabetes. Results Compared with non-diabetic control mice, the db/db mice demonstrated disrupted brain function as revealed by losses in episodic and spatial memory and this occurred concomitantly with dysfunctional NVU, neuronal circuitry and cerebral atrophy. scRNA-seq of db/db mouse cerebral cortex revealed cell population changes in neurons, glia and microglia linked to functional regulatory disruption including neuronal maturation and altered metabolism. These changes were validated through immunohistochemistry and protein expression analysis not just in the db/db mouse cerebral cortex but also in post-mortem sections and protein isolates from individuals with type 2 diabetes (74.3 ± 5.5 years) compared with non-diabetic control individuals (87.0 ± 8.5 years). Furthermore, metabolic and synaptic gene disruptions were evident in cortical NVU cell populations and associated with a decrease in vascular density. Conclusions/interpretation Taken together, our data reveal disruption in the cellular and molecular architecture of the cerebral cortex induced by diabetes, which can explain, at least in part, the basis for progressive cognitive decline in individuals with type 2 diabetes. Data availability The single-cell sequencing data that supports this study are available at GEO accession GSE217665 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE217665). Graphical Abstract Supplementary Information The online version contains peer-reviewed but unedited supplementary material available at 10.1007/s00125-023-05935-2.


Introduction
Type 2 diabetes constitutes a huge burden on healthcare systems worldwide due to multi-morbidities such as blindness, renal failure and cardiovascular-related death [1]. There is mounting evidence that type 2 diabetes is also associated with cognitive decline [2][3][4]. Individuals with type 2 diabetes have an approximate twofold increased risk of developing Alzheimer's disease compared with non-diabetic, agematched individuals [3].
Clinical evidence and pre-clinical data suggest that common pathways may be involved in type 2 diabetes and cognitive decline, including glial dysfunction, neurodegeneration and inflammation [5,6], arising from dysfunctional insulin signalling, chronic hyperglycaemia and oxidative stress. These pathways may contribute to neurovascular unit (NVU) dysfunction in the brain and retina during type 2 diabetes [4]. However, the underlying pathways that drive neurodegeneration and cognitive decline in type 2 diabetes are not fully understood.
The NVU plays an important homeostatic role in the central nervous system and is formed by interdependency of vascular cells, glia, neurons and professional immune cells. Besides ensuring integrity of the blood-brain barrier, the NVU is important for immune privilege and regulating blood flow to meet the oxygenation and nutrient demands of the highly metabolically demanding neuropile. A recent study revealed human-specific transcriptomic signatures in the cerebrovasculature in Huntington's disease [7] and another study revealed several Alzheimer's disease risk genes mapping to endothelial and microglial cell populations [8]. Furthermore, neurodegenerative disease-linked shifts in brain cell populations have been reported [9]. Neuronal and glial population changes have been observed in Alzheimer's disease [10] and may co-occur with reduced cortical thickness in individuals [11].
Analysis from a single-cell RNA sequencing (scRNA-seq) dataset revealed profound changes to the transcriptome of vascular, glial and neuronal cells from the retina of db/db mice, showing a useful linkage to the pathogenesis of diabetic retinopathy [12]. The scRNAseq approach has also been used to characterise hippocampal changes in db/db mice, indicating the role of microglial populations in diabetes-induced cognitive decline [13]. The impact of type 2 diabetes on the brain NVU, particularly in the cortex, have not been fully explored. Moreover, whether the cognitive decline in type 2 diabetes is associated with disruption in brain cell populations remains unknown.
The aim of the present study was to provide novel insights into type 2 diabetes-associated cognitive decline and the underlying associated pathology in the cerebral cortex by combining state-of-the-art physiological and molecular measurements in db/db mice as a model and validation in human samples.

Methods
Animals db/db (C57BLKS/heterozygous db/+ mice [BKS.Cg-+Lepr db /+Lepr db /OlaHds]) (RRID: IMSR_ENV:HSD-173) were purchased from Harlan Laboratories (Netherlands) or Janvier (France). Since heterozygous (db/+) mice do not show a specific phenotype [14], wild-type (WT) and db/+ mice are commonly considered a single control group [15,16]. For For immunohistochemistry (IHC) and behavioural studies, WT mice (BKS.Cg-/OlaHds) belonging to the same colony were used as the control group. For transcriptional studies and cortical metabolic assessment, db/+ mice were used as the control group. Mice were housed in a standard pathogen-free experimental facility, exposed to a 12 h light-dark cycle and had free access to food and water. Five or six male/female mice per group aged 14-16 weeks (representing an established but not yet advanced stage of diabetes) were used. Metabolic determinations (Table 1) were carried out as described in electronic supplementary material (ESM) Methods.
Actimetry and new object discrimination task Episodic memory for 'what,' 'where' and 'when' paradigms was analysed, following procedures shown in ESM Methods. 'What' was the difference in time exploring familiar and recent objects, 'where' was the difference in time exploring displaced and non-displaced objects and 'when' was the difference between time exploring familiar non-displaced and recent non-displaced objects. All measurements were performed in triplicate and data were expressed as percentage of control values. Discrimination index was calculated as the percentage of time exploring recent objects/total exploration time.
Morris water maze Spatial learning and memory abilities were analysed in the Morris water maze (MWM) (see ESM Methods). Time required to locate the platform in the acquisition phase, number of entrances in quadrant 2 and swimming velocity during the retention phase were analysed using Smart software version 3.0.04 (Panlab S.L.U., Barcelona, Spain).
Tissue dissociation For scRNA-seq, one male homozygous (db/db) mouse and one male heterozygous (db/+) mouse were killed at 16 weeks of age. Whole cortex was dissected and dissociated using the papain dissociation system (Worthington, Lakewood, NJ, USA). Cortex samples were digested in papain (20 U/ml) for 45 min at 37°C. After treating with ovomucoid, cells were re-suspended in PBS with 0.04% wt/vol. BSA. Cell counts and viability was checked using Trypan Blue (Sigma-Aldrich, Missouri, USA).
scRNA-seq library preparation, sequencing and associated analysis scRNA-seq libraries were generated using the 10x Genomics Chromium Single Cell 3′ Reagent Kit v3 and sequenced on Illumina NovaSeq6000 S2 200 flow cell at the Genomics Core Technology Unit (GCTU), Queen's University Belfast. Analysis was completed as discussed in ESM Methods using Seurat R package v.4.2.0 (https:// satij alab. org/ seurat/; developed by P. Hoffman, S. Lab and collaborators, New York City, NY, USA). Identification of cell types, marker genes and differential expression analysis Following procedures detailed in ESM Methods, cortical cell types were identified by expression of established markers (Table 2). Differential gene expression, single-cell regulatory network, weighted gene correlation network (WGCN) [17] and pseudotime trajectory [18] analyses were performed.
Tissue preparation and staining Immediately after mice were killed via overdose of pentobarbital, db/db and control mouse brains were harvested and weighed. Left hemispheres were fixed in 4% wt/vol. paraformaldehyde for 1 week and 30 µm coronal brain sections were obtained in a Microm HM450 microtome (ThermoFisher, Spain). Sections were stored in 50% wt/vol. glycerol at −20°C until used. For staining procedures, consecutive sections were selected at 1.5, 0.5, −0.5, −1.5, −2.5 and −3.5 mm from Bregma. Staining was carried out for neuronal nuclear protein (NeuN) [19], BrdU and doublecortin (DCX) [20] as previously described.
Vascular staining Antigen retrieval was performed using citrate buffer at 95°C for 30 min in a water bath. Slides were permeabilised in 2% (wt/vol.) Triton-X for 30 min. Slides were blocked in 5% donkey serum in 0.2% (wt/vol.) Triton-X for 1 h at room temperature. Primary antibodies (ESM Table 1) were added and slides were incubated overnight at 4°C in 0.5% donkey serum and 0.2% Triton-X followed by incubation with secondary antibodies (ESM Table 1) for 2 h at room temperature, in PBS. Metabolic assessment of cortical brain punches The method published by Qi et al (2021) [22] was adapted in order to assess glycolysis in acute coronal slices from the cortex of five male db/db and five male db/+ mice, as described in ESM Methods. The 'glycolysis stress test' protocol from Agilent (USA) was optimised for use with cortex slices (ESM Table 2). Protocol length and injection timings were as follows: baseline, 4 cycles; glucose injection, 5 cycles; oligomycin A injection, 5 cycles; and 2-deoxy-d-glucose (2-DG), 10 cycles. Each cycle consisted of 3 min mix, 1 min wait, 2 min measure. Analysis of basal glycolysis, glycolytic capacity and glycolytic reserve was performed in Wave Software version 2.6.3.5 (Agilent).

Image analysis
Human tissue For IHC staining, post-mortem brain samples of the frontal cortex of eight individuals were provided by the Netherlands Brain Bank, Amsterdam. The brain samples were from donors with or without diabetes mellitus. Classification of the donors was based on clinical diagnosis and confirmed by autopsy. The staining procedures were performed as described in ESM Methods. Antibody information is included in ESM Table 3. The new object discrimination test showed that episodic memory was not affected in db/db mice when 'what' (p=0.441) or 'where' (p=0.484) paradigms were analysed. However, the 'when' paradigm was significantly impaired in db/db mice (*p=0.016). Data are representative of five mice. (c) Discrimination index was not affected in db/db mice (p=0.530). Data are representative of five or six mice. (d) Learning impairment was observed in db/db mice in the acquisition phase of the MWM. Times required to locate the platform were significantly longer in db/db mice than control mice in sessions 5-8 (session 1 p=0.729, session 2 p=0.637, session 3 p=0.294, session 4 p=0.168, session 5 ***p<0.001 vs control, session 6 ***p<0.001 vs control, session 7 **p=0.003 vs control, session 8 *p=0.025 vs control). Data are representative of five or six mice. White squares, mean value for db/db mice; grey squares, mean value for control mice. (e) The number of entries into quadrant 2 (where the platform was located along the acquisition phase) was significantly lower in db/db mice (*p<0.05). (f) Brain weight was reduced in db/db mice when compared with control mice (**p<0.001 vs control). (g) Hemisection size was significantly reduced in db/db mice (*p=0.039 vs control    For western blotting, post-mortem brain samples were obtained as frozen pieces of cortical tissue from the neurological tissue biobank of Hospital Clínic -IDIBAPS in Barcelona, Spain. Control donors did not have signs of neurological alterations. Samples from individuals with type 2 diabetes clinically diagnosed with dementia were excluded from the study. The most relevant associated comorbidities are presented in ESM Table 4. Proteins were extracted and western blots were run as described in ESM Methods. Antibody information is included in ESM Table 3.

Analysis and statistics
Randomisation and blinding at the experimental phase was not possible due to the obese nature of db/db mice. Quantification and analysis of mouse data was done by two independent, blinded researchers. Student's t test for independent samples, or Mann-Whitney test was used to compare control and db/db mouse groups. Two-way ANOVA (group × session) was used to analyse the acquisition phase in the MWM. IBM SPSS Statistics software version 29.0.0.0 (IBM Corporation, NY, USA) was used for all statistical analysis and results are expressed as mean ± SEM.

db/db mice exhibit altered memory function and reduced cortical thickness
The db/db mouse displays key characteristics of type 2 diabetes including obesity, hyperglycaemia and hyperinsulinaemia (Table 1) [21,23]. A schematic of the study is shown in Fig. 1a.
Episodic memory was compromised in db/db mice (Fig. 1b). While no differences were detected for 'what' or 'where' paradigms, a significant impairment was observed for 'when' paradigm. A similar discrimination index was observed when comparing the groups of mice (Fig. 1c). Assessment of spatial memory using the MWM revealed no significant group × session effect (F (1,7) =1.009, p=0.427), although individual session assessment revealed that db/db mice had great difficulty in learning the location of the hidden platform. This was reflected from session 5 onwards (Fig. 1d). The number of entries into the quadrant where the platform used to be located was significantly lower in db/ db mice (Fig. 1e). Swimming speed was reduced in db/db mice (ESM Table 5) and while this may impact the results of the MWM, we have also shown that walking speed (and therefore motor activity) is not compromised in these mice. At the time of death, brain weight was significantly lower in db/db mice (Fig. 1f) [24]. Cresyl violet staining revealed that hemisection sizes were also significantly smaller in db/ db mice (Fig. 1g). db/db mice presented significantly thinner cortices when compared with control mice (Fig. 1h,i).
Single-cell transcriptome of db/db mouse cortex reveals massive changes in identity and composition of various key cell types Single-cell transcriptome analysis of cortex from db/db mice revealed significant changes in neuronal and glial populations (Fig. 2a,b). There was a marked increase in the microglia population of the db/db mouse cerebral cortex accompanied by a loss of astrocytes and mature neurons (Fig. 2c). An increased fraction of immature neurons was observed, alongside an increase in oligodendrocyte progenitors and concomitant decrease in mature oligodendrocytes (Fig. 2c). Pseudotime trajectory on glia populations (astrocytes and oligodendrocytes together) (ESM Fig. 1) showed cellular heterogeneity and an altered lineage trajectory in db/db vs control mice. These results suggest altered neuronal development, glial development and maturation trajectory in the db/db mouse cortex.

Molecular analysis of cortical cell populations reveals critical changes in diabetic mouse brains
Focused molecular analysis on neuronal and glial specific transcripts ( Fig. 2d-g) demonstrated that genes encoding ribosomal protein S21 and S29 (Rps21 and Rps29, respectively) were downregulated in immature neurons as well as astrocytes, with upregulation of ribosomal protein SA (encoded by Rpsa) in microglia. Transcripts linked to inflammation and microglial activation were altered in db/db mouse microglia: expression of genes encoding apolipoprotein E (Apoe) [25] and prostaglandin D2 synthase (Ptgds) [26] was upregulated. Furthermore, downregulated genes in mature neurons were major mediators and regulators of synaptic neuronal functions such as the genes encoding glutamate ionotropic receptor AMPA type subunit 1 (Gria1) [27] and myelin basic protein (Mbp) [28]. Mature neurons showed downregulation of genes encoding maternally expressed gene 3 (Meg3) and β-actin (Actb). Astrocytes displayed downregulation of actin polymerisation regulating genes tetraspanin 7 (Tspan7) [29] and metallothionein III (Mt3) [30]. Immature neurons showed downregulation of the gene encoding early growth response-1 (Egr1), an important transcription factor in memory formation [31]. This occurred alongside upregulation of HbA-a2 (encoding a globin family protein) and Gad2, which are known to be linked to schizophrenia and bipolar disorder [32]. Analysis by the algorithm pySCENIC identified enhanced expression of the transcription factor genes (Hes5, Lhx2, Sox21, Sox2) and neurogenesis regulatory factors in db/db mouse astrocytes (Fig. 3a-e), suggesting a potentially ambiguous lineage trajectory between db/db and control mice. Top microglial regulons were slightly discrete between control and db/db mice (Fig. 3f-j). The mature neuronal population (db/db) showed top regulons that are known transcription factors (Elk1 and Egr4) regulating early immediate genes (ESM Fig. 2a-e), while immature neuronal cells (db/db) had increased expression of marker genes for inhibitory neuronal characteristics (Sp9 and Meis1) (ESM Fig. 2f-j). These outcomes suggest a shift in the gene regulatory networks and driver transcription factors underlying distinct cell fates in db/db mouse cortex. The corresponding scale independence and mean connectivity are shown in ESM Fig. 3.
Ex vivo investigation validates neuronal and glial population alterations in db/db mice Immunohistochemistry (Fig. 4a) was used to quantify the numbers of immature and mature neurons (Fig. 4b-g). NeuN + cell/DAPI ratio was significantly reduced in db/db mice (Fig. 4b). A slight increase in the density of DCX + and proliferating cells was observed in db/db mice (Fig. 4c,d). Since the cortex is not a classical neurogenic niche, proliferation and neurogenesis were also assessed in the subventricular zone (SVZ). Both DCX burden and BrdU + cell density in the SVZ were significantly increased in db/db mice when compared with control mice (Fig. 4e-g).
Microglia burden was increased in the cortex from db/db mice when compared with control mice (Fig. 4h), while no differences in astrocyte burden were observed (Fig. 4i). (Fig. 5a-d). Gene ontology (GO) analysis of correlated unique gene modules in cells revealed altered mRNA metabolism in immature neurons and changes in glycogen and lipid metabolism in astrocytes of diabetic mouse brains (Fig. 5e). Furthermore, pathways implicated in inflammation and phagocytosis in microglia were uniquely altered in diabetic cortex, whereas mature neurons had altered pathways relating to synaptic processes (Fig. 5e).

Vascular abnormalities observed in db/db mouse cortex
Previous reports have suggested that brain vasculature is affected in type 2 diabetes models [24]. We observed reduced laminin + vessels in db/db mice (Fig. 6a, d). 'Hotspots' of leakage were observed in the db/db mouse cortex but not in the control mice cortex (Fig. 6e). Expression of endothelial marker genes including Lgals1 (encoding galectin 1) and Lamc3 (encoding laminin γ subunit 3) was reduced in db/db mice (ESM Fig. 4). The top differentially expressed genes (DEGs) in db/db mouse endothelial cells included Malat (encoding metastasis-associated lung adenocarcinoma transcript 1), which may be involved in angiogenesis [33], and Cts3, which is dysregulated in cerebral amyloid angiopathy (Fig. 6h) [34]. There was an increase in acellular capillaries  . (a, b) Cluster dendrogram for gene modules showing modules of high interconnection in control (a) and db/db (b) mice. Height in cluster dendrograms indicates the inter-cluster correlation distance and modules are marked by different colours in the horizontal bar (grey represents unassigned genes). (c, d) The relationship between gene sets and cell types is shown for control (c) and db/db (d). Each column is a cell type and each row represents module eigengenes. Each unit is coloured in the correlation coefficient (R) on a scale of −1 to +1 (black, negative; red, positive; white, null). (e) Unique GO terms for correlated gene module in db/db mouse cell types. Cell images were created with BioRender.com. ABC, arachnoid barrier cells; ASC, astrocytes; CPC, choroid plexus cells; DC, dendritic cells; EC, endothelial cells; EPC, ependymocytes; ImmN, immature neurons; MAC, macrophages; MG, microglia; mNeur, mature neurons; OLG, oligodendrocytes; OPC, oligodendrocyte progenitor cells; PC, pericytes; VLMC, vascular leptomeningeal cells; VSMC, vascular smooth muscle cells ◂ Fig. 6 Vascular alterations in db/db mouse cortex. (a) Laminin + vessel area was significantly reduced at various levels of Bregma in the cortex of db/db mice (*p<0.05, **p<0.01). (b) A significant increase in acellular capillaries was observed in db/db mouse cortex (*p<0.05) at Bregma level +0.5 mm. (c) A significant loss of pericyte coverage (PDGFR-β) was observed in db/db mouse cortex at Bregma +0.5 mm (**p<0.01). (d-g) Representative images: laminin staining (purple) (d); 'hot-spot' (arrow) area of vessel leakage in db/db mouse cortex (blue, laminin; red, IgG; green, albumin) (e); lectin (green) and Col-IV (red) in control and db/db mouse cortex for assessment of acellular capillaries (arrows) (f); and lectin (green) and PDGFR-β (red) in the control and db/db mouse cortex for measurement of pericyte coverage (g). Scale bar, 50 µm. (h) DEGs in endothelial cells. (i) DEGs in pericytes. Expression levels for individual genes (all p<0.001) are presented as the normalised counts in the db/db mouse cortex (Fig. 6b,f) alongside a loss of pericyte coverage (Fig. 6c,g). The top DEGs in pericytes included Cald1 (encoding downregulation of caldesmon 1), which is involved in smooth muscle contraction [35]. Dysregulation of metabolic genes was observed in db/db mouse pericytes, with a loss of mitochondrial complex-related genes such as mt-Co1 and mt-Nd4 (encoding mitochondrial NADH dehydrogenase 4) (Fig. 6i).
Cortical metabolism is significantly altered in db/db mice, with dysregulated metabolic pathway genes observed in specific cortical populations Our results indicated that metabolic dysfunction in the type 2 diabetes brain may be a consequence of NVU impairment. To validate these findings in vivo, metabolic profiling of the cortex was performed using acute brain slices (Fig. 7a). Glycolytic function was increased in db/db mouse cortex compared with controls (Fig. 7b). Non-glycolytic acidification (Fig. 7c), basal glycolysis (Fig. 7d) and glycolytic capacity (Fig. 7e) were all significantly increased in the db/db mouse cortex; however, no significant difference was observed in glycolytic reserve (Fig. 7f). A number of metabolism-related genes were differentially expressed in the db/db mouse cortex (Fig. 7g). For example, astrocytes and mature neurons showed increased expression of electron transport chain subunit genes and NADH dehydrogenase subunits. Together, these data indicate that the metabolic processes of cortical NVU cell populations are altered in db/db mice.
Cortical cell populations are also changed in humans with type 2 diabetes To validate our db/db mouse model findings, we studied samples from people with type 2 diabetes. Neuronal and glial cell populations were assessed in a small population of cortex samples from non-diabetic individuals and from individuals with type 2 diabetes. Staining of NeuN in frontal cortex sections indicated a qualitative reduction in neuronal cells in the frontal cortex of individuals with type 2 diabetes (Fig. 8a), although semi-quantitative analysis of NeuN western blots from parietal, temporal and frontal cortex samples showed no significant difference (Fig. 8c). A similar observation was seen after analysis of postsynaptic density protein 95 (PSD95) (ESM Fig. 5a). Noticeably, a reduction in GFAP protein was observed in samples from type 2 diabetes temporal cortex only (Fig. 8b,d). GFAP staining of human cortex showed an increased localisation with blood vessels (stained with wheat germ agglutinin) (Fig. 8b). Iba-1 was not significantly changed in any cortical region in individuals with type 2 diabetes when compared with healthy control individuals. A reduction in zonula occludens-1 (ZO-1) protein levels was observed in individuals with type 2 diabetes in the parietal and temporal cortex (ESM Fig. 5b). Occludin was unexpectedly increased in the parietal cortex of individuals with type 2 diabetes (ESM Fig. 5b). A statistically significant loss of platelet-derived growth factor receptor (PDGFR) α/β was observed in the parietal cortex of individuals with type 2 diabetes (ESM Fig. 5c).

Discussion
Recent evidence has shown a relationship between type 2 diabetes and cognitive decline, dementia and Alzheimer's disease [2][3][4]. Hyperglycaemia, insulin resistance, hypertension and dyslipidaemia in type 2 diabetes have a profound effect on the vasculature, leading to complications such as stroke, cardiovascular disease and diabetic retinopathy. Type 2 diabetes also has a damaging effect on neurons and glia, often relating to cell-cell communication defects across the retinal NVU [12]. The current study has detailed how db/db mice manifest cognitive deficiencies in unison with reduced brain weight and cortex size. Here, for the first time, we have catalogued a map of transcriptional profiles of the cerebral cortex from db/db mice, with a focus on the cell populations of the NVU.
Our scRNA-seq analysis identified a decrease in mature neurons, alongside an increase in microglia and immature neurons in the db/db mouse cortex. These novel findings, including disruption of several key neuronal regulatory genes, were complemented by the observed brain atrophy and cognitive decline in the db/db mouse. Meg3 and Mbp both showed reduced expression in db/db mice, indicating functional disruption. Downregulation of Mbp in mature neurons may indicate that golli-Mbp [28] (a known protein found outside the myelin sheath), which is involved in neuronal development [36] and Ca 2+ homeostasis, may be affected in type 2 diabetes. We also observed some evidence of NeuN reduction in individuals with type 2 diabetes. Together, these data suggest that type 2 diabetes may affect neuronal maturation, thereby having a profound impact on cognition. This finding is reinforced by similar findings in Alzheimer's disease [37] and other neurodegenerative diseases associated with progressive cognitive decline [38].
Previous reports have suggested that glial GFAP + area is increased in the db/db mouse cortex [39]. Our findings in db/ db mice and humans with type 2 diabetes remain ambiguous. Nevertheless, we observed a distinct transcriptional profile of cortical astrocytes in db/db mice. A notable example is Mt3 downregulation in db/db mouse astrocytes. This change in expression has been related to reduced uptake of Aβ leading to extracellular accumulation in Alzheimer's disease [40], highlighting a link between Alzheimer's disease pathology and diabetes-induced cognitive decline. Mt1 and Mt3 are also involved in protection against oxidative stress [30] and their downregulation in db/db mouse astrocytes may contribute to oxidative damage and enhanced inflammation in the type 2 diabetes cortex.
Cortical astrocyte metabolism was affected in db/db mice, including upregulation of genes involved in the formation of mitochondrial complexes (e.g. mt-Co2, mt-Nd1, mt-Nd2) and dysregulated lipid metabolism. A previous study reported that alterations in astrocyte lipid metabolism may be linked to degeneration of motor neurons [41]. Astrocytes play an important role in NVU function, interacting with and supporting neurons and blood vessels [42] and these changes in metabolic function in cortical astrocytes may contribute to NVU dysfunction in db/db mice.
Microglia are important modulators of homeostasis in the cerebral cortex. We found an increased number of microglia in the cortex of db/db mice, occurring in unison with overexpression of inflammatory pathways and genes involved in myelin damage, indicating neuronal injury in the diabetic brain [43]. ApoE is an established genetic risk factor for Alzheimer's disease with roles in Aβ clearance, glucose metabolism and proinflammatory responses [44]. However, high levels of leptin can influence ApoE expression [45], therefore this finding should be explored in other diabetic models before a conclusion is made. In diabetic retinopathy, neurodegeneration is known to be an important early event (prior to vascular complications [46]), and is thought to be driven by chronic inflammation. A recent study showed that depleting microglia in a high-fatdiet-induced model of diabetes was able to improve cerebrovasculature deficits [47]. Further understanding of the neuroinflammatory pathways mediated by microglia in the type 2 diabetes cortex will be beneficial in understanding the role of neuroinflammation in the progress of diabetes-related cognitive decline.
Diabetes is known to have a profound effect on vascular function. Our IHC data indicated reduced laminin + vascular density in the db/db mouse cortex, mirrored by increased acellular capillaries and loss of pericytes. This is in line with findings in the db/db retina, where vascular alterations can be seen from 10 weeks [48]. We also observed evidence of vascular leakage in specific 'hot spots' in the db/db mouse cortex, although this requires further characterisation. Top DEGs in endothelial cells suggested alterations to angiogenesis and vessel disease, while smooth muscle contraction and metabolism were dysregulated in pericytes. We saw a statistically significant reduction of ZO-1 protein expression in the parietal and temporal cortex of individuals with type 2 diabetes. We also observed region-specific increases in occludin expression, although the cellular source of this upregulation remains to be clarified. Loss of vascular density and pericyte coverage have been reported in diabetic retinopathy [49] and also in the cortex in Alzheimer's disease [8], strengthening the assertion that progressive NVU dysfunction in the brain during type 2 diabetes is linked to neuronal depletion and cognitive decline. Together, these findings of altered neuronal maturation, astrocyte metabolism and microglial inflammation in the db/ db mouse cortex highlight the impact of complex neuroglial interactions in type 2 diabetes-induced cognitive decline.
We report that db/db mouse cortex showed a significant increase in glycolysis and alterations to a number of genes involved in oxidative phosphorylation (e.g. mt-Nd2 and mt-Nd5 were reduced in immature neurons). This result suggests that at least some cells in the type 2 diabetes cortex may switch to anaerobic glycolysis as the main supply of energy (Warburg effect). Interestingly, this Warburg effect metabolism shift has been shown to drive neuronal degeneration in sporadic Alzheimer's disease [50]. Metabolism was also dysregulated in mature neurons and Fig. 7 Metabolic profiling of db/db mouse cortex. (a) Schematic describing the process of obtaining acute coronal sections for metabolic profiling by Seahorse X FE 96. Male db/db or control mice aged 16 weeks (n=5 mice per group) were used to generate acute coronal slices (250 µm) using the Leica VT100s Vibratome. The Seahorse analyser was used to assess glycolytic function, via an adapted glycolytic stress test protocol. (b) Traces from n=5 db/db and n=5 control mice (representative of seven or eight punches per mouse) following the adapted glycolysis stress test protocol. (c) Non-glycolytic acidification was significantly increased in db/db mice. (d) Basal glycolysis was significantly increased in db/db mice. (e) Glycolytic capacity was significantly increased in db/db mice. (f) No significant changes were observed between db/db and control mice when comparing glycolytic reserve. n=5 mice per group; seven or eight cortical punches per mouse. *p<0.05, **p<0.01. Unpaired Student's t test. Data are presented as mean values ± SEM. (g) Metabolism-related DEGs (all p<0.001) in db/+ (control) and db/db mouse cortex. Heatmap depicts normalised expression counts scaled between +1 and −1. Schematic in (a) was created with BioRender.com. ECAR, extracellular acidification rate ◂ 1 3 (d) Western blotting of human cortex samples (parietal, temporal and frontal) shows the expression of GFAP and Iba-1 in each location. GFAP expression was reduced in the temporal cortex (*p<0.05 vs control samples), n=3 individuals per group. Western blot quantification was analysed by normalising the protein levels to the Ponceau stain. C, control; T2D, type 2 diabetes; WGA, wheat germ agglutinin astrocytes. Mitochondrial dysfunction in diabetes has been widely described [51]. We propose that altered metabolic function of key component cells of the brain NVU may play a major role in the loss of vascular integrity and homeostasis leading to increased risk of neurodegeneration and cognitive decline.
Our study reveals distinct changes to the cells of the cerebral cortex in type 2 diabetes. One limitation of our approach is the use of IHC quantification to validate changes to cortical NVU populations, as IHC is primarily a qualitative technique. Future analysis should focus on characterisation and quantitative validation of cortical cellular heterogeneity and subcellular vulnerability in type 2 diabetes. Our dataset facilitates opportunities for molecular analysis and their validation in cell-cell communication in diabetes-related cortical dysfunction. Future studies in such diabetes models will provide further insights into the therapeutic potential of targeting these gene alterations in the NVU for treatment of diabetic comorbidities of memory and cognition.