Obesity and diabetes cause cognitive dysfunction in the absence of accelerated β-amyloid deposition in a novel murine model of mixed or vascular dementia

Mid-life obesity and type 2 diabetes mellitus (T2DM) confer a modest, increased risk for Alzheimer’s disease (AD), though the underlying mechanisms are unknown. We have created a novel mouse model that recapitulates features of T2DM and AD by crossing morbidly obese and diabetic db/db mice with APPΔNL/ΔNLx PS1P264L/P264L knock-in mice. These mice (db/AD) retain many features of the parental lines (e.g. extreme obesity, diabetes, and parenchymal deposition of β-amyloid (Aβ)). The combination of the two diseases led to additional pathologies-perhaps most striking of which was the presence of severe cerebrovascular pathology, including aneurysms and small strokes. Cortical Aβ deposition was not significantly increased in the diabetic mice, though overall expression of presenilin was elevated. Surprisingly, Aβ was not deposited in the vasculature or removed to the plasma, and there was no stimulation of activity or expression of major Aβ-clearing enzymes (neprilysin, insulin degrading enzyme, or endothelin-converting enzyme). The db/AD mice displayed marked cognitive impairment in the Morris Water Maze, compared to either db/db or APPΔNLx PS1P264L mice. We conclude that the diabetes and/or obesity in these mice leads to a destabilization of the vasculature, leading to strokes and that this, in turn, leads to a profound cognitive impairment and that this is unlikely to be directly dependent on Aβ deposition. This model of mixed or vascular dementia provides an exciting new avenue of research into the mechanisms underlying the obesity-related risk for age-related dementia, and will provide a useful tool for the future development of therapeutics.


Introduction
Alzheimer's disease (AD) is a neurodegenerative disease affecting the elderly. There are two major neuropathologies associated with AD: extracellular plaques containing β-amyloid (Aβ) and intracellular neurofibrillary tangles composed of the microtubule-associated protein tau. The combined insults of Aβ and tau accumulation are thought to promote the progressive synaptic failure and neuronal loss, leading to memory loss and cognitive impairment [1][2][3][4]. While familial forms of AD exist, sporadic AD is far more common. Though the two forms of AD ultimately reflect similar pathologies, the underlying causes vary. Familial AD is linked to specific mutations in amyloid precursor protein (APP) or presenilin (PS1 or PS2), leading to accumulation of toxic β-amyloid species in the brain by mid-life. Sporadic AD manifests later in life, and the triggers are less clear and likely complex. Though there are genetic components associated with sporadic AD, environmental factors, such as lifestyle (e.g. diet and exercise), are also likely to impact disease onset and progression.
Obesity is a major worldwide public health problem, and is associated with the metabolic disorder type 2 diabetes mellitus (T2DM). Diabetes is associated with cognitive decline in both rodents and humans [5][6][7]. Due to improved treatments, T2DM patients are living longer, putting them at increased risk for age-related complications. Although simply living to an older age increases the risk of Alzheimer's disease, there is a well-known (albeit poorly understood) link between obesity, T2DM and dementia [8]. The form of dementia afflicting these individuals combines elements of vascular pathology, small strokes and AD-related neuropathology. In fact, the amount of AD pathology is essentially unchanged in cases with a history of T2DM, while cerebrovascular pathology increases [9,10]. Vascular dementia, or even cerebrovascular dysfunction as a general AD comorbidity, is a poorly understood condition with no viable treatment options. This is due to cerebrovascular dysfunction being understudied as a major cause of dementia and the lack of useful model systems in which to develop therapies or to study the disease process.
In this paper, we describe the creation of a novel mouse model combining the key features of obesity, diabetes, and AD. We crossed the obese and diabetic db/db mouse [11][12][13] with the APP ΔNL/ΔNL × PS1 P264L/P264L knock-in model of AD [14,15]. The resulting mice (which we have called db/AD) are morbidly obese, glucose intolerant, insulin resistant, and display parenchymal amyloid plaques, similar to the parental lines. In addition, although these mice had profound cognitive impairment and marked cerebrovascular abnormalities, this does not appear to be driven by Aβ deposition. The db/AD mice will be a useful tool with which to study the intersection of T2DM and dementia.

Genotyping
Tail snips were collected prior to weaning. For some of the db and APP genotyping, tail snips were sent to Transnetyx (Cordova, TN) for purification and analysis. For those analyzed in our lab, as well as PS1 genotyping, genomic DNA was isolated and purified from tail snips using the Promega Wizard Genomic DNA kit (Promega; Madison, WI). db genotyping was performed using a single nucleotide polymorphism Taqman® genotyping kit (Applied Biosystems Life Technologies; Grand Island, NY). APP and PS1 genotyping were performed by PCR as described previously [16] using GoTaq® Flexi DNA Polymerase (Promega).

Mouse groups
The mice used for this study were broadly divided by age and will be referred to as young (1-4 months old: 3.0 ± 0.8 months), middle-aged (5-9 months old: 7.2 ± 1.6 months), and older (10-14 months old: 12.2 ± 1.0 months) based on the predicted lifespan of the db/AD mice (~15-16 months).

Glucose and insulin tolerance tests
Mice were fasted 3-6 hours prior to the start of the glucose tolerance test (GTT) or insulin tolerance test (ITT). All glucose measurements were obtained via tail bleed using a Bayer Breeze 2 glucometer and test strips (Bayer; Tarrytown, NY). For the GTT, a baseline measurement was obtained after which the GTT was initiated by intraperitoneal injection of dextrose (2 mg/g: Hospira; Lake Forrest, IL). Subsequent measurements were recorded at 15, 30, 60, and 120 minutes post-injection. For the ITT, a baseline glucose measurement was taken, after which insulin (0.75 U//kg: Eli Lilly; Indianapolis, IN) was injected intraperitoneally. Subsequent measurements were recorded at 15, 30, 60, and 120 minutes post-injection. Any glucometer reading of "HI" was set to 700 mg/dL for data analysis.

Blood pressure measurements
Blood pressure (BP) was measured using a Kent CODA 8 BP machine (Kent Scientific; Torrington, CT). Animals were allowed to acclimate to the tail blood pressure cuff for five minutes on a warming platform before recording BP measures. The BP measures consisted of 20 cycles of diastolic/systolic measures, with a 20 second rest period between cycles. After finishing the data collection, the mice were immediately released back into their home cages. The rodent restraints, cuffs, and warming platform were cleaned between animals; female animals were always run after male animals to avoid any possible irritation of the males. BP measures were performed at the same time each day to account for the possible influence of circadian rhythms.

Plasma measurements
Blood was collected upon decapitation in the presence of EDTA, centrifuged (1500 × g, 10 min.), and the plasma collected. Plasma leptin was measured by a commerciallyavailable, species-specific ELISA (EMD Millipore; Billerica, MA), according to package instructions.
qPCR Tissue was homogenized in Trizol™ (Invitrogen; Grand Island, NY) in order to isolate RNA, followed by phenol/ chloroform extraction. When needed, RNA was further purified by RNeasy columns (Qiagen; Valencia, CA). Expression of ECE1 and ECE2 were determined by two-step qRT-PCR, using iScript (BioRad; Hercules, CA) reverse transcription, followed by qPCR with PerfeCTa FastMix™ (Quanta BioSciences; Gaithersburg, MD). The geometric mean of the C T values for RPL30, cyclophilin, and RNA polymerase IIJ was used as an internal control to calculate and compare relative expression (2 -ΔΔC T ). Gene specific primer sets were obtained from IDT (Coralville, IA).
Insulin degrading enzyme (IDE) activity was measured using a commercially-available kit (EMD Millipore) according to manufacturer's instructions. Briefly, hemibrains were homogenized in Tris buffer (100 mg/mL) supplemented with PMSF and E-64, centrifuged (20,800 × g, 30 min., 4°C), and the supernatant used for the activity assay. Samples were compared against rat IDE. Fluorescence was measured at an excitation wavelength of 320 nm and an emission wavelength of 405 nm.

MRI
T2*-MRI was performed using a horizontal bore Bruker Clinscan (7.0 T, 30 cm, 300 MHz: Billerica, MA) imager equipped with a triple-axis gradient (630 mT/m and 6300 T/m/s) and a helium-cooled 14 K quadrature head cryo-coil, cooled to 20°K. T2*-weighted images were acquired with a 2D GRE sequence with at 34 μm × 34 μm × 400 μm resolution, 15 mm FOV, 25 degree flip angle, 10 averages, TR 165 ms, and TE 15.3 ms. Mice were imaged under constant isofluorane anesthesia and their body temperature and respiration were continuously monitored. At least ten equally-spaced images were taken of each mouse brain. Asymmetrically-occurring dark spots on the images were considered indicative of vascular events (confirmed histologically, see below), whereas symmetrically-occurring dark areas were considered to be blood vessels and were excluded.

Vascular corrosion casting
Vascular corrosion casting was performed as described [24]. Briefly, mice were anesthetized using pentobarbital (100 mg/kg), followed by transcardial perfusion with heparinized saline (0.9%). Following a brief perfusion with para-formaldehyde (4%), the brains were perfused with the polyurethane resin Pu4ii (4 mL/min: VasQtec; Switzerland). After allowing the resin to cure for at least two days, the brains were incubated in KOH (7.5%, 50°C, 48 h), followed by formic acid (5%, 50°C, 24 h). The tissue was subsequently frozen, then lyophilized to macerate the soft tissue. Finally, the casted brains were sputter-coated in palladium and viewed by scanning electron microscopy (Hitachi S-4300: Schaumburg, IL), using the middle cerebral artery as a landmark. Endothelial cell density was determined by endothelial cell nuclear imprints measured directly using Image J software. Aneurysm pathology was assessed on a 4 point scale based on clear data break points (0 = none; 1 = 1 possible; 2 = 1-3 definite; 3 = 4+ definite. Vascular density was determined by rank order of representative images using three blinded, independent reviewers. Images were scored from 1 (most dense) -26 (least dense), and the ranks from the three reviewers averaged.

Histology
Tissue was harvested and fixed in PBS-buffered 10% formalin for at least 24 hrs. For Aβ immunohistochemistry, hemibrains were embedded in a matrix and sectioned (30 μm) by NeuroScience Associates (Knoxville, TN). For Prussian blue staining, hemibrains were embedded in paraffin and sectioned to 8 μm using a microtome. For free-floating sections, the hemibrains were incubated in sucrose (10%, 20%, 30% sequentially for 24 hours each) for cryoprotection, then sectioned on a sliding, freezing microtome to 25 μm.

Behavioral testing
Testing was performed by the UK Rodent Behavioral Core (http://www.rodentbehaviorcore.uky.edu/default.aspx/0_UK_ Rodent_Behavior_Core). Mice were tested using the Morris Water Maze paradigm. The maze consisted of a circular pool (134.5 cm diameter) filled with 25°C water. A circular platform (11 cm diameter) was placed in the northeast quadrant 1 cm below the surface of the water so that it was not visible. Nontoxic tempura paint was used to create opaque water, thus obscuring the platform. The pool was placed behind dark curtains holding external maze cues. The cues were rotated each day. There were five consecutive training/acquisition days. On each-training day the animals swam four trials (rotating initial placement each time), lasting one minute each, with a five minute interval between trials. After a 30 minute rest upon the conclusion of training on the fifth day, we performed a probe trial where the platform was removed from the pool. The animal's location in the pool was recorded for one minute and used to calculate the time spent in the target quadrant and the number of times crossing the platform area. After the completion of training, mice were tested for visual acuity during which the external cues were provided along with a visibly-raised platform. The mice were tested for visual acuity in four trials during one day. Water Maze data (e.g. swim speed, distance, latency to platform. etc.) were collected and analyzed using Etho-Vision XT software (Noldus Information Technology; Leesburg, VA).

Statistics
Weight data were analyzed by student's t-test at each age using Microsoft Excel, and the probability adjusted using the Holm-Bonferroni method [27]. All other data were analyzed with SPSS (Hewlett Packard; Palo Alto, CA) using the general linear model (GLM) module for ANOVA with the independent variables gender, db genotype, and AD genotype (for an explanation of this model, see http://pic.dhe.ibm.com/infocenter/spssstat/ v21r0m0/index.jsp?topic=%2Fcom.ibm.spss.statistics.help %2Fidh_glm_multivariate.htm). Post-hoc multiple comparisons were conducted using Tukey's test, Dunnett's test, or similar. Chi-square analyses were performed on the visual acuity measurements for the Morris Water Maze. We performed correlation analyses using either Pearson's r or Spearman's ρ (parametric and nonparametric values, respectively), and adjusted probability using the Holm-Bonferroni method. For nonparametric comparisons, we used a Kruskal-Wallis ANOVA, or Mann-Whitney U test, where appropriate. For most presented statistics, we note an overall effect of genotype (db or AD) across the data set. In some cases, we also present a direct comparison between two different genotypes.

β-amyloid
We have shown previously that leptin downregulates expression of the γ-secretase components, particularly presenilin [28]. We reasoned that leptin resistance, as seen in obesity and diabetes, would likely increase PS1 expression in the brain and, as a consequence, β-amyloid deposition in the db/AD mice. In order to visualize the AD pathology, brains from middle-aged mice were sectioned and stained for Aβ (Ab4G8specific for Aβ 17-24 ). As expected, only mice containing the AD-related mutations in APP and PS1 were positive for Aβ-containing plaques (Figure 2a-d). Quantitation of plaque burden indicated that the number of plaques present in db/AD mice decreased relative to AD mice (0.104 ± 0.012 vs. 0.168 ± 0.013 A.U.; p < 0.001; n = 9 AD, 10 db/AD; not Cerebral amyloid angiopathy (CAA), in which Aβ is deposited in the vasculature, is a co-morbidity in many AD patients [29]. We therefore hypothesized that excess Aβ was created, but was deposited in blood vessels rather than the brain parenchyma in the db/AD mice. To investigate this possibility, we pursued a variety of histopathological staining techniques. There was no appreciable Congo Red, Thioflavin S, or resorufin staining in any of the aged db/AD mice tested, nor did we detect any Aβ immunoreactivity in large or small blood vessels (not shown). Additionally, there was no co-labeling of amyloid (by Amylo-Glo) and anti-collagen IV-labeled vessels (Figure 3a). These data demonstrate a lack of vascular deposition and indicate that CAA is not a primary pathology in these mice. Additionally, we hypothesized that excess Aβ could have been produced, but cleared from the brain at an increased rate in the db/AD mice. We therefore measured plasma Aβ levels. Neither

Cognitive deficit
Since the diabetes phenotype did not significantly impact amyloid accumulation, we next determined if there was an effect on cognition in these mice. We tested older mice using a standard Morris Water Maze paradigm. While there was no difference in swim distance on the first day of training (Figure 4a: p < 0.38), the db/AD mice had significantly longer swim paths on each of the subsequent days (p ≤ 0.04 on each of days 2-5), indicating a learning and/or memory impairment. Neither db nor AD mice showed any significant impairment at this age (p ≥ 0.13 for each genotype). Swim speeds were not significantly different on the first training day (WT 374 ± 18; AD 356 ± 21; db 368 ± 18; db/AD 352 ± 18 mm/s: p ≥ 0.4 for all comparisons: not shown) demonstrating that db/AD mice are capable of swimming as well as other genotypes. After the fifth day of acquisition trials, we performed a probe trial during which the platform was removed. In this trial, db/AD mice spent significantly less time in the target quadrant (Figure 4b: p < 0.04) and less time in proximity to the platform location (p < 0.01). Since diabetics are prone to retinopathy and blindness, we wanted to exclude a profound visual impairment that would affect performance on the Morris Water Maze. We, therefore, performed a visual acuity test in which the platform location was visible and cued. All animals, with the exception of one db/AD mouse, (See figure on previous page.) Figure 1 db/AD characteristics. The diabetic Lepr db/db mouse was crossed to the APP/PS1 knock-in model of AD to create the db/AD line. (a) Mice homozygous for the db gene were obese (female, 9 months old; left: Lepr db/db mouse; right: Lepr db/+ mouse). (b) Lepr db/db mice showed an impaired response to glucose (F[2,30] = 38.2, p ≤ 0.0001 for the db genotype overall, n = 13 Lepr +/+ , n = 5 Lepr db/+ , n = 18 Lepr db/db ; ** = p ≤ 0.01, * = p ≤ 0.05, Tukey's LSD). The AD genotype did not affect the GTT (not shown). (c -d) Lepr db/db mice showed substantial weight gain from an early age. Weight was relatively stable after~7 months, and there was no effect of the AD genotype (* = p ≤ 0.01, t-test; Holm-Bonferroni [27] correction; n =~14 mice / genotype). Female (e) and male (f) Lepr db/db x APP ΔNL/ΔNL /PS1 p264L/P264L mice had reduced survivability compared with other genotypes, though the males had a particularly high attrition rate. (N = 367 F/359 M: Lepr db/db x APP ΔNL/ΔNL /PS1 p264L/P264L , n = 56 F/53 M; Lepr db/db x APP +/+ / PS1 +/+ , n = 73 F/60 M; Lepr +/+ x APP ΔNL/ΔNL /PS1 p264L/P264L , n = 54 F/44 M; Lepr +/+ x APP +/+ / PS1 +/+ , n = 56 F/61 M; Lepr db/+ × APP ΔNL/ΔNL /PS1 P264L/P264L , n = 69 F/61 M; Lepr db/+ × APP +/+ / PS1 +/+ , n = 59 F/80 M) In subsequent experiments, we focused on the four main genotypes. For simplicity we have named them WT, db, AD, and db/AD. (g) Lepr db/db (db and db/AD) mice were insulin resistant (F[15,177] = 4.64, p ≤ 0.0001 for the db genotype overall; WT, n = 14; db, n = 13; AD, n = 20; db/AD, n = 18;~6 months of age; * = p ≤ 0.05, ** = p ≤ 0.01; Dunnett's test, relative to Lepr +/+ (WT and AD)). (h) Blood pressure was not different, although there was a modest tendency (F[2,15] = 3.24, p ≤ 0.1) for it to be slightly lower overall in Lepr db/db mice regardless of the AD genotype, at least at this age (n = 5-7 mice / genotype,~7-8 months of age).
were able to locate the cued platform (χ 2 = 3.22: p < 0.36). We also analyzed the visual acuity data using an ANOVA: the db genotype had no effect on either swim distance (WT 2610 ± 2059; AD 3576 ± 2511; db 5805 ± 1757; db/ AD 9700 ± 3173 mm: p < 0.07: not shown) or latency to platform (WT 10.6 ± 3.4; AD 11.7 ± 3.6; db 13.45 ± 2.4; db/ AD 21.6 ± 4.5 seconds: p < 0.12: not shown), indicating that visual impairment could not account for the deficit. These data indicate that intersection of the both diabetes and AD is necessary for cognitive impairment. and AD (c) mice displayed amyloid-containing plaques, while db (b) and WT (d) mice did not. The db genotype significantly upregulated both PS1 expression (p < 0.002 for db overall: e) and tau phosphorylation (p < 0.003 standardized to total tau: f) in the brain (middle-aged: n = 10 F/ 12 M db wild-type; 13 F/11 M db homozygotes; 22 F/28 M heterozygotes). The results are similar for phosphor-tau when it is not standardized to total tau (not shown) (g) Aβ oligomers were significantly increased in diabetic (db and db/AD) mice (p < 0.001 for db overall, WT, n = 8; db, n = 8; AD, n = 8; db/AD, n = 7) compared to non-diabetic mice. Additionally, oligomers were significantly higher in db/AD compared to AD mice ( # = p < 0.0005).

Synapse loss
Because synaptic dysfunction and subsequent loss have been implicated in AD-associated memory impairment [2][3][4], we next measured the amount of the synaptic marker PSD95 in older mice. Neither diabetes nor the AD genotype significantly affected the level of PSD95 in the brain (p > 0.1: Figure 5), indicating that the number of synapses is not substantially reduced in the db/AD mice at the age at which we have observed learning and/ or memory deficiencies. The db/AD mice spent less time in the target quadrant, and less time in proximity to the platform location (* = p ≤ 0.05, ** = p ≤ 0.01. Tukey's test, relative to WT). All groups performed similarly on the cued version of the task (both distance (p < 0.07) and latency (p < 0.12), indicating that the db and db/AD mice did not have a profound visual impairment (not shown).

Figure 3
Aβ is not deposited in the vasculature. Large numbers of activated astrocytes could be seen both around amyloid cores and around some larger blood vessels (arrowheads) in older db/AD mice (a) There was no significant co-staining of amyloid and collagen IV, indicating that Aβ was not deposited in blood vessels. Red: GFAP, Green: Collagen IV, Blue: Amylo-Glo. db/AD mice did not have significantly more Aβ 40 (b) or Aβ 42 (c) in plasma, compared with AD mice (n = 18 F WT; 18 F db; 14 F AD; 13 F db/AD: 7-12 months old: p > 0.06), though mice containing the AD mutations had significantly more than those without (p ≤ 0.0001 for AD overall). Though the activities were significantly different in mice homozygous for the db mutation, neprilysin (d; p < 0.02) and insulin degrading enzyme (IDE: e; p < 0.003) activities were not increased relative to db WT mice (1-4 months old: n = 5 F/7 M Lepr +/+ ; 8 F/4 M Lepr db/db ; 6 F/6 M Lepr db/+ ). Endothelin converting enzyme (ECE1: f) protein expression was unaffected by the db (p < 0.65) and AD (p > 0.1) genotypes (7-12 months old: n = 14 F/7 M WT; 16 F/5 M db; 12 F/5 M AD; 12 F/6 M db/AD). Similarly ECE1 (g) and ECE2 (not shown) mRNA expression was unchanged in diabetic mice (p > 0.38 for the db genotype overall), though ECE1 expression was reduced in AD mice relative to WT (* = p < 0.03: n = 1 F/4 M WT; 2 F/4 M db; 1 F/3 M AD; 1 F/3 M db/AD).

Cerebrovascular abnormalities
Since the learning and memory deficit in db/AD mice was not obviously attributable to accumulation of Aβ or synapse loss, we next focused on changes in the cerebrovasculature. We examined the brain vasculature in older mice using vascular corrosion casting followed by scanning electron microscopy. WT and db mice had normal appearing vasculature (Figure 6a). By contrast, AD and db/AD mice displayed a marked pattern of cerebrovascular pathologies. We observed evidence of widespread saccular aneurysms, often occurring at the vessel branch points (Figure 6b). Some of the mice tested also presented with extensive clusters of apparent aneurysms along the arteries and arterioles (Figure 6c) as well as arterial blebbing that may represent weakened areas of the vessel wall (Figure 6d). We next scored the aneurysm pathology on a four-point scale (0 = none; 1 = 1 possible; 2 = 1-3 definite; 3 = 4+ definite) and found that aneurysms were significantly more numerous in mice with the AD genotype (AD and db/AD mice 1.7 ± 0.23 vs. WT and db mice 0.45 ± 0.24: F[1, 15] = 14.14, p < 0.002 for the AD genotype overall; N = 4-5 F/ genotype: not shown). The presence of the diabetes genotype did not significantly increase the number of aneurysms (p < 0.2 for the db genotype overall).
Because aneurysms are unstable and prone to rupture, we next looked for evidence of hemorrhage in the AD and db/AD mice using Prussian blue staining for hemosiderin. Prussian blue staining showed a significant incidence of microhemorrhages in older db/AD mice (n = 7; χ 2 = 4.75, p < 0.03); we did not find microhemorrhages in genotypes other than the db/AD (n = 9: Figure 6e-f ), including AD mice, which also displayed significant aneurysm pathology.
We scanned a separate cohort of older mice using small animal magnetic resonance imaging (MRI) in order to visualize areas of hemorrhage and infarcts. Indeed, the majority of the db/AD mice tested (11/15) showed evidence of multiple vascular events by MRI (Figure 7a-b, h-j). Histological staining of brains from the scanned mice was negative for Prussian blue staining (indicating the lack of hemorrhage). Moreover, micrographs from the same neuroanatomical level as the largest event detected by MRI showed obvious necrosis in the surrounding tissue and an obvious lack of Prussian blue positive staining (Figure 7c-d). The histological data suggest that the vascular events are likely ischemic strokes. By contrast, no WT (0/9: Figure 7e) or db (0/10: Figure 7f) and only a small number of AD (2/9: Figure 7g) mice presented with strokes, suggesting that the presence of both the db and AD genotypes is required to promote these events (χ 2 = 21.769; p ≤ 0.0001). In addition, there were multiple events present in the db/AD mice (Figure 7h-j), whereas only one or two were present in the AD mice positive for strokes.

Passive immunization
Our data suggest that the intersection of the db and AD genotypes is necessary to induce strokes in these mice.
In light of this data and the absence of diabetes-induced amyloid accumulation, we believe that the stroke pathology is unlikely to be due to Aβ accumulation. In order  (c -d) The brain was sectioned transversely to obtain confirmation of stroke extent. Prussian blue staining with neutral red counterstain showed no evidence of hemorrhage, indicating an ischemic stroke event. P -A: posterior / anterior, for orientation (N.B.: section is perpendicular to scanning axis). (e-j)~70% of db/AD mice (n = 11/15) had strokes (arrowheads); these were rare in AD mice (g: n = 2/9), and not found at all in WT (e: n = 9) or db mice (f: n = 10). All mice imaged were 12-14 months old. The db/AD mice (h -j) often have multiple incidents as opposed to the two AD mice, which only displayed one or two small strokes. Representative cases are shown (all scans are at about the same neuroanatomical level).
to test this hypothesis, we next performed a pilot study in which we immunized older db/AD mice with an Aβ antibody (Ab42.5) for two months. Parenchymal Aβ was significantly reduced via immunization compared to agematched, untreated db/AD mice (~17% decrease; n = 6 / group; p < 0.006: not shown). Though Aβ was significantly reduced in the brain, there was no evidence that the stroke phenotype was rescued. Most of the treated mice imaged by MRI showed evidence of stroke (4/6 treated vs. 11/15 untreated; p < 0.67: not shown).

Vascular density
γ-Secretase has been implicated in the regulation of VEGF-dependent angiogenesis [30][31][32]. Since we observed an upregulation of PS1 expression in the db/AD mice in the absence of Aβ accumulation, we hypothesized that PS1 might contribute to the observed vascular pathology through the regulation of angiogenesis. We therefore measured the amount of blood vessels present in the brains of older db/AD mice. Staining for smooth muscle α-actin indicated that the brains of db/AD mice were significantly more vascularized than those of WT mice (1.35 ± 0.52 vs. 0.31 ± 0.52: n = 4/ genotype: p < 0.04: Figure 8a-b). SEM images showed a similar increase in the density of the cerebrovasculature in the brains subjected to vascular corrosion casting (Figure 8c-d). Indeed, median vascular density scores of the SEM images from three blinded, independent raters indicated that the db genotype significantly increased vascular density (Figure 8e: N = 19: p < 0.02, Mann-Whitney U-test). The AD genotype had a marginal effect, (p ≤ 0.05). Similarly, db/AD mice had a greater number of endothelial cells than the other genotypes (p ≤ 0.05; Kruskal-Wallis ANOVA), as measured by the endothelial cell nuclear imprints. Endothelial cell density was correlated with vascular density (Figure 8f: p < 0.03). Direct measurement of cell size (68 ± 27 cells / animal) from the middle cerebral artery indicated that endothelial cells were smaller in diabetic mice (F[3, 8] = 17.9, p < 0.01), and size was inversely correlated with density (R 2 = 0.41, p < 0.02). Collectively, these data indicate that db/AD mice had an increase in the number of cerebral blood vessels, supportive of increased angiogenesis or arteriogenesis.

Discussion
The db/AD model We have created a unique mouse model that encapsulates features of both T2DM and AD-the db/AD mouse. These mice are morbidly obese and glucose intolerant at a young age (Figure 1a-d), and have a profound cognitive impairment by 12 months (Figure 4). The db/AD mice display decreased survival (Figure 1e-f), the cause of which is currently unknown. Male db/AD mice appear to be more susceptible to premature death, though sexual dimorphism has been noted in many AD models [33][34][35]. While their lifespan is shortened relative to control genotypes, we were able to routinely age the db/AD mice beyond 12 months, allowing significant Aβ accumulation, plaque formation, stroke pathology, and cognitive impairment.
Contrary to our expectations, the db/AD mice did not exhibit increased parenchymal Aβ accumulation compared with the normoglycemic AD mice (Figure 2h-k), in spite of the observed increase in PS1 expression (Figure 2e). Aβ oligomers were modestly elevated in both db and db/AD mice (Figure 2g), though the potential impact of this increase is unknown at this point. It is possible that the detected oligomers are formed from murine Aβ and, thus, are not toxic. The reason for the relative dearth of excess Aβ in db/AD mice is unclear, though it does not appear to be due to stimulation of clearance mechanisms. While we cannot rule out clearance by other enzymes, the major enzyme activities that proteolyze Aβ (neprilysin, IDE, and ECE) were not increased in db/AD mice (Figure 3d-g), nor was there an increase in peripheral Aβ in the plasma (Figure 3b-c). In addition, we found no evidence that Aβ is deposited in the vasculature (Figure 3a), despite using multiple different staining techniques. Based on this data, it is likely that excess Aβ is simply not made in db/AD mice. In addition, there is no evidence of a significant reduction in the number of synapses in older db/AD mice ( Figure 5). These findings indicate that neither CAA nor synaptic loss causes the cognitive decline observed in our mouse model.
The most striking feature of this mouse model is the severe vascular abnormalities that are present, apparently in the absence of a corresponding increase in Aβ deposition. Older AD and db/AD mice exhibited profound aneurysm pathology (Figure 6b-d) and db/AD mice had small strokes (Figure 7). Though we did observe a few areas of hemosiderin-positive staining in those animals with the largest number of vascular events (Figure 6e-f ), we did not see substantial numbers of hemorrhages in the db/AD animals. Indeed, it is possible that the more extensive pathologies observed by SEM are representative of ischemic stroke, but take on this appearance during the vascular corrosion casting process. In addition, the largest event observed by MRI (Figure 7a-b) did not stain positive for hemosiderin (Figure 7c-d) and was likely ischemic in nature. We feel that infarction is the likely cause of these events, but further characterization will be needed. This is broadly consistent with the type of cerebrovascular disease observed in human diabetics [36,37]. Given that the db/AD mice were the only genotype to exhibit both stroke pathology and cognitive impairment, we believe that it is these strokes that are responsible for the observed cognitive decline. (e) Median vascular density scores of SEM images from three blinded, independent raters (* = p ≤ 0.05, ** = p ≤ 0.01 compared to WT, Mann-Whitney U-test). Endothelial cells were also directly counted on five randomly-selected arteries / animal (* = p ≤ 0.05 compared to WT; Kruskal-Wallis ANOVA). Direct measurement of cell size (68 ± 27 cells / animal) from the middle cerebral artery indicated that db/AD endothelial cells were smaller (F[3,8] = 7.8, p < 0.01) than those from WT mice, and as expected size was inversely correlated with density (R 2 = 0.41, p < 0.02; not shown). (f) Endothelial cell density was also correlated with vascular density (p = 0.03).

Mechanism of vascular pathology
Based on our data, it is likely that the aneurysm and stroke pathologies are separable events. Aneurysms were prevalent in AD animals, regardless of diabetic phenotype and were not exacerbated by diabetes. This suggests that the aneurysms may be caused by some feature of the AD genotype. While aneurysms are not typically associated with AD in humans, increased blood vessel tortuosity, which is associated with aneurysms in other diseases, has been observed [38,39]. In addition, mutations in the presenilin substrate Notch are associated with thoracic aneurysms, likely through crosstalk with TGFβ signaling [40,41]. The mutation in PS1 present in the AD mice may also affect this Notch signaling pathway, resulting in the aneurysm pathology.
On the other hand, the intersection of the db and AD genotypes was necessary to induce strokes in these mice ( Figure 7). In light of this data and the absence of diabetes-induced amyloid accumulation, we believe that the stroke pathology is unlikely to be due to Aβ accumulation. This hypothesis was supported by preliminary data from our passive immunization study, which showed that stroke incidence was not reduced in db/AD mice treated with an anti-Aβ antibody, though brain Aβ levels did decrease. While interesting, a more extensive study will be needed for a more definitive conclusion.
Diabetes itself has profound effects on the vasculature. Obesity and diabetes are associated with hypertension and atherosclerosis [42]. In addition, diabetic rodents, including db/db mice, have increased neovascularization such as angiogenesis and arteriogenesis [43][44][45]. This neovascularization consists of immature, unstable blood vessels that display increased permeability of the blood-brain barrier. Similar pathologic angiogenesis occurs in diabetic retinopathy and is thought to involve presenilin and γ-secretase regulation of VEGF signaling [30,46]. We have evidence that PS1 expression increased in diabetic mice ( Figure 2e) regardless of the AD mutations present-as expected with the use of "knocked-in" genes under endogenous promoters. Consistent with PS1 upregulation, the db/AD mice have a significantly higher density of blood vessels in the brain than any of the other genotypes tested (Figure 8). Further studies will be needed to determine if neovascularization may indeed play a role in the strokes and/or cognitive impairment, or if some other diabetes-related phenomenon underlies these pathologies. We have shown previously that leptin downregulates PS1 expression in both in vitro and in vivo models [28]. It will be interesting to determine if leptin resistance in the db/AD mice contributes to neovascularization via regulation of the γ-secretase complex.

A unique model of mixed dementia
The form of dementia afflicting diabetic individuals combines elements of vascular pathology, small strokes and AD-related neuropathology. In fact, the amount of AD pathology is essentially unchanged in cases with a history of T2DM, while cerebrovascular pathology increases [9,10]. The db/AD mice share these features. One way of looking at this seemingly paradoxical observation is that cerebrovascular pathology lowers the threshold for incipient AD pathology to become unmasked as a clinical dementia as has been suggested elsewhere [10].
A small number of studies have examined the linkage between obesity, diabetes and dementia in rodent models [47,48]. The majority of these are focused on two paradigms: treatment with streptozotocin (STZ) and feeding a high fat, or typical Western, diet (TWD). STZ, a pancreatic islet toxin, is primarily used to model type I diabetes; thus it does not address the issue of obesity. Although TWD feeding induces obesity, and has some short-term effects on AD-related neuropathology in these models [49][50][51], these studies have failed to provide any detailed mechanistic insights into how obesity might influence the development of age-related neurologic disease. Further, TWD feeding does not have strong long-term effects on AD and vascular dementia-related neuropathology [52]. Studies utilizing genetic models of diabetes have been more limited. When Tg2576 mice, which overexpress APP ΔNL , are crossed with Irs2 −/− insulin resistant mice, the resulting animals show reduced amyloid pathology [53]. In addition, a recent study examined the outcome of a cross between leptin-resistant ob/ob mice and APP23 mice [54]. These animals showed a very early Morris Water Maze deficit (2-3 months old) unrelated to amyloid load, as the animals had no plaques and no differences in Aβ levels compared with non-diabetic controls. Even at the oldest age examined (12 months old), plaque pathology in these mice was virtually nonexistent, although there was some vascular amyloid in a very small number of animals (n = 3). The choice of parental mouse lines has a profound effect on the viability of the resulting mice as well: a cross between the ob/ob and Tg2576 lines yielded animals with significantly reduced viability [55]. While our data are in broadly supportive of these other studies, the db/AD mice are unique in that they have Aβ plaques, very little vascular-associated Aβ, and profound underlying vascular abnormalities, even in the absence of a high-fat diet.
In summary, the db/AD mouse is a unique model of mixed dementia, possessing both AD-related and vascular pathologies. Older mice present with extensive stroke pathology, arising from a combination of the diabetic and AD phenotypes, thus leading to significant cognitive impairment. While these data suggest that Aβ is not a primary factor in the observed cognitive impairment, we cannot exclude the possibility that a soluble form of Aβ, such as oligomers, may play a role in the cognitive decline. Future studies will focus on the mechanisms