All procedures involving animals were approved by the Danish National Committee on Health Research Ethics in accordance with the European Council’s Convention for the Protection of Vertebrate Animals Used for Experimental and Other Scientific Purposes and followed the Animal Research: Reporting In Vivo Experiments (ARRIVE) guidelines. The animal housing facility has been accredited by the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC) and the Federation of Laboratory Animal Science Associations (FELASA).

We used 5xFAD mice (Tg(APPSwFlLon,PSEN1*M146L*L286V)6799Vas/Mmjax; JAX strain: 006554) and age-matched wild-type littermate (WT) mice on the mixed B6/SJLF1J background (The Jackson Laboratory, Bar Harbor, ME, USA). The colony was maintained by backcrossing to B6/SJL hybrids at the department of Drug Design and Pharmacology, University of Copenhagen. The 5xFAD mice overexpress five familial AD mutations in human APP and PSEN1 under the control of the Thy1 promoter, resulting in rapid deposition of Aβ with no deposition of tau (Oakley et al., 2006). At the age of 6 months and older, 5xFAD mice develop substantial pathology expressed as Aβ plaques, alterations in behavior, and memory deficits (Oakley et al., 2006; Jawhar et al., 2012; Forner et al., 2021).

Female 5xFAD mice exhibit more severe Aβ pathology than males (Oakley et al., 2006), thus, only females were used in the study. In total, we used 21 5xFAD and 21 WT mice. Of those, eight 8–10-month-old (m.o.) mice of each genotype were used to measure AMT; ten 7–11-m.o. 5xFAD and WT mice were used to assess NVC. Paracellular leakage was measured in six 5xFAD and seven WT mice after the measurement of NVC; one 5xFAD mouse was excluded from the analysis of paracellular leakage because of a substantial drift along the optical axis. Three 5xFAD and three WT mice (all 7-m.o.) were used for immunohistochemistry (IHC) analysis of the fibrinogen leakage and pericyte density. Genotyping of 5xFAD mice was performed from tail clippings by a standard PCR protocol (JAX protocol: 23370) for the APP gene using the following primers: transgene forward: AGG ACT GAC CAC TCG ACC AG (olMR3610), transgene reverse: CGG GGG TCT AGT TCT GCA T (olMR3611), internal positive control forward: CTA GGC CAC AGA ATT GAA AGA TCT (olMR7338), internal positive control reverse: GTA GGT GGA AAT TCT AGC ATC ATC C (olMR7339).

In addition, we used nine 6-7-m.o. male C57BL/6J mice and one 6-m.o. transgenic reporter mouse expressing cytosolic green fluorescent protein (GFP) under the endothelial-specific receptor tyrosine kinase (Tie2) promoter [heterozygous Tg(TIE2GFP)287Sato/J (Tie2-GFP) mice; The Jackson Laboratory (JAX) #003658].

All mice were housed in ventilated cages under a 12 h light/12 h dark cycle, at 50 ± 10% relative humidity, at room temperature, with ad libitum access to food and water.

Mice underwent surgery as previously described (Kucharz and Lauritzen, 2018). Briefly, anesthesia was induced by intraperitoneal (i.p.) injections of xylazine (10 mg/kg) and ketamine (60 mg/kg). During surgery, mice were kept anesthetized by supplementary i.p. injections of ketamine (30 mg/kg) every 25–30 min. A 3-mm diameter craniotomy was drilled over the somatosensory barrel cortex, the dura was removed, and the opening was sealed with a lukewarm agarose solution and then covered with a coverslip. Artificial cerebrospinal fluid (aCSF) (in mM: NaCl 120; KCl 2.8; NaHCO3 22; CaCl2 1.45; Na2HPO4 1; MgCl2 0.876, and glucose 2.55; pH = 7.4) was applied to the cranial window to keep the brain moist. To allow the insertion of a recording microelectrode, an opening was left at the lateral side of the craniotomy between the rim of the skull bone and the coverslip. After surgery, the anesthesia was switched to α-chloralose (0.5 g/mL) infused through the venous catheter at the rate of 0.01 mL/h/10 g.

To study NVC in the somatosensory cortex, we stimulated the contralateral trigeminal nerve using custom-made bipolar electrodes, as previously described (Kucharz and Lauritzen, 2018). Briefly, the cathode was positioned relative to the hiatus infraorbitalis (IO), and the anode was inserted into the masticatory muscles. Thalamocortical IO stimulation consisted of a 15-s train of 1-ms pulses (pulse amplitude 1.5 mA) at the rate of 2 Hz (ISO-flex; A.M.P.I.). Stimulation was triggered using Spike2 software (v. 7.2.; Cambridge Electronic Design) and was synchronized with image acquisition by the two-photon microscope (Olympus FVMPE-RS, v01.02.01.27).

To monitor brain activation during whisker-pad stimulation, we recorded local field potential (LFP) as previously described (Jessen et al., 2015). Briefly, single-barreled borosilicate microelectrode (resistance 1.5–2.5MOhm) was inserted into the whisker-barrel cortex, and the reference electrode was placed under the skin of the neck of the mouse. The signal was first preamplified (gain 10x; low-pass cutoff frequency, 3,000 Hz; DP-311 differential amplifier, Warner Instruments), then additionally amplified and filtered (gain 100x, high-pass cutoff frequency 1 Hz; CyberAmp 320, Axon Instruments). Digital sampling was performed using a 1,401 mkii interface (Cambridge Electronic Design) connected to Spike2 software (v. 7.2.; Cambridge Electronic Design) at the sampling rate 20 kHz. We calculated the amplitude of the negative peak of the LFP (Jessen et al., 2015) and averaged the LFP amplitudes over all but the first LFP within a stimulation train.

We used a FluoView FVMPE-RS two-photon microscope (Olympus, v01.02.01.27) equipped with a tunable femtosecond laser (Mai-Tai DeepSee). The fluorescence was recorded using a 25x (NA = 1.05) water immersion objective and high-sensitivity GaAsP detectors.

To test our AMT protocol under controlled conditions, we measured AMT in mice with acute systemic inflammation, which is accompanied by an increase in AMT (Banks and Erickson, 2010; Wang et al., 2019). Inflammation was induced in three C57BL/6 J mice by administering 5 mg/kg i.p. lipopolysaccharide (LPS) (500X lipopolysaccharide from E. coli O111:B4; #00–4976-93, Thermofisher) (Qin et al., 2007). Because LPS triggers the release of proinflammatory factors into serum within 1–2 h (Tateda et al., 1996; Qin et al., 2007), we injected BSA-Alexa488 one hour after LPS administration and quantified AMT two hours after injection of BSA-Alexa488.

We generated two U-net-based models (Ronneberger et al., 2015) to quantify AMT automatically. Model-1 takes as input a single image, consisting of fluorescence and autofluorescence channels, and generates a binary mask of the blood vessel; Model-2 receives the same input as Model-1 and generates a binary mask of BSA-Alexa488 punctae (Supplementary Figure S2). The surface area of the vessel was calculated from the vessel mask as the number of vessel pixels times pixel area (0.124 × 0.124 μm²) multiplied by π, and the number of albumin punctae was calculated from the punctae mask. As in the manual analysis, the density of the punctae at each type of vessel was calculated as the total number of punctae divided by the total surface area obtained from all ROIs of this vessel type.

For the model training and validation, 77 ROIs of pial arterioles and pial venules were cut from the 2 PM hyperstack data recorded in two 5xFAD and three WT mice. Each ROI was maximum intensity projected along the optical axis resulting in a set of five frames per ROI and two channels: one channel captured fluorescence from BSA-Alexa488, and another channel captured autofluorescence of the brain tissue. By using both channels, Model-2 trained to recognize BSA-Alexa488 punctae but not autofluorescence (Supplementary Figure S3). The total number of projections for each channel was 385. BSA-Alexa488 punctae and blood vessels were manually annotated by drawing binary segmentation masks, which were used as ground truth during the training. Sixty-five ROIs (325 projections) were randomly allocated for the training of the models, and the remaining 12 ROIs (60 projections) were allocated for validation. The training and validation data were independent, in terms of not overlapping, with the data used to quantify AMT.

Model-1 and Model-2 were trained independently from each other. The training maximized the Dice similarity coefficient (DICE) until convergence in the validation set (Supplementary Figure S4). DICE was defined as follows (Dice, 1945; Asgari Taghanaki et al., 2021):

where yˆ and y are the prediction and the ground-truth segmentation masks, respectively, and yˆ × y is the element-wise multiplication of the predicted and ground-truth masks. The sums are calculated over all the indices in the masks. The DICE coefficient was chosen over other objectives to accommodate the large class imbalance between background and punctae pixels. The punctae and vessel segmentation models were trained for 200 and 100 epochs, respectively.

During training, the input projected ROIs were augmented by random rotating, scaling, shearing, contrasting, brightening, and transposing (Shorten and Khoshgoftaar, 2019).

Mice were anesthetized with Xylazine/Ketamine and transcardially perfused for 1–2 min with phosphate-buffered saline (PBS) at the rate of 10 mL/min, then for 4–5 min with ice-cold 4% paraformaldehyde (PFA) in PBS. Next, the brains were post-fixed in 4% PFA for 24 h at 4°C and subsequently cryoprotected in 25% sucrose and 0.1% sodium azide in PBS for 24 h at 4°C. The brains were frozen in dry ice and sectioned into 30 μm thick coronal sections using a sliding microtome Microm HM450 microtome (ThermoFisher Scientific).

To perform fibrinogen staining, we carried out antigen retrieval in Tris/EDTA buffer (pH9.0) for 30 min at 80°C, followed by a 20-min quenching step in 3% H₂O₂ and 10% methanol in Tris-buffered saline (TBS). The sections were then washed 3x in TBS and pre-incubated with 5% normal goat serum in 0.25% triton-X in TBS (TBS-T) for 1 h at room temperature (RT). Next, the sections were incubated overnight at RT with primary antibody anti-fibrinogen (1:2000, ab227063, Abcam) in 5% normal goat serum TBS-T solution. After the washing step of 3x in TBS-T for 10 min each, the sections were incubated in 1% BSA in TBS-T with a biotinylated secondary goat-anti-rabbit antibody (1:200; Vector Laboratories Inc.). Next, the sections were rewashed 3x in TBS-T for 10 min each, followed by incubation with an avidin-biotin-peroxidase complex solution for 1 h at RT, and washed 3x again in TBS-T for 10 min each. Finally, the staining was visualized using 3,3′-diaminobenzidine (DAB) and 0.01% H₂O₂ according to the manufacturer’s instructions. The sections were then mounted on chromatin-gelatin coated glass slides, stained in 0.1% cresyl violet solution for 3 min, dehydrated in increasing alcohol solutions (70, 95, 100%), cleared in xylene, and coverslipped using DPX mountant (Sigma-Aldrich) for imaging.

For the confocal imaging, after the initial washing step from the anti-freeze solution, the brain sections were preincubated with 5% normal donkey serum in PBS with 0.3% Triton-X-100 for 1 h at RT. The sections were next incubated overnight at RT in PBS with 0.025% Triton-X-100 (PBST) with 3% normal donkey serum solution containing primary antibody anti-CD31 (1:400, BD Biosciences, BD550274). After washing the sections with PBS 3x, an antigen retrieval step was carried out in citrate buffer (pH6) for 30 min at 80°C. Next, the sections were washed again 3x for 10 min in PBS, then incubated overnight at RT with the primary antibody anti-PDGFR-β (1:100, Cell Signaling #3169). Subsequently, the sections were rinsed in PBST and incubated with Cy3 (1:500; donkey anti-rat, Jackson ImmunoResearch, #712–165–153) and AlexaFluor 647 (1:500, donkey Anti-Rabbit, ab150075) secondary antibodies, diluted in 10% normal serum in PBST for 1 h at RT. Next, the sections were rinsed 3x for 10 min in PBST and stained with DAPI for 10 min, followed by a washing step in PBS. Finally, the brain sections were mounted on gelatin-covered glass slides and coverslipped using an antifading mountant (ProLong Diamond Antifade Mountant, Thermofisher Scientific, #P36965).

We used bright-field imaging to assess fibrinogen extravasation. The imaging was performed using BX53 upright transmitted light microscope (Olympus), equipped with 4 × 0.16NA UPLSAPO objective for collecting section overview images and 20 × 0.75NA UPLSAPO super apochromat objective for blood vessel imaging. The light was collected by a DP73 camera with an exposure time of 5.9 ms at ISO 200 and in 24-bit sRGB Color mode with the same microscope settings between all sections and genotypes.

We assessed the fibrinogen leakage by quantifying both the average fibrinogen within a set of defined brain regions and at the proximity of the brain vessels. First, the sRGB images were split in ImageJ into the channels with the highest specificity toward absorbance of cresyl violet (red channel), intermediate signal (green channel), and fibrinogen (blue channel). Next, the fibrinogen channel was inverted, so the increase in pixel intensity corresponds to the increase in the fibrinogen in the tissue [channels marked fib. (inv.)].

The overall level of fibrinogen in the brain was quantified from the low magnification overview images by averaging the signal within ROIs covering different brain regions: cerebral cortex, hippocampus, and hippocampal subregions (stratum lacunosum-moleculare (s.l-m) and the remaining hippocampal areas).

The local vessel-associated fibrinogen was quantified from the high-magnification images by comparing the intensity profile across the vessel wall between WT and 5xFAD. The vessel boundary was manually traced using the segmented line tool (ImageJ). A region around the trace (3 μm toward lumen and 11 μm toward parenchyma) was extracted and straightened into a rectangle image with dimensions of L (=vessel length) and x (= constant width). To minimize the interference of the cresyl violet nuclei labeling on the measured signal of fibrinogen, we manually set a threshold mask that encompassed nuclei and removed these pixels from the corresponding channel with fibrinogen. Subsequently, the data was projected along the L-axis to calculate the average signal profile across the vessel. This procedure has been performed for both sides of each vessel. For each animal, a weighted average profile was calculated to account for the different lengths of sampled vessels.

We performed confocal fluorescence imaging to quantify pericytes on the cortical microvasculature. The images were collected using DMi8 inverted fluorescence microscope (Leica Microsystems) equipped with HC PL APO 40 × 1.3NA oil-immersion objective. We used 405 nm, 552 nm, and 638 nm laser diodes to excite DAPI (nuclei), Cy-3 (endothelium), and AlexaFluor 647 (pericytes), and the emitted fluorescence was collected by a hybrid detector after 440–480 nm; 565–585 nm; and 655–685 nm bandpass, respectively. The data was collected as Z-stacks with triple line averaging and at the resolution of 1,024 × 1,024 pixels and Z-step = 2 μm, corresponding to the volume of 388 μm × 388 μm with average stack depth = ∼26 μm. To minimize the bleed-through between the channels, the imaging was performed for each channel separately, i.e., in sequential mode. In addition, we used a 488 nm excitation wavelength to collect the signal after a 655–685 nm bandpass, obtaining an autofluorescence image of the brain tissue. The images were exported to ImageJ as multi-channel .tiff files for further analysis.

We quantified the number of pericytes per unit length of capillaries. Ten Z-stacks were acquired from the somatosensory cortex (2–3 mm posterior from bregma) in 3–4 sections per animal. First, we subtracted the autofluorescence signal from the CD31 imaging channel. The vessel lengths were calculated manually using ImageJ. Only vessels that corresponded to the diameter of capillaries (<10 μm) were considered. Next, we counted pericyte cell bodies with the inclusion criteria: (i) the presence at the microvessel and (ii) typical “bump on a log” pericyte morphology with a clear DAPI signal in nuclei. For each Z-stack, we calculated the pericyte density as the number of capillary-associated pericyte cell bodies divided by the total length of the capillaries.

Statistical analysis was performed using Scipy 1.9.1 (Virtanen et al., 2020) in Python 3 (Van Rossum and Drake, 2009). Linear regression was performed using Python statsmodels 0.13.5 (Seabold and Perktold, 2010). Data are presented as mean ± standard error of the mean (SEM) unless otherwise stated. Statistical tests were performed on averages within individual mice, i.e., on statistically independent data. The sample sizes were based on estimates from our previous research (Mathiesen Janiurek et al., 2019; Zambach et al., 2021). Normality was tested by using the Shapiro–Wilk test. Differences between groups were tested by Student’s T-test or Mann–Whitney U test for normally and non-normally distributed data, respectively. The outcome of the tests was assessed at a significance level of 0.05. We avoided outlier testing and elimination to prevent inflation of the type-I error rate (see Supplementary material, section 4 for details). Descriptive and test statistics, as well as p-values, are listed in Supplementary Table S2. Blinded quantification of paracellular leakage and AMT was impossible because it involved manual placement of ROI or counting of BSA punctae, during which the autofluorescence of Aβ plaques was visible. Fibrinogen image collection and data analyses were performed by an experimenter blinded to the experimental conditions. Analysis of LFP and NVC was automated and blinded toward the genotype. The blinding was not possible during pericyte IHC assessment due to the clear presence of the Aβ aggregates on the fluorescence images in 5xFAD mice.

We characterized the BBB and NVC in 7–11-m.o. 5xFAD mice using 2 PM in vivo at cortical depth up to 250 um. 5xFAD is a widely used AD model because of a widespread Aβ plaque deposition, starting at 2 months (Oakley et al., 2006; Forner et al., 2021).

The mice underwent microsurgical preparations to monitor and maintain optimal systemic parameters during 2 PM imaging (Figure 1A). First, we ascertained the presence of Aβ plaques in the brain parenchyma by their autofluorescence (Kwan et al., 2009; Gao et al., 2019), and further confirmed their identity by in vivo labeling with thioflavin-S (Figure 1B; Supplementary Videos S1, S2). The Aβ plaques were visible in 5xFAD mice at all cortical depths, and the mean density of the plaques was 2,217 ± 238 plaques/mm³ (Figure 1C), consistent with previous reports (Giannoni et al., 2016; Yin et al., 2022; Supplementary material, section 2).

Next, we asked whether Aβ pathology in 5xFAD mice is associated with disinhibition of AMT, a vesicular transport of blood-borne molecules (e.g., albumin) across the BBB. The increase in AMT is evidenced in aging, neuroinflammation, and stroke and is suggested to underlie and accelerate neurodegeneration (Knowland et al., 2014; Sadeghian et al., 2018; Yang et al., 2020). Yet, AMT remained unstudied in AD.

To monitor AMT, we injected fluorescently labeled BSA-Alexa488 into the bloodstream (Knowland et al., 2014; Mathiesen Janiurek et al., 2019) and continuously recorded Z-stack images capturing all types of cerebral blood vessels, i.e., pial and parenchymal arterioles and venules, and capillaries for 120 min (Figure 2A). Following the injection, we observed a gradual formation of BSA punctae at the BBB, indicating vesicular uptake of circulating BSA-Alexa488 into the endothelium (Figure 2B; Supplementary Video S3). The punctae moved substantially along the BBB interface, indicating that they represent free vesicles rather than vesicular pits attached to the endothelial membrane (Supplementary Videos S4, S5). The localization of the punctae to the endothelium was additionally confirmed by overlapping fluorescence signals from GFP-labeled endothelium and BSA-Alexa594 punctae in Tie2-GFP mice (Figures 2C,D). BSA formed punctae irrespective of the type of conjugated fluorophore, i.e., AlexaFluor 488 or AlexaFluor 594 (Figure 2E).

The AMT was observed in both 5xFAD and WT mice (Figures 2F–H), and it was more prominent in arterioles than in capillaries and venules (Supplementary Figure S5), which is consistent with earlier reports (Mathiesen Janiurek et al., 2019), and with the expression of caveolin-1 that forms AMT transport vesicles (Chow et al., 2020). However, AMT was the same in WT and 5xFAD, both at the level of the whole microvascular tree (Figure 2G) and at the level of corresponding vessel types (Figure 2H). AMT is also known to increase with age (Yang et al., 2020). To test whether the differences in age within experimental groups (7–11 m.o.) influenced our AMT results, we assessed age-related changes in AMT within both WT and 5xFAD experimental groups. The age span of 4 months within the animal groups had no effect on AMT in all vessel types except penetrating arterioles; however, the slope of AMT versus age in penetrating arterioles was not different between WT and 5xFAD mice (Supplementary Figure S6A).

To control our method of AMT quantification, we exposed C57BL/6J mice to LPS, which induces neuroinflammation and a rise in AMT (Marchiando et al., 2010; Wang et al., 2019). Expectedly, LPS-treated mice exhibited AMT increase, specifically in pial arterioles and venules (Supplementary Figure S7). Noteworthy, the AMT level in the control C57BL/6 J (Supplementary Figure S7) was lower than in the WT mice (Figure 2), which may be related to the younger age of the C57BL/6 J mice (Yang et al., 2020) or to the difference in the genetic backgrounds. Overall, AMT was unchanged in 5xFAD mice, and the lack of differences between WT and 5xFAD mice is unlikely caused by a sensitivity limit of our in vivo assay system.

To support our findings, we implemented a machine learning algorithm. Automatic quantification of AMT by machine learning is more objective than manual quantification and provides a measure less prone to human bias. We designed two machine learning models: Model-1 to detect vessels and Model-2 to detect BSA-Alexa488 punctae. We used the two models in conjunction to automatically quantify AMT (Figure 3; Supplementary Figures S2, S4). The performance of the models was measured by the DICE coefficient [estimates the similarity between punctae detected by the model and punctae labeled by the user (see Methods)], which reached 0.98 for Model-1 and 0.68 for Model-2 [DICE ranges between 0 (no similarity) and 1 (perfect similarity)]. Because of the small size of punctae, DICE = 0.68 was regarded as satisfactory for the counting (Asgari Taghanaki et al., 2021). Punctae density quantified by our model tended to be smaller in arterioles than the manual estimates. This trend was present in both genotypes (Supplementary Figure S8) and could be explained by too dim puncta signal for model recognition. Overall, AMT quantified by manual and automated approaches was consistent (Figure 3B), and both methods showed that AMT was unchanged in 5xFAD mice.

An increase in paracellular leakage represents another mechanism of BBB dysfunction which can be studied by 2 PM (Mathiesen Janiurek et al., 2019). Accordingly, to study paracellular leakage, we injected into the mouse bloodstream sodium fluorescein (NaFluo), a small (376 Da) tracer molecule, which crosses the BBB between brain endothelial cells, i.e., via the paracellular route (Figure 4). Immediately after injection, we recorded fluorescence of blood-circulating and parenchymal NaFluo in time series of Z-stacks containing both pial and parenchymal vessels (Figures 4A,B). We quantified the NaFluo in the blood and its accumulation in the brain tissue by averaging the intensity within ROIs placed in pial vessels and in parenchyma at the constant, defined depth (Figures 4B,C). Following the injection, NaFluo gradually appeared in the brain parenchyma in both 5xFAD and WT due to paracellular leakage (Figure 4C; Supplementary Video S6; Supplementary material, section 3). Because the intensity of NaFluo circulating in the blood differed between WT and 5xFAD (Supplementary Figure S1), we normalized the intensity in the parenchyma by signal intensity in the blood (Figure 4D). We quantified the paracellular leakage as the ratio of the AUC of NaFluo in the parenchyma and in the blood. We found that the paracellular leakage was the same in 5xFAD and WT mice, regardless of the cortical depth (Figure 4E). In addition, paracellular leakage did not change significantly with the age of mice used in this study (Supplementary Figure S6B). Likewise, we observed no overt changes in the density of capillaries between WT and 5xFAD, which could affect the number of sources for the NaFluo to enter the brain (Supplementary Figure S9). Thus, despite the Aβ load, the paracellular barrier of the BBB was preserved in 5xFAD mice.

Extravasation of blood-borne fibrinogen is another sign of compromised BBB and a contributing factor to AD progression, observed in some patients (Ryu and McLarnon, 2009), and reported in several animal models of AD (Cortes-Canteli et al., 2012). We measured the extravasation of fibrinogen by IHC in 7 m.o. WT and 5xFAD mice (Figure 5). Given the contribution of hippocampal pathology to AD, we included the hippocampus in our analyses. Neither WT nor 5xFAD exhibited signs of fibrinogen extravasation in corresponding brain regions (Figures 5A,B). First, we measured the average signal from the brain tissue in the cortex and in the hippocampus, separating the latter into stratum lacunosum-moleculare (s.l-m) and remaining areas due to non-uniform distribution of the baseline signal, i.e., higher in the s.l-m (Figures 5A,C). Regardless of the brain region, we found no fibrinogen leakage in 5xFAD mice, as evidenced by consistent levels of the measured signal between WT and 5xFAD mice (Figure 5C). Although whole-brain analysis can provide the measure of the degree of fibrinogen leak across the BBB, it may not be sensitive for local, i.e., perivascular increase in fibrinogen content. Therefore, we next analyzed the fibrinogen intrusion at the microscale, i.e., in the areas proximal to the brain microvessels. We extracted the fibrinogen signal intensity profiles along the >20 μm diameter brain vessels, obtaining profile plots across the boundary regions between the vessel lumen and the neighboring brain tissue (Figure 5D). In accord with the results from wide-brain areas, we observed no trend toward a local increase in fibrinogen content, with profile plots across the BBB interface and neighboring parenchyma being similar between WT and 5xFAD mice for both cortical (3–7 vessels per animal; n = 3 WT, n = 3 5xFAD) and hippocampal microvessels (2–5 vessels per animal; n = 3 WT, n = 3 5xFAD; Figure 5E). Thus, the BBB in 5xFAD mice exhibited no increase in permeability toward blood-borne fibrinogen.

NVC ensures appropriate blood supply to satisfy an increase in metabolic demand in neurons following their activation (Hall et al., 2014; Cai et al., 2018; Zambach et al., 2021). Here, we assessed whether Aβ pathology affects NVC (Figure 6A). We stimulated the whisker pad and simultaneously monitored both evoked cortical LFP and vasodilation of individual microvessels in the somatosensory (i.e., barrel) cortex. To examine the synaptic input to evoke NVC responses, we compared the amplitudes of LFPs in WT and 5xFAD mice. The amplitude of the LFPs was the same in WT and 5xFAD mice (Median (25-th, 75-th percentiles): 1.03 (0.80, 1.24) vs. 1.02 (0.89, 1.22), p = 0.82, n (WT) = 8, n (5xFAD) = 10, two-tailed Mann–Whitney U test), confirming no change in 5xFAD mice (Figure 6B). This indicated equal synaptic input to evoke NVC in WT and 5xFAD mice. In parallel, we monitored NVC vasodilation by recording a time-series of Z-stack images (i.e., hyperstack) capturing penetrating arteriole, precapillary sphincter, 1st-order capillary, and 2nd-order capillaries (Figure 6C; Supplementary Video S7). We quantified the baseline diameter and the relative dilation during stimulation by placing rectangle ROI on vessel segments (Cai et al., 2018; Figure 6D). We found that the baseline diameter of the penetrating arterioles was significantly larger in WT than in 5xFAD mice (14.8 ± 0.41 µm vs. 12.8 ± 0.56 µm, p = 0.018, n (WT) = 9, n (5xFAD) = 10, two-tailed Student-t test; Figure 6E). However, the relative dilation was the same in 5xFAD and WT mice across all corresponding vascular segments (Figure 6F). Similar to BBB experiments, we assessed whether the measured variables depended on the age of animals, and we observed no significant changes in NVC and the baseline diameter with the mouse age (Supplementary Figures S6C, S10). Thus, our results show that NVC was preserved in 5xFAD mice.

Pericytes are essential components of the neurovascular unit, regulating the NVC responses, structural integrity and transport across the BBB (Armulik et al., 2010; Hall et al., 2014). Loss of pericytes and their coverage is an important predictor of vascular cognitive impairment and dementia (Uemura et al., 2020), and accelerates AD-like neurodegeneration in mice (Sagare et al., 2013). Therefore, we next quantified the density of capillary pericytes in 5xFAD mice using immunohistochemistry. The tissue was immunostained for a cluster of differentiation 31 (CD31; = platelet endothelial cell adhesion molecule; PECAM-1) and platelet-derived growth factor receptor beta (PDGFR-β), to visualize both microvasculature and accompanying pericytes, respectively (Supplementary Figure S11A). The capillary pericytes exhibited their typical morphology of a “bump on a log” with extensive projections along the microvasculature (Supplementary Figures S11A,B). Noteworthy, IHC in 5xFAD mice also revealed the clear presence of Aβ plaques, often in proximity to capillaries (Supplementary Figure S11C) and of similar morphology to aggregates observed using 2 PM in vivo (Figure 1; Supplementary Videos S1, S2). Despite the presence of Aβ plaques, we found no evident reduction of the capillary pericyte density in 5xFAD mice (n = 3 mice/group; n Z-stack images = 10/animal; Supplementary Figure S11D). Thus, the lack of NVC and BBB dysfunction in 5xFAD animals was accompanied by preserved number of pericytes on capillaries.

Overall, the results of our study show that the major functions of the neurovascular unit were preserved in adult female 5xFAD mice despite the widespread presence of Aβ plaques in the brain.