Neuropathological correlates of MRI-observed hypointense lesions in the APP23 mouse model of cerebral amyloid angiopathy

Abstract

Cerebral amyloid angiopathy (CAA) is a small vessel disease associated with hemorrhagic lesions including microbleeds and intracerebral hemorrhage. Growing evidence suggests vascular remodeling may be a critical step leading to CAA-related vessel rupture. The APP23 mouse model is one of few models of cerebral amyloidosis that develop spontaneous microbleeds, visible on T2*-weighted MRI as hypointense lesions. However, it remains unknown whether this model recapitulates the microbleed-associated vascular remodeling observed in human tissue. In this study, we used in-vivo 9.4 T MRI and histopathology to study the neuropathological correlates of MRI-observed hypointense lesions in 18- and 24-month-old APP23 transgenic (Tg) mice and wildtype (WT) littermates (12 Tg, 17 WT). Tg mice had more cortical and deep lesions (p < 0.001, p < 0.001)). Brains were serially sectioned from a subset of these mice (11 Tg, five WT) and lesions retrieved. All cortical lesions were microbleeds (and in Tg mice). However, 96% of deep lesions (Tg and WT) were calcifications, suggesting caution is needed when interpreting deep hypointense lesions as microbleeds. A qualitative analysis of histopathologically-confirmed microbleeds revealed vascular remodeling including vessel wall thickening and decreased vascular amyloid-β at most vessel rupture sites, resembling human findings, suggesting APP23 mice may be an appropriate model to study mechanisms of CAA-related hemorrhage.

Introduction

Cerebral amyloid angiopathy (CAA) is an age-related small vessel disease characterized by amyloid-β build-up in the walls of the cortical and leptomeningeal vessels of the brain. CAA is a leading cause of symptomatic intracerebral hemorrhage (ICH) and is associated with the occurrence of other hemorrhagic lesions including cerebral microbleeds and cortical superficial siderosis. These hemorrhagic lesions occur at late stages of disease progression, decades after initial amyloid-β deposition. ¹ Currently, there are no disease-modifying therapeutics available to halt or slow down the sequence of events leading to hemorrhage in CAA.

Recent studies using ex-vivo MRI-guided serial sectioning of individual microbleeds in brain tissue of patients with neuropathologically confirmed CAA have revealed decreased or absent amyloid-β deposition at the rupture site compared to surrounding vessels.^2,3 Additional signs of vascular remodeling have been observed at the sites of vessel rupture, in the form of hyaline thickening of the vessel wall, loss of smooth muscle cells, fragmentation, and fibrinoid necrosis. Vessels with evidence of vascular remodeling in CAA are known as Vonsattel grade 3 and 4 vessels, ⁴ and a recent study in human autopsy tissue found that the number of Vonsattel grade 3–4 vessels correlated with the number of microbleeds observed on ex vivo MRI. ⁵ These findings suggest an additional step of vascular remodeling may occur between initial amyloid-β deposition and hemorrhage. Moreover, remodeled vessels were associated with increased densities of reactive astrocytes and activated microglia in the parenchyma immediately surrounding the blood vessel, suggesting that local inflammation might play a role in the vascular remodeling process and subsequent hemorrhage in CAA. ⁵

Leveraging mouse models of CAA is crucial to longitudinally delineate the steps leading to vascular remodeling and hemorrhage as well as to develop targeted therapeutic strategies. The APP23 mouse model of amyloidosis develops both parenchymal amyloid-β plaques and vascular amyloid-β, with amyloid-β deposition starting at around 8 months of age.^6–8 APP23 mice develop spontaneous MRI-observed microbleeds at around 16 months of age. ⁹ As such, the APP23 mouse model may be a relevant model of CAA-related hemorrhage, suitable to study the sequence of events leading to vessel rupture. However, whether vascular remodeling plays a role in the pathophysiology of microbleed formation in this model remains poorly understood. Additionally, whether most hypointense lesions observed on MRI in APP23 mice correspond to microbleeds versus other age-related pathologies such as calcifications remains underexplored.

To address these knowledge gaps, we first aimed to determine the neuropathological correlates of MRI-observed hypointense lesions in the APP23 mouse model. Next, using MRI-guided serial sectioning, we assessed vessel wall integrity and amyloid-β burden at the rupture site of individual histopathologically verified microbleeds. Additionally, we screened the tissue for unruptured remodeled vessels (defined as Vonsattel grade 3–4) and assessed the presence of perivascular inflammation in the form of reactive astrocytes and activated microglia. Finally, using brain tissue clearance and light sheet microscopy, we assessed blood vessels in 3D near MRI-observed hypointense lesions, specifically assessing for smooth muscle actin (SMA) and amyloid-β coverage.

Methods

Animals

The APP23 mouse line is available at Jackson Laboratories (reference number #030504). For this study, we used mice that were bred and aged in-house (n = 22) or ordered individually and aged in-house (n = 7). APP23 mice overexpress human APP, containing the Swedish double mutation (APP751*K670N/M671L) associated with Alzheimer’s disease, under the Thy1 promotor. This mutation results in a neuronal overexpression of mutated human amyloid-β and gradual accumulation of parenchymal amyloid-β plaques in the neocortex, hippocampus, amygdala, and thalamus. Also, they develop significant vascular amyloid-β deposits in the pial, thalamic, cortical, and hippocampal vessels. Female mice appear to have a higher amyloid-β burden than male mice. ¹⁰ Both male (M) and female (F) mice were included in this study.

All animal procedures were performed with the approval of the Massachusetts General Hospital Animal Care and Use Committee and experiments were conducted in compliance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals. The ARRIVE guidelines were followed for reporting animal experiments. ¹¹ The ARRIVE Essential 10 checklist can be found in Supplemental Table 1. The mice were housed together with same sex littermates (two to four per cage depending on the litter size) in individually ventilated cages. The cages were provided with bedding, unlimited chow food and water, and a plastic igloo.

A total of n = 29 18- and 24-month-old transgenic (Tg) APP23 mice and age-matched wild type (WT) littermates underwent in vivo MRI scans once as follows:

• Eighteen-month-old: 14 mice (six Tg (three F), eight WT (four F))

• Twenty-four-month-old: 15 mice (six Tg (four F), nine WT (three F))

For the MRI component of the study, the goal sample size was six APP23 Tg mice per group based on prior microbleed data from APP23 mice. ¹ Available age-matched WT littermates were scanned as well. The scan order was determined based on when mice reached the age of 18 or 24 months old. Note, these mice were a sub-set of those imaged as a part of a different MRI study examining vascular function in APP23 mice. ⁶ No inclusion/exclusion criteria were set a priori, and no animals were excluded in the study.

A sub-set of these mice (n = 16) underwent histopathological evaluation as follows:

• Eighteen-month-old: seven mice (six Tg (three F), one WT (zero F))

• Twenty-four-month-old: nine mice (five Tg (three F), four WT (one F))

In vivo MRI protocol

Mice were scanned in-house on a pre-clinical 9.4 T MRI scanner with Bruker Paravision 6 software (Bruker, Ettlingen, Germany) with a phased array surface coil and a whole-body volume transmit coil. Anesthesia consisted of low isoflurane (2% induction, 1.1% maintenance) dissolved in oxygen-enriched air (30% oxygen). Temperature was maintained at 36.5 °C using a feedback-controlled hot air pump. Respiration rates were monitored using a pressure-sensitive sensor placed below the mouse, and the head was stabilized with a bite bar and ear bars. The MRI protocol included, among other sequences acquired for a different study, ⁶ a multi-gradient echo (MGE) T2*-weighted sequence to detect hypointense lesions. The MGE sequence was acquired in coronal orientation, with the following parameters: repetition time (TR) = 31 ms, six echoes (3.5–21 ms, with 3.5 ms intervals), flip angle = 14°, 100 × 100 matrix, 0.175 × 0.175 × 0.35 mm³ resolution, 25 consecutive slices, 16 repetitions, and a total scan time of 15.5 min.

MR image processing and lesion scoring

The MGE scans were loaded into MATLAB using MP3, which is a freely available MATLAB-based image analysis software. ¹² The 16 repetitions were averaged together before rating. The assessment of hypointense lesions on MGE MRI was performed by two raters (MLB and MGK), blinded to mouse genotype, age, and sex. Probable lesions were defined as homogenous ovoid or round foci of low signal intensity. A final score was established during a consensus meeting.

Brain processing

Approximately 1–2 h after undergoing MRI, the mice were euthanized through CO₂ asphyxiation after which they were transcardially perfused with 20 ml phosphate buffered saline (PBS). Then, the brain was extracted from the skull. The brains were fixed in 4% paraformaldehyde with 15% glycerol in PBS for several months, processed, and embedded in paraffin in coronal orientation. Coronal serial sections of 6 μm thickness were cut with a microtome.

Histology

The Perls’ Prussian Blue method was used to visualize ferric iron in hemosiderin. We screened the tissue for microbleeds by staining every fifth section with the Prussian Blue method. Briefly, sections were deparaffinized and rehydrated through xylene and a series of ethanol (100%, 95%, 70%) and water. Then, the sections were incubated in a 1:1 ratio of 5% hydrochloric acid and 5% potassium ferrocyanide (for 30 min) at room temperature. Filtered neutral red was used for background staining. Sections adjacent to sections with microbleeds were stained with standard hematoxylin and eosin (H&E) to confirm the presence of hemosiderin-containing macrophages in Prussian Blue-positive sections. In addition, every 10th section was stained for H&E to screen for calcifications. If a suspected calcification was identified on H&E, adjacent sections were stained with the Von Kossa stain to confirm the presence of calcium in the tissue. These sections were incubated in 5% silver nitrate under a lamp (for 60–90 min) and then incubated in 1% pyrogallol (for 5 min). After a final incubation in 5% sodium thiosulfate (for 3–5 min), the background was stained with filtered neutral red. Between every incubation the sections were washed in demi water.

Immunohistochemistry

Adjacent sections of verified microbleeds underwent immunohistochemistry against amyloid-β, ionized calcium-binding adapter molecule 1 (Iba-1), glial fibrillary acidic protein (GFAP), fibrin(ogen), and smooth muscle actin (SMA). The sections were deparaffinized and rehydrated through xylene and graded series of ethanol (100%, 95%, 70%) and water. After that, the sections were treated with 3% hydrogen peroxide to quench endogenous peroxidase activity. Antigen retrieval was performed with heat induced epitope retrieval in citrate buffer (pH 6.0). For amyloid-β, we used either 100% formic acid (5 min) or heated citrate buffer as antigen retrieval (20 min). The sections were blocked using either normal goat serum (GFAP, Iba-1, fibrin(ogen)) or normal horse serum (SMA, amyloid-β) diluted in tris-buffered saline (TBS) from a Vectastain ABC kit (Vector Laboratories, cat.nr. # PK-4001 or PK-4002; for 1 h). Next, the sections were incubated overnight in a 4 °C cold room with primary antibodies against GFAP (rabbit polyclonal; Sigma Pharmaceuticals, cat.nr. #G9269, 1:1000), fibrin(ogen) (rabbit polyclonal; Dako, cat.nr. #A0080, 1:500), Iba-1 (rabbit polyclonal; Wako, cat.nr. #019-19741, 1:500), smooth muscle actin (SMA; mouse monoclonal; Dako, cat.nr. #M0851, 1:250), and amyloid-β (mouse monoclonal; Dako, cat.nr. #M0872, 1:200) diluted in TBS. The next day, a biotinylated secondary antibody was applied from a Vectastain ABC kit (Vector Laboratories). After that, we applied an avidin–biotin complex and visualized the signal using the chromogen 3,3′-diaminobenzidine (DAB; Vector Laboratories, cat.nr. #SK-4100). The sections were counterstained with hematoxylin. The sections were washed using TBS before and after every secondary antibody incubation and after avidin–biotin application. Sections were then dehydrated through a series of ethanol (70%, 95%, 100%), cleared through xylene, and cover slipped using Permount mounting medium (Fisher Chemical, cat.nr. #SP15-100).

After imaging (discussed below), the Prussian Blue-positive sections were re-stained for SMA. To remove the neutral red background stain, the sections were soaked in water until the signal was gone. Additionally, a few sections in which the rupture site was identified on H&E, underwent re-staining for amyloid-β. To strip the H&E stain, the sections were soaked in 5% acid alcohol until the signal was gone. After that, they underwent immunohistochemistry for amyloid-β following the same protocol as outlined above.

Image analysis

Sections were imaged using the Hamamatsu NanoZoomer Digital Pathology (NDP)-HT scanner (C9600-12; Hamamatsu Photonics K.K., Japan) at 20× magnification. The viewing platform NDP.View (version 2.6.13) was used to visualize the digitized sections. To identify microbleeds, the Prussian Blue sections were screened for clusters of iron-positive depositions. To confirm the presence of hemosiderin-containing macrophages, adjacent H&E-stained sections were then evaluated for clusters of brown hemosiderin deposits. The H&E sections as well as adjacent amyloid-β sections were also used to assess vessel pathology at the rupture site of microbleeds. In addition, H&E sections were screened for calcifications. As discussed in the methods, if a suspected calcification was identified on H&E, adjacent sections were stained with the Von Kossa stain for confirmation. Assessments for microbleeds and calcifications on histopathology were performed with a goal of identifying histopathological correlates of hypointense lesions visualized on MRI and were therefore not performed blinded.

All H&E-sections with confirmed microbleeds were also screened for the presence of unruptured remodeled vessels without evidence of red blood cell extravasation or hemosiderin deposits by a single rater (MLB). In addition, five random sections of every mouse brain were also screened for remodeled vessels. To determine inter-rater reliability, a second rater independently assessed 25% of the brain sections for the presence of remodeled vessels (SJV). Vascular remodeling was defined on H&E as hyaline thickening of the vessel wall, with loss of smooth muscle cells, and evidence of vessel wall fragmentation or fibrinoid necrosis (i.e. Vonsattel grade 3–4).

Tissue clearing

To study the relationship between microbleeds and remodeled vessels in 3D, two additional male 24-month-old Tg APP23 mice (both bred in house) underwent scanning on the same pre-clinical 9.4 T MRI scanner with the same MGE sequence. Afterwards, the mice were euthanized, and their brains were processed for tissue clearing.

We used tissue libraries to find the optimal conditions for delipidation and multiplexed immunohistochemistry with SMA, Glut1, and amyloid-β, following previously published methods. ¹³ The optimal protocol we found was as follows. Mice were transcardially perfused with ice-cold PBS, followed by 4% paraformaldehyde (Electron Microscopy Sciences) in PBS. The brains were then removed from the skull and placed in 4% paraformaldehyde in PBS for 48 h at 4 °C, then rinsed in PBS for 24 h at 4 °C. Each hemisphere was incubated with a hydrogel crosslinking solution containing 4% (wt/vol) acrylamide, 0.25% (wt/vol) VA-044 thermal polymerization initiator in PBS for 3 days at 4 °C. Tissue was then placed in fresh crosslinking solution and heated at 37 °C for 3 h under vacuum with an X-CLARITY polymerization system (Logos Biosystems, South Korea). The tissue was then rinsed with 50 ml PBS five times over 3 h. Tissue was then placed in a delipidation solution (SDS 200 mM, sodium sulfite 20 mM, sodium borate 20 mM, sodium hydroxide to pH 8.5–9) at 37 °C for 3 weeks. The tissue was then rinsed with 50 ml PBS-T at least 5× over 24 h.

The tissue was then incubated with 1:300 dilutions of antibodies against amyloid-β conjugated to Alexa-Fluor 488 (Cell-Signaling Technology, cat.nr. #51374), SMA conjugated to Alexa-Fluor 647 (Millipore-Sigma, cat.nr. #C6198), and Glut-1 conjugated to Alexa-Fluor 555 (EMD Millipore, cat.nr. #07-1401) in PBS-T for 3 weeks at 4 °C with gentle shaking. The tissue was then rinsed with PBS for 1 week at 4 °C with gentle shaking.

Tissue was then transferred to Easy-Index (refractive index of 1.52; Lifecanvas Technologies) and equilibrated prior to imaging with a Zeiss Lightsheet.Z7 Microscope (Carl Zeiss Microscopy). Images were stitched using Zen software (Carl Zeiss Microscopy), then 3D reconstructions were visualized using Arivis (version 4.2) and Imaris (version 9.9) and co-registered to MRI images. An example of the reconstruction is provided in Supplemental Figure 1.

Statistics

We performed inter-rater reliability assessments for the detection of hypointense lesions on MGE MRI and unruptured remodeled vessels on H&E. An intraclass correlation coefficient (ICC) was calculated for the agreement on the number of hypointense lesions on MRI. The Cohen’s κ score was calculated for the agreement on the presence of remodeled vessels on H&E. To compare the number of hypointense lesions found on MRI between the WT and Tg mice, a negative binomial regression model was fitted in R (function glm.nb, package MASS) with genotype, age and sex as predictors. Models with interaction terms for age, genotype and sex were also fitted, but because these models did not converge, the model without interaction terms was used. Statistics were performed in GraphPad Prism (version 9.0.2.) and RStudio (Build 496). Given the low numbers of lesions identified on histopathology, statistical analyses were not performed on histopathological data.

Results

Aged APP23 Tg mice have higher numbers of hypointense lesions on MGE MRI compared to WT littermates

A total number of 29 mice underwent in vivo 9.4 T MRI. In total, 46 cortical and 57 deep hypointense lesions were found on MGE MRI (Supplemental Table 2). All lesions were homogenous ovoid or round foci of low signal intensity, and no lesions suggestive of macrohemorrhage or cortical superficial siderosis were observed. The inter-rater reliability was considered moderate for cortical lesions (ICC = 0.76) and good (ICC = 0.86) for deep lesions. Figure 1 summarizes the number of cortical and deep hypointense lesions observed on MRI in all mice in the study. To assess the independent effects of age, genotype, and sex on the number of cortical and deep hypointense lesions observed, negative binomial regression was used. In the cortex, genotype had a significant effect on the lesion count (β = 2.57 (95% CI: 1.45–3.86), z = 4.25, p < 0.001), but age and sex did not have an effect (β = −0.07 (95% CI: −0.25 to 0.11), p = 0.46 and β = 0.52 (95% CI: −0.53 to 1.62), p = 0.33, respectively). In the deep brain regions, genotype also had a significant effect on the lesion count (β = 1.24 (95% CI: 0.50–2.02), z = 3.37, p < 0.001). Age had a significant effect as well (β = 0.22 (95% CI: 0.10–0.36), z = 3.43, p < 0.001), but sex did not (β = 0.53 (95% CI: −0.20 to 1.30), z = 1.47, p = 0.14).

Figure 1.

Aged Tg APP23 mice have higher numbers of hypointense lesions on MGE 9.4 T MRI compared to WT littermates: (a) a representative example of an MGE image in a 24-month-old Tg APP23 male mouse, showing three cortical (white squares) and one deep (orange square) hypointense lesions, (b) the number of cortical hypointense lesions on MGE MRI was higher in the Tg mice compared to the WT littermates (β = 2.57 (95% CI: 1.45–3.86), z = 4.25, p < 0.001), and (c) the number of deep hypointense lesions on MGE MRI was also higher in the Tg mice compared to the WT littermates (β = 1.24 (95% CI: 0.50–2.02), z = 3.37, p < 0.001). These effects are corrected for age and sex using a negative binomial regression model (n = 29 mice). Circles and triangles indicate female and male mice, respectively.

MGE: multi-gradient echo; WT: wildtype; Tg: transgenic.

Deep hypointense lesions correspond to calcifications and cortical hypointense lesions to microbleeds

The brains of 16 mice were processed for histopathology, generally selected as the brains with higher numbers of MRI hypointense lesions. In these mice, a total of 40 cortical and 41 deep hypointense lesions were observed on MRI, all of which we attempted to find with histopathology (Supplemental Table 3). After screening all serial sections, a total of 14 microbleeds were identified on the Prussian Blue-stained sections and verified on H&E (Supplemental Figure 2 and Supplemental Table 3). One microbleed was in the thalamus and 13 microbleeds were in the cortex. In addition, a total number of 27 calcifications were identified on the H&E-stained sections and verified on the Von Kossa stain. All calcifications observed were in deep areas of the brain (Figure 2). While some MRI lesions were not identified on histopathology, no lesions other than microbleeds or calcifications were observed in these regions on histopathology. Microbleeds were only observed in Tg mice on histopathology, while calcifications were found in both Tg and WT mice (Supplemental Table 3). Calcifications identified on histopathology could be reliably matched to MRI hypointense lesions, whereas cortical lesions were more difficult to correlate between modalities (Supplemental Table 4), likely due to the relatively small size of microbleeds on histopathology and the distance between Prussian Blue-stained sections (30 μm). Note, no cortical lesions (microbleeds, calcifications, or other lesions) were identified on histopathology in WT mice.

Figure 2.

Histopathological correlates of cortical and deep hypointense lesions: (a) MGE image of a cortical hypointense lesion (white arrow) in a 24-month-old Tg APP23 female mouse that corresponded to an iron-positive microbleed with associated hemosiderin deposits (black arrows on H&E) on histopathology (a’: adjacent PB and H&E stains shown), (b) MGE image of a deep hypointense lesion (white arrow) in a 24-month-old Tg APP23 male mouse that corresponded to an iron-positive microbleed with associated hemosiderin deposits (black arrows on H&E) on histopathology (b’: adjacent PB and H&E stains shown), and (c) MGE image of two deep hypointense lesions (white arrows) in a 24-month-old Tg APP23 female mouse (same as in (a)) that corresponded to calcifications on histopathology (c’: adjacent H&E and VK stains shown).

MGE: multi-gradient echo; PB: Perls’ Prussian Blue; H&E: hematoxylin & eosin; VK: Von Kossa.

Microbleeds in Tg mice show signs of vascular remodeling and lower vascular amyloid-β burden at the rupture site

The vessels involved in 9/14 microbleeds were identified on H&E, while the culprit vessel or a clear rupture site could not be identified on H&E in the other five (Supplemental Figure 2). Seven out of nine culprit vessels identified on H&E showed signs of vascular remodeling (defined as vessel wall thickening and loss of typical cellular architecture in the vessel wall; Figure 3 and Supplemental Figure 2). Culprit vessels were identified for 7/14 microbleeds on amyloid-β stains; no clear culprit vessel or rupture site could be identified on the other seven. In the seven microbleed culprit vessels identified on amyloid-β stains, four showed absent or minimal vascular amyloid-β burden, whereas three had circumferential vascular amyloid-β at the rupture site (Figure 3 and Supplemental Figure 2). Evidence of reactive astrocytes and activated microglia in the form of GFAP-positive and Iba-1-positive cells as well as blood–brain barrier leakage in the form of fibrin(ogen)-positive staining was observed surrounding some but not all histopathologically-confirmed microbleeds (Supplemental Figure 3).

Figure 3.

Microbleeds in Tg APP23 mice show signs of vascular remodeling and lower vascular amyloid-β burden at the rupture site: (a, b) examples of two separate microbleeds that demonstrated signs of vascular remodeling (arrows in (a, b)), and absence of amyloid-β at the presumed rupture site (arrows in (a′, b′)) and (c, d) examples of two separate microbleeds (arrows in (c, d) point at hemosiderin deposits) for which the culprit vessel or the rupture site could not be identified (c′, d′).

H&E: hematoxylin & eosin; Aβ: amyloid-β.

Remodeled vessels are present yet infrequently observed in Tg mice

A total number of 80 H&E-stained sections were screened for unruptured remodeled vessels (five sections from each of the 16 mouse brains, 11/16 Tg mice). We found six remodeled vessels across five different sections in two Tg APP23 mice (one 18-month-old M (four vessels), one 24-month-old F (two vessels)). Note, these mice had relatively high numbers of cortical hypointense lesions on MRI: 13 and 6, respectively. The inter-rater reliability was perfect (κ = 1.00). All remodeled vessels were located at the pial surface (i.e. leptomeningeal blood vessels) and were Vonsattel grade 4 (Figure 4). No Vonsattel grade 3 vessels were observed. Four out of six unruptured remodeled vessels were positive for fibrin(ogen) and Iba-1, indicative of blood–brain barrier leakage and perivascular inflammation (Figure 4).

Figure 4.

Remodeled vessels are observed in Tg APP23 mice: (a) an example of a leptomeningeal blood vessel at the level of the pial surface with evidence of vascular remodeling, (b) lower burden of amyloid-β compared to neighboring vessels, (c) deposition of fibrin(ogen) in the wall, suggestive of blood–brain barrier leakage, and (d) perivascular inflammation in the form of Iba-1-positive cells (arrows).

H&E: hematoxylin & eosin; Aβ: amyloid-β; Iba-1: ionized calcium-binding adapter molecule 1.

Exploring blood vessels associated with microbleeds in 3D in Tg mice

Using brain tissue clearing and Lightsheet microscopy, we explored the structural integrity of blood vessels in 3D in regions of MRI-observed hypointense lesions in two Tg APP23 mice. In one mouse, we were able to identify the microbleed region ex vivo, which revealed local blood vessels with evidence of decreased SMA in segments with amyloid-β deposition (Figure 5) adjacent to regions with intact SMA expression, serving as a proof-of-concept that these transitions occurring in individual blood vessels can be traced in 3D. Additionally, we observed occasional apparent discontinuities in the blood vessels in which no Glut1 staining assessing for vascular endothelium was observed (Figure 5 and Supplemental Movie 1), suggesting regions of blood vessel rupture and/or local endothelial dysfunction associated with reduced Glut1 expression. These regions appeared to be adjacent to segments of the blood vessel with amyloid-β present but did not have local amyloid-β in the “missing vessel segment,” suggesting potential decreased amyloid-β at the site of vessel rupture, in line with observations in 2D in the serial sections.

Figure 5.

3D visualization of cortical blood vessel in a microbleed region: (a) MGE image of microbleed in a 24-month-old Tg APP23 male mouse, (b) MGE image co-registered to 3D Lightsheet microscopy image in region of microbleed. Box surrounding vessel shown in (c/d). Vessel with discontinuity (arrow) in which no amyloid-β, SMA, or Glut1 staining was observed (c, d). Rotation of blood vessel in 3D is shown in Supplemental Movie 1.

MGE: multi-gradient echo; SMA: smooth muscle actin; Glut1: glucose transporter 1.

Discussion

In this MRI and histopathological study, we identified microbleeds in histopathological brain tissue from APP23 Tg mice and traced individual ruptured blood vessels. We observed signs of vascular remodeling, including a thickened vessel wall and an obliterated lumen, at the vessel rupture site of 7/9 microbleeds in which a culprit vessel could be identified on H&E. This finding is in line with prior findings in neuropathological studies of brain tissue from patients with CAA. ³ In addition, we observed reduced amyloid-β burden at the rupture site of some, but not all, microbleeds identified on amyloid-β stains (4/7). Examining unruptured remodeled vessels, we also observed a reduced amyloid-β burden as well as evidence of blood–brain barrier leakage and perivascular inflammation, in line with prior findings in human tissue. ⁵ The number of these remodeled vessels was low and notably grade 4 remodeled vessels are rare in human histopathological studies as well. ⁵

Vascular remodeling may play a central role in the pathophysiology of CAA-related hemorrhage. Importantly, this process seems to be a late-stage manifestation of CAA, with microbleeds in the APP23 model occurring many months after initial CAA deposition, ⁶ and microbleeds occurring in patients with CAA decades after initial CAA deposition. ¹ Patients with CAA frequently present at these late-stages—after hemorrhage events have begun to occur—making vascular remodeling a potentially clinically relevant target for hemorrhage prevention. Given our study demonstrates consistent vascular remodeling associated with bleeding events in APP23 Tg mice, we propose that the APP23 mouse model may be a relevant tool to further study vascular remodeling’s contribution to hemorrhage in CAA in vivo.

Notably, the presence of CAA has been implicated as an important risk factor for the development of amyloid-related imaging abnormalities (ARIA) secondary to anti-amyloid immunotherapy, with multiple proposed inflammatory-mediated mechanisms, including removal of amyloid-β from the vessel wall leading to blood–brain barrier leakage and rupture.^14,15 Our findings of reduced amyloid-β at vessel rupture sites in APP23 mice, paired with prior findings from ex-vivo human studies,^3,5 suggest amyloid-β removal could occur as a part of a vascular remodeling process in CAA, potentially paralleling processes that occur at a larger scale in cases of ARIA and suggesting that the APP23 model may be useful for future studies of ARIA mechanisms.

Of note, a prior MRI-guided histopathology study in brain tissue of patients with CAA found reduced amyloid-β at 96% of ruptured vessel sites, ³ and this percentage was lower in the current mouse study (57%). This finding may relate to several factors. First, identifying the precise vessel rupture site is more challenging within mouse tissue given the smaller size of the vasculature. Ruptured vessel segments typically have dense amyloid-β deposition up- and downstream of the rupture site and some of our findings may reflect slightly off-target hemorrhage sites. Additionally, the APP23 mouse model is a model of amyloid-β overproduction, while sporadic CAA is considered a disorder of amyloid-β clearance rather than increased build-up, a fundamental difference which may be reflected in ongoing amyloid-β deposition at vascular remodeling sites.

Albeit exploratory, through visualizing blood vessels near a microbleed site in 3D, we also observed individual blood vessels with discontinuities in Glut1 staining suggestive of vessel rupture and/or endothelial dysfunction associated with decreased Glut1 expression. Of note, decreases in Glut1 expression have been previously observed in brain tissue and brain-derived circulating endothelial cells of patients with Alzheimer’s disease.^13,16 These vessel “breaks” did not have amyloid-β but appeared to be flanked on either side by vessel segments with amyloid-β and no SMA staining. Further studies in larger datasets are needed to investigate whether these regions represent microbleed-related vascular remodeling, potentially including local amyloid-β removal from the vessel wall.

These exploratory findings represent a proof-of-concept that this technique can be used to track individual blood vessels affected by CAA in 3D. Of note, a recent study in human post-mortem brain tissue that underwent tissue clearing observed that vessels associated with microbleeds were focally dilated near the site of rupture, with evidence of depletion of vascular smooth muscle cells, in line with our findings. ¹⁷ Interestingly, the authors furthermore reported that in a total number of 78 microbleeds examined, amyloid-β was present in all, but that in vessels with severe focal dilations, the deposits were disrupted into “shards” of amyloid-β. They note that the pathological dilation of remodeled blood vessels could give the impression that amyloid-β is proportionally decreased in 2D histopathological sections. This study demonstrates the added value of examining the structural properties of blood vessels associated with hemorrhage in CAA in 3D.

The observation of MRI-visible hypointense lesions in cortical and deep areas of the brains of APP23 Tg mice is in line with previous studies.^8,9 We found however, that most hypointense lesions on MGE MRI located in the deep regions of the brain are in fact calcifications, and that these were present in WT and Tg mice. Calcifications do not correspond to prior bleeding events, but rather chronic build-up of calcium. These calcifications were more consistently detected on both MRI and histopathology, in contrast to cortical lesions, likely in part due to their larger size. Calcifications have been observed in previous studies of aged WT mice,^18,19 and are a common age-related finding in humans. We note that while hypointense lesions were observed in WT mice on MRI, our histopathological analysis did not identify any microbleeds in WT mice, suggesting that these hypointense lesions were indeed calcifications (or possible rater error). As shown in prior studies, we observed a higher number of calcifications in older mice. ¹⁸ Future studies using MRI as a screening tool for microbleeds in mice should take this into account, for example, by generating phase data to help distinguish between the diamagnetic and paramagnetic properties of calcifications and microbleeds (iron), respectively. ²⁰

This study has several limitations. First, because of the exploratory nature of the study, no formal power calculation was performed and only a small number of animals were included. The number of microbleeds and remodeled vessels available for histopathological analysis was therefore also low, particularly as is difficult to identify the culprit vessel and a clear rupture site in mouse tissue. Additionally, the microbleeds characterized on histopathology were from a subgroup of mice. Given these sample size limitations, we were unable to perform statistical analyses on the histopathology data, which are therefore descriptive only. Future studies using larger sample sizes and leveraging tissue clearing and 3D analysis are needed to confirm our observations of vascular remodeling and decreased local amyloid-β burden associated with microbleeds in the APP23 mouse model. Additionally, future studies of larger cohorts should address the potential effects of age and sex. Second, due to the cross-sectional nature of this histopathological study, we were restricted to looking at a single point of time. Consequently, we were unable to draw any conclusions regarding the sequence of events leading to microbleed formation. Future studies, potentially using in vivo imaging techniques, are needed to fully delineate this process in APP23 mice. Third, we found that cortical lesions observed on histopathology were not reliably observed on MRI, likely due to the low resolution and proximity of these lesions to the edge of the cortex (a region in which imaging artifacts are more commonly observed). Fourth, some MRI-detected lesions were not confirmed on histopathology, likely due to the small size of lesions (particularly microbleeds) in mice. While we attempted to account for the relatively small size of microbleeds as compared to calcifications by staining more sections for iron, our sampling strategy still left gaps of 30 µm between sections assessing for microbleeds (Prussian blue stain) and 60 µm between sections assessing for calcifications (Von Kassa stain). Fifth, as the aim was to identify histopathological correlates of hypointense lesions on MRI, the histopathological scoring was not performed blindly. Although the scoring relied on objective staining criteria, this may nonetheless have introduced observer bias. Finally, we did not observe any lesions consistent with macrohemorrhage or cortical superficial siderosis (both common findings in patients with CAA) on either MRI or histopathology in the APP23 mice, a limitation in using this model to study types of CAA-related hemorrhage other than microbleeds.

In conclusion, microbleeds in the APP23 mouse model have a similar histopathologic appearance as microbleeds in human CAA, including signs of vascular remodeling. These findings suggest that the APP23 mouse may be an appropriate model for future mechanistic studies into the pathophysiology of microbleeds to uncover potential therapeutic targets for hemorrhage prevention in CAA.