Multi-parametric longitudinal imaging of cerebral biomarkers in a rodent model of Alzheimer’s disease

Abstract

This study introduces a method for longitudinally monitoring Alzheimer’s disease (AD)-related biomarkers in a rodent model utilizing a dual-modality imaging system combining photoacoustic microscopy (PAM) and confocal fluorescence microscopy (CFM). Using a cranial window transparent to both light and ultrasound, we examined cerebral vasculature, blood flow speed, oxygen saturation (sO₂), and amyloid-β (Aβ) deposition with single capillary resolution in genetically modified AD mice longitudinally over three months. Empowered by the high-resolution multimodal imaging, the analysis showed consistent changes of small vessel density decrease and Aβ deposition increase in AD mice compared to the control group. Meanwhile, a decrease in sO₂ was observed in AD group near the end of the observation period, highlighting the potential importance of functional imaging of hemodynamics that PAM facilitates. This multimodal system, with its longitudinal imaging capability, could provide valuable insight into the temporal dynamics and interrelationships of multiple biomarkers in neurodegenerative diseases.

1. Introduction

Alzheimer’s disease (AD), a predominant form of dementia, is marked by the presence of amyloid-β (Aβ) extracellular plaques and neurofibrillary τ tangles in the brain [1]. While the etiology remains uncertain, the pathological alterations associated with AD, including Aβ and τ deposition, brain atrophy and cerebrovascular dysfunction, have been documented [2], [3]. Although Aβ deposition is a major hallmark of AD, the disease is considered as a multifactorial disorder in which an early imbalance between Aβ production and clearance leads to its aggregation in brain tissue and cerebral vessels, further compromise the clearance of Aβ by exacerbate cerebrovascular dysfunction [4]. A growing number of studies have shown that vascular abnormalities—such as abnormal vessel morphology, reduced vessel density, capillary stalling, and diminished perfusion—contribute to cerebral hypoxia, metabolic stress, and cognitive impairment in AD mouse models [5], [6], [7], [8]. Correspondingly, reduced cerebral blood flow in cortical and hippocampal regions has been consistently reported, and this hypoperfusion has been implicated in promoting Aβ overproduction and accelerating memory deficits [9]. Importantly, these vascular and molecular alterations can emerge decades before clinical symptoms of cognitive decline become evident [10], [11]. While no effective treatment currently exists for AD, studies have shown that early intervention on AD could slow pathological progress and cognitive decline [12], [13]. These observations underscore the need for longitudinal monitoring of AD-related pathological changes and vascular hemodynamics to support both mechanistic studies and the evaluation of emerging therapeutic strategies.

Micron-scale imaging modalities are essential as most of the plaque sizes and affected vessel groups are on the order of ∼10 μm [3], [14]. Currently, many microscopic studies focusing on the progression of AD have been endpoint studies using genetically modified rodent models of AD [8], [15]. While ex vivo, endpoint studies allow unlimited imaging depth and higher resolution, they are subject to individual variability, require large numbers of animals across different age groups, and are usually time- and resource-intensive. More importantly, it cannot capture certain information on cerebral hemodynamics such as blood oxygen saturation (sO₂) or flow speed. Noninvasive modalities including functional magnetic resonance imaging (MRI) and ultrasound imaging provide access to deep-brain hemodynamics [16], [17], [18], but lack the resolution to interrogate capillary, which is thought to be the most affected group of vessels [19]. In contrast, optical imaging with much higher resolution, such as photoacoustic (PA) microscopy (PAM), single or multi-photon fluorescence microscopy, and optical coherence tomography (OCT) angiography (OCTA), offer much higher resolution for cortical vasculature imaging [20], [21], [22]. Molecular biomarkers such as Aβ and neurofibrillary τ, could be imaged in vivo with deep-penetrating techniques such as positron emission tomography (PET) [23], [24] and photoacoustic computed tomography (PACT) [25], [26], while high-resolution methods such as single or multi-photon fluorescence microscopy and PAM [27], [28] enable visualization of single plaque development.

Among these techniques, PAM has proven its unique capability in small animal brain imaging [29], [30]. PAM relies on the detection of the acoustic pressure wave generated by the rapid thermal expansion induced by a short pulsed-light excitation. In brain imaging, PAM provides unique value in vasculature and hemodynamic mapping, offering high resolution (1–10 µm) in a label-free manner [31]. It leverages the intrinsic high contrast between the absorption of hemoglobin and brain tissue in the visible to near-infrared wavelengths. Functional PAM can measure hemodynamic parameters such as sO₂ and flow speed with the resolution of single capillaries, which may enhance the research of AD. For PAM imaging of small animal brains, the cranial window must be transparent for both optical and acoustic waves, which require clear materials with biocompatibility and acoustic impedance matching. To date, several studies have developed chronic PAM compatible windows, lasting up to 22 weeks [32], [33], [34], yet none has been proven in a longitudinal study involving biomarker monitoring and quantitative analysis.

Despite significant progress in imaging AD-related biomarkers [5], [16], [25], [35], [36], longitudinal imaging remains challenging, especially for the high-resolution optical solutions. This is partly due to the requirement to create a transparent window on the skull, which poses challenges in maintaining the window clearness and avoiding complication for the animals. To date, only a few studies have described methods for longitudinal monitoring of genetically modified AD mice over several months [14], [21], and only a few aspects of biomarkers were covered in these studies, limiting investigation into how different biomarkers correlate with each other. None of the studies was able to track Aβ deposition, vascular structure and function simultaneously in a longitudinal protocol, partly because of the complexity in requirement for multiple imaging systems and contrast agents.

To address these needs, we previously proposed a platform combining multiple imaging modalities, including PAM, OCT, and fluorescence microscopy to monitor several AD related biomarkers simultaneously [37], [38]. In this study, we extend this platform to a longitudinal, multiparametric system and introduce two key advancements beyond our prior study: (1) the incorporation of blood flow and sO₂ at single-capillary level as functional biomarkers, and (2) the ability to track multi-parametric cerebrovascular and pathological changes over a few months period in vivo. This integrated system allows repeated, spatially co-registered imaging of microvascular structure, hemodynamic function, and Aβ deposition within the same cortical region of mice. By eliminating inter-system transfer and enabling label-free measurement of vascular structure and function, the platform reduces experimental complexity while ensuring intrinsic co-registration across biomarkers.

The objective of this study is to establish and validate this longitudinal imaging framework, and to quantify age-dependent trajectories of vascular density, flow speed, oxygen saturation, and Aβ deposition in genetically modified mice with AD versus age-matched wild-type (WT) controls. We specifically aim to evaluate whether AD mice exhibit microvascular degeneration and increasing Aβ accumulation, accompanied by impaired vascular function along the observation period. By demonstrating cranial window stability and multiparametric imaging repeatability and sensitivity to these anticipated pathological trends, this work provides a technically and biologically meaningful extension of our previous single-timepoint study and establishes a platform for future mechanistic and interventional investigations in neurodegenerative disease models.

2. Materials and methods

2.1. Animal preparation and cranial window

All the animal procedures were performed in accordance with the National Institutes of Health guidelines and reviewed and approved by the Institutional Animal Care and Use Committee (IACUC) of the University of Michigan (Protocol No.: PRO00011735). Female 5xFAD mice (B6SJL-Tg(APPSwFlLon,PSEN1*M146L*L286V)6799Vas/Mmjax) obtained from Mutant Mouse Resource & Research Centers (MMRRC) and female wild-type (WT, C57BL/6 J, Jackson Lab) were subjected to craniotomy at the age of 40 weeks. The age of observation starting point was determined based on our previous study, where we found a significant difference in Aβ deposit density between 9 month-old mice and 12 month-old mice [37].

The craniotomy and cranial window installation were modified from a method described by Goldey et al. [39], with the glass window replaced by a soft, thin-film window to facilitate acoustic transmission. Briefly, a 3D-printed imaging well (Figure S1A) was affixed to the skull, centered over the region of interest (ROI) (Figure S1B). A 5-mm-diameter circular bone flap over the left parietal cortex was removed following the craniotomy procedure. While keeping the cortex moistened, a large piece of single layer polyvinyl film (i.e., plastic food wrap, No. 208733, Costco) was gently laid onto the exposed brain. To minimize the gap between the film and brain—a critical step for a successful brain window—a temporary glass pressor was fabricated by bonding 5-mm and 8-mm round cover glass (0.15 mm thick; No. 64–0700, 64–0701, Harvard Apparatus). The pressor was positioned over the film to gently compress it against the cortex and skull (Figure S1C), and excess fluid was removed with surgical sponges. The Vetbond tissue adhesive (No. 1469SB, 3 M) was then applied between the film and the skull. After the adhesive had cured, the glass pressor was carefully removed and excess film was trimmed. Mice were then single-housed and monitored for one week before the first imaging session.

For both the craniotomy and the imaging sessions, anesthesia was induced under 5 % isoflurane in oxygen and maintained with 1–4 % isoflurane. During anesthesia, the mice were placed on a 37 ℃ warm pad, with their heads fixed on a fixation stage. All the animals with satisfactory window quality at the end of the study were included in the data analysis, with n = 5 for AD mice and n = 7 for WT mice, the excluded animals were described in the supplementary note 1.

2.2. Imaging procedure

The dual-modality imaging system combining PAM and CFM was described in our previous publication [37], [40] (See supplementary note 2). For each mouse, five imaging sessions were taken at 1, 4, 7, 10, 14 weeks after the surgery. One hour before the imaging session, the mouse was administered (retro-orbital injection) with CRANAD-3 (4 mg/kg, 1 mg/ml solution in 15 % Dimethylsulfoxide, 15 % Cremophor, and 70 % phosphate buffered saline (PBS)), an Aβ-targeting fluorescence dye (CRANAD-3, Sigma Aldrich) [41]. CRANAD-3 absorbs light from 520 nm to 670 nm (FWHM), with a peak at around 600 nm, and emits light from 600 nm to 750 nm (FWHM). Upon binding with Aβ, CRANAD-3 has an enhanced fluorescence emission which makes it a sensitive and specific dye for Aβ imaging. The head of the mouse was secured by clamping the handle of a ring-shaped frame (Supplementary note 1) to ensure that the optical focal plane was parallel with the region of interest (ROI) on the brain surface. The imaging well formed by the frame was filled with 4 mm of coupling gel (Systane lubricant eye gel) for both CFM and PAM imaging.

Four images were acquired for each mouse, namely, a CFM image for Aβ deposition, a PAM image for vasculature, a dual-wavelength PAM image for sO₂, and a PAM flow-speed image. For all four images, the same area of a 2 mm-by-2 mm region at the center of the cranial window was scanned. The scanning pattern was 512 × 512 (equivalent to 3.9 µm step size), except for PAM flow-speed image, which was reduced to 256 × 256 due to the limited buffer of the data acquisition system.

CFM imaging was performed one-hour post-injection, using excitation light with 558 nm wavelength, 1.5 ns duration, and 100 nJ energy pulsed laser at 50 kHz repetition rate, while 700 nm ± 20 nm band was selected for emission. The focal plane was localized by adjusting the height of the animal, the image was selected when most of the spot-like features were displayed sharply.

PAM imaging was performed next, using the excitation light at 532 nm and 558 nm wavelengths, both with 1.5 ns duration, 200 nJ energy, and 50 kHz repetition rate. The two wavelengths were interleaved by 2 μs. Signal was acquired with a customized needle-shaped piezoelectric transducer with a central frequency of 30 MHz and 50 % bandwidth [42]. The 532 nm signal was used for vasculature imaging while both 532 nm and 558 nm signals were used for sO₂ imaging.

In flow speed imaging, 1.5 ns, 532 nm, 200 nJ laser pulse of 80 kHz repetition rate was used. The scanning pattern was set to be 256 × 256 with a step of 7.8 µm, while each B-line was scanned repeatedly 50 times. Due to the redundant return time for the scanning mirror after each B-scan that was equivalent to 50 consecutive points, the interval between two A-lines of the same pixel was 306/80 kHz ≈ 3.8 ms. Overall, the scan time for flow speed imaging was about 50 s, and for all other imaging was about 5 s.

2.3. Vascular structure analysis

The vasculature analysis utilized the 532 nm PAM imaging that provided high contrast between brain tissue and hemoglobin, hence the photoacoustic signal amplitude reflected the presence of vessels perfused with blood. For each image, 512 × 512 A-lines were acquired, the pixel value was represented by the peak-to-peak amplitude. Due to the large quantities of raw data, automatic processing was necessary to extract meaningful results. The vasculature analysis utilized a modified version of an open-source toolbox, OCTAVA [43]. We modify the software to accommodate the analysis for cerebral vessels where vessel diameter has a wide range (<10 μm to > 100 μm) [37]. Utilizing the software (see supplementary note 3), we obtained demographic of vessel segments that included the length and diameter. We further categorized the vessel segments into different size groups and calculated the length density (i.e., total length divided by the imaging area) of each group.

2.4. Functional PAM data processing

In functional PAM analysis, we quantified the flow speed and sO2. The flow speed was quantified through photoacoustic correlation spectroscopy (PACS) [44], [45]. For each pixel of the 256 × 256 points, 50 consecutive measurements with 3.8 ms interval were taken. Autocorrelation function was calculated for each pixel using the equation: $G\left(\tau \right)=\frac{<\delta P\left(t\right)\bullet \delta P\left(t+\tau \right)>}{{<P(t)>}^2}$, where $\delta P\left(t\right)=P\left(t\right)-<P(t)>$ is the fluctuation of PA amplitude $P(t)$ and < > denotes ensemble average. $\tau$ represents the interval, which equals 3.8 ms here. The flow speed was estimated through

$\frac{G\left(\tau \right)}{G\left(0\right)}=\exp \mathrm{}\left(-{\left(\frac{\tau }{{\tau }_0}\right)}^2\right)$, where ${\tau }_0=w/v$, $w$ is the width of the laser focus and $v$ is the flow speed. To link the autocorrelation coefficient to absolute speed, we ran a calibration experiment where we created a phantom blood vessel with human blood flowing in a 300 μm diameter tube. We measured the autocorrelation coefficient with several designed speed, and used cubic fitting between speed and autocorrelation coefficient to map the data from the brain measurements. After we obtained the absolute speed map, we ran the vasculature analysis described in the prior section on the structure of the images, which allows us to extract the diameter, length, and skeleton coordinates of each vessel segment. Thus, we could extract the averaged flow speed of each vessel segment and add it as another metric to the demographic of vessels.

To establish a mapping between velocity and the correlation coefficient, we ran a calibration experiment on a phantom blood vessel with human blood flowing in a 300 μm diameter tube. We measured a series of velocities, controlled by a precision syringe pump (NE1000, New Era Pump Systems), with each measurement repeated 4–13 times. The relationship between the average correlation coefficient and velocity was fitted with a cubic polynomial, which was then used for continuous mapping between blood flow velocity and correlation coefficient. After we obtained the speed map and _SO₂ map, we could extract the averaged flow speed and _SO₂ of each vessel segment and add it to the demographic of the vessels.

Blood oxygen saturation measurements were facilitated by dual-wavelength PA signals at 532 nm and 558 nm. Given the extinction coefficients $\epsilon$ of oxygenated (HbO₂) and deoxygenated hemoglobin (HbR), the relative concentrations $C$ of the two chromophores can be calculated by solving the following equation [46]: $\left[\begin{array}{c}\stackrel{\circledR }{P_{532}}\\ \stackrel{\circledR }{P_{558}}\end{array}\right]=\left[\begin{array}{cc}{\epsilon }_{\mathit{Hb}{O}_2}(532)& {\epsilon }_{\mathit{HbR}}(532)\\ {\epsilon }_{\mathit{Hb}{O}_2}(558)& {\epsilon }_{\mathit{HbR}}(558)\end{array}\right]\left[\begin{array}{c}{C}_{\mathit{Hb}{O}_2}\\ C_{\mathit{Hb}R}\end{array}\right]$, where $\stackrel{\circledR }{{\mathit{P}}_{\lambda }}$ is the PA peak-to-peak amplitude at $\lambda$ wavelength after power normalization. Then, sO2 can be calculated as $s{O}_2=\frac{C_{\mathit{Hb}{O}_2}}{C_{\mathit{Hb}{O}_2}+C_{\mathit{Hb}R}}$. After obtaining sO₂, the same process was used as in vessel segment flow estimation described above, averaged sO₂ for each segment was added to the vessel demographic.

2.5. CFM data processing

The Aβ deposition analysis relied on the CFM images with the assistance of CRANAD-3. The pixel amplitude of the image was the peak of the pulsed emission. The images were first normalized so that the highest intensity was 1 and the background noise was 0. Then, the images were processed through Radial Symmetry Fluorescent in situ hybridization (RS-FISH) method [47] to automatically detect spot-like features. Aβ deposits highlighted with CRANAD-3 exhibit spot-like fluorescent signals, and RS-FISH detects patterns with radial symmetry in grayscale images, thus became our choice for automatic Aβ detection algorithm. RS-FISH Parameters (e.g., radius of spot, detection threshold, inlier ratio, etc.) were manually adjusted for the representative image from AD group to extract most of the plaques while minimizing false positive features. Once the parameters were determined, the same set of parameters was applied to all the data.

2.6. Immunohistochemistry (IHC) preparation

After the last imaging session (at the age of 55 weeks), the brain of each mouse was harvested and fixed in 10 % buffered formaldehyde solution for one week. The brain was then embedded in paraffin and sliced into 4 µm slices in the coronal plane. Slices from three different parts of the brain (See supplementary Figure S3A) were included in the histological analysis. Two different antibodies were used: CD31 for labeling the endothelial cells to indicate the blood vessels and 6E10 for labeling Aβ plaques [48]. Hematoxylin and eosin (H&E) staining was also performed. In this process, one of the wild type (WT) mice was excluded due to the unsuccessful harvesting of the brain, six WT type and five AD type mice were included in the analysis.

Brain slices were imaged on a Leica Aperio AT2 digital slide scanner (Leica Biosystems) at a resolution of 0.5 µm/pixel (20x objective). Quantitative assessment of the IHC-stained images was performed using the open-source program QuPath v0.5.1 (github.com/qupath/qupath). Cortical regions of the brain images were first manually segmented, as defined by the first 400 μm from the brain surface (see supplementary Figure S3A). For analyzing IHC signals for CD31 and Aβ, the images were subjected to the pixel thresholder, which separated IHC-positive pixels from the background. The amyloid-positive areas, the CD31-positive areas, and the number of detected CD31-positive objects, were calculated via QuPath (see supplementary Figure S3B). To quantify the capillary density, each CD31-positive object with area below the threshold (80 µm², roughly equivalent to 10 µm diameter vessels) was counted and divided by the total area of the cortex to represent the density. The Aβ density was calculated as Aβ-positive area divided by the total area of the cortex.

3. Results

We present exemplary images from one of the AD mice and one of the WT mice in Fig. 1. From left to right are observations from the 1st week, 4th week, 7th week, 10th week, and 14th week, corresponding to images of mice from 41 weeks old to 55 weeks old. For all the time points, all four metrics including vasculature, flow speed, sO₂, and Aβ deposition were presented without deteriorating image quality. This demonstrates the durability of the thin film cranial window for longitudinal study.

Fig. 1.

Representative images from four metrics at five time points. From left to right: week 1, week 4, week 7, week 10, and week 14. (A) PAM images from an AD mouse showing vasculature. (B) Flow speed images from the same mouse as (A). (C) _SO₂ images from the same mouse as (A). (D) CFM images from the same mouse as (A), showing the deposition of Aβ. (E) PAM images from an WT mouse showing vasculature. (F) Flow speed images from the same mouse as (E). (G) _SO₂ images from the same mouse as (E). (H) CFM images from the same mouse as (E), showing the deposition of Aβ. Scale bars represent 200 µm.

3.1. Vessel length density analysis

The representative vasculature images of an AD mouse and a WT mouse are shown in Figs. 1A and 1E. While small vessels were abundant in both AD and WT mice, we noticed that the density of the small vessels in the AD mouse decreased over time, whereas no clear trend was evident in the WT mouse. To quantify the result, we categorized each vessel segment by diameter and grouped them into five groups: below 8 μm, between 8 and 20 μm, between 20 and 40 μm, between 40 and 60 μm, and larger than 60 μm. For each group, we calculated the vessel length density (VLD), defined as total length divided by imaging area, shown in Fig. 2 A and 2B. While no clear trends of VLD change with time was observed in other groups, we did notice that in the 8–20 μm group, AD mice and WT mice exhibited the opposite trend where the AD group had a decreasing VLD while WT group had an increasing VLD (Fig. 2 A and 2B).

Fig. 2.

(A) Vessel size distribution from AD group. (B) Vessel size distribution from WT group. Error bar represents standard deviation (SD). (C) Mixed-effect linear regression on the relative change of vessel length density (VLD) of 8 – 20 µm size group with time and genotype as two variables. Ribbons represent 95 % confidence interval. (D) Absolute VLD change comparison between AD group and WT group with baseline subtracted. Error bars represent SD, *: p < 0.05, ****: p < 0.001. (E, F) The segment count of the 8 – 20 µm size group from (E) AD mice and (F) WT mice, both the individual value and the averaged value are presented. Error bars represent SD. See supplementary Fig. S4 for the segment count of all size groups. In all figures, the vessel length density (VLD) is defined as the total length of the vessels divided by imaging area, n = 5 for AD group and n = 7 for WT group.

In the following analysis, we focus on the 8 – 20 μm vessel group. To focus on the relative changes with aging and how the changes differed between AD and WT groups, we normalized the VLD of each group by the averaged VLD within the group of the first week. As a result, the averaged VLD of the first week becomes 1, and the values of the following weeks are percentage changes compared to that of the first week. As shown in Fig. 2 C, the normalized VLD of 8 – 20 μm group from all the individual animals are presented with two linear fittings performed for each group, the ribbon represents the 95 % confidence interval. We used linear mixed-effect (LME) regression to examine the effect of genotype, time and their interaction on VLD. The LME analysis indicates that the VLD of WT group increases with age at a non-significant rate of 1.06 % per week (SE = 0.57 %, p = 0.072). Importantly, the VLD of AD group decreases with age at a rate of −2.15 % per week, which is significantly lower compared with WT group (SE = 0.89 %; p = 0.019). We also performed unpaired t-tests on the absolute VLD change from the first observation, as shown in Fig. 2D. We found that although the VLD of the AD group consistently decreased, the difference from the WT group started to become significant only at week 10.

Similar decreasing trend was observed in the vessel segment count of the AD group compared to WT group, as shown in Figs. 2E and 2 F. Averaged segment number in the AD group went from 1051 in the first week to 803 in the 14th week, while in the WT group the value did not change too much (977–1010, less than 5 % change). Although segment count from the 14th week in the AD group was lower, statistically it was not significantly lower when compared to either the number from the first week in the AD group (p = 0.0527), or the number from the 14th week in the WT group (p = 0.0587). This indicates that the VLD decrease might be a combination effect of less vessel segments and shortening of the segments. The vessel segment count for all size groups is shown in supplementary Figure S4.

3.2. Aβ deposition analysis

The Aβ deposition analysis was facilitated by CFM, with the exemplary images presented in Figs. 1D and 1H. The CFM modality of our system could visualize the Aβ plaques as marked by the bright spots (Fig. 1D). As shown in Fig. 1D, AD mouse at all five time points exhibited presence of numerous bright spots, while limited bright spots were presented in WT mouse. More importantly, the AD mouse exhibited an obviously increasing trend in the number of Aβ plaques. The quantification of the Aβ plaques was performed using RS-FISH (see details in Materials and Methods), the performance of the algorithm is shown in Fig. 3 A. We presented examples from both AD group and WT group at the last observation time point and observed reasonable detection with visually low false-positive and false-negative rate.

Fig. 3.

(A) RS-FISH recognition performance with two examples from AD group and WT group each. Scale bar represents 200 µm. (B) Mixed-effect linear regression on plaque density with time and genotype being two variables, ribbons represent 95 % confidence interval. (C) Unpaired t-test results between AD group and WT group at each time point, *: p < 0.05, ***: p < 0.005, ****: p < 0.001, *****: p < 0.0005, error bars represent SD. For both (B) and (C), n = 5 in AD group and n = 7 in WT group. (D) Correlation analysis between VLD of 8 – 20 µm size group and plaque density, each animal contributes 5 data points from 5 different observation times. For group AD, n = 25 (5 animals at 5 time points), while for group WT, n = 35 (7 animals at 5 time points). The solid line is the least square linear regression of each group, while the r and p are the Pearson’s correlation coefficient and statistical significance, respectively. Ribbons represent 95 % confidence intervals.

Next, we used LME regression to examine the effect of genotype, time, and their interaction on Aβ plaques density, defined as number of plaques divided by imaging area. As shown in Fig. 3B, two linear fittings were performed on each group, the ribbons represent 95 % confidence interval. We are interested in both the intercept and the rate of change. The WT group has an intercept of 4.79 mm⁻² (SE = 2.00 mm⁻², p = 0.020), which means that at the first observation, WT mice had a positive plaque density that was significantly different from 0. It should be noted, however, that this may not indicate WT mice express a significant amount of Aβ. This could be a mixed effect of false-positive recognition of plaques and occasional expression of Aβ in WT mice.

The main effect of the AD showed significantly higher plaque density in the first week by 6.31 mm⁻² (β₀=6.31 mm⁻², SE = 3.11 mm⁻², p = 0.0472) compared to WT group. This indicates that by the time of observation, AD mice already expressed significantly more Aβ plaques in the cortical region than the WT mice. More importantly, while no clear change over time on plaque density was observed on WT mice, (β=-0.0014 mm⁻²/week, SE = 0.26 mm⁻²/week, p = 0.9958), AD group exhibited significantly higher increasing rate of 1.60 mm⁻²/week (β=1.60 mm⁻²/week, SE = 0.415 mm⁻²/week, p = 0.0003) compared to WT group. When comparing plaque density of the two groups at each individual time points (Fig. 3 C), we noticed that while the density was increasingly higher in AD group compared with WT group, the trend seemed to slow down between 10th week and 14th week of observation (51-week-old and 55-week-old). This might indicate the late stage of AD where Aβ deposition reached its maximum.

After quantifying both small-vessel VLD and plaque density, we performed a correlation analysis between these two metrics (Fig. 3D). The plaque density of each animal at each of the five observations was plotted against the corresponding VLD within the 8 – 20 μm vessel group. Both biomarkers exhibited a significant altering trend in AD group as compared to WT controls. In the AD group, the correlation coefficient was r = -0.364, suggesting that the increase in Aβ load was associated with a decline in small-vessel density. The correlation did not reach statistical significance (p = 0.08), likely reflecting the limited sample size (n = 5 × 5). The same analysis in WT mice—characterized by markedly lower and relatively stable Aβ levels—yielded a weaker negative correlation (r = –0.255, p = 0.14), consistent with the minimal change in plaque load over time. Collectively, these findings illustrate the potential of our longitudinal imaging framework to reveal dynamic relationships among pathological and vascular biomarkers in neurodegenerative disease models.

3.3. Functional PAM analysis

Blood vessel analysis could be greatly enhanced by functional PAM as it provides two additional metrics that are crucial to understanding hemodynamics. Representative flow speed images are shown in Figs. 1B and 1F. Functional PAM can clearly visualize the flow speed of large and small vessels at the same resolution as conventional PAM. To quantify the flow speed, we grouped vessels with different diameters and calculated the averaged flow speed in each category (see details in Materials and Methods). While observing a positive correlation between vessel diameter and flow speed overall (Fig. 4 A and 4B), we observed neither a difference in temporal pattern nor a group-averaged difference between the AD and WT group. The largest non-significant difference was observed in vessel group of 40 – 60 µm, with average flow speed in AD group 10 %, 11 %, 12 %, 16 %, and 8 % lower at week 1, 4, 7, 10, and 14, respectively, when comparing to WT group. The validity of the measurement was verified through the calibration measurements. As shown in Fig. 4 C, the parameter derived from the correlation coefficient showed a good linear relationship with the velocity (R² = 0.97), which proved that our measurement is consistent with the PACS theory. The blood flow velocity value was derived from the mapping between velocity and correlation coefficient, where the dependence of velocity on correlation was fitted with a cubic polynomial ($y=-12.26x^3+22.98x^2-16.03x+4.886$, see Fig. 4D).

Fig. 4.

(A) Flow speed distribution by vessel size in AD group (n = 5). (B) Flow speed distribution by vessel size in WT group (n = 7). Error bar represents standard deviation in (A)(B). (C) Linear fitting of flow speed and intermediate parameter derived from correlation coefficient, used for PACS theory validation. (D) Correlation measurement over selected flow speed, fitted with a cubic polynomial ($\mathit{y}\mathit{=}\mathit{-}\mathit{12}\mathit{.}\mathit{26}{\mathit{x}}^{\mathit{3}}\mathit{+}\mathit{22}\mathit{.}\mathit{98}{\mathit{x}}^{\mathit{2}}\mathit{-}\mathit{16}\mathit{.}\mathit{03}\mathit{x}\mathit{+}\mathit{4}\mathit{.}\mathit{886}$). The fitted curve was used for in vivo flow speed mapping. Error bar represents the minimum and maximum value.

The sO₂ was analyzed in a similar manner as flow speed, and shown in Fig. 5 A and 5B, except that we excluded artery (sO₂ > 0.85) from the analysis. For each size group and timepoint combination, we compared the sO₂ between AD group and WT group. We found that on the last observation timepoint, sO₂ was significantly lower in all size groups of AD when comparing to WT, except for the vessels larger than 60 µm (Fig. 5 C).

Fig. 5.

(A) Averaged sO₂ for veins in different size groups and time points of AD animal group (n = 5). (B) Averaged sO₂ for veins in different size groups and time points of WT animal group (n = 7). Error bars in (A) (B) represent standard deviation. (C) Box plot for all size groups of vessels of the observation at week 14. Unpaired t-test performed for each size group between AD type and WT type, *: p < 0.05, **: p < 0.01.

3.4. Histology analysis

Immunohistochemistry (IHC)-processed slices imaged with microscope are presented in Fig. 6A-D. We present the slices in the coronal plane of the brains harvested at the endpoint from one AD mice (Fig. 6 A, 6B) and one WT mice (Fig. 6 C, 6D). Slices were stained with 6E10 (Fig. 6 A, 6 C) to label Aβ and CD31 (Figs. 6B, 6D) to label endothelial cells. It is obvious that in the Aβ-labeling slices, the one for AD shows distribution in all regions of the brain, including the cortical region (Fig. 6 A). The severity of the Aβ deposition also indicate that the brain is overloaded with plaques. On the contrary, the one for WT shows sporadic to no deposition of Aβ (Fig. 6 C). We also noticed that the size of each Aβ deposit (on the order of 10–20 µm, Fig. 6 A right image) coincided with that in the in vivo images (Fig. 3 A). The quantified Aβ plaque density also reflects the significant group-wise difference in the brain cortex between AD group and WT group (Fig. 6 F), which matches with the conclusion from the in vivo analysis.

Fig. 6.

(A)-(D) Selected histological images of slices from the mouse brains harvested at the endpoint of the in vivo study. The right images are the enlarged selection from the black box in the left images. Scalebar on the left images represent 1 mm, scalebar on the right images represent 100 µm. (A) Brain slice from an AD mouse, stained with 6E10 to highlight the Aβ deposition. (B) Brain slice from the same AD mouse as in (A), stained with CD31 to highlight the endothelial cells. (C) Brain slice from a WT mouse, stained with 6E10 to highlight the Aβ deposition. (D) Brain slice from the same AD mouse as in (C), stained with CD31 to highlight the endothelial cells. (E) Boxplot of the vessel count density of the two groups, the unpaired t-test did not show a statistically significant difference. (F) Boxplot of the Aβ plaque density of the two groups, the unpaired t-test showed a statistically significant difference. In (E) and (F), n = 5 for AD and n = 6 for WT, one WT mouse was removed from analysis due to failure in brain harvesting.

The CD31-stained slices highlighted the presence of blood vessels, as shown with the brownish color in Figs. 6B and 6D. From the area of blood vessel cross-section, we could focus on the small vessels (defined as vessels with area smaller than 80 µm², corresponding to roughly 10 µm diameter). We quantified the vessel density by dividing the total count of small vessels in the cortex by the total area of the cortex. From the quantified vessel density of the cortical region (Fig. 6E), however, we did not observe a statistical difference between the WT group and the AD group. This could be a multi-factorial result, for example, increased string vessel (non-functional capillary without blood perfusion) accompanied by high Aβ load has been reported [49]. While the endothelial cells of such vessel could be labeled with IHC analysis, the vessels themselves might not appear under in vivo PA imaging. In addition, the alteration of vasculature was much more subtle than the Aβ deposition as it reflected the mixed effect of disease and age, as well as individual variation and ROI selection. By looking at a snapshot of the endpoint, the statistical difference between AD group and WT group may require a large number of animals. Longitudinal study, however, could amplify the age-dependent alterations and avoid the individual difference by observing the same region of the same mouse continuously.

4. Discussion

This study successfully demonstrated the methodology of longitudinal brain imaging and biomarker analysis in a live mouse model of AD, facilitated by a dual-modality imaging system that combined PAM and CFM. Each modality provided distinct yet complementary information on brain biomarkers. We were able to monitor multiple biomarkers longitudinally for three months with uncompromised image quality through a cranial window transparent to both light and ultrasound. The observation length was determined by the study plan rather than limited by the durability of the window. The CFM modality enabled visualization of Aβ plaque deposition at single-plaque resolution, whereas the PAM modality provided detailed insights into the vasculature and hemodynamics at single capillary resolution. Notably, this is the first study to demonstrate functional PAM imaging of blood flow and sO₂ in a mouse model of AD. The multiparametric analysis, performed in the frame of a longitudinal dataset, enabled investigation of biomarker correlations with low variance, high consistency and high efficiency, which is critical to neurodegenerative disease research that typically involves large quantities of animals and complicated development of pathological parameters.

Our observation window captured the rapid growth of Aβ plaque deposition in 9-month-old 5xFAD mice, with a slowing down trend noticed at the age of 12 months (Figs. 3B, 3 C). This pattern was consistent with the findings from previous studies [15], [50], where the 5xFAD mice typically experienced onsets of Aβ accumulation in the brain cortex at around 6-month-old, and rapid growth of the Aβ density until 12-month-old, where the brains were loaded with Aβ deposits and the mice started to decease.

Cerebral vasculature related dysfunction and biomarkers in AD involve multiple intercorrelated factors, such as impaired blood-brain barrier (BBB) integrity, decreased vessel length and density, abnormal vessel morphology, and decreased cerebral blood flow (CBF) [3], [5], [35]. These biomarkers may also behave differently in different phenotypes. As discussed in our previous publication, due to the enlarging effect of 0 – 16 μm resulting from the finite beam size of the excitation beam (8 μm FWHM) [37], we believe that the majority of the vessels of our measurements in the 8–20 μm group belong to small vessels with size < 10 μm. As a technique developing study rather than a biology discovering study, our vasculature analysis focused on the rather straightforward metric, vessel length density. Our results in the vasculature align with most transgenic mouse models of AD where cortical vascular density declines with the progression of the disease [51]. While most studies utilized ex vivo immunohistology staining to evaluate cerebral vasculature of AD mice, only a limited number of studies utilized in vivo imaging techniques such as two-photon fluorescence imaging [52], high-frequency micro-Doppler imaging [16], MRI [53], and OCT angiography (OCTA) [21], with the last one utilizing OCTA being the only study implemented longitudinal monitoring with single capillary resolution. Interestingly, the aforementioned OCTA study did not show a significant decreasing trend in capillary density with age in the AD group compared to the WT group [21].

The advantage of longitudinal study lies in the low variance that leads to a consistent trend throughout the observation window. We estimated that by observing the same ROI at different time point, the variance of VLD stemming from individual difference can be greatly reduced, while considering misalignment between measurements (see supplementary note 4). The continuous temporal profile that enables capture of subtle age-dependent changes, which, combined with changes in all other biomarkers, can infer correlation between biomarkers. In our study, we observed a linear decreasing trend in vessel density in AD group, which indicated that the observation time window was within a vessel loss period. In order to capture the onset of the process, or to see if this process lasts until a certain point, a longer observation window would be necessary. It is worth noting that we observed an increasing trend of absolute small vessel density in WT group, which might be associated with the angiogenic response under the cranial window.

Baseline CBF reduction have been widely reported in both AD patients and AD animals, with decreases ranging from 15 % to 35 % [5], [54]. The majority of these findings were measured by arterial spin labeling MRI (ASL-MRI), which quantifies arterial perfusion at a macroscopic scale. While correlated, the blood flow velocity in our measurement was not equivalent to CBF. We observed a flow speed decline of 8 % - 16 % in the 40 – 60 μm vessel size category of AD mice compared to WT mice; however, none of the differences reached statistical significance, and the influence of age was not obvious. Notably, pathological vascular behaviors are model-dependent; for example, 5xFAD was specifically reported to have no significant change in CBF [55]. Still, several factors may have contributed to the modest differences detected in our measurements. First, the categorization by size may diminish some of the differences as flow reduction is usually accompanied by reduction in vessel size [56], [57]. A more rigorous approach may involve registration and tracking of the same vessels and their function. Second, the flow speed measurement could be improved in several aspects. Improvements such as increasing the correlation length and normalizing pulse-to-pulse fluctuations during PACS acquisition may yield more accurate estimates. Additionally, the reliability of flow speed mapping, especially at small vessels (< 20 μm) warrants further investigation, given that the calibration phantom used was much larger (300 μm). When vessel size is approaching the laser beam size, the autocorrelation function will also change due to the reduced number of particles (red blood cells) within the beam volume [58]. Future verification should therefore account for small-vessel limits, as well as red blood cell density and spatial distribution, to ensure accurate flow quantification.

The observation of the decrease in venous sO₂ at the last observation point might be another indication of vascular dysfunction in AD mice. The baseline sO₂ was less studied, while controversial results of both increasing and decreasing have been reported [59], [60]. We deduce that the disappearance of small vessels, which could be a result of capillary stalling, leads to the hypoperfusion of the brain tissue and further to the reduction of venous sO₂.

From the longitudinal dataset of the multi-parametric biomarkers obtained, we not only could study the alteration patterns of neurodegenerative diseases, but also could investigate the interactions between different biomarkers. For example, we were able to present the correlation between VLD and plaque density (Fig. 3D), suggesting a potential interaction between vascular and Aβ load. Although the present work does not include interventional experiments to establish causal links, the baseline associations identified here underscore the capability of our longitudinal framework to probe mechanistic connections among Aβ deposition, vascular structure, and vascular function.

It should be noted that there were limitations in these studies that should be considered for future improvement. Firstly, based on the longitudinal data, the observation window should be adjusted or prolonged. In the cortical region of 5xFAD mice, the Aβ accumulation, vasculature and cerebral blood flow change could occur as early as 6-month-old [15], [17]. Meanwhile, extra care should be taken into implantation of the cranial window for higher success rate and durability. Secondly, several system-wise improvements can be made. For example, the relatively low numerical aperture of the objective lens limited the z-axis resolution of the CFM modality, which, if improved to around 10 μm, could enable the spatial co-registration of the imaged Aβ plaques and vasculature. In addition, the field of view could be further expanded to the whole brain while maintaining consistent resolution and sensitivity across all biomarker measurements [33], [61]. Such expansion could reduce the variance caused by the region-of-interest selection, while providing information across different cortical regions. Thirdly, analysis based on more advanced algorithms could further enhance the research. Many studies had found vessel constriction or stall in AD mouse models [5], [19], [21], which requires tracking of the morphology and existence of the same vessel longitudinally. Automatic vessel categorization and registration longitudinally should be considered for the future development of algorithms. On the top of that, a rich library of parameters could be extracted from the structural and functional maps of the vessels [21], [62], such as flow, diameter, tortuosity by type of vessels (e.g., venule, arteriole, penetrating vessel, and capillary), vessel resistance, perfusion level, and metabolic rate of oxygen.

In summary, we have described a method of longitudinal, dual-modal, and multi-parametric imaging approach for monitoring key pathological features in a mouse model of AD, targeting Aβ deposition, vasculature, blood flow speed, and sO₂. This integrated methodology enables the extraction of a spatially co-registered comprehensive dataset from a single imaging system, reducing both cost and experimental complexity. Our results in Aβ deposition and vasculature were consistent with established findings by previous studies, validating the reliability of our approach. Notably, by leveraging the functional imaging of PAM, we observed a reduction in venular sO₂ in 55-week-old mice, which coincided with marked Aβ accumulation, suggesting a potential link between impaired oxygen metabolism and pathological progression. This research goes beyond the capability of PAM in the vascular aspect of AD research, and provides a multi-modal platform for investigating AD etiology that connects the vasculature with the widely adopted pathological biomarkers in a simultaneous, longitudinal manner. This multi-parametric approach may not only enhance our understanding of AD pathophysiology but also provide a foundation for assessing therapeutic interventions targeting both vascular and Aβ-related biomarkers.

CRediT authorship contribution statement

Daniel A. Lawrence: Writing – review & editing, Validation, Supervision, Resources, Methodology, Funding acquisition, Conceptualization. Geoffrey G. Murphy: Writing – review & editing, Validation, Supervision, Resources, Methodology, Funding acquisition, Conceptualization. Enming Joseph Su: Methodology, Investigation. Chenshuo Ma: Writing – review & editing, Visualization, Methodology, Investigation, Conceptualization. Wei Zhang: Writing – review & editing, Visualization, Validation, Supervision, Software, Methodology, Investigation, Conceptualization. Yannis M. Paulus: Writing – review & editing, Supervision, Resources, Funding acquisition, Conceptualization. Linyu Ni: Software, Investigation, Data curation. Xueding Wang: Writing – review & editing, Validation, Supervision, Resources, Project administration, Methodology, Funding acquisition, Conceptualization. Tianqu Zhai: Writing – review & editing, Writing – original draft, Visualization, Validation, Software, Methodology, Investigation, Formal analysis, Data curation, Conceptualization.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Biographies

Tianqu Zhai received Ph.D. degree from Electrical Engineering at the University of Michigan in 2024, with a focus on spectroscopy, fiber laser, photoacoustic imaging and multi-modality imaging. His primary research interest includes development of optical systems and sensors for biological sensing in the context of healthcare monitoring or disease model research.

Wei Zhang received the Graduate degree from Sun Yat-Sen University, Guangzhou, in 2012. He received the master’s degree and Ph.D. degree in the institution of Biomedical Engineering, Peking Union Medical College, Beijing, China. He is currently a research investigator in Department of Biomedical Engineering at University of Michigan. His current research interest includes photoacoustic imaging, multi-modality imaging and ionizing radiation acoustic imaging.

Chenshuo Ma is currently working as a Research Investigator at University of Michigan (MI, United States), focused on photoacoustic microscopy and photoacoustic tomography imaging. She was a Research Scientist at Duke University (NC, United States). She received her Ph.D. (2018) degree in Biomedical Engineering from Macquarie University (Sydney, Australia).

Linyu Ni she received her Ph.D. in Biomedical Engineering at the University of Michigan, Ann Arbor. Her primary research interests are photoacoustic microscopy and clinical applications of photoacoustic imaging.

Yannis M. Paulus, M.D., F.A.C.S, is an academic vitreoretinal surgeon and clinician scientist that loves applying optics, photonics, biomedical engineering, and nanoparticles to develop novel retinal imaging and therapies. He is the Helmut F. Stern Career Development Professor and Associate Professor with Tenure, Department of Ophthalmology and Visual Sciences and Department of Biomedical Engineering and Medical Director of the Grand Blanc ACU at the University of Michigan. He has published over 160 peer reviewed publications in leading journals and received numerous awards, including the Macula Society Gragoudas Award, Alcon Young Investigator Award, and the ARVO Early Career Clinician-Scientist Research Award.

Enming Joseph Su’s goal is to investigate the cellular and molecular mechanisms that have an impact on the progression of neurovascular diseases. With over twenty-five years of experience in vascular biology and vascular surgery in murine models, he has developed a number of in vivo models to study vascular diseases in CNS, which enabled us to develop novel treatments for new clinical trials. Specifically, he plans to determine the roles that platelet derived growth factor C (PDGF-CC) and its receptor PDGFRα play in neuro-vascular diseases, such as stroke (Nat. Med. 2008;14:731–737), TBI (Front Cell Neurosci. 2015 Oct 7;9:385) and Alzheimer disease.

Geoffrey G. Murphy, Ph.D. is the David F. Bohr Collegiate Professor in Physiology in the Department of Molecular & Integrative Physiology and a founding faculty member of the Michigan Neuroscience Institute. Dr. Murphy received his BS in neurobiology from the University of California at Berkely and his Ph.D. from the University of California, Los Angeles. Dr. Murphy’s lab is focused on understanding the neurobiological substrates of memory under normative conditions as well as in the context of neurological and psychiatric disease states.

Daniel A. Lawrence, Ph.D. is the Frederick G L Huetwell Professor of Basic Research in Cardiovascular Medicine at the University of Michigan. He received his Ph.D. in Molecular Biology from Umeå University, Umeå Sweden in 1989, followed by Postdoctoral training at the University of Michigan. His laboratory studies the role of proteases and their inhibitors in health and disease. Primary areas of interest focus on the vascular biology of the CNS, and on the mechanisms of fibrotic disease. He is the founder of MDI Therapeutics a pharmaceutical company developing novel therapies for the treatment of fibrosis and fibroproliferative diseases.

Xueding Wang, Ph.D. is a Jonathan Rubin Collegiate Professor of the Department of Biomedical Engineering and the Professor of the Department of Radiology at the University of Michigan School of Medicine. Before working as an independent principle investigator, Dr. Wang received his Ph.D. at Texas A&M University and Postdoctoral training at the University of Michigan. Dr. Wang has extensive experience in development of medical imaging and treatment technologies, especially those involving light and ultrasound. Sponsored by NIH, NSF, DoD and other funding agencies, his research has led to over 200 peer-reviewed journal papers. Dr. Wang was the recipient of the Sontag Foundation Fellow of the Arthritis National Research Foundation in 2005, the Distinguished Investigator Award of the Academy of Radiology Research in 2013, and was elected to the fellow of AIMBE in 2020 and the fellow of SPIE in 2021. He is also sitting on the editorial boards of scientific journals including Photoacoustics, Ultrasonic Imaging, and Journal of Biomedical Optics.

Appendix A. Supplementary material

Supplementary material

Data availability

The data that support the findings of this study are available upon reasonable request.