Capillary function progressively deteriorates in prodromal Alzheimer’s disease: A longitudinal MRI perfusion study

Abstract

Cardiovascular risk factors are associated with the development of Alzheimer’s disease (AD), and increasing evidence suggests that cerebral microvascular dysfunction plays a vital role in the disease progression. Using magnetic resonance imaging, we investigated the two-year changes of the cerebral microvascular blood flow in 11 mild cognitively impaired (MCI) patients with prodromal AD compared to 12 MCI patients without evidence of AD and 10 cognitively intact age-matched controls. The pAD-MCI patients displayed widespread deterioration in microvascular cerebral perfusion associated with capillary dysfunction. No such changes were observed in the other two groups, suggesting that the dysfunction in capillary perfusion is linked to the AD pathophysiology. The observed capillary dysfunction may limit local oxygenation in AD leading to downstream β-amyloid aggregation, tau hyperphosphorylation, neuroinflammation and neuronal dysfunction. The findings are in agreement with the capillary dysfunction hypothesis of AD, suggesting that increasing heterogeneity of capillary blood flow is a primary pathological event in AD.

Introduction

Alzheimer’s disease (AD) is the most common type of dementia with an estimated 5–7 million new cases worldwide every year and remains a major healthcare concern as no prevention or satisfactory treatment is currently available [17], [38], [49]. AD is characterised by abnormal protein aggregation in the brain manifesting as extracellular deposition of β-amyloid (Aβ) plaques and intracellular neurofibrillary tangles formed by hyperphosphorylated tau. These processes are thought to induce neurodegeneration leading to widespread cerebral atrophy in AD [37]. The build-up of Aβ oligomers and fibrils has long been considered the disease-initiating factor of AD and, as a result, extensive efforts have been made to stop disease progression by preventing Aβ deposition or removing Aβ aggregates from the brain. Disappointingly, this approach has had limited clinical success, implying that other disease mechanisms, acting either in conjunction or in parallel with Aβ aggregation, may contribute to the development and progression of the disease [25]. Indeed, AD is now considered to be the result of multiple accumulating risk factors rather than a single cause [3]. Epidemiological studies of AD reveal that essentially all known risk factors for AD, such as age, history of stroke, diabetes mellitus, atherosclerosis or inheritance of the Apolipoprotein E ε4 (ApoE4) gene, are associated with a vascular component [11]. In this article, we provide evidence that disturbances in brain microvasculature are an early and progressive pathological feature of AD. These findings may add new perspectives to the pathogenesis of the disease.

Vascular damage in Alzheimer’s disease

It is well established that cerebral hypoperfusion presenting as a macroscopic reduction in cerebral blood flow (CBF), is an early manifestation of AD and already observable at the mild cognitive impairment (MCI) stage of the disease [1], [20], [27], [39]. Furthermore, AD is associated with blood–brain barrier (BBB) dysfunction and morphological changes in brain capillaries that include basement membrane thickening, luminal narrowing and pericyte degeneration [11], [43].

The two-hit vascular hypothesis of AD posits that hypoperfusion and BBB breakdown (hit 1) precipitates Aβ production and retention (hit 2), which in turn causes perfusion and BBB function to further deteriorate [23], [51]. Meanwhile, studies suggest that Aβ oligomers cause capillary constriction at the location of capillary pericytes, but not arteriolar diameter changes, in AD [32]. The extent to which constriction of a proportion of tissue capillaries may limit CBF by increasing vascular resistance across the capillary bed in the human brain is unclear, and there is hence a need to understand whether cerebral amyloidosis affects CBF, capillary flow, or both, in MCI patients. According to the capillary dysfunction hypothesis of AD, capillary flow disturbances may cause sizeable reductions in tissue oxygenation, even in the absence of parallel CBF reductions [21], [34], [35]. Both arteriolar and capillary components of vascular damage related to Aβ, age and other risk factors may therefore be crucial to understand AD pathophysiology. Detection of capillary flow disturbances is possible using magnetic resonance imaging (MRI). Specifically, using dynamic susceptibility contrast (DSC) MRI, it is possible to extend the measurements of “classical” perfusion metrics, such as CBF, cerebral blood volume (CBV), and mean transit time (MTT), to also include estimates of capillary transit time heterogeneity (CTH), the standard deviation of blood’s microvascular transit times [29], [35]. Combining MTT and CTH, the balance between local oxygen availability and normal brain oxygen demands can be indexed in terms of the tissue oxygen tension (P_tO₂) using biophysical models [4], [21].

We have previously presented evidence of capillary dysfunction in MCI individuals with raised cortical Aβ levels in a cross-sectional study [31]. Here, we extend these findings by evaluating the longitudinal changes in microvascular perfusion in the same study cohort. Based on the capillary dysfunction hypothesis of AD, we hypothesise that microvascular perfusion will further deteriorate in MCI individuals with elevated cortical Aβ, representing prodromal AD (pAD-MCI), compared to subjects with suspected non-Alzheimer’s pathophysiology (SNAP)-MCI, i.e. individuals with low cortical Aβ, and age-matched healthy controls (HC). In particular, we hypothesise increasing heterogeneity of microvascular blood flow accompanied by prolonging of microvascular blood transit times in the pAD-MCI individuals.

Methods

This study is part of a longitudinal study designed to evaluate the temporospatial relationships between the different pathologies of AD, including brain perfusion, amyloid and tau deposition, neuroinflammation, and cognition. The study was approved by the Central Denmark Region Ethics Committee [1–10–72–116–13] and all subjects gave their informed written consent. Subjects were cognitively tested and examined with multiple positron emission tomography (PET) and MRI scans at baseline and two-year follow-up.

Study subjects

MCI subjects were recruited from national memory clinics and by advertisement applying the inclusion criteria described in Parbo et al. [36]. HC subjects were recruited by advertisement with similar inclusion criteria but had no subjective memory complaints or overt cognitive impairment. All subjects had to have an estimated glomerular filtration rate (eGFR) ≥ 60 to safely undergo the perfusion MRI scan. In this report, we included those MCI subjects who had completed perfusion MRI and amyloid ¹¹C-Pittsburgh compound B (¹¹C-PiB) PET scans both at baseline and at two-year follow-up. HC subjects who completed perfusion MRI at baseline and follow-up were also included. Due to upgrades in the MRI protocol, some participants recruited at the beginning of the study did not receive perfusion MRI, and hence, were not included in this report. The Montreal Cognitive Assessment (MoCA) and the Clinical Dementia Rating Sum of Boxes (CDR-SOB) scales were used to evaluate the cognitive status of the MCI subjects at the start and the end of study.

MRI acquisitions

MRI was performed on a 3 T Skyra system (Siemens Healthcare, Erlangen, Germany) using a 32-channel head coil. Structural 3D T1-weighted MP2RAGE [26] was acquired with 1 mm isotropic voxels. T2-FLAIR was acquired with 0.7 × 0.7 × 3.0 mm voxels. Both images were evaluated to exclude structural abnormalities or lesions.

To estimate brain perfusion, two DSC-MRI sequences were acquired with echo-planar imaging (EPI) (TR = 1.5 s); the first, using gradient echo (GE) (300 volumes with 3 mm isotropic voxels in 29 slices, no gap), and the second using spin echo (SE) (300 volumes with 3 mm isotropic voxels in 19 slices, 1 mm slice gap). Gadobutrol (Gadovist, Bayer HealthCare Pharmaceuticals, Berlin) was used as contrast agent with a concentration of 0.1 mmol/kg for GE and 0.2 mmol/kg for SE, each followed by 20 mL of saline with an injection rate of 5 mL/s. GE DSC-MRI is sensitive to contrast agent in vessels of all sizes, while SE DSC-MRI is relatively more sensitive to contrast agent in capillary-sized vessels and so is used to assess the microvascular flow [7], [29] Whole-brain coverage was not possible using DSC-MRI due to limitations in image acquisition, and so, superior (e.g. motor cortex) and inferior parts (base of the brain and cerebellum) were outside the field of view (FOV).

Perfusion analysis

DSC-MRI was processed and analysed using in-house software [29], using MATLAB (The MathWorks, Inc.) and SPM (The Wellcome Trust Centre for Neuroimaging, University College London). First, images were motion and slice-time corrected and the corresponding T1-weighted image was co-registered to the mean perfusion image. Then, volumes were truncated 60 s after the start of the contrast bolus, and automatic arterial input function (AIF) selection was performed (described in details in Appendix A, AIF selection). Parametric maps of CBF, CBV, MTT and CTH were calculated by parametric deconvolution [29]. CBF and CBV were normalised relative to normal-appearing white matter (NAWM). Lastly, voxel-wise P_tO₂ was estimated utilising a biophysical model of brain metabolism [12]. The model estimates the P_tO₂ required to support normal resting cerebral metabolic rate of oxygen (CMRO₂) given the local hemodynamics, as determined by SE-based MTT and CTH measurements. The CMRO₂ was assumed to be 2.5 mL/100 mL/min and oxygen extraction fraction (OEF) to be 0.3 [12], [44]. All parametric maps were smoothed with a 3 × 3 mm full width at half maximum (FWHM) Gaussian kernel using a smoothing mask containing only white matter (WM) and grey matter (GM) tissue to reduce overlap between the signal from brain tissue and the signal from cerebrospinal fluid and other tissues.

The perfusion analysis is only dependent on the dynamic bolus passage of the contrast agent. Thus, the parameter estimation is not sensitive to potential slow leakage of the BBB that might occur in subjects with AD pathophysiology. For the same reason, well-mixed circulating contrast agent from the first DSC-MRI does not affect the parameter estimation of the second DSC-MRI.

Amyloid imaging

Aβ load was evaluated with PET using a High-Resolution Research Tomograph (HRRT) (CTI/Siemens, Knoxville, TN) as previously described [36]. Briefly, ¹¹C-PiB was injected intravenously giving a mean dose of 391 MBq followed by PET acquisition in list mode at 40–90 min post-injection. After reconstruction, the 60–90 min averaged ¹¹C-PiB image was used to calculate a standard uptake value ratio (SUVR) image with cerebellar GM as reference. Finally, the image was smoothed using a 3 × 3 × 3 mm FWHM Gaussian kernel.

Amyloid status

To identify subjects with indications of early AD, the MCI subjects were dichotomised into a group with high cortical Aβ load, categorised as pAD-MCI, and a group with low or no cortical Aβ load, categorised as suspected non-Alzheimer’s pathophysiology (SNAP)-MCI. The Aβ load of each subject was evaluated from the ¹¹C-PiB PET at two-year follow-up, sampling a composite region of interest consisting of the inferior temporal gyrus, the fusiform gyrus, the posterior cingulate gyrus, the parietal operculum and orbitofrontal areas. These areas show early deposition of Aβ and increased ¹¹C-PiB retention in these areas suggest pAD [16]. The distribution of composite ¹¹C-PiB SUVRs in the MCI cohort was clearly bimodal and a cut-off value of 1.49 SUVR was established by two standard deviations from the mean ¹¹C-PiB uptake in 11 HC subjects. To evaluate the progression of amyloid load in the two MCI groups, a paired t-test was conducted using the mean ¹¹C-PiB SUVR value in the composite ROI.

Cortical surface extraction

Cortical perfusion and Aβ load were analysed using surface-based statistics. The cortical surface corresponding to the middle cortical layer was estimated from the high-resolution T1-weighted image using Fast Accurate Cortex Extraction (FACE) [13]. The cortical surface was moved to parametric native space by utilising a rigid-body co-registration between T1-weighted image and the parametric images (perfusion MRI and ¹¹C-PiB PET). Then, the signal of each parametric image was interpolated and mapped onto the cortical surface. To perform statistical analyses, individual surfaces were co-registered and moved to the cortical surface of an average non-linear template in MNI-space using a feature-driven surface registration algorithm [14], [15]. Lastly, the signals on the cortical surface were smoothed using a 20 mm FWHM geodesic Gaussian kernel. As the FOV of the perfusion images was limited, only areas available in all subjects were used for further analysis.

Statistical analysis

All analyses were performed using Python 3.7 (Python Software Foundation). Subject demographics were evaluated between the pAD-MCI, SNAP-MCI and HC groups using ANOVA for continuous data and χ2 test for binary data. Results are presented with a 95 % confidence interval. A linear mixed-effect model (LMM) was used to investigate and compare the longitudinal change of the perfusion parameters of the three groups. The fixed effects model the average group trajectories over the two-year follow-up period, while the random effects model that of individual subjects, given their different intercepts and slopes. The model was controlled for sex and ApoE4 differences. The statistical model was fitted at each vertex point on the cortical surface. Given the multiple comparisons performed, results were family-wise error rate (FWER) corrected (α = 0.05) using cluster-extent-based thresholding with two levels of primary cluster-defining threshold; p < 0.01 and p < 0.001. Visbrain was used for the visualisation of the cortical surfaces [9].

Results

A total of 43 MCI (25 pAD-MCI) and 24 healthy age-matched control subjects were recruited in the original study. A subgroup of 23 MCI (11 pAD-MCI) and 10 control subjects completed MRI perfusion at both baseline and two-year follow-up and was included in this report. An overview of subject demographics is presented in Table 1. The three groups had a similar education level, proportion of ApoE4 carriers, and age. In addition, the groups had similar proportion of prescribed medicines (antihypertensives, anti-diabetics, statins, aspirin and non-steroidal anti-inflammatory drugs (NSAIDs)). The sex ratio was significantly different between the groups due to a relatively higher proportion of males in the pAD-MCI group. Furthermore, there was a significant difference in the mean MoCA and CDR-SOB scores between the groups at both baseline and two-year follow-up. Using the guidelines from O’Bryant et al. [33], four pAD-MCI subjects and one SNAP-MCI subject were considered to have progressed to mild dementia during the two-year follow-up period.

Table 1.

Characteristics of participants.

	pAD-MCI (n = 11)	SNAP-MCI (n = 12)	HC (n = 10)	P-value
Sex (male/female)	9/2	5/7	3/7	0.04a
Age, baseline (years)	71.1 [67;75.2]	65.8 [61.9;69.7]	70.4 [66.1;74.7]	0.12b
Education (years)	11.5 [9.7;13.3]	12.4 [10.7;14.1]	14.0 [12.1;15.9]	0.16b
ApoE4 (Yes/No)	7/4	3/8*	5/5	0.23a
Antihypertensives (Yes/No)	6/5	4/8	3/7	0.45a
Anti-diabetics (Yes/No)	0/11	1/11	0/10	0.41a
Statins (Yes/No)	3/8	1/11	2/8	0.49a
Aspirin (Yes/No)	4/7	0/12	3/7	0.07a
NSAIDs (Yes/No)	5/6	1/11	3/7	0.13a
MoCA
Baseline	24.0 [21.1;26.9]	26.1 [24.3;27.9]	27.6 [25.5;29.7]	0.04b
Follow-up	21.0 [18.8;23.2]	25.1 [23;27.2]	26.2 [23.9;28.5]	<0.01b
CDR-SOB
Baseline	1.8 [1.2;2.4]	1.1 [0.6;1.6]	0 [-]	<0.01b
Follow-up	3.6 [2.4;4.8]	1.5 [0.4;2.6]	0.2 [0;0.4]	<0.01b

Results are presented with a 95 % confidence interval.

Abbreviations: pAD, prodromal Alzheimers disease; SNAP, suspected non-Alzheimers pathophysiology; MCI, mild cognitive impairment; HC, healthy controls; NSAIDs, Non-steroidal anti-inflammatory drugs; MoCA, the Montreal Cognitive Assessment; CDR-SOB, Clinical Dementia Rating-Sum of Boxes; ApoE4, Apolipoprotein E4.

^*One subject missing ApoE status.

^a${\chi }^2$.

^bANOVA.

Changes in perfusion parameters

The SE-based perfusion weighted MRI images (Fig. 1) showed an increase in MTT and a decrease in P_tO₂ over the two-year follow-up period throughout most of the cortex in the pAD-MCI group. In addition, the pAD-MCI group showed an increase in CTH in temporal and frontal areas and a decrease in CBF in frontal and parietal areas. No significant changes were observed in CBV (Fig. B1 in Appendix B). The SNAP-MCI group showed a decrease in SE-based CBF in parietal and frontal areas and cingulate cortex. No other SE-based perfusion parameters showed significant changes. The HC group displayed no changes in SE-based perfusion parameters (Fig. B1 in Appendix B).

Fig. 1.

Changes in spin echo perfusion. Two-year changes in the spin echo-based perfusion weighted magnetic resonance imaging (MRI) of 11 prodromal Alzheimer’s disease subjects (pAD-MCI) and 12 suspected non-Alzheimer’s pathophysiology subjects (SNAP-MCI). Statistical t-value maps were adjusted for sex and ApoE4 status. Positive t-values (red colours) indicate significant increases in the parameter over two years while negative t-values (blue colours) indicate significant decreases in the parameter over two years. Widespread increase in microvascular mean transit time and capillary transit time heterogeneity accompanied by decrease in tissue oxygen tension and smaller areas of decrease in cerebral blood flow in pAD-MCI subjects compared to SNAP-MCI. Statistical maps were family-wise error rate corrected (α = 0.05) using cluster-extent-based thresholding with two levels of primary cluster-defining threshold: p < 0.01 and p < 0.001. Areas surrounded by a white line indicate clusters surviving p < 0.001. Dark grey areas indicate areas outside the MRI field of view. Abbreviations: SE, spin echo; CBF, cerebral blood flow; MTT, mean transit time; CTH, capillary transit time heterogeneity; P_tO₂, tissue oxygen tension. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Results from GE-based perfusion (Fig. 2) showed small areas of significant increase in MTT and CTH along with small areas of decrease in CBF and CBV in the pAD-MCI group. The SNAP-MCI group showed a decrease in GE-based CBF in the right frontal and parietal regions, along with a decrease in CBV in the right superior temporal gyrus (hidden in the lateral fissure). No other perfusion parameters showed significant changes in the SNAP-MCI group. The HC group displayed a focal increase in CBF in the right parietal lobe (Fig. B2 in Appendix B).

Fig. 2.

Changes in gradient echo perfusion. Two-year changes in the gradient echo-based perfusion weighted magnetic resonance imaging (MRI) of 11 prodromal Alzheimer’s disease subjects (pAD-MCI) and 12 suspected non-Alzheimer’s pathophysiology subjects (SNAP-MCI). Statistical t-value maps were adjusted for sex and ApoE4 status. Positive t-values (red colours) indicate significant increases in the parameter over two years while negative t-values (blue colours) indicate significant decreases in the parameter over two years. Focal increase in mean transit time and transit-time heterogeneity is observed in the pAD-MCI group. Further, the pAD-MCI group show a decrease in cerebral blood volume in the left precuneus while the SNAP-MCI group show a decrease in cerebral blood volume in the right superior temporal gyrus (hidden in the lateral fissure). The pAD-MCI and SNAP-MCI groups both show small areas of decrease cerebral blood flow. Statistical maps were family-wise error rate corrected (α = 0.05) using cluster-extent-based thresholding with two levels of primary cluster-defining threshold: p < 0.01 and p < 0.001. Areas surrounded by a white line indicate clusters surviving p < 0.001. Dark grey areas indicate areas outside the MRI field of view. Abbreviations: GE, gradient echo; CBF, cerebral blood flow; MTT, mean transit time; CTH, capillary transit time heterogeneity; CBV, cerebral blood volume. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Change in amyloid load

A significant increase of the mean ¹¹C-PiB SUVR within the composite ROI was observed in both the pAD-MCI group (p = 0.006) and the SNAP-MCI group (p = 0.009). However, as displayed in Table 2, the mean increase in the pAD-MCI group was three times larger than in the SNAP-MCI group (0.15 compared to 0.05), and increasing from a markedly higher baseline level (2.12 compared to 1.23).

Table 2.

Change in amyloid load.

Group	Baseline mean 11C-PiB SUVR	Follow-up mean 11C-PiB SUVR	Difference	P-value
pAD-MCI	2.12 [1.85;2.38]	2.28 [2.01;2.55]	0.15 [0.06;0.26]	0.006
SNAP-MCI	1.23 [1.18;1.29]	1.29 [1.22;1.35]	0.05 [0.02;0.08]	0.009

Mean ¹¹C-PiB SUVR uptake within a composite ROI, consisting of cortical areas associated with increased amyloid load in Alzheimer’s disease. Results are presented with a 95 % confidence interval. P-values are calculated using a two-sided paired sample t-test. Abbreviations: pAD, prodromal Alzheimer’s disease; SNAP, suspected non-Alzheimer’s pathophysiology; MCI, mild cognitive impairment; PiB, Pittsburgh Compound B; SUVR, standard uptake value ratio.

Discussion

In this study, we have investigated the temporal and spatial development of cerebral perfusion changes in subjects with pAD-MCI compared with MCI subjects with SNAP and healthy elderly controls. The main finding is that pAD-MCI subjects show worsening of microvascular perfusion over two years measured with DSC-MRI sensitive to capillary-sized vessels compared to SNAP-MCI and HC subjects.

Previously, we reported that pAD-MCI is associated with widespread cortical microvascular flow disturbances compared to control subjects in the same study population [31]. Results from the current study, extend these findings and provide evidence of progressive disturbances in microvascular blood flow in pAD. Considering that the changes in microvascular transit times were only observed in the pAD-MCI group, suggests that these changes are specific to the AD pathology and not a common pathological feature of MCI or due to normal ageing.

Changes in brain microvasculature are increasingly acknowledged as an intimate part of AD pathophysiology alongside the well-known Aβ proteinopathy. According to the two-hit vascular hypothesis of AD [51], genetic, vascular, and environmental risk factors can lead to neuronal dysfunction through both Aβ-independent pathways (hit 1) and Aβ-dependent pathways (hit 2). The two pathways may independently and/or synergistically lead to neuronal dysfunction and neurodegeneration. Hit 1 comprises vascular injury that leads to BBB breakdown, constriction of arterioles and capillaries, and hypoperfusion. This can directly cause neuronal damage through toxic accumulates and insufficient oxygen delivery. Furthermore, reduction in capillary blood flow can cause focal hypoxia or ischaemia, which has been shown to further contribute to the build-up of Aβ [25], [45], [50]. Hit 2 comprises increased production and reduced clearance of Aβ resulting in Aβ accumulation in the brain. Insufficient CBF and elevated Aβ can each lead to hyperphosphorylation of tau and increased neuroinflammation, resulting in neuronal dysfunction [23]. While the two-hit vascular hypothesis of AD posits that hypoperfusion and reduced oxygen delivery is the primary vascular insult in AD, the capillary dysfunction hypothesis of AD suggests that diminished oxygen extraction caused by dysregulation of blood flow through the capillary bed is the initial pathological event [35]. This is based on the extended flow-diffusion equation, describing the oxygen extraction of capillaries [21]. The model recognises the importance of the distribution of blood flow through the capillaries for optimal oxygen extraction. The term CTH refers to the level of heterogeneity of blood flow through the capillary bed. In the normal resting state, CTH is high, introducing ”shunts” of oxygenated blood through the tissue, thereby limiting the oxygen extraction. During functional brain activation, reduction of CTH, i.e. homogenisation of the capillary blood flow, increases oxygen extraction efficacy without increasing macroscopic CBF [35]. In conditions associated with capillary dysfunction, the resting CTH gradually increases and the ability of the microvasculature to homogenise the blood flow (reduce CTH), during episodes of functional hyperaemia or hypoxia is limited. This dysregulation of the capillaries may happen due to constriction, blockage or other morphological insults to segments of the capillaries, similar to the risk factors leading to hit 1 of the two-hit vascular hypothesis [51]. The extended flow-diffusion model predicts that initially, the increased level of CTH can be compensated by a parallel increase in macroscopic CBF. However, as CTH continues to rise, CBF must be attenuated to prevent excessive shunting of oxygenated blood and thereby maintain sufficient oxygen extraction. At this stage, the oxygen extraction becomes increasingly dependent on the oxygen concentration gradient resulting from lower P_tO₂. Even though a reduction in P_tO₂ induces higher oxygen extraction, the resulting focal hypoxia also activates multiple pathological pathways, including Aβ accumulation, inflammation and BBB breakdown. As a direct consequence of increased CTH, capillary dysfunction is a source of much more severe microscopic tissue hypoxia than one might infer from the parallel macroscopic CBF changes [34], [35].

The changes we observed in the pAD-MCI group during this study fits well with the capillary dysfunction hypothesis. The cortical decrease in CBF is not as widespread as the changes in CTH, MTT and P_tO₂. No changes were observed in microvascular CBV, implying that the capillary density remained unchanged during the two-year follow-up period. Finally, the fact that no changes in transit times were observed in perfusion MRI sensitive to contrast agent in vessels of all sizes, implies that the vascular changes affecting the transit time of blood are due to capillary changes.

In essence, the capillary dysfunction hypothesis differs from the two-hit hypothesis by proposing that capillary flow disturbances and impeded oxygen extraction, rather than hypoperfusion and reduced oxygen delivery, represent the initial vascular contributions to AD. In fact, the extended flow-diffusion model predicts that a reduction in CBF may act as a compensatory mechanism to optimise the time available for oxygen exchange in the capillaries, thereby increasing the oxygen extraction [21], [35].

Thus far, it remains unclear whether AD-related capillary dysfunction is secondary to toxic Aβ and tau aggregation or is a primary event. Evidence suggests that BBB breakdown and degeneration of capillary pericytes play an important role in both capillary dysfunction and Aβ accumulation [18], [22], [40], [51]. Breakdown of the BBB has been shown to occur early in the development of AD and other neurodegenerative diseases [48]. Additionally, in both older ApoE4 carriers and individuals with early cognitive impairment, BBB dysfunction and capillary damage has been observed in the hippocampus and medial temporal lobe independent of Aβ or tau pathology [28], [30]. Meanwhile, increased Aβ load has been found to correlate with pericyte reduction in the hippocampus of AD patients compared to controls [43]. Pericytes maintain BBB integrity and play a key role in the clearance of Aβ from the brain [46]. Furthermore, oligomers of Aβ have been shown to cause pericyte contraction and subsequent capillary constriction in both human and mouse tissue [32]. In the current study, the pAD-MCI group showed an increase in cortical Aβ load over the two years, indicating that the capillary changes are at least occurring in parallel with Aβ accumulation in the early stages of AD. In summary, capillary dysfunction, BBB dysfunction and Aβ aggregation appear to be closely connected and may act to reinforce each other. However, the initiating cause and timing of each pathology remains to be determined.

Besides Aβ deposition, circulating leukocytes have been shown to disturb the capillary flow in AD mouse models by plugging a distinct part of the capillary [10]. We have previously shown an increased level of neuroinflammation in pAD-MCI subjects compared with HC subjects [36]. As neuroinflammation involves both microglia and the recruitment of circulating leukocytes [42], this suggests that an increased level of leukocytes together with capillary constriction could cause capillary blockage. Blockage of some capillaries would cause increased flow heterogeneity through the capillary bed, contributing to the capillary dysfunction. Thus, the role of neuroinflammation in AD-related capillary dysfunction require further investigation.

It is well-established that reduced CBF is a feature of AD [2], [24], [41]. Furthermore, progressive decrease in CBF is evident in both the MCI and dementia stages, primarily in parietal, cingulate and temporal regions [1], [6], [19]. Similar patterns of decreased capillary CBF were observed in the current study in both pAD-MCI and, to a lesser extent, SNAP-MCI subjects. The cognitive impairment in the SNAP-MCI subjects is of unknown aetiology and not necessarily due to a neurodegenerative disease. Studies suggest that decrease in regional CBF is associated with cognitive decline in normal ageing without evidence of dementia [5], [8], [47]. The decrease in CBF observed in the SNAP-MCI group was not seen in the healthy control group and may reflect underlying vascular pathology, α-synuclein aggregation, argyrophilic grain disease, TDP-43 proteinopathy or hippocampal sclerosis. A longer follow-up could reveal whether some SNAP-MCI cases will develop AD dementia. However, the fact that no changes were seen in any other perfusion parameters suggest that the observed decrease in CBF might be due to non-degenerative causes such as reduced metabolic demand associated with an affective disorder.

The proportion of ApoE4 carriers in the HC group was surprisingly large (5 out of 10). This is by chance and we found no apparent explanation for this. Even though ApoE4 is the main commonly found genetic risk factor for AD, the development of AD is most likely a result of multiple accumulating risk factors [3], [38]. One might speculate that these older ApoE4 carriers without evidence of cognitive decline, have fewer other AD risk factors, however, we do not have the data to confirm this. In future research of ApoE4 carriers, it would be interesting to study differences in other AD risk factors between individuals who develop AD and those that remain cognitively intact despite their ApoE4 genotype.

The current study has some limitations. The primary limitation is the relatively low number of subjects limiting the statistical power of the study. Consequently, the findings in this study must be validated in larger cohorts preferably with additional follow-up scans to be able to follow individual disease trajectories more closely. Additionally, examination of the BBB was not part of the experimental protocol so we were not able to investigate the association between cortical microvascular disturbances and BBB dysfunction. To achieve a sufficient temporal and spatial resolution, the FOV of perfusion MRI was limited and did not include the cerebellum, inferior parts of the temporal and occipital lobes, and superior parts of the frontal and parietal lobes. Even without whole-brain coverage, the voxel size remained relatively large and subject to partial volume effects. Consequently, calculation of vessel size index based on the combination of SE and GE perfusion images proved to be infeasible. Furthermore, investigation of perfusion in subcortical GM regions was not feasible due to pulsation and susceptibility imaging artefacts in the DSC-MRI data.

In summary, capillary dysfunction, defined as increasing heterogeneity of blood flow and prolonging of blood transit times through the capillaries, can be detected in the brains of prodromal AD individuals. By utilising a biophysical model of normal oxygen consumption, we also found a significant decrease in tissue oxygenation over two years in these individuals. The results are in agreement with the capillary dysfunction hypothesis of AD, which proposes that a drop in tissue oxygen tension is necessary for sufficient oxygen extraction as a response to increasing heterogeneous blood flow. The importance of capillary flow homogenisation could be an overlooked property that might explain hypoperfusion as a compensatory mechanism to maintain sufficient oxygen extraction rather than the primary cause of vascular injury. Our results show that capillary dysfunction is present at the MCI stage of AD where the cortical Aβ load is already elevated. Future studies need to investigate whether capillary dysfunction can be detected prior to cognitive impairment or even before evidence of cortical Aβ build-up in at-risk subjects. Such studies might include elderly cognitively intact ApoE4 carriers at increased risk of AD. Additionally, studies of the underlying mechanism of CTH are needed to determine whether increased CTH is indeed the primary pathological insult in AD and possibly how it can be treated or prevented.

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.

Appendix A. Supplementary data

The following are the Supplementary data to this article: