Cerebral microvascular changes in healthy carriers of the APOE-ɛ4 Alzheimer's disease risk gene

Abstract

APOE-ɛ4 is a genetic risk factor for Alzheimer's disease (AD). AD is associated with reduced cerebral blood flow (CBF) and with microvascular changes that limit the transport of oxygen from blood into brain tissue: reduced microvascular cerebral blood volume and high relative transit time heterogeneity (RTH). Healthy APOE-ɛ4 carriers reveal brain regions with elevated CBF compared with carriers of the common ɛ3 allele. Such asymptomatic hyperemia may reflect microvascular dysfunction: a vascular disease entity characterized by suboptimal tissue oxygen uptake, rather than limited blood flow per se. Here, we used perfusion MRI to show that elevated regional CBF is accompanied by reduced capillary blood volume in healthy APOE-ɛ4 carriers (carriers) aged 30–70 years compared with similarly aged APOE-ɛ3 carriers (noncarriers). Younger carriers have elevated hippocampal RTH and more extreme RTH values throughout both white matter (WM) and cortical gray matter (GM) compared with noncarriers. Older carriers have reduced WM CBF and more extreme GM RTH values than noncarriers. Across all groups, lower WM and hippocampal RTH correlate with higher educational attainment, which is associated with lower AD risk. Three days of dietary nitrate supplementation increased carriers' WM CBF but caused older carriers to score worse on two of six aggregate neuropsychological scores. The intervention improved late recall in younger carriers and in noncarriers. The APOE-ɛ4 gene is associated with microvascular changes that may impair tissue oxygen extraction. We speculate that vascular risk factor control is particularly important for APOE-ɛ4 carriers' healthy aging.

Introduction

The ɛ4 allele of the APOE gene is associated with several-fold elevated risk of Alzheimer's disease (AD), so while only about 15% of the general population carry one or two versions of the ɛ4 allele, it is found in roughly 50% of patients with AD (1, 2). Patients with AD typically reveal reduced resting cerebral blood flow (CBF) (3) and impaired CBF responses to neuronal activation (4), but asymptomatic APOE-ɛ4 carriers reveal regions of elevated CBF during rest (2, 5–9) and higher task-related CBF responses than people with two versions of the common ɛ3 allele (10–12), independent of task performance (5, 13, 14), and dementia risk (15, 16).

Keeping in mind that brain tissue's oxygen uptake is the product of CBF and the oxygen extraction fraction (OEF), healthy APOE-ɛ4 carriers' elevated CBF could, in principle, compensate for age- and risk factor-related changes that limit oxygen extraction (17–19).

Biophysically, the extraction of oxygen from a single capillary is limited by blood's capillary transit time: the time available for blood–tissue oxygen diffusion exchange before blood returns to the heart (20) (see Fig. 1A). At the macroscopic level, blood's mean capillary transit time (MTT) can be calculated as the ratio between local capillary blood volume (CBV, milliliter blood per milliliter tissue) and CBF (milliliter blood per milliliter tissue per minute), respectively (21). In normal brain tissue, CBV and CBF are both proportional to local metabolic demands during rest (22) and thus provide uniform MTT values across the brain, facilitating the extraction of oxygen from blood. During brain activation, however, CBF increases without proportional increases in CBV (23).

Figure 1A and B illustrates how capillary blood flows homogenize as CBF increases during brain activation (24–26) and thereby facilitate oxygen uptake. The accompanying reduction of capillary transit time heterogeneity (CTH) can indeed account for the brain's efficient oxygen extraction when metabolic demands and CBF increase, although CBV shows little change (27, 28). In fact, CTH—quantified as the SD of individual capillary transit times—decreases in proportion to their mean (MTT) in passive, compliant microvascular networks (27), providing a passive mechanism by which OEF remains high although MTT decreases during functional activation. The CTH/MTT ratio is termed the relative transit time heterogeneity (RTH) below.

Figure 1C illustrates how irreversible age- or risk factor-related capillary changes, which affect the resistance to blood flow through individual capillary segments, are expected to disturb the (re)distribution of capillary blood flows during rest (cf. Fig. 1A) and the homogenization of capillary transit times as CBF increases during brain activation (Fig. 1B). Note that while such changes limit blood flow through some capillary paths, they shorten blood's transit times through others, impairing oxygen extraction (20, 28, 29).

Figure 1D illustrates how a gradual increase in CTH, caused by the accumulation of such capillary changes over time, is predicted to reduce tissue oxygen levels if oxygen utilization and CBF remain constant. Figure 1E illustrates how increasing CBF is predicted to mitigate such tissue oxygen level reductions—until CTH reaches thresholds, where the benefit of higher blood supply, as a means to meet metabolic demands during rest and functional activation at physiological tissue oxygen levels, is outcompeted by its transit time-shortening effects (19, 20, 28, 29). Instead, subnormal CBF values yield longer MTT and permit more efficient oxygen extraction (higher OEF)—albeit at the expense of dwindling tissue oxygen levels (19). In other words, biophysical models predict that elevated resting CBF may compensate for (i.e. render asymptomatic) mild microvascular changes, whereas attenuation of CBF, paradoxically, is necessary to meet metabolic demands for more severe microvascular changes.

It is unknown whether the APOE-ɛ4 gene is associated with microvascular changes in the form of either reduced CBV or elevated RTH, and if so, whether this phenomenon is accompanied by altered CBF.

Dynamic susceptibility contrast (DSC) MRI (30) sensitized to capillary-sized vessels (31, 32) allows estimation of microvascular hemodynamics in humans (30). Cross-sectional (33–36) and longitudinal (37, 38) studies of healthy subjects at high risk of AD, subjects with mild cognitive impairment, and patients with AD, respectively, link the disease to loss of cortical microvessels (reduced CBV), decreasing CBF, and increasing CTH over time, particularly in cortical regions with amyloid-β (Aβ) aggregation (34–36), a pathological hallmark of AD that antedates memory symptoms by a decade or more (39). Aβ oligomers constrict human capillaries in AD by interfering with contractile capillary pericytes (40) and may thereby contribute to microvascular flow disturbances prior to any memory symptoms. Conversely, microvascular flow disturbances are believed to reduce the normal clearance of Aβ protein from brain tissue (41) and stimulate Aβ aggregation (42). To address APOE-ɛ4-related microvascular changes, subjects should therefore be examined prior to when Aβ aggregates begin to form in their brain tissue; thus, 15% of asymptomatic APOE-ɛ4 homozygotes are estimated to be Aβ-positive at age 40 years, while the corresponding age is 55 years for heterozygotes with the common APOE-ɛ3 allele (43, 44).

This study compared microvascular hemodynamics in healthy APOE-ɛ4 and APOE-ɛ3 carriers aged 30–70 years for signs of reduced CBV, elevated RTH, and accompanying CBF changes. To address whether microvascular changes are reversible and affect brain functioning, we quantified these parameters before and after 3 days of vasodilator administration: dietary nitrate equivalent to a large, daily salad serving.

Results

Subjects with at least one APOE-ɛ4 allele (n = 38; 29 ɛ4/ɛ3 heterozygotes and 9 ɛ4/ɛ4 homozygotes) are referred to as “carriers” below and APOE-ɛ3 homozygotes (n = 38) as “noncarriers.” See Fig. S1 for study groups and Table S1 for participant demographics.

Participants' CBF, CBV, and RTH were determined by spin echo DSC MRI. To the extent permitted by the spin echo MRI sequence's brain coverage, these parameters were first determined for three predefined regions of interest (ROIs): gray matter (GM), white matter (WM), and hippocampus (Fig. S2 B–D), and for cortical anatomical subregions for comparison with reports of elevated resting CBF in healthy APOE-ɛ4 carriers by others. In the analyses of GM, WM, and hippocampus, subjects were dichotomized according to mean age so that “younger” and “older” refer to subjects aged ≤ 52 and > 52 years, respectively. In the analyses of cortical subregions, age was used as a continuous variable for comparison.

APOE-ɛ4 is associated with elevated CBF but reduced CBV across cortical subregions

Carriers' average resting CBF values were higher in 11 cortical subregions compared with noncarriers (P < 0.01), all of which overlapped with anatomical regions where this phenomenon was reported in the past (2, 5–7, 9) (see Fig. S3). Carriers' CBV was lower than in noncarriers in 11 cortical subregions (P < 0.01; Fig. S4). Five of these subregions were among those displaying elevated CBF (Fig. S3). The APOE-ɛ4 gene is thus associated with microvascular changes in the form of reduced CBV and accompanied by either preserved or elevated CBF.

Partial volume effects may interfere with CBF and microvascular density measurements, typically by underestimating values obtained in regions where the cortex is thin. Participants' cortical thickness correlated negatively with age (P < 0.01) across subregions, but APOE genotype did not affect cortical thickness, nor interfere with this correlation (Fig. S5).

APOE-ɛ4 is associated with widespread microvascular changes in all carriers and low WM CBF in older carriers

In the subregion analysis, CBV was reduced in the hippocampus of APOE-ɛ4 carriers compared with noncarriers (Fig. S4), while their average RTH was elevated (Table 1 and Fig. S7). This phenomenon was not accompanied by changes in hippocampal CBF (Fig. S3). See Figs. S6 and S7 for all age- and gene-specific ROI comparisons of RTH.

For WM and GM ROIs, average RTH values were identical across age and carrier groups, but compared with noncarriers, the spatial variability of RTH values was elevated across all tissue types in young carriers and in the GM of older carriers, implying that the corresponding ROIs contain voxels with more extreme RTH values (see Table 2). In the subregion analysis (above), we noted a tendency for some anatomical GM regions to yield higher RTH values in carriers than in noncarriers (0.01 < P < 0.05; Fig. S7) and no overlap with regions showing elevated CBF (Fig. S3). The WM ROI showed lower CBF in older carriers (n = 23) compared with older noncarriers (n = 20) (Table 1) but no significant differences in average RTH (Table 1) or RTH spatial variability (see Table 2).

We characterized the increased spatial variability of RTH values further by means of probability density functions that describe the distribution of RTH values within subjects' WM (Fig. 2A) and GM (Fig. 2B) ROIs, respectively, in the two APOE and age groups. Note that young noncarriers display the narrowest distribution of RTH values in both WM and GM. Figure 2C and D shows the gene-related difference between these distributions for younger and older subjects, respectively. Note how these differences mostly relate to carriers having more tissue with higher RTH values. The black lines highlight RTH ranges in which these differences are significant as determined by curve χ² differences. Figure 2E illustrates the location of voxels with RTH values within two such ranges (interval 3: 1.05 < RTH < 1.23 and interval 4: 1.31 < RTH < 1.41) for WM and GM cross younger noncarriers (left) and carriers (right). Thus, warmer colors indicate more subjects having RTH values within this range at the corresponding brain location. Keeping in mind that, biophysically, higher RTH is associated with poorer oxygen extraction, deep WM appears most severely affected by microvascular flow disturbances in younger APOE-ɛ4 carriers. See Fig. S8 for the spatial distribution of values in other RTH ranges that displayed significant carrier–noncarrier differences.

White matter hyperintensities (WMHs) are radiological hallmarks of cerebral small vessel disease (cSVD) (45), and such lesions are known to affect RTH and CBF relative to surrounding normal-appearing WM in older subjects (46). Studies show WMHs to be more frequent in APOE-ɛ4 carriers than noncarriers (47, 48), and we therefore examined the subjects with T₂-weighted images (n = 59) to determine WMH number and Fazekas scores (49). This analysis revealed similar WMH numbers and cSVD severity in carriers and noncarriers and low WMH load in younger subjects (see Fig. S9A and B). Group differences in RTH and CBF could therefore not be attributed to differences in WMH load.

High educational attainment is associated with low RTH

Across age and carrier groups, more years of schooling was associated with lower RTH in the hippocampus, WM, and several cortical gray matter regions (see Figs. S6 and S7). This finding is notable in that more years of schooling is independently associated with lower AD risk (50, 51).

Partial volume effects may interfere with regional hemodynamic parameter estimates, and education attainment-related differences in tissue volumes could therefore bias RTH estimates toward these findings. However, we found no relation between educational attainment and regional cortical thickness (Fig. S5).

Carriers and noncarriers show opposite CBF responses to nitrate supplementation

Nitric oxide (NO) is a mediator of vasodilation in brain arterioles and capillaries (52), and NO depletion within the vascular wall contributes to neurovascular dysfunction: the attenuated vasodilator responses observed in aging, hypertension, amyloidosis, and APOE-ɛ4 genotype (53–56). Oxygen availability limits NO synthesis from L-arginine via endothelial NO synthase, and hence, the vasodilation needed to restore oxygen levels, under hypoxic conditions (52). Mammals, however, can also produce NO from dietary nitrate via reactions that are particularly efficient at low oxygen levels (57, 58). Nitrate supplementation has proven safe in humans (59).

To examine whether modulation of perivascular NO levels affects CBF, CBV, and RTH across age and carrier groups, we subjected participants to 3 days of nitrate supplementation, equivalent to one daily, large salad serving, a dose previously shown to improve mitochondrial function in human muscle (60). Accordingly, subjects were given 0.1 mmol NaNO₃ sodium nitrate per kilogram bodyweight per day for 3 days in a randomized, double-blinded, placebo-controlled (NaCl ingestion) cross-over design (see Fig. S2A).

The intervention increased plasma nitrite and nitrate levels while lowering systolic blood pressure as expected (Tables S2 and S3) (61) and caused WM CBF to increase in carriers, whereas WM CBF tended to decrease in noncarriers (see Table 1 and Fig. S10). Whole-brain statistical analysis showed significant increases in carrier CBF across occipital, temporal, and frontal cortices (Fig. S2E).

Nitrate supplementation modifies blood's microvascular distribution

Regional CBV values increased in response to nitrate supplementation (Fig. S13), whereas mean RTH remained unaffected (Fig. S12).

Figure S8 shows the WM and GM distributions of RTH values before (panels A and C) and after (panels E and G) nitrate ingestion across age and APOE groups. Note how these distributions are more similar after the intervention for GM (panel G), while the WM RTH distribution in young carriers seems to include more extreme RTH values after nitrate supplementation (panels E and F). Comparing the χ² differences between the nitrate and control session (Fig. S11), we notice that nitrate ingestion mostly affects the distributions of RTH values in carriers.

Nitrate supplementation and neurocognitive functions: paradox responses in older carriers

All subjects underwent a battery of six neuropsychological tests after the nitrate and placebo (NaCl) conditions, respectively, immediately before and after their MRI examinations. The resulting 24 scores were subjected to principal component analysis (PCA) to reduce score interdependency across test batteries, resulting in six largely distinct clusters (see Tables S4 and S5). The neuropsychological tests did not reveal major changes in subjects' performance after nitrate supplementation, with notable exceptions: Nitrate supplementation improved the ability to recall and redraw complex figures after 3 and 30 min in the Rey–Osterrieth complex figure (ROCF) test (Table S4, cluster 6) (62) in all participants except the older carriers, who performed worse (Table S5). Also, nitrate supplementation significantly prolonged the time taken by older carriers to complete the trail making tests A and B (63) compared with the placebo condition (Table S5).

Discussion

The first main finding of this study is that APOE-ɛ4 carriers reveal microvascular changes that are thought to limit the transport of oxygen from blood into brain tissue: reduced CBV and high RTH. Thus, carriers had cortical subregions with reduced CBV (Fig. S13) and elevated RTH in gray matter as a whole (Table 2). Consistent with earlier reports (2, 5–9), carriers revealed 12 cortical regions with elevated resting CBF, five of which corresponded to regions with low CBV, consistent with the notion that CBF may be increased to compensate for the resulting shortening of capillary transit times. Our findings are consistent with earlier findings of reduced microvascular density in the neocortex, corpus callosum (55) and hippocampus (64, 65) of mice carrying the human APOE-ɛ4 gene.

Younger carriers, who are not yet suspected of amyloid pathology (43, 44), reveal microvascular flow disturbances in the form of elevated RTH values in their hippocampi (Table 1) and widespread, abnormally high RTH values in image voxels across their GM (Table 2 and Fig. 2D) and WM (Table 2 and Fig. 2C). These observations suggest that the microvascular changes are characteristic of the APOE-ɛ4 gene, rather than results of early build-up of Aβ protein, which acts as a capillary constrictor (40). While amyloid pathology generally spares the hippocampus in AD, the hippocampal microcirculation is thought to be particularly vulnerable (66).

The second main finding is that WM CBF is reduced in healthy, older APOE-ɛ4 carriers. The dysregulation of microvascular transit times observed in younger carriers' WM (Table 2 and Fig. 1C and E) could herald severe microvascular flow disturbances in later life, suggesting that older carriers' lower WM CBF, paradoxically, may counteract the biophysical effects of such changes (Fig. 1). The defining feature of compensatory reductions of CBF to counteract poor oxygen extraction, as opposed to a primary reduction in CBF due to neurovascular dysfunction, is that “normalization” of CBF is expected to reduce tissue oxygenation in the former. Older carriers' resting WM CBF was indeed lower than that of noncarriers, but their neuropsychological scores deteriorated although their WM CBF increased after nitrate supplementation. This is consistent with a paradox reduction of oxygen extraction that negatively affects neuronal network activity—contrary to the current understanding of the relation between blood flow and tissue oxygenation and warrants further study. Nonetheless, if microvascular dysfunction is accompanied by tissue oxygen depletion, it could contribute to the WM vulnerability associated with the APOE-ɛ4 allele (55), including the propensity of carriers to develop WMHs (47, 48).

Previous studies show evidence of progressive and severe, cortical microvascular flow disturbances in patients with AD (33, 38), and recent studies of patients with mild cognitive impairment suggest a transition from mild to severe microvascular flow disturbances around the time that subjective memory symptoms appear, linked in part to the vasoconstrictor properties of amyloid protein (34, 35, 37). The current study suggests that APOE carrier status and WM microvascular function may be important to our understanding of the earliest stages of AD, and to future, targeted prevention strategies.

In terms of risk factors, the study unexpectedly found that RTH in the hippocampus, WM, and several cortical gray matter regions was inversely correlated with participant education years, irrespective of carrier and age group, possibly offering a physiological contributor to the puzzling relation between high educational attainment and low AD risk. Learning introduces profound, adaptive changes in both WM and GM cytoarchitecture (67, 68), paralleled by dynamic adaptations of the microcirculation to meet the associated energy costs, particularly of maintaining synapses (69). Such adaptations not only include angiogenesis and higher capillary density in cortical layers with high synaptic density (22), but also changes in capillary bed topology that permit more homogenous capillary blood flow and higher OEF in those cortical layers, relative to neighboring layers that share the same blood supply and tissue oxygen tension (70). We speculate that learning and memory-induced microvascular adaptations include such topological optimizations in relation to the cellular populations involved. Widespread and continuing microvascular adaptation to higher oxygen extraction yields (lower RTH), in turn, might offer resilience to the accumulation of age- and risk factor-related capillary damage (increasing RTH), which seemingly plays a role in the development of AD.

Effects of nitrate supplementation on vascular function and cognitive performance

The effects of dietary nitrate supplementation on neuropsychological functioning and on cerebral hemodynamics and blood oxygenation have been examined in several studies, yielding inconclusive results (59). These studies included relatively few healthy participants (10–40) in different age groups (the mean age ranging from 20 to ≥ 70) and a wide range of nitrate doses (5–24.2 mmol/day) and intervention durations (90 min after a single dose—10 weeks). Our results suggest that age and APOE-ɛ4 genotype, combined, affect the direction of changes in regional CBF and neurocognitive function after nitrate supplementation.

It should be kept in mind that neurovascular dysfunction is characterized not only by reduced NO bioavailability, but also oxidative injury to the microcirculation. The potential benefits of dietary nitrate supplementation may therefore be greatest before neurovascular dysfunction becomes apparent, at which time microvascular changes may still be reversible. Dietary supplementation should therefore be tested in well-powered trials controlled for age, APOE-ɛ4 genotype, hypertension, and other risk factors that might interact with capillary function.

Limitations to the study

This study measured parameters that, biophysically, limit the transport of oxygen from blood into brain tissue (reduced CBV and high RTH), rather than oxygen extraction per se. Direct measurements of OEF in cortical layers with high metabolic demands confirm that more homogenous (resting) capillary blood flows allow tissue to extract a higher proportion of bloods oxygen content without lowering tissue oxygen tension (70)—consistent with the biophysical importance of capillary transit times (20, 28). The dependence of OEF upon blood's microvascular distribution in human brain, as measured by gold-standard positron emission tomography OEF and the MRI method applied in this study, respectively, was confirmed in patients with large-vessel disease (71) but should be verified in larger studies.

The impact of bloods microvascular distribution on brain oxygenation was only recognized recently, and assessment of this hemodynamic property in humans is inherently difficult. Characterizing blood's capillary transit times based on the passage of exogenous contrast media (CM) passages through MRI voxels that contain up to 200,000 capillary segments totaling ∼12 m in length (72) therefore rely on a number of assumptions that are discussed in (30). Notably, relative flow heterogeneity (analog to RTH) is believed to be fractal across the vascular tree, that is, apply to the distribution of blood flow from the level of capillary ensembles to entire organs (73), supporting the interpretation of voxel-wise, or even organ-wide, transit time distributions in terms micrometer-scale hemodynamics. Indeed, RTH estimates based on the indicator dilution method used in this study have yielded similar and consistent results when applied to patients with AD and controls using DSC MRI (33) and to AD models and controls using two-photon laser microscopy at the level of individual microvessels (74).

Our study sample was relatively small, which may have affected our ability to detect microvascular flow disturbances in WM and hippocampi further. Future studies should also examine a sufficient number of APOE-ɛ4/ɛ4 homozygotes to examine whether the extent of capillary flow disturbances displays a gene dose–response similar to their AD risk.

Hippocampi were the brain structures most affected by microvascular flow disturbances in young carriers. Our DSC MRI methods are somewhat sensitive to brain motion and susceptibility artifacts in this brain region. Also, due to the oblong shape and limited volume of the hippocampi, image voxels in this region tend to contain contributions from surrounding tissue and fluids—so-called partial volume effects. Therefore, far fewer DSC voxels are classified as belonging to the hippocampi than are voxels within WM and GM, respectively. We speculate that this limited our sensitivity to detect CBF and RTH changes in these critical brain structures.

Materials and methods

Ethical approval and informed consent

The study was approved by the Ethics Committee of the Central Denmark Region (Project-ID: M-2013-239-13) and conducted in accordance with the Helsinki Declaration. All participants gave written informed consent prior to their participation.

Participant recruitment

Initially, 283 healthy participants between 30 and 70 years of age were recruited from the public through advertisements in a local newspaper. All participants were required to (i) have Danish as their main language, (ii) have no diagnosed neuropsychiatric disorders, nor use prescription drugs, (iii) have no prostheses, pacemakers, etc., that could interfere with the MR scans, (iv) not suffer from claustrophobia, (v) not be obese, (vi) not suffer from anxiety related to blood or needles, (vii) have normal kidney function, (viii) not be pregnant, (ix) not have a history of frequent or major allergic reactions, and (x) have no gross anatomical abnormalities on brain MRI.

All participants underwent genetic screening to establish their APOE allele status. Out of the 283 participants, we included 52 carriers of least one APOE-ɛ4 allele (14 ɛ4/ɛ4 and 38 ɛ3/ɛ4) and 48 carriers of two APOE-ɛ3 alleles. The two groups were balanced in terms of gender and distribution of age across the four decades from 30 to 70 years of age. The mean age of 52 years instead of 50 years was caused by late participant cancelations and older participants more readily accepting to show up.

Participant dropout

Five participants did not complete the study. Reasons included participants not showing up to be scanned and participants feeling uncomfortable. The dropouts are described in Fig. S1. Of the remaining 95 participants, four participants were excluded due to incorrect or missing nitrate intake as indicated by their blood nitrate/nitrite samples, and two participants were excluded because of lost samples/lost data. Hence, 89 participants were included in the analyses of cognitive tests, blood pressure, and blood samples. For the MR perfusion analyses, data from 76 participants were included in the analyses. Reasons for excluding patients from MR examinations included insufficient kidney function (estimated glomerular filtration rate [eGFR] < 60 ml/min/1.73 m²) as proper kidney function is a prerequisite for perfusion MRI using gadolinium (Gd)-based CM. Also, participants whose head size exceeded the dimensions of our MR systems head coil were excluded.

Study drug and placebo

The study was a randomized, double-blinded, placebo-controlled cross-over study, during which participants ingested either sodium nitrate (0.1 mmol NaNO3 per kg bodyweight per day, Sigma-Aldrich, MO, USA) or sodium chloride (0.1 mmol NaCl per kg bodyweight per day as a control) for 3 days. This nitrate dose mimics a large serving of nitrate-rich vegetables and resembles that of a study of beetroot juice, CBF, and cognitive performance in normal volunteers (75, 76).

Tests and imaging were carried out on the third day of ingestion, and the drug/placebo studies were separated by a washout period of a minimum of 10 days (Fig. S2A). The saline solutions were mixed with concentrated fruit juice to disguise differences in taste. The saline solutions were prepared by R.A. and their content blinded by a researcher not otherwise involved in the experiment. Solutions were tested for nitrite every day, and no sign of conversion was seen.

Food restrictions

A food regimen was implemented to prevent participants from ingesting vegetables with high levels of nitrate or nitrite (77) and food sources containing high levels of vitamin C, which may augment the conversion rate of nitrite into NO. Accordingly, during the 3-day periods of salt solution ingestion, participants were to abstain from eating green leafy vegetables, brassicas, beet root, pepper fruits, and radish. Participants were also asked to avoid eating strawberries, citrus fruits, rhubarb, nuts, and processed meats.

Blood data

Genotyping

For the initial screening for APOE genotype, venous blood samples were collected into ethylenediaminetetraacetic acid (EDTA) tubes and stored on ice. Centrifugation (2,000g, 10 min, 4°C) and separation of plasma, buffy coat, and red blood cells were completed within an hour, and samples were stored at −80°C until genotyping. APOE genotype was determined at Statens Serum Institut, Copenhagen (MC).

Blood biomarkers

In the main experiment, one blood sample was collected prior to each 3-day salt ingestion period and an additional blood sample was acquired immediately before the MR scanning session was commenced on day 3. Blood samples were analyzed at Aarhus University Hospital's Department of Clinical Biochemistry. Total plasma cholesterol, plasma high-density lipoprotein (HDL) cholesterol, and plasma creatinine levels were measured on a Cobas 6000 C system. Plasma low-density lipoprotein (LDL) cholesterol was calculated as the difference between total plasma cholesterol and HDL cholesterol. The eGFR was calculated from plasma creatinine levels and corrected for age and gender.

Nitrate/nitrite

Before each saline ingestion period and immediately before and after each MR scanning session, blood samples for plasma nitrate and nitrite quantification were collected in low-nitrite EDTA tubes, manufactured for this purpose and provided by the Karolinska Institute, Stockholm, Sweden (EW). Samples were immediately centrifuged (5,000 g, 5 min, 4°C) and after separation into red blood cells, buffy coat, and plasma instantaneously placed on ice and moved to −80°C storage. All measurements of nitrite and nitrate content in the plasma fraction were taken at the Department of Physiology and Pharmacology, Section of Anesthesiology and Intensive Care, Karolinska Institute, Stockholm, Sweden (EW). The nitrate and nitrite levels during MR were calculated as the average before and after the MR scan.

Physiological data

Blood pressure

Blood pressure was measured in the MR scanner using a Medrad Veris MR Vital Signs Monitor (Warrendale, USA). Measurements were taken before and after each perfusion scan. Participants were lying down quietly in the scanner for at least 45 min prior to blood pressure measurements and perfusion scans.

Transcutaneous blood gas measurements

The oxygen saturation (SpO₂) and carbon dioxide tension (PCO₂) in the blood was measured noninvasively and continuously across the skin just below the eye of the participants. This was done using a transcutaneous Sentec V-Sign sensor (SENTEC-AG, Therwil, Switzerland) adapted for use in the MRI environment as described in Fig. S1 of Rasmussen (78). The data were recorded using a Brain Vision BrainAmp ExG MR Amplifier (Gilching, Germany) connected to a PC.

Neuropsychological assessment

Sporadic AD is associated with deteriorating cognitive abilities, including executive functioning, visual spatial functioning, and episodic, working and spatial memory (79, 80). To probe these domains, participants underwent a battery of neuropsychological tests, including (i) Wechsler Adult Intelligence Scale-Fourth Edition digit span (forward/backward and sequencing), vocabulary (in Danish), and block design subtests (81), (ii) ROCF test (copy, recall after 3 and 30 min) (62), (iii) two text recall subtests from Wechsler Memory Scale-III (82), (iv) Ruff light trail learning test I and II (83), (v) Raven's progressive matrices (84), and (vi) trail making test A and B (63). The tests were carried out either before or after the MRI scans by staff supervised and trained by experienced neuropsychologists (SdP).

Magnetic resonance imaging

All participants were imaged in a supine position using a Magnetom Skyra 3.0-T MRI system (Siemens, Erlangen, Germany).

Anatomical T1 MRI

Anatomical images were recorded using a magnetization-prepared two rapid acquisition gradient echoes (MP2RAGE) sequence with 1-mm³ isotropic voxels using time of echo (TE) = 2.98 ms, time of repetition (TR) = 5,000 ms, inversion times 700 and 2,500 ms, acquisition matrix = 240 × 256 × 176, flip angle (FA) = 4° and 5°, echo spacing 7.1 ms, and iPAT factor 2 (GRAPPA).

Perfusion MRI

To measure cerebral perfusion, we performed DSC MRI, using a spin echo sequence, which, for short echo times and acquisition windows, is weighted toward microvessels in the 5–15-μm range (31, 85).

Dynamic spin echo echo planar imaging (EPI) MR images were recorded during the bolus injection of 0.2 mmol/kg per kg bodyweight Gadovist, a Gd-based CM (gadobutrol, Bayer, Denmark), followed by a 30-ml saline flush. All injections were carried out at a speed of 5 mL/s using a Medrad (Warrendale, PA, USA) Spectris Solaris EP MR injection system, controlled by an in-house trigger system, which allows us to link CM injection to a certain point in the dynamic image acquisition. The imaging sequence comprised 300 dynamic acquisitions, 200 of which were acquired prior to CM injection. The imaging parameters were: TE = 60 ms, TR = 1,530 ms, acquisition matrix = 64 × 64 × 19, FA = 90° and 180°, and voxel size 3 × 3 × 3 mm3 with a 1-mm gap between slices. The spin echo EPI images were acquired with an iPAT factor of 3 and an echo spacing of 0.5 ms, resulting in a readout window of (64/3) × 0.5 ms = 10.7 ms.

FLAIR images

The FLAIR images used for quantifying the WMH originated from a diffusion kurtosis imaging (DKI) measurement taken during the second examination of each subject. To avoid cerebrospinal fluid (CSF) contamination in the DKI measurements, we applied a 2,100-ms inversion pulse prior to radiofrequency (RF) excitation. Accordingly, the non-diffusion weighted images had contrast similar to a conventional FLAIR image. Using EPI images for FLAIR contrast has recently been found adequate (86). The sequence parameters were as follows: TE = 101 ms, TR = 24.8 s, acquisition matrix 96 × 96 × 76, FA = 180°, 90°, and 180°, and voxel size 2.3 × 2.3 × 2 mm³ with no gap between slices. The images were acquired with both blip-up and blip-down phase encoding and were subsequently distortion corrected using TOPUP (87) from the FSL software (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/topup).

Data analysis tools

All data were processed in MATLAB (MathWorks, Massachusetts, USA), release R2016b. For MR data analysis, we used SPM12 (release 6685, University College London, UK) and an in-house data processing pipeline for perfusion MRI analysis.

Analysis of blood data and physiological data

Missing data—nitrate/nitrite

Missing preingestion values (n = 4) were assigned the median value of all preingestion values. Missing pre-MRI values and missing post-MRI values (n = 3) were assigned the median value of the subject's group (carrier/noncarrier), time (pre/post), and treatment (NaCl/NaNO3). We chose this approach because nitrate/nitrite levels were used as indicators of correct saline ingestion rather than as an outcome measure.

Missing data—blood pressure

Participants with missing data were excluded from the analyses. The minimal value of the three blood pressure measurements taken during each scan session was used in the statistical analyses.

Alignment—transcutaneous SpO₂ and PCO₂ and perfusion MRI

The recordings were used as continuous covariates in the voxel-wise SPM analysis of perfusion data to control for their influence on cerebral hemodynamics. To obtain a representative value for SpO₂ and PCO₂ during the CM bolus passage used to calculate perfusion indices, we used their mean value starting at the time of CM injection and throughout the subsequent 30 dynamic image volumes (∼45 s).

Analysis of neuropsychological tests

At every experimental session, each subject's cognitive performance was characterized by 24 test scores, each summarizing the subject's performance on subtests taken from the six neuropsychological tests. To disentangle the interdependency of test scores and to group test scores according to separate aspects of neurocognitive function, we applied PCA across the two sessions in all subjects. Using this data-driven approach, we identified the six subtest clusters given in Table S4. The observed clustering was expected based on the nature of the subtests and the neurocognitive domains they probe, except the ROCF copying task scores, which grouped with fluid intelligence scores rather than the remaining ROCF scores. To summarize cognitive performance within each cluster, we applied PCA within each cluster to the mean-centered test scores scaled by their standard deviation from all subjects. PCA thus provides the linear combination of subscores along which within-cluster subscore variance is maximized. For each cluster, we retained the first principal component to use in the statistical analyses. See Table S4 for the percentage of within-cluster subscore variance accounted for by the six PCA-derived composite scores based on cluster subscores.

Perfusion MRI analysis

Preprocessing and concentration curves

Dynamic DSC image volumes were motion corrected to a reference volume defined by the participants' head location immediately before CM injection. The images volumes were slice-time corrected and co-registered to anatomical (T1-weigted) images. The relative CM concentration time curve, C(t), was calculated in each voxel based on its average intensity during the baseline scans prior to bolus passage $St0$ and the resulting signal time curve

where TE is the EPI sequences TE and S(t) the T₂-weighted signal intensity in the voxel at time t.

Arterial input function selection

Perfusion data analysis relies on the choice of an arterial input function (AIF) to represent the shape of the CM concentration time curve in the arterial supply of tissue within image voxels (88). Comparisons of perfusion values among subjects or across sessions in single subjects are therefore sensitive to the fidelity of the AIF, which in turn depends on the choice of arterial voxels (89, 90). AIF selection was thus performed as a two-step to ensure that AIFs were recorded at comparable locations in the two sessions. In the first step, an initial voxel selection was conducted. 500 voxels from each session with the fastest time to peak and the greatest area under the concentration time curve until peak was identified as potential AIF voxels. To do so, we used the algorithm previously described by Mouridsen et al. (91). In the second step, voxels identified as potential AIFs in both sessions were subjected to the same algorithm again, but only allowing six voxels to remain in each session. These six voxels were used to construct a mean AIF curve for each session and used to calculate perfusion maps. See Fig. S14 for a summary image of the AIF voxel positions.

Calculating perfusion maps

According to indicator dilution theory, the concentration CM in any voxel, C(t), can then be seen as a convolution of $Ca(t)$

where R(t) denotes the residue function, that is, the fraction of CM still present in the voxel at time t after an infinitely narrow impulse injection into its feeding artery (88). R(t) can be estimated using a parametric approach based on an anatomical model of the microvasculature (92). The approach approximates the transport function h(t) for CM as it passes through the vasculature by a family of gamma distributions

where

and α and β are shape and scale parameters, respectively, related to our physiological variables as MTT = α·β and CTH = β√α. We further computed the coefficient of variation RTH = CTH/MTT (30).

Finally, we computed relative blood volume within each image voxel as:

and CBF according to the central volume theorem (21) as CBF = CBV/MTT.

MR DSC ROI analyses

Anatomical T₁ MRI data were segmented in MATLAB using SPM12 to obtain individual, nonoverlapping masks of WM, GM, hippocampi, ventricles, and CSF. In each subject and each session, CSF voxels and the 5% voxels with the highest CBV values were excluded from the masks before applying them to the perfusion map. For each ROI, their mean and mean absolute deviation (MAD) were used in the statistical analyses. MAD was preferred as it is a robust measure of variability little influenced by outliers.

MR DSC SPM analyses

For the SPM analyses, all perfusion images were resampled to 1 × 1 × 1 mm³ and segmented into WM, GM, and CSF maps. Each segmented perfusion image was smoothed with a 9 × 9 × 9 mm³ Gaussian kernel and divided by its similarly smoothed mask to restore edge mass. The appropriate masks were applied to each segmented image and all brain segments reassembled to a single perfusion image. Finally, all images were warped into MNI space to facilitate spatial comparison across subjects.

Anatomical MRI analyses

The participants T1 images were used in the anatomical analyses of cortical volume and thickness. Cortical thickness and volume were calculated in CAT12 with default parameters (93). Cortical thickness ROIs were specified by the Destrieux cortical atlas and left and right ROIs summed. As this atlas is a surface atlas, it does not match with the Neuromorphometrics atlas. WMHs were visually scored and counted by a trained and blinded medical researcher (L.Ø.).

Statistical analyses of GM/WM and hippocampus ROI's, neuropsychology, blood pressure, and blood tests including nitrate/nitrite level

Participants were divided into a “younger” and an “older” subject group, dichotomized at the mean age of the participants (52 years). Statistical description of group characteristics (Table S1) was performed by two-sample t test or χ² test as appropriate. The statistical analyses were based on linear mixed models. Initially, a full model comprising the following terms was established: y ∼ age × APOE × nitrate intake + age × APOE × session + BMI + Gender + Group + Educational years + (1I Participant), where y is the response variable. The fixed effects of age group, APOE and nitrate intake, as well as their interactions were our primary interest. The term age × APOE × session was included to model out session effects in the age and APOE groups. BMI (or weight and height separately in the perfusion analysis as CM was administered according to weight, and height is related to brain size (94)), gender, and educational years were included as they were assumed important in modeling effects related to the brain and vasculature. Group was included to model out effects of inequalities in the two groups used to create the cross-over part of the study.

APOE, age group, nitrate intake, gender, group, and session were treated as categorical variables belonging to one of the following two groups: APOE (ɛ4 carrier or control); age group (younger or older); nitrate intake (NaNO₃ or NaCl); session (first or second); group (A or B); and gender (male or female). BMI (kg/m²), height (cm), weight (kg), and educational years (years of education) were computed as normalized continuous variables. This full model was iteratively reduced by defining a P-threshold of 0.05 and successively removing the model term with the largest supra-threshold P-value until all terms remained below the P-threshold. Final significance of the remaining model terms was judged based on a second P-threshold to account for multiple tests (details provided below).

Stepwise regression analyses for sub-ROIs

To complement our ROI analysis restricted to WM, GM, and hippocampus, we performed a stepwise regression analysis on smaller sub-ROIs defined by CAT 12 Neuromorphometrics atlas. As we had no hypothesis regarding sidedness, left and right hemisphere ROIs were, when applicable, grouped. This left us with 64 ROIs. Only voxels with between 0.8 and 1.0 tissue probability were used, and the ROI category (GM or WM) had to match across all participants. To be less influenced by potential outliers in these smaller ROIs, we used medians to summarize metrics across them. Also, an outlier removal step was included, and participant ROIs at more than 0.5 Cook's distance were not considered. The initial model for interrogating age and APOE effects was simplified and formulated as: y ∼ age × APOE + nitrate intake + LDL + BMI + Gender + Educational years, where y is the response variable, age is the age in years, APOE grouped in carrier or control, nitrate intake is NO₃ or NaCl intake, LDL is the low-density lipoprotein in mmol/l, BMI = kg/m², gender as male or female, and educational years as years of education. APOE, gender, and nitrate intake were computed as categorical variables and the rest as normalized continuous variables. The decision to include LDL was taken retrospectively as it was seen to improve our models. The criterion for model reduction was change in the value of the Akaike information criterion. As a final check, only reduced models statistically different (< 0.05) from a null/intercept model were retained for comparison. To be able to compare the final reduced model from the stepwise reduction with the LME models from the initial analysis, the final reduced models was also calculated in an LME version with participants included as random factors. To analyze cortical thickness, the same stepwise regression pipeline was used.

Thresholds for statistical significance

We applied Bonferroni’s correction to account for multiple tests:

Blood samples—nitrate/nitrite

The P-threshold was set to 0.05/2 = 0.025 as both nitrite and nitrate were tested.

Neuropsychological tests

The P-threshold was set to 0.05/6 = 0.0083 as six independent tests were conducted.

Blood pressure

The P-threshold was set to 0.05/2 = 0.025 as both diastolic and systolic blood pressure was tested.

MR DSC ROI analysis

As two main parameters (α and β) are the basis for all the perfusion maps computed across three tissue types (WM, GM, and hippocampi), a threshold of 0.05/6 = 0.0083 was considered as the P-threshold for significance.

MR DSC SPM analyses

Voxel-wise factorial analyses were carried out in SPM12. The design matrix was based on eight main groups created based on APOE status, nitrate intake, and age group. Nitrate ingestion was considered as a dependent factor with equal variance across levels, whereas APOE status and age group were considered as independent factors with unequal variance across levels. Weight, gender, APOE allele status, age, plasma nitrate, and transcutaneous PCO₂ and SpO₂ were included as group mean-centered regressors. No other intensity normalization or scaling procedures were carried out on the data. To be considered significant, the SPM maps had to contain clusters with a family-wise error (FWE) adjusted P-value ≤ 0.01, in maps with an uncorrected voxel-wise threshold ≤ 0.001.

Stepwise regression analyses for sub-ROIs

These analyses were used to support the initial analyses and to investigate the factors impacting several sub-ROIs. Therefore, no specific threshold for statistical significance was specified, but the standard limits of P < 0.05 and P < 0.01 were used to filter the results and clarify their patterns.

Supplementary Material

Supplementary material is available at PNAS Nexus online.

Funding

This study was supported by the VELUX Foundation (Healthy Ageing with Nitrate, grant no. 19784, R.A. and P.M.R.), the Lundbeck Foundation (Professorship grant no. R310-2018-3455, R.A. and L.Ø.), and the Danish Ministry of Science, Technology and Innovation's UNIK program (MINDLab, T.E.L.).

Author Contributions

R.A., T.E.L., and L.Ø. conceived the study. R.A., S.P., T.E.L., and L.Ø. developed the study design. R.A., K.S.A., and T.E.L. acquired the data. E.W. determined plasma nitrate and nitrite content. M.C. determined APOE genotype. R.A. drafted the manuscript, and L.Ø. made substantive revisions. All authors contributed to the analysis, interpreted the data, revised the manuscript, and approved the submitted version.

Data Availability

The authors are happy to share all data upon request and written agreement.

References

Author notes