CEREBROVASCULAR DAMAGE MEDIATES RELATIONS BETWEEN AORTIC STIFFNESS AND MEMORY

Abstract

Aortic stiffness is associated with cognitive decline. Here, we examined the association between carotid-femoral pulse wave velocity and cognitive function and investigated whether cerebrovascular remodeling and parenchymal small vessel disease damage mediate the relation. Analyses were based on 1820 (60% women) participants in the Age, Gene/Environment Susceptibility – Reykjavik Study. Multivariable linear regression models adjusted for vascular and demographic confounders showed that higher carotid-femoral pulse wave velocity was related to lower memory score (standardized β: −0.071±0.023; P=0.002). Cerebrovascular resistance and white matter hyperintensities were each associated with carotid-femoral pulse wave velocity and memory (P<0.05). Together, cerebrovascular resistance and white matter hyperintensities (total indirect effect: −0.029; 95% CI: −0.043, −0.017) attenuated the direct relation between carotid-femoral pulse wave velocity and memory (direct effect: −0.042; 95% CI: −0.087, 0.003; P=0.07) and explained approximately 41% of the observed effect. Our results suggest that in older adults, associations between aortic stiffness and memory are mediated by pathways that include cerebral microvascular remodeling and microvascular parenchymal damage.

Several community-based studies have shown an association of aortic stiffness and flow pulsatility with alterations in brain structure and function.^1–5 Recent studies have linked carotid-femoral pulse wave velocity (CFPWV), the reference standard measure of aortic stiffness, with cognition.^{1, 6–8} Pathways involving large and small artery remodeling and microvascular damage have been proposed as mechanisms by which arterial stiffness impacts cognition.^9–11 Aging is associated with stiffening of the aorta and a disproportionate increase in aortic as compared to muscular artery impedance. Consequently, wave reflection is reduced, thus impairing a mechanism that normally protects the microcirculation from excessive pulsatility.^{12, 13} Resulting microvascular damage and dysfunction may contribute to an association between dementia and aortic stiffness. We have shown that higher aortic stiffness was related to lower memory scores in older adults.¹ Moreover, associations between aortic stiffness and memory were attenuated when regression models of cognition were adjusted for MRI measures of brain and cerebrovascular structure,¹ suggesting that measureable structural changes via microvascular injury may mediate the link between aortic stiffness and memory. Here, we formally test the hypothesis that direct measures of cerebrovascular function mediate the association of aortic stiffness and memory.

Methods

Study Participants and Analytical Sample

The design and rationale for the Age, Gene/Environment Susceptibility–Reykjavik (AGES-Reykjavik) Study have been described.¹⁴ Participants attending a 5-year follow-up visit (n=3317; 2007 – 2011) were eligible for this analysis. AGES-Reykjavik was approved by the National Bioethics Committee in Iceland, which acts as the institutional review board for the Icelandic Heart Association (approval number VSN-00-063), by the Icelandic Data Protection Authority, and by the National Institute on Aging Intramural Institutional Review Board. Written informed consent was obtained from each study participant.

Participants were excluded for the following reasons: contraindications to MRI (n=292), refusal to participate in MRI (n=279), missing or incomplete MRI (n=261), missing hemodynamic data (n=361), missing cognitive data (n=256), and missing covariate data (n=48). The resulting sample included 1820 participants (1092 [60%] women). Table 1 presents demographic and clinical measures for included participants. Comparison of these measures for included and excluded participants is presented in Table S1.

Table 1.

Baseline demographic and clinical measures (N = 1820)

Variable	Value*
Age, years	80±5
Women, N (%)	1092 (60)
Height, cm	168±9
Weight, kg	75±13
Body mass index, kg/m2	26.5±3.9
Heart rate, beats/min.	61±10
Blood pressure, mmHg
Systolic	144±22
Diastolic	64±10
Pulse	80±25
Mean	94±14
Cholesterol level (mmol/L)
Total	5.3±1.2
HDL	1.6±0.5
LDL	3.1±1.0
Medical history
Anti-hypertension treatment, N (%)	1298 (71)
Lipid-lowering medication, N (%)	673 (37)
Prior cardiovascular disease, N (%)	468 (26)
Current smoker, N (%)	168 (9)
Diabetes mellitus, N (%)	211 (12)
Depressive symptoms, N (%)	106 (6)
Education, N (%)
Elementary	362 (20)
Secondary	948 (52)
Junior college	290 (16)
College or university	220 (12)

HDL, high-density lipoprotein. LDL, low-density lipoprotein.

^*All values are mean ± standard deviation except as noted.

Hemodynamic Data Acquisition and Analysis

The hemodynamic protocol has been described previously.¹⁵ With the participant supine, auscultatory brachial blood pressure was assessed via a computer-controlled device. Arterial tonometry with simultaneous electrocardiogram was acquired from the brachial, radial, femoral, and carotid arteries using a custom transducer (Cardiovascular Engineering, Inc.). Tonometry waveforms were signal-averaged using the electrocardiographic R-wave as a fiducial point. Systolic and diastolic brachial cuff pressures were used to calibrate peak and trough of the signal-averaged brachial waveform. Diastolic and integrated mean brachial pressures were then used to calibrate carotid, radial and femoral waveforms. CFPWV was calculated, accounting for parallel transmission in the brachiocephalic artery and aortic arch.¹⁶

Brain MRI and Analysis

The imaging protocol has been described.^{14, 17} Brain volumes were computed automatically with a modified algorithm based on the Montreal Neurological Institute pipeline.¹⁷ WMH were considered present in regions where signal intensity was higher than that of normal white and grey matter on both T₂-weighted and fluid attenuated inversion recovery (FLAIR) images. Total parenchymal volume was derived as the sum of grey and white matter and WMH volume. Tissue volumes were expressed as a percentage of total intracranial volume. Infarct-like lesions were defined as defects in brain parenchyma with signal intensity equal to cerebrospinal fluid on all pulse sequences and associated areas of high signal intensity on FLAIR and T₂-weighted images.¹

Cerebral blood flow was measured via non-triggering, phase contrast MRI.¹⁸ Total cerebral blood flow represents the blood flows through the basilar and internal carotid arteries. Cerebrovascular resistance (CVR) was calculated as mean arterial pressure (MAP) divided by mean cerebral blood flow and reported as dyne × sec/cm⁵.

Cognitive Function Testing

The cognitive battery was administered to the participants as described previously:¹ the memory composite included the California Verbal Learning Test immediate and delayed recall; the processing speed composite included the Digit Symbol Substitution Test, Figure Comparison and the Stroop Test, Parts I (word naming) and II (color naming); and the executive function composite included Digits Backward and the Stroop Test, Part III (word–color interference). Composite scores for three cognitive domains (memory, processing speed, and executive function) were calculated.^19–21 The 15-item Geriatric Depression Scale (GDS-15) was used to assess depressive symptoms; participants with a score greater than six were classified as having depressive symptoms.^{22, 23} Participants with suspected dementia were diagnosed by consensus according to international guidelines by a geriatrician, neurologist, neuropsychologist, and neuroradiologist.²⁴

Other Covariates

We adjusted for other factors in our analyses. Age, sex, education, use of antihypertensive and lipid-lowing medications, prior cardiovascular disease, and smoking (current vs. nonsmoker) were assessed via questionnaires. Height and weight were assessed during the examination. Heart rate and blood pressure were assessed during tonometry. Serum cholesterol and triglyceride levels were measured from a fasting blood test. Diabetes mellitus was defined as self-reported history of diabetes, use of blood glucose–lowering drugs, or fasting blood glucose level ≥ 7.0 mmol/L.

Statistical Analysis

Multivariable linear regression analysis was used to assess relations between arterial stiffness as measured by CFPWV and continuous measures of cognitive function. All cognitive function models were adjusted for the following covariates: age, height, weight, heart rate, diabetes mellitus, prior cardiovascular disease, use of antihypertensive and lipid-lowering medication, total and high-density lipoprotein cholesterol levels, triglycerides, smoking, education level, and depressive symptoms. We tested interactions of both CVR and WMH with CFPWV by incorporating corresponding interaction terms in the analysis. Logistic regression analysis was used to assess relations of arterial stiffness with subclinical infarcts and microbleeds, adjusting for all of the above-mentioned covariates except education level and depressive symptoms. All continuous variables were entered as sex-specific z-scores. CFPWV was inverted to normalize the distribution and limit heteroskedasticity; the inverted value was multiplied by −1000 in order to convert units to ms/m and rectify directionality of the association with aortic stiffness. CVR and WMH were natural log transformed in order to normalize the distribution. Values are presented as mean±SD except as noted. A two-sided P<0.05 was considered significant. All analyses were performed with SPSS version 20.

Mediating variables are hypothesized to provide a link between an independent variable (A) and a dependent variable (B); identifying such variables can generate hypotheses about disease pathways. Quantifying their statistical effects on the relationships of interest is accomplished through formal mediation analyses (path analyses), which gives estimates of the variance of the association between A and B explained by the mediator.^{25, 26} Presently, potential mediators were identified and measured a priori. Initially, we used partial correlations to assess the association of the potential mediator variables with either stiffness measures or cognitive function. Potential mediator variables that showed significant associations with both stiffness measures and cognitive scores were selected for regression analysis. Potential mediators were entered stepwise into a linear regression model with cognitive score as the dependent variable that additionally included CFPWV and covariates as independent variables. Mediator variables that were included in the final model and led to an attenuation of the relation between aortic stiffness and cognitive scores were used to produce conceptual mediation models, which were analyzed by using the PROCESS macro for SPSS.²⁶ A bias-corrected 95% bootstrapped confidence interval for the indirect effect using 10000 bootstrap samples was used with the above-mentioned covariates. Effects with bootstrapped confidence intervals that did not cross zero were considered significant.

Results

Hemodynamic and brain MRI variables are presented in Table 2. Aortic stiffness was high (CFPWV>12 m/s), indicating that this cohort is at high risk for cardiovascular disease events.

Table 2.

Hemodynamic and MRI brain structural measures (N=1820)

Variable	Value*
Hemodynamic measures
Mean cerebral blood flow, ml/s	9.9±2.1
Cerebrovascular resistance, DSC	13311±3443
Carotid-femoral pulse wave velocity, m/s	13.6±4.6
MRI structural measures
Brain volumes (ml)
Parenchymal	1065±100
Grey matter	675±60
White matter	369±48
White matter hyperintensities, median (25th, 75th percentile)	15 (8, 29)
Normalized brain volumes, %ICV
Parenchymal	72±4
Grey matter	45±3
White matter	25±2
White matter hyperintensities, median (25th, 75th percentile)	1.5 (0.5, 1.9)
Cortical infarcts, n (%)	267 (15)
Subcortical infarcts, n (%)	189 (10)
Cerebellar infarcts, n (%)	503 (28)
Microbleeds, n (%)	524 (29)

DSC, dyne × sec/cm⁵. ICV, intracranial volume.

^*All values are mean ± standard deviation except as noted.

Higher CFPWV was related to lower memory score (P=0.002) but was not related to processing speed (P=0.97) or executive function (P=0.90) in a model that adjusted for vascular risk factors. In logistic regression analysis that adjusted for vascular risk factors, higher CFPWV was associated with higher odds of prevalent subcortical but not cortical or cerebellar infarcts or microbleeds (Table 3). Regression models demonstrating the relations between brain lesions and cognitive function are presented in Table S2.

Table 3.

Multivariable adjusted relations between CFPWV and presence of infarcts and microbleeds (N=1820)

Lesion Type	OR	Lower CI	Upper CI	P
Cortical Infarct	1.00	0.86	1.16	0.999
Subcortical Infarct	1.30	1.09	1.56	0.004
Cerebellar Infarct	1.11	0.99	1.25	0.082
Microbleeds	1.12	1.00	1.26	0.054

CFPWV, carotid-femoral pulse wave velocity. CI=95% confidence intervals. Variables were standardized by sex. Models adjusted for age, height, weight, heart rate, diabetes mellitus, prior cardiovascular disease, use of antihypertensive and lipid-lowering medication, blood glucose, total and high-density lipoprotein cholesterol levels, triglycerides, and smoking.

We examined adjusted partial correlations using candidate mediators of the relation between CFPWV and memory (Table 4). CVR and WMH fraction were the only mediator candidates related to both CFPWV and memory score. MAP and segmental brain volumes were associated with CFPWV or memory score, but not both measures; these variables were not included as mediators or confounders in further analysis. CVR was also related to WMH fraction (r=0.103, P<0.001) in a model that adjusted for covariates included in Table 4. Additional stepwise linear regression models were examined to assess the effect of mediators on the relation between CFPWV and memory (Table 5). In a base risk factor-adjusted model, CFPWV was related to memory (Model 1). When WMH fraction entered the model, the relation between memory and CFPWV was attenuated but persisted. When CVR entered the model, the relation between CFPWV and memory was attenuated to no longer significant (P=0.07). Prevalent subcortical infarcts did not enter the model. Results of the foregoing regression models suggest potential mediation of the relation between CFPWV and memory by measures of microvascular tissue damage and remodeling of the cerebral microcirculation. We found no evidence of an interaction of CFPWV and CVR (P=0.25) or CFPWV and WMH (P=0.27), and therefore did not pursue moderation analysis.

Table 4.

Matrix of partial correlations for exposure, candidate mediator, and outcome variables (N=1820)

Variable	CFPWV		Memory Score
	r	P	r	P
MAP	0.32	<0.001	−0.03	0.19
Cerebrovascular Resistance	0.19	<0.001	−0.10	<0.001
GM Fraction	−0.009	0.71	0.26	<0.001
WM Fraction	−0.039	0.10	0.15	<0.001
WMH Fraction	0.10	<0.001	−0.16	<0.001
Parenchymal Fraction	0.002	0.92	0.23	<0.001

CFPWV, carotid-femoral pulse wave velocity. MAP, mean arterial pressure. GM, grey matter. WM, white matter. WMH, white matter hyperintensities. Variables were standardized by sex. Models adjusted for age, height, weight, heart rate, diabetes mellitus, prior cardiovascular disease, use of antihypertensive and lipid-lowering medication, blood glucose, total and high-density lipoprotein cholesterol levels, triglycerides, smoking, education level, and depressive symptoms.

Table 5.

Relations of memory with CFPWV and potential mediators (N=1820)

Model	Variables	β±SE	(LCI, UCI)	P	Adjusted R2
1	CFPWV	−0.071±0.023	(−0.116, −0.026)	0.002	0.19
2	CFPWV	−0.056±0.023	(−0.101, −0.011)	0.023	0.20
	WMH Fraction	−0.143±0.021	(−0.185, −0.101)	<0.001
3	CFPWV	−0.042±0.023	(−0.087, 0.003)	0.067	0.21
	WMH Fraction	−0.136±0.021	(−0.178, −0.095)	<0.001
	CVR	−0.072±0.021	(−0.114, −0.030)	0.001

CFPWV, carotid-femoral pulse wave velocity. WMH, white matter hyperintensities. CVR, cerebrovascular resistance. Variables were entered as z-scores standardized by sex. Models adjusted for age, height, weight, heart rate, diabetes mellitus, prior cardiovascular disease, use of antihypertensive and lipid-lowering medication, blood glucose, total and high-density lipoprotein cholesterol levels, triglycerides, smoking, education level, and depressive symptoms. Candidates for entry into the base model included prevalent subcortical infarcts, WMH fraction, and CVR.

In adjusted serial multiple mediation models containing both potential mediator variables (CVR and WMH fraction), all three indirect paths were significant, whereas the direct path between CFPWV and memory was not after accounting for mediated effects (Figure 1). By adding the mediators CVR and WMH, the direct effect of the relation between CFPWV and memory was reduced by 41% (from −0.071 to −0.042). When excluding 82 participants with dementia or 468 participants with prevalent CVD, the results were not substantively different. An unadjusted pathway analysis is present in Figure S1.

Figure 1. Pathway analysis for the effect of CFPWV on memory (N=1820).

CFPWV, carotid-femoral pulse wave velocity. CVR, cerebrovascular resistance. WMH, white matter hyperintensities. TIE, total indirect effect. Model numbers are denoted by path subscripts. Residual direct effects are labeled as path A in each model, and the indirect effects are labeled with relevant groupings of letters B-F as needed. Variables were entered as z-scores standardized by sex. The effect size ± bootstrapped SE and bias-corrected bootstrapped 95% confidence intervals are reported for all paths. Models adjusted for age, height, weight, heart rate, diabetes mellitus, prior cardiovascular disease, use of antihypertensive and lipid-lowering medication, blood glucose, total and high-density lipoprotein cholesterol levels, triglycerides, smoking, education level, and depressive symptoms.

Discussion

We investigated potential mechanisms that may underlie the association of aortic stiffness with cognitive function in the AGES-Reykjavik Study. Elevated CFPWV was associated with lower memory scores, higher WMH volume and CVR, and higher odds for the presence of subcortical infarcts. We used mediation analysis to examine the potential role of MRI structural measures as mediators of the relation between CFPWV and memory. Mediation analysis revealed that CVR and WMH separately and jointly mediated components of the relation between CFPWV and memory. Our results indicate that in older persons, elevated aortic stiffness is associated with remodeled cerebral microcirculation, microvascular brain parenchymal damage, and lower performance on memory tests.

Prior Studies of Relations between Aortic Stiffness and Cognitive Function

Several prior studies have shown that higher levels of measures of aortic stiffness and pressure and flow pulsatility, such as higher CFPWV, central pulse pressure and carotid pulsatility index, are associated with higher prevalence and extent of brain lesions (lacunar infarcts, WMH, or microbleeds) and lower cognitive scores.^{1–4, 8} Recent studies have examined the relations between CFPWV and domain-specific cognitive function via an array of neuropsychological tests.^{1, 6–8} While other studies have shown associations between aortic stiffness and other cognitive domains, previously we have shown that CFPWV was related to lower scores on quantitative testing of memory in a subsample of AGES-Reykjavik Study participants evaluated 5 years earlier.¹ We have extended those findings using established and novel markers of microvascular structure and mediation analysis with a larger sample and show that microvascular brain lesions mediate relations between higher aortic stiffness and lower memory performance. In addition, consistent with our previous findings, CFPWV was found to be independently associated with memory but not with executive function or processing speed using multivariable models. Previous work by Elias et al. suggests that the effect of aortic stiffness is modified by advanced age.⁸ Thus, the effects of CFPWV on memory in the present study may be specifically relevant to persons of advanced age.

Associations of Aortic Stiffness with Cerebrovascular Remodeling and Damage

CFPWV was associated with cerebrovascular and parenchymal alterations in our aged cohort. When pulsatile energy is transmitted into and dissipated within the microcirculation of the brain increases, small vessels may constrict or hypertrophy in response to persistent elevation of pulsatility, consistent with our observation that higher CFPWV was associated with higher CVR, a marker of cerebrovascular tone or remodeling. We posit that in order to maintain adequate perfusion, a compromise is reached wherein microvascular remodeling provides some damping of pulsatility just upstream of the fragile capillary beds at the expense of reduced basal mean flow, which increases susceptibility to ischemia. A study in younger individuals found reduced cerebrovascular reactivity in participants carrying the ε4 isoform of a neuronal repair gene ApoE: a well-known genetic risk factor for Alzheimer’s disease and dementia.²⁷ Microvascular reactivity is impaired in the presence of higher aortic stiffness.²⁸ Additionally, individuals with stiff arteries have labile blood pressure.²⁹ In middle aged to older individuals, cognitive function was shown to mediate the relation between blood pressure and physical activity.³⁰ Transient hypotension has been shown to further magnify the effect of arterial stiffness on the risk of cognitive impairment in the elderly.³¹ The combination of blunted microvascular reactivity and labile blood pressure increases susceptibility to repeated episodes of cerebral ischemia. Recurrent, transient ischemia in the course of normal daily activities could lead to cumulative tissue damage that may ultimately manifest as increased WMH and reduced memory performance as observed in the present study. Although crosstalk between small and large vessels has been described,³² we have depicted a unidirectional path by which aortic stiffness affects memory. CVR is unlikely the dominant resistance leading to modulation of systemic MAP; however, to the extent that other microvascular beds remodel their resistances and modulate MAP, aortic stiffness could be affected, resulting in a bidirectional relation.³¹

Potential Mechanisms Contributing to Decline in Memory

Consistent with our previous study,¹ we found the strongest relations between CFPWV and memory. Previous studies suggest that the loss of fiber tracts due to ischemic cerebrovascular injury in deep white matter may affect neuronal integrity associated with memory.^33–36 In older people, hippocampal volume in particular is lower in individuals with higher intracranial arterial flow pulsatility.³⁷ Thus, excessive flow pulsatility may contribute to microcirculatory damage in the hippocampus, resulting in reduced performance on memory tests.

Our results also suggest that diffuse microvascular ischemia and white matter damage (indicated by CVR and WMH) may be important contributors to memory function. CVR and WMH attenuated the relation between CFPWV and memory, acting largely through parallel paths accounting for approximately 52% and 41% of the indirect effect, respectively. Yet, the path involving both WMH and CVR in series also contributed to approximately 7% of the indirect effect. Our finding that higher CFPWV was associated with WMH and lower memory scores suggests that white matter tracts may be important in maintenance of memory and are susceptible to pulsatile microvascular damage. Additionally, white matter and the hippocampus may share a susceptibility to pulsatile microvascular damage.^{1, 37} Most of the cerebral cortex and subcortical white matter is perfused by perforating branches of the superficial circulation from the outside inward. The long, circuitous pial branches may act as a low pass filter that dampens pressure and flow pulsatility in superficial tissue.³⁸ In contrast, deep white matter and the hippocampus are perfused primarily by relatively short branches that arise from the circle of Willis. Therefore, deep white matter and the hippocampus are more directly exposed to the adverse effects of abnormal central hemodynamics and may be less protected from excessive pressure and flow pulsatility in individuals with increased aortic stiffness. We did not examine an association of CFPWV with hippocampal structure; thus, further investigation of regional brain volumes is warranted. However, we show that elevated CFPWV was associated with higher odds for prevalent subcortical infarcts (P=0.004) but not cortical infarcts (P=0.999). Moreover, a recent study of autopsy-confirmed Alzheimer’s disease patients revealed that the most prominent cerebrovascular change in participants was atherosclerosis of circle of Willis.³⁹ These data, along with the present study, underscore the importance of vascular contributors to all types of dementia.

Limitations

The limitations of our study should be considered. We employed a cross-sectional design; this type of observational study limits our ability to establish temporal relations among CFPWV, mediators, and memory. Thus, additional work employing longitudinal designs observing the mediating effects of MRI variables on the relation between aortic stiffness and memory decline is warranted. We did not acquire data on ApoE ε4 allele for this exam visit; therefore, we were unable to assess the potential mediating or confounding effects of this genetic risk factor for dementia. We did not perform triggering, phase contrast MRI and could not assess cerebrovascular flow pulsatility. We studied an exclusively older sample (mean age of 80 years). The age relations of CFPWV are much steeper after 50 years of age;⁴⁰ thus, additional studies in younger participants are needed in order to establish whether these associations apply to younger individuals. In addition, because our cohort is comprised of white participants of European descent, our findings may not be generalizable to other ethnic groups.

Perspectives

In this community-based sample of older individuals, CVR and WMH – markers of cerebrovascular remodeling and damage – mediated the relation between CFPWV (aortic stiffness) and lower performance in memory function tests. Beyond midlife, aortic stiffness increases rapidly and exposes the cerebral microcirculation to abnormal pulsatile mechanical forces that are associated with microvascular damage and remodeling. Our results indicate that microvascular and white matter damage associated with excessive aortic stiffness may contribute to impaired memory function with advancing age. Preliminary studies have suggested that aortic stiffening may be preventable and partially reversible, which makes it a good candidate for a primary or secondary prevention trial. Our findings join an ever growing body of work that implicates vascular etiologies as contributing to memory impairment, a key deficit in Alzheimer-type cognitive impairment and dementia. Thus, interventions targeting aortic stiffness among the elderly may delay the onset of memory decline. On a population-level, a lower prevalence of mild cognitive impairment decreases economic and societal burdens and increases healthspan in an aging population.

Novelty and Significance.

What is new?

• The relation between aortic stiffness and memory was mediated by measures of cerebrovascular remodeling and damage.

• Although the association between arterial stiffness and cognition has been described, our report of a mechanism by which elevated aortic stiffness is associated with lower memory function is novel.

What is relevant?

• As memory is the key cognitive domain impaired in amnestic mild cognitive impairment and Alzheimer’s disease, these findings have relevance to identifying potential targets for their prevention.

Summary

• Our results suggest that microvascular and white matter damage contribute to the observed association between higher aortic stiffness and lower memory function in older adults.