Decomposed arterial stiffness and NfL–cognition association in older adults: The IGNITE study

Abstract

BACKGROUND

Neurofilament light chain (NfL) indexes neuroaxonal injury and relates to poorer cognition. Higher carotid–femoral pulse wave velocity (cfPWV) arterial stiffness strengthens NfL–cognition associations, but reflects both pressure‐standardized structural stiffness (S–PWV) and transient hemodynamic load‐dependent pulse wave velocity (LD–PWV). We tested which component drives this interaction.

METHODS

Plasma NfL and cfPWV were measured in 552 cognitively unimpaired older adults. cfPWV was decomposed into S–PWV and LD–PWV. Five latent cognitive domains were modeled. Linear regression tested PWV moderators of NfL–cognition associations.

RESULTS

NfL correlated with all domains (r = −0.116 to −0.185; P < 0.01) and was unrelated to S–PWV or LD–PWV. Total–PWV moderated NfL–working memory (β = −0.268, P = 0.019). NfL × S–PWV interactions were significant for episodic memory, working memory, and processing speed (P < 0.040), but did not meet false discovery rate correction (q ≈ 0.067). NfL × LD–PWV interactions were non‐significant (P > 0.09).

DISCUSSION

Nominal interactions were more consistent for S–PWV. Findings did not provide multiplicity‐adjusted evidence of component‐specific moderation.

ClinicalTrials.gov: NCT02875301

Highlights

• Nominal neurofilament light chain (NfL) × structural stiffness pulse wave velocity (S–PWV) interactions were observed for episodic memory, working memory, and processing speed, but did not survive false discovery rate correction.

• Load‐dependent pulse wave velocity (LD–PWV) did not show nominal evidence of moderating NfL–cognition associations across cognitive domains.

• Findings suggest that carotid–femoral pulse wave velocity (cfPWV) decomposition may provide complementary physiological information, but additional longitudinal research is needed to determine whether decomposed PWV components improve neurocognitive prediction or mechanistic interpretation beyond conventional cfPWV.

1. INTRODUCTION

Vascular health is a major determinant of cognitive aging, and arterial stiffness—typically measured using carotid–femoral pulse wave velocity (cfPWV), is a leading mechanism linking cardiovascular aging to cognitive decline. ¹ , ² , ³ As large elastic arteries stiffen, reduced buffering capacity allows excess pulsatile energy to reach the cerebral microvessels. ⁴ Deep white‐matter arterioles are especially vulnerable, contributing to endothelial dysfunction, impaired neurovascular coupling, axonal degeneration, blood–brain barrier disruption, and microstructural white‐matter injury. ⁵ , ⁶ These changes often appear clinically as deficits in processing speed, working memory, and attentional/executive control domains that depend on distributed white matter networks. ⁶ , ⁷

cfPWV reflects both structural and load dependent physiology. Structural PWV (S–PWV) is a model‐derived, conceptually defined estimate of pressure‐standardized component of arterial stiffness intended to reflect chronic arterial remodeling, including elastin fragmentation and collagen accumulation. Load‐dependent PWV (LD–PWV) is a model‐derived estimate of the portion of cfPWV attributable to prevailing hemodynamic load, including blood pressure, vascular tone, heart rate, and acute arterial loading conditions during assessment. ⁸ , ⁹ , ¹⁰ In the Multi‐Ethnic Study of Aging, S–PWV but not LD–PWV predicted longitudinal cognitive decline, reduced gray matter volume, and greater white matter hyperintensity (WMH) burden. ¹¹ Structural stiffness also showed stronger associations with dementia and mortality, whereas load‐dependent stiffness was more closely linked to chronic kidney disease. ¹² Together these findings suggest that S–PWV may capture cumulative vascular risk relevant to brain vulnerability, whereas LD–PWV reflects more dynamic physiological conditions present at the time of assessment.

A downstream consequence of vascular dysfunction is elevation in neurofilament light chain (NfL), a biomarker of neuroaxonal injury. NfL increases in response to cumulative vascular risk exposure, ischemia, and inflammation ¹³ , ¹⁴ , ¹⁵ and has been associated with poorer cognitive performance in older adults. ¹⁶ Elevated NfL is also associated with WMHs, lacunes, and increased dementia risk, supporting its relevance to cerebral small vessel disease (SVD), in which arterial stiffness and hypertension‐related microvascular injury may converge to influence cognitive performance. ¹³ , ¹⁷ , ¹⁸ Because NfL reflects neuroaxonal injury, higher plasma NfL may index subclinical neural vulnerability even among cognitively unimpaired older adults. Greater structural arterial stiffness may reduce cerebrovascular resilience through chronic arterial remodeling, impaired buffering of pulsatile pressure, and downstream microvascular injury. Thus, higher S–PWV may strengthen the association between NfL and cognitive performance by increasing the likelihood that subclinical neuroaxonal injury and neurodegeneration is accompanied by measurable cognitive differences. Prior research from the Investigating Gains in Neurocognition in an Intervention Trial of Exercise (IGNITE) cohort demonstrated that higher conventional cfPWV strengthened associations between higher plasma NfL and poorer cognitive performance. ¹⁹ Because cfPWV reflects both intrinsic arterial wall properties and load‐dependent hemodynamic conditions, the component most aligned with this pattern remained unclear. PWV decomposition was therefore used as a complementary approach to examine whether the prior cfPWV × NfL pattern was more apparent for S–PWV, LD–PWV, or both.

RESEARCH IN CONTEXT

Systematic review: Previous studies suggest that higher arterial stiffness may exacerbate associations between neurodegenerative biomarkers and cognition, but the composite nature of carotid–femoral pulse wave velocity has limited insight into whether these relationships are more closely aligned with structural arterial remodeling or load‐dependent hemodynamic influences.

Interpretation: In this study, nominal neurofilament light chain (NfL)–cognition interaction patterns were more consistently observed for structural, pressure‐standardized arterial stiffness than for load‐dependent stiffness. However, these findings did not survive false discovery rate correction and should be interpreted as preliminary evidence that chronic arterial remodeling may contribute to neuroaxonal injury–related cognitive vulnerability.

Future directions: Longitudinal studies are needed to define temporal relationships among arterial stiffening, NfL, and cognitive decline, and to determine whether interventions targeting vascular remodeling modify dementia risk or neurodegenerative trajectories.

This distinction may also be cognitively relevant because brain regions differ in susceptibility to vascular and neurodegenerative processes. Animal and human studies show that the hippocampus is particularly sensitive to hypoperfusion and microvascular injury. ²⁰ , ²¹ , ²² Hippocampal vascular supply has been associated with medial temporal lobe–related cognitive performance and may provide vascular reserve in older adults with cerebral SVD. ²¹ In healthy older adults, cerebral arterial pulsatility has also been linked to hippocampal microvascular function and episodic memory, supporting a relationship between altered hemodynamic forces and hippocampal function in aging. ²² However, episodic memory performance may remain intact in cognitively unimpaired adults until substantial medial temporal pathology accumulates. ²³ , ²⁴ In contrast, working memory, supported by frontoparietal networks situated in arterial watershed territories, ²⁵ , ²⁶ and processing speed, which relies on long‐range white‐matter connectivity, may be more immediately affected by perfusion deficits and microvascular injury. ²⁷ , ²⁸ These domain‐specific vulnerabilities underscore the importance of considering the differential effects of vascular mechanisms on cognitive systems. Whether vascular–neurodegenerative coupling differs between S–PWV and LD–PWV across these domains is unknown.

We tested whether plasma NfL interacts with S–PWV and LD–PWV components of cfPWV in relation to domain‐specific cognitive performance in cognitively unimpaired older adults. Given prior evidence linking structural stiffness to cognitive decline, brain atrophy, and WMH burden, ⁸ we hypothesized that NfL–cognition associations would be stronger at higher levels of S–PWV, particularly for processing speed and executive/attentional control domains that are sensitive to white matter and microvascular injury. LD–PWV was examined as a complementary model‐derived component of cfPWV because it reflects prevailing hemodynamic load conditions that may influence cerebral perfusion and pulsatile transmission independent of neuroaxonal injury. We examined whether LD–PWV was associated with cognition independent of NfL, and whether PWV x NfL interaction patterns would emerge for S–PWV, particularly in executive function/attentional control and processing speed domains.

2. METHODS

2.1. Participants

IGNITE was a multisite randomized clinical trial evaluating the effects of exercise on cognitive outcomes in 648 cognitively unimpaired older adults aged 65 to 80 years; details of IGNITE methodology and eligibility criteria have been published previously ²⁹ (ClinicalTrials.gov: NCT02875301). The present analyses are cross‐sectional and included individuals with valid measures of arterial stiffness, plasma NfL, cognitive outcomes, and covariates. In brief, the full cohort excluded individuals scoring ≤ 25 on the Telephone Interview of Cognitive Status. Neuropsychological testing was conducted and only cognitively unimpaired participants who met National Institute on Aging–Alzheimer's Association (NIA‐AA) criteria were retained. Cognitive status was determined after a detailed neuropsychological evaluation and adjudicated through a consensus conference involving board‐certified geriatric neuropsychologists (Supplementary Material S1.1 in supporting information).

All participants included in the present analyses were clinically classified as cognitively unimpaired according to IGNITE eligibility criteria. Analyses of cognitive outcomes therefore reflect continuous variation in domain‐specific cognitive performance within a clinically unimpaired older adult sample. Detailed eligibility criteria, including medical, psychiatric, neurological, magnetic resonance imaging–related, and physical activity exclusions, are provided in Supplementary Material S1.2 in supporting information. All participants provided written informed consent, and study procedures were approved by the institutional review boards at each participating site (University of Pittsburgh, University of Kansas Medical Center, and Northeastern University).

2.2. PWA/cfPWV

cfPWV is a widely used, non‐invasive, and well‐validated method for assessing central arterial stiffness. ³⁰ , ³¹ All assessments were conducted after a 10‐minute rest period to ensure hemodynamic stabilization. Participants were instructed to abstain from alcohol for at least 12 hours, tobacco and caffeine for at least 4 hours, and to fast for at least 6 hours before measurement, although a light meal was permitted if fasting could not be achieved. Pulse wave analysis (PWA) was performed after a 10‐minute supine rest using the SphygmoCor XCEL left arm brachial cuff and before cfPWV assessment. The mean resting brachial values from accepted PWA recordings were used in the PWV decomposition calculations. PWA recordings were accepted only when the built‐in quality control indicator was met (i.e., “green checkmark”).

For the PWV measure, path length was measured using the SphygmoCor XCEL subtraction method. Specifically, carotid‐to‐sternal‐notch, sternal‐notch‐to‐superior‐thigh‐cuff, and femoral‐artery‐to‐superior‐thigh‐cuff distances were measured in millimeters and entered into the SphygmoCor XCEL system. Carotid pressure waveforms were acquired by applanation tonometry, and femoral waveforms were acquired simultaneously using the thigh cuff. In some cases, a trained research technician manually initiated and recorded the cfPWV acquisitions (e.g., confirming stable waveforms, and saving the recording). cfPWV values were generated by the SphygmoCor system after both waveforms met internal quality‐control (QC) criteria, indicated by an acceptable QC indicator. Two cfPWV recordings were collected; if either failed quality standards, the measurement was repeated. If the two values differed by < 0.5 m/s, their average was used. If the difference exceeded 0.5 m/s, a third measurement was obtained, and the median of the three values was used.

2.3. Decomposition of cfPWV to structural and LD–PWV

cfPWV reflects PWV under observed hemodynamic conditions and is influenced by both arterial wall properties and prevailing distending pressure. Because PWV increases non‐linearly with blood pressure, measured cfPWV can be decomposed into conceptually distinct structural and load‐dependent components. We used a previously described pressure‐standardization approach to distinguish pressure‐standardized structural stiffness from load‐dependent stiffening. ⁸ , ⁹ , ¹¹ , ¹² , ³² In this framework, S–PWV represents cfPWV estimated at a standardized reference pressure, reducing the influence of acute blood pressure at the time of measurement. LD–PWV represents the residual portion of measured cfPWV attributable to prevailing hemodynamic load and was calculated as the difference between total cfPWV and S–PWV. Because this decomposition approach is grounded in the pressure dependence of arterial stiffness and the non‐linear pressure–artery relationship of elastic arteries, S–PWV and LD–PWV were interpreted as complementary model‐derived components of cfPWV, rather than as direct biological measurements. Detailed procedures for the decomposition approach are provided in Supplementary Material S1.3 in supporting information.

2.4. Plasma NfL

Detailed procedures for plasma NfL collection, processing, assay platform, and QC procedures are provided in Supplementary Material S1.4 in supporting information.

2.5. Cognition

A comprehensive cognitive evaluation was conducted at baseline using an extensive neuropsychological battery administered across 2 days by annually certified psychometricians. The battery included traditional paper‐and‐pencil measures, tasks from the National Institutes of Health Toolbox and other computerized assessments. ³³ Raw scores from all tests were standardized and entered into a confirmatory factor analysis, which yielded five latent cognitive domains: episodic memory, processing speed, working memory, visuospatial processing, and executive function/attentional control ³⁴ (see Supplementary Material S1.1).

2.6. Statistical analyses

All statistical analyses were conducted in R version 4.4.1 using standard analytical packages. Data distributions were examined for normality, and NfL values were log transformed to address skewness. We excluded 16 participants for very low cfPWV values < 5.5 m/s that met the prespecified low–cfPWV exclusion range. ³⁵ Additionally, participants were excluded if they lacked NfL data (n = 37), cfPWV data (n = 38), were missing both biomarkers (n = 3), or had extreme NfL values exceeding 12 median absolute deviations above the sample median (n = 2). ³⁶ Our final analytic sample included 552 participants.

Pearson correlation coefficients were used to examine relationships among all three PWV variables, NfL, and each of the five cognitive domain scores. To test whether total PWV, S–PWV, or LD–PWV moderated the association between plasma NfL and cognitive performance, we fit separate multiple linear regression models for each cognitive domain. In each model, PWV components were entered either as continuous variables or as categorical groups defined by median splits (higher vs. lower PWV), and moderation was evaluated using PWV × NfL interaction terms. High versus low PWV groups were defined using sample‐specific median splits to support simple‐slopes interpretation across relatively higher and lower stiffness burden and to maintain consistency with our prior IGNITE analysis using conventional cfPWV. ¹⁹ These groups were not intended to represent clinically validated thresholds. When a significant interaction was detected, we analyzed the simple slopes to estimate the association between NfL and cognition separately within each PWV component category. Our analytic models included NfL, dichotomized PWV variables (lower vs. higher based on median split), and their interaction term as primary predictors. Models were adjusted for demographic variables known to influence cognitive performance—age, sex, site, years of education—as well as waist circumference. Waist circumference was included to reduce potential bias in arterial stiffness measurement related to increased central adiposity. Because blood pressure values are incorporated into the decomposition, they were not included as covariates. Additional cardiometabolic and hemodynamic variables were considered but were not uniformly added to primary models because they may represent upstream determinants.

3. RESULTS

Participants were aged 69.8 ± 3.8 years, and 70% were female (Table 1). Compared to included participants (N = 552), those excluded from analysis (n = 96) had lower cognitive performance scores (all P < 0.05) and cfPWV (P < 0.001), but did not differ significantly in demographic characteristics, other anthropometric measures, or NfL levels (all P > 0.05).

TABLE 1.

Demographics.

Variable	Full sample N = 6481	Analytic sample N = 552 1	Excluded N = 961	p value 2
Demographics
Age (years, SD)	69.9 (3.7)	69.8 (3.8)	70.3 (3.5)	0.12
Sex (%)				0.4
Male	187 (29)	163 (30)	24 (25)
Female	461 (71)	389 (70)	72 (75)
Education (years, SD)	16.32 (2.21)	16.36 (2.17)	16.04 (2.42)	0.2
Race/ethnicity (%)				0.2
Black	123 (19)	95 (17)	28 (29)
Asian	10 (1.5)	8 (1.4)	2 (2.1)
White	491 (76)	427 (77)	64 (67)
Other/did not specify	8 (1.2)	8 (1.2)	0
Bi‐racial	16 (2.5)	14 (2.5)	2 (2.1)
Structural PWV group				0.6
Low	296 (50%)	275 (50%)	21 (54%)
High	295 (50%)	277 (50%)	18 (46%)
Load‐dependent PWV group				0.4
Low	296 (50%)	279 (51%)	17 (44%)
High	295 (50%)	273 (49%)	22 (56%)
Anthropometrics
Waist circumference (cm)	98 (14)	98 (14)	100 (15)	0.1
MAP	95 (10)	96 (10)	95 (12)	0.8
Systolic, mmHg	134 (16)	134 (16)	135 (18)	0.9
Diastolic, mmHg	76 (9)	76 (9)	76 (10)	0.5
cfPWV, m/s	9.05 (2.03)	9.22 (1.85)	7.46 (2.84)	<0.001
Structural PWV	10.65 (1.88)	10.67 (1.90)	10.31 (1.43)	0.3
Load‐dependent PWV	0.23 (0.66)	0.23 (0.66)	0.29 (0.71)	0.5
Cognition
Episodic memory	0.00 (0.61)	0.02 (0.61)	−0.11 (0.61)	0.048
Working memory	0.00 (0.54)	0.02 (0.54)	−0.14 (0.55)	0.005
EF/attentional control	0.00 (0.59)	0.03 (0.59)	−0.15 (0.62)	0.012
Processing speed	0.00 (0.62)	0.03 (0.62)	−0.16 (0.63)	0.011
Visuospatial	0.00 (0.56)	0.02 (0.56)	−0.14 (0.57)	0.01
Biomarkers
NfL (pg/ml)	17 (7)	17 (7)	17 (11)	0.2
Study site				0.026
Pittsburgh	219 (34%)	194 (35%)	25 (26%)
Kansas	214 (33%)	171 (31%)	43 (45%)
Northeastern	215 (33%)	187 (34%)	28 (29%)

Abbreviations: cfPWV, carotid femoral pulse wave velocity; MAP, mean arterial pressure; NfL, neurofilament light chain; PWV, pulse wave velocity; SD, standard deviation.

¹Mean (SD); n (%).

²Excluded versus analytic sample.

3.1. Correlations

We applied the previously established latent cognitive factors described in earlier work ³⁵ (Supplementary Material S1.1). Consistent with prior findings from IGNITE, ¹⁸ , ²¹ NfL demonstrated significant inverse correlations with all five cognitive domains (r = −0.116 to −0.185; P < 0.01). No correlations were observed between NfL and S–PWV or LD–PWV (Table 2).

TABLE 2.

Correlations.

Outcome	r	t	p	95% CI
Neurofilament Light
Episodic memory	−0.15	−3.69	<0.001	[−0.23, −0.07]
Working memory	−0.18	−4.57	<0.001	[−0.26, −0.11]
Attentional control	−0.18	−4.36	<0.001	[−0.25, −0.10]
Processing speed	−0.19	−4.63	<0.001	[−0.26, −0.11]
Visuospatial ability	−0.12	−2.87	0.004	[−0.19, −0.04]
Structural PWV	0.01	0.27	0.787	[−0.07, 0.10]
Load‐dependent PWV	−0.01	−0.28	0.780	[−0.10, 0.07]
Total PWV
Episodic memory	−0.14	−3.32	<0.001	[−0.21, −0.06]
Working memory	−0.11	−2.76	0.006	[−0.19, −0.03]
Attention / control	−0.09	−2.23	0.026	[−0.17, −0.01]
Processing speed	−0.12	−2.81	0.005	[−0.19, −0.03]
Visuospatial ability	−0.09	−2.20	0.028	[−0.17, −0.10]
Structural PWV
Episodic memory	−0.12	−2.87	0.004	[−0.20, −0.04]
Working memory	−0.09	−2.30	0.022	[−0.17, −0.01]
Attention / control	−0.09	−2.10	0.036	[−0.17, −0.01]
Processing speed	−0.11	−2.74	0.006	[−0.19, −0.03]
Visuospatial ability	−0.07	−1.74	0.082	[−0.15, 0.01]
Load dependent PWV
Episodic memory	−0.11	−2.66	0.008	[−0.19, −0.03]
Working memory	−0.10	−2.46	0.014	[−0.18, −0.02]
Attention / control	−0.05	−1.30	0.193	[−0.13, 0.03]
Processing speed	−0.06	−1.39	0.165	[−0.14, 0.02]
Visuospatial ability	−0.09	−2.25	0.025	[−0.17, −0.01]

Note: Negative correlations indicate poorer cognitive performance with higher NfL or arterial stiffness measures. Confidence intervals represent 95% intervals for Pearson correlation coefficients.

Abbreviation: CI, confidence interval; NfL, neurofilament light chain; PWV, pulse wave velocity.

Continuous Total–PWV (r = −0.09 to −0.14; P < 0.05) was negatively correlated with all five cognitive domains (r = −0.094 to −0.118; P < 0.05). S–PWV was negatively correlated with four cognitive domains except for visuospatial processing (r = −0.07 to −0.12; P < 0.05). LD–PWV was negatively correlated with three domains: episodic memory, working memory, and visuospatial processing (r = −0.092 to −0.109; P < 0.05; Table 2).

3.2. NfL x PWV interactions across cognitive domains

3.2.1. Total–PWV

Relative to our prior work, which identified NfL × conventional cfPWV interactions for working memory and episodic memory, ²¹ the present analyses computed an overall arterial stiffness measure using Total–PWV, the parent PWV metric defined within the current decomposition framework. Using Total–PWV, higher NfL was more strongly associated with poorer working memory among individuals with high Total–PWV (β = −0.268, standard error [SE] = 0.114, P = 0.019), whereas no significant association was observed at low Total–PWV. This interaction was considered nominal and did not survive false discovery rate (FDR) correction. Episodic memory, executive/attentional control, processing speed, and visuospatial ability showed similar directionality of associations for the NfL × higher Total–PWV interaction, but these did not reach statistical significance (all P > 0.08; Table 3).

TABLE 3.

Interaction models.

Domain	PWV type	Term	β	SE	t	P	FDR P	95% CI
Episodic memory	Total	NfL × high PWV	−0.214	0.132	−1.627	0.104	0.131	−0.472, 0.044
	Structural	NfL × high PWV	−0.270	0.131	−2.058	0.040	0.067	−0.528, −0.012
		Simple slope (low)	0.019	0.095	0.200	0.842	—	−0.168, 0.206
		Simple slope (high)	−0.253	0.095	−2.660	0.008	—	−0.438, −0.064
	Load	NfL x high PWV	−0.145	0.132	−1.093	0.275	0.275	−0.405, 0.115
Working memory	Total	NfL × high PWV	−0.268	0.114	−2.355	0.019	0.094	−0.491, −0.044
		Simple slope (low)	0.009	0.083	0.108	0.914	—	−0.154, 0.171
		Simple slope (high)	−0.260	0.082	−3.171	0.002	—	−0.421, −0.098
	Structural	NfL × high PWV	−0.269	0.114	−2.362	0.019	0.067	−0.492, −0.045
		Simple slope (low)	0.085	0.083	1.024	0.306	—	−0.154, 0.171
		Simple slope (high)	−0.260	0.083	−3.133	0.002	—	−0.422, −0.098
	Load	NfL x high PWV	−0.218	0.115	−1.900	0.058	0.125	−0.443, 0.007
Processing speed	Total	NfL × high PWV	−0.231	0.133	−1.735	0.083	0.131	−0.492, 0.031
	Structural	NfL × high PWV	−0.295	0.133	−2.216	0.027	0.067	−0.556, −0.033
		Simple slope (low)	−0.043	0.097	−0.443	0.658	—	−0.232, 0.147
		Simple slope (high)	−0.338	0.097	−3.485	0.001	—	−0.527, −0.148
	Load	NfL × high PWV	−0.221	0.134	1.646	0.100	0.125	−0.485, 0.043
Visuospatial	Total	NfL × high PWV	−0.171	0.120	−1.428	0.154	0.154	−0.407, 0.064
	Structural	NfL × high PWV	−0.149	0.119	−1.251	0.212	0.212	−0.385, 0.085
	Load	NfL × high PWV	−0.200	0.121	−1.660	0.098	0.125	−0.437, 0.037
Executive/attentional control	Total	NfL × high PWV	−0.211	0.126	−1.683	0.093	0.131	−0.458, 0.036
	Structural	NfL × high PWV	−0.237	0.125	−1.886	0.059	0.075	−0.483, 0.009
	Load	NfL × high PWV	−0.221	0.126	−1.748	0.081	0.125	−0.469, 0.027

Note: Values are beta coefficients (β) with SEs from linear regression models testing the interaction between plasma neurofilament light (NfL) and arterial stiffness on cognitive domain scores. FDR‐adjusted P values are shown only for the 15 PWV × NfL interaction tests. Simple slopes are follow‐up estimates derived from the same interaction models and were not included in the FDR correction family. PWV was examined as a categorical moderator (high vs. low, median split). Simple slopes reflect the estimated association between NfL and cognition at low and high levels of PWV, derived from the same models. t statistics and two‐sided P values for simple slopes were calculated as β/SE using the model residual degrees of freedom (df = 542). All models were adjusted for age, waist circumference, education, sex, and study site. Negative coefficients indicate poorer cognitive performance with higher NfL. Statistically significant effects (P < 0.05) are bolded in the table.

Abbreviations: CI, confidence interval; FDR, false discovery rate; NfL, neurofilament light chain; PWV, pulse wave velocity; SE, standard error.

3.2.2. S–PWV

Nominally significant NfL × high S–PWV interactions were observed for episodic memory (β = −0.270, SE = 0.131, P = 0.040), working memory (β = −0.269, SE = 0.114, P = 0.019), and processing speed (β = −0.295, SE = 0.133, P = 0.027). For each domain, simple slopes analyses indicated that NfL was not associated with cognitive performance at low S–PWV, whereas NfL was significantly associated with poorer performance among individuals with high S–PWV (all P ≤ 0.008; Figure 1). Executive/attentional control and visuospatial ability showed a similar pattern, but were not significant (Table 3).

FIGURE 1.

Nominal NfL × structural PWV interactions across cognitive domains. Model‐based predicted values illustrate associations between log‐transformed NfL episodic memory, working memory, and processing speed, stratified by low (orange) and high (blue) structural PWV. Linear regression models adjusted for waist circumference, age, education, sex, and study site. Shaded regions represent 95% confidence intervals. Across episodic and working memory, and processing speed, higher PWV was associated with a steeper negative association between NfL and cognitive performance, consistent with greater vulnerability to neuroaxonal injury in the context of elevated structural arterial stiffness. NfL, neurofilament light chain; PWV, pulse wave velocity

3.2.3. LD–PWV

Although interaction estimates were generally in the negative direction for LD–PWV models across all cognitive domains, none reached conventional significance thresholds (all P > 0.09; Table 3).

3.3. Multiplicity corrections and sensitivity analyses

Benjamini‐Hochberg FDR correction was applied across the five prespecified S–PWV × NfL interaction tests. Because the full set of 15 PWV × NfL tests may also be viewed as a broader multiplicity context, we also applied FDR correction within each PWV family and across all 15 interactions. The nominal Total–PWV × NfL working memory interaction did not survive within‐family correction (q = 0.094). The nominal S–PWV × NfL interactions for episodic memory, working memory, and processing speed also did not meet FDR correction (q ≈ 0.067); LD–PWV interactions were non‐significant, and no interaction survived correction across all 15 tests.

Thus, PWV × NfL interaction findings are interpreted as preliminary and hypothesis generating rather than definitive evidence of component‐specific moderation.

Sensitivity analyses re‐included participants previously excluded for low cfPWV (< 5.5 m/s) who had complete data, increasing the analytic sample from 552 to 567 (Table 4). In these analyses, Total–PWV, S–PWV, and LD–PWV were modeled using median split high/low PWV groups. Across cognitive domains, the overall pattern of results was consistent with the primary analyses. These analyses evaluated the influence of the exclusion criterion but could not independently verify the original path‐length entries for all unusually low values.

TABLE 4.

Model comparisons.

		Excluding cfPWV values <5.5 m/s N = 552		Including cfPWV values <5.5 m/s N = 567
Cognitive domain	PWV type	NfL × PWV (high/low) β (SE)	p	NfL × PWV (high/low) β (SE)	p	NfL × PWV (continuous) β (SE)	p
Episodic memory	Total	−0.214 (0.132)	0.104	−0.204 (0.130)	0.115	−0.015 (0.025)	0.530
	Structural	−0.270 (0.131)	0.040	−0.260 (0.130)	0.045	−0.020 (0.027)	0.465
	Load	−0.145 (0.132)	0.275	−0.151 (0.130)	0.250	−0.016 (0.100)	0.870
Working memory	Total	−0.268 (0.114)	0.019	−0.243 (0.113)	0.032	−0.035 (0.021)	0.100
	Structural	−0.269 (0.114)	0.019	−0.246 (0.113)	0.030	−0.036 (0.024)	0.131
	Load	−0.218 (0.115)	0.058	−0.198 (0.114)	0.082	−0.113 (0.087)	0.198
EF/attentional control	Total	−0.211 (0.126)	0.093	−0.190 (0.125)	0.129	−0.042 (0.024)	0.076
	Structural	−0.237 (0.125)	0.060	−0.217 (0.125)	0.084	−0.043 (0.026)	0.102
	Load	−0.221 (0.126)	0.081	−0.196 (0.126)	0.120	−0.118 (0.097)	0.223
Processing speed	Total	−0.231 (0.133)	0.083	−0.221 (0.132)	0.096	−0.047 (0.025)	0.062
	Structural	−0.295 (0.133)	0.027	−0.284 (0.132)	0.032	−0.049 (0.028)	0.082
	Load	−0.221 (0.134)	0.100	−0.207 (0.133)	0.120	−0.130 (0.102)	0.203
Visual–spatial	Total	−0.171 (0.120)	0.154	−0.140 (0.118)	0.235	−0.002 (0.022)	0.931
	Structural	−0.150 (0.120)	0.212	−0.121 (0.118)	0.301	0.004 (0.025)	0.864
	Load	−0.200 (0.121)	0.098	−0.173 (0.119)	0.145	−0.098 (0.091)	0.280

Note: Values are beta coefficients (β) with SEs in parentheses from linear models examining the interaction between log‐transformed NfL and PWV variables across cognitive domains. Models were estimated excluding PWV values < 5.5 m/s (N = 552) and those including all PWV values (N = 567) comparing high versus low PWV (grouped across medians) and continuous PWV metrics. PWV metrics include Total PWV, pressure‐standardized structural PWV, and load‐dependent PWV. All models were adjusted for waist circumference, age, education, sex, and study site. Negative interaction coefficients indicate stronger inverse associations between NfL and cognitive performance at higher PWV. P values are two sided; statistically significant effects (P < 0.05) indicate moderation of the NfL–cognition association by PWV.

Abbreviations: cfPWV, carotid–femoral pulse wave velocity; EF, executive function; FDR, false discovery rate; NfL, neurofilament light chain; PWV, pulse wave velocity; SE, standard error.

We also examined whether stiffness moderated NfL–cognition associations differently by sex using three‐way interactions (NfL x stiffness x sex). There was no evidence of sex‐specific moderation across any cognitive domains (all P > 0.10). Because the sample was predominantly women (71%), these analyses were treated as exploratory. Sex‐stratified models are provided in Supplementary Material S1.5 in supporting information.

Last, age moderation was tested using three‐way interactions (log NfL × S–PWV group × age [median split and continuous]). Three‐way terms were not significant for episodic memory, working memory, or processing speed (median split: all P ≥ 0.57; continuous age: all P ≥ 0.48). Despite non‐significant three‐way interaction tests, sex‐stratified models are presented in Supplementary Material S1.5 to provide a complete record and support interpretations.

4. DISCUSSION

Building on prior IGNITE findings that higher conventional cfPWV strengthened NfL–cognition associations, ¹⁹ the present analysis explored whether the cfPWV × NfL pattern was more apparent for S–PWV, LD–PWV, or both. Nominal NfL × S–PWV interactions were observed for episodic memory, working memory, and processing speed, whereas LD–PWV interactions were not nominally significant. However, the S–PWV findings did not survive FDR correction, and continuous interaction models were non‐significant; therefore, they should be interpreted as preliminary and hypothesis generating rather than definitive component‐specific moderation. Thus, decomposition is best viewed as complementary to conventional cfPWV, not as a replacement for the aggregate parent metric or as evidence that S–PWV is the sole component underlying the prior cfPWV × NfL finding.

cfPWV decomposition may provide physiologically meaningful information about structural and load–dependent arterial stiffness mechanisms, but the present findings do not establish incremental statistical value for neurocognitive outcomes after multiple comparisons correction. Replication in longitudinal cohorts is needed to determine whether decomposed PWV components improve prediction or mechanistic interpretation beyond conventional cfPWV. These components may also differ in response to intervention: pressure‐lowering strategies may influence LD–PWV by reducing prevailing hemodynamic load, as suggested by recent Systolic Blood Pressure Intervention (SPRINT) findings, ³² , ³⁷ whereas interventions targeting extracellular matrix remodeling, vascular smooth muscle function, inflammation, calcification, glycation, or collagen–elastin turnover could plausibly influence S–PWV. This distinction may be particularly relevant for brain aging, given links among structural stiffness, cerebrovascular and neurodegenerative vulnerability, white matter lesion burden, cortical thinning, and cognitive decline. ⁸ , ¹¹

Mechanistically, chronic aortic stiffening may alter ventricular–vascular coupling, increase characteristic impedance, and promote earlier wave reflection, ³⁸ , ³⁹ thereby increasing transmission of pulsatile energy into cerebral microvascular beds. Unlike LD–PWV, which reflects prevailing hemodynamic load, S–PWV represents a pressure‐standardized estimate intended to capture more stable arterial wall properties, including extracellular matrix alterations, vascular smooth muscle cell mechanotransduction, and epigenetic reprogramming. ¹¹ , ⁴⁰ Elevated S–PWV may therefore reflect a vascular context in which neural injury is more closely linked to cognitive performance through chronic pulsatile stress and microvascular burden, ⁴¹ consistent with evidence that the brain may be vulnerable to excessive pulsatile flow and hyperperfusion‐related injury. ⁴¹ , ⁴²

This framework may explain why higher NfL was more strongly associated with poorer episodic memory, working memory, and processing speed among participants with higher S–PWV, whereas associations were negligible at lower S–PWV. In contrast, Total–PWV showed a nominal interaction only for working memory and did not reproduce the broader three‐domain S–PWV pattern. Therefore, decomposition may provide complementary physiological information by separating structural and load‐dependent contributions that may not be fully apparent in aggregate Total–PWV, consistent with recent reports that S–PWV and LD–PWV may differ across cognitive and biomarker outcomes. ⁴³ However, because these effects did not survive multiplicity correction, they should not be interpreted as proof that S–PWV is the specific driver of NfL–related cognitive vulnerability.

Together, these findings support a model in which chronic arterial remodeling—characterized by elastin degradation, collagen deposition, medial thickening, and reduced vessel compliance ³ , ⁴⁴ —may heighten vulnerability to the cognitive consequences of neuroaxonal degeneration particularly in domains reliant on large‐scale white matter networks. Working memory and processing speed depend on vulnerable frontoparietal pathways, ²⁶ , ²⁷ providing a plausible substrate for these interactions. Episodic memory also showed nominal moderation, consistent with evidence that hippocampal function is sensitive to impaired vascular compliance and altered perfusion. In contrast, executive/attentional control and visuospatial processing did not show significant moderation, suggesting that the observed pattern may be domain specific rather than global.

LD–PWV did not show moderation effects, suggesting that a single‐occasion LD–PWV stiffness may not be the primary driver of cognitive vulnerability in this sample. Structural stiffening, by increasing forward pressure wave transmission and reducing damping capacity, may therefore be more closely aligned with the cognitive consequences of neuroaxonal degeneration, whereas load‐related changes may be insufficient to breach cerebrovascular protective mechanisms. ³ , ⁴¹ , ⁴² , ⁴⁵ Nevertheless, the null findings should not be interpreted as evidence that load–dependent mechanisms are unimportant. Single‐occasion LD–PWV may be less aligned with NfL–cognition coupling, or its relevance may differ by outcome, blood pressure treatment context, or study design.

4.1. Sensitivity analyses

Sensitivity analyses that included participants with lower cfPWV values (< 5.5 m/s) yielded largely consistent patterns, supporting the robustness of the findings and indicating that exclusion of individuals with low cfPWV did not materially influence results. However, source documentation did not permit independent verification of original path‐length entries for all unusually low values. Continuous interaction models were not significant for any domain, suggesting that the observed effects may reflect differences at higher versus lower stiffness burden rather than linear variations across the full PWV range. Individuals excluded from the primary analyses for low cfPWV values or missing data performed worse across multiple cognitive domains, indicating the analytic cohort represents a comparatively higher functioning subset (Table 1). Nominal S–PWV × NfL interactions within this healthier group therefore provide a conservative test of the hypothesized pattern.

Although an interaction with sex was not significant, exploratory sex‐stratified analyses suggested a more pronounced NfL × S–PWV pattern among females—particularly for episodic memory and working memory—whereas estimates in males were weaker and largely non‐significant. Exploratory analyses showed no sex interaction for LD–PWV. One possible explanation is that widespread use of antihypertensive medications in older adults ⁴⁶ may preferentially attenuate load‐dependent stiffness, ³² while leaving more stable structural indices relatively less affected. Because these stratified analyses were exploratory, these sex‐specific patterns should not be overinterpreted and warrant evaluation in larger samples designed to test sex‐dependent moderation.

4.2. Strengths and limitations

Decomposition of cfPWV provides a complementary, model‐derived distinction between S–PWV and LD–PWV hemodynamics. Additionally, use of plasma NfL as a marker of neuroaxonal injury and neurodegeneration allowed testing whether vascular stiffness modifies cognitive correlates of subclinical neural injury in a largely cognitively unimpaired cohort. Detection of these patterns prior to clinical impairment underscores the relevance of stiffness–injury interactions for early risk stratification and suggests vascular vulnerability may influence cognition well before overt disease. Last, examining multiple cognitive domains revealed domain‐varying patterns.

Limitations include the cross‐sectional design, which precludes causal inference and temporal ordering among arterial stiffness, neuroaxonal injury, and cognition. Longitudinal studies with repeated cfPWV, NfL, and cognitive measures are needed to determine whether structural stiffening precedes or accelerates NfL‐related cognitive decline. Exclusions for missing or low cfPWV values may limit generalizability. Residual confounding is possible because hypertension burden (e.g., duration, severity, variability) and treatment effects were not completely characterized. Other cardiometabolic and clinical factors (e.g., type 2 diabetes, chronic kidney disease, smoking history, lipid levels, and statins), and medications with potential cognitive effects (e.g., anticholinergics, benzodiazepines)—were also not characterized. However, many of these factors may contribute to arterial stiffening; therefore, extensive adjustment could introduce overadjustment by controlling for variables along the causal pathway rather than independent confounders. Moreover, although race and ethnicity were reported descriptively, the study was not designed to evaluate racial or ethnic differences or the social and structural exposures these categories may reflect.

The SphygmoCor system has methodological constraints, which could attenuate associations and reduce sensitivity to detect moderation. ⁴⁷ , ⁴⁸ Additionally, the decomposition approach assumes a constant blood density, consistent with the Bramwell–Hill framework in which PWV is proportional to the inverse square root of blood density. ⁴⁹ Although blood density varies across individuals, reported whole blood values generally fall within a narrow physiologic range, ≈ 1.043 to 1.060 g/mL, ⁵⁰ suggesting a small proportional influence on derived PWV estimates relative to blood pressure, arterial wall properties, and measured cfPWV. Because participant‐specific blood density, hematocrit, and hemoglobin values were unavailable, we could not directly evaluate the sensitivity of S–PWV and LD–PWV estimates to individual blood density differences. This assumption should be considered when interpreting decomposed PWV metrics as model‐derived estimates.

Last, concomitant brain imaging markers of SVD were not included, limiting mechanistic interpretation because SVD relates to both cognition and circulating NfL levels. ¹³ , ¹⁸ Future analyses are needed to determine whether SVD burden mediates or modifies the observed associations among PWV components, NfL, and cognition.

4.3. Summary

Nominal NfL × S–PWV interactions were observed for episodic memory, working memory, and processing speed, whereas LD–PWV interactions were not nominally significant. However, these findings did not survive FDR correction, and continuous interaction models were non‐significant. Thus, results should be interpreted as hypothesis generating rather than definitive evidence of S–PWV–specific moderation. Additional longitudinal research is needed to determine whether decomposed PWV components improve neurocognitive prediction or mechanistic interpretation beyond conventional cfPWV.

AUTHOR CONTRIBUTIONS

Concept and design: Amani M. Norling, Kirk I. Erickson. Acquisition, analysis, or interpretation of data: Amani M. Norling, Kelsey R. Sewell, Audrey M. Collins, Lu Wan, George Grove, Haiqing Huang, Lauren E Oberlin, Chaeryon Kang, Thomas K. Karikari, Jill K. Morris, Eric D. Vidoni, Sandra A. Billinger, Jeffrey M. Burns, Anna L. Marsland, Chaeryon Kang, Kirk I. Erickson. Drafting of the manuscript: Amani M. Norling, Kirk I. Erickson. Critical review of the manuscript for important intellectual content: all authors. Statistical analysis: Amani M. Norling. Obtained funding: Kirk I. Erickson.

CONFLICTS OF INTEREST STATEMENT

None for: Amani M. Norling, Audrey M. Collins, Kelsey R. Sewell, Lu Wan, George Grove, Haiqing Huang, Timothy M. Hughes, Akira Sekikawa, Lauren E Oberlin, Jill K. Morris, Daniel E. Forman, Arthur F. Kramer, Charles H. Hillman, Eric D. Vidoni, Sandra A. Billinger, Jeffrey M. Burns, Anna L. Marsland, and Chaeryon Kang. Adam D. Gepner has submitted applications related to methods of arterial stiffness calculation. John M. Jakicic is on the scientific advisory board for Wondr Health, Inc. Thomas K. Karikari has consulted for Quanterix Corporation, SpearBio Inc., Neurogen Biomarking LLC, and Alzheon; and has served on advisory boards for Siemens Healthineers and Neurogen Biomarking LLC, outside the submitted work. Thomas K. Karikari has received royalties from Bioventix for the transfer of specific antibodies and assays to third party organizations. Thomas K. Karikari is an inventor on patents and provisional patents regarding biofluid biomarker methods, targets, and reagents/compositions that may generate income for the institution and/or self should they be licensed and/or transferred to another organization. Kirk I. Erickson is on the scientific advisory board for MedRhythms, Inc. and NeoAuvra, Inc. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

All participants provided written informed consent prior to data collection, with approval obtained from the institutional review boards at each participating institution: University of Pittsburgh, University of Kansas Medical Center, and Northeastern University.

Supporting information

Supporting Information

Supporting Information