Plasma NfL and cognitive functioning in older adults: The moderating role of HDL cholesterol

Abstract

BACKGROUND AND OBJECTIVES

Plasma neurofilament light chain (NfL) is a marker of neuroaxonal injury associated with cognitive decline. High‐density lipoprotein (HDL) cholesterol has neuroprotective properties, but its interaction with neurodegeneration remains unclear. This study examined whether HDL moderates the association between NfL and cognitive performance.

METHODS

Baseline data from 417 participants in the Aging Adult Brain Connectome study were analyzed. Plasma NfL and HDL were measured via Simoa and enzymatic assays; cognition was assessed using Montreal Cognitive Assessment (MoCA) and Preclinical Alzheimer Cognitive Composite (PACC). Generalized linear models were used to evaluate NfL and HDL interactions, adjusting for demographics. Sensitivity analyses included apolipoprotein E ε4, body mass index, total cholesterol, LDL, and triglycerides.

RESULTS

Significant interaction effects were observed: MoCA (β = −1.86×10⁻⁴, P = 0.006) and PACC (β = −4.0×10⁻⁵, P = 0.004), indicating HDL moderates the negative association between NfL and cognition.

DISCUSSION

These findings suggest that HDL modifies the cognitive impact of neurodegeneration, highlighting the importance of metabolic–neurological interactions.

Highlights

• High‐density lipoprotein (HDL) cholesterol moderates the negative association between plasma neurofilament light chain (NfL) and cognition.

• Higher HDL levels intensify the negative effect of NfL on cognitive performance.

• Findings challenge the assumption of HDL's uniformly protective role.

• Results support the integrated use of metabolic and neurodegenerative biomarkers.

1. BACKGROUND

Cognitive decline is a hallmark of progressive neurodegenerative diseases that impairs memory, executive function, and activities of daily living, ultimately compromising an individual's independence and quality of life. ¹ , ² As the global population continues to age, the burden of cognitive impairment and dementia is rapidly increasing, posing significant challenges for public health systems and caregivers alike. ³ , ⁴ In the United States alone, the number of individuals affected by Alzheimer's disease (AD) dementia is projected to increase to 13.8 million by 2060. ⁵ , ⁶ This trend reflects a broader global trajectory, with an estimated 10 million new dementia cases occurring annually and a projected global prevalence of 152 million by 2050. ⁷ , ⁸

Underlying this rise in dementia prevalence is the complex and multifactorial process of neurodegeneration. Characterized by progressive synaptic and neuronal loss, neurodegeneration contributes to age‐related cognitive decline and various forms of dementia. ⁹ , ¹⁰ One measurable marker of this process is neurofilament light chain (NfL), a cytoskeletal protein released into the extracellular space and cerebrospinal fluid (CSF) after axonal injury or degeneration. ¹¹ NfL levels are also known to increase with age and may vary by sex. ¹² Higher NfL levels serve as a sensitive indicator of neuroaxonal damage and have been associated with worsening cognitive function and a greater risk of dementia, even in the preclinical stages. ¹³ As neuronal damage accumulates, cognitive impairment intensifies, leading to increased dependence and reduced quality of life. ¹³ , ¹⁴

AD accounts for the majority of dementia cases, contributing to 60% to 80% of diagnoses, followed by vascular dementia, frontotemporal degeneration, hippocampal sclerosis, and dementia with Lewy bodies. ¹⁰ , ¹⁴ Mixed pathologies are common, underscoring the need for research approaches that consider multiple interacting mechanisms. Among the emerging biomarkers, plasma NfL has received substantial attention for its utility in capturing neurodegenerative processes non‐invasively. However, neurodegeneration does not occur in isolation. Cognitive performance in aging is influenced by a constellation of factors, including vascular and metabolic health. ¹⁵ Importantly, NfL itself is a non‐specific marker that can be affected by several systemic factors such as inflammation, renal function, and lipid metabolism, including cholesterol levels. ¹⁶

One such factor is high‐density lipoprotein (HDL) cholesterol, traditionally viewed as cardioprotective but increasingly recognized for its neuroprotective properties. HDL has been shown to exert antioxidative, anti‐inflammatory, and endothelial‐supporting effects, all of which may be relevant in preserving cognitive function. ¹⁷ , ¹⁸ Multiple studies have demonstrated that higher HDL levels are associated with better cognitive performance and a reduced risk of dementia. ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ Conversely, low cholesterol levels in older age have been linked to worse cognitive performance, particularly in the absence of cardiovascular risk factors. ²¹ , ²² , ²³

Despite this evidence, recent studies have introduced conflicting findings. Some large‐scale cohort studies report that very high HDL cholesterol levels may be associated with an increased risk of all‐cause dementia, suggesting a potential U‐shaped or non‐linear relationship between cholesterol and cognition. ²² , ²⁴ , ²⁵ , ²⁶ , ²⁷ These contradictory results highlight the complexity of lipid metabolism in the aging brain and the need for a more nuanced understanding of how metabolic and neurodegenerative pathways interact.

Most investigations have examined neurodegenerative and metabolic biomarkers independently. The relationship between plasma NfL and HDL cholesterol remains poorly understood. Exploring this interaction across the adult lifespan, particularly from midlife through older adulthood, could offer new insights into the combined influence of neurodegeneration and lipid metabolism on cognitive performance.

RESEARCH IN CONTEXT

1. Systematic review: We reviewed prior literature on PubMed and recent conference proceedings to understand the independent associations between plasma neurofilament light chain (NfL), high‐density lipoprotein (HDL) cholesterol, and cognition. While plasma NfL is a well‐established marker of neuroaxonal injury, the role of HDL in moderating neurodegenerative biomarkers has not been adequately explored.

2. Interpretation: Our findings indicate that HDL cholesterol significantly moderates the relationship between plasma NfL and cognitive performance, such that higher HDL levels amplify the negative association between NfL and cognition. This challenges the commonly held view of HDL as universally neuroprotective.

3. Future directions: Future studies should evaluate the mechanistic pathways underlying this interaction using longitudinal data and include other metabolic and inflammatory markers. Investigating whether interventions that modify HDL levels impact the relationship between neurodegeneration and cognition will be critical for developing precision‐based strategies in dementia risk management.

This study investigates the influence of HDL cholesterol on the relationship between plasma NfL concentrations and cognitive performance in older adults by assessing whether the interaction remains significant after adjusting for demographic and clinical covariates. We hypothesized that higher plasma NfL levels would be associated with poorer cognitive performance and that this association would be attenuated in individuals with higher HDL levels. We anticipated that this moderating effect will be consistent across both global cognitive measures included in our analysis and will be independent of relevant covariates.

2. METHODS

2.1. Study design and participants

This cross‐sectional study used baseline data from the Aging Adult Brain Connectome (AABC) study, a multi‐site longitudinal cohort conducted at Washington University in St. Louis, Harvard/Massachusetts General Hospital in Boston, UCLA David Geffen School of Medicine in Los Angeles, and the University of Minnesota in Minneapolis. Participants were community‐dwelling adults aged ≥ 36, recruited through academic centers and outreach at four US sites; the sample was voluntary and not population based. As part of the parent study, participants completed clinical, cognitive, and biomarker assessments every 2 years. For the current analysis, data from the baseline time point were used.

Sociodemographic information, including age, sex, education level, and apolipoprotein E (APOE) ε4 genotype, was collected. Eligible participants were ≥ 36 years, physically healthy, fluent in English, and able to travel to the study site. Individuals with any psychiatric or neurological disorders, severe medical illnesses limiting life expectancy (e.g., advanced cancer), recent substance abuse, current use of medications known to affect cognition, or a history of traumatic brain injury involving loss of consciousness were excluded (see Supplementary Flowchart 1 in supporting information). No formal mild cognitive impairment (MCI) diagnosis was conducted; cognitive status was assessed cross‐sectionally using the Montreal Cognitive Assessment (MoCA) and Preclinical Alzheimer Cognitive Composite (PACC). As brief cognitive tests may be influenced by age, education, or fatigue, low scores alone are insufficient to confirm MCI.

The study protocol was approved by the institutional review board at Washington University in St. Louis (IRB #: 202201003‐1001) and complied with the Declaration of Helsinki. Written informed consent was obtained from all participants. This study adheres to the Strengthening Reporting of Observational Studies in Epidemiology guidelines for cross‐sectional studies.

2.2. Biomarker assessments

Blood samples were collected in K₂ ethylenediaminetetraacetic acid tubes and centrifuged at 2000 g for 5 minutes. Plasma was transferred to low‐retention polypropylene cryovials within 4 hours and stored at −80°C until analysis. Plasma NfL concentrations were measured using the ultra‐sensitive single‐molecule array (Simoa) platform (Quanterix). HDL cholesterol was quantified via standardized enzymatic colorimetric methods (Roche Diagnostics). Additional metabolic biomarkers, including low‐density lipoprotein (LDL), total cholesterol, triglycerides, and creatinine, were assessed similarly. APOE ε4 genotype was determined using polymerase chain reaction–based assays.

2.3. Cognitive assessments

Cognitive performance was evaluated using PACC, which included tests of episodic memory, processing speed, executive function, and semantic memory. ²⁸ Participants completed the Rey Auditory Verbal Learning Test (RAVLT) to assess episodic memory, the Trail Making Test Parts A and B (TMT‐A and TMT‐B) ²⁸ to assess processing speed and executive function, and the National Institutes of Health Toolbox Picture Vocabulary Test to assess semantic memory. ²⁹ The MoCA was also administered to evaluate global cognitive function. ³⁰ A single composite score, PACC, was computed using standardized (z score–transformed) scores from the delayed recall score of RAVLT, TMT‐A, TMT‐B, and the uncorrected standard score from the Picture Vocabulary Test.

2.4. Statistical analysis

Statistical analyses were conducted using R software (version 4.2.2). Descriptive statistics summarized demographic, clinical, cognitive, and biomarker characteristics. Continuous variables were compared using one‐way analysis of variance (for ≥ 3 groups) or independent t tests (for 2 groups), while categorical variables were assessed using chi‐squared or Fisher exact tests when expected frequencies were < 5. For the analysis, race was categorized into two groups: White and non‐White (nW). The nW race groups included individuals racialized as Black or African American, Hispanic, Native American, and Asian due to the limited number of individuals in each of these groups.

Generalized linear models (GLM) were used to examine whether the relationship between plasma NfL concentrations and cognitive performance was modified by HDL cholesterol across two cognitive outcomes: MoCA and PACC. All models included an interaction term between NfL and HDL and were adjusted for age, race, sex, and years of education. Interaction effects were visualized using interaction plots.

2.5. Sensitivity analysis

Sensitivity analyses were conducted to evaluate the robustness of findings by incorporating additional covariates, including body mass index (BMI), APOE ε4 genotype, total cholesterol, LDL, triglycerides, and creatinine. These variables were included to account for potential confounding effects that may influence the relationships among NfL, HDL, and cognitive performance. Friedewald LDL is a linear function of total cholesterol, HDL, and triglycerides; ³¹ , ³² therefore, we used non‐redundant lipid sets to avoid structural multicollinearity: Set A (HDL + LDL + triglycerides; total cholesterol excluded) and Set B (HDL + total cholesterol + triglycerides; LDL excluded). Multicollinearity was assessed with variance inflation factors (VIF < 5 acceptable) using the check_collinearity() function in R Studio, with GVIFs converted to VIF equivalents for multi‐df factors. To test for non‐linear associations between HDL and cognition, we added a quadratic HDL term (HDL²) to unadjusted and fully adjusted models (including age, sex, race, education, NfL, BMI, and APOE ε4). Model fit was evaluated using Akaike information criterion (AIC), Bayesian information criterion (BIC), and likelihood ratio tests, with ΔAIC or ΔBIC ≥ 2 indicating improved fit.

3. RESULTS

3.1. Baseline characteristics and biomarker‐based stratification

A total of 417 participants were included (Table 1), stratified into three age groups: 36 to 45 years (n = 57), 45 to 65 years (n = 154), and ≥ 65 years (n = 206). The sample had a mean age of 66.5 years (standard deviation [SD] = 16.8; range 37.7–90.0), was 55.4% female, and 76.1% White (χ²[2] = 49.59, P < 0.001). The mean BMI was 27.3 kg/m² (SD = 5.09), with higher values in younger groups (F[2, 415] = 8.62, P < 0.001). Years of education averaged 17.5 years (SD = 2.20) and did not differ by age (F[2, 415] = 0.06, P = 0.945). Mean plasma NfL levels increased markedly with age (36–45: 74.9 pg/mL; 45–65: 85.6 pg/mL; ≥ 65: 218 pg/mL; F[2, 415] = 42.67, P < 0.001). Among metabolic biomarkers, HDL varied significantly (F[2, 412] = 4.12, P = 0.0169), lowest in the 36 to 45 group (53.4 mg/dL) and higher in older groups (60.1–60.7 mg/dL). LDL was significantly lower in the ≥ 65 group (106 mg/dL; F[2, 411] = 6.64, P = 0.00145), with total cholesterol showing a similar trend (F[2, 410] = 3.97, P = 0.0195); triglycerides did not differ (F[2, 412] = 0.81, P = 0.444). Cognitive performance declined with age: MoCA scores decreased from 26.8 (36–45) to 24.8 (≥ 65; F[2, 413] = 15.82, P < 0.001), and PACC scores from 0.185 (36–45) to –0.158 (≥ 65; F[2, 415] = 15.35, P < 0.001).

TABLE 1.

Baseline demographic, cognitive, cardiometabolic, and plasma biomarker characteristics stratified by age.

Participants stratified by age
	Total (N = 417)	36–45 (N = 57)	45–65 (N = 154)	65 and above (N = 206)	Statistic †
Age (years)
Mean (SD)	66.5 (16.8)	41.4 (2.11)	55.6 (6.19)	81.6 (7.23)	F(2, 415) = 1241.35, P  < 0.001
Sex
Female	231 (55.4%)	32 (56.1%)	91 (59.1%)	108 (52.4%)	χ2(2) = 1.73, P = 0.421
Male	186 (44.5%)	25 (43.9%)	63 (40.6%)	98 (47.6%)
Race
White	318 (76.1%)	32 (56.1%)	99 (63.3%)	187 (90.8%)	χ2(2) = 49.59, P < 0.001
non‐White (nW)	99 (23.7%)	25 (43.9%)	55 (35.7%)	19 (9.2%)
Years of education
Mean (SD)	17.5 (2.20)	17.4 (2.10)	17.5 (2.23)	17.5 (2.21)	F(2, 415) = 0.06, P = 0.945
APOE ε4 status †
Control	316 (75.8%)	44 (77.2%)	106 (68.8%)	166 (80.6%)	χ2(2) = 4.31, P = 0.116
Case	88 (21.1%)	11 (19.3%)	40 (26.0%)	37 (18.0%)
Missing	13 (3.1%)	2 (3.5%)	8 (5.2%)	3 (1.5%)
PACC score †
Mean (SD)	0.0004 (0.598)	0.185 (0.371)	0.144 (0.498)	−0.158 (0.673)	F(2, 415) = 15.35, P < 0.001
MoCA mean (SD) †	25.6 (2.84)	26.8 (2.16)	26.2 (2.73)	24.8 (2.88)	F(2, 413) = 15.82, P < 0.001
BMI mean (SD) †	27.3 (5.09)	28.9 (5.56)	28.0 (5.77)	26.3 (4.16)	F(2, 415) = 8.62, P < 0.001
HDL mean (SD) †	59.5 (17.0)	53.4 (14.2)	60.1 (18.0)	60.7 (16.7)	F(2, 412) = 4.12, p = 0.0169
Missing	3 (0.7%)	2 (3.5%)	0 (0%)	1 (0.5%)
LDL mean (SD) †	111 (32.8)	116 (26.6)	117 (31.3)	106 (34.5)	F(2, 411) = 6.64, P = 0.00145
Missing	4 (1.0%)	2 (3.5%)	1 (0.6%)	1 (0.5%)
Triglycerides mean (SD)	102 (55.4)	110 (84.8)	102 (54.9)	99.6 (45.0)	F(2, 412) = 0.81, P = 0.444
Missing	3 (0.7%)	2 (3.5%)	0 (0%)	1 (0.5%)
Total cholesterol mean (SD)	191 (38.0)	191 (32.9)	198 (34.6)	186 (41.0)	F(2, 410) = 3.97, P = 0.0195
Missing	3 (0.7%)	2 (3.5%)	0 (0%)	1 (0.5%)
NfL mean (SD) †	149 (164)	74.9 (147)	85.6 (32.2)	218 (197)	F(2, 415) = 42.67, P < 0.001

^† Note: Test statistics are reported as P (df1, df2) for analysis of variance and χ²(df) for chi‐squared tests, with corresponding P values.

Abbreviations: APOE, apolipoprotein E; BMI, body mas index; HDL, high‐density lipoprotein; LDL, low‐density lipoprotein; MoCA, Montreal Cognitive Assessment; NfL, neurofilament light chain; PACC, Preclinical Alzheimer's Cognitive Composite; SD, standard deviation.

3.2. Regression analysis

In the model predicting PACC, the NfL and HDL interaction was statistically significant (β = −0.00, standard error [SE] = 0.00, t = −2.89, P = 0.0041), indicating that higher NfL levels were more strongly associated with lower PACC scores among individuals with higher HDL. The interaction remained significant after adjusting for age (β = −0.01, SE = 0.00, t = −6.74, P < 0.001), race (β = 0.24, SE = 0.06, t = 3.61, P = 0.0004), and education (β = 0.06, SE = 0.01, t = 5.08, P < 0.001). Main effects of HDL (β = 0.01, P = 0.0040) and NfL (β = 0.00, P = 0.029) were also significant, while sex was not (P = 0.65). The model explained 24.6% of the variance (N = 414, R ² = 0.246; Table 2).

TABLE 2.

General linear models predicting cognitive score from demographic, cardiometabolic, and plasma biomarkers.

	PACC score				MoCA score
Predictor	Beta (SE)	95% CI (LL, UL)	t statistic	P value	Beta (SE)	95% CI (LL, UL)	T statistic	P value
Demographic factors
Age (years)	−0.01 (0.00)	[–0.01, –0.00]	−6.7384	0.0000	−0.0629 (0.0095)	−0.0814, –0.0443	−6.6425	0.0000
Race (White)	0.24 (0.06)	[0.11, 0.37]	3.6055	0.0004	1.0169 (0.3308)	0.3685, 1.6652	3.0742	0.0023
Years of education	0.06 (0.01)	[0.03, 0.08]	5.0831	0.0000	0.1782 (0.0595)	0.0615, 0.2948	2.9941	0.0029
Sex (Male)	−0.02 (0.05)	[–0.14, 0.08]	−0.4580	0.6472	−0.4079 (0.2869)	−0.9701, 0.1544	−1.4217	0.1559
Biomarkers
HDL	0.01 (0.00)	[0.00, 0.01]	2.8948	0.0040	0.0295 (0.0128)	0.0045, 0.0545	2.3116	0.0213
NfL	0.00 (0.00)	[0.00, 0.00]	2.1861	0.0294	0.0080 (0.0040)	0.0002, 0.0158	2.0007	0.0461
Interaction term
HDL × NfL	−0.00 (0.00)	[–0.00, –0.00]	−2.8919	0.0041	−0.0002 (0.0001)	−0.0003, –0.0001	−2.7319	0.0066

Notes: Model N: 414 (PACC), 412 (MoCA). Model R²: 0.246 (PACC), 0.216 (MoCA). PACC: a composite cognitive score designed to detect subtle cognitive changes in the preclinical stage of Alzheimer's disease. MoCA: a 30‐point cognitive screening tool used to detect mild cognitive impairment. HDL cholesterol, measured in milligrams per deciliter (mg/dL). NfL: a blood‐based biomarker of neurodegeneration, measured in picograms per milliliter (pg/mL). Beta (SE): estimated regression coefficient with its standard error. 95% CI (LL, UL): 95% CI showing the LL and UL of the estimate. t value: test statistic from the general linear model. P value:  probability of obtaining the observed results if the null hypothesis is true. Values < 0.05 are typically considered statistically significant. HDL × NfL: interaction term reflecting whether HDL levels moderate the relationship between NfL and cognition.

Abbreviations: CI, confidence interval; HDL, high‐density lipoprotein; LL, lower limit; MoCA, Montreal Cognitive Assessment; NfL, neurofilament light chain; PACC, Preclinical Alzheimer Cognitive Composite; SE, standard error; UL, upper limit.

In the MoCA model, the NfL and HDL interaction was again significant (β = −0.0002, SE = 0.0001, t = −2.73, P = 0.0066), with higher NfL concentrations associated with lower scores among those with higher HDL. The effect remained after adjusting for age (β = −0.063, SE = 0.0095, t = −6.64, P < 0.001), race (β = 1.02, SE = 0.33, t = 3.07, P = 0.0023), and education (β = 0.18, SE = 0.0595, t = 2.99, P = 0.0029). Both HDL (β = 0.030, P = 0.021) and NfL (β = 0.0080, P = 0.046) were significant main effects, while sex was not (P = 0.16). The model explained 21.6% of the variance (N = 412, R ² = 0.216; Table 2).

In age‐stratified analyses, the NfL and HDL interaction was significant for PACC in participants ≥ 65 (β = −4.77 × 10⁻⁵, SE = 2.22 × 10⁻⁵, P = 0.033) but not for MoCA (P = 0.077). No significant interactions were observed in the 36 to 45 or 45 to 65 age groups. However, when age was modeled continuously, the interaction remained significant for both outcomes, supporting the robustness of age as a modifier.

To further visualize the interactions, we plotted stratified interaction plots depicting the association between plasma NfL levels and cognitive performance, adjusting for age, race, and years of education, and stratified by HDL cholesterol, which was dichotomized into low (< 60 mg/dL) and high (≥ 60 mg/dL) groups, based on the cohort mean (59.5 mg/dL) and clinical guideline thresholds. ³³ , ³⁴ , ³⁵ Across all outcomes (MoCA and PACC), individuals in the high HDL group exhibited a steeper decline in cognitive scores with increasing NfL compared to those with low HDL (Figures 1A and 1B). To enhance transparency and illustrate variability around the regression lines, data points and 95% confidence intervals were added (Figures S1A and S1B in supporting information). These findings visually reinforce the statistical interaction and suggest that higher HDL levels may amplify the adverse cognitive effects of elevated NfL. To further examine the moderating effect of HDL, the interaction effect was evaluated using moderated path models for both MoCA and PACC, which showed the same interaction effects (Figures S1C and S1D).

FIGURE 1.

A, Interaction between plasma NfL concentration and HDL levels on PACC scores. B, Interaction between plasma NfL concentration and HDL levels on MoCA scores. HDL, high‐density lipoprotein; MoCA,: Montreal Cognitive Assessment; NfL, neurofilament light chain; PACC, Preclinical Alzheimer Cognitive Composite

3.3. Sensitivity analysis

Across all sensitivity models, the interaction between plasma NfL and HDL cholesterol remained statistically significant and consistently negative. This interaction effect persisted after adjusting for genotype status, BMI, total cholesterol, LDL, and triglycerides. In centered, non‐redundant lipid models, VIFs were low for all covariates (≈ 1–2), including the c_hdl:c_nfl interaction (≈ 1.05), indicating negligible collinearity. The HDL and NfL interaction remained negative and statistically significant across all models. For PACC, this effect was consistent across specifications (Base: t = −2.89, P = 0.0041; Set A: t = −3.03; Set B: t = −3.03; LDL only: t = −2.98; total cholesterol only: t = −3.01; triglycerides only: t = −2.99; all P < 0.005; N = 388). MoCA results were similar (Base: t = −2.73; Set A/B: t = −2.78; LDL/total cholesterol/ triglycerides only: t ≈ −2.75 to −2.79; all P < 0.007; N = 388). AIC values were comparable across models (PACC ≈ 605.8–607.3; MoCA ≈ 1834.8–1838.3; see Tables S1 and S2 in supporting information). Multicollinearity diagnostics showed stable estimates: for PACC, Set A VIFs ranged from 1.05 to 1.52; Set B from 1.05 to 1.91. MoCA followed a similar pattern. In uncentered models, VIFs were inflated (≈ 27–29) but dropped to ≈ 5.2 to 5.4 after adjustment and ≈ 1.05 after centering (see Tables S1 and S2).

The quadratic HDL term (HDL²) was small, non‐significant (all P > 0.69), and did not improve model fit (ΔAIC and ΔBIC < 2), with no change in the direction or significance of the HDL–NfL interaction supporting the use of a linear HDL specification (Table S3 in supporting information). These findings confirm that HDL consistently modifies the NfL–cognition relationship, highlighting the importance of metabolic context in interpreting neurodegenerative effects.

4. DISCUSSION

Currently, understanding how metabolic markers interact with neurodegenerative processes to influence cognitive function in aging adults is an important area in dementia research. In this study, we found that the association between plasma NfL levels and cognitive performance was significantly modified by HDL cholesterol. Although HDL is often considered neuroprotective, our results revealed that higher HDL levels were linked to a stronger negative association between plasma NfL and cognitive performance, suggesting a more complex role for HDL in the context of neurodegeneration. This consistent interaction suggests that the cognitive effects of neuroaxonal injury, as reflected by plasma NfL, may be amplified rather than mitigated in the context of higher HDL. These findings point to a complex interplay between metabolic and neurodegenerative processes in aging, indicating that the role of HDL in cognitive health may be more nuanced than previously thought.

For PACC, the NfL and HDL interaction was significant only among participants aged ≥ 65, aligning with literature that highlights the convergence of neurodegenerative and metabolic vulnerabilities in later life. For MoCA, the effect was similar in direction and magnitude but not statistically significant. When age was treated continuously, the interaction remained significant for both outcomes, suggesting age is a continuous effect modifier, with subgroup analyses likely limited by smaller sample sizes.

Mean plasma NfL concentrations in our cohort were relatively high, likely reflecting factors such as age distribution, vascular and metabolic burden, sample type, and inter‐assay variability. Because no standardized reference materials exist, absolute levels are not directly comparable across laboratories. Nevertheless, uniform processing and single‐batch analysis preserved internal validity and ensured reliable within‐cohort comparisons. Regression coefficients are unstandardized, reported in native biomarker units (e.g., NfL pg/mL, HDL mg/dL). Although the numerical value of the interaction term is small, such magnitudes are expected in observational studies involving continuous biomarkers, and even modest effects may signal biologically meaningful processes contributing to cognitive aging, particularly when examined longitudinally or at the population level.

HDL cholesterol is traditionally regarded as protective in both cardiovascular and neurological contexts due to its anti‐inflammatory, antioxidative, and vasoprotective properties. ¹⁸ , ³⁶ HDL is known to modulate endothelial function, suppress oxidative stress, and participate in amyloid beta clearance, all of which are thought to support cognitive health. ³⁷ , ³⁸ , ³⁹ , ⁴⁰ Observational studies have linked higher HDL levels with better cognitive performance in midlife and old age. ¹⁹ , ²⁶ , ³⁷ However, these associations have not been consistent, and emerging evidence indicates that the relationship between HDL and cognition may be non‐linear or context dependent. ²² , ²⁴ To address this possibility, we tested a quadratic HDL term (HDL²), but found no evidence of a non‐linear association in unadjusted or adjusted models. Accordingly, HDL was modeled as a linear predictor in all analyses.

Recent cohort studies have reported that very high HDL levels are associated with an increased risk of cognitive impairment and dementia. ²⁷ , ⁴¹ , ⁴² These findings suggest that HDL may not universally confer protection and may, under specific biological conditions, become dysfunctional or even harmful. Our study supports this emerging perspective by demonstrating that higher HDL levels are associated with worse cognitive performance in the presence of higher NfL, a marker of neuroaxonal damage. Rather than mitigating the effects of neurodegeneration, HDL appears to amplify its cognitive consequences. This paradox is further complicated by evidence from aging cohorts, which reveals variability in HDL trajectories and functionality over time. Cross‐sectional studies in older adults report stable or higher HDL, likely reflecting survivor effects, whereas longitudinal studies show gradual declines with advancing age. ²⁷ , ⁴³ , ⁴⁴ , ⁴⁵ , ⁴⁶ At the same time, aging is associated with changes in HDL functionality, including reduced antioxidant and cholesterol efflux capacity, ⁴⁵ , ⁴⁷ , ⁴⁸ suggesting that HDL concentration does not fully capture protective potential. In this context, our finding that higher HDL intensified the negative NfL–cognition association suggests that elevated HDL in late life may represent a maladaptive rather than protective state. Taken together, these observations indicate that HDL's impact on cognition is context dependent, influenced by vascular health, medication use, and comorbidities.

Several mechanisms may underlie these conflicting findings. HDL particles are heterogeneous, and their functionality can vary with age, inflammation, metabolic status, and disease. ³⁸ , ⁴⁵ , ⁴⁷ , ⁴⁸ , ⁴⁹ Under oxidative or inflammatory stress or metabolic diseases, HDL may lose its antioxidative and anti‐inflammatory properties and become pro‐inflammatory. ⁴⁵ , ⁵⁰ This conversion may reduce HDL's protective capacity or contribute to neuronal injury in aging populations and those with neurodegeneration. This dysfunctional HDL phenotype is associated with reduced capacity for cholesterol efflux, impaired vascular support, and diminished ability to modulate immune responses. ⁴⁵ , ⁴⁷ , ⁴⁸ , ⁴⁹ , ⁵⁰

In our study, higher HDL levels were associated with a stronger negative association between NfL and cognition, indicating that the functional role of HDL in cognitive aging may differ in the context of elevated neurodegeneration. Our findings imply it may function more as a marker of metabolic and vascular vulnerability than a uniformly protective target, serving simultaneously as a modifiable factor, a risk biomarker, and an indicator of systemic health.

This study has several strengths. First, it used validated plasma biomarkers, including plasma NfL, a sensitive indicator of neuroaxonal damage, and HDL cholesterol, a key metabolic marker. This enabled a novel investigation into how metabolic and neurodegenerative processes may interact to influence cognitive performance. Second, cognitive function was assessed using well‐established and complementary outcomes, which captured multiple cognitive domains and enhanced the comprehensiveness and relevance of the findings. Third, the use of clinically meaningful HDL cutoffs in stratified visualizations improved the interpretability and potential translational relevance of the interaction effects.

However, we recognize several limitations. First, the absence of detailed data on clinical factors, including cholesterol‐lowering medications (e.g., statins) and cardiovascular comorbidities (e.g., hypertension, diabetes, stroke, and atherosclerosis), limits our ability to distinguish high HDL from favorable metabolic health versus pharmacologically altered profiles and prevents evaluation of their potential confounding or moderating roles. Second, the cross‐sectional design precludes causal inference or assessment of longitudinal changes. Third, HDL was dichotomized for visualization, which may not fully capture the complexity of its relationship with NfL and cognition. Fourth, systemic factors such as metabolic (e.g., hemoglobin A1c [HbA1c]) and inflammation (e.g., interleukin 6 [IL‐6], tumor necrosis factor alpha [TNF‐α], and C‐reactive protein [CRP]) status were not measured, reducing insight into broader pathways influencing HDL function and cognition.

Future research should use longitudinal designs with larger, more diverse samples, incorporate lifestyle and medication data, include additional metabolic (e.g., HbA1c) and inflammatory biomarkers (e.g., IL‐6, TNF‐α, and CRP), and collect repeated biomarker measurements across disease stages. It will also be important to evaluate extreme HDL levels and apply Mendelian randomization to clarify causal pathways, while trials should test whether enhancing HDL functionality rather than simply raising concentration yields cognitive benefits and supports integration of HDL into risk stratification for vulnerable individuals.

Our study demonstrates that HDL cholesterol modifies the association between plasma NfL and cognition in older adults, with higher HDL strengthening the negative relationship. These findings suggest that HDL may not be uniformly protective in neurodegeneration and highlight the importance of considering metabolic context when evaluating cognitive risk. Because HDL is measurable in blood and responsive to medications, physical activity, and diet, it represents both a clinical biomarker and a modifiable target. Clinicians should consider integrating metabolic markers such as HDL with plasma biomarkers like NfL to improve early identification and guide personalized dementia prevention strategies.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest. All authors have disclosed any relevant relationships/activities/interests related to the content of this manuscript. The authors declare no conflicts of interest. Funding sources (NIH/NIA) are detailed in the Acknowledgments section, and no sponsor was involved in study design, data analysis, or manuscript preparation. Author disclosures are available in the supporting information.

CONSENT STATEMENT

All participants provided written informed consent at the time of enrollment. This study was approved by the institutional review board at Washington University in St. Louis and was conducted in accordance with the Declaration of Helsinki.

Supporting information

Supporting information

Supporting information

Singh RK, Bekena S, Zhu Y, et al. Plasma NfL and cognitive functioning in older adults: The moderating role of HDL cholesterol. Alzheimer's Dement. 2025;17:e70205. 10.1002/dad2.70205