Quantifying Brain-Age Gap in People with HIV on Antiretroviral Therapy Using Plasma Neurofilament Light Chain

Abstract

Objective:

To estimate the number of additional years above chronological age (“brain-age gap”) at which observed plasma neurofilament light chain (NfL) levels in cognitively unimpaired people with HIV (PWH) correspond to published age-stratified standards from cognitively unimpaired individuals in the general population.

Design:

Secondary analysis of the multicenter observational cohort study (ACTG HAILO) involving older, cognitively unimpaired PWH on antiretroviral therapy (ART).

Methods:

Plasma NfL concentrations were quantified using the Quanterix^® Simoa assay in 340 PWH aged ≥45 years. Values were compared with two age-stratified reference standards from cognitively unimpaired general-population controls. The brain-age gap was estimated using a generalized logistic model with repeated measures so that 5% of observed NfL concentrations in PWH would exceed the reference upper 95^th percentile for the imputed age.

Results:

Median plasma NfL concentration was 10.1 pg/mL (IQR: 7.4–13.5). Using Simrén et al. standards, 24.1% (95% CI: 19.8–29.0) of observed values exceeded the age-specific upper 95^th percentile, and 10.0% (95% CI: 7.2–13.7) using Bornhorst et al. standards. Model-based results suggested brain age was 11.0 years (95% CI: 9.1–13.7) or 6.4 years (95% CI: 5.1–8.5) higher than chronological age, depending on the standard. The estimated brain-age gap was significantly greater in individuals with subsequent cognitive decline than in those without (Simrén: 13.0 vs. 8.4 years; Bornhorst: 7.6 vs. 5.0 years).

Conclusion:

Plasma NfL reflects a substantial brain-age gap in cognitively unimpaired PWH on ART, supporting evidence of advanced aging. The brain-age gap was higher in PWH with subsequent cognitive decline.

Introduction

Cognitive impairment in older people with HIV (PWH) on antiretroviral therapy (ART) is an increasingly important clinical concern. Current U.S. clinical guidelines recommend incorporating neurocognitive assessments into routine care for PWH with cognitive concerns [1,2]. These guideline recommendations are supported by epidemiologic data, from Kaiser Permanente and Medicare beneficiaries demonstrating an increased prevalence of dementia diagnoses with advancing age in PWH [3,4], as well as by modeling studies, including a recent microsimulation model projecting that the cumulative incidence of age-associated dementia by age 80 is higher among PWH than in the population without HIV [5]. This cognitive susceptibility is consistent with a broader pattern of premature age-associated comorbidities observed in PWH, such as in cardiovascular disease, and may reflect premature aging observed in other chronic conditions [6–9]. To operationalize primary care guidelines for aging PWH, clinicians will need to characterize early cognitive changes and monitor brain health, enabling identification of individuals at risk for age-associated impairment and supporting focused prevention and management strategies.

Neurofilament light chain (NfL) is a neuronal cytoskeletal protein released following neuroaxonal injury and enters cerebrospinal fluid (CSF) and systemic circulation [10–12]. CSF and plasma concentrations correlate strongly and are elevated across a range of central and peripheral nervous system disorders [10,13–15]. NfL levels increase with age even in the absence of neurological disease, reflecting the natural accumulation of neuroaxonal injury across the lifespan [15–17]. Elevated plasma NfL is associated with brain atrophy, reduced brain activity, and cognitive decline [16,18–21]. A large-scale proteomic study showed that plasma NfL is a top-weighted feature in proteomic brain-aging clocks, which predicted mortality and cognitive decline better than chronological age alone [22]. This broader aging-related relevance is reinforced by findings from elderly cohorts, including centenarians and nonagenarians, where higher plasma NfL predicted all-cause mortality with performance comparable to or exceeding multi-item cognitive function scales beyond chronological age [23]. Because many soluble biomarkers rise with age, comparing NfL to age-stratified reference standards provides a means of quantifying premature brain aging in PWH.

Evidence from other biological markers supports that PWH have a vulnerability to premature brain aging. DNA-methylation and neuroimaging studies show accelerated aging and older-appearing brains [20,24]. Sleep EEG-derived brain-age indices show patterns characteristic of older individuals without HIV [25]. Collectively, these findings suggest brain-aging biomarkers in PWH shift toward levels seen in older individuals without HIV.

In this study, we evaluated whether plasma NfL levels in cognitively unimpaired PWH on ART aged ≥45 years are consistent with published age-stratified reference values from cognitively unimpaired individuals in the general population. We estimated an NfL-based brain-age gap, defined as the difference between chronological age and the NfL-implied brain age. This quantifies premature brain aging and allows us to examine whether a larger brain-age gap is associated with subsequent cognitive decline, information that may help clinicians identify individuals at heightened risk before clinical impairment becomes apparent.

Methods

Cohort Description

This secondary analysis used clinical, cognitive, and plasma NfL data from the multisite HIV Infection, Aging and Immune Function Long-term Observational Study (HAILO), within the Advancing Clinical Therapeutics Globally for HIV/AIDS network [26]. The analysis focused on a subset of HAILO participants: PWH aged ≥45 years without cognitive impairment at initial assessment, on ART with HIV-1 RNA <200 copies/mL. Participants were followed longitudinally, with neurocognitive assessments conducted annually. Thirty-two individuals with stage 3 or higher chronic kidney disease (estimated glomerular filtration rate <60 mL/min/1.73m²) were excluded due to the potential influence of renal dysfunction on NfL levels [27]. All participants provided informed consent, and the study protocol was approved by the institutional review board at each participating site [28].

Cognitive Assessment

Cognition was assessed using four neuropsychological tests (Trail Making Tests A and B, the Wechsler Adult Intelligence Scale-Revised Digit Symbol subtest, and the Hopkins Verbal Learning Test–Revised total learning trials 1–3), with each score normalized for age, sex, and education. Z-scores from each test were averaged into a composite “NPZ-4” z-score, and longitudinal NPZ-4 slopes were calculated for each participant [26,28]. Participants were excluded if they were cognitively impaired at baseline, defined as a z-score 2.0 or more standard deviations (SD) below the mean on any one neuropsychological test or 1.0 or more SD below the mean on two or more tests. Subsequent cognitive decline (i.e., worsening neuropsychological test performance following a cognitively unimpaired baseline) was indicated by a negative NPZ-4 slope (slope <0) [26].

Plasma NfL Quantification

Plasma NfL was measured using the Simoa Neurology 2-plex B kit (Quanterix^®, Item 103520) on a fully automated HD-X analyzer according to the manufacturer’s specifications. Samples were diluted 1:4 and measured in duplicate. Quantification was performed at Massachusetts General Hospital. The duplicate coefficient of variation (CV) was 5.1 ± 5.2%, and the inter-plate CV was 9.8 ± 2.8% [26].

Reference NfL Values

We identified two published sources of age-specific reference values for plasma NfL levels using the Quanterix^® Simoa kits. The first source (Simrén et al.) provided upper 95th percentile values derived from 1,724 “neurologically healthy” individuals aged 5–90 years from eight different cohorts with “no history or clinical symptoms or signs of neurologic disorder” [29]. The second source (Bornhorst et al.) reported upper 95th percentile values from 1,100 “cognitively normal” individuals aged 20–95 years [30]. This study included individuals aged 20–49 without “cognitive impairment” and individuals aged ≥50 years who were classified as “cognitively normal” based on a consensus review after neurological and neuropsychometric testing (see Table 1). The 95th percentile is a standard threshold commonly used to define the upper limit of normal in clinical practice; it represents a value below which 95% of measurements from a healthy reference population fall, as reported in Simrén et al. and Bornhorst et al. [29,30].

Table 1.

Summary of study designs and age-stratified plasma neurofilament light chain (NfL) reference standards.

	Simrén et al. (2022)	Bornhorst et al. (2022)
Cohort Size (n)	1,724	1,100
Study Population	Eight cohorts from Sweden, Spain, Belgium, Italy; ages 5–90	Single-site adult cohort; ages 20–95
Cohort Description	“Neurologically healthy individuals”	“Cognitively normal individuals”
Health Exclusions	Vary by cohort; generally excluded neurological diseases, dementia, psychiatric/neurological illness, or Aβ pathology	No chronic kidney disease (eGFR >90 mL/min/1.73m2), neurological conditions, stroke, myocardial infarction, or atrial fibrillation, BMI >30
NfL Assay Platform	Quanterix® Simoa HD-X or HD-1 (Simoa® NF-light™ Kit)	Quanterix® Simoa HD-X (Simoa® NF-light™ Advantage Kit)
Age (years)	Upper 95% NfL Limit for Age Range (pg/mL)*
5 to <18	7	N/A
20 to <30	10	≤7.9
30 to <40		≤10.5
40 to <50		≤14.0
50 to <60	15	≤18.5
60 to <70	20	≤24.6
70 to <80	35	≤32.7
80+		≤43.3

This table compares methodological characteristics and age-stratified neurofilament light chain (NfL) reference limits reported in Simrén et al. (2022) and Bornhorst et al. (2022). Cohort size, population characteristics, health exclusions, sample types, and assay platforms differ between the two studies. The NfL values shown represent the upper 95% reference limit for each age range, based on plasma measurements (with serum values adjusted for two cohorts in Simrén et al.)

Abbreviations: Aβ, amyloid-beta; BMI, body mass index; eGFR: estimated glomerular filtration rate; NfL, neurofilament light chain; Simoa, single molecular array.

^*NfL reference values reflect the 95% upper limit for the age range, a commonly used clinical threshold indicating the value below which 95% of healthy reference populations fall.

Statistical Methods

Our objective was to estimate the brain-age gap, which we define as the difference between chronological age and implied age for the entire population. We estimated the brain-age gap such that 5% of the observed NfL values in the PWH cohort exceeded the upper 95% reference value for implied age; equivalently, 95% of observed values fell below this threshold. If the brain-age gap were zero, approximately 95% of observed NfL values would fall below the 95% population reference value for participants’ chronological age. For example, the 95% upper limit of normal from Simrén et al. is 15 pg/mL for individuals between ages 50 and 59 years old. A 45-year-old (chronological age) person with a plasma NfL of 15 pg/mL would correspond to an implied age of 50 or higher for the NfL value to be considered normal. Thus, this individual’s brain-age gap would be 5 years or more for their NfL level to fall within the normal range.

Using the PWH cohort participants’ chronological age, we first calculated the proportion of NfL values above the upper 95% reference value, based on published reference values [29,30]. We then estimated the proportion of PWH with NfL values above the upper 95% reference value by considering a range of potential values for the brain-age gap. We report these estimated percentages with exact (Wilson) 95% confidence intervals for the range of potential values considered. When approximately 5% of the population exceeds the reference values (or, equivalently, 95% fall below the 95% reference value), we obtain an estimate of the brain-age gap: how ‘old’ participants’ neuronal health is biologically compared with their chronological age.

Finally, we constructed a multi-observation dataset per individual, which included whether the NfL value exceeds the upper 95% age-matched reference. For each individual, we included the observed data and imputed data for a brain-age gap of one year older, a brain-age gap of two years older, and so on, up to the maximum value included in the analysis. We then used a generalized logistic regression model (repeated measures, no intercept, robust estimates of the standard error to compute the 95% confidence limit) to estimate the brain-age gap. Imputed data, details and figure for this model-based analysis are in Supplementary Figure 1.

The statistical significance of covariates on the estimated brain-age gap was evaluated using a generalized linear model with repeated measures and an intercept. No adjustment for multiple testing was applied. For subgroup analyses, the original model without an intercept was used to estimate the brain-age gap. Covariates considered included sex at birth, race, ethnicity, smoking, duration on ART, ARV medication type, current use of tenofovir disoproxil fumarate [TDF] or tenofovir alafenamide [TAF], any exposure to d-drugs [stavudine (d4T), dideoxycytidine (ddC), or didanosine (ddI)], current CD4+ count, CD4+ nadir, chronic kidney disease [stages 1–2], and subsequent cognitive decline [31,32]. All analyses were performed in SAS version 9.4 (Cary, NC).

Results

Study Demographics

The study population consisted of 340 PWH on ART with a median age of 52 years (IQR: 48–57 years) (see Table 2). The majority were male at birth (82.6%) and identified as either non-Hispanic white (56.8%) or non-Hispanic Black (28.2%). Participants had a median of 14 years of education (IQR: 12–16 years) and a median plasma NfL concentration of 10.1 pg/mL (IQR: 7.4–13.5 pg/mL). Nearly all participants (95.9%) achieved viral suppression, defined as HIV-1 RNA <50 copies/mL [2]. The median current CD4+ count was 644 cells/μL (IQR: 475–849 cells/μL), and the median duration of ART was 8.6 years (IQR: 5.9–12.0 years). Roughly one-third of participants were on a non-nucleoside reverse transcriptase inhibitor (NNRTI; 133/336, 39.6%) or on a protease inhibitor (PI; 112/336, 33.3%), with the remainder (75/336, 22.3%) on an integrase strand transfer inhibitor (INSTI). Only 4.8% of participants (16/336) were on more than one of the three antiretroviral classes. Current TDF or TAF use was reported in 78.8% of participants (267/339), and 13.0% (44/339) had ever been exposed to d-drugs (d4T, ddC, or ddI). Half of the participants (172/340, 50.6%) experienced subsequent cognitive decline.

Table 2.

Demographic and clinical characteristics of the PWH study population.

	Total (n=340)
Age, years [median (IQR)]	52 (48–57)
Sex at birth [n (%)]
Female	59 (17.4)
Male	281 (82.6)
Race/ethnicity [n (%)]
Non-Hispanic Black	96 (28.2)
Non-Hispanic white	193 (56.8)
Hispanic (regardless of race)	51 (15.0)
Ethnicity [n (%)]
Hispanic or Latino	51 (15.0)
Non-Hispanic or Latino	289 (85.0)
Years of education, years [median (IQR)]	14 (12–16)
Smoking* [n (%)]
Current	74 (21.8)
Prior	151 (44.4)
Never	114 (33.5)
eGFR, mL/min/1.73m2 [median (IQR)]	91.1 (79.5–103.2)
Chronic kidney disease** [n (%)]
Stage 1 (normal)	178 (52.4)
Stage 2 (mild)	162 (47.6)
HIV viral load, copies/mL*,*** [median (IQR)]	39 (20–40)
HIV viral load, copies/mL*,*** [n (%)]
<50	325 (95.9)
≥50	14 (4.1)
Current CD4+ count, cells/μL* [median (IQR)]	644 (475–849)
Current CD4+ count, cells/μL* [n (%)]
≥500	245 (72.3)
<500	94 (27.7)
CD4+ nadir, cells/μL* [median (IQR)]	188 (66.5–301.5)
CD4+ nadir, cells/μL* [n (%)]
≥350	53 (15.6)
200–349	108 (31.8)
<200	178 (52.4)
Duration of ART, years [median (IQR)]	8.6 (5.9–12.0)
Duration of ART, years [n (%)]
<5	65 (19.1)
5–10	141 (41.5)
>10	134 (39.4)
Current ART Class Medication† [n (%)]
INSTI	75 (22.3)
NNRTI	133 (39.6)
PI	112 (33.3)
More than one class	16 (4.8%)
Current TDF/TAF use*,‡ [n (%)]
Yes	267 (78.8)
No	72 (21.2)
Ever use of d-drug*,§ [n (%)]
Yes	44 (13.0)
No	295 (87.0)
Plasma NfL concentration, pg/mL [median (IQR)]	10.1 (7.4–13.5)
Year of plasma collection [n (%)]
2013–2024	216 (63.5)
2015–2016	124 (36.5)
Subsequent cognitive decline‖ [n (%)]
Present	172 (50.6)
Absent	168 (49.4)

Abbreviations: ART, antiretroviral therapy; EFV, efavirenz; eGFR, estimated glomerular filtration rate; HIV, human immunodeficiency virus; INSTI, integrase strand transfer inhibitor; NNRTI, non-nucleoside reverse transcriptase inhibitor; NfL, neurofilament light chain; PI, protease inhibitor; TAF, tenofovir alafenamide; TDF, tenofovir disoproxil fumarate.

^*Unknown, n=1.

^**Chronic kidney disease stages are defined based on eGFR (mL/min/1.73 m²) as follows: Stage 1, ≥90 without evidence of kidney damage; Stage 2, 60–89 without evidence of kidney damage.

^***All individuals are on ART and considered virally suppressed (<200 copies/mL).

^†Four participants were not included. One had no regimen information, one was on 3TC/ZDV/NFV, and two were on ABC/3TC/ZDV/TDF.

^‡Combined as only seven individuals are on TAF.

^§Ever use of any of stavudine (d4T), dideoxycytidine (ddC), or didanosine (ddI).

^‖Defined as worsening neuropsychological test performance following a cognitively unimpaired baseline. [26]

Estimates of the Brain-Age Gap

Using Simrén et al. standards, 24.1% of participants (95% CI: 19.8–29.0%) had NfL levels above this threshold; using the Bornhorst et al. standards, 10.0% (95% CI: 7.2–13.7%) exceeded the reference value. Figures 1A and 1B show the proportion of participants above the 95% reference standard at different brain-age gaps for each set of reference values.

Figure 1.

Proportion of cognitively unimpaired PWH exceeding the upper 95% neurofilament light chain (NfL) reference values from general cognitively unimpaired populations according to A) Simrén et al. and B) Bornhorst et al.

The figure shows how the estimated proportion of cognitively unimpaired PWH above (A) the Simrén et al. upper 95% reference values and (B) the Bornhorst et al. upper 95% reference values changes (Y-axis) at different values of the brain-age gap (X-axis). Each panel displays the observed proportion above the upper 95% reference value (marked as “X,” assuming no brain-age gap). If there were no brain-age gap in PWH, the observed proportion should be close to the 5% reference line drawn in the figure. Results for the proportion that would be above the 95% upper limit are represented for different values of the brain-age gap by solid dots; pointwise 95% confidence intervals are given for each estimate represented by open circles. In each panel, the line at y=1 shows the results of the mixed-model analysis, indicated by a black square, and the 95% confidence interval of the estimate.

By adding years to participants’ chronological age until approximately 5% of the population exceeded the 95% reference value, we estimated the brain-age gap to be 11–12 years older than participants’ chronological age according to Simrén et al. standards (11 years: 19/340, 5.6%; 12 years: 15/340, 4.4%), and 7 years older according to Bornhorst et al. standards (17/340, 5.0%).

Because the estimated brain-age gap changed depending on the number of years included in the reference model for the model-based approach, we performed a sensitivity analysis using imputed datasets spanning multiple years of brain-age gaps. (see Supplementary Technical Note). For the Simrén et al. reference, including data up to 25 years resulted in an estimated brain-age gap 11.0 years (95% CI: 9.1–13.7 years), and for the Bornhorst et al. reference, including 15 years resulted in an estimated brain-age gap of 6.4 years (95% CI: 5.1–8.5 years). These spans were selected, because the model-based estimates leveled off at approximately twice the estimate of the brain-age gap based on the proportion of participants above the 95% reference value, indicating that including additional years beyond this range minimally affected the results (see Supplementary Figure 1). These sensitivity analyses demonstrate that, although the precise estimate varies slightly depending on the reference age ranges, the brain-age gap remains consistently elevated relative to chronological age.

Other Correlates of the NfL-associated Brain-Age Gap

In generalized linear models, subsequent cognitive decline was consistently and statistically significantly associated with the estimated brain-age gap (see Table 3). Using Simrén et al. reference standards, PWH who experienced subsequent cognitive decline had a brain-age gap of 13.0 years (95% CI: 10.4–17.4 years) compared to a gap of 8.4 years (95% CI: 6.4–12.3 years) for PWH without subsequent decline (p=0.02). Similarly, with Bornhorst et al. reference standards, the brain-age gap was higher in PWH with cognitive decline (7.6 years [95% CI: 5.8–10.8 years] vs. 5.0 years [95% CI: 3.5–8.8 years], p=0.05).

Table 3.

Estimated brain-age gap across demographic and clinical subgroups in the PWH cohort, using Simrén et al. and Bornhorst et al. reference standards.

	Simren et al. Reference		Bornhorst et al. Reference
	Estimated Brain-Age Gap [years (95% CI)]	p*	Estimated Brain-Age Gap [years (95% CI)]	p
Total population	11.0 (9.2–13.6)		6.4 (5.1–8.5)
Sex at birth		0.28		<0.01
Female	8.9 (6.2–15.0)		2.2 (1.2–9.8)
Male	11.4 (9.3 –14.5)		7.0 (5.6–9.4)
Race/ethnicity		0.51		0.67
Non-Hispanic Black	10.0 (7.1–17.2)		7.2 (5.0–12.5)
Non-Hispanic white	11.4 (9.1 – 15.4)		6.5 (4.9–9.6)
Hispanic (regardless of race)	10.7 (7.4–18.8)		4.4 (2.7–12.1)
Ethnicity		0.63		0.37
Hispanic or Latino	10.7 (7.4–18.8)		4.4 (2.7–12.1)
Not Hispanic or Latino	11.0 (9.0–14.1)		6.7 (5.3–9.1)
Smoking		0.07		0.08
Current	14.4 (10.9–21.3)		8.8 (6.3–14.2)
Prior	9.3 (6.9–14.1)		4.7 (3.0–10.6)
Never	10.3 (7.6–15.8)		6.3 (4.5–10.4)
Duration of ART, years		0.47		0.86
<5	11.1 (7.9–18.4)		5.2 (3.4–11.0)
5–10	12.2 (9.4–17.5)		6.8 (4.9–11.1)
>10	9.4 (7.1–13.9)		6.5 (4.8–10.3)
ARV medication type		0.32		0.39
INSTI	13.6 (9.5–24.0)		7.8 (5.2–15.5)
NNRTI	7.6 (6.0–10.3)		4.7 (3.3–8.5)
PI	12.4 (9.6–17.8)		7.5 (5.5–11.8)
More than one class	6.7 (4.4–14.8)		3.2 (1.6–35.1)
Current TDF/TAF use		0.93		0.89
Yes	11.3 (9.2–14.6)		6.5 (5.1–9.0)
No	9.8 (7.5–14.3)		6.2 (4.1–12.4)
Ever use of d-drugs**		0.63		0.75
Yes	8.3 (5.7–15.0)		6.2 (3.6–21.2)
No	11.3 (9.4–14.3)		6.5 (5.1–8.7)
Current CD4+ count, cells/μL		0.02		0.35
≥500	12.0 (9.9–15.2)		6.7 (5.3–9.3)
<500	7.8 (5.0–17.7)		5.6 (3.7–11.7)
CD4+ nadir, cells/μL		0.96		0.35
≥350	10.9 (7.6–18.7)		4.6 (2.8–13.1)
200–349	11.8 (8.7–18.2)		6.1 (4.1–12.2)
<200	10.5 (8.2–14.5)		7.1 (5.4–10.1)
Chronic kidney disease***		0.84		0.39
Stage 1 (normal)	10.7 (8.3–15.0)		5.5 (4.0–8.7)
Stage 2 (mild)	11.2 (8.8–15.4)		7.3 (5.5–10.8)
Subsequent cognitive decline‡		0.02		0.05
Present	13.0 (10.4–17.4)		7.6 (5.8–10.8)
Absent	8.4 (6.4 –12.3)		5.0 (3.5–8.8)

Abbreviations: ARV, antiretroviral; ART, antiretroviral therapy; INSTI, integrase strand transfer inhibitor; NNRTI, non-nucleoside reverse transcriptase inhibitor; PI, protease inhibitor; TAF, tenofovir alafenamide; TDF, tenofovir disoproxil fumarate.

^*p-values indicate the statistical significance of differences between subgroups.

^**Ever use of any of stavudine (d4T), dideoxycytidine (ddC), or didanosine (ddI).

^***Chronic kidney disease stages are defined based on estimated glomerular filtration rate (mL/min/1.73 m²) as follows: Stage 1, ≥90 without evidence of kidney damage; Stage 2, 60–89 without evidence of kidney damage.

^‡Defined as worsening neuropsychological test performance following a cognitively unimpaired baseline. [26]

Among the remaining 11 covariates, two demonstrated statistically significant associations with the brain-age gap, each with only one of the two reference standards. Using Simrèn et al. standards, participants with a current CD4+ count ≥500 cells/μL had a significantly greater brain-age gap than PWH with CD4+ count <500 cells/μL. Using the Bornhorst et al. values, male sex at birth was associated with a significantly greater brain-age gap than female sex at birth. For both covariates, the direction of association was similar when using the other reference standard, although the differences were not significant.

Discussion

In a cohort of 340 cognitively unimpaired PWH on ART, we quantified plasma NfL and compared values to published age-stratified reference standards. This allowed us to estimate a brain-age gap, reflecting the degree to which participants exhibited an increased NfL-derived brain age relative to their chronological age, using NfL as a proxy for neuroaxonal damage. Our analyses revealed an estimated brain-age gap of 11.0 years (95% CI: 9.1–13.7 years) with Simrén et al. standards and of 6.4 years (95% CI: 5.1–8.5 years) using Bornhorst et al. standards. Across both reference standards, the brain-age gap was significantly larger in PWH with subsequent cognitive decline (Simrén: 13.0 vs. 8.4 years; Bornhorst: 7.6 vs. 5.0 years). We identified two additional statistically significant covariates, each observed with only one of the reference standards, which warrant further investigation in future studies but may represent chance findings given the number of statistical tests performed.

Our findings add to mounting evidence of premature aging in some PWH across multiple biological systems. Epigenetic analyses using a DNA-methylation clock demonstrate an approximately 12-year increase in cellular aging in PWH compared to chronological age [24], although the contribution of specific epigenetic sites varies between individuals, highlighting biological heterogeneity of aging [33]. Structural brain magnetic resonance imaging studies using machine-learning algorithms show PWH have brain structures that appear 2.2 years older on average, correlating with deficits in psychomotor speed, processing speed, executive function, and memory [20,21]. Complementing these structural measures, a machine-learning model applied to sleep EEG estimated an average 3.4-year increase in functional brain age. This was primarily driven by reduced slow-wave activity during non-REM sleep, a stage that supports memory and cognitive function [25]. Despite variability in effect sizes across modalities and cohorts, these convergent findings support a broader profile of premature molecular, structural, functional, and neuronal aging in PWH.

Although NfL reflects neuroaxonal injury rather than directly estimating biological age and is not equivalent to information from neuroimaging or EEG, its strong correlation with chronological age, particularly in older adults, makes it an informative indicator of overall neurological health [29]. By quantifying cumulative burden, plasma NfL can be interpreted within the broader concept of premature aging, similar to how lipid profiles summarize long-term cardiovascular risk, highlighting the potential for clinical implications, including quantifying risk for cognitive impairment in PWH. This elevated burden when participants were still cognitively unimpaired may be clinically meaningful, as significantly larger brain-age gaps were observed in PWH with subsequent cognitive decline, highlighting the potential utility of plasma NfL as a prognostic biomarker and monitoring tool for older PWH when appropriately adjusted for confounders.

Routine plasma NfL testing is not currently recommended in primary care, and current dementia guidelines do not recommend cognitive screening for asymptomatic individuals [35–38]. Nevertheless, plasma NfL is increasingly recognized as a broadly informative neurodegeneration (“N”) biomarker within the ATX(N) classification system, which organizes biomarkers of amyloid (A), tau (T), neurodegeneration (N), and other relevant Alzheimer’s disease (AD) processes (X) [35]. While AD-specific biomarkers such as Aβ42/Aβ40 and p-tau217 provide information about amyloid and tau pathology, respectively, NfL provides complementary information about overall neurodegenerative burden, albeit with lower specificity [35,39]. This broad sensitivity makes NfL a potentially useful marker for capturing cumulative neuronal injury across diverse conditions. Plasma NfL is pre-analytically robust, well-validated, and measurable in clinical laboratories [36]. When interpreted alongside AD-specific markers such as p-tau217, plasma NfL could help contextualize overall neurodegenerative burden and aid differentiation of rarer dementias from AD.

The use of plasma NfL in the primary care of PWH remains an area of ongoing discussion. Unlike age-associated dementias in people without HIV, where combinatorial markers are often validated by PET or CSF, standardized algorithms for PWH are not yet developed [26]. Elevated plasma NfL could reflect cumulative or historical neuronal injury, but its clinical interpretation is unclear. It may warrant a more intensive clinical evaluation, or require interpretation with a specific correction factor, as sometimes done with CSF NfL [10,26]. Existing NfL reference standards do not adequately account for factors common in PWH, such as chronic inflammation, ART exposure, and co-infections. If plasma NfL is clinically used, integrating it with neuropsychological testing and other diagnostics, such as neuroimaging, could improve risk stratification and management, although longitudinal studies are needed to clarify the stability and predictive value of NfL in PWH [26,36].

Beyond HIV, non-HIV exposures such as social determinants of health have been associated with elevated systemic inflammation and may contribute to neuronal injury [40–43]. Studies of PWH on ART and people without HIV on pre-exposure prophylaxis (PrEP) report elevated inflammation and CSF NfL compared to people without HIV not on PrEP, suggesting that non-HIV factors influence NfL levels [44]. Similarly, PWH on ART and lifestyle-matched people without HIV show comparable monocyte activation and inflammation [45]. These findings underscore the importance of considering cumulative neuronal injury alongside aging, immunological, behavioral, and socioeconomic factors when assessing brain health in PWH. Psychosocial resilience and stress-related pathways, including hypothalamic-pituitary-adrenal axis regulation, may also contribute to neuroinflammatory processes and represent important areas for future investigation. Proactive interventions, including cognitive rehabilitation or lifestyle modifications, could help maintain neurocognitive health, support effective clinical communication, and mitigate long-term impact of cognitive decline in PWH [46]. Incorporating plasma NfL into risk assessments alongside neuropsychological testing or imaging could support personalized, biomarker-informed monitoring and earlier identification of individuals at heightened risk for cognitive decline.

Limitations

There are several limitations to our analysis. First, we did not analyze NfL in people without HIV, preventing the development of internal reference values for a general population in our own laboratory. Although both reference cohorts were described as “neurologically healthy” and “cognitively normal,” it is unclear whether PWH were specifically identified and excluded. If PWH were included, the reference standards would reflect a general population rather than a population without HIV, potentially leading to a slight underestimation of the true brain-age gap. Second, the relatively broad age ranges reported in the reference values (e.g., 50–<60 years) may affect the estimated brain-age gaps, as the same upper 95% reference value may not accurately represent both a 51-year-old and a 59-year-old. However, age-specific reference values are not available and would be cumbersome to implement clinically. Third, one reference standard (Simrén et al.) was derived from a non-U.S. population, which may differ from our U.S.-based cohort in factors that could affect NfL levels. To reduce confounding, we limited our analysis by excluding participants with factors known to affect NfL (e.g., chronic kidney disease or baseline cognitive impairment). Finally, in addition to potential differences between the populations included in the reference standards and our PWH cohort, using external reference values introduces potential technical assay variability between laboratories and batches, which may affect comparability. The use of a standardized, commercially available kit helped reduce this assay-related variability.

Conclusion

This study highlights that plasma NfL levels are higher in cognitively unimpaired PWH compared to cognitively unimpaired individuals from the general population based on published reference standards; these findings support the hypothesis that premature brain aging occurs in PWH. The increase in brain-age gap using NfL values was greater in PWH who declined on subsequent cognitive testing, suggesting plasma NfL may detect early neuroaxonal injury underlying premature brain aging and cognitive trajectories in aging PWH. Future research should leverage pooled NfL data from independent global cohorts, using assays comparable to those used in this analysis, to validate these findings in larger, more diverse populations of PWH. Such efforts are essential to confirm the generalizability of these results across different demographic and clinical subgroups. Ultimately, longitudinal studies are needed to determine if NfL can serve as a reliable surrogate marker to be used in neuroprevention trials and to determine whether targeted clinical interventions can successfully slow or halt premature brain aging.