People living with HIV (PLWH) are living longer, healthier lives due to the effectiveness of antiretroviral therapy (ART) and sustained viral suppression.^1–4 Studies conducted in Europe and North America indicate that individuals who initiate ART early and achieve adequate immune recovery can have a life expectancy comparable to that of the general population.^1–4 Some studies highlight survival rates extending into the late 70s for those who sustain long-term viral suppression.^1,4 However, as the majority of PLWH enter their 50s and 60s, attention is increasingly being given to the intersection of HIV and brain health, with growing concern about cognitive decline driven by chronic effects of HIV infection, immunological dysfunction and ageing process.^5–7

Chronological age alone does not fully capture individual differences in ageing, as some PLWH experience premature (earlier onset), accentuated (higher frequency of occurrence) or accelerated (faster progression) ageing patterns.^8–14 Although the prevalence of HIV-associated dementia has reduced substantially in the ART era, milder HIV-related cognitive impairment persists.^7,15,16 As PLWH age, the risk for Alzheimer’s disease (AD) and other degenerative disorders is likely to become more imminent and clinically salient compared with people living without HIV (PLWoH).¹⁶

There remains a gap in our understanding of causal factors that drive cognitive impairment in PLWH, even with effective ART. Multiple mechanisms have been proposed to underlie cognitive impairment in PLWH, including chronic neuroinflammation, neuronal injury, persistent peripheral immune activation and vascular dysfunction.^5,17 These processes are usually investigated in conjunction with biomarkers, such as acute-phase reactants (markers of non-specific inflammation), tryptophan catabolites (markers of indoleamine 2,3-dioxygenase pathway activation), neurofilament light chain (NfL; marker of neuronal injury), monocyte activation markers such as soluble cluster of differentiation 14 (CD14) and cluster of differentiation 163 (CD163), and endothelial activation markers such as intercellular adhesion molecule 1 and vascular cell adhesion molecule 1, which may reflect active central nervous system (CNS) injury.^5,17–19 However, the pathophysiology remains incompletely understood, and no targeted pharmacologic interventions adjunctive to ART are approved to treat cognitive disorders in PLWH.

Currently, a diagnosis of cognitive impairment in PLWH relies primarily on comprehensive neuropsychological (NP) testing, which is time-consuming, resource-intensive and may not be accessible in all clinical settings.²⁰ Furthermore, symptoms of cognitive impairment in PLWH, such as deficits in attention, executive function and memory, can mimic early-stage AD and other neurodegenerative conditions, making differential diagnosis more challenging.^21–24 These limitations underscore the need for accessible, scalable diagnostic tools that can support earlier detection, monitor disease progression and help differentiate cognitive impairment related to HIV from other aetiologies.

Of the numerous biomarkers that have been studied to this end, NfL and glial fibrillary acidic protein (GFAP) have more recently been investigated in largely virally suppressed cohorts.^25–28 These markers may reflect underlying neuronal and astrocytic damage, respectively, and could serve as indicators of subclinical neurodegenerative processes.^29,30

This review synthesizes findings from two recent US cohort studies examining the association between plasma levels of NfL and GFAP and cognitive impairment in PLWH. The first analysis examines cognitive performance longitudinally across multiple timepoints, with a single baseline measurement of the biomarkers.³¹ The second analysis is a longitudinal assessment of NfL and GFAP at two timepoints and their association with cognitive performance.³² Here, we highlight key similarities and differences between the two recent studies within the context of current literature and discuss future directions to optimize the clinical utility of these biomarkers for monitoring cognitive wellbeing in PLWH.

Biomarker insights: Neurofilament light chain and glial fibrillary acidic protein in ageing people living with HIV

Neurofilament light chain: Evidence from recent studies

NfL, a structural protein in myelinated axons, is released into cerebrospinal fluid (CSF) and subsequently into the bloodstream following neuronal injury.²⁹ NfL levels in CSF are higher than in blood and more reflective of CNS pathology, making it a sensitive marker for neuronal damage in the brain and spinal cord.³³ Its implementation in research and clinical use is limited by the invasive nature of lumbar puncture. With advancements in technology that address the challenge of NfL concentrations being approximately 50 times lower in blood than in CSF, there is growing interest in blood-based NfL as a less-invasive alternative for detecting neuronal injury-related cognitive impairment in PLWH.²⁶ Multiple studies across PLWH and other populations show moderate-to-strong correlations between blood and CSF NfL, supporting the use of plasma NfL measurements as a surrogate marker for CNS pathology.^25,34–37 For instance, data from the COmorBidity in Relation to AIDS (COBRA) cohort of virally suppressed PLWH suggest moderate correlations between plasma and CSF NfL levels (r=0.52), while a separate study of both virally suppressed and unsuppressed PLWH showed strong correlations (r=0.89).^26,37

Evidence from multiple cohorts suggests that higher plasma NfL levels are associated with poorer cognitive performance in PLWH. In a USA–based cohort, higher plasma concentrations of NfL were cross-sectionally associated with worse cognitive performance.²⁵ Specifically, each 10-fold increase in plasma NfL corresponded to an 11.5-point reduction in NP test scores.²⁵ Notably, ART initiation in the same cohort led to reductions in plasma HIV viral load and NfL levels and improved cognitive performance, suggesting that uncontrolled HIV replication could contribute to ongoing neuronal injury and subsequent cognitive deficits.²⁵ Yet, individuals on ART continue to experience milder forms of cognitive impairment, necessitating ongoing investigation of NfL as a biomarker in PLWH on ART.¹⁵ Supporting this, two US-based cohorts of virally suppressed PLWH found that people with cognitive impairment had higher plasma NfL levels compared with those without cognitive impairment, with small-to-moderate effect sizes (Cohen’s d=0.21 and 0.45, respectively).^9,28

To differentiate ongoing neuronal injury from prior damage before ART initiation, longitudinal analyses provide particularly valuable insights. One recent US-based longitudinal study reported that serum NfL levels increased by 46.5% over 8 years in women living with HIV (WLWH), compared with a 24.4% rise in women living without HIV (WLWoH), suggesting potential ongoing neuroaxonal damage despite viral suppression.³⁸ Additionally, elevated baseline NfL levels were significantly associated with poorer processing speed in WLWH, with a β of -0.22 standard deviations per unit increase in serum NfL.³⁸

While plasma NfL offers a less-invasive method than CSF analysis or is less expensive than magnetic resonance imaging (MRI) for detecting neuroaxonal damage, its interpretation remains complex. Plasma NfL is not entirely CNS specific; elevations may reflect peripheral neuropathy, which is common among PLWH, and systemic disorders such as renal or cardiovascular disease.^39–41 These factors, along with assay platform variation, can make it challenging to compare results across studies and clinical practices. Nevertheless, plasma NfL remains a potentially valuable tool, offering complementary information to screen individuals at risk of ongoing neuronal damage, monitor longitudinal changes and differentiate between active CNS injury and past static damage.

Glial fibrillary acidic protein: A window into astrocytic injury in HIV

GFAP, a marker of astrocytic injury, has been studied in neurological conditions such as traumatic brain injury, AD and multiple sclerosis.^30,42–44 In the context of PLWH, GFAP has been evaluated as a potential CSF and blood biomarker for cognitive impairment.^45–48 Unlike NfL, whose blood levels may be confounded by peripheral neuropathy, as shown in an individual cohort and in a broader systematic review of neuropathy biomarkers, some studies report that blood GFAP shows stronger associations with CNS pathology than CSF GFAP, suggesting that plasma measurements may reliably indicate astrocytic injury.^39,49–51

Several studies have reported elevated GFAP levels in plasma and CSF in PLWH compared with PLWoH.^45,46,52 A study examining neuroimaging and GFAP found that plasma GFAP correlated with both cognitive performance and white matter microstructural abnormalities, as reflected by modest-to-large correlations with MRI diffusion metrics in PLWH.⁴⁸ A large CSF proteomic survey of PLWH found GFAP levels to be comparable with PLWoH controls in both treated and untreated groups, but markedly elevated in those with severe HIV-related cognitive impairment.⁵³ Studies analysing extracellular vesicle (EV)-associated GFAP (EV-GFAP) reinforce a graded relationship between GFAP and cognitive-impairment severity.^47,54 One study reported significantly higher plasma and CSF EV-GFAP levels in PLWH with severe cognitive impairment compared with unimpaired individuals, with a subsequent analysis confirming this trend in CSF EVs.^47,54 In contrast, other studies suggest that GFAP levels do not statistically differentiate between mild cognitive impairment and unimpaired PLWH, indicating that subtle cognitive changes may not be captured by GFAP measurements.^45,46,52 For example, a study restricted to mild impairment found no significant differences in EV-GFAP between groups.⁵² Collectively, these findings suggest that the utility of soluble or EV-associated GFAP in detecting milder forms of cognitive impairment in virally suppressed cohorts remains unclear.

Overall, while GFAP reflects astrocytic injury, a key component of HIV neuropathogenesis, its clinical implementation as a standalone biomarker for cognitive impairment appears limited, particularly compared with NfL or AD biomarkers like phosphorylated tau217 (p-tau217).^28,55 Future investigations would be valuable in assessing GFAP within multi-biomarker panels, including as a complement to neuroimaging or to enhance cohort selection, stratification and characterization of cognitive-impairment biotypes, ultimately supporting more targeted interventions in PLWH.

Comparing two cohorts in the USA: Insights into plasma biomarkers and cognition

The HAILO cohort: Plasma biomarkers and cognitive outcomes (Mukerji et al., 2025)³¹

The HAILO (HIV Infection, Aging, and Immune Function Longitudinal Observational) cohort was a prospective, multi-centre observational study of 1,035 older PLWH ≥40 years of age, who initiated ART through an AIDS Clinical Trials Group (ACTG) clinical trial and were followed longitudinally with clinical, behavioural and immunological parameters. At semi-annual visits, data were collected through medical chart abstraction, questionnaires, anthropometric measurements, neurocognitive evaluation and laboratory testing. The study was a subset of the HAILO cohort, which assessed 503 participants aged 45 years who were virally suppressed (HIV RNA, <200 copies/mL) (Table 1) and had repeat NP tests; plasma NfL and GFAP were assessed at the earliest available plasma sample on or after age 45 years.^31,32

The study addressed two key questions.

• Are baseline plasma levels of NfL and GFAP associated with cognitive performance in virally suppressed PLWH?

• Are baseline plasma levels of NfL and GFAP associated with future cognitive decline in virally suppressed PLWH?

Cross-sectionally, higher levels of both biomarkers were associated with worse global cognition, with a stronger association observed for NfL. Specifically, a one-unit increase in log-transformed NfL corresponded to a -0.76 decrease in NP composite score (β=-0.76, p<0.01), while log-transformed GFAP was associated with a -0.44 decrease (β=-0.44, p=0.02) (Table 2).^31,32 These findings suggest that markers of both neurodegeneration and astrocytic injury are related to cognitive performance in PLWH.^31,32

Longitudinally, over a median follow-up of approximately 6 years, higher baseline NfL was significantly associated with greater cognitive decline (β=-0.08 NP composite score units/year, p<0.01). However, baseline GFAP levels were not significantly associated with longitudinal decline (β=-0.03, p=0.08). Individual neuropsychological Z-score composite of four tests (NPZ-4) scores showed that higher baseline NfL was significantly associated with future declines in processing speed (Trail Making A: β=-0.10, p=0.01; Digit Symbol: β=-0.10, p=0.02) and learning (Hopkins Verbal Learning Test-Revised: β=-0.08, p=0.04). GFAP, on the other hand, was only associated with decline in processing speed (Digit Symbol: β=-0.11, p<0.01) (Table 2).

While statistically significant associations were observed for both biomarkers after adjusting for relevant covariates, the effect sizes for NfL and GFAP were modest, potentially limiting their clinical utility among PLWH on ART.

Women’s Interagency HIV Study cohort: Longitudinal plasma biomarker trends in women living with HIV (Li et al., 2023)³²

The Women’s Interagency HIV Study (WIHS), part of the MACS/WIHS Combined Cohort Study (ClinicalTrials.gov identifier: NCT00000797), is a long-standing, prospective cohort of WLWH and WLWoH across diverse urban centres in the USA. The study included 209 WLWH and 98 WLWoH aged ≥40 years. The WLWH group was primarily composed of women from racial and ethnic minority backgrounds, with 72% on ART and 87% achieving viral suppression (HIV RNA, <80 copies/mL) (Table 1).

The study addressed two key questions.

• Are plasma levels of NfL, GFAP and AD biomarkers, including amyloid beta 40 (Aβ40) and amyloid beta 42, total tau (t-tau) and phosphorylated tau231, associated with poor cognitive performance in WLWH and WLWoH?

• Do 1-year changes in these biomarkers correlate with poor cognitive performance?

This review focuses on findings specific to WLWH. It should also be noted that, since the primary scope of this review is NfL and GFAP, AD biomarker results from WIHS are summarized only as contextual information and are not directly compared with HAILO.

At baseline, WLWH exhibited worse performance in memory, learning, verbal fluency and global performance compared with WLWoH, with elevated levels of NfL, GFAP and Aβ40, even after adjusting for clinically relevant covariates. Notably, baseline NfL and GFAP levels were not significantly associated with cross-sectional cognitive performance in WLWH, which contrasts with findings from the HAILO study. However, repeated measurement over 1 year suggested that elevations in plasma NfL were associated with poor performance in processing speed (Digit Symbol and Stroop Color Test: β=-0.05, p=0.03), while no significant associations were found between 1-year changes in GFAP and cognitive performance (Table 2). Additional findings from the study specific to WLWH showed that increases in Aβ40 over 1 year were significantly associated with poor learning (Hopkins Verbal Learning: β=-0.07, p=0.01), memory (Hopkins Verbal Learning Delayed Recall: β=-0.06, p=0.02) and global cognitive performance (Composite Score: β=-0.03, p=0.01), while increases in t-tau were related to poor executive function (Trail Making B: β=-0.34, p=0.04).

These findings suggest that non-specific biomarkers like NfL and AD-related biomarkers may provide complementary information in understanding cognitive impairment among ageing WLWH.

Although findings from WIHS support the idea of using plasma NfL, Aβ40 and t-tau for monitoring cognitive impairment among PLWH, the observed effect sizes were small, which may limit their clinical use.

Comparative synthesis: Key similarities and differences

Understanding the relationship between plasma NfL and GFAP and cognitive impairment in PLWH is complex, and both HAILO and WIHS substudies provided valuable but contrasting insights, shaped by unique study designs, participant characteristics and statistical strategies.

The HAILO substudy focused exclusively on PLWH, while the WIHS substudy broadened the scope by including WLWoH as a comparison group. This inclusion offered a deeper context for interpreting NfL and GFAP findings. However, the timing and frequency of assessments demonstrated fundamental differences between the two studies. In HAILO, plasma biomarkers were measured once at baseline, and then cognitive performance was assessed through repeated NP assessments over years. This design was well suited to assessing whether a single biomarker measurement of NfL or GFAP could predict long-term cognitive decline. WIHS, on the other hand, measured biomarkers at baseline and after 1 year, but assessed NP tests only once at the study’s end. This approach was more attuned to detecting short-term changes in biomarker values, rather than mapping cognitive trajectories over time. These methodological distinctions, while reflecting the specific aims of each study, make direct comparisons challenging.

Participant demographics and clinical profiles also differed markedly between the two cohorts (Table 1). The HAILO cohort was 80% male and 53% White, and all participants were on ART and virally suppressed. In contrast, the WIHS cohort consisted of women, was 53% Black and 72% were on ART. Such differences in gender, race, viral suppression and baseline cognitive performance introduce important variables that can influence both biomarker levels and cognitive outcomes. For example, gender and race may affect neuroinflammatory responses, while ART adherence can impact the degree of HIV control and, by extension, neurocognitive risk.^56–58

How cognitive function was measured further distinguished the two studies. In HAILO, a smaller set of four NP tests combined into a single composite score was used, providing a broad overview of cognitive status. In WIHS, seven cognitive domains were assessed, allowing for a more nuanced understanding of specific cognitive deficits. Interestingly, while HAILO found associations between biomarkers and global cognition, WIHS did not, a difference that may stem from how cognitive performance was aggregated and analysed from differences in sample sizes (n=503 HAILO; n=209 WHIS).

A key distinction between the studies lies in biomarker selection. The WIHS substudy included AD-related biomarkers such as Aβ40 and t-tau, enabling exploration of whether cognitive impairment in PLWH overlaps with or diverges from AD patterns. However, because cognitive decline in PLWH often fluctuates rather than follows a steady trajectory, single-time NP assessments may not fully capture these dynamics and remain a limitation for future research.⁵⁹ HAILO’s extended 6-year follow-up allowed for the detection of subtle, cumulative cognitive changes, most of which indicated stable performance, while WIHS’s shorter follow-up, paired with more frequent biomarker measurements, was better suited to identifying early signals before clinical symptoms emerged.

In summary, while HAILO and WIHS each contribute important pieces to our understanding of cognitive impairment in HIV and blood biomarkers, their differences in design, participant demographics, cognitive assessment and biomarker selection highlight the complexity of this field. Together, they point to the need for more integrated, longitudinal research that can capture the dynamic interplay between biomarkers, cognitive function and the evolving clinical landscape of HIV.

Discussion

The extended lifespan of PLWH has shifted clinical and research priorities towards understanding ageing-related comorbidities, particularly cognitive decline. While PLWH who initiate ART early approach life expectancies comparable to the general population, ageing introduces complex interactions between HIV-associated neuropathology, immunosenescence and neurodegenerative processes. Chronic neuroinflammation, neuronal injury and vascular dysfunction persist even in virally suppressed individuals, driving milder yet clinically significant cognitive impairment. As in the general ageing population, where plasma NfL and GFAP increase with age and are linked to cognitive decline and dementia, PLWH appear to experience a convergence of HIV-specific and age-related pathological processes.^{33,50,60–62} These overlapping influences underscore the importance of situating biomarker findings within the context of both HIV and healthy ageing trajectories.

The HAILO and WIHS substudies provide complementary insights into plasma NfL and GFAP dynamics. HAILO’s longitudinal NP assessments over 6 years demonstrated that baseline NfL was associated with global cognitive decline, particularly in processing speed and learning, while GFAP showed weaker associations with cognitive outcomes across studies, suggesting its role may be context specific. In contrast, WIHS’s repeated biomarker measurements and inclusion of AD-related markers showed that short-term NfL increases correlated with poor performance in processing speed, and Aβ40/t-tau changes were linked to memory and executive deficits. These findings suggest that NfL may serve as one predictor of gradual cognitive decline and be used as a screening or monitoring aid alongside neuroimaging, rather than as a standalone diagnostic tool, whereas AD biomarkers could capture distinct pathological processes. However, both studies reported small-to-modest effect sizes, emphasizing that standalone biomarkers are likely insufficient for clinical decision-making.

Differences in cohort composition and covariate adjustments further complicate comparisons. HAILO enrolled only PLWH and adjusted for CD4 nadir, neuropathy and antidepressant use, whereas WIHS included WLWH and WLWoH and modelled current CD4 count and depressed mood. Neither study appeared to control for duration of HIV infection, further complicating direct comparison. Such differences may influence observed associations and highlight the need for harmonized covariate strategies in future studies.

Sex differences are also important in studies of cognitive impairment among PLWH, with several reports showing higher prevalence in women.^63–65 One study found this disparity attenuated after adjusting for educational quality, while another study reported a threefold higher risk of impairment in women even after accounting for sociodemographic factors.^63,64 These findings suggest that both biological and sociocultural influences must be considered when interpreting sex differences in biomarker–cognition relationships. Given that WIHS enrolled only women and HAILO predominantly men, sex composition likely contributed to differences in biomarker associations.

Substance use, including injection drug use, is another potential confounder. Substance exposure has been associated with worse neurocognitive outcomes in PLWH and may independently elevate neuronal injury biomarkers such as NfL.¹⁵ Intravenous drug use has been linked to increased risk of HIV-associated cognitive decline, while broader studies in non-HIV populations demonstrate elevated NfL levels in individuals with substance use disorders.^66–68 Both HAILO and WIHS incorporated measures of substance exposure, but differences in how substance use was assessed and adjusted for could influence biomarker associations.

We anticipate that in the next 3–5 years, new longitudinal, multi-modal frameworks to address gaps in knowledge about cognitive disorders in PLWH will be used in trials and potentially clinics. Combining NfL with AD-specific biomarkers (e.g. p-tau217), inflammatory markers, such as monocyte chemoattractant protein-1 (MCP-1)/chemokine (C-C motif) ligand 2 (CCL2), monocyte chemoattractant protein-2 (MCP-2)/chemokine (C-C motif) ligand 8 (CCL8), interleukin 23 (IL-23), tissue inhibitor of metalloproteinases 1 (TIMP-1), tumor necrosis factor alpha (TNF-α) and interleukin 1 beta (IL-1β), and neuroimaging may enhance diagnostic specificity.^69–73 Recent studies in PLWH have shown that inflammatory markers such as CCL8, TIMP-1 and IL-23 are associated with cognitive domains including memory, processing speed and executive function, while MCP-1 has been linked to HIV-associated cognitive decline clusters.^71,72 In AD cohorts, p-tau217 and NfL combinations improve prediction of cognitive decline.^69,70 Machine learning-based approaches that integrate these multimodal data streams may help identify biomarker clusters predictive of distinct cognitive decline subtypes in HIV. Stratifying cohorts by sex, age, viral suppression status, comorbidities and ART regimens will enhance biomarker interpretation, and validating population-specific reference ranges is essential for clinical translation.

Although this review focuses on two large longitudinal cohorts, their limited scope highlights the need for more integrative studies. Clinically, plasma NfL’s consistent associations with processing speed, albeit modest, suggest utility in risk stratification or therapeutic monitoring, whereas GFAP’s role may remain secondary unless paired with complementary tools. Establishing validated thresholds and assessing biomarker responsiveness to interventions will be critical next steps. As PLWH age into longevity, resolving these challenges will be pivotal to safeguarding cognitive health in this resilient yet vulnerable population.