Association of glymphatic dysfunction, free water, white matter integrity and long-term memory performance in aging autistic adults

Abstract

Background

Autistic adults are at elevated risk of accelerated cognitive aging and Alzheimer’s disease and related dementias (ADRD), yet the underlying neurobiological mechanisms remain poorly understood. Dysfunction in the glymphatic system—a brain-wide network responsible for clearing waste via interstitial fluid flow—may contribute to this vulnerability by promoting extracellular free water (FW) accumulation and white matter (WM) degeneration.

Methods

A total of 113 autistic and 90 age- and sex-matched neurotypical (NT) adults (aged 18–71 years) underwent multimodal MRI scanning and episodic memory assessments. Diffusion tensor image analysis along the perivascular space (DTI-ALPS) index, alongside FW maps, and fractional anisotropy (FA) maps were computed for each participant. Group comparisons, correlations, and mediation analyses were performed.

Results

Autistic adults showed significantly lower DTI-ALPS values, higher fornix FW, lower fornix FA, and poorer episodic memory scores compared to NT adults. Age-related hippocampal FW accumulation was more pronounced in autistic adults. Mediation analyses revealed that fornix FW mediated the relationship between DTI-ALPS and both fornix FA and hippocampal FW. Long-term episodic memory scores correlated with fornix FA, as well as whole-brain gray matter FW and WM FA in autistic adults.

Limitations

The cross-sectional design precludes causal inference regarding glymphatic function, free water accumulation, WM integrity, and cognition. In addition, our sample was not evenly balanced by sex and excluded individuals with co-occurring intellectual disability, which may limit generalizability to the broader autistic population.

Conclusions

Our results suggest that glymphatic dysfunction and FW accumulation may contribute to aberrant WM microstructure and episodic memory challenges in autistic adults across a broad age range. These findings point to potential biomarkers for identifying and intervening in the cognitive aging process in autism.

Background

Autism is a neurodevelopmental condition characterized by challenges in social communication, restricted interests, and repetitive behaviors [1, 2]. While much of the research on autism has historically focused on early development and childhood, there is a growing recognition of the need to understand how this condition evolves across the lifespan, particularly during aging [3]. As the population of autistic individuals reaches middle and older age, emerging evidence suggests that the aging process in autism may differ significantly from that in neurotypical populations, involving accelerated cognitive decline [4], alterations in brain structure and function [5], and increased likelihood of developing Alzheimer’s disease (AD) [6]. However, the mechanisms underlying these differences remain poorly understood, necessitating advanced cognitive and neuroimaging studies to elucidate the aging trajectory in autism.

There is a growing body of literature implicating increased risk of AD in autistic adults. For instance, Medicaid records show autistic adults are 2.6 times more likely to receive early-onset AD and Related Dementia (ADRD) diagnoses [6], compared to non-autistic adults, and Medicare records indicate a 35% prevalence of ADRD in elderly autistic adults [7]. Caregivers of older adults with cognitive impairment have reported those with high autistic traits to have younger age at dementia onset and more severe impairment [8]. Furthermore, healthcare records in California show higher prevalence of dementia in autistic adults [9], compared to non-autistic counterparts. Another large study also found increased prevalence of cognitive disorders, including dementia, in autistic adults on Medicaid [10]. For primary data collection, our group was the first to publish longitudinal cognitive aging trajectories in middle-age and older autistic adults, which identified accelerated decline of long- and short-term memory and hippocampal volume, compared to matched controls [4, 11]. A longitudinal autism and aging cohort study in the Netherlands found heterogeneity in aging outcomes for autistic adults that produce distinct subgroups indicating risk and resilience phenotypes [12] (but no group differences in longitudinal cognitive trajectories between autistic and non-autistic adults [13]). Further research is needed to determine the biological indicators of adverse cognitive aging outcomes in autistic adults.

Emerging evidence suggests that disruptions in the glymphatic system may play a role in the accelerated cognitive aging among autistic adults. Glymphatic function, essential for removing neurotoxic proteins like β-amyloid and tau—hallmarks of ADRD pathology—and pro-inflammatory cytokines, diminishes with normal aging [14] and exhibits marked deficits in individuals with ADRD [15, 16]. The glymphatic system relies heavily on slow-wave sleep [17, 18], a phase often disrupted in autistic individuals across the lifespan [19]. In high-risk infants and young autistic children, studies have identified increased extra-axial cerebrospinal fluid (CSF) [20, 21] and enlarged perivascular spaces [22, 23]—key conduits of glymphatic flow—as early markers of atypical neurodevelopment, with these anomalies correlating with autism severity and later sleep deficits [21, 23]. While these findings shed light on the role of glymphatic dysfunction in early brain development in autism, its impact on cognitive aging and dementia risk in autistic adults remains largely unexplored. Investigations in this regard may help elucidate neural mechanisms underlying the accelerated cognitive aging and elevated ADRD vulnerability in this population.

The diffusion tensor image analysis along the perivascular space (DTI-ALPS) index has recently been suggested as a noninvasive imaging marker of glymphatic activity [24], quantifying the diffusivity of water molecules along perivascular pathways that support interstitial fluid transport and metabolic waste clearance [25]. Reduced DTI-ALPS values indicate diminished perivascular fluid movement and are increasingly interpreted as evidence of glymphatic dysfunction [26]. By linking microstructural diffusion properties to physiological processes of brain fluid homeostasis, DTI-ALPS provides a valuable tool for investigating age- and disease-related changes in glymphatic efficiency. Importantly, this index has been successfully applied to assess glymphatic dysfunction across multiple neurological disorders, including Alzheimer’s disease [27] and frontotemporal dementia [16], providing mechanistic insights into how impaired perivascular fluid transport contributes to neurodegeneration.

Extracellular free water (FW), derived from diffusion tensor imaging (DTI) data, has emerged as a sensitive marker of neurodegeneration, neuroinflammation, and cognitive decline in aging populations [28]. In neurotypical aging and mild cognitive impairment, elevated FW in white matter (WM) has been associated with poorer cognition and accelerated cognitive decline [28–30]. Notably, increases in FW often co-occur with WM microstructural abnormalities and are thought to arise, in part, from impaired interstitial fluid clearance [27, 31], pointing to glymphatic system impairment. Furthermore, FW in WM has been shown to mediate the relationship between glymphatic function and cognition across multiple aging cohorts [32]. These findings have particular relevance for autistic adults, who show elevated free water in both frontal transcallosal WM [33] and the hippocampus [4], with hippocampal FW levels correlating with accelerated memory decline in midlife [4]. However, no studies to date have directly examined the inter-relationships among FW accumulation, WM integrity, and glymphatic system efficiency in this population. Elucidating these relationships may uncover autism-specific pathways of brain aging and identify potential targets for early detection and intervention.

The primary aim of this study was to characterize the associations among glymphatic dysfunction (measured by DTI-ALPS index), FW accumulation, WM integrity (measured by fractional anisotropy [FA]) and long-term memory performance in 113 autistic adults compared with 90 neurotypical (NT) adults across a broad age range (18–71 years). Based on existing literature [4, 11, 33, 34], we hypothesized that (1) compared with NT adults, autistic individuals would exhibit lower long-term memory scores, DTI-ALPS values, higher FW in the hippocampus and fornix, lower fornix FA, and a steeper age-related relationship in these measures; (2) DTI-ALPS would be associated with hippocampal and fornix microstructural features (FW and FA), which would in turn mediate the relationship between DTI-ALPS and long-term memory performance in autistic adults. We also explored global measures of gray matter (GM) and WM FW and WM FA in these analyses to assess the specificity of effects to the hippocampal system.

Methods

Participants

The study included 203 participants, consisting of 113 autistic adults (75 male, 38 female) and 90 NT adults (52 male, 38 female), aged 18 to 71 years. Participants were recruited between 2014 and 2024, with some overlap from prior studies. All autistic participants had their diagnosis formally verified at the Southwest Autism Research and Resource Center (SARRC) using the Autism Diagnostic Observation Schedule-2 (ADOS-2) Module 4 as well as a brief psychiatric history interview, and Autism Diagnostic Interview-Revised if needed. Using a DSM-5 checklist, a psychologist with 25 years of autism spectrum disorder (ASD) diagnostic experience reviewed assessment results and confirmed that participants with ASD met diagnostic criteria. All participants underwent identical screening and enrollment procedures and met the following inclusion criteria: (1) a score above 70 on the Kaufman Brief Intelligence Test (KBIT-2), (2) a score greater than or equal to 26 on the Mini-Mental State Exam, and (3) no self-reported history of neurological illness, head injury with loss of consciousness, or known genetic disorders that affect brain and behavior. NT participants were required to have a self-report Social Responsiveness Scale–Second Edition (SRS-2) scaled score below 65, no self-reported suspected or established ASD diagnosis, and no first-degree relative with an ASD diagnosis. Participants in the ASD group were additionally required to have a self-report SRS-2 scaled score above 65, indicating elevated autism symptomatology. Then, all autistic participants had their diagnosis formally verified at the Southwest Autism Research and Resource Center (SARRC) using the Autism Diagnostic Observation Schedule-2 (ADOS-2) Module 4 as well as a brief psychiatric history interview, and Autism Diagnostic Interview-Revised if needed. Using a DSM-5 checklist, a psychologist with 25 years of autism spectrum disorder (ASD) diagnostic experience reviewed assessment results and confirmed that participants with ASD met diagnostic criteria. Autistic participants were primarily recruited via SARRC’s database of individuals across the lifespan who voluntarily enrolled at local community events and consented to be contacted for future studies. Participants were also recruited via distribution of flyers, media releases, and participant registries through Arizona State University (ASU) and Banner Alzheimer’s Institute. The study was approved by the ASU Institutional Review Board and all participants provided written informed consent.

Long-term memory assessment

To assess long-term memory performance, we utilized two standardized neuropsychological measures: the Rey Auditory Verbal Learning Test (AVLT) and the Visual Reproduction subtest of the Wechsler Memory Scale–Third Edition (WMS-III). The AVLT is a verbal learning and memory task in which participants are presented with a 15-word list across five learning trials (A1–A5), followed by an interference list and a delayed free recall trial. Long-term verbal memory was measured using AVLT-A7, which reflects the number of words accurately recalled after a 20–30-minute delay. The WMS Visual Reproduction (VR) subtest assesses visual memory by requiring participants to reproduce abstract geometric designs from memory both immediately and after a 30-minute delay (WMS-VR-II); the WMS-VR-II score serves as an index of long-term visual memory retention. These measures were selected because long-term memory is among the cognitive domains most vulnerable to age-related and pathological decline, including in conditions such as mild cognitive impairment and AD.

MRI acquisition

All MRI data were collected on a 3-Tesla Philips Ingenia MRI scanner at the Barrow Neurological Institute, St. Joseph’s Hospital and Medical Center, Phoenix, AZ. T1-weighted MRI data were acquired with a 3D magnetization-prepared rapid acquisition gradient echo sequence, using the following parameters: 256 × 256 in-plane resolution, a 240-mm field of view, and 170 sagittal slices of 1.2-mm thickness. DTI data were obtained using a gradient-echo echo-planar imaging sequence with these parameters: echo time/repetition time = 101/7850 ms, bandwidth = 2621 Hz/pixel, and voxel size = 1.41 × 1.41 × 3 mm. The DTI acquisition included 32 directions with a b-value of 2500 s/mm² in the axial plane and a 3-mm slice resolution.

DTI-ALPS analysis

The DTI-ALPS index was derived from DTI data to evaluate glymphatic function for each participant, as described in the publicly available code repository (https://github.com/gbarisano/alps) [35]. Specifically, the DTI data underwent denoising, Gibbs ringing artifact removal, and correction for head motion and eddy current distortions. Subsequently, FA and diffusivity maps along the x-, y-, and z-axes were generated with a diffusion tensor model [36] and transformed into the Montreal Neurological Institute (MNI) space using a combination of linear and nonlinear registration techniques. The projection fibers (superior corona radiata [SCR]) and association fibers (superior longitudinal fasciculus [SLF]) at the level of the lateral ventricle body were identified using the JHU-WM atlas [37]. Four spherical masks, each 5 mm in diameter, were placed bilaterally over the SCR and SLF regions and overlaid onto the diffusivity maps. Mean diffusivity values along the x-axis (Dx), y-axis (Dy), and z-axis (Dz) were extracted from these masks for both projection and association fibers, denoted as Dxproj, Dyproj, Dzproj, Dxassoc, Dyassoc, and Dzassoc, respectively. The DTI-ALPS index was calculated as the ratio [(Dxproj+Dxassoc)/(Dyproj+Dzassoc)] [24]. Using the JHU-WM atlas, we further calculated the global mean FA across all WM tracts and the mean FA specifically for the fornix (merging labels 6, 39, and 40) to evaluate their integrity.

Hippocampus segmentation

T1-weighted MRI data were processed using FreeSurfer (version 7.4.1; https://surfer.nmr.mgh.harvard.edu/) [38, 39] to obtain the bilateral hippocampal volumes and intracranial volume (ICV) for each participant. The normalized hippocampal volume was calculated as the ratio of the average bilateral hippocampal volume to the ICV. To create a hippocampal mask in MNI space, the T1-weighted image “MNI152_T1_1mm.nii.gz” was processed using FreeSurfer’s preprocessing pipeline. The bilateral hippocampi were extracted and merged into a single hippocampal mask.

FW mapping

All diffusion volumes were preprocessed using MRtrix3 (https://www.mrtrix.org) [40], including denoising with Marchenko-Pastur PCA, Gibbs-ringing suppression, correction for eddy-currents, susceptibility distortion, and subject motion, outlier slice (> 3 SD) replacement based on a weighted combination of neighboring non-outlier slices, and gradient-table re-orientation. Intensity inhomogeneity was removed, and brain masks were generated. FW maps were calculated from a bi-tensor model with custom Matlab scripts [41, 42]. In this model, FW refers to isotropically diffusing extracellular water, which is estimated as a separate compartment from the tissue-restricted diffusion signal. This is achieved using a mean diffusivity-based constraint, spatial regularization scaled by tissue-class priors, and the exclusion of voxels identified as CSF [43, 44]. We used FW maps in MNI space for subsequent analyses.

For each participant, we computed the global mean FW across all GM voxels delineated by the Automated Anatomical Labeling atlas [45], and mean hippocampal FW using the aforementioned hippocampal mask. Using the JHU-WM atlas, we computed the global mean FW across all WM tracts, as well as the mean FW specifically for the fornix.

MRI quality control

Intermediate processing outputs of diffusion MRI data were visually inspected in both native and MNI spaces to ensure data integrity, accurate preprocessing, and proper spatial alignment. FreeSurfer-derived subcortical segmentations were reviewed following the ENIGMA quality control protocol to confirm accurate hippocampal delineation. Participants with segmentation errors were reprocessed or excluded from further analyses.

Statistical analyses

Group differences in long-term memory scores and imaging parameters, as well as their relationships with age, were evaluated by fitting a linear model in R. The model included main effects of group, sex, and age, as well as an age-by-group interaction term. When the interaction term was not statistically significant, it was removed from the final model to improve model fit and reduce unnecessary complexity. We then used Spearman’s correlation coefficient to examine the relationship between DTI-ALPS and other imaging parameters in each group. To control for multiple testing, false discovery rate (FDR) correction was applied. To assess group differences in the association between DTI-ALPS and specific imaging parameters, we fit a linear model with main effects of group, age, sex, the imaging parameter, and an interaction term between the imaging parameter and group. Within the autistic group, we further explored the relationship between imaging parameters and long-term memory scores using Spearman’s correlation coefficient.

To explore the potential mechanisms underlying the associations among the imaging measures and long-term memory performance, we conducted a series of mediation analyses within the autistic individuals using the PROCESS macro (version 4.1) for SPSS (available at http://www.afhayes.com). Specifically, we first planned to test three serial mediation models (Model 6) to examine whether: (1) fornix FW and fornix FA, (2) fornix FW and hippocampal FW, and (3) whole-brain WM FW and WM FA sequentially mediated the association between the DTI-ALPS index and memory performance (WMS-VR-II). Bias-corrected 95% confidence intervals (CIs) for indirect effects were obtained using 5,000 bootstrap samples. If the total and serial indirect effects in these serial mediation analyses were not significant, we planned to further conduct simplified single-mediator models (Model 4) using the first three variables of the corresponding serial mediation model (DTI-ALPS → FW → FA for the first and third models, and DTI-ALPS → FW → hippocampal FW for the second model) to specifically evaluate whether FW mediated the association between DTI-ALPS and FA (or hippocampal FW in the second model).

Results

Demographic and neuropsychological differences

Autistic and neurotypical adults were well matched in demographic and cognitive characteristics (Table 1). There were no significant group differences in age (t = 0.44, p = 0.66), sex distribution (χ² = 1.23, p = 0.27), or cognitive ability as measured by the KBIT (t = − 0.53, p = 0.60). Compared to NT individuals, autistic adults demonstrated significantly lower scores on the WMS-VR-II (t=-2.46, FDR-corrected p = 0.015, Fig. 1A) and AVLT-A7 (t=-2.60, FDR-corrected p = 0.015, Fig. 1B). These models included the main effects of group, age, and sex, but did not include the non-significant age × group interaction.

Table 1.

Demographic and clinical data of the autistic and NT adults

	Autistic adults (n = 113)	NT adults (n = 90)	Statistics
Age, y	40.26 ± 15.86 (median 42.0, IQR 30.00)	41.28 ± 16.90 (median 43.0, IQR 31.75)	t = 0.44, p = 0.66
Sex, M/F	75/38	52/38	χ²=1.23, p = 0.27
KBIT	106.77 ± 15.28 (median 107.0, IQR 20.50)	107.83 ± 12.45 (median 108.0, IQR 17.00)	t=-0.53, p = 0.60
ADOS-2 SA	10.23 ± 3.03 (median 10.0, IQR 4.00)
Framewise Displacement	1.54 ± 0.42 (median 1.48, IQR 0.43)	1.61 ± 0.53 (median 1.51, IQR 0.52)	t=-1.14, p = 0.26

Note: KBIT= Kaufman Brief Intelligence Test; ADOS-2 SA = Social Affect score from the Autism Diagnostic Observation Schedule, Second Edition; Framewise displacement values were estimated from diffusion-weighted images, representing the head motion parameters; Values were presented as mean ± standard deviation (SD) with median and interquartile range (IQR) in parentheses. Continuous variables were analyzed using independent-samples t tests, while categorical variables were compared using chi-square tests

Fig. 1.

Differences in long-term memory scores and neuroimaging markers between autistic and NT adults. All p-values were corrected for multiple comparisons using the false discovery rate (FDR)

Differences in DTI-ALPS, FW and FA

Compared to NT individuals, autistic adults demonstrated significantly lower DTI-ALPS index (t=-2.50, FDR-corrected p = 0.013, Fig. 1C), higher FW (t = 2.53, FDR-corrected p = 0.012, Fig. 1D) and lower FA in the fornix (t=-2.98, FDR-corrected p = 0.006, Fig. 1E). These models included only the main effects of group, age, and sex; the non-significant age × group interaction was not included. A group-by-age interaction was evident for hippocampal FW (t = 2.30, FDR-corrected p = 0.044), with a stronger positive correlation between age and hippocampal FW observed in the autistic group (r = 0.37, p < 0.001, Fig. 1F) than the NT group (r = 0.25, p = 0.018). There were no significant group-by-age interactions for DTI-ALPS, fornix FW/FA, normalized hippocampal volume, GM FW or WM FW/FA. There were no significant group differences in global GM FW or WM FW/FA.

Associations among imaging parameters and memory performance

In both groups, the DTI-ALPS index was negatively correlated with GM (Fig. 2A) and hippocampal FW (Fig. 2B). There was a significant diagnosis group interaction for the relationship between DTI-ALPS and fornix FW (t=-2.02, p = 0.045), such that there was a significant negative correlation with fornix FW in the autistic group, and no significant correlation in the NT group (Fig. 2C). There was also a diagnosis group interaction for the relationship between DTI-ALPS and fornix FA (t = 2.11, p = 0.036), such that there was a stronger positive correlation in the autism group than the NT group (Fig. 2D). Additionally, the DTI-ALPS index was significantly correlated with normalized hippocampal volume (r = 0.29, FDR-corrected p = 0.0056), WM FW (r=-0.20, FDR-corrected p = 0.0588) and FA (r = 0.23, FDR-corrected p = 0.0228) in the autistic group, but not in the NT group (p > 0.05). There was no significant DTI-ALPS by group interaction for normalized hippocampal volume.

Fig. 2.

Associations of DTI-ALPS with FW and FA measures in autistic and NT adults. All p-values were corrected for multiple comparisons using the false discovery rate (FDR)

In autistic adults, the WMS-VR-II scores were significantly correlated with GM FW (Fig. 3A), as well as FA in WM (Fig. 3B) and the fornix (Fig. 3C). By contrast, the AVLT-A7 scores were not significantly correlated with any imaging parameters, although a trend toward a positive correlation with fornix FA was observed (p = 0.08; Fig. 3D).

Fig. 3.

Associations of long-term memory scores with FW and FA measures in autistic adults

Mediated relationships among imaging parameters and memory performance

In the serial mediation analysis that examined fornix FW and FA as sequential mediators of the association between DTI-ALPS and WMS-VR-II scores, the total effect of DTI-ALPS on WMS-VR-II was not significant (β = 0.134, p = 0.181). The serial indirect effect via fornix FW and fornix FA was also non-significant (β = 0.093, 95% CI [–0.008, 0.245]). Similarly, the serial mediation analysis testing fornix FW and hippocampal FW as sequential mediators revealed a non-significant total effect (β = 0.134, SE = 0.099, p = 0.181) and a non-significant serial indirect effect (β = 0.0231, 95% CI [–0.037, 0.098]). In the third serial mediation analysis, which tested whole-brain WM FW and WM FA as sequential mediators, both the total effect (β = 0.134, SE = 0.099, p = 0.181) and the serial indirect effect (β = 0.0228, 95% CI [–0.0023, 0.0761]) were non-significant.

As the serial mediation effects were non-significant in all three analyses, we conducted simplified single-mediator analyses by focusing on the first three variables of each pathway. In the analysis of DTI-ALPS → fornix FW → fornix FA, the bootstrapped indirect effect of DTI-ALPS on fornix FA via FW was significant (β = 0.083, 95% CI [0.031, 0.126]). In the analysis of DTI-ALPS → fornix FW → hippocampal FW, the bootstrapped indirect effect of DTI-ALPS on hippocampal FW via fornix FW was significant (β = − 0.100, 95% CI [–0.171, − 0.036]). In the analysis of DTI-ALPS → WM FW → WM FA, the bootstrapped indirect effect of DTI-ALPS on WM FA via WM FW was not significant (β = 0.016, 95% CI [–0.002, 0.039]). These results indicate that fornix FW significantly mediates the relationships between DTI-ALPS and both fornix FA (Fig. 4A) and hippocampal FW (Fig. 4B), whereas whole-brain WM FW does not mediate the association with WM FA.

Fig. 4.

Mediating role of fornix FW in the associations of DTI-ALPS with fornix FA and hippocampal FW in autistic adults. Analyses were based on 5000 bootstrap samples; 95% confidence intervals are reported. Path coefficients are non-standardized. Solid arrows indicate significant paths (*p < 0.05, **p < 0.01, ***p < 0.001). (A) DTI-ALPS → fornix FW → fornix FA. The total effect of DTI-ALPS on fornix FA was 0.123 (95% CI [0.071, 0.175]). The direct effect of DTI-ALPS on fornix FA, controlling for fornix FW, was 0.041 (95% CI [0.002, 0.080]). The indirect effect of DTI-ALPS on fornix FA through fornix FW was 0.083 (95% CI [0.031, 0.126]), indicating partial mediation. (B) DTI-ALPS → fornix FW → hippocampal FW. The total effect of DTI-ALPS on hippocampal FW was − 0.215 (95% CI [-0.299, -0.131]). The direct effect of DTI-ALPS on hippocampal FW, controlling for fornix FW, was − 0.115 (95% CI [-0.193, -0.037]). The indirect effect of DTI-ALPS on hippocampal FW through fornix FW was − 0.100 (95% CI [-0.171, -0.036]), indicating partial mediation

Discussion

In this study, we examined the association among glymphatic dysfunction, FW accumulation, WM integrity, and long-term memory performance in autistic adults compared with matched NT adults. Autistic adults exhibited lower scores on the WMS-VR-II and AVLT-A7, indicating challenges in long-term visual and verbal memory, alongside lower DTI-ALPS index, greater FW, and lower FA in the fornix, suggesting altered glymphatic function and WM integrity. A notable diagnosis-by-age interaction highlighted a stronger positive relationship between increasing age and higher hippocampal FW in the autistic group, potentially reflecting accelerated neurodegenerative processes. The DTI-ALPS index showed distinct correlation patterns with other structural brain metrics. We observed negative correlations between DTI-ALPS and GM FW (hippocampal and whole-brain) and positive correlations with fornix FA in both groups. For DTI-ALPS and fornix measures (FW and FA), we observed group interactions driven by stronger relationships between these measures in the autism group than the NT group. Mediation analyses further confirmed that fornix FW significantly mediated relationships between DTI-ALPS and FA, and DTI-ALPS and hippocampal FW, underscoring the role of glymphatic dysfunction in hippocampal system structural alterations across the adult lifespan. Cognitive performance, particularly long-term verbal memory, was closely linked to global measures of GM FW, global WM FA, and fornix FA in autistic adults. Collectively, these findings suggest that autistic adults may experience distinct neurobiological aging trajectories, characterized by compromised glymphatic function, greater FW, and disrupted WM integrity, which may contribute to memory challenges and accelerated brain aging.

The lower DTI-ALPS index observed in autistic adults highlights significant glymphatic dysfunction, which may contribute to a unique neurobiological aging profile in this population. To our knowledge, this study is the first to characterize glymphatic dysfunction in autistic adults using DTI-ALPS. Our finding aligns with previous reports of lower DTI-ALPS indices in autistic children [34, 46]. Elevated extra-axial CSF volumes in infants and toddlers at high risk for autism suggest early disruptions in CSF dynamics that may compromise glymphatic function [20, 21]. Enlarged perivascular space, particularly in the dorsolateral prefrontal cortex, are associated with autism severity and sleep disturbances, indicating potential constraints on glymphatic clearance pathways [22, 23, 47]. Taken together, these findings suggest a lifelong persistence of glymphatic dysfunction in autism.

The mechanisms underlying the glymphatic dysfunctions are not yet fully understood, but several inter-related factors may contribute. First, excessive extra-axial CSF in early development could compromise fluid dynamics critical for glymphatic clearance [48]. Second, enlarged perivascular space may obstruct perivascular pathways, hindering waste removal [49]. Third, sleep dysregulation, prevalent in autism [50, 51], likely impairs glymphatic efficiency, as clearance relies on sleep-dependent CSF flow [17, 52]. Fourth, cerebral blood flow deficits, commonly reported in autism, may reduce arterial pulsatility essential for glymphatic transport [14, 53], with prior studies documenting hypoperfusion in frontal and temporal regions. Notably, the glymphatic system shows a gradual decline in function with normal aging [14, 54], characterized by slower clearance of metabolic waste and structural changes such as enlarged perivascular space. This reduction in clearance capacity may contribute to the accumulation of neurotoxic proteins and subtle cognitive changes seen in older adults [54]. In ADRD, these impairments become more severe, with disrupted perivascular pathways playing a key role in the buildup of amyloid and tau pathology and associated cognitive decline [27, 29]. However, the specific role of glymphatic alterations in the aging trajectory of autistic adults remains unclear. Although current evidence suggests persistent glymphatic impairment throughout the lifespan in autism, how these disruptions shape age-related behavioral and neurobiological changes—or whether they represent a core mechanism underlying atypical aging in this population—remains to be determined.

The absence of significant differences in global measures of GM FW and WM FW/FA between autistic adults and NT individuals suggests that autistic aging may not involve the diffuse neurodegenerative changes commonly observed in dementia-related conditions like AD. This preservation of global brain integrity indicates that aging in autism may follow a distinct trajectory, with less emphasis on widespread tissue deterioration and greater prominence of targeted, region-specific alterations. Notably, autistic adults exhibited significantly more FW and lower FA in the fornix, a critical WM tract essential for hippocampal connectivity and memory function [55, 56]. These localized differences likely reflect aberrant WM microstructure driven by mechanisms such as neuroinflammation, axonal injury, or disrupted fluid clearance, all of which are implicated in age-related neuropathology [56]. Furthermore, the group-by-age interaction showing a stronger positive correlation between age and hippocampal FW in autistic adults suggests a possible accelerated aging process in memory-related circuits [41]. Elevated FW has been linked to tau pathology and neuroinflammation in AD [57], emphasizing its potential as an early marker of pathological changes. Further, greater brain FW has been associated with cognitive decline and executive dysfunction in normal aging [56, 58]. This pattern of selective vulnerability in the hippocampus and fornix in aging autistic adults, contrasted with preserved global metrics, emphasizes the need to focus on specific neural pathways in autistic aging research. FW imaging emerges as a sensitive biomarker for detecting these early, localized changes, providing critical insights into the mechanisms of cognitive decline in autism and supporting the development of targeted interventions to mitigate cognitive challenges.

In this study, we observed that the DTI-ALPS index was significantly associated with FW content and WM microstructural integrity, particularly in the autistic group. These findings are consistent with prior research highlighting FW as a sensitive biomarker reflecting neuroinflammation and tissue degeneration [28, 59]. In autistic aging, greater FW may be partially caused by impaired glymphatic system function, leading to reduced clearance of metabolic waste products and subsequent WM microstructural disruption. Within the autistic group, fornix FW mediated both the association between the DTI-ALPS index and fornix FA and the relationship between the DTI-ALPS index and hippocampal FW. These findings indicate a selective disruption in the memory network, where glymphatic impairment undermines fornix’s structural integrity, subsequently impacting hippocampal function. This fornix-hippocampal circuit disruption likely contributes to memory impairments, a prominent feature of autistic aging, highlighting the critical dependence of cognitive networks on effective glymphatic clearance [15]. Collectively, these findings underscore the central role of glymphatic system impairment in driving WM and memory network pathology in autistic aging. These insights advocate for targeted interventions to enhance glymphatic function, which could mitigate the progression of age-related brain changes in autism and support cognitive resilience.

The functional relevance of these findings is underscored by correlations between fornix FA and long-term visual memory and long-term verbal memory that reached and nearly reached significance, respectively. Autistic adults across our broad age-range scored lower on both of types of long-term, episodic memory and had reduced fornix FA compared to NT adults, which is consistent with previous publications on subsets of this cohort [11, 60]. Further, we have previously published accelerated longitudinal decline in visual long-term memory middle-age and older autistic adults compared to NT adults [4]. Similarly, an autism cognitive aging study in the Netherlands identified a “deviant cognitive profile” in two independent samples of autistic adults that is primarily driven by lower performance on verbal and visual episodic memory test [61], but no differences in longitudinal memory trajectories [13]. Interestingly, global GM FW and WM FA also significantly correlated with long-term visual memory, in the absence of group differences in these metrics. It is possible that autistic adults’ hippocampal system is compromised for a variety of reasons (e.g., glymphatic system functioning, stress, genetic vulnerability) [60], which leads to compensatory recruitment of extrahippocampal circuitry to support long-term memory. For example, autistic adults show greater prefrontal activation for some types of memory compared to NT adults [62]. Therefore, integrity of these extrahippocampal GM and WM regions is likely captured in our global measures and linked with long-term memory performance in autistic adults.

Limitations

This study has several limitations. First, the cross-sectional design limits causality inferences between DTI-ALPS, free water accumulation, WM integrity, and cognitive performance, as well as age-related change interpretation. Longitudinal studies are necessary to explore how these factors evolve over time and whether they precede age-related cognitive decline. Second, there are other metrics of glymphatic function besides the DTI-ALPS index that may provide complementary information and should be investigated in the future. Third, although we included sex as a covariate in our analyses, our sample was not evenly balanced in terms of sex distribution, which is common in autism research. Prior work has highlighted sex differences autistic individuals and cognitive aging suggesting that future studies with larger samples are needed to investigate potential interactions with sex. Fourth, although age was included as a covariate in inter-group comparisons to minimize potential confounding effects, it was not controlled for in within-group correlation or mediation analyses. This approach may not fully account for age-related variance, which represents a limitation of the present study. Fifth, the fornix is a small, thin structure located close to CSF/ventricular spaces, making it susceptible to partial volume effects and potential CSF contamination. Although the diffusion-derived free water metric is specifically designed to account for extracellular partial volume contributions, we cannot completely rule out the possibility that these measurements were influenced by CSF signal. Finally, this study did not include participants with co-occurring intellectual disability. Thus, our findings may not generalize to the broader autistic population, particularly those with lower intellectual functioning. Future research should examine how intellectual disability interacts with aging processes in autism.

Conclusions

In conclusion, this study provides compelling evidence that autistic aging involves a distinct neurobiological profile, marked by glymphatic dysfunction, greater FW accumulation, and selective WM microstructural abnormalities, particularly in the fornix and hippocampus, all of which may converge to impact long-term memory. FW’s mediating role between glymphatic dysfunction and WM integrity underscores disrupted fluid clearance as a critical driver of brain differences in aging autistic adults. These findings not only deepen our understanding of how autism interacts with the aging process but also highlight the importance of considering brain clearance mechanisms and structural integrity as critical factors in the lifelong trajectory of this condition.

Abbreviations

ADRD

Alzheimer’s disease and related dementias

FW

Free water

WM

White matter

GM

Gray matter

CSF

Cerebrospinal fluid

NT

Neurotypical

DTI

Diffusion tensor imaging

DTI-ALPS

Diffusion tensor image analysis along the perivascular space

FA

Fractional anisotropy

AD

Alzheimer’s disease

ADOS-2

Autism Diagnostic Observation Schedule-2

ASD

Autism spectrum disorder

AVLT

Auditory Verbal Learning Test

WMS-III

Wechsler Memory Scale–Third Edition (WMS-III)

Author contributions

B.B.B, Y.Z., and E.O. designed the study, analyzed data, and wrote the manuscript. K.C., S.A.H., M.V., S.G., K.G., F.J., L.B., and B.W. reviewed the manuscript. B.B.B, K.G., S.A.H, M.V. and F.J. collected the data. K.C. provided critical feedback on the statistical analysis. All authors reviewed and approved the final version of the paper to be published.

Funding

This work was supported in part by the National Institute of Mental Health (Grant Numbers: R01MH132746, K01MH116098, F31MH138112, F31MH122107), the Department of Defense (Grant Numbers: W81XWH-20-1-0171, W81XWH-15-1-0211), and the Arizona Biomedical Research Commission (Grant Number: ADHS16-162413).

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Ethical approval and consent to participate

All participants provided informed consent. The study was approved by the ASU Institutional Review Board.

Competing interests

The authors declare no competing interests.