White matter lesions disrupt cholinergic pathways important for cognition in Parkinson’s disease

Abstract

White matter lesions are associated with cognitive impairment in Parkinson’s disease, but the underlying neurobiological mechanisms remain unclear. One hypothesis is that strategically located white matter lesions disrupt long-range white matter pathways crucial for cognitive functioning, particularly cholinergic projections from the basal forebrain, contributing to cognitive decline. Using volumetric measures of basal forebrain nuclei (T1 MRI proxy for brain cholinergic projection nuclei), quantitative white matter lesion assessments (fluid-attenuated inversion recovery MRI) and a presynaptic cholinergic marker (¹⁸F-FEOBV PET), we tested a mechanistic model of white matter lesion effects on basal forebrain cholinergic projections and nuclei, and how these disruptions relate to cognition.

We analysed data from 127 mid-to-advanced-stage patients with Parkinson’s disease without dementia. Periventricular and deep white matter lesion burdens were quantified with the UBO Detector pipeline. Cognitive function was assessed across five domains using z-scores from a control group matched for age and education. Voxel-wise analyses examined associations between white matter lesion burden and cholinergic synaptic density, followed by mediation analyses testing whether cholinergic alterations mediated the relationship between white matter lesions and cognition, while correcting for sex, levodopa equivalent dose and disease duration.

Voxel-wise analyses revealed that higher periventricular lesions burden was associated with reduced ¹⁸F-FEOBV uptake in an insular–limbic–frontal–cingulum cluster, while deep lesions showed more topographically limited associations that disappeared after adjusting for periventricular burden. Periventricular lesions affected lateral and external Ch4 cholinergic projection pathways (81%–88% of patients), more frequently than deep white matter lesions (25%–47% of patients), except for the external capsule, which was equally affected by both. Cognitive performance (global cognition, memory, executive function and attention) was associated with both periventricular lesion burden and ¹⁸F-FEOBV uptake within the insular–limbic–frontal–cingulum cluster, as well as with basal forebrain volume. Mediation analyses indicated that periventricular lesions had a direct effect on cortical terminal integrity (¹⁸F-FEOBV) and an indirect, presumably retrograde degeneration effect via basal forebrain atrophy. The effect of periventricular lesions on insular–limbic–frontal–cingulum cholinergic terminal density was a strong mediator of the association between lesion burden and cognitive performance. A weaker mediation effect via the basal forebrain emerged for global cognition, whereas mediation was more robust for memory and executive functioning.

Our findings indicate that periventricular white matter lesions contribute to cognitive vulnerability in Parkinson’s disease without dementia through impacts on basal forebrain cholinergic projections. White matter lesion burden may represent a mechanistically meaningful biomarker for predicting cognitive trajectories and informing targeted interventions in Parkinson’s disease.

Carli et al. show that periventricular white matter lesions contribute to cognitive decline in Parkinson’s disease by disrupting cholinergic projections from the basal forebrain. Lesion burden may serve as a mechanistically meaningful biomarker for predicting cognitive trajectories and informing targeted interventions.

See Murphy et al. (https://doi.org/10.1093/brain/awag295) for a scientific commentary on this article.

See Murphy et al. (https://doi.org/10.1093/brain/awag295) for a scientific commentary on this article.

Introduction

Parkinson’s disease (PD) is a neurodegenerative disorder characterized by neuronal accumulation of misfolded α-synuclein.¹ Although PD is traditionally conceptualized as a grey matter disorder, growing evidence indicates that white matter alterations also play an important role.² White matter changes may reflect both comorbid conditions, such as small vessel cerebrovascular disease, and intrinsic neurodegenerative processes, such as demyelination and oligodendrocyte dysfunction.³ White matter hyperintensities, signal abnormalities typically observed on MRI sequences such as fluid-attenuated inversion recovery (FLAIR) or T2-weighted images, represent a reliable imaging biomarker of white matter lesions (WMLs).⁴ Neuropathological studies, though limited in number, suggest that WMLs have multifactorial aetiologies.^5-7 WMLs may reflect combinations of pathological mechanisms, including small vessel disease, leading to chronic hypoperfusion, microglial activation, gliosis, increased blood–brain barrier permeability as well as demyelination and axonal loss secondary to neurodegenerative processes.⁸ Studying WMLs in PD offers a unique opportunity to investigate pathologic processes beyond the traditionally studied grey matter alterations (e.g. atrophy or cortical synaptic dysfunction).

Jellinger,⁹ in a post-mortem study of autopsy-confirmed idiopathic PD cases and age-matched controls, reported that the prevalence of cerebrovascular lesions, including lacunes, amyloid angiopathy, WMLs and old or recent ischaemic infarcts and haemorrhages, was higher in PD (44.0%) than in controls (32.8%). Subsequent in vivo studies assessing WMLs have largely failed to replicate these findings,^10-15 with only a few exceptions.¹⁶ A recent meta-analysis reported no significant difference in the total WML burden between PD patients and controls,¹⁷ suggesting that WMLs may represent a non-specific age-related process rather than a PD-specific feature. An emerging body of work, however, indicates that this may not apply uniformly across all PD patients. Substantial evidence shows that WMLs are markedly more severe in PD patients with cognitive deterioration. In particular, individuals with PD and dementia exhibit significantly higher WML burden compared with cognitively normal PD patients.¹⁸ Findings in PD with mild cognitive impairment point in a similar direction, though with greater heterogeneity.¹⁸ These findings suggest an important role of leukoaraiosis in the cognitive impairment syndrome of PD. Across studies, both periventricular WMLs, lesions adjacent to the lateral ventricles often involving long-range projection fibres, and deep WMLs, lesions located within the deep white matter closer to cortical grey matter, are increased in cognitively impaired PD populations.¹⁸ This pattern suggests that in PD patients with cognitive decline, WMLs are unlikely to represent solely age-related changes. Instead, they may reflect additional neurodegenerative processes (e.g. neuroinflammation and/or axonal demyelination), vascular comorbidities or an interaction between the two, ultimately contributing to a more severe clinical phenotype.^18,19 Although WMLs hold promise as a biomarker for stratifying cognitive heterogeneity in PD, we know very little about the neurobiological mechanisms underlying this association. This knowledge gap limits the clinical applicability of assessing WMLs, not only as diagnostic or prognostic biomarkers but also for informing preventive or therapeutic strategies.

One plausible mechanistic link is that WMLs may disrupt white matter pathways that are critical for cognitive functioning and vulnerable to PD-related neurodegeneration. Strategically located WMLs may interfere with the cholinergic projections of basal forebrain nuclei,²⁰ deteriorating cholinergic pathways.^21,22 The basal forebrain cholinergic system comprises several nuclei, most prominently the nucleus basalis of Meynert (Ch4), which provides cholinergic innervation to the cortical mantle and paralimbic cortices, and the medial septum and diagonal band (Ch1–Ch2), which project primarily to the hippocampus and medial temporal structures.^23,24 These cholinergic projection systems support attentional control, learning and memory, visuospatial processing and executive function.²⁵ In PD, degeneration of these cholinergic pathways is well documented²⁶ and is strongly linked to several clinical manifestations,²⁷ particularly cognitive dysfunction.^25,28-31 Additional disruption from WMLs could further degrade cholinergic signalling and exacerbate cognitive impairments.

Emerging literature supports this mechanistic view, suggesting that cholinergic tract disruption is a key neurobiological pathway linking WML burden to cognitive deficits in PD. Some studies have shown that greater WML accumulation along cholinergic pathways is associated with worse cognitive outcomes, as evidenced by the Cholinergic Pathways Hyperintensities Scale (CHIPS). The CHIPS was developed by Bocti and colleagues³² based on immunohistochemical tracings of human cholinergic pathways.³³ It quantifies WML burden by visually assessing selected T2 or FLAIR MRI slices that encompass key anatomical regions traversed by cholinergic fibres. Since its validation, CHIPS has been widely used as an in vivo proxy for WML burden affecting cholinergic projections to the cortical mantle in neurodegenerative diseases,^34-38 including PD.^39-41 Prior findings indicate that PD patients with dementia show higher total CHIPS scores than healthy controls,^39,40 PD with mild cognitive impairment³⁹ and cognitively normal PD patients.³⁹ Although these findings are promising, molecular level evidence for the relationship between the cholinergic system, WMLs and cognition in PD is lacking.

The vesicular acetylcholine transporter is the presynaptic protein responsible for packaging acetylcholine into synaptic vesicles. Vesicular acetylcholine transporter expression can be quantified in vivo using the highly specific PET radiotracer ¹⁸F-FEOBV. Reduced vesicular acetylcholine transporter expression, reflected by lower ¹⁸F-FEOBV binding, is interpreted as a marker of cholinergic synaptic (density) denervation, providing a molecular measure for the presynaptic integrity of cholinergic projections. ¹⁸F-FEOBV has been successfully used in PD subjects to investigate cholinergic neurobiology underlying several clinical manifestations, including cognitive symptoms.^27,42 In the present context, it provides a selective molecular marker to study how WMLs may affect cholinergic synaptic integrity.

By integrating volumetric measures of the basal forebrain (T1 MRI proxy for Ch4 and Ch1–Ch2 nuclei), quantitative WML burden assessments (FLAIR) and a PET measure of presynaptic cholinergic integrity (¹⁸F-FEOBV), we tested a mechanistic model of how WMLs affect basal forebrain cholinergic projections, both downstream at the synapse (¹⁸F-FEOBV binding) and upstream at the projection nuclei (T1 MRI volumetry) (Aim 1), and how these disruptions relate to cognition in PD subjects (Aim 2). We examined these mechanisms in patients with mid-to-advanced-stage PD without dementia to capture early, subtle cognitive changes. We hypothesized that strategically located WMLs, intersecting cholinergic projection fibres, reduce synaptic cholinergic density and, through retrograde effects on perikarya, impair basal forebrain cholinergic nuclei. This dual disruption was expected to degrade cognitive performance in PD.

Materials and methods

Participants

This study was conducted on a convenience sample of 127 patients with mild-to-advanced PD without dementia, recruited from the Movement Disorders Clinics at the University of Michigan and the affiliated VA Ann Arbor Healthcare System (see Table 1 for more details). All PD participants met the UK Parkinson’s Disease Society Brain Bank Clinical diagnostic criteria.⁴³ Typical nigrostriatal dopaminergic denervation was confirmed in 90 PD participants using either ¹¹C-dihydrotetrabenazine vesicular monoamine transporter type 2 brain PET (n = 89) or ¹¹C-PE2I dopamine transporter brain PET (n = 1). Subjects with evidence of large vessel stroke or other intracranial lesions were excluded. No enrolled participants had dementia, nor were they using anticholinergic agents or cholinesterase inhibitor drugs.

Table 1.

Participant features

Variable	Overall
N	127
PET scanners
HR+	72
True Point	55
Age, mean (SD)	67.57 (7.29)
Education, mean (SD)	16.09 (2.77)
Sex, n (%)
Female	29 (22.8)
Male	98 (77.2)
Clinical duration, years, mean (SD)	5.78 (4.41)
LED, mean (SD)	605.26 (395.77)
H&Y, mean (SD)
1	7 (5.5)
1.5	6 (4.7)
2	33 (26.0)
2.5	50 (39.4)
3	28 (22.0)
4	3 (2.4)
MCI, n (%)
Yes	53 (41.7)
No	74 (58.3)
Z-scores Global Cognition, mean (SD)	−0.38 (0.74)
Z-scores Memory domain, mean (SD)	−0.33 (0.88)
Z-scores Attention domain, mean (SD)	−0.53 (0.94)
Z-scores Executive domain, mean (SD)	−0.44 (1.00)
Z-scores Language domain, mean (SD)	−0.32 (1.03)
Z-scores Vision domain, mean (SD)	−0.29 (0.93)
Obesity (BMI > 30), n (%)	48/126 (38.1%)
Diabetes, n (%)	14/126 (11.1%)
Heart problems, n (%)	10/126 (7.9%)
High blood pressure, n (%)	51/126 (40.5%)
pWML burden, mean (SD)	1.42 (1.53)
dWML burden, mean (SD)	0.43 (0.54)

BMI = body mass index; d = deep; H&Y = Hoehn and Yahr; LED = levodopa equivalent dose; MCI = mild cognitive impairment; n = number; p = periventricular; SD = standard deviation; WML = white matter lesion.

Ethical compliance

Written informed consent was obtained from all participants in accordance with the Declaration of Helsinki. This study was approved by the Institutional Review Boards of the University of Michigan School of Medicine and VA Ann Arbor Healthcare System.

Cognitive assessment and cognitive domains

Participants underwent a detailed neuropsychological test battery that evaluated five cognitive domains²⁸: memory, attention, executive function, language and visual skills (including one visuospatial and one visuoconstructive assessment). Each cognitive domain assessment included at least two subtests from the battery. Test scores were converted to z-scores using a non-PD control cohort (n = 77; male/female = 36/41; Montreal Cognitive Assessment = 26.74 ± 2.63), matched for age (69.03 ± 7.27 years) and education (16.06 ± 2.39 years). We also identified the number of patients classified as having mild cognitive impairment according to the Movement Disorder Society Level II criteria.⁴⁴ We then derived five cognitive domain scores by averaging the z-scores obtained across the individual tests corresponding to each previously described domain (Supplementary Table 1)²⁸: memory (California Verbal Learning Test immediate and delayed free recall scores, and Weschler Memory Scale total score), attention (Stroop 2: colour test time, Delis–Kaplan Executive Function System Trail Making Test 2: number sequencing time, Symbol Digit Modalities Test and Wechsler Adult Intelligence Scale: digit backwards task scores), executive function (Stroop 4: adjusted colour word test time, Delis–Kaplan Executive Function System Trail Making Test 4: letter–number sequencing time, Wechsler Adult Intelligence Scale: matrix reasoning score and letter fluency), language (Boston Naming test score and verbal fluency animal naming test) and visual function (Judgement of Line Orientation test total score and Parkinson’s Disease–Cognitive Rating Scale: draw clock). Z-scores for tests in which higher values indicated poorer performance (Stroop and Trail Making Test times) were multiplied by −1 to ensure that higher z-scores uniformly indicated better performance. A measure of global cognition was derived by averaging the five cognitive domain scores and was used as the primary outcome measure. Individual cognitive domain scores were analysed as secondary outcomes.

Imaging acquisition and preprocessing

Brain MRI was performed on a 3 T Philips Achieva system. Multiple MRI sequences were acquired for each participant, namely high-resolution T1-weighted and FLAIR images. The T1-weighted sequence consisted of a three-dimensional inversion recovery–prepared turbo field echo acquired in the sagittal plane with the following parameters: repetition time (TR)/echo time (TE)/inversion time (TI) = 9.8/4.6/1041 ms, turbo factor = 200, single average, field of view (FOV) = 240 × 200 × 160 mm and an acquisition matrix of 240 × 200 × 160, reconstructed to yield 1 mm isotropic resolution. The FLAIR sequence was acquired using TR = 4800 ms, TE = 125 ms, TI = 1650 ms, flip angle = 40°, FOV = 256 mm, matrix = 256 × 256 and slice thickness = 1.0 mm.

PET imaging was performed in 3D imaging mode with a Biograph 6 TruPoint PET/CT scanner (Siemens Molecular Imaging, Inc.) or an ECAT Exact HR+ tomograph as previously reported.^21,45 To ensure accurate and unbiased results from two PET scanners, an inter-scanner normalization method was implemented, and images were corrected for scatter and motion.⁴⁶ In addition, we performed dedicated sensitivity regression analyses to evaluate the potential effect of scanner type on PET-related main outcomes (Supplementary material). ¹⁸F-FEOBV was prepared as described previously by Shao et al.^{47 18}F-FEOBV delayed dynamic imaging was performed over 30 min (in six 5 min frames) starting 3 h after an intravenous bolus dose injection of 8 mCi ¹⁸F-FEOBV.⁴⁸ FreeSurfer (Charlestown, MA, USA) was used to segment the brain MRI into cortical and subcortical regions. The FreeSurfer ScLimbic pipeline was used to segment the basal forebrain, including, but not limited to, Ch1–Ch4, in each subject’s native T1-weighted structural MRI space.⁴⁹ The resulting basal forebrain volume was normalized to the intracranial volume and used as a measure in the statistical analyses. We co-registered participants’ MRI scans and segmentations to their individual PET images. Parametric ¹⁸F-FEOBV PET images were generated using a supratentorial white matter reference tissue approach.²¹ The reference region was defined from a FreeSurfer-derived white matter mask by excluding voxels below the ventricles and near the cortex, followed by a 3 mm spherical erosion (Supplementary Fig. 1). Validation studies have shown that non-specific binding in this region is comparable between Lewy body disease spectrum patients and controls.⁵⁰ Parametric (intensity-normalized) images were then generated by dividing voxel values in the respective activity images of six delayed PET frames by the mean activity in this reference region. The Müller-Gärtner partial volume correction method was applied to correct for the partial volume effect on the parametric PET images.⁵¹ Individual participant partial-volume-corrected parametric ¹⁸F-FEOBV studies were normalized to the study-specific template in Montreal Neurological Institute space using high-dimensional Diffeomorphic Anatomical Registration Through Exponentiated Lie Algebra (DARTEL) normalization. See Supplementary Fig. 2 for a summary graphical overview of multimodal imaging pipelines.

White matter lesion burden in periventricular and deep spaces

Previous literature on WMLs in PD exhibits substantial variability in WML assessment tools, ranging from visual rating scales to semi-automated and fully automated approaches.¹⁸ Most fully automated algorithms are in-house validated pipelines that, although potentially replicable, are not distributed as shared, ready-to-use tools, limiting the reproducibility of findings. In order to ensure that our results are replicable, we selected UBO Detector among the available WML segmentation tools for its reliable performance without the need for a training dataset.⁵² UBO Detector is a cluster-based, fully automated pipeline that employs a k-nearest neighbour algorithm for WML mask identification.⁵² This tool has demonstrated high correlations with gold standard manually defined WMLs and with clinically meaningful scales for lesion burden evaluation.⁵³ For our analyses, we applied a probability threshold of 0.8 and set the number of nearest neighbours to k = 5. Following the recommendations of its creators,⁵² we used a 12 mm threshold from the ventricular border to define the periventricular WMLs. The algorithm classifies all cerebral non-periventricular lesions as deep WMLs (excluding brainstem and cerebellum). Lesion burden was expressed as a percentage, calculated as volume of white matter hyperintensities/total white matter volume × 100. WML burden variables showed significantly positive skewness and were therefore cube-root transformed to facilitate investigation of their relationships with cognition and cortical ¹⁸F-FEOBV uptake (Supplementary Fig. 3).

Cholinergic pathways white matter lesion burden

To explore the roles of periventricular and deep WMLs in cholinergic deficits, we examined whether these lesion types affect white matter tracts containing basal forebrain cholinergic projections. We used the CHIPS scale to identify lateral, external and medial Ch4 projection pathways, including the cingulum (cingulate and hippocampal—Ch4 medial pathway), external capsule (Ch4 lateral and external pathway) and anterior, superior and posterior corona radiata (Ch4 lateral and external pathway) from the Johns Hopkins University white matter tractography atlas (Fig. 1A).^54-56 Although the corona radiata is not an exclusively basal forebrain projection bundle, its three subregions (anterior, posterior and superior) were included because the lateral cholinergic pathway traverses this major white matter tract. We quantified the proportion of each white matter tract affected by periventricular and deep WMLs. To do this, we computed voxel-wise overlap between the volumes of interest (VOIs) derived from the Johns Hopkins University white matter tractography atlas and the periventricular/deep WML masks of each participant. Specifically, we used the periventricular and deep WML masks generated by the UBO Detector in Montreal Neurological Institute space, co-registered and resliced them to tract-based VOIs, and then calculated overlap. The percentage of affected tract volume was then calculated as (number of overlapping voxels/total tract voxels) × 100. Differences in lesion prevalence between periventricular and deep WMLs within each tract (CHIPS scale-derived VOIs) were assessed using McNemar’s test for paired binary data.

Figure 1.

CHIPS-derived VOIs and WMLs. (A) CHIPS-derived VOIs representing the lateral, external and medial Ch4 cholinergic projection pathways, as defined using the Johns Hopkins University white matter tractography atlas in MNI space. The lateral and external pathway comprises three corona radiata subregions, reflecting the anatomical course of basal forebrain projections through the corona radiata and external capsule. (B) Prevalence of lesion overlap (number of patients showing overlap) between pWML (salmon pink) and dWML (light blue) masks and CHIPS-derived VOIs. PWMLs and dWMLs affected the external capsule in a comparable proportion of patients. In contrast, pWMLs affected (showing any degree of overlap) the remaining CHIPS tracts significantly more frequently (i.e. in a higher number of patients) than dWMLs. Differences in prevalence were tested using the McNemar test (*P < 0.05, ***P < 0.001). CHIPS = Cholinergic Pathways Hyperintensities Scale; dWMLs = deep white matter lesions; MNI = Montreal Neurological Institute; pWMLs = periventricular white matter lesions; VOIs = volumes of interest; WMLs = white matter lesions.

White matter lesion burden voxel-wise ¹⁸F-FEOBV brain PET correlates

Two voxel-wise regression models were performed with partial volume corrected ¹⁸F-FEOBV brain images as the dependent variable and cube root transformed periventricular and deep WML burden as independent variables, correcting for levodopa equivalent dose and disease duration. Because periventricular and deep WML burden can be associated, we also constructed two additional models to disentangle specific associations with cholinergic synaptic density. In these models, periventricular WML burden was examined while controlling for deep WMLs, and vice versa. An explicit mask, including only regions innervated by the basal forebrain cholinergic nuclei, was applied (Supplementary Fig. 4). The statistical threshold for significance was set at P < 0.05, corrected voxel-level for family-wise error for multiple comparisons, and cluster extent of 100 voxels (K > 100). The significant clusters resulting from these analyses were combined into dichotomous VOIs (see the ‘Cholinergic integrity and cognition: cognitive cholinergic molecular correlates’ section).

Statistical analyses

White matter lesion burden, ¹⁸F-FEOBV brain PET and basal forebrain volume: correlations

We examined the associations between periventricular and deep WML burden, the corresponding clusters identified in voxel-wise analyses (see previous section) and basal forebrain volume. Correlation analyses were performed using Pearson’s correlation coefficient for normally distributed variables and Spearman’s rank correlation coefficient for non-normally distributed variables. Statistical significance was set at P < 0.05, with adjustment for multiple comparisons using the Holm correction.

White matter lesion burden, cholinergic integrity and cognition: correlations

We further investigated the associations between periventricular and deep WML burden, the related voxel-wise clusters and basal forebrain volume with global cognition (primary outcome) and domain-specific cognitive measures (secondary outcomes). Pearson’s or Spearman’s correlation coefficients were applied depending on the distribution of the variables. Statistical significance was defined as P < 0.05, corrected for multiple comparisons using the Holm method.

Cholinergic integrity and cognition: cognitive cholinergic molecular correlates

For cognitive scores that showed significant associations with both WML burden variables and the corresponding ¹⁸F-FEOBV cluster binding, we ran voxel-wise regression models, correcting for sex, levodopa equivalent dose and disease duration. Significant clusters were converted into dichotomous VOIs and intersected with clusters associated with WML burden to identify brain regions most strongly linked to both WMLs and the specific cognitive domain. These intersection-derived VOIs were then used to extract ¹⁸F-FEOBV uptake values from the partial volume corrected parametric images in Montreal Neurological Institute space. The extracted values were subsequently used in the correlation and mediation analyses.

White matter lesion burden, cholinergic integrity and cognition: mediation

We tested three mechanistic models (Fig. 2):

Figure 2.

Mechanistic models tested. The orange arrow denotes the portion of the model testing whether white matter lesions (WMLs) affect cholinergic terminal density directly and indirectly via retrograde effects on the basal forebrain. The dark-pink arrow represents the model testing whether the effect of WMLs on cholinergic terminal density mediates the association between WMLs and cognition. The green arrow denotes the model testing whether the effect of WMLs on basal forebrain volume mediates the association between WMLs and cognition. Dotted lines represent mediation paths.

1. We assessed whether the association between WML burden and terminal cholinergic density in brain regions identified through voxel-wise analyses (see the ‘¹⁸F-FEOBV correlates of periventricular and deep white matter lesion burden’ section) was mediated by basal forebrain volume. This model was operationalized with WML burden as the independent variable (X), WML burden-related ¹⁸F-FEOBV cluster uptake as the dependent variable (Y) and basal forebrain volume as the mediator (M).

2. We examined whether the relationship between WML burden and cognitive function was mediated by cholinergic terminal density. This model was operationalized with WML burden (X), cognitive scores (Y) and ¹⁸F-FEOBV extracted from regions showing overlapping effects of WML burden and each cognitive variable (i.e. WML–cognition intersection-derived VOIs) (M).

3. We tested whether the relationship between WML burden and cognition was mediated by basal forebrain volume. This model was operationalized with WML burden (X), cognitive scores (Y) and basal forebrain volume (M).

Each variable was standardized before being entered into the models. All models were adjusted for levodopa equivalent dose, disease duration and sex. Mediation effects were estimated using ordinary least squares regression with 5000 non-parametric bootstrap samples (mediation package).⁵⁷ Model diagnostics included checks for multicollinearity (variance inflation factor), heteroscedasticity (Breusch–Pagan test) and residual normality (Shapiro–Wilk test). Although bootstrap-based mediation is relatively robust to assumption violations, we repeated analyses using the more stringent robust mediation approach (ROBMED)⁵⁸ framework when assumptions were not met. ROBMED replaces ordinary least squares with robust regression while maintaining bootstrapping, providing more conservative estimates resistant to outliers and distributional deviations.⁵⁸ Because ROBMED yields more conservative estimates, we used it as a complementary validation step rather than the primary analytic approach in order to retain the ability to detect weaker, but potentially meaningful, patterns that may inform new hypothesis generation. Statistical significance was set at a two-tailed alpha level of 0.05.

Results

Demographic, clinical and cognitive features of the cohort

Our cohort included 127 PD patients (mean age: 67.6 ± 7.3 years; 29 female/98 male) with a mean disease duration of 5.8 ± 4.4 years and a mean levodopa equivalent dose of 605.3 ± 395.8. Fifty-three patients (41.7%) were classified as having mild cognitive impairment. Regarding potential aetiological factors contributing to WMLs, 38.1% of participants were obese (body mass index > 30), 11.1% had diabetes, 7.9% reported a history of heart disease and 40.5% had hypertension (Table 1).

Periventricular and deep white matter lesions and cholinergic tracts

Periventricular WMLs affected lateral/external and medial projection pathways more frequently compared to deep WMLs, as reflected by a higher proportion of patients with overlap within CHIPS VOIs (Fig. 1B). The posterior, anterior and superior corona radiata were the most frequently affected tracts by periventricular WMLs (88.2%, 83.5% and 81.1% of patients, respectively). The deep WMLs involved these tracts in 25.2%, 37.0% and 47.2% of patients. Similarly, periventricular WMLs more frequently involved the cingulum (hippocampal subdivision: 6.3% versus 0.8%, seven versus zero discordant cases, P = 0.023; cingulate subdivision: 7.9% versus 2.4%, eight versus one discordant cases, P = 0.046). In contrast, the external capsule was equally affected by periventricular and deep WMLs (23.6% versus 29.1%, 14 versus 21 discordant cases, P = 0.310) (Supplementary Table 2).

¹⁸F-FEOBV correlates of periventricular and deep white matter lesion burden

Statistical Parametric Mapping (SPM) voxel-wise analyses revealed that higher periventricular WML burden was associated with lower ¹⁸F-FEOBV PET uptake in the right parahippocampal gyrus and temporal pole, bilateral insula, operculum, prefrontal cortices, anterior and middle cingulate, calcarine and lingual gyri, as well as the left inferior parietal cortex and the bilateral pre- and postcentral gyri (Fig. 3). To facilitate interpretation of effect size, statistical maps were converted to correlation coefficient (r) maps using the CAT12 toolbox and shown in Supplementary Fig. 5. Given predominant involvement of limbic, cingulum, insular and frontal regions, this topography will hereafter be termed the insular–limbic–frontal–cingulum cluster. The correlation between periventricular and deep WML burden was significant (r = 0.796; P < 0.001). When controlling for deep WML burden, the topography of the periventricular WML burden cholinergic association became limited to the temporal pole, insula (right more than left) and right prefrontal cortex (Supplementary Fig. 6). Deep WML burden showed a more restricted pattern of negative correlations with ¹⁸F-FEOBV brain uptake, primarily involving the left more than right insula, the middle cingulate cortex, the bilateral precentral and postcentral gyri, and the left calcarine cortex (Fig. 3 and see Supplementary Fig. 5 for correlation coefficient maps). However, when controlling for periventricular WML burden, no significant clusters survived for the deep WML burden.

Figure 3.

Periventricular and deep WMLs and cholinergic synaptic density.  Top: Examples of periventricular (red) and deep (green) WML segmentation in a patient with extensive WML burden (Subject 1, left) and in a patient with mild WML burden (Subject 2, right), overlaid on individual FLAIR images normalized to MNI space. Bottom: Statistical parametric maps showing the results of voxel-wise Statistical Parametric Mapping regression analyses, with periventricular WML and deep WML burden entered as independent variables and regional ¹⁸F-FEOBV PET binding as the dependent variable. All models were adjusted for disease duration and levodopa equivalent daily dose. Clusters demonstrating negative correlations with periventricular WML burden (left) and deep WML burden (right) are displayed. The red–yellow colour scale reflects correlation strength, with yellow indicating stronger negative associations. Threshold: voxel-level family-wise error corrected P < 0.05; cluster extent k > 100 voxels. dWMLs = deep white matter lesions; FLAIR = fluid-attenuated inversion recovery; MNI = Montreal Neurological Institute; pWMLs = periventricular white matter lesions; VOIs = volumes of interest.

White matter lesion burden, ¹⁸F-FEOBV brain PET and basal forebrain volume: correlation results

Periventricular WML burden [ρ = −0.397; P (Holm) < 0.001] and ¹⁸F-FEOBV uptake in the associated insular–limbic–frontal–cingulum cluster [ρ = 0.530; P (Holm) < 0.001] showed significant negative and positive correlations, respectively, with the volume of the basal forebrain (Supplementary Fig. 7). Both deep WML burden [ρ = −0.285; P (Holm) = 0.019] and ¹⁸F-FEOBV uptake in the related cluster [ρ = 0.510; P (Holm) < 0.001] showed significant associations with basal forebrain volume (Table 2).

Table 2.

Correlations with cognitive domains and basal forebrain volume

X variables	Y variables	Method	n	Coeff.	P (Holm)
Basal forebrain volume	Global cognition (z)	Spearman	126	0.373	0.001**
	Memory (z)	Spearman	125	0.305	0.011*
	Attention (z)	Spearman	122	0.325	0.006**
	Executive function (z)	Spearman	121	0.277	0.034*
	Language (z)	Spearman	125	0.258	0.039*
	Vision (z)	Spearman	127	0.200	0.145
Deep WML burden (transf)	BF volume	Spearman	127	−0.285	0.019*
	Global cognition (z)	Spearman	127	−0.258	0.039*
	Memory (z)	Spearman	125	−0.143	0.255
	Attention (z)	Spearman	122	−0.200	0.145
	Executive function (z)	Spearman	121	−0.180	0.194
	Language (z)	Spearman	125	−0.215	0.111
	Vision (z)	Spearman	127	−0.244	0.052
Deep WML–18F-FEOBV cluster	BF volume	Spearman	127	0.510	<0.001***
	Global cognition (z)	Spearman	127	0.409	<0.001***
	Memory (z)	Spearman	125	0.369	0.001***
	Attention (z)	Spearman	122	0.326	0.006**
	Executive function (z)	Spearman	121	−0.180	0.194
	Language (z)	Spearman	125	0.153	0.255
	Vision (z)	Spearman	127	0.269	0.034*
Periventricular WML burden (transf)	BF volume	Spearman	127	−0.397	<0.001***
	Global cognition (z)	Spearman	127	−0.382	<0.001***
	Memory (z)	Spearman	125	−0.265	0.038*
	Attention (z)	Spearman	122	−0.343	0.003**
	Executive function (z)	Spearman	121	−0.271	0.038*
	Language (z)	Spearman	125	−0.294	0.015*
	Vision (z)	Spearman	127	−0.261	0.038*
Periventricular WML–18F-FEOBV cluster	BF volume	Spearman	127	0.530	<0.001***
	Global cognition (z)	Spearman	127	0.424	<0.001***
	Memory (z)	Spearman	125	0.407	<0.001***
	Attention (z)	Spearman	122	0.351	0.002**
	Executive function (z)	Spearman	121	0.373	0.001**
	Language (z)	Spearman	125	0.155	0.255
	Vision (z)	Spearman	127	0.243	0.052

All correlations were computed using Spearman’s rank correlation because, in each case, at least one variable was not normally distributed. Bold text indicates significant correlation. BF = basal forebrain; Coeff. = coefficient; pWML = periventricular white matter lesion; transf = cube-root transformed variable; WML = white matter lesion; z = z-scores.

^* P ≤ 0.05.

^** P ≤ 0.005.

^*** P ≤ 0.001.

White matter lesion burden, cholinergic integrity and cognition: correlation results

White matter lesion burden

Periventricular WML burden showed significant negative correlations with global cognitive [ρ = −0.382; P (Holm) < 0.001], memory [ρ = −0.265; P (Holm) = 0.038], attention [ρ = −0.343; P (Holm) = 0.003], executive [ρ = −0.271; P (Holm) = 0.038], language [ρ = −0.294; P (Holm) = 0.015] and visuospatial [ρ = −0.261; P (Holm) = 0.038] cognitive scores. Deep WMLs showed a significant negative correlation with global cognition [ρ = −0.258; P (Holm) = 0.039], but not with cognitive domain scores.

Cholinergic synaptic density (¹⁸F-FEOBV)

¹⁸F-FEOBV periventricular WML-correlated cluster uptake showed positive correlations with global [ρ = 0.424; P (Holm) < 0.001], memory [ρ = 0.407; P (Holm) < 0.001], attention [ρ = 0.351; P (Holm) = 0.002] and executive cognitive [ρ = 0.373; P (Holm) = 0.001] scores. ¹⁸F-FEOBV deep WML-correlated cluster uptake showed positive correlations with global [ρ = 0.409; P (Holm) < 0.001], memory [ρ = 0.369; P (Holm) = 0.001], attention [ρ = 0.326; P (Holm) = 0.006] and visual [ρ = 0.269; P (Holm) = 0.034] cognitive functions.

Basal forebrain

Basal forebrain volume showed significant correlations with global cognitive [ρ = 0.373; P (Holm) = 0.001], memory [ρ = 0.305; P (Holm) = 0.011], attention [ρ = 0.325; P (Holm) = 0.006], executive ρ = 0.277; P (Holm) = 0.034] and language [ρ = 0.258; P (Holm) = 0.039] scores, but not with visuospatial abilities [ρ = 0.200; P (Holm) = 0.145].

See Table 2 for detailed statistical results and Supplementary Figs 8–10 for the graphical representation of the correlations with the primary cognitive outcome (global cognition).

¹⁸F-FEOBV voxel-wise correlates of global cognition, memory, executive and attentive function

Global cognition showed a significant positive correlation with ¹⁸F-FEOBV uptake in the bilateral temporal pole, parahippocampal and hippocampal gyri, insula, orbitofrontal cortex, superior lateral and medial frontal cortices, anterior and middle cingulum, the right precentral gyrus, and the left postcentral gyrus (P uncorrected < 0.001; cluster extent > 100 corrected for family-wise error) (see Supplementary Fig. 5 for correlation coefficient maps). Spatial overlap with the periventricular WML-related cluster was located in the bilateral insula, hippocampus and middle cingulum. These overlapping regions were combined to form the WML–cognition intersection-derived VOIs from which uptake values were extracted for the mediation analyses (Fig. 4).

Figure 4.

Global cognition, periventricular WMLs and cholinergic synaptic density. Statistical Parametric Mapping voxel-wise correlations with global cognition and their spatial overlap with the periventricular white matter lesion (WML)-associated cholinergic cluster. Left: The cluster positively associated with global cognitive scores; middle: the cluster associated with periventricular WML burden; right: the intersection between the two clusters (binary). The red–yellow scale indicates the strength of the correlation, with yellow indicating stronger associations.

We performed similar voxel-wise regression analyses for the three cognitive domains that showed correlations with both periventricular WML burden and ¹⁸F-FEOBV uptake: memory, attention and executive function (Table 2). All three domains exhibited partially overlapping topographies, with significant positive correlations between ¹⁸F-FEOBV uptake and cognitive performance in the bilateral insula, hippocampus and the anterior and middle cingulum. Memory showed the most widespread correlation pattern, involving orbitofrontal regions, bilateral precentral and postcentral gyri, and bilateral insula, hippocampus, and the anterior and middle cingulum. Executive function showed a similar, though more restricted, correlation pattern extending into right frontoparietal areas. Attention displayed the most spatially circumscribed correlations, largely confined to the insula and cingulum (Supplementary Fig. 5 for correlation coefficient maps). For each cognitive domain, WML–cognition intersection VOIs were derived. These domain-specific intersection VOIs were then used to extract uptake values for the mediation analyses (Supplementary Fig. 11).

Mediation analysis results

We ran mediation models using only periventricular WML burden, as deep WML burden showed weak associations with cognition, yielding a significant correlation only with global cognitive score, but not for individual domain scores. This pattern was further supported by group comparisons between mild cognitive impairment and cognitively normal PD patients, in which only periventricular WML burden was significantly higher in the mild cognitive impairment group (Supplementary Fig. 12). Diagnostic statistics for the following mediation models are reported in Supplementary Table 3.

White matter lesion burden, basal forebrain and white matter lesion burden-related ¹⁸F-FEOBV cluster uptake

Basal forebrain volume partially mediated the association between periventricular WML burden and ¹⁸F-FEOBV uptake in the insular–limbic–frontal–cingulum cluster. The indirect effect was significant [estimate = −0.168, 95% confidence interval (CI) (−0.266, −0.090), P < 0.001], accounting for approximately 30% of the total effect [95% CI (0.165, 0.490), P < 0.001]. The direct effect remained significant [estimate = −0.389, 95% CI (−0.532, −0.230), P < 0.001]. Results were confirmed using a more robust mediation approach (ROBMED) (Supplementary Table 4).

White matter lesion burden, ¹⁸F-FEOBV uptake and cognition

Mediation analysis indicated that ¹⁸F-FEOBV uptake in brain regions associated with both periventricular WML burden and global cognition partially mediated the relationship between periventricular WML burden and global cognitive performance [proportion mediated = 43%, 95% CI (0.173, 0.930), P < 0.001]. When examining specific cognitive domains, ¹⁸F-FEOBV regional uptake acted as a full mediator for both memory [proportion mediated = 82%, 95% CI (0.33, 2.16), P = 0.002] and executive function [proportion mediated = 82%, 95% CI (0.37, 2.66), P = 0.004], as the direct effects of periventricular WML burden on these domains were no longer significant. For attention, ¹⁸F-FEOBV regional uptake partially mediated the effect, with 44% of the total effect mediated [95% CI (0.101, 1.13), P = 0.011], and the direct effect showed a trend towards significance (P = 0.080). See Table 3 for the models’ estimates and P-values. All models demonstrated stability, as results were confirmed using ROBMED (Supplementary Table 4).

Table 3.

Mediation results: (X) periventricular WML burden, (M) ¹⁸F-FEOBV uptake and (Y) cognitive domains

Domain	n	ACME	ACME P-value	ADE	ADE P-value	Total	Total P-value	Prop Med	Prop Med P-value
Global z-scores	126	−0.166	<0.001***	−0.222	0.037*	−0.389	<0.001***	0.428	<0.001***
Memory z-scores	125	−0.203	0.004**	−0.044	0.597	−0.247	0.001**	0.822	0.002**
Attention z-scores	122	−0.155	0.011*	−0.196	0.080	−0.351	<0.001***	0.441	0.011*
Executive z-scores	121	−0.222	<0.001***	−0.047	0.683	−0.269	0.004**	0.824	0.004**

The causal mediation models were run with non-parametric bootstrap confidence intervals using the percentile method (5000 bootstrap). ACME = Average Causal Mediated Effect; ACE = Average Direct Effect; n = number of patients; Prop Med = proportion mediated.

^* P ≤ 0.05.

^** P ≤ 0.005.

^*** P ≤ 0.001.

White matter lesion burden, basal forebrain volume and cognition

Mediation analyses using total basal forebrain volume as the mediator indicated that periventricular WML burden influenced cognitive performance through basal forebrain atrophy, although this pathway appeared more heterogeneous than the one observed for synaptic density (¹⁸F-FEOBV uptake). For global cognition, the non-parametric bootstrapped ordinary least squares-based mediation model showed a significant indirect effect [proportion mediated = 41%, 95% CI (0.149, 0.840), P < 0.001]. However, this effect was not confirmed when applying the more robust and conservative ROBMED approach (indirect effect P = 0.249). A similar pattern emerged for the attention domain, where only the ordinary least squares-based model supported a mediation effect. In contrast, for both the memory domain [proportion mediated = 54%, 95% CI (0.108, 1.53), P = 0.008] and the executive domain [proportion mediated = 44%, 95% CI (0.050, 1.60), P = 0.025], total basal forebrain volume significantly and fully mediated the relationship between periventricular WML burden and cognitive performance, with the direct effect becoming non-significant. These mediation effects were confirmed by the ROBMED approach, indicating greater robustness and stability for these domains. See Table 4 for the ordinary least squares-based models’ estimates and P-values, and Supplementary Table 4 for the corresponding ROBMED results.

Table 4.

Mediation results: (X) periventricular WML burden, (M) basal forebrain volume and (Y) cognitive domains

Domain	n	ACME	ACME P-value	ADE	ADE P-value	Total	Total P-value	Prop Med	Prop Med P-value
Global cognition	126	−0.159	<0.001***	−0.230	0.02*	−0.389	<0.001***	0.408	<0.001***
Memory	125	−0.134	0.006*	−0.113	0.193	−0.247	0.001***	0.543	0.008*
Attention	122	−0.106	0.012*	−0.245	0.010*	−0.351	<0.001***	0.301	0.012*
Executive function	121	−0.119	0.021*	−0.150	0.172	−0.269	0.004**	0.444	0.025*

The causal mediation models were run with non-parametric bootstrap confidence intervals using the percentile method (5000 bootstrap). ACME = Average Causal Mediated Effect; ACE = Average Direct Effect; N = number of patients; Prop Med = proportion mediated.

^* P ≤ 0.05.

^** P ≤ 0.005.

^*** P ≤ 0.001.

Discussion

Higher WML burden is consistently associated with cognitive impairments in PD, particularly in the dementia stage.¹⁸ The neurobiological mechanisms underlying this association remain poorly understood, especially in patients with PD without dementia. In this study, we tested the mechanistic hypothesis that strategically located WMLs disrupt long-range neurotransmitter pathways that are important for cognition and vulnerable in PD, specifically cholinergic basal forebrain projections.^20,59 Such disruptions could accelerate cholinergic system degeneration and aggravate cognitive impairment. To evaluate this hypothesis, we combined multimodal imaging, including T1-weighted and FLAIR MRI and ¹⁸F-FEOBV PET, and assessed cognition using continuous cognitive z-scores in a cohort of patients with mid-to-advanced PD without dementia. Our findings indicate that periventricular WMLs exert a greater impact on cholinergic pathways than deep WMLs, most likely affecting fibres in the lateral and external Ch4 projection pathway. Periventricular WML burden was associated with global cognition and all cognitive domains. Periventricular WMLs also showed the strongest relationship with cholinergic denervation, reflected by lower ¹⁸F-FEOBV uptake in an insular-limbic–frontal–cingulum cluster, surviving after correction for contributions of deep WML burden. Cholinergic synaptic density (¹⁸F-FEOBV uptake) in this cluster was positively correlated with global cognition, executive function, memory and attention. Mediation analyses suggested that periventricular WML burden affected cholinergic terminals both directly through axonal injury and indirectly through retrograde degeneration effects on basal forebrain projection nuclei (basal forebrain volume). The effects of periventricular WMLs on insular–limbic–frontal–cingulum cholinergic terminal density strongly mediated the relationship between lesion burden and cognitive performance. A similar, although weaker, mediation pattern was observed through basal forebrain volume involvement for global cognition and attention, whereas the mediation was compelling for memory and executive function. Together, these results support the view that periventricular WMLs are mechanistically involved in the neurobiology of cognitive deterioration in PD by contributing to the degeneration of basal forebrain cholinergic neuron projections. These findings underscore the need to incorporate WML burden into predictive models of cognitive trajectories in PD and highlight WMLs, and their potentially modifiable vascular contributors, as promising therapeutic targets for mitigating cognitive decline in PD.

The topography emerging from our voxel-wise analyses closely follows the projections of the lateral cholinergic pathway (Ch4), which includes a peri-Sylvian division that innervates the insula, operculum and superior temporal gyrus, and a capsular division projecting through the external capsule to the frontal, parietal and temporal neocortex.²⁴ This is consistent with our observation that the corona radiata, a major conduit for Ch4 lateral (external) pathway fibres,^60,61 is the tract most commonly affected in our cohort. The involvement of the cingulum bundle suggests some, although more limited, impact on the medial cholinergic pathway. The predominance of periventricular WMLs within the corona radiata may help explain the stronger association of periventricular, compared with deep, WML burden with cognitive domain performance. This interpretation is consistent with previous findings showing that the superior corona radiata microstructural integrity is related to cognitive performance in PD, especially memory,⁶² supporting its role as a key anatomical substrate for cognitive function. Although caution is warranted when interpreting CHIPS-derived results, given that some of the white matter tracts (e.g. corona radiata) included in the scale are not exclusively composed of cholinergic axons, the converging findings nonetheless strengthen the view that periventricular WMLs (by intersecting a strategic white matter bottleneck) disproportionately disrupt basal forebrain cholinergic projections and contribute to cognitive decline in PD, even prior to the onset of dementia.

Mediation analyses allowed us to evaluate a two-component mechanistic model underlying the association between WMLs and cognitive deficits in PD: WML burden impacts the cholinergic system both downstream (e.g. perikaryal atrophy/degeneration) and upstream (e.g. axonal degeneration), and these components mediate relationships with cognitive performance. This suggests that long-range cholinergic fibres are vulnerable to periventricular lesions, propagating downstream via terminal loss and upstream towards perikarya via retrograde degeneration, promoting synaptic and exacerbating irreversible neurodegeneration. Although the precise mechanisms of these inferred processes cannot be determined from our data, several mechanisms could be operative. Periventricular WMLs may result from vascular comorbidities^63-65 such as hypertension, diabetes or cardiac disease. Alternatively, periventricular lesions may reflect neurodegeneration-related processes, such as neuroinflammation,⁶⁶ which could plausibly drive some of the retrograde effects observed in the basal forebrain in a feedforward manner.

Periventricular WML burden showed the most consistent associations with cognitive performance compared with deep WMLs. Although deep WML burden correlated with global cognition, it was not associated with any specific cognitive domain and did not differ between patients with mild cognitive impairment and normal cognition. These results align with recent data suggesting that periventricular lesion burden is specifically associated with cognitive performance in PD.⁶⁷ Given the relatively low deep WML burden in our sample, it is possible that deep WMLs may contribute to cognitive impairment only when they become more extensive, reflecting more advanced leukoaraiosis.

The ¹⁸F-FEOBV binding cluster associated with periventricular WML burden (insular–limbic–frontal–cingulum) correlated with global cognition, memory, executive function and attention, but not with language or visuospatial performance. Importantly, this lack of association should not be interpreted as evidence of the absence of cholinergic involvement in these domains. Language performance was significantly related to basal forebrain volume, whereas visuospatial performance was significantly associated with the deep WML-related cluster. Overall, these findings indicate that cholinergic cortical denervation contributes to cognitive functioning across domains. However, attention, memory and executive functions showed greater topographical overlap with the cholinergic projections disrupted by periventricular WML burden. Notably, language and visuospatial functions were relatively preserved in this cohort, suggesting that stronger associations with periventricular WML-related cholinergic vulnerability may emerge in samples with more pronounced impairments in these domains. The voxel-wise analyses for global, executive, attentive and memory cognitive domains revealed cholinergic topographies that substantially overlapped with the cluster associated with periventricular WMLs. This suggests that these regions are critical, and partly shared across cognitive functions, for supporting global cognition, consistent with prior literature.^28,30,68 Basal forebrain volume showed similar correlations, consistent with prior evidence linking basal forebrain degeneration to both global and domain-specific cognitive deficits, particularly in attention and executive function, which is consistent with the known biology of basal forebrain cholinergic systems.^25,31 Mediation effects were most stable for the pathway involving cholinergic terminals, suggesting that cholinergic dysfunction within the insular–limbic–frontal–cingulum network may be particularly relevant for cognitive performance.⁶⁸ Basal forebrain volume showed the most consistent mediation effects for memory and executive functions, whereas findings for global cognition and attention were less conclusive. Our data are consistent with the hypothesis that, in PD patients without dementia, WML-related effects on cholinergic terminals directly degrade cognitive function, whereas WML impacts on the integrity of basal forebrain projection nuclei (presumed retrograde effects of WMLs) may become more impactful with disease progression and in more severe cases with overt dementia. For the attentional domain, we observed only partial mediation, suggesting a modest cholinergic contribution that likely interacts with other neurotransmitter systems, particularly dopaminergic pathways. Supporting this interpretation, a recent molecular imaging study found that ventral striatal dopaminergic terminal integrity mediates the relationship between WML burden and performance on attentional and working-memory tasks in PD.⁶⁹

Several limitations should be considered when interpreting our findings. First, comorbidities were extracted from medical records without pathophysiological biomarkers, limiting our ability to determine whether vascular, metabolic or neurodegenerative factors contributed to WML burden. Future studies should incorporate biomarker-based comorbidity assessments to clarify the mechanisms driving WML development in PD. Second, to better capture the full spectrum of WML-related cognitive effects, studies should include a broader range of cognitive severity, ideally spanning patients with PD with and without dementia. Third, we used total basal forebrain volume as a proxy for cholinergic projection nuclei. However, this measure is inherently noisier than our ¹⁸F-FEOBV synaptic terminal marker because the basal forebrain contains mixed neuronal populations. Similarly, the white matter tracts studied via the CHIPS scale contain both cholinergic and non-cholinergic basal forebrain efferents. Some of the effects on cognition could be secondary to dysfunction of non-cholinergic effects. Additionally, our mechanistic framework was simplified for mediation testing and does not model interactions with other neurotransmitter systems, although we partially account for dopaminergic involvement by adjusting for the levodopa equivalent dose. Multitracer imaging will be necessary to characterize these multisystem interactions. Despite these limitations, our results highlight the vulnerabilities of cholinergic pathways to periventricular WMLs, which contribute to cognitive dysfunction in PD.

Conclusions

Our study demonstrates that periventricular WMLs contribute to cognitive vulnerability in PD by disrupting long-range cholinergic pathways and accelerating cholinergic degeneration. Cognitive trajectories in PD are highly heterogeneous,⁷⁰ and there is substantial effort in the literature to identify the neurobiological factors that determine which trajectory a patient follows, particularly the mechanisms underlying early cognitive decline and dementia. Our findings position WMLs, especially periventricular lesions intersecting cholinergic projection fibres, as a meaningful contributor to this heterogeneity. Beyond their mechanistic role, WMLs may serve as a useful biomarker for predicting cognitive trajectories when integrated with other molecular and structural markers. Importantly, because WMLs often arise from potentially modifiable processes, they may also represent a promising therapeutic target. Together, these results support incorporating WML burden into models of cognitive prognosis in PD and underscore its relevance in strategies aimed at preventing or slowing cognitive decline.

Data availability

The dataset used and analysed during the present study can be made available by the corresponding author or the last author upon reasonable request from qualified researchers. Because the dataset includes highly personal neuroimaging data, access will require a formal data sharing agreement approved by the relevant institutional ethics bodies. To balance participant protection with open science, we have made the group-level t-maps derived from the voxel-wise analyses publicly available in the Supplementary material. These include: (i) the periventricular WML-associated cholinergic cluster mask; (ii) the deep WML-associated cholinergic cluster mask; and (iii) the cognitive-domain masks, so that other groups may use these results in future research.

Funding

Supported by P50NS123067, P50NS091856, R01AG073100, I01RX003397, I01RX001631, K99NS146570, Parkinson’s Foundation, and Michael J Fox Foundation.

Competing interests

P.S. receives grant funding to the institution from Atonco and Radionetics and is a consultant for GE Healthcare and Radionetics.

Supplementary material

Supplementary material is available at Brain online.