Regional cortical network atrophy predicts progression to dementia in the Lewy body diseases

Abstract

INTRODUCTION

The timeline from symptom onset to the loss of independent functioning varies across patients with Lewy body disease (LBD). Here, we investigate whether the magnitude of cortical atrophy within functional networks that subserve cognitive and affective functions has utility in predicting progression to dementia in LBD.

METHODS

Forty‐six LBD patients with intact instrumental and basic activities of daily living at baseline underwent brain magnetic resonance imaging scans and longitudinal clinical assessments. Cortical atrophy was estimated in LBD patients relative to an amyloid‐negative control group.

RESULTS

Multivariate Cox regression analysis demonstrated that atrophy in the affective salience and limbic networks, but not in cognitive, motor, or visual networks, predicted progression to dementia after controlling for disease duration, diagnosis, amyloid beta status, and baseline cognitive severity.

DISCUSSION

Baseline atrophy in the salience and limbic networks is an important predictor of progression to dementia and may have value in identifying early‐stage LBD patients at risk for faster progression.

Highlights

• In this longitudinal study of the Massachusetts Alzheimer's Disease Research Center's Memory and Aging Cohort, although the magnitude of cortical atrophy in the early stages of Lewy body disease (LBD) was highly variable at the individual level, atrophy in the salience–limbic network predicted progression to dementia in LBD.

• In the early stages of LBD, each one standard deviation increase in salience–limbic atrophy doubled the risk of dementia.

• These findings were robust to adjustment for disease duration, diagnosis, amyloid beta status, and baseline cognitive severity.

1. BACKGROUND

Dementia with Lewy bodies (DLB) and Parkinson's disease (PD) are two important neuronal α‐synuclein diseases (NSD) and are also known as the Lewy body diseases (LBDs), on the basis of Lewy bodies and Lewy neurites rich in misfolded α‐synuclein. ¹ , ² , ³ Although cognitive impairment is universal in DLB and is common later in the course of PD as well, the interval from the onset of cognitive impairment to loss of independent functioning (dementia) varies widely among patients with LBD. ⁴ Identifying those who exhibit rapid cognitive decline is critical for clinical trials, highlighting the need for reliable biomarkers that can predict and monitor progression to dementia in the early stages of LBD.

To date, few biomarkers that predict progression in LBD have been identified. Although fibrillar α‐synuclein assays in the cerebrospinal fluid (CSF) or skin biopsy (seed amplification assay; immunohistochemistry) are associated with high diagnostic accuracy, ⁵ , ⁶ , ⁷ , ⁸ , ⁹ both are dichotomous measures, and their prognostic value remains uncertain. Neurofilament light chain (NfL) in blood or CSF tracks motor and cognitive decline in PD and DLB but lacks specificity and is influenced by systemic factors. ¹⁰ , ¹¹ , ¹² In addition, common co‐pathologies in LBD, including Alzheimer's disease (AD), transactive response (TAR) DNA‐binding protein 43 (TDP‐43) deposition, and cerebrovascular pathology, can affect both clinical phenotype and disease progression. ¹³ , ¹⁴ Of these, biomarkers of amyloid beta (Aβ) and tau deposition in LBD have been found both to correlate with cognitive impairment cross‐sectionally and to predict faster cognitive decline. ¹⁵ , ¹⁶ , ¹⁷ Biomarkers for Aβ and tau do not capture underlying Lewy body pathology; however, modalities that assess downstream, cumulative changes resulting from the combined pathologies present in patients with LBD may provide more effective predictors of progression.

Magnetic resonance imaging (MRI)–based structural biomarkers that quantify brain volume and cortical thickness are highly sensitive techniques that have proven useful in studies of LBD. We and others have reported distinct patterns of cortical and subcortical atrophy in patients with LBD who are affected by AD co‐pathology. ¹⁸ , ¹⁹ , ²⁰ Previous research further highlights the utility of baseline imaging features to predict the trajectory of clinical decline in various neurodegenerative syndromes. ²¹ , ²² , ²³ , ²⁴ Recent advances in neuroimaging have identified markers of large‐scale brain networks that subserve multiple domains of cognitive and affective regulation, including executive functions, language, memory, socio‐affective processes, and motivation. ²⁵ These networks are critical for independence in advanced and instrumental activities of daily living ²⁶ and have gained utility in prognostication in neurodegenerative diseases. ²⁷ , ²⁸ In primary progressive aphasia, baseline atrophy in cognitive and affective networks was a key predictor of progression to dementia. ²⁷ In another study with patients with posterior cortical atrophy (PCA), baseline atrophy in the dorsal attention network predicted the rate of clinical progression. ²⁸ Atrophy within affective and salience networks is associated with dysfunction in mood, arousal, and goal‐directed behavior, ²⁹ symptoms frequently observed in LBD, and this network involvement may accelerate decline in functional independence. Thus an approach focusing on these networks may have value in predicting progression in LBD as well.

Here, we tested the hypothesis that in LBD, baseline atrophy within large‐scale brain networks subserving cognitive and affective functions can predict progression to dementia. As cognitive impairment is often accompanied by parkinsonism and hallucinations in LBD, we also compared the utility of cortical atrophy in cognitive‐affective networks and in motor and visual networks to predict progression to dementia in patients with LBD.

2. METHODS

2.1. Participants

A total of 46 patients, including 35 with PD without dementia and 11 with mild cognitive impairment (MCI) with Lewy bodies (MCI‐LB), were recruited from the Massachusetts Alzheimer's Disease Research Center's Memory and Aging Cohort (MADRC). Among the PD patients without dementia, there were 26 with normal cognition and 9 with MCI (PD‐MCI). All patients with PD met the clinical diagnostic criteria of the UK Parkinson's Disease Society Brain Bank, ³⁰ and those with PD‐MCI fulfilled Level II Movement Disorder Society Task Force guidelines. ³¹ A diagnosis of PD‐MCI required preserved independence of activities of daily living in association with impairment in at least two cognitive domains assessed with the National Alzheimer's Coordinating Center (NACC) Uniform Data Set (UDS) detailed cognitive testing. All patients diagnosed with MCI‐LB satisfied the consortium criteria for probable MCI‐LB with MCI and preserved independent daily living activities. ³² , ³³ Exclusion criteria included cerebral vascular disease, brain tumors, brain traumatic injury, and neurosurgery. Sixteen patients (34.8%) came to autopsy, and all had LBD.

RESEARCH IN CONTEXT

1. Systematic review: The pattern of cortical atrophy in Lewy body dementia is widespread across multiple cortical networks, including cognitive–affective networks and Lewy body–specific networks, such as motor and visual networks, whereas the somatosensory region is usually spared. The timeline from the onset of symptoms to the loss of independent functioning varies among patients with Lewy body disease (LBD). In this study, we set out to assess whether the magnitude of cortical atrophy within specific functional networks in LBD can serve as a predictor of progression to dementia.

2. Interpretation: In the early stages of LBD, cortical atrophy in the salience–limbic networks was the strongest predictor of progression to dementia, outperforming the predictive utility of other networks examined. For every one standard deviation increase in atrophy within the salience–limbic networks, the patient's risk of progression to dementia was about twofold higher. No additional variance was explained by baseline clinical severity, suggesting that the salience–limbic networks contribute uniquely to the progression to dementia beyond what is predicted by clinical severity measures.

3. Future directions: The measurement of salience network atrophy at baseline in LBD has value for the prediction of progression to dementia. The utility of this prognostic biomarker in preclinical and prodromal LBD patients warrants evaluation.

All patients underwent clinical assessment and detailed neuropsychiatric examination, ³⁴ , ³⁵ including the Montreal Cognitive Assessment (MoCA) or the Mini‐Mental State Examination (MMSE), and all patients were followed for at least 6 months and up to 15 years. At each visit, participants completed the UDS neuropsychological test battery assessing multiple cognitive domains, along with the Neuropsychiatric Inventory (NPI) for behavioral symptoms. Functional independence was evaluated using the Clinical Dementia Rating (CDR) and Functional Assessment Questionnaire (FAQ) scales, incorporating both participant and informant input. At baseline, all patients had independent daily living functioning. Cognitive severity was evaluated by MoCA scores, and in those where MoCA scores were not available, MMSE scores were converted to MoCA scores as described previously. ³⁶ Disease duration was defined based on the symptom onset reported in clinical notes by patients or their caregivers. For each patient, we measured the time between the first visit and the follow‐up visit at which progression to dementia was documented or the last observation if the patient remained nondemented. Dementia was defined as a clinical state involving deficits across multiple cognitive domains that substantially impair independent functioning in occupational or daily activities. ³⁷

All patients provided informed consent under protocols approved by the Mass General Brigham Institutional Review Board. Anonymized data are available for sharing with qualified investigators in compliance with MADRC data sharing protocol.

2.2. Neuroimaging

At baseline, all patients underwent a structural MRI scan on a Siemens Tim Trio 3T system equipped with a 12‐channel phased array head coil as previously described. ¹⁸ High‐resolution structural T1‐weighted magnetization‐prepared rapid gradient‐echo (MPRAGE) sequences were acquired with the following parameters: repetition time (TR) = 2.3 ms, echo time (TE) = 2.98 ms, flip angle = 9°, field of view = 240 mm × 256 mm, and slice thickness = 1 mm and TR = 2.53 ms, TE = 1.64 ms, flip angle = 7°, field of view = 256 mm × 256 mm, and slice thickness = 1 mm. Head movement was minimized using expandable foam cushions, and automated scout and shimming procedures were performed for all patients.

Each patient's three‐dimensional (3D) T1‐weighted MRI scan was analyzed using FreeSurfer version 6 to estimate cortical thickness (http://surfer.nmr.mgh.harvard.edu). All MRI data were visually inspected for gross artifacts (e.g., subject motion) and evaluated on image quality prior to data processing as described previously. ³⁸ Once they passed quality control, each participant's 3D T1 data underwent intensity normalization, skull stripping, and automated segmentation of cerebral white matter to locate the gray matter/white matter boundary via FreeSurfer v6.0, which is documented and freely available for download online (http://surfer.nmr.mgh.harvard.edu). Defects in the surface topography were corrected, and the gray/white matter boundary was deformed outward using an algorithm designed to obtain an explicit representation of the pial surface. We visually inspected each participant's cortical surface reconstruction for technical accuracy. Whole‐brain maps of cortical thickness were registered to template surface space (fsaverage) and smoothed geodesically with a full‐width‐half‐maximum (FWHM) of 15 mm. ³⁸

We defined the large‐scale functional cortical networks of interest based on the Yeo atlas ²⁵ and previous literature ³⁹ (Figure 1). The networks analyzed were the default mode network (DMN), dorsal attention network (DAN), ventral attention network (VAN, largely overlapping with the salience network), and the frontoparietal (FPN), limbic, visual (VN), and somatomotor (SMN) networks. The somatosensory network (SSN) was included as a comparison network with no predictive value. The mean cortical thickness within each network label was then calculated for every participant by averaging thickness estimates at all vertices falling within its boundaries. To account for global individual differences in cortical thickness, we converted each participant's cortical thickness estimates to W‐scores, which are analogous to Z‐scores adjusted for specific covariates of no interest, which in this study were age and sex (for more details, see Touroutoglou et al., 2023 ³⁸ ). Briefly, we first performed a multiple regression analysis using mean cortical thickness data obtained from 25 Aβ‐negative cognitively normal participants (mean age ± SD: 67.4 ± 4.8, sex: 13M/12F), which resulted in beta coefficient values for age and sex as well as individual values of residuals. Using these parameters, we then computed W‐scores for all LBD patients with the following formula:

$$W_{\mathrm{ik}}=\frac{T_{\mathrm{ik}}-\widehat{T_{\mathrm{ik}}}}{SD\mathrm{i}}$$

FIGURE 1.

Brain networks of interest. The visual, somatomotor, dorsal attention, ventral attention (largely overlapping with the salience network), limbic, and frontoparietal networks are defined by the Yeo atlas. The somatosensory region extracted from the Glasser atlas served as a control region.

 Where $T_{\mathrm{ik}}$ = the observed mean cortical thickness of label i and patient k, $\widehat{T_{\mathrm{ik}}}$ = the predicted mean cortical thickness of label i and patient k based on age and sex of this patient as well as beta coefficients obtained from cognitively normal participants, and SD = the standard deviation of the individual residuals obtained from cognitively normal participants for label i. Disease duration and laterality were not incorporated due to unavailability in the control reference group. To facilitate interpretation in subsequent analyses, we inverted W‐scores into atrophy scores, such that higher atrophy scores in these networks reflect more pronounced cortical atrophy.

[11C] Pittsburgh compound B (PiB)‐PET scans were performed on a Siemens/CTI ECAT HR + system operating in three‐dimensional mode (63 transaxial slices; 15.2 cm axial field of view; 5.6 mm in‐plane resolution; 2.4 mm slice spacing). After an intravenous bolus of 8.5–15 mCi [11C] PiB, dynamic emission data were acquired continuously for 60 min. Parametric images were generated as distribution‐volume‐ratio (DVR) maps using cerebellar gray matter as the reference tissue. Cortical amyloid burden was summarized with a composite region of interest encompassing frontal, lateral temporal, and retrosplenial cortices (FLR). Positive Aβ load was defined as an FLR DVR ≥1.32 after partial‐volume correction, as described previously. ⁴⁰ When multiple PiB scans were available, the examination closest to the baseline visit was selected for analysis.

2.3. Statistical analyses

We conducted survival analyses with time‐to‐event data collected in the MADRC's Memory and Aging Longitudinal Cohort. To compare baseline demographic and clinical features between MCI‐LB and PD subgroups, continuous dependent variables were analyzed with two‐tailed t‐tests for normally distributed variables or Wilcoxon rank‐sum tests for non‐normally distributed variables. Bivariate relations among categorical variables were assessed using the chi‑square test or Fisher's exact test when any expected cell count was below five. To test our first hypothesis regarding the relationship between cortical network atrophy and the likelihood of progression to dementia, we built a univariate Cox regression model separately for each network. The outcome variable was the time between the baseline and progression to dementia or last available visit. The predictor variables were demographic and clinical features, including age, sex, years of education, disease duration, MoCA scores at baseline, clinical diagnosis (MCI‐LB vs PD), and atrophy scores of each network.

To investigate our second hypothesis regarding the independent and synergistic contributions of functional networks to dementia, we constructed a multivariate Cox regression model informed by the findings of univariate Cox regression analyses. Given the very high degree of collinearity among atrophy scores for the networks, we performed an exploratory factor analysis to delineate the contributions of atrophy in latent factors derived from the seven functional networks with respect to progression to dementia. An orthogonal rotation method was employed, ensuring that the extracted factors were uncorrelated following rotation. In accordance with Kaiser's Criterion, factors with eigenvalues exceeding 1 were retained, as they account for more variance than a single variable. The chi‐square test in factor analysis, which compares the observed correlation matrix with the matrix reproduced by the factor model, identified several latent factors in the dataset. In the subsequent multivariate Cox regression model, we included the latent network factors as predictors, alongside disease duration, clinical diagnostic groups, and baseline cognitive severity (MoCA scores) as covariates, with the time from baseline to either the onset of dementia or the most recent follow‐up visit serving as the outcome variable. Hazard ratios (HRs) and the corresponding 95% confidence intervals (CIs) were reported for each univariate or multivariate Cox regression model. All Cox regression models were evaluated for proportional hazards assumptions, influential observations, and potential non‐linear effects using the “survival” and “survminer” packages. All analyses in this study were conducted using R version 4.4.0. All tests were performed at a significance level of p < 0.05. False discovery rate (FDR) correction was used to control for multiple comparisons, and the p‐value threshold for FDR was set at 0.05.

3. RESULTS

3.1. Participant characteristics

Baseline demographic, clinical, and cognitive characteristics of the 46 participants, including 35 PD without dementia and 11 MCI‐LB, are summarized in Table 1. The age at baseline ranges from 56 to 84 years (70.8 ± 6.8) and was comparable between the PD and MCI‐LB groups (p = 0.14). As expected, women constituted a lower proportion (28%) of participants with LBD than men. MoCA scores ranged from 16 to 30 in the PD without dementia group (25.2 ± 2.7) and ranged from 14 to 27 in the MCI‐LB group (21.4 ± 4.4). Parkinsonism was present in all LBD participants, whereas visual hallucinations were less common (37%). The average follow‐up duration was 5.3 (SD: 3.6) years. At the last available follow‐up visit, 20 of 46 participants had progressed to dementia. PiB‐PET scans were available in all patients, and 17 (37%) of them showed Aβ positivity.

TABLE 1.

Demographic and baseline clinical characteristics of all patients with LBD.

Characteristics	Total LBD	PD without dementia	MCI‐LB	p‐value
N	46	35	11	–
Age, years	70.75 (6.83)	70.16 (7.56)	72.61 (3.25)	0.14
Female, %	13 (28)	12 (34)	1 (9)	0.15
Education, years	16.33 (2.75)	16.34 (2.88)	16.27 (2.41)	0.76
Disease duration, years	8.54 (5.13)	9.64 (5.38)	5.01 (1.50)	0.003 *
MoCA score	24.29 (3.56)	25.21 (2.73)	21.36 (4.41)	0.005 *
CDR‐SOB	0.97 (1.13)	0.44 (0.6)	2.64 (0.71)	<0.001 *
Aβ positivity (Y/N)	17/29	10/25	7/4	0.071
Progression to dementia (Y/N)	20/26	11/24	9/2	0.005 *
Duration of follow‐up after baseline, years	5.31 (3.58)	6.08 (3.74)	2.85 (1.25)	0.008 *
Parkinsonism (Y/N)	46/0	35/0	11/0	–
Visual hallucination (Y/N)	17/29	9/26	8/3	0.01 *
Levodopa‐equivalent daily dose, mg	621 (454)	772 (401)	141 (218)	<0.001 *
Cholinesterase inhibitors (Y/N)	15/31	4/31	11/0	<0.001 *

Abbreviations: CDR, Clinical Dementia Rating; LBD, Lewy body disease; MCI‐LB, mild cognitive impairment with Lewy bodies; MoCA, Montreal Cognitive Assessment; PD, Parkinson's disease. Mean (standard deviations) are shown.

^* p < 0.05.

3.2. The pattern of baseline cortical atrophy within large‐scale brain networks

The atrophy scores for each of the seven brain networks and the comparison network are displayed in Figure S1. Higher atrophy scores in these networks correspond to more pronounced cortical atrophy. The average atrophy scores in each network were close to zero, indicating that at the group level, the magnitude of cortical atrophy in these networks did not differ substantially from amyloid‐negative cognitively normal controls after adjusting for the effects of age and sex in the non‐demented stage of LBD. However, at the individual level, the extent of cortical network atrophy varied considerably among participants. For instance, the dorsal attention network shows scores ranging from −2.9 to 3.7, the visual network from −2.6 to 2.4, and the motor network from −2.8 to 2.9. In contrast, the variation in the somatosensory comparison network was minimal, with an SD of 0.87 and a range from −1.5 to 1.5, suggesting that the somatosensory region is typically spared.

We also calculated the mean vertex‐wise, whole‐cortex atrophy score map across all LBD patients, revealing a relatively preserved and sparse pattern of cortical atrophy (Figure 2A), consistent with our previous study. ¹⁸ To characterize the distributed topography of atrophy in terms of functional networks, we investigated the percentage of all vertices with values beyond an atrophy score threshold of 0.5 belonging to each of the seven networks of interest. This analysis showed that subtle atrophy was present in multiple functional networks, including not only cognitive–affective networks but also motor and visual networks affected in LBD (Figure 2B). When we stratified patients into amyloid‐positive and amyloid‐negative groups, as expected, we found that a more pronounced pattern of cortical atrophy was evident in the amyloid‐positive group in contrast with the amyloid‐negative group (Figure S2).

FIGURE 2.

Spatial topography of cortical atrophy in the Lewy body diseases. (A) Cortical network atrophy scores (inverted W‐scores) are displayed at each vertex on the cortical surface maps. Higher atrophy scores indicate greater atrophy at baseline in patients with Lewy body disease relative to amyloid‐negative cognitively normal controls. (B) Bar graph indicates the percentage of all vertices showing atrophy scores above 0.5 at the group level falling within the boundaries of each functional network of interest.

3.3. Cortical network predictors of progression to dementia

Univariate Cox regression models of age, sex, education, and disease duration indicated that these variables did not significantly predict time to dementia (p > 0.05). In contrast, univariate Cox regression models of baseline MoCA scores (HR: 0.84, 95% CI: 0.75–0.94, p = 0.003), Aβ positivity (PiB+ vs PiB‐HR: 3.13, 95% CI: 1.2–8.17, p = 0.019), and diagnostic groups (MCI‐LB vs PD HR: 7.1, 95% CI: 2.48–20.3, p < 0.001) showed that each of these variables significantly predicted time to dementia (Table 2). We constructed a series of univariate Cox regression models to test the effect of cortical atrophy in different functional networks at baseline on subsequent progression to dementia in all patients with LBD. The magnitude of baseline cortical atrophy in all functional networks except the somatosensory region was predictive of a greater rate of progression to dementia in LBD (Table 2; Figure 3). All of the seven networks survived correction for multiple comparisons (FDR‐corrected p < 0.05).

TABLE 2.

Univariate Cox regression models to test demographic, baseline clinical characteristics, or brain network atrophy in predicting time of progression to dementia in LBD.

	χ2	HR	95% CI	p‐value
Age	3.19	1.07	0.99–1.15	0.074
Female vs male	0.47	0.68	0.23–2.03	0.49
Education	1.81	1.12	0.95–1.31	0.18
Disease duration	0.17	0.98	0.88–1.09	0.68
MoCA score	8.83	0.84	0.75–0.94	0.003 *
PiB+ vs PiB–	5.44	3.13	1.2–8.17	0.019 *
MCI‐LB vs PD	13.4	7.1	2.48–20.3	<0.001 *
Levodopa‐equivalent daily dose	0.92	1.00	0.99–1.01	0.34
Cholinesterase inhibitors	16.5	14.2	3.9–51.3	<0.001 *
Magnitude of baseline atrophy in brain networks
Visual network	8.94	2.29	1.33–3.93	0.003 * , a
Default mode network	8.57	2.07	1.27–3.37	0.004 * , a
Limbic network	8.58	1.99	1.26–3.15	0.003 * , a
Somatomotor network	8.44	1.94	1.24–3.04	0.004 * , a
Ventral attention network	5.27	1.83	1.09–3.05	0.022 * , a
Frontoparietal network	4.9	1.63	1.06–2.5	0.027 * , a
Dorsal attention network	6.26	1.53	1.09–2.06	0.012 * , a
Somatosensory region	2.39	1.52	0.89–2.57	0.12

Abbreviations: CI, confidence interval; FDR, false discovery rate; HR, hazard ratio; LBD, Lewy body disease; MCI‐LB, mild cognitive impairment with Lewy bodies; MoCA, Montreal Cognitive Assessment; PD, Parkinson's disease; PiB, Pittsburgh compound B.

^aSignificant after controlling for multiple comparisons across eight brain networks (false discovery rate [FDR] p < 0.05).

* p < 0.05.

FIGURE 3.

Hazard ratios from univariate Cox regression analyses to predict time of progression to dementia in Lewy body disease. (See Figure 1 for a description of networks.)

To examine the partial independence or synergistic contributions of atrophy in different networks to progression to dementia, we constructed a multivariate regression model. Due to a high degree of collinearity between atrophy scores of networks (Figure S3), we performed an exploratory factor analysis and examined the contributions of atrophy in the three latent factors: (1) The visual‐motor‐DAN networks, (2) the DMN‐frontoparietal network (DMN‐FPN) networks, and (3) the salience–limbic networks (Figure 4). These factors jointly explained 84.1% of the total variance, a substantial proportion of the overall variability in the data. These factors represent the contributions of multiple networks as named.

FIGURE 4.

Latent network factors identified by factor analysis of atrophy scores in seven functional brain networks. Factor analysis identified three uncorrelated latent factors in the dataset: (1) the visual‐motor‐DAN factor, (2) the DMN‐FPN factor, and (3) the salience–limbic factor. These three factors jointly explained 84.1% of the total variance. DAN, dorsal attention network; DMN‐FPN, default mode network–frontoparietal network.

Atrophy in the salience–limbic networks was found to be the strongest predictor of progression to dementia, after controlling for disease duration, clinical diagnosis, Aβ positivity, and baseline cognitive severity (HR: 1.86, 95% CI: 1.04–3.35, p = 0.038). For every 1 SD increase in atrophy within the salience–limbic networks, the risk of progression to dementia was about twofold higher. Participants with MCI‐LB had much faster progression to dementia compared to participants with PD without dementia (HR: 7.34, 95% CI: 1.73–31.18, p = 0.007). No additional variance was explained by the DMN‐FPN networks, visual‐motor‐DAN networks, baseline MoCA scores, Aβ burden, or disease duration (Table 3). These results suggest that the atrophy in the salience–limbic networks is a better predictor of progression to dementia than baseline MoCA score. These findings also identify the differential contribution of the salience–limbic networks over the visual‐motor‐DAN or DMN‐FPN networks in predicting progression to dementia. Incidence predictive curves for dementia risk stratified by atrophy in the salience–limbic networks in PD without dementia and MCI‐LB groups are plotted in Figure 5.

TABLE 3.

Multivariate Cox regression models to test the partial independence or synergistic contributions of atrophy in different networks to progression to dementia in LBD.

	χ2	HR	95% CI	p‐value
Overall	23.01			0.002 *
Disease duration		1.12	0.98–1.29	0.09
MoCA score		0.92	0.81–1.06	0.27
PiB Aβ positivity		1.66	0.48–5.73	0.42
MCI‐LB vs PD		7.34	1.73–31.18	0.007 *
Salience–limbic		1.86	1.04–3.35	0.038 *
Visual‐motor‐DAN		1.44	0.74–2.78	0.28
DMN‐FPN		0.88	0.54–1.43	0.61

Abbreviations: CI, confidence interval; DAN, dorsal attention network; DMN‐FPN, default mode network–frontoparietal network; HR, hazard ratio; LBD, Lewy body disease; MCI‐LB, mild cognitive impairment with Lewy bodies; MoCA, Montreal Cognitive Assessment; PD, Parkinson's disease; PiB, Pittsburgh compound B.

^* p < 0.05.

FIGURE 5.

Predictive value of atrophy in the salience–limbic networks for progression to dementia in Lewy body disease. Multivariate Cox regression model controlling for disease duration, clinical diagnosis, amyloid positivity, and baseline cognitive severity. Individuals with baseline cortical thickness in this network more than one standard deviation below controls are those at highest risk, as indicated with the red line.

4. DISCUSSION

In this study, we found that greater brain network atrophy in non‐demented LBD participants is predictive of faster transition to dementia. This measure was more predictive than standard clinical severity measures alone. We also identified atrophy in the salience–limbic networks, in particular, as a robust predictor of progression to dementia that was superior to other cognitive, motor, and visual networks in LBD.

The salience–limbic networks have particular relevance in LBD. The salience network, also known as the ventral attention network (Yeo atlas), includes the dorsal anterior cingulate cortex, anterior insular cortex, temporoparietal junction, and middle frontal gyrus. ²⁹ , ⁴¹ , ⁴² This network underlies a range of functions, spanning attention, interoception, and maintenance of tonic alertness and sympathetic drive (e.g., connectivity to adrenal medulla), as well as subjective awareness, affect, and subjective arousal experiences. ²⁹ , ⁴² , ⁴³ , ⁴⁴ Recent studies have also suggested that this network may act as a switch between goal‐directed behavior (frontoparietal control network) and self‐referential, emotional, and memory processing (DMN). ⁴⁵ , ⁴⁶ , ⁴⁷ As such, it may serve as a generalized mechanism for detecting both internal and external salient events and for initiating and allocating resources toward appropriate behavioral responses. ⁴⁸ Dysfunction of this network could account for core features of DLB, such as interruptions in attention to stimuli, ultimately impairing functional abilities and contributing to cognitive decline.

The limbic network, encompassing the temporal pole and medial temporal cortex, is crucial for memory processes ⁴⁹ and is closely linked to the initiation of tau pathology ⁵⁰ and the spread of amyloid pathology. ⁵¹ At the transitional Lewy body stage of LBD, Lewy bodies and Lewy neurites typically accumulate within limbic network regions—as well as in the anterior cingulate cortex. ⁵² In this context, our observation linking atrophy in the limbic network to dementia progression at the early stage of LBD likely reflects the interplay of multiple underlying pathologies that commonly arise in LBD. Although the earliest cognitive deficit of LBD is often impaired executive function rather than memory loss, ⁴ , ⁵³ it is possible that the emergence of memory impairment may exert a pivotal influence in the timeline to dementia. Although we observed cortical atrophy in DMN, FPN, visual, motor, and DAN networks, these functional networks did not enhance predictions of dementia onset in LBD after accounting for atrophy scores in other networks and covariates. These findings suggest that cortical thickness measures from a combination of both the salience and the limbic networks contribute uniquely to the progression to dementia in LBD.

Consistent with prior studies, we found that participants with DLB progress more rapidly to dementia than those with PD. ⁵⁴ In this regard, it is noteworthy that most PD patients in this study had intact cognition at baseline, whereas all patients with DLB presented with MCI. Poorer baseline cognitive performance has been linked to accelerated cognitive decline in both PD and DLB, ⁵⁵ , ⁵⁶ and our univariate Cox regression model using the MoCA score as a predictor yielded similar findings. Consequently, both clinical diagnostic groups and baseline cognitive performance were incorporated into the final multivariate Cox regression model that identified a strong predictive effect of atrophy in the salience–limbic networks on the time to dementia in ,   LBD.

Key strengths of this study include its longitudinal design, the inclusion of clinically well‐characterized LBD patients, and the use of age‐ and sex‐adjusted atrophy scores. There were also a number of limitations. First, the strong collinearity among atrophy scores hampered the detection of independent and synergistic effects in each network. We used factor analysis to address this issue, which yielded robust findings on the predictive value of the salience–limbic networks. Second, data on amyloid and tau status were available only in a subset of patients, thereby limiting our ability to perform further analyses. Even so, this did not detract from our primary aim focused on predicting dementia in LBD with MRI‐derived atrophy scores, which may capture downstream effects of not only AD co‐pathology but other pathologies as well. Third, consistent with the natural history of LBD, disease duration varied considerably across individuals, prompting its inclusion as a covariate in the multivariate model to account for heterogeneous disease progression. After controlling for disease duration, atrophy scores in the salience–limbic networks remained significant. An additional limitation is the absence of complementary biomarkers of LBD and neuronal loss, including, but not limited to, α‐synuclein biomarkers, dopaminergic and cholinergic PET tracers, neuromelanin MRI, and plasma NfL, which could further substantiate and extend the present findings in future investigations. Overall, the results of this study are exploratory and warrant validation in larger future studies, including those focusing on prodromal or preclinical LBD populations.

Together, the results of this study identify baseline atrophy in cortical salience and limbic networks subserving cognitive–affective function as a predictor of progression to dementia in LBD. This measure may prove useful to identify patients with early‐stage LBD who are at risk for faster ,   progression to dementia.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

CONSENT STATEMENT

All patients provided informed consent under protocols approved by the Mass General Brigham Institutional Review Board. The studies were performed in accordance with the Declaration of Helsinki and its later amendments.

Supporting information

Supporting Information

Supporting Information

Ye R, Pereira TP, Locascio JJ, et al. Regional cortical network atrophy predicts progression to dementia in the Lewy body diseases. Alzheimer's Dement. 2025;21:e70963. 10.1002/alz.70963