CSF proteome profiling reveals biomarkers to discriminate dementia with Lewy bodies from Alzheimer's disease

Dementia with Lewy bodies (DLB) is a common dementia after Alzheimer's disease (AD), with overlapping clinical and pathological features that complicate differential diagnosis. Although seed amplification assays can detect α-synuclein proteinopathy with high accuracy, α-synuclein pathology is not unique to DLB and is present in a substantial fraction of AD cases. Classical AD CSF biomarkers (Aβ1-42, total tau, phosphorylated tau) have limited accuracy for distinguishing DLB from AD because 25–40% of DLB patients show abnormal AD biomarker profiles due to comorbid pathology. There is a need for additional CSF markers capturing distinctive aspects of DLB pathophysiology for use in diagnosis, prognosis, disease monitoring, and clinical trials. CSF proteome profiling enables broad in vivo assessment of biological processes and has identified biomarker candidates in AD. Prior DLB proteomic studies were limited by small sample sizes. This study aimed to define CSF proteomic changes underlying DLB and to identify, develop, and validate multiplex biomarker assays to aid specific DLB diagnosis.

Previous CSF studies of α-synuclein reported conflicting results, though newer seed amplification assays can distinguish DLB from controls and AD with high accuracy; however, α-synuclein pathology frequently co-occurs in AD. Core AD CSF biomarkers (Aβ42, tTau, pTau) are insufficient to discriminate DLB from AD because many DLB patients have abnormal profiles. Earlier DLB CSF proteomic studies, often with 30–40 samples per group, identified few candidates, likely due to heterogeneity and limited power. In AD, proteomic profiling has successfully revealed biomarkers and pathophysiological pathways, motivating a similar high-throughput, scalable approach in DLB.

Design: A discovery CSF cohort (n=534; DLB=109, AD=235, controls=190) was profiled using Olink proximity extension assay (PEA) panels (11 panels; initial 979 assays). Quality control retained 665 proteins (642 unique) with ≥85% detectability. Samples were randomized across plates with manufacturer QC, run in two rounds with bridging samples for normalization. Statistical analysis of differential abundance used nested linear models adjusting for age and sex, controlling False Discovery Rate at q≤0.05 per comparison. Functional enrichment used Metascape on GO Biological Processes. Classification modeling: Penalized generalized linear models (elastic net via glmnet) with age and sex covariates were used to construct minimal-marker signatures for DLB vs control and DLB vs AD. Grids of mixing parameters and maximum selected features (up to 21) were compared. Optimal models were chosen by 10-fold stratified cross-validation based on AUC and parsimony; final signatures were ridge-regularized (penalty 0.1) and internally validated by repeated 5-fold cross-validation (1000 repeats). Performance was assessed by ROC/AUC with bootstrap 95% CIs. Custom assay development and external validation: Custom multiplex PEA assays were developed for six of the seven selected biomarkers (DDC, CRH, MMP-3, ABL1, MMP-10, THOP1). Assays showed mean intra-/inter-assay CVs of 5%/9% and >90% detectability. Three independent cohorts were analyzed: Validation 1 (54 DLB, 55 AD, 55 controls), Validation 2 (55 DLB, 55 AD, 55 controls; SPIN cohort), and an autopsy-confirmed AD/DLB cohort (17 aDLB, 30 aAD, plus 29 non-autopsy controls). Classification models and single-marker DDC were tested by ROC/AUC. Sensitivity analyses examined effects of parkinsonian medication, DAT scan status, and REM sleep behavior disorder (RBD) positivity. Pathophysiological associations: Exploratory correlations assessed relationships of CSF DDC, FCER2, CRH, and MMP-3 with UPDRS-III, post-mortem α-syn load across brain regions, α-syn Braak stage, and DLB stage where available. PD cohorts: Public CSF PEA data from Parkinson’s Progression Markers Initiative were analyzed (AMP-PD set: 93 controls, 44 prodromal PD, 33 PD; PPMI set: 37 controls, 36 PD). Differential abundance used nested linear models with FDR control; classification performance for DDC and the DLB panel was evaluated. Clinical/biomarker covariates: AD CSF biomarkers (Aβ42, tTau, pTau) were measured locally with standard platforms; harmonization used Passing–Bablock transformations where needed. Group comparisons for demographics used ANOVA or Kruskal–Wallis as appropriate; chi-square for categorical variables; multiple testing via Bonferroni for non-proteomic analyses.

• Differential proteins in DLB vs controls: 14 proteins significant after FDR (q<0.05); 6 upregulated (DDC, GH, IDUA, PRCP, KYNU, ENTPD5) and 8 downregulated (CRH, FCER2, MMP1, COL4A1, WIF1, PAM, VEGFA, CTSC). Up to 90 proteins showed nominal differences (p<0.05).
• Strongest effect: DDC (β=0.95; fold-change 1.9; top-ranked), followed by GH, MMP1, FCER2, CRH (fold-changes >1.5 among top hits).
• Specificity to DLB: 49 proteins (≈55% of nominally altered) uniquely dysregulated in DLB; 24 proteins differed across DLB vs AD and AD vs control, often with opposite directions between DLB and AD. Enrichment implicated negative regulation of myelination and pathways including Notch signaling and steroid metabolism.
• Discrimination performance (discovery cohort):
  • DDC alone: DLB vs control AUC 0.91 (95% CI 0.88–0.94); DLB vs AD AUC 0.81 (95% CI 0.76–0.86).
  • Seven-protein panel (DDC, FCER2, CRH, MMP-3, ABL1, MMP-10, THOP1): DLB vs control AUC 0.95 (95% CI 0.89–0.99); DLB vs AD AUC 0.93 (95% CI 0.86–0.98). Performance unaffected by AD CSF biomarker comorbidity within DLB.
• Medication effects: FCER2 decreased only in DLB cases on parkinsonian medication; DDC and CRH remained altered in medication-naïve DLB; panel maintained high accuracy in medication-free subsets.
• Comparison with AD CSF biomarkers: The DLB panel outperformed Aβ42/tTau for DLB vs control and had similar AUCs for DLB vs AD.
• Correlations (discovery): DDC and MMP-3 weakly negatively correlated with MMSE; CRH and MMP-3 moderately positively correlated with (p)Tau; DDC weakly negatively correlated with Aβ42. AD-specific markers (ABL1, MMP-10, THOP1) correlated most with AD biomarkers and MMSE.
• External validations with custom assays:
  • Validation 1: High accuracies similar to discovery (AUCs >0.86 for DDC and panel in both comparisons).
  • Validation 2: Lower accuracies, especially DLB vs AD (DDC AUC 0.59; panel AUC 0.68). Subset with RBD showed performance similar to discovery; DAT scan status did not improve accuracy.
  • Autopsy cohort: DDC AUCs 0.95 (DLB vs control) and 0.86 (DLB vs AD); panel AUCs 1.00 (DLB vs control) and 0.90 (DLB vs AD).
• Pathophysiological associations: CSF DDC correlated with UPDRS-III in Validation 1 (r=0.76, p<0.001), with α-syn load in specific brain regions, and increased across DLB stages and α-syn Braak stages. Associations for FCER2, CRH, MMP-3 with pathophysiology were less consistent across cohorts.
• PD cohort findings: In AMP-PD, DDC and panel discriminated prodromal PD and PD from controls (DDC AUCs 0.80 and 0.83; panel AUCs 0.87 and 0.90). In PPMI set, DDC AUC 0.71 and panel AUC 0.84 for PD vs controls.

This study addressed the clinical need for DLB-specific CSF biomarkers by large-scale proteomic profiling and translation to a practical multiplex panel. The strongest marker, DDC, reflects dopamine biosynthesis and aligns with nigrostriatal degeneration in DLB. The seven-protein panel, integrating DLB-associated markers (DDC, CRH, MMP-3, FCER2) and AD-associated markers (ABL1, MMP-10, THOP1), improves discrimination of DLB from AD relative to single markers and standard AD CSF biomarkers. Functional enrichment suggests a role for demyelination and hypothalamic–pituitary–adrenal axis dysregulation in DLB. DDC associated with motor severity, α-syn pathology burden, and disease stages, supporting utility for staging and monitoring. Validation across independent clinical and autopsy-confirmed cohorts demonstrates robustness, though clinical heterogeneity (e.g., absence of RBD, neuropsychiatric symptom differences) can impact performance. The dysregulation of these markers in PD, including the prodromal phase, indicates they track α-synuclein-related and dopaminergic pathophysiology, while remaining specific for DLB within the dementia context. Collectively, findings support application of DDC and the multiplex panel as diagnostic tools and potential quantitative measures for disease monitoring and clinical trial endpoints.

Comprehensive CSF proteome profiling identified DLB-specific protein alterations and yielded a six-plex custom assay (covering six of seven discovery markers) that accurately discriminates DLB from controls and AD, validated in multiple independent and autopsy-confirmed cohorts. DDC emerged as a strong single marker with pathophysiological correlates to motor severity and α-syn load, suggesting value for staging and monitoring. The antibody-based, scalable workflow facilitates translation to clinical settings. Future research should include longitudinal, highly phenotyped cohorts with neuropathology, integration with α-syn seed amplification assays and imaging biomarkers, evaluation across α-synucleinopathies and parkinsonian disorders, and studies of treatment effects and trajectories to define contexts of use for diagnosis, prognosis, and therapeutic monitoring.

• Parkinsonian medication modestly influenced some CSF proteins (notably FCER2), underscoring the need to account for treatment effects; medication data were limited in number.
• Lack of CSF α-synuclein measures in the discovery and validation data sets precluded direct comparison with α-syn assays.
• Potential diagnostic misclassification due to clinicopathological overlap with AD, although cohorts were well-characterized and included DAT-supported and autopsy-confirmed cases; sensitivity analyses mitigate this concern.
• Heterogeneity of DLB clinical features affected performance (e.g., lower accuracy in a cohort with fewer RBD cases and differing neuropsychiatric profiles).
• Some subgroup analyses (e.g., stratification by pathophysiological stages) were limited by small sample sizes.
• Markers are likely not specific across all α-synucleinopathies or dopamine-deficiency conditions, as dysregulation was observed in PD.