Two-step detection of Lewy body pathology via smell-function testing and CSF α-synuclein seed amplification

Abstract

Cerebrospinal fluid (CSF) α-synuclein (α-syn) seed amplification assays (SAAs) can detect Lewy body pathology (LBP) with high accuracy but are invasive and costly. To address these challenges, this study evaluated a two-step workflow combining prescreening via smell-function testing with confirmatory CSF α-syn SAA testing only in individuals with reduced smell, for predicting postmortem LBP status. Among 358 autopsied participants, the two-step workflow predicted brain LBP with high accuracy overall (94%), and within clinical subgroups (clinical parkinsonism=95%; clinical Alzheimer's disease [AD]=94%; clinically unimpaired [CU]=93%). It reduced the need for confirmatory CSF testing by 43% overall (23% clinical parkinsonism; 35% clinical AD; 80% CU). In an independent in vivo cohort (N=1209), the workflow predicted CSF α-syn SAA status with 79% accuracy and reduced CSF testing by 26%. This approach may reduce invasive CSF testing, alleviating patient burden and lowering healthcare costs.

Fig. 1. Two-step workflow design and results summary.

a Design of a conditional two-step workflow to detect Lewy body pathology (LBP). Step 1 consists of UPSIT-based risk stratification into high- and low-risk groups having cortical LBP (LBP_ctx). Step 2 includes confirmatory CSF α-syn seeding amplification assay (SAA) testing in high-risk patients identified in step 1. b The association between postmortem LBP, postmortem CSF α-syn SAA, and UPSIT scores (n = 358). c Summary of the proportion of individuals selected as high–risk, LB-positive based on postmortem CSF, LB-positive based on postmortem neuropathology in the cortex, and the reduction in CSF tests, for each of the four scenarios. Boxplots show the median, lower, and upper quartiles with whiskers representing minimum and maximum values. Source data are provided as a Source Data file. Comparisons were performed with two-sided linear regression models adjusted for multiple comparisons (tukey method). * p < 0.05; ** p < 0.01; *** p < 0.001. αsyn alpha-synuclein, CSF cerebrospinal fluid, ctx cortex, LBP Lewy body pathology, SAA seed amplification assay, UPSIT University of Pennsylvania Smell Identification Test.

Fig. 2. Performance of a two-step workflow for detecting cortical LB pathology.

Performance and reduction in CSF α-syn SAA tests across the whole cohort and subgroups of individuals with clinical signs of parkinsonism, clinical Alzheimer’s disease, and clinically unimpaired individuals. For illustrative purposes, probabilities and datapoints correspond to the median of 1000 iterations and thresholds corresponding to 95% sensitivity were used. a–c The median accuracy, positive predictive value (PVV) and negative predictive value (NPV) of the two-step, CSF-only, and UPSIT-only approaches. d The observed median percentage of reduction in CSF tests using the two-step workflow. Error bars correspond to 95% CIs based on 1000 iterations of model development and classification. αsyn alpha-synuclein, AD Alzheimer’s disease, CSF cerebrospinal fluid, NPV negative predictive value, PPV positive predictive value, SAA seed amplification assay, UPSIT University of Pennsylvania Smell Identification Test.

Fig. 3. Application of the two-step workflow to an in vivo clinical cohort.

The two-step workflow applied to an in vivo clinical cohort. a Distribution of model-derived probabilities for cortical Lewy body pathology (LBP) based on a logistic regression model including UPSIT scores, age, and sex as predictors, trained on the autopsy dataset. A probability threshold corresponding to 95% sensitivity, derived from the autopsy dataset, was used to classify individuals as low- (blue dots) or high- (red dots) risk for cortical LBP. b Performance metrics of the UPSIT-based risk classification groups for predicting CSF α-syn SAA status, including accuracy, positive predictive value (PPV), and negative predictive value (NPV). c Reduction in the number of CSF tests required using the two-step workflow. Error bars correspond to 95% CIs based on 1000 iterations of model development and classification. 1209 PPMI participants were included in the analyses. αsyn alpha-synuclein, CSF cerebrospinal fluid, NPV negative predictive value, PPV positive predictive value, SAA seed amplification assay.