Regional Patterns of Tau Pathology Predict Domain-Specific Cognitive Performance
Poster No:
213
Submission Type:
Abstract Submission
Authors:
Stephanie Doering1, Nicole McKay2, Peter Millar1, Shaney Flores3, Austin McCullough1, Diana Hobbs3, Jason Hassenstab1, John Morris4, Brian Gordon4, Tammie Benzinger4
Institutions:
1Washington University School of Medicine, Saint Louis, MO, 2Washington University, St Louis, MO, 3Washington University in St. Louis School of Medicine, St. Louis, MO, 4Washington University in St. Louis, St. Louis, MO
First Author:
Co-Author(s):
Introduction:
Cognitive decline in Alzheimer Disease (AD) is strongly coupled to the spatiotemporal pattern of tau neurofibrillary tangles. Neuroimaging summary measures of tau typically focus on tau burden within a standard set of early-impacted regions of interest (ROI); however, observed regional patterns of tau spread are heterogenous within patient cohorts and correlate with specific cognitive domain deficits. Our previous work (Doering et al., 2024) demonstrated that a measure of tau spread extent across the brain was more sensitive to preclinical attention/processing speed deficits than summary estimates of tau burden. These results suggest utility in evaluating a wider scope of ROIs based on a patient's cognitive profile. In this work, we leverage a machine learning framework to evaluate in which ROIs tau pathology best predicts impairment for different cognitive domains.
Methods:
529 older adult participants were recruited with tau positron emission tomography (PET) and neuropsychological testing from the Washington University in St. Louis Knight Alzheimer Disease Research Center (Knight ADRC). Tau PET scans were parcellated using the Schaefer atlas with 200 ROIs and 7 networks. ROI SUVRs were calculated as the voxel-averaged SUVR. Cognitive domain composites were calculated using performance on specific neuropsychological tests. We implemented a random forest regression within a repeated k-fold cross-validation with 100 iterations predicting cognitive score from the 200 ROI SUVR inputs. The model framework was implemented for 5 cognitive domain measures: Knight Preclinical Alzheimer Cognitive composite (Knight PACC), Episodic Memory composite, Semantic Memory composite, Attention/Processing Speed composite, and the Benson Figure Copy Test (visuospatial assessment). Feature importance (FI) for each ROI was extracted from the resulting model based on mean decrease in impurity. Regional patterns of FI were evaluated for each model based on functional network assignment, neuroanatomical location, and hemisphere lateralization.
Results:
Spatial distributions for FI of each model differed by cognitive measure (Figure 1). For the Knight PACC, high FI was found across several functional networks, primarily in the inferior temporal lobe, and bilateral. For episodic memory, high FI was found in the limbic network, primarily in the medial temporal lobe, and bilateral. For semantic memory, high FI was found in the limbic network, primarily in the inferior temporal lobe, and left-lateralized. For attention/processing speed, high FI was found across several functional networks, primarily in the medial parietal and lateral temporal lobes, and bilateral. For the Benson Figure Copy test, high FI was found in the dorsal attention network, primarily in the superior parietal lobe, and right-lateralized.
Conclusions:
Overall, performance across cognitive domains was best predicted by different sets of brain regions. Tau pathology in the temporal lobe is highlighted in several domains, likely attributed to early development of tau in the medial temporal lobe; however, the specific regions are distinct between cognitive measures, even within those testing for memory. The attention and visuospatial measures demonstrate large parietal lobe involvement despite lower levels of tau in this region. The left-lateralization for semantic memory may be attributed to language production in the semantic memory tests, which has been previously shown to be left-lateralized. The right-lateralization for the visuospatial test has been previously demonstrated in functional connectivity studies of visuospatial attention. The regions identified to be predictive of each cognitive measure generally align with studies of functional connectivity, but are constrained to regions that demonstrate tau pathology in AD. These results indicate a high sensitivity of our cognitive measures to regional tau pathology and the utility of evaluating distinct sets of brain regions depending on cognitive profile.
Disorders of the Nervous System:
Neurodegenerative/ Late Life (eg. Parkinson’s, Alzheimer’s) 1
Modeling and Analysis Methods:
PET Modeling and Analysis 2
Keywords:
Cognition
Degenerative Disease
Machine Learning
Positron Emission Tomography (PET)
1|2Indicates the priority used for review