Abstract Body: INTRODUCTION: Post-stroke fatigue is prevalent and significantly impacts quality of life chronically after stroke. Its underlying biological mechanisms remain largely unknown. Here we set out to understand the persistence of post-stroke fatigue and to use plasma proteomics to identify candidate mechanisms.

METHODS: We acquired neurocognitive assessments and blood draws from 250 stroke survivors from two sites at baseline (median 8 mo after stroke, range 5-120 mo), and 131 participants one year later. The Functional Assessment of Chronic Illness-Fatigue (FACIT) scale was used to assess fatigue, with no fatigue defined as FACIT>41, any fatigue as FACIT≤41, and severe fatigue as FACIT<31. We used binary logistic regression to examine associations with clinical variables. Aptamer based proteomics (Somalogic) was used to quantify 6344 proteins in 97 participants at baseline and 51 one year later. We employed machine learning with rigorous feature selection to predict FACIT from proteomics. We performed Gene Set Enrichment Analysis (GSEA) comparing no fatigue to any fatigue or to severe fatigue at both visits.

RESULTS: At the baseline visit, fatigue was present in 44% of our cohort with no difference between sites. 41% were female and the median age was 67 [IQR 58,74]. When adjusted for age, sex and NIHSS, diabetes (OR=8.4, 95% CI=3.1-22) and a prior diagnosis of cancer (OR=4, 95% CI=1.2-12) were associated with fatigue. If a patient had fatigue at baseline, they were more likely to have fatigue at year 1 (OR=32, 95% CI=10-97). Our machine learning model achieved high predictive performance of FACIT at baseline, with an AUROC of 0.81 for distinguishing fatigue and a Spearman correlation of 0.62 for direct FACIT score prediction. GSEA comparing no fatigue with either any fatigue or severe fatigue revealed fatigue-associated increases in pathways involved in stress granule formation and clearance (9 and 8 of the top 20 pathways, respectively). Excitingly, one year later 7 and 11 of the top 20 fatigue-associated pathways were also directly related to stress granules.

CONCLUSION: We observed high prevalence and persistence of post-stroke fatigue and proteomics demonstrated associations with stress granules at two consecutive visits. Stress granules broadly dysregulate protein translation and are associated with both neurodegeneration and myopathy. Future studies will seek to replicate these findings, identify affected cell types, and define therapeutic targets.