Proteomics reveals associations between inflammation and chronic depression in a prospective study of post‐stroke cognition

Background Depression affects approximately 40% of people with dementia, and is critical to quality of life. Here we examined the relationship between chronic depression and cognition in people with ischemic stroke and used plasma proteomics to get at potential mechanisms and identify novel therapeutic targets. Method We recruited subjects 5 months to 9 years after ischemia, age >40, who completed a 60 minute battery of cognitive tests. Composite scores of memory, processing speed, working memory, spatial functioning, and a global cognition score (the average of all cognitive tests) were created based on age‐normed performance. Mood was assessed with the Stroke Impact Scale (SIS3), transformed to a 100‐point scale. Linear regression models analyzed the relationship between depression and cognitive scores in 99 subjects. Plasma was analyzed by O‐link proteomics for 1011 proteins in 85 subjects. Additional regression models were constructed to estimate SIS3 using proteomics. Models were subject to bootstrapping for robustness, and cross‐validation to ensure results were reported on subjects blinded during model training. Pearson correlation identified linear associations between individual proteins and mood. We also investigated differences in proteins with subjects dichotomized into non‐depressed (SIS3>63) or depressed (SIS3≤63) groups. Result Mood was significantly associated with global cognitive score (R=0.368; p<0.001), visuospatial score (R=0.255; p=0.014), and processing speed (R=0.430; p<0.001), but not memory (p=0.441) or working memory (p=0.166). Proteomics alone predicted SIS3 in multivariable models. A total of 180 proteins correlated significantly with SIS3. Plasma levels of IL‐6 ( p =0.0325), and TRIM5 ( p =0.0011) were significantly elevated in subjects with depression, while HPGDS was significantly reduced ( p <0.001). There was no difference in plasma levels of candidate proteins IL‐1ß ( p = 0.0830) or TNF ( p =0.5287) between depressed and non‐depressed subjects. Conclusion We report that depression is associated with cognitive outcomes in a population with ischemia, and that machine learning models can predict mood from comprehensive plasma proteomics. We also identified proteins of interest including HPGDS (produces Prostaglandin D) and TRIM5 (upstream of NFkB) that point to an immune association with ischemia‐triggered depression. Future studies are needed to replicate these findings, and preclinical models may help uncover mechanistic relationships that could lead to new therapies.