An OMICS‐based analytical framework to identify metabolic predictors of the cognitive response to a multimodal intervention: preliminary findings from the PENSA Study

Abstract

AbstractBackgroundNot all individuals may respond to a cognitive decline preventive intervention. However, the low frequency of cognitive assessments during short‐lasting interventions hampers the study of the heterogeneity of treatment effects (HTE). We have recently developed and validated an innovative method for remote monthly monitoring of executive functions, which leverages the performance indicators from cognitive training games (CTG) (Figure 1). Here, we aimed to examine the HTE to a multimodal intervention using this CTG measure and to identify which baseline glycaemic, lipidic, and inflammatory factors are associated with the cognitive response profiles.MethodsWe included N = 98 individuals (65.3% females; 67.2 ±4.6 years) APOE‐ɛ4 carriers with Subjective Cognitive Decline who had completed eight months of the PENSA Study multimodal intervention on January 2023. We analysed P = 62 baseline exposures that included: anthropometric factors, biochemical parameters, endocannabinoids, and plasma metabolites of the tryptophan and glutamine metabolism. First, we examined the HTE with the CTG measure as the outcome (6 repeated measures/subject) using growth mixture models. Second, we assessed the partial correlations between exposures using graphical LASSO with stability selection. Finally, we identified the most contributing predictors to the cognitive response profiles using stability selection‐based LASSO multinomial regression, fit on 500 random 80% subsamples of the data using the R package ‘sharp’.ResultsWe identified and labelled three different cognitive response trajectories: ‘average respondents’ (75.6%), ‘super respondents’ (16.3%), and ‘non‐respondents’ (8.2%) (Figure 2A). The network of exposures showed higher intra‐family than inter‐family correlations (Figure 2B). Finally, four out of 62 exposures were identified as stable predictors of the cognitive response profiles (Figure 2C). Compared to ‘average respondents’ the ‘super‐respondents’ and ‘non‐respondents’ presented higher carnitine and serotonin concentrations. However, fasting plasma glucose and total cholesterol were lower in ‘super‐respondents’ (OR = 0.63 and OR = 0.69) and higher in ‘non‐respondents’ (OR = 6.07 and OR = 2.40).ConclusionsThis study shows that not all individuals at high risk of Alzheimer’s disease respond to an intensive lifestyle‐based multimodal intervention, and that baseline cholesterol and glycaemic homeostasis could determine the success of preventive interventions for cognitive decline. Further cholesterol‐ and glycaemic‐related metabolites should be analysed to better understand the underlying mechanisms of these findings.