AGE-DEPENDENT EFFECTS OF SHORT-CHAIN POLYSIALIC ACID ON LEARNING, REVERSAL, AND RULE INTERFERENCE IN THE ATTENTIONAL SET SHIFTING PARADIGM REVEALED BY A TRIAL-WISE MIXED-EFFECTS MODEL

Attentional set shifting (ASST) is a core assay of cognitive flexibility, yet classical trials-to-criterion readouts and commonly used low-stringency stopping criteria can obscure stage-specific learning and may terminate stages prematurely by chance. Although linear in the predictors, trial-wise generalized linear mixed-effects models (GLMMs) can recover substantially richer, stage-resolved dynamics than conventional summaries.
We developed a novel binomial trial-wise GLMM that decomposes ASST performance into acquisition and reversal learning and, during extradimensional shifting, transfer and extinction of the previously rewarded stimulus dimension. We fit this model to trial-wise choices using phase-specific trial counters that treat acquisition and reversal as continuous processes and explicitly separate the rule transfer from extinction, yielding interpretable age-by-treatment effects for each ASST stage.
Using this framework, we tested male C57BL/6J mice (young, old, very old) treated with intranasal short polysialic acid fragments (DP12) before each ASST stage. DP12 was reported to inhibit extrasynaptic GluN2B-mediated signaling. DP12 improved acquisition and reversal in compound discrimination in very old mice and increased reversal learning rates in old mice. During extradimensional shifting, DP12 increased the transfer of the previously learned rule in very old mice, yet accelerated its extinction in young mice. Beyond ASST, DP12 improved hippocampus-dependent spatial memory in very old mice and attenuated age-elevated microglial lysosomal load and lysosomal vGLUT1/Homer1 cargo in CA1 dendritic layers, consistent with reduced synaptic stress–driven microglial uptake of excitatory synapses.
Given limited trials per stage, this parsimonious GLMM-based approach yields robust, well-identified estimates and an accessible framework for routine analysis.