NETWORK-BASED ANALYSIS OF WEIGHTED FUNCTIONAL CONNECTIVITY ACROSS MEMORY LOADS IN MARMOSETS

Working memory is the ability to maintain and manipulate information over short time scales and has a limited capacity that decreases as memory load increases. This limitation may arise from constraints imposed by interactions within neuronal networks. Here we test whether changes in memory load are reflected in network-level reconfigurations measurable with complex network analysis. We recorded population activity from lateral prefrontal cortex of common marmosets using multielectrode arrays while animals performed a delayed non-match-to-position working memory task with varying numbers of remembered locations. Behavioral performance declined with increasing memory load. From neuronal recordings we constructed weighted functional connectivity networks, generating multiple network realizations for each experimental condition with a shared set of neuronal nodes. We first assessed within-condition consistency by comparing network realizations using weighted similarity measures to determine whether stable connectivity patterns emerged for each condition. Representative networks were then derived to summarize reproducible structure and reduce variability across repetitions. At the population level, representative networks were compared across memory loads without enforcing one-to-one matching between conditions. Network descriptors and low-dimensional representations were used to identify systematic shifts in network organization and recurrent patterns. To further characterize load-dependent structure, we built second-order networks for each memory load in which nodes represent experimental conditions and edges represent similarity between representative networks. This multilayer-inspired meta-network quantifies coherence and redundancy of network patterns within each load. We find that network configurations differ across remembered locations and reorganize with increasing memory load (see Table), indicating load- and item-dependent network dynamics.

Memory load	Prototype type	Entanglement intensity	Homogeneity
memory_load_2	consensus	3.788	8.98e-06
memory_load_2	medoid	3.156	8.59e-04
memory_load_2	probabilistic	3.686	6.39e-05
memory_load_1	consensus	5.332	1.02e-04
memory_load_1	medoid	4.311	4.96e-04
memory_load_1	probabilistic	5.262	4.61e-05
memory_load_0	consensus	3.566	1.22e-04
memory_load_0	medoid	3.089	4.47e-04
memory_load_0	probabilistic	3.512	1.27e-04