Cognitive strategy accounts for failure on a hippocampal relational memory task in non-human primates

Relational memory, the ability to flexibly encode and retrieve associations among distinct elements, is critically dependent on the hippocampus and declines with age in humans. The Transverse Patterning (TP) task is designed to probe relational memory by requiring learning of hierarchical, circular stimulus relationships (e.g., A+ B-, B+ C-, C+ A-), a structure akin to rock-paper-scissors. In humans, TP performance is reliably impaired by hippocampal damage and aging. In non-human primates, however, findings have been inconsistent with some studies demonstrating clear hippocampal dependence, while others report no impairment, or even improvements, following hippocampal lesions. This raises the possibility that species differences in cognitive strategy use may underlie these divergent outcomes. We hypothesized that non-human primates rely on an elemental learning strategy, supported by corticostriatal systems, even when relational memory is required. To test this, we trained young and aged common marmosets (Callithrix jacchus) on the TP task and several control tasks designed to isolate elemental versus configural learning. Marmosets successfully acquired reward contingencies for individual stimulus pairs but failed when success required integrating all three stimulus relationships. In contrast, all animals readily acquired control tasks solvable via simple stimulus-response associations. Notably, there was no evidence of age-related impairment on TP or control task performance. Given the early vulnerability of the hippocampus to aging and the relative preservation of striatal systems, this pattern further supports the conclusion that marmosets rely on a habit-based learning strategy that is poorly suited to relational demands. These findings suggest that humans and non-human primates may approach the same tasks using different cognitive strategies. This has critical implications for interpreting cross-species differences in memory performance and highlights the need to validate which neural systems a task engages in each species before using it as a translational model of hippocampal function or cognitive aging.