RETROSPLENIAL–TEMPORAL NETWORK SPECIALISATION GATES LANDMARK-BASED RESCUE OF A VULNERABLE GRID CODE IN APOE4 CARRIERS

Spatial navigation relies on flexibly weighting internal path integration (PI) signals and external landmark cues, yet how this balance is shaped by genetic risk for Alzheimer’s disease and large-scale network organisation remains unclear. We tested whether a landmark-based strategy emerges from intrinsic specialisation and whether it compensates for grid-code vulnerability in APOE4 carriers. Seventy-seven older adults performed an fMRI PI task either relying purely on visual flow (pure PI, PPI) or with an additional local landmark (landmark-supported PI, LPI), after a resting-state fMRI. Behaviourally, only landmark users (n = 44) improved performance in LPI, with APOE4 carriers (n = 19) showing the largest reduction in rotation errors. Segregation of the resting-state retrosplenial–temporal network (RSTN) predicted landmark use in APOE4 carriers, indicating that strategy choice reflects stable network architecture rather than preference. During navigation, APOE4 carriers who used the landmark exhibited efficient coupling within entorhinal, retrosplenial, parietal, and hippocampal circuits, whereas non-users showed maladaptive hyperconnectivity linked to poorer performance. Grid-like representations in posterior-medial entorhinal cortex were robust, yet APOE4 carriers showed reduced magnitude in PPI, indicating vulnerability of internally driven grid coding. In LPI, this deficit was attenuated in APOE4 carriers who used the landmark, suggesting that external cues modulate the grid code in a compensatory direction. These findings identify a pathway whereby RSTN segregation enables landmark-based compensation of vulnerable grid signals, determining whether individuals at genetic risk can stabilise navigation performance. This work reveals strategy as a network-constrained phenotype with relevance for early Alzheimer-related changes in navigation.