State-dependent fornix-hippocampal communication and the impact of deep brain stimulation: Insights from supervised learning, granger causality and causal discovery analyses.

BACKGROUND: Deep brain stimulation (DBS) of the fornix (FX) is being explored as a therapy for memory impairment in Alzheimer's disease, but clinical trials have yielded mixed results. The physiological mechanisms underlying these outcomes, particularly the state-dependent dynamics of fornix-hippocampal (HC) communication, remain poorly understood. OBJECTIVE: To dissect the state-dependent dynamics of FX-HC communication and assess the impact of continuous DBS using chronic local field potential (LFP) recordings, spectral analysis, machine learning, and advanced causal inference techniques. METHODS: Three ovine subjects were implanted with DBS leads targeting the FX and HC. Chronic LFP recordings were obtained during awake and overnight periods, with and without stimulation. Spectral analysis identified theta band (4-8 Hz) as the frequency of interest. Unsupervised k-means clustering and linear discriminant analysis classified high-theta and low-theta states. This state definition was used to differentiate state-dependent dynamics in communication between the fornix and hippocampus. Granger causality and causal discovery methods quantified the directionality and strength of FX-HC interactions. RESULTS: Bidirectional, state-dependent communication was observed: during high-theta states HC→FX drive was dominant over FX- > HC, while in low-theta states FX→HC drive was enhanced over high-theta states. Continuous FX stimulation disrupted this balance, overdriving FX→HC connectivity and attenuating HC→FX feedback, effectively creating an "electrical lesion." Causal discovery analyses corroborate these findings. CONCLUSIONS: Continuous fornix DBS disrupts physiological bidirectional communication within the FX-HC network, which may underlie the limited efficacy seen in clinical trials. These results highlight the need for adaptive or cycled stimulation paradigms to preserve endogenous network function and optimize therapeutic outcomes.