Frequency matters: how changes in hippocampal theta frequency can influence temporal coding, anxiety-reduction, and memory

Abstract

Hippocampal theta frequency is a somewhat neglected topic relative to theta power, phase, coherence, and cross-frequency coupling. Accordingly, here we review and present new data on variation in hippocampal theta frequency, focusing on functional associations (temporal coding, anxiety reduction, learning, and memory). Taking the rodent hippocampal theta frequency to running-speed relationship as a model, we identify two doubly-dissociable frequency components: (a) the slope component of the theta frequency-to-stimulus-rate relationship (“theta slope”); and (b) its y-intercept frequency (“theta intercept”). We identify three tonic determinants of hippocampal theta frequency. (1) Hotter temperatures increase theta frequency, potentially consistent with time intervals being judged as shorter when hot. Initial evidence suggests this occurs via the “theta slope” component. (2) Anxiolytic drugs with widely-different post-synaptic and pre-synaptic primary targets share the effect of reducing the “theta intercept” component, supporting notions of a final common pathway in anxiety reduction involving the hippocampus. (3) Novelty reliably decreases, and familiarity increases, theta frequency, acting upon the “theta slope” component. The reliability of this latter finding, and the special status of novelty for learning, prompts us to propose a Novelty Elicits Slowing of Theta frequency (NEST) hypothesis, involving the following elements: (1) Theta frequency slowing in the hippocampal formation is a generalised response to novelty of different types and modalities; (2) Novelty-elicited theta slowing is a hippocampal-formation-wide adaptive response functioning to accommodate the additional need for learning entailed by novelty; (3) Lengthening the theta cycle enhances associativity; (4) Even part-cycle lengthening may boost associativity; and (5) Artificial theta stimulation aimed at enhancing learning should employ low-end theta frequencies.