Frequency-specific network connectivity increases underlie accurate spatiotemporal memory retrieval Andrew J Watrous, Nitin Tandon, Chris R Conner, Thomas Pieters & Arne D Ekstrom

Neuroimaging and patient lesion studies implicate the MTL, FL, and PL in episodic memory. Low frequency (3-12 Hz) activity coordinated between the hippocampus and the PFC is associated with learning a spatial navigation task. Recall of an episodic memory requires retrieving multiple types of different context. Where something occurred (spatial context) and when something occurred (temporal context).

Spectral fingerprint hypothesis argues that different cognitive operations display frequency-specific patterns of phase synchronization between regions in large-scale neural networks. Frequency-specific phase interactions are a candidate for coordinating distributed cell assemblies in parallel. This may underlie the rapid retrieval of contextual information from a specific experience.

Retrieval Theories Top: # and strength of connections are the same. Networks differ on interconnections. Middle: same regions are interconnected but the strength varies by performance. Bottom: connectivity pattern and strength are the same. Distinct phase-coupled frequencies to accomplish different processes.

Testing theoretical perspectives on the neural basis of episodic memory retrieval. 1) Does accurate episodic memory retrieval depend on a disproportionate contribution of a specific brain region (top) and/or is it accomplished by changes in global connectivity (middle)? 2) How can different contexts be retrieved through interactions in the same set of brain regions (bottom)?

Electrocorticography (ECoG) Electrodes are placed on exposed brain to record electrical activity. 6 epileptic patients participated and underwent ECoG to localize seizures.

MTL: Parahippocampal, perirhinal and entorhinal cortices

PFC: SFG, MFG, pars triangularis (IFG)

PL: SPL, IPL, precuneus (medial SPL)

Behavioral Paradigm Picked up passengers from the center of the virtual environment and delivered them to specific locations. Presented with a store cue (cookie shop) Temporal: which store was closer in time to the delivery to the cue-store Spatial: which store was closer in space to the cue-store C: free navigation with a fixed number of stops D: interim testing involved localizing stores to a grid (till %100 in 2 consecutive rounds or completing 8 rounds) The 2 s cue interval was analyzed because it was hypothesized that this period would reliably contain consistent changes in activity.

Measuring Phase synchronization with PPC Computed relative phase angle difference between electrodes on each trial. The cosine was computed between all pairwise combinations. Pairwise phase consistency (PPC) was the mean of these cosine values. PPC: Values -1 to 1 with positive values indicating phase synchronization.

Neural networks Functional connection (edge) exists between 2 subregions (nodes) if there was a significant difference between conditions across all electrode pairs. 1) Connectivity of a single node was determined by the total number of connections between a given node and all other connected nodes over time. 2) The relative strength between 2 nodes was determined by the total number of connections between those 2 nodes over time. 3) All connections in the network at a specific frequency were summed over time to provide a frequency-specific measure of global functional connectivity. Percent connectivity: # of observed edges out of the total # of edges.

Phase synchronization for correct trials between the PHG and the IPL and for the PHG and the MFG A & B: low-frequency phase consistency C & D: individual electrode pairs (A & B) E & F: population results

A: because greatest PHG connectivity was in the low- frequency bin, subsequent analysis was restricted to low- frequency (1-10 Hz). B & C: more global connectivity during correct retrieval for every frequency up to 9 HZ. PHG: greatest # of connections with nodes at all low frequencies, thus serves as a hub. radius of each node shows the relative number of connections for that frequency and condition in a specific network

Phase coherence for correct spatial retrieval occurs at a lower frequency than temporal retrieval

B: patient C: population

Spatial retrieval occurs between 1 – 4 Hz Temporal retrieval occurs between 7-10 Hz SFG, MFG, IFG, PHG, PCN, IPL Temporal

Significant condition X frequency interaction for PHG connectivity PHG with MFG, SFG, and PCN PHG with MFG, SFG, and IPL

Black box = edge (functional connection) Spatial (2 Hz) and temporal (8 Hz) retrieval display different clustering of connectivity.

More connectivity for spatial retrieval (2 Hz) Early (0-1 s): More connectivity for spatial retrieval (2 Hz) Late (1-2 s): More connectivity for temporal retrieval (8 Hz)

F: Connectivity in the spatial (2 Hz) network was significantly more clustered in time than the temporal (8 Hz) retrieval. D: spatial and temporal networks did not differ in their total # of connections. Black dashed line indicates 95 percentile using monte carlo

Summary Correct retrieval is accompanied by increases in network functional connectivity via low-frequency (1 -10 Hz) phase synchronization . PHG serves as a “hub” by having the most interconnectivity with other regions. Regional and global hybrid: correct retrieval is characterized by an increase in functional connectivity across different regions but this is mediated by the MTL. Frequency-specific connectivity for spatial (1-4 Hz; 0-1 s) and temporal (7-10 Hz; 1-2 s) supports the spectral fingerprint hypothesis.

Discussion MTT: cortical areas represent features that are bound to experience-specific contextual representations in the hippocampus. Low-frequency coordinated oscillations between the MTL and FL and PL may be the neural mechanisms suggested by MTT. It has been suggested that synchronized oscillations may facilitate precise depolarization's between cells in communicating regions. LTP occurs when the post-synaptic cell fires less than 10-20 ms after the presynaptic cell. MTL acting as a hub communicating with the FL and PL during episodic retrieval may explain why FL and PL lesions do not lead to major memory deficits whereas MTL lesions do. MTL may index memory traces while PL and FL interacting with the MTL may facilitate and augment retrieval of memory traces.