The illusion of sharp-wave sequence replay

Zachary Roth1, 2, Yingxue Wang1, Vladimir Itskov2, and Eva Pastalkova1, 3

1 Janelia Research Campus, HHMI, Ashburn, VA, USA

2 Department of Mathematics, The Pennsylvania State University, University Park, PA, USA

3 Department of Biology, Eastern Mennonite University, Harrisonburg, VA, USA


Alternate title

The anisotropic structure of hippocampal SPW sequences

One-sentence summary

Sharp wave sequences correlate only positively with each other despite replaying running sequences forward and backward.

Evidence against bidirectional sharp-wave replay.


Hippocampal sharp-wave (SPW) sequences are believed to contribute to the encoding of episodic memories by “replaying” experience on a faster timescale. This belief stems from the observation that some SPW sequences correlate significantly with templates generated from an animal’s running experience. As these correlations are both positive and negative, it is believed that SPW sequences replay experience bidirectionally. We compared SPW sequences directly without the use of running templates. Surprisingly, correlations between SPW sequences were significantly positive, with negative correlations occurring at chance level. This suggests that SPW sequences are not activated bidirectionally. Further, this observation held regardless of how SPW sequences correlated with running sequences, suggesting that SPW sequences do not replay experience as previously believed.


The hippocampus is a brain region necessary for episodic memory (Scoville 1996, Morris 1982). Sequential firing patterns of hippocampal neurons are believed to serve as the physiological substrate of episodic memory (Eichenbaum 2014, Buzsáki 2015). Specifically, when an animal moves through space, position-selective neurons called place cells (O’Keefe 1978) are activated sequentially, forming so-called place-cell firing sequences (Fig. \ref{fig:setup}A). Similarly, when an animal runs on a running wheel during a memory task, neurons called episode cells (Pastalkova 2008) are activated sequentially, forming so-called episode-cell firing sequences. Both of these types of running sequences evolve over the timescale of seconds.

Another type of sequential firing is observed during brief bursts (Fig. \ref{fig:setup}B, C) of hippocampal network activity called sharp waves (SPWs), which occur throughout stationary behaviors such as eating, grooming, and drinking (Buzsáki 1983, Buzsaki 1992). SPWs are characterized by a distinct change of the local-field potential (LFP) across the pyramidal layer of CA1 in the hippocampus induced by synchronized CA3 input (Fig. \ref{fig:setup}B, C top; (Buzsáki 1983, Buzsáki 1986, Sullivan 2011)). The accompanying SPW sequences are usually brief (50–150 ms; (Nguyen 2009)) but often consist of spikes from a large number of neurons (Buzsáki 1986). Interestingly, during some SPWs, place/episode cells fire in an order similar to that in which they fired while the animal was running in a maze or wheel (Fig. \ref{fig:setup}B, C bottom). This phenomenon is referred to as replay of running sequences during SPWs (Foster 2006, Jackson 2006, ONeill 2006). It has been suggested that the role of SPW sequences is to aid in the storage of running experiences in memory (Pavlides 1989, Eschenko 2008, Girardeau 2009, Karlsson 2009, Singer 2009, Dupret 2010, Ego-Stengel 2009, Jadhav 2012, Girardeau 2014) by reactivating the running experience of an animal on a faster time scale and, thus, inducing experience-dependent changes in synaptic plasticity within the local and downstream networks (Buzsáki 1989, ONeill 2008, O’Neill 2010, Carr 2011, Atherton 2015, Buzsáki 2015). Interestingly, replay during SPWs has been observed (Foster 2006, Diba 2007) in both the forward and backward directions relative to the original running sequence (Fig. \ref{fig:setup}D, red and blue arrow). This observation led to belief that SPW sequences can be activated bidirectionally (i.e. forward and backward). However, the bidirectionally of SPW sequence activation was never confirmed directly (Fig. 1D, black arrow).

Prior methods for the analysis of SPW sequences have relied on averaging spiking information from entire recordings to obtain a place/episode-cell sequence template/model (Wilson 1994, Nádasdy 1999, Foster 2006, Jackson 2006, Diba 2007, Davidson 2009, Pfeiffer 2013, Dragoi 2010, Grosmark 2016). We developed a novel method that enabled us to compare pairs of raw spiking sequences without the use of averaging across sets of sequences. Using this method, we tested the hypothesis that SPW sequences are bidirectional by inspecting correlation values between SPW sequences. Surprisingly, we found that negative correlations occurred only at the chance level and that positive correlations occurred well-above the chance level regardless of whether the SPW sequences were significantly correlated to the running sequences. Further inspection of the lack of negative correlations among SPW sequences suggests that SPW sequences may not be bidirectional as previously thought (Diba 2007, Buzsáki 2015). This raises a question of the validity of the inference of backward replay from the observation of negative correlations between running sequences and SPW sequences. If SPW sequences are indeed not bidirectional, then the observation and interpretation of backward replay must be reevaluated. Similarly, if this anisotropic characteristic of SPW sequences holds true, evaluation of SPW sequences relative to an external template may lead to misleading results. Moreover, the prevalence of positive correlations between pairs of SPW sequences regardless of how the SPW sequences were correlated with running sequences suggests that the characterization of SPW sequences as “forward replay” or “backward replay” may not be the most useful. This may suggest that behavioral experience does not have so direct of a causal effect on SPW sequence generation as was previously proposed.