Hippocampal Sequences During Exploration: Mechanisms and Functions — TEI-tagged view

DrieupersName, ZugaropersName (2019date). “Hippocampal Sequences During Exploration: Mechanisms and Functions.” Frontiers in Cellular Neurosciencetitle·j.

Object of study — possible mechanisms and functions of hippocampal theta sequencesterm during exploration

Research question — whether phase precessionterm is a cause or a consequence of the formation of theta sequencesterm

Contribution — a closer examination of the available data suggests a more complex picture, and the reviewterm·mth points out the current inconsistencies and overlooked findings

Full text — tagged

The reference list is not part of the text we hold for this paper. Our copy was extracted from the publisher's page and the bibliography did not survive that extraction, so the article below ends before its references. Nothing has been removed by us.

Hippocampal Sequences During Exploration: Mechanisms and Functions Céline DrieupersName Céline DrieupersName 1 Center for Interdisciplinary Research in Biology, Collège de France, CNRS UMR 7241, INSERM U 1050, PSL Research University, Paris, France 1, *, † , Michaël ZugaropersName Michaël ZugaropersName 1 Center for Interdisciplinary Research in Biology, Collège de France, CNRS UMR 7241, INSERM U 1050, PSL Research University, Paris, France 1, * 1 Center for Interdisciplinary Research in Biology, Collège de France, CNRS UMR 7241, INSERM U 1050, PSL Research University, Paris, France Edited by: Francesca Cacucci, University College London, United Kingdom Reviewed by: Romain Goutagny, UMR7364 Laboratoire de Neurosciences Cognitives et Adaptatives (LNCA), France; Caswell Barry, University College London, United Kingdom ✉ *Correspondence: Céline DrieupersName cdrieu1@jhu.edu * Michaël ZugaropersName michael.zugaropersName@college-de-france.fr †Present Address: Céline DrieupersName, Department of Psychological and Brain Sciences, Krieger School of Arts and Sciences, Johns Hopkins University, Baltimore, MD, United States This article was submitted to Cellular Neurophysiology, a section of the journal Frontiers in Cellular Neuroscience Received 2018 Nov 9; Accepted 2019 May 8; Collection date 2019.

Copyright © 2019 DrieupersName and ZugaropersName.

This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

PMCID: PMC6584963 PMID: 31263399 Abstract Although the hippocampusterm plays a critical role in spatial and episodic memories, the mechanisms underlying memoryterm formation, stabilization, and recall for adaptive behavior remain relatively unknown. During exploration, within single cycles of the ongoing thetaterm rhythm that dominates hippocampal local field potentials, place cellsterm form precisely ordered sequences of activity. These neural sequences result from the integration of both external inputs conveying sensory-motor information, and intrinsic network dynamics possibly related to memoryterm processes. Their endogenous replayterm during subsequent sleep is critical for memoryterm consolidationterm. The present reviewterm·mth discusses possible mechanisms and functions of hippocampal theta sequencesterm during exploration. We present several lines of evidence suggesting that these neural sequences play a key role in information processing and support the formation of initial memoryterm traces, and discuss potential functional distinctions between neural sequences emerging during thetaterm vs. awake sharp-wave ripples.

Keywords: place cellterm, network dynamics, phase precessionterm, theta sequencesterm, awake replayterm, memoryterm 1. Introduction The ability of brain circuits to generate precise sequences of neuronal activity may underlie certain forms of sensory-motor processing and learning (Skaggs et al., 1996 ; Riehle et al., 1997 ; Stopfer et al., 1997 ; Petersen et al., 2001 ). While in some systems these sequences may be driven by external stimuli or motor signals, reports of internally generated neuronal sequences in the hippocampusterm, under either minimal (e.g., sleep; Lee and Wilson, 2002 ; Ji and Wilson, 2007 ) or invariant (Pastalkova et al., 2008 ; Wang et al., 2015 ) sensory-motor inputs, have identified a potential evolutionary mechanism for planningterm and memoryterm consolidationterm (Lisman and Redish, 2009 ; Foster and Knierim, 2012 ; Buzsáki and Moser, 2013 ; Wikenheiser and Redish, 2015a ; Zielinski et al., 2017 ).

The hippocampal network receives highly integrated inputs from distributed areas (Witter et al., 1989 ; Andersen et al., 2007 ; van Strien et al., 2009 ) and is endowed with auto-associative properties (Gilbert and Brushfield, 2009 ), which allows it to integrate and retrieve spatial as well as non-spatial information (Buzsáki and Moser, 2013 ; Eichenbaum, 2017 ). Bilateral lesions of the hippocampusterm result in long term deficits in episodic memoryterm (Scoville and Milner, 1957 ) as well as in strong impairments in spatial memoryterm (Morris et al., 1982 ). Indeed, the hippocampal formation has been proposed to support a cognitive map of the environment (Tolman, 1948 ; O'Keefe and Nadel, 1978 ), based on the remarkable property of its “place cellsterm” that discharge when the animal occupies restricted locations in the environment (“place fields,” O'Keefe and Dostrovsky, 1971 ). During the exploration, the animal sequentially crosses successive place fields, and the corresponding place cellsterm fire in sequence, reflecting the ongoing trajectoryterm at the behavioral timescaleterm. Strikingly, embedded in these slow sequences, time-compressed sequences emerge in each cycle of the ongoing thetaterm rhythm (O'Keefe and Recce, 1993 ; Skaggs et al., 1996 ), a strong and regular ~8 Hz oscillationterm of the hippocampal local field potential (LFP; Jouvet, 1969 ; Vanderwolf, 1969 ). These sequences are known as “theta sequencesterm.” As a result of this fast timescaleterm organization, the spikesterm emitted by successive cell assembliesterm occur within dozens of milliseconds, and this pattern is continually repeated in successive thetaterm cycles, which is conducive to Hebbian synaptic plasticityterm (Skaggs et al., 1996 ; Magee and Johnston, 1997 ; Stuart, 2001 ; Buzsáki and Draguhn, 2004 ; Wójtowicz and Mozrzymas, 2015 ). Therefore, theta sequencesterm have been proposed to allow the formation of spatial and episodic memoryterm traces, by anchoring spatial events in their temporal context (Skaggs et al., 1996 ; DrieupersName et al., 2018 ).

Interestingly, fast time-compressed trajectories are also represented by neuronal sequences that occur during brief pauses and prolonged immobility (Foster and Wilson, 2006 ; Diba and Buzsáki, 2007 ), when thetaterm oscillationsterm are absent. Instead, these replayterm events occur during sharp-wave ripple complexes (O'Keefe, 1976 ; Buzsáki et al., 1983 , 1992 ), a transient LFP pattern associating fast (~200 Hz) oscillationsterm in stratum pyramidale and a large deflection in stratum radiatum . These awake replayterm events have been involved in spatial working memoryterm (Jadhav et al., 2012 ). Further replayterm of behavior-related neural sequences endogenously occurs during sleep (Wilson and McNaughton, 1994 ; Lee and Wilson, 2002 ), and has been shown to play a critical role in memoryterm consolidationterm (Girardeau et al., 2009 ; Ego-Stengel and Wilson, 2010 ; Maingret et al., 2016 ; van de Ven et al., 2016 ).

Hippocampal sequences may thus represent a general neurophysiological mechanism supporting memoryterm formation, consolidationterm and retrieval. The present reviewterm·mth focuses on hippocampal theta sequencesterm. We first describe the cellular and network mechanisms proposed to give rise to the theta-related dynamics of hippocampal place cellsterm. We then report evidence for a role in information processing and memoryterm trace formation. Finally, we discuss possible functional distinctions between neural sequences emerging during thetaterm and awake sharp-wave ripples.

2. Emergence of Theta Sequencesterm The sequential activation of successive place cellsterm along a given trajectoryterm, at the behavioral timescaleterm, does not trivially imply that in overlapping regions of space these same cells should fire in the same order at the much faster, “thetaterm” timescaleterm. How do theta sequencesterm emerge in the hippocampal network? During locomotion, the hippocampal dynamics are under the combined influences of extrinsic inputs from multiple sensory and locomotion-related structures, as well as the intrinsic properties of the hippocampal network, modulated by elevated acetylcholine levels. Thetaterm oscillationsterm dominate the hippocampal LFP and powerfully shape both pyramidalterm cell and interneuronterm firing.

2.1. Thetaterm During Locomotion The thetaterm rhythm includes a range of low and regular frequency oscillationsterm from 4 to 12 Hz ( Figure 1A ). During wakefulness, thetaterm oscillationsterm arise during locomotion and epochs of active engagement in the environment such as navigating, exploring objects, rearing and preparing for movement (Jouvet, 1969 ; Vanderwolf, 1969 ; Foster et al., 1989 ). While the precise mechanisms responsible for the generation of thetaterm remain to be elucidated (Buzsáki et al., 2002 ), thetaterm oscillationsterm are thought to result from an interplay between numerous factors ( Figure 1B ), including rhythmic cholinergic modulation and feed-forward inhibition from the medial septumterm and diagonal band of Broca (MSDB; Petsche et al., 1962 ; Kramis et al., 1975 ; Buzsáki et al., 1986 ; Vanderwolf et al., 1988 ; Kocsis et al., 1999 ; Müller and Remy, 2014 ; Zutshi et al., 2018 ), rhythmic inputs from layers II and III of the entorhinalterm cortex targeting DG/CA3term and CA1term, respectively (Buzsáki et al., 1983 ; Kamondi et al., 1998 ), activation of CA1term pyramidalterm cell dendrites by CA3term collaterals (Konopacki et al., 1987 ; Kamondi et al., 1998 ; Fisher and Blum, 1999 ), interneuronal network dynamics within (Goutagny et al., 2009 ) and between (Jackson et al., 2014 ) hippocampal fields, rhythmic feedback inhibition from hippocampusterm long-range interneuronsterm to MSDB (Alonso and Köhler, 1982 ; Tóth et al., 1993 ), as well as intrinsic resonant properties of hippocampal, entorhinalterm and medial septal neurons mediated by various voltage-dependent channels (Ca 2+ , K + , Na + ) which provide membrane resonance and subthreshold oscillationsterm at thetaterm frequency (Kamondi et al., 1998 ; Evstratova et al., 2011 ; Hansen et al., 2014 ).

Figure 1.

Open in a new tab Thetaterm oscillationsterm. (A) Example local field potential (LFP) trace during track running. (B) Simplified schematic of thetaterm generation. The medial septum-diagonal band of Broca (MSDB) sends theta-rhythmic GABAergic (g), cholinergic (c) and glutamatergic projections to the hippocampusterm (glutamatergic neurons appear to entrain other septal neurons, but not hippocampal thetaterm, and are not represented here). Rhythmic inhibition (red, GABA) of highly interconnected perisomatic interneuronsterm (g) constrains the timing of pyramidalterm cell spikesterm. The entorhinalterm cortex (EC) also sends theta-paced outputs to the hippocampusterm through the perforant path. CA3term outputs targeting stratum radiatum constitute another likely intrahippocampal thetaterm generator. In addition, the dendritic membrane of CA1term pyramidalterm cells is endowed with resonant properties at thetaterm frequency. Finally, hippocampal long-range backpropagating interneuronsterm target the MSDB, arguing for cyclical (non-unidirectional) mechanisms in the generation of thetaterm. (C) Power spectral density of dorsal CA1term LFP as a function of running velocity (logarithmic scale; average over four rats). At low velocities, power is dominated by a single peak located between 6 and 7 Hz. As velocity increases, the peak quickly narrows, shifts toward 8 Hz, while additional peaks (harmonics) develop at frequencies n × 8 Hz (up to n = 6 for dorsal CA1term region, n = 2 for intermediate CA1term). Reproduced from Sheremet et al. ( 2016 ).

The thetaterm rhythm appears to act as an internal clock: hippocampal pyramidalterm cells fire maximally at the trough of thetaterm cycles, when perisomatic inhibition by parvalbumin (PV) and cholecystokinin (CCK) basket cells, as well as axo-axonic cells, is minimal; and different classes of GABAergic neurons have different preferred firing phases (Freund and Buzsáki, 1996 ; Csicsvari et al., 1999 ; Klausberger and Somogyi, 2008 ; Müller and Remy, 2014 ). Local GABAergic inhibition may fine tune the spiketerm timing of pyramidalterm cells relative to thetaterm [e.g., Ego-Stengel and Wilson, 2007 ; Geisler et al., 2007 ; Maurer and McNaughton, 2007 ; Royer et al., 2012 ; Grienberger et al., 2017 ; see the network modelterm·mth by Ferguson et al. ( 2015 )]. Given the strong and reliable modulation of hippocampal activity by thetaterm oscillationsterm, behavior-related changes in the thetaterm profile (e.g., spectral components) may play a crucial role in hippocampal dynamics.

Many structures activated during locomotion modulate thetaterm (Vertes et al., 2004 ; Kiehn, 2016 ). Consistently, thetaterm is primarily influenced by ambulatory signals, compared to e.g., optic flow or vestibular self-motion cues (Czurkó et al., 1999 ; Terrazas et al., 2005 ). Hence, numerous subcortical structures, which may not be directly involved in the actual generation of thetaterm, have nonetheless been reported to modulate thetaterm. These are part of the “synchronizing ascending system” (Vertes and Kocsis, 1997 ; Vertes et al., 2004 ; Orzeł-Gryglewska et al., 2015 ; Kiehn, 2016 ). Efferences of several subnuclei of the brainstem reticular formation and the mesencephalic locomotor region can trigger the emergence (onset, offset) and the regulation (power, frequency) of thetaterm (Grastyán et al., 1965 ; Peck and Vanderwolf, 1991 ; Sinnamon, 1993 ; Vertes and Kocsis, 1997 ; Sńska and Kasicki, 1998 ; Jackson et al., 2008 ; Bland et al., 2016 ; Ma et al., 2017 ).

How is thetaterm altered by locomotion? In the spectral domain, a 16 Hz oscillationterm, described as the second harmonic of thetaterm (Harper, 1971 ), has been reported to emerge with increasing locomotion velocity ( Figure 1C ; Czurkó et al., 1999 ; Terrazas et al., 2005 ; Sheremet et al., 2016 ). More generally, a chain of harmonics significantly phase coupled to the thetaterm oscillationterm appear to be related to the velocity of the animal (Sheremet et al., 2016 ). In the time domain, these speed-related nonlinearities are manifested as an increase in skewness and asymmetry in the shape of thetaterm cycles (Buzsáki et al., 1983 ; Terrazas et al., 2005 ; Sheremet et al., 2016 ), which decrease along the septo-temporal axis of the hippocampusterm (Maurer et al., 2005 ; Hinman et al., 2011 , 2013 ; Sheremet et al., 2016 ). In addition, the frequency of thetaterm and its harmonics increase with running velocity (McFarland et al., 1975 ; Sńska and Kasicki, 1998 ; Maurer et al., 2005 ; Bender et al., 2015 ; Sheremet et al., 2016 ), in a consistent manner across the septo-temporal axis (Maurer et al., 2005 ; Sheremet et al., 2016 ). Intriguingly, thetaterm phase shifts progressively from the septal to the temporal pole of the hippocampusterm, and thus thetaterm can be seen as a “traveling wave” (Lubenov and Siapas, 2009 ; Patel et al., 2012 ).

Changes in thetaterm power and frequency coincide with systematic changes in the spiking dynamics of pyramidalterm cells (Song et al., 2005b ; Terrazas et al., 2005 ; Lu and Bilkey, 2010 ), and a strong correlation has been reported between thetaterm nonlinearity and the intensity and timing of the spiking activity of both pyramidalterm cells and interneuronsterm (Maurer et al., 2005 ; Sheremet et al., 2016 ). Thus, speed-related changes in thetaterm oscillationsterm have a strong impact on the precise spiketerm timing of hippocampal cells. As described below, the precise phase at which spikesterm are emitted is directly related to the position of the animal in the environment (O'Keefe and Recce, 1993 ), which may constitute one of the clearest examples of temporal coding in the brain.

2.2. Theta Phase Precessionterm·def As a rat walks through the firing field of a given place cellterm, the firing rate of this cell progressively increases until the animal reaches the center of the place field, then decreases until the animal leaves the field. Thus, the firing rate of the place cellterm appears to code for the location of the animal in the environment. Strikingly, O'Keefe and Recce ( 1993 ) discovered that spatial information can also be provided by the precise time at which the place cellterm fires relative to the ongoing thetaterm rhythm. Indeed, the place cellterm fires at progressively earlier phases across successive thetaterm cycles: firing occurs at late phases when the animal enters the place field, then progressively shifts back in phase as the animal progresses through the field, and finally occurs at the beginning of the thetaterm cycle when the animal leaves the field. This phenomenon is referred to as “phase precessionterm” ( Figure 2A ; O'Keefe and Recce, 1993 ). While the range over which spiketerm thetaterm phase precesses may appear to span as much as 360° when pooling data over multiple field crossings, single-trial phase range is typically ~180°, and never exceeds a full cycle (O'Keefe and Recce, 1993 ; Maurer et al., 2006 ; Schmidt et al., 2009 ). At a given (fixed) running speed, the rate of phase precessionterm is inversely related to the size of the place field: in particular, place cellsterm with small fields in the dorsal hippocampusterm phase precess faster than place cellsterm with large fields in the ventral hippocampusterm (Maurer et al., 2005 ). On the other hand, phase precessionterm slope increases with running speed, ensuring that the phase-position relationship is maintained regardless of running speed (Bose and Recce, 2001 ; Geisler et al., 2007 ).

Figure 2.

Open in a new tab Temporal coding in the hippocampusterm and emergence of theta sequencesterm. (A) Thetaterm spiketerm phase precessionterm. Top, schematic representation of the temporal organization of successive place cellterm bursts relative to thetaterm as a rat runs along a linear track. When the rat enters the firing field (pink ellipse), the cell fires near the end of the thetaterm cycle, i.e., with a phase of ~360°. As the rat progresses through the field, bursts occur on earlier and earlier phases (red vertical ticks, action potentials; black numbered dots, mean burst times within individual thetaterm cycles). Bottom, phase precessionterm plot. For each spiketerm, phase is represented as a function of position (tilted black line, best linear-circular regression line). (B) Top, four schematic overlapping firing fields. Middle, in their overlap region the cells discharge in sequence (A, AB, ABC, etc.) within each thetaterm cycle, reflecting the order of field traversal at a compressed time-scale. Bottom, these past-present-future “sweeps” may require coordinationterm between individual phase precessing cells.

Phase precessionterm has been reported not only in all subfields of the hippocampusterm (Skaggs et al., 1996 ; Johnson et al., 2007 ; Mizuseki et al., 2009 ), but also in neighboring parahippocampal structures projecting to the hippocampusterm, including the subiculum (Kim et al., 2012 ) and medial entorhinalterm cortex (MEC, Hafting et al., 2008 ), as well as in downstream structures, such as the medial prefrontal cortex (Jones and Wilson, 2005 ) and ventral striatum (van der Meer and Redish, 2011 ). Understanding the mechanisms of hippocampal phase precessionterm is thus expected to provide invaluable insights about a general coding scheme present in multiple brain areas (Harris, 2005 ; Malhotra et al., 2012 ). Recent evidence of phase precessionterm in the human hippocampusterm (Qasim and Jacobs, 2016 ) further confirms that theta phase precessionterm·def remains relevant beyond rodent species.

2.2.1. Mechanisms of Phase Precessionterm: Computational Modelsterm·mth and Experimental Data The neurophysiological mechanisms of phase precessionterm are difficult to explore experimentally, but have been extensively investigated through computational modeling (for reviewsterm·mth, see Maurer and McNaughton, 2007 ; Burgess and O'Keefe, 2011 ). While no modelterm·mth appears to fully account for all aspects of phase precessionterm, a continued dialogue between experimental and modeling approaches has provided constraints on potential generation mechanisms. Below, the three main classes of modelsterm·mth of phase precessionterm are briefly reviewed and discussed in light of experimental data.

The detuned oscillators modelterm·mth was originally proposed by O'Keefe and Recce ( 1993 ) in their initial report of phase precessionterm, and remains one of the most elegant hypotheses to date 1 ( Figure 3A ). This modelterm·mth considers a place cellterm that oscillates at a slightly faster frequency than thetaterm. These two oscillators, synchronized at the beginning of the place field, slowly shift due to their frequency difference. Because the place cellterm fires at the peaks of its own membrane potential oscillationterm, i.e., its more depolarized states, this occurs earlier and earlier relative to the slower thetaterm oscillationterm, which results in phase precessionterm. In addition, the frequency of the membrane potential oscillationterm is assumed to be proportional to running speed, so that the phase advance of spikesterm reflects the distance traveled through the place field, rather than the time spent in the field. However, because the field size is maintained despite changes in running speed, it has been proposed that the frequency of both oscillators increased with running speed to compensate for the decrease in numbers of thetaterm cycles at high running speed (Geisler et al., 2007 , 2010 ). The modelterm·mth requires neurons with cosine directional tuning, which have been identified in the hippocampusterm, medial septumterm and anterior thalamus (Welday et al., 2011 ). Similar cells were documented in entorhinalterm cortex (Krupic et al., 2012 ), although the finding remains controversial (Krupic et al., 2015 ; Navratilova et al., 2016 ).

Figure 3.

Open in a new tab Computational modelsterm·mth of phase precessionterm. (A) In the detuned oscillators modelsterm·mth, the membrane potential ( V m ) of the neuron (light blue) oscillates faster than the local thetaterm rhythm (dark blue) within the firing field. Spiketerm bursts (black dots numbered from 1 to 5) occur at positive peaks of V m and thus anticipate thetaterm cycles, resulting in phase precessionterm. (B) In the somato-dendritic interference modelsterm·mth with asymmetric excitation ramps, spikesterm are emitted whenever excitation (red) exceeds somatic inhibition (light blue), which is paced by thetaterm oscillationsterm (dark blue). Because excitation increases, it exceeds inhibition earlier and earlier, resulting in phase precessionterm. (C) In the asymmetric connections modelterm·mth, successive place cellsterm (labeled from A to E) with overlapping fields are connected (gray arrows) in such a way that activity propagates unidirectionally within the network. Thus, as the rat enters the field of cell E, it happens to be located at the center of the field of cell A. Cell A is therefore maximally excited, and transmits its activity via the chain of intermediate cells until it reaches cell E, which fires late in the thetaterm cycle as a result of transmission delays. But as the rat progresses through the field, there are fewer and fewer intermediate steps and the transmission delays decrease, so that cell E fires earlier and earlier in the thetaterm cycle. Note that in the present Figure, phase precessionterm is shown from 360° to 180° to simplify the schematic representation.

The generation sites of these two oscillators has remained unresolved. Intrahippocampal oscillators under the influence of medial septumterm and CA3term recurrent inputs are inconsistent with the finding that thetaterm reset by brief single-pulse electrical stimulation of the ventral hippocampal commissure does not reinitialize phase precessionterm and leaves the phase-position relationship intact (ZugaropersName et al., 2005 ). Extrahippocampal oscillators (e.g., Yamaguchi, 2003 ) would be supported by phase precessionterm in MEC grid cells (in layer II and to a lesser extent in layer III) and conjunctive cells (in layer III) (Hafting et al., 2008 ; Reifenstein et al., 2012 , 2014 , 2016 ; Climer et al., 2013 ; Ebbesen et al., 2016 ), which persists during pharmacological inactivation of the hippocampusterm (Hafting et al., 2008 ). Conversely, hippocampal phase precessionterm is disrupted after bilateral lesions of the MEC (Schlesiger et al., 2015 , but see Ormond and McNaughton, 2015 ). Further, a recent study by Robinson et al. ( 2017 ) has shown that phase precessionterm in object-responsive CA1term cells persists during transient optogenetic inactivation of the MEC, suggesting a possible complementary role for the LEC, which predominantly provides non-spatial information (Deshmukh and Knierim, 2011 ; Tsao et al., 2013 ). Several modelsterm·mth have suggested that entorhinalterm oscillators might generate phase precessionterm (Burgess et al., 2007 ; Climer et al., 2013 ; Jeewajee et al., 2014 ). On the other hand, relatively large phase shifts between EC and CA1term/CA3term spikesterm (Mizuseki et al., 2009 ) remain at odds with the notion that phase precessionterm could be directly transferred from the EC to the hippocampusterm. In addition, the mechanism proposed to account for phase precessionterm in EC requires that entorhinalterm head direction cells signal movement direction, which is inconsistent with experimental findings (Cei et al., 2014 ; Raudies et al., 2015 ).

Somato-dendritic interference modelsterm·mth ( Figure 3B ; Kamondi et al., 1998 ; Magee, 2001 ; Harris et al., 2002 ; Mehta et al., 2002 ; Losonczy et al., 2010 ) posit that the timing of action potentials is determined by the combination of oscillatory somatic inhibition and transient ramp-like dendritic excitation. As the animal enters the place field, the cell fires when excitation overcomes inhibition, which occurs earlier and earlier because of the ramp-like shape of the dendritic depolarization. The excitation ramp has been proposed to be either symmetrical (Harris et al., 2002 ) or asymmetrical (Mehta et al., 2002 ). The asymmetrical depolarization shape mirrors the asymmetrical development of place field with experience (Mehta et al., 1997 , 2000 ). To remain consistent with the observation that spiketerm phase decreases but does not subsequently increase, the symmetric ramp modelterm·mth posits that the place cellterm stops firing at the peak of the ramp. This was supported by in vitro data obtained following symmetrical-shaped depolarizing current injection through the pyramidalterm cell dendrites (Harris et al., 2002 ), which suggested that an adaptive mechanism allowed the cell to precisely stop firing at the ramp peak.

Both intracellular in vitro (Magee, 2001 ; Losonczy et al., 2010 ) and extracellular in vivo (Kamondi et al., 1998 ; Harris et al., 2002 ; Mehta et al., 2002 ) data are consistent with the somato-dentritic interference modelsterm·mth. Notably, recordings of intracellular activity in awake behaving animals confirmed that within firing fields, the membrane potential of place cellsterm had an asymmetric depolarization curve (Harvey et al., 2009 ; Epsztein et al., 2011 ; this was also observed in grid cells, Domnisoru et al., 2013 ; Schmidt-Hieber and Häusser, 2013 ). Epsztein et al. ( 2011 ) further documented an increase in the excitability of place cellsterm during field traversal, compared to silent cells. While this supports the somato-dentritic interference modelsterm·mth, the same studies also reported that spikesterm remain locked to the peaks of the membrane potential, a critical prediction of the detuned oscillators modelsterm·mth. And because recent work has shown that even network connectivity modelsterm·mth (see below) can account for these findings (Romani and Tsodyks, 2015 ), it becomes unclear whether intracellular data do discriminate between different classes of modelsterm·mth of phase precessionterm.

In the network connectivity modelsterm·mth , contrary to the above cellular modelsterm·mth in which no particular synaptic connectivity is assumed, phase precessionterm results from transmission delays between asymmetrically connected cells ( Figure 3C ; Jensen and Lisman, 1996 ; Skaggs et al., 1996 ; Tsodyks et al., 1996 ; Romani and Tsodyks, 2015 ). According to these modelsterm·mth, place cellsterm are unidirectional and asymmetrically connected to each other. Feed-forward sensory inputs arriving when the animal reaches the end of the place field drive the cells to fire at early phases of the thetaterm oscillationterm. Activity then spreads to place cellsterm with fields located further along the trajectoryterm (via asymmetric connections), which fire later in the cycle due to transmission delays. Hence, place cellsterm fire earlier and earlier as the animal progresses through their field, resulting in phase precessionterm. Asymmetric connections can be hard-wired (Dragoi and Tonegawa, 2011 , 2013 ) or result from experience (Silva et al., 2015 ), consistent with the experience-dependent asymmetric expansion of place fields (Mehta et al., 1997 , 2000 ). An alternative version of the modelterm·mth posits that phase precessionterm results from phase differences between CA3term and entorhinalterm inputs (Chance, 2012 ; Fernández-Ruiz et al., 2017 ). In both cases, the proposed mechanisms are difficult to reconcile with the experimental finding that the relation between phase and allocentric spatial position changes during backward movement (Cei et al., 2014 ; Maurer et al., 2014 ). Further, this change occurs abruptly, ruling out the possibility that asymmetry could reverse via slow plasticityterm processes (Cei et al., 2014 ). Hence, these modelsterm·mth may need refining to account for these experimental findings (but see Chadwick et al., 2015 , 2016 ).

While phase precessionterm has been mainly described in principal cells, there have been reports of phase precessionterm in hippocampal interneuronsterm, in particular those receiving putative monosynaptic connections from place cellsterm recorded on the same tetrode. Within the field of their partner place cellsterm, these interneuronsterm share the same rate of phase precessionterm (Maurer et al., 2006 ; Ego-Stengel and Wilson, 2007 ; Geisler et al., 2007 ). Because pyramidalterm cells generally fire before interneuronsterm, the spatial selectivity of interneuronsterm was suggested to be inherited from local pyramidalterm cells (but see Chadwick et al., 2016 ). Functionally, these interneuronsterm could inhibit competing cell assembliesterm and therefore participate in the segregation of cell assembliesterm in space and time (Maurer et al., 2006 ; Cutsuridis and Hasselmo, 2012 ; Royer et al., 2012 ; Dupret et al., 2013 ). This is supported by recent data by Grienberger et al. ( 2017 ), who coupled LFP and intracellular recordings in head-restrained mice running on an enriched treadmill, and showed that optogenetic hyperpolarization of CA1term interneuronsterm reduces both the rate and temporal coding of spatial location in CA1term place cellsterm by suppressing out-of-field excitatory inputs.

2.2.2. Functional Role of Phase Precessionterm While phase precessionterm may code for allocentric position on linear tracks (O'Keefe and Recce, 1993 ; Jensen and Lisman, 2000 ), it is unclear how this code would generalize to two dimensional environments. Contrary to linear tracks, in open fields animals can enter place fields from multiple directions, and place cellsterm precess in a similar manner regardless of the direction of motion. As a result, place cellsterm fire at the same thetaterm phase for multiple locations (Skaggs et al., 1996 ; Huxter et al., 2008 ). Thus, phase may instead code for the relative distance from the place field center (Jeewajee et al., 2014 ). Similarly, the reversal of the phase-position relation observed during backward travel (Cei et al., 2014 ; Maurer et al., 2014 ) indicates that phase does not code for the allocentric position of the animal, but more likely for the distance traveled through the field.

Phase precessionterm has been observed in a variety of contexts including running on a wheel (Hirase et al., 1999 ; Pastalkova et al., 2008 ) or a treadmill (Royer et al., 2012 ; Cei et al., 2014 ; Grienberger et al., 2017 ), jumping (Lenck-Santini and Holmes, 2008 ), in virtual reality (Chen et al., 2013 ; Ravassard et al., 2013 ; Aghajan et al., 2015 ), and during fixation (Takahashi et al., 2014 ). While all of these conditions correspond to thetaterm states, the animal does not necessarily change location. Consistently, several studies have reported object- and sound-modulated CA1term phase-precessing cells (Aronov et al., 2017 ; Robinson et al., 2017 ) and phase-precessing cells toward reward locations in the ventral striatum (van der Meer and Redish, 2011 ). This suggests that phase precessionterm may not be uniquely linked to space, but rather to the full episodic component of the ongoing experience (e.g., reviewed in Jaramillo and Kempter, 2017 ). An even more drastic departure from the spatial coding hypothesis is the view that perhaps the main role of theta phase precessionterm·def is the formation of neuronal activity sequences, segregated by thetaterm cycles ( Figure 2B ; e.g., Skaggs et al., 1996 ; Schmidt et al., 2009 ).

2.3. Mechanisms of Theta Sequencesterm Whether phase precessionterm is a cause or a consequence of the formation of theta sequencesterm remains unresolved. As discussed above, phase precessionterm could result from a variety of mechanisms that do not require sequential connectivity between neighboring place cellsterm ( detuned oscillators and somato-dendritic interference modelsterm·mth). Independent initiation of phase precessionterm in successive cells would automatically yield theta sequencesterm (Skaggs et al., 1996 ). Thus, phase precessionterm would be a causal mechanism of theta sequencesterm. An alternative view is that hippocampal sequences result from specific patterns of connectivity within the hippocampal network ( network connectivity modelsterm·mth; see also Dragoi and Tonegawa, 2011 ). At the single cell level, this would be manifested as phase precessing spikesterm. Thus, phase precessionterm would be a consequence of the formation of theta sequencesterm.

Several studies have challenged the idea that phase precessionterm could entirely account for theta sequencesterm, by showing that a moderate addition of time jitterterm·mth to experimental spiketerm trains resulted in preservation of phase precessionterm but a reduced prevalence of theta sequencesterm (Dragoi and Buzsáki, 2006 ; Foster and Wilson, 2007 ; Itskov et al., 2008 ). This suggested that theta sequencesterm required precise coordinationterm between cell assembliesterm in addition to independent phase precessionterm mechanisms. Two recent studies have provided further support for this view. Feng et al. ( 2015 ) analyzed single-trial phase precessionterm and theta sequenceterm formation in rats exploring a novel linear track, and reported that while single place cellsterm exhibit phase precessionterm from the first lap, theta sequencesterm do not emerge until the second lap. Middleton and McHugh ( 2016 ) analyzed the temporal coding of CA1term and CA3term neurons during spatial navigation in a transgenic mouse modelterm·mth lacking CA3term synaptic transmission (CA3term tetanus toxin mouse line). They showed that the absence of CA3term inputs abolished the formation of theta sequencesterm at the CA1term ensemble level despite the persistence of phase precessionterm in individual neurons. They suggested that CA3term may be instrumental for temporal coordinationterm of the CA1term neuronal population. In summary, the above studies point to an intermediate mechanism, where phase precessionterm plays an important but non-exclusive role in generating theta sequencesterm, complemented by temporal coordinationterm processes which may require specific functional connectivity.

To formally test the requirement for additional coordinationterm mechanisms, Chadwick et al. ( 2015 ) developed phenomenological modelsterm·mth of independent and coordinated place cellterm activity during navigation, and compared their predictions in light of experimental data. They concluded that the independent coding modelterm·mth is sufficient to replicate the data previously interpreted as evidence for the coordinated assembly hypothesis. First, although it has been shown that predictions of individual CA1term place cellterm spiketerm timing is improved by information conveyed by their peer cells (Harris et al., 2003 ), suggesting that these cells directly interact with each other consistent with the coordinated assembly hypothesis (Harris, 2005 ), Chadwick et al. ( 2015 ) showed that this no longer holds when both the modulation of firing phase by travel direction (Huxter et al., 2008 ; Climer et al., 2013 ; Cei et al., 2014 ; Jeewajee et al., 2014 ; Maurer et al., 2014 ) and the modulation of firing rate by running speed (e.g., Czurkó et al., 1999 ; Huxter et al., 2003 ; Geisler et al., 2007 ) are taken into account to predict the activity of a given cell. As a matter of fact, peer information even decreases the average prediction, consistent with the independent coding modelterm·mth. Second, the independent coding modelterm·mth also replicated the finding that spiketerm jitterterm·mth can degrade theta sequencesterm but spare the phase-position relation (Dragoi and Buzsáki, 2006 ; Foster and Wilson, 2007 ; Itskov et al., 2008 ), and the covariance in firing rate between cell pairs across successive running laps (Foster and Wilson, 2007 ). Finally, the independent coding modelterm·mth also accounted for global remapping data that the coordinated assembly modelterm·mth failed to replicate.

Incidentally, both the independent coding and coordinated assembly modelsterm·mth can account for internal sequence generation (Pastalkova et al., 2008 ; Wang et al., 2015 ), proposed to support cognitive processes. One potential issue with the independent coding modelterm·mth, which is not directly related to the generation of theta sequencesterm, is that it remains unclear how sequences could reactivate during non-theta states (see also Chadwick et al., 2016 ). A possible solution would be that once theta sequencesterm emerge due to independent phase precessionterm, the repeated activation of successive neurons induces synaptic changes that result in asymmetric network connectivity; this would allow trajectoryterm replayterm during sleep (DrieupersName et al., 2018 ; see section 3.3).

While direct interactions between pyramidalterm neurons may not be required, specific classes of hippocampal interneuronsterm may be involved in coordinating sequential information. As stated earlier, despite the fact that interneuronterm activity usually displays weak spatial modulation (McNaughton et al., 1983 ), many interneuronsterm fire spikesterm that precess in phase against the thetaterm rhythm, tightly coupled to monosynaptically connected pyramidalterm cells (Maurer et al., 2006 ; Ego-Stengel and Wilson, 2007 ; Geisler et al., 2007 ). While this suggested that interneuronterm phase precessionterm may be inherited from place cellsterm (Maurer et al., 2006 ; Geisler et al., 2007 ), Chadwick et al. ( 2016 ) recently proposed an alternative modelterm·mth of theta sequencesterm where phase precessing interneuronsterm play a crucial role in coordinationterm of place cellterm spiketerm timing and the emergence of theta sequencesterm. In their modelterm·mth, phase precessionterm occurs dynamically whenever a place cellterm is driven by external inputs due to the transient functional coupling between place cellsterm and interneuronsterm. Consequently, phase precessionterm and theta sequencesterm are generated de novo within the network, and slow input sequences are automatically compressed into theta sequencesterm in networks of interacting pyramidalterm cells and interneuronsterm. This is consistent with the findings of Royer et al. ( 2012 ), who reported that optogenetic silencing of PV-expressing interneuronsterm shifted the phase of pyramidalterm cell spikesterm toward the trough of thetaterm (conversely, silencing of SOM-expressing interneuronsterm increased the burstiness of pyramidalterm cell spikesterm without altering their thetaterm phase). This suggests that somatic inhibition may play a major role in theta phase precessionterm·def and segregation of cell assembliesterm (Royer et al., 2012 ; Dupret et al., 2013 ; Chadwick et al., 2016 ).

3. Theta Sequencesterm, Navigation, and Memoryterm Both theoretical considerations and experimental data indicate that theta phase precessionterm·def and theta sequencesterm could underlie cognitive functions. While the seminal report of O'Keefe and Recce ( 1993 ) emphasized the probable role of phase precessionterm in spatial coding, the authors also immediately recognized that phase precessionterm placed conditions on the possibility that synaptic connections could be modified. This was shortly followed by suggestions that a critical function of phase precessionterm was to bring neurons to fire within brief time windows conducive to synaptic plasticityterm (Burgess et al., 1994 ; Lisman and Idiart, 1995 ; Skaggs et al., 1996 ). Effectively, theta sequencesterm would allow the hippocampal network to store temporally ordered information, such as events in their spatio-temporal context, thus constituting a substrate for episodic-like memoryterm trace formation. Consistently, there is evidence that degradation of theta sequencesterm impairs sleep replayterm underlying memoryterm consolidationterm (DrieupersName et al., 2018 ), and is accompanied by memoryterm deficits (Lenck-Santini et al., 2008 ; Robbe and Buzsáki, 2009 ). In addition, recent data indicate that theta sequencesterm reflect ongoing cognitive processes (Wikenheiser and Redish, 2015b ), suggesting that they could also be involved in planningterm and decisionterm making.

3.1. Theta-Gamma Neural Code During thetaterm states, nested gammaterm oscillationsterm shape the activity of cell assembliesterm so that largely non-overlapping ensembles are sequentially activated in the 10–30 ms time windows of successive gammaterm cycles which occur at successive thetaterm phases (Lisman and Idiart, 1995 ; Harris et al., 2003 ; Harris, 2005 ). This would enable the hippocampal network to represent four to eight items per thetaterm cycle (corresponding to the number of nested gammaterm cycles), which has been related (Lisman and Idiart, 1995 ) to the “magic number” of 7 ± 2 items that humans can typically hold in short term memoryterm (Miller, 1956 ). Hence, theta-gamma coupling has been hypothesized to form a neural code representing multiple items in a temporal frame ( Figure 4A ; Jensen and Lisman, 2005 ; Lisman and Buzsáki, 2008 ; Lisman and Jensen, 2013 ). Buzsáki ( 2010 ) even proposes the concept of a neural syntax, in which cell assembliesterm, the fundamental computational unit (letters), are arranged in specific orders by oscillationsterm, external or internal excitatory inputs and local inhibition, in order to form words.

Figure 4.

Open in a new tab Theta-gamma neural code. (A) Individual memoryterm representations (cell assembliesterm, identified by letters) are repeatedly activated within each new thetaterm cycle. The gammaterm rhythm separates the respective representations in time. The number of gammaterm cycles per thetaterm cycle determines the span of working memoryterm. (B) If discrete locations (i.e., A–G) are encoded within individual gammaterm cycles, longer paths would be encoded during fast gammaterm than during slow gammaterm. Alternatively, slow (but not fast) gammaterm cycles encode sequences of locations, rather than single locations. This results in longer paths encoded during slow gammaterm than during fast gammaterm.

Evidence for a role of thetaterm in information chunking has been provided in a number of experimental studies. By quickly changing light cues on the walls and floor of a recording chamber, Jezek et al. ( 2011 ) induced rapid switches between two perceived environments, and reported that CA3term representations flickered between the two maps. These network flickers tended to occur at the timescaleterm of one or a few thetaterm cycles. Notably, activity within a single thetaterm cycle correlated with either environment, but rarely with both simultaneously, suggesting that thetaterm cycles constituted temporal units of information. Similarly, Gupta et al. ( 2012 ) recorded CA1term activity while rats performed a decisionterm task on a eight-shape maze with two reward sites, and reported that neural sequences expressed during single thetaterm cycles reflected cognitive segments of the task. This supported the view that thetaterm cycles may underlie the cognitive chunking of experience. Further, they reported that the length of the path encoded in a given thetaterm cycle was directly related to the length of the cycle and to the number of gammaterm cycles it contained, suggesting that gammaterm cycles constitute information processing steps (see also Zheng et al., 2016 ).

The role of gammaterm oscillationsterm in organizing spiketerm timing has been supported by several recent studies. During thetaterm oscillationsterm, at least two types of nested gammaterm oscillationsterm can occur in short bursts (Bragin et al., 1995 ), although it remains unclear whether these can occur within single thetaterm cycles (Belluscio et al., 2012 ) or they are restricted to different thetaterm cycles (Colgin et al., 2009 ). In CA1term, slow gammaterm (gammaterm S , ~30–50 Hz) preferentially occurs on the descending phase of thetaterm. It is thought to arise from CA3term and to propagate along the Schaffer collaterals to the stratum radiatum . Medium gammaterm oscillationsterm (gammaterm M , ~50–90 Hz) tend to occur near the thetaterm peak and may emerge from MEC inputs targeting the stratum lacunosum-moleculare (Bragin et al., 1995 ; Csicsvari et al., 2003 ; Colgin et al., 2009 ; Belluscio et al., 2012 ; Schomburg et al., 2014 ; Buzsáki and Schomburg, 2015 ).

Takahashi et al. ( 2014 ) studied hippocampal responses during fixation periods in a delayed spatial-alternation task. Neurons selective for the alternation sequence phase precessed during the entire fixation period, shifting their spikesterm from the peak to the trough of thetaterm oscillationsterm. Concomitantly, gammaterm M power (60–90 Hz) arising at the thetaterm peak decreased, while gammaterm S power (30–45 Hz) arising at the thetaterm trough increased toward the end of the fixation period. The authors hypothesized that CA1term received predominant inputs from the EC at fixation onset, providing information about ongoing context. This signal would then elicit auto-associative intra-hippocampal processing in CA3term to recall the required action, or to look up in memoryterm the appropriate task sequence given the ongoing context. This scenario, reminiscent of the modelterm·mth of Chance ( 2012 ), is consistent with the finding that gammaterm S power is reduced and theta sequencesterm are disrupted in inducible KO mice lacking CA3term inputs (Middleton and McHugh, 2016 ). Different gammaterm frequency bands would thus reflect different types of information processing, e.g., externally cued vs. internally generated.

This is in line with the notion that thetaterm cycles have two complementary functional modes (Hasselmo et al., 2002 ): an encoding phase at the peak of thetaterm (as recorded from CA1term pyramidalterm layer), largely mediated by EC inputs, and a retrieval phase at the thetaterm trough mediated by CA3term recurrent activity. Incidentally, this could explain why Jezek et al. ( 2011 ) reported a better separation of the two hippocampal maps during the second half of the thetaterm cycle, when CA3term retrieval mode would dominate. A similar dissociation would be expected in the study of Gupta et al. ( 2012 ), where theta sequencesterm representing trajectories ahead of the animal, reflecting memoryterm retrieval, could be linked to higher gammaterm S band power, whereas trajectories behind the animal, reflecting memoryterm encoding, could be linked to higher gammaterm M band power. Unfortunately, this was not investigated, but two recent studies reported a similar dissociation in rats running on a linear track.

In the first study, Bieri et al. ( 2014 ) found that late thetaterm phases were dominated by gammaterm S , when spikesterm appear to anticipate the upcoming field location (“prospective” firing, Battaglia et al., 2004 ). In contrast, early phases were dominated by gammaterm M , when spikesterm lag behind field location (“retrospective” firing). These results suggest that alternating gammaterm S and gammaterm M states allow the hippocampusterm to switch between prospective and retrospective modes, possibly to prevent interference between memoryterm retrieval and encoding.

In the second study, Zheng et al. ( 2016 ) performed Bayesian reconstruction of the trajectoryterm of the animals at the thetaterm timescaleterm, and showed that theta sequencesterm more accurately represented ongoing trajectories during thetaterm cycles dominated by gammaterm M than during thetaterm cycles dominated by gammaterm S ( Figure 4B ). Gammaterm S -associated theta sequencesterm were found instead to represented relatively long paths extending ahead of the current location. In these sequences, spiketerm phases relative to gammaterm S also conveyed spatial information (Senior et al., 2008 ). This again argues that encoding and retrieval could be mediated by different gammaterm subtypes: relatively long gammaterm S periods may allow successive items within the stored sequence to be retrieved in a time-compressed manner; in contrast, short periods of gammaterm M would encode spatial memories in real time and may ensure chunking of cell assembliesterm. Note however that in this study, entire thetaterm cycles were considered to contribute either gammaterm S or gammaterm M sequences (depending on the overall dominating gammaterm band), but not both (although see e.g., Belluscio et al., 2012 ), leaving unexplored the possibility that encoding and retrieval could both occur within single thetaterm cycles (Hasselmo et al., 2002 ).

The spatial strategy of the animal may also bias the balance between gammaterm S and gammaterm M . Cabral et al. ( 2014 ) trained mice to find food rewards in a five-arm maze. During training, mice always started exploring the maze from the same location. But during test trials, they started from a different location, which allowed to assess if the animals used a place strategy (navigation to the same place in space) vs. a sequence strategy (performing the same series of left or right turns). When the mice used a sequence strategy, gammaterm S was most prominent and place maps were more similar to training trials (after realignment), compared to when the mice used a place strategy, when spikesterm emitted during gammaterm S and gammaterm M periods contributed equally to the hippocampal representation. A possible interpretation is that during sequence strategy trials, the hippocampal network retrieves sequences of cell assembliesterm stored in CA3term, while during place strategy trials, hippocampal activity would be equally driven by the recall of memoryterm stored in CA3term and by landmark-related sensory information conveyed by the EC.

3.2. Theta Sequencesterm and Trajectoryterm Planningterm As hinted previously, theta sequencesterm do not appear to be entirely driven by sensori-motor signals. For instance, Gupta et al. ( 2012 ) reported that “look-ahead” or “look-behind” thetaterm sweeps were preferentially expressed in different maze segments, and were modulated by the running velocity of the animal ( Figure 5 ). Bieri et al. ( 2014 ) found that prospective spikesterm tended to occur when rats were leaving reward locations (the ends of a linear track), whereas retrospective spikesterm tended to occur as rats approached reward locations.

Figure 5.

Open in a new tab Theta sequencesterm and trajectoryterm planningterm. (Left) A rat runs on the central arm of a maze (blue dashed box) and must decide whether to turn left of right at a choice point (red dashed box). Letters represent hippocampal cell assembliesterm and colored ellipses illustrate their respective firing fields on the maze. Center. During deliberation (vicarious trial and error, VTEterm·def) at the choice point, sequences of cell assembliesterm alternatively sweep toward the two possible paths (top). Extra-field firing can occur before the animal turns back toward the choice point during error correction trials (bottom). (Right) When the animal runs along a segment of the maze, theta sequencesterm can represent positions centered on the animal (“in-between”), behind the animal (“more behind”) or ahead (“more ahead”). The spatial representation embedded in theta sequencesterm is influenced by multiple factors, including the velocity and acceleration of the animal, as well as the location of surrounding landmarks.

Convincing and direct evidence that cognitive mechanisms can regulate theta sequencesterm was provided by Johnson et al. ( 2007 ), who recorded CA3term activity in rats performing two different versions of the T-maze decisionterm task. At decisionterm points, when the rat had to choose to turn left or right, theta sequencesterm quickly and repeatedly swept forward toward the reward sites, alternatively representing the two possible paths. Similar anticipatory sequences occurred upon recovery from an error ( Figure 5 ). The animals performed two different tasks: a multiple-T task, in which reward locations were fixed within a given session but changed from day to day, and a cued-choice task, in which a tone indicated for each lap which arm would be rewarded. In the multiple-T task, the number of sweeps down the unrewarded arm decreased across trials. In contrast, in the cued-choice task, sweeps kept representing both directions throughout the session. Interestingly, at the decisionterm point, the rats often expressed a typical behavior, consisting in pausing and orienting the head toward the left or right corridors, as if deliberating over the choice. This process is referred to as “vicarious trial and error” (VTEterm·def; Muenzinger and Gentry, 1931 ). Subsequent studies further linked VTEterm·def to decision-making processes (Schmidt et al., 2013 ; Amemiya and Redish, 2016 ; for a complete reviewterm·mth see Redish, 2016 ). Of particular note, Wikenheiser and Redish ( 2015b ) showed that theta sequencesterm predicted longer trajectories when the animal was actually planningterm to reach a more distant goal. Thus, theta sequencesterm also convey information about ongoing goals or intentions.

3.3. Theta Sequencesterm and Memoryterm Because the mechanisms underlying theta sequencesterm remain poorly understood and may involve many interacting factors (including single-cell properties, synaptic connectivity patterns, local inhibition, theta-paced inputs, etc.), interfering with theta sequencesterm is challenging. To the best of our knowledge, only three studies have reported task performance deficits linked to a degradation of theta sequencesterm.

In the first study, Lenck-Santini et al. ( 2008 ) investigated hippocampal thetaterm dynamics in rats, following status epilepticus , a condition that leads to mesial temporal lobe epilepsy (MTLE). These rats had altered phase precessionterm and temporal organization of firing among pairs of neurons, along with deficits in spatial learning in the Morris water maze. This suggests that memoryterm impairments stemmed from impaired temporal coding, consistent with a role for theta sequencesterm in normal memoryterm function.

In the second study, Robbe and Buzsáki ( 2009 ) examined the involvement of cannabinoid receptors in the dynamics of the hippocampal network (see also Robbe et al., 2006 ). Rats were trained to perform a hippocampus-dependent delayed spatial alternation task ( Figure 6A ). Administration of the cannabinoid receptor agonist CP55940 profoundly and reversibly impaired task performance. While the hippocampal place representation remained essentially intact, both theta phase precessionterm·def and spike-timing coordinationterm between place cellsterm at the thetaterm timescaleterm were reversibly reduced. Notably, the percentage of correct trials on the task was correlated with the theta-scale pairwise coordinationterm (assessed as the “compression indexterm·stat,” i.e., the ratio of the distance between the place fields of pairs of place cellsterm and the average time lag between their spiketerm trains at the thetaterm timescaleterm). These results strongly argue for a role of phase precessionterm and theta sequencesterm in task performance. One potential caveat however is that CP55940 administration decreases ripple power (Robbe et al., 2006 ). Since place cellterm activity during awake ripples may play a crucial role in working memoryterm (Jadhav et al., 2012 ), it is conceivable that the performance deficit could also result from altered place cellterm activity during awake ripples.

Figure 6.

Open in a new tab Theta sequencesterm and memoryterm. (A) Selective perturbation of phase precessionterm, but not firing maps, by injection of cannabinoid receptor agonist CP55940 induces performance deficits in a spatial memoryterm task. Top. Firing maps for an example place cellterm recorded in control (left), drug (center), and recovery (right) conditions. Bottom left. Average power of the cross-correlograms for all neuron pairs with overlapping fields during control (blue), drug (red), and recovery conditions (magenta; dotted lines: s.e.m.) Bottom center. Correlation between behavioral performance (percentage of correct trials) and theta-scale sequence compression indexterm·stat in one rat. The different symbols refer to data obtained in two different experiments. Bottom right. Average performance (correct trials) for all sessions and rats. (B) Medial septumterm inactivation by muscimolterm·mth infusion impairs theta sequencesterm and behavioral performance. (C) Perturbation of theta sequencesterm impairs subsequent sleep replayterm. Top. Experimental protocol. The rats traveled on a modelterm·mth train. In successive sessions, a miniature treadmill was either turned off (passive travel; red and yellow) or on (active travel; green) to selectively disrupt or maintain theta sequencesterm, while leaving behavioral timescaleterm sequences intact. Bottom left. Firing fields (left column) remained stable in all conditions (red, green and yellow curves), but theta sequencesterm (right column) emerged only during active travel (green box). Bottom center. Trajectories represented by hippocampal sequences were assessed using a combination of measures, namely trajectoryterm score (linearity) and slope (velocity). During passive travel and subsequent sleep (red and yellow boxes), a very low proportion of both thetaterm and replayterm sequences emerged in the hippocampal network (upper right quadrants are mostly black). On the contrary, during active travel and subsequent sleep (green boxes), a dramatically higher number of theta sequencesterm and replayterm events were observed (red dots in upper right quadrants). Bottom right. Similarity of thetaterm and replayterm sequences (P1, P2: passive travel sessions; A: active travel session; SA, SP1, SP2: sleep following respective travel sessions). Theta sequencesterm were clear and self-consistent only during active travel (red zone at [A,A]). Sleep replayterm mirrored theta sequencesterm only following active travel (red zone at [A,SA]). Panel (A) adapted from Robbe and Buzsáki ( 2009 ); panel (C) adapted from DrieupersName et al. ( 2018 ), with permission from AAAS.

Wang et al. ( 2015 ) also reported impaired task performance following perturbation of theta-timescale activity ( Figure 6B ). In rats performing a delayed spatial alternation task, medial septumterm inactivation by muscimolterm·mth injection abolished theta sequencesterm both on the maze and in the wheel where the rats ran during delay periods (although measures of thetaterm cycles and therefore theta sequencesterm may not be as reliable when thetaterm amplitude vanishes following muscimolterm·mth injection). On the other hand, the place-cell representation was spared when the rats were navigating the maze. Yet, performance was impaired. This suggests that theta sequencesterm normally formed on the maze would have been necessary to achieve optimal levels of performance. One cannot entirely rule out that the observed performance deficits could be directly due to disruption of thetaterm oscillationsterm (Pan and McNaughton, 1997 ; Hasselmo et al., 2002 ) induced by medial septumterm inactivation (~80% decrease in thetaterm power, Wang et al., 2015 ).

In summary, the above studies are consistent with the view that theta sequencesterm are involved in the formation and recall of spatial memoryterm, possibly by enabling the synaptic changes that underlie the storage of initial memoryterm traces (Burgess et al., 1994 ; Skaggs et al., 1996 ). Yet, an alternative possibility is that these performance deficits resulted from impairments in planningterm and decisionterm making, which also appear to involve theta sequencesterm (Johnson et al., 2007 ; Schmidt et al., 2013 ; Wikenheiser and Redish, 2015b ; Amemiya and Redish, 2016 ), rather than from deterioration of learning and memoryterm processes. Thus, are theta sequencesterm actually necessary for memoryterm formation and recall? Bittner et al. ( 2017 ) recently uncovered a novel form of synaptic potentiationterm occurring at the behavioral timescaleterm, and proposed that this could suffice to store entire sequences of events that took place several seconds before and after initiation of plateau-potentials in CA1term pyramidalterm cells. This effectively undermines the theoretical framework that theta sequencesterm are required to bring spiketerm trains in close temporal proximity and enable synaptic potentiationterm, underlying hippocampus-dependent memoryterm encoding.

Because during exploration, theta sequencesterm are nested within behavioral timescaleterm sequences, disentangling their respective roles in memoryterm encoding requires a protocol that instantly and selectively blocks or releases theta sequencesterm throughout the hippocampal system, but spares the firing fields of place cellsterm (and therefore, behavioral timescaleterm sequences). Our previous work has established that passive travel alters phase precessionterm in most place cellsterm (see also Song et al., 2005a ; Terrazas et al., 2005 ), except if the rats actively run on a treadmill while being transported (Cei et al., 2014 ). Importantly, switching the treadmill on or off does not induce global remapping of place cellsterm. We have recently leveraged this protocol to selectively impair phase precessionterm (treadmill off) or leave it intact (treadmill on), while maintaining the spatial layout of place fields in the environment (DrieupersName et al., 2018 ). The rats were tested in an entirely novel environment, to stimulate spatial learning, network coordinationterm and synaptic plasticityterm (Li et al., 2003 ; Leutgeb et al., 2004 ; Cheng and Frank, 2008 ; O'Neill et al., 2008 ), as well as to ensure that previously formed memoryterm traces would not bias our observations. As expected, impaired phase precessionterm during passive travel (treadmill off) was accompanied by an almost complete absence of theta sequencesterm. On the other hand, behavioral timescaleterm sequences were unaffected. During subsequent sleep, the hippocampal network failed to replayterm trajectories. Only following “active” travel (treadmill on), when theta sequencesterm emerged nested within behavioral timescaleterm sequences, did the hippocampal network generate genuine replayterm events during sleep (DrieupersName et al., 2018 ) ( Figure 6C ). This indicates that theta sequencesterm play a causal role in establishing initial memoryterm traces, which can later be reinstated during sleep for memoryterm consolidationterm.

Note that in this study, the functional role of theta sequencesterm was not assessed from behavior (task performance), but from the occurrence of dynamic network patterns with a well-documented function, namely trajectoryterm replayterm underlying memoryterm consolidationterm. As a consequence, this alleviates the ambiguity discussed earlier, that behavioral impairments could result either from memoryterm deficits, or from impairments in planningterm and decisionterm making.

Interestingly, additional evidence for a link between theta sequencesterm and sleep replayterm was recently provided by Muessig et al. ( 2019 ), who recorded from rat pups at different post-natal stages. They reported that before weaning (P21), ripple-associated activity represents stationary locations in space, and does not start to represent bona fide trajectories until theta sequencesterm progressively emerge between P17 and P32.

Finally, over the last 15 years, an ever growing number of studies have addressed the network substrates of memoryterm using causal approaches. However, these studies have generally failed to consider the temporal dimension of neuronal activity, such as manifested in cell assembliesterm and sequences. Experimental exploration of engrams (Semon, 1921 ) has benefited from the development of optogenetics (Boyden et al., 2005 ) and gene expression control techniques combined with immediate early gene (IEG) expression signals, to artificially manipulate memoryterm traces (e.g., Reijmers et al., 2007 ; Liu et al., 2012 ; Cowansage et al., 2014 ; Redondo et al., 2014 ; Trouche et al., 2016 ). These strategies allow tagging and subsequent manipulation of neuronal populations activated during learning (for recent reviewsterm·mth, see Silva et al., 2009 ; Tonegawa et al., 2015 ; Guan et al., 2016 ; Kim et al., 2016 ; Poo et al., 2016 ). Thus, selective activation or inhibition of tagged neuronal ensembles have been shown to induce (Liu et al., 2012 ; Cowansage et al., 2014 ; Kim et al., 2014 ; Redondo et al., 2014 ) or block (Han et al., 2009 ; Trouche et al., 2016 ) the expression of memoryterm, respectively.

One important issue with most of these studies is that they artificially activate neurons in a hyper-synchronous manner, and neglect any fine-timescale temporal information embedded in endogenous spiking patterns. In addition, because the time window for genetic labeling of neuronal networks ranges from several hours to days, the tagged population is likely to include both memory-specific neurons as well as numerous irrelevant neurons that were activated by unrelated events during the relatively long labeling time window. In any case, it remains unclear whether the tagged neurons represent a memoryterm trace or constitute relays toward the relevant ensembles. For example, memoryterm recall could actually result from downstream circuit computations rather than from the experimental activation of the tagged neurons per se . This would potentially restore the temporal patterns obliterated by the synchronous experimental stimulation. A parallel can be made with fRMI studies showing robust and consistent activation patterns in response to similar stimuli, despite the low temporal and spatial resolution of the technique (Guest and Love, 2017 ). Thus, while certain types of computation can still be detected or elicited even without having access to the fine time scale of neuronal computation, it is uncertain that they reflect actual brain computation during learning and memoryterm. Further experimental and computational work will undoubtedly shed light on the complex issue of the temporality of circuit computations underlying memoryterm.

4. Hippocampal Sequences During Exploration: Distinct and Complementary Functions 4.1. Awake Replayterm During Sharp-Wave Ripples Theta sequencesterm are not the sole pattern of sequential activity in the awake hippocampusterm. While offline replayterm of neuronal sequences was first documented during sharp-wave ripples (SPW-Rs) of slow wave sleep (Pavlides and Winson, 1989 ; Wilson and McNaughton, 1994 ; Lee and Wilson, 2002 ), replayterm also occurs during SPW-Rs of quiet awake states, such as when the animal pauses in an environment ( Figure 7 ; Buzsáki, 1989 ; Foster and Wilson, 2006 ; Jackson et al., 2006 ; O'Neill et al., 2006 ; Csicsvari et al., 2007 ; Diba and Buzsáki, 2007 ; Dupret et al., 2010 ). Foster and Wilson ( 2006 ) recorded CA1term pyramidalterm cell activity while rats ran back and forth between reward sites, and showed that compared to the location of their firing fields on the track, place cellsterm tended to reactivate in reverse order during awake replayterm, contrary to replayterm during slow wave sleep (although more recent studies have since documented overlooked reverse replayterm during sleep as well; see e.g., Wikenheiser and Redish, 2013 ; Silva et al., 2015 ; Grosmark and Buzsáki, 2016 ; Ólafsdóttir et al., 2016 ; O'Neill et al., 2017 ; DrieupersName et al., 2018 ). Replayterm events were more readily observable on a new track, where they occurred even after the very first lap (Foster and Wilson, 2006 ). In addition, Diba and Buzsáki ( 2007 ) reported that replayterm tended to occur in the reverse order upon arrival at the reward site, but in the forward order (“preplayterm”) when rats then faced the track in anticipation of the next run ( Figure 7 ). Interestingly, the authors documented a strong correlation between the temporal offsets of neuron pairs during forward events (preplayterm) and during thetaterm oscillationsterm when the animals were running between reward sites. In contrast, theta sequencesterm and reverse replayterm events were less strongly correlated, suggesting a selective link between offline preplayterm and neural activity during run. The balance between replayterm and preplayterm may also depend on task contingencies, as Ólafsdóttir et al. ( 2017 ) found a greater proportion of forward replayterm immediately (< 5 s) before and after movement or during short stops (< 10 s), but no directional bias during longer periods of immobility (although one cannot exclude that the apparent inconsistency with the results of Diba and Buzsáki ( 2007 ) may stem from the limited number of replayterm events examined in the Ólafsdóttir et al. ( 2017 ) study).

Figure 7.

Open in a new tab Awake replayterm. Place cellterm sequences experienced during behavior are generated in both the forward and reverse direction during awake SPW-Rs (Diba and Buzsáki, 2007 ). Spiketerm trains of place cellsterm on the track are shown before, during and after a single traversal. Sequences that occur during track running are reactivated during SPW-Rs both before and after the run, when the rat remains immobile. Forward preplayterm (Left) tends to occur as the animal is getting ready to run on the track, while reverse replayterm (Right) tends to occur shortly after run. Black traces, CA1term local field potential.

Foster and Wilson ( 2006 ) hypothesized that reverse replayterm could be associated with a slowly decaying dopamine signal, effectively storing a value gradient path that the animal could follow during subsequent goal-finding behavior. Conversely, forward preplayterm has been hypothesized to be involved in goal-directed planningterm (Diba and Buzsáki, 2007 ). Accordingly, awake replayterm can be influenced by reward locations (Foster and Wilson, 2006 ; Singer and Frank, 2009 ; Ambrose et al., 2016 ), contributes to mapping novel environments (O'Neill et al., 2006 ; Cheng and Frank, 2008 ; Roux et al., 2017 ), helps establishing goal-related cell assembliesterm (Dupret et al., 2010 ; Csicsvari and Dupret, 2014 ), supports decisionterm making (evaluation of choices, prospection and planningterm) (Pfeiffer and Foster, 2013 ; Singer et al., 2013 ; Yu and Frank, 2015 ), and plays a role in fear (Wu et al., 2017 ) and episodic-like (Takahashi, 2015 ) memoryterm retrieval. In addition, awake replayterm can span multiple SPW-R events to represent extended trajectories (Davidson et al., 2009 ) and link familiar paths into novel trajectories never experienced before (Gupta et al., 2010 ). Finally, selective suppression of awake SPW-Rs during a spatial alternation task fails to impair memoryterm consolidationterm, and instead appears to alter working memoryterm and decisionterm making (Jadhav et al., 2012 ). In summary, awake replayterm has been proposed to mediate both spatial and non-spatial functions (see O'Neill et al., 2010 ; Buhry et al., 2011 ; Carr et al., 2011 ; and for a recent reviewterm·mth Roumis and Frank, 2015 ).

Because both forward and reverse replayterm have a tendency to start at the current location of the animal, awake replayterm has been proposed to be biased by a residual, place-selective spatial tuning drive (Foster and Wilson, 2006 ; Csicsvari et al., 2007 ; Diba and Buzsáki, 2007 ). However, awake replayterm can also be modulated by non-local cues such as rewards (Pfeiffer and Foster, 2013 ; Singer et al., 2013 ) and contextual memories (Takahashi, 2015 ; Wu et al., 2017 ), or even be initiated at remote locations (Davidson et al., 2009 ; Karlsson and Frank, 2009 ; Gupta et al., 2010 ). This is at odds with the seminal modelterm·mth which posits that replayterm (in particular, reverse replayterm) can be accounted for by decreasing excitability over time (Buzsáki, 1989 ; Foster and Wilson, 2006 ). Instead, it has been proposed that awake replayterm could emerge from auto-associative properties of the CA3term network, in which the activation of a single cell can initiate the reactivation of an entire trajectoryterm sequence due to recurrent connectivity established or reinforced during theta-state exploration (Lee and Wilson, 2002 ; O'Neill et al., 2008 ; Dupret et al., 2010 ; Silva et al., 2015 ), similar to the classical modelterm·mth of sleep replayterm. In turn however, this modelterm·mth cannot account for the (weak) link between the time spent exploring different areas of the environment and the propensity of replayterm to represent the corresponding trajectories (Gupta et al., 2010 ; but see O'Neill et al., 2008 ), the construction of never-experienced sequences (Gupta et al., 2010 ), nor the differential balance of forward and backward replayterm across behavioral states (Wikenheiser and Redish, 2013 ). It is therefore likely that replayterm results from a combination of these mutually non-exclusive mechanisms. Hence, Atherton et al. ( 2015 ) have suggested that the lingering excitability modelterm·mth largely dictates local replaying sequences during awake behavior, while the synaptic plasticityterm modelterm·mth contributes to subsequent nonlocal awake replayterm (and sleep replayterm). Finally, neuromodulatory activity may also exert an important role (Atherton et al., 2015 ): dopamine release during novel maze exploration and reward-driven spatial tasks may promote the strengthening of new place cellterm assemblies and place-reward associations (Benchenane et al., 2010 ; Dupret et al., 2010 ; McNamara et al., 2014 ; de Lavilléon et al., 2015 ; Ambrose et al., 2016 ; Takahashi et al., 2016 ; Takeuchi et al., 2016 ), while cholinergic tone may influence the direction of replayterm in different behavioral states (Marrosu et al., 1995 ; Csicsvari et al., 2007 ; Wikenheiser and Redish, 2013 ; Saravanan et al., 2015 ).

4.2. Functional Dissociation Between Theta Sequencesterm and Awake Replayterm?

The above discussion suggests a functional dissociation between theta sequencesterm and awake replayterm, where theta sequencesterm would store initial memoryterm traces about ongoing experiences as well as recall previously formed memories and enhance them to include goal-directed information, while awake replayterm would play an instrumental role in learning goal locations and planningterm upcoming trajectories. Complementing these mechanisms, sleep replayterm would sustain memoryterm consolidationterm. As it turns out, a closer examination of the available data suggests a more complex picture. Below, we point out some of the current inconsistencies and overlooked findings.

4.2.1. Memoryterm Maintenance and Retrieval Wang et al. ( 2016 ) inactivated the medial septumterm in rats trained on a delayed spatial alternation task. They found that theta sequencesterm were disrupted and performance was impaired. Surprisingly, awake replayterm was preserved. The authors concluded that theta sequencesterm on the maze (and in the running wheel during delays), but not awake replayterm, are required to recall episodic memories even in well-trained animals. Alternatively, awake replayterm would be required only for initial learning of a task but not once the environment and the task have become familiar. This would be consistent with the finding that suppression of awake SPW-Rs during the first 8 days of learning impairs working memoryterm in a spatial alternation task (Jadhav et al., 2012 ). Yet another possible explanation for the apparently predominant role of theta sequencesterm over awake replayterm may be that the rats spent the delay period running in the wheel, which may have favored neuronal events occurring in the thetaterm state. In any case, these results still argue that theta sequencesterm and awake replayterm do not play equivalent, interchangeable roles.

Wu et al. ( 2017 ) trained animals in a foot-shock avoidance task and reported that avoidance of the shock zone was concomitant with awake replayterm, but not theta sequencesterm, of paths leading from current position to the foot-shock zone. This further supports distinct functional roles for theta sequencesterm and awake replayterm in long-term contextual memoryterm retrieval.

4.2.2. Planningterm and Decision-Making Another notable difference between theta sequencesterm and awake replayterm is that theta sequencesterm seem to be restricted to short distances always encompassing current location, occasionally extending to further locations (Johnson et al., 2007 ; Wikenheiser and Redish, 2015b ), whereas awake replayterm can span much longer distances (up to ≥10 m; Davidson et al., 2009 ) and be decoupled from current experience (Karlsson and Frank, 2009 ; Gupta et al., 2010 ). Therefore, while theta sequencesterm may be relevant for spatial memoryterm, in particular to allow trajectoryterm encoding and retrieval (Hasselmo and Eichenbaum, 2005 ; Dragoi and Buzsáki, 2006 ; Foster and Wilson, 2007 ; Robbe and Buzsáki, 2009 ; Wang et al., 2015 ; DrieupersName et al., 2018 ), awake replayterm may be necessary to associate paths and events disconnected in time and space, allowing flexibility and adaptive behaviors (Gupta et al., 2010 ).

During deliberative behavior at a choice point (Muenzinger and Gentry, 1931 ; Redish, 2016 ), theta sequencesterm reflect possible paths (Johnson et al., 2007 ). Papale et al. ( 2016 ) reported that the occurrence rate of VTEterm·def at the choice point was inversely correlated to that of SPW-Rs at the reward sites on the same lap. Further, the SPW-R rate on a given lap inversely correlated with the likelihood of subsequent VTEterm·def events. Finally, interruption of SPW-Rs during learning led to an increased prevalence of VTEterm·def. This suggests that SPW-Rs are involved in ongoing consolidationterm, while VTEterm·def sequences participate in a decision-making process. On the other hand, SPW-Rs are also known to occur at decisionterm points during learning (Jadhav et al., 2012 ), which could reflect an implication in decision-making processes, blurring the proposed functional distinction between theta sequencesterm and awake replayterm. Disentangling these matters is further hindered by the fact that ripple disruption in Jadhav et al. ( 2012 ) was not triggered with spatial specificity: ripples were disrupted irrespective of the location of the rat. Therefore, the specific functional contribution of ripples occurring at decisionterm points remains to be elucidated.

4.2.3. Learning and Consolidationterm Both thetaterm and SPW-Rs are propitious to synaptic potentiationterm between sequentially activated cell assembliesterm (Skaggs et al., 1996 ; Magee and Johnston, 1997 ; Markram et al., 1997 ; Dragoi et al., 2003 ; Wójtowicz and Mozrzymas, 2015 ; Sadowski et al., 2016 ; Cobar et al., 2017 ; DrieupersName et al., 2018 ). Therefore, in principle neuronal sequences in both states could lead to learning and reinforcement of sequential activity.

In the machine learning literature, reverse replayterm has been repeatedly reported to enhance reinforcement learning, in particular temporal difference learning (Cazé et al., 2018 ; Mattar and Daw, 2018 ). Interestingly, awake replayterm during brief pauses in exploration and at reward sites largely occur in reverse order (O'Neill et al., 2008 ; Wikenheiser and Redish, 2013 ), identifying a potential mechanisms to support learning of trajectories and goal locations (Singer et al., 2013 ). In addition, working memoryterm deficits have been observed following suppression of SPW-Rs that occurred mainly at reward locations (Jadhav et al., 2012 ). Theta sequencesterm on the other hand mainly reflect currently experienced trajectories in the forward order. Yet, a small fraction of thetaterm cycles do contain significant reverse theta sequencesterm (Gupta et al., 2012 ; Zheng et al., 2016 ; our own unpublished observations). The function of reverse theta sequencesterm remains unclear. Similarly, although this has been overlooked in initial descriptions of sleep replayterm, recent studies have reported abundant reverse replayterm during sleep as well (Wikenheiser and Redish, 2013 ; Silva et al., 2015 ; Grosmark and Buzsáki, 2016 ; Ólafsdóttir et al., 2016 ; O'Neill et al., 2017 ; DrieupersName et al., 2018 ). Again, these considerations dim the usual theoretical distinction between (forward) theta sequencesterm for memoryterm encoding, (reverse) awake replayterm for goal learning, and (forward) sleep replayterm for memoryterm consolidationterm.

An additional complexity is that SPW-R incidence and replayterm strength and accuracy vary according to the length, shape and complexity of the maze, as well as the cognitive demand of the task (Jackson et al., 2006 ; Dupret et al., 2010 ; Pfeiffer and Foster, 2013 ; Singer et al., 2013 ). Increases in both SPW-R occurrence rate and replayterm strength can even be observed in familiar environments (Jackson et al., 2006 ; Dupret et al., 2010 ; Pfeiffer and Foster, 2013 ; Singer et al., 2013 ), perhaps because awake replayterm helps stabilize hippocampal place fields when new configurations are learned within a familiar map (Roux et al., 2017 ).

5. Conclusion Sequences of hippocampal cell assembliesterm formed during individual thetaterm cycles represent the time-compressed ongoing trajectoryterm of the animal. These theta sequencesterm are thought to emerge from ensembles of phase precessing cells, although it is currently unclear whether this requires additional coordinationterm between participating neurons. Numerous computational modelsterm·mth have proposed cellular and network mechanisms to account for theta sequencesterm, but the precise neurophysiological mechanisms of phase precessionterm remain elusive. While initial reports emphasized the presence of sequential activity during single thetaterm cycles, recent studies have increasingly recognized that a notable fraction of thetaterm cycles contain non-canonical activity, including seemingly random noise and even reverse sequences.

Theta sequencesterm chunk information into time windows conducive to classical synaptic plasticityterm mechanisms, and appear to support the initial storage of memoryterm traces. Within familiar environments, theta sequencesterm allow the maintenance (and possibly the update) of planned trajectories in working memoryterm. During VTEterm·def, extended neuronal sweeps may support deliberative processes. Awake replayterm could also support trajectoryterm planningterm, on-line consolidationterm and reconsolidation, as well as on-line retrieval of long-term memoryterm. Thus, it remains currently unclear which specific functions are sustained by theta sequencesterm, awake replayterm, or both.

Finally, a growing number of studies have provided evidence that hippocampal activity can also convey non-spatial information, including about objects, odors, sound sequences, time, goals, fear, etc. (e.g., Wood et al., 1999 ; Moita et al., 2003 ; Macdonald et al., 2011 ; Aronov et al., 2017 ). This suggests a general mechanism for encoding continuous, task-relevant variables. Recent findings indicate that phase precessionterm may also represent a potential code for non-spatial features (Terada et al., 2017 ). Future work will be required to determine how non-spatial information is integrated at the fine timescaleterm during hippocampal computation.

Author Contributions All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

Conflict of Interest Statement The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Glossary Cell assemblyterm·def. Group of neurons that transiently co-activate during a certain function or mental process. Place cellsterm with highly similar fields can fire within a brief window of time (common thetaterm phase, often grouped within a gammaterm cycle) and form a cell assemblyterm·def. Reciprocal connectivity between group members may or may not be implied (e.g., cell assembliesterm in CA3term vs. CA1term).

Theta sequenceterm. Sequential activation of hippocampal cell assembliesterm paced by the ongoing thetaterm rhythm (~8 Hz) and representing the ongoing trajectoryterm at a highly accelerated rate. For instance, when a rat successively traverses place fields A, B, and C (which lasts on the order of a second), in the overlapping region, ongoing thetaterm cycles contain spikesterm A, B, and C in the same order (within only ~150 ms). Theta sequencesterm typically span one thetaterm cycle, but longer sequences have also been described.

Replayterm. Reinstatement during offline states of the sequential activity observed during behavior. Replayterm is associated with sharp wave ripple complexes and can occur both in the forward or backward direction.

Theta phase precessionterm·def. The time at which a cell fires relative to the ongoing thetaterm cycle (“phase”) decreases as the animal moves through space. Thus, the cell initially fires at the end of a thetaterm cycle, then slightly before the end of the next cycle, until it eventually fires its last spikesterm near the beginning of the ongoing thetaterm cycle. This phenomenon is called “phase precessionterm.” Initially described in place cellsterm, phase precessionterm has also been reported in other hippocampal cells, including interneuronsterm, “episode,” and sound-responsive cells (in which case the movement is considered in a more abstract space), as well as in other brain structures such as the medial prefrontal cortex or the ventral striatum.

Nested oscillationsterm·def. Intermingled oscillationsterm of different frequencies, especially when the phase of the slower oscillationterm modulates the amplitude of the faster oscillationterm. For instance, slow, medium and fast gammaterm bursts tend to occur at different phases of thetaterm (phase-amplitude modulation).

Nested sequencesterm·def. As a rat explores an environment, place cellterm assemblies activate in sequence and thus form “behavioral” (slow time scale) sequences representing the ongoing trajectoryterm in real time. Simultaneously, the same assemblies form thetaterm (fast time scale) sequences that describe the same trajectoryterm, nested within the behavioral sequences.

1 Incidentally, in its “uncorrected” version, this modelterm·mth predicted that place cellsterm should fire in regular spatial patterns long before the discovery of grid cells (see e.g., Lengyel et al. ( 2003 ) for an influential modelterm·mth which eliminates repeated fields).

Funding . This work was supported by the Agence Nationale de la Recherche (ANR-15-CE16-0001-02) and Collège de France (CD).