Potential factors influencing replay across CA1 during sharp-wave ripples

Abstract

Sharp-wave ripples are complex neurophysiological events recorded along the trisynaptic hippocampal circuit (i.e. from CA3 to CA1 and the subiculum) during slow-wave sleep and awake states. They arise locally but scale brain-wide to the hippocampal target regions at cortical and subcortical structures. During these events, neuronal firing sequences are replayed retrospectively or prospectively and in the forward or reverse order as defined by experience. They could reflect either pre-configured firing sequences, learned sequences or an option space to inform subsequent decisions. How can different sequences arise during sharp-wave ripples? Emerging data suggest the hippocampal circuit is organized in different loops across the proximal (close to dentate gyrus) and distal (close to entorhinal cortex) axis. These data also disclose a so-far neglected laminar organization of the hippocampal output during sharp-wave events. Here, I discuss whether by incorporating cell-type-specific mechanisms converging on deep and superficial CA1 sublayers along the proximodistal axis, some novel factors influencing the organization of hippocampal sequences could be unveiled.

This article is part of the Theo Murphy meeting issue ‘Memory reactivation: replaying events past, present and future’.

1. Introduction

The hippocampus is crucial for episodic memory, an ability that relies on the formation of a cognitive map [1]. According to this theory, a set of hippocampal pyramidal cells fire selectively to map relationships between locations where events happen. During encoding, firing sequences of hippocampal cells are coordinated by theta (4–12 Hz) and gamma oscillations (40–80 Hz) and activated in a manner ordered by experience [2]. In a given theta cycle, spikes from place cells organize along the past, present and future events [3–7]. This mechanism allows behavioural sequences to be chunked into theta sequences [4,8,9]. Presumably, theta sequences facilitate encoding of information by plasticity mechanisms relying on the relative timing between spikes [10,11].

Offline reinstating of this activity has been traditionally viewed as a form of replaying sequences for their subsequent consolidation via high-frequency induced plasticity [12–14]. Replay was initially described during sleep in association with fast oscillations or ripples, but it was also observed during exploratory pauses [15–17]. Moreover, the content of replay (i.e. the organization and identity of neuronal sequences) was shown to vary substantially across states and behaviour [18]. In the light of this new evidence, it was proposed that replay may serve a wider range of cognitive functions than originally thought [19,20].

Here, I first review recent data on hippocampal replay studied in rodents to illustrate how our conceptions have evolved to accommodate a more complex perspective. In spite of this conceptual shift, we still lack a physiological understanding on how different forms of replay are generated. A major difficulty in updating this view is that our current models have not yet considered the cell-type and region specificity of hippocampal microcircuits. However, emerging data suggest a specific organization of the hippocampus along the dorso-ventral (or septo-temporal), proximodistal (transverse) and deep–superficial (radial) axes [21–26]. I discuss whether, by considering physiological, microcircuit and neuromodulatory factors in the transverse and radial axes of the dorsal hippocampus, different mechanisms may help to explain biases of hippocampal replay.

2. Complexities of retrospective hippocampal replay

Hippocampal replay of neuronal sequences is a rather complex phenomenon. Replay is typically associated with a particular physiological event called the sharp-wave ripple [27]. It consists of short multi-neuronal sequences (approx. 40–100 ms) carrying information about the temporal organization of experience (figure 1). Originally viewed as a basic mechanism underlying memory consolidation during sleep [12], retrospective replay reflects ordered firing from a set of neuronal ensembles that were previously activated by experience [28,29].

Figure 1.

Miscellaneous patterns of replay potentially underlying different processes of memory consolidation, retrieval, planning and imagination. As the animal experiences a succession of events (represented by letters), some hippocampal neurons fire selectivity to build an abstract representation or cognitive map. During periods associated with sleep and immobility, sequences of these ‘place cells’ are co-activated in an orderly manner coordinated by sharp-wave ripples (SPW-ripples). The order of replay reveals the multifactorial influence of brain state and microcircuit physiology, as well as other procedural and cognitive factors. According to use, replay can be retrospective (engaging sequences already experienced) or prospective (engaging sequences ahead in time). According to the order, replay can be forward (in the same order as experienced) or reverse (opposite to experienced). Some sequences are not linked to experience and reflect a sort of preplay of events never seen before. Finally, according to the subject ‘location’ in the event space, replay can be local or remote. (a) Forward retrospective replay occurring remotely during sleep was the first form of replay reported in the literature [28,29]. Later reports showed it is also present locally during awake immobility and exploratory pauses [30] (b) Reverse retrospective replay is more typically present in awake conditions [6,30] and strongly influenced by novelty, reward values or goal-oriented tasks [31]. Forward and reverse replay can be concatenated along several sharp-wave ripples to accommodate extended experience [32]. (c) Forward prospective replay of already experienced neuronal sequences is typically seen before running for a goal or during choice learning [30,33,34]. (d) Preplay depicts sequences never experienced before and is correlated to or predictive of the activity during the future experience [35,36]. Preplay, which can be forward or reverse, is detectable in awake rest and in slow-wave sleep, and it has been proposed to play a major role in rapid encoding of novel information. (Online version in colour.)

The organization of replay was early challenged by the discovery that sequence order is influenced by the brain state [6,30]. During sleep, sequential firing is typically organized forward (i.e. in a similar order to that experienced; figure 1a), but when replayed during awake immobility the order can reverse (figure 1b). Physiologically, different neuromodulatory influences during sleep and awake states help to explain part of this bias [37]. The role of place fields and place field tails suggested a contribution of the animal's current location in the emergence of awake replay [30,38], but remote places were also seen to trigger sequences [39]. The terms centripetal (away from the animal) and centrifugal (towards the animal) were introduced to identify replay from a more egocentric perspective (see [40] for an early reference to the terms).

Additional complexities were noted in these studies. When rodents were trained to run back and forth along a linear track, the forward and reverse orders were dissociated [30]. At the end of the track, when the animal was drinking and immobile, sequences replayed forwardly, while after running and preceding drinking they replayed in reverse order. This quasi-directional effect suggests a higher level of organization emerging from complex associations among different ensembles [41]. In subsequent studies, it was noted that forward and reverse sequences intermixed in long tracks to incorporate the spatial intricacies of the maze [32,42]. Sequence integrity was not uniform, suggesting that extended replay (up to 600 ms) during concatenated sharp-wave ripples may depend specifically on direct inputs from layer III of the entorhinal cortex [43,44].

In general, forward replay appears more frequent than reverse replay [30,32]. However, additional variables can influence the organization of retrospective sequences. Sensory interference (sounds) can bias the content of replay during sleep to reactivate memory traces associated with auditory cues [45,46]. Reverse replay can be modulated by changing rewards [31], a manipulation that also affects the rate of sharp-wave ripples [47]. This brings reverse replay to a unique position to integrate reward-predictive information into specific neuronal sequences. Consistently, during spatial learning, sequences reorganize to represent new goal locations [48] under the influence of dopaminergic signals from the ventral tegmental area (VTA), locus coeruleus and the dorsal raphe nucleus [40,49–51]. Thus, replay and sharp-wave ripples can actually scale to a system level to serve different cognitive abilities in interaction with other brain regions (e.g. the striatum [52,53], visual cortex [54], cingulate cortex [55], prefrontal cortex [56], auditory cortex [46], amygdala [57], mediodorsal thalamus [58]). Accordingly, forward and reverse retrospective replay can contribute to retrieving information more dynamically during awake states [59,60].

3. Prospective hippocampal replay defies the rule

Research on the intriguing nature of replay gained a new impetus with the discovery of prospective coding during sharp-wave ripples (figure 1c). Prospective coding is a more general phenomenon [61,62], but identification of similar processes during ripples challenged the dominant view. Using spatial alternation tasks requiring short-term memory, it was found that forward replay occurring just before making a decision depicted future place cell sequences ahead of the animal [33]. Moreover, increased activation of these ripple-associated neuronal sequences predicted correct choices, whereas during incorrect trials synchrony remained at chance levels. A role of the animal's location in this form of replay was proposed based on the observation that sequences started in place and ended at the goal sites [34]. This form of replay is dependent on the memory for recent experiences and therefore it may reflect an interaction between neuronal processes underlying encoding, retrieving and planning [20].

However, using a complex maze with several bifurcations and detours in a choice task with daily changing contingencies, it was noted that the replay content may not be determined solely by the recency of events [35]. Up to that moment, it was believed that replay was intimately associated with sequences built up by experience (theta sequences) [4,11]. Surprisingly, the new experiments revealed sequences of trajectories never experienced before by the animal [35]. This more complex organization was related to subsequent observations of preplay (figure 1d), a phenomenon by which future place cell sequences can be recorded during sharp-wave ripples preceding exploration [36]. Pre-existing ensembles were proposed to underlie the formation of sequences that can be repurposed for acquiring new experiences while others are generated de novo [63–67]. The existence of these rigid and reconfigurable hippocampal ensembles may be supported by the skewed distribution of relevant physiological parameters such as synaptic weights and firing rate [64,68,69]. Thus, replay may also contribute constructive cognitive roles such as imagination, a hippocampal-dependent ability that can be conceptualized as a mental travel through the cognitive map [20,70].

In the view of all the evidence reviewed so far, replay can be considered a multifactorial event and hippocampal sharp-wave ripples the underlying physiological mechanism. However, our understanding is still fragmented. How can a palette of sequences ordered back and forth in time and space emerge from an apparently unidirectional event (i.e. from CA3 to CA1) such as the sharp-wave ripple?

4. Can replay organize differently along CA1?

Sharp-waves reflect a population synaptic event presumably emerging from the recurrent CA3 microcircuit, whereas ripples can be generated locally in CA1 [27]. During states dominated by a lower cholinergic tone (i.e. immobility and slow-wave sleep), CA3 neurons are released from sustained GABAergic inhibition and population bursts emerge from transient increases in synchrony [71,72]. These synchronous discharges from CA3 pyramidal cells concurrently activate synaptic currents in the apical dendrites of CA1 neurons, resulting in a transient sharp potential at the stratum pyramidale [73]. Together with CA1 pyramidal cells, some GABAergic interneurons discharge during sharp-wave ripples [74]. Thus, dedicated pyramidal–interneuron interactions shape the local expression of ripples [75]. The relative ‘independency’ between the sharp-wave and the ripple presumably allows for the expression of several but coherent events along the proximodistal and septo-temporal axes [76,77]. The term independency refers to different local ripple mechanisms associated with a given sharp-wave, as well as their independent occurrence at different loci. Thus, multiple neuronal sequences can be activated during sharp-waves, reflecting different ensemble patterns of CA3–CA1 neurons [78,79].

Consistently with this view, intracellular recordings of CA1 pyramidal cells typically show depolarizing responses shaped by phasic excitation/inhibition during sharp-wave ripples both in awake and anaesthetized conditions [73,80–82]. However, this picture was challenged by reports of intermixed hyperpolarized responses able to silence some CA1 neurons [83,84]. Morphological reconstruction revealed a specific organization across the deep (closer to oriens) and superficial (closer to the stratum radiatum) sublayers [83]. Part of this selection can be explained by a radial gradient of perisomatic inhibition interacting with behavioural and state-dependent effects to bias neuronal firing during sleep and awake ripples [83]. Other microcircuit factors may equally contribute to sublayer segregation of hippocampal activities [26,83,85].

While investigating the mechanisms for selective reactivation of morphologically identified CA1 cells, some additional differences were noted [86]. Unsupervised clustering of sharp-wave ripple events allows us to evaluate how variability of single-cell firing could be explained by the participation of different ensembles, as captured by extracellular signatures [86,87] (figure 2a). When applied to recordings of morphologically identified cells, it was noted that deep cells participated less than superficial cells [83,86] and that their spike timing along sharp-waves was more variable (figure 2b,c). For instance, the cell shown in figure 2b fired earlier during sharp-wave ripples in the orange cluster and later for ripples in the blue cluster. This wider range of firing variability may suggest more flexibility for forward and reverse replay in deep than superficial cells in a sequence (figure 2d).

Figure 2.

Flexibility of neuronal firing from CA1 pyramidal cells during sharp-wave ripples (SPW-ripples). (a) Unsupervised classification of individual sharp-wave ripple events clusters of similar local field potential signatures to be identified. The number of events in each group is indicated. Clusters of groups with topological similarities in the high-dimensional space are identified by colours in the inset scheme. (b) Single CA1 pyramidal cells were recorded juxtacellularly during sharp-wave ripples and their firing was grouped per cluster. Note different timing of the same cell in the orange and blue cluster indicated before. (c) Juxtacellular recorded cells are labelled for morphological identification with streptavidin, and classified as deep or superficial depending on their location within the calbindin-positive sublayer. (d) Distribution of the preferred firing phase during the sharp-wave from different clusters in deep and superficial cells. Note wider distribution in deep cells indicating more flexibility. Data from [86]. (Online version in colour.)

These data suggest that ripple-associated replay may proceed differently across CA1 sublayers. A radial organization of CA1 neuronal firing was early noted [88]. Deep CA1 cells fire at higher rates and have broader place fields as compared with superficial cells [88,89]. The nature of these fields is also qualitatively different: deep cells are more tightly linked to somatosensory landmarks and influenced by rewards, while superficial cell firing is more contextual [89,90]. Superficial place fields apparently provide a more stable representation of a given context while deep cells are more flexible. Their ability to form theta sequences is strongly determined by different phase precession dynamics: deep cells exhibit a wider phase range than superficial cells and can even shift phases during rapid eye movement (REM) sleep [88,91]. Therefore, physiological mechanisms may be in place to influence neuronal activity across CA1; whether they could explain the variety of hippocampal replay is unknown.

5. Potential biases of hippocampal replay across deep–superficial sublayers

By considering genetic, microcircuit and behavioural factors, different mechanisms could be proposed to bias replay. First, cell-type and region-specific microcircuits wire differentially along sublayers (figure 3a). Deep cells at CA1 regions closest to CA2 (proximal CA1) receive more inputs from the medial entorhinal cortex, whereas superficial cells located closer to subiculum (distal CA1) are better connected to the lateral entorhinal cortex [99]. Therefore, the way sensory modalities integrate into the CA1 region should be determined by this dedicated wiring. Consistently, non-spatial and spatial information segregates radially [89,90,94] and proximodistally in CA1 [100–102]. Intrahippocampal circuits also wire differently across sublayers. Superficial cells are more responsive to CA3 inputs than deep cells owing to differences of feed-forward inhibition [83], which together with the proximodistal organization of Schaffer collaterals will determine how contextual information from dentate gyrus (DG) and CA3 enters into CA1 [103–105]. Deep cells, in contrast, are more strongly activated by CA2 [106,107], which may help to accommodate other cognitive representations such as social memory and delay signals during immobility [108–111].

Figure 3.

Biases of hippocampal replay can affect deep and superficial (sup.) CA1 pyramidal cells differently. (a) Known deep superficial local microcircuit motifs. Deep cells are more strongly activated by inputs from CA2 (at basal dendrites) and the medial entorhinal cortex. Inputs from CA3 cells onto deep cells are strongly interfaced by feed-forward inhibition. Superficial cells receive more innervation from the lateral entorhinal cortex and direct CA3 inputs. Importantly, lateral and medial entorhinal inputs to deep and superficial cells organize proximodistally along the traverse CA1 axis. Superficial cells mainly recruit PV+ basket cells whereas deep cells are biased for SOM+ interneurons. In return, innervation by PV+ basket cells is larger over deep cells while CCK+ basket cells preferentially target superficial cells. (b) Differential transcriptomic expression of serotoninergic (Htr) and cholinergic receptor genes (Chrn) along the deep and superficial layers. Normalized gene expression values from three different replicates are shown. Data from https://hipposeq.janelia.org/ [92]. (c) Multifactorial axes biasing the content and organization of replays during sharp-wave ripples. Different influences on deep and superficial CA1 pyramidal cells have been reported along these axes (grey boxes). The relative axis length is not necessarily informative [6,30,34,36–38,40,43,45–51,57,58,68,78,88–91,93–98]. PC, pyramidal cell, SOM+, somatostatin-positive interneuron; PV+, parvalbumin-positive basket cell; CCK+, cholecystokinin-positive basket cell; MECIII, medial entorhinal cortex layer III; LECIII, lateral entorhinal cortex layer III. (Online version in colour.)

Different innervation by local GABAergic interneurons also shapes the dynamics of CA1 cells [112], especially across behavioural states [113–115]. Parvalbumin (PV) and cholecystokinin (CCK) basket cells target deep and superficial cells differently [83,116] so that their feed-forward and feedback activation can gate information differently across sublayers [91,116–118]. In addition, transcriptomic differences regulate the laminar expression of neuromodulatory receptors such as G-protein coupled receptor 5-HT1a for serotonin (enriched at superficial cells) or the nicotine receptor for acetylcholine (enriched in the deep sublayer) [92] (figure 3b).

Therefore, data support integration of distinct physiological influences across CA1. Under this scenario, it is tempting to consider whether these mechanisms could provide explanatory axes for a variety of replays (figure 3c). For instance, state (including brief exploratory pauses) has a major influence [30,119], and this is supported by the physiology of sharp-wave ripples in response to different neuromodulators [120]. Sensory inputs [45,46] as well as the animal's location can bias the replay content, a property that presumably depends on entorhinal inputs segregated along CA1 [99]. Accordingly, co-activations of neuronal sequences in entorhinal layers can independently control the input/output flow of the hippocampal replay [95,121]. Novelty and reward can reverse replay in time [6,47] owing to the influence of the VTA, nucleus accumbens and medial prefrontal cortex [40,49,122,123]. Aversive inputs from the amygdala can also affect hippocampal replay [51,57].

Along many of these axes, a deep–superficial organization of neuronal firing has been described and can therefore influence sequence dynamics (figure 3c; grey boxes). State-dependent shifts of the preferred theta phase and wider precession dynamics are reported for deep CA1 cells [88,91]. Place field of deep cells is more typically linked with somatosensory and olfactory inputs [89,94] while superficial maps are more contextual and allocentric [90]. The activity of deep cells is more affected by goals, rewards and anxiety [90,96]. Given that sharp-wave ripple physiology is influenced by all these factors [27,124,125], I hypothesize that the deep–superficial hippocampal axis may contribute to shape replay. Are there microcircuit substrates for such an effect?

6. A proximodistal and deep–superficial perspective of hippocampal replay

According to the classical view, sharp-wave ripples represent a complex event built from two different processes [27]. Sharp-waves result from convergent depolarization of the dendrites of CA1 pyramidal cells in response to CA3 firing distributed through Schaffer collaterals [78]. By contrast, ripples emerge locally in CA1 from feedback interactions between pyramidal cells and interneurons [75]. It is believed that the recruitment process starts more typically at the distal CA3a region (close to CA2) and runs as a neuronal avalanche towards CA1.

However, sharp-wave activity can also backpropagate to the proximal CA3c area (close to DG) before invading CA1 [71,78]. Driven by CA3c firing, mossy cells from the hilus are recruited and integrated into the flow [126]. Moreover, recent evidence suggests that some CA2 cells may independently support the generation of sharp-wave ripples markedly during awake states [97], possibly owing to recurrent connectivity [127] and in interaction with a specific subset of CA3a cells [128]. Therefore, different initiating regions distribute along the hilus–CA3–CA2 axis to trigger local ripples in CA1 [105,124,97,129]. If diversity of replay can be explained by any physiological mechanism, the multiple sharp-wave initiating loci must be one key. In support of this view, individual ripple events vary along CA1 [77]. Such high-dimensional dynamics can be disentangled with appropriate methods ([86,87,130,131]; see also figure 2).

I propose bringing the focus to models of sharp-wave ripples consisting of multiple initiating loci instead of considering a single triggering area. By considering different spots along upstream pre-synaptic regions, the activity can flow differently in CA1 to control the replay order more precisely by discharging cells at different timing (figure 4). Under this perspective, the proximodistal hilus–CA3–CA2–CA1 axis represents a primary source of modulation, but entorhinal inputs as well [43,132]. Adherence of individual cells to particular ripple clusters (i.e. replay content) reflects that the underlying orthogonalization may also segregate across CA1 sublayers [75,83,86]. Therefore, the proximodistal organization of pre-synaptic ensembles interacts radially in CA1. For instance, sharp-wave events initiated by CA3c cells would more likely result in the feed-forward inhibition of CA2 cells [83,133] and consequently would discharge mostly superficial CA1 cells. By contrast, sharp-wave ripples initiated by CA3a–CA2 pyramidal cells could rather flow through deep CA1 sublayers first (figure 4). Connectivity between CA3 cells, hilar mossy cells and dentate granule cells should further contribute to shaping firing content during these events by incorporating additional sequences and influencing reactivation [105,126,129]. Selection of different sequences can be triggered by single cells [71,134], which together with plasticity can help to reconfigure and to establish new ensembles [135]. Recent evidence supports the idea that such a process may indeed occur randomly [66].

Figure 4.

Potential mechanisms for sequence orthogonalization across deep and superficial CA1 sublayers. Different responsiveness of deep and superficial (sup.) cells to a collection of pre-synaptic inputs from CA2 to CA3 regions may support different timing for activation in a sequence, which together with dedicated inhibitory control will determine firing selection across CA1 sublayers. Entorhinal inputs converging differently in deep and superficial cells across the proximodistal axis could additionally bias firing depending on the animal location and salient sensory information. State and emotional factors represent additional orthogonalizing factors.

Given the ‘independency’ of ripples from the sharp-wave event, local CA1 circuits could also contribute to the way sequences can be replayed. While recurrent connectivity between CA1 pyramidal cells is very low, di-synaptic inhibitory circuits interface between sublayers [25,26]. Thus, driven by specific spatio-temporal patterns of GABAergic inhibition the content and order of replay could further unfold [74,136,137]. Interneuronal interactions, but possibly GABAergic projecting cells as well [115,118,138], could organize spatially separated ensembles into coherent sequences [75]. For example, superficial CA1 cells preferentially activate PV+ basket cells while deep cells apparently do so with SOM+ interneurons [91,116]. Therefore, cell-type-specific local ripple generators can provide additional mechanisms for orthogonalization (figure 4). Indeed, optogenetic manipulations able to induce artificial CA1 ripples suggest that deep and superficial sublayers may actually form distinct ensembles [139].

Computational models of replay support the existence of multiple attractors resulting from different buffers of activity [5,140]. In some of these models, direct inputs bypassing the intrahippocampal circuit are required to favour encoding of reversed association [5,141]. In most cases, the experience is critical in establishing the organization of auto-associative ensembles and thus the order and content of replay. While retrospective and prospective firing can be simulated with these premises, evidence of preplay has questioned whether sequence reactivation during sharp-wave ripples requires encoding or rather it may result from existing ensembles [67,142]. Pre-determined cell-type-specific connectivity can favour pre-configured attractors. For instance, a single CA3 pyramidal cell typically gives rise to a number of collaterals projecting in different directions onto its neighbouring cells [143–145], with the pattern of connectivity depending on the location of the pre-synaptic soma [146,147]. Novel synapse labelling methods have revealed Schaffer collateral connections with CA1 are enriched between neurons sharing similar developmental periods [148]. Strikingly, CA3 connectivity with early developed deep CA1 pyramidal cells appears more highly structured and non-random than with late developed superficial cells [149]. This suggests there is a level of determinism that provides additional functional constraints. The way in which pre-configured and learned sequences coexist and interact remains to be understood.

Overrepresentation of reward, goal and aversive contexts bias replay for specific events [48,51,150–153]. The hippocampal sharp-wave ripple generator interacts brain-wide to hierarchically orchestrate these representations. New evidence supports that hippocampal–amygdala and hippocampal–hypothalamic interactions are more likely to involve subpopulations of deep cells [96,116] and that deep CA1 cells are more influenced by goals and rewards [90]. Similarly, state-dependent changes may affect differently deep and superficial layers given their different innervation by some GABAergic interneurons [113] and gradient expression of receptors for some neuromodulatory transmitters [154]. Consistently, deep cells phase shift during theta oscillations associated with REM sleep [88]. Therefore, different behavioural and emotional states could also contribute to orthogonalize replay (figure 4).

7. Conclusion

Understanding replay is challenging. Most of the work done so far has adopted conceptually separated views. On the one hand, sequence analysis has enabled identifying the organization of neuronal ensembles during replay and using them to decode neuronal representations. On the other hand, a physiological perspective has helped to pinpoint microcircuit mechanisms responsible for the basic field potential signatures of sharp-wave ripples. We need to fill the gap between the mechanisms and function of these events.

To better understand the complexities of replay, an updated view of sharp-wave ripple physiology has to incorporate the critical influence of cell-type-specific subcircuits that wire brain-wide the proximodistal and radial axes of CA1. We then need to exploit this specificity to evaluate how replay dynamics can be precisely controlled. Here, I have reviewed the evidence and discussed recent data supporting that firing of deep and superficial CA1 cells can be influenced differently during sharp-wave ripples. I suggest that cell-type- and input-specific connectivity together with radial expression of receptors and intrinsic properties may provide substrates for biasing hippocampal replay back and forth. How these variables can specifically affect the content and order of replay remains to be examined.

I propose relying on these mechanisms to force conceptual shifts regarding our understanding of the way replay is established and used to guide behaviour. Based on this review and emerging cell-type and region-specific data the following research questions and hypotheses can be addressed:

• —Are sharp-wave ripples multifocal? I hypothesize events may arise from different loci and engage different sets of neuronal sequences along the septo-temporal, proximodistal and deep–superficial axes of the hippocampus.
• —What is the role of cortical and subcortical inputs? Based on the different connectivity with deep and superficial pyramidal cells, contrasting and dynamic effects across CA1 sublayers can be predicted.
• —What determines the direction of propagation of sharp-wave ripples and how does it relate with replay order and content? I hypothesize that the generating loci, as well as participating GABAergic interneurons and dynamically fluctuating inputs, will bias replay in a predictable way.
• —To what extent can experience modify replay of deep and superficial subcircuits? Given their different developmental origin I predict different connectivity and plasticity rules determining preplay ability across sublayers.
• —Can manipulations of the replay content differently affect cognition? Given different brain-wide connectivity of deep and superficial cells, specific behavioural effects can be predicted.

The advent of super-resolution optoelectrodes and multicellular voltage imaging [139,155–158], in combination with single-cell transcriptomic public atlases and gene editing techniques [92,154,159–161], will certainly many of these questions to be addressed at unprecedented levels of detail in forthcoming years [162–164].

Data accessibility

This article has no additional data.

Competing interests

I declare no competing interests.

Funding

My research on sharp-wave ripples is currently supported by the Spanish Ministry of Science, Innovation and Universities (grant no. RTI2018-098581-B-I00).