The two tales of hippocampal sharp-wave ripple content: the rigid and the plastic

Abstract

Sharp-wave ripples, prominently in the CA1 region of the hippocampus, are short oscillatory events accompanied by bursts of neural firing. Ripples and associated hippocampal place cell sequences communicate with cortical ensembles during slow-wave sleep, which has been shown to be critical for systems consolidation of episodic memories. This consolidation is not limited to a newly formed memory trace; instead, ripples appear to reactivate and consolidate memories spanning various experiences. Despite this broad spanning influence, ripples remain capable of producing precise memories. The underlying mechanisms that enable ripples to consolidate memories broadly and with specificity across experiences remain unknown. In this review, we discuss data that uncovers circuit-level processes that generate ripples and influence their characteristics during consolidation. Based on current knowledge, we propose that memory emerges from the integration of two parallel consolidation pathways in CA1: the rigid and plastic pathways. The rigid pathway generates ripples stochastically, providing a backbone upon which dynamic plastic pathway inputs carrying novel information are integrated.

1. Introduction

The hippocampus is the center for episodic memory formation. While it performs crucial roles in encoding and consolidation of memories, it arguably has a limited role in long-term storage of memory, particularly the schematic form of memory that lacks contextual details of the original experience (Squire 1992, Squire and Bayley 2007, Winocur, Moscovitch et al. 2010, Preston and Eichenbaum 2013). Instead, long-term schematic memories are thought to be stored in distributed networks of cortical neuronal ensembles – small subsets of neurons that encode components of memory traces (Josselyn, Kohler et al. 2015, Tonegawa, Liu et al. 2015). Hippocampal reactivation is necessary to transform hippocampal dependent memories into cortical neural networks through its active role in a process called systems consolidation (Winocur, Moscovitch et al. 2010, Born and Wilhelm 2012). Systems consolidation involves coordinated, network-wide reactivation of neuronal ensembles during slow-wave sleep (SWS), which strengthens the synaptic connections between neuronal ensembles to allow for long-term storage of memory (Stickgold 2005, Klinzing, Niethard et al. 2019). Despite this understanding, the mechanisms underlying hippocampal reactivation during systems consolidation remain unclear.

The discovery of sharp-wave ripples, neural oscillations most prominent in hippocampal CA1 during SWS and immobility, have begun to uncover these mechanisms (Buzsáki, Horváth et al. 1992, Buzsaki 2015). Sharp-wave ripple is a complex field potential pattern composed of two interdependent and temporally linked events, the sharp wave and ripple. Sharp waves reflect large excitatory inputs to CA1 apical dendrites primarily in the stratum radiatum; and ripples are brief, 110–180 Hz oscillations that reflect alternating perisomatic inhibitions and synchronous spikes in the stratum pyramidale (Liu, Henin et al. 2022). Ripples accompany the reactivation of firing sequences of CA1 pyramidal neurons that occurred during wakefulness, a phenomenon commonly referred to as replay (Lee and Wilson 2002, Foster and Wilson 2006, Diba and Buzsáki 2007, Davidson, Kloosterman et al. 2009). This ripple-associated replay of wakefulness neural sequences is believed to consolidate hippocampal-dependent experiences into long-term storage in the cortex. In support, prolonging ripple duration or enhancing ripple coupling with cortical delta waves and spindles post-learning has been shown to improve memory (Maingret, Girardeau et al. 2016, Fernández-Ruiz, Oliva et al. 2019). Conversely, disruption of ripple and associated activity post-learning leads to memory impairments (Girardeau, Benchenane et al. 2009, Ego-Stengel and Wilson 2010, Jadhav, Kemere et al. 2012, Wang, Yau et al. 2015, Gridchyn, Schoenenberger et al. 2020). These findings, along with other studies that found pronounced hippocampal-cortical communication during ripples, provide compelling evidence supporting the important role of hippocampal ripples in systems consolidation (Siapas and Wilson 1998, Sirota, Csicsvari et al. 2003, Isomura, Sirota et al. 2006, Peyrache, Khamassi et al. 2009).

Ripple content – the specific set of CA1 neurons that fire during a ripple event – varies depending on which memory traces or trajectories are being reactivated (Karlsson and Frank 2009, Gupta, van der Meer et al. 2010, Pfeiffer and Foster 2013, Gillespie, Astudillo Maya et al. 2021). Memory specificity thus appears to emerge from the unique temporal arrangements of CA1 neuronal activity within different ripple content during consolidation. It is still unknown, however, how individual CA1 neurons are selectively recruited and incorporated into ripples and how ripple content is amended following learning. In this review, we aim to elucidate the mechanism by which ripple content evolves following learning and how this contributes to systems consolidation. We propose the existence of a dual-pathway framework wherein ripples are generated through a stochastic rigid pathway that encodes backbone information through preconfigured neural sequence motifs, while novel and salient experiences are incorporated into these motifs through a dynamic plastic pathway.

2. Stochastic generation of sharp-wave ripples

2.1. Sharp-wave ripples are generated within the CA3-CA1 network

A main feature of ripples is their routine generation during SWS and immobility regardless of the presence or absence of immediate prior experiences (Ylinen, Bragin et al. 1995, Dragoi and Tonegawa 2011). Researchers have characterized in vitro spontaneous generation of ripple events in hippocampal CA1 in the absence of extrahippocampal inputs, suggesting that a local circuit within the hippocampus is self-sufficient in generating ripples (Behrens, van den Boom et al. 2005). Excitatory pyramidal neurons in the CA3 of the hippocampus likely drive the generation of ripples, as they send dense projections to CA1 pyramidal neurons and are necessary for inducing ripple events during SWS and immobility (Nakashiba, Buhl et al. 2009, Buzsaki 2015). Importantly, acute optogenetic silencing of CA3 terminals in the CA1 drastically suppresses ripple occurrence in vivo, providing direct evidence supporting CA3’s key role in ripple generation (Davoudi and Foster 2019).

During SWS there is baseline tonic activity within CA3, likely due to a combination of intrinsic properties of CA3 neurons and external inputs such as cortical Up-state inputs (Hahn, McFarland et al. 2012, Schlingloff, Kali et al. 2014). This tonic activity leads to the accumulation of excitatory post-synaptic currents (EPSCs) across reciprocally connected CA3 neurons, and thus gradually raises their membrane potential to a depolarization threshold, which results in stochastic population-level activation (Behrens, van den Boom et al. 2005, de la Prida, Huberfeld et al. 2006, Schlingloff, Kali et al. 2014). At the molecular level, this gradual buildup of depolarization is mediated by the AMPA and kainate glutamate receptors, as blockade of them eliminates or reduces ripple generation in vitro (Maier, Nimmrich et al. 2003, Colgin, Kubota et al. 2004, Behrens, van den Boom et al. 2005). Further supporting CA3’s role in ripple generation, researchers found that EPSCs of CA3 neurons accumulate exponentially ~50 ms before ripple onset in vitro (Behrens, van den Boom et al. 2005, Schlingloff, Kali et al. 2014). In summary, CA3 EPSC accumulation results from the buildup of baseline tonic excitatory inputs. Recurrent excitation in CA3, therefore, allows for the conversion of incremental EPSCs into propagating population-wide excitability of CA3 neurons. Once the depolarization threshold is reached, CA3 population excitability activates CA1 neurons, intermittently triggering the generation of sharp-wave ripples. Notably, this is largely a stochastic process due to the gradual buildup of EPSCs in CA3 neurons.

Although CA3 is thought to be the primary driver of ripple generation, additional hippocampal subregions, such as CA2 and the dentate gyrus, also facilitate ripple generation. It has been shown that optimized optogenetic activation of the CA2 triggers ripple generation and facilitates social memory (Oliva, Fernández-Ruiz et al. 2020). Additionally, the dentate gyrus is necessary for CA3-mediated generation of ripples that support goal-directed behavior and memory (Sasaki, Piatti et al. 2018). More recently, the subiculum has been implicated as another brain region to generate ripple that can backpropagate to the CA1 (Imbrosci, Nitzan et al. 2021). Despite the complexity of neural circuits, CA3 remains the major contributor in sharp-wave ripple generation (Liu, Henin et al. 2022).

2.2. Ripple generation is regulated by interneurons and subcortical inputs

Ripples are not continuous; instead, they occur at variable intervals throughout SWS and immobility. Following a ripple event, there is a short refractory period where additional ripples are not generated due to hyperpolarization (Buzsaki 2015). This refractory period outlasts the refractory period of individual CA1 or CA3 neurons, suggesting that an additional mechanism is responsible for regulating the process of ripple generation (Patel 2015). Interneurons emerge as important candidates due to their known ability to drive oscillatory patterns in the brain (Freund and Buzsaki 1996, Buzsaki 2015). In particular, CA1 parvalbumin (PV) interneurons play a key role in the generation of ripples: selective activation of PV interneurons triggers ripple generation, whereas suppression of PV interneurons perturbs ripple oscillation and impairs memory (Schlingloff, Kali et al. 2014, Gan, Weng et al. 2017, Ognjanovski, Schaeffer et al. 2017). Other types of interneurons, notably the axo-axonic and oriens-lacunosum/moleculare (OLM) neurons, may also regulate ripple generation. These axo-axonic and OLM interneurons turn silent during ripples, raising the possibility that their activation suppresses ripple generation (Klausberger and Somogyi 2008). Together, CA3 recurrent excitation likely triggers coordinated activation of multiple CA1 interneuron subtypes that can initiate, pause, or terminate ripple activity.

Subcortical inputs to the CA1, including projections from the median raphe and medial septum, may also play a role in regulating stochastic ripple generation. In fact, optogenetic activation of the median raphe or septal cholinergic neurons drastically suppresses ripple occurrence, indicating an extra-hippocampal regulation of ripple generation (Vandecasteele, Varga et al. 2014, Wang, Yau et al. 2015). Septal cholinergic neurons exhibit state-dependent activity with maximal activation during active wakefulness and paradoxical sleep, yet almost complete silence during SWS (Lee, Hassani et al. 2005). These cholinergic inputs act on presynaptic cholinergic/muscarinergic receptors, leading to a suppression of CA3 recurrent excitation (Buzsaki 2015). Therefore, septal cholinergic activity may block ripple generation altogether during active wakefulness and paradoxical sleep. During SWS, the median raphe dominates the regulation of ripple generation: its activation drastically decreases ripple occurrence, whereas its suppression increases ripple occurrence (Wang, Yau et al. 2015). This effect on ripple generation is largely mediated by a subset of median raphe neurons (Wang, Yau et al. 2015), which exclusively target hippocampal non-PV interneurons (Miettinen and Freund 1992, Szonyi, Mayer et al. 2016). This subset of raphe neurons was confirmed as Vglut3 neurons in a later study (Huang, Ikemoto et al. 2022). Notably, these Vglut3 neurons decrease their activity prior to ripple occurrence but increase their activity at the termination of ripple events, which likely suppresses the immediate occurrence of another ripple event (Wang, Yau et al. 2015). Therefore, raphe Vglut3 activity appears to separate adjacent ripple events by imposing a brief silent period between events. This potentially prevents distinct ripple contents from intermixing during consolidation. Together, these findings provide direct evidence that the raphe and septal projections to the CA1 regulate the stochastic generation of ripples in a state-dependent manner.

3. Rigid and plastic CA1 neurons

Growing evidence suggests that functional diversity in ripple content arises from specific neuron types in the CA1 (Mizuseki, Diba et al. 2011, Danielson, Zaremba et al. 2016, Geiller, Fattahi et al. 2017, Li, Xu et al. 2017, Masurkar, Srinivas et al. 2017, Sun, Sotayo et al. 2017, Sharif, Tayebi et al. 2021). Two distinct populations of CA1 pyramidal neurons have been identified based on their firing properties during ripples, designated as rigid and plastic neurons. Rigid neurons tend to activate during ripples, and their activity exhibits limited change between pre- and post-learning sleep (Grosmark and Buzsaki 2016, Chenani, Sabariego et al. 2019). Therefore, rigid neurons appear to function as a ripple backbone that is not modified by new experiences. We speculate that the ripple backbone consists of a group of functionally connected neurons that represent either an internally generated sequence or a shared architecture of multiple traversed paths from prior experiences. By contrast, plastic neurons are adaptable and exhibit dynamic changes following novel experiences. These neurons not only alter their firing activity from pre- to post-learning sleep but may also undergo group reorganization, indicating their importance in encoding newly acquired memories (Grosmark and Buzsaki 2016, Liu, Sibille et al. 2018, Chenani, Sabariego et al. 2019). These findings support our theoretical two-module framework of ripple-mediated memory, where rigid neurons represent a stable, unchanging, ripple backbone that serves as a foundation for incorporating adaptive plastic neurons.

In addition to the rigid/plastic classification of CA1 pyramidal neurons, which relies on firing patterns, there is also classification system based on anatomical differences: CA1sup and CA1deep neurons, located in the superficial and deep pyramidal sublayers, respectively (Mizuseki, Diba et al. 2011). CA1sup and CA1deep neurons, although not entirely analogous to rigid and plastic neurons, overlap greatly. CA1sup neurons tend to exhibit rigid-like characteristics, with more stable firing rates and fewer place fields, whereas CA1deep neurons tend to exhibit plastic-like characteristics, with less stable firing rates, more place fields, and stronger modulation by memory demand (Mizuseki, Diba et al. 2011, Danielson, Zaremba et al. 2016, Geiller, Fattahi et al. 2017, Harvey, Robinson et al. 2022). Furthermore, CA1deep neurons are strongly activated by spatial landmarks, enriched cues, and sensory inputs during learning, whereas CA1sup neurons show little response to the same stimuli (Danielson, Zaremba et al. 2016, Geiller, Fattahi et al. 2017, Sharif, Tayebi et al. 2021). Importantly, CA1deep neuronal activity has been shown to correlate with success rate during learning. Researchers found that activity of CA1deep, but not CA1sup neurons, corresponds with reward zones in a goal learning task and predicts task performance, indicating a role of CA1deep neurons in learning (Danielson, Zaremba et al. 2016). These findings suggest that CA1sup neurons demonstrate rigid-like dynamics that are not modified by experiences, while CA1deep neurons exhibit a plastic-like phenotype that is important for new memory formation.

4. Rigid and plastic pathways emerging from CA1 sublayer differences

4.1. Rigid and plastic pathways

At the circuit level, CA1sup and CA1deep neurons receive distinct inputs, potentially contributing to their different characteristics and functions (Valero, Cid et al. 2015, Grosmark and Buzsaki 2016, Valero, Averkin et al. 2017). CA1sup neurons are more responsive to CA3 inputs than CA1deep neurons, whereas CA1deep neurons are more responsive to CA2 inputs (Kohara, Pignatelli et al. 2014, Valero, Cid et al. 2015, Grosmark and Buzsaki 2016, Valero, Averkin et al. 2017, Fernández-Ruiz, Oliva et al. 2019). Moreover, CA3 neurons are most active in response to optogenetic stimulation of the dentate gyrus (DG), while CA2 neurons are active in response to optogenetic stimulation of either the DG or entorhinal cortex (Kohara, Pignatelli et al. 2014, Sun, Sotayo et al. 2017). These findings lead us to characterize two distinct pathways, the DG→CA3→CA1sup (rigid pathway) and entorhinal/CA2→CA1deep (plastic pathway), which influence different sublayers of CA1 neurons (Figure 1).

Figure 1. Two major neural pathways for ripple-mediated memory consolidation.

Red, plastic pathway: Neurons originating from the cortex, mostly through the entorhinal cortex, either directly target CA1deep neurons or indirectly via CA2 neurons. Red dashed line, DG projects to the CA2, which in turn projects to CA1deep neurons, representing another likely plastic pathway. Blue, rigid pathway: Neurons originating from the cortex, through the entorhinal cortex, project to the DG which in turn projects to the CA3. Neurons in the CA3 form Schaffer collaterals and send excitatory projections to CA1 pyramidal and interneurons. PV interneurons preferentially receive CA1sup neuronal inputs and synapse onto CA1deep neurons. As a net effect, CA3 outputs activate CA1sup neurons but suppress CA1deep neurons. Social memory, dorsal CA2 neurons project to ventral CA1 which conveys social information.

The existence of rigid and plastic pathways raises the possibility that hippocampal ripple contents are influenced by their crosstalk. Indeed, a differential CA3 influence on CA1 sublayers leads to opposing effects between the rigid and plastic pathways. Although the CA3 projects comparably to CA1 deep and superficial sublayers, the feedforward inhibition via PV interneurons preferentially targets the deep layer, leading to a greater inhibition of CA1deep neurons compared to their superficial counterparts (Mizuseki, Diba et al. 2011, Kohara, Pignatelli et al. 2014, Lee, Marchionni et al. 2014, Stark, Roux et al. 2014, Valero, Cid et al. 2015). Moreover, CA1deep neurons can be further inhibited by CA1sup neurons which synapse on PV interneurons (Lee, Marchionni et al. 2014). These connectivity patterns manifest in increased hyperpolarization and, thus, decreased activation of CA1deep neurons during ripples. Therefore, CA3-induced ripple generation is predisposed to excite CA1sup neurons but suppress CA1deep neurons. This may reflect a mechanism which, at baseline, facilitates the rigid pathway activation while restricting integration of the plastic pathway into ongoing ripples when memory saliency is low.

4.2. Cortical control of the plastic pathway

CA1deep neurons, therefore, must rely on additional excitatory inputs, external to that of CA3, to overcome PV interneuron-mediated inhibition. Indeed, CA2 drives activation of CA1deep neurons; however, these CA2 neurons are hyperpolarized by CA3 inputs via feedforward inhibition (Kohara, Pignatelli et al. 2014, Valero, Cid et al. 2015, Fernandez-Lamo, Gomez-Dominguez et al. 2019). Therefore, only cases in which strong excitatory input to the CA2 can circumvent CA3-mediated inhibition will CA1deep neurons become activated. Outside of CA2, the medial and lateral entorhinal cortices, which have dense cortical connectivity, act as another mechanism to influence CA1deep neurons. The medial entorhinal cortex not only excites CA2, but also directly synapses onto CA1deep neurons (Kohara, Pignatelli et al. 2014, Masurkar, Srinivas et al. 2017). Lesion of the medial entorhinal cortex affects the activity of plastic but not rigid CA1 neurons, further indicating a selective influence on CA1deep neurons (Chenani, Sabariego et al. 2019). In parallel, the lateral entorhinal cortex projects to the CA2, which in turn targets ventral CA1 neurons, forming a separate circuitry that conveys social information required for the formation of social memories (Kohara, Pignatelli et al. 2014, Meira, Leroy et al. 2018, Lopez-Rojas, de Solis et al. 2022). Although the CA2 projects non-selectively to ventral CA1 sublayers, only the deep sublayer seems to mediate social memory via its projection to the nucleus accumbens and perhaps medial prefrontal cortex (Okuyama, Kitamura et al. 2016, Meira, Leroy et al. 2018, Xing, Mack et al. 2021). Together, CA2 and the entorhinal cortices potentially serve as relays, enabling broad cortical influence to circumvent CA3 baseline inhibition for direct and selective excitation of CA1deep neurons during ripples.

One possible mechanism by which cortical drive is strengthened to initiate plastic pathway activation is through learning-induced long-term potentiation (LTP) and cortical tagging (Teyler and DiScenna 1987, Lesburguères, Gobbo et al. 2011, Girardeau, Cei et al. 2014). In essence, cortical tagging posits that molecular and synaptic changes of cortical and connected neurons during learning form the foundation for future reactivation (Lesburguères, Gobbo et al. 2011, Girardeau, Cei et al. 2014). In support, it has been shown that selective lesion of cortical inputs to the hippocampus disrupts memory consolidation (Remondes and Schuman 2004). Following learning, the rate and length of ripples are increased, indicating a higher frequency of memory reactivation, as well as an incorporation of additional CA1 neural activity into ripple contents (Girardeau, Cei et al. 2014, Fernández-Ruiz, Oliva et al. 2019). These learning-induced ripple changes are reliant upon LTP, as blocking LTP prevents ripple rate alterations (Girardeau, Cei et al. 2014). These alterations most likely reflect specific changes in the recruitment of CA1deep/plastic but not CA1sup/rigid neurons because the latter exhibit little response to learning (Danielson, Zaremba et al. 2016, Grosmark and Buzsaki 2016). Therefore, LTP and cortical tagging during learning likely represent the fundamental elements that can later drive plastic pathway activity during ripples.

Direct evidence of a cortical influence on ripple content comes from a recent study that performed dual-site recording from the CA1 and auditory cortex (Rothschild, Eban et al. 2017). They found that selective auditory cortical neurons, following an auditory association learning paradigm, became reactivated during SWS and that this activation can predict CA1 firing during ripples (sublayer difference in the CA1 was not examined in this study). Although the auditory cortex does not project directly to CA1, it could influence the CA1 plastic pathway indirectly via its connection with the entorhinal cortex. Other sensory cortices, including the olfactory, visual, and somatosensory cortices, likely also influence ripple contents, given their implied roles in memory consolidation (Sirota, Csicsvari et al. 2003, Ji and Wilson 2007, Rasch, Buchel et al. 2007, Rudoy, Voss et al. 2009, Bendor and Wilson 2012). These findings, along with studies that show increased recruitment of CA1deep/plastic neurons during ripples post-learning, suggest that learning-induced cortical tagging is important for integrating plastic pathway activity onto preconfigured rigid frameworks.

5. Integration of rigid and plastic pathways

The rigid pathway, thus far, has been described as a stochastic framework that plastic information is built upon, but what exactly is the rigid pathway encoding? During ripples, hippocampal place cells replay sequences in a manner similar to how they were activated during wakefulness (Lee and Wilson 2002, Foster and Wilson 2006, Diba and Buzsáki 2007, Davidson, Kloosterman et al. 2009). Recently, this concept has been expanded with studies showing that ripple can replay sequences either in forward or reverse direction (Foster and Wilson 2006, Diba and Buzsáki 2007). In studies investigating ripple contents during awake immobility, researchers identified CA1 sequences that may even predict future paths in goal-directed behavioral paradigms (Pfeiffer and Foster 2013, Gillespie, Astudillo Maya et al. 2021). Most unexpectedly, ripples also preplay sequences that have yet to be experienced (Dragoi and Tonegawa 2011, Dragoi and Tonegawa 2013, Farooq and Dragoi 2019). Notably, these CA1 preplay motifs are skeleton-like, such that preplay events often recruit smaller sets of CA1 neurons compared to post-learning memory replays, reflecting a more cursory neural scaffold of information during preplays (Dragoi and Tonegawa 2013, Grosmark and Buzsaki 2016, Farooq, Sibille et al. 2019). These findings reflect a crucial development in the field that ripple contents are, at least in part, composed of preconfigured CA1 sequence motifs (Farooq and Dragoi 2019, McKenzie, Huszar et al. 2021). Based on one estimation, hippocampal CA1 neurons form at least 15 preplay sequence motifs (Dragoi and Tonegawa 2013). And more preplay sequence motifs could be identified as memory demand increases (Rich, Liaw et al. 2014, Liu, Sibille et al. 2021). We speculate that these preconfigured CA1 motifs represent an encoding of ripple backbone by the low-threshold rigid pathway. Functionally, this ripple backbone structure could enable a fast information coding, a notion supported by recent findings that replay of CA1 sequences requires minimal experience (Bittner, Milstein et al. 2017, Berners-Lee, Feng et al. 2022).

We acknowledge that there is controversy surrounding the notion of preplay. Notably, research from the Foster group reported no preplay in multiple settings, and that blocking NMDA receptors during experience disrupts nearly all subsequent replays (Silva, Feng et al. 2015, Foster 2017), which calls into question the original findings and analyses (Dragoi and Tonegawa 2011, Dragoi and Tonegawa 2013). This failed detection of preplay or replay sequences seems to be at least partly explained by the usage of different statistical criteria; nonetheless, additional follow-up studies with modified analyses have since replicated the original preplay findings (Grosmark and Buzsaki 2016, Farooq, Sibille et al. 2019). Additionally, a potential concern of the CA3→CA1sup rigid pathway model stems from a report that CA3 neurons form statistically independent representations of many environments, which indicates a de novo generation of CA3 place cell maps rather than the existence of preconfigured CA3 motifs (Alme, Miao et al. 2014). This finding, in theory, does not necessarily argue against the existence of preconfigured CA1 motifs, because each CA3 mapping may activate only one of a large repertoire of CA1 motifs. In support, a recent study implies the existence of a large repertoire of CA1 network pre-configurations existing across many environments (Liu, Sibille et al. 2021).

Essential for rigid and plastic pathway integration is temporal synchronization of neuronal activity, as synchronized activity within a short time window is critical for information binding and Hebbian plasticity. A number of cortical regions, including the auditory cortex, visual cortex, somatosensory cortex, entorhinal cortex, prelimbic cortex, cingulate cortex, retrosplenial cortex, etc., potentially influence ripple contents due to their pre-ripple activation (Sirota, Csicsvari et al. 2003, Isomura, Sirota et al. 2006, Ji and Wilson 2007, Wang and Ikemoto 2016, Rothschild, Eban et al. 2017, Opalka, Huang et al. 2020, Nitzan, Swanson et al. 2022). Notably, these cortical regions exhibit cyclic Up and Down states that are largely synchronous during SWS (Battaglia, Sutherland et al. 2004, Isomura, Sirota et al. 2006, Narikiyo, Mizuguchi et al. 2020). This synchrony, thus, enables a coordinated cortical influence on ripples that occur predominantly during the Up states (Sirota, Csicsvari et al. 2003, Isomura, Sirota et al. 2006, Nir, Staba et al. 2011, Wang and Ikemoto 2016, Opalka, Huang et al. 2020).

Building on the above findings, we propose a dual-pathway framework that integrates novel information into pre-existing ripple motifs for memory consolidation. We speculate that a global cortical Up state triggers ripple generation and simultaneously influences ripple contents. Specifically, the Up state activates the low-threshold DG→CA3→CA1sup rigid pathway that intermittently triggers stochastic ripple generation. Meanwhile, the same Up state reactivates distinct tagged cortical neuronal ensembles, particularly those embedded within sensory cortices, which are subsequently incorporated into ripple motifs through the entorhinal/CA2→CA1deep plastic pathway (Figure 2).

Figure 2. Theoretical concept of rigid and plastic pathway integration.

Cortical Up states trigger ripple onset and synchronize the timing of each pathway to be activated during Up states and silenced during Down states. Activation of rigid/plastic pathway depends on strength of inputs. Basal cortical activity from low-saliency events (such as exploring a familiar environment) is sent to the hippocampus, where it is sufficient in triggering rigid pathway activation of CA1sup neurons but fails to circumvent PV mediated inhibition of CA1deep neurons. In contrast, cortical tagging stemming from high-saliency events (such as receiving foot-shocks) provides strong excitatory signal that not only triggers rigid pathway activation but is also sufficient in activating plastic CA1deep neurons, which can be incorporated into preconfigured CA1sup motifs. On right, depicts two different ripple events varying in content. Top, rigid pathway ripple which contains skeleton-like motif of CA1sup neuronal spiking (blue). Bottom, plastic pathway ripple which incorporates plastic CA1deep neuronal spiking (red) into preconfigured CA1sup motif.

6. Conclusions

Central to memory is the brain’s ability to encode and consolidate a seemingly limitless range of experiences. Whether it be something as insignificant as breakfast or as momentous as our wedding day, our brain possesses a near effortless ability in consolidating memories of wide-spanning saliency. This requires a robust consolidatory process that can efficiently encode memories from a multitude of experiences without losing much specificity. How exactly does the brain accomplish this task? Here, we have argued that the hippocampus possesses two distinct memory pathways which work in tandem to integrate complex ongoing information onto a rigid preconfigured network of motifs for efficient and specific memory consolidation.

Converging evidence suggests that sharp-wave ripples are a key component in systems consolidation. Ripples have been robustly linked to the reactivation of CA1 place cells, as well as cortical oscillations like delta waves and spindles. In this review, we establish two distinct pathways which simultaneously influence hippocampal ripple contents. The first, is the rigid pathway composed of DG→CA3→CA1 projections that preferentially target the CA1sup neurons, which tend to exhibit stable firing properties across ripples. The second, is the plastic pathway composed of entorhinal/CA2→CA1 projections that preferentially target CA1deep neurons, which tend to exhibit dynamic firing properties between pre- and post-learning ripples. Essentially, we propose that the rigid pathway forms a framework consisting of preconfigured CA1 sequence motifs that are stochastically activated by low-threshold cortical inputs. These preconfigured motifs act as a backbone for plastic pathway integration. Salient and novel experiences strengthen cortical-hippocampal connectivity which drives the activation of plastic-pathway CA1deep neurons. This learning-based novel information becomes incorporated on to rigid motifs through cortical Up/Down state synchronization. We speculate that this dual-framework process is adaptive due to its ability to incorporate new information rapidly and efficiently. The stochastic activity of preexisting ripple motifs establishes ongoing communication with cortical networks which enables efficient and fast integration of plastic activity rather than generation of a new network-communication. As it stands, outstanding questions remain with future investigations needed to provide causal evidence to further establish and uncover these distinct ripple pathways (Box 1).

Box 1.

Outstanding Questions
What is the mechanism for specific preconfigured CA1 motifs to be selected for information coding by CA3 stochastic activation?
Do cortical tagging associated neuronal ensembles selectively communicate with CA1 plastic neurons? As a prediction, the post-learning ripple associated CA1deep neuronal activity can be decoded by cortical population spikes.
How do CA1 rigid and plastic neurons differ in terms of efferent projections? Do they have rigid and plastic counterparts in their targeted brain regions?
CA1deep and CA1sup neurons express different levels of neuromodulator receptors such as acetylcholine and serotonin. How might these neuromodulators influence the rigid and plastic pathways?

Highlights.

• Sharp-wave ripples are generated within the CA3-CA1 network stochastically

• CA1 pyramidal neurons can be separated into rigid and plastic neurons

• Rigid and plastic pathways emerge from CA1 sublayer differences

• Cortical Up/Down states integrate rigid and plastic pathways