Is reactivation of neuronal firing patterns limited by brain systems contributions?

(1)

Is reactivation of neuronal firing patterns

limited by brain systems contributions?

13.02.2015

Lea Himmer (106 522 72) University of Amsterdam

MSc Brain and Cognitive Sciences, Cognitive Neuroscience track

Supervisor:

Dr. Monika Schӧnauer

University of Tübingen, Psychology and Behavioral Neurobiology

Co-Assessor:

Dr. Carien Lansink

(2)

1 Offline reprocessing of memory traces as a consolidation mechanism

The hippocampus has been a central structure in memory research for more than 50 years now. This interest was raised by reports on patient HM which suggested that his hippocampal lesions had led to specific memory deficits (Scoville & Milner, 1957). More studies showed that lesions in the medial temporal lobe (MTL) in humans as well as in other primates cause memory deficits and anterograde amnesia (Bohbot et al., 1998; Mishkin, 1978). These findings indicate that the hippocampus may be crucial for the formation of declarative memories and are the basis of models on declarative memory consolidation (Eichenbaum, 2001; Squire, 1992; Squire et al., 2004). Declarative information is thought to be encoded by a process of long-term potentiation (LTP) in the hippocampus, which then serves as an intermediate-term memory buffer. Interestingly, patients with hippocampal lesions show a gradient in their retrograde amnesia. Older memories often remain undisrupted compared to newer memories (Squire et al., 1989; Teng & Squire, 1999). It has therefore been suggested that memories become independent of the hippocampus over time. At the time, however, possible underlying mechanisms could only be speculated on (Squire, 1992).

McClelland et al. (1995) proposed the systems consolidation model to explain how this may occur on the basis of earlier suggestions on memory processes by Marr (1971), Squire et al. (1984) and Teyler and DiScenna (1984). The systems consolidation model describes a neocortical and a hippocampal system that interact during the course of memory formation. While the hippocampus represents a highly plastic fast learning storage, the neocortex needs more time to acquire and store information. High plasticity enables the hippocampus to acquire new information after a single learning trial, which is conceivably advantageous, but also entails the risk of rapid overwriting of fresh memories by new interfering learning. The two-storage model of systems consolidation solves this dilemma. During a learning experience, memory traces, i.e. changes in the synaptic connections between involved neurons, are formed in both systems. The neocortical representation needs to be additionally strengthened over time in order to be maintained independently, but is then also less susceptible to rapid interference. Memory trace reactivation, i.e. the recurring activation of neuronal firing patterns during memory acquisition continuing into later off-line periods, has been proposed as a mechanism which can lead to such a strengthening of neocortical memory traces and ultimately to an exclusive representation of memories in the neocortex.

Offline reprocessing or reactivation as a memory process had been suggested earlier already by Marr (1971). He proposed that the hippocampus may facilitate reprocessing in the neocortex. By strengthening cortico-cortical connections in this way, such a process may enable long-term memory storage in neocortical sites. This hypothesis was not only supported by Marr’s (1971) computational model of memory networks but also by structural findings (Teyler & DiScenna, 1984). Strong connections of the hippocampus to the neocortex could enable interaction of these structures and Teyler and DiScenna (1984) argued in their work on hippocampal anatomy that the hippocampus might use these projections to reactivate neocortical networks in order to strengthen memory.

Both the systems consolidation model and Marr’s (1971) hypotheses postulate that reprocessing of memory traces primarily takes place during sleep (McClelland et al., 1995). Changes in neuronal network

(4)

dynamics such as widely synchronized firing during slow oscillations (Diekelmann & Born, 2010) and changes in neurotransmitter levels (Hasselmo, 1999) support interregional communication. Acetylcholine inhibits feedback connections of the hippocampus to the neocortex during the wake state. During sleep these levels drop which may lead to increased communication between these regions and thus enable strengthening of memory traces in the neocortex (Hasselmo, 1999). The increased connectivity of hippocampus and neocortex during sleep offers an optimal milieu for memory consolidation since it may enable feed-forward memory reactivation and a concurrent systems consolidation. Therefore, consolidation mechanisms during sleep and the reactivation of memory traces in particular have received great interest over the last years and have been investigated extensively in different brain systems and species (for reviews see Diekelmann, 2014; Gais & Born, 2004; Genzel et al., 2014; Inostroza & Born, 2013; Maquet, 2000; O'Neill et al., 2010; Rasch & Born, 2013; Stickgold, 2013; Stickgold & Walker, 2013; Walker & Stickgold, 2006). However, the question whether the hippocampus is always involved in this process has not been addressed yet. By analyzing published results on memory reactivation this review aims to investigate which role the hippocampus plays in neuronal trace reactivation.

The main question of this review is thus, whether hippocampal involvement is essential for reactivation of neuronal activity during sleep to occur or whether reactivation can in principal emerge in any neural system. As part of this question, I will also investigate in which other brain structures reactivation has been observed and whether they form a coherent network. Because the hippocampus has been studied extensively and its role in memory is well known, I expect the hippocampus to be crucial for reactivation when different components of a memory that could be stored in different brain regions have to be bound together. However, I also propose, that because reactivation is a phenomenon that can be observed outside of the hippocampus, neuronal replay in other regions is not bare of functional relevance for memory consolidation.

(5)

2 Reactivation in rodents

Reoccurring neuronal firing patterns (reactivation) were first observed and investigated in rat and mice models. In rodents place cell activity in the cornu ammonis region CA1 can be recorded which holds the advantage that these cells fire selectively for certain positions in a given spatial field (O'Keefe, 1979). This creates very specific neuronal firing patterns that can be identified and investigated during subsequent resting periods.

Hippocampal activity is dominated by two distinct patterns. During exploring behavior and running, theta activity is the most prominent part of the neural signal. This activity needs an external pacemaker, whereas sharp waves that occur in the absence of theta are internally generated in the hippocampus. They lead to synchronous bursting of neurons in CA3 to CA1 and the subiciulum. This firing is synchronized in both hippocampi and across their longitudinal axes (Buzsaki, 1989). Buzsaki (1989) early proposed a two-stage model, that stresses the probable importance of the hippocampus for memory consolidation. In a first stage, relatively labile memory traces are formed during theta activity. These traces are then strengthened during sharp wave activity which is optimally suited to induce LTP. Sharp wave ripples (SWRs) can be detected across the whole hippocampal-entorhinal output pathway which connects the hippocampus to the neocortex. This connectivity is one reason why hippocampal SWRs might have a capacity to promote offline memory consolidation processes. They might induce long-lasting plasticity changes within the hippocampus and also in target structures of hippocampal efferents (Buzsaki, 1989; Chrobak & Buzsaki, 1996) .

Reactivation as well as SWRs have been found during sleep and wake behavior (Carr et al., 2011; Diekelmann, 2014). This review will focus on reactivation findings during sleep as these are more comparable to research on memory consolidation in humans. Because SWRs and reactivation are strongly coupled (Lansink et al., 2008; Lansink et al., 2009; Lee & Wilson, 2002; Nadasdy et al., 1999; Pennartz et al., 2004; Wilson & McNaughton, 1994), SWRs themselves can be viewed as a sign for reactivation and have been suggested to be involved in a hippocampal-neocortical dialogue that may integrate information across brain systems (Buzsaki et al., 1992; Chrobak & Buzsaki, 1996; McNaughton et al., 1983). In this section, findings on reactivation in rodents will be reviewed in respect to where they occur and which characteristics this reactivation shows.

2.1 Reactivation in the hippocampus

Several studies have tried to find whether replay of previous neuronal activity occurs in offline states in the hippocampus, when this can be found and what characteristics such replay shows. Pavlides and Winson (1989) showed that restricting the firing of place cells in CA1 by restricting movement of animals during wakefulness led to increased firing and bursting rates during subsequent slow wave sleep (SWS) for cells that were active during the task compared to silent cells. This increased activity declined over the time of sleeping. While such a finding does not necessarily show that memory is reactivated during sleep, they clearly show that experiences preceding sleep influenced neuronal activity during sleep. Following studies tried to investigate reactivation more closely by looking at firing patterns of cell pairs in hippocampus. Analyses here focused on whether cells that fire together during a spatial task are more likely to also fire together during subsequent resting intervals than other cell pairs. Wilson and

(6)

McNaughton (1994) found that such ongoing correlated firing occurs during SWS and rapidly decays over time. Also, the similarity of correlated of cell-pair firing during SWS compared to wake activity increased during the occurrence of SWRs.

These results already show two of the main characteristics of hippocampal replay that can be identified. One of these is, that reactivation has been found to occur mainly during SWS. Sleep is characterized by an alternation of different sleep stages. The most important distinction is usually made between SWS and REM-sleep (Diekelmann & Born, 2010). Reactivation in the hippocampus has also been found to occur during SWS in other studies. Kudrimoti et al. (1999) showed that correlations of cell pair firing were more similar to wake activity during SWS than during rapid-eye-movement (REM) sleep and were most robust during SWRs. Reactivation declined over time. Lee and Wilson (2002) investigated the replay of longer sequences of 4 or more spikes. Significant time-compressed reactivation of these was found during SWS after a spatial task.

Reactivation windows during SWS fit well with the two-stage model Buzsaki (1989) proposed. REM-sleep is characterized by theta activity (Diekelmann & Born, 2010). According to the two-stage model proposed by (Buzsaki, 1989), theta rhythms within the hippocampus accompany the formation of labile memory traces. It is sharp waves however, that occur outside of these windows of theta activity, i.e. among others in SWS, that contribute to LTP and memory strengthening. SWS is supposed to be beneficial for hippocampus-dependent memory consolidation in humans (Maquet, 2001), and reactivations taking place during these time windows might have a functional role in strengthening memory traces.

Also, reactivation is generally observed during the beginning of sleep intervals and then declines over minutes to hours (Kudrimoti et al., 1999; Pavlides & Winson, 1989; Wilson & McNaughton, 1994). Shen et al. (1998) showed that increased firing and increased pairwise co-firing in a sustained temporal order during sleep could not only be observed in CA1, but also in the dentate gyrus, where these effects declined rapidly over time. The underlying cause of this decline is still unknown but could possible originate from changes in synaptic strength during sleep in a process of homeostasis and a decrease of excitability (O'Neill et al., 2008; Olcese et al., 2010).

A third characteristic often observed in reactivation is a preserved order of firing.Skaggs and McNaughton (1996) demonstrated that the temporal sequence in which cell pairs fired together during a spatial task was sustained during following sleep, an effect that has also been found by (Shen et al., 1998). The same effect has also been found for longer sequences (Lee & Wilson, 2002; Louie & Wilson, 2001; Nadasdy et al., 1999). Nadasdy et al. (1999) examined longer spike trains of 4 to 13 CA1 neurons during exploration behavior and also found time-compressed reoccurrences of these spike trains with a preserved firing order during subsequent sleep and in quiet resting state where they were often accompanied by SWRs.

It should be noted here that in the wake state also reversed replay of sequences has been observed which is supposed to play a functional role in reinforcement learning (Foster & Wilson, 2006). The sustained order of replay during sleep points to a reactivation of memory traces following the structure

(7)

they had been encoded in. Replay could thus strengthen synaptic connections of neurons in a specific manner that depends on previous experiences and also codes for temporal relationship between single events. Strengthened connections could thus reflect experiences in a very precise way. It must be noted that early models of a two-stage consolidation process predicted a reverse rather than forward replay of memory traces in rest periods (Buzsaki, 1989). It is therefore surprising that replay during sleep shows a sustained forward firing order. However, more recent research has revealed that forward sequential activation of neuronal firing patterns occurs already during theta related behavior in a time compressed fashion during single theta cycles. This phenomenon has been termed theta phase precession since neurons that reflect a certain location in an environment fire progressively earlier in the theta cycle when the animal moves closer to its respective place field. This could explain why such forward replay can then also be observed in later rest periods (Sutherland & McNaughton, 2000).

While the order of firing is typically preserved, changes in the timescale of reactivation during offline periods have been analyzed in a number of studies investigating hippocampal reactivation. Replay of sequences has been found to be compressed in time in some cases (Lee & Wilson, 2002; Nadasdy et al., 1999). Why replay events are usually compressed in time has not been explained so far. The time compression might simply reflect the similar faster time-scale on which sequential firing of place cells that code for successive place field can be observed during theta phase precession during behavior (Sutherland & McNaughton, 2000). It can be speculated however, that this compression can also aid the strengthening of memory traces by representing an entire memory trace in a very short time window that is suitable for induction of LTP.

Some studies, however, do not show the same characteristics common to many other findings. Louie and Wilson (2001) found recurring spike sequences during REM-sleep after rats had repeatedly been running on a circular track. Sequences observed during waking behavior were replayed in a sustained temporal order. Reactivation here occurred during REM-sleep and not SWS and did not show the typical decline over time. REM-sleep and SWS show great differences regarding molecular changes, field potentials and their time of occurrence during sleep (Diekelmann & Born, 2010) and these changes might explain that Louie and Wilson (2001) did not find the typical decline of reactivation over time. Also, replay was not time-compressed but was slowed down in this study when compared to sequences in wake behavior. Again, these differences might be caused by the occurrence of these phenomena during different sleep stages. This study investigated minute-long sequences of neuronal activity while the other studies reviewed here typically investigated sequences in the scale of seconds. The difference in length of sequence might be a reason for the prominent difference in offline replay characteristics of the reactivation they detected.

Even though not as frequently investigated as the other characteristics given here, reactivation strength has also been shown to be modulated my memory trace strength (Kudrimoti et al., 1999; O'Neill et al., 2008). O'Neill et al. (2008) compared reactivation of neural activity during learning in novel and familiar environments as well. They found increased co-firing of cell pairs in the dorsal hippocampus during SWS following a spatial task, where these cells fired together. Furthermore, they showed that these correlations were elevated when more time was spent in the corresponding place fields of the place cells in the pair and if joint firing occurred with a short inter-spike interval. This modulating effect of

(8)

behavior was higher for novel than for familiar environments. This indicates a functional role of reactivation in memory consolidation. Stronger memory traces can represent more relevant sequences, that would be more likely to be reactivated during sharp wave events, as already proposed in the two-stage model (Buzsaki, 1989).

Studies investigating neuronal replay in rodents have used many different designs and reached diverse conclusions on the process of reactivation. Still, their results allow an analysis of characteristics of hippocampal replay that is typically found. Taken together, findings show that reactivation in the hippocampus does not occur at random times or in a random fashion. Its timing, occurrence during specific stages in sleep, the sustained order of firing and the modulation by memory strength indicate that reactivation is a very systematic mechanism that may potentially be involved in memory consolidation.

2.2 Reactivation in other brain areas

Early research on reactivation focused on the hippocampus. However, memory replay has also been found in other cortical and subcortical areas. This replay shares many of the characteristics of replay events discussed above. It must be noted, however, that this replay does not necessarily reflect the same reactivation process that is observed in hippocampal circuits.

Reactivation in somatosensory cortex, thalamus, striatum and visual cortex have been shown to occur mostly during SWS (Ji & Wilson, 2007; Lansink et al., 2008; Lansink et al., 2009; Pennartz et al., 2004; Ribeiro et al., 2004). Ribeiro et al. (2004) measured activity in hippocampus, primary somatosensory cortex, ventral posteromedial thalamus and putamen while rats explored familiar or novel environments. They found higher similarity of neuronal firing to task activity in subsequent SWS compared with sleep before the task in all regions. This difference in similarity was higher for new than for already familiar environments. Using unit recordings in both hippocampus and cortical areas, Ji and Wilson (2007) also showed replay of multicell firing sequences during sleep in both in the hippocampus and the visual cortex during SWS. This replay showed a sustained temporal order and was compressed in time. Three studies investigating replay events in the ventral striatum also found reactivation during SWS following a spatial navigation and reward task (Lansink et al., 2008; Lansink et al., 2009; Pennartz et al., 2004).

Also, the temporal order of replay in striatum, parietal and visual cortex as well as mPFC events has been shown to follow the temporal order observed in wake behavior (Euston et al., 2007; Ji & Wilson, 2007; Lansink et al., 2009; Qin et al., 1997). Qin et al. (1997) recorded neuronal activity in CA1 and the parietal cortex in rats during a maze task and during subsequent sleep. They analyzed correlated firing of cell pairs within and between these structures. This study found recurring correlated cell-pair firing both within the hippocampus, within the parietal cortex and for hippocampus-parietal cortex cell pairs. The temporal order of firing was sustained within structures. Euston et al. (2007) investigated neural activation in the medial prefrontal cortex (mPFC) during sleep periods before and after a sequential running task. They found higher similarity between cell firing during the task and firing during subsequent sleep compared to sleep before the task. Cell pairs in mPFC also fired in the same order as they did during the task. The time compression often observed in hippocampal replay has been reported

(9)

for the mPFC and visual cortex (Euston et al., 2007; Ji & Wilson, 2007) while another study found no time compression of replayed sequences in multiple brain areas (Ribeiro et al., 2004).

Three studies that examined reactivation in the ventral striatum extensively can shed further light on this topic (Lansink et al., 2008; Lansink et al., 2009; Pennartz et al., 2004). Pennartz et al. (2004) analyzed correlations of cell firing within the ventral striatum during a reward based navigation task and during preceding and succeeding periods of sleep. They found higher similarity of neuronal activity in SWS after the task in the ventral striatum that did not decline over time. Neurons whose firing activity during behavior was influenced by the timing of SWRs showed more reactivation, which suggests that this replay might not have occurred independently of the hippocampus. These findings were extended by a study by Lansink et al. (2008) who again observed reactivation in the ventral striatum after a spatial task with specific reward locations. They found reactivations in the ventral striatum during SWS as well as during quiet rest after the task that stayed stable over time. Neurons in the ventral striatum that showed strong firing responses to rewards during the task contributed more strongly to reactivations later. Reactivations were shown to be particularly strong when they occurred within a 200ms time window after hippocampal SWR onsets. A possible coordination of reactivation between hippocampus and ventral striatum was studied more precisely in a third experiment (Lansink et al., 2009). Here, replay events across CA1 and the striatum were investigated after rats were trained in a similar setup as in the preceding study. The results showed cross-regional reactivations of cell-pairs during post-task SWS. These reactivations were strongest when the hippocampal cells within the pair coded for place fields on the track and the striatal neuron was reward-related and when CA1 replay preceded the reactivation in the ventral striatum. This temporal bias of cell pairs was preserved during SWS in this study in CA1-striatum, striatum-striatum and CA1-CA1 cell pairs. Furthermore, the number of reactivation was correlated with the number of repetitions of the task and session progression. While these findings on reactivation are similar to findings on hippocampal replay considering sleep stage and temporal order of firing, they report a considerable difference in decay of reactivation over time. A rapid decline of reactivations over time is typical for hippocampal cells, whereas no such decrease has been found in neurons in the ventral striatum (Lansink et al., 2008; Lansink et al., 2009; Pennartz et al., 2004).

Lastly, a similar modulation of reactivation by the number of repetitions of a behavior that O'Neill et al. (2008) reported for hippocampal replay was found in the ventral striatum and mPFC (Lansink et al., 2008; Peyrache et al., 2009). Peyrache et al. (2009) investigated reactivations in hippocampus and mPFC. They extracted principle components of neuronal activity in mPFC during the task and found reoccurrences of these components during sleep. The reactivation events coincided with and peaked about 40ms SWR onset. Also, patterns that were observed more frequently compared to other patterns during wake behavior, were replayed more often.

Even considering the differences pointed out for reactivation decay in the striatum, replay patterns in multiple brain regions have been shown to occur during the same sleep stage as hippocampal replay and to preserve the firing order determined during behavior. Thus, it is likely that these replay events are triggered by a similar mechanism as hippocampal replay or that they are somehow connected to reactivation in the hippocampus. Also, this reactivation seems to be as systematic as hippocampal replay and can thus be hypothesized to be reprocessing some of the neuronal memory traces formed during

(10)

behavior. The strong similarity of the characteristics of reactivation in other brain areas and the hippocampus raises the question whether both reflect the same process and whether reactivation in other brain areas is bound to hippocampal replay or vice versa.

The question remains, however, whether reactivation in areas outside the hippocampus reprocesses different information than reactivation within hippocampus. The modulation of reactivation event number by behavior that was found in the ventral striatum as well as in the mPFC (Lansink et al., 2008; Peyrache et al., 2009) suggests that some of the experiences during behavior might be reprocessed in those areas. However, neither these findings nor the speculations made by Qin et al. (1997) that reactivations in the hippocampus process spatial information, while cortical areas integrate motor-sensory information, are proof of the nature of information that gets reprocessed in these areas. A more specific answer can be given for reactivation in the striatum which Pennartz et al. (2004) hypothesized to contribute a motivational value to memory. This hypothesis was investigated by Lansink et al. (2008) who analyzed replay of neurons that fired in response to rewards during behavior compared to neurons whose firing was related to rewards. They found, that reward related units not only showed more replay but also that replay events were more strongly temporally aligned to SWRs in the hippocampus. Because the striatum is regarded to be a crucial structure in motivational behavior (Mogenson et al., 1980) this could point to a domain-specific mechanism of reactivation, i.e. information is being reprocessed in the structure it has been encoded in (Lansink et al., 2008). An additional indication for domain-specific reprocessing is the finding in a following study, that found the strongest replay events in CA1-striatum cell pairs, in which the hippocampal unit fired for spatial fields and the striatal unit represented rewards during behavior (Lansink et al., 2009). It is still unclear however, whether domain specific reprocessing is a general mechanism of reactivation. Should this be the case, then replay across multiple brain structures would have to be coordinated by a central orchestrator. This orchestrator would be necessary to bind these different components of a single experience that is represented in multiple different memory traces in different brain structures. The structure most often proposed to fulfill this role is the hippocampus (Ji & Wilson, 2007; Lansink et al., 2008; Lansink et al., 2009; Pennartz et al., 2004; Peyrache et al., 2009).

Many studies have already tried to link reactivation in other brain areas to reactivation in the hippocampus. One approach to do so is showing that replay events are temporally closely coupled to SWRs in hippocampus, which has been found in a few studies (Ji & Wilson, 2007; Lansink et al., 2008; Peyrache et al., 2009). Showing a temporal coupling of these events that occur in the same sleep stage, however, is not sufficient to show a causal relationship of these events, i.e. concluding that the hippocampus is generating or facilitating replay in other regions. For instance, mPFC activity was generally increased during SWRs in a study by Peyrache et al. (2009) which increases the probability for replay events to occur and thus concur with SWRs. Ji and Wilson (2007) additionally showed that the reactivations observed in visual cortex and the hippocampus at the same time were replays of the same sequence observed during a behavioral experience in those two regions. While this does show that these events are probably connected, no conclusion can be drawn about where reactivation originated or how these events were coordinated.

(11)

A way to look at actually connected replay in different regions is to show the reactivation of sequences that include neuronal units from different brain regions. Qin et al. (1997) observed replay of parietal cortex - hippocampal pairs but were unable to show a sustained temporal bias of these reactivations. In contrast, Lansink et al. (2009) showed hippocampal-striatal cell pair correlations during SWS whose replay showed a sustained temporal bias. Additionally, they found, that these cross-regional cell-pairs show more replay if the hippocampal unit fired before the striatal unit during behavior. Thus, these results not only demonstrate that replay across striatum and hippocampus was coupled, but also indicate that hippocampal reactivation is likely to precede replay in the striatum. This timing bias is a requirement for the hippocampus to take a role as a central orchestrator, as reactivation would be initiated there and then induced in other brain structures. While this connection has ultimately only been shown for the striatum, this finding indicates that the hippocampus is able to trigger reactivation in another brain structure. Thus, replay events observed in various brain structures might not occur independently but rely on the hippocampus as an orchestrator.

2.3 Reactivation and field potentials

Another way of researching events related to reactivation and underlying brain areas is looking at field potentials related to SWRs. This approach is less direct than investigating neuronal replay in various brain structures with unit or cell recordings, because field potentials are caused by activity of huge numbers of neurons and the signal cannot always be traced back to a single structure initiating it. However, this technique offers the great advantage of being transferrable to research in humans because imaging techniques like electroencephalography (EEG) and magnetoencephalography (MEG) provide similar signals to field potential recordings in rodents. Also, neural activity during sleep shows very characteristic field potentials like sleep spindles and slow oscillations (Diekelmann & Born, 2010) in both rodents and humans. Furthermore, coupled network oscillations are likely to facilitate interactions between brain structures as they form fast, precise and persistent patterns across brain regions (Sirota & Buzsaki, 2005).

One phenomenon SWRs are frequently linked to is sleep spindles. Sleep spindles are short oscillatory events with a frequency of 7 Hz to 14 Hz that can be observed in neocortical and thalamic areas (Contreras et al., 1996) and originate by an interaction of thalamic and thalamocortical loops (Steriade et al., 1985). Siapas and Wilson (1998) found a close temporal relationship of neocortical spindles and SWRs during non-REM sleep in rats. SWRs slightly preceded spindle events, which can either point to a third region driving both these phenomena or to SWRs biasing the initiation of neocortical sleep spindles. An orchestrating role of the hippocampus cannot be concluded from these findings, as no causal connection was shown. Furthermore, both SWRs as well as spindles are events frequently observed during non-REM sleep (Diekelmann & Born, 2010) and thus very likely to occur at the same time. The functional significance of sleep spindles for memory consolidation in rats has been shown by Eschenko et al. (2006), who found higher spindle densities during sleep in rats after they learned odor-reward associations. These findings mirror the increase in SWR density after learning that (Ramadan et al., 2009) found and show that spindle density, which is frequently investigated in human memory consolidation research is connected to memory trace reactivations (Clemens et al., 2005; Gais et al., 2002; Meier-Koll et al., 1999; Mölle et al., 2002; Schmidt et al., 2006).

(12)

Slow oscillations have also been linked to SWRs. Slow oscillations occur at frequencies between 0.5 Hz and 4 Hz and originate in frontal areas. They then travel in posterior direction and later reach the hippocampus and subcortical regions (Riedner et al., 2011; Wolansky et al., 2006). Slow oscillations occur when neurons in the neocortex are synchronized in large areas and oscillate between an up-state, which reflects cortical depolarization and a down-state when neurons are mostly hyperpolarized (Battaglia et al., 2004). Up-states are probably comparable to frames of elevated neuronal activity during sleep that have been reported in rodent studies (Ji & Wilson, 2007; Louie & Wilson, 2001; Wilson & McNaughton, 1994). It should be noted here, that the same bias of temporal correlation as explained above for sleep spindles applies for these events. Slow oscillations as well as SWRs are neural events frequently occurring during non-REM sleep and are thus very likely to take place at similar times. Battaglia et al. (2004) investigated the temporal relationship of SWRs and slow oscillations. Slow oscillations were found to arise in a highly synchronous manner across the whole cortex and SWRs tended to coincide with transitions from down- to up-states. Such a temporal coupling indicates that SWRs originating in the hippocampus may trigger reactivation in the neocortex during the up-state of slow oscillations (Battaglia et al., 2004). SWRs and slow oscillations have also been investigated in anesthetized rodents. Isomura et al. (2006) recorded neuronal activity in the cortex, entorhinal cortex, CA1 and the subiculum. They found slow oscillations that not only spread over the whole cortex but also reached into the entorhinal cortex and subiculum. Additionally, membrane potentials of neurons in CA1 and CA3 were influenced by cortical up- and down-states. Most SWRs observed in this study occurred during cortical up-states and usually followed down-to-up-state transitions with a delay of about 100ms. The authors suggested that this temporal relationship might reflect a cortical input to the hippocampus that influences neuronal reactivation in this region, which consecutively enables feedback of the hippocampus to the neocortex via SWRs (Isomura et al., 2006). The finding that slow oscillations reached into hippocampal subfields and influence membrane potentials of hippocampal cells demonstrates a more causal connection of slow oscillations and SWRs than had been shown by pure timing correlations previously. This indicates that neocortical phenomena might influence when reactivation is initiated within the hippocampus. These findings do however not ultimately proof a feedback mechanism from the hippocampus to the neocortex.

Similar conclusions had been drawn before from a study investigating SWRs, sleep spindles and slow oscillations in rats and mice (Sirota et al., 2003). The authors recorded activity in somatosensory areas and the hippocampus and examined the temporal coupling of neuronal events on a short and a long timescale. On a long timescale of 1s to 2s, SWR events were associated with higher power in spindle and delta frequencies. On a short timescale, SWRs were predominantly observed about 50ms after spindle troughs and slow oscillation troughs. These robust temporal relations suggest a strong interaction between the different brain regions during sleep, which may facilitate strengthening of memory traces (Sirota et al., 2003). These results agree with findings by (Isomura et al., 2006). SWRs occurred shortly after the transition from the cortical down-to-upstate, which again indicates a neocortical influence on hippocampal activity, and thus possibly the timing of memory replay. Hippocampal SWRs could then provide feedback to the neocortex, where neurons would still be in the up-state and thus very excitable (Isomura et al., 2006; Sirota et al., 2003). While causal conclusions cannot necessarily be drawn from these findings, the robust temporal relationships found in multiple studies match very well across

(13)

different studies and are unlikely to be based on coincidental timing. Additionally, at least neocortical influence on hippocampal neuronal activity by slow oscillations has been demonstrated (Isomura et al., 2006).

Mölle et al. (2009) investigated the influence of learning on sleep spindle activity, SWRs and slow oscillations in humans and rats. Learning before sleep led to increased amplitudes of cortical up-states as well as an increase in spindle and SWR density in rats. However, they did not find a temporal relationship between spindles or SWRs and slow oscillation events. Sleep spindles tended to coincide with SWRs during sleep before learning, whereas they shortly succeeded SWRs after sleep. These results confirm the coupling of spindles and SWRs. However, the absence of coordination by slow oscillations is startling, especially because the same study found such a relationship for slow oscillations and sleep spindles in humans (Mölle et al., 2009).

The temporal coordination of different field potentials suggests close interactions between the brain regions they originate in. To achieve a more precise understanding of this process, direct interactions of cell pairs across CA1 and mPFC during sleep have also been investigated with regard to SWRs (Wierzynski et al., 2009). Wierzynski et al. (2009) found that SWRs preceded firing in mPFC neurons by up to 100ms in 70% of all the cell pairs they recorded during sleep. Moreover, they calculated the strength of SWRs by dividing the number of spikes within a burst by the number of CA1 neurons in the dataset, and then correlated this with the multiunit response in mPFC. They found that small bursts in the hippocampus induce a single peak of prefrontal activity within a short timeframe, whereas more powerful bursts generate the same early prefrontal response, but also additional activity about 100ms later. This additional activity showed increased power in the spindle-band. This study thus not only demonstrates a temporal relationship of these phenomena but also a dependency of neuronal responses in mPFC on the strength of hippocampal input that is provided by SWRs.

Overall, these findings point to a critical involvement of the hippocampus in interregional communication during memory reactivation. SWRs originate in CA3 and CA1 (Chrobak & Buzsaki, 1996) and are functionally involved in neuronal trace reactivations (O'Neill et al., 2006). The number of SWRs is increased during sleep after learning experiences (Ramadan et al., 2009) and suppressing them impairs subsequent memory performance (Ego-Stengel & Wilson, 2010; Girardeau et al., 2009). Also other field potentials like sleep spindles and slow oscillations, which originate in neocortical and cortico-thalamic circuits, have been connected to SWRs and neuronal replay, which indicates an active interaction of these areas. SWRs typically occur during cortical up-states or the transition from down-to-up-states (Battaglia et al., 2004; Isomura et al., 2006; Sirota et al., 2003). It has been proposed before that this temporal interaction supports neocortical events by influencing the timing as well as the content of replay in the hippocampus (Sirota & Buzsaki, 2005). Actual modulation of hippocampal SWR activity by slow oscillations has also been shown (Isomura et al., 2006). Replay in the hippocampus, as reflected by neuronal trace replay and SWRs, could then give feedback to the neocortex, at a time when neurons in these areas are in a more excitable up-state (Isomura et al., 2006). This feedback hypothesis is supported by the findings of Wierzynski et al. (2009) who showed a differentiated response of mPFC neurons to hippocampal SWRs. The role of sleep spindles is less clear. Spindle density increases after training (Mölle et al., 2009) and in close temporal proximity to or following SWRs (Mölle et al., 2009;

(14)

Sirota et al., 2003; Wierzynski et al., 2009), so a causal involvement of spindles in memory reactivation can be hypothesized. As most studies have shown that spindles follow shortly after SWRs, spindles can be speculated to reflect neuronal activity in the neocortex that is triggered by feedback from the hippocampus. A great advantage of these methods used in these studies is that similar events can be observed in humans so that the functional impact of the described field oscillations on memory can be assessed. Also, the hypothesis that cortical activity can bias hippocampal replay is tested in humans by presenting external triggers that are processed cortically during sleep and examining effects on memory performance.

2.4 The hippocampus as a possible orchestrator of reactivation

In summary, unit recordings have shown that replay across brain regions exists and that the hippocampus might take on a leading role during this reactivation (Lansink et al., 2009; Qin et al., 1997). Evidence studies investigating field potentials allows less conclusions but suggests close interactions of hippocampal SWRs and neocortical and thalamic phenomena like slow oscillations and sleep spindles. Both of these fields hypothesize the hippocampus to play a central role in initiating and orchestrating reactivation (Isomura et al., 2006; Ji & Wilson, 2007; Lansink et al., 2008; Lansink et al., 2009; Pennartz et al., 2004; Peyrache et al., 2009; Sirota et al., 2003).

This hypothesis is not only based on the findings reviewed above but also based on the special network specifics of the hippocampus. The hippocampus receives inputs from widespread areas and also sends projections back there. While the main input comes from the entorhinal cortex, the subiculum as the major output pathway of the hippocampus sends efferent connections to multiple brain areas. These areas include thalamic nuclei, mammillary bodies, the nucleus accumbens, parts of the olfactory nucleus and widespread cortical areas like the retrosplenial cortex and the parahippocampal gyrus (Rosene & Van Hoesen, 1977; Swanson & Cowan, 1977; Teyler & DiScenna). The anatomy of the hippocampus thus allows an initiation of reactivation in other structures and therefore also an orchestration of these events.

Sharp waves, that are spontaneously initiated in hippocampus are hypothesized to be able to aid LTP (Buzsaki, 1989). SWRs are suggested to be involved in a hippocampal-neocortical dialogue that may integrate information across brain systems (Buzsaki et al., 1992; Chrobak & Buzsaki, 1996; McNaughton et al., 1983) and have been found to coincide with replay events in multiple rodent studies (Lansink et al., 2008; Lansink et al., 2009; Lee & Wilson, 2002; Nadasdy et al., 1999; Pennartz et al., 2004; Wilson & McNaughton, 1994). They could possible lead to a feed-forward excitation that could be aided by changes in neurotransmitter levels that occur during sleep. Lower levels of acetylcholine during SWS lead to an increased communication of the hippocampus to the neocortex and this could promote feed-forward excitation by the hippocampus (Gais & Born, 2004; Hasselmo, 1999).

How relevant this excitatory input is for reactivation in other brain areas can be illustrated at the example of the striatum. The striatum serves as the primary input structure of the basal ganglia and consist predominantly of GABAergic interneurons and spiny projections (Bennett & Bolam, 1993; Tunstall et al.). Replay is unlikely to be initiated spontaneously in this system of inhibitory neurons. Excitatory input from a different structure could thus be crucial for reactivation to be observed in this

(15)

structure. Because replay has been observed in the striatum (Lansink et al., 2008; Lansink et al., 2009; Pennartz et al., 2004) and this replay has been shown to be connected to replay in the hippocampus it is very likely that this structure is initiating the necessary excitation for replay to occur.

To conclude, research in rodents so far has shown that reactivation does not only occur within the hippocampus, but can be found in multiple cortical and subcortical structures. However, many studies suggest an important role for the hippocampus in the process. Reactivations in other structures are repeatedly reported to co-occur with hippocampal SWRs (Ji & Wilson, 2007; Lansink et al., 2008; Peyrache et al., 2009; Wilson & McNaughton, 1994). Also, reactivation has been shown to take place in cell pairs whose cells are located in different brain structures. The hippocampus seems to adapt a leading and initiating role in these cell pairings (Lansink et al., 2009; Qin et al., 1997). Furthermore, reactivation events across the hippocampus and other structures have been shown to reflect the same event (Peyrache et al., 2009) or to contain specific components of memories (Lansink et al., 2009). These findings suggest that the hippocampus could be an orchestrator of reactivation that coordinates memory trace replay in other regions. Striking findings on field potentials furthermore suggest that hippocampal replay timing can be influenced by slow oscillations. The hippocampus could then be initiating replay events in other areas in the brain by providing an excitatory input (Isomura et al., 2006; Sirota et al., 2003). These findings are in agreement with the hippocampus’ well known role in memory (Scoville & Milner, 1957; Squire et al., 2004) and thus a central role for the hippocampus in this consolidation mechanism seems plausible.

(16)

3 Memory reactivation in humans

Although it has long been known that sleep is beneficial for memory performance in humans (Heine, 1914; Jenkins & Dallenbach, 1924), the exact mechanisms that mediate these effects had remained unclear. Reactivation of memory traces during sleep as observed in electrophysiological recordings in animals can potentially strengthen memories. However, research in rodents rarely investigates whether reactivation of memory traces during sleep has a beneficial effect on later task performance. Research on memory reactivation in humans tried to investigate whether it is plausible to assume a similar network and processes of memory reactivation in humans as in animals. The imaging techniques used in research in humans are limited. Surface EEG and functional magnetic resonance imaging (fMRI) are readily available imaging methods. Also local field potential activity recorded from intracranial electrodes in e.g. epilepsy patients has been analyzed with regard to memory reprocessing during sleep. However, these methods’ spatial and temporal resolutions make it impossible to measure brain activity on the level of single cell firing, and the obtained results are therefore difficult to compare with findings from animal research. It is still unclear whether reactivation measured on the level of brain regions in humans reflects the same underlying processes as reactivation measured on the single cell level in animals. Even though reactivation of memory traces can only be measured on such a coarser level in humans, the results might still give additional insight into the underlying processes and neural systems. Most importantly, however, research in humans has aimed to associate the reprocessing of learned material during sleep with memory performance, elucidating the functional relevance of reactivation for memory formation.

3.1 Field potentials show comparable temporal coordination in humans

Studies on field potentials associated with reactivation of neuronal activity patterns give insight into involved brain structures and their interactions during this process. Neural oscillations can be interpreted as signs of reactivation, even though they do not represent the replay of formerly encoded memory traces directly. The most distinct neuronal events that accompany memory trace reactivations are SWRs. SWRs have not only been observed in rodents but also in humans (Axmacher et al., 2008; Bragin et al., 1999) and are associated with sleep spindles and slow oscillations, two phenomena that occur frequently during sleep (Diekelmann & Born, 2010). Coupled network oscillations are likely to facilitate interactions between brain structures as they form fast, precise and persistent patterns across brain regions (Sirota & Buzsaki, 2005). The investigation of field potentials that have been previously connected to memory trace reactivation by using electroencephalography (EEG) can therefore shed further light on the reactivation process in humans.

3.1.1 Intracranial EEG

SWRs cannot be detected in scalp EEGs because they originate deep within the brain in the hippocampus. Thus, SWR activity in the human hippocampus has been investigated using deep electrode recordings in epilepsy patients. Bragin et al. (1999) recorded neuronal activity in multiple brain areas in epilepsy patients and found SWRs in the hippocampus and entorhinal cortex. Similar to findings in animals, these ripples only emerged during quiet rest and sleep and occurred largely simultaneously over hemispheres. Extending these results, another study investigated SWRs and slow oscillations in epilepsy patients by simultaneous deep electrode measurements in the medial temporal

(17)

lobe and scalp EEG recordings during sleep after a memory task (Axmacher et al., 2008). SWRs were found in the hippocampus and the rhinal cortex and were usually observed during the up-state peak of slow oscillations. Interestingly, this study found a correlation of the number of rhinal SWRs and recognition performance in the morning. If ripples in fact reflect spontaneous memory reactivation, this indicates that reactivation during sleep is related to later memory strength. The same correlation with performance was not observed for hippocampal SWRs, which might have been an effect of the small number of contacts in the medial temporal lobe detecting only a small number of SWR events. A third study that recorded activity in the parahippocampus using intracranial electrodes and a simultaneous scalp EEG studied SWR sleep spindle interactions in epilepsy patients (Clemens et al., 2011). Spindle activity was observed over the whole cortex and within the parahippocampus. SWR activity was locked to the trough of sleep spindles and more SWRs were found before than after spindle peaks. These findings mirror the findings by Sirota et al. (2003) who found a similar temporal relationship of sleep spindles and SWRs. Taken together, these studies not only show that SWRs occur in humans, but also that the temporal dynamics of SWRs, slow oscillations and sleep spindles in humans mirror the findings in rodents (Isomura et al., 2006; Sirota et al., 2003). SWR events are locked to the up-state of slow oscillations and to the troughs of sleep spindles. It is thus likely that these field potentials are similarly involved in neuronal trace reactivations in humans and enable cross-regional neuronal communication.

3.1.2 Surface EEG

Because sleep spindles and slow oscillations can be observed in scalp EEGs, they have received a lot of interest in memory research. Meier-Koll et al. (1999) compared the number of sleep spindles after a spatial learning task to a period of sleep without preceding learning experience and found that spatial learning increases the number of spindles. This suggests an involvement of sleep spindles in memory consolidation. Similar findings were obtained by Gais et al. (2002) who found a higher number of sleep spindles during sleep after participants had done a declarative learning task compared to controls who received similar visual input. The highest increase in spindle-density was observed in a fronto-central position where the number of spindles also positively correlated with performance in the memory task before as well as after sleep (Gais et al., 2002). These results point to a causal role of sleep spindles in memory consolidation, which has also been proposed by Clemens et al. (2005). They found that the number of spindles in the left frontal lobe correlates with the overnight retention of memory in a verbal memory task. The number of sleep spindles is not only influenced by the occurrence of learning per se, but also additionally modulated by the kind of task (Schmidt et al., 2006). Increasing difficulty in a memory task led to an increase in the number of spindles in the left frontal cortex. The number of spindles again correlated with memory performance after sleep. These findings all point to a functional role of sleep spindles in memory consolidation. The strong temporal coordination of SWRs and sleep spindles indicates that they may be involved in the memory reactivation process. It cannot be determined from this, whether the reactivation process or the spindle activity itself are beneficial for performance. However, the findings stress the importance of reactivation-associated field potentials for memory consolidation. Schmidt et al. (2006) found that especially spindles in frontal and posterior regions correlated with behavior. Thus, these regions may be particularly involved.

(18)

Mölle et al. (2002) investigated the interaction of slow oscillations and sleep spindles. They recorded EEG signals in sleeping participants and found that while the number of spindles was increased during cortical up-states they were suppressed during down-states. Mölle et al. (2009) who investigated the influence of learning experiences on field potentials in both humans and rats (as reported above) found an increased amplitude of slow oscillations during sleep after a word association task. Moreover, they found increased spindle activity during up-states while it was decreased during down-states. This temporal relationship was enhanced after learning and more sleep spindles occurred during the transition to and during up-states. The fact that foregoing learning influences the temporal relationship between spindles and slow oscillations further supports the idea that it holds an important role in the consolidation process. To test functional relevance of slow oscillations for memory consolidation, (Marshall et al., 2006) induced slow oscillations during sleep by applying transcranial direct current stimulation after a procedural and a declarative task(Marshall et al., 2006)(Marshall et al., 2006)(Marshall et al., 2006)(Marshall et al., 2006) (Marshall et al., 2006). The stimulation led to better performance in the declarative task the next morning and not only induced slow oscillations in fronto-central sites, but also increased spindle power in this. This again corroborates the view that both slow oscillations and spindles are involved in memory consolidation and their occurrence is intricately linked and orchestrated.

The temporal orchestration of field potentials reveals a network of cortical and subcortical structures that show coordinated activity during sleep. Findings in humans show that field potentials over multiple brain regions are connected to memory consolidation and SWRs (Axmacher et al., 2008; Bragin et al., 1999; Clemens et al., 2011). The temporal coupling of these events in humans matches that in animals closely (Axmacher et al., 2008; Clemens et al., 2011; Marshall et al., 2006; Mölle et al., 2009; Mölle et al., 2002). Both sleep spindles and slow oscillations seem to have a causal role in memory consolidation in humans (Clemens et al., 2005; Gais et al., 2002; Marshall et al., 2006; Schmidt et al., 2006). They are tightly connected to SWRs, which stresses the involvement of the hippocampus coordinating interregional communication. This interplay may coordinate memory trace reactivations, which occur in a synchronized manner across brain regions. Because sleep spindles and slow oscillations originate in cortical and thalamic circuits these areas are also likely to contribute to memory reactivations and memory systems consolidation. The exact roles of the thalamus and neocortex in this process, however, have not yet been investigated.

SWRs originate in the hippocampus in both humans and rats (Axmacher et al., 2008; Bragin et al., 1999; Chrobak & Buzsaki, 1996; Clemens et al., 2011). They have been consistently found to be temporally connected to slow oscillations and sleep spindles (Axmacher et al., 2008; Battaglia et al., 2004; Clemens et al., 2011; Isomura et al., 2006; Marshall et al., 2006; Mölle et al., 2009; Mölle et al., 2002; Sirota et al., 2003; Wierzynski et al., 2009). This temporal coordination might be regulated by hippocampal activity. Slow oscillations up-states slightly precede hippocampal SWRs, which suggests that cortical areas influence the timing and possibly also the content of replay in the hippocampus. SWRs could then enable the hippocampus to give feedback to the still excited neocortical areas and thus trigger reactivation outside of the hippocampus. This framework proposed by Sirota et al. (2003) and Isomura et al. (2006) is not only indicated by evidence in rodents but also supported by highly similar findings in

(19)

human memory research. Field potentials offer an ideal mechanism to enable communication between brain structures because their close and robust temporal association allows fast and precise communication (Sirota & Buzsaki, 2005). This may enable the strengthening of an integrated memory trace in the neocortex and thus explain how memories become more and more independent of hippocampal involvement as proposed in the systems consolidation model (Dudai, 2004; McClelland et al., 1995).

3.2 Reactivation on the level of brain systems in humans

Research on memory reactivation in humans is confined by available imaging methods. Positron emission tomography (PET) and functional magnetic resonance imaging (fMRI) can measure elevated neural activity based on regional cerebral blood flow (rCBF) or on blood-oxygen levels. However, these techniques are not comparable to cell recordings because of their lower spatial and temporal resolution. They can only reveal a “re-activation” on the level of brain regions. Still, attempts to show neural reactivations in humans have been made.

Maquet et al. (2000) measured rCBF levels during sleep after participants had been trained on a serial reaction time task (SRTT). Participants had to press buttons in sequences that followed probabilistic rules without them being aware of this sequential structure. The authors compared which brain areas where more active during sleep after learning than during sleep with no preceding learning experience. Elevated activity was found in the cuneus, the adjacent striate cortex, left premotor cortex and within the mesencephalon, regions that had also been active during task training.

A follow-up study refined this design by investigating whether this elevated activity reflected probabilistic sequence learning that is induced by the SRTT or whether it was induced by basic visuomotoric learning (Peigneux et al., 2003). Instead of comparing sleep after training with sleep after no training, one group performed the SRTT while a control group performed a similar task tapping only random sequences. They found elevated activity during subsequent sleep in participants who trained the SRTT in the cuneus and with a more liberal threshold in premotor cortex and the mesencephalon, suggesting that reactivation of these areas reflected higher order learning. These findings greatly resemble the findings by Maquet et al. (2000). It is interesting to note that even though the hippocampus has been shown to be involved in SRTTs (Albouy et al., 2008; Rieckmann et al., 2010; Schendan et al., 2003), no ongoing task-related activity during sleep after learning was found in the hippocampus in both studies (Maquet et al., 2000; Peigneux et al., 2003). This is surprising, since models of memory reactivation during sleep developed from findings in rodents suggest a crucial role of the hippocampus in initiating the reactivation process.

A third PET study compared sleep after a spatial navigation task to sleep without preceding learning experiences and to sleep after performing the SRTT (Peigneux et al., 2004). They found elevated activity in the hippocampus and parahippocampus during the spatial navigation task compared to the SRTT. Also during post-training sleep, they found higher hippocampal and parahippocampal activity during sleep compared to wake after spatial learning compared to after the SRTT. The overnight improvement of task performance correlated with rCBF in hippocampus and parahippocampus during sleep for spatial

(20)

learning. The authors interpreted these findings as the hippocampus being involved in offline reprocessing of previous experiences.

Yotsumoto et al. (2009) investigated recurring neural activity during sleep after training on a procedural task that is acquired without hippocampal involvement. The texture discrimination task (TDT) induces simple visual learning and later performance has been shown to benefit from sleep (Gais et al., 2000; Stickgold et al., 2000). Participants learned to discriminate the orientation of briefly presented stimuli that were always shown in one part of the screen. This leads to the selective adaptation of the corresponding retinotopic region in V1. Subjects were scanned using fMRI during sleep preceding and succeeding training. They showed elevated neural activity during post-task sleep compared to sleep before training within the area in V1 that had been trained by the TDT. The difference of neural activation in the trained region of V1 to the untrained region was bigger during post-task sleep compared to before. The activations in the trained V1 area during sleep correlated with performance while activity in untrained V1 areas did not. These results show a very selective reactivation of areas involved in a task preceding sleep and could reflect offline reprocessing during sleep (Yotsumoto et al., 2009). Because the task used in this design is completely independent of the hippocampus, these results argue against a crucial involvement of the hippocampus in neural reactivation. This study thus indicates independent reactivation in the cortical areas in which a memory trace has been formed during encoding. This finding is thus in line with hypothesis made in rodent research on reactivation occurring in domain-specific regions(Lansink et al., 2009; Qin et al., 1997).

Overall, studies in humans are difficult to interpret due to methodological difficulties. While one study found ongoing task-related activity in the hippocampus during sleep following a spatial task (Peigneux et al., 2004) other studies found no hippocampal involvement (Maquet et al., 2000; Peigneux et al., 2003) or argue against it (Yotsumoto et al., 2009). However, the available methods do not allow direct conclusions. Higher activation in these studies reflects elevated rCBF or blood-oxygen levels and might thus simply indicate elevated energy demand in certain areas. This might be caused by neuronal trace replay but it could also reflect use-dependent ongoing activity, synaptic downscaling (Tononi & Cirelli, 2003, 2006), long term potentiation, protein synthesis, homeostasis or any other process that demands increased metabolism. For instance, Yotsumoto et al. (2009) stated that their results could be a sign for offline reprocessing but that it could not be excluded that these activations reflect synaptic consolidation. Evidence for hippocampal involvement in neural trace reactivation in research on humans thus remains inconclusive.

New analysis techniques like multivariate pattern analysis (MVPA) approaches for fMRI data that allow more specific conclusions about what activations in the brain reflect are beginning to overcome these methodological difficulties. Tambini and Davachi (2013) let participants rest in a wake state for 8 minutes after encoding pictures of faces or scenes inside an MRI scanner. Analysis of functional data in the hippocampus revealed that activity during the resting periods was more similar to the activations caused by the previously learned picture category than the remaining category (Tambini & Davachi, 2013). However, this effect could have been driven by active rehearsal of learnt contents. A second study investigated reoccurring patterns during a delay period between encoding and recall sessions (Staresina et al., 2013). Participants were occupied with doing an odd-even-number judgment task for 2

(21)

minutes after encoding. The degree of reactivation measured by representational similarity analysis during this delay period in entorhinal and retrosplenial cortex was predictive for whether or not the respective picture was forgotten. Active rehearsing is an implausible explanation for these effects, and thus, these findings can be interpreted as some kind of reactivation of earlier encoding patterns. However, it remains unclear whether these events in a very short period after encoding reflect reactivation and memory consolidation as it occurs during sleep as they could also reflect ongoing task-related activity. A third study by Deuker et al. (2013) attempted to measure reactivations in fMRI during sleep and wake resting periods after participants did an object-location memory task. While they were able to detect some reactivation in the hippocampus in wake periods, they could not find reactivation during sleep. This, however, might have been caused by insufficient sleep duration of subjects in the MRI scanner as only 5 out of 17 subjects reached deep sleep stages or REM sleep. Also, they examined spontaneously occurring reactivations, which might be difficult to measure as their timing is unknown (Deuker et al., 2013). MVPA approaches are a promising tool to investigate memory reactivation in humans and allow further conclusions about the brain areas involved. The hippocampus, entorhinal cortex and retrosplenial cortex have been found to exhibit reactivated patterns during wake resting periods (Deuker et al., 2013; Staresina et al., 2013; Tambini & Davachi, 2013), which correlate with memory performance (Staresina et al., 2013). These findings all indicate an important role of the hippocampus and adjacent regions during reprocessing of previously learnt information. However, until now, no study has shown memory reactivation during sleep using these methods.

3.3 External manipulation of the reactivation process

Another way of studying memory trace replay is by externally triggering reactivation processes. This approach additionally allows the investigation of behavioral effects of reactivation on memory consolidation.

Barnes and Wilson (2014) showed in a recent study that it is possible to directly induce neuronal replay by electrical stimulation of neocortical areas involved in learning. They implanted electrodes in the piriform cortex in rats. The piriform cortex receives direct input from the olfactory bulb (Bekkers & Suzuki, 2013) and thus stimulation of the piriform cortex can be used as a conditioned stimulus in fear conditioning instead of presenting odors. After conditioning the rats with piriform cortex stimulation and foot shocks, the piriform cortex was stimulated again during sleep or the wake state, thus inducing artificial replay. Stimulation enhanced memory performance when it was administered during sleep but not during wake. Against the predictions, stimulation was not most effective at the peak of piriform sharp waves which could indicate that external manipulation might have interfered with naturally occurring reactivation at these points in time. This study shows that reoccurring neuronal activity during sleep in the brain region responsible for encoding increases performance in a memory task. The involvement of the hippocampus or other regions was not investigated in this study. However, the piriform cortex projects directly to the lateral entorhinal cortex (Kerr et al., 2007) so input by the piriform cortex could potentially have influenced the content of hippocampal replay which could have triggered replay in other brain regions.

Is reactivation of neuronal firing patterns limited by brain systems contributions?