Our project in centred around a model of how salient memories are preferentially consolidated by hippocampus. The mechanism we have proposed is that spontaneous cortical activity is biased towards specific networks during non-REM sleep upstates (preceding replay) , which we have described in our recent review (Lewis & Bendor 2019). One important aspect of the model, is that it provides an explanation for why Targeted Memory Reactivation (TMR) is able to influence memory consolidation- that it mirrors the normal biasing of hippocampal memories by cortical activity. Targeted memory reactivation is a phenomenon where task-related sounds played during sleep bias the hippocampus to replay the portion of the task associated with that sound preferentially (Bendor and Wilson 2012).
Following from this work, one important prediction related to this model, is that neurons in auditory cortex are capable of persistent activity. During TMR, the biasing effects caused by the sound on the hippocampus last many seconds beyond the termination of the sound, presumably because some neurons are capable of persistently firing, and providing a strong long-lasting input to the hippocampus. We discovered that auditory cortical neurons were indeed capable of persistent activity during passive listening, and outside the context of a behavioural task, indicating that this key mechanism was available for our model (Cooke et al, Scientific Reports, 2020).
Exploring this further, we examined the role of excitation and inhibition, combined with synaptic depression and facilitation in representing information within auditory cortex, as these are likely mechanisms that could be involved in cortical biasing of replay. These ideas were applied to a previously developed model in the lab which was used to investigate how auditory cortex encoded repeated sounds, finding that the strength and delay between excitation and inhibition could explain why only a subset of neuron expressed a temporal representation for slowly repeating sounds but were suppressed by quickly repeating sounds. However this previous model could not explain the phenomenon of rate coding that had been observed in many neurons that also used this temporal representation (Bendor and Wang 2007). Our model explains that this can occur simply by differences in the rate of synaptic depression between excitation and inhibitory neurons. If excitatory neurons adapt faster, than these neurons will decrease their response for fast temporal rate. If inhibitory neurons adapt faster, these neurons will show the opposite affect- increasing their response for fast temporal rates (Lee et al, PLoS Comput Biol, 2020).
To develop a behavioural model for examining cortico-hippocampal interactions during replay, we trained rats on a auditory discrimination task, combined with spatial navigation, while recording from the hippocampus using silicon probes. As a first step, we analysed theta oscillations during behaviour, and this dataset was later combined with a parallel dataset in ferrets, in a collaboration with the Bizley Lab (UCL). This was the first recordings in ferret hippocampus, which was directly compared to the dataset generated from our lab in rats, and recently published on BioRxiv (Dunn et al. 2021). We found that ferrets (unlike rats) did not have prominent LIA (large irregular activity) and sharp wave ripples during rest, but instead had a persistent form of theta (type 2 theta) that differed from the type of theta that normally occurs during locomotion. This suggests that ethological differences must be considered when trying to generalise observed phenomenon between species.
To record from large number of single neurons, we next developed several versions of hyperdrives (3D printed mechanically controlled tetrodes, targeting the hippocampus and other cortical regions). We first targeted the hippocampus to examine how information was represented by replay, followed by recordings from hippocampus and cortex simultaneously. Our hippocampal experiments examining replay, and how replay was affected by experimental variables were extremely fruitful. We identified that reward and recency (Tirole et al., in prep), as well as experience (Huelin et al., in prep) can influence how much an experience replays for during sleep. These experiments were well controlled by having the rat run on 2-3 different linear tracks, which varied in the amount of reward or experience the animal received, allowing us to directly compare replay rates during sleep for each track. In doing these experiments, we also identified that rate coding present during locomotion (firing rate changes for a place cell between tracks) is reinstated during offline replay.
Lastly, we have also developed methods for molecular-genetic interrogation of the causal interactions between cortex and hippocampus within the framework of memory consolidation (Aim 3). As a first step, we successfully tested complete hippocampal inactivation using DREADDs (Varela et al (PLoS One, 2016).
