Abstract
Memory reactivation during NonREM-ripples is thought to communicate new information to a systems-wide network. We now show a learning-specific increase in cortical ripple power associated with decreased hippocampal power across a hippocampal-prefrontal-parietal network, and increases in connectivity measured by Granger Causality. Disruption of either sleep or ripples impairs long-term memory consistent with a role for these ripples in memory consolidation.
Main Text
We only retain memories of what is new and relevant to updating our model of the world. But what makes new information salient enough for triggering memory consolidation processes? We recently proposed that dopamine coming from LC and VTA to the hippocampus could determine the fate of memories 1. Novel experiences sharing some commonalities with past ones (‘common novelty’) would activate the VTA and promote semantic memory formation via increased reactivation during sleep and ensuing systems consolidation1,2. By contrast, experiences that bear only a minimal relationship to past experiences (‘distinct novelty’) are thought to activate the LC to trigger strong sleep-independent, initial memory consolidation in the hippocampus, resulting in vivid and long-lasting episodic memories 1 2,3.
To test this, we compared different behaviors (Fig. 1A) to a non-learning Baseline condition: (1) Foraging, a mix of open-field foraging and track running with small chocolate rewards spread along a track. This controls for the effect of food rewards but contains no novelty. (2) Novelty, exploration of a new environment with novel cues/textures, a form of ‘distinct novelty’. (3) Plusmaze, training to a new reward location, which tests abstractions across multiple events (16 trials) in a familiar environment (with updated cues). Based on our hypothesis the ‘common novelty’ of the plusmaze condition should specifically lead to changes in post-behavioral sleep1,4.
We recorded 4h in rats after these four different conditions (Baseline, Foraging, Novelty, Plusmaze) and compared characteristics of NonREM-ripple events in the hippocampus. Surprisingly, after Plusmaze fewer ripples (across thresholds, Fig. S2) were detected than in the other conditions. Ripple count, rate of occurrence as well as duration showed a significant effect but the average frequency stayed the same (Fig. 1B). Next, we took the 300-500 largest ripple events (nr derived from maximum-count in Plusmaze for each animal, all ripples Fig.S3), and compared the corresponding oscillatory power in ECoCgs placed above the prefrontal and parietal cortex targeting the ripple-range (100-250 Hz) 5. A rmANOVA across conditions and brain areas (±25ms of ripple peak) showed a significant interaction (Fig. 1C, D). After Plusmaze a decrease in hippocampal and increase in cortical ripple power was detected.
Next, parametric Granger Causality analysis (GC) on the same ripple events, including all causality flows between hippocampus (HPC), prefrontal (PFC) and parietal (PAR) cortices (Fig. 2A; non-parametric GC Fig. S4) showed learning-specific effects when comparing Plusmaze to the other conditions. In the slower frequency ranges (0-20Hz) PFC→HPC and PFC→PAR showed an increase and PAR→PFC a decrease in GC values. In the faster frequency ranges (20-300Hz) both HPC→PFC/PAR and PFC/PAR→HPC showed increases.
The above analysis suggests that learning a new goal location in a Plusmaze changes cortico-hippocampal networks during NonREM-ripples. But is this activity necessary for long-term memory performance? To test this, we implanted animals with additional stimulating electrodes to the ventral hippocampal commissure (AP-1.3, ML 1, DV 3.8). Using similar methods others6,7 have shown that disrupting ripple activity daily (1h/d) slowed down learning in tasks trained over many days. Our one-session Plusmaze task allowed us to target a longer sleep period (4h) and compare this to a separate sleep deprivation group. Specifically, we compared 4h of sleep and sleep deprivation in unimplanted animals (within-subject, n=16, Fig. 3A) as well as sharp-wave-ripple-disruption (SWR-D), control-disruption (200ms after SWR, Con-D) and baseline (no stimulation) in implanted animals (within-subject, n=6, Fig. 3C). Animals performed above chance at 24h test (no food present) but performance fell to chance if sleep deprived after learning (Fig. 3B). SWR-D could mimic the sleep-deprivation effect, while both Baseline and Con-D showed above chance performance (Fig. 3D). Thus sleep and more specifically NonREM-ripples are necessary for long-term memory performance in this task.
In sum, we could show that ‘common novelty’ – extracting a new goal location in a familiar maze across multiple trials – induces changes across the hippocampal-prefrontal-parietal network during NonREM ripples in contrast to ‘distinct novelty’ or very familiar behaviors (Baseline, Foraging). Analysis revealed that ‘common novelty’ decreased the amount and power of ripples in the hippocampus, but increased the cortical response. Granger analysis of these ripple events showed increased prefrontal connectivity to both hippocampus and parietal cortex in the slower frequency ranges. In the faster frequencies both hippocampus to cortex and cortex to hippocampus was increased (Fig. 2C). Finally, ripple activity and sleep after learning in the Plusmaze is necessary for long-term memory performance in this task.
Interestingly, the observed network reconfiguration takes the shape of a bi-directional increase in LFP predictability, between the hippocampus and the neocortex, this may correspond to not only increased flow of replayed information from hippocampus to neocortex, but also in greater control exerted by the neocortex on the timing – and potentially the information content – of hippocampal ripples 8-10. The cortico-hippocampal interplay involve ripple-frequency LFP in neocortex as well as in the hippocampus. This may reflect variability in neocortical population activity, or the hippocampal ability to trigger local neocortical modes giving rise to neocortical ripples 5 and conversely the power of neocortical ripple to “broadcast” and influence hippocampal activity.
These results are the first direct experimental support for the hypothesis that different types of novelty affect sleep related consolidation differently1,2,11. Reactivations during sleep-ripples are thought to allow memory abstraction across multiple events, such as multiple trials or sessions in a learning task, and thus the consolidation from initial hippocampal to long-term cortical memory storage when we encounter something new that fits into what we know 12.
Acknowledgments
This work was funded by a Branco Weiss Fellowship – Society in Science to Lisa Genzel