The emphasis placed on the role of REM sleep in episodic memory encoding calls for a closer examination of neural circuitry involved in both episodic memory encoding and rapid eye movement (REM) sleep. A meta-analysis by Spaniol et al. (Reference Spaniol, Davidson, Kim, Han, Moscovitch and Grady2009) of functional magnetic resonance imaging (fMRI) studies has demonstrated robust involvement of hippocampus and other temporal lobe structures (e.g., amygdala, inferior temporal gyrus) in episodic memory encoding. Furthermore, this meta-analysis showed involvement of the left lateral prefrontal cortices in episodic memory encoding, both in ventral and in dorsal areas, compared with episodic memory retrieval. It is also of interest that pioneering fMRI work (Maguire et al. Reference Maguire, Valentine, Wilding and Kapur2003) in superior memorizers has shown the involvement of the retrosplenial cortex (below the posterior cingulate) besides the hippocampus during the visuospatial episodic memory strategies (“mental walk”) mentioned in the target article. If episodic memory encoding were a key feature of REM sleep, one would expect increased activity in hippocampus, lateral prefrontal cortices, and posterior cingulate/retrosplenial cortex.

Contrasting REM sleep with slow-wave sleep (SWS) or wakefulness reveals increased activity in basal ganglia, anterior cingulate, and anterior temporal lobe areas (Braun et al. Reference Braun, Balkin, Wesensten, Carson, Varga, Baldwin, Selbie, Belenky and Herscovitch1997; Maquet et al. Reference Maquet, Peters, Aerts, Delfiore, Degueldre, Luxen and Franck1996). Increased activity has been observed for hippocampus during REM sleep (although more pronounced in other regions) in the study by Braun et al. (Reference Braun, Balkin, Wesensten, Carson, Varga, Baldwin, Selbie, Belenky and Herscovitch1997), whereas Maquet et al. (Reference Maquet, Peters, Aerts, Delfiore, Degueldre, Luxen and Franck1996) did not observe increased hippocampus activity. In contrast, lateral prefrontal areas and posterior cingulate cortex show a decrease in activity from SWS to REM sleep, which we find difficult to reconcile with the main idea proposed in the target article.

These findings in relevant neural circuitry of episodic memory encoding are in accord with neural circuitry involved in more general memory processing, which is strongly correlated to hippocampus activity in the resting state. The hippocampus is embedded within a functionally related brain network referred to as the default mode network (DMN), comprising anterior and posterior cingulate, precuneus/retrosplenial cortex, and inferior parietal lobules. The hippocampal formation (hippocampus and adjacent temporal lobe regions) is strongly connected to the DMN in wakefulness, particularly during episodic memory processing and future imagination (Buckner et al. Reference Buckner, Andrews-Hanna and Schacter2008). It is essential to keep in mind that REM sleep recruits only part of this episodic memory network, specifically the anterior part (Braun et al. Reference Braun, Balkin, Wesensten, Carson, Varga, Baldwin, Selbie, Belenky and Herscovitch1997; Maquet et al. Reference Maquet, Peters, Aerts, Delfiore, Degueldre, Luxen and Franck1996).

The relevance of findings on neural circuitry are stressed by cognitive effects in the same direction: there are fewer episodic memory traces after waking from REM sleep than from non-REM sleep. In contrast, semantic and abstract self-references are more frequent memory sources in REM sleep (Baylor & Cavallero Reference Baylor and Cavallero2001). These observations require further attention in the light of the proposed hypothesis in the target article.

We would further like to emphasize the importance of distinguishing activity of a brain region from its connectivity. Even if hippocampus activity is increased in REM sleep, this would not necessarily equate to increased functional connectivity between hippocampus and neocortex, which could represent increased flow from a large variety of neocortical regions to hippocampus. Findings in healthy human subjects providing whole-brain results are still lacking because of the difficulty of investigating REM sleep with the combined neuroimaging methodology required for high spatial and temporal resolution (Wehrle et al. Reference Wehrle, Kaufmann, Wetter, Holsboer, Auer, Pollmächer and Czisch2007). Interestingly, neurophysiological studies in rodents have shown the greatest hippocampus synchrony during non-REM sleep hippocampal sharp-wave ripples, and lowest intrahippocampus synchrony in REM sleep (Grosmark et al. Reference Grosmark, Mizuseki, Pastalkova, Diba and Buzsáki2012). It is intriguing that these authors further observed an increase in synchrony from non-REM period to non-REM period, which was correlated with the power of theta activity during intervening REM episodes. This can be taken as evidence for synaptic downscaling processes occurring in REM sleep.

Studies on the effects of REM sleep deprivation are also in line with the aforementioned findings. For instance, preclinical work has shown that REM sleep deprivation in rats results in impairments in hippocampus-independent cued fear extinction but not in hippocampus-mediated contextual fear extinction (Silvestri Reference Silvestri2005). This suggests affective processing of visuospatial aspects (e.g., light was paired with foot shocks but is now safe) rather than of contextual aspects. Alternatively, it could be taken as an argument for processing of generic rather than specific features. With increasing evidence for a role of SWS in declarative and episodic memory processing (Diekelmann & Born Reference Diekelmann and Born2010), a function of REM sleep may well lie in more generic processing of information to enable more efficient reorganization into schemas (as has recently been proposed for SWS [Lewis & Durrant Reference Lewis and Durrant2011]). As such, comparison/integration of information with existing schemas in neocortex appears more plausible than linking episodic traces in hippocampus. The role of the hippocampus in REM sleep could be to provide incidental output to neocortex through sporadic bursts of activity (Montgomery et al. Reference Montgomery, Sirota and Buzsáki2008). Such a view would be more in line with the aforementioned imaging, cognitive, neurophysiological, and functional findings. Further experimental work is warranted, and the literature has provided elegant solutions for testing associative processes in the light of REM sleep (Cai et al. Reference Cai, Mednick, Harrison, Kanady and Mednick2009), and in our opinion these solutions are more informative and better to combine with imaging methods than are subjective or nonstandardized measurements.

Finally, the graph theory approaches mentioned in the target article have become popular for analyzing functional connectivity data, reducing the brain's complexity to a network of nodes and edges (the connections between nodes). Llewellyn refers to the increased connectivity observed in light non-REM sleep, which we would like to specify as increased corticocortical connectivity, since thalamocortical connectivity was strongly reduced in light non-REM sleep because of the thalamus being removed from the whole-brain network at sleep onset (Spoormaker et al. Reference Spoormaker, Schröter, Gleiser, Andrade, Dresler, Wehrle, Sämann and Czisch2010). This demonstrates how the behavior of one critical hub can change the behavior of the whole-brain network; alternating hippocampus-neocortex connectivity from SWS to REM sleep could critically impact general network functioning and information processing. To date, all we can say is that more experimental neuroimaging work in humans is needed.