The target article makes a parallel between rapid eye movement (REM) sleep dreaming and principles of the ancient art of memory (AAOM), thus proposing a role for REM sleep dreaming state in elaborative encoding of episodic memories. First, in the target article, Llewellyn argues that REM sleep dreaming renders recent memories more distinctive through novel and meaningful associations with emotionally salient and remote memories. Second, Llewellyn goes on to predict how recent episodic memories are further configured in the brain through dreaming imagery. This concept further challenges the enigmatic nature of REM sleep and its dreaming. Similar to previous theories of sleep, the inevitable reductionism inherent in the concept of the target article leaves unresolved questions. The critical issue for milestone hypotheses for REM sleep dreaming is therefore to avoid as much as possible “surprise,” “free energy” costs, or redundant “prediction errors,” as posited by another recent concept, according to which REM sleep dreaming and phenomenology may subserve basic adaptive functions beyond memory reprocessing such as free energy minimization (Hobson & Friston Reference Hobson and Friston2012, p. 87).

Even if the AAOM principles are regarded as a focus in the cognitive domain in the context of already existing more fundamental theories of REM sleep (e.g., Hobson Reference Hobson2009; Hobson et al. Reference Hobson, Pace-Schott and Stickgold2000; Jouvet Reference Jouvet1998), there is still experimentally driven and fact-based information that needs to be reconciled.

Recent human studies indicate that REM sleep electroencephalographic (EEG) signatures are not merely associated with memory encoding and consolidation. Whereas theta (5- to 7-Hz) REM sleep electroencephalographic (EEG) activity has been associated with both dream recall from REM sleep (Marzano et al. Reference Marzano, Ferrara, Mauro, Moroni, Gorgoni, Tempesta, Cipolli and De Gennaro2011) and consolidation of only emotional memory (Nishida et al. Reference Nishida, Pearsall, Buckner and Walker2009), gamma (more than 25- to 30-Hz) REM sleep EEG activity has been assumed to be associated with affective mental states (Marshall et al. Reference Marshall, Kirov, Brade, Mölle and Born2011; van der Helm et al. Reference van der Helm, Yao, Dutt, Rao, Saletin and Walker2011b). Hence, frequency-specific EEG activities of human REM sleep (Cantero et al. Reference Cantero, Atienza, Stickgold, Kahana, Madsen and Kocsis2003; Llinás & Ribary Reference Llinás and Ribary1993) may support its functions dissimilarly. A recent rodent study indicates that theta oscillations during REM sleep may produce synaptic downscaling in the hippocampus (Grosmark et al. Reference Grosmark, Mizuseki, Pastalkova, Diba and Buzsáki2012). Along with the observed association between human theta EEG rhythm and successful dream recall in REM sleep (Marzano et al. Reference Marzano, Ferrara, Mauro, Moroni, Gorgoni, Tempesta, Cipolli and De Gennaro2011), this finding conforms well with either the kind of elaborative encoding proposed in the target article or the need for room to minimize the free energy (Hobson & Friston Reference Hobson and Friston2012). Yet, specifically regarding the target article's concept, a possibility remains that sleep-dependent memory reprocessing may be an epiphenomenon of more basic functions of both non-REM sleep (Tononi & Cirelli Reference Tononi and Cirelli2006) and REM sleep (Hobson & Friston Reference Hobson and Friston2012).

Sleep studies of human insight have demonstrated that whereas early-night slow-wave sleep (SWS) and its EEG signatures support the transformation of implicit knowledge into explicit insight to a hidden regularity in a complex generation task, late-night REM sleep preserves only the previously generated implicit knowledge without engaging specific REM sleep EEG signatures (Yordanova et al. Reference Yordanova, Kolev, Verleger, Bataghva, Born and Wagner2008; Reference Yordanova, Kolev, Wagner, Born and Verleger2012). These results show that only SWS and its EEG signatures may be linked to successful human heuristic creativity and do not suggest memory reconstruction during REM sleep.

Critically, nearly all psychiatric disorders are characterized by a REM sleep overdrive, yet these psychiatric conditions can not be linked to successful memory formation (Benca et al. Reference Benca, Obermeyer, Thisted and Gillin1992; Walker Reference Walker2010). Further, it has been shown that augmented REM sleep in children with attention-deficit/hyperactivity disorder is associated with the core psychiatric symptoms of inattention and hyperactivity/impulsivity, whereas the greater amount of REM sleep in normally developing children is associated with better performance intelligence (Kirov et al. Reference Kirov, Kinkelbur, Banaschewski and Rothenberger2007; Reference Kirov, Uebel, Albrecht, Banaschewski and Rothenberger2011). These findings suggest at least a bidirectional role of REM sleep, depending on presence or absence of psychopathology.

From a developmental perspective, infants experience much more REM sleep, which decreases through childhood and adolescence, than adults (Roffwarg et al. Reference Roffwarg, Muzio and Dement1966). Thus, considering that REM sleep neurobiology and the corresponding dream generation are tightly coupled (Stickgold et al. Reference Stickgold, Hobson, Fosse and Fosse2001), explanations are needed for the following: (1) What memory sources may create the presumably enormous dreaming during infancy and early childhood? (2) What cognitive and/or adaptive functions does such dreaming subserve? (3) How does this risky homeostatic state (e.g., Hobson & Friston Reference Hobson and Friston2012) predominant during early development relate to evolutionary advantages? (4) What kind of memory processes may be subserved by REM sleep dreams incorporated in nightmares, a common feature of both normal ontogenesis and developmental psychopathology (Brand & Kirov Reference Brand and Kirov2011; Kirov & Brand Reference Kirov and Brand2011)? Although it has been assumed that REM sleep subserves brain maturation during ontogenesis through internally generated and genetically grounded stimulation of neuronal assembles (Marks et al. Reference Marks, Shaffery, Oksenberg, Speciale and Roffwarg1995), this assumption mandates experimental updates by applying new paradigms accounting for REM sleep hypotheses such as genetic programming (Jouvet Reference Jouvet1998), threat simulation (Revonsuo Reference Revonsuo2000), and protoconsciousness (simple awareness of perception and emotions provided by genetic equipment ensuring generation of a virtual world) or Bayesian learning and inference (Hobson Reference Hobson2009; Hobson & Friston Reference Hobson and Friston2012).

Human REM sleep occurs periodically, thus distinguishing sleep cycles from one another, with the frequency and duration of REM sleep periods increasing progressively across overnight sleep (Broughton Reference Broughton, Niedermeyer and Lopes da Silva1987). These temporal dynamics of REM sleep and dreaming can not be readily linked to the concept of elaborative encoding through the AAOM principles forwarded in the target article. Instead, it seems more relevant that in combination with concomitant neuroendocrine (Steiger Reference Steiger2002, Reference Steiger2007) and cardiorespiratory processes (Gastaut & Broughton Reference Gastaut and Broughton1964), the temporal dynamics of REM sleep and dreaming could supply vitally important homeostatic functions together with many cognitive adaptive processes.

I argue that the periodic occurrence of REM sleep and its dreaming may be regarded as a recurrent adaptive interference that may incorporate previously encoded and consolidated memories into a broader vital context comprising residuals of hypotheses testing, related emotions, basic needs, and individual genetic traits.