A reciprocal influence between sleep and memory has been suggested, insofar as consolidation of newly acquired information is facilitated by previous sleep periods, whereas acquisition of new information alters sleep pattern (Maquet, 2001; Stickgold et al., 2001; Smith et al., 2004; Walker and Stickgold, 2004, 2005). Additional evidence of this relationship has been given by human and animal studies demonstrating that training in different memory tasks increases sleep time (Lucero, 1970; Smith et al., 1980; Smith and Lapp, 1986; Portell-Cortés et al., 1989; Smith and Rose, 1997) and that either sleep- (SD) or REM sleep deprivation (REMSD) before training impairs the performance of animals in numerous hippocampal-dependent tasks, such as inhibitory avoidance (Stern, 1971; Bueno et al., 1994, 2000; Gruart-Masso et al., 1995; Moreira et al., 2003; Dubiela et al., 2005), multiple trial inhibitory avoidance (Moreira et al., 2010; Ota et al., 2013), Morris water maze (Smith and Rose, 1996; Guan et al., 2004) and fear conditioning (Hicks et al., 1988; Dametto et al., 2002; McDermott et al., 2003; Tiba et al., 2008). However, results from studies using pre-training protracted REMSD protocols are difficult to interpret because the animals are under an altered sleep-waking pattern both before (sleep-deprived) and after training (when sleep rebound is taking place) thus, precluding conclusions as to whether this manipulation affects acquisition and/or consolidation of the information in these memory tasks. In addition, REMSD-induced memory impairment may be due to its effect on attention, since this is disrupted by 72 or 96 h of REMSD and recovered by 24 h of sleep rebound (Godoi et al., 2005), suggesting that acquisition may be compromised. Indeed, sleep-deprived rats require more trainings in the multiple inhibitory avoidance task than control rats, and even so, their performance is poorer when tested 24 h later (Moreira et al., 2010; Ota et al., 2013). Finally, the possibility exists that impairment of performance in the abovementioned studies is due to a phenomenon known as state-dependent learning. This phenomenon implies in learning a particular response under a given stimulus-situation that involves a conditioning aspect; in test situations, when the training stimulus is replaced by another one, there is no transfer of the learned response (Overton, 1974). Therefore, in state-dependent learning a particular response is displayed if, and only if, there exists a similarity of the animal’s inner state between the training and testing conditions (Colpaert et al., 1976).

Studies employing passive avoidance task show that 24 h of sleep recovery are sufficient to restore the performance of 96 h REM sleep-deprived rats (Dubiela et al., 2005, 2010), suggesting that SD-induced performance deficit may not be due to impairment of consolidation, but rather to some other unspecific effect, such as altered motor performance. In fact, REMSD induced-increased locomotor activity in rats has been reported by some authors in rats (Suchecki et al., 2002; Tartar et al., 2009) and mice (Armani et al., 2012), but not by others (Dubiela et al., 2011); in the case of passive avoidance task, altered motor activity could be responsible for the short- (2 h post-training), but not for the long-term (24 h post-training) retention deficit (Dubiela et al., 2005), indicating, once again, that acquisition is impaired by REMSD.

One strategy to circumvent the impact of SD on acquisition is to sleep-deprive animals after training, so acquisition is guaranteed. The studies that employed this approach resulted in controversial results, with impairing (Graves et al., 2003; Palchykova et al., 2006), as well as improving effects (Gisquet-Verrier and Smith, 1989; Smith, 1996; Smith et al., 1998). This discrepancy may be attributed to the length (5 h/6 h × 24 h) and to the type of SD (total × REMSD). Moreover, the interval between SD and testing is an important methodological issue that should be considered when evaluating its effects in different memory stages. For instance, Silva et al. (2004) showed that 72 h of REMSD after training did not induce memory impairment in passive avoidance test or in the plus maze discriminative test when testing took place immediately after the end of SD, but when animals were re-tested 1 week later, they displayed impaired performance. Whether this impairment represents a facilitated extinction of the fear memory was not explored in this study, but there is evidence that 6 h of REMSD retard extinction of cued fear memory (Silvestri and Root, 2008).

Given the controversy regarding the effects of protracted SD on mnemonic process, the aim of the present study was to investigate the consequences of this manipulation, applied in different moments of memory formation in two related tasks that recruit distinct neural circuits, contextual fear conditioning (CFC), a hippocampal-dependent task, and tone fear conditioning (TFC), a hippocampal-independent task. Moreover, we also tested whether the performance impairment observed in these tasks could be explained by the state-dependent learning.

The results of the present study showed that SD affected memory acquisition and retrieval, but not consolidation; this effect was not induced by a state-dependent learning and exposure of animals to SD only immediately before the first retrieval session impaired the extinction of a hippocampus-dependent task.

In Experiments 1, 2 and 3, we investigated whether SD-induced impairment of recall in aversively motivated tasks, in the present case, CFC and TFC, resulted from interference with memory consolidation and/or recall per se. The results of these experiments strongly suggest that SD impairs recall, but not consolidation, because the performance of sleep-deprived rats in the TFC task was restored to control values after 24 h of sleep recovery, whereas in the CFC, the same effect was obtained between 24 and 96 h of sleep recovery. Interestingly, previous studies have shown that short periods of SD immediately (Graves et al., 2003; Palchykova et al., 2006) or 1 h (Prince et al., 2014) after training in hippocampus-dependent tasks impair memory consolidation and, consequently, recall. One possible explanation for the lack of SD effect on consolidation, may reside in the amount of associations made between footshock and either context or tone; in Graves et al. (2003), mice received one footshock of 1.5 mA, whereas in the present one, rats received five footshocks of 0.8 mA, and there is a possibility that memory consolidation begins already after the first pairing. Nonetheless, this possibility appears to be true for CFC, whereas data on the effects of SD on TFC are controversial, with studies showing this task to be resistant to the disrupting effects of REMSD (Bueno et al., 1994; Ruskin et al., 2004; Ruskin and Lahoste, 2009) or SD (Graves et al., 2003) and others showing that it is also affected by SD (Dametto et al., 2002; Kumar and Jha, 2012). These findings may explain why a shorter period of sleep recovery was already enough to restore the animals’ performance in TFC, whereas in CFC, a period between 24 h and 96 h was required for animals to express their recall-related behavior.

It has been shown previously that 96 h of SD by the platform method leads to intense sleep rebound in the 24 h following the deprivation period (Machado et al., 2004, 2008, 2013) resulting in disorganized sleep pattern characterized by excessive compensation of REM sleep and reduced non-REM sleep. Impairment of recall could also be attributed to this altered sleep architecture as recent studies have shown that non-REM sleep is also important for memory performance in emotional tasks (Giuditta et al., 1995; Fogel and Smith, 2006; Hellman and Abel, 2007). By allowing animals to sleep for 96 h between SD and testing, sleep pattern as well as the performance of sleep-deprived rats returned to normal on both tasks, being identical to that of control rats, thus evidencing that SD affects retrieval, but not memory consolidation. In addition, these data indicate a dissociative effect of SD on CFC and TFC, reinforcing the proposition that the impact of SD is not unspecific (as, for instance, on motor activity) since both tasks require the same response: freezing behavior.

The results of Experiment 4 indicated that memory impairment in both CFC and TFC was not due to a state-dependent learning. If it were, then animals that were sleep deprived before training and before testing should display a performance similar to that of control animals. However, the performance in both tasks was impaired in animals sleep-deprived either before training and/or testing. These results are in contrast with a previous one (Patti et al., 2010), that investigated the state-dependent effects of 72 h of REMSD in mice. Possible sources of controversy may be the use of different species, length of SD, but most of all, differences in the experimental protocol, for in Patti et al.’s study, testing was performed 7 days after training, a time sufficient for the animals to sleep recover and, according to the present results, to restore the performance in hippocampus-dependent tasks. In the present study, there was no interval between training and testing other than the period involving SD.

Just as REMSD applied before training may impair performance by altering attention state during acquisition so can the same phenomenon be true for retrieval. As, to the best of our knowledge, no study had ever been performed to evaluate the effects of SD on attentional process during retrieval, this possibility cannot be ruled out. Thus, even if acquisition and consolidation phases were guaranteed (Experiments 1B and C), pre-test SD might still affect unspecific aspects of performance rather than impairing retrieval per se. Importantly, during retrieval, memories become labile and may undergo two different processes: one that weakens old memories as a result of extinction and another that strengthens the memory as a result of additional consolidation, a process known as reconsolidation. These processes share common mechanisms but also recruit distinct circuits (for review, see (Alberini, 2005). Carrying out subsequent tests, in which the animals would be no longer sleep-deprived, could be a useful strategy to assure that normal performance was restored and to evaluate specific aspects of memory retrieval. Thus, in Experiment 5, animals were sleep-deprived either immediately before or after the first test, in order to investigate whether the manipulation would interfere with post-learning mechanisms. The results confirmed the impairing effects of SD before testing, but fear memory was expressed again in a subsequent test of both CFC and TFC, when rats had already recovered from SD (test 2). As expected, all groups, except from that sleep-deprived before test 1 of CFC exhibited low levels of freezing behavior upon repeated exposures to the CS, indicating that SD before recall of hippocampus-dependent fear memory had a long-term effect on memory processing. This result is unlikely to reflect SD-induced fear memory erasure, since SD when not followed by testing (G4 in Experiment 5) did not affect memory extinction. Interestingly, although freezing behavior was not expressed by rats of G2, their memory became labile, allowing some kind of alteration to take place leading to persistence of the fear memory. Several possibilities can be offered to explain the effect of SD conducted just prior to the first test. The first one refers to the acute effect on memory retrieval: although this may be the case, it is not possible to dissociate this effect from the confounding effect of SD on motor performance or emotionality. Usually a retrieval deficit can be inferred when the manipulation affects only one task, but in the present experiments both tasks were affected. Nonetheless, the long lasting effect of SD selectively observed in CFC cannot be attributed to an acute influence on recall or motor performance. The second one refers to the impairing effect of SD on extinction or, alternatively, on reconsolidation, or both. As control rats extinguished the conditioned response it is reasonable to assume that sleep manipulation interfered with this phenomenon. Because stressful situations disrupt reconsolidation (Cai et al., 2006) and SD induces sustained increase of corticosterone plasma levels (Galvão et al., 2009), it is unlikely that SD would produce a positive effect on reconsolidation of fear memories.

Pre-test SD affected memory extinction only for CFC, but not for TFC task, a result that is in contrast with previous work reporting that REMSD affects only TFC but not CFC (Silvestri, 2005; Fu et al., 2007). Some methodological differences may explain this contradiction. In Silvestri’s work, rats were submitted to 6 h of REMSD immediately after training, whereas testing/extinction training occurred 48 h later. Sleep-deprived animals did not extinguish fear behavior in TFC, but freezing was reduced in CFC. In a subsequent test, no difference was found in TFC, and all animals expressed similar levels of fear behavior, whereas there was a tendency to increase freezing in REMSD animals in CFC. This later result would agree with the present data, in which SD animals presented impaired extinction in subsequent tests only in CFC. Importantly, in Silvestri’s work training/testing and extinction for TFC were carried out in the same chamber, while in the present study TFC test and extinction was performed in a chamber different from the training one. The importance of context on modulation of extinction performance has been reviewed elsewhere (Bouton et al., 2006). Although Fu et al.’s work took this aspect into consideration, the same result was reported, i.e., impaired extinction in the TFC but not in the CFC after SD. There are still some methodological differences among these studies that could account for the distinct outcomes reported, including habituation prior to training (used in both studies–20 and 30 min respectively—but not in the present one), that could facilitate contextual extinction, masking possible effects of REMSD on extinction of this task, different lengths of SD (6 h after training in both studies), interval between training and testing (shorter periods in the previous studies and 96 h in the present one) and different methods of SD. Naturally, it is only possible to establish which factor(s) could influence the outcomes with certainty, by performing a study that compares all these variables.

Nevertheless, despite the inconsistencies between the abovementioned studies and the present one, the outcomes seem to convey the message that sleep loss can be detrimental for extinguishing aversive memories. This conclusion is in accordance with reports in human beings who had REMS periods during a 90-min nap and displayed less reaction to fearful facial expressions and more positive reactions to happy faces (Gujar et al., 2011) and a report that individuals who were involved in automobile accidents were more likely to develop depression and posttraumatic stress disorder (PTSD) when experienced sleep disturbances on the 2 weeks prior to the traumatic event (Bryant et al., 2010); conversely, individuals who displayed consolidated REM sleep in the aftermath of a traumatic situation are less likely to develop PTSD (Mellman et al., 2007). Collectively, these findings agrees with the proposition that REM sleep facilitates reframing of negative experiences and restores emotional reactivity (Walker and Van Der Helm, 2009). Therefore, experimental protocols such as those used in Silvestri (2005), Fu et al. (2007) and in Experiment 3 are promising tools to study the neural circuits involved in the relationship between REM sleep and extinction of fear memory.

In conclusion, 96 h of SD immediately after training of aversive memory tasks impaired the recall of fear memory both in context and TFC and this effect was reversed after a period of sleep recovery, but did not seem to result from a state-dependent learning process. However, when SD took place immediately before testing, there was an impairment of extinction of the aversive memory only in CFC.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.