Predictability is an important factor in the effects of stress and a preference for predictability has been repeatedly demonstrated (French et al. 1972; Gliner 1972; Miller et al. 1974). For example, animals given the opportunity to determine whether shocks delivered to them will be signaled or unsignaled typically choose to spend their time in the signaled conditions regardless of whether the shock is escapable or inescapable [reviewed in (Badia et al. 1979)]. The strong behavioral effects suggest that predictability may also have a role in the modulating effects of stress on sleep. In fact, stressor predictability is a significant component in shock avoidance training in a shuttlebox, a paradigm in which laboratory rats are signaled of imminent shock and can learn to prevent shock from being delivered. Variants of this paradigm have often been used in studies of learning and sleep and have typically found increases in the amount of REM sleep at various latencies after training that have been interpreted as indicating a role for REM sleep in memory consolidation (e.g. Smith and Lapp 1986; Datta 2000). Unfortunately, the potential role of predictability in modulating sleep after stress has received very little attention.

One study of stressor predictability in mice examined sleep after training with signaled escapable shock (SES) and signaled inescapable shock (SIS) (Yang et al. 2011a). Compared to mice experiencing SIS, those experiencing SES showed significantly increased REM sleep after each of two shock training sessions whereas compared to mice experiencing SES, those experiencing SIS showed significantly increased NREM sleep after both shock sessions. Interestingly, groups receiving either SES or SIS showed reduced REM sleep in response to cue presentation alone. In another study, mice exposed to either predictive or non-predictive auditory cues during training with ES also showed post-stress increases in REM sleep (Machida et al. 2013). However, a subgroup of mice (around 35 %) trained with the predictive auditory cue failed to improve their escape performance from the first to second day of training. Those mice that did not improve also did not show enhanced REM on either shock training day, suggesting a learning component in the alterations in REM sleep.

It is useful to compare these results to those obtained using non-signaled escapable and IS used to model controllable and uncontrollable stress as described above (Sanford et al. 2010). Without predictive cues, the relative differences in post-stress REM after escapable and IS were more pronounced. Contexts associated with non-signaled escapable and IS also produced directionally different changes in REM similar to those seen when shock was presented (Sanford et al. 2010), whereas predictive cues associated with either escapable or IS produced similar reductions in REM sleep.

While at this point the data are limited, these findings suggest that contexts and auditory cues associated with different shock training conditions may carry different, and potentially competing, types of information regarding the stressful situations. This difference is more pronounced in training with ESs as both contextual and cued fear associated with uncontrollable stress have similar effects on sleep in mice and both reduce REM (Sanford et al. 2003a, c). Thus, competing cued and contextual information associated with ESs may have interacted during training resulting in competing influences on REM, thereby suggesting that stressor predictability and controllability may interact in complex ways to modulate the changes in subsequent sleep.

In addition to producing direct physiological effects, stressful situations provide an opportunity for learning as the individual responds to the stressor and seeks to use available information to cope with the ongoing challenge the stressor imposes. While in humans this could involve a variety of activities, including higher order cognitive processing; in rodents, the simplest behavioral responses to a stressor may be avoidance or escape attempts. In this case, stress-related parameters such as controllability and predictability may provide useful information that shapes avoidance and escape behaviors thereby facilitating successful coping. By comparison, stressors that are uncontrollable or occur unpredictably do not provide information that can guide the animal to learn successful avoidance or escape behaviors. In these situations, the animal may still engage in escape attempts, but its behavior will not alter the presentation of the stressor or facilitate coping.

The impact of stressor controllability and predictability on behavior are central to a number of well-established learning paradigms that are motivated by stressful events. Of these, fear conditioning and related paradigms are beginning to demonstrate that the learning options an animal has in a stressful situation play a significant role in determining the impact of stress.

Experimental fear conditioning is a learning paradigm in which an animal makes an association between an uncontrollable stressor (usually footshock) and previously neutral cues (typically auditory) or contextual information (the test box and experimental room). Afterwards, presenting the fearful cues and contexts alone elicit physiological and behavioral fear responses similar to those produced by the initial uncontrollable stressor. Fear conditioned alterations in sleep are now also established though these can vary with the amount of training and with the strain of rats or mice that is studied. In agreement with the data in previous sections showing that uncontrollable footshocks reduce REM sleep, the primary and most consistent effect of extensive training with inescapable footshock is a marked reduction in REM sleep that occurs both after the shock training and after presentation of shock associated fearful cues and contexts (Sanford et al. 2003a, c; Tang et al. 2005d). This reduction in REM sleep has been reported across species and across strains (Sanford et al. 2001, 2003a, c) and can occur without the rebound or recovery REM sleep that has been reported for most stressors (Sanford et al. 2003a, c). Changes in NREM sleep in fear conditioning studies appear to be less consistent. Some studies have reported increases in (light) NREM sleep (Adrien et al. 1991), whereas others have shown strain-dependent reductions in NREM sleep after shock training and fearful contexts (Sanford et al. 2003a). There also may be relatively less NREM sleep EEG delta power (slow wave activity) in animals that show greater fear conditioned changes in sleep (Tang et al. 2006). Critically, these studies demonstrate that fear conditioned memories of stressful events can produce mostly the same changes in sleep as those produced by the stressful event itself and indicate the importance of learning in both the immediate and lasting effects that stress can have on sleep.

Fear extinction is another important type of stress-related learning. While fear conditioning can be involved in the long term, negative effects of stress, it also can underlie adaptive behavior that occurs only so long as the fear-inducing stimulus is predictive of, or associated with, an aversive event (Kishimoto et al. 2000; Pitman et al. 2001). Repeated presentation of a fearful cue or context without shock results in fear extinction, a type of new learning that inhibits subsequent fear behavior without erasing the original memory for fear conditioning (Bouton 2004). It is the failure of extinction that has been linked to stress-related psychopathology, particularly posttraumatic stress disorder (PTSD) (Myers and Davis 2007).

The processes that make fear behaviors resistant to extinction remain mostly unknown though there appears to be a relationship between fear extinction and post-training REM sleep. Post-training REM sleep deprivation has been reported to impair extinction (as indicated by freezing) for light cues (Silvestri 2005), but not for auditory cues (Fu et al. 2007) previously paired with shock. REM sleep-deprived rats did show greater spontaneous recovery of freezing on a second day with presentation of the fearful auditory cue alone. Neither of these studies found that post-training REM sleep deprivation significantly altered contextual fear extinction learning or spontaneous recovery of freezing on a second day of testing (Silvestri 2005; Fu et al. 2007). However, sleep (both NREM and REM) following extinction of contextual fear does return to normal, whereas rats that continued to show fear exhibited reductions in REM sleep (Wellman et al. 2008).

Genetic differences in vulnerability and resilience are recognized as important factors in the development of stress-related pathology. For example, approximately 20–30 % of individuals who experience traumatic events may develop PTSD whereas others do not appear to suffer significant long-lasting effects (Cohen et al. 2003; Kerns et al. 2004). Attempts to develop animal models that better represent individual differences in clinical populations have included the selection of low and high responders to stressors in outbred rat strains (Cohen et al. 2003; Kerns et al. 2004). There is also evidence that differences in vulnerability are a factor in the impact of stress on sleep, but the significance of individual differences has not been fully appreciated in either studies of stress or in studies of sleep in general (Irmis et al. 1971, 1974; Tang et al. 2007).

Some of the best evidence for the role in resilience and vulnerability in the impact of stress on sleep comes from studies comparing inbred strains of rodents, which are genetically identical within strain but which vary genetically and phenotypically across strain. Work in mice and rats has demonstrated that strains that exhibited greater anxiety-like behaviors in response to challenges in wakefulness exhibited correspondingly greater and longer duration alterations in sleep after training with IS and after fearful cues (Sanford et al. 2003c) and contexts (Sanford et al. 2003a) associated with IS. In general, vulnerable mouse strains (e.g., BALB/cJ mice compared to more resilient C57BL/6 J mice) also showed greater decreases in sleep after stressful situations with unlearned responses, including exposure to an open field (Tang et al. 2004), cage change, and novel objects placed in the home cage (Tang et al. 2005b). Moreover, BALB/cJ mice also do not show a significant REM sleep increase during recovery from restraint stress, whereas C57BL/6 J mice do (Meerlo et al. 2001b).

In addition to genetic determinants of individual resilience and vulnerability, environmental factors and prior experiences with stress also play a major role in shaping future responses to stressful challenges. One such factor is neonatal stress that can be induced by disruptions of mother–infant relationship and that can have repercussions for adult behavior (Levine 2005) though the specific effects of maternal separation on sleep have varied across studies. For example, 3 h of maternal separation from postnatal days 2–14 in rats has been reported to increase both spontaneous baseline REM sleep and cold-stress-induced changes in REM sleep in males (Tiba et al. 2004) and females (Tiba et al. 2008). Similarly, neonatal rats maternally separated for 3 h and exposed to chronic mild stress as adults were reported to show longer sleep time, more episodes of REM sleep, and more episodes of NREM sleep transitioning to REM sleep (Mrdalj et al. 2013). By comparison, 6 h of neonatal maternal deprivation reduced the time spent in REM, without changes in NREM sleep when the rats attained adulthood (Feng et al. 2012). In addition to differences in experimental procedure, another important aspect that may differentiate these studies is the strain of rats used. While Feng and co-workers used Sprague-Dawley rats, the other studies used Wistar rats, which display more maternal behavior upon reunion with their litters (Lehmann and Feldon 2000), possibly buffering potential harmful effects of the separation procedure.

The regulation of sleep and arousal involves multiple neurotransmitter systems as well as excitatory and inhibitory amino acids, peptides, purines, and neuronal and non-neuronal humoral (i.e., cytokines and prostaglandins) modulators (Steiger and Holsboer 1997; Steiger et al. 1998; Jones 2005; Steiger 2007; Luppi 2010; Espana and Scammell 2011). Many of these same neurotransmitters and neuromodulators are also influenced by and/or mediate the effects of stress and are likely involved in the effects of stress on sleep. This section will briefly review some of the major neurochemical systems that link stress and sleep.

As discussed in the above section on stress mediators, there are several points of overlap in the neural regions/systems involved in stress and those directly involved in arousal. This section will focus on the amygdala and medial prefrontal cortex (mPFC), two regions not typically considered as direct regulators of arousal and sleep but which play significant roles in mediating the effects of stress on sleep and arousal (see Fig. 2).

Both NREM and REM sleep have putative roles in regulating neuronal plasticity and synaptic strength (Benington and Frank 2003; Tononi and Cirelli 2006; Meerlo et al. 2009; Havekes et al. 2012). Stress-induced changes in sleep and sleep architecture might lead to alterations in these plasticity processes and ultimately brain function. In fact, some of the changes in plasticity and brain function traditionally linked to stress may in part be related to alterations in sleep.

Work on stress and plasticity has distinguished the effects of acute and chronic stress. Acute stress can impact functional plasticity whereas chronic stress can differentially alter structural plasticity across brain regions. For example, chronic stress results in dendritic atrophy and reductions in spine density in the hippocampus (Magarinos and McEwen 1995a, b; Magarinos et al. 1997; Sandi et al. 2003; Stewart et al. 2005) and prefrontal cortex (Wellman 2001; Cook and Wellman 2004; Radley and Morrison 2005; Liston et al. 2006; Radley et al. 2006). Similar types of chronic stress produce increased dendritic arborization and increased spine density in BLA spiny neurons (Mitra et al. 2005; Vyas et al. 2006) and spine down-regulation in the medial amygdala (Bennur et al. 2007). Some stress-induced changes in the hippocampus (Sandi et al. 2003; Stewart et al. 2005) and prefrontal cortex appear to be reversible whereas those in the amygdala are not [Reviewed in (Christoffel et al. 2011)]. Acute restraint, tail shock, and environmental stress impair long-term potentiation (LTP) in the hippocampus (Foy et al. 1987) and acute environmental stress can enhance long-term depression (Xu et al. 1997). However, stress-induced impairment in hippocampal LTP was significantly less in rats allowed to escape footshock than in yoked controls receiving identical shock, but not allowed to escape (Shors et al. 1989). This control mediated attenuation occurred even after a week of daily training sessions with relatively intense shock (30 trials, 1 mA, 1.5 s mean duration).