In contrast to forward genetics, in which the strategy is to go from the observed phenotype to the underlying genotype, reverse genetics starts with a disrupted or altered gene of interest and examine its effect on a phenotype or phenotypes of interest. In humans, reverse genetic studies most often concern natural occurring functional polymorphisms while the use of animal models allows for targeting specific genes. Transgenic animals can be used to study the consequences of overexpression, ectopic expression, time- and tissue-specific expression, and gain or loss of function of a specific gene. To date, a large number of studies in sleep genetics have utilized reverse genetic approaches by knocking-out a gene of interest (see Sect. 5.1). The delineation between reverse and forward genetic approaches is not a formal one and depends on the scale at which they are used. Transgenic strategies can also been used in forward genetic studies, in which hundreds of mutated or knockout (KO) lines of mice (Knight and Abbott 2002), or flies (Cirelli 2003) are screened. The International Knockout Mouse Consortium is currently creating a collection of mouse in which all protein-coding genes are mutated and will therefore provide a population of animals for genome-wide screens (Ringwald et al. 2011).

Molecular genetics is an approach that evaluates changes in gene expression related to the trait of interest; it is an unbiased method that makes use of several techniques such as DNA microarrays, proteomics, and RNA sequencing. Molecular approaches in sleep studies are based on the assumption that the expression of genes change as a function of time spent awake or asleep (Cirelli et al. 2004; Mackiewicz et al. 2009; Thompson et al. 2010). Rhyner et al. were the first to use a molecular approach to identify genes that change their expression after SD in the rat, and identified the protein neurogranin to be decreased in the rat forebrain after 24 h of SD (Rhyner et al. 1990). These methods have been helpful in identifying the brain expression of genes involved in sleep or modulated by sleep (see Sect. 8). However, these approaches cannot reveal a causal relationship between a gene or a set of genes and a particular behavior or phenotype. For example, a gene that does not show transcriptional modification may nonetheless play an important role in the process under investigation. On the other hand, a transcript that does increase with increased sleep need might merely be driven by the sleep–wake distribution while not playing a role in the homeostatic sleep process itself.

It is clear that the above approaches are complementary; e.g., the function of candidate genes identified with QTL or molecular genetics approaches have to be confirmed in KO animals (an example is provided in Sect. 8.1). Future efforts should therefore combine genetics approaches in animal models and GWAS in humans to facilitate uncovering the molecular pathways that underlie sleep homeostasis. However, other approaches and techniques will remain important in making progress in elucidating the complex physiology of sleep; genetic dissection of sleep can be used in conjunction with state-of-the-art electrophysiological, neuroanatomical and pharmacological techniques already used with great success in sleep research.

A sleep homeostatic gene could be defined as a gene that modifies the sleep–wake dependent dynamics of EEG delta power or a gene that affects the (compensatory) increase in NREM or REM sleep duration after SD . Establishing whether a gene variant affects a sleep homeostasis process is, however, not always straightforward and several considerations have to be taken into account in the analysis. Here we illustrate some of the general problems that can occur with interpreting the results of SD studies that should be considered before claiming a homeostatic phenotype is observed.

Although, sleep homeostatic processes could, in principle, be assessed and quantified under baseline conditions, usually sleep homeostatic responses are studied after experimentally challenging sleep need through; e.g., keeping subjects awake for a certain duration at a time-of-day sleep is present under undisturbed baseline conditions. In most studies, sleep variables measured during recovery sleep after enforced waking are then contrasted to the individual levels reached in baseline. Subsequently, these relative changes are compared between genotypes to assess the effect of a disrupted or mutated gene of interest on sleep homeostasis (Fig. 2a). A slightly more elaborate variant of such analysis are the so-called “gain-loss” time course analyses in which the sleep loss incurred over the course of a SD is analyzed by accumulating the differences of sleep observed during the SD minus the sleep duration observed during the same time-of-day during baseline (Fig. 2b′). This accumulated sleep deficit then serves as the starting point of the sleep gained during the recovery period following a SD and is assessed in the same way (i.e., as relative differences from matching times during baseline). Genotype differences in ‘gain-loss’ dynamics are then taken as evidence of an altered homeostatic regulation of sleep. An important short-coming of such analyses is that they implicitly assume that the sleep obtained during baseline is the amount a subject “needs” or, in other words, the duration of sleep that is homeostatically defended. For instance, the deficit in sleep duration in the gain-loss analyses introduced above exclusively reflects the sleep duration obtained during the baseline period the SD took place (given that no sleep is obtained during the SD protocol). Also the recovery dynamics are obtained by contrasting recovery values to corresponding baseline values. In the hypothetical example shown in Figs. 2b and 2b′, the recovery dynamics of sleep is exactly the same; i.e., each hour the same duration of extra sleep is obtained making the recovery curves run in parallel. Nevertheless, a researcher can claim the discovery of a homeostatic phenotype because of statistical genotype differences that, in this case, are due only to a difference of sleep expressed during baseline at the time SD was performed. In another example, recovery was made to be exactly proportional to the sleep duration the animal was deprived of and this amount of extra sleep was linearly distributed of the 18 h depicted (example in Fig. 2b′). As a result, in this exercise, recovery is “complete” by hour 24 (i.e., after 18 h of recovery) reaching the zero deficit level in both genotypes. Nevertheless, a researcher can again claim that homeostatic regulation is different between genotypes because the SD resulted in different relative deficits and the slopes of the gain process differed, while it could equally be claimed that recovery of sleep time lost is perfect and not different in both because in one case more sleep was deprived of and thus a higher pressure for sleep was accrued leading to more extra sleep. Many other scenarios could be construed pointing out these or related problems such as the so-called “ceiling effect.” This effect is sometimes alluded to account for a smaller increase attained during recovery at times-of-day when during baseline the subject already sleeps a lot. It is clear, and all will agree, that the level of a sleep variable expressed during baseline cannot be attributed to homeostatic need alone. For example, in a study examining sleep need in short and long sleepers that differed in habitual sleep duration by more than 3 h, it was concluded that the dynamics of the sleep homeostat did not differ and that short sleepers somehow could resist higher levels of sleep pressure (Aeschbach et al. 1997, 2001).

The problem becomes even more complex when the variable cannot be directly analyzed in absolute terms (i.e., in the units they are measured) because, due to a range of confounds (e.g., electrode placement, impedance) variability among individuals is large thereby decreasing statistical power. Quantitative EEG variables are therefore often expressed relative to an individually calculated reference value to reduce variability among subjects. Differences in absolute levels should also be corrected for variables for which only the sleep–wake dependent changes are considered informative. Our genetic analyses of the sleep–wake dependent relative changes in EEG delta power and of the contribution of EEG delta power to NREM sleep EEG revealed that these two aspects are under the control of different genetic factors (Franken et al. 2001; Maret et al. 2005). Similarly, EEG delta power contributes more prominently to the NREM sleep EEG in female compared to male mice; nevertheless, the sleep–wake dependent dynamics of the changes in EEG delta power differ with sex such that the build-up in females is slower than in males, and therefore higher absolute EEG delta power not necessarily reflect higher homeostatic sleep need (Franken et al. 2006). Studies in humans arrive at similar conclusions. EEG analyses in subjects carrying specific polymorphisms has demonstrated that large genotype effects on the prevalence and amplitude of EEG delta waves (absolute values) do not necessarily translate in differences in the sleep–wake dependent relative changes in EEG delta power (i.e., homeostasis; reviewed in Franken 2012). As a last example illustrating this duality between these two aspects of EEG delta power , the effect of benzodiazepines can be mentioned. These sleep-promoting drugs are known to reduce EEG delta power while leaving the sleep–wake dependent dynamics unaffected (Borbely and Achermann 1991). Therefore, one should be cautious with making claims on sleep homeostasis and sleep depth based on absolute EEG values. Assessing other, non-EEG indexes of NREM sleep depth or quality such as the fragmentation of NREM sleep (Franken et al. 1991a, 1999) or arousal thresholds (Neckelman and Ursin 1993; Wimmer et al. 2012) could be used to strengthen such claims.

When investigating the homeostatic regulation of sleep, the relative changes in EEG delta power related to the sleep–wake distribution are analyzed. The individual reference value for EEG delta power should be chosen such that effects of eventual differences in the sleep–wake distribution and sleep duration do not affect the result. As the sleep–wake distribution is the main determinant of the changes in EEG delta power , genotype differences in sleep duration as well as sleep distribution can greatly impact average levels of EEG delta power while not (necessarily) impacting the dynamics of the underlying homeostatic process. This can be illustrated in SCN-lesioned animals (Trachsel et al. 1992). In our hypothetical example (Fig. 2c), removing the SCN results in an increase in sleep and in a “flat” distribution of sleep over the 24 h day mainly due the lack of the consolidated period of wakefulness after dark onset. The bout of spontaneous wakefulness results in high levels of EEG delta power in subsequent NREM sleep in the intact animals, levels that are never reached in SCN-lesioned animals. Therefore, when taking the 24 h mean value of EEG delta power as a reference, as is often done, this reference value will be higher in the intact animals compared to the same reference in the arrhythmic animals. As a result, the relative values calculated using this reference, will be lower in the intact animals and when compared to the lesioned animals could lead to erroneous statements concerning sleep homeostasis because the difference in the level in EEG delta power reached after SD are an artifact of selecting the reference. As an alternative, we have proposed to use the mean EEG delta power reached in the last portion (4 h) of the rest phase (or light phase in nocturnal species) as a reference (Franken et al. 1999). Levels reached at this time are less dependent on eventual differences in the sleep–wake distribution because EEG delta power decreases exponentially and EEG delta power differences at sleep onset gradually disappear over the course of the rest phase. In our hypothetical case of the SCN-lesioned animal, EEG delta power does not appreciatively deviate from this low level (Fig. 2c).

Another issue concerning the interpretation of EEG delta power seen in the literature is to contrast EEG delta power measured after SD to its value at corresponding clock time during baseline. Already minor genotype differences in the sleep–wake distribution during baseline can lead to significant differences in EEG delta power . Therefore, by expressing values reached after SD as a percentage of values reached during particular times during baseline can lead to wrong conclusions. To circumvent the problem of having to contrast values obtained after a SD to a baseline reference, “dose–response” experiments could be considered. In such studies, SD s of different durations could be used to quantify directly the dynamics of the relationship between the time animals were kept awake and the response variable (sleep duration, EEG delta power etc.; e.g., Tobler and Borbely 1986; Franken et al. 2001; Seugnet et al. 2006). To quantify the relationship between the sleep–wake distribution and EEG delta power , computer simulations can be used as an alternative approach (Franken et al. 2001).

In conclusion, the definition of an altered sleep homeostasis is not trivial. Besides some of the conceptual issues outlined above, differences in methods to perform SD such as the sleep time allowed during the SD , recording and analyzing the EEG further add to the problem of unambiguously establishing whether a genotype difference in the response to sleep loss qualifies as a homeostatic sleep phenotype.

Many different cells, molecules, and signaling pathways have been investigated through reverse genetic studies for their possible role in the control of sleep homeostasis (Table 1). In the Sect. 5 through 8, we give an overview of the pathways that have been linked to sleep homeostasis , with an emphasis on clock genes and adenosine . The list provided is, however, non-exhaustive and many other systems are currently under investigation. Even though these studies brought essential information in the understanding of sleep homeostasis, a unifying picture on its genetic control has yet to emerge.

Although the genes referred to as clock genes are involved in many pathways they are best known (and named) for their role in circadian rhythm generation. Clock genes are transcriptional regulators engaged in negative feedback loops that underlie the molecular circuitry of the circadian clock machinery (Ko and Takahashi 2006). Positive elements of this feedback loop are three factors: CLOCK, NPAS2, and BMAL1. NPAS2::BMAL1 and CLOCK::BMAL1 heterodimers can drive transcription of many genes including the Period (Per1 and –2) and Cryptochrome (Cry1, and –2) genes. PER::CRY protein complexes suppress CLOCK/NPAS2::BMAL1-mediated transcription and thus their own transcription thereby constituting the negative elements in the feedback loop. Additional interactions between these core clock genes and other actors (e.g., REV-ERBα, RORα) at the level of transcription, translocation back into the nucleus, post-translational modifications add further complexity and stability. In Drosophila, the molecular oscillator relies on very similar timing systems using orthologs to the mammalian clock genes (Benito et al. 2007). The central autoregulatory feedback loop is composed of the two factors CLOCK and CYCLE, which are the homologues of the mammalian CLOCK and BMAL1, respectively. The CLOCK::CYCLE heterodimers activates the transcription factors PERIOD and TIMELESS which act as negative regulators like the mammalian PER and CRY proteins, respectively (Tomioka and Matsumoto 2010).

The circadian process and the homeostatic process appear independent (for review see Franken 2013). Rendering animals arrhythmic through lesioning the SCN (Trachsel et al. 1992; Easton et al. 2004) or a light pulse (Larkin et al. 2004), did not affect the increase in EEG delta power after SD . Moreover, studies in humans showed that the sleep–wake dependent dynamics of EEG delta power are little affected by circadian factors (Dijk and Czeisler 1995). Analyses of sleep homeostasis in humans, mice, and flies carrying polymorphism or targeted disruptions of clock genes demonstrate, however, that at least at the molecular level, circadian rhythms and sleep homeostasis are difficult to dissociate (Shaw and Franken 2003; Franken and Dijk 2009). In the following, we will discuss the different clock gene mutants/polymorphisms that have been studied and suggesting a non-circadian role of clock genes in sleep homeostasis .

Adenosine is an inhibitory neuromodulator that has been proposed to act as a homeostatic regulator of sleep and to link humoral and neural mechanisms of sleep–wake regulation (Porkka-Heiskanen et al. 1997, 2000; Basheer et al. 2004; Kalinchuk et al. 2011). In mammals, four subtypes of G-protein coupled receptors mediate the effects of adenosine : A₁R, A_2AR, A_2BR, and A₃R and two of them have been investigated for their role in sleep homeostasis (Fredholm et al. 2001). It is thought that A₁R are responsible of the sleep effects of adenosine (Rainnie et al. 1994; Benington and Heller 1995), but a careful study performed in A₁R KO mice showed that the homeostatic aspect of sleep regulation was unaltered in animals lacking A₁R (Stenberg et al. 2003). This study revealed that mice lacking A₁R had normal baseline sleep–wake distribution and responded normally to sleep pressure , with NREM sleep rebound and EEG delta power rebound being similar to WT animals. More recently, a conditional central nervous system KO of this adenosine receptor was created. A1R^-/- mice were found to have reduced EEG delta power in NREM sleep during baseline, as well as during sleep restriction (Bjorness et al. 2009) whereas the sleep–wake distribution and amount were preserved. More importantly, when the mice were allowed to sleep for 2 h following 4 h of sleep restriction, the relative increase in EEG delta power during NREM sleep above baseline levels was smaller in the KO mice although the amount of time spent in NREM sleep was similar to WT. The authors suggested that the elevated sleep need signaled by adenosine is, at least in part, mediated through the A₁R. However, the implemented sleep restriction protocol allows mice to recover during the 2 h sleep opportunity windows between the 4 h SD s, importantly affecting the level of EEG delta power . Moreover, the authors observed a general decrease in EEG delta power when calculated over all states (wakefulness, NREM and REM sleep) which might indicate a specific effect of the gene on general electrical brain activity rather than an effect on sleep homeostasis . Given the contradictory findings, the role of A₁R in sleep homeostasis remains unclear. It might be interesting to generate inducible KO animals to study the loss of A₁R in the adult stage only, thereby circumventing several potential confounds such as developmental compensation. The accumulated evidence indicates that besides A₁R also A_2AR contributes to the effects of adenosine on sleep. A preliminary report indicated that A_2AR KO mice do not show a NREM sleep rebound following 6 h SD , revealing an altered sleep homeostasis in these mice (Urade et al. 2003). Changes in EEG delta power were not reported in these mice. Along the same line, a human study revealed that a distinct polymorphism in the gene encoding the A_2AR, Adora2, modulates individual sensitivity to subjective and objective effects of caffeine on sleep (Rétey et al. 2007). It would be of interest to test the effect of this polymorphism on the homeostatic process by submitting individuals carrying the different polymorphism to a SD . In Drosophila, one single adenosine receptor gene has been identified, dAdoR, that is most closely related to the mammalian Adora2 gene. Study in dAdoR mutant flies revealed that sleep was not affected by the mutation, neither in baseline, nor in recovery from SD (Wu et al. 2009). These results suggest that in flies, adenosine receptors are not required to maintain sleep homeostasis .