INTRODUCTION

TOP ABSTRACT INTRODUCTION CLASSICAL GENETICS AND SLEEP MOLECULAR GENETICS AND SLEEP CANDIDATE GENE APPROACH DISCOVERY OF NEW "SLEEP... QTL ANALYSIS OF SLEEP... QTL ANALYSIS OF THE... FROM QTL TO MOLECULAR... CONCLUSIONS REFERENCES

WHY WE SLEEP IS A LONG-STANDING question that modern neuroscience is still unable to answer. However, substantial progress has been made in our understanding of the mechanisms underlying different aspects of sleep physiology. We now know most of the neuronal substrates, neurochemical messengers, and regulatory processes involved in initiating and maintaining sleep. How can we further our understanding of sleep toward its function? Can molecular genetics bring new insights? Here, we first give a historical perspective on the genetics of sleep and then continue to review the available methods for the genetic dissection of sleep as an experimental approach that can be used in conjunction with state-of-the-art electrophysiological, neuroanatomic, and pharmacological techniques already used with great success in sleep research. We also caution against making premature interpretations of molecular findings of complex traits that should only be attempted within the general physiology of an intact organism.

	CLASSICAL GENETICS AND SLEEP

TOP ABSTRACT INTRODUCTION CLASSICAL GENETICS AND SLEEP MOLECULAR GENETICS AND SLEEP CANDIDATE GENE APPROACH DISCOVERY OF NEW "SLEEP... QTL ANALYSIS OF SLEEP... QTL ANALYSIS OF THE... FROM QTL TO MOLECULAR... CONCLUSIONS REFERENCES

Before the introduction of molecular biology, genetic studies of traits or phenotypes was based on either the observation of a resemblance between related individuals or crossing experiments between experimental individuals carrying divergent phenotypes. Simple laws have been discovered, according to which simple monogenic autosomal or recessive phenotypes segregate and are inherited on the basis of the assumptions of their molecular substrates, i.e., genes. However, these types of traits are far less prevalent than those that cannot be attributed to the influence of single genes. Nevertheless, shortly after the introduction of electroencephalography (EEG), Vogel (82) was one of the first to notice that even a complex phenotype like the EEG is under strong genetic control. Segregation analysis in pedigrees carrying rare EEG variants was concordant with classical genetic laws, suggesting the presence of single gene effects (83-85). Results from twin studies also indicated that the EEG patterns of monozygotic (MZ) twins have a much higher resemblance than those of dizygotic (DZ) twins or unrelated subjects (33, 82, 92), again confirming that this highly complex functional brain phenotype might be tightly controlled by genes and little (if any) affected by environment.

The study of the genetics of sleep can be approximately dated back to a publication by Geyer in 1937 (26) in which he reported on a higher concordance between sleep habits of MZ twins than DZ twins. Gedda (23) reported rare cases of concordant long sleepers (up to 15 h) in MZ twins. Gedda and Brenci (24) first estimated that the heritability (percentage of variance explained by the additive effects of genes) of sleep duration is over 30%. These authors later confirmed that sleep duration is highly similar in twins living apart (25), discounting the influence of possible environmental effects. More recent observations in twins indicated that even the pattern of rapid eye movements present higher concordance between MZ than between DZ twins (8) and that between 40 and 50% of the variance in sleep duration and the presence of a sleep disorder can be accounted for by genetic effects (30, 50). Zung and Wilson (94) performed the first polygraphic sleep recordings in twins in 1966; they found the temporal sequence of sleep stages to be almost completely concordant between MZ twins. Between 1983 and 1998, detailed polygraphic analyses were performed in twins; all of these studies reported surprising similarities between MZ twins (31, 39-41, 87). Experimental genetics of sleep in mice was pioneered by Valatx in the early 1970s. Over a time span of 15 years, Valatx's group (72-77) conducted several crossing experiments and recorded sleep in hundreds of inbred, recombinant inbred (RI), and hybrid mice, mainly to follow the segregation of paradoxical sleep (PS) amount. These studies, together with a diallelique (reciprocal crossing between several inbred strains) sleep experiment in mice performed by Friedmann (22), clearly indicated that, although some aspects of sleep may follow a simple segregation, the classical genetic laws cannot predict most others.

	MOLECULAR GENETICS AND SLEEP

TOP ABSTRACT INTRODUCTION CLASSICAL GENETICS AND SLEEP MOLECULAR GENETICS AND SLEEP CANDIDATE GENE APPROACH DISCOVERY OF NEW "SLEEP... QTL ANALYSIS OF SLEEP... QTL ANALYSIS OF THE... FROM QTL TO MOLECULAR... CONCLUSIONS REFERENCES

Although these now classical genetic sleep studies favor a strong genetic basis for sleep, we still have no clue about possible molecular mechanisms. The first attempts to identify the molecular basis of sleep have been based on the assumption that there must be genes that change their expression according to the behavioral state (sleep vs. wake) and/or time spent in a particular state. Closely related examples come from the circadian field where dramatic changes in the expression of several clock genes occur according to a strict circadian pattern (reviewed in Ref. 86). The two-process model of sleep regulation (10), by its remarkable predictions on the homeostatic aspects of sleep, motivated the first molecular investigation of sleep. Because sleep need and intensity are homeostatically regulated, deprivation of sleep should lead to a change in the expression of those genes, of which the product(s) is necessary for recovery or is (at least) involved in sleep regulation. Rhyner et al. (54) performed the first substractive hybridization in sleep-deprived vs. control rats and isolated several clones with increased or decreased relative expression, among which neurogranin and dendrin were later identified (47-48). The sleep-deprivation paradigm has since become the method of choice in gene expression experiments aimed at uncovering molecular bases of sleep regulatory processes (reviewed in Ref. 9). Our laboratory (56) has already discussed several conceptual problems of this approach. In summary, there are two major limitations. First is the fact that the expression of many genes varies with vigilance state rather than themselves being actively involved in vigilance state regulation. This weakness of correlative gene expression can only be overcome by confirming the involvement of the identified genes by loss-of-function and gain-of-function experiments. Second, the very high complexity of mammalian brain mRNA populations is a technical limitation; however, this can be overcome by recent developments in gene expression profiling (see below). Nevertheless, the identification of cortistatin and orexin (12, 13) with specific actions on sleep shows that gene expression investigation can be highly successful.

Transgenic methods based on random gene integration and homologous recombination gave rise to a strong hope to uncover the molecular bases of many phenotypes, including the most complex ones. These techniques can be used to address the consequences of overexpression, ectopic expression, time- and tissue-specific expression, and gain or loss of function of a candidate gene. Potential advantages and problems have been addressed in several reviews (for a recent review, see Ref. 88). With the exception of the orexin/hypocretin system (see below), to the best of our knowledge not a single gene manipulation has been performed specifically to investigate its consequences on sleep. However, many investigators in the field benefit from the availability of such transgenic models to answer sleep-related questions. The accumulated results indicate that all transgenic mice studies published thus far (see Table 1) report some sleep-related abnormalities, strongly suggesting either the nonspecific effects of gene manipulations and/or the complex integrative nature of sleep that makes it highly sensitive to many physiological changes.

	CANDIDATE GENE APPROACH

TOP ABSTRACT INTRODUCTION CLASSICAL GENETICS AND SLEEP MOLECULAR GENETICS AND SLEEP CANDIDATE GENE APPROACH DISCOVERY OF NEW "SLEEP... QTL ANALYSIS OF SLEEP... QTL ANALYSIS OF THE... FROM QTL TO MOLECULAR... CONCLUSIONS REFERENCES

The most straightforward approach is to consider a candidate gene based on previous evidence of its implication in sleep. As indicated above, sleep physiology and pharmacology studies have identified most of the essential systems from which candidate genes can be chosen. With the use of this approach, studies of the differential expression of tyrosine hydroxylase (53), growth hormone-releasing hormone (6), interleukin-1 (43), somatostatin (63), and brain-derived neurotrophic factor (5) have shown that all of these play a potential role in sleep. Sleep studies in transgenic mice (Table 1) have also used this approach. This powerful approach can also be combined with more recent techniques for gene identification. The major disadvantage of this approach is that obviously nothing new can be learned in terms of sleep gene discovery.

A unique "coincidental" sleep-related gene discovery concerns the orexin system. Orexin-A and -B are hypothalamic neuropeptides acting on orexin 1 and 2 receptors and first thought to be involved in feeding behavior (13, 55). Mignot's group (38) identified, through linkage analysis and positional cloning, mutations in the orexin 2 receptor gene as the cause of canine narcolepsy, an animal model of human sleep disorder narcolepsy. Independently and almost simultaneously, Yanagisawa's group (7), who were interested in the role of orexins in feeding behavior, discovered in the mouse a phenotype similar to canine and human narcolepsy after a targeted deletion of the preproorexin gene. More recently, the orexin system has also been implicated in the etiology of human narcolepsy; narcoleptic patients have undetectable orexin A levels in the cerebrospinal fluid (49), and in a small number of postmortem cases a dramatic reduction in the number of orexin-containing neurons was observed in the hypothalamus (51, 65). These unexpected observations led Sakurai's group (29) to generate the first transgenic mice to specifically investigate the role of the orexin system in vigilance states. These transgenic mice carry the promoter of the human preproorexin gene ligated to a truncated human ataxin-3, a gene that can induce cell death. The specific expression of this transgene construct in orexin-containing neurons induces apoptosis between 4 and 15 wk of age. Adult orexin/ataxin-3 mice show narcolepsy, hypophagia, and obesity (29). However, although the causal implication of the orexin system in narcolepsy is now established, its direct and causal involvement in normal sleep regulation, as is often claimed (3, 14, 28, 34, 64, 89, 91), remains to be elucidated.

	DISCOVERY OF NEW "SLEEP GENES"

TOP ABSTRACT INTRODUCTION CLASSICAL GENETICS AND SLEEP MOLECULAR GENETICS AND SLEEP CANDIDATE GENE APPROACH DISCOVERY OF NEW "SLEEP... QTL ANALYSIS OF SLEEP... QTL ANALYSIS OF THE... FROM QTL TO MOLECULAR... CONCLUSIONS REFERENCES

The systematic search for genes affecting a particular phenotype needs to cover the whole genome of an organism. The genome-wide search or mapping experiments make no a priori assumptions on gene systems involved; although this approach may lead to already known physiological mechanisms, its strength is that unknown and/or unexpected systems may be discovered. The orexin success story in canine narcolepsy is the unique and the best example of this approach in the field of sleep research. Therefore, genome-wide search constitutes the method of choice if we are to discover new "sleep genes."

The fundamental basis of this approach is the linkage disequilibrium between closely localized genes. If a gene is affecting a trait, there must be in addition to this gene other genetic material tightly linked to it and segregating from generation to generation together with the trait under study. Therefore, instead of testing all genes of an organism to check for cosegregation with a trait (linkage), one needs to test only a small panel of genes at regular intervals throughout the genome to obtain sufficient information on the large chromosomal pieces that are transmitted as a single unit. To follow the segregation of genes, they must be polymorphic (nucleotide variation from individual to individual); this may not be the case, or the sequence of the gene may be unknown. A powerful substitute is provided by highly polymorphic elements (restriction fragment length polymorphisms or RFLPs, simple sequence length polymorphisms or SSLPs, single nucleotide polymorphisms or SNPs, and so forth), mostly found in the noncoding genetic material. The more polymorphic markers are tested, the more accurate the mapping will be. A major question is how big the effect of a gene should be to be detected. Mutations in a single gene (e.g., orexin 2 receptor) can produce remarkable changes in a trait that is readily recognized (e.g., the monogenic autosomal recessive canine narcolepsy), whereas natural variations in genes controlling complex traits may have only subtle effects. Therefore, genome-wide approaches to dissect complex traits should be highly sensitive to detect all genes with variable effects on the trait.

Basically, two major approaches are available: mutagenesis and quantitative trait loci (QTL) analysis. The mutagenesis approach is straightforward: a mutagen like N-ethyl-N-nitrosurea is used to mutate at random the whole genome; this is followed by high-throughput screening of all mutant offspring to detect a major effect on a given phenotype. Both dominant and recessive mutations can be screened in the same way as a single gene mutation in a pathological condition. Once the mutated gene is localized, candidate gene or positional cloning approaches are used for its identification, and ultimately its functional analysis can be performed by gain or loss of function. The remarkable demonstration of this technique is provided by the isolation of Clock, one of the key mammalian circadian clock genes (1, 35, 81). However, genetic screens in this approach are for fully penetrant dominant and recessive mutations and therefore cannot identify small-effect sequence variations that may turn out to be essential for some aspects of the phenotype. In addition, with currently available technologies, recording and analyzing sleep of thousands of mice in a mutant screen does not seem feasible, at least not in a single academic laboratory. QTL analysis has been proposed as a powerful approach in the genetic dissection of complex traits (11, 17, 36). Although natural allelic variation of genes with small effect can be mapped through QTL analysis, the final identification of sequence variants (quantitative trait nucleotides or QTNs; Ref. 42) in the QTL region with biologically significant effects on the phenotype may represent prohibitive efforts in terms of both phenotyping and genotyping. An excellent example of how QTL analysis can further our understanding of complex traits was recently provided by Takahashi's group (57) for the circadian behavior in mice. Although most of the molecular machinery of the circadian timekeeping system has been discovered mainly by direct molecular techniques and mutagenesis, the identified genes do not explain the complexity of the observed circadian behavior. For instance, none of the clock genes has been found to be involved in the approximately 1-h difference in free-running period between BALB/c and C57BL/6 inbred mouse strains. Takahashi's group (57) used QTL analysis in a BALB/c × C57BL/6 (CXB) intercross and discovered several new loci with epistatic interaction.

	QTL ANALYSIS OF SLEEP AMOUNT AND DISTRIBUTION

TOP ABSTRACT INTRODUCTION CLASSICAL GENETICS AND SLEEP MOLECULAR GENETICS AND SLEEP CANDIDATE GENE APPROACH DISCOVERY OF NEW "SLEEP... QTL ANALYSIS OF SLEEP... QTL ANALYSIS OF THE... FROM QTL TO MOLECULAR... CONCLUSIONS REFERENCES

The complex nature of sleep is reflected in high intra- and interspecies phenotypic variability. For instance, the 24-h amount of sleep shows highly significant differences between inbred mouse strains (21, 22, 74). We have shown that AKR mice sleep ~3 h more than DBA mice over a 24-h period (21). As for the differences in the period of circadian rhythms, not a single gene but many genes with complex interactions may be found to account for this difference. We therefore initiated a series of experiments to dissect different phenotypic aspects of sleep in mice through QTL analysis. Table 2 summarizes the main quantitative aspects of sleep for those inbred strains of mice that differed the most. Note that most differences between high and low strains are consistent between the two studies performed 25 years apart (usually, differences between two extreme strains expressed in standard deviations are twice as large in Ref. 74 due to smaller within-strain standard deviations based on repeated measures; this is because data in Ref. 74 are based on 5-day continuous recordings instead of single 24-h recordings in Ref. 21).

QTL analysis can be performed in several segregating mouse populations, including intercross and backcross, advanced intercross and backcross, RI, and heterogeneous stocks (11, 63). Basically, two inbred mouse strains differing in the trait of interest are crossed and their F1 offspring are either intercrossed to generate F2 or backcrossed to one of the progenitor strains to generate backcross populations. Further random intercross or backcross generations can be performed to generate advanced intercross and backcross populations. To generate RI sets, F2 mice are brother × sister mated until full homozygosity, thereby "fixing" unique sets of recombinations in several inbred lines. Heterogeneous stocks are generated by intercrossing several inbred mouse strains over many generations and therefore represent higher rates of recombination and polymorphism useful for fine mapping (44, 63). Although RIs are usually not suitable for QTL mapping due to their limited progenitor strains and number, for QTLs of large effects, they may provide significant mapping accuracy because of the fourfold increase in recombination compared with a F2 population (11). Also, for complex phenotypes with high variability, the use of RIs is advantageous because several individuals per strain are tested instead of, for example, a single F2 animal. Therefore, as a first step, we have performed QTL analysis in two RI sets. In seven CXB RIs, QTLs were identified for the amount of PS on chromosomes 5, 7, 12, and 17 (61). Toth and Williams (71) conducted a similar study in 13 CXB RIs and reported provisional QTLs on chromosomes 4, 16, and 17 for the amount of PS during the light period. Because of a small number of CXB RIs (n = 7-13), none of the QTLs identified could satisfy standard significance levels (37). Also, the discrepancy between the two referenced works indicates that, as in any linkage study, replication is needed, especially when the findings are based on a limited number of animals and/or strains (for a recent review, see also Ref. 69). QTL analysis in 25 C57BL/6 × DBA/2 (BXD) RIs identified the first significant QTL for the amount of PS in the 12-h light period on chromosome 1 (60). We have estimated in BXD RIs that between 40 and 60% of variance in sleep amounts can be accounted for by the additive effects of 6-15 genes, confirming a polygenic basis for differences in sleep amounts.

	QTL ANALYSIS OF THE SLEEP EEG

TOP ABSTRACT INTRODUCTION CLASSICAL GENETICS AND SLEEP MOLECULAR GENETICS AND SLEEP CANDIDATE GENE APPROACH DISCOVERY OF NEW "SLEEP... QTL ANALYSIS OF SLEEP... QTL ANALYSIS OF THE... FROM QTL TO MOLECULAR... CONCLUSIONS REFERENCES

Quantitative EEG analysis by means of fast Fourier transform analysis is a robust technique to evaluate functional changes of the central nervous system, including changes in vigilance states (2). Slow-wave sleep is characterized by oscillations in the delta (1-4 Hz) and spindle (11-15 Hz) frequency ranges, whereas PS, especially in rodents, is characterized by theta oscillations (5-9 Hz). Wakefulness encompasses a variety of behaviors, each with a characteristic EEG pattern. Waking EEG analysis in twins demonstrated that spectral profiles are among the most heritable traits in humans (reviewed in Ref. 79), with heritability estimates ranging from 75 to 90% (80). We have shown that several spectral characteristics of sleep (Fig. 1) differ significantly between inbred mouse strains (20). Among these, the theta peak frequency during PS showed the highest difference between inbred strains, and 80% of interstrain variability could be attributed to genetic effects. We have established in intercross and backcross CXB mice that a single gene may explain most of this phenotypic variance (a major gene) (62). A high-resolution QTL mapping is needed to localize the underlying gene.

View larger version (45K): [in this window] [in a new window]   Fig. 1.   Bottom: a 10-s electroencephalogram (EEG) sample for paradoxical sleep (PS; left) and slow-wave sleep (SWS; right) from 1 representative mouse of the inbred strains for which EEG characteristics differed the most. AK, AKR/J; BR, C57BR/cdJ; C, BALB/cByJ; D2, DBA/2J. During PS, the EEG is dominated by theta oscillations (5-9 Hz). During SWS, the EEG is composed of delta (1-4 Hz), theta, and sigma (11-15 Hz) oscillations. Top: relative contribution and frequency of these EEG components can be quantified by spectral analysis. Average theta frequency during PS is ~1.5 Hz faster in BR compared with AK mice. The spectral profiles of the SWS EEG in C and D2 mice display large differences in the relative contribution of the delta and sigma frequency components. Spectral profiles are based on all 4-s epochs scored as PS or SWS in a baseline 12-h light period (n = 7/strain).

Another aspect of the sleep EEG, with considerable importance for its homeostatic regulation, is the change in delta power (i.e., EEG power in the 1- to 4-Hz range) as a function of the prior wakefulness duration. Relative delta power can be used to index both sleep need and sleep intensity in all mammalian species studied so far. Sleep loss induces a proportional and predictable increase in delta power, which differs significantly between inbred mouse strains (18, 21). We have used the delta power rebound after a 6-h sleep deprivation to map the underlying genes in BXD RIs (18). The results confirmed that the rate at which sleep need accumulates varies greatly with genotype. One significant QTL was identified on chromosome 13 that accounted for 50% of genetic variance in this trait. These two examples suggest that quantitative sleep EEG parameters are under stronger genetic control than sleep amounts; therefore, these are more suitable for quantitative genetic dissection. Sleep amounts in mammals may be approximated by techniques other than EEG, such as the use of piezoelectric films, which would make a high-throughput genetic analysis feasible. However, success may be limited due to the highly polygenic nature of sleep amounts, suggesting the presence of many genes with small effects. On the other hand, the quantitative EEG analysis dramatically limits its wide-range use (because this involves implantation, recovery, adaptation, cabling, and time-intensive analysis) but may turn out to be more successful due to the limited number of underlying genes and the presence of major genes.

	FROM QTL TO MOLECULAR BASIS

TOP ABSTRACT INTRODUCTION CLASSICAL GENETICS AND SLEEP MOLECULAR GENETICS AND SLEEP CANDIDATE GENE APPROACH DISCOVERY OF NEW "SLEEP... QTL ANALYSIS OF SLEEP... QTL ANALYSIS OF THE... FROM QTL TO MOLECULAR... CONCLUSIONS REFERENCES

QTL analysis does not map a gene but a genetic effect in a large chromosomal region (usually between 20 and 30 cM). These large regions may contain a single major gene or several genes with small effects. Also, QTLs may interact with each other (epistasis), an effect that is difficult to detect in QTL mapping experiments. The first step is to make sure that the region contains QTLs of large enough effect. One possibility is to transfer the QTL region from one inbred strain background to another inbred background through repeated backcrossing and selection to make a congenic strain. We have reviewed sleep amount data available in eight congenic strains generated by transferring minor histocompatibility genes from the inbred strain BALB/c to the inbred background of C57BL/6 (60). In almost all cases, the results indicated that, even if the transferred pieces of chromosomes contained a large-effect QTL detected in CXB RIs, a clear effect could not be observed in the resulting congenic strain. This finding strongly suggests that 1) many genes are involved, 2) they are in interaction (epistasis), and 3) even if a major QTL is identified, its effect may vary from genetic background to genetic background due to the action of modifier genes (45). Once the QTL region is verified, the next step is to finely map the QTL down to the smallest chromosomal region (11, 17), amenable to candidate gene analysis. The final identification of functional QTNs (sequence variants) is the most difficult part, since QTNs may be found every few kilobases. Therefore, a combination of several approaches is necessary for mapping and candidate gene analysis. High-resolution QTL mapping in conjunction with future developments in high-throughput phenotyping and genotyping and the availability of whole genome sequences of several major mouse strains should identify candidate genes to be investigated. All of those candidate genes can then be mutated either by classical homologous recombination (knockout) or by the newly developed serial nested chromosomal deletions (59). Systematic crosses between increasing and decreasing QTL allele and the complete set of mutants should physically identify the functional QTN. Because most QTNs will probably be involved in gene regulation rather than being mutations affecting the protein function, further gene expression profiling with high-throughput genotyping technologies (e.g., microarray or TaqMan) and gene translation and posttranslational protein analyses should be used to uncover the molecular mechanisms involved. Note that finding the molecular basis in a mutagenesis experiment follows approximately the same time- and labor-intensive procedure.

	CONCLUSIONS

TOP ABSTRACT INTRODUCTION CLASSICAL GENETICS AND SLEEP MOLECULAR GENETICS AND SLEEP CANDIDATE GENE APPROACH DISCOVERY OF NEW "SLEEP... QTL ANALYSIS OF SLEEP... QTL ANALYSIS OF THE... FROM QTL TO MOLECULAR... CONCLUSIONS REFERENCES

The challenge of finding molecular bases of sleep requires demanding efforts, not only in terms of genetic and genomic resources but also in terms of future developments of high-throughput phenotyping techniques. If enough effort and financial resources are given to develop simplified and automated sleep recording devices, this major limitation should be overcome. Both high numbers of QTL and mutant screen experiments will thus become feasible. Once the entire mouse genome sequence becomes available, genome-wide assessment of spatial and temporal expression of genes and identification of every conceivable sequence variation will provide speedy molecular analysis of complex traits such as sleep. The genetics of sleep may also be complemented by the genetic analysis of rest activity in flies (27). Considering that we still know very little about the molecular basis of sleep and how ineffective the current treatments for common sleep disorders such as insomnia or hypersomnia are, we believe that molecular genetic approaches are still our best hope for developing better therapies and uncovering sleep functions. In this review, we have tried to show that many techniques and their combinations can be used to answer basic questions. However, relying solely on large-effect genes as revealed through mutagenesis or candidate gene approaches may bring erroneous interpretations of the complex polygenic nature of sleep mechanisms.