Introduction: Finding new clock components; past and future

Abstract

The molecular mechanism of circadian clocks has been unraveled primarily by the use of phenotype-driven (forward) genetic analysis in a number of model systems. We are now in a position to consider what constitutes a clock component, whether we can establish criteria for clock components, and whether we have found most of the primary clock components? This perspective discusses clock genes and how genetics, molecular biology and biochemistry have been used to find clock genes in the past and the future.

The modern era of ‘clock genetics’ began in 1971 with the discovery by Ron Konopka and Seymour Benzer of the period (per) locus in Drosophila (Konopka and Benzer, 1971). At the time, even the existence of single-gene mutations affecting a process as complex as circadian rhythms was met with great skepticism. The eminent phage geneticist, Max Delbruck, is purported to have said, “No, no, no. It is impossible,” when told of Konopka’s results. Benzer’s reply: “But Max, you don’t understand. They’ve already been found” (Greenspan, 1990). Much has changed since the golden age of phage genetics, but much has also stayed the same. Thirty-three years later, we have dozens of examples of single genes that affect behavior (Kyriacou and Hall, 1994; Takahashi et al., 1994; Bucan and Abel, 2002). We have lived through decades when it was hard to clone a gene – now, we can manipulate genes at will. We have complete genome sequences for most of the important model organisms.

In the circadian field, we have identified cell autonomous circadian pathways that share a conserved transcriptional autoregulatory feedback mechanism across two domains of life (Dunlap, 1999; Young and Kay, 2001; Reppert and Weaver, 2002; Lowrey and Takahashi, 2004). And, we can already anticipate that circadian clock proteins will ultimately yield to structural biology (Uzumaki et al., 2004; Vakonakis et al., 2004; Ye et al., 2004). So where will this end, and what issues remain? In a sense, this is where much has stayed the same. We are upon the 50^th anniversary of Pittendrigh’s analysis of temperature compensation of circadian oscillations in Drosophila pseudoobscura (Pittendrigh, 1954), yet we know very little more about this mechanism today. [Ironically, the D. pseudoobscura genome is being sequenced. Even Pittendrigh would not have expected this.] That is not to say that we haven’t learned much. To the contrary, we are approaching a complete catalog (‘parts list’) of circadian oscillator components in a number of organisms, and we even think we know how many of these parts work and interact with one another. But, a rigorous description of clock mechanism that includes an understanding of kinetics and stoichiometry still eludes us. Moreover, we need validation of such models in vivo (not in transfected cells!). Finally, we have no deep understanding of how cell autonomous circadian oscillators interact in multicellular organisms ultimately to regulate behavior (Low-Zeddies and Takahashi, 2001).

In the beginning there was per

Benzer’s bold foray into the genetics of brain and behavior in Drosophila remains one of the most remarkable achievements in biology - by unraveling complex systems using systematic screens focused on a judicious array of related phenotypes (Benzer, 1973). The period gene story remains the exemplar for genetic dissection of behavior (Hall, 2003); however, one should not forget the significance of the other Benzer mutants. The Shaker mutant defined the first potassium channel at the molecular level (Kaplan and Trout, 1969; Kamb et al., 1987; Papazian et al., 1987). The dunce mutant provided an entrée into the genetics of learning and memory (Dudai et al., 1976; Chen et al., 1986). The sevenless mutant defined a developmental pathway for photoreceptor determination (Banerjee et al., 1987; Hafen et al., 1987). And recent mutants such bubblegum and methuselah have uncovered genes in neurodegeneration and lifespan (Lin et al., 1998; Min and Benzer, 1999). So what is my point here? The discovery of circadian rhythm genes rests largely (indeed almost exclusively) upon the use of forward genetic screens to isolate circadian mutants, which in turn were used to identify the causative gene. Including per (Bargiello et al., 1984; Zehring et al., 1984), almost every important circadian locus has been defined by so called ‘clock mutants.’ This list includes frequency (frq), timeless (tim), Clock, KaiABC, and double-time which were first identified by positional cloning (McClung et al., 1989; Myers et al., 1995; Antoch et al., 1997; King et al., 1997a; Ishiura et al., 1998; Kloss et al., 1998). A second list includes cycle and tau which provided the primary phenotypic evidence that the underlying gene was a ‘core component’ (see below) (Rutila et al., 1998; Lowrey et al., 2000).

Phenotype, phenotype, phenotype

In retrospect, a compelling ‘phenotypic profile’ can be described to define what we now consider ‘core clock genes.’ First and foremost is an extreme phenotype: all of these key loci can be defined by mutations that alter period length by more that 15% (3–4 hours) or lead to a complete loss of circadian rhythms. When multiple alleles are available, period shortening and lengthening mutations are observed, as well as, loss-of-rhythm mutants. Ironically, per met all of these criteria from the beginning. Subsequently, frq, tim and KaiABC also produced this extreme range of phenotypes. In the absence of multiple alleles or in cases where null mutations are lethal, an extreme period phenotype appears to be sufficiently compelling to define a clock gene of interest. The mouse Clock, Drosophila double-time and hamster tau mutants fall into this category in which mutations caused ≥4-hour period changes and null mutations were not available.

One could of course argue that this analysis is biased – we have only gone after the extreme mutations and so these are the ones that we know about. This is fair, but if we look more broadly at targeted mutations in which we begin with the gene and then ask what the phenotype might be, the ‘extreme phenotypic profile’ still appears to hold, especially if we allow the inclusion of double mutants for paralogous genes such as Period and Cryptochrome in mice. With the exception of the Bmal1 (also known as Mop3) knockout, which has an extreme loss-of-rhythm phenotype by itself (Bunger et al., 2000), most gene-targeted null mutations have produced less extreme (more subtle) circadian phenotypes. The next best knockout is Per2 which causes a 1.5 hour period shortening and eventual loss of rhythms in the majority of mice in constant darkness (Zheng et al., 1999). Knockouts of Per1 and Per3 fall into the subtle (0–1 hour period changes) category at best (Bae et al., 2001). Fortunately, the Per1/Per2 double knockout has strong effects, causing a complete loss of rhythms in constant darkness (Bae et al., 2001; Zheng et al., 2001). The Cryptochrome knockouts are also interesting. The Cry1 null mutation causes a relatively subtle 1-hour shortening of period, while the Cry2 null mutation causes a 1-hour lengthening of period (van der Horst et al., 1999; Vitaterna et al., 1999). Again, the Cry1/Cry2 double knockout has a profound phenotype with a complete loss of circadian rhythms in constant darkness (van der Horst et al., 1999; Vitaterna et al., 1999).

In contrast to this strong class of mutant circadian phenotypes, there are also examples of circadian genes that are involved in the clock mechanism but are not essential for the generation of circadian rhythms. This second ‘nonessential category’ is exemplified by Rev-erbα (Preitner et al., 2002). This gene is activated by the CLOCK-BMAL1 complex and is in turn regulated by PER/CRY-mediated repression. REV-ERBα then feeds back on Bmal1 (and Clock) to drive cycles of transcription that are out of phase with Per and Cry (Preitner et al., 2002). This pathway forms a second feedback loop in the circadian network that has been labeled as a “positive feedback” loop (which we now know is the result of two inhibitory steps). Null mutations of Rev-erbα abolish the feedback effects of this pathway on Bmal1 and transcription of Bmal1 is constitutively elevated. However, circadian rhythm generation persists in the absence of REV-ERBα, and there are only subtle effects on the period and stability of circadian rhythms in the behavior of knockout mice (Preitner et al., 2002). This specific example illustrates the quandry that exists when we define clock components: REV-ERBα is clearly part of the native circadian oscillator network, yet it is not essential for the generation of circadian oscillations.

Additional examples of genes that would not be expected to produce extreme phenotypes include other elements of the circadian oscillator system such as those involved in input and output pathways. Again, it is difficult to make definitive statements here, but some generalizations apply. Inputs to the circadian oscillator can exert significant phase-shifting and period effects; however, loss-of-function mutations should not lead to arrhythmicity. Rather such mutations should lead to defects in entrainment as seen with the cryptochrome^b (cry^b) mutation in Drosophila (Stanewsky et al., 1998), or with compound mutations of the mammalian visual system involving melanopsin (Hattar et al., 2003; Panda et al., 2003). Output mutations can produce overt arrhythmicity, but strong effects on period or phase should not occur. Molecular readouts of core oscillator components (such as Per) can then be used to assess whether the circadian oscillator is intact. Good examples of the effects of loss-of-function mutations on output components are Dbp, a direct target of CLOCK-BMAL1 (Lopez-Molina et al., 1997; Ripperger et al., 2000), and Pdf, a peptide output signal in Drosophila (Renn et al., 1999). Caution is warranted, however, because feedback effects of output pathways such as locomotor activity can exert phase-shifting effects, subtle period changes, and modulation of clock gene expression (Van Reeth and Turek, 1989; Edgar et al., 1991; Maywood et al., 1999).

In spite of the complexities addressed above, we can define a set of criteria for primary (core or essential) clock components. The strongest case can be made when there are ‘strong’ (i.e., 3–4 hour) period shortening and period lengthening mutations, as well as, arrhythmicity for null mutations. Drosophila per and tim are the prototypical examples (Konopka and Benzer, 1971; Konopka et al., 1994; Sehgal et al., 1994; Rothenfluh et al., 2000). When homozygous null mutations are lethal, as in the case of double-time in Drosophila, the spectrum of extreme period mutations maybe compelling (Price et al., 1998). However, the lack of a loss-of-function null mutation tempers our ability to conclude that the gene is essential in the classic genetic sense. In mammals, where paralogous genes are common, compound null mutations are required to achieve a strong loss-of-rhythm phenotype (van der Horst et al., 1999; Vitaterna et al., 1999; Bae et al., 2001; Zheng et al., 2001). Finally, as seen for Clock and tau, substantially abnormal period mutants (caused by antimorphic mutations) may be sufficient to implicate the gene in circadian function in the absence of null mutations (Vitaterna et al., 1994; King et al., 1997a; King et al., 1997b; Lowrey et al., 2000). In addition to period phenotypes, Clock and Per2 mutant mice also become progressively arrhythmic in constant darkness, which are interesting examples of mutant alleles that have a combined period and loss-of-rhythm phenotype (Vitaterna et al., 1994; Zheng et al., 1999).

If an allelic series is not available, mutant phenotypes alone may not be sufficient to define core clock components. However, elucidating the specific biochemical functions of the proteins combined with a strong circadian phenotype can be ultimately compelling. For example in mammals, CLOCK was shown to interact with BMAL1 to activate transcriptionally the Per and Cry genes (Gekakis et al., 1998; Kume et al., 1999); and Casein kinase 1 epsilon (CK1ε) was shown to phosphorylate the PERIOD, CRYPTOCHROME and BMAL1 proteins (Lowrey et al., 2000; Eide et al., 2002). Thus, a second set of criteria (beyond an extreme circadian phenotype in mutants) involves defining the biochemical function of putative clock components within circadian pathways.

What are clock components anyway?

Traditionally, circadian expression patterns or circadian oscillations in ‘activity’ of a putative clock component have been thought to be critical criteria for inclusion (Aronson et al., 1994). However, real examples have clearly shown that this assumption only applies to a subset of core clock components. Theory has defined state variables of limit cycle oscillators as clock components, but under this definition, only the feedback elements of the oscillator would fulfill this criterion. (Two examples of mathematical models of the circadian oscillator incorporating recent molecular information have appeared recently (Forger and Peskin, 2003; Leloup and Goldbeter, 2003).) While state variables obviously oscillate; and clock components such as PER, TIM, and CRY match this definition, other critical elements as defined above do not necessarily oscillate in an obvious manner. Clock and tau are two examples in which circadian expression may not be necessary for their function. Although for these specific cases, one could enter into a semantic discussion concerning whether the ‘activity’ of CLOCK or CK1ε may be circadian. In any case, we know that circadian expression patterns are neither restrictive nor compelling since literally hundreds of downstream target genes exhibit circadian patterns (Akhtar et al., 2002; Panda et al., 2002; Storch et al., 2002; Ueda et al., 2002) and many components of input pathways to the circadian system can have circadian patterns (Kornhauser et al., 1990; Ginty et al., 1993; Emery et al., 1998; Heintzen et al., 2001; Merrow et al., 2001).

Perturbation analysis has been considered the gold standard for the identification of clock components. The classical view has been that if a putative component is part of the oscillator mechanism, then perturbation of the expression level or activity of the component must change the steady-state phase of the oscillator for pulse treatments, or may cause period changes for continuous treatments. This criterion is valid for distinguishing output elements, but has never been able to discriminate input pathways from effects on the oscillator itself. We now have clear examples of input components that oscillate and could fulfill the perturbation test, but do not appear to be clock components (Emery et al., 1998; Heintzen et al., 2001). In addition, we have examples of output pathways such as locomotor activity that can feedback and influence the period and phase of circadian systems (albeit with relatively weak phenotypic effects). So unfortunately the complexity of real circadian systems has compromised our desire to have simple criteria to define oscillator components.

So how do we define core clock components? Is this a valid concept, or does the complexity of circadian networks as in the case of genetic regulatory networks allow too many degrees of freedom to exist (Greenspan, 2001; Roenneberg and Merrow, 2003)? I would argue that the situation is not hopeless. We can distinguish among different levels of importance of clock components. As discussed above, it appears that only a limited number of genes can be mutated to cause strong (3–4 hour) period changes for missense mutations and loss-of-rhythm phenotypes for null mutations. This allelic series definition incorporates both the classic genetic loss-of-function criterion for necessity, as well as, the perturbation criterion when applied to strong period changes. Defining the position of a putative component in a regulatory network (classically known as defining a pathway) describes how the component interacts with other elements. Moreover, describing the biochemical function of the putative element can further define its role within a circadian network. Some elements of the network may be essential – their removal (loss-of-function mutations) will lead to a loss-of-rhythm phenotype. Other elements of the network may not be essential either because there are alternate paths for the network to function or there may be true cases of redundancy or overlapping function. In these cases, the loss-of-function phenotypes will be subtle and compound mutations may be required to produce strong phenotypes. If a strong phenotype cannot be produced by compound loss-of-function mutations for a paralogous gene family, then these elements would be considered nonessential.

Ultimately, we would want to apply another level of perturbation tests for core clock components. This would require conditional expression of clock components. We would want to apply this test for a number of reasons. First, it is important to demonstrate that the phenotypic effects seen in classical mutants are due to the ‘ongoing’ or cycle-to-cycle effects of the mutation rather than some developmental effect (since classical mutants develop in this context). Second, conditional experiments can test whether rhythmic or constitutive expression is required for the component. Third, conditional expression in a wild-type background can be used to apply perturbation tests (phase shifts or period changes). Finally, conditional experiments can be used in a tissue-specific manner to test for cell autonomy, hierarchical regulation and other types of interactions. To date, only a few clock components have been tested in this way. Aronson et al. applied this test early on in the field (Aronson et al., 1994). In retrospect, the only deficit in this analysis was definitive measurement of the expression levels of the conditionally expressed genes which was performed later (Merrow et al., 1997). Similar analyses have been accomplished in Cyanobacteria (Ishiura et al., 1998). Tissue-specific and constitutive expression patterns have also been tested in Drosophila (Yang and Sehgal, 2001; Zhao et al., 2003). Thus far, no critical tests of this type have been performed in mammals particularly in the context of the intact animal. We are still a long way from critical testing of known clock components especially in mammals.

Are there more new clock components?

The answer is a definite “yes.” There are many reasons for this opinion. (1) Even though we have complete genome sequences in humans, mice and rats (Lindblad-Toh, 2004), we are still far from understanding the function of a large fraction (~40%) of genes. We would be naively egotistical to think that we have already found all the core circadian clock genes just on the basis of our ignorance of gene annotation and function. (2) There are a significant number of candidate clock genes that have been proposed to be components of the circadian clock mechanism that require genetic evidence of strong circadian phenotypes before they can be admitted to the ‘inner circle.’ An example is the recent description of the Dec1(Stra13) and Dec2 genes which have circadian expression patterns in the SCN and have strong inhibitory effects on CLOCK-BMAL1 activation in cell transfection/reporter gene assays (Honma et al., 2002; Grechez-Cassiau et al., 2004). What is still missing, however, is in vivo evidence that mutations can produce strong circadian phenotypes (loss-of-function or strong period mutants). (3) Quantitative genetic experiments analyzing circadian phenotypes in different inbred strains of mice reveal that the majority of quantitative trait loci (QTLs) do not map to known circadian genes (Shimomura et al., 2001). These allelic variants contribute to striking differences in circadian behavior in different inbred strains of mice and suggest that a large number of circadian modifier genes remain to be identified. (4) Very few forward genetic screens for circadian mutations have reached saturation. In mice, we have barely scratched the surface. Indeed in a recent pilot screen in which two circadian mutants were recovered, one mutant was a new allele of a known locus and a second mutant defines a new circadian gene because no known circadian genes coincide with its map location. New large-scale recessive mutagenesis screens in mice promise to deliver many new circadian mutants (Moldin et al., 2001). Furthermore, because saturation mutagenesis can only be assessed empirically by recovery of multiple, new alleles only at previously known loci, it is likely that this level of screening has only been accomplished in Cyanobacteria (Kondo et al., 1994; Ishiura et al., 1998). In addition, even saturation mutagenesis is ultimately limiting because it may not be possible to recover circadian mutants for ‘all’ genes. We can already anticipate this result in the case of paralogous genes such as Per where compound mutations may be required to produce strong circadian phenotypes. (5) Microarray experiments have uncovered literally hundreds of cycling genes (Akhtar et al., 2002; Duffield et al., 2002; Panda et al., 2002; Storch et al., 2002; Ueda et al., 2002). In mice, about 10% of all detectable transcripts in any tissue express circadian profiles. A significant set of genes appear to cycle in the majority of tissues and most of these genes have not been previously assigned a function in the circadian network. It seems likely that at least a few of these cycling genes could define additional circadian clock genes. In addition, more comprehensive microarrays interrogating the entire transcriptome are now available so additional cycling genes will undoubtedly be uncovered (Su et al., 2004). (6) Biochemical interaction screens have identified a large number of proteins that can interact with known circadian clock proteins. For example, using chromatin immunoprecipitation (ChIP) assays on the Dbp promoter, Ripperger et al. found many proteins associated in a complex with CLOCK, most of which have not been identified (Ripperger et al., 2000). Yeast two hybrid screens have also yielded about 20 bonafide CLOCK-interacting proteins which remain to be explored (Gekakis et al., 1998). Such protein interaction screens have only begun to be pursued in earnest and so much remains to be explored in this arena using both traditional biochemistry as well as modern proteomics approaches. (7) Finally, many large-scale approaches such as genome-wide siRNA screens or overexpression screens with genome-wide cDNA collections are underway and are likely to reveal additional circadian clock genes. Thus, the future is bright. A daunting array of tools are being applied to circadian gene discovery and each of these is likely yield new insight into the circadian mechanism.

The Paradigm Shift: What have we lost and what have we gained?

The application of genetics, biochemistry, molecular biology and genomics to the study of circadian biology in the last decade has certainly led to a paradigm shift in the field. We have moved from analysis of circadian rhythms primarily at the formal level to a much more concrete mechanistic level. Much like the transition from classical to molecular genetics, the field of circadian rhythms has flourished while still maintaining a common language and set of problems. Perhaps it is somewhat ironic that genetics on the one hand and circadian rhythms on the other hand are so complementary. Circadian rhythms stands as one of the great success stories of classical genetic approaches to the study of behavior; while genetics and molecular biology have been the key to unraveling the network of genes that compose the circadian oscillator system.

So where are we today? As we stated at the beginning of this piece, we are well on our way to a complete description of circadian clock components in a number of model systems. And, we think we know how the core autoregulatory transcriptional feedback loop operates. However, we are still far from a quantitative understanding of any circadian oscillator with respect to kinetics, stoichiometry and cell biology in the context of real cells and organisms. We still have little knowledge concerning how this molecular oscillator regulates the large number of pathways in the cell that are under circadian control – much less, how these cells communicate circadian information more broadly. A clear view has emerged that virtually every cell in the body has the capability of expressing circadian oscillations (Balsalobre et al., 1998; Yamazaki et al., 2000; Yamaguchi et al., 2003; Yoo et al., 2004). Now we need to know why. Are circadian rhythms so fundamental to metabolic efficiency that we cannot do without them (Rutter et al., 2002)? Or, are cellular circadian rhythms a vestige of a unicellular past where metabolism and environmental insult are more proximal to survival (Ouyang et al., 1998; Johnson and Golden, 1999)? We can only speculate. Recent links between circadian rhythms and cancer (Fu and Lee, 2003) and the cell cycle (Matsuo et al., 2003) in mice suggest that the relationship is more than vestigial.

How has this molecular view of circadian rhythms changed us? For a while, some of us may have lost some of our focus on systems and integrative questions (while we engaged in sprints to find new genes). But, with new tools such as circadian promoter::reporter gene technology and the knowledge that we are composed of populations of circadian oscillators (Millar et al., 1992; Plautz et al., 1997; Balsalobre et al., 1998; Yamazaki et al., 2000; Yamaguchi et al., 2003; Yoo et al., 2004), a renaissance in circadian systems biology is on the horizon. Lost forever is that black box that once represented the circadian clock.

Figure 1.

Photograph of Ron Konopka (left) and Seymour Benzer (right) taken in October 2000 in Pasadena, CA. Photo courtesy of Howard Hughes Medical Institute, Holiday Lectures on Science, December 2000, Clockwork Genes: Discoveries in Biological Time.