Abstract
Circadian clocks drive daily oscillations in a variety of biological processes through the coordinate orchestration of precise gene expression programs. Global expression profiling experiments have suggested that a significant fraction of the transcriptome and proteome is under circadian control, and such output rhythms have historically been assumed to rely on the rhythmic transcription of these genes. Recent genome-wide studies, however, have challenged this long-held view and pointed to a major contribution of posttranscriptional regulation in driving oscillations at the messenger RNA (mRNA) level, while others have highlighted extensive clock translational regulation, regardless of mRNA rhythms. There are various examples of genes that are uniformly transcribed throughout the day but that exhibit rhythmic mRNA levels, and of flat mRNAs, with oscillating protein levels, and such observations have largely been considered to result from independent regulation at each step. These studies have thereby obviated any connections, or coupling, that might exist between the different steps of gene expression and the impact that any of them could have on subsequent ones. Here, we argue that due to both biological and technical reasons, the jury is still out on the determination of the relative contributions of each of the different stages of gene expression in regulating output molecular rhythms. In addition, we propose that through a variety of coupling mechanisms, gene transcription (even when apparently arrhythmic) might play a much relevant role in determining oscillations in gene expression than currently estimated, regulating rhythms at downstream steps. Furthermore, we posit that eukaryotic genomes regulate daily RNA polymerase II (RNAPII) recruitment and histone modifications genome-wide, setting the stage for global nascent transcription, but that tissue-specific mechanisms locally specify the different processes under clock control.
Circadian clocks drive daily oscillations in a variety of biological processes. These cell-autonomous timekeepers are widespread across the tree of life (Bell-Pedersen et al., 2005) and are thought to confer a selective advantage, by allowing organisms to anticipate predictable variations resulting from the constant transitions between days and nights and thus adjusting their physiology and behavior accordingly (Yerushalmi and Green, 2009).
The description of diverse rhythmic processes in different organisms eventually prompted researchers to “pop the hood” and genetically dissect various model systems to characterize their molecular basis. This led to the identification of the first clock mutants during the 1970s, followed by the subsequent cloning of clock genes (i.e., of genes whose product is involved in the inner workings of the clock) and, later, by the detailed study of the mechanisms underlying these clocks (reviewed in Dunlap, 2008). Hand-in-hand with this detailed characterization came the daunting task of identifying the processes (and genes) that are under clock control in each of these organisms (the so-called output of the clock), to ultimately gain insights regarding their adaptive value.
The quest for identifying which genes are under clock control has been undertaken in all major clock model systems, most commonly by globally assessing the steady-state level of messenger RNAs (mRNAs), and these studies have suggested that ~5% to 20% of the transcriptome, in a single tissue, may be under circadian regulation (Covington et al., 2008; Kula-Eversole et al., 2010; Filichkin et al., 2011; Hardin, 2011; Mohawk et al., 2012; Hurley et al., 2014; Zhang et al., 2014). Such transcript rhythms arise from a combination of both clock transcriptional control and circadian regulation at the mRNA level, the latter including splicing, alternative polyadenylation, poly(A) tail length regulation, and transcript stability, among others (reviewed in Kojima et al., 2011; Staiger and Koster, 2011; Kojima and Green, 2015). In addition, a handful of studies have globally addressed the question of how many proteins display daily rhythms, suggesting that 5% to 20% of them may be under circadian control in a single tissue (reviewed in Robles and Mann, 2013; Mauvoisin et al., 2015), which analogously may result from both rhythmic protein production or daily regulation of stability (Kojima et al., 2011). In addition, circadian control of ribosome biogenesis and rhythmic expression of genes involved in both ribosome biogenesis and protein synthesis have been reported in different systems (Correa et al., 2003; Jouffe et al., 2013), suggesting broad clock regulation of translation. Altogether, this shows that circadian control takes place at multiple levels, affecting all stages of gene expression, and this is thought to result in a robust and accurate time-keeping mechanism that allows for specific activities to take place at appropriate times.
Rhythms in steady-state transcript levels have commonly (and historically) been assumed to result from rhythmic transcription, despite the fact that these approaches, which rely on mRNA profiling, fail to differentiate whether the observed rhythm is the result of either transcriptional or posttranscriptional regulation (or a combination of both). Such assumption is derived from the fact that for core clock components, in different eukaryotic systems, regulation has been shown to rely largely (although not exclusively) on rhythmic transcription (Mackey, 2007). In addition, such transcriptomic approaches leave out clock regulation at the translational level, which is noteworthy, considering that in many cases, rhythmic proteins appear to derive from nonrhythmic mRNAs (Morse et al., 1989; Reddy et al., 2006; Mauvoisin et al., 2014), and hence, additional novel processes to those identified solely by using mRNA profiling may exhibit clock control.
Determining the Relative Contribution of Transcriptional and Posttranscriptional Mechanisms in output Rhythms: the Jury is Still out
All the aforementioned considerations raise the question as to what the exact contribution of each stage of gene expression is to the total number of cycling mRNAs/proteins and, consequently, to the different processes under clock control within the cell. Recent studies have partly addressed this question globally, evaluating the relative contribution of transcriptional versus posttranscriptional control in animal systems (Koike et al., 2012; Le Martelot et al., 2012; Menet et al., 2012; Rodriguez et al., 2013). As a whole and contrary to the aforementioned expectation, these studies have pointed to a significant contribution of posttranscriptional regulation to output circadian/diurnal mRNA cycling. By comparing nascent and precursor RNAs to mature mRNAs in mouse liver, it was shown that less than a third of the rhythmic mRNAs had strong corresponding rhythms in transcription (Koike et al., 2012; Menet et al., 2012). An additional study (conducted in flies) also supported this notion (Rodriguez et al., 2013), identifying several genes for which the mRNA amplitude was significantly higher than that at the nascent level. Similarly, by evaluating the temporal relationship between RNA polymerase II (RNAPII) occupancy and mRNA accumulation in mice liver, extensive posttranscriptional control on mRNA cycling was also suggested (Le Martelot et al., 2012), although to a lesser degree than in the other two studies. Even though posttranscriptional clock control is nothing new (Kojima et al., 2011; Staiger and Koster, 2011; Kojima and Green, 2015), its widespread contribution to global mRNA cycling, as suggested by these studies, was unexpected.
While such global transcript profiling studies suggest a prominent role for posttranscriptional control in rhythmic mRNA levels (as a significant amount of output cycling mRNAs appear to lack strong rhythms in transcription), there are similarly many cases (20%-50%) in which rhythmic proteins in that same tissue (mouse liver) appear to derive from nonoscillating mRNAs (Reddy et al., 2006; Mauvoisin et al., 2014; Robles et al., 2014), suggesting extensive translational regulation and highlighting its importance in determining rhythmic activities in the cell.
The task of determining the relative contributions of transcriptional, posttranscriptional, and translational mechanisms in ultimately controlling the various output rhythmic processes within the cell (and their phase) is definitely a complex one. The different relative roles we have discussed (the specific percentages) may apply only to the mouse liver, as it has become quite evident that rhythmicity is extremely condition dependent. In many organisms, the catalogue of rhythmic genes (and the point of clock control) reported in each study is heavily dependent on growth conditions, tissues assayed, entrainment protocols, sampling frequency, and algorithms used to identify rhythmic genes, among other factors (reviewed in Doherty and Kay, 2010). Interestingly, meta-analyses integrating data from different tissues and conditions have shown that it is not simple to classify a gene as either rhythmic or not, as some genes may be detected as rhythmic (at any level of gene expression) only under certain experimental conditions and in particular tissues (Ptitsyn et al., 2006; Keegan et al., 2007; Miller et al., 2007; Covington et al., 2008; Hurley et al., 2014; Patel et al., 2014; Zhang et al., 2014). Notably, the latter has been shown to be true even for core clock components (Hardin, 1994). This, then, suggests that contributions may be quite dynamic, changing between tissues and conditions (and susceptible to experimental design and analysis), and it may be too early to suggest that a particular stage in gene expression reigns over others in the determination of clock molecular outputs.
In addition, if rhythm amplitudes are low, and the intrinsic noise of the system is high (due to biological reasons or systematic technical errors), this could yield poor overlap between the data sets assessing transcriptional, posttranscriptional, and translational regulation, further complicating analyses and hindering a clear definition of rhythmic sets in each tissue. Indeed, even when the same tissue is sampled and the same molecule is assayed (e.g., mRNA levels), the overlap of genes catalogued as rhythmic in different studies is usually low (reviewed in Kojima and Green, 2015), possibly (among other factors) due to underpowered experimental designs (for experimental considerations for studying circadian rhythms using genome-wide transcriptional profiling, see Li et al., 2015) and differences in data processing (Deckard et al., 2013). Similarly, a recent study in yeast shows that between-study correlations for either mRNA or protein levels are usually very modest, suggesting the presence of large systematic errors between studies (Csardi et al., 2015). This is important not only for the comparison of rhythmic gene lists between studies, as mentioned above, but also when comparing lists of genes that exhibit rhythms at one step of gene expression to those in another, in a single tissue, as in the mammalian studies we have discussed. For instance, even though it has usually been suggested, in a wide variety of organisms, that steady-state mRNA levels determine roughly only 30% to 50% of the variation in protein levels, suggesting a significant role for protein-level regulation, new studies show that experimental noise due to, for instance, measurement biases and errors have reduced the real value of such correlation and underestimated the role of transcription (Li et al., 2014; Csardi et al., 2015; Li and Biggin, 2015). Indeed, through the use of noise-robust analyses of yeast data, for instance, it has been reported that mRNA levels actually explain more than 80% of the variation in protein levels (Csardi et al., 2015). Such analyses reveal that transcriptional and posttranscriptional regulation appear to act in a concerted and nonindependent manner to determine protein levels. Similarly, when experimental error is accounted for, transcription rises as a significant player in the determination of protein levels (Li et al., 2014). It is possible, then, that experimental noise might have confounded the appreciation of the specific impact that each step of gene expression has in determining the set of rhythmic molecular processes in the cell.
TThe multiple comparisons of circadian Nascent-seq or RNAPII ChIP-seq experiments, mRNA measurement studies, and proteomic approaches, might be complicated by systematic biases and experimental errors that have not been taken into account. In addition, even within-technique comparisons might suffer from experimental errors, and without oversampling (i.e., higher time resolution to evaluate whether the used sampling times have been appropriate) and defining the proper parameters for significant cycling detection (e.g., sampling density, number of replicates, read depth and signal-to-noise ratio, use of proper algorithms, consideration of method biases) in these new data sets (rhythmic Nascent-Seq, ChIP-seq, proteomics, etc.), the lack of overlap could easily be a consequence of comparing multiple noisy data sets, ones with missing data or ones that have been affected by batch effects. For instance, improving sampling resolution (using 1- or 2-h sampling protocols instead of the conventional 4- and 6-h ones) dramatically increases the number of identified cycling transcripts (Hughes et al., 2009; Li et al., 2015). Furthermore, low overlap between circadian mRNA profiling and proteomics studies may derive from lower sensitivity and accuracy, as well as bias to high-abundance proteins in the latter, together with the fact that amplitude appears to be lower at the protein level, making their detection as rhythmic more challenging. In other words, many more genes may currently be able to be detected as rhythmic in mRNA-based assays than through proteomic approaches, even when the corresponding proteins do in fact oscillate. Tellingly, neither of the recent large-scale circadian proteomic studies (Mauvoisin et al., 2014; Robles et al., 2014) identified any core clock components among the rhythmic proteins. Such technical issues as a raison d’être for the low overlap between the genes that appear to exhibit rhythmic mRNA to those that display oscillating protein levels are supported by the fact that when only proteins that are detected as rhythmic are considered, correlation between mRNA and protein rhythms can be as high as 80% (Robles et al., 2014). Last, reproducibility for such circadian proteomic studies, evaluated simply by comparing the list of proteins that were determined to be rhythmically expressed between the aforementioned 2 recent proteomic studies assaying the mouse liver, is low, similar to the case with mRNA profiling studies, as previously mentioned. Similarly, the genes identified to be rhythmically transcribed in the same tissue through both Nascent-seq and intron counts are also low (Koike et al., 2012; Menet et al., 2012). When all such technical limitations are eventually taken into account, together with differences in experimental design and analysis, as discussed above, the scenario may turn out to be different, and claims of a major contribution of a particular stage of gene expression in determining output molecular rhythms may need to be revised.
On the Importance of Transcriptional Regulation and Coupling in Determining output Rhythms in Gene Expression
While the recent aforementioned mammalian studies (Koike et al., 2012; Menet et al., 2012) have suggested a major contribution of posttranscriptional regulation to output rhythms (due to identifying a large number of genes displaying rhythmic mRNA levels but flat transcriptional profiles), these reports did also identify genes that exhibited rhythms at both the transcriptional and mRNA levels and genes that cycled only at the transcriptional level. By using the nomenclature from Menet et al. (2012), we refer to the first genes as AR/R and to the others as R/R and R/AR, respectively (R: rhythmic; AR: arrhythmic).
For the genes in the R/R category, transcriptional rhythms do appear to be the main determinants of mRNA (and, most likely, protein) level rhythms, and not surprisingly, most clock genes are in this gene set (Koike et al., 2012; Menet et al., 2012). What about the AR/R genes? These genes are usually argued to exhibit rhythms at the mRNA level due to exclusive and independent regulation at the posttranscriptional level, as their transcription is arrhythmic and assumed not to contribute to the oscillation.
These studies also identified a significant number of genes that displayed rhythms in transcription but no robust cycling at the mRNA level (R/AR set): over half of the genes that had cycling nascent RNAs lacked a rhythm in the corresponding mRNA. Similar findings were echoed in Drosophila (Rodriguez et al., 2013). This phenomenon has been reported previously (Wuarin and Schibler, 1990; Millar and Kay, 1991; Wuarin et al., 1992) and represents a true phenomenon, although its exact prevalence (the high percentage suggested by these recent studies) may be affected by technical issues, as discussed above.
Analogous to the situation of the AR/R genes, recent circadian proteomics assays have shown that many oscillating proteins do not derive from rhythmic mRNAs (Mauvoisin et al., 2014; Robles et al., 2014), and while rhythmic translational regulation is not new (Morse et al., 1989), these recent large-scale and quantitative proteomic analyses have largely extended earlier global studies (Reddy et al., 2006).
As a whole, these studies epitomize that clock control can take place at any stage of gene expression. Is, however, circadian regulation at each level independent from regulatory processes occurring on a previous one? Does it follow that a flat upstream process and a rhythmic downstream one (e.g., flat transcription and oscillating mRNA levels) mean that rhythmic regulation is determined exclusively and independently at the downstream level? Specifically, can events that take place during transcription play any role in the observed mRNA rhythms of the genes in the AR/R category? Can coupling, a functional connection between different steps of gene expression, in a scenario in which regulation arises at the level of transcription, play a role in ultimately determining rhythmic protein levels, regardless of whether mRNA levels oscillate?
In the rest of this Commentary, we speculate that even when arrhythmic, processes occurring during transcription may still play a role in defining circadian output through the functional connection between transcription and downstream processes affecting the resulting mRNA and protein levels.
Delving into the AR/R Set: Let’s Not Dismiss Transcriptional Regulation Just Yet
Posttranscriptional regulation has been shown to play an important role in circadian biology, and such mechanisms have been called into play to try to explain the AR/R gene set. Daily changes in RNA stability, for instance, could account for such rhythms in a way that even though the gene may be transcribed uniformly throughout the day, the stability of the resulting mRNA would change daily and would be detected as a cycling mRNA. This could be mediated, for instance, by specific proteins/noncoding RNAs (ncRNAs) that interact with the target mRNAs and mediate its rhythmic decay (for examples, see Kojima et al., 2011). In any case, the different actors involved in mediating the expression profiles of the genes in the AR/R group are, for the most part, unknown and are currently the subject of intensive research.
While it is possible that for some of the AR/R genes, low-amplitude (but still rhythmic) profiles or even the shape of the expression pattern of the nascent RNA (Michael et al., 2008; Deckard et al., 2013) might have led to the misclassification of some R/R genes into the AR/R category, the latter represents a bona fide biological phenomenon (i.e., posttranscriptional clock control). A search of overrepresented motifs in the sequence of the RNAs in this group might shed a mechanistic light on their regulation (Ray et al., 2013).
Should we stop thinking, then, about any transcriptional role for the rhythmic profiles of the mRNAs in this gene set? Do (seemingly) flat transcriptional profiles negate its role in the rhythmic behavior of the corresponding mRNA? In other words, are transcriptional and posttranscriptional regulatory mechanisms independent from one another in the context of circadian rhythms?
It has long been acknowledged that different stages in the gene expression pathway are coupled, so that control associated with one step could somehow regulate the next one, which has been suggested to result in a versatile system with maximized efficiency and specificity. Transcription has been shown to affect other nuclear processes like capping, splicing, and polyadenylation (Maniatis and Reed, 2002; Komili and Silver, 2008). In addition, an interesting link exists between transcriptional and posttranscriptional events, such that the cytoplasmic fate of the corresponding mRNA may already be determined in the nucleus (Giorgi and Moore, 2007; Dahan et al., 2011; Dahan and Choder, 2013). Akin to our discussion, accumulating evidence shows coupling between transcription and processes like mRNA stability, localization, and translation, predominantly by the cotranscriptional recruitment of RNA-binding proteins (RBPs) to the nascent mRNA (reviewed in Choder, 2011; Tirosh, 2011; Haimovich et al., 2013). It has been speculated that such regulatory coupling between the different stages of gene expression may allow biological systems to better respond to genetic and environmental perturbations (Dahan et al., 2011).
Notably, specific cis promoter elements have been shown to affect cytoplasmic mRNA decay in yeast (Bregman et al., 2011; Trcek et al., 2011), as suggested earlier in mammals (Enssle et al., 1993), providing a fascinating example of such interaction between 2 steps in gene expression. Coupling of transcription and mRNA degradation appears to be widespread in eukaryotes and may involve multiple mechanisms (Dori-Bachash et al., 2012). For instance, several transcripts in yeast display coupled synthesis and decay regulation via the RNAPII heterodimer subunit Rbp4/7 (Goler-Baron et al., 2008). In the most characterized examples, such coupling relies on the “imprinting” of the nascent mRNA with specific proteins, recruited to the site of mRNA synthesis via interaction with members of the transcriptional machinery, which ultimately mediate the cytoplasmic destiny of the RNA (Dahan et al., 2011; Tirosh, 2011; Haimovich et al., 2013). Transcription, then, is not merely a passive mechanism for mRNA synthesis, but instead, this step can result in transcript imprinting, which could allow for the coordination of different stages of gene expression, influencing the fate of the corresponding mRNA.
Interestingly, this mechanism has been reported to be at play in the regulation of the mRNA stability of 2 genes that exhibit cell cycle–dependent decay kinetics in yeast, SWI5 and CLB2 (Trcek et al., 2011). Such control depends on the promoter sequence and is independent of cis elements in the mRNA. The authors propose a model in which specific regulatory proteins are recruited to the promoter and cotranscriptionally loaded onto the mRNA. These proteins mediate the acceleration of mRNA decay at the appropriate time in the cell cycle (Trcek et al., 2011). Such cell cycle–dependent decay kinetics are relevant for the sharp transitions of mRNA levels between consecutive cycles.
This mechanism would often operate in another cellular process that involves sharp transitions of gene expression patterns, such as differentiation and response to stress (Haimovich et al., 2013), and it is easy to imagine how it could also play a role in the context of the circadian clock. In this scenario, the mRNA would be cotranscriptionally loaded with a specific set of proteins, in such a way that they may mediate time-of-day–dependent decay kinetics. The loaded proteins may recruit decay factors (e.g., deadenylation regulators) at appropriate times of the day, leading to rhythmic mRNA expression profiles. Although this would be independent of rhythmic transcription, these rhythmic mRNA profiles would in fact require specific promoter elements (as in Bregman et al., 2011; Trcek et al., 2011) or, alternatively, may be modulated by specific RNAPII subunits (Harel-Sharvit et al., 2010; Dahan and Choder, 2013) during transcription. In this way, even though the genes in the AR/R set may be argued to be clock-regulated at the posttranscriptional level, some may not be completely independent of their transcriptional regulation (albeit arrhythmic) to attain rhythmicity at the mRNA level. Furthermore, if different genes share the promoter elements involved in the aforementioned coupling mechanism, their stability could be regulated as a group, as an RNA regulon (Keene, 2007), which would be promoter determined. Note that the RNAs in such cluster would not need to share any sequence motif within the transcribed region. It might be useful, then, to search for overrepresented elements in the promoter regions of the genes in the AR/R group. Similarly to the experiments in Trcek et al. (2011), one could propose an experiment in which a) the promoter region of some of the AR/R genes is replaced, to test whether the mRNA still retains rhythmicity (i.e., full posttranscriptional clock control), or b) one of these promoter regions is placed upstream of a reporter gene, to see if this gene now displays the rhythmic expression profile of the endogenous one (coupling).
With such mechanism in mind, one could argue, however, that the only thing required for the rhythmic profile displayed by these genes in the AR/R group is a protein or ncRNA to bind to the mRNA and somehow mediate its rhythmicity: there would be no need to require this to happen cotranscriptionally or even in the nucleus. While this may be true for a number of genes, the yeast She2p case is enlightening in this regard. She2p is a yeast RBP that recognizes specific cis elements (“zip codes”) on target mRNAs, and this interaction is necessary for the localization of these molecules to the yeast bud tip (Olivier et al., 2005). Interestingly, localization of these mRNAs appears to be determined during transcription: She2p interacts with RNAPII (via elongation factors) and is proposed to “hop” onto the nascent RNAs cotranscriptionally (Shen et al., 2010). Even though She2p can interact with the target mRNA in the cytoplasm, localization of target genes and other regulatory events are altered when its nuclear-cytoplasm shuttling is impaired, suggesting that its nuclear role is important (reviewed in Trcek and Singer, 2010). Also, even though Rbp4p, which we mentioned previously, is in vast excess over RNAPII, its role as a coordinator of posttranscriptional mRNA processes depends on its prior association with RNAPII, suggesting that its cotranscriptional loading is relevant (discussed in Dahan and Choder, 2013; Haimovich et al., 2013). Cotranscriptional loading of regulatory factors may allow the interacting partners to be placed in close proximity, which may be relevant for efficient binding. Also, this mechanism may provide the partners with binding sites that could later be unavailable, thus affecting specificity temporally (Tirosh, 2011; Haimovich et al., 2013).
A mechanism similar to the one just described could also be employed by genes in the R/R set: in this case, decreased stability, accompanied by rhythmic promoter shutoff, can lead to very sharp transitions in mRNA levels between circadian cycles. This has indeed been shown to play a role for a few rhythmic genes in different species (So and Rosbash, 1997; Lidder et al., 2005; Guo et al., 2009). Whether these varying decay kinetics are mechanistically coupled to their transcription, however, as for SWI5 and CLB2, is unknown. Some mRNAs may simply attract cytoplasmic factors that can rhythmically modulate their decay. In any case, coordinating transcriptional control and mRNA decay appears to be a widespread regulatory mechanism for maintaining appropriate levels of cellular mRNAs (reviewed in Haimovich et al., 2013), and we speculate that some genes in the AR/R set may exploit this type of coordination, such that even though their transcription rates do not oscillate strongly on a daily basis, transcriptional regulation may still be relevant for their rhythmicity at the mRNA level.
It should be noted that several genes in the AR/R category in the aforementioned mammalian studies display highly variable transcription rates throughout the day (which were not, however, deemed rhythmic by the analyses performed), and only a small fraction of cycling mRNAs (~14%) actually exhibited relatively constant transcription (Menet et al., 2012). Interestingly, many of the AR/R genes displayed higher levels of transcription preceding the peak in mRNA cycling (Koike et al., 2012; Menet et al., 2012). Furthermore, for most AR/R genes, this increased variability correlated with rhythmic mRNA expression. This led the authors to suggest that this transcriptional variability plays a significant role in the rhythmic mRNA profile displayed by the genes in this category (Menet et al., 2012; Partch et al., 2013), so their rhythmic expression profile at the mRNA level might still be somehow related to their transcriptional regulation. Although some of these may have simply been misclassified as AR/R, it is tempting to suggest that for some of these genes, this variable transcription might be associated with the activity of different transcriptional regulators, which could modulate daily mRNA levels (e.g., by cotranscriptional loading of regulators, differential choice of transcriptional start sites, splice sites or poly(A) sites). Of note, not all arrhythmically transcribed genes with variable transcription had rhythmic mRNA profiles (i.e., the mechanism would be gene specific), and not all genes in the AR/R group had variable transcription (although most did).
Rhythmic Transcription and Nonoscillating mRNA Levels: Not Quite a Dead End
One interesting set of genes found in the aforementioned global studies is the R/AR set: genes that display rhythmic transcription but whose corresponding mRNAs do not cycle. Although some of the genes in this group have been suggested to display such trait simply from cycling read-through transcription from a single adjacent gene (Menet et al., 2012; Rodriguez et al., 2013), they appear to be a real group, supported by the fact that genes exhibiting such profiles have also been identified in earlier studies (Wuarin and Schibler, 1990; Millar and Kay, 1991). These genes have usually been suggested to display such profiles due to high stability at the mRNA level (Wuarin et al., 1992), which poses an interesting question regarding the reason behind the presence of rhythmicity at the transcriptional level in the first place: why would a gene be rhythmically transcribed only to be later stabilized and lose daily variation? (See the next section.) Is such rhythmic information lost and rendered irrelevant in these cases?
For the sake of discussion, let us propose a practical (and very loose) definition of a clock-controlled gene. This would include all those genes that are shown to have rhythmic RNA or protein levels or that the function/activity of the final gene product (whether it is RNA or protein) is controlled by the clock. In this scenario, the R/AR genes would not be classified as clock controlled, unless there was information regarding clock regulation on the final product. This raises the obvious question as to whether nonoscillating mRNAs can lead to rhythmic protein expression. The answer is yes.
Various studies, using the mouse liver and other tissues as a model systems, have reported that a significant number of proteins appear to derive from nonoscillating mRNAs (Reddy et al., 2006; Deery et al., 2009; Mauvoisin et al., 2014; Robles et al., 2014). In addition, and also by probing the mouse liver, the Gachon group reported circadian clock control of ribosome biogenesis, together with the rhythmic translation of a subset of mRNAs, some of which lacked rhythmic mRNA abundance (Jouffe et al., 2013). These results suggest that either translation or protein stability is subjected to daily regulation. As a whole, these studies show that oscillating mRNA levels are not a prerequisite for rhythmic protein accumulation. Could rhythmic transcription play a role in any of these cases?
Again with the mouse liver as a research system, the Green lab reported robust rhythms in poly(A) tail length of hundreds of mRNAs and showed that this rhythmicity is closely correlated with rhythmic protein expression (Kojima et al., 2012)—that is, poly(A) tail length rhythms are a good predictor for rhythmic protein levels. Their results suggest that ~80% of these genes exhibit such rhythmic poly(A) tail length as a result of nuclear adenylation coupled with rhythmic transcription, regardless of the steady-state levels of the corresponding message. Indeed, while ~48% of the genes exhibiting rhythmic poly(A) tails also showed rhythmicity at the pre-RNA (a proxy for studying regulation at the transcriptional level) and at the steady-state mRNA level, ~32% of them displayed rhythms in transcription but flat steady-state mRNA levels. These genes are rhythmically transcribed, and the phase of poly(A) tail rhythms correlates well with pre-mRNA rhythms, which led the authors to suggest that they are rhythmically polyadenylated with long initial poly(A) tails as they are rhythmically transcribed in the nucleus, likely by the canonical poly(A) polymerase, which also appears to be rhythmically expressed (Kojima et al., 2012). Rhythmic transcription, then, could lead to rhythmic protein production, regardless of whether the mRNA oscillates under the same conditions. This suggests that some R/AR genes may indeed have rhythmic protein levels as a result of their oscillating transcription (and hence be classified as clock-controlled genes), and the results advise against dismissing this gene set when studying the circadian proteome and the mechanisms underlying it. As promoters (and its associated transcription factors) can also regulate splicing patterns (Kornblihtt, 2005) and selection of polyadenylation sites or poly(A) tail length (Nagaike et al., 2011; Oktaba et al., 2015), transcriptional regulation can have profound effects on protein levels regardless of total mRNA levels.
Recently, promoter elements have also been shown to affect not only subcellular localization but also translation efficiency under particular conditions in yeast (Zid and O’Shea, 2014). In this work, the authors observed that upon glucose starvation, two classes of transcriptionally upregulated genes could be observed, which differed in their subcellular distribution and in the translational efficiency of the corresponding mRNAs. Genes in both classes were induced, at the transcript level, to similar degrees upon stress, but the resulting mRNAs had different fates. Notably, it was shown that specific promoter cis elements were necessary and sufficient for these differences in localization and protein output levels, which the authors propose may allow for selective translation aimed at producing proteins that are needed for adaptation to stress conditions, when overall translation is usually reduced (Zid and O’Shea, 2014). As in the aforementioned cited examples, it is possible that transcription/promoter-level regulation may mediate transcript imprinting or regulation of the mRNA molecule itself (e.g., of the poly(A) tail length), ultimately determining transcript fate. These results further support that transcriptional regulation can affect protein levels regardless of the steady-state levels of the resulting mRNAs.
How prevalent are rhythms in transcription?
Our discussion has centered on the fact that transcriptional regulation could affect the circadian transcriptome/proteome in a variety of ways (some less intuitive than others) and that disregarding a functional connection between each of the different steps in gene expression is very likely to misrepresent their relative contributions to determining molecular rhythms. We have made the case that transcription might have a more relevant role in determining such rhythms as currently estimated based on mRNA and protein levels, but a pressing question in such argument would then be to determine how much exactly of the genome is rhythmically transcribed.
While the 3 main studies on mammalian systems that have prompted this discussion suggest that only about 5% to 15% of genes appear to be rhythmically transcribed in the liver (Koike et al., 2012; Le Martelot et al., 2012; Menet et al., 2012), it seems that overall, thousands of genes exhibit rhythmic recruitment of RNAPII and daily variation in histone marks, regardless of whether rhythmic transcription occurs for those genes (Koike et al., 2012; Le Martelot et al., 2012; Vollmers et al., 2012). In addition, the distribution of peak phases for rhythmically expressed genes is less broad than for rhythmic mRNAs, which led the authors to propose that a global circadian regulation of the novo transcription might exist (Koike et al., 2012; Partch et al., 2013). Such histone modification marks may set the stage for genome-wide recruitment of RNAPII at a specific time of the day, which may result in daily fluctuations in transcriptional activity. It may be possible, then, that the circadian clock may affect gene expression in a global manner, such that specific clock-associated transcriptional regulators allow for the daily changes in histone modifications and global recruitment of RNAPII, preparing the way for gene expression on a genome-wide scale. A similar scenario of a specific time window for global circadian regulation has been suggested to take place at the translational level (Jouffe et al., 2013). In a way, such generalized transcriptional regulation would be similar to the situation in cyanobacteria, in which the whole genome appears to be rhythmically transcribed, as determined by luciferase reporter constructs (Liu et al., 1995), although only 30% to 60% of mRNA abundances appear to be rhythmic (Johnson et al., 2011). In fact, the presence of extensive rhythms in transcriptional regulation in eukaryotes that appear not be extended to the mRNA level has been suggested to reflect evolutionary remnants of such global regulation (Stempfl et al., 2002). Such time-specific global processes (transcription and translation) may alternatively have a functional origin, in that they may result from commitment of the specific machineries to operate at a time of day when resources are more readily available or when interference with other cellular processes is minimized (Jacobshagen et al., 2008). This would then have a general function and may not be appreciated, or make particular sense, at the single gene level (i.e., why is this particular gene rhythmically transcribed/translated only to result in a flat mRNA/protein?).
With all this in mind, it could be proposed that the circadian clock may somehow regulate RNAPII recruitment and histone modifications genome-wide, at a specific time of the day, but for some genes, additional actors, operating at enhancers and/or promoters, may ultimately modulate whether a particular gene is rhythmically transcribed. These regulators might be expressed only in particular tissues/conditions, so each cell may only manifest a small fraction of the circadian transcriptional regulome. In any case, as with mRNA profiling, for which consideration of an increasing number of different tissues and algorithms leads to the identification of daily rhythms at the mRNA level for most genes in mice (Ptitsyn et al., 2007; Ptitsyn and Gimble, 2011; Patel et al., 2014; Zhang et al., 2014), similarly, such approaches may unveil a widespread rhythmicity at the transcriptional level. Most likely, such meta-analyses will show the tissue dependency of the R/R, R/AR, and AR/R gene group memberships. Despite the fact that the circadian clock could in principle have pervasive control of gene expression, tissue/condition-specific regulators, modulating each step in gene expression (e.g., transcription, mRNA steady-state levels, translation), would ultimately define which genes are expressed rhythmically (or determine their phase), so that the genes that would actually be required to oscillate in a particular tissue and at a specific time of day do so appropriately.
Concluding Remarks
In this Commentary, we have urged caution in prematurely assigning relative roles to specific stages in gene expression in the determination of the circadian transcriptome and proteome in eukaryotes. Further studies aimed at evaluating experimental and statistical design—considering noise, reproducibility, and/or systematic errors and biases—are required to critically evaluate current estimates and should be addressed before we can truly assess the contribution of each of these steps in specific tissues/conditions. In addition, as has been shown in other systems, it is quite possible these different steps work cooperatively, rather than competitively, in determining gene expression levels (Csardi et al., 2015), opposing a frequent (and rather naive) approach that usually plays transcriptional and posttranscriptional variation against one another, commonly with the purpose of assigning unnecessary hierarchies.
Rhythms at a single step in gene expression do not necessarily imply rhythms in the next one, nor does lack of a rhythm at one step imply such absence in the following one. We propose that for some genes, events that occur during transcription, even if arrhythmic, can still play a role in rhythmic gene expression. Functional connections between different stages of gene expression (i.e., coupling), particularly between transcription and downstream processes—as discussed throughout this Commentary—may have functional circadian consequences. Thus, the clock might impinge on all steps of gene expression in complex and sometimes unintuitive ways.
While we do not propose this to be a generalized mechanism responsible for most mRNA and protein rhythms, we suggest that it is possible that regulation at the transcriptional level, through specific cis elements and their associated transcription factors, can determine the circadian fate of the resulting mRNA and protein levels, with implications for our definition of “clock-controlled genes.”
Footnotes
Acknowledgements
We thank John B. Hogenesch (University of Pennsylvania) and Editor William J. Schwartz (University of Massachusetts Medical School) for critical and insightful comments. Work in our laboratory is funded by Millennium Nucleus for Fungal Integrative and Synthetic Biology (NC120043) and Fondo Nacional de Desarrollo Científico y Tecnológico (FONDECYT 1131030).
Conflict of Interest Statement
The author(s) have no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.