Multiple circadian transcriptional elements cooperatively regulate cell-autonomous transcriptional oscillation of Period3, a mammalian clock gene

Cell-autonomous oscillation in clock gene expression drives circadian rhythms. The development of comprehensive analytical techniques, such as bioinformatics and ChIP-sequencing, has enabled the genome-wide identification of potential circadian transcriptional elements that regulate the transcriptional oscillation of clock genes. However, detailed analyses using traditional biochemical and molecular-biological approaches, such as binding and reporter assays, are still necessary to determine whether these potential circadian transcriptional elements are actually functional and how significantly they contribute to driving transcriptional oscillation. Here, we focused on the molecular mechanism of transcriptional oscillations in the mammalian clock gene Period3 (Per3). The PER3 protein is essential for robust peripheral clocks and is a key component in circadian output processes. We found three E box–like elements located upstream of human Per3 transcription start sites that additively contributed to cell-autonomous transcriptional oscillation. However, we also found that Per3 is still expressed in a circadian manner when all three E box–like elements are functionally impaired. We noted that Per3 transcription was activated by the synergistic actions of two D box–like elements and the three E box–like elements, leading to a drastic increase in circadian amplitude. Interestingly, circadian expression of Per3 was completely disrupted only when all five transcriptional elements were functionally impaired. These results indicate that three E box–like and two D box–like elements cooperatively and redundantly regulate cell-autonomous transcriptional oscillation of Per3.

Cell-autonomous oscillation in clock gene expression drives circadian rhythms. The development of comprehensive analytical techniques, such as bioinformatics and ChIP-sequencing, has enabled the genome-wide identification of potential circadian transcriptional elements that regulate the transcriptional oscillation of clock genes. However, detailed analyses using traditional biochemical and molecular-biological approaches, such as binding and reporter assays, are still necessary to determine whether these potential circadian transcriptional elements are actually functional and how significantly they contribute to driving transcriptional oscillation. Here, we focused on the molecular mechanism of transcriptional oscillations in the mammalian clock gene Period3 (Per3). The PER3 protein is essential for robust peripheral clocks and is a key component in circadian output processes. We found three E box-like elements located upstream of human Per3 transcription start sites that additively contributed to cell-autonomous transcriptional oscillation. However, we also found that Per3 is still expressed in a circadian manner when all three E box-like elements are functionally impaired. We noted that Per3 transcription was activated by the synergistic actions of two D box-like elements and the three E box-like elements, leading to a drastic increase in circadian amplitude. Interestingly, circadian expression of Per3 was completely disrupted only when all five transcriptional elements were functionally impaired. These results indicate that three E box-like and two D box-like elements cooperatively and redundantly regulate cell-autonomous transcriptional oscillation of Per3.
Almost all organisms exhibit circadian rhythms in physiology and behavior, which are driven by the circadian clock. The circadian clock enables maximum expression of genes at appropriate times of the day, allowing organisms to adapt to the earth's rotation. The circadian clock is phase-adjusted in response to environmental cues such as light and nutrients in a flexible but phase-dependent manner. The suprachiasmatic nucleus functions as a standard clock and orchestrates peripheral clocks to prevent internal desynchronization (1). The molecular machinery for the circadian clock is cell-autono-mous and ubiquitous throughout the body (2,3). The circadian clockwork consists of a negative feedback loop of transcription, which drives circadian gene expression (4). In mammals, the BMAL1 and CLOCK complex activates transcription of clock and clock-related genes, such as Period (Per) and Cryptochrome (Cry) via E-box elements. Subsequently, PER together with CRY, a potent transcriptional inhibitor, function to negatively regulate this complex (5,6). Recent studies have suggested that the PER-mediated transient neutralization of CRY may cause a time delay in CRY-mediated transcriptional repression, which may help to extend the transcriptional oscillation period (7)(8)(9).
The mammalian clock gene, Period3 (Per3), one of three homologs of the Drosophila gene, Period, was cloned in 1998 (10,11). Studies have shown that the circadian period length of locomotor activity rhythms in Per3-deficient mice are ϳ0.5 h shorter than that in wild-type mice, whereas other circadian characteristics examined are mostly normal compared with mice lacking Per1 and Per2 (12,13). Therefore, it has been hypothesized that PER3 protein has a dispensable role in the clock oscillator machinery. However, recent studies indicate that PER3 protein may be a tissue-specific component of peripheral clocks (14,15). In addition, genetic analyses of human sleep disorders have revealed that circadian rhythm sleep disorders, such as delayed sleep phase syndrome, are associated with polymorphisms in the coding region of Per3 (16,17), suggesting that PER3 may contribute to circadian physiological and behavioral outputs. Furthermore, recent studies indicate that Per3 polymorphisms with amino acid changes are associated with familial advanced sleep phase accompanied with mood disorders (18). Additionally, PER3 protein may be an inhibitor of adipocyte cell fate, because Per3-deficient mice have greater adiposity compared with wild-type mice (19). Therefore, better molecular understanding of PER3 function would benefit further research into the prevention and treatment of these circadian-related disorders.
Because cell-autonomous oscillation in clock gene expression is the central machinery that drives circadian rhythms in physiology and behavior, deciphering the molecular mechanisms of the transcriptional oscillation of Per3 will contribute to the general understanding of the circadian system. Although a few studies have examined parts of this mechanism using comprehensive analytical approaches, there is currently no consensus on the mechanism underlying Per3 transcriptional oscillation. For example, Bmal1-deficient mice show drastically reduced circadian amplitude of Per3 expression, indicating an indispensable role of E box-mediated regulation in Per3 transcriptional oscillation (20). In support of this result, a ChIP-seq analysis using antibodies targeting circadian transcription factors showed their binding to the upstream region of Per3 transcription start sites (21). However, the specific location of the binding sites and the functional significance of this binding remain unclear. In contrast, a study that combined bioinformatics with experimental validations reported that transcriptional oscillation of the Per3 gene may be regulated by two D-box transcriptional elements (22). However, it remains unclear whether or not functional impairment of these transcriptional elements impairs Per3 circadian transcription.
To clarify the mechanism by which the Per3 gene is expressed in a cell-autonomous circadian manner, we investigated whether individual potential circadian transcriptional elements were actually functional for cell-autonomous circadian transcription using traditional biochemical and molecular-biological approaches.

The ؊779 to ؊170 transcription-regulatory region is essential for the cell-autonomous circadian transcription of Per3
Because various extracellular factors can affect clock gene expression in vivo, real-time monitoring of clock-driven bioluminescence in culture is a powerful method for investigating the molecular mechanisms underlying cell-autonomous circadian transcription.
To narrow down the genomic region essential for cell-autonomous circadian transcription of Per3, deletion constructs of Per3 (Ϫ4792 to ϩ170) Ϫluciferase (luc) were generated using restriction enzymes (Fig. 1A) and transfected into U2OS cells. Under culture conditions, individual cellular clocks are probably desynchronized from each other due to the absence of suprachiasmatic nucleus-driven orchestration (23), resulting in a low amplitude of transcriptional rhythms at the cell-population level. Dexamethasone (DEX), 2 which transiently activates the clock gene Period1 via glucocorticoid receptors, is a well-known synchronizer for peripheral clocks (24). Therefore, after synchronization of cellular clocks by DEX treatment, real-time and continuous monitoring of bioluminescence was performed for a few days (Fig. 1B). Comparisons among the six constructs from Per3 (Ϫ4792 to ϩ170) to Per3 (Ϫ779 to ϩ170) revealed clear circadian rhythms of bioluminescence in all cases, although a deletion-induced minor decrease in circadian amplitude was observed. In contrast, cells containing Per3 (Ϫ170 to ϩ170) Ϫluc and Per3 (Ϫ22 to ϩ170) Ϫluc showed no obvious circadian rhythms in bioluminescence. These circadian rhythmicity evaluations were further assessed using detrended data, which were generated by subtracting the 24-h moving average from the raw data. Cosine curve fitting for detrended data demonstrated a significant circadian oscillation 2 The abbreviations used are: DEX, dexamethasone; EЈ1, EЈ2, and EЈ3, EЈ box 1, 2, and 3, respectively; DBP, albumin D-box-binding protein. After synchronization by dexamethasone treatment (50 nM for 2 h), bioluminescence was measured in the presence of luciferin and integrated for 1 min at 15-min intervals. The peak value of the first curve was set to 1. Relative bioluminescence (y axis) was measured in cpm. The data are from two independent experiments. C, detrended oscillation curves of the data sets in B obtained by subtracting the 24-h simple moving average from the respective raw data. D, circadian transcriptional oscillation of clock genes, hBmal1, mPer2, and hPer3, in U2OS cells transfected with the adenovirus reporter vectors indicated in the top panel. After synchronization by dexamethasone treatment (50 nM for 2 h), bioluminescence was measured in the presence of luciferin and integrated for 1 min at 15-min intervals. Maximum peak values were set to 1. The acrophase was calculated using cosine curve-fitting analysis and compared among the three clock genes. Data were from a single assay.

Multiple circadian elements regulate Per3
in cells containing the six constructs from Per3 (Ϫ4792 to ϩ170) to Per3 (Ϫ779 to ϩ170). In contrast, analysis of detrended data confirmed that circadian rhythms were not detectable in cells containing Per3 (Ϫ170 to ϩ170) Ϫluc and Per3 (Ϫ22 to ϩ170) Ϫluc (Fig. 1C). These results clearly suggest that the genomic region, Per3 (Ϫ779 to Ϫ170), plays a pivotal role in the cell-autonomous circadian transcription of Per3.
We also examined whether Per3-driven bioluminescence rhythms accurately reflect the expression characteristics of endogenous Per3. U2OS cells were infected with adenovirus vectors containing Per3-, Per2-, or Bmal1-luc, and after synchronization by DEX treatment, real-time and continuous monitoring of bioluminescence was performed for a few days (Fig. 1D). Previous studies have shown that the circadian phase of endogenous Per2 mRNA levels is slightly delayed compared with that of Per3, whereas endogenous Bmal1 mRNA levels are antiphasic to that of Per2 and Per3 (25). The circadian phase relationship among these three genes was highly consistent between endogenous mRNA and bioluminescence levels. We compared the phase intervals of bioluminescence rhythms obtained using three adenovirus vectors in this study with those of endogenous mRNA rhythms reported previously (25); the phase intervals of bioluminescence and endogenous mRNA rhythms between Bmal1 and Per3 were 12.1 and 12.2 h, whereas those between Bmal1 and Per2 were 13.2 and 13.3 h, respectively. This result strongly suggests that the genomic region used in this study contains almost all endogenous transcriptional elements essential for cell-autonomous circadian transcription of Per3.

Three E box-like transcriptional elements cooperatively regulate the cell-autonomous circadian transcription of Per3
An in silico search revealed that the Per3 (Ϫ779 to Ϫ170) region contains three E box-like sequences (in order from upstream to downstream): EЈ box 1 (EЈ1, CACGCG), EЈ box 2 (EЈ2, CACGCG), and EЈ box 3 (EЈ3, CTCGAG) ( Fig. 2A). Previous studies indicate that the circadian transcription factors BMAL1 and CLOCK bind to canonical E box (CACGTG) as a heterodimer to activate transcription (26). A comprehensive ChIP-seq analysis revealed that BMAL1 binds around Ϫ335 upstream of the transcription start site of Per3 (21), which corresponds to the location of the three EЈ-boxes within the region Ϫ445 to Ϫ389. To determine whether BMAL1 and CLOCK activate these three EЈ-boxes, we performed a reporter assay (Fig. 2B) using various deletion mutants of the Per3 (Ϫ779 to ϩ170) region, as schematically shown in Fig. 2A. BMAL1-CLOCK-mediated transcriptional activation was reduced by the loss of EЈ1 and EЈ3 but not EЈ2. As expected, double-deletion mutants, such as ⌬(EЈ1, EЈ2) and ⌬(EЈ1, EЈ3), exhibited further reduced transcriptional activity compared with singledeletion mutants. Unexpectedly, however, the double deletion mutant ⌬(EЈ2, EЈ3) showed increased transcriptional activity compared with the single deletion mutant ⌬EЈ3. In the absence of all three EЈ-boxes, BMAL1 and CLOCK could no longer activate transcription. These results demonstrate that the three EЈ-boxes function as circadian transcriptional elements in an independent and cooperative manner.
The physical interaction between BMAL1-CLOCK and EЈboxes was examined by an EMSA (Fig. 2C). Incubation of affinity-purified Myc-tagged BMAL1 and FLAG-tagged CLOCK with a 32 P-labeled DNA probe containing EЈ1 resulted in an expected mobility-shifted band containing a complex of proteins and DNA by electrophoresis. This band was abolished by the addition of an excess amount of a non-labeled competitor. Co-incubation of anti-Myc or FLAG antibody induced a supershifted band, confirming specific binding. No mobility-shifted band was detected when using a probe containing nucleotidesubstituted EЈ1. Interestingly, whereas no shifted band was detected using probes containing EЈ2 or EЈ3 alone, both a shifted and supershifted band were observed when using a probe containing both EЈ2 and EЈ3. This suggests that BMAL1 andCLOCKbindtotandemlyarrangedEЈ2andEЈ3inaninterdependent manner.
To investigate whether the EЈ boxes are indispensable for the cell-autonomous circadian transcription of Per3, real-time and continuous monitoring of bioluminescence was performed in cells transfected with a construct lacking EЈ boxes in the transcription-regulatory region in various combinations (Fig. 2D). Whereas ⌬EЈ1 and ⌬EЈ3 showed reduced Per3 expression levels, ⌬EЈ2 showed unexpected enhanced expression and a shallow circadian trough, indicating the action of a transcription repressor on EЈ2 that is independent of EЈ3. When EЈ3 is dysfunctional, we hypothesize that whereas BMAL1-CLOCK cannot bind to EЈ2, this transcription repressor can; for example, whereas BMAL1-CLOCK activates neither the ⌬(EЈ1, EЈ2) nor ⌬(EЈ1, EЈ3) construct, the transcription repressor can regulate the latter, leading to differing expression levels of the two constructs. This hypothetical transcription repressor may also explain the higher transcriptional activity arising from the ⌬(EЈ2, EЈ3) construct compared with ⌬EЈ3 (Fig. 2, B and D). Detrended data more clearly demonstrate the reduced transcriptional amplitude in ⌬EЈ1 and ⌬EЈ3 and the shallow trough in ⌬EЈ2 (Fig. 2E). The interdependency of EЈ2 and EЈ3 in transcriptional activation and the independency of EЈ2 in transcriptional repression can explain all phenotypic outputs observed using constructs lacking EЈ boxes in various combinations. For example, whereas both the ⌬(EЈ1, EЈ2) and ⌬(EЈ1, EЈ3) constructs exhibited low amplitude transcription because they were not activated by BMAL1-CLOCK, the latter exhibited a deeper circadian trough via the circadian action of a transcription repressor on EЈ2, leading to differing circadian robustness between these constructs. Importantly, circadian rhythms in bioluminescence remained detectable even in the absence of all three EЈ boxes, strongly suggesting the action of other circadian transcriptional elements.
Cosine curve fitting was performed using detrended data to calculate circadian robustness, amplitude, and phase ( Fig. 2, F-H). Although circadian robustness was over 75% for all constructs, it was significantly lower for constructs lacking functional EЈ2 compared with the others due to the shallow circadian trough. Circadian amplitude was reduced in ⌬EЈ1 and ⌬EЈ3 but not ⌬EЈ2 constructs. Differences in amplitude among double deletion mutants may be explained by the hypothesis that neither circadian transcription factors nor repressor(s) can bind to ⌬(EЈ1, EЈ2), only circadian transcription repressor(s)

Multiple circadian elements regulate Per3
can bind to ⌬(EЈ1, EЈ3), and only circadian transcription factors can bind to ⌬(EЈ2, EЈ3). The difference in amplitude correlates with the results of our reporter assay experiments (Fig. 2B). Circadian phase for all deletion constructs, except ⌬EЈ1, was significantly delayed compared with wild type. The constructs lacking all functional EЈ boxes, ⌬(EЈ1, EЈ2) and ⌬(EЈ1, EЈ2, EЈ3), exhibited delays of Ͼ2 h. This phase delay may be due to the presence of other types of circadian transcriptional elements, because deletion of EЈ boxes in the case that circadian transcription of Per3 is controlled by only E box-type elements should result in a reduced amplitude but not a phase shift. Although our results from reporter assays shown in Fig. 2B suggest the absence of other BMAL1-CLOCK-responsive elements in the Per3 region (Ϫ779 to ϩ170) apart from the three EЈ boxes, circadian rhythms in bioluminescence remained detectable when all three EЈ boxes were deleted, supporting the hypothesis that other types of circadian transcriptional elements are present.

Two D box-like transcriptional elements regulate the cell-autonomous circadian transcription of Per3
An in silico analysis identified two D box-like sequences within the Per3 region Ϫ779 to Ϫ170 (Fig. 3A): a canonical D box (Ϫ589, TTATGTAA) and a non-canonical DЈ box (Ϫ548, TTATATAA). Although a previous report demonstrated that these D box-like sequences can regulate the SV40 promoter in a circadian manner (22), it remains unknown whether they are actually required for cell-autonomous circadian transcription of Per3. To examine this, we generated Per3 (Ϫ779 to ϩ170) constructs lacking these two sites and examined their function using reporter assays (Fig. 3B). Overexpression of DBP induced minor (ϳ2-fold) activation of Per3. Deletion of either of the D/DЈ boxes resulted in similar or reduced transcriptional activation, whereas deletion of both D/DЈ boxes completely abolished DBP-induced activation. Interestingly, overexpression of BMAL and CLOCK elevated DBP-mediated activation by Ͼ9-fold, suggesting their synergistic cooperation in regulating Per3 transcription.
We examined the contribution of the D/DЈ boxes to the circadian transcription of Per3 (Fig. 3C). Deletion of either of the D/DЈ boxes resulted in reduced expression of Per3. The contribution of the DЈ box to expression levels was higher than that of the D box. Circadian rhythms in bioluminescence remained detectable in the absence of both D/DЈ boxes. Detrended data highlighted the contribution of the D/DЈ boxes to circadian amplitude and revealed changes in the circadian phase in deletion mutants (Fig. 3D). To evaluate the effect of D/DЈ-box deletions on circadian characteristics in detail, cosine curve fitting was performed using detrended data to calculate circadian robustness, amplitude, and phase (Fig. 3, E-G). Interestingly, whereas deletion of the D/DЈ boxes resulted in a drastic reduction in amplitude, circadian robustness remained close to 100%, indicating that the D/DЈ boxes contribute to increasing amplitude rather than driving oscillation. Loss of either or both of the D/DЈ boxes advanced the circadian phase of Per3 transcription, an opposing phenotype to that observed after the deletion of EЈ boxes (Fig. 2H).

Three E boxes and two D/D boxes cooperatively control the cell-autonomous circadian transcription of Per3
Because the separate deletion of either all EЈ boxes or all D/DЈ boxes did not completely abolish the circadian transcription of Per3, we examined the effect of simultaneously deleting all three EЈ boxes and two D/DЈ boxes on circadian transcription (Fig. 4A). We constructed a total of 20 reporter vectors by introducing single or double deletions of D/DЈ boxes into the Per3 (Ϫ779 to ϩ170) region lacking EЈ boxes in different combinations. Our data clearly demonstrate the contribution of the D/DЈ boxes to Per3 expression levels. Importantly, deletion of the D/DЈ boxes in the absence of all three EЈ boxes did not affect Per3 expression levels. This indicates that D/DЈ box-mediated transcriptional activation requires functional EЈ boxes, which is consistent with our data showing that simultaneous overexpression of DBP and BMAL1-CLOCK results in the synergistic activation of Per3 transcription. Detrended data clearly demonstrated that deletion of the D/DЈ boxes decreased the amplitude White circles indicate individual data points from replicates. Mean values were compared with that of wild-type (Ϫ779 to ϩ170) constructs containing BM ϩ CL as controls using Dunnett's test, and asterisks indicate statistical significance (p Ͻ 0.05). A t test at ␣ ϭ 0.05 was performed for two combinations: wild type (Ϫ779 to ϩ170) versus wild type (Ϫ779 to ϩ170) ϩ [BM1 ϩ CL] or wild type (Ϫ779 to ϩ170) versus ⌬ (EЈ1, EЈ2, EЈ3) ϩ [BM1 ϩ CL], which are indicated by the short and long brackets, respectively. # and NS, statistical significance and no statistical significance, respectively. C, electrophoretic mobility shift assay. A 32 P-labeled double-stranded oligonucleotide containing the EЈ-box sequence(s) indicated in the top panel was used as a probe. Non-radiolabeled EЈ-box consensus probes were used as competitors. The boxed and shaded nucleotide sequences indicate the targeted EЈ-box and mutated nucleotides, respectively. Heterodimerized protein of Myc-BM1 and FLAG-CL were purified by immunoprecipitation using FLAG antibody-conjugated beads from HEK293A cells co-transfected with Myc-BM1 and FLAG-CL expression vectors. The BM1 ϩ CL-bound probes are indicated by the arrowhead labeled Band shift. The addition of c-Myc or FLAG antibodies resulted in supershifted BM1 ϩ CL-bound probes. Probes not bound to the proteins appeared as free probe. D, real-time monitoring of bioluminescence in U2OS cells transfected with the EЈ box-deleted constructs indicated in A. Graphs show representative results from three replicates of a single assay. After synchronization by dexamethasone treatment (50 nM for 2 h), bioluminescence was measured in the presence of luciferin and integrated for 1 min at 15-min intervals. Bioluminescence (y axis) was measured in cpm. Schematic representations of the deleted EЈ-box elements (light gray cross-marked boxes) are indicated above each graph. Light gray curves represent the wild-type (Ϫ779 to ϩ170) construct. E, relative detrended data sets obtained across a 2-day period were obtained by subtracting the 24-h simple moving average from three replicates of raw data. Light gray curves represent the wild-type (Ϫ779 to ϩ170) construct, and the peak value of the first curve was set to 1. F-H, robustness, amplitude, and acrophase were calculated from the data set in E by curve fitting using the Cosinor software provided by Dr. Refinetti. The amplitude and acrophase of the wild-type (Ϫ779 to ϩ170) construct were set to 1 and 0, respectively. The data indicate mean Ϯ S.D. (n ϭ 3). White circles in F and G indicate individual data points from replicates. The mean of each factor was compared with that of the wild type (Ϫ779 to ϩ170) using the Dunnett's test, and asterisks indicate statistical significance (p Ͻ 0.05).
Cosinor analysis using detrended data revealed that deletion of the D/DЈ boxes had no effect in ⌬EЈ1 but significantly reduced circadian robustness in ⌬EЈ3 (Fig. 4C). Deletion of EЈ2 led to relatively low robustness due to a shallow circadian trough, as already shown in Fig. 2 (E and F), and additional deletion of the D/DЈ boxes further decreased robustness. Multiple deletion constructs, such as ⌬(D, DЈ, EЈ2), ⌬(D, DЈ, EЈ2, EЈ3), and ⌬(D, DЈ, EЈ1, EЈ2, EЈ3), reduced robustness to less than 50%. ⌬(D, DЈ, EЈ2, EЈ3) and ⌬(D, DЈ, EЈ1, EЈ2, EЈ3) were classified as arrhythmic, as they showed non-circadian periods of Per3 transcription (34 h or more) in two of three or all measurements, respectively. Although the negative effect of D/DЈ-box deletion on circadian amplitude was very strong, the effect in White circles indicate individual data points from replicates. Among cells containing the wild-type (Ϫ779 to ϩ170) construct, Dunnett's test (␣ ϭ 0.05) was used to analyze differences in mean values compared with the control, which contained no transcription factor (white bar), and asterisks indicate statistical significance (p Ͻ 0.05). # and NS, statistically significant and not significant differences determined by the t test at ␣ ϭ 0.05 compared with controls (white bars), respectively. C, real-time monitoring of bioluminescence in U2OS cells transfected with the D/DЈ box-deleted constructs indicated in A. Representative curves indicate data of three replicates from a single assay. After synchronization by dexamethasone treatment (50 nM for 2 h), bioluminescence was measured in the presence of luciferin and integrated for 1 min at 15-min intervals. Bioluminescence (y axis) was measured in cpm. Schematic representation of the deleted D/DЈ-box elements (light gray cross-marked boxes) are indicated above each graph. Light gray curves represent the wild-type (Ϫ779 to ϩ170) construct. D, relative detrended data sets obtained across a 2-day period were obtained by subtracting the 24-h simple moving average from three replicates of raw data. The light gray curves represent the wild-type (Ϫ779 to ϩ170) construct, and the peak value of the first curve was set to 1. E-G, robustness, amplitude, and acrophase from the data set in D were calculated by curve fitting using the Cosinor software provided by Dr. Refinetti. Amplitude and acrophase of the wild-type (Ϫ779 to ϩ170) construct were set to 1 and 0, respectively. Data indicate the mean Ϯ S.D. of three replicates, and white circles in E and F indicate individual data points from replicates. Mean values were compared with that of the control using the Dunnett's test, and asterisks indicate statistical significance (p Ͻ 0.05).

Multiple circadian elements regulate Per3
the absence of all three EЈ boxes was small or unclear due to the lack of DBP-mediated transcriptional activation (Fig. 4D). Consistent with the results in Fig. 3G, additional deletion of the D/DЈ boxes in EЈ-box mutants resulted in a phase advance, which was statistically significant only for ⌬EЈ1 constructs (Fig.  4E). The lack of statistically significant differences in phase may be attributed to the reduced reliability of the calculated phases in EЈ-box mutants due to the low robustness observed for all ⌬EЈ constructs except for ⌬EЈ1 after the additional deletion of D/DЈ boxes (Fig. 4C).

Discussion
Although PER3 is a component of peripheral clocks and has indispensable roles in circadian outputs, the mechanism underlying the cell-autonomous circadian transcription of Per3 remains unclear. To elucidate this mechanism, we conducted detailed analyses of circadian transcriptional elements. The transcription-regulatory region examined in this study, Per3 (Ϫ4792 to ϩ170), contains almost all of the circadian transcriptional elements required for cell-autonomous circadian transcription of Per3. Although we focused on and further analyzed the core region Per3 (Ϫ779 to Ϫ170), comparison of Per3 (Ϫ4792 to ϩ170) and Per3 (Ϫ779 to ϩ170) constructs (Fig. 1, B and C) showed that removal of a genomic region upstream of Per3 (Ϫ779) resulted in a slight reduction in amplitude. This suggests that although this region may not be essential, it does play a role in cell-autonomous circadian transcription. In support of this, we found potential E box-like sequences within this region.
We identified three EЈ-box sequences (EЈ1, EЈ2, and EЈ3) within the core transcription-regulatory region, Per3 (Ϫ779 to Ϫ170), and showed that they were regulated by BMAL1-CLOCK and contributed to the circadian expression of Per3, as expected. A non-canonical E box has been reported to play an indispensable role in the cell-autonomous circadian transcription of another Per gene, Per2 (27,28). Interestingly, the three EЈ boxes identified within the transcription-regulatory region of Per3 have different characteristics. First, an in silico comparison of these EЈ boxes among mammals revealed that EЈ1 is only found in humans. This indicates that the circadian expression of human Per3 requires more robust regulation compared with that in other mammals, which supports the frequent reports of a genetic association between human Per3 and circadian-related disorders. Second, EЈ2 and EЈ3 are interdependent, with the simultaneous binding of BMAL1-CLOCK to both EЈ boxes required for transcriptional activation. Tandemly repeated E/EЈ boxes have been reported to contribute to the amplitude and non-linearity of circadian transcription (29 -32). It remains unclear why both EЈ2 and EЈ3 are required for the binding of BMAL1-CLOCK despite the fact that the core sequences in EЈ2 and EЈ1 are exactly the same. It is possible that nucleotide sequences surrounding EЈ boxes may have a role in determining the different characteristics of these elements. For example, previous reports demonstrated a strong preference of circadian transcription factors for an adenine nucleotide in the flanking region of E boxes (32,33); therefore, it is possible that the lack of a flanking adenine may be responsible for the reduced binding affinity between BMAL1-CLOCK and EЈ2. Further studies are required to confirm this. Third, there is strong evidence from reporter assays and bioluminescence monitoring indicating that a transcription repressor binds to EЈ2 independently of EЈ3 in the absence of BMAL1-CLOCK. Consistent with this hypothesis, an in silico analysis showed that inhibitory basic helix-loop-helix transcription factors were more likely to bind to EЈ2 and its surrounding sequences, compared with other EЈ boxes. DEC1 and DEC2, inhibitory basic helix-loop-helix factors with circadian transcription repressor action, may constitute such repressors (34).
We showed that two D/DЈ boxes underlie the detectable circadian transcription of Per3 in the absence of the three EЈ boxes. Importantly, access to the D/DЈ boxes by DBP may be dependent on the activation state of the EЈ boxes. In other words, BMAL1 and CLOCK may function to convert the transcription-regulatory region of Per3 into an active state to enable DBP to access the D/DЈ boxes and synergistically activate Per3 transcription, leading to a drastic increase of circadian amplitude. Several lines of evidence support this hypothesis. First, overexpression of DBP alone resulted in only minor activation of Per3 transcription, but in the presence of BMAL1 and CLOCK overexpression, it induced synergistic activation of Per3 transcription (Fig. 3B). Second, loss of the two D/DЈ boxes did not affect circadian robustness but resulted in a drastic decrease in circadian amplitude (Fig. 3C). Third, loss of the two D/DЈ boxes did not affect circadian amplitude in the absence of all three EЈ boxes (Fig. 4A).
The role of each of the three EЈ boxes and two D/DЈ boxes is summarized in Fig. 5A. The EЈ1, EЈ3, D, and DЈ boxes contribute to the maintenance of circadian amplitude, whereas EЈ2 and EЈ3 activate transcription in an interdependent manner. We hypothesize that a deep circadian trough may result from the binding of a transcription repressor to EЈ2 in the absence of BMAL1-CLOCK; however, further studies are required to confirm this. Interestingly, loss of the EЈ boxes and D/DЈ boxes delays and advances the circadian phase of Per3 transcription, respectively. The mechanisms underlying these differential Representative graphs show data from three replicates from a single assay. After synchronization by dexamethasone treatment (50 nM for 2 h), bioluminescence was measured and integrated for 1 min at 15-min intervals. Bioluminescence (y axis) was measured in cpm. Schematic representations of the deleted EЈ-and D/DЈ-box elements (light gray cross-marked boxes) are indicated above each graph. Light gray curves represent the wild-type (Ϫ779 to ϩ170) construct. B, relative detrended data sets obtained across a 2-day period were obtained by subtracting the 24-h simple moving average from three replicates of raw data. Light gray curves represent the wild-type (Ϫ779 to ϩ170) construct, and the peak value of the first curve was set to 1. C-E, robustness, amplitude, and acrophase from the data set in B were calculated by curve fitting using the Cosinor software provided by Dr. Refinetti. Amplitude and acrophase of the wild-type (Ϫ779 to ϩ170) construct (Fig. 2) were set to 1 and 0, respectively. Data indicate mean Ϯ S.D. (error bars) (n ϭ 3). White circles in C and D indicate individual data points from replicates. The mean values were compared with that of the wild-type D and DЈ boxes (darkest gray bars or circles) within each ⌬EЈ group using the Dunnett's test, and asterisks indicate statistical significance (p Ͻ 0.05).

Multiple circadian elements regulate Per3
phenotypes in circadian phase are summarized in Fig. 5B. When the transcription-regulatory region of Per3 contains only EЈ boxes and no D/DЈ boxes, Per3 expression is correlated with the circadian pattern of BMAL1-CLOCK activity. In contrast, when only D/DЈ boxes and no EЈ boxes are present, Per3 expression is correlated with DBP activity, although the circadian amplitude is very low due to the limited access of DBP to the D/DЈ boxes. The circadian phase of DBP activity is delayed compared with that of BMAL1-CLOCK activity because Dbp transcription is controlled by BMAL1-CLOCK (35). When the transcription-regulatory region of Per3 contains both types of transcriptional elements, active BMAL1-CLOCK initiates Per3 transcription through the EЈ boxes. Accessibility of DBP to the D/DЈ boxes subsequently increases following the activation of EЈ boxes, and BMAL1-CLOCK-induced DBP accelerates Per3 transcription in a synergistic manner. This cooperative activation of Per3 transcription by EЈ-and D/DЈ boxes underlies the observed enhanced amplitude and delayed phase compared with those arising due to activation of EЈ boxes alone. A gradual increase in the amount of the nuclear PER-CRY complex subsequently inactivates the EЈ boxes, and DBP-mediated acceleration of Per3 transcription is simultaneously silenced even in the presence of a sufficient amount of intracellular DBP, due to limited access to the D/DЈ boxes. The relative amount of Per3 transcripts therefore begins to decrease earlier than that observed for D/DЈ boxes alone. This theory may also explain the regulation of other clock genes.
Future studies should examine the potential role of surrounding nucleotides on the different characteristics of EЈ boxes, identify unknown transcription repressor(s) that bind to EЈ2, and elucidate ways to control DBP accessibility to D/DЈ boxes.

Plasmid construction
The transcription-regulatory region (Ϫ4792 to ϩ170) of the human Period3 gene was excised from a BAC clone (RP3-467L1) using XbaI and XhoI and subcloned into the pGL3 basic vector (Promega). Vectors containing varying lengths of Per3 transcription-regulatory regions (from Ϫ3852/ϩ170 to Ϫ22/ ϩ170) were constructed by digesting the excised hPer3 sequence (Ϫ4792 to ϩ170) with several restriction enzymes (EcoRV, KpnI, PstI, Tth111I, PmlI, BssHII, and SacI) and subsequently subcloning them into the pGL3 basic vector. The 5Јand 3Ј-regions of the cloned inserts were checked by sequencing. Site-directed deletion mutagenesis of hPer3 (Ϫ779 to ϩ170) ϪpGL3 was conducted according to the KOD-Plus mutagenesis kit protocol (TOYOBO, Osaka, Japan). A PCR product of the entire hPer3 (Ϫ779 to ϩ170) ϪpGL3 vector was obtained using primers designed to introduce a deletion of the target elements between their 5Ј termini. After digesting the template plasmid vector with DpnI enzyme, which digests methylated DNA, self-ligation of the PCR products was performed using T4 polynucleotide kinase and ligase. The ligated plasmid DNA was transformed into competent Escherichia coli cells (E. coli DH5␣ competent cells; Takara, Tokyo, Japan). Plasmid DNA was purified from the transformed E. coli cells using an automated DNA isolation system (PI-80X, KURABO, Osaka, Japan). To check that the target elements were deleted, plasmid DNA that was digested with restriction enzymes was electrophoresed, and the consistency of band lengths was confirmed. The plasmid DNA was purified again using the QIAquick PCR purification kit (Qiagen) before transfection experiments. The mutated plasmid constructs had deletions of 4 -9 base pairs overlapping the wild-type target sequences. The
For the luciferase assay and EMSA experiments, the mBmal1, mClock, and hDbp coding sequences were subcloned into the pcDNA3 expression vector (Invitrogen). BMAL1 and CLOCK protein were expressed as fusion proteins with c-Myc and FLAG tags, respectively.

Adenovirus vector construction
Adenoviruses carrying the luciferase gene driven by circadian promoter/enhancer elements controlling the hBmal1, mPer2, and hPer3 clock genes were constructed as follows. The transcription-regulatory region (Ϫ4792 to ϩ170) of the human Period3 gene was excised from a BAC clone (RP3-467L1) using XbaI and XhoI and subcloned into the pGL3 basic vector (Promega). The resulting vector was digested using KpnI and selfligated to produce hPer3 (Ϫ3112 to ϩ170) ϪpGL3. In addition to hPer3, the previously constructed vectors, mPer2 (Ϫ2811 to ϩ110) ϪpGL3 and hBmal1 (Ϫ3465 to ϩ57) ϪpGL3, were used for adenovirus vector construction (27,36). Together with the luciferase gene, the transcription-regulatory region of these three genes was excised from each pGL3 vector, and the inserts were subcloned into the pENTR-1A vector (Thermo Fisher Scientific). The restriction enzymes used for digestion of the pGL3 vectors were as follows: hPer3-luc, Asp718 (KpnI) and SalI; mPer2-luc, Asp718 (KpnI) and SalI; hBmal1-luc, SalI (two restriction sites, one at the 3Ј-end of the SV40 late poly(A) signal sequence and one inside hBmal1). The resulting vectors were hPer3 (Ϫ3112 to ϩ170) Ϫluc-pENTR-1A, mPer2 (Ϫ2811 to ϩ110) Ϫluc-pENTR-1A, and hBmal1 (Ϫ1683 to ϩ57) Ϫluc-pENTR-1A. These entry vectors were deposited into the RIKEN BioResource Center (RDB15084, RDB15083, and RDB15085, respectively). Before introduction into a destination vector, each pENTR-1A vector was transfected into NIH3T3 fibroblasts and synchronized by a DEX shock (50 nM), and circadian bioluminescence rhythms were confirmed by real-time monitoring with a photomultiplier tube (LM2400, Hamamatsu Photonics, Hamamatsu, Japan) in the presence of luciferin. pENTR-1A inserts were introduced into the pAd/ PL-DEST vector using LR recombination. After PacI digestion, the resulting pAd/PL-DEST vectors were transfected into HEK293A cells to produce an adenoviral stock. The recombinant adenoviruses were generated using the ViraPower adenoviral gateway expression kit (K4940-00, Thermo Fisher Scientific), according to the manufacturer's instructions.

Luciferase assay
U2OS cells were cultured in 24-well plates at 37°C with 5% CO 2 and subsequently transfected with a combination of vectors (an hPer3-luc reporter vector; Renilla luciferase vector, used as an internal control; and clock protein expression vectors). The total amount of DNA used for transfection per well was adjusted to 0.5 g by adding empty pcDNA3. The transfection was performed using Lipofectamine 3000 and P3000 reagents (Invitrogen). After incubation for about 20 h, the cells were lysed and centrifuged at 15,000 rpm for 15 min. The supernatant (20 l) was assayed using the Dual-Luciferase assay system (Promega). Luminescence was measured by a GloMax 20/20 luminometer (Promega). Measurement parameters were set as follows: injection volume, 100 l; delay time after injection, 2 s; integration time, 10 s. Reported data represent firefly luciferase activity normalized to Renilla luciferase activity in relative light units. Dunnett's and Student's t test were used for comparisons. Differences were considered to be significant when the p value was Ͻ0.05.

Real-time monitoring of bioluminescence in living cells and data analysis
U2OS cells were cultured in 35-mm diameter dishes and transfected with the various mutant Per3 promoter/enhancerluciferase constructs using FuGENE6 (Promega). The total amount of DNA used for transfection per dish was adjusted to 1 g by adding empty pcDNA3. After incubation for 24 h, circadian synchronization was conducted by adding DEX (final concentration 50 nM) and further incubating for 2 h. After washing the cells in DMEM (Nacalai Tesque, Kyoto, Japan), the culture medium was changed to luciferin (0.1 mM)-containing DMEM supplemented with fetal bovine serum and penicillin/streptomycin antibiotics. The bioluminescence was measured in real time and integrated for 1 min at 15-min intervals with a photomultiplier tube (LM2400, Hamamatsu Photonics, Hamamatsu, Japan). Detrended data sets were obtained by subtracting the 24-h running average from the raw data. Relative data were calculated by setting the first curve peak to 1. Circadian rhythm acrophase, amplitude, and robustness were calculated using the detrended data sets in the Cosinor software provided by Dr. Refinetti.
Adenovirus vectors expressing circadian-driven luciferase were examined according to the same procedure except that the infection was conducted by adding the adenovirus stocks (1.24 -1.47 ϫ 10 9 infectious units/ml, HEK293A) to culture medium at 1:100 dilution and incubating for 1 h.

EMSA
Heterodimerized proteins of Myc-BMAL1 and FLAG-CLOCK were prepared by immunoprecipitation with FLAG antibody-conjugated beads (Sigma). HEK293A cells in 10-cm diameter dishes were co-transfected with Myc-BMAL1 and FLAG-CLOCK expression plasmids using Lipofectamine and Plus reagents (Invitrogen). Transfected HEK293A cells were lysed in 1.2 ml/dish of immunoprecipitation buffer containing 50 mM Tris-HCl (pH 8.0), 10% glycerol, 100 mM NaF, 50 mM NaCl, 2 mM EDTA, 2 mM sodium orthovanadate, 10 mM sodium pyrophosphate, 25 mM ␤-glycerophosphate, and 1% Nonidet P-40. Lysates was centrifuged at 15,000 rpm for 15 min. Beads were prewashed twice in immunoprecipitation buffer, and 100 l of 50% suspended beads was added to the supernatant from cell lysate. Immunoreactions were conducted on a rotator at 4°C for 2 h, and the beads were washed three times for 5 min. The immunoprecipitants were eluted with 200 g/ml FLAGpeptide (Sigma). Purification was confirmed by silver staining the SDS-polyacrylamide gels using a Silver Stain MS kit (Wako, Osaka, Japan). A double-stranded oligonucleotide containing the EЈ-box consensus (or mutated) sequence was used as an EMSA probe. The double-stranded oligonucleotide (25 pM) was labeled with [␥-32 P]ATP using polynucleotide kinase

Multiple circadian elements regulate Per3
(TOYOBO, Osaka, Japan). G-50 Spin Oligo columns (GE Healthcare) were used to remove free ATP from the labeled probe. Non-radiolabeled EЈ-box consensus probes were used as competitors. Purified clock proteins were mixed with 0.5 nM radiolabeled probe (and competitor) in buffer containing 15 mM Tris-HCl (pH 7.5), 15 mM NaCl, 1.5 mM EDTA, 1.5 mM dithiothreitol, 7.5% glycerol, 0.3% Nonidet P-40, and 1 g/ml BSA and incubated for 10 min. Finally, c-Myc or FLAG antibodies were added to perform supershift experiments. All reactions were incubated at 25°C. The reactions were loaded onto 4% nondenaturing polyacrylamide gels and resolved at 150 V for 2 h at 4°C. The gels were dried and exposed to X-ray film.
Author contributions-R. M. performed experiments, analyzed data, and wrote the manuscript. M. A. conceived and supervised the project and wrote the manuscript.