Activation of 5′-AMP-activated Kinase with Diabetes Drug Metformin Induces Casein Kinase Iϵ (CKIϵ)-dependent Degradation of Clock Protein mPer2*

Metformin is one of the most commonly used first line drugs for type II diabetes. Metformin lowers serum glucose levels by activating 5′-AMP-activated kinase (AMPK), which maintains energy homeostasis by directly sensing the AMP/ATP ratio. AMPK plays a central role in food intake and energy metabolism through its activities in central nervous system and peripheral tissues. Since food intake and energy metabolism is synchronized to the light-dark (LD) cycle of the environment, we investigated the possibility that AMPK may affect circadian rhythm. We discovered that the circadian period of Rat-1 fibroblasts treated with metformin was shortened by 1 h. One of the regulators of the period length is casein kinase Iϵ (CKIϵ), which by phosphorylating and inducing the degradation of the circadian clock component, mPer2, shortens the period length. AMPK phosphorylates Ser-389 of CKIϵ, resulting in increased CKIϵ activity and degradation of mPer2. In peripheral tissues, injection of metformin leads to mPer2 degradation and a phase advance in the circadian expression pattern of clock genes in wild-type mice but not in AMPK α2 knock-out mice. We conclude that metformin and AMPK have a previously unrecognized role in regulating the circadian rhythm.

Metformin is one of the most commonly used first line drugs for type II diabetes. Metformin lowers serum glucose levels by activating 5-AMP-activated kinase (AMPK), which maintains energy homeostasis by directly sensing the AMP/ATP ratio. AMPK plays a central role in food intake and energy metabolism through its activities in central nervous system and peripheral tissues. Since food intake and energy metabolism is synchronized to the light-dark (LD) cycle of the environment, we investigated the possibility that AMPK may affect circadian rhythm. We discovered that the circadian period of Rat-1 fibroblasts treated with metformin was shortened by 1 h. One of the regulators of the period length is casein kinase I⑀ (CKI⑀), which by phosphorylating and inducing the degradation of the circadian clock component, mPer2, shortens the period length. AMPK phosphorylates Ser-389 of CKI⑀, resulting in increased CKI⑀ activity and degradation of mPer2. In peripheral tissues, injection of metformin leads to mPer2 degradation and a phase advance in the circadian expression pattern of clock genes in wild-type mice but not in AMPK ␣2 knock-out mice. We conclude that metformin and AMPK have a previously unrecognized role in regulating the circadian rhythm.
Animal behavior, including spontaneous locomotion, sleeping, eating, and drinking, follows a 24-h light-dark (LD) 2 cycle of the environment. The master pacemaker for the rhythmic behavior lies in the suprachiasmatic nucleus (SCN) of the hypothalamus (1). The SCN neurons, cued by the LD cycle, orchestrate the circadian rhythms of peripheral clocks that reside in most cells of the body. The pacemaker, both in the SCN and in peripheral tissues, consists of a self-sustaining near 24-h rhythm in the expression of core clock genes. A central component of this pacemaker is the negative-feedback loop, which results from Per and Cry proteins suppressing their own transcription with a precisely timed lag. A key regulator of the period length is casein kinase I⑀ (CKI⑀) (2). CKI⑀ and its homolog CKI␦ regulate the circadian period by phosphorylating mammalian Per proteins (3)(4)(5)(6). The role of CKI⑀ in mammalian circadian rhythm is best illustrated by the semidominant mutation in hamster CKI⑀, tau (7). CKI⑀ tau is a highly specific gain-of-function mutation that increases the CKI⑀ kinase activity on Per proteins. CKI⑀-mediated phosphorylation induces proteasome-mediated degradation of Per proteins, leading to circadian phase advance and shortened period length (8).
AMPK, by sensing the rise in AMP level under energy-deprived conditions (9), maintains energy homeostasis by stimulating ATP production and suppressing ATP-consuming processes such as synthesis of macromolecules (10). The catalytic subunit of AMPK has two isoforms, ␣1 (11) and ␣2 (12). Mice with a knockout (KO) of either AMPK ␣1 or AMPK ␣2 are viable (13,14), but mice with a KO of both ␣1 and ␣2 are not viable, 3 indicating that the two isoforms have partially redundant functions. In the hypothalamus, AMPK maintains energy homeostasis at the whole body level by stimulating food intake (15). Activation of AMPK in peripheral tissues such as liver and muscle decreases serum glucose level. Metformin, which activates AMPK (16), is one of the most widely used drugs for treatment of type II diabetes and is, in most cases, the agent of choice for initial therapy (17).
Here, we report that metformin and AMPK have a surprising role in modulating the circadian rhythm. We find that CKI⑀ is activated by AMPK-dependent phosphorylation and stimulates mPer2 degradation. Activation of AMPK by metformin injection shifted the circadian expression pattern of clock genes in wild-type mice but not in AMPK␣2 KO mice.

MATERIALS AND METHODS
Real-time Luminescence Reporting-Rat-1 cells with mPer2luciferase reporter (18) were grown to confluency. Rat-1 cells were then treated with forskolin, and 30 min later, the medium was replaced with the recording medium, containing 10% of fetal bovine serum and HEPES-buffered Dulbecco's modified Eagle's medium. Luminescence was measured for 6 days by counting the photon every 60 s. The period length was determined with regression analyses by using ClockLab (Actimetrics). Four to six peaks detected from the baseline-subtracted * This work was supported by the Intramural Research Program, National Institutes of Health, National Heart Lung and Blood Institute. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. 1  data (the baseline drift determined by an adjacent averaging method with 24 h was corrected by subtracting it from the original data) were used for the period determination. A more detailed description of this technique has been previously published (19).
Mice-Animals were maintained in barrier facilities with sterile food, water, and bedding on a 12-h light/dark (6 a.m./6 p.m.) cycle and were allowed free access to food. All experiments were approved by NHLBI Animal Care and Use Committee.
In Vitro AMPK Kinase Assay-GST-CKI⑀ fusion protein fragments and mutants were expressed and purified using the GST purification kit (Amersham Biosciences) as described by the manufacturer. Reactions (20 l) were performed in kinase buffer (20 mM HEPES-NaOH, pH 7.0, 0.4 mM dithiothreitol, 0.01% Brij-35 with or without 300 M AMP and 15 Ci of [␥-32 P]ATP with 50 milliunits of AMPK (Millipore) and 2 g of GST-CKI⑀ fusion protein substrates. The mixture was incubated 30°C for 20 min and analyzed by SDS-PAGE.
Immunoprecipitation Kinase Assay-Myc-tagged CKI⑀, transiently expressed in 293T cells, was immunoprecipitated with the anti-Myc antibody and 10 l of protein A-agarose beads in 200 l of total volume overnight after preclearing the lysates with protein A-agarose beads for 2 h at 4°C. For the kinase reaction, a His-tagged mPer2 (amino acids 550 -763) fragment was expressed in and purified from Escherichia coli. Kinase reactions were performed as described previously (3).
In Vivo 32 P Labeling Assay-V5-tagged CKI⑀ fragment 2 and its S389A mutant were transiently expressed in 293T cells as described above. 36 h after transfection, cells were preincubated with phosphate-free Dulbecco's modified Eagle's medium for 1 h and labeled with 500 Ci/ml [ 32 P]orthophosphate (Amersham Biosciences) for 2 h. Cells were washed with phosphate-buffered saline, after which they were lysed in radioimmune precipitation buffer containing protease and phosphatase inhibitors. Equal amounts of protein were precleared by incubating them with protein A-agarose beads for 2 h at 4°C. The samples were immunoprecipitated with V5 antibody and protein A-agarose bead overnight at 4°C. The precipitates were washed four times with the cell lysis buffer and dissolved in SDS sample buffer for SDS-PAGE.

RESULTS AND DISCUSSION
Since food intake and energy metabolism are synchronized to the LD cycle, we suspected that some pathways regulated by food intake and energy metabolism may affect the circadian rhythm. Since AMPK plays a central role in food intake and energymetabolism,weinvestigatedthepossibilitythatanAMPKdependent pathway can affect circadian rhythm. The ability to visualize circadian oscillations of gene expression in vitro has been facilitated by the development of a cell culture-based luminescent reporter assay that can monitor the expression levels of the reporter in real time (20). To monitor the circadian variation in the activity of the mPer2 promoter, we used Rat-1 fibroblasts that have been stably transfected with an mPer2-luciferase reporter. After these cells were synchronized with forskolin (21), we treated them with metformin and monitored the luciferase activity in real-time. As shown in Fig. 1, metformin shortened the period length by approximately 1 h (p Ͻ 0.01).
Since proteasomal degradation of mPer2 that is stimulated by CKI⑀ phosphorylation can shorten the circadian period length (4), we explored the possibility that AMPK regulated the period length by stimulating CKI⑀-mediated degradation of mPer2. To do this, we transiently transfected NIH3T3 cells with expression vectors for V5-tagged mPer2 and either CKI⑀ or dominant-negative CKI⑀ (K38R) and treated them with 5-aminoimidazole-4-carboxamide ribonucleoside (AICAR) (22), an activator of  AMPK. As shown in Fig. 2A, mPer2 protein levels decreased dramatically in the presence of CKI⑀, but this was blocked by dominant-negative CKI⑀ (K38R), suggesting that AICAR treatment decreased mPer2 levels in a CKI⑀-dependent manner. To demonstrate that the reduction in mPer2 levels was caused by AMPK activation, we overexpressed constitutively active (CA) AMPK␣2 and dominant-negative (DN) AMPK␣2 in HeLa cells by infecting them with adenovirus carrying the expression vector for CA AMPK or DN AMPK. The levels of transiently expressed V5-mPer2 were significantly reduced by infection with CA AMPK adenovirus but were increased by infection with DN AMPK adenovirus (Fig. 2B). To further establish that activation of AMPK reduces mPer2 levels, we examined AMPK-dependent reduction in mPer2 level using murine embryo fibroblasts (MEFs) developed from AMPK␣1/␣2 double KO and WT embryos (23). As shown in Fig. 2C, activation of AMPK with metformin reduced mPer2 levels in WT MEFs but not in AMPK␣1/␣2 double KO MEFs. Activation of AMPK with metformin also reduced mPer1 levels in WT MEFs but not in AMPK␣1/␣2 double KO MEF (data not shown). To demonstrate that AMPK reduced mPer2 levels through proteasome-dependent degradation, we repeated the experiment shown in Fig. 2B in the presence of the proteasome inhibitor MG132 (Fig. 2D). Indeed, MG132 blocked the reduction of mPer2 level induced by CA AMPK. Consistent with this, AICARinduced reduction in mPer2 level was also blocked with MG132 (Fig.  2E). Taken together, these results suggest that activation of AMPK induces CKI⑀-mediated degradation of mPer2 by proteasome.
We investigated whether AMPK stimulated CKI⑀-mediated degradation of mPer2 by directly phosphorylating CKI⑀. To test this possibility, we performed in vitro kinase reactions using fragments F1 (aa residues 1-230) and F2 (aa 211-416) of CKI⑀ as AMPK substrates (Fig. 3A). AMPK did not phosphorylate F1 but did phosphorylate F2 very strongly in an AMP-dependent manner. In vitro kinase assays using fragments derived from F2 as substrates of AMPK indicated that the AMPK phosphorylation site resides in aa 377-404 near the C terminus of CKI⑀ (Fig. 3B). Examination of the amino acid sequence in this region revealed that Ser-389 is part of a motif that resembles the consensus AMPK phosphorylation site ((S/T)XXXL) (Fig. 3C). Ser-389 and the sequence surrounding it are highly conserved in all vertebrate CKI⑀, including that of chicken and frog, and also in CKI␦. To test whether Ser-389 could be the AMPK phosphorylation site, we used F2 containing a Ser-389 3 Ala mutation (S389A) as the AMPK substrate in a kinase reaction. The S389A mutation almost completely abolished AMPK phosphorylation (Fig. 3B), suggesting that in vitro, Ser-389 is the primary AMPK phosphorylation site in CKI⑀. To determine whether Ser-389 is phosphorylated in vivo, we transiently expressed F2 and F2 (S389A) in 293T cells and incubated them with [ 32 P]orthophosphate in phosphate-free medium. Because phosphate-free medium activated AMPK by decreasing ATP (Fig. 3D, top), we did not treat cells with an AMPK activator. As shown in Fig. 3D (bottom), the S389A mutation dramatically decreased 32 P labeling of CKI⑀ in cells, indicating that Ser-389 is phosphorylated in vivo.
To determine the role of Ser-389 on CKI⑀ activity, we immu- noprecipitated CKI⑀ containing the S389A mutation and performed a kinase reaction using mPer2 as substrate (Fig. 3E).
Although the basal activity of CKI⑀ (S389A) was similar to that of CKI⑀, the activity of CKI⑀ (S389A) did not change with AICAR treatment, whereas the activity of CKI⑀ consistently increased by about 3-fold within 2 h of AICAR treatment (p Ͻ 0.05). If activation of AMPK stimulates mPer2 degradation by phosphorylating Ser-389 of CKI⑀, mPer2 degradation should be blocked in S389A mutant. Indeed, AICAR-induced degradation of mPer2 did not occur when CKI⑀ (S389A) was overexpressed (Fig. 3F). The activity of the endogenous CKI⑀ immunoprecipitated from Rat-1 fibroblasts (Fig. 1) was also increased after metformin treatment (Fig. 3G), and mPer2 in Rat-1 fibroblasts was degraded after metformin treatment (Fig. 3H).
Since the oscillation in mPer mRNA expression results from mPer proteins suppressing their transcription through negative feedback, degradation of mPer2 during the period when mPer2 transcription is increasing should cause the mPer2 mRNA level to rise faster and peak earlier. To test this possibility, we injected WT mice with metformin, which has a short half-life (2.7 h) in mice (24), at 8 a.m. and measured mRNA levels of core clock genes mPer1, mPer2, Clock, and Bmal1 at 8 a.m. (no injection), 1 p.m., 6 p.m., 10 p.m., 2 a.m., and 7 a.m. in the heart, skeletal muscle, and fat. Consistent with our earlier results, metformin injection induced AMPK activity and led to mPer2 degradation (Fig. 4A). Since metformin injection causes transient hypoglycemia, we also injected insulin to another group to evaluate the effect of hypoglycemia alone. Metformin and insulin both decreased serum glucose by ϳ40% after 1 h (data not shown). The peak mPer1 and mPer2 mRNA levels in heart occurred at 6 p.m. in saline-and insulin-injected mice, but it occurred at 1 p.m. in metformin-injected mice (Fig. 4B). The expression patterns of Bmal1 and Clock mRNA were also shifted forward in the tissues of metformin-injected mice (Fig.  4B). Phase advance was also induced in skeletal muscle (Fig. 4C) and fat (data not shown). To determine whether AMPK␣2 was required for the metformin-induced phase advance, we performed these experiments in AMPK␣2 KO mice. As shown in Fig. 4D, the levels of mPer2 protein in AMPK␣2 KO mice were higher when compared with WT mice but did not change with metformin injection. Moreover, metformin injection did not significantly alter mPer1 and mPer2 mRNA expression patterns in AMPK␣2 KO mice as it did in WT mice (p ϭ 0.2-0.9 between saline-and metformin-injected samples) (Fig. 4E). These results indicate that metformin-induced mPer2 degradation and phase advance were mediated primarily by AMPK␣2.
In summary, we provide evidence that activation of AMPK can modulate the circadian rhythm by targeting the CKI⑀-mPer pathway. From what is known about metformin or AMPK, their linkage to the CKI⑀-mPer pathway and circadian rhythm was unexpected. However, previous work has shown that altering the eating behavior affects the circadian rhythm. For example, leptin-deficient mice, which overeat and develop morbid obesity, have altered sleep regulation and circadian rhythmicity (25). Moreover, locomotor activity and the circadian clock in the peripheral tissues can be entrained by feeding (26). It would be interesting to test whether AMPK functions as the feeding sensor for this type of entrainment.
Although metformin caused a dramatic shift in the circadian phase in the peripheral tissues, we have not been able to demonstrate that the circadian rhythm of behavior can be shifted by FIGURE 4. Activation of AMPK with metformin in WT mice shifts the expression pattern of clock genes. A, stimulation of AMPK activity and of mPer2 degradation in the WT heart was detected at 1 p.m. after saline (Sal) or metformin (Met) was injected at 8 a.m. (left panel). Quantification of the decline in mPer2 protein level after saline or metformin injection in arbitrary units (right panel). n ϭ 4. Results are mean Ϯ S.E.; **, p Ͻ 0.01. B, the effect of metformin injection on clock gene expression pattern. Saline, insulin, or metformin was injected at 8 a.m., and the mRNA levels of mPer1, mPer2, Bmal1, and Clock were measured by real-time PCR at 8 a.m. (no injection), 1 p.m., 6 p.m., 10 p.m., 2 a.m., and 7 a.m. The relative levels of mRNA are presented in arbitrary units. Only mPer1 and mPer2 mRNA levels in the heart (1 p.m. and 6 p.m.) were measured in mice injected with insulin. n ϭ 4 -5 for each time point for each type of injection. Results are mean Ϯ S.E.; *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001 between metformin-and saline-injected samples. C, the expression pattern mPer2 and mPer1 mRNA in skeletal muscle in mice from panel B. D, the experiment shown in panel A was repeated in AMPK␣2 KO mice. Quantification of mPer2 protein levels from WT and AMPK␣2 mice is shown (in arbitrary units). *, p Ͻ 0.05 between WT and AMPK␣2 KO mice injected with saline. n ϭ 3. E, the effect of metformin injection on mPer1 and mPer2 mRNA expression pattern in AMPK␣2 KO mice (heart). The differences in the levels of mPer1 and mPer2 mRNA between saline and metformin-injected samples at 8 a.m., 1 p.m., 6 p.m., and 10 p.m. were not statistically significant (p ϭ 0.2-0.9) (n ϭ 3 for each time point). metformin because metformin does not cross the blood brain barrier. However, our work provides a proof of principle that chemical manipulation of the circadian phase, at least in the peripheral tissues, may be possible.