Characterization of AMP-activated protein kinase beta and gamma subunits. Assembly of the heterotrimeric complex in vitro.

There is growing evidence that mammalian AMP-activated protein kinase (AMPK) plays a role in protecting cells from stresses that cause ATP depletion by switching off ATP-consuming biosynthetic pathways. The active form of AMPK from rat liver exists as a heterotrimeric complex and we have previously shown that the catalytic subunit is structurally and functionally related to the SNF1 protein kinase from Saccharomyces cerevisiae. Here we describe the isolation and characterization of the two other polypeptides, termed AMPKbeta and AMPKgamma, that together with the catalytic subunit (AMPKalpha) form the active kinase complex in mammalian liver. Sequence analysis of cDNA clones encoding these subunits reveals that they are related to yeast proteins that interact with SNF1, providing further evidence that the regulation and function of AMPK and SNF1 have been conserved throughout evolution. The amino acid sequence of the beta subunit is most closely related to SIP2 (35% identity), while the amino acid sequence of the gamma subunit is 35% identical with SNF4. We show that both AMPKbeta and AMPKgamma mRNA and protein are expressed widely in rat tissues. We show that AMPKbeta interacts with both AMPKalpha and AMPKgamma in vitro, whereas AMPKalpha does not interact with AMPKgamma under the same conditions. These results suggest that AMPKbeta mediates the association of the heterotrimeric AMPK complex in vitro, and will facilitate future studies aimed at investigating the regulation of AMPK in vivo.

There is growing evidence that mammalian AMP-activated protein kinase (AMPK) plays a role in protecting cells from stresses that cause ATP depletion by switching off ATP-consuming biosynthetic pathways. The active form of AMPK from rat liver exists as a heterotrimeric complex and we have previously shown that the catalytic subunit is structurally and functionally related to the SNF1 protein kinase from Saccharomyces cerevisiae. Here we describe the isolation and characterization of the two other polypeptides, termed AMPK␤ and AMPK␥, that together with the catalytic subunit (AMPK␣) form the active kinase complex in mammalian liver. Sequence analysis of cDNA clones encoding these subunits reveals that they are related to yeast proteins that interact with SNF1, providing further evidence that the regulation and function of AMPK and SNF1 have been conserved throughout evolution. The amino acid sequence of the ␤ subunit is most closely related to SIP2 (35% identity), while the amino acid sequence of the ␥ subunit is 35% identical with SNF4. We show that both AMPK␤ and AMPK␥ mRNA and protein are expressed widely in rat tissues. We show that AMPK␤ interacts with both AMPK␣ and AMPK␥ in vitro, whereas AMPK␣ does not interact with AMPK␥ under the same conditions. These results suggest that AMPK␤ mediates the association of the heterotrimeric AMPK complex in vitro, and will facilitate future studies aimed at investigating the regulation of AMPK in vivo.
A number of recent studies have led to the proposal that in mammals an AMP-activated protein kinase (AMPK) 1 plays a major role in the response to metabolic stress Hardie, 1994;Hardie et al., 1994). AMPK was first identified through its role in the phosphorylation and inactivation of a number of enzymes involved in lipid metabolism (Carling et al., 1987;Hardie et al., 1989;Hardie, 1992), and subsequently was shown to phosphorylate enzymes in other metabolic pathways ). AMPK has been purified from a number of species, including human, rat, and pig, and in each case is activated allosterically by micromolar concentrations of AMP Mitchelhill et al., 1994;Sullivan et al., 1994). The kinase is itself regulated by reversible phosphorylation, being phosphorylated and activated by a distinct AMPK kinase (AMPKK), thereby forming a protein kinase cascade (Carling et al., 1987;Weekes et al., 1994). The phosphorylation and activation of AMPK is markedly stimulated by AMP (Moore et al., 1991;Hawley et al., 1995), making AMPK extremely sensitive to changes in the intracellular concentration of AMP. These findings have led to the proposal that one of the primary roles of AMPK is to conserve ATP during periods of excessive ATP utilization, when AMP levels are elevated Hardie et al., 1994).
We recently reported that the deduced amino acid sequence of the catalytic subunit of rat liver AMPK is remarkably similar to the sequence of the yeast protein kinase SNF1 . In a further study we went on to show that SNF1 is functionally related to mammalian AMPK . In vitro, SNF1 phosphorylates a specific peptide substrate for AMPK, and there is good evidence that SNF1 phosphorylates and inactivates acetyl-CoA carboxylase in vivo. Furthermore, like AMPK, SNF1 is inactivated by protein phosphatases and can be reactivated by a partially purified preparation of mammalian AMPKK, suggesting functional conservation of the upstream kinases . The SNF1 protein kinase from Saccharomyces cerevisiae is required for the expression of glucose repressed genes in response to glucose starvation (Celenza and Carlson, 1986;Estruch et al., 1992;Gancedo, 1992), e.g. the SUC2 gene, which encodes invertase (Carlson and Botstein, 1982). snf1 mutants are unable to utilize a wide range of non-glucose sugars (Carlson et al., 1981;Estruch et al., 1992). In addition, snf1 mutants have been shown to be defective in other aspects of cell growth, e.g. glycogen synthesis and sensitivity to heat stress (Thompson-Jaeger et al., 1991). SNF1 is physically associated with a 36-kDa polypeptide, termed SNF4 , which is itself required for expression of many glucose-repressible genes. SNF4 is thought to function as an activator of SNF1 , although the mechanism by which SNF4 activates SNF1 is not known.
A number of yeast proteins, which interact with SNF1 in vivo, termed SNF1 interacting proteins or SIPs, have been identified using the two-hybrid system . Two of these proteins, SIP1 and SIP2, share significant amino acid sequence identity, particularly at their C termini (Yang et al., * 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. The nucleotide sequence ( ʈ To whom correspondence should be addressed.  The abbreviations used are: AMPK, AMP-activated protein kinase; AMPKK, AMPK kinase; PAGE, polyacrylamide gel electrophoresis; RACE, rapid amplification of cDNA ends; PCR, polymerase chain reaction; kb, kilobase pair(s); bp, base pair(s); X-Gal, 5-bromo-4-chloro-3-indoyl ␤-D-galactoside; Tricine, N-[2-hydroxy-1,1-bis(hydroxymethyl)ethyl]glycine. 1992). Furthermore, the amino acid sequence of SIP2 is 52% identical to GAL83 (Erickson and Johnston, 1993;Yang et al., 1994). GAL83 is involved in the glucose repression of GAL genes, and genetic evidence suggests that GAL83 is involved in the SNF1 pathway (Matsumoto et al., 1981;Erickson and Johnston, 1993). SIP1, SIP2, and GAL83 have been shown to co-immunoprecipitate with SNF1, and all three proteins are phosphorylated in an immune complex SNF1 kinase assay . In the same study it was shown that the C-terminal 80 amino acids of SIP2, termed the ASC domain, were sufficient to mediate interaction with SNF1. The functions of SIP1, SIP2, and GAL83 remain unknown, although it has been proposed that they may act as modulators of SNF1, targeting the kinase to specific intracellular locations and/or substrates .
Recently, AMPK has been purified to apparent homogeneity from both rat and pig liver Davies et al., 1994). Two other polypeptides co-purified with the catalytic subunit (molecular mass 63 kDa), and biochemical analysis of the purified kinase complex indicated that AMPK isolated from rat liver exists as a heterotrimer (Davies et al., 1994). We subsequently reported that the catalytic subunit of AMPK isolated from rat skeletal muscle did not appear to be associated with any other polypeptides and that this observation might account for the low activity of AMPK detectable in skeletal muscle (Verhoeven et al., 1995). In this paper we report the isolation and cDNA cloning of two AMPK subunits from rat liver which we refer to as AMPK␤ and AMPK␥ (the catalytic subunit is designated AMPK␣) following the terminology of Kemp and colleagues . The ␤ subunit is most closely related to SIP2 and contains a region at its C terminus, which is 50% identical with the ASC domain of SIP1/SIP2/GAL83 . The ␥ subunit has a high degree of amino acid sequence identity with SNF4, and this conservation of sequence suggests that, like SNF4, it is necessary for the catalytic activity of AMPK. We show here that AMPK␤ interacts with both AMPK␣ and AMPK␥ and that this mediates the assembly of the ternary complex in vitro. The similarity between the mammalian AMPK complex and the SNF1 complex from yeast emphasizes the likelihood that the role of these kinases have been highly conserved throughout evolution.

MATERIALS AND METHODS
Purification and Amino Acid Sequencing of AMPK Subunits-AMPK was partially purified from rat liver up to and including the DEAE-Sepharose ion-exchange step, as described previously . AMPK was further purified by immunoaffinity chromatography using affinity-purified antibodies raised against a synthetic peptide based on the deduced sequence of AMPK␣ . Approximately 200 mg of partially purified AMPK was incubated overnight at 4°C with 5 mg of affinity-purified antibody that had been cross-linked to protein A-Sepharose (Harlow and Lane, 1988). Following extensive washing of the resin with 50 mM Tris/HCl, pH 7.5, 50 mM NaF, 5 mM sodium pyrophosphate, 1 mM EDTA, 1 mM dithiothreitol, 10% (v/v) glycerol (buffer A), protein was eluted with 5 bed volumes of 0.1 M glycine, pH 2.5. The eluate was immediately neutralized by the addition of 0.1 volumes of 1 M Tris/HCl, pH 8, and concentrated in a Centricon-30 microconcentrator (Amicon). Proteins were resolved by SDS-PAGE on a 10% polyacrylamide gel and visualized by staining with Coomassie Blue. The polypeptides migrating at apparent molecular masses of 38 kDa (AMPK␤) and 36 kDa (AMPK␥) were excised from the gel and the proteins cleaved in the gel slice by overnight incubation with CNBr in 90% formic acid at room temperature. The supernatant was removed and dried in a Speed-Vac. The residue was washed twice with water, dried, and resuspended in SDS-gel loading buffer. Peptides were resolved by SDS-PAGE using a Tricine buffer system (Schagger and von Jagow, 1987) and transferred onto a Problot membrane (Applied Biosystems) for sequencing using an Applied Biosystems model 475 sequencer.
Autophosphorylated AMPK was prepared by incubating the immune complex with 0.2 mM [␥-32 P]ATP and 0.2 mM AMP at 30°C for 30 min. Unincorporated ATP was removed by extensive washing with buffer A. Protein was eluted from the resin as above. Following SDS-PAGE the gel was dried and subjected to autoradiography at Ϫ70°C. Amplification of cDNAs Encoding the ␤ and ␥ Subunits-Degenerate oligonucleotide primers based on the peptide sequences obtained from either AMPK␤ or AMPK␥ (shown in Table I) were synthesized as follows: ␤ forward primer 1, CCNGARAARGARGARTT; forward primer 2, AARGARGARTTYYTNGC; reverse primer 1, ACRTAYTTYTTYT-GRTA; reverse primer 2, TGRTANC(GT)NACNGTNGC (primers corresponding to the reverse and complemented sequence of these oligonucleotides were also synthesized but did not yield a distinct product after amplification, indicating that peptide 38CB3 is N-terminal to peptide 38CB1); ␥ forward primer 1, GAYTTYAT(ACT)AAYAT(ACT)YT; forward primer 2, YTNCAY(AC)GNTAYTAYAA; reverse primer 1, TGYT-GNACRAA(AGT)ATNCC; reverse primer 2, CCNARNGCNACRT-ANAC. For both subunits, rat liver cDNA (Clontech) was amplified using forward primer 1 and reverse primer 1 for 30 cycles of 94°C, 1 min; 50°C, 1 min; and 72°C, 1 min. In each case, an aliquot (0.1 l) of the products of the reaction were used for a second round of amplification using forward primer 2 and reverse primer 2 (same cycles as before). The products from the reaction were purified by agarose gel electrophoresis, cloned into a pGEM-T vector (Promega), and sequenced to confirm their identity.
Isolation of cDNA Clones Encoding the ␤ and ␥ Subunits-Standard molecular biology techniques were used (Sambrook et al., 1989). The products isolated from amplification of rat liver cDNA were used separately to screen a rat liver cDNA library ( UniZAP, Stratagene). Hybridization conditions were: 50 ng of probe (ϳ10 9 cpm/g; 10 6 cpm/ ml) in 5 ϫ SSPE, 100 g/ml sonicated and denatured salmon sperm DNA, 2 ϫ Denhardt's, 0.1% SDS at 65°C for 12 h. Filters were washed with 2 ϫ SSC, 0.1% SDS at room temperature for 1 h, followed by 2 ϫ SSC, 0.1% SDS at 65°C for 30 min and autoradiographed at Ϫ70°C for 18 h, with intensifying screens. Twelve positive clones encoding AMPK␤ were isolated from approximately 4 million plaques and one positive clone for AMPK␥ was isolated from approximately 1 million plaques. Plasmids were recovered by in vivo excision from the phage (Short et al., 1988) and the inserts sequenced manually by dideoxy chain termination (Sanger et al., 1977) using vector and cDNA specific primers.
In order to obtain the 5Ј end of the cDNA encoding the ␥ subunit, antisense oligonucleotide primers were synthesized based on ␥ cDNA sequence and used to perform 5Ј RACE-PCR (Frohman et al., 1988). (AMPK␥: RACE primer 1, TGGCGTAGGTGCCAATCTG; RACE primer 2, ATCTGTAGCTCTTCCAGAG). Rat liver RACE-ready cDNA (Clontech) was used as a template for amplification using primer 1 and the anchor primer for 30 cycles of 94°C, 1 min; 58°C, 1 min; 72°C, 1 min. A second round of amplification on an aliquot (0.1 l) of the reaction products was performed using primer 2 and the anchor primer under the same conditions as before. Products from the second round of amplification were isolated by agarose gel electrophoresis, cloned into pGEM-T vector, and sequenced. In order to construct a cDNA encoding the entire amino acid sequence of the ␥ subunit oligonucleotide primers spanning the initiating methionine (GCCAAGGTCGACGGCCGGGT-GCTAGCAATG) and downstream of the stop codon (GGCCACTAGTC-GACTCCGTTCTCTCAGG) and containing a SalI restriction site (underlined) were synthesized and used to amplify rat liver cDNA. The product (1.1 kb) was cloned into pGEM-T vector to yield pGEM-␥. The inserts from each of three independent clones were sequenced to confirm their authenticity.
Northern Analysis-A rat multiple tissue Northern (Clontech) was probed with either a random primed (Feinberg and Vogelstein, 1983) 1.9-kb fragment encoding the ␤ subunit or a random primed 1.1-kb cDNA fragment encoding the ␥ subunit according to the manufacturer's instructions. Following hybridization the blot was washed with 2 ϫ SSC, 0.5% SDS at room temperature for 1 h, followed by 0.2 ϫ SSC, 0.5% SDS at 65°C for 2 ϫ 20 min and autoradiographed at Ϫ70°C for 2-5 days.
Antibody Production-The entire coding sequence of AMPK␥ (330 residues) was expressed in Escherichia coli as a fusion protein with glutathione S-transferase. A polypeptide containing the C-terminal 217 residues of the ␤ subunit was expressed in E. coli as a fusion protein with glutathione S-transferase. The fusion proteins were purified on a glutathione-agarose column (Pharmacia) and used to immunize male New Zealand White rabbits. Following three rounds of immunization, antiserum was collected and used for Western blot analysis and immunoprecipitations. Antiserum against the catalytic subunit of AMPK was obtained as described previously .
Western Blotting of Tissue Lysates-Female Wistar rats (250 -300 g body weight) were killed by stunning and cervical dislocation, tissues removed, and immediately frozen in liquid nitrogen. Approximately 0.5 g of frozen tissue was ground to a fine powder using a pestle and mortar and homogenized in 10 ml of buffer (50 mM Tris/HCl, pH 7.5, 50 mM NaF, 5 mM sodium pyrophosphate, 1 mM EDTA, 1 mM dithiothreitol, 0.25 M sucrose, 0.1 mM phenylmethylsulfonyl fluoride) using a Polytron homogenizer. After homogenization, SDS was added to a final concentration of 0.5% and the homogenate was boiled for 15 min. The homogenate was centrifuged at 14,000 ϫ g for 15 min and the supernatant removed. The protein concentration of the supernatant was determined using the Lowry assay. 100 g of protein for each tissue was resolved by SDS-PAGE and transferred to a polyvinylidene membrane. The membrane was blocked by incubation in 10 mM Tris/HCl, pH 7.5, 1 M NaCl, 0.5% Tween, 5% nonfat milk powder (w/v) for 1 h at room temperature. Primary antibody was added to this buffer and the blot incubated for another 2 h. The blot was washed extensively in 10 mM Tris/HCl, pH 7.5, 1 M NaCl, 0.5% Tween and then incubated with an anti-rabbit antibody conjugated to horseradish peroxidase. Antibodies were detected using enhanced chemiluminesence (Boehringer Mannheim).
Interactions Using the Two-hybrid System-Vectors expressing the GAL4 DNA binding domain (pGBT9) and the GAL4 activation domain (pGAD) from the ADH1 promoter (Clontech) were used to construct fusion proteins with each of the AMPK subunits. Plasmids containing the entire coding sequence of the catalytic subunit were constructed by insertion of a 1.9-kb EcoRI fragment of p63 cDNA  into the EcoRI site of pGBT9 or pGAD, producing G BD -␣ or G AD -␣. Plasmids containing the entire coding sequence of the ␥ subunit were made by insertion of the 1.1-kb SalI fragment from pGEM-␥ into the SalI site in pGBT9 or pGAD to produce G BD -␥ and G AD -␥. In order to construct vectors expressing the ␤ subunit, the following oligonucleotide, GACTGCGAATTCCGGTCGACGGTGAATGAGAAAGCC, was used with the T7 promoter primer to amplify AMPK␤ cDNA from a clone contained in pBluescript. The product was digested with EcoRI (underlined sequence in the oligonucleotide) and XhoI (contained within the polylinker of Bluescript) and ligated into either pGBT9 or pGAD that had been digested with EcoRI and SalI. This creates an in-frame fusion between either the GAL4 DNA binding domain or GAL4 activation domain and the C-terminal 201 amino acids of the ␤ subunit. Vectors expressing fusions of the GAL4 activation domain with SNF1, SNF4, SIP1, and SIP2  were generously provided by Dr. Marian Carlson, Columbia University.
Yeast (strain SFY526 harboring a GAL1-lacZ reporter gene) were transformed with various combinations of the vectors and grown on selective media. To test for interactions, several colonies from each transformation were patched onto selective plates and grown for 2 days at 30°C. Colonies were transferred to nitrocellulose filters and cells were permeabilized by freeze-thawing in liquid nitrogen. The filters were incubated in the presence of X-Gal at 30°C for 1-2 h in order to determine any blue coloration. For quantitative analysis, transformants were grown to mid-log phase in selective liquid culture and ␤-galactosidase activity was determined in permeabilized cells. In every case, activities were measured in at least two transformants and assays were performed in triplicate. Values are expressed in Miller units (Miller, 1972) with a standard error of less than 20% of the mean.
In Vitro Translation and Immunoprecipitation of AMPK Subunits-cDNA containing the entire coding sequence of AMPK␣ was inserted into pET-21a (Novagen), while cDNAs containing the entire coding sequence of AMPK␤, AMPK␥ and the C-terminal 204 amino acids of AMPK␤ were inserted into pET-14b. The full-length AMPK␤ construct does not have any additional protein sequence, whereas the other constructs contain additional protein sequence encoded by the vector (including a polyhistidine sequence). RNA transcripts were synthesized using T7 polymerase and translated in reticulocyte lysates using a coupled transcription/translation system (TNT system, Promega) in the presence of [ 35 S]methionine. Total labeled products of translation were analyzed by SDS-PAGE and fluorography. For immunoprecipitations, the volume of the lysate was adjusted to 1 ml with phosphate-buffered saline containing 1% Triton X-100 and incubated with preimmune serum and protein A-Sepharose for 2 h at 4°C. The precleared reticulocyte lysate was incubated with antibodies against the ␣, ␤, or ␥ subunits and protein A-Sepharose for 2 h at 4°C. The immune complex was collected by centrifugation and washed extensively with with phosphate-buffered saline containing 1% Triton X-100 and either 0.2 M NaCl (␣ antibody), 1 M NaCl (␤ antibody), or 0.5 M NaCl (␥ antibody) and analyzed by SDS-PAGE and fluorography.

RESULTS
Purification of AMPK Subunits from Rat Liver-Affinitypurified antibodies raised against the catalytic subunit of AMPK  were covalently cross-linked to protein A-Sepharose and used to purify AMPK from rat liver. Fig. 1 shows that in addition to the 63 kDa catalytic subunit two other polypeptides, with apparent molecular masses of 38 and 36 kDa, as judged by SDS-PAGE, were identified (a faint band corresponding to IgG heavy chains could also be detect-FIG. 1. Co-purification of two polypeptides with the catalytic subunit of rat liver AMPK. Partially purified AMPK from rat liver was purified using protein A-Sepharose cross-linked to affinity-purified antibodies raised against AMPK␣. Following extensive washing, the antibody-protein A-Sepharose resin was incubated with 0.2 mM [␥-32 P]ATP and 0.2 mM AMP at 30°C for 30 min. Proteins were eluted from the antibody-protein A-Sepharose resin, resolved by SDS-PAGE, and visualized by staining with Coomassie Blue (lane A). An autoradiograph of the same gel is shown in lane B. The migration of molecular mass standards is shown on the right of the figure.

TABLE I Amino acid sequence obtained from peptides derived from CNBr cleavage of AMPK␤ and AMPK␥
Positions where no amino acid residue could be assigned are marked X. Residues in parentheses indicate tentative amino acid assignments. Where two amino acids are shown in parentheses, both were alternate tentative assignments. The amino acid sequences of all the peptides were found in cDNA clones encoding AMPK␤ and AMPK␥ (see Fig. 2), except for the differences indicated below the peptide sequence.
. We refer to these polypeptides as the ␤ subunit (38 kDa) and ␥ subunit (36 kDa) of AMPK, with the catalytic subunit being designated the ␣ subunit. As can be seen from Fig. 1, both the ␣ and ␤ subunits undergo autophosphorylation, whereas there is no detectable incorporation of phosphate into the ␥ subunit. In order to characterize the ␤ and ␥ subunits further, the purified polypeptides were transferred to a Problot membrane and subjected to N-terminal amino acid sequence analysis. On two separate occasions, no sequence was derived from the N terminus of either AMPK␤ or AMPK␥. However, cleavage of the gel-purified ␤ and ␥ subunits with cyanogen bromide yielded several peptides from which amino acid sequence was obtained ( Table I). Analysis of the Swiss-Prot data base with the peptide sequences from the ␥ subunit revealed that they were most closely related to sequences within yeast SNF4. No significant identity with any sequences in the data base was found with the peptides from the ␤ subunit. Isolation of AMPK␤ cDNA-A partial cDNA sequence encoding AMPK␤ was obtained by PCR using rat liver cDNA and degenerate oligonucleotide primers corresponding to potential sequences encoding the peptide sequences 38CB1 and 38CB3 shown in Table I (see "Materials and Methods"). A product of approximately 600 bp was isolated and sequenced, revealing that it encoded peptide sequences corresponding to 38CB2 and 38CB4. The residue predicted to be immediately N-terminal to peptide 38CB4 was found to be an aspartic acid, rather than a methionine as expected following CNBr cleavage, indicating that peptide 38CB4 was produced by acid cleavage of an Asp-Pro bond. The PCR product was used to screen a rat liver cDNA library in order to isolate full-length clones. Twelve positive hybridizing clones for AMPK␤ were isolated from approximately 4 million plaques. Sequence analysis of the longest clone showed that the insert contained an in-frame methionine, which was preceded by a stop codon in the same frame, followed by an open reading frame of 269 amino acids that included all of the peptides obtained by amino acid sequencing of the purified protein from rat liver. The open reading frame was followed by approximately 1 kb of untranslated sequence, including a polyadenylation site consensus sequence and a poly(A) tail ( Fig. 2A). The predicted molecular mass of AMPK␤ from the deduced protein sequence is 30 kDa, which is significantly lower than the apparent mass observed by SDS-PAGE analysis (38 kDa). However, in vitro translation of RNA synthesized from AMPK␤ cDNA (beginning at the first methionine) produced a protein that exactly co-migrated on SDS-PAGE with the ␤ subunit immunoprecipitated from rat liver (Fig. 3). This result confirms that there is no additional coding sequence upstream of the first methionine in the cDNA we isolated. At present we do not know the reason for the anomalous migration of the ␤ subunit on SDS-PAGE.
Isolation of AMPK␥ cDNA-A partial cDNA sequence encoding AMPK␥ was obtained by amplifying rat liver cDNA using degenerate oligonucleotide primers corresponding to potential sequences encoding the peptide sequences 36CB1 and 36CB2 shown in Table I (see "Materials and Methods"). A single product of approximately 400 bp was isolated, and sequence analysis confirmed that it encoded amino acid sequence within the peptides derived from the ␥ subunit. This cDNA product was used to screen a rat liver cDNA library. One positive hybridizing clone was isolated from approximately 1 million plaques. Sequence analysis of the clone revealed that it contained a 900-bp insert with an open reading frame of 200 amino acids, which encoded the sequence of peptide 36CB2. However, this clone did not encode the N-terminal region of ␥ including peptide 36CB1. In order to isolate the 5Ј region of cDNA encoding the ␥ subunit, we performed 5Ј RACE (Frohman et al., 1988) with rat liver cDNA. A unique product was isolated, and sequence analysis showed that it contained an in-frame methionine residue that was followed downstream by sequence corresponding to peptide 36CB1. A composite of the overlapping nucleotide sequences, together with the deduced protein sequence, is shown in Fig. 2B. The deduced mass of the protein encoded by the cDNA is 37 kDa, which is in reasonable agreement with the apparent mass observed from SDS-PAGE analysis of the purified protein. The open reading frame is followed by approximately 300 bp, but there is no obvious polyadenylation signal or poly(A) tail.
The ␤ and ␥ Subunits of AMPK Share Significant Amino Acid Sequence Identity with Yeast Proteins That Interact with SNF1-A search of the SwissProt data base with the amino acid sequence of the ␤ subunit showed that it is most closely related to SIP2 from S. cerevisiae, sharing 35% sequence identity overall. The highest conservation of sequence occurs at the C termini of the polypeptides, where there is 50% identity over a stretch of 68 amino acids (Fig. 4A, sequence beginning PPILPP to end). A similar, but slightly lower, degree of sequence identity was found with two other yeast proteins, SIP1 and GAL83, which have been shown previously to be related to SIP2 (Erickson and Johnston, 1993). A search of the SwissProt data base with the deduced protein sequence of the ␥ subunit identified the yeast protein SNF4 as having the highest degree of identity. The amino acid sequences of AMPK␥ and SNF4 are 35% identical overall (Fig. 4B). No other significant similarities were identified with either of the subunits.
Tissue Distribution of AMPK␤ and AMPK␥-Poly(A)-rich RNA isolated from a number of rat tissues was probed with cDNA encoding the ␤ and ␥ subunits. Fig. 5 shows the results of the Northern analysis and compares the expression of the ␤ and ␥ mRNA with the expression of AMPK␣, which we have reported previously (Verhoeven et al., 1995). A strongly hybridizing band of approximately 2.4 kb was detected in all of the tissues tested following labeling with a cDNA probe specific for AMPK␤. A weakly hybridizing signal at approximately 4.5 kb, present in all tissues, could also be detected with the AMPK␤ probe. When a ␥ subunit-specific probe was used, a single hybridizing band at approximately 1.8 kb was detected in all tissues, although only a faint signal was detected in testis. A band at approximately 2.4 kb was also detected with mRNA isolated from brain. These results indicate that the messages for both subunits are expressed in a wide number of rat tissues. In contrast to the ␤ and ␥ subunits, the mRNA expression of the ␣ subunit shows marked differences in tissue distribution and is most highly ex-pressed in skeletal and cardiac muscle (see Verhoeven et al. (1995)).
Antibodies raised against fusion proteins of AMPK␤ or AMPK␥ with glutathione S-transferase were used to determine the expression of the polypeptides in various rat tissues. We also examined the expression of AMPK␣ (the catalytic subunit) using antibodies raised against ␣ specific peptides . Fig. 6 shows the expression of the polypeptides in a number of tissue lysates. All three polypeptides were detected in every tissue tested, although there appeared to be some variation in the relative amounts of the three subunits present in different tissues (for instance compare the expression of the ␣ and ␤ subunits in brain and skeletal muscle). As we have noted previously, there is a small but detectable shift in the mobility of the ␣ subunit between different tissues, which we believe may reflect differences in the phosphorylation state of the enzyme (Verhoeven et al., 1995).
Interaction of the ␣, ␤ and ␥ Subunits in the Two-hybrid System-We employed the yeast two-hybrid system (Fields and Song, 1989) in order to examine the interaction of the three AMPK subunits in more detail. The entire coding sequence of the ␣, ␥, or the C-terminal 201 amino acids of ␤ were expressed as fusion proteins with either the GAL4 DNA binding domain (G BD ) or the GAL4 transcriptional activation domain (G AD ). Table II shows the activity of ␤-galactosidase in yeast transformed with various combinations of the fusion proteins. Transformation of yeast with G BD -␣ and G AD -␤ or G BD -␤ and G AD -␥ resulted in significant ␤-galactosidase activity, and this was confirmed by the appearance of blue colonies in the presence of X-Gal (Table II). Similar results with these combinations were obtained when the G BD and G AD fusions were switched. The activity detected in transformants with G BD -␣ and G AD -␥ or G AD -␣ and G BD -␥ was very low in comparison with the other combinations, which could indicate that the interaction between AMPK␣ and AMPK␥ is weaker than the other interactions. Very low levels of ␤-galactosidase activity were detected when yeast were transformed with the same subunit expressed with both G BD and G AD , and these transformants remained white in the presence of X-Gal, indicating that the subunits do not form homodimers.
Since the two-hybrid system is carried out in yeast, it was important to determine any interactions between the mammalian subunits and their yeast counterparts. We therefore extended the study to determine interactions between AMPK subunits and SNF1, SNF4, SIP1, and SIP2. Table III shows the results of the various combinations of rat and yeast proteins in FIG. 5. Northern blot analysis of AMPK␣, ␤, and ␥ mRNA. Approximately 2 g of poly(A)-rich RNAs isolated from the indicated tissues were separated on a 1.2% agarose gel under denaturing conditions, transferred to a charged-modified nylon membrane, and probed separately with either a 1.9-kb fragment of AMPK␤ cDNA or a 1.1-kb fragment of AMPK␥ cDNA. In each case the blot was washed under stringent conditions (0.2 ϫ SSC, 0.5% SDS at 65°C) and exposed for either 2 days (AMPK␤) or 5 days (AMPK␥) at Ϫ70°C. For comparison a Northern blot of the ␣ subunit is shown (Verhoeven et al., 1995). The migration of RNA markers are indicated.
FIG. 6. Western blot analysis of AMPK␣, ␤, and ␥. Approximately 100 g of tissue lysate (14,000 ϫ g supernatant) from the indicated tissues were separated by SDS-PAGE and transferred to a polyvinylidene membrane. Separate blots were probed with polyclonal antibody to AMPK␣, AMPK␤, or AMPK␥. Primary antibody was detected using a goat anti-rabbit antibody conjugated to horseradish peroxidase and visualized by enhanced chemiluminescence. the two-hybrid system. We were unable to detect any interaction of AMPK␣ with any of the yeast proteins. However, AMPK␤ gave a signal with both SNF1 and SNF4 and AMPK␥ interacted with SNF1, SIP1, and SIP2. Furthermore, AMPK␥ interacted with all of the SIP1 and SIP2 fusions tested, including a fusion expressing the C-terminal 120 amino acids of SIP2 (G AD -SIP2 296 -415 ; Yang et al. (1994)).
Association of the AMPK Complex in Vitro-In order to confirm the results of the interactions determined by the twohybrid system, we looked at the association of the subunits in vitro. The ␣, ␤, and ␥ subunits were translated in rabbit reticulocyte lysates either alone or with one, or both, of the other subunits in the presence of [ 35 S]methionine. Translated products were immunoprecipitated with anti-␣, anti-␤, or anti-␥ antibodies bound to protein A-Sepharose, and the [ 35 S]methionine-labeled subunits in the immune complexes were resolved by SDS-PAGE and detected by fluorography. The products of the translations before immunoprecipitation (panel A) or following immunoprecipitation with AMPK subunit specific antibodies (panel B) are shown in Fig. 7. When ␣ or ␥ were cotranslated with ␤, they could be co-precipitated with anti-␤ antibodies. In contrast, however, the ␥ subunit was not detected in anti-␣ immunoprecipitates, nor was the ␣ subunit detected in anti-␥ immunoprecipitates, from lysates programmed with both the ␣ and ␥ subunits. All three subunits were immunoprecipitated from lysates programmed with ␣, ␤, and ␥ using anti-␣, anti-␤, or anti-␥ antibodies. These results indicate that when all three subunits are co-translated, a ternary complex between ␣, ␤, and ␥ is formed, rather than a mixture of ␣␤ and ␤␥ dimers. For clarity we have shown the results using a truncated form of AMPK␤ since full-length ␤ has a similar mobility to ␥ on SDS-gels. These results were consistent with those obtained using the full-length form of AMPK␤ (data not shown).
The results of the in vitro translation studies demonstrate that the ␣ and ␤ subunits and the ␤ and ␥ subunits interact with each other forming relatively stable complexes, whereas there is no evidence for a stable complex between the ␣ and ␥ subunits. If the ␣ and ␥ subunits do interact, then their association must be either transient or weak, or both, and does not survive immunoprecipitation. The translations and immunoprecipitations are carried out in buffers lacking protein phosphatase inhibitors and under conditions that would not be expected to cause activation of endogenous AMPKK, which may be present in the reticulocyte lysate. It is likely that AMPK␣ is in the dephosphorylated form following translation, suggesting that phosphorylation by AMPKK is not necessary for formation of the ternary complex. This may also explain why we have been unable to detect AMPK activity in any of the translations.

DISCUSSION
Full-length cDNA clones encoding AMPK␤ were isolated by conventional library screening, and a composite cDNA clone encoding the full-length sequence of AMPK␥ was constructed from overlapping clones isolated by a combination of library screening and 5Ј RACE. The deduced amino acid sequence of AMPK␤ predicts a protein with a molecular mass of 30 kDa. This is considerably lower than the apparent mass of the ␤ subunit isolated from rat liver, as judged by SDS-PAGE. In vitro translation of RNA synthesized from AMPK␤ cDNA produced a major product, which exactly co-migrated with rat liver AMPK␤ following SDS-PAGE. This finding, coupled with the fact that an in-frame stop codon is present upstream of the first methionine, confirms that the AMPK␤ cDNA reported here is full-length. The reason for the anomalous electrophoretic mobility of the ␤ subunit in denaturing gels is unclear, but if it is due to post-translational modification of the polypeptide the reticulocyte lysate system must be competent in carrying out the modification.
The finding that AMPK␤ and AMPK␥ share sequence identity with yeast proteins that interact with SNF1 strengthens FIG. 7. Association of AMPK subunits in vitro. A, [ 35 S]methionine-labeled AMPK subunits were generated in rabbit reticulocyte lysates programmed with the indicated RNAs and an aliquot of the products was analyzed by SDS-PAGE and fluorography. The AMPK␤ RNA used for these studies lacked sequence encoding the N terminus. AMPK␤, therefore, migrates with a lower apparent molecular mass compared to AMPK␥, allowing the subunits to be clearly resolved. The migration of molecular mass markers is shown on the left. B, interaction of the subunits was determined by co-immunoprecipitation from the lysates (2 ϫ volume used in A). Following incubation with preimmune serum and protein A-Sepharose, lysates were immunoprecipitated with subunit-specific antibodies as indicated. The immune complexes were washed extensively (as described under "Materials and Methods"), boiled with SDS-sample buffer, and resolved by SDS-PAGE. Labeled products were detected by fluorography.

TABLE II
Interaction of AMPK subunits using the two-hybrid system The color of transformants (yeast strain SFY526, Clontech) was determined by filter assay. ␤-Galactosidase activities are the average values from two transformants (assays performed in triplicate for each transformant) with standard errors of less than 20% of the mean. DNA-binding hybrid Activation hybrid Color ␤-Galactosidase activity TABLE III Interaction of AMPK subunits and their yeast counterparts using the two-hybrid system G AD -SIP1 201-863 , etc., has codon 201 of SIP1 fused to the GAL4 activation domain . Color of transformants was determined by filter assay. the proposal that the functions of the two kinases have been highly conserved throughout evolution . Although the function of SNF4 is not known, it is necessary for the protein kinase activity of SNF1 and it seems likely that AMPK␥ will have a similar role in the activity of AMPK. Despite the obvious similarity between the amino acid sequences of AMPK␥ and SNF4, we have not been able to complement snf4 mutants by expression of AMPK␥. 2 In a previous study, we reported that we were unable to complement snf1 mutants by expression of the catalytic subunit of AMPK, which shares 47% amino acid sequence identity with SNF1 Woods et al., 1994). Taken together these results indicate that, although the AMPK and SNF1 complexes are highly related, significant differences between the two complexes must exist. One notable difference that is already known is that, while the mammalian kinase is markedly activated by AMP , no measurable effect of AMP on SNF1 activity has been demonstrated Woods et al., 1994). Whether this difference alone is sufficient to explain the inability of the catalytic subunit to rescue snf1 mutants, and AMPK␥ to rescue snf4 mutants, is not yet clear. In order to address this question, a detailed comparison of the structures of the mammalian and yeast kinase complexes, e.g. from crystallographic studies, and the elucidation of the regulation of the kinases are required.
Western blot analysis of rat tissue lysates shows that AMPK␤ and AMPK␥ subunits are expressed in every tissue examined. Although the blots are not quantitative, there does appear to be some variation in the relative expression of the three subunits in different tissues (Fig. 6). AMPK purified from rat liver appears to exist entirely as a heterotrimeric complex of AMPK␣, ␤, and ␥ (see Fig. 1 and Davies et al. (1994)). Immunoprecipitation of ␣ from skeletal muscle, however, suggests that it exists predominantly as a monomer, with no kinase activity (Verhoeven et al., 1995), even though AMPK␤ and AMPK␥ are present in this tissue. These findings raise the interesting possibility that the association of the catalytic subunit with AMPK␤ and AMPK␥ is regulated and that the structure of the complex may vary between different tissues and/or different conditions. At present the mechanism that leads to the association of the subunits is not known, although our results suggest phosphorylation is not required. Dephosphorylation of the active AMPK complex to an inactive form in vitro does not result in dissociation of the subunits (data not shown), and it is interesting to note that in yeast the association of SIP1 or SIP2 with SNF1 does not require SNF1 protein kinase activity or the presence of SNF4 .
The results from the two-hybrid experiments indicate that AMPK␤ interacts with both AMPK␣ and AMPK␥, but that the interaction of AMPK␣ with AMPK␥ is very weak. We also tested the interaction of the mammalian subunits with their yeast counterparts in the two-hybrid system. In this case we found evidence for the interaction of AMPK␤ with both SNF1 and SNF4 and AMPK␥ with SNF1 and SIP1 and SIP2. However, there was no detectable interaction between AMPK␣ and any of the yeast proteins. These results do not rule out the possibility that the AMPK␥-SNF1 interaction is indirect and could be mediated by an AMPK␤ homologue in yeast, e.g. SIP1/SIP2/GAL83, which could act as a bridging protein in a ternary complex.
The results from the in vitro study show that AMPK␤ interacts with both AMPK␣ and AMPK␥ forming stable complexes. However, under the same conditions, we could not detect stable complexes between AMPK␣ and AMPK␥. These results suggest that the formation of the ternary complex between ␣, ␤, and ␥ is mediated by the ␤ subunit. We have not been able to detect AMPK activity in lysates programmed with all three subunits, or in immune complexes from these lysates. This may be due to the dephosphorylated form of the kinase and/or the lack of sufficient protein in the translation system to allow detection of kinase activity.
Two different models for the association of the three subunits can be predicted based on the results of this study. The first model is one in which the ␤ subunit links the ␣ and ␥ subunits (Fig. 8A). The second model involves a conformational change in either the ␣ or ␥ subunit, or both, upon binding of the ␤ subunit, which would then allow direct interaction between the ␣ and ␥ subunits (Fig. 8B). It is interesting to note the similarities between the AMPK complex and the heterotrimeric complex formed between CDK-activating kinase, cyclin H, and p36/MAT1 (RING finger protein) (Fisher et al., 1995;Devault et al., 1995). Although there is no significant amino acid sequence identity between the AMPK subunits and the CDK-activating kinase subunits, the regulation and association of the two kinase complexes bear some obvious resemblances.
Although the functions of AMPK␤ and AMPK␥ remain unclear, a possible insight can be gained by comparison with a proposed model for the SNF1 complex in yeast. SNF4 is necessary for SNF1 kinase activity in vitro  and may therefore fulfill a similar function to cyclins in the activation of cyclin dependent kinases (Jeffrey et al., 1995). AMPK␥ by analogy would play a similar role in the activation of AMPK. Biochemical evidence suggests that SNF1 forms a relatively stable complex with SNF4, and that this complex has protein kinase activity in vitro . The SNF1⅐SNF4 complex interacts with one of a number of related proteins, which include SIP1, SIP2, and GAL83 . It has been proposed that these proteins act as adaptors or targeting subunits, directing the kinase to specific intracellular substrates . In this model, the formation of different SNF1⅐SNF4⅐adaptor complexes would allow selective phosphorylation of the downstream targets of SNF1, if the adaptor proteins recognize different substrates. It is not clear whether the assembly of the SNF1⅐SNF4⅐adaptor complex 2 P. C. F. Cheung and D. Carling, unpublished results. Our results demonstrate that AMPK␤ interacts with both AMPK␣ and AMPK␥ and that this mediates the formation of a stable trimeric complex. AMPK␤ could act as a bridge linking the other two subunits (A), or alternatively AMPK␤ could cause a conformational change in AMPK␣, AMPK␥, or both (depicted by the different labeling of the subunits) allowing the ␣ and ␥ subunits to interact (B). AMPK␣ is phosphorylated by an upstream kinase (AMPKK), leading to activation of the complex (Weekes et al., 1994;Hawley et al., 1995). Our preliminary results indicate that AMPKK can only phosphorylate AMPK␣ when it is in the heterotrimeric form, implying that association of the complex is a prerequisite for phosphorylation. AMP activates the complex both allosterically and by promoting the phosphorylation of AMPK␣ by AMPKK, although it is not known whether AMP has any direct effect on the association of the complex. is regulated, or how the adaptor proteins act to promote phosphorylation of target substrates. However, there is clear evidence from other systems that one mechanism for regulating the phosphorylation of a protein is to regulate the distribution of the kinase, or phosphatase, acting on that protein via specific targeting subunits (Hubbard and Cohen, 1993;Coghlan et al., 1995). Could AMPK␤ act as a targeting subunit for the AMPK complex? It seems unlikely that the function of AMPK␤ is merely to bring together the ␣ and ␥ subunits, especially given the fact that in yeast there appears to be a family of proteins related to AMPK␤. It will be interesting to determine whether or not there is a family of proteins related to AMPK␤ in mammalian cells, and whether these proteins act as adaptors for AMPK.
Finally, it is interesting to note that, although the interaction of SNF1 and SNF4 is often used as a model for the twohybrid system, our results imply that this interaction could in fact be indirect. Given the similarities between the mammalian AMPK and yeast SNF1 complexes it is possible that the interaction between SNF1 and SNF4 is mediated by a member of the SIP1/SIP2/GAL83 family of proteins. In addition to SNF4, a number of other polypeptides were found to co-purify with SNF1 . Although SIP1 was not identified, it is possible that some of these co-purifying polypeptides are members of the SIP1/SIP2/GAL83 family, which could mediate the association of SNF1 and SNF4 in a ternary complex, analogous to AMPK␤ in the mammalian complex.