Characterization of the AMP-activated protein kinase kinase from rat liver and identification of threonine 172 as the major site at which it phosphorylates AMP-activated protein kinase.

We have developed a sensitive assay for the AMP-activated protein kinase kinase, the upstream component in the AMP-activated protein kinase cascade. Phosphorylation and activation of the downstream kinase by the upstream kinase absolutely requires AMP and is antagonized by high (millimolar) concentrations of ATP. We have purified the upstream kinase >1000-fold from rat liver; a variety of evidence indicates that the catalytic subunit may be a polypeptide of 58 kDa. The physical properties of the downstream and upstream kinases, e.g. catalytic subunit masses (63 versus 58 kDa) and native molecular masses (190 versus 195 kDa), are very similar. However, unlike the downstream kinase, the upstream kinase is not inactivated by protein phosphatases. The upstream kinase phosphorylates the downstream kinase at a single major site on the alpha subunit, i.e. threonine 172, which lies in the "activation segment" between the DFG and APE motifs. This site aligns with activating phosphorylation sites on many other protein kinases, including Thr177 on calmodulin-dependent protein kinase I. As well as suggesting a mechanism of activation of AMP-activated protein kinase, this finding is consistent with our recent report that the AMP-activated protein kinase kinase can slowly phosphorylate and activate calmodulin-dependent protein kinase I, at least in vitro (Hawley, S. A., Selbert, M. A., Goldstein, E. G., Edelman, A. M., Carling, D., and Hardie, D. G. (1995) J. Biol. Chem. 270, 27186-27191).

The AMP-activated protein kinase (AMPK) 1 is a protein kinase that phosphorylates multiple targets both in vitro and in vivo (1). These targets include several biosynthetic enzymes such as hydroxymethylglutaryl-CoA reductase (sterol/isoprenoid synthesis) (2, 3), acetyl-CoA carboxylase (fatty acid synthesis) (4 -6), and glycogen synthase (glycogen synthesis) (7). The kinase is activated by elevation of the AMP:ATP ratio during cellular stress, and a current hypothesis for its physiological role is that it preserves ATP by switching off biosynthesis whenever cellular ATP levels are compromised by adverse cellular environments (8).
Purified AMPK consists of three subunits of 63, 38, and 35 kDa (9), now referred to as ␣, ␤, and ␥, respectively (10). The catalytic (␣) subunit (11) contains an N-terminal kinase domain and is closely related to the product of the SNF1 gene from the yeast Saccharomyces cerevisiae (12,13). A recent report (14) has identified a novel isoform, termed ␣ 1 , which is widely distributed, while the form originally cloned (Refs. 12, 15, and 16; now ␣ 2 ) is highly expressed in heart, liver, and skeletal muscle. The functions of the ␤ and ␥ subunits remain unknown, but both may also exist as multiple isoforms (14) and are related to the SIP1/SIP2 and SNF4 gene products, respectively, from S. cerevisiae (10). Snf4p has been shown to form a stable complex with Snf1p (17), while both Snf4p (18) and either Sip1p or Sip2p (19,20) form complexes with Snf1p in vivo as assessed by the two-hybrid assay. Thus all three subunits of mammalian AMPK have counterparts in S. cerevisiae. Genetic evidence shows that functional SNF1 (and SNF4) genes are required for the response to glucose starvation (17,21), which in yeast could be regarded as an acute stress situation. In particular, Snf1p is required for derepression of genes that are normally repressed in the presence of glucose (21), while it is also required for inactivation of acetyl-CoA carboxylase, which occurs in response to starvation (22).
At present little is known about regulation of Snf1p, but the regulation of mammalian AMPK is better characterized. 5Ј-AMP not only allosterically activates the kinase but is also required for phosphorylation and activation by an upstream protein kinase termed AMP-activated protein kinase kinase (AMPKK) (23)(24)(25). Phosphorylation by AMPKK results in at least 50-fold activation, which can be further amplified by the 5-fold allosteric effect of AMP (25). We have also recently reported that micromolar concentrations of AMP dramatically inhibit dephosphorylation and inactivation of AMPK by protein phosphatase-2C (26).
Initially it was unclear whether the effect of AMP on phosphorylation of AMPK by AMPKK was due to binding of the nucleotide to the substrate (AMPK), to the enzyme (AMPKK), or to both. Although we reported that the former mechanism could explain the effect, at least in part (24), more recently we have presented evidence that AMP also activates AMPKK directly, i.e. that the latter is itself an AMP-activated protein kinase (27). An answer to this question was possible because we found that AMPKK would also phosphorylate and activate (at least in vitro) calmodulin-dependent protein kinase I, an enzyme that was not itself sensitive to AMP.
These findings show that the upstream protein kinase AMPKK plays an active role in the overall activation of the cascade by AMP rather than merely being a passive, constitutively active partner. This makes the further characterization of AMPKK an important goal. Here we report on the purification and some physical and regulatory properties of AMPKK, and we also identify the major site at which it phosphorylates and activates AMPK.
Enzyme Assays-PP2A was assayed using phosphorylase a (31). One unit of PP2A dephosphorylates 1 mol of substrate/min at 30°C. AMPK was assayed as described (34) except that, unless stated otherwise, the AMARA peptide (200 M) replaced SAMS. One unit of AMPK phosphorylates 1 mol of peptide/min at 30°C. AMPKK was assayed by its ability to reactivate dephosphorylated AMPK; a single preparation of AMPK was used throughout these studies. The preparation was stored frozen at Ϫ70°C in small aliquots at 10 units/ml in buffer C. Before use it was diluted to 2 units/ml in buffer C containing 50% glycerol; in this form it could be stored at Ϫ20°C for up to a month. AMPK (0.013 units/ml in buffer E) was inactivated by incubation with PP2A C (6 milliunits/ml final concentration) at 30°C for 10 min. PP2A was inhibited by the addition of okadaic acid to 100 nM final. Three volumes of dephosphorylated AMPK were added to 1 volume of ATP, MgCl 2 , and AMP in buffer E, and 3 volumes of this mixture was added to 1 volume of AMPKK in buffer E. Final concentrations were 1 mM ATP, 5 mM MgCl 2 , and 100 M AMP. The reaction was incubated at 30°C, and 5-l aliquots were removed at 0 and 25 min for the AMPK assay as described above. One unit of AMPKK was that amount which increased the activity of dephosphorylated AMPK by 1 unit/ml (calculated at the dilution of the stock AMPK) per min at 30°C.
Protein Analysis and Electrophoresis-Protein concentration was determined by the method of Bradford (35). SDS-PAGE was on 10% gels using the method of Laemmli (36) in a Bio-Rad Mini-Protean II gel apparatus unless stated otherwise.
Purification of AMPK-AMPK was purified using modifications of the previous method (11). All procedures were at 4°C unless stated otherwise. Livers from male Wistar rats were left at room temperature for 1-2 min to become hypoxic (this elevates AMP and activates AMPK (5)), chopped coarsely, and washed in buffer A. A 2.5-6% polyethylene glycol 6000 (PEG) fraction was prepared as described (11), resuspended in buffer B using a ground glass homogenizer, and applied to a DEAE-Sepharose (Fast Flow) column (6 ϫ 15 cm). The column was washed until A 280 Ͻ 0.05, and protein was eluted with buffer B plus 0.2 M NaCl. Active fractions were loaded directly onto blue-Sepharose (5 ϫ 7 cm) and washed extensively, and protein was eluted with buffer B plus 0.8 M NaCl. Active fractions were precipitated with ammonium sulfate, resuspended, and dialyzed as for the previous procedure (11) and then clarified by centrifugation (20,000 ϫ g, 10 min). The preparation was applied to a Q-Sepharose HiLoad 16/10 column at 3 ml/min, the column was washed until A 280 was Ͻ0.05, and protein was eluted with a linear gradient from 0 to 450 mM NaCl in buffer B. Active fractions were pooled, concentrated to Ͻ0.5 ml in Centricon-30 concentrators, and applied to a high performance Sephacryl S-200 HiLoad column in buffer C at 0.5 ml/min. Active fractions were pooled, concentrated to 10 units/ml in Centricon-30 concentrators, frozen in small aliquots in liquid N 2 , and stored at Ϫ70°C.
Purification of AMPKK-All procedures were at 4°C unless stated otherwise. The 6% PEG supernatant from the AMPK preparation above was made to 10% (w/v) PEG and centrifuged (30,000 ϫ g, 20 min). The pellet was resuspended in buffer B and applied to a DEAE-Sepharose (Fast Flow) column (6 ϫ 15 cm). The column was washed until A 280 was Ͻ0.05, and protein was eluted with buffer B plus 0.2 M NaCl. Active fractions were loaded directly onto Blue-Sepharose (5 ϫ 7 cm), and the flow-through was applied to a high performance Q-Sepharose HiLoad 16/10 column in buffer B plus 200 mM NaCl at 3 ml/min. The column was washed until A 280 was Ͻ0.05 and then AMPKK was eluted with a 96-ml linear gradient from 200 to 500 mM NaCl in buffer B. Active fractions were pooled, concentrated to Ͻ0.5 ml in Centricon-30 concentrators, diluted 3-fold in buffer C, and applied to a Mono Q (HR 5/5) column at 1 ml/min. The column was washed until A 280 was Ͻ0.05, and then AMPKK was eluted with gradients from 0 to 50 mM MgCl 2 (1 min) and 50 -100 mM MgCl 2 (24 min) in buffer C. Active fractions were pooled, concentrated to Ͻ0.5 ml in Centricon-30 concentrators, and then diluted 3-fold in buffer D. An ATP-␥-Sepharose (9) column (1.5 ml) was equilibrated in buffer D. AMPKK was applied, stirred into the matrix, and allowed to stand for 20 min. The column was washed at ϳ1 ml/min with 40 ml of buffer D and then 25 ml of buffer D plus 5 mM ATP. Active fractions were pooled, concentrated in Centricon-30 concentrators, frozen in small aliquots in liquid N 2 , and stored at Ϫ70°C. Aliquots of the Q-Sepharose fraction were stored in the same way and used in some experiments (Figs. 1 and 4 -10).
Determination of Physical Parameters of AMPKK-AMPKK (250 l, 200 units/ml, Q-Sepharose preparation) was subject to gel filtration on a Sephacryl S-200 column (60 ϫ 1.6 cm) in buffer B, both in the presence and absence of 0.2 M NaCl. The column was calibrated prior to use, using markers of known Stokes radii (Sigma gel filtration markers). For gradient centrifugation, AMPKK (amount and purity as for gel filtration) was layered with or without marker proteins (8.9 S sweet potato ␤-amylase and 4.4 S bovine serum albumin) onto a glycerol gradient (10 -20% (v/v), 14 ml) in buffer B. A third tube contained marker proteins alone. The tubes were centrifuged in a Beckman SW 28 rotor (100,000 ϫ g, 42 h, 4°C). Fractions (0.5 ml) were collected from the bottom of the tube using a Nyegaard gradient unloader and analyzed for protein (A 280 ) and AMPKK activity.
Labeling of AMPKK Using [␥- 32  by the addition of SDS-PAGE sample buffer. Autoradiography (ATP labeling) was performed on dried gels at Ϫ70°C using Hyperfilm-␤ max with Kodak X-Omatic intensifying screens. Phosphor imaging (FSBA labeling) was performed using a Molecular Dynamics PhosphorImager.
Phosphorylation and Immunoprecipitation of AMPK for Sequencing Studies-AMPK was purified to the gel filtration step as described above, except that NaF and sodium pyrophosphate were omitted from the gel filtration buffer. AMPK (14,000 units) was dephosphorylated using PP2A C (500 milliunits) for 15 min at 30°C in 8 ml of buffer C. The reaction was stopped by adding okadaic acid to 100 nM, and then incubation was continued for 8 min in the presence of AMP (100 M), MgCl 2 (5 mM), and [␥-32 P]ATP (200 M; 400 cpm/pmol) in the presence or absence of AMPKK (210 units). Reactivation was terminated by the addition of EDTA to 20 mM, and then AMPK was immunoprecipitated by five sequential additions of anti-AMPK-␤ subunit antibody. In each case, 200 l of a 50% (v/v) slurry of protein A-Sepharose in buffer B containing 1% (w/v) Triton X-100 was added, and the resulting mixture was incubated for 2 h at 4°C with constant shaking. The immunocomplex was collected by centrifugation and washed extensively with buffer B containing 1% (w/v) Triton X-100. The immunoprecipitates were combined, boiled with SDS-PAGE sample buffer, resolved by SDS-PAGE on 10% gels, and transferred electrophoretically to polyvinylidene difluoride membranes (37).
Tryptic Digestion of AMPK and Purification of 32 P-Peptides for Sequencing Studies-Peptides were digested on polyvinylidene difluoride membranes using trypsin (37). The membranes were stained with sulforhodamine (Eastman Kodak Co.) (38). Regions of the membrane containing the ␣ and ␤ subunits were excised, cut into ϳ4 ϫ 4-mm pieces, and incubated with sufficient trypsin solution (100 g/ml in 200 mM NH 4 HCO 3 containing 1% (w/v) n-octyl ␤-D-glucopyranoside) to fully wet the membranes. After 24 h at room temperature, 0.1 volume of 100 mM dithiothreitol in buffer A was added, and incubation continued for a further 24 h. The membrane pieces were extracted once with 0.6 -1.0 ml of buffer A, followed by transfer of the supernatant to clean polypropylene tubes. This first extract contained about 70% of the total counts; subsequent extraction of the membrane pieces with the same volume of buffer A and then the same volume of 50% formic acid:methanol resulted in extraction of a further 15 and 8% respectively. These extracts were dried on a rotary evaporator and resuspended in 200 -300 l of 0.1% (v/v) trifluoroacetic acid. In order to keep the octyl glucoside concentration below 2% (w/v), to 50-l aliquots of these samples 10 l of 100 mM dithiothreitol was added followed by 50 l of 0.1% (v/v) trifluoroacetic acid and then 10 l of 10% (v/v) trifluoroacetic acid. Aliquots were then subjected to reversed phase HPLC using a 2.1 ϫ 100-mm Aquapore OD300 C18 column (Applied Biosystems) at a flow rate of 250 l/min. Fractions were collected at 1-min intervals. The column was washed with 0.1% (v/v) trifluoroacetic acid in acetonitrile:water (3:97) for 5 min, followed by a gradient (5-90 min) from 3 to 60% of 0.1% (v/v) trifluoroacetic acid in acetonitrile:water (9:1). Some of the radioactive peaks from these initial runs were pooled, dried by rotary evaporation, and resuspended in 50 l of buffer F (10 mM ammonium acetate pH 6.5) followed by 10 l of 10 mM dithiothreitol and 60 l of 200 mM ammonium acetate, pH 6.5. Peptides were separated by reversed phase HPLC as before except using buffer F in place of 0.1% trifluoroacetic acid. Recovery of radioactivity from HPLC in either buffer was 50 -70%.
Amino Acid Sequencing-Fractions containing radiolabeled peptides were pooled and dried down onto a Sequelon arylamine membrane disc (Millipore Corp.). The peptides were coupled to the disc according to manufacturer's instructions, and the discs were washed with 1 ml of 50% (v/v) methanol followed by 1 ml of 0.1% (v/v) trifluoroacetic acid in 50% (v/v) acetonitrile. Coupling efficiencies, as determined by counting of radioactivity, were ϳ90%. The membrane discs were subjected to N-terminal sequencing using a modified ABI 475 gas phase sequencer, which either injected 100 out of 120 l of the PTH-derivative into the HPLC analyzer or delivered the whole fraction into a fraction collector. Solvent S3 was 70% (v/v) acetonitrile. The radioactivity in the PTHderivative fraction was determined by scintillation counting using a Wallac 1218 counter.

Establishment of an Assay for AMP-activated Protein Kinase
Kinase-Purification and characterization of AMPKK required the development of a sensitive and quantitative assay. We utilized its ability (24) to catalyze the time-and MgATP-dependent reactivation of purified AMPK, which had been inactivated by pretreatment with the purified catalytic subunit of protein phosphatase-2A (PP2A C ). We found it necessary to perform the protein phosphatase treatment immediately prior to the AMPKK assay, since the dephosphorylated form of AMPK appears to be unstable on storage. After the protein phosphatase treatment, okadaic acid was added to prevent the protein phosphatase interfering with the two subsequent kinase reactions. Fig. 1 shows that this reactivation is suitable for a quantitative assay using a crude postmitochondrial supernatant fraction from rat liver, a PEG precipitate prepared from it, or a much more highly purified fraction (Q-Sepharose). With crude fractions, the assay became nonlinear with protein concentration when the degree of reactivation was Ͼ30%. Remarkably, after purification through four steps (Q-Sepharose fraction) the assay was essentially linear right up to 100% reactivation.
In the initial studies, we assayed AMPK after reactivation using as substrate the synthetic SAMS peptide (HMRSAMS-GLHLVKRR), which is based on the primary phosphorylation site on rat acetyl-CoA carboxylase (34). However, as we purified AMPKK further this substrate began to give high blanks, i.e. there was significant phosphorylation of the peptide even if dephosphorylated AMPK was not added. This was eventually traced to the presence of protein kinase(s) that co-purified with AMPKK. The problem of the high blanks could be overcome by replacing SAMS with the peptide AMARAASAAALARRR, which we developed as a basis for specificity studies on AMPK and its homologues (32). Not only is this a better substrate for AMPK (V/K m 5-fold higher), but it was not a substrate at all for the SAMS kinase(s) that contaminated the AMPKK preparation. All subsequent studies were conducted using this peptide to assay the reactivated AMPK.
Purification of AMPKK-In previous studies (24) AMPKK had been detected as a byproduct of the AMPK preparation. It had been partially purified from the 2.5-6% PEG precipitate used in the AMPK purification, being resolved from AMPK at the third step (blue-Sepharose). With our new more quantitative assay we reexamined the recovery of AMPKK in the early stages of the procedure and found that only a small proportion of the total activity was recovered in the 2.5-6% PEG precipitate. The majority of AMPKK was in the 6 -10% PEG precipitate, which was therefore used as the starting point for further purification. A six-step procedure, utilizing affinity chromatography on ATP-␥-Sepharose (9, 39) as the final step, resulted in Ͼ1000-fold purification of AMPKK from a postmitochondrial supernatant fraction of rat liver (Fig. 2, Table I). Despite prolonged efforts, we were unable to achieve further purification by conventional means. Further studies were therefore conducted with this preparation.
Molecular Mass of AMP-activated Protein Kinase Kinase and Its Catalytic Subunit-SDS-PAGE revealed that the preparation still contained several polypeptides, but with a species of apparent molecular mass 58 kDa being a major component (Fig. 3, lane 1). When incubated with [␥-32 P]ATP, this polypeptide was essentially the only labeled species (Fig. 3, lane 2). We also conducted experiments aimed at identification of the catalytic subunit using the reactive ATP analogue FSBA. This reagent inactivates the catalytic subunit of cyclic AMP-dependent protein kinase by reacting with the lysine residue (40) that is conserved in the ATP binding site of all other protein kinases (41). As expected, FSBA caused a time-dependent, exponential decay of AMPKK activity, with half-times of 18 min at 250 M, and 43 min at 100 M FSBA (data not shown). Using [ 14 C]FSBA, the 58-kDa polypeptide was the only significant radioactive species detectable after an SDS-PAGE gel was analyzed by PhosphorImager analysis (Fig. 3, lane 3). The radioactivity migrating at the dye front was also evident in a control incubated without AMPKK (not shown) and is due to unreacted [ 14 C]FSBA or its breakdown products.
We also estimated the native molecular mass (42) of AMPKK by combining the Stokes radius determined by gel filtration (5.5-5.7 nm; Fig. 4) and the sedimentation coefficient determined by glycerol gradient centrifugation (8.4 -8.6 S; Fig. 5), which yielded a value of 196 Ϯ 5 kDa. The frictional ratio (f/f 0 ) was 1.44 -1.47, indicating that AMPKK has a highly asymmetric structure. Prolate or oblate ellipsoids of this frictional ratio would have axial ratios of ϳ8 and ϳ10, respectively. The migration of AMPKK on gel filtration was not affected by the presence or absence of 0.2 M NaCl in the buffer (not shown).
Dependence of AMPKK on AMP and ATP and Lack of Effect of Protein Phosphatases-We have previously reported that reactivation of AMPK by AMPKK is absolutely dependent on the presence of 5Ј-AMP (24). In the present study we examined whether the effect of AMP on the AMPKK reaction could be antagonized by high concentrations of ATP. Fig. 6 (open circles) shows the effect of varying the ATP concentration, at a constant AMP concentration of 100 M, during the AMPKK assay. Rather than the simple hyperbolic curve one might expect for a protein kinase reaction, a more complex curve was obtained, with ATP completely inhibiting the reaction at high concentration. The data could be fitted to a model in which ATP bound with positive cooperativity at the catalytic site(s) (half-maximal activation at 16 Ϯ 6 M; Hill coefficient (h) ϭ 1.6 Ϯ 0.5) but also bound at higher concentration at inhibitory allosteric site(s) (half-maximal inhibition at 360 Ϯ 80 M; h ϭ 1.9 Ϯ 0.4). These results show that, as for the allosteric effect of AMP on AMPK itself (11,25), the activation of AMPK by AMPKK in the presence of AMP is antagonized by high concentrations of ATP. When the experiment was performed at a higher AMP concentration (500 M; Fig. 6, filled circles), the whole curve was shifted to the right; half-maximal activation was now at 54 Ϯ 4  Since results in this study suggested that AMPKK and AMPK have several remarkable similarities, we considered the possibility that, like AMPK, AMPKK might be regulated by reversible phosphorylation. To examine this possibility, we incubated AMPKK with purified preparations of protein serine/ threonine phosphatases (PP2A C , PP2A 1 , PP2C␣, PP1␥ C ), a protein-tyrosine phosphatase (LAR), and a dual specificity protein phosphatase (CL-100). Although all were active against known substrates (see "Experimental Procedures" for further details), none had any effect on the activity of AMPKK (not shown).
Identification of the Activating Phosphorylation Site Phosphorylated by AMPKK on AMPK-In order to determine the site of activating phosphorylation, AMPK was dephosphorylated with PP2A C and incubated with AMPKK in the presence of [␥-32 P]ATP. The AMPKK preparation was not homogeneous and gave a fairly high background phosphorylation when incubated on its own (see Ref. 27). We therefore reisolated AMPK after phosphorylation, by immunoprecipitation using anti-␤ subunit antibody. When AMPK was dephosphorylated and incubated with AMP and [␥-32 P]ATP alone, autophosphorylation occurred principally on the ␣ subunit, with traces on the ␤ and ␥ subunits (Fig. 7); there was no activation under these conditions (not shown). When AMPKK was added to the incubation, phosphorylation of the ␣ subunit increased (although this may not be obvious in Fig. 7 due to the high autophosphorylation) and was accompanied by Ͼ95% reactivation of AMPK as expected. There was also increased labeling of the ␤ subunit.
When AMPK was incubated with AMPKK in the presence of AMP and [␥-32 P]ATP, an additional labeled polypeptide with an apparent molecular mass of ϳ58 kDa became evident in the anti-␤ subunit immunoprecipitate (Fig. 7). This corresponded to p58, the major autophosphorylated polypeptide in the AMPKK preparation (Fig. 3). The labeled p58 was not precipitated by the anti-␤ subunit antibody in the absence of AMPK (not shown). Unfortunately, p58 was not clearly resolved from the immunoglobulin heavy chain on the Coomassie-stained gel, so it was not possible to estimate its abundance relative to the AMPK subunits.
The ␣ subunit labeled in the presence or absence of AMPKK was purified by SDS-PAGE followed by electrophoretic transfer to polyvinylidene difluoride membrane. The polypeptide was digested on the membrane with trypsin as described under "Experimental Procedures." Fig. 8 shows the resolution of phosphopeptides obtained by reversed phase HPLC in 0.1% trifluoroacetic acid for the AMPKK-treated sample and for the autophosphorylation control. In addition to a small peak of radioactivity in the column breakthrough, autophosphorylation of AMPK gave rise to four radiolabeled peptide peaks (labeled 1-4 in Fig. 8, with peak 1 being a shoulder on the leading edge of peak 2). Peaks 1-4 were also seen in very similar amounts when AMPK was incubated with AMPKK, but . Max is the theoretical activity when the activating site is fully occupied and the inhibitory site is empty, A 0.5 is the concentration giving a half-maximal effect at the activating site, I 0.5 is the concentration giving a half-maximal effect at the inhibitory site, and n1 and n2 are Hill coefficients. Estimates, with standard errors, of these parameters are given under "Results." Curve fitting was carried out using Kaleidagraph (Abelbeck Software).
three new peaks (peaks 5, 6, and 7) were also seen eluting later in the gradient. Similar results were obtained in three experiments with different preparations of both AMPK and AMPKK, except that the distribution of radioactivity between trifluoroacetic acid peaks 5, 6, and 7 was somewhat variable, with peaks 6 and 7 being less prominent in other runs.
Trifluoroacetic acid peaks 5, 6, and 7 were each further purified by reversed phase HPLC in ammonium acetate, pH 6.5. In each case up to three closely eluting radioactive peaks were obtained, termed AA1-AA3 (not shown). Trifluoroacetic acid peak 5 yielded predominantly AA1, trifluoroacetic acid peak 6 a mixture of AA1 and AA2, and trifluoroacetic acid peak 7 a mixture of AA2 and AA3. The AA1 peaks from trifluoroacetic acid peaks 5 and 6 were combined, coupled through the C terminus to a Sequelon membrane, subjected to solid phase sequencing, and found to yield the single sequence IAXF-GLSXMMSXGXFLR (Fig. 9A), corresponding unequivocally to the N-terminal sequence of the predicted tryptic peptides from Ile 155 to Arg 171 in AMPK-␣ 2 (12) or AMPK-␣ 1 (14). Three of the unassigned residues could be accounted for by predicted acidic amino acids (Asp 157 , Asp 166 , Glu 168 ), which would be expected to yield blank cycles due to the coupling of the peptide to the filter disc through carboxyl groups. Counting of the PTH-derivative fractions from this run (Fig. 9A) showed that there was a small peak of radioactivity at cycle 7, corresponding to Ser 161 , followed by a larger peak at cycle 18, corresponding to Thr 172 . Trypsin apparently had not cleaved the peptide at Arg 171 , probably due to the proximity of the phosphate group on Thr 172 .
Although the radioactivity in cycle 7 in Fig. 9A might appear to be fairly substantial, after correction for the amounts of PTH-derivative being recovered at the respective stages of the sequencer run, it only accounted for 6% of the radioactivity loaded, whereas cycle 18 accounted for ϳ80%. When the amount of radioactivity in each of these cycles was converted into picomoles and compared with the pmol of PTH-derivative recovered, the stoichiometry of phosphorylation was ϳ3% at Ser 161 and Ͼ45% at Thr 172 . The low stoichiometry of phosphorylation at Ser 161 is also confirmed by the recovery of substantial amounts of PTH-serine in cycle 7 (where the amino acid is serine, the recovery of PTH-serine is usually Ͻ50% of that obtained for most other amino acids, whereas if it is phosphoserine, no PTH-serine is recovered). In addition, in another preparation of AA1 no radioactivity was recovered at cycle 7 (Fig. 9B).
Some radioactivity was also recovered in cycle 19 and, with one preparation, in cycle 20 (Fig. 9), corresponding to Ser 173 and Cys 174 , respectively. While it is not possible to rule out a slight phosphorylation of Ser 173 , we think these results are most likely explained by carry over of Thr 172 into cycles 19 and 20, due to the repetitive yield being less than 100%. In the run shown in Fig. 9A, the carry over of the PTH amino acid from cycle 15 into cycle 16 was Ͼ30%.
Sequencing of peak AA2 (pooled from runs derived from trifluoroacetic acid peaks 6 and 7) yielded the same major sequence as for peak 5. Once again the radioactivity was released in cycle 18 during solid phase sequencing (data not shown). There was insufficient material in peak AA3 to allow sequence determination. Fig. 10 shows separation of phosphorylated tryptic peptides The immunoprecipitate was analyzed by SDS-PAGE (10% gel), and the gel was dried and exposed to x-ray film. Migration of marker proteins (prestained markers, Life Technologies) is indicated on the right.

FIG. 8. Reversed phase HPLC of 32 P-peptides derived from the AMPK ␣ subunit phosphorylated using [␥-32 P]ATP in the presence (filled circles) and absence (open circles) of AMPKK.
Tryptic digests were prepared as described under "Experimental Procedures" and analyzed on a C18 column in 0.1% (v/v) trifluoroacetic acid using a gradient from water to acetonitrile. The flow rate was 250 l/min, and 250-l fractions were collected. Radioactivity was determined by Cerenkov counting of fractions.

FIG. 9. Recovery of radioactivity (filled circles) and semilogarithmic plot of recovery of PTH-derivatives (open circles) versus cycle number during solid phase sequencing of peptide 5. Panel
A shows results obtained from the preparation depicted in Fig. 8; the AA1 peaks derived from trifluoroacetic acid peaks 5 and 6 were combined for analysis. Panel B shows results from another preparation of trifluoroacetic acid peak 5 where the second (pH 6.5) HPLC run was omitted and only the recovery of radioactivity was determined. The abscissa is marked with the cycle number (B refers to counting of a reagent blank) and the residue, which is the single letter code of the predicted N-terminal sequence of the tryptic peptide from Ile 154 to Ala 190 of AMPK-␣ 1 or AMPK-␣ 2 : residues of this sequence identified are shown by open circles in panel A). Aspartic and glutamic acid residues were not detected because the peptide was coupled to the filter via carboxyl residues, so cycles 3, 12, and 15 yielded blanks as expected. The residue at cycle 8 (Asn expected) was not identified due to a problem with injection of the PTH-derivative onto the HPLC. The straight line shows fitting of the PTH-amino acid recovery to an exponential decay with a repetitive yield of 86%. derived from the ␤ subunit by reversed phase HPLC. In addition to a large peak in the breakthrough of the column, there were five labeled peptides resolved. Detailed analysis of these peptides will be presented elsewhere, but it is relevant here that, although some peaks (particularly 4 and 5) became more radioactive after treatment with AMPKK, the same peaks were also observed in the absence of AMPKK.

DISCUSSION
Although we did not quite achieve purification to homogeneity, the purification procedure for AMPKK described here is a great improvement over that previously available (24). We had originally observed that AMPK reactivating activity was still present in the AMPK preparation at the second step (DEAE-Sepharose) but that AMPK and AMPKK were completely resolved during the third step (blue-Sepharose). However, the development of the more quantitative AMPKK assay described here allowed us to monitor the recovery of AMPKK during the early stages of the preparation and revealed that the majority of the AMPKK activity was not in the 2.5-6% PEG pellet used to purify AMPK but rather in the 6% PEG supernatant in the first step, which we had previously discarded. Purifying AMPKK from the 6 -10% PEG pellet resulted in a much greater recovery of active AMPKK. The final preparation was still contaminated with protein kinase(s), which slowly phosphorylated the SAMS peptide, but AMPK was completely absent by several criteria. Utilization of AMARA for monitoring reactivation of AMPK during the AMPKK assay overcame the problem of the high blanks obtained due to the presence of the SAMS kinase(s).
Remarkably, the degree of reactivation of AMPK using highly purified preparations of AMPKK was linear with the amount of AMPKK added, right up to 100% reactivation (Fig.  1). This suggests that AMPKK has a very low apparent K m for AMPK, so that the rate does not fall although the concentration of dephospho-AMPK is dropping. The fact that AMPK and AMPKK seem to form a stable complex as judged by co-immunoprecipitation (see below) may be another reflection of high affinity binding between AMPK and AMPKK. Using cruder preparations of AMPKK (postmitochondrial supernatant or 6 -10% PEG precipitate), the assay became nonlinear when reactivation was Ͼ30% (Fig. 1). This may either be due to the presence of inhibitors of reactivation or due to the fact that at these early stages in the preparation (prior to the blue-Sepharose) AMPKK is still contaminated with AMPK and that this is not completely corrected by running blanks in which exogenous dephosphorylated AMPK is omitted.
Utilizing AMPKK purified as far as step 5 (Mono Q) in the present preparation, we have already shown that reactivation of AMPK is completely dependent on the presence of AMP (25). We now report that, using a fixed concentration of 100 M AMP, ATP inhibited the AMPKK reaction at high concentrations. This demonstrates for the first time that high concentrations of ATP inhibit the activation of AMPK by phosphorylation as well as causing direct allosteric inhibition of AMPK by binding at the allosteric site as reported previously (11,25). Although alternative models are possible, the results in Fig. 6 could be interpreted via a model in which ATP bound to the catalytic site(s) on AMPKK at low concentrations (half-maximal effect at 16 Ϯ 6 M; this is within the range of apparent K m values for ATP for other protein kinases) and at an inhibitory site at high concentrations (half-maximal effect at 360 Ϯ 80 M). However, with the present data it is not possible to distinguish between enzyme (AMPKK)-mediated and substrate (AMPK)-mediated effects of ATP. Elsewhere we have demonstrated that AMPKK is directly activated by AMP (27), raising the possibility that it is related to AMPK and that both AMPK and AMPKK may have two sites (catalytic and allosteric) for ATP. The biphasic curve in Fig. 3 may therefore reflect binding of ATP to multiple sites on both AMPKK and AMPK. Whatever the exact mechanism, these results show that high ratios of AMP to ATP would promote phosphorylation of AMPK by AMPKK (as long as there was sufficient ATP to satisfy its requirements as a substrate for AMPKK), whereas low ratios of AMP to ATP would inhibit phosphorylation of AMPK. The present observations in a cell-free system therefore parallel previous observations in intact cells that heat shock, arsenite, and other stresses that elevate the AMP:ATP ratio stimulate the activation of AMPK by phosphorylation (8).
Although conclusive identification of the catalytic subunit of AMPKK will have to await cDNA cloning and sequencing, the data in Figs. 3 and 7 strongly suggest that the catalytic subunit is a 58-kDa polypeptide (p58). Although other polypeptides could be detected in the preparation, p58 was the only polypeptide that was significantly labeled using [␥-32 P]ATP and [ 14 C]FSBA, both of which were used previously to identify the catalytic subunit of AMPK (11). In addition, 32 P-labeled p58 co-precipitated with AMPK, using antibody against the ␤ subunit of the latter (Fig. 7). Precipitation of p58 did not occur unless AMPK was present, showing that it was not due to a chance cross-reaction between p58 and the AMPK-␤ antibody. As well as co-precipitating p58, the anti-AMPK-␤ antibody precipitated AMPKK activity, based on two separate analyses (data not shown): (a) in four separate incubations, 42 Ϯ 8% (mean Ϯ S.E.) of AMPKK activity was depleted from the supernatant after immunoprecipitation; (b) when a mixture of dephosphorylated AMPK and AMPKK were immunoprecipitated in the absence of MgATP, the subsequent addition of MgATP to the precipitate resulted in 4-fold reactivation of the AMPK, relative to a control in which AMPKK was not included. Thus, both p58 and AMPKK activity could be co-precipitated with AMPK, suggesting that the two are related. Whatever their relationship, these data demonstrate that AMPK and AMPKK form a complex that is stable during extensive washing of the immunoprecipitate.
We have previously reported that both AMPKK and AMPK are directly activated by AMP (27). This study reveals further similarities between the upstream and downstream protein kinases. The molecular masses of the catalytic subunits (AMPKK versus AMPK, 58 versus 63 kDa), the Stokes radii Tryptic digests were prepared as described under "Experimental Procedures" and analyzed on a C18 column in 0.1% (v/v) trifluoroacetic acid using a gradient from water to acetonitrile. The flow rate was 250 l/min, and 250-l fractions were collected. Radioactivity was determined by Cerenkov counting of fractions. the possibility that, like AMPK, which consists of ␣ (63-kDa), ␤ (38-kDa) and ␥ (35-kDa) subunits, AMPKK might also contain additional subunits as well as the 58-kDa catalytic subunit. No obvious candidates for these were seen in a Coomassie-stained gel of the immunoprecipitate shown in Fig. 7, but they could have been obscured by the subunits of AMPK and/or by the heavy or light chains of the immunoglobulin. Whatever the structure of the AMPKK oligomer, migration on a gel filtration column was not affected by the presence or absence of 0.2 M NaCl in buffer B, indicating that the oligomer is stable under these conditions.
In this study we identify threonine 172 as the major site phosphorylated by AMPKK on AMPK-␣, leading to activation of AMPKK. We believe that all three radioactive peptide peaks that were labeled only after incubation with AMPKK are derived from this site. Trifluoroacetic acid peaks 5-7 eluted very close together (Fig. 8) and also yielded three closely eluting peaks on HPLC in ammonium acetate pH 6.5 (AA1-AA3, not shown). Although we did not have sufficient AA3 for sequence studies, AA1 and AA2 gave identical sequences, corresponding to the tryptic peptide commencing at Ile 155 in AMPK-␣ 1 or AMPK-␣ 2 , and in both cases the major radioactive amino acid was at cycle 18, corresponding to Thr 172 . We suspect that trifluoroacetic acid peaks 5-7 (corresponding to AA1-AA3) are generated by differing degrees of oxidation of the two methionines within the peptide.
At least four other 32 P-labeled tryptic peptide peaks (Fig. 8,  peaks 1-4) were separated from AMPK-␣, but these were not affected by the presence of AMPKK and appear to be autophosphorylation sites that are phosphorylated independently of the state of activation of AMPK.
New information to be taken into account was the report (14), published just as we were about to submit this work, that rat liver AMPK catalytic subunit was encoded by not one but at least two genes, termed ␣ 1 and ␣ 2 . The ␣ 2 isoform seems to be expressed at a high level in rat liver, but it appears to have a low SAMS peptide activity, with ␣ 1 apparently accounting for at least 90% of the SAMS peptide kinase activity in this tissue (14). The AMPK used for our identification of the phosphorylation site may have been a mixture of both isoforms, although because it had been purified on the basis of activity it presumably contained a substantial proportion of ␣ 1 . However, the sequence around Thr 172 is perfectly conserved between ␣ 1 and ␣ 2 , suggesting that AMPKK is likely to activate both forms. The predicted tryptic peptides containing this site (even allowing for failure to cleave at Arg 171 ) are also identical for both isoforms. Since the anti-␤ subunit antibody used for immunoprecipitation depletes all of the AMPK activity from even relatively crude preparations of rat liver (data not shown) it seems likely that it would precipitate both ␣ 1 and ␣ 2 . Although we therefore cannot be certain whether the amino acid sequence containing Thr 172 was derived only from ␣ 1 or from both ␣ 1 and ␣ 2 , our overall conclusions are not affected.
In contrast to the autophosphorylation sites on AMPK-␣, the ␤ subunit is more highly phosphorylated in the presence of AMPKK (Fig. 7). More detailed studies on the phosphorylation of the ␤ subunit will be presented elsewhere, but current evidence 2 favors the idea that this is due to autophosphorylation and is not catalyzed by AMPKK.
Although further studies (e.g. site-directed mutagenesis) will be required to conclusively demonstrate that activation of AMPK by AMPKK is accounted for entirely by phosphorylation of Thr 172 , comparison with other protein kinases lends consid-erable support to this hypothesis. AMPK joins a long list of protein kinases known to require phosphorylation within the "activation segment" (43) between the "DFG" motif (subdomain VII (41)) and the "APE" motif (subdomain VIII) for activity. The phosphorylation site on protein-serine/threonine kinases is usually a threonine but can also be tyrosine (Gsk3), threonine and tyrosine (mitogen-activated protein kinases), or two serines (mitogen-activated protein kinase kinases). In some cases, such as cyclic AMP-dependent protein kinase, these phosphate groups may be inserted during protein synthesis and appear to be stable in vivo. In many other cases such as Cdc2, mitogen-activated protein kinases, mitogen-activated protein kinase kinases, p90 ribosomal protein S6 kinase and AMPK, the phosphorylation can be readily reversed and has important regulatory consequences in vivo. In the crystal structure of the C␣ subunit of cyclic AMP-dependent protein kinase, the phosphorylated threonine (Thr 197 ) interacts with a positively charged cluster (His 87 , Arg 165 , and Lys 189 ), thus permitting proper orientation of the pseudosubstrate peptides for phosphate transfer from ATP (43). Two of these three basic residues are conserved in AMPK-␣1 and -␣2 (12,14).
The yeast (S. cerevisiae) homologue of AMPK, Snf1p, is inactivated by treatment with protein phosphatases and can be reactivated by treatment with a crude preparation of mammalian AMPKK (22), indicating that it may be regulated by phosphorylation at the site equivalent to Thr 172 . A threonine residue (Thr 210 ) is conserved at this position in Snf1p, and the sequence around it is also highly conserved, with a few conservative replacements on the N-terminal side. Estruch et al. (44) have already shown that a T210A mutant is nonfunctional in yeast in vivo with respect to growth on raffinose or expression of invertase. This is entirely consistent with Thr 210 being the site on Snf1p phosphorylated by a putative "Snf1p kinase," which is currently uncharacterized.
Calmodulin-dependent protein kinase I (CaMKI) is activated by phosphorylation by an upstream protein kinase (CaMKI kinase), and CaMKI kinase can also phosphorylate and activate calmodulin-dependent protein kinase IV, at least in vitro. The sites of regulatory phosphorylation modified by CaMKI kinase have been shown to be Thr 177 (CaMKI (45)) and Thr 196 (calmodulin-dependent protein kinase IV (46)), which exactly align with Thr 172 on AMPK. We have recently shown that, although AMPKK and CaMKI kinase are distinct enzymes, AMPKK can slowly phosphorylate and reactivate CaMKI, while CaMKI kinase can slowly phosphorylate and reactivate AMPK (27). These observations are entirely consistent with our identification of Thr 172 as the regulatory site on AMPK, given the relatively high sequence similarity of AMPK and CaMKI within the activation segment. Whether AMPKK can phosphorylate and activate other protein kinases more distantly related to AMPK than CaMKI remains to be determined.