Our results provide strong evidence that LKB1:STRAD:MO25 complexes represent the major upstream kinases acting on AMPK, although they do not rule out the possibility that the complex might contain additional components. The key evidence may be summarized as follows. First, during previous extensive efforts to purify from rat liver extracts activities that activate dephosphorylated AMPK, ( and subsequent unpublished work), we have not detected any activities other than AMPKK1 and AMPKK2, at least under the assay conditions used. Second, both AMPKK1 and AMPKK2 purified from rat liver contained LKB1, STRADα and MO25α that were detectable by western blotting and whose presence correlated with AMPKK activity across the column fractions (Figure 1). Third, the ability of the AMPKK1 and AMPKK2 fractions to activate AMPK was almost completely eliminated by immunoprecipitation with anti-LKB1 antibody, but not a control immunoglobulin. Activity was also detected, along with the LKB1, STRADα and MO25α polypeptides, in the anti-LKB1 immunoprecipitates but not in the control immunoprecipitates (Figure 2). Fourth, the AMPKK activity in AMPKK1 and AMPKK2 was not a contaminant that co-precipitated with anti-LKB1 antibody, because recombinant complexes of GST-LKB1, STRAD and MO25 expressed in 293 cells and purified on glutathione-Sepharose also activated the AMPKα1 catalytic domain efficiently, and phosphorylated Thr172 (Figure 3). Complexes formed from a catalytically inactive mutant LKB1 failed to activate or phosphorylate AMPK. Phosphorylation of the AMPKα1 catalytic domain by this recombinant complex occurred exclusively at Thr172, because the wild-type AMPKα1 catalytic domain, but not a T172A mutant, could be phosphorylated using [γ-32P]ATP and the GST-LKB1:STRADα:MO25α complex. Fifth, although most of the experiments in this study were conducted using the bacterially expressed AMPKα1 catalytic domain as substrate, AMPKK1, AMPKK2 and recombinant LKB1-STRADα-MO25α complexes also efficiently activated heterotrimeric AMPK complexes, both the α1β1γ1 and α2β1γ1 isoforms (Figure 4). Sixth, HeLa cells, unlike HEK 293T cells, do not express LKB1 (Figure 5) and therefore represent a natural 'knockout' cell line. The drugs AICA riboside and phen-formin, which activate AMPK in other cell types via distinct mechanisms [24, 27], did not activate AMPK in HeLa cells. In cells stably transfected with DNA that expressed wild-type LKB1 (but not a catalytically inactive mutant), however, the ability of AICA riboside and phenformin to activate AMPK, to phosphorylate Thr172 on the AMPKα subunit, and to cause phosphorylation of a downstream target (ACC) was restored (Figure 6). This experiment proves that (in the presence of STRADα and MO25α) LKB1 is sufficient for AMPK activation, but does not prove that it is necessary, because expression of upstream kinases other than LKB1 might also be defective in HeLa cells. Figure 5 also confirms that STRADα and MO25α are necessary to generate an active complex because, although the LKB1 polypeptide was greatly overexpressed in the stably transfected HeLa cells compared to the endogenous level in 293 cells, the AMPKK activity, and the amounts of STRADα and MO25α, in the anti-LKB1 immunoprecipitate, were actually less. This suggests that the amount of active LKB1 was limited by the availability of STRADα and MO25α. Seventh, in LKB1
+/+ MEF cells, AMPK became activated in response to both AICA riboside and phenformin. In LKB1
-/- MEF cells, however, the basal activity of AMPK was lower and AICA riboside and phenformin failed to activate AMPK. These results show that LKB1 is both necessary and sufficient for AMPK activation, at least in MEF cells.
All of the assays of the activity of LKB1 and its complexes described in this article, whether using the AMPKα1 catalytic domain or AMPK heterotrimers as substrate, utilized MgATP as the co-substrate. Previous studies of the kinase activity of LKB1, whether utilizing autophosphorylation , or p53  or myelin basic protein  as substrates, had used MnATP as co-substrate and had reported that there was no activity with the more physiological MgATP complex. Thus, unlike previously used substrates, AMPK is a good substrate for LKB1 complexes even using the physiologically relevant divalent metal ion.
We have previously reported that the activation of AMPK by AMPKK1 (called at that time AMPKK) was stimulated by AMP, and presented evidence favoring the hypothesis that AMP acted not only by binding to the downstream kinase and making it a better substrate, but also by activating the upstream kinase . The results in Figure 4c support the first hypothesis but do not support the second. Using α1β1γ1or α2β1γ1 AMPK complexes as substrate, activation by AMPKK1, AMPKK2 or LKB1 was stimulated from 2-to 3.5-fold by AMP. When using the AMPKα1 catalytic domain as substrate, however, AMP had no effect, or even slightly inhibited activation. The AMPKα1 catalytic domain is not allosterically activated by AMP , and AMP binding appears to be a function of the γ subunit (, and unpublished observations). Taken together with previous results [6, 24, 31], these data support the idea that the effects of AMP on the kinase cascade are all mediated through binding to the downstream kinase, AMPK. The previous report that AMP stimulated the upstream kinase was obtained using a less pure AMPKK preparation , and we have been unable to reproduce this with the more purified preparations utilized here.
Both AMPKK1 and AMPKK2 appear to contain LKB1, STRADα and MO25α, and thus it is not clear at present why they resolve on Q-Sepharose chromatography. One interesting difference is that the LKB1 polypeptide in AMPKK1 migrated significantly faster on SDS gels than that in AMPKK2 (Figures 1 and 2). LKB1 is known to be phosphorylated at up to eight sites, and is also farnesylated at Cys433, near the carboxyl-terminus [23, 28, 32, 33], suggesting that the difference in mobility might be due to a difference in covalent modification. It did not appear to be due to differential phosphorylation, however, because neither incubation with MgATP, nor protein phosphatase treatment, produced a shift in mobility of the LKB1 polypeptides in either AMPKK1 or AMPKK2 (Figure 1c). Another difference between AMPKK1 and AMPKK2 was their Stokes radii estimated by size exclusion chromatography (5.7 versus 5.2 nm respectively). By combining estimates of Stokes radius and sedimentation coefficient, we previously estimated the molecular mass of AMPKK1 to be 195 kDa , and assuming a similar shape our estimate of the Stokes radius of AMPKK2 would suggest a mass of about 175 kDa. These values are close to, although slightly larger than, the calculated mass of 140 kDa for a 1:1:1 LKB1:STRADα:MO25α complex. Although we cannot rule out the possibility that AMPKK1 and/or AMPKK2 contain additional associated protein(s) other than LKB1, STRADα/β and MO25α/β, it is also possible that differences in covalent modification might affect the shape of the complex and hence the Stokes radius. Whatever the reason for the difference in electrophoretic and chromatographic behavior of AMPKK1 and AMPKK2, a clear conclusion from Figure 4 is that, for the same amount of LKB1, STRADα and MO25α polypeptides, the former was more active than the latter. Although further work is required to explain these differences, they might be caused by the same covalent modifications that alter the mobility on SDS gels. Figure 4 also shows that, for the same amount of LKB1, STRADα and MO25α polypeptides, both AMPKK1 and AMPKK2 were much more active than the recombinant complex. The low activity of the latter might be explained by the presence of the purification tag on each subunit, by imperfect folding or assembly, or by an altered level of covalent modification, when the complex is overexpressed. As mentioned above, our data do not rule out the possibility that the recombinant LKB1 complex may be lacking additional subunit(s) present in the endogenous AMPKK1 and AMPKK2 complexes.
Our present results confirm, using a probable physiological substrate, previous findings using an artificial substrate (myelin basic protein) that a STRAD subunit stimulates the kinase activity of LKB1 , and that the MO25 subunit stimulates the activity further, probably by stabilizing the LKB1:STRAD complex . No AMPKK activity was obtained with recombinant LKB1 unless a STRAD subunit was also expressed, and the activity was increased substantially by the additional presence of a MO25 subunit (Figure 3). It was also noticeable that the amount of STRADα and STRADβ that co-precipitated with LKB1 was greatly enhanced by the co-expression of MO25α or MO25β (Figure 3), consistent with previous findings . Another new result in this article is that STRADα protein (unlike MO25α) was not detectable in HeLa cells unless either wild-type or kinase-dead LKB1 was stably expressed (Figure 6b). These results suggest that STRADα is normally complexed with LKB1 in the cell, and that STRAD is unstable in the absence of LKB1. The exact mechanism by which the STRAD and MO25 subunits activate LKB1 remains unclear, but these accessory subunits introduce scope for additional regulation of the kinase. It is already known that LKB1 phosphorylates STRADα at two distinct sites , and that STRADα and MO25α form a complex that retains LKB1 in the cytoplasm .
People with PJS are heterozygous for mutations in LKB1, and further work is required to establish whether loss of one allele of LKB1 could affect AMPK activation in these patients. An interesting unanswered question is whether activation of AMPK can explain the ability of LKB1 to act as a tumor suppressor and to arrest cell growth and proliferation. This certainly seems plausible, because apart from the fact that AMPK is a general inhibitor of biosynthesis [1, 2], there is accumulating evidence that it can regulate cell proliferation and apoptosis. For example, activation of AMPK inhibits proliferation of HepG2 cells by stabilizing p53 . Interestingly, expression of LKB1 in G361 cells that normally lack expression of the kinase causes an arrest in G1 phase of the cell cycle that is associated with an induction of p21 and is dependent on p53 [12, 13].
Another exciting possibility is that LKB1:STRAD:MO25 complexes might also act as upstream kinases for other protein kinases, in the same manner that PDK1 phosphorylates threonine residues in the activation loop of a number of kinases of the 'AGC' subfamily . A dendrogram showing the relationships between catalytic domain sequences of 518 human protein kinases encoded in the human genome  shows that the AMPKα1 and AMPKα2 subunits lie on a small sub-branch also containing eight other protein kinases (NuaK1, NuaK2, BrsK1, BrsK2, SIK, QIK, QSK and MELK), most of which either have not been studied previously or have very little known about them. An alignment of the activation loop sequences of these kinases is shown in Additional data file 2 with the online version of this article and show that, as well as conservation of the threonine residue phosphorylated by LKB1 in AMPK, they have other conserved motifs that are not present in other protein kinases known to be activated by other upstream kinases. It remains to be determined whether the other kinases in the AMPK subfamily are activated by phosphorylation of the conserved threonine residue by LKB1:STRAD:MO25 complexes, but if this is the case these other kinases might mediate some of the tumor suppressor functions of LKB1.
A significant number of inherited forms of PJS found in certain families do not exhibit mutations in the LKB1 gene [37, 38], indicating that there could be other causative loci for PJS. On the basis of the results presented here it would be very interesting to examine whether mutations in the genes encoding STRADα or β, MO25α or β, or any of the subunits of AMPK or of the AMPK-like subfamily of kinases, are found in PJS patients who do not have mutations in the LKB1 gene.
While this article was under review, two papers were published that are relevant to our results. Hong et al.  reported that FLAG-tagged LKB1 expressed in, and purified from, COS7 cells would activate a recombinant AMPK heterotrimer, and phosphorylate the α subunit at Thr172, in cell-free assays. This is consistent with our results, although these authors provided no evidence that LKB1 acts on AMPK in vivo, or that LKB1 required the presence of the STRAD and MO25 subunits to phosphorylate AMPK. Spicer et al.  reported evidence, based on expression of recombinant LKB1 in cultured cells, suggesting that it might act upstream of the PAR1A protein kinase. PAR1A (also known as MARK-3 ) lies with three closely related protein kinases (MARK-1, MARK-2, MARK-4) on a branch of the human kinase tree  immediately adjacent to AMPK-α1 and -α2 and the eight AMPK-like kinases discussed above. Although Spicer et al.  did not provide evidence that LKB1 directly phosphorylated PAR1A, the results of our study suggest that this might be the case. The sequence of the activation segment of PARIA is given in Additional data file 2 (available with the online version of this article).