Multisite and Hierarchal Protein Phosphorylation*

Phosphorylation of proteins at Ser, Thr, and Tyr residues is one of the most frequent forms of posttranslational modification in eukaryotic cells and is linked to the control of a multitude of cellular functions (1, 2). The historical prototype for proteins controlled by phosphorylation is mammalian glycogen phospho- rylase, an enzyme activated by modification of a single Ser per subunit by a single protein kinase (3). As the number of examples of phosphorylated proteins has escalated in recent years, it has become apparent that the majority of phosphoproteins contain multiple sites. This review seeks to survey the phenomenon of multisite' phosphorylation and to evaluate its significance.

Phosphorylation of proteins at Ser, Thr, and Tyr residues is one of the most frequent forms of posttranslational modification in eukaryotic cells and is linked to the control of a multitude of cellular functions (1,2). The historical prototype for proteins controlled by phosphorylation is mammalian glycogen phosphorylase, an enzyme activated by modification of a single Ser per subunit by a single protein kinase (3). As the number of examples of phosphorylated proteins has escalated in recent years, it has become apparent that the majority of phosphoproteins contain multiple sites. This review seeks to survey the phenomenon of multisite' phosphorylation and to evaluate its significance.

Which Proteins Are Multiply Phosphorylated?
In assessing any specific role of multiple, as opposed to single site, phosphorylation, an immediate question is whether the proteins so modified fall into any particular classes. Almost certainly, only a fraction of multiply phosphorylated proteins have yet been identified, and efforts to survey multiply phosphorylated proteins, as in Fig. 1, inevitably involve a restricted data base. In compiling Fig. 1, the guidingprinciple was to identify proteins for which the location of phosphorylations was either known precisely or at least localized to some specific domain of the protein. ' The best response to the question posed at the outset of this paragraph is that the occurrence of multiple phosphorylation appears no more restricted than the occurrence of phosphorylation in general.
The survey of Fig. 1 does reveal some interesting features. First, the number of phosphorylation sites observed in proteins varies from 1 to over 100 ( Fig. 1) and no special numerology emerges. The only distinction that can perhaps be made is between proteins that contain relatively few sites and those that are heavily phosphorylated. The egg yolk protein phosvitin, neurofilaments, and the COOH-terminal tails of the large subunit of eukaryotic RNA polymerases are good examples of the latter phenomenon. Second, Ser(P), Thr(P), and Tyr(P) can be found in the same protein. Third, multiple phosphorylation sites tend not to be randomly distributed and are usually concentrated in relatively short segments of the polypeptide chain. Often, these phosphorylated regions are located at the extreme NH2 or COOH termini of proteins. Since phosphorylation requires interaction with protein kinase(s), regions of phosphorylation will be defined in part by their accessibility, such as on the surface of a globular protein. In addition, their location must also be related to function which will vary from protein to protein. These restrictions alone, however, are unlikely to explain the occurrence of clustered sites. Another perspective on the clustering of sites in short segments is an evolutionary one. An interesting comparison is between the Grants DK27221 and DK42576. (Y and regulatory subunits of phosphorylase kinase (6). These polypeptides have significant overall sequence homology, but 9 out of the 10 phosphorylation sites are in sequences specific to one subunit or the other. The a-subunit contains an -100-residue insertion that harbors some seven sites while the P-subunit has an NH2-terminal extension that contains two sites. In this regard, it is relevant that related proteins in different species or tissues often differ most at their termini. Glycogen synthase is an enzyme with its phosphorylation sites localized to the termini of the polypeptide (7). In comparing mammalian isoforms and the yeast enzymes, the greatest divergence is seen in precisely these regions. The yeast versions lack entirely the NHz-terminal regulatory domain, and the other major differences are in the region of COOH-terminal phosphorylation sites. Two messages for acetyl-CoA carboxylase have been detected that differ as to the presence or absence of sequences encoding an 8-amino acid segment just upstream of an important phosphorylation site (8). The insertion disrupts the ability of CAMP-dependent protein kinase to phosphorylate this site (Ser-1200 in the shorter message).

How Are Proteins Multiply Phosphorylated?
Any discussion of mechanism leads to consideration of the specificity of protein kinases. These enzymes range from being highly specific, designed to phosphorylate even a single substrate, to having very broad substrate specificity. The latter class of protein kinase is faced with recognizing specific determinants that can be duplicated in several or many protein substrates. Studies of the specificity of protein kinases, based both on surveys of natural sites and analysis of synthetic peptide substrates, have defined, in several cases, local sequence motifs that appear to accomplish this goal (4,5,9). Many protein kinases have requirements for positively or negatively charged groups in the vicinity of the modified residue and have been termed basotropic or acidotropic accordingly (10).
There are two fundamental mechanisms of multisite protein phosphorylation (Fig. 2). The simplest is one in which the substrate protein has multiple copies of the recognition determinants for one or more protein kinases. The density at which multiple phosphates can be introduced in this way is obviously linked to the amount of sequence information involved in the recognition, enzymes with very simple requirements being able to introduce more phosphate groups within a given stretch of the polypeptide chain. Protein kinases of the cdc2 family, which appear to recognize simple -S/T-Por -K-S/T-P-motifs, may be examples here (9). It is interesting that some very heavily phosphorylated proteins are modified in regions of sequence repeats. Examples are the neurofilaments NF-H and NF-M which contain multiple copies of a -K-S-Psequence (11,12) and the RNA polymerase I1 large subunit which, depending on species, can have as many as 52 repeats of a heptameric sequence unit (13,14). Microtubuleassociated protein 2 has been reported to contain more than 30 phosphates per polypeptide (15) and is a substrate i n vitro for numerous protein kinases (see Ref. 16). Identification of these sites is incomplete, but the multiplicity of sites in this case is not related to any sequence repeat but rather to the presence of numerous Ser and Thr residues in a generally polar protein. Many other proteins ( Fig. 1) contain more moderate numbers of sites whose modification can be explained by the independent action of one or more protein kinases and do not involve global sequence features like sequence repeats.
A mechanism unique to multiply phosphorylated proteins is when the introduction of phosphate groups influences the subsequent phosphorylation reaction(s) (Fig. 2). The first example worked out in molecular terms was that of the phosphorylation 14139 -E m y m e s of glycogen synthase by glycogen synthase kinase-3 (GSK-3)3 and casein kinase I1 (7). Totally dephosphorylated glycogen synthase is not a substrate for GSK-3. Upon introduction of phosphate at one specific site by casein kinase 11, the protein becomes a substrate for GSK-3, which sequentially modifies 4 Ser residues.

Minireview: Multisite and Hierarchal Protein Phosphorylation
The results are explained if GSK-3 recognizes sites in the motif -S-X-X-X-S(P)-(17). In this example, another feature of the reaction is that GSK-3 introduces multiple phosphates, due to the presence of adjacent repeats of the -S-X-X-X-S-motif. As one phosphate is introduced, a new GSK-3 site is generated. At the time of writing, seven examples of hierarchal phosphorylation involving more than a dozen GSK-3 sites have been recorded (Table I). In three instances the primary protein kinase is CAMPdependent protein kinase and in three cases casein kinase 11. The important feature for recognition by GSK-3 is the presence of a phosphate in an appropriate site and not the kinase that introduced it. Another enzyme that can act as a secondary protein kinase in the sense above is casein kinase I which has a particular selectivity for sites in the motif -S(P)-X-X-S-(10). Again, sites in glycogen synthase provide the best examples to date. The Golgi "The abbreviation used is: GSK-3, glycogen synthase kinase-3.

S-X-E/SiPj ~ ' s -x -x -x -S ( P ) BARK
E-X-S casein kinase (9, 29) and casein kinase I1 (30, 31) are acidotropic kinases that phosphorylate sites in motifs containing either Glu, Asp, or Ser(P) (Table 11). Thus, these enzymes can act either as primary or as secondary protein kinases and have the potential t o phosphorylate initially on the basis of existing Asp or Glu residues and subsequently on the basis of the covalent phosphate introduced. For example, the sequence -S(P)-X-S(P)-S(P)-S(P)-E-E-, found in caseins (33) and riboflavin-binding protein (34), can be explained by the ordered action of an enzyme recognizing the motif -S-X-E/S(P)-. One possibility raised by the occurrence of hierarchal phosphorylation is for "cross-talk" becween protein Ser/Thr kinases and protein Tyr kinases. The protein Tyr kinases appear often, although not always, to be acidotropic, and one could thus ask whether Ser(P)/Thr(P) could be involved in recognition. Likewise, could Tyr(P) be recognized by protein Ser/Thr kinases? No physiological example of either is known, but Pinna and colleagues (35) have shown that Tyr(P) in a synthetic peptide can act as a recognition determinant for casein kinase 11. A potential example of protein tyrosine kinases in a hierarchal scheme is the autophosphorylation of the insulin receptor in which the modification of 3 tyrosine residues involved in control of kinase activity is reported to be ordered (36).
Most of the known interdependent phosphorylations involve a positive role for the initial phosphorylation though a priori there is no reason why phosphorylation might not impair the action of a second kinase. For example, hormone-sensitive lipase (27) has two phosphorylation sites, Ser-563 (site 1) and Ser-565 (site 2). Phosphorylation of Ser-563 by CAMP-dependent protein kinase activates the enzyme whereas Ser-565 modification is without effect. However, phosphorylation of the two sites is mutually exclusive so that modification of Ser-565 could control activity indirectly by reducing phosphorylation at Ser-563. A similar situation holds for acetyl-coA carboxylase. Phosphorylation of Ser-77 and Ser-1200 by CAMP-dependent protein kinase prevents phosphorylation of Ser-79 by the AMP-dependent protein kinase (28).
The exact role of the phosphate in hierarchal phosphorylation schemes has not been defined. One possibility is that the phosphoserine itself is involved in recognition contacts at the active site of the secondary kinase. This idea fits with the fact that most enzymes so far identified as acting in a secondary manner are acidotropic, their recognition motifs characterized by acidic residue(s) close to the modified amino acid (Table 11). In the case of GSK-3, it should be noted that -S-X-X-X-S(P)-is the shortest unit recognition sequence and in two substrates, phosphatase inhibitor 2 and CAMP-dependent protein kinase RII subunit, the phosphates are more distant in the linear sequence. One can speculate that the folding of the protein might bring the target and the recognition serine phosphate into similar juxtaposition as in a -S-X-X-X-S(P)-unit. Two multiply phosphorylated GSK-3 substrates, c-Jun and c-Myb, do not fit the recognition criteria noted above and are reported to be phosphorylated without prior phosphorylation (37).
A second possibility is that the introduction of a phosphate group changes the conformation of the substrate into one that is recognized by a secondary protein kinase without the phosphate group itself participating in kinase-substrate contacts. One potential example is the enhanced phosphorylation of Thr-34 in DARPP32 by CAMP-dependent protein kinase once Ser-45 and/ or Ser-102 has first been phosphorylated by casein kinase I1 (26). The CAMP-dependent protein kinase typically recognizes basic motifs such as -R-R-X-S-, and a phosphate group would not be expected to participate directly in recognition. Hierarchal phosphorylation mediated by conformational changes in the substrate would not require the secondary kinase to be acidotropic.

Why Are Proteins Multiply Phosphorylated?
There are two related but distinct perspectives on the possible role of multisite phosphorylation. One is at the level of the structural changes elicited by the introduction of phosphate groups and the subsequent effects on protein function. Multiple phosphorylations could correlate with the generation of a variety of protein forms, in which one or more properties are altered. Different phosphorylations could thus be linked to distinct protein functions or graded effects on a single function. Alternatively, multiple phosphates might be necessary to cause one critical conformational change. The other perspective is at the level of the regulation of the different phosphorylations. Obviously, if the actions of more than one kinase influence the functional status of a target protein, more complex regulation could be exerted relative to the action of a single kinase. Examples below are selected to illustrate some of the features that may be exclusive to multiple phosphorylations.
Hormones regulate the phosphorylation of glycogen synthase, and the multisite phosphorylation is in part linked to the occurrence of the hierarchal mechanism described above (7). An important feature is that the primary phosphorylations, such as mediated by casein kinase I1 and CAMP-dependent protein kinase, at best have moderate effects on activity; effective inactivation requires the occurrence of the secondary phosphorylations catalyzed by GSK-3 or casein kinase I. Thus, some phosphorylations function to alter activity whereas others influence kinase recognition. Not all the details linking hormone action to the control of phosphorylation have been worked out, but an important aspect of glycogen synthase control is that, mechanistically, the dozen or more phosphorylations do not occur totally independently and a smaller number of multiply phosphorylated units can be defined.
Acetyl-coA carboxylase is another metabolic enzyme whose multiple phosphorylation is regulated by hormones (28,38). In this protein, there is evidence that different phosphorylation sites have different influences on the kinetic properties of the enzyme. From site-directed mutagenesis studies, it appears that phosphorylation of Ser-77 and Ser-79 is linked primarily to decreases in V,,, whereas modification of Ser-1200 increases the KO for the allosteric activator citrate: It has also been reported that phosphorylation of the enzyme by casein kinase I1 at site 6 (Ser-29), itself without effect on activity, may influence dephosphorylation of other sites that do control activity (39). The tyrosine protein kinase pp60"-"" provides an example of a protein in which different tyrosine phosphorylations can potentially modulate protein function (40). The autophosphorylation site, Tyr-416, may activate the kinase, whereas Tyr-527 is an inactivating site that lies, interestingly, in precisely the region of the molecule missing in the retroviral transforming gene product pp60""". The viral protein therefore lacks the negative control, and the unconstrained activity of pp60""" is thought to be linked to transformation by Rous sarcoma virus. pp60"~"" is also phosphorylated at several Ser and Thr residues close to the NH2 terminus, including sites for CAMP-dependent protein kinase and protein kinase C that may also be linked to activation of the kinase. The activity of pp60""" may thus depend on the phosphorylation of multiple sites.
The 0-adrenergic receptor undergoes phosphorylation in two sets of functionally distinguishable sites (41), modified by CAMPdependent protein kinase and the 0-adrenergic receptor kinase, respectively. Receptor activation promotes its own phosphorylation by CAMP-dependent protein kinase which is thought to mediate short-term desensitization of the receptor by low levels of agonist, so-called heterologous desensitization. Exposure to high agonist levels results additionally in phosphorylation in the extreme COOH terminus at sites for P-adrenergic receptor kinase. The 0-adrenergic receptor kinase only recognizes receptor occupied by agonist, leading to what has been termed homologous desensitization. The precise mechanisms of desensitization have still to be elucidated but potentially involve altered interactions with the G-protein G, and/or the accessory molecule arrestin. Both types of phosphorylation represent a feedback control on receptor function, but the phosphorylations are distinguishable both in the feedback circuit utilized and possibly the exact functional effect of the phosphorylation.
Some phosphorylations are irreversible in the sense that they occur once in the lifetime of a protein, and the phosphate groups themselves must be required for protein function (33). For example, caseins are proteins in which the phosphate is important for the structure of the casein micelles that maintain otherwise insoluble ions, notably calcium, in suspension. Phosvitin and riboflavin-binding protein are synthesized in the liver of laying hens and are transported to the egg. In phosvitin, the most highly phosphorylated protein known with over 100 phosphates per polypeptide, the covalent phosphorylation serves as a physical source of phosphate and perhaps also to bind and transport cations. There is also evidence for both proteins that phosphorylation is needed for their proper deposition in egg yolk and the phosphate could be involved in receptor recognition (42).
Other situations in which it is tempting to speculate a requirement for multiple phosphorylations are when large numbers of phosphates are clustered in a segment of a protein (Fig. 1). Some of the more heavily phosphorylated proteins come from the ranks of the intermediate filaments. Neurofilaments NF-L and NH-M, for example, are phosphorylated in both their NH-terminal heads (43, 44) and more densely in the COOH-terminal tails (11,12). Phosphorylation has been implicated in determining the structures of intermediate filaments as well as interactions with other cytoskeletal elements. Though a precise need for multisite phosphorylation has not yet been demonstrated, one could surmise that the high degree of phosphorylation correlates with major protein structural changes which a single phosphorylation might not be sufficient to invoke. When massive phosphorylation occurs, one could also ask whether a precise set of sites is involved or whether there is a degree of randomness, the evolved goal being simply to introduce a sufficient number of phosphates.

Summary
Multisite phosphorylation is a prevalent form of protein modification whose full implications are just beginning to be understood. Multiple protein modifications expand the repertoire of structural changes that can be elicited in proteins and permit more intricate regulatory circuits to operate. and to Bruce Kemp and Richard Pearson for allowing me access to their reviews