Exceptional Disfavor for Proline at the P+1 Position among AGC and CAMK Kinases Establishes Reciprocal Specificity between Them and the Proline-directed Kinases*

To precisely regulate critical signaling pathways, two kinases that phosphorylate distinct sites on the same protein substrate must have mutually exclusive specificity. Evolution could assure this by designing families of kinase such as basophilic kinases and proline-directed kinase with distinct peptide specificity; their reciprocal peptide specificity would have to be very complete, since recruitment of substrate allows phosphorylation of even rather poor phosphorylation sites in a protein. Here we report a powerful evolutionary strategy that assures distinct substrates for basophilic kinases (PKA, PKG and PKC (AGC) and calmodulin-dependent protein kinase (CAMK)) and proline-directed kinase, namely by the presence or absence of proline at the P+1 position in substrates. Analysis of degenerate and non-degenerate peptides by in vitro kinase assays reveals that proline at the P+1 position in substrates functions as a “veto” residue in substrate recognition by AGC and CAMK kinases. Furthermore, analysis of reported substrates of two typical basophilic kinases, protein kinase C and protein kinase A, shows the lowest occurrence of proline at the P+1 position. Analysis of crystal structures and sequence conservation provides a molecular basis for this disfavor and illustrate its generality.

Phosphorylation is a prevalent modification in cells that controls many functions such as signaling transduction, proliferation and apoptosis. It is estimated that at least one-third of all proteins in eukaryotic cell are phosphorylated at any given time (1,2). More than 500 human protein kinases have been identified so far (3). A high degree of selectivity in substrate phosphorylation is necessary to maintain functional integrity of this very complicated signaling environment. Precision in phosphorylation is particularly critical when a substrate protein is phosphorylated at two (or more) phosphorylation sites, and those sites 1) are phosphorylated by distinct upstream kinases and 2) confer distinct properties on that substrate. To assure fidelity of signaling in this common situation, each upstream kinase must show high specificity by phosphorylating only the relevant phosphorylation site and not the inappropriate site(s). This requirement poses a major challenge in evolutionary design of kinase peptide specificity, since those upstream kinases are usually recruited to the substrate, and such recruitment can overcome much of the barrier provided by peptide specificity (4).
These considerations raise the important issue as to what elements in kinase peptide specificity confer the strongest reciprocal specificity between kinases, i.e. which prevent one kinase from phosphorylating substrates phosphorylated by another. For Ser/Thr kinases, we propose that much of this reciprocal specificity is provided by the evolution of three broad classes (5): basophilic kinases that phosphorylate sites with clustered positive charges, acidophilic kinases that phosphorylate sites with clustered negative charges, and proline-directed kinases that phosphorylate sites in which Ser/Thr is followed immediately by a proline (i.e. proline at the Pϩ1 position). This classification by peptide specificity corresponds generally with classification of Ser/Thr kinases based on sequence similarity. Based on studies of a sampling of family members, two large families of kinases appear to be largely included in the basophilic group: the 61 gene AGC 1 (PKA, PKG and PKC) family (6 -8) and the 83 gene CAMK (calmodulin-dependent protein kinase) family (5,9,10). Proline-directed kinases such as GSK, CDK, and Erk belong to CMGC (CDK, MAPK, GSK3, CLK kinases) superfamily of kinases, which consists of 61 kinases (6,(11)(12)(13). The acidophilic kinase group is much smaller, including, for example, casein kinase I and casein kinase II. The basophilic kinases and acidophilic kinases would be expected to be largely non-overlapping in peptide specificity; the electrostatic interactions that promote phosphorylations of basic sites by basophilic kinases will create an aversion for acidic sites (and vice versa). Moreover, proline-directed kinases will eschew substrates of the other two which lack proline at Pϩ1.
However, a missing element in the foregoing paradigm is one that prevents basophilic kinases or acidophilic kinases from phosphorylating sites preferred by proline-directed kinases. Such an element may be greater importance for basophilic kinases, since their preference for basic residues is shared in a limited fashion by some proline-directed kinases (e.g. Cdk1, Cdk2, Cdk5, Dyrk, GSK3) (5,6,13,14). We have identified an evolutionary strategy widely used by basophilic kinases that addresses that problem. We find a disfavor for proline at the Pϩ1 position among AGC and CAMK kinases. That disfavor is stronger than for any other residues at positions around the phosphorylation site and is unusually consistent among members of these kinase families. As a result, proline at the Pϩ1 position serves as a "veto residue" for phosphorylation by these kinases; this strategy established by kinase evolution provides tight reciprocal specificity between basophilic and proline-directed kinases.

MATERIALS AND METHODS
Degenerate Peptides and Kinases-Biotinylated single sequence peptides and degenerate peptides were synthesized as C-terminal amides on Mimotopes (Clayton, Australia) SynPhase Rink amide acrylicgrafted polypropylene solid support (loading 7.5 mol) using conventional Fmoc (N-(9-fluorenyl)methoxycarbonyl) chemistry as described previously (15). Seven catalytically active kinases were used. PKC-␦, PKC-, PKA, and PKG were purchased from Calbiochem, and PKD2 and Cdk2 were purchased from Upstate Biotechnology. A 6 ϫ Histagged construct corresponding to residues 615-874 of human PRK1 (GenBank TM accession number BC040061) was expressed in 293T cell by calcium phosphate transfection and the corresponding protein purified by nickel affinity chromotography.
In Vitro Kinase Assay-Peptides were phosphorylated by in vitro kinase assay as described previously (15). In brief in vitro phosphorylation in the presence of ␥-32 P-labeled ATP (in 100 M cold ATP) was performed in 50 l of solution under standardized conditions resulting in stoichiometry of phosphorylation less than 10%. After reaction termination, 50 pmol of substrate was transferred to streptavidin-coated plates and emissions counted after extensive washing. Kinase buffer for PKC-␦, PKC-, PKD2, PKG, PKA, and PRK1 is 100 mM HEPES, 0.05% Triton X-100, 1 mM CaCl 2 , 20 mM MgCl 2 . For kinase PKC-␦, PKC-, and PKD2, 0.2 mg/ml phosphatidylserine (Avanti Polar Lipids) and 100 ng/ml phorbol 12-myristate 13-acetate were added to the kinase buffer immediately before the assay. Kinase buffer for CAMK II is 25 mM HEPES, 10 mM MgCl 2 , 1 mM CaCl 2 , 8 g/ml calmodulin, and 1 mM dithiothreitol, which was minimally modified from Madhavan (34).
Data Analysis-Data were analyzed as described previously (15). All results shown represent at least duplicate assays, and usually quadruplicate assays. Data analysis was largely done in Microsoft Excel spreadsheets. Functionality of embedded spreadsheet formulas was augmented by automation using Microsoft Visual Basic and data storage using Microsoft SQL Server. The PSSM (position-specific scoring matrix) Logo was generated using postscript files generated by Visual Basic code; some postscript code was adapted from Tom Schneider's "makelogo" 8.69 (www.bio.cam.ac.uk/cgi-bin/seqlogo/logo.cgi).

RESULTS
During studies of PKC isoform specificity (15) a strong disfavor for proline at the Pϩ1 position was noted. Given the potential biological importance of Pϩ1 proline disfavor, analysis was expanded to other AGC family members such as PKG (Fig. 1A). Markedly different levels of phosphorylation by PKG are observed between peptides that differ only by a single amino acid at their Pϩ1 position. Since each peptide is a pool of about 1 billion individual peptides, the preference (or disfavor) observed must reflect a preference (or disfavor) of that kinase for many peptides in that huge pool. When Pro is present at the Pϩ1 position, the amount of phosphorylation is much lower than with any other residue. Specifically, phosphorylation of peptide #13 (Pro at Pϩ1) is about 1 ⁄10 of the average for all peptide pools. Thus, Pro must be unfavorable at the Pϩ1 position for most of the phosphorylatable peptides in this degenerate peptide pool #13.
We investigated whether the strong disfavor for Pro at Pϩ1 was a characteristic shared with other kinases. Four additional members of the AGC kinase family (PKA, PKC-␦, PKC-, and FIG. 1. Proline at the position P؉1 is strongly disfavored by five AGC kinases and two CAMK kinases. A, determination of PKG specificity at the Pϩ1 position. Biotinylated peptides were synthesized corresponding to the 13 sequences shown. All peptides share three fixed residues: P0, shown as "-S-" and an R at PϪ2 and PϪ3. The amino acid fixed in the position Pϩ1 systematically varied between peptides. The remaining positions are degenerate ("d"), meaning they are synthesized with a mixture of 19 amino acids at that position. The 13 peptides were subjected to in vitro phosphorylation by PKG resulting in 32 P incorporation. Results are shown as: raw counts (cpm); ratio-to-mean calculated as (cpm)/(geometric mean CPM for all 13 peptides); and log score calculated as log 2 (ratio-to-mean). B, PSSM Logos for the Pϩ1 position of five different AGC kinases and two CAMK kinases. In a PSSM Logo (15), each stack of letters represents preferences of a single kinase for 13 different residues at the Pϩ1 position. The height of each letter is proportional to the absolute value of the residue's log score, and the positions of the letters in the stack are according to the sorted log score in ascending value. The most favored residue is at the top and the most disfavored at the bottom. Residues that are neither favored nor disfavored are largely invisible in the PSSM Logo. Each letter is colored according to the physico-chemical properties of the residue to which it corresponds; basic residues are blue, acidic residues are red, hydrophobic are black, and Pro is yellow. Value are shaded red for peptides having ratio-to-mean greater than 1.5; values are shaded blue for peptides having ratio-to-mean less than 2/3 (i.e. 1/1.5). PRK1) were tested and each showed strong disfavor for Pro at Pϩ1. One way to visualize this comparison of kinases is with a PSSM Logo showing preferences of those kinases at the Pϩ1 position (Fig. 1B). Consider first the results for PKG (Fig. 1B, left column, which corresponds to data shown in Fig. 1A). The parameter represented in the PSSM Logo is log2 of the ratio to mean for each residue; this mathematical conversion gives favored residue positive scores and disfavored residues negative scores. The PSSM Logo shows the residue most favored by PKG at the top (I) and the most disfavored residue (P) at the bottom. The columns corresponding to other kinases show that the strong disfavor for proline at the Pϩ1 position is observed for each of the four other AGC kinases (PKA, PKC-␦, PKC-, and PRK1). To broaden the analysis, we also examined two members of the CAMK family (CAMK II and PKD2); since they share the preference for Arg at PϪ3 (and PϪ2) (5, 6) they can be analyzed with the same sets of degenerate peptides. The results show that CAMK II and PKD2 also have strong disfavor for Pro at Pϩ1 (Fig. 1B).
One useful way to further characterize the proline disfavor at Pϩ1 is to compare it with the preferences for residues in other positions near the phosphorylation site. Such a comparison is shown for the five basophilic kinases studied in further detail (Fig. 2). Consider first the results for PKG (Fig. 2, left  panel). The distribution of scores is approximately centered on zero, and most residue preferences are close to zero. The outlier far to the left corresponds to Pro at Pϩ1; its separation from the rest of the distribution demonstrates that this is a singular disfavor, i.e. distinctly stronger than the disfavor for any residue at any other position studied. At the opposite extreme, there are several outliers in the favored direction; these correspond to Arg and Lys at PϪ2 and PϪ3, as would be expected based on the known very strong PKG preference for those residues in those positions (5). Thus, disfavor for Pro at Pϩ1 position is by far the single strongest disfavored residue for PKG. The frequency distribution of scores for these others kinases revealed that for each of them Pro at Pϩ1 is the most disfavored residue among all residues at all positions tested. Scores for Pro at position Pϩ1 range from Ϫ2.2 to Ϫ3.1; these scores are both more extreme than any other disfavored residue we have studied and more consistent between AGC/CAMK kinases than any other disfavored residue.
The foregoing studies of in vitro peptide phosphorylation indicate that Pro at the Pϩ1 position is a disfavored residue for many kinases of the AGC and CAMK family. If this characteristic is relevant not only to peptides but also to proteins, then the frequency of Pro at the Pϩ1 position in reported substrates should be much less than the frequency of Pro in human proteins generally (6.5%). Such a comparison can be convincing only when a large number of substrates have been reported, as is the case for PKA and PKC. We have accumulated a list of 124 PKC substrates from review of the literature (15); the list was not biased by our hypothesis of Pro disfavor, since it was accumulated before that hypothesis was formulated. Analysis of the number and position of Pro residues in these PKC substrates strongly supports the prediction. The overall frequency of Pro residues in these substrates (from position PϪ8 to Pϩ8) was 4.8% and thus not very different from the proteome at large. Notably the frequency of Pro at Pϩ1 was very low; only 1 of 124 reported substrates had Pro at Pϩ1, which represents a frequency of only 0.8%. It is useful to convert results of residue frequency analysis in substrates into a log score by determining the log base 2 of (observed frequency/expected frequency); that log2 score for Pro at Pϩ1 from reported PKC substrate analysis is Ϫ3.1, which is comparable with our score of Ϫ2.6 derived from degenerate peptide analysis. Analysis was expanded to determine proline favor/disfavor at residue position PϪ7 to Pϩ6 (Fig. 3A). There is generally good agreement between the scores derived from residue frequency in reported substrates and scores derived from degenerate peptides. We performed a similar analysis on PKA sites compiled by Shabb (8); of those 136 were mammalian sites plausibly assigned to PKA based on the references cited. The frequency of Pro at Pϩ1 in these substrates was 1.3%, which is again much lower than expected based on overall frequency in those substrates (5.5%) or human proteins generally (6.5%). As was seen with PKC, the position-specific scores for Pro frequency in substrates are in general agreement with scores from degenerate peptides (Fig.  3B). Thus, there is a very low Pro frequency at the Pϩ1 position in reported PKC and PKA substrates and the Pϩ1 position is the only one at which Pro disfavor occurs in both PKC and PKA.
The potential importance of this disfavor is magnified when understood in the context of the virtual requirement for Pro at Pϩ1 among proline-directed kinases (5,11,12), consisting primarily of the CMGC family of kinases, such as Cdk2. As noted in the introduction, at least five proline-directed kinases have a modest basophilic preference, which creates the potential risk of poor discrimination between basophilic AGC/CAMK kinases and proline-directed kinases. We therefore tested whether the disfavor for Pro at Pϩ1 is an important element in discrimination of AGC/CAMK sites from sites recognized by proline-directed kinases. As a proof of principle we chose "proteomic" peptides corresponding to basic sites in the proteome that our algorithms suggested could be good substrates for basophilic kinases in all respects except for their Pro at Pϩ1. We then synthesized peptides with wildtype sequence (Pro at Pϩ1) or with substitutions to two alternative residues at the Pϩ1 position. One proteomic peptide was particularly informative for comparison of PKC-␦ versus Cdk2 (Fig. 4A). PKC-␦ was efficient in phosphorylating the peptide whose Pϩ1 residues was Phe (PKC-delta PSSM score 1.0, also see Fig. 1B); less efficient with an His at Pϩ1 (PSSM score 0.1, also see Fig. 1B) and almost inactive with a Pro at Pϩ1 (PSSM score Ϫ3.1, also see Fig. 1B). In contrast Cdk2 efficiently phosphorylated only the peptide with Pro at Pϩ1 and neither of the others. Thus, the specificity differences between Cdk2 and PKC-␦ are as much a result of the aversion of PKC-␦ for Pro as the preference of Cdk2 for Pro. Analysis of the second proteomic peptide (and Pϩ1 variations thereof) was particularly informative for comparison of CAMK II with Cdk2 (Fig. 4B). CAMK II can efficiently phosphorylate the peptide with Phe at Pϩ1 but is very inefficient in phosphorylating the peptide with Pro at the Pϩ1 position. Note that phosphorylation of the peptide with Pro at Pϩ1 (the most disfavored residue at Pϩ1 for CAMK II, Fig. 1B) is much less than the peptide having Gln at Pϩ1 (the second most disfavored residue, Fig. 1B); thus Pro has an exceptional capacity to veto phosphorylation by CAMK II. Based on the foregoing studies, we introduce the term "proline-aversive" kinase to describe those with a strong disfavor for Pro at Pϩ1 and thereby highlight the reciprocity of their specificity to that of proline-directed kinases.
The structural basis for the difference between the prolinedirected kinases and proline-aversive kinase can be understood based on differences between the solved structures of prolinedirected kinases with bound peptide (e.g. PDB code 1QMZ) and AGC/CAMK kinases with bound peptide (e.g. PDB code 1ATP). One of the regions of the kinase domain that is most important for substrate recognition region is the activation loop, a flexible region of the kinase that 1) lies along the catalytic cleft, 2) is involved in binding C-terminal residues of substrate, and 3) is subject to extensive conformational regulation (16 -18). In the activation loop there is a critical residue which we refer to as the toggle residue whose conformation differs dramatically between proline-aversive and proline-directed kinases. In the two proline-aversive kinases with solved structure (AGK family kinase PKA and the CAMK member PHK) the toggle residue is a glycine, whose backbone carbonyl is oriented toward the catalytic cleft (Gly 200 (G200, Fig. 5, A and B); that carbonyl facilitates binding of substrate by the formation of a H-bond with the backbone amide of the substrate Pϩ1 residue (19,20). Substrates having Pro at Pϩ1 cannot form that H-bond because the proline amide is not an H-bond donor. In contrast, in proline-directed kinases (Erk2, Cdk2, GSK3) the carbonyl group of their toggle residue is oriented away from the catalytic cleft (21)(22)(23)(24)(25) (Val 164 (V164), Fig. 5, C and 5D). So binding of all residues except Pro at Pϩ1 position will be disfavored because of an uncompensated H-bond from the main-chain Pϩ1 amide of the substrate (23,24).
The orientation of the toggle residue in proline-directed kinases is enforced by a strategically placed arginine (Arg 169 (R169), Fig. 5D), which we refer to as a toggle-regulating residue. That arginine is located in the Pϩ1 loop, which is a loop adjacent to the activation loop (that is firmly attached to the stable C-lobe of the kinase domain). That arginine forms an H-bond with the carbonyl of the toggle residue, thereby maintaining the usual conformation of that toggle residue (23, 24) (Fig. 5D). In the foregoing model, the single residue most pivotal to creating the two distinct conformations of the toggle residue is this toggle-regulating arginine. Sequence analysis of human kinases reveals a striking pattern of conservation of the toggle-regulating residue (Table I) reported PKC substrates was determined and converted into a score for each substrate position from PϪ7 to Pϩ6; the "Log2 O/E" value that is plotted is the Log base 2 of (observed Pro frequency)/(expected Pro frequency of 6.5%). This score (on the x axis) was compared with the score for proline at the position derived from analysis of degenerate peptides with PKC-␦ (y axis). B, comparison of Pro disfavor in reported PKA substrates with Pro disfavor determined from degenerate peptide analysis of 136 PKA substrates from position PϪ7 to Pϩ5 (8).
FIG. 4. Pro at the position P؉1 is a veto residue for AGC and CAMK kinases. A, in vitro phosphorylation by PKC-␦ and Cdk2 of peptides having or lacking proline at Pϩ1. Peptides were synthesized that correspond to a peptide from LOXL1 residue 83-99 (AQQRRSHG-S-PRRRQAPS) and two variations thereof that differed in the Pϩ1 position: His (H), predicted to be neither favored nor disfavored by PKC-delta; Phe (F), predicted to be favored by PKC-␦. x axis represents substrate concentration and y axis is measured phosphorylation (as cpm). B, in vitro phosphorylation by CAMK II and Cdk2 of peptides having or lacking proline at Pϩ1. Peptides were synthesized that correspond to a peptide from SFRS2 residue 83-99 (AQQRRSHG-S-PRRRQAPS) and two variations thereof that differed in the Pϩ1 position: Gln (Q), predicted to be the second most disfavored residue by CAMK II; Phe (F), predicted to be favored by CAMK II. 324 kinases in the AGC, CAMK, STE, TK, and TKL families have arginine at that position; instead they have acyclic hydrophobic residues, which cannot provide an H-bond to the carbonyl of the toggle residue. Therefore, this amino acid position has the property of absolute conservation within the CMGC family and complete discordance from AGC/CAMK and other kinases. Recently Kannan and Neuwald (26) noted that this arginine residue, which they refer to as the "CMGC-arginine," is the most characteristic feature of the CMGC kinase family.
Analysis of sequence conservation of the toggle residue of the proline-aversive kinases is likewise informative. Nearly all CAMK and AGC kinases have A or G at this position (Table I). This makes sense from a structural perspective. Given the importance outlined above for hydrogen bond formation between the carbonyl group of the toggle residue and the Pϩ1 backbone amide, it is essential that toggle residue have sufficient conformational flexibility to engage the backbone amide. This flexibility would be limited by spatial constraints imposed by a substantial side chain. To test these structural hypotheses we have mutationally swapped the toggle residue and toggleregulating residues between Cdk2 and PKC-␦; however, both constructs are catalytically inactive (data not shown), indicating that these residues function normally only in the context of other kinase family-related residues. Nevertheless, these patterns of conservation of the toggle residue and toggle-regulating residue are fully consistent with their critical role in establishing a major difference between CMGC kinase and AGC/ CAMK kinases. DISCUSSION These results demonstrate that Pro at the Pϩ1 position is a veto residue which virtually precludes phosphorylation by multiple AGC and CAMK kinases. As a result, Pro at Pϩ1 provides tight control of reciprocal substrate specificity between prolinedirected kinases and AGC/CAMK kinases. These findings for peptides are also relevant to proteins, since we find that Pro   Toggle-regulating  ILMV  0  95  59  98  65  Ala  0  2  30  0  3  Thr  0  0  1  0  23  Arg  100  0  0  0  0   Toggle  AG  21  95  87  94  26  Gln  20  0  0  2  0  FILMV  46  3  4  0  61  No. of kinases  61  63  81  48  132 occurs at a much reduced frequency at the Pϩ1 position among reported substrates for PKA and PKC. A structural explanation for this disfavor for proline is evident in crystal structures of AGC/CAMK in which a carbonyl group is strategically oriented toward the catalytic cleft. The ability of all substrate Pϩ1 residues except proline to provide a backbone amide for Hbonding to that carbonyl gives such kinases a singular disfavor for proline. The foregoing analysis focuses on substrate binding but does not explicitly consider substrate positioning. A cardinal feature identified by structure/function analysis of protein kinases is that proper orientation of the activation loop and peptide is essential for catalysis (16 -18). For proline-directed kinases substrate positioning is provided by docking of proline into a proline-selective pocket (Fig. 5C). Remarkably those residues (Val 63 and Glu 162 ) form this proline-selective pocket only when the activation loop is phosphorylated at Thr 160 and the kinase domain is bound to cyclin A; in the unphosphorylated inactive complex the conformations of these two residues are very different, and there is no such tight and specific docking site for proline (27). For proline-aversive kinases positioning is provided in a different manner; the H-bond between activation loop and the Pϩ1 backbone amide positions the substrate into an orientation suitable for phosphorylation (Fig. 5A). This Hbond also stabilizes the Pϩ1 residue in a preferred orientation in which its side chain is oriented into the hydrophobic Pϩ1 pocket of PKA; this orientation is observed with isoleucine at Pϩ1 in 1ATP and is consistent with the preference for hydrophobic residues at the Pϩ1 position for PKA (6). But in the special case of peptides with proline at Pϩ1, this positioning mechanism cannot operate for PKA (or other CAMK/AGC kinase). We hypothesize that proline at Pϩ1, since it lacks the orientation/constraint of a H-bond, can find another binding site(s) in the wide hydrophobic pocket formed by Leu 205 , Leu 198 , or even Phe 187 (Figs. 5A). Improper binding of the proline residue can be quite stable but fails to result in catalysis because of incorrect positioning of the substrate (and phosphorylatable residue) in the catalytic cleft or conformation changes resulting from ectopic binding of proline. Extensive in silico and experimental analysis will be required to properly test this hypothesis.
Proline is unique among natural amino acids in having a substituted amide and resulting conformational rigidity. Those properties make it uniquely suited for particular structural roles. For example, it is an important element in the peptide sequences recognized by proline-specific binding domain such as SH3, WW, and EVH1 domains (28 -30). The role of proline in favoring recognition by proline-directed kinases has been well established but not its reciprocal role as a veto residue at Pϩ1 for basophilic kinases. Prior to the present study, a disfavor for proline at Pϩ1 was noted for PKA (6) and ␥-PAK (31). The proline disfavor of ␥-PAK is particularly interesting since it is a basophilic kinase that belongs to a distinct STE subfamily of kinase (3). It suggests that disfavor for proline at Pϩ1 extends beyond the AGC and CAMK families, as is suggested by our sequence analysis tabulated in Table I.
The present findings raise the possibility that disfavor for proline at other substrate positions may also contribute to fine peptide specificity of kinases. For example, PKC has a moderate disfavor for proline at the Pϩ2 and Pϩ3 positions evident both in our peptide specificity determination and in analyses of reported substrates (Fig. 3A). In contrast PKA shows disfavor for proline at PϪ2 by the same two criteria (Fig. 3B). Moreover, a strong disfavor for proline at PϪ4 has been reported for a plant kinase belonging to the SNRK subfamily of CAMK (32). These findings indicate that proline residues at substrate po-sitions other than Pϩ1 can contribute to fine peptide specificity differences among basophilic kinases. Detailed structural understanding of such preferences is not available; however, it is plausible that PKCs disfavor for prolines at Pϩ2 and Pϩ3 in substrates reflects disruption of anti-parallel ␤ sheet formation between activation loop and substrate positions Pϩ1 to Pϩ3 that has been observed in the Ser/Thr kinase PHK (20) and the tyrosine kinase IRK (33).
These results highlight the importance of disfavored residues in kinase peptide specificity. They demonstrate the singular importance of Pro at the Pϩ1 position as a veto residue for diverse kinases of the AGC and CAMK families. The reciprocity of this pattern to proline-directed CMGC kinases constitutes a powerful evolutionary strategy to assure non-overlapping phosphorylation by these large and important families of kinases.