Research ArticleBiochemistry

Engineering allosteric regulation in protein kinases

See allHide authors and affiliations

Science Signaling  06 Nov 2018:
Vol. 11, Issue 555, eaar3250
DOI: 10.1126/scisignal.aar3250

Hotspots of kinase regulation

The addition of phosphate moieties can alter protein conformation, function, abundance, and localization, thereby regulating a myriad of cellular behaviors and responses. These sites are often mutated in cancer and other diseases. Pincus et al. mapped the residue alignment and patterning of phosphoregulatory sites in eukaryotic kinases and determined that kinases share a conserved architecture, such that introducing a target motif of the kinase PKA at surface sites within two different yeast kinases altered their activity, pathway interactions, and subcellular localization, as well as the overall response of the yeast to pheromone and osmotic signals, in a PKA-dependent manner. These findings, which reveal insights into kinase evolution, also have implications for biological engineering and medicine.


Phosphoregulation, in which the addition of a negatively charged phosphate group modulates protein activity, enables dynamic cellular responses. To understand how new phosphoregulation might be acquired, we mutationally scanned the surface of a prototypical yeast kinase (Kss1) to identify potential regulatory sites. The data revealed a set of spatially distributed “hotspots” that might have coevolved with the active site and preferentially modulated kinase activity. By engineering simple consensus phosphorylation sites at these hotspots, we rewired cell signaling in yeast. Using the same approach with a homolog yeast mitogen-activated protein kinase, Hog1, we introduced new phosphoregulation that modified its localization and signaling dynamics. Beyond revealing potential use in synthetic biology, our findings suggest that the identified hotspots contribute to the diversity of natural allosteric regulatory mechanisms in the eukaryotic kinome and, given that some are mutated in cancers, understanding these hotspots may have clinical relevance to human disease.


Phosphoregulation provides a dynamic and reversible means for the regulation of proteins. In allosteric phosphoregulation, the addition of a phosphate group at a site distinct from the active site is able to modulate protein activity. Introducing such regulation (by either engineering or evolution) would seem to require satisfaction of two main properties: First, like any form of allostery, allosteric phosphoregulation requires the cooperative action of multiple amino acids to functionally link the phosphorylated site to a spatially distinct active site. Second, the addition of a phosphate group has to somehow engage or activate this underlying cooperative network. Regarding the former, several lines of work indicate that proteins have a latent capacity for allosteric regulation at a diversity of surfaces. For example, it is possible to engineer synthetic allosteric switches through domain insertion at certain surface sites (15), and screens for small molecules that modify protein function sometimes identify cryptic allosteric regulatory sites (6, 7). In addition, experimental analysis of regulation in orthologs of the yeast mitogen-activated protein kinase (MAPK) Fus3 indicates that the capacity for allosteric regulation existed well before the regulatory mechanism evolved (8). Together, these findings suggest that proteins have an internal architecture in which multiple sites on the protein surface are functionally “prewired” to provide control of protein activity and that these sites could serve as hotspots for the introduction of new regulation (5).

The question then becomes how placing a phosphate at one of these surfaces might engage the underlying allostery. Previous work from Ferrell and colleagues provides a potential solution: Phosphoregulation might evolve simply by mutating an allosterically precoupled negatively charged residue (Asp/Glu) to a phosphorylatable residue (Ser/Thr/Tyr) (9). Thus, a constitutive negative charge at a latent allosteric site can be transformed into a regulated negative charge in a potentially stepwise manner (10). Here, we experimentally tested the proposal that new phosphoregulation can be introduced at negatively charged surface sites and conducted sequence analyses to understand what properties distinguish sites with regulatory potential.


Eukaryotic protein kinases are regulated at a diversity of sites

We felt that an excellent model to test this proposal is the family of eukaryotic protein kinases, a family that has diversified to control a vast array of cellular signaling activities. Eukaryotic protein kinases themselves catalyze the transfer of a phosphate group from adenosine triphosphate (ATP) onto a Ser/Thr/Tyr residue of a substrate protein and are regulated by different mechanisms at distinct surface regions. To illustrate this, we mapped known regulatory sites from a diversity of kinases to a single representative kinase structure (Fig. 1, A and B). Sites for regulation are distributed across the kinase surface and mechanistically include protein-protein interactions, autoinhibition, dimerization, and posttranslational modification (11). This indicates that despite the complex intramolecular cooperativity required, allostery evolves readily at multiple distinct locations in the kinases (12). The diversity of regulation that has evolved across the kinome suggests the possibility that individual kinases might harbor a latent capacity for regulation at many surfaces.

Fig. 1 Regulatory diversity in the eukaryotic protein kinases.

(A) Unanchored dendrogram of the human kinome, illustrating the diversity of the eukaryotic protein kinase superfamily and subfamilies. Individual subfamily members with functional mutations shown in a subsequent figure and included in table S7 are listed. TK, tyrosine kinase; TKL, TK like; STE, STE7/11/20; CK1, casein kinase 1; AGC, protein kinase A/G/C; CAMK, calmodulin kinase; CMGC, cyclin-dependent kinase (CDK)/MAPK/glycogen synthase kinase (GSK)/CDK-like kinase (CLK). (B) Allosteric regulatory sites from diverse kinases mapped to a single representative structure: yeast CDK Pho85 [Protein Data Bank (PDB): 2PK9, shown as space-filled surface]. Regulatory surfaces were identified by structural alignment of the kinase of interest to Pho85; all Pho85 positions within 4 Å of the interaction surface are colored. Color coding is the same as in (A). Bright yellow spheres indicate known phosphoregulatory sites.

An alanine scan of negatively charged, solvent-accessible residues identifies functionally coupled positions distributed across the surface of Kss1

We chose the prototypical yeast CMGC kinase Kss1 as our experimental model (text S1). Kss1 is a homolog of human extracellular signal–regulated kinase (ERK) and is involved in signal transduction pathways that regulate yeast filamentous growth and the mating response (1316). Kss1 activity can be quantitatively monitored in living yeast cells by its ability to specifically activate fluorescent transcriptional reporters of the mating pheromone response in the absence of its paralog, Fus3 (Fig. 2A). We conducted an alanine scan of all 40 Asp/Glu residues on the surface of Kss1 to determine which residues (hereafter, positions) are functionally coupled to kinase activity. We integrated the resulting 40 Kss1 mutants as the only copy of Kss1 in the yeast genome, tagged at their C terminus with a 3×FLAG epitope (tables S1 and S2). To test their activity, we assayed for induction of the pheromone-responsive AGA1pr-YFP reporter at four concentrations of the α-factor (αF) mating pheromone by flow cytometry. Although all mutants maintained wild-type–like amounts of expression, nine mutations altered kinase activity in yeast (Fig. 2, B and C, and fig. S1, A and B). Three of these positions were identified as Kss1 mutants with a functional effect in previous studies (Asp117, Asp156, and Asp321; table S3) (17, 18). Although enriched in the N-terminal half of the primary Kss1 sequence, these nine mutations occur at positions distributed broadly over the Kss1 atomic structure—consistent with the notion that multiple, specific surface sites are prewired to allosterically influence active site function.

Fig. 2 Alanine scan of acidic residues on the solvent-accessible surface of yeast MAPK Kss1.

(A) Schematic of the Kss1-dependent yeast pheromone pathway. The αF mating pheromone binds to a G protein–coupled receptor (GPCR), leading to activation of a signaling cascade culminating at the MAPK Kss1. Kss1 then activates the Ste12 transcription factor to induce the mating transcriptional program, which can be monitored by fusing the promoter of the target gene AGA1 to a yellow fluorescent protein (YFP) reporter. (B) Ribbon diagram of a Kss1 homology model (30) with the 40 solvent-accessible Asp/Glu residues shown as spheres. All 40 positions were mutated individually to alanine to remove negative charge. The DFG motif and activation loop are indicated in light blue. (C) The 40 resulting yeast strains along with wild-type (WT) and kss1Δ controls were assayed for activation of the AGA1pr-YFP reporter by flow cytometry following treatment with 0, 0.01, 0.1, and 1 μM αF for 4 hours. Bars represent the average of the median YFP fluorescence from three biological replicates normalized to the untreated kss1Δ cells, and error bars are the SD of the biological replicates. Mutations at red and green positions resulted in significantly reduced or increased YFP expression [P < 0.05, as scored by one-way analysis of variance (ANOVA)] in response to at least two doses of αF, respectively. Yellow positions indicate that the mutation had no effect in this assay. The color coding is identical in (B).

Negatively charged and functionally coupled surface sites can be used to engineer new regulation of Kss1 activity

We next tested whether these positions can support new regulation of Kss1 through phosphorylation by another yeast kinase in vivo. In principle, this gain of function can effectively rewire signaling through the mating response pathway. We chose to engineer regulation of Kss1 by protein kinase A (PKA) because the PKA substrate consensus motif, RRxS/T, requires minimal local modifications, its activity in yeast cells is orthogonal to the pheromone pathway, and it can be hyperactivated in yeast via ectopic expression of Ras2G19V (mutant Ras-like protein 2 precursor; Fig. 3A and fig. S2, A and B). We selected three of the nine mutationally sensitive positions (Asp8, Glu68, and Glu70), with the highest PKA substrate scores predicted by the computational tool pkaPS (19). To claim PKA-mediated allosteric regulation of Kss1, one must demonstrate (i) that Kss1 retains functionality after introduction of a local PKA consensus motif (RRxD or RRxE, termed pka-D and pka-E, respectively), (ii) that Kss1 loses activity when the charge is neutralized (RRxA, termed pka-A), and (iii) that Kss1 displays PKA-dependent activity in yeast upon introduction of a phosphorylatable residue (RRxS, termed pka-S) (fig. S2C). In this manner, a functional surface-negative charged residue might neutrally acquire a substrate consensus sequence for a kinase and become a phosphoregulatory site with one step of variation.

Fig. 3 Engineering allosteric control of Kss1 by PKA phosphorylation.

(A) Cartoon of the engineered PKA- and Kss1-dependent yeast pheromone pathway. In this schematic, Kss1 activation requires both activation loop phosphorylation by the upstream MAP2K Ste7 and phosphorylation by PKA at an allosterically coupled surface. To experimentally increase PKA activity, expression of constitutively activated RAS2G19V is induced by addition of estradiol, which, in turn, activates adenylate cyclase (AC) to generate cyclic adenosine monophosphate (cAMP) from ATP to activate PKA. (B) Kss1 mutants with PKA phosphorylation site consensus motifs introduced near position 8 (pka-X8, top) or position 70 (pka-X70, bottom) were assayed for expression of the AGA1pr-YFP reporter by flow cytometry after treatment with αF (0, 0.01, 0.1, or 1 μM) for 4 hours. Again, bars represent the average of the median YFP fluorescence from three biological replicates normalized to the untreated kss1Δ cells, and error bars are the SD of the biological replicates. “X” stands for the amino acid at position 8 or 70 as denoted under the bar graphs. The images below the bar graphs show morphology and expression of the AGA1pr-YFP reporter in yeast cells bearing the indicated Kss1 mutants in the presence of 1 μM αF following growth in the presence or absence of 20 nM estradiol. The percent cells shmooing is indicated and was quantified by counting cells in three 512 × 512 pixel frames in each condition. n > 40 cells for each sample. (C) 3×FLAG-tagged wild-type Kss1 and pka-X8 and pka-X70 mutants were immunoprecipitated from untreated cells or cells that had been treated with both 20 nM estradiol and 1 μM αF. IP eluates were analyzed by Western blotting for total Kss1 and Kss1 phosphorylated on its activation loop (Phospho act. loop) or at the engineered PKA site (Phospho pka site). Merged images show that all mutants can be phosphorylated on their activation loop in the presence of αF, but only pka-S8 and pka-S70 can be phosphorylated by PKA in the presence of estradiol.

Introduction of pka-E at position 68 (replacing positions 65 to 68, the sequence YFHE, with RRHE) resulted in Kss1 loss of function (fig. S2, D and E). This indicates that in this instance, the mutation of positions 65 to 66 to arginine to introduce the PKA site was not neutral. However, introducing the PKA consensus motif at positions 8 and 70 (in place of positions 5 to 8 and positions 67 to 70, respectively) showed the complete expected pattern of activity for gain of phosphoregulation. For both sites, introduction of the two arginine residues upstream was near neutral, mutation of the negatively charged residue caused loss of function, and Kss1 pka-S activity depended on enhanced PKA activity via estradiol-induced expression of Ras2G19V (Fig. 3B). Immunoprecipitation (IP) of the 3×FLAG-tagged Kss1 mutants followed by Western blot analysis supported this finding. Both the pka-A and pka-S variants displayed activation loop phosphorylation when treated with αF, indicating that they remain substrates of the upstream MAP2K Ste7. However, only Kss1–pka-S8 and Kss1–pka-S70 were recognized by an antibody specific for phosphorylated PKA substrates when purified from cells treated with estradiol (Fig. 3C). Moreover, both Kss1–pka-S8 and Kss1–pka-S70 were able to induce the morphological response to pheromone—the mating projection known as the “shmoo”—in an αF- and PKA activity–dependent fashion (Fig. 3B). Thus, the transcriptional and physiological outputs of Kss1 can be rewired to depend on an orthogonal input by a stepwise process of introducing a phosphorylation site at latent allosteric surface sites.

Functionally coupled Kss1 positions are associated with a network of conserved, coevolving amino acids in the kinases

What distinguishes the nine surface, negatively charged amino acids that are allosterically coupled to Kss1 activity? Are the functionally coupled positions idiosyncratic to Kss1 or conserved across the kinase family? To address these questions, we examined the relationship between our mutational data and the pattern of amino acid conservation and coevolution in protein kinases. The motivation is that conservation should provide an indication of functionally important positions, and coevolution can provide insights into the pattern of coupling or interaction between amino acids. More specifically, previous work using an approach called statistical coupling analysis (SCA) showed that coevolving networks of amino acids form sparse yet physically contiguous networks in the protein tertiary structure, termed “sectors.” Sectors tend to link protein active sites to distantly positioned allosteric surface sites and are proposed to represent a cooperative physical mechanism embedded within the protein (2024). Experiments demonstrated that sector-connected surface sites are “hotspots” for introducing new allosteric regulation by domain insertion, suggesting that the sector endows particular locations on the protein surface with a latent capacity for allosteric regulation (5, 25). Following on from this previous work, we (i) hypothesize that multiple sites on the protein kinase surface are sector connected and allosterically coupled, (ii) expect that the precise subset of coupled sites at which regulation is achieved will depend on details of the perturbation (choice of mutation or regulation introduced), and (iii) propose that functional mutations will be found with strong statistical preference at sector-connected surface sites. Thus, we wanted to test whether the functionally important Asp and Glu (D/E) residues identified in Kss1 had a statistically significant association to coevolving and/or conserved kinase positions. If so, it would suggest that these regulatory sites not only are relevant for Kss1 but also represent conserved allosteric hotspots across the CMGC kinases or even the entire kinome.

We applied SCA to two multiple sequence alignments of the kinase catalytic domain: one encompassing all eukaryotic protein kinase subfamilies and a second focused alignment of the CMGC subfamily that includes the MAPKs (Fig. 4, A and B). Using SCA, we computed the conservation of individual positions (as a Kullback-Leibler relative entropy, Di) and an estimate of coevolution between position pairs (as a conservation-weighted covariance matrix). The resulting pattern of coevolution was analyzed using standard approaches from linear algebra to identify the sector (described further in Materials and Methods). Regardless of alignment choice, the sector defined for the kinases includes ~30% of amino acid positions and forms a physically contiguous network within the three-dimensional structure (Fig. 4, C to E, and table S4). The sector defined for the CMGC alignment includes the MAPK insert, a feature that is not conserved across the kinome-wide alignment (Fig. 4C). Because the SCA calculation of coevolution emphasizes correlations at conserved positions, we observed a strong overlap between individually conserved amino acids and coevolving sector positions (Fig. 4D; see also text S1 for additional discussion).

Fig. 4 SCA of the eukaryotic protein kinases.

The analysis was performed for two different multiple sequence alignments of the kinase catalytic domain: one specific to the CMGC kinases (635 sequences) and one containing 7128 kinases sampled across the kinome. (A) Histogram showing the distribution of pairwise sequence identities computed across all pairs of sequences in the CMGC alignment. (B) As described in (A) but for the kinome-wide alignment. Both alignments show a unimodal distribution with a mean pairwise sequence identity near ~25%. (C) Sector positions derived from the CMGC alignment (blue) or kinome-wide alignment (yellow) are distributed along the primary and secondary structure of the CMGC/MAPK ERK2. Subfamily-specific regions, such as the MAPK insert, are only part of the sector derived from the CMGC alignment. (D) Relationship between the sector and positional conservation (computed as the Kullback-Leibler relative entropy, Di) for both the CMGC and kinome-wide alignments. Sector positions are highlighted in blue or yellow for the CMGC and kinome-wide alignments, respectively. Red stars indicate highly conserved positions (defined as Di > 2.0 in the kinome-wide alignment). (E) Kinome-wide and CMGC-specific sectors (yellow and blue transparent surfaces, respectively) mapped on human ERK2 (gray ribbon) (PDB: 2ERK). Conserved positions are shown as red spheres.

Consistent with expectation, both the sector and conserved amino acids were enriched for positions associated with kinase function. The sector encapsulated several structural motifs well known to be associated with kinase activation including the αC-helix, the DFG motif, and the catalytic and regulatory spines (Fig. 4, C to E, and fig. S3, A to C) (26, 27). Further, comparison to a deep mutational scan of human ERK2 (28) showed a clear, statistically significant association between sites associated with loss of function and either the sector or conserved positions (fig. S4, A and B, and table S5). Thus, like for other proteins, analysis of conserved coevolution of amino acids in eukaryotic protein kinases provides a sparse, distributed model for the functionally relevant energetic connectivity of amino acids (29).

Although there is no crystallographic structure for Kss1, an available homology model permits comparison of our mutational data to the structural distribution of sector and conserved positions (30). Kss1 D/E mutations with a functional effect showed a statistically significant association with conserved and coevolving positions (Fig. 5, A to C; fig. S5, A and B; and table S6). Eight of the nine functionally coupled surface D/E residues in Kss1 were observed to contact the coevolving sector, including the two that yield new PKA-dependent phosphoregulation (Asp8 and Glu70). Position 8 is located in the N-terminal loop and is proximal to two potential hydrogen-bonding partners within the sector: Lys43 and Tyr91 (Fig. 5, D and E). Both Lys43 and Tyr91 were moderately conserved in our alignments (46% of kinases in the CMGC alignment contain K at position 43; 37% contain Y at position 91). Mutation at a position homologous to Kss1 Tyr91 in PKCβ (Y417H) is known to reduce kinase activity and is associated with liver cancer, indicating that this region is functionally important in other kinase family members (31). The second PKA regulatory site—position Glu70—was found to contact the sector, faces outward from the surface of the kinase, and appears to interact with Arg129 (Fig. 5, D and F). The P124S mutation in MEK2 (MAPK kinase 2) (homologous to Glu70 in Kss1) is a gain of function and is associated with melanoma, also suggesting a functional role for this region in another kinase (32). Together, these data indicate that the gain of new regulatory function in Kss1 occurs at sites that are not idiosyncratic but that interact with an allosteric network that coevolves across the kinase family.

Fig. 5 Association of Kss1 D/E mutations with conserved and coevolving positions.

(A) Space-filling diagram of a Kss1 homology model (30). The CMGC sector, defined as positions that coevolve across the CMGC kinases, is indicated in blue. Acidic surface residues with a neutral, activating, or inactivating effect on kinase function upon mutation to alanine are shown as yellow, green, or red spheres, respectively. (B) Fisher’s exact table demonstrating statistically significant enrichment of acidic surface residues with a functional effect upon mutation at sector-connected positions. To be sector connected, a position must have at least one atom within 4 Å of the sector. Both a P value and estimated false discovery rate (FDR) are indicated. (C) Same as described in (B), but considering conservation-connected rather than sector-connected positions. The cutoff for conservation was chosen to give a similar number of positions as the sector. A detailed analysis of the effect of conservation and sector cutoffs is in the supplement (text S1, tables S5, S6, and S8). (D) Homology model of Kss1 illustrating the relationship of positions Asp8, Glu68, and Glu70 (cyan sticks) to the sector (blue surface). The kinase backbone is in gray cartoon; the activation loop and DFG active site motif are colored yellow. Residues proximal to Asp8, Glu68, and Glu70 are shown in gray sticks. (E and F) Close-up view of the region surrounding positions 8 (E) and 68 to 70 (F). Color coding as in (D).

Introduction of phosphorylation sites at sector-connected surfaces on Hog1 can alter signaling dynamics

If the sector reflects a conserved allosteric mechanism within the protein kinases, sector-connected surface sites should be hotspots for the engineering of regulation in other kinase homologs. Further, in addition to simple phosphorylation-mediated restoration of function, we hypothesized that introducing phosphorylation sites at sector-connected surface positions could allow cells to evolve more complex regulatory schemes. For example, phosphorylation could tune the magnitude and dynamics of kinase activation rather than serving as an ON/OFF switch. To test these ideas, we extended our experimental analysis to a kinase that involves an additional layer of regulation: Hog1. Like Kss1, Hog1 is a yeast MAPK (homolog of human p38 MAPK), but unlike Kss1, it transiently relocalizes from the cytosol to the nucleus upon activation by hyperosmotic stress, presenting the opportunity to modulate its localization and activation dynamics (33, 34).

To attempt to introduce novel functional phosphoregulation, we first identified sector-connected D/E surface residues in Hog1, reasoning that these sites are likely to be enriched for latent regulatory potential and can tolerate negative charge. We directly inserted pka-S motifs (RRxS) at all seven positions that satisfied these requirements and integrated the Hog1-pka mutants as the only copy of Hog1 in the cell (Fig. 6A). To assay for Hog1 activity, we treated cells with 0.5 M NaCl for 90 min in the presence or absence of enhanced PKA activity (by estradiol-induced Ras2G19V expression) and measured induction of HOR2pr-GFP (green fluorescent protein), a transcriptional reporter of Hog1 activity, by flow cytometry (Fig. 6B). Six of the seven Hog1-pka mutants showed altered activity, with three showing constitutive loss of function and three showing PKA-dependent effects (pka-S80, pka-S274, and pka-S277; Fig. 6C). Hog1–pka-S274 behaved similarly to Kss1–pka-S8 and Kss1–pka-S70 in that it displayed diminished responsiveness that was rescued by synthetic activation of PKA. By contrast, Hog1–pka-S80 and Hog1–pka-S277 showed the opposite PKA-dependent behavior: Increased PKA activity resulted in reduced activation (Fig. 6C).

Fig. 6 Modulation of Hog1 activation dynamics by introduction of PKA phosphorylation sites at sector edges.

(A) Sector-connected D/E residues are indicated on a homology model of Hog1. (B) Schematic of the engineered PKA/Hog1 pathway. Note that the Sln1/Ypd1 branch of the endogenous HOG pathway is omitted for clarity. (C) The seven Hog1-pka yeast strains along with wild-type and hog1Δ controls were grown in the presence and absence of 20 nM estradiol and assayed for activation of the HOR2pr-GFP reporter by flow cytometry after treatment with 0.5 M NaCl for 2 hours. Data are the means of the median GFP fluorescence ± SD from three biological replicates normalized to the untreated hog1Δ cells. **P < 0.01 by one-way ANOVA. (D) Immunoblots probed with antibodies against FLAG and phospho-p38 to show total amounts of the Hog1-pka mutants and their respective abundance of activation loop phosphorylation under basal conditions and after treatment with 0.5 M NaCl for 15 min. (E) Quantification of Hog1-mKate nuclear localization dynamics in wild-type and Hog1-pka mutants. Cells were imaged under basal conditions and after treatment with 0.5 M NaCl for 5, 20, or 60 min. n > 50 cells for all mutants at all time points.

Immunoblot analysis of 3×FLAG-tagged Hog1-pka mutants demonstrated that all proteins are stably expressed in cells but differentially phosphorylated on their activation loops in response to NaCl as determined by an anti–phospho-p38 antibody (Fig. 6D and fig. S6A). Activation loop phosphorylation was not perfectly correlated with the transcriptional reporter, suggesting that these normally concerted processes have been decoupled. Moreover, although hog1Δ showed complete growth inhibition in the presence of NaCl, none of the PKA-dependent mutants showed any differences in proliferation compared to wild type (fig. S6B). This indicates that Hog1 has retained its essential function in all of these mutants despite their altered patterns of activation loop phosphorylation and transcriptional output. Thus, as in the case of Kss1, the sector-connected surface sites are highly functional. However, unlike Kss1, PKA activity enabled both activation and inhibition of Hog1-dependent transcriptional induction depending on the location of the PKA consensus motif.

To test whether these PKA-dependent differences in Hog1 activity could be due to altered signaling dynamics, we fluorescently tagged wild-type Hog1 and the three Hog1-pka mutants that showed PKA-dependent effects (pka-S80, pka-S274, and pka-S277). In the absence of estradiol, all mutants and wild type localized to the nucleus upon acute activation with NaCl (fig. S6C). In the presence of estradiol, each of the three PKA-dependent mutants showed distinct, altered localization patterns (Fig. 6E). Despite not showing constitutive activation of the transcriptional reporter, Hog1–pka-S80 was enriched in the nucleus in ~60% of cells under basal conditions and then showed persistent nuclear localization upon treatment with NaCl in ~80% of cells for the remainder of the 60-min time course. By contrast, Hog1–pka-S277 showed a muted nuclear localization response, maxing out at less than 40% of cells with nuclear enrichment and receding with wild-type–like kinetics. Hog1–pka-S274 translocated into the nucleus with kinetics similar to wild type but remained enriched in the nucleus longer than wild type. Together, these results demonstrate that addition of phosphorylation sites at sector-connected positions can functionally modulate Hog1 signaling dynamics to generate novel regulatory schemes.

Functional mutations sampled across the kinome associate with the sector

The data for Kss1 and Hog1 support a model that new regulation preferentially emerges in proteins at surface sites that are evolutionarily conserved across protein families. If so, all natural kinases should follow the principle that functionally sensitive and physiologically relevant allosteric sites, regardless of mechanism, should be found with statistical preference at conserved and sector-connected surfaces. The sector-connected surfaces would then provide an explanation for the diversity of regulatory sites observed in extant kinases (Fig. 1). To investigate this, we constructed a curated database of mutations sampled across the kinase superfamily (those listed in Fig. 1A and table S7). These mutations were selected because they were experimentally demonstrated to disrupt kinase regulation and/or function and, in many cases, are also associated with disease. Comparison of these mutations to the sector revealed a clear pattern: The mutations cluster around conserved positions and the sector edges with strong statistical preference (Fig. 7, A to C, and table S8). These data support the idea that the kinases share a conserved allosteric architecture that allows for information transmission to the protein active site (Fig. 7D).

Fig. 7 Sector-connected surface sites are associated with function in multiple kinases.

(A) The eukaryotic protein kinase superfamily-wide sector (blue spheres) mapped to the CMGC yeast kinase Pho85 (PDB: 2PK9, gray cartoon and surface). Red positions are sites collected from the literature known to alter kinase function when mutated in a functional study or human disease context (see table S7). (B) Fisher’s exact table demonstrating statistically significant enrichment of the functional mutations shown in (A) at sector-connected positions. (C) Fisher’s exact table demonstrating statistically significant enrichment of the functional mutations shown in (A) at conservation-connected positions. (D) Model for the evolution of regulatory diversity. Latent allosteric sites distributed across the protein surface (red circles) are connected to the active site via a protein sector (blue arrows). These sites are poised for the acquisition of new regulation via evolutionary, disease, or engineering processes. In any particular family member, only a subset of sites may be used, and the regulatory mechanism need not be conserved across homologs. PTM, posttranslational modification.


We showed that negatively charged, functionally coupled surface residues in Kss1 can be converted to sites of phosphoregulation by introducing a PKA consensus motif. Further, Kss1 positions with regulatory potential were statistically associated with the sector. This supports the proposal of Ferrell and colleagues that negatively charged surface positions are poised to become sites of phosphoregulation (9), and synthesizes this idea with work indicating that cooperative networks of amino acids within a protein endow certain sites with latent allosteric potential (5, 35). Moreover, in the case of Hog1, we found that more complex regulatory mechanisms can be instantiated by phosphorylation of sites at sector edges. Overall, these results suggest a general strategy for engineering new cell signaling pathways—in vivo phosphoregulation can, in principle, be introduced into any soluble protein by targeting negatively charged residues at sector-connected surfaces (35).

Beyond Kss1 and Hog1, comparison of the sector to kinome-wide mutational data indicated that sector-connected positions are enriched for mutations that perturb kinase function. The analysis of conservation and coevolution presented here may thus provide a context for interpreting kinase mutations involved in disease and suggest possible cryptic sites for the development of allosteric inhibitors (7). The results of our sequence analyses were robust to details of alignment construction and statistical cutoffs for determining sector positions and conservation (fig. S5 and tables S5, S6, and S8). We note that the sector positions and conservation are strongly overlapping and show an equivalent association with the functional data, meaning that it is difficult to distinguish between the functional statistical significance of conserved residues and sector residues (36). However, our goal is not to test the sector as an exclusive model for allosteric networks in proteins, but rather to examine whether particular surface sites have a capacity for regulation that might be conserved across the kinase family. The sector provides one way to examine this and, unlike single-site conservation, leads naturally to the interpretation that these positions form a cooperative network embedded within the protein structure.

In terms of regulatory mechanism, one possibility is that mutations at or near highly conserved positions modulate activity through global coupling to kinase stability (for example, by unfolding the kinase) rather than a more subtle rearrangement of thermodynamic interactions within the network of sector positions. Although our data cannot exclude the first possibility, we observed that (i) the surface D/E mutations do not strongly affect Kss1 expression, (ii) the insertion of a PKA consensus motif at 9/10 positions tested on Kss1 and Hog1 does not strongly affect expression, and (iii) Kss1–pka-A8 and Kss1–pka-A70 undergo activation loop phosphorylation (although they lack a negative charge at positions 8 and 70 and contain the PKA motif). These observations strongly support the idea that regulation is occurring through allosteric coupling among a defined group of residues, rather than global kinase destabilization at these sites.

Last, we note that these results are consistent with the model that sector-connected surfaces facilitate the evolution of regulatory diversity. In any individual kinase, multiple latent allosteric sites are present, and a handful are harnessed by evolution to yield functional regulation. Thus, regulation that appears idiosyncratic across a protein family might be explained by a conserved underlying mechanism (Fig. 7D). Overall, this model provides a path for understanding how complex regulatory systems evolve, and suggests that sector edges provide a substrate for generating variation in cellular signaling and communication.


Yeast strains and plasmids

Yeast strains and plasmids used in this work are described in tables S1 and S2, respectively. All strains are in the W303 genetic background. Gene deletions were performed by one-step polymerase chain reaction as described (37). All Kss1 mutants were integrated into yeast genome as a single copy expressed from the endogenous KSS1 promoter.

Site-directed mutagenesis

Site-directed mutagenesis was performed with QuikChange according to the manufacturer’s directions (Agilent).

Cell growth and treatment with αF and NaCl

All cells were grown in synthetic complete medium with dextrose (SDC). Three single colonies from each Kss1 or Hog1 strain were inoculated in 1 ml of SDC in 2-ml 96-well deep-well plates and serially diluted 1:5 three times. Plates were incubated overnight at 30°C. In the morning, cells from the row that had been diluted 1:25 were typically found to have OD600 (optical density at 600 nm) ~ 0.5. These cells were diluted 1:5 in four rows of a 96-well U-bottom microtiter plate in a total volume of 180 μl and incubated for 1 hour at 30°C. For Kss1 strains, in each row, cells were treated with different concentrations of αF: 0, 0.01, 0.1, and 1 μM (10× stocks of αF were prepared, and 20 μl was added to 180 μl cells). Treated cells were incubated for an additional 4 hours at 30°C before translation was stopped by addition of cycloheximide (50 μg/ml). Cells were incubated for an additional hour at 30°C to allow time for fluorophores to mature. For Hog1 strains, 0 or 0.5 M NaCl was added for 2 hours before cycloheximide arrest. For experiments with estradiol, everything is the same except that all media contained 20 nM estradiol for the duration of the overnight growth and throughout the experiment.

Flow cytometry

The AGA1pr-YFP and HOR1pr-GFP reporters were measured by flow cytometry by sampling 10 μl of each sample using a BD LSRFortessa equipped with a 96-well plate high-throughput sampler. Data were left ungated, and FlowJo was used to calculate median YFP fluorescence. Bar graphs show the average of the median of the three independent colonies that were assayed, and error bars are the SD.

Confocal microscopy

Ninety-six–well glass bottom plates were coated with concanavalin A (100 μg/ml) in water for 1 hour, washed three times with water, and dried at room temperature. For Kss1 experiments, 80 μl of cells that had been treated with pheromone at the indicated concentrations for 3 hours was diluted to OD600 ~ 0.05 and added to a coated well. Cells were allowed to settle and attach for 15 min, and unattached cells were removed and replaced with 80 μl of SDC medium. Hog-mKate experiments were performed by adding 80 μl of untreated cells to the wells. After the cells were attached and basal images were captured, at t = 0, the medium was removed and replaced with SDC + 0.5 M NaCl and cells were imaged at 5, 20, and 60 min. Nuclear localization was scored manually. Imaging was performed at the W.M. Keck Microscopy Facility at the Whitehead Institute using a Nikon Ti microscope equipped with a 100×, 1.49–numerical aperture objective lens, an Andor Revolution spinning disc confocal setup, and an Andor electron-multiplying charge-coupled device camera. Images were analyzed in ImageJ.

IP of 3×FLAG-tagged Kss1 and mutants

Cultures (2 × 250 ml) of each strain were grown to OD600 = 0.8 at 30°C with shaking, one in SDC and the other in SDC + 20 nM estradiol. The SDC culture was left untreated, whereas the SDC + estradiol culture was treated with 1 μM αF for 30 min. Samples were collected by filtration, and filters were snap frozen in liquid N2 and stored at −80°C. Cells were lysed frozen on the filters in a coffee grinder with dry ice. After the dry ice was evaporated, lysate was resuspended in 1 ml of IP buffer [50 mM Hepes (pH 7.5), 140 mM NaCl, 1 mM EDTA, 1% Triton X-100, 0.1% deoxycholate, and complete protease inhibitors], transferred to a 1.5-ml tube, and spun to remove cell debris. Clarified lysate was transferred to a fresh tube, and serial IP was performed. First, 25 μl of anti-FLAG magnetic beads (50% slurry, Sigma) was added, and the mixture was incubated for 2 hours at 4°C on a rotator. Beads were separated with a magnet, and the supernatant was removed. Beads were washed five times with 1 ml of IP buffer, and bound material was eluted twice with 25 μl of 3×FLAG peptide (1 mg/ml; Sigma) in IP buffer by incubating at room temperature for 10 min. Beads were separated with a magnet, and the two eluates were pooled in a fresh tube. Ten microliters of eluate was analyzed by Western blotting.

Western blotting

Total protein was tricholoracetic acid purified from cells as described (38). Ten microliters of each sample was loaded into 4 to 15% gradient SDS–polyacrylamide gel electrophoresis gels (Bio-Rad). Gels were run at 25 mA for 45 min and blotted to polyvinylidene difluoride membrane at 225 mA for 40 min. After 1-hour blocking in LI-COR blocking buffer, membranes were incubated with anti-FLAG primary antibody (Sigma, F3165), anti–phospho-PKA substrate, anti–phospho-p38, anti–phospho-p44/42 (Cell Signaling, 9101), and/or anti–phosphoglycerate kinase 1 (22C5D8) overnight at 4°C on a platform rotator (all 1:1000 dilutions in blocking buffer). Membranes were washed three times with tris-buffered saline and Tween 20 and probed by anti-mouse or anti-rabbit IRDye-conjugated immunoglobulin G (LI-COR, 926-32352, 1:10,000 dilution). The fluorescent signal was detected with the LI-COR Odyssey system.

Dilution series spot assays (frogging)

Cells were grown to OD600 = 0.3 in SDC, diluted to 0.06, and then serially diluted 1:5. Four microliters of each dilution was spotted on yeast extract peptone dextrose (YPD) + 20 nM estradiol and YPD + 20 nM estradiol + 0.8 M NaCl plates and incubated for 36 and 72 hours, respectively.

Statistical coupling analysis

SCA was performed as described in (29) using PySCA 6 ( for two different multiple sequence alignments of the kinase catalytic domain: one specific to the CMGC kinases (635 sequences) and one containing 7128 kinases sampled across the kinome. The CMGC alignment was constructed by searching kinbase ( Sequences were filtered for a length of 250 to 350 amino acids and aligned by Promals3D (39) including the PDB identifiers 2B9H, 1BI8, 1Q97, 2ERK, 2F49, 2F9G, 2IW8, and 2R7I as reference structures. The kinome-wide alignment was previously constructed by the Shokat laboratory and was downloaded from (40). After alignment processing and the application of sequence weights [as described in (29)], the alignments contained 464 and 380 total effective sequences for the CMGC and eukaryotic protein kinase alignments, respectively. Both alignments are restricted to the catalytic domain and do not include regulatory domains. For both alignments, we followed an identical procedure for defining the sector. Briefly, we compute a conservation-weighted covariance matrix between all pairs of amino acid positions (see Supplementary Text for discussion of the relationship between the sector, conservation, and allosteric hotspots). This matrix provides a statistical description of the “evolutionary coupling” between all pairs of amino acid positions. We then analyze this matrix by conducting principal components analysis and rotating the top eigenmodes using independent component (IC) analysis. The top ICs are used to define sectors. For both kinase alignments, we define a single sector that includes all positions contributing to the top four ICs. The group of positions contributing to each IC groups is defined by fitting an empirical statistical distribution to the ICs and choosing positions above a defined cutoff (default, >95% of the continuous distribution function). The full analysis of both families can be downloaded from (

Defining sector-connected solvent-accessible surface sites

We computed the relative solvent-accessible surface area (RSA) over a homology model of Kss1 (30) using Sanner et al.’s MSMS, with a probe size of 1.4 Å, excluding all water and heteroatoms (41). A cutoff of 20% RSA was used to define solvent-exposed surface positions (5). “Sector-connected” is defined as a position where any atom is within 4.0 Å of a sector position.


Text S1. The relationship between the sector, conservation, and allosteric hotspots.

Text S2. Background on Kss1.

Fig. S1. The effect of surface Asp/Glu-to-Ala mutations on Kss1 expression.

Fig. S2. Experimental approach to introduce a PKA phosphorylation site that controls MAPK Kss1 activity.

Fig. S3. The kinase sector encompasses the catalytic and regulatory spines.

Fig. S4. ERK2 mutations within the kinase sector are enriched for loss of function.

Fig. S5. The relationship of negatively charged surface positions to the kinome-wide eukaryotic protein kinase sector.

Fig. S6. Phosphorylation, phenotype, and nuclear localization of Hog1-pka mutants.

Table S1. Plasmids.

Table S2. Yeast strains.

Table S3. Comparison of Kss1 point mutations from the literature with our data and the CMGC sector.

Table S4. Sector positions mapped to several representative kinase structures.

Table S5. Statistical association between the sector, conservation, and ERK2 mutational data.

Table S6. Statistical association between the sector, conservation, and KSS1 D/E surface mutations.

Table S7. Functional kinase mutations sampled across the kinome.

Table S8. Statistical association between the sector, conservation, and functional mutations sampled across a diversity of kinases.

References (4259)


Acknowledgments: This collaboration was initiated at the 2013 q-bio conference held at St. Johns College, Santa Fe, NM. We would like to thank R. Ranganathan for discussion and comments on the manuscript and H. Zhang (UTSW bioinformatics) for review of our statistical methods. We are grateful to the Whitehead Institute FACS facility and the Keck Microscopy facility for technical assistance. Funding: This work was supported by an NIH Early Independence Award (DP5 OD017941-01 to D.P.), the Green Center for Systems Biology, the Gordon and Betty Moore Foundation’s Data-Driven Discovery Initiative (grant GBMF4557 to K.A.R.), a Merck Postdoctoral Fellowship from the Helen Hay Whitney Foundation (to P.C.), and an NIH Pathway to Independence Award (K99/R00 CA226393 to P.C.). Author contributions: K.A.R., D.P., and O.R. conceptualized the study; K.A.R., D.P., O.R., and P.C. developed the methodology; D.P., J.P.P., Z.A.F., O.R., and K.A.R. performed the experimental investigations; D.P. and K.A.R. wrote, edited, and revised the manuscript; D.P., J.P.P., O.R., and K.A.R. reviewed and edited the manuscript; D.P. and K.A.R. supervised the study. Competing interests: The authors declare that they have no competing interests. Data and materials availability: The SCA analysis for both kinase alignments is available from GitHub ( All other data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials.
View Abstract

Navigate This Article