Many major protein–protein interaction networks are maintained by ‘hub’ proteins with multiple binding partners, where interactions are often facilitated by intrinsically disordered protein regions that undergo post-translational modifications, such as phosphorylation. Phosphorylation can directly affect protein function and control recognition by proteins that ‘read’ the phosphorylation code, re-wiring the interactome. The eukaryotic 14-3-3 proteins recognizing multiple phosphoproteins nicely exemplify these concepts. Although recent studies established the biochemical and structural basis for the interaction of the 14-3-3 dimers with several phosphorylated clients, understanding their assembly with partners phosphorylated at multiple sites represents a challenge. Suboptimal sequence context around the phosphorylated residue may reduce binding affinity, resulting in quantitative differences for distinct phosphorylation sites, making hierarchy and priority in their binding rather uncertain. Recently, Stevers et al. [Biochemical Journal (2017) 474: 1273–1287] undertook a remarkable attempt to untangle the mechanism of 14-3-3 dimer binding to leucine-rich repeat kinase 2 (LRRK2) that contains multiple candidate 14-3-3-binding sites and is mutated in Parkinson's disease. By using the protein-peptide binding approach, the authors systematically analyzed affinities for a set of LRRK2 phosphopeptides, alone or in combination, to a 14-3-3 protein and determined crystal structures for 14-3-3 complexes with selected phosphopeptides. This study addresses a long-standing question in the 14-3-3 biology, unearthing a range of important details that are relevant for understanding binding mechanisms of other polyvalent proteins.
Discovered in 1967 , 14-3-3 proteins became one of the first known protein modules recognizing protein phosphorylation [2,3]. Although being practically unable to bind unphosphorylated forms of protein clients (rare exceptions exist [4,5]), 14-3-3 proteins are well-known for their ability to interact with a variety of cellular proteins containing phosphorylated Ser/Thr residues within certain sequence contexts. Biochemical studies revealed several optimal 14-3-3-binding sequences flanking the central phospho-Ser/Thr, that can be roughly subdivided into three main classes — internal motif I (R/K)-[S/F/Y/W]-[H/+]-p(S/T)-[L/E/A/M]-(P/G), motif II R-X-[S/Y/F/W]-(+)-p(S/T)-[L/E/A/M]-[P/G], and the C-terminally located motif III p(S/T)X1-2-COOH (where X denotes any amino acid, ‘+’ denotes positively charged amino acid, and p(S/T) denotes phosphorylated Ser or Thr) [6–8]. The C-terminal motif III 14-3-3-binding sites may overlap with the C-terminal PDZ domain-binding motifs (PBMs) and the recent data suggest that phosphorylation of certain PBMs can play a role of a 14-3-3/PDZ-binding switch that links the two protein–protein interaction (PPI) networks [9–11].
14-3-3-binding phosphomotifs tend to locate in the intrinsically disordered protein regions (IDPRs) , which brings sufficient plasticity for their accommodation within the binding grooves of 14-3-3 dimers that have a rather rigid structure . 14-3-3 binding to phosphorylated sites of client proteins may have different outcomes including (i) the regulation of their enzymatic activity; (ii) masking of recognition motifs for binding to other proteins and, therefore, modulation of PPI; (iii) the regulation of intracellular localization of client proteins, and, more speculatively, (iv) mediation of protein interactions via bringing together two interacting entities on a ditopic 14-3-3 scaffold  (see also below).
14-3-3-binding motifs appear upon phosphorylation by different kinases, e.g. of the AGC and CAMK families phosphorylating (R/K)XX(S/T) sequences , and are found in hundreds of different proteins . This explains the involvement of 14-3-3 proteins in sophisticated PPI networks [17,18], in the regulation of many different processes [19,20], and in the development of neurodegenerative diseases and cancer [21–23]. The high biomedical relevance of 14-3-3 complexes has gained much attention during the recent years, and those complexes are considered promising targets for modulation by small molecules in drug discovery and therapy [24,25].
Phosphorylation of client proteins allows members of the 14-3-3 family to ‘read’ this modification. Forming rather stable α-helical homo- and heterodimers from several 14-3-3 isotypes, that are usually named by Greek letters and co-expressed in eukaryotic organisms, 14-3-3 proteins harbor two identical phosphopeptide-binding amphipathic grooves , and, therefore, by definition are ditopic. Each subunit of the antiparallel 14-3-3 dimer contains one such phosphopeptide-binding site, where the conserved Lys49, Arg56, Arg127, and Tyr128 residues (human 14-3-3ζ numbering) coordinate the phosphate moiety , presenting some filter that selectively accepts phosphorylation. Noteworthy, the same unphosphorylated motifs are usually ignored by 14-3-3, despite the presence of the same positional determinants of binding surrounding the central phosphorylatable Ser/Thr. The optimal consensus 14-3-3-binding motifs derived from peptide libraries emphasize the favorable effect of a positively charged Arg (or Lys) residue upstream of pSer/pThr and a Pro/Gly residue in position +2 relative to pSer/pThr . However, knowledge accumulated in more than 20 years of research says that phosphopeptide sequences with the experimentally confirmed 14-3-3 binding may substantially differ from the consensus motifs I–III, making a sequence-based prediction of binding affinities and even likelihood of binding problematic. This questions the applicability of existing prediction algorithms (e.g. [27,28]) and underlines the remaining importance of an individual experimental approach to 14-3-3/client interactions.
This problem lies deeper because many clients of 14-3-3 proteins bear more than one phosphorylatable 14-3-3-binding site. It was shown that doubly phosphorylated clients bind to 14-3-3 dimers with an increased affinity, exceeding a simple additive effect [8,29] — the occupation of both phosphopeptide-binding grooves of 14-3-3 leads to client stabilization and solidifies the complex. In line with this, the ‘molecular anvil’ and ‘gatekeeper phosphorylation’ hypotheses were put forward . According to the latter, the presence of two binding centers allows stepwise phosphorylation and binding to 14-3-3, whereas the former hypothesis suggests that such a bidentate binding can deform the doubly phosphorylated client on the rather rigid 14-3-3 dimer (an ‘anvil’), thereby regulating client's structure and function . Cdc25B binding to 14-3-3 using one high- and two low-affinity sites limits the access to the catalytic site of this phosphatase and seems to support such a scenario . Further development of the abovementioned ideas led to the insightful suggestion that ditopic 14-3-3 proteins may operate as ‘logic gates’ and ‘coincidence detectors’ that integrate different signaling inputs, i.e. site-specific phosphorylation by different protein kinases, into a timely and weighted output . Indeed, the concerted action of two kinase cascades may be required to trigger the downstream effects mediated by 14-3-3 binding, such as in the case of proapoptotic BAD protein phosphorylation by ERK1/2, PKB/Akt, and PKA  or glucose uptake-regulating Rab GTPase-activating protein AS160 (Akt substrate of 160 kDa) phosphorylation by PKB/Akt and AMPK protein kinases [33,34]. In such a context, 14-3-3 proteins by right may be considered as readers and translators of the phosphorylation code that is installed by a range of protein kinases (writers) in response to various intra- or extracellular stimuli.
Another aspect of ditopic 14-3-3 protein binding is their so-called scaffolding (or adaptor) function that is fulfilled upon tentative simultaneous binding to two different partners, each containing a 14-3-3-binding phosphosite, whereby 14-3-3 could mediate transient interactions and cross-talk between proteins that do not appreciably interact in the 14-3-3 absence. Although appealing, this hypothesis lacks direct structural confirmations, as ternary complexes formed via a 14-3-3 dimer have not been sufficiently well evidenced and described yet.
Most likely dependent on a particular phosphoprotein, the problem of the stoichiometry of 14-3-3 complexes cannot be neglected. Variations include structurally characterized scenarios when (i) 14-3-3 dimer is bound to some dimer composed of singly phosphorylated polypeptide chains (stoichiometry 2 : 2, such as seen in crystal structures of 14-3-3 with phosphorylated HSPB6 dimer  or B-RAF dimer ); (ii) 14-3-3 dimer bound to two non-interacting monomers of a client (stoichiometry 1 : 2 : 1, such as seen in the crystal structure of 14-3-3 with serotonin N-acetyltransferase (AANAT) ); (iii) 14-3-3 dimer bound to a doubly phosphorylated client (stoichiometry 2 : 1, such as seen in the crystal structure of yeast 14-3-3 with neutral trehalase Nth1 , Figure 1). Remarkably, most of these structures have been published in the last 3 years, i.e. the long time after the first structure of the 14-3-3:client complex was published in 2001 . This reflects a great challenge that the community has been facing in studies of 14-3-3 complexes that are enriched with IDPRs and, therefore, difficult for structural biology approaches [12,39,40].
Stoichiometry variations in 14-3-3 complexes.
A completely different challenge is presented by binding of principally ditopic 14-3-3 protein to client proteins with multisite phosphorylation, where stoichiometry and binding mechanism is impossible to predict without a careful systematic analysis. These examples include 14-3-3 binding to (i) Tau protein involved in the development of Alzheimer's disease and other tauopathies (at least three to seven 14-3-3-binding sites [41,42]), (ii) cystic fibrosis transmembrane conductance regulator (CFTR) involved in the development of the associated genetic disease (nine 14-3-3-binding sites ), and (iii) Parkinson's disease-related LRRK2 (six 14-3-3-binding sites ).
Some trivial considerations should be taken into account when studying the interaction of multivalent phosphoproteins with 14-3-3. If the phosphosites are located too close to each other in sequence, they may interfere upon 14-3-3 binding. To contribute to 14-3-3 binding, they have to be separated by at least 15 residues not forming rigid structural elements, otherwise, the antiparallel 14-3-3 dimer can accommodate only one of them. The sites that are too far away from each other, both in sequence and in space (for example, in large multi-domain proteins like ∼2500 amino acid residues-long leucine-rich repeat kinase 2 (LRRK2) ), can serve as docking points for different 14-3-3 molecules unless the protein architecture permits sufficient 14-3-3 dimer-mediated deformation. Synergistic binding of three 14-3-3 dimers to the C-terminal tails of the plant plasma membrane H+-ATPase dimers was shown to convert the enzyme into the active hexamer [45,46]. Although somewhat speculative in the absence of direct structural information about 14-3-3 complexes with full-length multivalent proteins, the presence of IDPRs should be especially beneficial for complex formation with large multivalent client proteins composed of several rigid domains. Multivalency may lead to various 14-3-3 binding configurations involving different combinations of binding sites and each combination may provide distinct output. This emphasizes the importance of studies aimed to delineate the binding process that basically decodes the complex multisite phosphorylation signal via 14-3-3.
Multisite phosphorylation indeed creates some complexity that can be interpreted by reader proteins as the phosphorylation code, and this complexity can be expressed in numbers. In the example depicted in Figure 1B, 14-3-3 dimer binding to a hexavalent phosphoprotein generates the 6!/2!*4! = 15 unique combinations of outputs taking into account ditopic binding, plus 6 possible single-site binding events, 21 variants in total. Likewise, for a nonavalent partner, there are 36 unique possible outputs of the ditopic binding, plus 9 single-site binding events (45 in total). Even if some of these variations are not realized due to a low affinity or steric hindrance, the repertoire appears to be pretty rich. Taking into account differences in binding affinity at individual sites, the multivalency creates a very complex hierarchy of the binding events, which complicates the binding mechanism and makes its deciphering an enthralling challenge.
In their recent work, Stevers et al.  have systematically studied the 14-3-3 interaction with human LRRK2. This large enzyme contains seven domains and a range of residues, upon phosphorylation of which recruitment of 14-3-3 is achieved. Among 20 PKA phosphorylation sites , the six candidate 14-3-3-binding sites (Figure 2) are found at the end of ANK domain (S860), in the long disordered loop connecting ANK and LRR domains (S910, S935, S955, S973), and in the ROC domain (S1444) . The latter site, 1441-RASpSSP-1446, represents motif I 14-3-3-binding site and includes residue Arg1441 that is subject to mutation in Parkinson's disease, which impairs S1444 phosphorylation and 14-3-3 binding . Although being distant from the other five residues, S1444 was earlier shown to be a high-affinity binding site for 14-3-3 (KD ∼ 0.2 µM ). As with other 14-3-3 partners, the sites unlikely contribute to 14-3-3 binding unless they are phosphorylated; however, apart from S1444 and S935, the sequences of the rest four tentative 14-3-3-binding sites in LRRK2 are not optimal (Figure 2). Nevertheless, Stevers et al.  clearly demonstrated that at least three combinations, involving S910 + S935 (KD ∼ 11 nM), S935 + S1444 (KD ∼ 3 nM), and S910 + S1444 (KD ∼ 6 nM), are among the highest-affinity 14-3-3 binders known to date (Figure 2). Nanomolar affinities were measured by two independent experimental methods, fluorescence polarization and isothermal titration calorimetry, increasing the robustness of the assessment. The authors selected the protein-peptide binding approach and overall used 20 synthetic peptides representing singly and different combinations of doubly phosphorylated peptides to make an insight into the hierarchy of binding events. In addition, crystal structures trapping the 14-3-3 interaction with the S910 (PDB code 5MYC) and S935 (PDB code 5MY9) high-affinity sites were obtained . Although these cannot be fully representative of the 14-3-3/LRRK2 molecular interface, especially given that the S1444 highest-affinity peptide could not be crystallized , these atomic resolution structures revealed several remarkable points. First, although binding at a very similar affinity (∼8 µM KD determined by fluorescence polarization ), the S910 and S935 peptides show substantially different occupancy and order in the binding grooves of 14-3-3 — in the S935 case, only 4 out of 13 residues are seen in the electron density maps in spite of the relatively high resolution of 1.33 Å. It is especially surprising given that the S935 site is much more similar than the S910 site to the consensus motifs that are thought to be optimal for 14-3-3 binding (Figure 2). This strongly points to the lack of correlation between the peptide sequence, its binding affinity, and the apparent stability of its interface with 14-3-3. Second, it is intriguing that the long peptide co-crystallized with 14-3-3 (VKKKSNSISVGEFYRDAVLQRCSPNLQRHSNSLGPIFD, phosphosites are underlined; PDB code 5MYC), encompassing the two sites S910 + S935 that give a nanomolar affinity to 14-3-3, led to only one site of 14-3-3 occupied (12 residues in the interpretable electron density). The observed peptide conformation features a short α-helix, which is very atypical of the 14-3-3 bound phosphopeptides  and raises an interesting question of whether it folds upon 14-3-3 binding or pre-exists in the LRRK2 structure.
Summary of the protein-peptide binding approach used by Stevers et al. [
44] to characterize the 14-3-3/LRRK2 interaction.
The inability of the 14-3-3 complex with the highest-affinity peptide around S1444 to form crystals is also remarkable and hints to potential problems with the solubility of this peptide, as stated by the authors . The set of the mentioned issues (i.e. partial binding occupancy, low peptide solubility, unstable binding stoichiometry even for the high-affinity peptides) is very often encountered in crystallographic studies of 14-3-3/peptide complexes and can to some extent be alleviated by direct tethering of the client peptides to the C terminus of 14-3-3 and co-expression of such fusion proteins in Escherichia coli in the presence of a protein kinase or by in vitro phosphorylation following fusion protein purification . This fortunate opportunity is provided by a natural orientation of phosphopeptides in the binding grooves of 14-3-3, which brings the C-terminal end of 14-3-3 very close to the N-terminal end of the phosphopeptide, and is directly supported by an exotic 14-3-3 structure with the self-bound C-terminal peptide spontaneously phosphorylated upon production in E. coli (PDB code 3EFZ ).
For multivalent clients, first binding events inevitably lead to an increase in the effective local concentration of other, yet unbound phosphosites, which can substantially improve the binding efficiency for even low-affinity sites. The great attempt to model the complexity of multivalent phosphoprotein binding to 14-3-3 was undertaken recently by the same research group . Curiously, depending on the set of affinities of the individual binding sites within the polyvalent client, which could only be obtained using the robust systematic approach demonstrated by Stevers et al. for LRRK2  and CFTR , the binding regime may be dramatically different. In case of a too large difference in KD for individual sites, only a limited set of those will likely be realized and removal of one of the highest-affinity sites drastically influences the overall affinity (as predicted for LRRK2). For the multiple binding sites with similar affinity, a remarkable buffering may take place, when the low-affinity sites may collectively contribute to the appreciably tight binding and removal of some sites is compensated by the remaining sites, which maintains the overall affinity (as predicted for CFTR) . Importantly, in both CFTR and LRRK2 cases, the stabilization of specific binding sites much more efficiently affects the binding affinity than site-specific inhibition .
The focused attention of the community to the issues highlighted above is expected to help elucidate binding mechanisms involving other polyvalent proteins having wide-ranging roles in cellular signaling.
The author declares that there are no competing interests associated with this manuscript.
This work was supported by the Russian Science Foundation (grant no. 19-74-10031).
I am grateful to Prof. Alfred A. Antson (YSBL, The University of York, U.K.) for critical reading of the manuscript.