E6 PBMs show parallel binding profiles to 14-3-3 isoforms

Among all 225 HPV E6 proteins curated in the PaVE database (https://pave.niaid.nih.gov/, Accessed December 9, 2020), 31 E6 proteins from mucosal α-genera HPV possess a C-terminal PBM with the class 1 consensus (X(S/T)X(L/V/I/C)-COOH, where X is any amino acid residue^30,31). E6 PBMs are phosphorylatable by protein kinases at their conserved antepenultimate S/T residue^15,16,32. This phosphosite is preceded by arginine residues in most of the HPV E6 PBM sequences with recognizable basophilic kinase substrate consensus motifs, R(X/R)X(S/T) and RXRXX(S/T)^33,34. The E6 PBMs can be classified in three subgroups: subgroups 1 and 2 prone to phosphorylation by the basophilic kinases and orphan subgroup 3 with a less predictable phosphorylation propensity (Supplementary Fig. 1). In line with earlier observations^15,28,35, the phospho-PBM sequences from subgroups 1 and 2 ideally match the C-terminal 14-3-3-binding motif III⁹ (Fig. 1a and Supplementary Table 1).

The seven full-length human 14-3-3 isoforms, produced as fusions to a maltose-binding protein (MBP), were analyzed for their interaction with four phospho-PBMs from E6 proteins of HPV types 16, 18, 33, and 35 belonging to subgroups 1 and 2 (as defined in Supplementary Fig. 1). For comparison, we also measured two non-viral phospho-PBMs originating from protein kinase RSK1²⁸. We used a competitive fluorescence polarization (FP) assay that measures the displacement of a fluorescent tracer phosphopeptide (here, derived from the HSPB6 protein) bound beforehand to 14-3-3, by an increasing amount of the peptide of interest. All binding curves are shown in Supplementary Fig. 2a.

All phospho-PBMs (p16E6, p18E6, p33E6, p35E6, RSK1_-1P, and RSK1_-2P) detectably bound to 14-3-3 proteins, in sharp contrast to their unphosphorylated counterparts. The interactions between E6 phospho-PBMs and 14-3-3 proteins spanned very wide affinity ranges, from just below 1 μM (p33E6–14-3-3γ) to above 300 μM (Fig. 1b and Supplementary Fig. 2a). Such large binding affinity differences are noteworthy since the four E6 PBM sequences are very similar (Fig. 1a), and all 14-3-3 isoforms share highly conserved phosphopeptide-binding grooves^1,4.

Remarkably, the six phospho-PBMs obeyed a consistent hierarchized profile in their relative binding preferences toward the seven 14-3-3 isoforms, albeit with an overall shift in affinity from one peptide to another. For each phosphopeptide, the seven 14-3-3 isoforms systematically clustered as four groups of decreasing affinity, in a conserved order from the strongest to the weakest phospho-PBM binder: gamma, eta, zeta/tau/beta, and epsilon/sigma (γ, η, ζ/τ/β, and ε/σ) (Fig. 1b). These conserved relative affinity shifts can be quantified by calculating, for two distinct 14-3-3 isoforms, their differences of free energy of binding (ΔΔG) toward each individual phosphopeptide, then calculating the average difference (ΔΔG_av) with its standard deviation (Fig. 1c). Between the strongest and the weakest binders (isoforms γ and σ, respectively) the average phosphopeptide-binding energy difference is ΔΔG_av = –5.1 ± 1.3 kJ/mol, roughly corresponding to a 11-fold K_D ratio.

The seven 14-3-3 isoforms also showed consistent profiles in their relative binding preferences toward the four E6 phospho-PBMs. For each 14-3-3 isoform, the four phospho-PBMs systematically rank the same way from the strongest to the weakest binder: p33E6, p18E6, p16E6, and p35E6 (Fig. 1c). The average 14-3-3 binding free energy difference between p33E6 and p35E6 was ΔΔG_av = –10.9 ± 0.7 kJ/mol, roughly corresponding to a 100-fold K_D ratio.

Of note, the presence of the MBP tag did not affect the relative affinity differences observed for the 14-3-3 isoforms since selected untagged 14-3-3 isoforms obeyed the same trend from the strongest to the weakest (tau/beta and epsilon) (Supplementary Fig. 2c) observed for the MBP-tagged variants (Fig. 1b, c). The untagged 14-3-3 variants also preserved toward the four selected HPV E6 PBMs the preferences observed in more detail for the full-length MBP-tagged 14-3-3 isoforms (Fig. 1).

Atomic structure reveals the 14-3-3ζ–18E6 PBM interface

To get structural insight into the 14-3-3ζ interaction with 18E6 PBM, we determined a crystal structure of the 14-3-3ζ–18E6 phospho-PBM complex at a 1.9 Å resolution using a previously reported chimeric fusion strategy³⁶ (Table 1, Fig. 2, and Supplementary Figs. 3 and 4). The phosphopeptide establishes multiple polar interactions with the basic pocket in the amphipathic groove of 14-3-3 (Supplementary Fig. 5), largely reminiscent of previously solved structures of 14-3-3–phosphopeptide complexes^8,36. The conformation of 18E6 phosphopeptide bound to 14-3-3ζ within the chimera is practically identical (RMSD = 0.17 Å upon superimposition of Cα atoms of the peptides) to the 14-3-3σ-bound conformation of a synthetic 16E6 phosphopeptide reported very recently at a lower resolution (Fig. 2b)²⁸. The observed conservation of most interface contacts within the two complexes suggests that these crystal structures can serve as templates to build accurate homology models of 14-3-3 complexes for other E6 phospho-PBMs or, more generally, other C-terminal motif III peptides phosphorylated on the antepenultimate position.

Nonetheless, a few noteworthy differences appear in a subset of the crystallographic conformers of 14-3-3/16E6 and 14-3-3/18E6 complexes. On the one hand, in one of the four conformers observed in the asymmetric unit of the 14-3-3σ/16E6 crystal, the side chains of Arg –7 (Gln in 18E6) and Glu –3 form an additional in-cis salt bridge (Fig. 2b). On the other hand, Arg –6 of 18E6 (Thr in 16E6) mediates a bipartite interaction with 14-3-3 in most of the observed conformers. It simultaneously interacts with the carbonyl of Asp223 and participates in a water-mediated interaction with Asn224 (Fig. 2b and Supplementary Fig. 5).

Rational design rescues the weakest E6–14-3-3 interaction

Next, we investigated possible causes of the remarkable 14-3-3 binding affinity differences observed between the four E6 phospho-PBMs.

In principle, the affinity of a series of variant peptides for a given protein may be modulated by two types of atomic contacts: intermolecular and intramolecular contacts within the formed complexes, and intramolecular contacts in the free unbound peptides.

As concerns contacts within the 14-3-3/E6 complexes, the crystal structures have shown that Arg –6 can mediate more interactions than Thr –6 with the generic 14-3-3 interface (Fig. 2b). Interestingly, position –6 is an Arg in the two strongest 14-3-3 binders (18E6 and 33E6) versus a Thr in the weakest ones (35E6 and 16E6).

As concerns possible contacts within the unbound peptides, we noticed that all E6 phospho-PBMs have a delicate charge distribution, with an acidic C-terminal segment (that includes the C-terminal -COOH group and the natural acidic or phosphorylated residues) and a basic N-terminal segment (that is also involved in recognition by kinases). These local charged segments may form, within the unbound phosphopeptide, transient in-cis interactions often referred to as charge clamps³⁷. We speculated that Glu –1 in p35E6, the weakest 14-3-3 binder, might participate in such a charge clamp, thereby disfavoring its binding to 14-3-3.

To address these potential mechanisms, we synthesized three variants of the weakest 14-3-3 binder, p35E6. The first variant contained a T-6R substitution, which in principle could allow a more stable bound conformation, but may also stabilize charge clamps in the free form of the motif. The second variant contained an E-1A substitution, which in principle could destabilize in-cis charge clamps. A third variant contained both substitutions. All substitutions turned out to reinforce the binding affinities of 35E6 without altering the apparent preferences of the different 14-3-3 isoforms (Figs. 1b and 2c). Taken individually, T-6R moderately increased binding (ΔΔG_av = –1.1 ± 0.5 kJ/mol, 1.5-fold K_D ratio), while E-1A strongly reinforced it (ΔΔG_av = –5.1 ± 0.2 kJ/mol, 11-fold K_D ratio). When combined, the two substitutions synergistically increased binding (ΔΔG_av = –8.7 ± 0.4 kJ/mol, 35-fold K_D ratio), thereby turning p35E6 into a strong 14-3-3 binder, just below p33E6 (Figs. 1b and 2c). These results indicate that the two above-stated mechanisms act in combination to generate the wide 14-3-3-binding affinity range displayed by distinct E6 phospho-PBMs despite their high sequence conservation.

The 14-3-3/E6 PBM interaction is druggable by fusicoccin

Fusicoccin (FSC) is a commonly used stabilizer of 14-3-3 complexes, when its insertion within the 14-3-3/phosphopeptide interface is allowed by phosphopeptide side chains of the amino acids in downstream positions relative to the phospho-residue^38,39,40,41. This is especially the case with motif III phosphopeptide complexes of 14-3-3 having only one residue after the phosphosite^39,41,42. Nevertheless, the effect of FSC on interaction of longer motif III phosphopeptides with 14-3-3 is less characterized (Supplementary Table 1).

We performed FP experiments to measure equilibrium binding affinity constants of complexes between the four HPV E6 phosphopeptides and 14-3-3 isoforms ζ and γ, in the presence of FSC (Fig. 3a and Supplementary Fig. 2b). The addition of FSC consistently decreased by 1.5- to 2-fold the affinities of all eight interactions (ΔΔG_av = –1.3 ± 0.5 and –1.8 ± 0.4 kJ/mol for ζ and γ, respectively) without altering the apparent preferences of the different peptides (Fig. 3a, b).

a Affinities of four selected HPV E6 phospho-PBMs toward human 14-3-3ζ and 14-3-3γ in the absence and presence of FSC as determined by FP using FITC-labeled HSPB6 phosphopeptide as a tracer. Protein names are boldfaced for clarity. Apparent K_D values determined from competitive FP experiments are presented. Standard deviation (std) values are indicated. The binding curves are shown in Supplementary Fig. 2. b Averaged ΔΔG values between 14-3-3–E6 phospho-PBM pairs in the absence or presence of FSC, calculated based on their observed order of binding affinities (from weakest to strongest). Individual K_D values from Supplementary Fig. 2 were first converted into ΔG values (at T = 295 K; excluding cases when K_D > 300 μM), then average ΔΔG values (ΔΔG_av) were calculated for the E6-binding affinity changes of the indicated 14-3-3 isoforms in the absence or presence of FSC. c An overall view on the ternary complex between 14-3-3ζ (subunits are shown by surface using two tints of gray), 18E6 phosphopeptide (cyan sticks), and FSC (pink sticks). FSC was soaked into the 14-3-3ζ–18E6 chimera crystals. 2F_o-F_c electron density maps contoured at 1σ are shown for the phosphopeptide and FSC only. d The effect of FSC binding. Conformational changes in the 9th α-helix of 14-3-3 and in the C-terminal part of the 18E6 phosphopeptide upon FSC binding are shown by red arrows, a significant rise of the local B-factors of the phosphopeptide is shown using a gradient from blue to red as indicated. The amplitudes of the conformational changes are indicated in Å by dashed arrows.

Next, we used a soaking approach to crystallize the ternary 14-3-3ζ/18E6 PBM/FSC complex and solved its structure at 1.85 Å resolution (Fig. 3c, Table 1, and Supplementary Figs. 4 and 5). While FSC binding in the well-defined cavity did not disrupt the overall assembly (Fig. 3c and Supplementary Fig. 4), it induced a hallmark ~4 Å closure of the last α-helix of 14-3-3ζ (Fig. 3d) as observed for other 14-3-3 complexes containing FSC⁴³. Also, FSC binding reoriented the C-terminal carboxyl group and caused local destabilization of the very C-terminal residues of the phosphopeptide, increasing their temperature factors and dispersing the local electron density (Fig. 3d and Supplementary Fig. 5). As a result of FSC binding, the water network around the phospho-PBM C-terminus significantly changed (Supplementary Fig. 6).

Nevertheless, the simultaneous binding of FSC and E6 PBM in the amphipathic groove of 14-3-3 indicates that such ternary complex can be used as a starting point to design both stabilizers and inhibitors of 14-3-3/E6 interactions.

Hierarchized peptide-affinity profiles are a general feature of human 14-3-3 isoforms

Former studies have measured the binding of the seven human 14-3-3 isoforms to unrelated phospho-motifs derived from Cystic Fibrosis Transmembrane Conductance Regulator (CFTR), Leucine-Rich Repeat Kinase 2 (LRRK2), Potassium channel subfamily K members (TASK1/3), C-Raf, the p65 subunit of the NF-κB transcription factor, and from Ubiquitin carboxyl-terminal hydrolase 8 (USP8), representing a wide variety of different 14-3-3-binding motifs, including C-terminal, internal, monovalent, or divalent motifs^{40,44,45,46,47,48}. These phosphorylated motifs from different origins have a strikingly wide affinity range, spanning from low nanomolar to low millimolar detectable dissociation constants (Fig. 4a–c). For instance, for 14-3-3γ, the K_D ratio between the strongest and the weakest-binding phosphopeptide is almost 625-fold in the present work, and 39,000-fold when taking into account affinities from the literature (Fig. 4b).

a Affinity maps of 14-3-3 interactions based on experimentally determined dissociation constants against the 14-3-3ome, as obtained in this study (white background) and in others^{40,44,45,46,47,48} (gray background). The color scale is either based on affinity values or on K_D ratios. √ denotes affinities weaker than the limit of quantitation of the experimental assays. b Same map as in a, normalized to the strongest 14-3-3-binding motif. An average 34,000-fold K_D ratio is observed between the strongest and weakest 14-3-3-binding peptide. c Same map as in a, normalized to the strongest phosphopeptide-binding 14-3-3 isoform. Note that all peptides follow very similar affinity trends between the different 14-3-3 isoforms, with an average 12-fold K_D ratio between the strongest and weakest-binding 14-3-3 isoform. d Number of unique partners detected according to the BioPlex database²⁹ for each 14-3-3 isoform, taken individually (left) or grouped in three subsets (right) following their relative affinity trends (strong, intermediate, and weak binders). e Number of 14-3-3 partners in BioPlex, which bound to 1, 2, 3, 4, 5, 6, or all 7 isoforms, respectively. Within each bar, the proportion of partners that bound to each individual isoform is indicated (same isoform color code as in d). f Venn diagram showing repartition of the 14-3-3 partners from BioPlex among the strong, medium, and weak phosphopeptide-binding subsets, defined as in d. g Correlation between the number of binders of 14-3-3 isoforms, according to BioPlex, and ΔΔG_av between the strongest phosphopeptide-binder, 14-3-3γ, and all individual isoforms (same color code as in d). ΔΔG values were calculated from the average K_D ratios from c. h The average amounts of prey proteins from BioPlex (normalized to the amount captured using 14-3-3γ) that interact with at least five different 14-3-3 baits, deduced from their PSM values, also show a correlation with the ΔΔG_av values of the same proteins. i Correlation of the sequence identity of human 14-3-3 isoforms relative to 14-3-3γ with the ΔΔG_av values from g. Source data are provided as a Source Data file.

Conversely, the hierarchized relative binding profile of the seven human 14-3-3 isoforms observed herein for E6 and RSK1 phosphopeptides is remarkably confirmed in most published data that have also measured affinities for all these seven isoforms^{40,44,45,46,47,48}, with 14-3-3γ and 14-3-3η consistently being the strongest binders and 14-3-3σ and 14-3-3ε being the weakest binders, independently of the nature of the target motif (Fig. 4c). Furthermore, the average maximal K_D ratio between the strongest-binding and the weakest-binding 14-3-3 in the literature is around 12-fold, like in our present work (~11-fold).

Moving further, we wondered whether the observed trends would be conserved at a full proteome-wide scale. Recently, a massive parallel affinity-purification approach coupled to mass-spectrometry (AP-MS) was applied to decipher the complexomes of more than 10,000 recombinantly expressed bait proteins in two orthogonal cell lines²⁹. We retrieved from the BioPlex 3 database (a compendium of the abovementioned complexome study) the numbers of detected interaction partners for each 14-3-3 isoform (Fig. 4d–g). In total, 547 unique proteins were detected as an interaction partner of at least a single 14-3-3 isoform. Out of those, 14-3-3γ and 14-3-3η had the highest number of interaction partners, followed by a second group including 14-3-3β, 14-3-3ζ, and 14-3-3τ, and a third group comprising 14-3-3σ and 14-3-3ε (Fig. 4d). Most of these interaction partners were found to bind more than a single 14-3-3 isoform (Fig. 4e, f). While the strongest-binding isoforms (γ and η) do not share ~30% of their interactome with the other isoforms, they interact with more than 85% of the binders of the mild-binding isoforms (β, ζ, and τ) and more than 90% of the binders of the weak-binding isoform 14-3-3ε. Indeed, out of the 75 detected binders of 14-3-3ε, only 1 (below 2% of the total) is unique to 14-3-3ε. By contrast, the other weak-binding isoform, 14-3-3σ, has a distinct behavior. Out of its 51 detected binders, 26 interactions are unique to 14-3-3σ (above 50% of all its binders).

Furthermore, we observed a remarkable linear correlation (R² = 0.91) between the numbers of binders detected by the BioPlex project²⁹ for each 14-3-3 isoform, and their relative affinity (ΔΔG_av) as compared to the strongest phosphopeptide-binding isoform, 14-3-3γ (Fig. 4g).

In the AP-MS experiments, interaction partners (and 14-3-3 proteins in particular) can be either baits or preys. Baits are recombinantly expressed in the cells using the same promoter, which should ensure a relatively even expression for all 14-3-3 isoforms. By contrast, the preys are proteins naturally expressed by the cells, so that the distinct 14-3-3 preys should be present in different amounts, depending on their intrinsic levels of expression in the host cells. In BioPlex, six out of seven 14-3-3 isoforms (with the exception of ε) were among the tested recombinant bait proteins. This allowed us to distinguish, among the 14-3-3 binders, those identified as baits retaining 14-3-3 preys, from those identified as preys retained by 14-3-3 baits. In both situations, the linear correlation of the numbers of 14-3-3 binders with the relative phosphopeptide-binding affinity scale of 14-3-3 proteins remained very strong (R² = 0.96 and R² = 0.90, respectively) (Supplementary Fig 7a).

In further support of these interrelations, the number of 14-3-3 prey-binding baits also indicated correlation with the affinity trend of 14-3-3 isoforms when using data from a recent independent study (https://sec-explorer.shinyapps.io/Kinome_interactions/ and⁴⁹) that used AP-MS to uncover the interactions of more than 300 protein kinases (R² = 0.64) (Supplementary Fig. 7a).

The BioPlex database²⁹ also contains all peptide-spectrum matches (PSM) values for all the preys retained by each and every bait. PSM values bear information about the enrichment of prey proteins on resins carrying particular baits. The higher the PSM values measured for a given prey precipitated by a particular bait, the more enriched the prey, and, therefore, the stronger the affinity of the corresponding bait–prey complex. We retrieved and summed up the PSM values of the 114 preys captured in the BioPlex experiment²⁹ by five different 14-3-3 isoforms. These values show good agreement (R² = 0.66) with the affinity trends of the different 14-3-3 isoforms (Fig. 4h). Even more remarkably, when those 114 individual preys are ranked from their highest to lowest PSM values relative to 14-3-3γ (Supplementary Fig. 7b), one observes the same bi-directional decreasing intensity pattern as seen in our experiments (Fig. 1b) as well as in the low-throughput data from literature (Fig. 4a).

Altogether, these analyses indicate that both the numbers and the PSM enrichment values of partner proteins of 14-3-3 isoforms detected by proteome-wide interactomic studies are remarkably correlated with their relative phosphopeptide-binding affinity trends.

Notably, the level of the overall sequence divergence of 14-3-3 isoforms, using 14-3-3γ as a reference (γ < η < β ≈ ζ < τ < σ < ε; i.e., ε is the most divergent from γ), also correlates very well with their hierarchized affinity differences (Fig. 4i). However, the peptide-affinity trend cannot be explained merely by features of the phosphopeptide-binding regions of the 14-3-3 isoforms, which in fact are identical in all the seven human 14-3-3 proteins (Supplementary Fig. 8). Indeed, even the extreme isoforms on the peptide-affinity scale, 14-3-3γ and 14-3-3σ, have only minor sequence variations in their phosphopeptide-binding grooves and only at their periphery (Supplementary Fig. 8), which are unlikely to dictate the phosphopeptide-binding differences. Interestingly, the sequence divergence trend relative to 14-3-3γ (Fig. 4i) remains conserved when considering diverse sub-regions of the sequence (Supplementary Fig. 8). This indicates that the general target affinity differences arise from fine conformational effects spanning the entire structure, rather than a defined sub-region.

Prediction of cellular 14-3-3/phosphotargets complexomes

14-3-3 proteins are highly expressed. Therefore, their abundances in all human tissues have been reliably quantified. According to the integrated whole human body dataset of the PAXdb (https://pax-db.org and³), 14-3-3ε is the 48th most abundant human protein (2479 ppm) and 14-3-3ζ is the 72nd (1680 ppm) out of 19,949 proteins. Considered as a whole, the cumulated seven 14-3-3 isoforms even rank within the top 20 (i.e., top 0.1%) most abundant human proteins. However, 14-3-3 isoforms are not uniformly distributed across tissues. Each human cell type displays a specific distribution of the 14-3-3 family (Fig. 5a).

We took advantage of the quantified hierarchized affinity profile of 14-3-3 isoforms to build a predictor tool that estimates the fraction of a given phosphoprotein that is engaged with each distinct 14-3-3 isoform (Supplementary Data File 2). As an input, the predictor requires (i) the K_D of that phosphoprotein for at least one 14-3-3 isoform, and (ii) the cellular concentrations of the seven 14-3-3 isoforms and of the phosphoprotein of interest. The concentrations of a given protein species in a given cell type can be roughly estimated from protein abundance databases (such as PAXdb³), by using a simple conversion rule (see Methods).

We used this approach to predict the proportions of each 14-3-3 isoform among the overall 14-3-3/phosphoprotein complexes formed in various tissues, including uterus, rectum, and oral cavity, which are all susceptible to hrm-HPV infection, as well as in five other organs (esophagus, skin, lung, brain, and heart) (Fig. 5b and Supplementary Fig. 9).

Conversely, the free fraction of each phosphoprotein depends on absolute affinity constants. HPV-positive cell lines have been estimated to produce an average of ~1 ng of E6 per 10⁶ cells, corresponding to an approximate intracellular concentration of 25 nM⁵⁰. In a situation where the E6 PBM would be fully phosphorylated, we can estimate, using the predictor, that 87%, 35%, 14%, or 5% of phosphorylated 33E6, 18E6, 16E6 and 35E6, respectively, would be engaged in 14-3-3 complexes in cells containing the average 14-3-3 concentrations found in human cells (estimated from integrated human data in PAXdb database³) (Supplementary Fig. 9).