Allostery in Protein Tyrosine Phosphatases is Enabled by Divergent Dynamics

Abstract

Dynamics-driven allostery provides important insights into the working mechanics of proteins, especially enzymes. In this study we employ this paradigm to answer a basic question: in enzyme superfamilies where the catalytic mechanism, active sites and protein fold are conserved, what accounts for the difference in the catalytic prowess of the individual members? We show that when subtle changes in sequence do not translate to changes in structure, they do translate to changes in dynamics. We use sequentially diverse PTP1B, TbPTP1, and YopH as the representatives of the conserved Protein Tyrosine Phosphatase (PTP) superfamily. Using amino acid network analysis of group behavior (community analysis) and influential node dominance on networks (eigenvector centrality), we explain the dynamic basis of catalytic variations seen between the three proteins. Importantly, we explain how a dynamics-based blueprint makes PTP1B amenable to allosteric control and how the same is abstracted in TbPTP1 and YopH.

Introduction

Correlations between enzyme function and protein dynamics are gaining increasing interest as researchers explain biological function¹, explore therapeutic opportunities², and expand enzymes to new chemistries³. The classical view asserts the role of dynamics in influencing the stability of enzyme active sites⁴, participation in substrate binding⁵, and/or modulation of thermal stability of enzyme:substrate transition states⁶. A more recent view emphasizes the role of concerted amino acid networks in playing a role in catalysis and is inclusive to detailing distal sites that may have the ability to modulate enzyme function^{5, 7–10}. While the exact nature of protein dynamics’ contribution to enzyme catalysis can be debated^11–12, it is apparent that cell signaling makes great use of protein dynamics, both enzymatically and otherwise¹³. Signaling enzymes, including protein kinases and protein phosphatases, are allosterically modulated, often exploiting dynamics-based components of their catalytic properties^{8, 14–18}. In this study we focus on identifying the dynamics-based distinctions between three Protein Tyrosine Phosphatases (PTPs) that share a near-identical active site but are varied in their catalytic rates.

PTPs are a superfamily of protein phosphatases^19–20, that are principally responsible for the cleavage of phosphorylated tyrosine residues into tyrosine and inorganic phosphate. Of these, the Class I Classical PTPs (hereinafter referred to as just PTPs) are well-appreciated for their crucial roles in the modulation of human cell signaling^21–22 and are targets of much scrutiny into their atomic details of function and regulation^23–25. Recent work regarding the mobile loops that make up the canonical active site as well as the dynamic allostery inherent in Class I Classical PTPs highlights their prominence as models for investigating the impact of conformational dynamics on catalysis^{15–16, 24, 26–37}. The conserved PTP domain is defined by a set of 10 conserved motifs²⁰ (M1–M10) (Figure 1A, 1B, S1A, S1B); including 6 structural motifs (M2–M7) that form the core of the domain and 4 active site motifs (M1, M8–M10) that play a direct role in catalysis (Figure 1B). M1, the phosphotyrosine (pY)-binding loop or pY-loop, is defined by the sequence NxxKNRY/F and contains an aromatic residue (Y46 in PTP1B, Y51 in TbPTP1 and F229 in YopH, (Figure 1E,1F,1G)) for binding the incoming pY substrate³⁸. M8, the WPD-loop, is defined by the sequence (Y/F)xxWPDxGxP and houses the general acid catalyst aspartic acid (D181 in PTP1B, D199 in TbPTP1 and D356 in YopH, (Figure 1E,1F,1G)) required for catalysis^39–40. This loop can adopt multiple conformations such that open conformation(s) keep the enzyme in an inactive state, while the closed conformation is catalytically competent⁴¹. Rate of opening-closing of this loop is linked to PTP activity^{29, 31, 36} and blocking its closure by allosteric modulators is an effective strategy to inhibit PTP function^{15–16, 42}. M9, the Phosphatase-loop or P-loop, is defined by a 13 amino-acid sequence containing the HCx₅R motif that is a signature of a PTP active site⁴³. It contains the nucleophilic cysteine residue (C215 in PTP1B, C229 in TbPTP1 and C403 in YopH, (Figure 1E,1F,1G)). that is stabilized in a deprotonated form for efficient catalysis^44–45. Finally, M10, the Q-loop, is defined by the sequence QTxxQYxF that contains a glutamine residue (Q262 in PTP1B, Q275 in TbPTP1 and Q446 in YopH, (Figure 1E,1F,1G)) important for coordinating catalytic waters at the PTP active site^45–46. The outer periphery of the PTP active site is marked by a conserved Glutamate²⁰ (E115 in PTP1B, E126 in TbPTP1 and E290 in YopH) that resides in a sequence variable E-loop (Figure S18). While the E-loop is not a part of the ten motifs detailed above, its functionality has garnered recent interest in PTP catalysis¹⁵ and redox regulation⁴⁷.

Figure 1.

PTP domain organization and an overview of studied systems. (A) Sequences of the conserved PTP motifs as seen in PTP1B, TbPTP1, and YopH. Catalytic residues are highlighted in gray (PTP1B), red (TbPTP1), and magenta (YopH). Sequence highlighted in Cyan show divergence from the canonical motif. (B) Ten conserved motifs mapped to the catalytic domain of PTP1B. (C) A) Superimposition of the three starting structures for PTP1B (PDB ID: 1PTV in gray), TbPTP1 (PDB ID: 3M4U in red), and YopH (PDB ID: 2YDU in pink). (D) Kinetic properties of PTP1B, TbPTP1, and YopH for hydrolyzing para-NitroPhenyl Phosphate (pNPP) (obtained from Ref^48–49). (F,G,H) Active site of the PTP1B, TbPTP1, and YopH binding a Phosphotyrosine residue, as seen in the starting structures of our pY-bound simulations. Motifs harboring these active site residues are indicated in teal.

All PTPs use a conserved catalytic mechanism that proceeds in a two-step fashion (Figure S2B)^{22, 50–51}. In the first step, the tyrosine residue of the pY-loop interacts with the substrate pY. The sulfur atom of the deprotonated active site cysteine residue acts as a nucleophile and attacks the substrate phosphorous atom while the protonated aspartic acid residue of the WPD-loop donates it proton to the substrate tyrosine oxygen atom to create a better leaving group. This creates a cysteinyl-phosphate intermediate in the PTP active site, which is hydrolyzed in the second step. Here, the glutamine residue of the Q-loop coordinates catalytic water molecules while the now-deprotonated aspartate of the WPD-loop abstracts a proton, activating the water molecule. This nucleophile then attacks the phosphorous of the cysteinyl-phosphate intermediate, returning the enzyme to its initial state⁵². Given the conserved structure and mechanism of PTPs, it is easy to overlay details of molecular mechanics from one more popularly studied enzyme member to other less studied family members; diversity between members is often overlooked and even discounted while analyzing a family of enzymes. However, the evolution of individual enzyme members has led to distinct catalytic prowess to suit the needs of the organism. How are these differences in catalytic activity achieved in a conserved active site that employs a conserved mechanism? Here, protein dynamics can provide the vital information to connect the gap between sequence, structure, and function paradigm.

In the presented study, we investigate the protein dynamics of the prototypical PTP1B, alongside its two most divergent PTP homologues with a solved structure, YopH from Yersinia pestis⁵³ and TbPTP1 from Trypanosoma brucei⁵⁴ (Figure 1C, S1A, S2A, S2A, S2C, S2D). PTP1B is a mammalian enzyme, ubiquitously expressed in multiple tissues including the liver, skeletal muscle, adipose tissue, and the brain⁵⁵. It is an important participant of the cellular signaling machinery and is most well known as a negative regulator of insulin and leptin signaling^55–56. YopH is a secreted, highly-active PTP, that is a virulence determining factor for the nonhuman Yersinia pestis⁵⁷. YopH is an intriguing antibacterial drug target for treatment of bubonic plague and gastrointestinal syndromes^58–59. Finally, TbPTP1 controls life cycle events and inhibits differentiation of Trypanosoma brucei in mammalian hosts, preventing premature loss of immune evasion abilities^{49, 60}. PTP1B, TbPTP1, and YopH share a conserved PTP domain that superimposes with a cumulative RMSD of ~2.5Å over their Cα backbone structures (Figures 1a, S1). TbPTP1 and YopH, being most divergent from PTP1B, have distinctions in sequence from PTP1B throughout their sequence. These include variations of the E-loop, between M6 and M7, as well as some positions on the WPD-loop (Figures 1A, S1A, S1B). These variations in YopH and TbPTP1 mirror the non-conserved nature of these same positions in the PTP family as garnered by a sequence alignment of the PFAM PTP family (Figure S2A, S2C S2D). Some protein specific regions of the three include the secondary pY binding site in PTP1B⁶¹, the four trypanosome-specific motifs⁶², and the extended E-loop of YopH⁶³ (Figure 1B). PTP1B has an additional helix at the C-terminus of the PTP domain. This helix, termed the α7 helix, is ordered in WPD-loop closed structures of PTP1B and disordered otherwise¹⁶. Despite the small differences, the active site of the three proteins is identical (Figure 1E,1F,1G) and uses the exact same two-step mechanism for hydrolysis of the phosphotyrosines^{26, 60, 64}.

The sharing of a highly conserved fold and catalytic motifs between the three proteins is somewhat incongruous with their vastly different catalytic activities (Figures 1D). YopH is the most active known PTPs with a k_cat of 600s⁻¹ towards para-nitrophenol phosphate (pNPP)⁴⁸, a pan-PTP substrate that mimics a pY residue. TbPTP1 is also highly active with a k_cat of 375s⁻¹ towards pNPP⁶²; PTP1B, while quite active compared to other human PTPs, has a relatively low k_cat of only 25s⁻¹ for pNPP⁴⁸. All three enzymes have an affinity (K_m) for pNPP (Figure 1D) in the low millimolar range as pNPP accesses their active sites without a contribution from protein surface residues. With a conserved fold, catalytic residues, and catalytic mechanism, this disparity between the activities of the three enzymes can only be explained by a difference in their dynamics. Our investigation into the evolutionarily diverse members of the PTP family, PTP1B (human), YopH (Y. pestis), and TbPTP1 (T. brucei) seeks to understand the fast-timescale motions of these enzymes using molecular dynamics (MD) simulations of both a pY-bound and ligand-free apo form (total six states). We analyze the essential dynamics of the six states using Principle Component Analysis (PCA)⁶⁵ and their dynamics-based allostery by network analysis^66–68. We characterize protein communities using the Girvan Newman approach^{5, 8, 69} for each state and identify the most influential nodes using Eigenvector centralities^70–71. In comparing these three divergent enzymes, the similarities in dynamic features constitute the base requirements for PTP functionality, while the differences give rise to the variation in catalytic rates.

Results and Discussion

Essential Dynamics as seen in TbPTP1, PTP1B and YopH

Principal component analysis (PCA) and root-mean-square fluctuations (RMSF) of the peptide backbone and Cα atoms display the overall motions of the three enzymes in their apo and pY-bound states (Figures 2, S9–18). In PTP1B apo state, the E-loop is the most mobile structural element, the motion of which is seen in the first 5 modes of PCA (Figure 2A, top). Mode 1 of the PCA shows concurrent motions of the E-loop and pY-loop, encompassing a ~4Å and ~2Å backbone movement, respectively. Furthermore, the RMSF of the Cα atoms of the α7 helix does not show increased motion in the apo state, instead remaining structured despite the opening of the WPD-loop, likely due to the nature of helix unfolding occurring on a much longer timescale than the course of this simulation (Figure S18). This provides a unique perspective of PTP1B in an immediately pre- or post-catalytic state. The PTP1B+pY system shows similar overall motions as apo PTP1B, but with decreased magnitude (Figure 2A, bottom). The E-loop motions only encompass 3 of the first 5 modes of the PCA, with mode 1 showing a similar concurrent motion between the E- and pY-loops. Mode 2 shows a concurrent motion of several structural regions of PTP1B, including the pY-, E-, and WPD-loops, as well as a motion of similar magnitude in loop 16, which lies sequentially between the WPD- and P-loops (Figures S10–S13).

Figure 2.

Essential dynamics, C_α backbone fluctuations and Principal Component Analysis. Putty representation of C_α root-mean-squared fluctuations (RMSF) seen in the simulations of PTP1B(A), TbPTP1(B) and YopH(C). Corresponding panels also show RMSF plots for the first 5 eigenvectors of PCA for PTP1B(A), TbPTP1(B) and YopH(C) systems. Positions for catalytic motifs are highlighted: pY-loop (yellow), E-loop (red), WPD-loop (orange), P-loop (green), and Q-loop (blue). For the overlay of first 5 individual modes of motion for each system and a Cosine similarity index for each pair of vectors, see Supporting Information Figures S12–18.

In the nonhuman PTPs, motions seem generally more conserved than in PTP1B. In TbPTP1 (Figures 2B, S14), the sole difference between the apo and pY-bound states is a striking rigidity of the E-loop, which becomes much more mobile in the presence of the ligand and contributes to modes 1 and 2 of the PCA. Other fluctuations are almost identical in location, differing only in magnitude of the motion in the PCA. YopH displays a highly mobile pY-loop in both the apo and pY-bound states, contributing to all of the first 5 PCA modes (Figures 2C, S15). Additionally, the loop 16 region of YopH apo is highly mobile, showing a ~4Å displacement in mode 1. The apo form of YopH displays a highly mobile WPD-loop where Cα fluctuations peak in the residues Gln357, Thr358, and Ala359 (Figures 2C, S17). This increased mobility is likely due to the lack of Pro residues in the sequence of the WPD-loop as seen in PTP1B and TbPTP1, and could partially explain the substrate promiscuity typical in YopH⁷². Also, much like in PTP1B and TbPTP1, both YopH systems show very high fluctuations in a region adjacent to the pY-loop (Figures 2A,2C, S13, S15). However, due to their prevalence both apo and pY-bound systems, it is unlikely that these fluctuations play a role in regulating catalysis. Instead, due to their proximity to the pY-loop, it is possible that these highly mobile regions are important for interactions with broader protein substrates in a region distinct from the substrate pY residue.

To further compare the two states of each enzyme, we have calculated the cosine similarity index for each pair of their first five eigenvectors (Figure S16). This yields a value of 1 if perfectly parallel, −1 if perfectly antiparallel, and 0 if orthogonal. Due to the high dimensionality of these vectors, the highest value seen was at 0.46 in TbPTP1. The increased similarity of the modes seen in TbPTP1 is likely due to the prevalence of major motions in the loop adjacent to the pY-loop in both states. This implies that these motions are not strictly relevant to the binding or catalysis of the substrate pY itself; instead, due to the proximity of this loop and the pY-loop, it may be important for the binding of the broader protein substrate. Comparatively, in PTP1B and YopH, there is not a similar degree of similarity between the modes. The closest in between vector 5 in both PTP1B+pY and PTP1B-apo. These modes capture inverse motions of the E-loop, with one moving away from the active site and the other moving toward the active site.

Active Site Dynamics Rationalize Kinetic Properties

To assess differences in active site dynamics between the different enzymes, we analyzed the depth of the active site cleft, measuring the distance between the C_α atoms of the catalytic residues of the pY-loop (Tyr51 in TbPTP1, Tyr46 in PTP1B, and Phe229 in YopH), P-loop (Cys229 in TbPTP1, Cys215 in PTP1B and Cys403 in YopH) and the Q-loop (Gln275 in TbPTP1, Gln262 in PTP1B and Gln446 in YopH) over the length of MD simulation runs (Figures 3A, S19–20). In PTP1B, there is a single high-probability density conformation regardless of state. This is not the case in either TbPTP1 or YopH. In TbPTP1, a small conformational change is seen in the Q-C distance, showing a small deformation in the Q-loop. This is also seen in YopH, with the addition of the pY-loop motions increasing the Y-C distance in the apo form.

Figure 3.

Active site dynamics. (A) Active site loops as observed in the conserved PTP domain. The orientation of the loops is depicted in both Apo and pY bound complexes. (B) Probability density for the distance between the C_α atoms of residues on the Q-loop (PTP1B:Q262, TbPTP1:Q275, YopH:Q446) and P-loop (PTP1B:C215, TbPTP1:C229, YopH:C403) (d1 / y-axis) and residues on the pY-loop (PTP1B:Y46, TbPTP1:Y51, YopH:F229) and P-loop (PTP1B:C215, TbPTP1:C229, YopH:C403) (d2 /x-axis) for PTP1B (top), TbPTP1 (middle), and YopH (bottom). Both Apo (purple) and pY-bound (orange) forms are shown. (C) Probability density for the distance between the C_α atoms of residues of the P-loop (PTP1B:C215, TbPTP1:C229, YopH:C403) and WPD-loop (PTP1B:D181, TbPTP1:D199, YopH:D356) (d3 / y-axis) (d3) and of the pY-loop (PTP1B:Y46, TbPTP1:Y51, YopH:F229) and WPD-loop (PTP1B:D181, TbPTP1:D199, YopH:D356) (d4 / x-axis) for the Apo (purple) and pY-bound (orange) PTP1B (top), TbPTP1 (middle), and YopH(bottom). Darker colors indicate a higher density, or more frequent occurrence, of that pair of distances. Also see Supporting Information Figure S19–20.

The WPD-loop dynamics show a much more distinct conformational change between apo and pY-bound states (Figure 3B, S20). When comparing apo states, TbPTP1 adopts a single high-density WPD-loop open conformation, unlike PTP1B, which samples two similar conformations, or YopH, which shows diffuse density, implying a mobile WPD-loop in the apo form. The additional density seen in the top-right corner of the YopH-apo plot is due to the deformation of the pY-loop, not the hyper-open conformation mentioned in literature¹⁵. Interestingly, the WPD-loop of YopH adopts a single high-density conformation in the pY-bound state, whereas the other two enzymes adopt two high-density conformations, as well as other low-density conformations.

When considering these conformations in the context of the two distinct reaction steps (Figure S2B), they provide insight into the difference in catalytic rates between PTPs. The rates for the two steps, k_cleavage and k_hydrolysis, differ between the enzymes PTP1B and YopH with the substrate pNPP. In PTP1B, k_cleavage is 270s⁻¹ and k_hydrolysis is 28s⁻¹ from one study²⁷. Comparatively, YopH was found to have k_cleavage and k_hydrolysis values of 343s⁻¹ and 76s⁻¹, respectively⁷³. While no such mechanistic studies exist for TbPTP1, the k_cat value for the overall reaction lies in-between PTP1B and YopH (Figure 1D). The conformations of TbPTP1’s catalytic motifs mirror characteristics of both PTP1B and YopH, in which the Q-loop behaves similarly in TbPTP1 and YopH (Figure 3A), but the WPD-loop of TbPTP1 behaves more akin to PTP1B (Figure 3B). The alternative conformation of the Q-loop in YopH and TbPTP1 could allow for increased interactions with the solvent and the recruitment of the catalytic water, increasing the rate of k_hydrolysis, whereas the alternative conformations of the WPD-loop in PTP1B and TbPTP1 may indicate closed, but inactive, conformations that reduce the rate of k_cleavage.

Hydrogen-bond dynamics reveals the conserved role of Arg in the Cx₅R motif

We analyzed the hydrogen bond dynamics at the interface of the WPD-, P-, and E-loops due to the highly polar nature of PTP active sites and to examine the proposed role of the E-loop in WPD-loop dynamics (Figure 4A–C). This analysis considers a hydrogen bond present between two atoms less than 3Å apart with a donor atom–hydrogen atom–acceptor atom angle greater than 135°. When analyzing the prevalence of hydrogen bonds in the active site, through determining the proportion of frames in the trajectory where the criteria is met, we identified dramatic changes in the hydrogen bond network that occur after ligand binding. Due to the possibility of having more than one donor-acceptor pair per residue (i.e., the guanidinium group on arginine and the phosphate group on phosphotyrosine), it is possible that the proportion of frames where an H-bond is present is greater than one.

Figure 4.

Hydrogen bond dynamics at the PTP active site (A, C, E) Hydrogen bond interactions between residues of the P-, WPD-, and E-loops for PTP1B (right, gray), TbPTP1 (middle, red), and YopH (right, pink). The arrow points from hydrogen bond donor to hydrogen bond acceptor, with “bb” indicating the involvement of backbone atoms. The top, black number indicates the proportion of frames in which that hydrogen bond is present in the apo form. The bottom, colored number indicates the proportion of frames in which that hydrogen is present in the pY-bound form, indicating either an increase (green) or decrease (red) in proportion from the apo state. Note that many residues can have more than one donor or acceptor atom, thus the proportion of frames can be greater than 1. (B,D,F) Probability density for the distances observed between the C_ζ of the conserved P-loop arginine (PTP1B:R221, TbPTP1:R235, YopH:R404) as the hydrogen donor and E-loop residues (back bone oxygen of PTP1B:L110, TbPTP1:L121, YopH:L285) as the hydrogen acceptor are shown at the top. Middle panel shows the corresponding probability distributions for the distances observed between the C_ζ of the conserved P-loop arginine (PTP1B:R221, TbPTP1:R235, YopH:R404) as the hydrogen donor and C_δ of the conserved Glutamate of the E-loop (PTP1B:E115, TbPTP1:E126, YopH:E290) as the hydrogen acceptor atom. Bottom panel shows probability distributions for the distances observed between the C_ζ of the conserved P-loop arginine (PTP1B:R221, TbPTP1:R235, YopH:R404) as the hydrogen donor and backbone oxygen of the conserved Glutamate of the WPD-loop (PTP1B:W179, TbPTP1:W197, YopH:W354) as the hydrogen acceptor atom. Both Apo (purple) and pY-bound (orange) forms are shown are shown for the three PTPs. Also see Supporting Information Figure S21 for the dihedral plots of the P-loop arginine (PTP1B:R221, TbPTP1:R235, YopH:R404).

This analysis highlights the importance of the P-loop arginine residue (R235 in TbPTP1, R221 in PTP1B and R405 in YopH) that is a part of the conserved HCx₅R active site motif. In all three enzymes in the apo state, this arginine makes a prevalent H-bond with the backbone of a conserved leucine at the base of the E-loop (L121 in TbPTP1, L110 in PTP1B and L285 in YopH) (Figure 4A–C). This arginine also makes H-bonds with the conserved Glu of the E-loop (E126 in TbPTP1, E115 in PTP1B and E290 in YopH), but the prevalence of this interaction is much higher in YopH (1.58) and TbPTP1 (1.84) when compared to PTP1B (0.39). This interaction seems important for the mobility of the E-loop. Upon ligand binding, the proportions for this interaction change to 0.92, 1.38, and 0.15 for PTP1B, YopH, and TbPTP1, respectively, implying an inverse relationship between the prevalence of this interaction and the mobility of the E-loop.

To corroborate the hydrogen bonding analysis, we measured the distances of the Cζ atom of R221/R235/R404 to the corresponding hydrogen bond partner (Figure 4D–F), as well as the dihedral angles of R221/R235/R404 to detect conformational changes of this residue (Figure S21). With the exception of the R221_Cζ –L110_O distance in PTP1B (Figure 4d, top), there are dramatic differences in the distributions of distances. Additionally, the distribution of its sidechain χ-angles shift in the presence of the ligand, particularly in χ₃ and χ₄. Since the Cζ atom of arginine is not actually the hydrogen donor, instead being a proxy for the nitrogen atoms in the guanidinium group, the similar distributions of the PTP1B R221_Cζ –L110_O distance may induce a hydrogen bond in one case, but not another, especially when considering the different χ-angle distributions (Figure S21).

When considering this alongside the H-bonds made between this arginine and the substrate pY, it is possible that the E-loop interaction is important for R221 to adopt a conformation suitable for ligand binding. A key change that occurs in the presence of the ligand pY is the H-bond between the conserved arginine and the backbone tryptophan of the WPD-loop (W197 in TbPTP1, W179 in PTP1B and W354 in YopH) (Figure 3C). This increase in prevalence is seen in all three enzymes in the pY-bound state, and when compared with the H-bonds present between (PTP1B Numbering) R221 and L110 in the apo form, suggests a “switch” mechanism in which a ligand-induced conformational change of R221 creates an H-bond with W179, aiding in the closure of the WPD-loop in all three enzymes. Further interactions are seen between many residues on the E-, WPD-, and P-loops, but differ in trend and prevalence between the enzymes. These residues surely play a role in determining the rate of closure or flexibility of the WPD-loop for that enzyme, but it is likely that these enzymes use a conserved mechanism for the basal WPD-loop closure.

Diverse PTPs contain Distinct Dynamic Networks

We employed a network-based “Violin Model”^{8, 74} to gain a more detailed view of the dynamics of these enzymes. From a cross-correlation-based network that separately determines the correlation for a residue’s main chain and side chain, we utilized the Girvan-Newman algorithm⁷⁵ to detect “communities” of residues. In each community, the nodes (consisting of either the main chain atoms or sidechain atoms) within that community are highly interconnected with more correlated motion compared to residues outside of that community. Detected communities generally align with structural features, and similar communities across systems are colored the same (see Figures 5 & 6). For communities containing the PTP active site loops, the coloring is as follows: WPD-loop (orange); Q-loop (purple); E-loop (dark red); pY-loop (violet); P-loop (yellow). Other similar communities are present in several systems, with the following coloring: α2 helix (red); α3 helix (brown); α4 helix (teal); β4-β7 (lavender).

Figure 5.

Network analysis for the Apo and pY-bound TbPTP1 (top) and YopH (bottom). Like in Figure5, force atlas maps have been colored according to communities obtained from Girvan-Newman clustering (B,D, H, J). Inset of each map shows node with high eigenvector centrality values (C, E, I, K). Community structure (A, F, G, L) shows communities as connected to each other by edges computed on betweenness. Edge values and bridging residues are presented in Supporting Information Tables S2, S3, S5 and S6. Also see Supporting Information Figure S23–24 for community structure of the Apo and pY-bound active sites of TbPTP1 and YopH.

Figure 6.

Network analysis for assessing the role of dynamics-based allostery in PTP1B’s Apo (top) and pY-bound (bottom) forms. Force atlas maps as seen for the two states are shown in (A) and (E). Networks have been clustered and colored into communities using the Girvan-Newman algorithm. Community structure as nodes and edges is shown for the Apo and pY-bound states in (C) and (F). Here, the size of each community corresponds to the number of nodes it contains. Edges connected the communities are weighted (thickness of the connection) based on betweenness. Edge values and bridging residues are presented in Supporting Information Tables S1 and S4. Communities as mapped onto the structures of PTP1B are shown in (B) and (D). Inset of (A) and (E) show the highly weighted of nodes as represented by their eigenvector centrality measures. Also see Supporting Information Figure S22 for community structure of the Apo and pY-bound PTP1B active site.

Community detection for the nonhuman PTPs yielded 11 communities for TbPTP1-apo (Figure 5A), 16 for TbPTP1+pY (Figure 5F), 14 for YopH-apo (Figure 5G), and 11 for YopH+pY (Figure 5L). The partitioning of specific active site features into communities shows a few trends among the nonhuman PTPs, namely the WPD-loop being principally distinct from other active site loops, whereas the residues in the P-loop fall into several different communities, positing this loop as a communication hub the dynamics of PTPs (Figures S22–S24).

To gain a residue-level view of the dynamic network, we employed the ForceAtlas2 algorithm to visualize the network, coloring nodes by their assigned community and weighting the size of the node by its eigenvector centrality (EC) (Figure 5B,D,H,J). This revealed that the nonhuman PTPs share a cluster of influential residues, irrespective of state. These clusters contained residues mostly from the buried core of the PTP fold, belonging to communities for the P-loop (yellow), E-loop (dark red), α2 helix (red), β4-β7 (lavender), and occasionally WPD-loop (orange) (Figure 5C,E,I,K).

In PTP1B, the community partitioning reveals similar architecture to the nonhuman PTPs, with PTP1B-apo containing 15 communities and PTP1B+pY containing 14 (Figure 6C,F). Furthermore, these communities show similar structural distribution (Figure 6B,D). However, on a residue level, the networks diverge substantially. In PTP1B-apo, there is no connection between the WPD-loop and E-loop communities, a feature unique to the system (Figure 6C, Table S1), implying a lack of communication between the dynamics of the two loops. Furthermore, the emergence of the Loop 11 community, containing Y152, a known allosteric residue¹⁶, is unique to the PTP1B+pY system. In all other cases, this region is assigned to communities containing either the WPD-loop or α3 helix (Figures 5, 6, Table S1). Y152 itself is split, its main and side chain nodes being assigned to the Loop 11 and α3 communities, respectively (Table S4).

The network visualization for PTP1B-apo shows a cluster of high EC nodes at the interface of the WPD-loop (orange), α3 (brown), and α6 (dark green) (Figure 6A). Intriguingly, the substrate-bound form of PTP1B shares characteristics of both PTP1B-apo and the nonhuman PTPs. The network reveals two clusters of high-EC nodes (Figure 6E), one similar to the cluster from the nonhuman enzymes in the core of the enzyme, containing α2 and P-loop community residues, and a second, containing WPD-loop and α3 community residues, as seen in PTP1B-apo.

Influential Nodes as Drivers of PTP Behavior

Further interrogation of the EC of nodes within the dynamic network yields a model for the reduced activity of PTP1B compared to TbPTP1 and YopH (Figure 7). In PTP1B, the two distinct clusters are shown as surfaces of atoms whose corresponding nodes have EC > 0.05 (Figure 7A). In PTP1B+pY, the highest EC nodes are seen principally in motifs 4 (residues 81–87), 5 (91–101), 6 (107–111), 7 (120–126), and 9 (210–223), making up the red-outlined cluster on the structure. The second cluster, outlined in blue, is shared in both the apo and pY-bound forms of PTP1B, containing several residues of motif 8 (residues 176–185). In the bound form, this cluster shows a reduced influence, as seen by the lower EC values (Figure 7A, S25). This cluster in PTP1B-apo contains nodes corresponding the sidechains of Y152, Y176, F191, F194, T224, and F269, which have been previously implicated in the regulation of WPD-loop motion in PTP1B^{16–17, 24, 27, 33, 76}. Additionally, the sidechains of L110 and Y176, as well as the main- and sidechains of S190, show influence on the network, and likely play a role in transferring allosteric signals that impact WPD-loop mobility.

Figure 7.

Influential nodes driving dynamics-based allostery in the three PTPs. Eigenvector centrality plots for the Apo and pY-bound PTP1B (top, A), TbPTP1 (middle, B), and YopH (bottom, C) are shown for each backbone (C_α) (orange) and side chain (C_β) (purple) node. Nodes with EC > 0.10 are shown. High eigenvector centrality nodes are mapped onto the structures of the three PTPs. TbPTP1 and YopH show a cluster surrounding the E-loop in both the Apo and pY-bound forms. PTP1B shows a distinct influential cluster around the base of the WPD-loop in both Apo and pY-bound forms. In the pY-bound form, PTP1B recapitulates the influential residues as seen around the E-loop in TbPTP and YopH. Also see Supporting Information Figures 25–26.

In the present simulations, the α7 helix remains structured, likely due to the longer timescales required for helix folding/unfolding as compared to loop motions (Figure S18). This means that the identified high EC nodes are influential in the context of a structured α7 helix. When this helix is deleted, the catalytic activity is reduced; the same effect is seen in the literature for mutations of several residues with high EC values (Table 1)^{16–17, 24, 33, 77}.

Table 1:

PTP1B variants showing change in percent k_cat of the wildtype

Variant	% kcat	Reference
M109A	13.04	Hjortness 2018
Y152A	65.91	Choy 2017
Y152F	72.73	Choy 2017
Y176G	31.82	Choy 2017
Y176A	23.21	Cui 2017
W179A	0.21	Cui 2017
P180A	158.93	Cui 2017
P185G	1.82	Choy 2017
P185A	55.36	Cui 2017
F225I	59.65	Torgeson 2022
F225L	66.67	Torgeson 2022
F225Y	178.07	Torgeson 2022
Δα7	63.64	Choy 2017
Δα7	75.44	Torgeson 2022

Conversely, the high-EC clusters seen in TbPTP1 (Figure 7B) and YopH (Figure 7C) are likely the residues responsible for the basal, catalytic dynamics of the PTP fold. These nodes lie within motifs 4, 5, 6, 7, and 9, similar to one of the clusters seen in PTP1B+pY. While no residues in the variable region of the E-loop, located between β3 and β4, show high EC values, nodes located at this loops base, such as M120, L124, Y135, and W136 (TbPTP1) and L282, Y301, and F302 (YopH), are highly influential. This implies an indirect effect of the E-loops motion of catalysis, although no direct influence of these dynamics was seen in this study. Furthermore, the high degree of similarity between the two states of the nonhuman PTPs likely play a role in the increased catalytic rate seen in YopH and TbPTP1 (Figure S25).

Conclusions

As crucial regulators of cell signaling, PTPs have been the subject of extensive research to characterize their functionality and activity. However, this work has illuminated the seemingly paradoxical dichotomy of vastly different enzymatic activity with a highly conserved fold, which posits the dynamic motions of PTPs as the mechanism for divergent catalytic rates. As such, the dynamics of PTP1B and YopH have been the subject of numerous studies, both in the context of catalytic activity^{15, 24, 26–32} and in PTP1B’s allosteric regulation^{16, 24, 33–35}, which likely exploits the fast-timescale motions of PTPs.

In this work, we examine the changes in protein dynamics that occur between the apo and pY-bound, precatalytic states in three evolutionarily diverse PTPs. We explore the changes in backbone RMSF and PCA (Figure 2) seen in active site features, such as the WPD-, pY-, and E-loops, that occur upon substrate binding. We note an increased flexibility of the pY-loop of YopH, explaining the enhanced substrate promiscuity of that enzyme, as well as increased mobility of the WPD-loop in the open state. The mobility of the E-loop depending on state is contrasted between PTP1B and TbPTP1, where the human PTP1B is more mobile without a ligand, whereas the T. brucei TbPTP1 E-loop is rigid in the apo state, mirroring the more active YopH.

Through analysis of the probability distributions of active site distances (Figure 3), we detected an alternative conformation of the active site loops in the more active PTPs, possibly creating the increased rate of the second catalytic step. Relevant to the rate of the first step, we saw multiple high-density conformations for the WPD-loop in PTP1B and TbPTP1, explaining the depressed catalytic rate compared to YopH. We also analyzed the hydrogen bonding network (Figure 4) of the highly basic PTP active site, identifying a hydrogen bond switch in the R221-L110-E115/R221-W179-E115 interaction (PTP1B numbering) that selectively stabilizes either the WPD- or E-loops.

Community detection and network analysis from the cross-correlation network (Figures 5, 6) solidifies the similar dynamics of the nonhuman PTPs and elucidates divergent motions in PTP1B. The presence of conserved high-EC clusters in both states of TbPTP1 and YopH, as well as in PTP1B+pY, point to motions required for catalysis in the PTP fold. However, the identification of an allosteric cluster of high-EC nodes in PTP1B-apo and PTP1B+pY, reveals the mechanisms by which PTP1B is autoregulated by its dynamics.

The identification of influential nodes through the eigenvector centrality (Figure 7) illuminates potential regulatory mechanisms that have evolved in mammalian PTPs. In the nonhuman PTPs TbPTP1 and YopH, the distribution of influential nodes remains mostly consistent between the apo and pY-bound states. Particularly, residues in motifs 6 and 7, which flank the E-loop, and the histidine residue from motif 9 show high eigenvector centrality. This contrasts PTP1B, which shows different patterns of influential nodes between the two states. In the apo form, residues in and adjacent to the WPD-loop showed high influence. With the pY ligand, the influential nodes appear more similar to YopH and TbPTP1, with the allosteric cluster seen in PTP1B-apo showing decreased EC. In the context of the α7 helix remaining structured during the simulation, the presence of the secondary cluster of high-EC nodes provides a rationale for the decreased catalytic activity in PTP1B mutants with a removed α7 helix.

The catalytic activity of PTPs is clearly influenced by several different phenomena, making it difficult to point to a single conserved mechanism that determines the reaction rate for the dephosphorylation of substrates. However, the present work identifies conserved protein dynamics in the hydrophobic core of three diverse PTPs that is required for basal catalysis and is directly tied to the hydrogen bonding network responsible for WPD-loop closure through conserved positions such as L110 and R221. In PTP1B, the presence of a additional influential motions rationalizes the decreased activity, as certain conformations must be adopted in order for WPD-loop closure to occur, as has been shown in other studies for residues Y176, F191, I261, and F269^{16–17, 27, 77}

Taken altogether, the dynamic landscapes of these three enzymes mirror an evolutionary pressures behind their function. PTP1B is a crucial signaling regulator in human cells, modulating the action of kinases such as Janus kinase (JAK) and the insulin receptor (IR), and as such, is heavily regulated through a various mechanisms⁷⁸. On the opposite end of the spectrum is YopH, a secreted virulence factor from the plague-causing Yersinia pestis⁷², whose high activity and broad substrate promiscuity benefits its expressing organism^{73, 79}. TbPTP1 lies somewhere in the middle of this spectrum. As an enzyme important for preventing premature life cycle progression of Trypanosoma brucei^{54, 62, 80}, TbPTP1 must be highly active until it is no longer required, at which point mechanisms that are less nuanced than their mammalian counterparts can regulate this enzyme⁸¹. Our studies show that the dynamics of these enzymes portions their responsibilities of being critical catalysts with regulatory control. Motions or interactions that are precisely conserved across the three PTPs likely enable basal catalysis; less similar processes create the differences in catalytic activity, and the divergent motions determine how these enzymes are regulated in their specific cellular environments. Future work should investigate dynamics with these concepts in mind, to determine precise mechanisms for loop closure and transition state stabilization between PTPs, but also to determine how human PTPs could be targeted therapeutically through exploiting inherent dynamic processes.

Materials and Methods

Multiple Sequence Alignment and Phylogenetic Analysis

The sequence conservation of PTPs was assessed through a multiple sequence alignment of the PFAM⁸² family entry y_phosphatase (ID: PF00102) with the exclusion of sequences <200 amino acids to exclude PRL phosphatases, which yielded ~42,000 sequences. The database was aligned using MUSCLE⁸³ with the Super5 algorithm. The alignment was analyzed using Jalview⁸⁴, which also calculated the property conservation⁸⁵ and percent consensus at each residue position. The sequences of solved structures of PTPs were also attained from the PFAM database and aligned with MUSCLE, upon which a phylogeny was built using the Phylogeny.fr server⁸⁶ with the BioNJ distance algorithm⁸⁷ and 100 bootstrap steps.

System Preparation for Molecular Dynamics

This study investigated three enzymes: Trypanosoma brucei TbPTP1 (PDB ID: 3M4U)⁵⁴, human PTP1B (PDB ID: 1PTV)⁶¹, and Yersinia pestis YopH (PDB ID: 2YDU)⁵³. Missing loop regions from these structures were built using MODELLER⁸⁸. For PTP1B, the active site C215S mutation present in the crystal structure was reverted to cysteine. For pY-bound systems, the pY ligand was aligned in the pocket using the pose present in the PTP1B crystal structure (PDB ID: 1PTV) prior to initial minimization, which was done in the gas phase for 2000 steps of steepest decent (SD) and 200 steps of conjugate gradient (CG) minimization in UCSF Chimera⁸⁹. Model quality following minimization was assessed using PROCHECK⁹⁰. Protonation states for most residues were assigned using PROPKA^91–92, with the exception of the catalytic and active site residues aspartic acid (protonated) (TbPTP1: D199; PTP1B: D181; YopH: D356), cysteine (deprotonated) (TbPTP1: C229; PTP1B: C215; YopH: C403), and histidine (doubly-protonated) (TbPTP1: H228; PTP1B: H214; Yoph: H402) in order to simulate the enzymes in a catalytically competent state according to biochemical requirements and previous studies⁴⁷.

Molecular Dynamics Simulations

All MD simulations were performed using AMBER20⁹³ with the ff14SB force field⁹⁴. The phosphorylated tyrosine (pY) ligand was handled by the phosaa14SB force field⁹⁵. Hydrogen bonds were constrained using the SHAKE algorithm⁹⁶ and the Particle mesh Ewald summation was used to handle long-range electrostatics⁹⁷. Solvation was performed using the TIP3P⁹⁸ water model with 12 Å padding from the periodic boundary. A constant temperature of 300 K was maintained using Langevin dynamics⁹⁹ and constant pressure maintained by Berendson’s barostat¹⁰⁰.

Minimization following solvation was performed first on the solvent and ions only by restraining the solute for 2700 steps of SD and 300 steps of CG. This restraint was gradually relaxed in phases of 50 kcal mol⁻¹ Å⁻², 20 kcal mol⁻¹ Å⁻², and 5 kcal mol⁻¹ Å⁻² for 8000 SD steps and 2000 CG steps. The system was then gradually heated from 0 K to 300 K in six steps of 50 K with a solute restraint of 5 kcal mol⁻¹ Å⁻² under the NVT ensemble. Further minimization was then performed by reducing the solute restraint to 2, 0.1 and 0.05 kcal mol⁻¹ Å⁻² for 8000 SD and 2000 CG steps per restraint value under the NVE ensemble. Following this, the system is equilibrated in NPT with restraints of 0.5, 0.1, 0.04, and 0.01 kcal mol⁻¹ Å⁻² for 50, 100, 400, and 400 ps. Finally, the system undergoes a 10ns unrestrained run with a 2fs timestep before the 255ns production run. 100ns was discarded from the beginning of each run to avoid equilibration artifacts based on analysis of structural factors (Figures S3–S8). A total of 4 replicates was run for all six systems (three enzymes, with and without pY ligand), yielding a total of 620ns data for each system.

Trajectory Analysis

All analysis of MD trajectories was performed in CPPTRAJ¹⁰¹ unless stated otherwise. Standard structural factors of root-mean-squared deviation (RMSD), radius of gyration (R_G), and solvent-accessible surface area (SASA) were analyzed to determine the equilibration of each system (Figures S4–S8). Principal component analysis (PCA) was performed on the C_α atoms of the core PTP domain of each system, defined by the residues 14–291 (TbPTP1), 5–282 (PTP1B), and 184–391 (YopH). Eigenvector similarity was determined in Python by calculating the cosine similarity between vectors $A$ and $B$:

$${\text{S}}_{\text{c}}=\frac{{\sum }_{\text{i}=1}^{\text{n}}{\text{A}}_{\text{i}}{\text{B}}_{\text{i}}}{\sqrt{{\sum }_{\text{i}=1}^{\text{n}}{\text{A}}_{\text{i}}^2}\sqrt{{\sum }_{\text{i}=1}^{\text{n}}{\text{B}}_{\text{i}}^2}},$$

where $A_i$ and $B_i$ are components of vector $A$ and $B$, respectively. Structure figures were made using PyMOL, graphs were made using the python based visualization packages MatPlotlib¹⁰² and seaborn¹⁰³.

Network construction and Community analysis

The construction of the dynamic network was performed using the networkView¹⁰⁴ plugin of VMD¹⁰⁵ as has been reported elsewhere^66–68. The network is described as a set of nodes and edges, the nodes consisting of all amino acid residues split into two: the α node, containing the C_α, C, O, and N atoms of the main chain, and the β node, containing all heavy atoms of the sidechain. An edge is created between every pair of nodes if any heavy atoms of the two nodes are within a contact distance 4.5 Å. The edge weight, $w_{ij}$, is defined as a transformation of the dynamic cross correlation: $w_{ij}=-\ln \left|C_{ij}\right|$. A final restriction that no edge is created between nodes belonging to the same residue is applied to the adjacency matrix before further analysis. This weighted adjacency matrix is used in a Python NetworkX¹⁰⁶ implementation of the ForceAtlas2 algorithm¹⁰⁷ for network visualization and layout.

Community analysis was performed to detect groups of nodes that are highly interconnected and intercorrelated among each other, and not with the nodes of other groups. The Girvan-Newman algorithm⁷⁵ was employed to determine the community structures within the network. This algorithm functions in a top-down manner, iteratively removing edges that have the highest edge betweenness, which is the number of shortest paths between every pair of nodes that pass through a given edge. As edges are removed, the network is remembered whenever the number of communities, or groups of nodes that are disconnected from the rest of the network, increases. Once every edge is removed, the assignment of communities to nodes is selected that maximizes modularity, a measure of the probability difference of intra- and inter-community edges. Graphs were made using the open source package Gephi¹⁰⁸.

Eigenvector centrality (EC) was calculated to identify influential nodes within the network. EC serves as a powerful method to integrate direct centrality (measure of direct connections) and information flow through the nodes¹⁰⁹. This method is recently gaining traction for defining protein allosteric networks⁷⁰, protein-ligand interaction¹¹⁰ and locating influential nodes that determine protein stability and function^111–112. EC for a node $x$ is defined as:

$$\text{EC}\left(x\right)=\frac{1}{\lambda }{\sum }_{y\to x}w\cdot EC\left(y\right),$$

where $w$ is the edge weight, $y\to x$ indicates all nodes $y$ which are adjacent to $x$ in a graph $G=\left(V,E\right)$. This can be rewritten as $\lambda e=Ae$, in which $A$ is the weighted adjacency matrix, $\lambda$ is the largest eigenvalue of $A$, and $e$ is the EC of all nodes, such that the $i$th value of $e$ is the EC of the $i$th node in $V$¹¹³. The EC of every node $x$ is hence a measure of long-range interactions and connected to all other residues of the protein. Due to the recursive nature of the formula, the power iteration method included in the Python package NetworkX¹⁰⁶ was used with a convergence cutoff of 10⁻⁶. While ECs can be calculated per eigen solution arising from eigendecomposition of the adjacency matrix, only the solution corresponding to the highest eigenvalue is considered^{70, 114}. This is in accordance with the Perron-Frobenius theorem that attributes uniqueness to the leading eigenvalue and its corresponding eigenvector calculated for a non-negative real square matrix (such as the adjacency matrix $A$)^{70, 114–116}.

Data and Software Availability

Molecular dynamics topologies (.prmtop), initial coordinates (.inpcrd), and trajectories (.nc) of the four replicates of 155ns as well as input files and an example script for running simulations in AMBER20 are available for download from https://doi.org/10.5281/zenodo.8364009.