Calcium mediated static and dynamic allostery in S100A12: Implications for target recognition by S100 proteins

Abstract

Structure and functions of S100 proteins are regulated by two distinct calcium binding EF hand motifs. In this work, we used solution‐state NMR spectroscopy to investigate the cooperativity between the two calcium binding sites and map the allosteric changes at the target binding site. To parse the contribution of the individual calcium binding events, variants of S100A12 were designed to selectively bind calcium to either the EF‐I (N63A) or EF‐II (E31A) loop, respectively. Detailed analysis of the backbone chemical shifts for wildtype protein and its mutants indicates that calcium binding to the canonical EF‐II loop is the principal trigger for the conformational switch between ‘closed’ apo to the ‘open’ Ca²⁺‐bound conformation of the protein. Elimination of binding in S100‐specific EF‐I loop has limited impact on the calcium binding affinity of the EF‐II loop and the concomitant structural rearrangement. In contrast, deletion of binding in the EF‐II loop significantly attenuates calcium affinity in the EF‐I loop and the structure adopts a ‘closed’ apo‐like conformation. Analysis of experimental amide nitrogen (¹⁵N) relaxation rates (R ₁, R ₂, and ¹⁵N–{¹H} NOE) and molecular dynamics (MD) simulations demonstrate that the calcium bound state is relatively floppy with pico–nanosecond motions induced in functionally relevant domains responsible for target recognition such as the hinge domain and the C‐terminal residues. Experimental relaxation studies combined with MD simulations show that while calcium binding in the EF‐I loop alone does not induce significant motions in the polypeptide chain, EF‐I regulates fluctuations in the polypeptide in the presence of bound calcium in the EF‐II loop. These results offer novel insights into the dynamic regulation of target recognition by calcium binding and unravels the role of cooperativity between the two calcium binding events in S100A12.

1. INTRODUCTION

Biological sensors such as proteins require controlled modulation of their structure and/or dynamics to execute cellular signaling. Such dynamic or static modulation of the conformational space often entails cooperative interactions between different domains of the polypeptide. S100 family of proteins are exemplary models of regulatory molecules that undertake distinct architectural transformation to achieve target specificity upon calcium binding (Donato et al., 2013; Foell et al., ²⁰⁰⁶). Albeit 25%–60% sequence similarity, each S100 protein exhibits unique target selectivity triggered by calcium binding, which allows these proteins to regulate a diverse array of intra‐ and extracellular activities. S100 proteins execute these functionalities with a pair of EF‐hand motifs that assemble to form a four‐helix bundle. The EF loop at the C‐terminal (EF‐II) is a canonical 12 residue motif found in all calcium binding proteins, while the N‐terminal loop (EF‐I), composed of 14 amino acids, is specific to the S100 family (Grabarek, 2006). This pseudo‐EF loop is distinct from its canonical counterpart. Unlike the pentagonal bipyramidal Ca²⁺ coordination afforded by the sidechain carbonyl groups in the EF‐II loop, binding of calcium ions in the EF‐I loop takes place primarily via backbone carbonyl groups. While structurally distinct, the EF loops are spatially coupled allowing them to cooperatively regulate the structure and functions of S100 proteins (Szebenyi & Moffat, 1986). However, the contributions of the individual calcium binding domains and their extent in the exerted cooperativity is not fully understood. Furthermore, the functional relevance of biologically unique EF‐I loop remains unclear.

Most S100 proteins exist as (homo)dimers, allowing them to ligate four calcium ions per dimeric unit. A hallmark of S100 proteins is the distinct conformational rearrangement upon calcium binding, which exposes a hydrophobic surface known to be the site of target binding (Nelson & Chazin, ^1998a). Within the four‐helix bundle, the EF‐I loop tethers helices 1 and 2, while the EF‐II loop connects helix 3 and 4. The linker connecting helix 2 and 3 or the so‐called ‘hinge domain’ is the putative site for target binding in many S100 proteins. In calcium saturated state of S100 proteins, helix 3 undergoes a large reorientation compared to the apo protein, while the orientation of other helices remains relatively unchanged. Such independent reorganization of the different domains is unique to the S100 family of proteins (Grabarek, 2006; Potts et al., ¹⁹⁹⁵). Several seminal works in the literature have evaluated the cooperativity of calcium sequestration and the subsequent implications on structure and dynamics for some S100 and related proteins (Smith et al., 2012; Nelson & Chazin, ^1998b; Immadisetty et al., ²⁰²¹; Akke et al., ^1993a; Inman et al., ²⁰⁰¹; Pálfy et al., ²⁰¹⁶; Akke et al., ^1993b; Lee et al., ²⁰⁰⁰). So far we know that the high sequence and structural homology between the S100 family of proteins underlies a generic mechanism of function mediated by calcium binding, but the origin of the specificity of individual S100 proteins towards biological partners is not well understood. Analogous to other functionally versatile but sequentially conserved domains, the S100 family has also evolved unique mechanisms to expand the molecular recognition space (Lee et al., 2008; Bertini et al., ²⁰¹³; Ecsédi et al., ²⁰²¹; Wheeler & Harms, ²⁰²¹).

S100A12, a homodimeric member of the S100 family, is responsible for both intra‐ and extracellular regulatory functions. In addition to calcium buffering, S100A12 also sequesters Zn²⁺ ions from invading pathogens in the extracellular milieu during infection (Donato et al., 2013; Goyette & Geczy, ²⁰¹¹; Cunden & Nolan, ²⁰¹⁸; Zackular et al., ²⁰¹⁵). Zn²⁺ and Ca²⁺ binding to this protein is cooperatively coupled, but the atomistic details of this cooperativity are currently not available (Dell'Angelica et al., 1994; Moroz et al., ^2009b; Cunden et al., ²⁰¹⁶). S100A12, with its Zn²⁺ sequestration mediated antimicrobial functions is a prominent member of the human innate immune response along with S100A8/A9 heterodimer. We have previously reported on structural characterization of divalent metal ion bound S100A12 (Wang et al., 2019), augmenting existing structural information on this protein (Moroz et al., 2001, 2002, ^{2003b, 2009a}; Hung et al., ²⁰¹³). Additionally, we have demonstrated that calcium binding mediates dynamic coupling between the two EF‐loops in the micro–millisecond (μs–ms) timescale regime (Wang et al., 2022). Despite extensive characterization of the modular structure, parsing the contributions of the individual EF loops is challenging owing to the cooperativity between the EF loops.

Here, we evaluate the contributions of the individual EF‐loops towards modulating the structure and dynamics of S100A12. NMR characterization of S100A12 variants that selectively bind to calcium in the EF‐I or the EF‐II loop demonstrate that the EF‐II loop is the primary determinant of the overall architecture of the protein upon calcium binding. High frequency fluctuations of the amide bond vectors probed using NMR relaxation measurements provide insights into site specific pico–nanosecond (ps–ns) timescale motions in the presence and absence of bound calcium. The results indicate that calcium binding regulates fluctuations in functionally relevant domains of the protein. Molecular dynamics (MD) simulations recapitulating the experimentally observed high frequency fluctuations enable estimation of conformations entropies originating from individual calcium binding events, thereby providing quantitative insights into the cooperativity exerted by the EF loops. These findings, in addition to providing atomistic consequences of calcium binding to S100A12, also provide clues into general mechanism of target recognition by S100 proteins.

2. RESULTS

2.1. Deletion of calcium binding in the EF‐I loop

Calcium binds to the EF‐I loop via backbone carbonyl groups of S18, K21, H23, and T26 and the sidechain carboxylate moiety of E31. Based on studies on the homologous protein S100B, we hypothesized that a single mutation at position 14 (E31A) may be sufficient to eliminate calcium binding in the EF‐I loop (Markowitz et al., 2005). Calcium binding monitored by isothermal calorimetry (ITC) titrations revealed a single binding event for the E31A variant of S100A12 with a K _D = 20 ± 5 μM, indicating that this mutation indeed eliminates binding to the EF‐I loop while preserving calcium binding in the EF‐II loop (Figure 1a). Dissociation constant for Ca²⁺ binding to the EF‐II loop in S100 proteins typically range between 0.1 and 100 μM (Pietzsch & Hoppmann, 2009; Zimmer & Weber, ²⁰¹⁰) and K _D of ~50 μM has been reported for S100A12 extracted from pig granulocytes (Dell'Angelica et al., 1994). The relatively small variation in the Ca²⁺ binding affinity of E31A compared to the wildtype protein implies the cooperativity between the two sites has limited impact on the higher affinity site. Backbone chemical shift assignments were obtained from triple resonance experiments (3D HNCACB and CBCACONH) and 2D ¹H–¹⁵N HSQC experiments performed on uniformly ¹³C‐ and ¹⁵N‐labeled apo and Ca²⁺‐E31A mutant proteins respectively. We successfully assigned 82/92 and 88/92 residues for both the apo‐ and the calcium bound proteins, respectively. NMR titration performed with the aid of these chemical shift assignments yielded a K _D value in agreement with the ITC measurements (K _D ^NMR = 16 ± 3 μM; Figure 1b).

FIGURE 1.

Calcium binding to the E31A variant of S100A12. (a) Normalized change in heat versus equivalents of calcium added to E31A mutant obtained from ITC, (b) normalized peak intensities for select residues during calcium titration to E31A and site‐specific weighted ¹H, ¹⁵N CSPs (left) and the correlation between ¹H and ¹⁵N^H chemical shifts (right) between wildtype and E31A proteins in the absence (c) and presence (d) of calcium. An overlay of the crystal structures of apo (PDB: 2WCF) and calcium bound (PDB: 1E8A) S100A12 showing the reorientation of helix 3 (H3) upon calcium binding is shown in the top‐right inset. Only one subunit of the dimer is shown for clarity.

In the X‐ray structure of the apo protein, the bidentate carboxylate group of E31 forms a hydrogen bond with a water molecule in the vacant N‐terminal pseudo‐EF loop and S28 hydroxyl group (PDB: 2WCE) (Moroz et al., ^2009a). The residue specific profile of the ¹H, ¹⁵N chemical shift perturbations (CSPs) suggests the point mutation rearranges the configuration of the EF‐I loop but the helical core of the protein is still intact when compared to the wildtype apo‐protein (Figures 1c and S1a). Residue specific plots between backbone amide nitrogen (¹⁵N) and proton (¹H) chemical shifts of apo‐wildtype and E31A proteins exhibits a near perfect correlation, suggesting that the mutation does not alter the overall architecture of the protein (Figure 1c). The perturbation of multiple residues (55–75, 85–90) in the C‐terminal EF‐II domain (Figure 1c) in the mutated protein is consistent with strong coupling between the metal binding sites in the apo state (Wang et al., 2022).

Unlike the wildtype protein where both EF loops are functional, the addition of calcium is expected to disproportionately affect residues in the EF‐II loop from the mutated protein. This assumption is validated by the large deviations observed (>1 ppm) for residues 25–37 in the EF‐I loop and not the EF‐II loop in E31A compared to the wild‐type protein (Figure 1d). This is attributed to the difference between the metal free state in E31A versus Ca²⁺ coordination by EF‐I in the wildtype protein. In the absence of significant CSPs at residues (except E72) in the EF‐II loop directly involved in coordinating the Ca²⁺ ion, we conclude the canonical binding mode is not impacted adversely by the remodeling of the EF‐I loop. The cumulative chemical shift change in each EF loop is sensitive not only to the coordination of Ca²⁺ ion, but also to the extent of coupling between the loops in different coordination states. Therefore, in E31A, calcium binding in EF‐II is expected to perturb some residues (25–35) in the metal free EF‐I loop as seen in Figure S1b. Conversely, the difference in the CSPs of residues 68–70 in the Ca²⁺ bound EF‐II loop (Figure 1d) indicates the vacant EF‐I loop interacts differently with the metal bound EF‐II loop in the half‐saturated state of E31A. We note that the observed chemical shift changes cannot be attributed to inter‐subunit CSP effects because the Ca²⁺ binding sites are ~12 Å apart in the dimeric structure. In both apo‐ and Ca²⁺‐E31A, backbone chemical shift analysis by TALOS (Figure S1) shows the secondary structure in the two EF‐hands is not influenced by the mutation in the EF‐I loop. Additionally, residue specific backbone amide chemical shifts of apo and Ca²⁺ bound wildtype and E31A proteins are highly correlated, suggesting high degree of structural similarity (Figure 1d). Thus, structurally, the half‐saturated state of E31A mutant adopts a conformation closer to the fully saturated Ca²⁺‐bound state of the wildtype protein. To further decouple the role of the two metal binding loops, we generated a complementary mutation which eliminated Ca²⁺ binding exclusively in the EF‐II loop.

2.2. Deletion of calcium binding in the EF‐II loop

Calcium binding in the 12 residue canonical EF‐II loop primarily takes place via sidechain carbonyl groups (Moroz et al., ^2003a) and mutating any of the positions has a drastic effect on Ca²⁺ binding affinity (Liriano et al., 2012). In the EF‐II loop of S100A12, D61, N63, D65, Q67, D69, and E72 are involved in calcium binding. Given the proximity of D61 and E72 to the termini of helix 3 and 4 respectively, these residues were considered unsuitable as candidates for mutation. A better option was the highly conserved asparagine at position 3 in the EF‐II loop, with a flexible sidechain in the apo protein. Therefore, with the goal of eliminating Ca²⁺ binding with marginal impact on the protein structure, we selected the N63A mutation for further characterization by NMR. The ¹H–¹⁵N HSQC spectra of apo N63A and the wildtype protein are nearly superimposable (Figure S2a) which facilitated the assignment of 78/91 backbone amide resonances. The ¹H, ¹⁵N CSPs displayed in Figure 2a indicate the most significant changes map to the EF‐II loop (>0.2 ppm) compared to minor changes in the remaining protein. Like the E31A mutation, remote perturbation in N63A confirms the presence of interactions between the EF‐hand domains (Figure 2a). Two key observations from calcium titrations on N63A include: (i) unlike the wildtype and E31A proteins, high concentrations of calcium (~100 equivalents) were required to fully saturate the protein, and (ii) in the presence of a large excess of calcium, N63A is structurally similar to the apo state (Figure 2b) and the backbone resonances deviate significantly from the wild type Ca²⁺ bound protein in both EF‐hands (Figure 2c). The largest CSPs in the C‐terminal EF hand (helix 3 and helix 4; Figure 2b) can be attributed to Ca²⁺ binding in the mutated protein. NMR titrations monitoring changes in chemical shifts provide an estimated Ca²⁺ binding affinity of ~35 mM at the EF‐I binding site (Figure S2b) which is more than an order of magnitude weaker than the native protein (0.1–1 mM; Figure S5).

FIGURE 2.

Site‐specific weighted ¹H, ¹⁵N CSPs between apo‐wildtype and (a) apo N63A, (b) Ca²⁺‐N63A. (c) Site specific chemical shift perturbations between apo and calcium bound wildtype (blue) and N63A (orange) proteins.

To summarize, the structural comparison between the two mutations and their calcium dependent conformational response presents strong evidence for cooperative interaction between the two sites where the higher affinity EF‐II loop is essential for enabling Ca²⁺ binding in the EF‐I loop. Secondly, the calcium induced conformational switch in the C‐terminal EF‐hand required for ligand interaction is triggered primarily by Ca²⁺ binding in the canonical EF‐II loop while binding in the EF‐I loop has marginal impact on the activation of the protein (Figure 2c).

2.3. Picosecond to nanosecond motions in wildtype S100A12

Motions of the backbone amide N—H bond vectors in picosecond–nanosecond (ps–ns) timescale were probed by ¹⁵N^H longitudinal (R ₁), transverse (R ₂) spin relaxation rates and, steady‐state nuclear Overhauser enhancements (¹⁵N–{¹H} NOE) for apo and calcium bound S100A12. Backbone resonance assignments of these forms of the protein have been previously reported (Wang et al., 2019; Hung et al., ²⁰¹³). With the aid of these assignments, 86 out of 91 resonances in the apo and 83 resonances in the calcium bound protein can be unambiguously assigned in the 2D ¹H–¹⁵N correlation spectra of relaxation measurements. The cross‐peaks assigned to functionally important residues in the canonical EF‐II loop (11, apo; 11: Ca²⁺ bound) and the hinge region (16, apo; 15, Ca²⁺ bound out of 16) exhibited adequate signal‐to‐noise for data analysis. Owing to severe exchange broadening the data for only a subset of residues in the S100 specific EF‐I loop (9, apo; 2, Ca²⁺ bound) were analyzed. R ₁, R ₂, R ₂/R ₁, and NOE versus residue numbers for both states of the protein are presented in Tables S2 and S3. Motional parameters for each state were determined with the extended Lipari–Szabo model‐free formalism (Lipari & Szabo, ^{1982a, 1982b}) implemented in the Relax software (d'Auvergne & Gooley, ^{2008a, 2008b}) from data acquired at two fields, 11.7 T (apo, Ca²⁺), 18.8 T (Ca²⁺), and 21.1 T (apo), respectively.

2.3.1. Apo protein

Relaxation analysis for 15 out of 91 amino acids could not be performed either due to weak or overlapping peak intensities in the 2D ¹H–¹⁵N planes. Most of these residues belong to the EF‐I loop and EF‐II loops. The average values of the generalized order parameter, S ² _avg, for the entire polypeptide chain was 0.83, indicating that the protein backbone is generally rigid. The hinge domain, EF‐II and the C‐terminal region of the protein exhibit small fluctuations in the ps–ns timescale regime while the rest of the polypeptide remains relatively rigid. The available order parameters for residues in the EF‐I suggest that the N‐terminal S100 specific loop is more rigid than the canonical EF‐II loop (Table S1 and Figure 3). The calculations show that except for the C‐terminal, majority of residues undergoing motions such as in the hinge domain and the EF‐II loop, exhibit internal correlation times, ${\tau }_e$ < 100 ps (Figure 3b).

FIGURE 3.

(a) and (b) Site‐specific generalized order parameter, S ² (filled circles) and ${\tau }_e$ (bars) for apo (blue) and calcium bound (red) S100A12. (c) Residue‐wise internal correlation times at slower timescales, ${\tau }_s$, for Ca²⁺‐S100A12.

2.3.2. Ca²⁺‐S100A12

Optimized motional parameters for calcium bound S100A12 were similar to that of the apo protein as determined by the averaged order parameters of the four helices (S ² _helix1 = 0.86 ± 0.04, S ² _helix2 = 0.85 ± 0.05, S ² _helix3 = 0.84 ± 0.04, and S ² _helix4 = 0.87 ± 0.03). Figure 3a shows a comparison of the order parameters of apo and calcium bound protein. While order parameters for most residues in the EF‐I loops are not available, motions in the EF‐II loop were attenuated upon calcium binding. In contrast to the EF loops, calcium binding manifests elevated motions in the C‐terminal residues. Unequivocal increase in backbone fluctuations were monitored for C‐terminal residues such as H85 and T88 in the presence of calcium. Interestingly in the hinge domain, the N‐terminal residues experience increased mobility (E39, L40, I44, and N46), while the C‐terminal residues become more rigid (K48, K50, and I53) with the notable exception of A51 and V52. As shown in Figure 3b, overall, the fast timescale motions (<50 ps) detected in the hinge domain, EF‐II loop and the C‐terminal backbone are comparable to similar amplitude motions in the apo protein. However, unlike the apo protein, significantly slower sub‐nanosecond motions ~500 ps $\left({\tau }_s\right)$ were identified exclusively in the Ca²⁺ bound protein (Figure 3c). These much slower timescale motions (~ 500 ps) identified in the hinge domain and C‐terminal residues are broadly consistent with conformational fluctuations between the calcium bound ‘open’ structures but absent in the ‘closed’ apo state.

2.4. Comparison of faster timescale motions in wildtype and mutants

It is well established that the variation in the heteronuclear {¹H}–¹⁵N NOEs correlates with the dynamic trends represented by the S ² values (Figure S3). Therefore, to evaluate the role of dynamics in the cooperative interaction between the EF loops we measured the backbone amide {¹H}–¹⁵N NOEs for apo and Ca²⁺ bound E31A and N63A mutants. A comparison of site‐specific {¹H}–¹⁵N NOEs of apo and Ca²⁺ bound proteins is shown in Figure 4. Overall, the helices and the Ca²⁺ loops are less flexible compared to the hinge region and the C‐terminal residues. In the wildtype protein, the already dynamic C‐terminal increases further in flexibility upon Ca²⁺ binding along with the hinge region linking the two EF hand domains (Figure 3). This enhanced mobility at the principal ligand binding surface (Xie et al., 2007) appears to be finely tuned by the ability to bind Ca²⁺ at one or both EF loops. For instance, in E31A with Ca²⁺ bound at the EF‐II loop, the dynamic response is very similar to the wildtype protein in contrast to the absence of change in N63A where the EF‐II loop is functionally defective. Collectively, the dynamics agrees with the conclusions drawn from the chemical shift analysis of the mutants, which showed the importance of the coordination state of the EF‐II loop in mediating the conformational transformation of the protein.

FIGURE 4.

Site‐specific {¹H}–¹⁵N NOEs for apo (blue) and calcium bound (pink) S100A12 proteins acquired at 25°C and 11.7 T. For clarity, difference in NOE values between the calcium bound and apo states are shown as sticks in each panel.

2.5. MD simulations of S100A12 and its variants

To gain further insight into the conformational space of the different metal ligated states, we performed MD simulations of apo and calcium bound forms of wildtype and E31A S100A12 proteins. Owing to the structural perturbation introduced by the E31A mutation in both EF‐hand domains (Figure 1c) we also performed MD simulations on a half‐saturated state of wildtype S100A12 bearing Ca²⁺ only in the EF‐II loop (Ca²⁺(EF‐II)‐S100A12). Apo and Ca²⁺ bound wildtype proteins maintained their structural integrity throughout the 0.5 μs simulations. Order parameters were determined from these MD trajectories using variable averaging times ranging between 100 ps and 10 ns. As shown in Figure 5a,c, S ² values derived from the shorter averaging times of ca. 500 ps are in better agreement with the experimental data compared to longer times. These results are also in agreement with model‐free calculations yielding ${\tau }_{\mathrm{e}}$ and ${\tau }_{\mathrm{s}}$ of ca. 100 ps and 2 ns, respectively (Figure 3b,c). MD derived S ² values recapitulate motions in the C‐terminal tail and the hinge domain (residue 38–52), while reproducing the overall rigidity of the remaining helices. Additionally, the simulations exhibit the expected trend of dampening of motions in the EF loops upon calcium binding.

FIGURE 5.

Site‐specific generalized order parameter computed from 0.5 μs MD trajectories with variable averaging time windows (from 100 ps; blue −10 ns; red) for (a) apo S100A12; (b) Ca²⁺‐E31A (left) and Ca²⁺‐(EF‐II)‐S100A12 (right) and; (c) Ca²⁺‐S100A12. Experimental model‐free S ² values are shown in open circles. The worm representation of the backbone shows the S ² values (top: apo; bottom: Ca²⁺‐S100A12) mapped on the structure of the protein with the radius of the tube proportional to the timescale of motion (only the monomeric chain is shown for the sake of clarity).

Given the excellent correlation between experimental and theoretical S ² values, we turned our attention to E31A model structure loaded with calcium in both EF loops. Within the first 50 ns of the trajectory, the calcium ions in the EF‐I loop were released to the water environment, in agreement with our experimental data demonstrating that E31A mutation inhibits calcium binding to the EF‐I loop. Subsequent MD simulations were performed on E31A and the wildtype proteins with calcium bound to the EF‐II loop only. In these simulations, the calcium ions remained in the EF‐II loop throughout the trajectory for both proteins with the structural integrity of the protein preserved throughout the simulation. For E31A, the NMR chemical shifts predicted using ShiftX (Han et al., 2011) on the structure extracted from this MD trajectory are in reasonable agreement with experimental values when compared to the predicted chemical shifts for the wildtype protein (Figure S4), suggesting that the overall architecture of MD equilibrated structure is in agreement with the protein in solution observed by NMR. The order parameters derived from these simulations are presented in Figure 5b. As expected, the EF‐I loop shows greater mobility, similar to the apo‐wild type protein. Furthermore, the C‐terminal tail shows depressed S ² values, consistent with the dip in the experimentally observed NOE data for the mutant reflecting a highly flexible backbone.

2.6. Evaluating the entropic contribution to calcium binding affinity

The change in the conformational entropy derived from generalized order parameters offers a powerful approach to parse residue specific contributions to the free energy of binding ligands (Jarymowycz & Stone, 2006). A pertinent application of this methodology quantified the entropic contributions of individual calcium binding EF loops and showed synergistic binding between EF‐I and II loops in calbindin D_9k (Akke et al., ^1993a; Akke et al., ^1993b; Linse & Chazin, ¹⁹⁹⁵; Wimberly et al., ¹⁹⁹⁵). These works utilized the (Cd²⁺)₁ state as a proxy for calcium binding to the EF‐II loop and a mutant (N56A) to enable binding exclusively in the EF‐I loop (Wimberly et al., 1995). By measuring the binding affinities between the apo, two half‐saturated (Cd²⁺ at site 2 and Ca²⁺ at site 1) and the doubly (Ca²⁺)₂ saturated states (Akke et al., ^1993b; Linse & Chazin, ¹⁹⁹⁵), the authors were able to demonstrate that the dominant contribution of the free energy comes from Ca²⁺ binding to the EF‐II loop (Akke et al., ^1993b; Linse & Chazin, ¹⁹⁹⁵). As an extension to these previous works, below we have quantified the entropic contributions of the individual EF loops towards the overall conformational entropy of S100A12.

Instead of considering the experimental amide bond order parameters, the present analysis relies on calculating change in conformational entropy (ΔS) using order parameters derived from the MD trajectories. Given the excellent agreement in the trend between experimentally measured and MD derived S ² values for the wildtype protein (Figure 5), the use of the MD results for our analysis bears some benefits. For instance, we can sample the order parameters of residues not amenable to direct observation by NMR owing to extensive exchange broadening in the calcium binding loops. Also, by comparing the MD derived entropic changes in the calcium binding loops between the apo, half and fully saturated states, we can parse the contributions of individual calcium binding events.

Our analysis rests on the assumption that the calcium binding events occur sequentially, first to the EF‐II and then to the EF‐I loop. The half‐saturated state, resulting from the binding of calcium to the EF‐II loop, can be modeled either by Ca²⁺(EF‐II)‐S100A12 or Ca²⁺‐E31A. While the half‐saturated state represented by the E31A mutant is experimentally amenable, the overall conformational space of this protein may also be influenced by the mutation, as evident from the observed CSPs. In this scenario, the MD analysis considering the half saturated state as Ca²⁺‐E31A would account for both the calcium binding events and perturbations introduced by the mutation. Since these two effects are entangled, here we focus on ΔS occurring along the apo → Ca²⁺(EF‐II)‐S100A12 → Ca²⁺‐S100A12 pathway in Table 1 and the results obtained for the E31A pathway are reported in Table S4.

TABLE 1.

Change in conformational entropy (in J/K/mol) for domains of S100A12.

	ΔS II apo → Ca2+(EF‐II)	ΔS II,I Ca2+(EF‐II) → Ca2+‐S100A12	ΔG c = –TΔS TOT
Helix 1	11.1 (−3.3)	7.4 (−2.2)	−5.5
EF‐I	15.3 (−4.6)	−53.0 (15.8)	11.2
Helix 2	9.3 (−2.8)	−10.6 (3.2)	0.4
Hinge	31.8 (−9.5)	−25.0 (7.5)	−2.0
Helix 3	−3.4 (1.0)	−11.0 (3.3)	4.3
EF‐II	−48.5 (14.5)	−2.0 (0.6)	15.1
Helix 4	9.9 (−3.0)	−10.3 (3.1)	0.1
C‐terminal	5.6 (−1.7)	10.8 (−3.2)	−4.9

Note: The last column gives values of ΔG _c = –TΔS ^TOT at 298 K (in of kJ/mol), ΔS values are reported in the units of J/K/mol; –TΔS values at 298 K are presented parentheses in the units of kJ/mol.

The ΔS analysis of our MD derived order parameters, based on the formalism developed by Kay and coworkers (Yang et al., 1997), shows that the two calcium binding events induce distinct domain‐dependent changes in the conformational entropy of S100A12 (ΔG _c). The coordination of the first Ca²⁺ ion leads to a significant loss in conformational entropy in the EF‐II loop but increases floppiness in (almost) all the other domains, including the EF‐I loop and the putative target binding hinge domain. The net free energy change (−9.4 kJ/mol) is negative, indicating that the first binding event to the high affinity EF‐II loop is enthalpically driven (Figure 1a) with favorable entropic contributions. In contrast, the second binding event leads to significant entropy losses in all domains, except in helix 1 and C‐terminal, amounting to an overall free energy penalty of 28.1 kJ/mol. Notably the coordination of Ca²⁺ ion to the EF‐I loop has marginal effect on the dynamics in the EF‐II loop.

In the alternate pathway involving the E31A mutation, similar trends are observed in the entropic contributions to the free energy calculated for EF‐II, C‐terminal, and hinge domains. The first calcium binding causes the configurational entropy to increase in both the EF‐II and EF‐I loops, leading to an overall increase of the configurational free energy by about 9 kJ/mol. The second binding event affects mostly the mobility of helix 3; it leads to almost no conformational entropy changes in the EF‐I loop but increases the mobility of EF‐II. Consequently, the summation of the configurational free energy increases by 9 kJ/mol. Key takeaways from all these results are: (i) the dynamics of the calcium binding loops are strongly correlated where binding to the EF‐II loop modulates the dynamics of the EF‐I loop and vice versa, and (ii) calcium binding in the EF‐II loop increases the entropy of the hinge domain and the C‐terminal residues which may facilitate optimal target interaction. Despite the possibility of the influence of the mutation in E31A, we can draw similar conclusions from the data related to these key conformational changes in the loop and target binding site, which reinforces the functionally calibrated dynamic response triggered by the two calcium binding events. Overall, both sets of entropy changes derived by considering either the half‐saturated state of S100A12 or the mutant as intermediate states between apo and Ca²⁺‐S100A12 indicate that the two calcium binding events are correlated, and they induce striking dynamical changes throughout the whole protein.

3. DISCUSSION

The characterization of S100A12 and its variants described here demonstrates that the two EF loops bind Ca²⁺ cooperatively with long‐range effect on the structure and dynamics of the protein. To evaluate the extent of cooperativity between the two EF loops we have used a combination of NMR, ITC, and MD simulations to characterize the half and fully saturated states of the protein. In biological milieu, the high affinity EF‐II loop in the S100 proteins binds Ca²⁺ with K _D ≈ 0.1–100 μM, while the EF‐I loop exhibits a lower Ca²⁺ affinity ranging between 0.1 and 1 mM (Pietzsch & Hoppmann, 2009). The distinct binding affinities of the two EF loops are key to the biological functions of S100 proteins, particularly in the cytoplasm. Typical basal Ca²⁺ concentrations (~100 nM) are not sufficient to fully saturate the EF loops of cytoplasmic S100 proteins, many of which are present in micromolar concentrations. This prevents the depletion of free Ca²⁺ ions in the intercellular space. The activation of S100 proteins is regulated by a spike in the intracellular Ca²⁺ levels (~500 nM) in concert with enhanced target association initiating specific cellular response (Young et al., 2022). This activation mechanism has been characterized for several S100 proteins such as S100A1, S100B and S100A4 (Malashkevich et al., 2008; Wright et al., ^{2005, 2009}; Prosser et al., ²⁰⁰⁸; Charpentier et al., ²⁰¹⁰; Rustandi et al., ¹⁹⁹⁸; Garrett et al., ²⁰⁰⁸; Duelli et al., ²⁰¹⁴). For S100A12, while modulation of calcium sequestration upon target binding as not been established, several intracellular target proteins have been identified (Hatakeyama et al., 2004). An understanding of Ca²⁺ modulated structural and dynamical attributes of S100A12 is therefore needed for evaluating the interdependence of S100A12‐target interaction and calcium binding.

The point mutations E31A (EF‐I) and N63A (EF‐II) have vastly different effects on Ca²⁺ binding affinity and the accompanying conformational changes in the protein. The loss of Ca²⁺ binding in the EF‐I loop has marginal effect on suppressing calcium binding in the canonical EF‐II motif (K _D ~ 20 μM) compared to the wild‐type protein. This result is contrary to previously reported studies on S100B where the E31A mutation completely abrogates binding in the EF‐II loop (Markowitz et al., 2005). The N63A mutant, which lacks the ability to bind calcium in the EF‐II loop, exhibits significantly lower affinity (~35 mM) than expected from the pseudo loop (0.1–1 mM, Figure S5), hinting at the presence of positive cooperativity between the two binding sites. The more than two orders of magnitude difference in the Ca²⁺ binding affinities in the two EF loops in S100A12 implies sequential binding starting with the saturation of the EF‐II site before the second EF‐I site is fully occupied. Therefore, any positive cooperativity is expected to enhance the affinity at the EF‐I site once the EF‐II loop is saturated with calcium. This conclusion is in general agreement with other multi‐step calcium binding proteins like the N‐terminal domain of Calmodulin and Calbindin D_9k (Linse & Chazin, 1995; Beccia et al., ²⁰¹⁵).

The coupling of low and high affinity Ca²⁺ binding sites is the structural foundation of the dual functionality of S100s as Ca²⁺ sensors in cell signaling (EF‐II) and homeostasis of metal ions (EF‐I) (Donato et al., 2013). The entropy calculations reveal the tentative mechanism of a feedback loop whereby reduced mobility of the Ca²⁺ bound EF‐II loop modulates the dynamics of the EF‐I loop and the target binding site both of which become more malleable. While the conformational flexibility in the structure aids target selectivity it can also reduce Ca²⁺ affinity by increasing the entropic cost of rearranging the binding loops. However, one could argue that once the higher affinity EF‐II site is occupied, the solvent exposed target binding site is poised for binding the target. It is likely that the subsequent rigidification of the structure (Liriano et al., 2012) in the target bound state plays an important role in shifting the equilibrium of the dynamic ensemble in favor of better‐defined structures which could lower the entropic penalty for Ca²⁺ binding at both loops. This is also consistent with the results from the MD simulations of the wild‐type protein which indicate Ca²⁺ bound EF‐I loop makes the backbone more rigid. The role of target mediated conformational dynamics in tuning calcium affinities is an open question for S100A12 and a topic for future studies.

The major structural differences between apo and Ca²⁺‐S100A12 originates from the reorientation of helix 3 triggered by calcium binding in EF‐II loop. Based on chemical shift data (Figure 2), the N63A mutation favors a conformation like the apo protein both in the absence and presence of calcium. In apo‐E31A protein, we see the evidence of local conformational changes in the EF‐I loop propagated to helix 1. Because EF‐I and EF‐II loops are coupled, the C‐terminal EF‐hand domain also undergoes some rearrangement in the apo state. Except for the binding loops, the calcium bound state is indistinguishable from the wild‐type protein in the C‐terminal EF‐hand domain. As seen in the X‐ray structures of S100A12 with different coordination states and by NMR in this study, the N‐terminal EF hand has limited conformational flexibility. In contrast, in the C‐terminal EF‐hand both NMR chemical shifts and MD simulations revealed an ‘open’ to ‘close’ conformational switch driven primarily by calcium binding to EF‐II loop. This raises an important question, whether the difference in the calcium dependent response of the EF hands is encoded by the intrinsic flexibility of the loop sequence or dictated by the stability of the helical motif. There is some evidence from calbindin D_9k, a member of the S100 family, that showed replacing EF‐I with EF‐II loop failed to modify the response of N‐terminal domain (Malmendal et al., 1998). Thus, the activation of the S100s is tunable not just by modulating the Ca²⁺ binding affinity but the dynamic response depends on the rigidity of the hydrophobic core.

In addition to introducing structural modifications, calcium binding to S100A12 also influences internal dynamics of the polypeptide. Barring the EF‐II loop, the relatively rigid apo protein presents a stark contrast to the dynamic allostery detected in the calcium bound state of the protein. The rigidification of the EF‐II loop in the Ca²⁺ bound state causes increased backbone mobility at the target binding site formed by the C‐terminal and hinge region of the protein on the faster timescale. The presence of fast timescale (ps–ns) motions in several S100 and related proteins has been evaluated in the literature (Akke et al., ^1993b; Inman et al., ²⁰⁰¹; Pálfy et al., ²⁰¹⁶; Koerdel et al., ¹⁹⁹²; Dutta et al., ²⁰⁰⁸; Nowakowski et al., ²⁰¹³; Spyracopoulos et al., ¹⁹⁹⁸; Gagné et al., ¹⁹⁹⁸). Generally, the C‐terminal domain in S100B, S100A4, S100A1, calbindin D_9k, and troponin have been shown to be flexible in the fast timescale regime, which is allosterically coupled to Ca²⁺ binding. This plasticity at the binding site is not only vital for targeting multiple binding partners (Bhattacharya et al., 2004), but plays a key role in the synchronization of Ca²⁺ uptake and release linked with target association and dissociation in signaling (Wang et al., 2022; Rustandi et al., ¹⁹⁹⁸; Zhang et al., ²⁰¹⁷).

The findings of this work reinforce major conclusions drawn from a common paradigm of calcium mediated function of S100 family of proteins. The remarkable similarity of the proteins presents a challenge to understand how each member fine tunes its function for different signaling pathways. All S100 protein undergo a structural rearrangement upon calcium binding driven primarily driven by EF‐II loop. The significance of the EF‐I loops, and its marginal impact on the conformational changes in the protein is less understood. In this study, we have decoded the cascade of dynamic changes triggered by the simple act of calcium binding in the loops. Thus, the local entropy of C‐terminal residues at the binding site increases with near equal contributions from both calcium binding events. However, the hinge domain experiences opposing contributions from the two binding events resulting in overall a modest increase in the conformation entropy of this critical region in the fully calcium saturated state. Being the putative target binding sites, such complementary changes in the backbone dynamics may be a key component of molecular recognition: whereby the mobility of the C‐terminal domain could facilitate ligand binding via the classical induced fit model and, restricted motions in the hydrophobic hinge domain could impart some degree of target specificity. Like calmodulin and its many homologues, such dual mechanism is very effectively employed in the conformational selection of diverse targets (Westerlund & Delemotte, 2018) in signaling pathways without enabling promiscuous association.

In summary, in this report we evaluate the individual and cooperative contributions of the EF loops towards modulating the structure and dynamics of S100A12. The conformational transformation of the protein is not an isolated event dominated by calcium binding to the canonical EF‐II loop but part of a cascade of structural and dynamic changes involving the pseudo‐EF loop and the target binding site that are all coupled by a single signal. Despite the rigidity of the N‐terminal domain and much lower calcium affinity, the EF‐I loop is equally relevant and plays a critical supporting role in modulating the response of the canonical EF‐hand domain.

4. METHODS

4.1. Protein expression and purification

Unenriched, ¹⁵N and ¹⁵N, ¹³C labeled recombinant wildtype was expressed and purified as previously reported (Wang et al., 2019). Single point mutagenesis to generate the E31A and N63A mutants was accomplished by polymerase chain reaction with the following primers: E31A forward 5′‐ AAAGGCGCACTGAAACAGCTGCTG; E31A reverse: 5′‐TTTCAGTGCGCCTTTGCTCAGCGT; N63A forward 5′‐GGATGCGGCCCAGGACGAACAAGTG; N63A reverse 5′‐CCTGGGCCGCATCCAGACCCTG. The identity of the plasmid upon mutagenesis was confirmed by DNA sequencing. The mutant plasmids were transformed into Escherichia coli BL21(DE3) cells. The mutants were expressed and purified using the same protocol as for the wildtype protein. For preparation of NMR samples, purified ¹⁵N‐ or ¹³C, ¹⁵N labeled proteins were concentrated to 0.3–1 mM in aqueous buffer containing 20 mM MES (pH 6.0), 100 mM sodium chloride, 10% D₂O, and 200 mm 4,4‐dimethyl‐4‐silapentane‐1‐sulfonic acid (DSS). Typical protein concentration for NMR measurements varied in the range of 300–800 μM. Ca²⁺ bound wildtype and E31A samples were prepared by adding 20 equivalents of Ca²⁺ to the protein. Ca²⁺‐N63A was generated by adding 100 equivalents of calcium to the protein.

4.2. Isothermal titration calorimetry

Isothermal titration calorimetry (ITC) experiments were conducted using a MicroCal VPN‐200 titration calorimeter, measuring at 25°C. Both the protein samples and the concentrated calcium ions were dissolved in the buffer solution (20 mM HEPES at pH 7, 100 mM sodium chloride). The protein concentration in the reaction chamber was adjusted to 300 μM. During the experiment, calcium, serving as the ligand, was introduced into the system at 3‐min intervals. A total of 40 injections were administered within a 120‐min timeframe. Each injection consisted of 2 μL of solution, except for the initial injection, which utilized a volume of 0.3 μL. Numerical fitting of the data was performed using Origin software and associated macros provided by the manufacturer.

4.3. NMR spectroscopy

NMR measurements were conducted on Bruker 500 MHz Ultrashield and 800 MHz US² spectrometers equipped with 5 mm TXI and 5 mm TCI CryoProbe, respectively. 2D ¹H − ¹⁵N HSQC (Cavanagh et al., 2007) and three‐dimensional (3D) ¹H − ¹⁵N − ¹³C CBCACONH (Grzesiek & Bax, ^{1992a, 1992b}) experiments for resonance assignments were performed at 11.7 T and 37°C, while NMR titrations employed via ¹H − ¹⁵N HSQC experiment were performed at 25°C. The laboratory frame R ₁, R ₂, and ¹⁵N–{¹H} NOE relaxation spectra on apo (11.7 and 21.1 T) and Ca²⁺ bound (11.7 and 18.8 T) wildtype protein was recorded at two static fields and 37°C sample temperature according to established methods (Farrow et al., 1994). The pseudo 3D experiments were acquired with variable relaxation delays for N¹⁵–R ₁ (~0, 80, 200 × 3, 400, 600, 800, and 1000 ms) and N¹⁵–R ₂ (0.0, 16, 32 × 3, 48, 64, 80, 96, and 128 ms) measurements. The heteronuclear ¹⁵N–{¹H} NOE experiments were acquired with 4 s saturation plus 1 s recycle delay and the reference experiment with 5 s recycle delay. Each experiment was repeated two to three times to get an estimate of the standard deviation. The two‐dimensional datasets were processed in NMRPipe (Delaglio et al., 1995), and the relaxation rates extracted from the mono‐exponential decay of the intensity using rate analysis tools in NmrViewJ 8.0.3 (Johnson, 2004). The R ₁ and R ₂ data at high field (21.1 T) was scaled to account for small temperature offset from data acquired at lower field (11.7 T) (Gill et al., 2016). Experimentally obtained relaxation parameters at both magnetic fields are reported in Appendix S1. The field dependent relaxation rates were analyzed in the RELAX 4.0.3 program (d'Auvergne & Gooley, ^{2008a, 2008b}) using the extended model‐free formalism (Lipari & Szabo, ^{1982a, 1982b}; Chen & Tjandra, ²⁰⁰⁸). Owing to exchange broadening in the calcium binding loops and the presence of sub‐nanosecond motions, the data was best fitted with a local diffusion model (Schurr et al., 1994). The amide nitrogen relaxation rates measured for each residue at two fields are fit up to 10 different motional model, with and without slow exchange (R _ex) contribution to ¹⁵N–R₂ on the micro‐millisecond timescales. The dynamic model that best fits the experimental data is selected based on Akaike information criterion (AIC) implemented in RELAX 4.0.3. The results of the best fitted model for each residue are summarized in Figure 3. For the two mutants ¹⁵N–{¹H} NOE relaxation spectra on apo and Ca²⁺ bound state was acquired at a single static field (11.7 T) and 37°C sample temperature. The corresponding spectra were processed using the same methodology as described for the wildtype protein.

4.4. MD simulations

MD simulations were carried out using the GROMACS software package (version 2022.3) (Abraham et al., 2015). Atomistic models of S100A12 homodimers were constructed based on the PDB files 1GQM (calcium‐bound, chains G and H) and 2WCF (calcium‐free, chains A and B). S100A12 variants was obtained from the PDB file 1GQM, after replacing the amino acid E31 with alanine. Protonation states of titratable residues were assigned based on pK _a values assuming pH 6.0. In particular, the protonation state of the histidine residues in each symmetric chain of the homodimer was assigned according to the previous experimental findings (Wang et al., 2020), that is, His6 (neutral), His15 (protonated), His23 (neutral), His85 (protonated), His87 (protonated), and His89 (protonated). Each protein was inserted in a cubic box whose dimensions were determined by imposing a buffer width of 15 $\AA$. The protein was then solvated in explicit water and neutralized by adding sodium and chlorine ions to have a concentration of 0.15 M. Overall, the atomistic models considered in this study encompassed 66,357 (calcium‐bound), 58,802 (apo), and 69,372 (mutation) atoms. We used the AMBER99SB force field (Hornak et al., 2006) to describe proteins and ions, and the TIP3P model (Jorgensen et al., 1983) to treat the water environment. The systems were equilibrated at 300 K and 1 atm using standard procedures (Wang et al., 2022). After the equilibration step, the production run consisted of a MD simulation of 0.5 μs in the NPT ensemble at 300 K and 1 atm. MD simulations were carried out by using a time step of 2 fs using the particle mesh Ewald summation method with a grid spacing of 1.6 $\AA$ to calculate electrostatic interactions, the modified Berendsen thermostat to control the temperature, and the Parrinello–Rahman barostat to control the hydrostatic pressure. Bonds involving H atoms were constrained using the LINCS algorithm.

To calculate the order parameters ($S^2$) of the backbone amide bonds, we relied on the Lipari–Szabo model‐free approach (Lipari & Szabo, ^1982a) and adopted the truncated average approximation (Bowman, 2016):

$$s^2=\frac{1}{\tau }{\int }_{{\tau }_m}^{{\tau }_m+\tau }C\left(t\right)\mathit{dt},$$

where $C\left(t\right)$ is the angular correlation function describing the rotational dynamics of a N–H bond, ${\tau }_m$ is the isotropic rotational correlation time, and $\tau$ is 1 ns (Bowman, 2016). For completeness, we remark that $C\left(t\right)$ is calculated as follows:

$$C\left(t\right)=⟨P_2\left(\widehat{\mu }\left(0\right)\bullet \widehat{\mu }\left(t\right)\right)⟩=\frac{1}{t_{\mathit{MD}}-t}{\int }_0^{t_{\mathit{MD}}-t}{P}_2\left(\widehat{\mu }\left(\tau \right)\bullet \widehat{\mu }\left(t+\tau \right)\right)\mathit{d\tau }$$

where $\widehat{\mu }$ is the unit vector pointing along the N–H bond, $P_2\left(x\right)=\left(3x^2-1\right)/2$ is the second Legendre polynomial, and $t_{\mathit{MD}}$ is the temporal length of the MD simulation.

AUTHOR CONTRIBUTIONS

Rupal Gupta: Conceptualization; investigation; funding acquisition; writing – original draft; writing – review and editing; formal analysis; data curation; supervision; project administration. Qian Wang: Conceptualization; data curation; formal analysis. Christopher DiForte: Conceptualization; formal analysis; data curation. Aleksey Aleshintsev: Conceptualization; formal analysis; data curation. Gianna Elci: Data curation; formal analysis. shibani bhattacharya: Conceptualization; writing – original draft; writing – review and editing; formal analysis; supervision; data curation. Angelo Bongiorno: Conceptualization; funding acquisition; writing – original draft; writing – review and editing; formal analysis; supervision.

Supporting information

FIGURE S1. (a) Comparison of TALOS predicted secondary structures from NMR chemical shifts for apo (left) and calcium bound (right) wildtype and E31A proteins and, (b) site specific chemical shift perturbations between apo and calcium bound wildtype (blue) and E31A (pink) proteins.

FIGURE S2. (a) Overlay of ¹H–¹⁵N HSQC spectra of wildtype and N63A proteins in the presence and absence of calcium; and (b) change in NMR peak intensities during calcium titration for select residues of N63A. The lines represent hyperbolic fit to the data corresponding to K _D values of 35 mM (L32), 38 mM (G30), and 37 mM (E31).

FIGURE S3. Overlay of site‐specific {¹H}–¹⁵N NOEs (pink) and S2 values (blue) for the apo and calcium bound wildtype S100A12.

FIGURE S4. (a and b) Correlation plots showing agreement between experimental and ShiftX calculated NMR chemical shifts for Ca carbon atoms; (c) ¹³Ca chemical shift perturbations between ShiftX predicted and experimental chemical shifts for Ca²⁺‐wildtype (blue) and Ca²⁺‐E31A (orange) proteins.

FIGURE S5. Normalized change in heat vs equivalents of calcium added to the wildtype protein obtained from ITC experiments. Numerical fitting of the data was performed using Pytc software using a sequential binding model yielding K _D values of 5 μM and 0.4 mM for the first and second binding events.

TABLE S1. Generalized order parameter (S2) for EF loops, hinge and C‐terminal (CTD) in apo and calcium bound S100A12.

TABLE S2. Residue specific ¹⁵N relaxation rates (R ₁, R ₂) and ¹⁵N–{¹H} NOE values for apo S100A12.

TABLE S3. Residue specific ¹⁵N relaxation rates (R ₁, R ₂) and ¹⁵N–{¹H} NOE values for Ca²⁺‐S100A12.

TABLE S4. Change in conformational entropy (in J/K/mol) for domains of S100A12. ΔS _E31A ^II and ΔS _E31A ^II,I represent the change in conformational entropy resulting from the binding events apo → Ca²⁺‐E31A and Ca²⁺ → S100A12, respectively. The last column gives values of ΔG _c = −TΔS ^TOT at 298 K (in of kJ/mol), corresponding to the entropic contribution to the Gibbs free energy change arising from the conformational change from apo to Ca²⁺‐S100A12; this quantity is the sum of ΔS _E31A ^II and ΔS _E31A ^II,I.

Wang Q, DiForte C, Aleshintsev A, Elci G, Bhattacharya S, Bongiorno A, et al. Calcium mediated static and dynamic allostery in S100A12: Implications for target recognition by S100 proteins . Protein Science. 2024;33(4):e4955. 10.1002/pro.4955