Propagation of Conformational Instability in FK506-binding Protein FKBP12

Abstract

FKBP12 is the archetype of the FK506 binding domains that define the family of FKBP proteins which participate in the regulation of various distinct physiological signaling processes. As the drugs FK506 and rapamycin inhibit many of these FKBP proteins, there is need to develop therapeutics which exhibit selectivity within this family. The long β₄-β₅ loop of the FKBP domain is known to regulate transcriptional activity for the steroid hormone receptors and appears to participate in regulating calcium channel activity for the cardiac and skeletal muscle ryanodine receptors. The β₄-β₅ loop of FKBP12 has been shown to undergo extensive conformational dynamics, and here we report hydrogen exchange measurements for a series of mutational variants in that loop which indicate deviations from a two-state kinetics for those dynamics. In addition to a previously characterized local transition near the tip of this loop, evidence is presented for a second site of conformational dynamics in the stem of this loop. These mutation-dependent hydrogen exchange effects extend beyond the β₄-β₅ loop, primarily by disrupting the hydrogen bond between the Gly 58 amide and the Tyr 80 carbonyl oxygen which links the two halves of the structural rim that surrounds the active site cleft. Mutationally-induced opening of the cleft between Gly 58 and Tyr 80 not only modulates the global stability of the protein, it promotes a conformational transition in the distant β₂-β_3a hairpin that modulates the binding affinity for a FKBP51-selective inhibitor previously designed to exploit a localized conformational transition at the homologous site.

1. Introduction

FKBP12 initially gained clinical attention as the cellular receptor for the immunosuppressant drugs FK506 and rapamycin and the resultant inhibition of calcineurin and TOR (target of rapamycin). Although FKBP12 catalyzes prolyl peptide isomerization, the clinical importance of that activity remains unclear [1]. One significant physiological function for both FKBP12 and the closely related FKBP12.6 involves their binding to the ryanodine receptors of skeletal and cardiac muscle [2, 3]. While their role in the regulation of these calcium ion channels has not yet been fully resolved, much attention has been placed upon the putative function of FKBP12.6 in suppressing aberrant calcium leakage during the diastolic phase of the cardiac cycle [4]. Recent high resolution cryo-EM structures of the ryanodine receptors have characterized in atomic detail the conformation of FKBP12/FKBP12.6 which resides near the periphery of the tetrameric receptor complex, well removed from the central calcium ion channel [5–7]. In addition to these analyses of structural interactions within the isolated receptor complex, other studies on the in vivo clustering of the individual receptors have revealed physiologically dynamic restructuring of these clustered microdomains as well as pathology-dependent variations in these microdomains [8, 9]. In particular, both FKBP12 and FKBP12.6 have been shown to reduce the size of the clustering domains for the cardiac ryanodine receptors with a corresponding change in the calcium ion transport kinetics [10].

The steroid hormone receptor regulation proteins FKBP51 and FKBP52 both contain FK1 domains that are structurally and evolutionarily related to FKBP12. Consistent with their known antagonistic roles in modulating the expression levels of genes that are controlled by the steroid hormone receptor protein, Riggs and colleagues demonstrated that a L119P mutation in the long β₄-β₅ loop of the FK1 domain in FKBP51 increases reporter gene expression while the complementary P119L mutation in FKBP52 yields a corresponding decrease in the reporter gene expression [11]. ¹⁵N relaxation studies found that the β₄-β₅ loop in the FK1 domain of FKBP51 exhibits extensive μs-ms conformational dynamics that are absent in the FKBP52 protein and are quenched when the L119P mutation is introduced into the FKBP51 protein [12]. Earlier studies indicating the importance of the β₄-β₅ loop in the regulation of steroid hormone receptor activity [13–15] have been directly affirmed by the high resolution cryo-EM structures of the hormone-containing glucocorticoid receptor:Hsp90 complex bound with either FKBP51 or FKBP52 [16]. In both cases, the FKBP protein wraps around the ligand binding domain of the glucocorticoid receptor with an alteration in the angle of contact between the FK1 domain and the rest of the complex appearing to influence the potentiation of glucocorticoid hormone binding. In its direct interactions with the ligand binding domain, the β₄-β₅ loop was found to adopt multiple conformations. In-cell photocrosslinking experiments have characterized the inter-domain interactions within the less well-ordered glucocorticoid receptor:Hsp90:FKBP51/52 complex that is formed in the absence of glucocorticoid hormone binding to demonstrate a qualitatively similar quaternary structure as that reported in the cryo-EM study of the hormone-bound form [17].

The corresponding β₄-β₅ loop of FKBP12 (Tyr 80 – Ala 95) binds to the human RyR2 cardiac ryanodine receptor with the tip of that loop directly interacting with both the NTD-C and Jsol domains [6]. Within the isolated FKBP12 protein, that loop exhibits a pattern of μs-ms conformational dynamics that qualitatively resembles that seen in the FK1 domain of FKBP51 [18]. Kinetic analysis of these conformational dynamics in FKBP12 have been obtained from ¹³C relaxation dispersion measurements on methyl labeled proteins [19] as well as with ¹⁵N relaxation dispersion studies [20]. Among the five ¹³C methyl positions monitored within the β₄-β₅ loop, the derived k_ex values ranged from 4.8 ± 2.1 × 10³ s⁻¹ to 12.0 ± 4.2 × 10³ s⁻¹. In noting the substantial uncertainties obtained for the individual residue values, those authors reported a dynamics analysis constrained to a single global k_ex value that refined to 7.9 ± 1.0 × 10³ s⁻¹ [19]. The analogous dynamics analysis of ¹⁵N relaxation dispersion data yielded a global k_ex value of 8.6 ± 0.8 × 10³ s⁻¹ for all dynamically exchanging residues throughout the protein structure [20]. While a subsequent NMR relaxation study indicated that there are three distinct regions of FKBP12 that undergo at least partially decoupled conformational transitions in the μs-ms timeframe [18], that study did not address the question of whether the monitored transition within the β₄-β₅ loop corresponds to a simple two-state exchange process.

The ability to characterize the μs-ms conformational dynamics within the β₄-β₅ loop of FKBP12 is enhanced by the observation that the H87V variant of that domain strongly suppresses those dynamics, despite giving rise to a strikingly similar crystallographic structure to that of the wild-type protein [18]. These observations appear consistent with the H87V structure corresponding to a mutationally stabilized conformational ‘ground state’ for the β₄-β₅ loop. Further structural insight into the conformational dynamics within the β₄-β₅ loop of FKBP12 was obtained from the correlation between measured ¹⁵N conformational linebroadening and structurally predicted differential ¹⁵N chemical shift values which were consistent with an ~180° flip of the peptide group linking Pro 88 and Gly 89 [21]. Substitution of an alanine at residue 89 yielded structurally conservative changes in the NMR spectral data that were consistent with the anticipated increase in population for the transiently sampled backbone conformation that has an energetically favored negative ϕ dihedral angle value.

While these earlier studies indicated that a flipping of the peptide unit between Pro 88 and Gly 89 is likely to be a major contributor to the observed conformational dynamics within the β₄-β₅ loop, they do not exclude the possibility of significant contributions to the dynamics of that loop from other localized structural transition(s). Recently, we reported that amide hydrogen exchange analysis can provide a powerful means for distinguishing between a two-state and three-state conformational exchange system with respect to transitions that occur in the β₂-β_3a hairpin of the FK1 domain in FKBP51 [22]. Crucial to the robustness of that analysis was the ability to determine the kinetic acidity of those backbone amides (i.e., log k_OH−) to within an overall accuracy of 0.04. Given the similar quality of hydrogen exchange data provided by FKBP12, in the present study an analogous approach was applied to the β₄-β₅ loop to assess whether at least two distinct transient conformational states might be required to explain the observed hydrogen exchange behavior. Further analysis would provide insight into how the conformational effects of these transitions within the β₄-β₅ loop propagate to other specific sites within the protein and impact both inhibitor binding affinity and global stability of the protein.

2. Materials and methods

2.1. Protein expression and purification

Genes for the mutational variants of human FKBP12 as well as the wild-type FKBP12.6 and FKBP51 FK1 domain proteins were chemically synthesized (Genscript) with codon optimization for expression in Escherichia coli [23]. The genes were cloned into the expression vector pET11a and then transformed into the BL21(DE3) strain (Novagen) for expression. The [²H,¹⁵N]labeled proteins were expressed and purified as previously described [12, 18]. For the CLEANEX-PM experiments, aliquots of the [U-²H,¹⁵N]-enriched protein were concentrated by centrifugal ultrafiltration and then exchanged into a series of buffers containing 6% ²H₂O in which the buffer concentration for each pH value was set to 25 mM, with reducing agents dithiothreitol and tris(2-carboxyethyl)phosphine hydrochloride at 1 mM, and sodium chloride added to a final ionic strength of 150 mM. For these experiments, a set of seven pH values were collected for each protein (six pH values for the less stable I76C/Y80V variant – Table S1). Sodium phosphate buffers were utilized for the three lowest pH samples, sodium borate buffers were utilized for the next two higher pH values, and sodium carbonate buffers were utilized for the two highest pH values. For the H/D exchange experiments, an exchange-in protocol was used [22]. The protein samples were initially back-exchanged into D₂O buffers and then equilibrated in phosphate buffers composed as described for the CLEANEX-PM experiments. These samples were then lyophilized. For binding experiments on the SAfit2 inhibitor (MedChemExpress), the protein samples were equilibrated against a pH 7.0 sodium phosphate buffer prepared analogously to those used in the hydrogen exchange measurements.

2.2. NMR data collection

NMR data for the hydrogen exchange kinetics were collected on a Bruker Avance III 600 MHz spectrometer equipped with a TXI cryoprobe at a probe temperature of 25°C with samples at protein concentrations of ~1.0 mM. For each sample, a series of four CLEANEX-PM experiments were collected with mix times of 6.49, 11.68, 21.41, and 38.93 ms using the fhsqccxf3gpph Bruker pulse program. Interleaved with these spectra were a complementary set of reference hard pulse spectra in which the e-PHOGSY selective water excitation pulse [24] was replaced with a high power 90° pulse to enable post-acquisition compensation for the relaxation effects that give rise to non-linearity in the CLEANEX-PM buildup curves [25]. Recycle delays of 4.0 s were used in all experiments except for an additional repeat of the shortest mix time hard pulse experiment that utilized an 8.0 s recycle delay which was used to estimate the fully relaxed peak intensities. Processing of the spectral data was carried out with Felix software (Felix NMR).

For the ¹H exchange-in experiments, the lyophilized protein samples were dissolved in 94% ¹H₂O-6% ²H₂O, and a time series of ¹H-¹⁵N 2D TROSY [26] spectra was collected. A CLEANEX-PM experiment was collected on each exchange-in sample to enable comparison with the previously analyzed pH series of CLEANEX-PM datasets to calibrate the pH values of that sample (Table S1). Data for the SAFit2 inhibitor binding experiments for the β₄-β₅ loop variants were collected on a Bruker Avance III 800 MHz spectrometer equipped with a TXI cryoprobe.

2.3. Hydrogen Exchange Analysis

For each mix time, the amplitude for each CLEANEX-PM crosspeak was normalized by the fractional signal loss due to the selective PHOGSY pulse sequence component and the relative number of scans used in the CLEANEX-PM and the hard pulse reference spectra. The normalized CLEANEX-PM intensity was then subtracted from the amplitude of the corresponding hard pulse reference crosspeak normalized to its fully relaxed intensity to yield an exponential decay with time constant ρ + k_ex, where ρ is the spin relaxation rate for the ¹H^N resonance. Under the condition of equivalent peak intensity noise across the mix time series, the individual deviations from the best fit line and the distribution of sampling times analytically determine the statistical uncertainty in the resultant k_ex values [22, 27]. With the residues for which the conformational exchange contributions were significant, the log k_OH− value derived from the highest pH utilized were reported.

In contrast to the more common protein hydrogen isotope exchange protocol carried out in deuterated water, isotope exchange of back-exchanged lyophilized protein samples dissolved into H₂O-containing buffers exhibits no deuterium isotope-induced shift either in the ionization behavior or in protein stability [28]. The only significant differential isotope effect that arises during the ¹H exchange-in experiments is the 0.08 unit shift in the log rate constants for exchange at N-²H amide bonds, relative to the corresponding N-¹H bonds [29]. The uncertainties arising from sample pH calibrations against the CLEANEX-PM data sets and the consistency between the exchange kinetics measurements for the two pH samples of the H87V variant yielded an estimated overall uncertainty of 0.08 for the log k_ex values derived from the exchange-in experiments.

2.4. Competitive inhibitor binding affinities among FKBP domains

[²H,¹⁵N]labeled samples of FKBP12.6 and the FK1 domain of FKBP51 were equilibrated in the inhibitor binding buffer with a saturating concentration of the inhibitor SAfit2 [30]. An aliquot from a 1 mM protein solution of the inhibitor saturated sample was mixed to an equivalent volume of a 1 mM solution of the competing FKBP domain. HSQC spectra were collected on the resultant mixture, and the relative peak volumes of the well-resolved resonances corresponding to both the apo- and inhibitor-bound forms of each protein were measured. A preliminary 2σ filtration was applied to the fractional saturation values derived from each residue to remove those resonances that might exhibit significant conformational linebroadening relaxation behavior between the unliganded and inhibitor-bound forms. The average saturation level for each protein was then used to determine the differential free energy of inhibitor binding for the two competing FKBP domains.

3. Results and discussion

3.1. Two-state analysis of hydrogen exchange reactivity in the β₄-β₅ loop of FKBP12

Quantitative interpretation of protein amide exchange reactivity in terms of thermodynamic properties requires the measurement of those data under the so-called EX2 kinetics condition. Under most circumstances, peptide hydrogen exchange is exclusively hydroxide ion-catalyzed for pH values above 4. Within that pH range, the EX2 condition implies that the log k_ex rate values for each backbone position will be directly proportional to the pH value of the solution. For the situation in which a protein amide hydrogen transitions between an unreactive ‘closed’ state and an exchange-active ‘open’ state, the EX2 kinetics condition requires that the rate of transition between the unreactive and reactive conformational states is rapid as compared to the hydroxide-catalyzed exchange reaction, insuring that the hydrogen exchange rate accurately samples the thermodynamic distribution of that conformational transition.

Simple peptides are known to exhibit diffusion-limited Eigen kinetics in hydroxide-catalyzed exchange reactions [31, 32]. The reaction rate of a peptide hydrogen with hydroxide ion is attenuated from the diffusion limit (10^10.3 M⁻¹ s⁻¹ at 25°C [32, 33]) by a factor of K_i/(K_i+1) where K_i is the equilibrium constant for the transfer of a hydrogen from the amide to the hydroxide ion. The brief lifetime (~10 ps) of the resultant peptide anion implies that the aggregate hydrogen exchange reactivity is well approximated by the sum of the conformer reactivities for the Boltzmann-averaged distribution of all conformations sampled by that peptide [23]. This insensitivity of amide acidities to the dielectric relaxation effects that can arise from protein conformational motion sharply differs from what is observed for protein sidechain ionizations that have much longer charge state lifetimes. The global conformational sampling behavior exhibited by amide exchange reactivities stands in contrast to the intrinsically dynamical NMR conformational linebroadening measurements that only report upon the modulation in chemical shift values which occur during relaxation-active conformational transitions. The practical utility of protein hydrogen exchange measurements for thermodynamic characterization largely arises during the characterization of changing conformational equilibria in which the intrinsic exchange reactivity of each conformational state is unaltered and only the population ratios of these states are affected. For circumstances in which the exchange reactivity of each conformational state is adequately known or predicted, detailed structural analysis of even transient conformational states can become feasible [22].

Hydrogen exchange measurements were carried out on the wild-type as well as H87V and G89A variants of the human FKBP12 protein. As noted above, a major contribution to the μs-ms conformational linebroadening dynamics in the β₄-β₅ loop is believed to arise from the flipping of the peptide linkage between Pro 88 and Gly 89 which is enhanced by the G89A substitution and suppressed by the H87V substitution. For six residues of the β₄-β₅ loop, log k_OH− rate constants were obtained in each of the wild-type, H87V, and G89A proteins (Fig. 1). In addition to the presence of proline residues at 88, 92, and 93, the spectra of the G89A variant failed to provide adequate data at several residue positions, primarily due to the increased level of resonance linebroadening that arises from the conformational exchange dynamics for that protein [21]. Under the assumption that the experimental hydrogen exchange data for the wild-type protein arises from a simple two-state mechanism, the hydrogen exchange reactivity values for each amide in the transient ‘excited’ state is sufficient to predict the experimental data for wild-type protein from the data of the dynamically quenched H87V protein at any given population value for the transient state T (p_T):

$${10}^{[\log k_{OH}(wt)]}=(1-p_T)*{10}^{[\log k_{OH}(H87V)]}+p_T*{10}^{[\log k_{OH}(T)]}$$ Eq. 1

Conversely, given the experimentally measured value of log k_OH(wt), the value of log k_OH(T) can be predicted as a function of p_T. Under the assumption of a two-state transition in which the same transient conformational state gives rise to both log k_OH(wt) and log k_OH(G89A), the value for the ratio of the population of the excited state for the G89A and wild-type proteins (p_T(G89A)/p_T(wt)) can then be deduced as a function of the assumed value for p_T(wt). Furthermore, since the experimentally observed exchange rate constants for both the wild-type and G89A proteins are significantly above those of the H87V variant for these residues, the transient state must largely dominate the observed exchange measurements. This implies that the ratio of p_T(G89A)/p_T(wt) will be largely insensitive to the assumed value of p_T(wt).

Fig. 1.

Observed and predicted hydrogen ion-catalyzed hydrogen exchange rate constants for residues of the β₄–β₅ loop. Experimental data were obtained for the wild-type (black) as well as H87V (blue) and G89A (red) variants. The diameter of the circles closely approximates the average 95% confidence limit for the individual data points. Assuming a shared two-state mechanism with the transient state population for the wild-type protein set to 0.5%, the corresponding hydrogen exchange rate constants for the transient state are displayed (open black circles). The putative value of transient state population for the G89A variant was then optimized to minimize the standard deviation between the transient state hydrogen exchange reactivities predicted from the experimental G89A exchange rate constants and those derived using the wild-type predicted transient state rate constants (open red circles), assuming all six residues (A), residues 83, 90, and 94 (B), or residues 80, 81, and 95 (C).

Based upon the report that the ¹³C relaxation dispersion curves for the methyls of FKBP12 could be fitted to a fast limit model if p_T was assumed to be less than 1% [19], a value of 0.5% for p_T(wt) was initially assumed for illustrating the proposed two-state analysis approach. The corresponding model log k_OH(T) values for the transient state were then derived from the experimental H87V and wild-type exchange rate constants. The value of p_T(G89A) was then optimized by minimizing the variance between the experimentally observed log k_OH− values for G89A and those predicted utilizing the model transient state log k_OH(T) values derived from the wild-type protein measurements. A p_T(G89A) value of 3.7% yielded a minimum rmsd of 0.31 for the transient state hydrogen exchange reactivities directly predicted from G89A experimental data and compared to those obtained assuming the wild-type derived log k_OH(T) values with p_T(wt) set to 0.5% (Fig. 1A). When the value for p_T(wt) is shifted from 0.5% to either 0.25% or 1.0%, the derived value for p_T(G89A)/p_T(wt) remained 7.4 ± 0.1.

The fact that the log k_OH− values for the wild-type protein predominantly arise from the transient state hydrogen exchange reactivities also implies that, for a given p_T(wt) value, the magnitude of the uncertainty in derived k_OH− values for the transient state will be equivalent to that for the measured wild-type k_OH− values. As a result, with an estimated average experimental accuracy for the measured log k_OH− values of 0.04, the nearly 8-fold larger discrepancy in the simple two-state model analysis implies that the optimal fit for that model, as applied to all six β₄-β₅ loop residues, lies well beyond the range of plausible statistical variation. As a result, the assumption of a simple two-state transition shared in common for the wild-type and G89A proteins would appear to be invalid.

Interestingly, for residues 83, 90, and 94 within the central region of the β₄-β₅ loop, the predictions for the transient state log k_OH− values based upon the experimental G89A exchange data for an occupancy of 3.7% consistently underestimated those derived from the experimental wild-type exchange data. In contrast, for the three monitored residues that occur at the ends of that loop (i.e., 80, 81, and 95), this pattern of systematic differences in the predicted transient state log k_OH− values was reversed. In the simplest circumstance, this disparity in hydrogen exchange behavior for these two regions of the β₄-β₅ loop might arise from two distinct local conformational transitions. The assumption that two such transitions within the same structural loop are energetically independent of each other is surely open to question. Nevertheless, applying that assumption to the analysis of these hydrogen exchange data can provide an approximation for the relative energetics of the two hypothesized transient conformational states. When the analysis of the predicted transient state hydrogen exchange reactivities was repeated with optimization only on the central region residues 83, 90, and 94, a p_T(G89A) value of 2.0% yielded an rmsd of 0.09 for the fit to the transient state hydrogen exchange reactivities (Fig. 1B). The analogous optimization utilizing the terminal loop residues 80, 81, and 95 yielded an optimal p_T(G89A) value of 6.9% with an rmsd of 0.07 (Fig. 1C). This substantial improvement in the statistical predictions provides support for the presumption of at least one additional form of dynamical transition within the β₄-β₅ loop that contributes to the observed hydrogen exchange kinetics.

The alanine substitution at position 89 yielded a 4-fold increase in hydrogen exchange reactivity for the residues near the center of the loop while inducing nearly a 14-fold enhancement for the residues at the stem of that loop. These results appear consistent with a second conformational transition occurring in the loop stem region that is enhanced by the sampling of the transient state at the Pro 88 – Gly 89 linkage.

3.2. Variations in amide hydrogen exchange reactivities arising from β₄-β₅ loop mutations

Hydrogen exchange measurements were carried out on a series of human FKBP12 variants bearing mutational substitutions in the β₄-β₅ loop (Fig. 2A). In addition to the wild-type and the H87V and G89A variants considered in the previous section, similar measurements were carried out on the A81G and H94G variants that lie within the stem region of the loop. In both instances the introduction of a glycine serves to increase the range of the backbone (ϕ,ψ) dihedral angle space that is energetically accessible to that residue. In the case of His 94, that residue assumes a positive ϕ value in the crystal structure which is a relatively rarely observed conformation for non-glycyl residues due to unfavorable local backbone interactions. Furthermore, the double mutant I76C/Y80V was incorporated into this study to provide a structural mimic of the homologous FKBP12.6 protein for the region surrounding the hydrogen bond between Gly 58 and Tyr/Val 80 which has previously been proposed to exhibit conformational heterogeneity [34].

Fig. 2.

Hydrogen exchange reactivities in FKBP12 as a function of mutational variants in the β₄–β₅ loop region. The ribbon structure of human FKBP12 with the positions of mutation selectively colored (A). The same color labeling is used to display the hydrogen ion-catalyzed hydrogen exchange rate constants for residues in each of the mutational variants with the wild-type data illustrate in black (B). The diameter of the circles closely approximates the average 95% confidence limit for the individual data points. All experimental data are derived from CLEANEX-PM measurements with the exception of a subset of the more slowly exchanging residues from the H87V variant that were monitored by isotope exchange experiments (open symbols).

For more rapidly exchanging protein amides, the rate at which selectively excited water-bound protons exchange onto backbone amide sites can be directly monitored by NMR methods. The associated experiments render it practical to collect data on a pH series of samples that provides the combined benefit of accurate kinetic measurements and direct verification of the proportionality between the pH values and the log k_ex rate values, or equivalently the stability of the derived log k_OH− rate constants. In the experimental analysis of various well-studied proteins, the vast majority of backbone amides exhibited log exchange rates that follow a simple linear pH dependence across the range of pH 6 to pH 10 [23]. By carrying out a series of measurements which span that pH interval, the hydroxide ion-catalyzed exchange rate constants can be readily determined over the range of 10⁴ to 10⁹ M ⁻¹s ⁻¹. Among the ionizable sidechains, it is typically only a subset of histidine residues that undergo significant charge state titration within this pH range, giving rise to two distinct electrostatic potential environments for the nearby amide sites. In such circumstances, corrections for the deviation from linearity in the log k_ex vs. pH plots for nearby amides can often be calculated from the sidechain pK value [23].

A relaxation-compensated version [25] of the widely used CLEANEX-PM experiment [35, 36] was applied. This modification allows for the extraction of an exchange rate measurement at each mixing time period used in the CLEANEX-PM experiment, thus enhancing the statistical robustness of the derived rate constants [22]. As the hydrogen exchange rates for several of the residues in the H87V variant protein were appreciably slower than those of the various other FKBP12 proteins analyzed, the more commonly applied isotope exchange approach was used to obtain kinetic values for those residues. Not surprisingly, the largest mutational variations in the log k_OH− values were observed within the β₄-β₅ loop (Fig. 2B, Table S2). In addition, a few comparatively large-scale mutational variations in the log k_OH− values were also observed at specific sites outside of the β₄-β₅ loop.

A useful indication of the quality of these hydrogen exchange reactivity measurements is provided by those residues for which the log k_OH− values are largely insensitive to the mutational substitutions. For 32 of the 54 residues characterized in these measurements, the aggregate rmsd value for the log k_OH− values among all six of the mutational variants for each of those 32 residues was 0.043. That level of consistency encompasses not only the statistical uncertainty for each sample measurement and the statistical uncertainty in correlating those individual sample measurements between pH values for a given mutational variant, this overall consistency also includes the variations that arise from the physical differences among these different structural variants. Across the full set of 54 residues, direct estimates of the experimental uncertainties were obtained from each set of individual sample measurements of the log k_OH− values (Table S2). The standard deviation for these uncertainties in the individual sample measurements were calculated for each of the protein variants (Table 1), which yielded an overall value of 0.027. Standard deviations among the individual protein variants were also calculated for the uncertainties that arose in correlating the log k_OH− values that were determined at different pH values (Table 1), yielding an overall value of 0.029. While these two sources of statistical uncertainty are surely not independent of each other, treating them as such yields a reasonable upper limit of 0.040 for the overall experimental uncertainty of these measurements. The similarity between this direct estimate of experimental uncertainty and the observed experimental variations between the 32 residues that exhibited seemingly negligible differences in their hydrogen exchange reactivities among the various mutational forms not only supports the overall validity of these error estimates, it also reinforces the qualitative conclusion that the hydrogen exchange-sensitive conformational flexibility for a substantial portion of the FKBP12 protein is essentially unperturbed by these β₄-β₅ loop mutations. It may be noted that quite similar statistical results have also been reported in the hydrogen exchange analysis of the homologous FK1 domain of FKBP51 [22].

Table 1.

Statistical uncertainty in log k_OH− values.

variant	intra-pH	inter-pH
wild-type	0.022a	0.028
H87V	0.023	0.028
A81G	0.028	0.028
H94G	0.026	0.027
I76C/Y80V	0.031	0.026
G89A	0.034	0.035

^astandard deviation for CLEANEX-PM data

With regards to the selective effects of the specific mutations introduced into the β₄-β₅ loop, the H94G mutation was found to have a comparatively uniform destabilizing effect upon the residues of that loop, as evidenced by an ~0.6 increase in their log k_OH− values, relative to the wild-type protein (Table S2). In contrast, the effects arising from the A81G substitution were more localized. While the 0.64 increase in the log k_OH− value at residue 81 is consistent with an apparent 4-fold destabilization of the hydrogen bonding interaction of that amide, the associated −0.62 decrease at residue 83 would similarly imply an apparent 4-fold stabilization of that residue’s hydrogen bonding. The intrinsic effects upon hydrogen exchange kinetics arising from the alanine-to-glycine substitution at that residue position are anticipated to be comparatively negligible [37]. The more selective effect of the A81G mutation suggests a significant alteration in the local conformational flexibility in the region of Ala 81 to Gly 83.

3.3. The modeling of conformational heterogeneity at the stem of the β₄-β₅ loop

The initial two-state analysis of the hydrogen exchange reactivity within the β₄-β₅ loop provided evidence for potentially distinct dynamical processes occurring in the loop and stem regions of this protein subdomain. Given the previously reported evidence for a relatively localized conformational transition involving the peptide group linking Pro 88 and Gly 89 [21], the stem region of the β₄-β₅ loop in the 0.92 Ǻ resolution structure of human FKBP12 (PDB code 2PPN [38]) was examined for potential sites that might give rise to conformational dynamics. Attention was immediately drawn to the bifurcated hydrogen bonding interactions of the Pro 78 carbonyl oxygen that binds to the amides of both Ala 81 and Gly 83 (Fig. 3). Also of potential interest is the amide of Tyr 82 which is hydrogen bonded to a partially buried water molecule that is seemingly stabilized by only two other hydrogen bonding interactions, one of which is with another structurally buried water. Along with the amide hydrogen of His 94, the second buried water molecule is hydrogen bonded to the carbonyl oxygen of Gly 83.

Fig. 3.

Bifurcated hydrogen bonding interactions in the stem of the β₄–β₅ loop of FKBP12. In the 0.92 Ǻ resolution crystal structure [38], the amides of Ala 81 and Gly 83 form hydrogen bonds to the carbonyl oxygen of Pro 78. The amide of Tyr 82 forms a hydrogen bond to one of two structurally buried water molecules. Along with the amide of His 94, the second of the two buried waters form hydrogen bonds to the carbonyl oxygen of Gly 83.

To examine whether other known structures of FKBP12 might provide evidence for differing hydrogen bonding arrangements around the carbonyl oxygen of Pro 78, the Protein Data Bank was surveyed for crystal structures with resolution limits better than 3.0 Ǻ that have sequence identity values within 85% of the human FKBP12 protein so as to include the closely related FKBP12.6 structures as well. Of the 93 such structures, 91 form hydrogen bonds between the amides of both Ala 81 and Gly 83 and the carbonyl oxygen of Pro 78 while lacking such a hydrogen bond for Tyr 82 (default relaxed constraints as specified by Chimera [39]). Furthermore, all 93 structures exhibited a hydrogen bond between the amide of residue 94 and the carbonyl oxygen of Gly 83. The only two deviating crystal structures were 1QPL (resolution 2.90 Ǻ) and monomer 8ER6a (resolution 2.81 Ǻ), both of which lack the canonical hydrogen bonding interaction between the amide of Gly 83 and the carbonyl oxygen of Pro 78 due to a small alteration in the backbone dihedral angles that directly border the Gly 83 amide. In contrast, the crystallographically non-equivalent monomers 8ER6b and 8ER6c do satisfy the default criterion for hydrogen bond formation.

As high resolution cryo-EM structures of the ryanodine receptor have recently been reported with resolution limits at 3.0 Ǻ or better, these structures were examined for evidence of conformational heterogeneity in this region of the FKBP12/FKBP12.6 structure. While the conformation of the FKBP12 protein within the 2.45 Ǻ resolution structure of the skeletal muscle RyR1 receptor exhibits no significant variation in the interaction geometries surrounding the Pro 78 carbonyl oxygen [7], substantial variations were observed among the conformations of the FKBP12.6 molecules found in a series of complexes for the wild-type and the pathogenic R2474S variant of the cardiac muscle RyR2 receptor [6]. Each of these RyR2 receptor structures yielded aggregate resolution limits below 3.0 Ǻ with the structure of the FKBP12.6 molecules being explicitly described as yielding atomic resolution detail. Each of these RyR2 receptor cryo-EM refinements were initiated from a model in which the FKBP12.6 conformation was derived from PDB code 6JI8 [40] which exhibits local geometries involving the Pro 78 carbonyl oxygen and Ala 81 to Gly 83 residues that closely match that of the 0.92 Ǻ structure of FKBP12 (PDB code 2PPN).

As illustrated for three of the high resolution cryo-EM structures reported for these RyR2 receptors (PBD codes 7UA3, 7UA4, and 7UA5 [6]), the backbone conformation at the ends of the β₄-β₅ loop where it merges into the β₄ and β₅ strands are closely superimposed (Fig. 4 - upper left). Similarly, the large majority of the central region of the β₄-β₅ loop is also closely superimposable among these three structures. On the other hand, the backbone conformation and hydrogen bonding interactions of residues Ala 81, Tyr 82, and Gly 83 markedly deviate among these structures. In a particularly notable deviation, the peptide group linking Ala 81 and Tyr 82 is flipped by ~180° between the 7UA3 and 7UA5 structures giving rise to an energetically unfavorable positive ϕ value for the highly conserved Tyr 82 in the 7UA5 structure. In the 7UA3 structure, this transition enables the formation of a hydrogen bond from Tyr 82 to the Pro 78 carbonyl oxygen while retaining hydrogen bonds from Ala 81 and Gly 83 to Pro 78. In contrast, the 7UA5 structure retains the bifurcated hydrogen bonding pattern to Pro 78 that is seen in the canonical 2PPN structure of FKBP12, while in the 7UA4 structure only the amide of Ala 81 forms a hydrogen bond to the carbonyl oxygen of Pro 78. Currently, there is no evidence to suggest that these local conformational differences in the FKBP12.6 structure are of physiological consequence (7UA3 and 7UA4 are the closed and open states of the PKA-phosphorylated R2474S variant with calmodulin bound, while 7UA5 is the wild-type dephosphorylated RyR2 in the closed state). Nevertheless, having been derived from the same starting conformation for FKBP12.6, these decidedly different local geometries would appear to indicate a significant degree of conformational plasticity at this site under the conditions studied.

Fig. 4.

Superposition of FKBP12.6 from three high-resolution cryo-EM structures of RyR2 [6]. The β₄-β₅ loop from PDB codes 7UA3 (green), 7UA4 (light blue), and 7UA5 (pink) is illustrated with the hydrogen bonding interactions displayed. While largely superimposed within the loop region in the lower right and the ends of the β₄ and β₅ strands in the upper left, the carbonyl oxygen of Ala 81 is oriented in the opposite direction in 7UA3 and 7UA5, while that of 7UA4 is perpendicular to both.

It is plausible to suggest that the varied patterns of hydrogen bond interactions observed in the stem region of the β₄-β₅ loop in the cryo-EM structures of the ryanodine receptor might serve to indicate the structural basis for an additional localized conformational transition within that loop (Fig. 1). In this regard, it may be noted that the flipping of the peptide group linking Ala 81 and Tyr 82 among the various RyR2 cryo-EM structures (Fig. 4) defines an arc that clearly delineates between the two sets of loop residues (80, 81, 95 and 83, 90, 94) that exhibited distinctly different patterns of reactivity behavior in that hydrogen exchange analysis (Fig. 1). As illustrated in our recent analysis of the conformational dynamics of the β₂-β_3a hairpin of the FK1 domain of FKBP51 [22], under favorable conditions it is practical to quantitatively test against detailed structural models of a three-state conformational transition by prediction of the observed hydrogen exchange measurements based upon the calculated electrostatic potential at the various sites of peptide ionization. Unfortunately, in contrast to that earlier study, the modeling of the two putative transient conformational states in the β₄-β₅ loop is confounded, in part, by the comparatively slow (μs-ms) dynamics involved which renders the accurate prediction of the associated conformational ensembles as comparatively problematic.

3.4. Spatial propagation of the β₄-β₅ loop conformational stability

Among the more rapidly exchanging backbone amides, the largest differential effects arising from the mutations in this study are principally within the β₄-β₅ loop (Fig. 2B, Table S2). For sites outside of this loop, a quite selective effect on the log k_OH− values was observed for the FKBP12.6-like I76C/Y80V variant which gave rise to substantial differences at Val 2 and Gln 3. The slower exchange rates observed at this site for the other mutational variants presumably reflects, at least in part, the impact of the shielding effect of the Tyr 80 phenol ring that stacks upon the N-terminus in the crystal structure of FKBP12 and is reflected in upfield shifts for the ¹H amide resonances of Val 2 and Gln 3, relative to those observed not only for the I76C/Y80V variant but for FKBP12.6 as well.

Two other sites of particular interest exhibited a far broader distribution in their sensitivity to the various β₄-β₅ loop mutations. Gly 58 exhibits a comparatively large dispersion in its log k_OH− values as a function of mutations in the β₄-β₅ loop. As noted above, this residue has attracted attention due to the fact that the hydrogen bond between the amide of Gly 58 and the carbonyl oxygen of Tyr 80 represents the only hydrogen bonding interaction that connects the β₄-β₅ loop to the remainder of the protein (Fig. 5A), beyond the previously mentioned interactions with the carbonyl oxygen of Pro 78. This hydrogen bond serves to help anchor together the two sides of the rim that surrounds the active site cleft. As this hydrogen bonding pair is inaccessible to the solvent phase in the crystal structure (Fig. 5B), the mutation-dependent variation in its hydrogen exchange reactivity strongly suggests a corresponding change in the fraction of time in which the stem region of the β₄-β₅ loop becomes separated from the start of the α-helix where Gly 58 resides. Previous to publishing their 0.92 Ǻ resolution structure of FKBP12 [38], Saven and colleagues reported a molecular dynamics analysis on a set of computational mutations at residue 60 that destabilize the interactions of a buried water molecule [34]. The computational removal of this buried water molecule predicted a displacement that resulted in a reversible increased separation between the amide of Gly 58 and the carbonyl oxygen of Tyr 80 of ~3 Ǻ.

Fig. 5.

Spatial propagation of correlated hydrogen exchange reactivities beyond the β₄-β₅ loop. Hydrogen bonding interactions are indicated within the β₄-β₅ loop and connected segments of those two strands. In addition, the hydrogen bond between Gly 58 and Tyr 80 is highlighted (A). The coloring of the residue mutation sites is displayed as in Fig. 2. Superposition of the remainder of the protein structure illustrates the conformational burial of the Gly 58 – Tyr 80 hydrogen bond and the close packing of the aromatic ring of Phe 36 upon the sidechains of Ile 90 and Ile 91 (B).

Correlation coefficients were determined between the log k_OH− values for Gly 58 and the various β₄-β₅ loop residues as a function of the β₄-β₅ loop mutations. Since the enhanced conformational exchange linebroadening processes at Gly 58 in the G89A variant protein precluded quantitation of the hydrogen exchange behavior at this position, correlation coefficients were calculated between the log k_OH− values for Gly 58 and sites within the β₄-β₅ loop for the other five mutational forms. As might be anticipated, the strongest correlation to the data of Gly 58 was observed at its hydrogen bond partner Tyr 80 (r = 0.97) as well as with the nearby Ala 81 and Ala 95. The level of correlation then decreased throughout the rest of the β₄-β₅ loop with the lowest correlations observed for Gly 83, Gly 86, and Ile 90 (Table 2).

Table 2.

Correlation coefficient for log k_OH− values.

residue	58	36
80	0.97	0.93
81	0.91	0.88
83	0.75	0.71
86	0.78	0.68
87	0.81	0.72
90	0.78	0.74
91	0.88	0.81
94	0.88	0.83
95	0.93	0.91

Another region of FKBP12 that exhibits substantial variation in hydrogen exchange reactivities as a function of mutations in the β₄-β₅ loop is the β₂-β_3a hairpin (Gly 28 – Asp 37) which is the key binding site for the first reported inhibitor that efficiently discriminates between the steroid hormone receptor regulation proteins FKBP51 and FKBP52 [41]. The largest differential hydrogen exchange reactivities in this hairpin region are observed for Phe 36, the position homologous to Phe 67 in the FK1 domain of FKBP51 which in its unliganded form transiently accesses a solvent exposed conformation from its more stable position structurally buried underneath the tip of the β₄-β₅ loop [22, 41–43]. Interestingly, the selective inhibitors that bind to FKBP51 far more tightly than to FKBP52 also bind to FKBP12 although at a somewhat lower affinity [44]. The current understanding of the selective binding interaction of FKBP51, relative to FKBP52, involves the absence of a specific hydrogen bonding interaction that connects the β₂ strand to the β_3a strand [22, 41, 42]. However, the interstrand hydrogen bond that appears to block FKBP52 from transitioning to the inhibitor-binding conformation is also present in the crystal structures of FKBP12. This seeming inconsistency suggested that additional structural insight into the mechanism of inhibitor binding by FKBP12 might prove to be of value.

Somewhat surprisingly, the mutational variation in hydrogen exchange reactivities at Phe 36 was found to correlate relatively poorly with those residues in the tip of the β₄-β₅ loop that lie directly over the buried phenyl ring of Phe 36 in the crystal structure. Despite the aromatic ring of Phe 36 being packed directly against the sidechains of Ile 90 and Ile 91 (Fig. 5B), the hydrogen exchange reactivity correlation coefficients for these residues are only 0.74 and 0.81, respectively. In contrast, the analogous correlation coefficients were significantly higher at the stem of the β₄-β₅ loop with an r value of 0.93 for residue 80 (Table 2). Most strikingly, the correlation was nearly perfect (r = 0.98) between hydrogen exchange reactivities for the amides of Gly 58 and Phe 36, despite these sites being separated by 16.5 Ǻ (Fig. 5A). Although less statistically robust, a similar pattern of correlation was observed for the β₂-β_3a hairpin residues Gly 28 and Asp 37. This correlation strongly suggests that the structural effects of disrupting the hydrogen bond between Gly 58 and Tyr 80 facilitate the differential conformational dynamics which occur in that hairpin region.

3.5. The effect of β₄-β₅ loop mutations on inhibitor binding in the β₂-β_3a hairpin

Hausch and colleagues have described a series of inhibitors that bind much more strongly to FKBP51 than FKBP52, thus providing a potential opportunity for clinical modulation of the steroid hormone receptor activity [45]. However, as a point of possible clinical concern, this set of inhibitors exhibit a much weaker selectivity with respect to FKBP12 and FKBP12.6. For the promising lead compound SAFit2, binding selectivity measurements for FKBP51 have reported values of >120 with respect to FKBP52, 3.17 ± 0.92 with respect to FKBP12, and 2.45 ± 0.85 with respect to FKBP12.6 [44].

Having observed significant mutationally-induced variations in the hydrogen exchange reactivities for residues in the homologous β₂-β_3a hairpin region of FKBP12, we investigated whether these β₄-β₅ loop mutations might give rise to corresponding alterations in binding affinity for SAFit2. In the simplest of circumstances, the mutationally-induced differential hydrogen exchange reactivities would directly correspond to changes in population for a transient drug binding-competent conformational state. With log k_OH− values of Phe 36 equal to 5.09 and 4.32 for the I76C/Y80V and H87V mutational variants, respectively (Table S2), a differential binding selectivity of 5.89 would be predicted.

As the predicted differences in binding selectivity among the various β₄-β₅ loop mutational variants are relatively small, it was clearly desirable to establish a competition assay which provided an accuracy comparable to that of the CLEANEX-PM measurements (i.e., 10^0.04 or ~10%). NMR samples were prepared containing, at concentrations of 0.5 mM, the SAFit2 inhibitor as well as [U-²H,¹⁵N]-enriched samples of the FK1 domain of FKBP51 and either FKBP12.6 or FKBP12. The resultant spectra present the four distinct species that arise from each of the two proteins in either the unliganded or inhibitor-bound form (Fig. 6). For those residues that yielded well resolved crosspeaks for both liganding states, the average level of saturation was determined for each protein. The selectivity of FKBP51 FK1 domain with respect to FKBP12 was 3.12 ± 0.25 while that for FKBP12.6 was 1.05 ± 0.08. Analogous competitive binding assays were carried out in which the binding selectivity of FKBP12.6 was compared to the FKBP12 variants I76C/Y80V and H87V (Fig. 7). Similar to the comparison between FKBP51 and FKBP12.6, the FKBP12.6-like mutations in the I76C/Y80V variant of FKBP12 yielded effectively indistinguishable affinities for SAFit2 (0.96 ± 0.13). In contrast, FKBP12.6 binds to this inhibitor more strongly than the H87V variant of FKBP12 by a factor of 4.62 ± 0.41.

Fig. 6.

Inhibitor binding competition between FKBP51 and FKBP12.6 or FKBP12. The arrows mark the conversion from the unliganded to the SAFit2-bound states. The relative intensities for all pairs of resonances are quite similar for both FKBP51 (black) and FKBP12.6 (pink) (A). In contrast, for the binding competition between FKBP51 (black) and the FKBP12 (pink), the SAFit2-bound resonances of FKBP51 are substantially more intense than the corresponding unliganded resonances, while the reverse pattern holds for the FKBP12 crosspeaks (B).

Fig. 7.

Inhibitor binding competition between FKBP12.6 and FKBP12 variants. The arrows mark the conversion from the unliganded to the SAFit2-bound states. The relative intensities for all pairs of resonances are quite similar for both FKBP12.6 (black) and the I76C/Y80V variant (orange) (A). An exception to that behavior was observed for the L74 resonances of the I76C/Y80V variant for which increased dynamical linebroadening occurred in the SAFit2-bound state. In the binding competition between FKBP12.6 (black) and the H87V (blue) variant, the SAFit2-bound resonances of FKBP12.6 are substantially more intense than the corresponding unliganded resonances, while the reverse pattern holds for the H87V crosspeaks (B).

Based upon the log k_OH− values for Phe 36, the simple model for hydrogen exchange via a transient drug-binding competent conformational state would predict a relative binding selectivity between the wild-type and I76C/Y80V variant of FKBP12 of 3.7 while the value deduced from these competition experiments indicates a value nearer to 3.1. For both that comparison as well as the one for the H87V variant, the simple transient drug-binding competent conformational state model appears to modestly overestimate the observed binding selectivity by ~20%.

3.6. The impact of mutations in the β₄-β₅ loop on the global stability of FKBP12

When the isotope exchange-in experiment on the H87V variant that was used to complete the log k_OH− determinations for the β₄-β₅ loop was carried out for longer time points to monitor slower rates of exchange, it was found that the cluster of most slowly exchanging residues exhibited an average time constant of a little more than nine days at pH 6.82. When repeated at pH 7.45, this slowest time constant was virtually unchanged, indicating that a transition from EX2 (hydroxide ion dependent) to EX1 (hydroxide ion independent) kinetics had occurred for the most slowly exchanging residues. A similar pair of isotope exchange-in experiments on the wild-type protein revealed qualitatively similar results in terms of a transition to EX1 kinetics in which the conformational transition has become rate-limiting for the most slowly exchanging residues. While, in principle, EX2 kinetics might potentially be obtained at substantially lower pH values, the issue of stability for some of the variants being studied appeared to render that option impractical. Nevertheless, the substantial difference in apparent global stability for the wild-type and H87V variants encouraged the collection of analogous isotope exchange-in measurements of the other mutational variants.

Each of the β₄-β₅ loop variants exhibited a pattern of three segments (23–25, 64–76, and 98–104) that contained the most slowly exchanging residues (Fig. 8, Table S3). This set of slowly exchanging residues closely correspond to those previously reported by Jackson and colleagues in their characterization of the global stability core of the wild-type FKBP12 protein [46]. The distribution of differential hydrogen exchange rates for the most slowly exchanging residues qualitatively mirrored the pattern observed for the CLEANEX-PM data of Gly 58. To examine this issue further, for each of the three slowly exchanging segments, the log k_ex values for the residues in the segment were averaged, following an initial 2σ filtering to remove severe outliers. These average log k_ex values were then correlated with the corresponding Gly 58 log k_OH− values to yield correlation coefficients of 0.97, 0.97, and 0.96 for the 23–25, 64–76, and 98–104 segments, respectively. It should be noted that for each mutational form, the log k_OH− values measured for Gly 58 and β₄-β₅ loop were well above the apparent log k_OH− values observed for the core residues, indicating that these sites are substantially less stable than the core region. As the isotope exchange-in data for the most slowly exchanging residues appear to arise predominately in the EX1 kinetic regime, it is the kinetic stability of the protein that is being experimentally monitored. That said, the protein sequence is identical in the core region of each of these mutational variants, suggesting that the relative thermodynamic stability values will presumably be similarly correlated. These results strongly suggest that the separation of the top of the α-helix from the stem region of the β₄-β₅ loop and the resultant disruption in the rim of the active site cleft plays a central role in the unfolding process of this protein.

Fig. 8.

Amide isotope exchange-in kinetics for the β₄–β₅ loop variants of FKBP12. Log exchange rates are displayed for H87V (blue), wild-type (black), A81G (purple), H94G (green), G89A (red), and I76C/Y80V (orange).

4. Conclusions

The quantitative measurement of protein amide hydrogen exchange reactivities can provide a valuable complement to other approaches for analyzing the coupling between dynamical processes within these macromolecules, most notably molecular simulation studies. Although not exploited in the present study, the ability to obtain moderately accurate predictions of the thermodynamic acidity of individual amide hydrogens from structure-dependent calculations of the electrostatic potential can be compared against accurately determined hydrogen exchange rate constants to provide powerful constraints upon the set of protein conformations capable of yielding that experimental data [22]. As demonstrated here, relatively small shifts in the local conformational flexibility across the structure of a protein can be useful for characterizing the propagation of long range interactions.

This study provides a plausible resolution to an ambiguity observed in the recent studies of FKBP51-selective inhibitor binding. In the analysis of the differential inhibitor affinities for FKBP51 and FKBP52, it was observed that the introduction of a T58K mutation into FKBP52 increased the binding affinity for a selective inhibitor by more than 200-fold, while the corresponding K58T mutation in FKBP51 reduces its binding affinity by nearly 60-fold [41]. The K58T variant of FKBP51 similarly suppresses the μs-ms linebroadening dynamics that are seen in the wild-type FK1 domain of FKBP51 but not in the corresponding FK1 domain of FKBP52 [42]. These results appear to imply that the presence of an inter-strand hydrogen bonding interaction between the sidechains of Thr 58 and Ser 69 in FKBP52 that is eliminated with the Lys 58 substitution in FKBP51 is crucial to hindering FKBP52 from reorienting its β₂ and β_3a strands so as to generate the inhibitor binding conformation [22, 41, 42]. Yet the high-resolution crystal structure of FKBP12 exhibits an inter-strand hydrogen bond homologous to that seen in FKBP52 while at the same time binding to the same inhibitors with affinities quite similar to those of FKBP51. The present evidence that the mutation-dependent modulation of the opening up of the active site cleft by the disruption of the hydrogen bonding interaction between Gly 58 and Tyr/Val 80 in FKBP12/FKBP12.6 selectively enhances the conformational flexibility around Phe 36 and the associated inhibitor binding affinities offers an explanation for this apparent conundrum.

Given the energetic coupling between the cleft opening transition at the Gly 58 – Tyr/Val 80 hydrogen bond and the transition to a SAFit2 inhibitor binding conformation in the β₂-β_3a hairpin, the resultant altered conformation in the region of that cleft opening might prove to be exploitable for FKBP12/FKBP12.6-selective inhibitor design. While the recently described series of inhibitors designed for FKBP51 do not efficiently discriminate against either FKBP12 or FKBP12.6, modification of such inhibitors at positions which interact with the Gly 58 – Tyr/Val 80 region of the protein might provide an additional level of discrimination among the larger family of FKBP binding domains.

Highlights.

• Cleft opening between the α-helix and the β₄-β₅ loop of FKBP12 mediates stability

• At least two localized transitions drive differential hydrogen exchange in this loop

• Cleft opening correlates with hydrogen exchange at a distant inhibitor binding site

• Inhibitor binding affinity correlates with the site’s differential hydrogen exchange

• These two correlated transient conformations may facilitate selective drug design

Data availability

Data will be made available on request