Long-range effects of substrate binding on GrsA and its active site mutant, W239S.
Previous studies highlight that the A-domain can adopt at least two distinct sequentially closed conformations to form two functionally different states [24-25]. The adenylate conformation is marked by aa and ATP binding, which react to form aminoacyl-AMP and PPi [26-27]. Subsequently, the AC rotates 140◦ relative to the AN subdomain [24–29] to adopt the thiolation conformation, to allow for the stabilization of the PPant-arm from the PCP domain inside a newly formed tunnel within the A-domain [27-29] (Fig. 1D). While the active-site residues for adenylation reaction and the subsequent thioester formation are both shared at the A-domain, they represent very different catalytic cavities. Without the presence of the PCP-domain and the structural remodeling of the AC-subdomain the thiolation conformation is unattainable. Modulation of this equilibrium in turn control the reactions catalyzed by the A-domain.
A previous HDX-MS study on GrsA-WT was performed only using ATP as a ligand and not its non-hydrolysable analog [23]. So, the true effects of binding remained occluded, and the resulting intermediates could have been entirely guided by the reactionary triggers. Consequently, the analysis of the apo-form and influence of substrate binding thereof remained an important mystery for the field. To assess the differential effects of substrate binding on inter-domain dynamics we performed HDX-MS experiments with apo-form of GrsA-WT as well as W239S and identified regions that underwent conformational changes (Fig. 2A, Fig. S1, HDX-MS data is summarized for all variants in Fig. S2D).
Focusing on the W239S mutant first, we observe that the apo-state is identical to the Tyr-bound state (its cognate aa) (Fig. 2B). We did not see any E-domain patches of altered D2O accessibility when the aa -bound dataset was compared to apo-form. The AMP-PCP-bound state was also identical to the Tyr-bound state with no change in the relative accessibility of D2O (Fig. S2). When the protein was incubated with both L-Tyr and AMP-PCP (dual condition) and compared against its apo-form, we noticed five distinct patches of altered D2O accessibility (Fig. 2A): 1) the linker region between AN and AC (residues 410 – 426, Fig. 2A magenta) , 2) the terminal helix of the AC subdomain (residues 488-505, Fig. 2A orange, 3) the linker region between the PCP and the E-domain (residues 605-621, Fig. 2A green), 4) residues 848-868, Fig. 2A teal) that stabilize from the E-PCP interaction and 5) residues (923-935 and 961-988, Fig. 2A blue) which encompass the E-domain’s active site (Fig. 2A, 2C-middle).
Most of these residues are neither in close proximity to the active site of the A-domain (Fig. 2A, black arrow), nor in any way in contact with the primary or secondary hydration shell of the active site (Fig. 2A). The solvent accessible, surface patches from the E-domain that become more accessible (residues 848-868) are known to be distal stabilizers of the PCP-domain [23, 30]. We observed very similar patterns of accessibility when the aa-bound dataset is compared to the dual condition, confirming that this is not an aberrant effect from aa binding to the active site of E-domain. All of our HDX-MS experiments were performed using non-hydrolysable analog of ATP (AMP-PCP), so that the allosteric changes observed can be unambiguously attributed to the effects of binding and not catalytic reactions.
These results clearly establish four primary conclusions. First, the allosteric effect of the binding of aa and AMP-PCP is observed specifically at few locations that are far away (up to 45 Å) from the active site of the A-domain (Fig. 2A). Our data suggest that the distal binding of substrates at the A-domain acts as the trigger for the E-domain to switch from a hydrophobic closed conformation to a more solvent accessible, relatively open one (Fig. 2C middle). This is the first experimental evidence for the characterization of such an intramolecular allosteric trigger and bears important relevance to the product-release mechanism, as the epimerized D-Phe would subsequently dissociate from the E-domain. While the ‘assembly-line’ field was aware of a checkpoint, the trigger for ‘product-release’ has remained elusive till date. Second, not all of these allosteric perturbations are induced by aa or AMP-PCP alone (Fig. 2 and Fig. S4). Meaning, that even if the A-domain is induced into a closed conformation by the binding of aa (which bear significantly higher kD than ATP, Fig. 6) [23], the relative solvent-accessibility of the downstream PCP and the E-domains are changed only in the presence of dual substrates (Fig. 2C, middle and right). Third, the hinge region of the A-domain is not a single residue, unlike suggested earlier [24, 27], but a patch of residues consisting of a β-hairpin fold becomes more solvent-accessible in the dual condition (Fig. 2C left, 2D). Upon closer look at the E-domain, we noticed that in dual condition, mimicking the pre-catalytic state of the enzyme, a major shift was observed in the linker region between PCP and E-domain (Fig. 2C, right). This region (residue 605-621) was shown to form extensive, ordered interactions along the E-domain in the published crystal structure (PDBID: 5ISW), where the A-domain is absent, suggesting that the specific topology of the PCP-E interface plays a key role in propagation of the allosteric signal (Fig. 2A, in pale green).
Interdomain allosteric communication is distinctly different in selectivity-switched mutant, W239S.
Our HDX-MS results also revealed several key differences between GrsA-WT and W239S (Fig. 3). In its apo-form, W239S (relative to GrsA-WT) showed greater solvent-accessibility at the active site and from several other residues of the A-domain (residues 187-426) (Fig, 3A inset). The same datasets showed no perturbations to the downstream domains (Fig 3A). Meanwhile, the global accessibility of the A-domain is massively reduced upon binding their cognate aa (Fig. 3A, 3B). This suggests that upon aa binding, the conformations of the interacting residues are similar to GrsA-WT—providing a key explanation for the mechanism of selectivity switch. We noticed several participating residues, especially from AC become relatively solvent-inaccessible, suggesting a tight binding of AC to AN in the presence of aa (Fig. 3B, inset). This validates the yet unknown reason for success of A-domain mutants compared to the chimeric constructs [31] and supports the general concept of selectivity-switch by which the preferential affinity of an active site pocket can be changed. Thus, for future applications of bio-engineering, if aa-bound forms provide similar accessibility of solvents, it can be used as a successful marker for alteration of selectivity.
Intriguingly, we found that in the presence of aa, the critical β-hairpin of the A-domain is found to be solvent-inaccessible in W239S as compared to GrsA-WT (Fig. 3B inset, dark green). The β-hairpin hinge-region (residues 425-444) (Fig. 2D), containing the conserved Asp residue located within the A8 motif was reported [16, 24, 27] to be important for the adenylate to thioester transitions and in turn regulates the reaction rate (Fig. 1D). Although away from the active site, this region becomes secluded from solvent in W239S compared to GrsA-WT (Fig 2B inset, green, residues 410-426), suggesting that this hinge mechanism is what is critically altered even in case of minor perturbations such as a point mutation. In line with our results, other work to rigidify the linker region with proline mutations showed deficits in the adenylation reaction, although still favoring the adenylate state [32]. Thus, In the absence of the correct synchrony of the β-hairpin motion with respect to AC subdomain motion, the rate of product formation is significantly decreased (Fig. 2D) suggesting that the conformational flexibility of this region is critical to relaying the allosteric information.
Differences start to emerge when the dual condition (aa + AMP-PCP) is compared between W239S and GrsA-WT (Fig. 3C). Interestingly, there is little known knowledge about the mutual influence of ATP binding on the mechanism of aa binding or vice versa. We also observed that in the dual condition for W239S, there is an increased preference towards flexibility of the participating residues from the E-domain, including the patch implicated in stabilizing the PCP domain patch (residues 848-868, Fig. 3C inset, yellow), highlighting the non-compatibility of the A-domain mutation in other functional states. Our observation that the W239S point-mutation leads to an increased conformational flexibility in downstream domains is in alignment with the previously recorded decrease in downstream catalytic rates, such as the thiolation reaction [18, 19] (Fig. 2D). Taken together, we can identify that the path of allostery while originating from the A-domain’s active site, is clearly ending at the E-domain, far and distal from its proximity (~700 residues away). This network of allosteric contacts consists of three distinct patches: the β-hairpin joining the AN and AC subdomain, the interface between the PCP and the E-domain and lastly, the active site of the E-domain. The insights gained from the HDX-MS results are congruent with the previous observations of kinetic sampling of the molecule [21, 33]. The implications of these changes at the mechanistic level or in the context of domain-domain contacts remains a mystery as structures of the full-length protein remains to be characterized.
Finding the resting state of GrsA using Small Angle X-ray Scattering
We pursued SAXS experiments to circumvent the problem of static measurements from crystallography and the averaging effects of cryo-EM in finding the resting state. Our CD spectroscopy experiments validated that W239S had a very similar compact structure like GrsA-WT (Fig. S1). So, we determined an ensemble of envelopes for W239S using SAXS, which showed better monodisperse profile as compared to GrsA-WT. Even in these experiments we used AMP-PCP so that the resultant observations were purely allosteric in nature, impacted by binding rather than catalytic reactions. Characterization of all-substrate bound states (in different combinations of AMP-PCP +/ Tyr +/ PPant) were performed similar to analysis of the apo-form (Table 1).
The resulting scattering envelopes were distinctly more elongated than predicted by the Alphafold model, indicating that neither the apo-form nor any of the substrate-bound conditions were truly represented by the Alphafold model. We noticed from the apo-form that the resting isoform of the molecule is manifested as a highly elongated state (called ‘elongated state’ henceforth), which cannot compact into any tight geometric configuration compatible with catalytic activity. This is a remarkable finding, as this is the first evidence of the NRPS resting-state which is uniquely different from any of its previously crystallized catalytic conformations [16, 17] and the first complete envelope of the full-length GrsA protein (Fig. 4A-D). Furthermore, we found that W239S-apo existed in a 2-state conformational equilibrium (~50% each) of non-catalytic conformations (Fig. 4E).
The substrate-bound envelopes (Fig. 4A-C) showed markedly different scattering densities. When the scattering envelopes from the substrate-bound experiments were compared to the apo-form, unaccounted densities were present around the regions of AC and PCP, representative of their significantly dynamic state, while the AN and the E-domains remain fairly rigid (Fig. 4A-C). To address the unaccounted densities, we made combinatorial composites of two models, which represented the envelopes accurately: including the alphafold, adenylate, thiolation and elongated states (Fig. 4). The resulting two-state ensembles for each complex showed statistically improved fits than either one-state or three-state models (Fig. 4E-K, Table 1). The dimensionless Kratky plot (Fig. 4H) showed that the W239S-apo form is similar in compactness to the Tyr-bound form but not to the dual substrate form, providing a secondary validation for our HDX experiments.
TABLE. 1. SEC-SAXS analysis of substrate-bound complexes of W239S reveal a two-state ensemble.
The Alphafold model by itself does not fit the scattering envelopes of the apo- or the substrate-bound forms. When we included both the Alphafold model and the elongated state in the statistical analysis using FoxS [36],[37], the resulting 2-state ensemble matched the scattering profile of the W239s-apo condition (Fig. 4A) with significantly improved fit (Fig. 4E, inset, 4I). Replacing the Alphafold model with the adenylate state leads to a drastic improvement of the fitting (Fig. 4F, Table 1), suggesting that the presence of substrate allows for this conformational change by which the molecular ensemble now is comprised of elongated and adenylate states. It is curious to note that the elongated-state does not disappear upon addition of substrates, suggesting that this is a key intermediate of the molecule, which is often sampled in solution and appear in each of our complexes (Fig. 4A-C, D, H). Interestingly, on comparing the holo-form (with dual substrates) to the apo-form , the dual condition was almost identical to their Tyr-bound form in terms of the relative presence of the elongated state. (Fig. 4A-C, Table 1, Fig. S2).
Our results show that the scattering envelope from the holo-form, containing the PPant moiety, preferentially sampled the thiolation state (Fig. 4C, G, K) instead of the adenylation state or the Alphafold model. This is unique—as the generally accepted paradigm assumes that the thiolation state is triggered only by the reactionary cue of PPi release. Here, we are observing a distinctly different result: even without the release of PPi, in the presence of dual substrates at the active site, a ‘near-thiolation’ conformation is preferentially sampled. While this conformation was the predominant species explaining the scattering profile, the elongated-state was still evident from the envelope, truly solidifying our hypothesis of this being the resting state of the molecule (Fig. 4G, 4K, 4L, Table 1).
Finding the influence of binding of non-cognate amino acids on the resting state
We realized that this resting state of the molecule, if omnipresent, will also have an impact in providing the discriminatory property of the A-domain. However, it is hard to gauge what happens to the resting state of the full-length molecule in the case of a non-cognate aa binding at the active site of A-domain. The residues that create the selectivity filter are well established, but to the best of our knowledge there is no known literature about the exact molecular mechanism of substrate discrimination and what happens in the case of a non-cognate aa. To understand if the resting state of the molecule is affected by the identity of the substrate, we performed several MD simulations on the A-domain of GrsA-WT (Fig. 5). During the initial modelling phase of the simulations, Glycine (Gly) was added to the active site of the A-domain in place of its cognate Phe and energy-minimized into the adenylate or Alphafold state (Fig 5 A-D, Table S3). The adenylate state was conformationally identical to the crystallized form (PDBID-1AMU), except for the identity of the aa.
During the course of the simulation, we observed a clear intermediate, representing the elongated state (Fig. 5D) of the A-domain, where the active site is ruptured by the action of the hinge residues (discussed earlier) leading to a partial unfolding of the AC from the AN.
This allows for the active site, which started as a minimally solvated cavity (Fig. 5C), to be completely flooded with solvent leading to an immediate release of the Gly (Fig. 5D). The dissociation takes place within 15 ± 5 ns suggesting that even if there is a binding event of a non-cognate aa at the active site, the reaction is less likely to proceed to next catalytic configurations. To validate our observation, we measured the average radius of gyration of the A-domain along the time-steps and found that they stably maintained throughout the course of the simulation (Fig. 5E). Comparing our CD spectroscopy results (Fig. S1) to our MD results, we inferred that the unique 2-step denaturation profile can arise from the disjoining of the AC-subdomain from AN separate from the global unfolding event associated with the loss of secondary structure from the entire protein molecule. The RMSD of the active site also stayed stable in both cases; with cognate (Phe) as well as non-cognate (Gly) aa (Fig. 5F, G). The Gly dissociation towards the end of the representative simulation was verified with an abrupt rise in the RMSD value (Fig. 5G), as well the distance from the conserved D235 (Fig. 5H). The conserved residues D235 and K517 belong to AN and AC subdomains respectively, providing us a unique distance that can be used to identify the hinge-based disjoining/unfolding of AC from AN.
To identify if there was a correlation between the aa unbinding and the hinge-based opening of the A-domain into its subdomains, we performed a dynamic cross-correlation analysis (Fig. 5J) of the distance vectors generated previously (Fig. 5H-I). The sliding dot product revealed that during the initial phase of the simulation, the aa and residues at the active site was highly correlated (close to 28%), which slowly declined before spiking again to mark another binding event. As the aa dissociated, the correlation of motion drops with respect to the motion of the unfolding event (Fig 5J). The Phe-bound active site in comparison stayed within the confidence interval of <95%, for almost the entirety of the simulation, showcasing that the motions associated with that active site are truly small and Brownian in nature, without any major concerted movements.
These MD results show how the disrupted active site of the A-domain leads to the formations of the elongated state and can be achieved even after the erroneous binding of a non-cognate aa, validating our hypothesis of a global non-reactionary resting state of the molecule. We performed similar MD simulations starting from the Alphafold model with the aa placed in the active site (Fig. S4) and noticed very similar patterns. Moreover, even when the AC subdomain rotated out, the cognate Phe did not dissociate leading us to hypothesize that the AC had minimal effect towards stabilization of that ligand in the active site.
It is understandable that the protein molecule stays in the elongated catalytically non-compatible configuration until the correct binding event occurs to avoid faulty initiation of the non-cognate aa from the cellular pool. Biophysically, our AN-only mutant (residues 1-430) also proved this hypothesis with similar binding kD in ITC experiments for Phe (Fig. S3G) as compared to the full-length GrsA-WT(Fig. 6E). Similarly, we wanted to verify our findings from HDX-MS for the dual condition with an orthogonal approach. The working hypothesis was that if there was a change in the binding determinants of the substrates in W239S, we should be able to capture that using kD measurements. Our ITC results validated that both GrsA-WT and W239S bear similar binding affinity for the cognate aa (1-5 µM) (Fig. 6E-H). The dissociation constants for the aminoacid were congruent with previously published values for GrsA-WT (Fig. 6E). However, the AMP-PCP binding affinity to GrsA-WT (Fig. 6A, B) is about a fold higher than in the presence of Phe. W239S binding to AMP-PCP is unaffected by the presence of Tyr (Fig. 5C, D), although at 2-fold lower affinity compared to GrsA-WT.