Gαi1 inhibition mechanism of ATP-bound adenylyl cyclase type 5

Conversion of adenosine triphosphate (ATP) to the second messenger cyclic adenosine monophosphate (cAMP) is an essential reaction mechanism that takes place in eukaryotes, triggering a variety of signal transduction pathways. ATP conversion is catalyzed by the enzyme adenylyl cyclase (AC), which can be regulated by binding inhibitory, Gαi, and stimulatory, Gαs subunits. In the past twenty years, several crystal structures of AC in isolated form and complexed to Gαs subunits have been resolved. Nevertheless, the molecular basis of the inhibition mechanism of AC, induced by Gαi, is still far from being fully understood. Here, classical molecular dynamics simulations of the isolated holo AC protein type 5 and the holo binary complex AC5:Gαi have been analyzed to investigate the conformational impact of Gαi association on ATP-bound AC5. The results show that Gαi appears to inhibit the activity of AC5 by preventing the formation of a reactive ATP conformation.


Introduction
The nucleotide adenosine triphosphate (ATP) plays a crucial role in metabolism, not only as primary energy carrier, but also in intracellular signal transduction, acting as a substrate for the formation of the second messenger cyclic adenosine monophosphate (cAMP). cAMP is produced through deprotonation and cyclization of ATP, catalyzed by the enzyme adenylyl cyclase (AC) [1,2]. Nine different membrane-bound isoforms of AC are known in nature, ranging from AC1 to AC9 [3]. Structurally, such isoforms consist of two membrane regions, M1 and M2, and two cytoplasmatic domains, C1 and C2, of which the C1/C2 interface harbours the enzyme's active site. Once cAMP dissociates from the C1/C2 interface, the second messenger is able to transduce a signal in the cell by, for example, activating protein kinase A and affecting ion channels located in the lipid bilayer [4][5][6]. Apart from their primary function as signal transducers, ACs may also work as signal integrators since these enzymes can act as decision functions, determining the amount and the time of cAMP release. A typical decision function, characteristic for several AC enzymes, is the detection of co-occuring signaling a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 template used for modelling the active rat AC5 (UniprotKB Q04400) conformation in the binary complex consists of the C1 and C2 domains (C1a from canine AC5 and C2a from rat AC2) from 1CJK [22]. The Gα i1 subunit was taken from reference [29], a modelled structure of the myristoylated Rattus norvegicus Gα i1 subunit (UniprotKB P10824) interacting with guanosine-5'-triphosphate (GTP) and a Mg 2+ ion. The active myristoylated Rattus norvegicus Gα i1 is referred to as Gα i1 because only a myristoylated form of Gα i1 was used in the simulations. Myristoylation, crucial for Gα i1 's inhibitory function [18,19], is a post-translational modification of the N-terminus of Gα i1 that results in the covalent attachment of a 14-carbon saturated fatty acid to the N-terminal glycine residue of Gα i1 via an amide bond. The modelled AC5 and Gα i1 structures were used for docking Gα i1 on AC5's C1 domain to obtain the initial binary complex conformation for simulation.
The HADDOCK web server [30] was used for docking ten conformations of the active Gα i1 subunit to AC5's catalytic domains in the holo form as described in references [26,29]. The active region of Gα i1 was defined in HADDOCK as a large part of the alpha helical domain (residues 112-167), the switch I region (residues 175-189) and the switch II region (residues 200-220), allowing for a large area on the Gα i1 subunit surface to be taken into account during docking. The active region of AC5's C1 domain was defined as the α1 helix (residues 479-490) and the C-terminal region of the α3 helix (residues 554-561) based on results from gel filtration that show that Gα i1 is unable to interact with C2 and mutagenesis experiments that have confirmed that Gα i1 's main interactions with AC is via the C1 domain [18]. Ten snapshots of Gα i1 were used for docking the Gα subunit to the catalytic domain of AC5. These snapshots were extracted at time intervals of 0.5 ns from the end of the classical MD trajectory of Gα i1 (around 1.9 μs) described in reference [29]. The same three criteria for complex selection as in reference [26] were applied: (1) the absence of overlap between the C2 domain and Gα i1 , (2) no overlap with the GTP binding region of Gα i1 and the C1 domain of AC5, and (3) presence of similar complexes in the top-ten docking results of the docking calculations performed for all ten used Gα i1 conformations. This last criterium increases the probability that the docked orientation of the selected complex is robust as similar AC5:Gα i1 complexes can be obtained using different conformations of Gα i1 . Besides meeting these criteria, the orientation of Gα i1 with respect to the membrane (absent in the simulation) was assessed to select a binary complex that would be consistent with a membrane environment. Moreover, the position of Gα i1 on the C1 domain was required to be similar to its position on C1 in the absence of ATP [26].

Classical molecular dynamics simulations
Holo Gα-free AC5 and the holo AC5:Gα i1 complex were solvated with *31 000 and *53 000 water molecules respectively. K + and Cl − ions were added to the system to obtain a physiological concentration of 150 mM. An excess of K + ions was used to neutralize the system. The force fields used for the protein and the water molecules are AMBER99SB [31] and TIP3P [32], which were employed by Gromacs 5.1.2 [33,34] to perform the simulations. The adjusted force field parameters for the Cl − and K + ions were taken from Joung et al. [35]. For GTP and ATP, the force field parameters generated by Meagher et al. were used [36]. The Mg 2+ ion parameters originated from Allnér et al. [37] and the parameter set for the myristoyl group was taken from reference [29]. Both the holo Gα-free AC5 and the holo AC5:Gα i1 complex were simulated for 3 μs at 310 K and 1 bar using a Nosé-Hoover thermostat [38,39] and an isotropic Parrinello-Rahman barostat [40]. A replicate of the holo AC5:Gα i1 system with different initial velocities was simulated for *0.7 μs. Electrostatic interactions were calculated with the Ewald particle mesh method [41] with a real space cutoff of 12 Å. Bonds involving hydrogen atoms were constrained using the LINCS algorithm [42]. The integration time step was set to 2 fs. Images of both protein systems were prepared with Pymol [43].

Stability and conformational analysis
In order to shed light on the molecular basis of the Gα i1 inhibition mechanism, the holo Gαfree AC5 protein and the holo AC5:Gα i1 complex have been simulated for 3 μs. The rootmean-square deviation (RMSD) of both systems (Fig 2) shows that the subunits in each complex are stable, yet, C1 domain converges between 0.15-0.20 nm, while C2 domain stabilizes around 0.30 nm. However, the stability of the secondary structures of C1 and C2 (see S1 and S2 Figs Concerning the RMSD of the overall complex, holo Gα-free AC5 appears to deviate significantly less from its initial structure than holo AC5:Gα i1 (S4 Fig). This difference is to be expected as the initial AC5 conformation in both systems originates from an AC:Gα s structure in the absence of Gα i1 . Hence, in AC5:Gα i1 , AC5 is affected by Gα s removal as well as Gα i1 association, inducing a larger perturbation on the AC5 protein than in the Gα-free AC5 condition. However, the decreased radius of gyration (Rg) of the binary complex (Fig 2) implies that interdomain interactions enhance during simulations. Indeed, the number of hydrogen bonds between AC5 and Gα i1 in the binary complex was found to increase along the simulation, which is consistent with the decrease in Rg, indicating an overall stabilization of the complex. Hence, these results suggest that the difference in overall RMSD could be due to an alteration in conformation, leading to a change in intermolecular interactions with respect to the starting structure with a possible reorientation of the system's subunits. Such a reorientation is shown by comparing the first and the last frame of the holo AC5:Gα i1 simulation using the DynDom software [44,45] (Fig 3). This analysis identified a rotational mode defined by a rotation of * 41˚of the Gα i1 protein around the rotational axis reported in Fig 3. This trend is not surprising as the initial conformation of AC5 in the binary complex originates from an AC:Gα s structure in the absence of Gα i1 .
The root-mean-square fluctuation (RMSF) per residue was also calculated for the C1a and C2a domains in both the Gα-free AC5 and binary complex (S5 Fig). Concerning the C1a domain, Gα i1 reduces the flexibility of the C-terminal part of the α3 helix (residues 550-560) compared to Gα-free AC5. This change in flexibility is not surprising since this region strongly interacts with Gα i1 , resulting in reduced mobility. In the C2a domain, the β4 0 −β5 0 strand (residues 1190-1200) is considerably less flexible in the holo AC5:Gα i1 complex than in the apo binary and ternary AC5 complexes [27], indicating that ATP association can stabilize the mobility of the C1/C2 interface in holo AC5:Gα i1 due to the positioning of the substrate.

Gα s binding site
A crucial site on C2 that could be affected by the conformational changes that occur within AC5, is the Gα s binding site. The effect of the presence of Gα i1 with respect to potential Gα s association was investigated by comparing the conformations sampled by helix α2' and α3' (constituting the Gα s binding site) in Gα-free AC5 and in the AC5:Gα i1 complex (Fig 4). In the Gα-free AC5 system, the α2'-α3' distance (Fig 4) stabilizes around *1.7 nm, thus showing an increase of about 2 Å with respect to the distance in the X-ray structure of the AC:Gα s complex (i.e. 1.48 nm, PDB ID 1CJK) [21]. However, when Gα i1 interacts with AC5, this distance is significantly decreased from *1.4 to *1.1 nm.
This conformational change in AC5:Gα i1 could potentially lower the probability of Gα s complexation due to a reduced accessibility of the Gα s binding site. A decreased α2'-α3' distance was also observed in the apo AC5:Gα i1 complex [27], suggesting that such conformational change is independent from ATP association.
Moreover, the suggested reduced ability of AC5:Gα i1 to associate with Gα s indirectly proposes a favourable route towards the possible formation of a holo ternary Gα s :AC5:Gα i1 complex. These results suggest that the most plausible route to form a ternary complex would be to first form holo AC5:Gα s and subsequently bind Gα i1 , which was also suggested for the apo form [46].

ATP binding site
Although a reduced probability in Gα s association can negatively impact AC5's catalytic activity as Gα s stimulates ATP conversion, sampling an ATP conformation that is able to convert to cAMP is imperative for AC5's catalytic function. The first part of the conversion of ATP to cAMP consists of the deprotonation of O3 � (Fig 5), which is followed by a nucleophilic attack of O3 � on the phosphorous atom of the α phosphate, P α , producing cAMP. It is suggested that the first step of the reaction mechanism, the deprotonation of O3 � , would require O3 � to coordinate to the neighbouring Mg 2þ A ion in the active site, which would lead to an O3 � À Mg 2þ A distance of *2.7-2.0Å [47][48][49][50]. Therefore, the distance between ATP's O3 � and Mg 2þ A plays a crucial role and was monitored along both holo trajectories to study the potential catalytic activity of the conformational ensemble of ATP (Fig 5).
In the Gα-free AC5 system, the O3 � -Mg 2+ distance was mainly found to oscillate around *5.4 Å (5Å<O3 � -Mg 2+ <6Å is sampled in 75% of the trajectory), close to the X-ray distance of 5.25 Å (Fig 5C) [21]. To a lesser extend, a second conformation with an O3 � -Mg 2+ distance larger than 6.0 Å (O3 � -Mg 2+ >6Å is sampled in 11% of the trajectory) was also sampled in Gαfree AC5. This conformation is suggested to be inactive as the nucleophilic attack of O3 � is unable to be performed at this distance ( Fig 5D). Occasionally, the O3 � -Mg 2+ distance decreased to a value lower than 3.5Å in the Gα-free AC5 simulation (O3 � -Mg 2+ <3.5Å is sampled in 0.33% of the trajectory) ( Fig 5B). Intriguingly, albeit poorly sampled along the simulation, this conformation could be close to the ATP structure undergoing the deprotonation step and can, therefore, be identified as the catalytically active or near-attack conformation, according to Hahn et al. [50]. ATP appears to adopt a different conformation in the AC5:Gα i1 complex. The O3 � -Mg 2+ distance increases up to *8.0 Å in the first 300 ns, and equilibrates to a stable value of *7.2 Å

Fig 5. ATP sampled conformations. The distance between ATP's ribosyl oxygen O3 � and the neighbouring Mg 2þ
A is reported as function of time for both holo free AC5 and the AC5:Gα i1 complex in image (A). The starting distance corresponding to the value found in the X-ray structure (PDB ID: 1CJK) is shown by a green dashed line. Snapshots of three representative conformations sampled by ATP and its neighbouring Mg 2+ ions are shown in stick (for ATP) and ball (for Mg 2+ ) representation and were extracted from the simulated trajectories of holo Gα-free AC5 at 500 ns (B) and the holo AC5:Gα i1 binary complex at 0 ns (C) and at 1μs (D). for the last 2.3 μs of the simulation (O3 � -Mg 2+ >6Å is sampled in 96% of the trajectory), which makes the cyclization reaction unlikely to occur (Fig 5D). The absence of O3 � À Mg 2þ A coordination seems to indicate an inhibited state of ATP in the binary complex as a near-attack conformation of ATP [21] cannot be sampled. Additionally, the main inactive conformation sampled by ATP in AC5:Gα i1 adopts a structure incapable of sampling other states, whereas in the Gα-free AC5 system, an equilibrium between the different ATP conformations is already present on the microsecond timescale. In order to check the reproducibility of the observed conformational change of ATP in AC5:Gα i1 , a second MD simulation of the holo AC5:Gα i1 complex was performed using different initial velocities. This simulation of *0.7 μs confirms the presence of an inactive conformation of ATP in AC5:Gα i1 since the same trend in conformational change was observed for the nucleotide as in the first AC5:Gα i1 simulation (S6 Fig). The inactive ATP conformation sampled in AC5:Gα i1 could be the result of conformational changes at the C1/C2 interface, perturbing interactions between ATP and AC5's active site. When comparing AC5:Gα i1 and Gα-free AC5, a crucial difference seems to be a change in ATP's interactions with the C2 domain around its adenosine group which takes place in the first 200 ns of the simulation. Whereas in the Gα-free AC5 system the interactions of the adenine moiety and the active site are stable (Fig 6), an irreversible loss of a hydrogen bond can be observed in the AC5:Gα i1 system between adenine and K1124, connected to a second hydrogen bond with D1198 (Fig 6, S7 and S8 Figs). The loss of the K1124-ATP and D1198-ATP hydrogen bonds appears to destabilize the position of the adenine ring in the active site, which reorients shortly after (Fig 6, S8 Fig). The conformational change of the adenine ring in turn seems to impact the interaction of ATP with Mg 2þ A as it appears to coincide with the increase in distance between O3 � and Mg 2þ A , leading to the suggested inactive state of ATP (Figs 5 and  6). This mechanism was also observed in the second MD simulation of AC5:Gα i1 (S6-S8 Figs).
Both K1124 and D1198 are conserved across species (rat, dog, human, rabbit and mouse) and are also conserved between all membrane-bound isoforms of AC (AC1-9 in mouse). The conservation of these two residues in the active site could indicate that both residues indeed play important roles in maintaining the catalytic function of adenylyl cyclase. The performed simulations suggest that K1124 and D1198 are crucial for stabilizing the conformation of ATP in the active site and when these interactions are lost, an inactive ATP state is sampled.
Hence, these results suggest the conformational change of the ATP substrate into an inactive state is promoted by Gα i1 :AC5 complex formation. Although Gα i1 association appears to inhibit the catalytic function of AC5 destabilizing the active conformation of ATP, it cannot be excluded that Gα i1 could also affect AC5's activity by impairing its substrate binding affinity. However, it was experimentally determined by Dessauer and coworkers [51] that Km does not significantly change upon Gα i or Gα s binding. Therefore, the Gα i1 -induced conformational change of ATP in AC5's active site seems to be the main effect of Gα i1 's inhibition mechanism on AC5.

Conclusions
Here, the structural and dynamical properties of the AC5:Gα i1 complex have been characterized in order to understand the inhibition mechanism triggered by Gα i1 association. The conformation adopted by the ATP substrate in the binary complex significantly reduces the probability of an S N 2 reaction between O3 � and Pα, potentially preventing the formation of cAMP. These findings demonstrate that although ATP is able to interact with the C1/C2 interface in both AC5:Gα i1 and Gα-free AC5 systems, the catalytic activity of AC5 seems to be impaired by Gα i1 association, affecting the conformation of the substrate. Additionally, these results show that, as already observed in the apo form, also in holo AC5, Gα i1 association induces a conformational change in the Gα s binding region on the AC5 C2 domain, thus resulting in a closure of this binding site. Based on this observation it is suggested that Gα i1 association prevents ternary complex formation via this route. This potential pathway, found in previous studies of the apo form, is also reported here for the holo form and confirms the existence of a potentially favourable route towards ternary complex formation from the AC5: Gα s complex.