Viral packaging ATPases utilize a glutamate switch to couple ATPase activity and DNA translocation

To establish that viral DNA packaging ATPases utilize a glutamate switch to couple ATP binding and DNA gripping, it is necessary to show the following:

To demonstrate 1, we carried out atomistic molecular-dynamics (MD) simulations of four different packaging ATPases and calculated the two-dimensional free-energy landscapes of their putative catalytic glutamate with respect to its (χ₁, χ₂) rotamer dihedral angles in different nucleotide-bound states. We assumed that if a glutamate switch were present, we would observe computed free-energy minima corresponding roughly to the “active” and “inactive” clusters seen experimentally (Fig. 1). To demonstrate 2, we generated in silico mutants of the ATPases targeting the putative glutamate switch and recomputed the free-energy landscapes. From how the landscapes changed with respect to the type of mutation, we could infer the ability of the wild-type residue to function as a glutamate switch. To rule out that the tight-binding transition positions multiple residues that each contribute a little energy toward the cumulative catalytic glutamate free-energy landscape, we also considered ATPases held in a “pre-tight-binding” pose. To demonstrate 3, we calculated mutual information (MI) from microsecond-long MD simulations. MI is a metric that quantifies correlation between two parties, in this case protein residues. From this correlation, we infer communication. If the glutamate switch mediates the signaling pathway, we would expect the glutamate switch to have high MI—and thus high communication—with both the ATPase active site and DNA-gripping residues.

While we performed computational studies on four viral packaging ATPases to generalize our predictions, we chose to experimentally test a subset of these ATPases. Specifically, our simulations predict that a helical motif between the ATPase active site and DNA-gripping residues shifts in the presence of ATP, but not ADP. Thus, we solved the structure of the P74-26 packaging ATPase bound with ADP to compare against the previously reported ATP-analog bound structure (11). Additionally, to interrogate the signaling pathway in an actively packaging assembly, we simultaneously characterized the ATPase activity and DNA translocation activity of mutant φ29 ATPases. This allowed us to determine if a mutant introduced in the signaling pathway would decouple these two otherwise tightly coupled activities.

Available crystal structures of the Sf6, φ29, and ascc-φ28 packaging ATPases show that the predicted ASCE catalytic glutamate residues point their carboxylate group away from the ATPase active site and toward analogous arginine residues (SI Appendix, Fig. S1). This is observed in the Sf6 crystal structures solved with either ATP or ATP analog. Because this type of interaction is indicative of a glutamate-switch mechanism in AAA+ proteins (Fig. 1), the crystal structures are the first indication that these arginine residues may act as glutamate switches.

Free-energy landscapes calculated for the proteins in the apo state show that the catalytic glutamate has two free-energy minima, which correspond to the identified inactive and active rotamer states. In general, there is preference for the inactive pose (SI Appendix, Fig. S2). This contrasts with the preference of an isolated glutamate residue in solution, which prefers the active pose (SI Appendix, Fig. S3), suggesting that neighboring residues influence the catalytic glutamate’s preference for the inactive pose in the apo ATPases.

Free-energy landscapes calculated for the φ29 (Fig. 2 A and E), Sf6 (Fig. 2 B and F), and ascc-φ28 (Fig. 2 C and G) ATPases in the ATP-bound and ADP-bound states show that the catalytic glutamate’s rotameric state varies with the presence or absence of the γ-phosphate of ATP. In the ATP-bound state, the catalytic glutamate prefers the active pose, pointing its carboxylate group toward ATP (Fig. 2 A, B, and C). In the ADP-bound state (Fig. 2 E, F, and G), the global free-energy minimum of the catalytic glutamate changes to prefer the inactive pose, pointing its carboxylate group toward the putative glutamate-switch arginines. This indicates that the presence of the γ-phosphate of ATP is necessary and sufficient to induce local conformational changes that stabilize the catalytic glutamate residue in the active pose. We note that the WA arginine residue moves into the binding pocket upon ATP binding and moves out of the binding pocket in the absence of the γ-phosphate. We have previously reported that this conserved WA arginine in phage λ functions analogously to the “sensor II motif” arginine found in AAA+ enzymes, whose role is to sense the presence of the γ-phosphate and induce further conformational changes (10). The results presented here further support this assignment.

We also investigated the packaging ATPase from the P74-26 thermophilic phage. This ATPase does not contain an analogous glutamate-switch arginine as in Sf6, φ29, and ascc-φ28 packaging ATPases described above. Instead, it contains a glutamine residue, which we suspected could act as the glutamate switch due to its similar position relative to the catalytic glutamate, despite significant divergence in local tertiary structure (SI Appendix, Fig. S5). Indeed, the catalytic glutamate (χ₁, χ₂) free-energy landscape in the P74-26 ATPase also varies with the presence or absence of the γ-phosphate of ATP (Fig. 2 D and H). In the ATP-bound state, the catalytic glutamate is only stable in the active pose (Fig. 2D). In the ADP-bound state, a metastability emerges in the inactive pose (Fig. 2H). This metastability is also found in the apo state of the P74-26 ATPase (SI Appendix, Fig. S2). While preference is not swapped between active and inactive like in the other three ATPases above, we note that this difference is perhaps expected because the P74-26 ATPase uses a polar residue for the glutamate switch as opposed to the positively charged residues in the φ29, Sf6, and ascc-φ28 ATPases. Thus, the nucleotide-bound state can be discriminated by the ATPase through the interaction between the catalytic glutamate and glutamate-switch residues.

To determine the role of the ATP tight-binding transition on the rotameric state of the catalytic glutamate, we calculated the catalytic glutamate (χ₁, χ₂) free-energy landscapes of the φ29, Sf6, and P74-26 packaging ATPases in the “pre-tight-binding” conformations, where ATP is present but the ATPase active site has not yet adapted to its presence. These pretight-binding poses are characterized by the WA arginine pointing out of the binding pocket and an increased distance between the WA and WB motif backbones (SI Appendix, Fig. S6), based on our two previous studies of tight-binding conformational changes (9, 10). We find that for φ29 (Fig. 3A) and Sf6 (SI Appendix, Fig. S7), the catalytic glutamate prefers the inactive pose prior to the ATP tight-binding transition. In the case of P74-26 (Fig. 3B), the metastability of the catalytic glutamate’s inactive position is strengthened. However, it remained unclear whether the preference for the inactive pose of the catalytic glutamate could be attributed solely to its interaction with the putative glutamate-switch residue or if ATP tight binding induces multiple small conformational changes that cumulatively modulate the catalytic glutamate’s preference. Thus, we repeated the above calculations with catalytic glutamate mutant ATPases φ29 R53(L/K) and Sf6 R51(L/K). Leucine was chosen as a variant to remove the charge group from the putative glutamate-switch residue while maintaining a similar bulk to the wild-type (WT) arginine. Lysine was chosen as a second variant in the hope that it would weaken, but not abrogate, its hold on the catalytic glutamate residue so that the mechanism could be probed experimentally. Under the same conditions, where the WT catalytic glutamate prefers the inactive pose, the R→K mutants’ catalytic glutamate also preferred the inactive pose (SI Appendix, Fig. S7). However, under these conditions the R→L mutants’ catalytic glutamate preferred the active pose (Fig. 3C and SI Appendix, Fig. S7). Thus, the positive charge donated by the arginine residue is necessary to hold the catalytic glutamate in the inactive pose prior to the ATP tight-binding transition, and the arginine is thus assigned to be a glutamate-switch residue.

Because the WT P74-26 packaging ATPase potentially uses a polar residue as its glutamate switch and the inactive pose is predicted to be metastable, it offers a unique opportunity to probe the effects of potentially strengthening, rather than weakening, the glutamate switch’s interactions with the catalytic residue. Thus, we chose to study the Q72K mutant to see if the change from polar to positively charged would cause the catalytic glutamate free-energy landscapes to be similar to other packaging ATPases. Indeed, the introduction of a positive charge in the P74-26 glutamate-switch position strengthens the catalytic glutamate’s preference for the inactive pose compared to the WT (Fig. 3D). In total, our free-energy landscapes calculated for mutant ATPases enable us to assign these residues as glutamate switches.

To probe the relationship between ATP tight binding and DNA interaction, we performed microsecond-long equilibrium MD simulations of ATP- and ADP-bound ATPases. In all cases, a helical motif between the ATPase active site and proposed DNA-gripping residues rotates upon ATP binding (Fig. 4). In the recently solved cryogenic electron microscopy structure of the φ29 stalled during packaging (15), Lys56 directly contacts the DNA phosphates. Lys56 is part of the rotated helical motif and is only a few residues downstream of the glutamate switch Arg53. Our simulations predict that relative to the ADP-bound state, ATP binding moves Lys56 (Fig. 4A) such that it would be donated into the pore of the ATPase ring. This suggests that the response of the catalytic glutamate and glutamate switch to ATP binding is coupled to key residues believed to participate in DNA binding. For both the Sf6 (Fig. 4B) and ascc-φ28 (Fig. 4C) ATPases, our simulations predict that shifting the helical motif also repositions nearby proposed DNA-gripping motifs in a neighboring beta hairpin through steric interactions. Thus, shifting this helical domain signals nucleotide occupancy to DNA-gripping residues. We note that it has not been experimentally determined which residues of the Sf6 packaging ATPase grip DNA. However, there are similar backbone shifts in Sf6 Arg82 as in its analog P74-26 Arg101, which has been demonstrated to grip DNA (23).

To test these predictions, we determined the crystal structure of the ADP-bound P74-26 packaging ATPase to a resolution of ∼1.9 Å, using the same procedure used earlier for determining its apo and ATP-bound structures (11). Our structure of the P74-26 ATPase bound to ADP appears nearly identical to the apo state (Fig. 4D); in contrast, the ATP-analog-bound state exhibits a pronounced displacement in the helical motif partitioning the ATPase active site from the DNA-gripping residues. Therefore, we infer that the presence of a γ-phosphate actuates the conformational change, in agreement with our computational predictions in the φ29, Sf6, and ascc-φ28 packaging ATPases. We note that the P74-26 ATPase is unique among the four ATPases studied here, as its glutamate switch is within this helical motif itself and not adjacent to the motif as in the φ29, Sf6, and ascc-φ28 ATPases; this difference in position may be related to the difference in the appearance of a polar residue as opposed to a positively charged residue in the glutamate-switch position.

To better understand the mechanism by which nucleotide occupancy is signaled through the helical motif and to the DNA-gripping residues, we carried out MI calculations. In particular, we calculated the MI between the rotamer/backbone dihedral angles of the identified glutamate-switch residue and all other residues in each protein using the CARDS (Correlation of All Rotameric and Dynamical States) method (24). We found that in the simplest case of φ29 (where the primary DNA-gripping residue is located within the helical motif), the glutamate switch has high MI with both the catalytic glutamate and the predicted DNA-gripping motifs (Fig. 5 and SI Appendix, Fig. S8). In the other cases (where the primary DNA-gripping residue neighbors but is not part of the helical motif), the glutamate switch has high MI with both the catalytic glutamate residue and the N-terminal of the rotated helical motif, and the DNA-gripping residues have high MI with at least one residue in the helical motif. However, the glutamate switch and DNA-gripping residues do not have high MI with each other. We note that whereas CARDS can capture both direct MI (between residues i and j) and indirect MI (residue i interacting with residue j by way of residue k), it relies on correlating rotamer and backbone dihedral angles. Because of this dependence on dihedral angles, the target-site analysis may not necessarily predict high MI between the two residues i and j if residue i induces a cartesian translation of j by changing backbone dihedral angles of residues j − 1 and j + 1. Thus, we assert that the two target-site analyses that predict that the glutamate switch and DNA-gripping residues both have high MI to different parts of the same helical motif still indicates communication between the two residues.

To test our computational predictions, we experimentally characterized select mutant enzymes predicted to disrupt the signaling pathway. For this purpose, we chose the φ29 system, which is one of the most well-developed packaging systems among those studied. Importantly, assays for this system can be performed on motors that are actively packaging DNA into viral procapsids. This allows for separate and simultaneous measurements of ATPase activity and DNA translocation activity. Thus, it is an ideal system to study the coupling between the ATPase active site and DNA gripping residues. We found that changes to the glutamate-switch residue itself either do not affect overall function (R53K) or abrogate function altogether (R53L), which may be due to slight changes in protein folding leading to defective oligomerization (Fig. 6). We also interrogated the signaling pathway without disturbing the glutamate-switch residue. Our MI calculations predict that the glutamate switch Arg53 interacts through Glu58 to regulate the position of DNA-gripping Lys56 (Fig. 5). Thus, we speculated that E58A would be a viable variant to disrupt the signaling pathway. MD simulations of an ATP-bound E58A variant predict that its DNA-gripping Lys56 is positioned similarly to the apo state in WT (Fig. 6). Thus, we predicted that the E58A variant may not competently bind DNA despite not changing any DNA-gripping residue. We then separately characterized the ATPase activity and packaging activity of an E58A variant and found that while the ATPase activity of the E58A variant is at near-WT levels, the DNA translocation activity is drastically reduced. This is ostensibly because the motor can no longer grip DNA tightly in response to ATP binding (Fig. 6). The experimentally observed decoupling of ATPase activity and DNA translocation supports the computational assignment that the glutamate switch signals DNA-gripping Lys56 through Glu58 in the φ29 packaging ATPase.

Initial structures for packaging ATPases were taken from their respective PDB depositions: P74-26 (4ZNI, apo; 4ZNL, ATP-analog bound) (11), Sf6 (4IDH apo; 4IEE, 4IFE, ATP-analog and ATP bound) (29), φ29 (5HD9, ATP-analog bound, ATP analog not resolved) (13), and ascc-φ28 (14). Simulations of the ADP-bound P74-26 structure start from the solved crystal reported herein (PDB: 7KS4). Mutant proteins were generated in silico by replacing side chains with rotamers chosen from the Dunbrack rotamer library (30) via swapaa in University of California, San Francisco Chimera (31). Because the C-terminal domain does not directly interact with the ATPase active site during packaging, the C-terminal domains of the Sf6 and ascc-φ28 packaging ATPases were truncated to reduce system size and computational cost; structures of P74-26 and φ29 only contain the ATPase domain.

All-atom MD simulations were performed with a 2 fs time-step integrator in the AMBER16 and AMBER18 packages using the Amber ff14SB force field (32) to describe protein interactions. The SHAKE algorithm was used to constrain bonds connecting hydrogen atoms to heavy atoms. Simulations were performed with graphical processing unit-accelerated sampling utilizing the particle mesh Ewald summation to correct for long-range interactions (33). All proteins were centered in an octahedral periodic box with a minimum 14 Å padding. The box was filled with explicit solvent using the TIP3P model for water. ATP parameters were taken from the AMBER parameter database (34). Each system was subjected to 300 steps of steepest descent and conjugate gradient energy minimization. Systems were heated slowly from 100 to 310 K over the course of 100 ps in the canonical ensemble. Then, each system was equilibrated for 100 ns in the isobaric-isothermal ensemble at 1 bar before free-energy sampling was performed. The Monte Carlo barostat and the Langevin thermostat were used to control pressure and temperature.

We used the combined approach of umbrella sampling and the weighted histogram analysis method (WHAM) to calculate the two dimensional free-energy landscapes. The (χ₁, χ₂) angles of the catalytic glutamate residue were harmonically restrained with spring constant k = 30 kcal/mol/rad² to windows spanning the range of the desired reaction coordinates. Because most of the distinction between active and inactive poses manifests as change in the χ₂ variable, we separated χ₂ windows by 7° and χ₁ windows by 20° to ensure ample sampling along the primary reaction coordinate. To hold an ATPase in the “pre-tight-binding” pose, the WA arginine side-chain dihedral angles were restrained by the same harmonic potential. Each window was simulated for 3 ns, and data points were collected every 100 fs. The potential of mean force was calculated using wham-2d (35), as described by Kumar et al. (36). To estimate the error in our calculations, we randomly split our sampling windows into three subwindows and calculated free-energy landscapes from each set of subwindows. From these independent landscapes, we calculated the SEM at each grid point. In general, the error in our calculated free-energy landscapes ranges from 0.01 to 0.1 kcal/mol.

To predict signaling pathways connected to the glutamate switch, we calculated the MI of the side-chain and backbone dihedral angles with every other residue in the protein. We used the Enspara package (37) to calculate the CARDS (24) from 1 μs long equilibrium simulations, saving frames every 25 ps. The holistic correlation $I_H$ between two residues $X$ and $Y$ is defined to be $I_H\left(X,Y\right)=\overline{I_{ss}\left(X,Y\right)}+\overline{I_{sd}\left(X,Y\right)}+\overline{I_{ds}\left(X,Y\right)}+\overline{I_{dd}\left(X,Y\right)}$. That is, CARDS calculates MI from concerted structural changes $\overline{I_{ss}\left(X,Y\right)}$ (a side-chain/backbone dihedral angle in one residue that influences a sidechain/backbone angle in another), structural changes that promote/hinder conformational disorder $\overline{I_{sd}\left(X,Y\right)}+\overline{I_{ds}\left(X,Y\right)}$ (a side-chain/backbone dihedral angle in one residue that allows another residue to become flexible, and vice versa), and concerted conformational disorder $\overline{I_{dd}\left(X,Y\right)}$ (flexibility in one residue correlating with flexibility in another residue). We then performed a target-site analysis, which aggregates all the MI with respect to an inputted target site. The calculated target MI was normalized by the highest MI calculated for a single residue–residue pair and then input into PDB files as the temperature factor for visualization.

The P74-26 ATPase domain was expressed and purified as previously described (11). Concentrated protein was diluted to 5 mg/mL in 25 mM Tris (pH 8.5), 125 mM sodium chloride, 10% (vol/vol) glycerol, 10 mM ADP, 10 mM magnesium chloride, and 10 mM DTT. ATPase domain crystals were grown using the hanging-drop vapor diffusion method, with drops consisting of equal volumes of the protein solution and crystallization well solution (100 mM sodium acetate:HCl pH 5.0, 0.5 M ammonium sulfate). Crystals were harvested and flash frozen after a 30 s soak in cryoprotectant containing ADP (50 mM sodium acetate:HCl pH 5.0, 0.8 M ammonium sulfate, 10 mM ADP, 10 mM magnesium chloride, and 30% ethylene glycol).

Crystallographic data were collected at the Advanced Light Source beamline 5.0.1. The data were processed with HKL3000 (38). Molecular replacement was accomplished with the SeMet model, as previously described (11). Molecular replacement and refinement were performed with Phenix (39) and model building with Coot (40). In addition to standard positional and B-factor refinement for the protein, we also directly refined the occupancy of the ADP molecule.

DNA-packaging and ATPase assays were run simultaneously using a randomized, predetermined order of reactions of two independently made, replicate batches for each protein. φ29 proheads, DNA-gp3, and gp16 were prepared as previously described (41). The in vitro DNA packaging assay is based on a DNase protection assay (42). The ATPase assay uses Malachite green to measure the production of inorganic phosphate (43).

Purified gp16 was diluted to 100 ng/mL (2.5 µM) with SUMO protease cleavage buffer containing 50 mM Tris ⋅ HCl (pH 8), 400 mM NaCl, 10% vol/vol glycerol, and 3 mM TCEP. sus8.5(900)-sus14(214)-sus16(300) proheads were diluted to 1 mg/mL (83 nM) buffer containing 50 mM Tris (pH 7.8) and 10 mM MgCl₂. φ29 DNA-gp3 was diluted to 250 ng/mL (21 nM) with sterile dH₂O.

Proheads and DNA-gp3 were combined in reaction buffer (100 mM Tris (pH 7.8), 20 mM MgCl₂, and 20 mM NaCl) to final concentrations of 10 nM and 62.5 nM, respectively. Reaction mixtures were prepared by adding 7 µL gp16 solution with 56 µL prohead/DNA-gp3 mixture, then incubated in ice for 15 min. Zero-time DNA packaging samples were made by combining 18 µL reaction mixture with 4 µL solution containing 5 ng/mL DNase I, 50 mM Tris (pH 8), 0.05 mM CaCl₂, 10 mM MgCl₂, and 50% glycerol, incubating in a room-temperature water bath for 15 min, and then chilled in ice.

DNA-packaging and ATPase assays were conducted simultaneously from the same samples. The reactions were started by combining 40.5 µL of the above reaction mixture with 4.5 µL 5 mM ATP. A zero-time ATPase sample was prepared by immediately removing 10 µL from the reaction, which was combined with 160 µL Malachite green reagent. After 1 min, the Malachite green reaction was quenched with 20 µL 34% wt/vol Na-citrate and allowed to remain at room temperature. The remaining 35 µL of the reaction was placed in a room-temperature water bath. At 15 min, a second ATPase sample was prepared by removing 10 µL from the reaction, which was combined with 160 µL Malachite green reagent and processed as above. Simultaneously, 15 min DNA packaging samples were prepared by combining 20 µL of the reaction with 2 µL solution containing 10 ng/mL DNase I, 50 mM Tris (pH 8), 0.05 mM CaCl₂, 10 mM MgCl₂, and 50% glycerol, incubated in a room-temperature water bath for 15 min, and then chilled in ice.

Zero-time and 15 min time DNA-packaging samples were further processed by combining each sample with 2 µL solution containing 0.25 M EDTA and 5 mg/mL proteinase K and incubating at 65 °C for 30 min. The packaged, undigested DNA was resolved in an agarose gel, and the densities of the packaged DNA and corresponding backgrounds were analyzed with Carestream Molecular Imaging Software 5.3.4.17821 (BioRad). The 100 µL aliquots of zero-time and 15 min time ATPase samples were transferred to 64-well plates, and absorption was measured at 620 nm with a BioTek plate reader. Zero-time measurements were subtracted from their respective 15 min time measurements, and the ATPase measurements were plotted on the DNA-packaging measurements.