Microsecond Dynamics of the Pyrophosphate Ion Release in SARS-CoV‑2 RNA Polymerase

Abstract

RNA-dependent RNA polymerase (RdRp) is the central molecular machine for viral replication and transcription. It catalyzes the nucleotide addition reaction, in which the release of the pyrophosphate ion (PPi) is a critical step in the resumption of the catalysis cycle. However, the precise mechanism by which PPi is released in the viral RdRp remains elusive, thereby impeding a comprehensive comprehension of the transcription and replication mechanisms of RdRp. In this study, we elucidated the molecular mechanism of PPi release in SARS-CoV-2 through the construction of a tailor-made Markov state model informed by extensive molecular dynamics simulations. We identified three metastable states, which demonstrate the stepwise release of PPi over a timescale of 3.6 μs. Additionally, we discovered that four highly conserved and positively charged residues, K551, R553, R555, and K798, can collectively mediate the release of PPi, suggesting their typical roles in the transcription and replication of coronaviruses. A two-step NTP incorporation assay and a PPi-mediated cleavage assay (pyrophosphorolysis) were developed and performed, validating their impact on PPi release by assessing both forward nucleotide incorporation efficiency and reverse RNA cleavage efficiency under exogenous PPi treatment. Further comparisons with the mechanisms utilized by other polymerases indicate that our study reveals a distinctive model of PPi release. In summary, our investigation contributes to advancing our understanding of the transcription and replication mechanism of viral RdRps by offering a glimpse into the dynamics of PPi release in SARS-CoV-2 RdRp.

Introduction

RNA-dependent RNA polymerase (RdRp) is essential for the life cycle of RNA viruses. By catalyzing viral transcription and RNA genome replication, it functions as a critical enzyme for viral assembly and propagation. − The nucleotide addition cycle (NAC) is a turnover that the RdRp employs to catalyze the incorporation of substrate nucleotide triphosphate (NTP) into the primer strand. A turnover of this nature is composed of several functional states and stages of conformational change. In particular, the active site would transition from an open to a closed conformation following the loading of NTP into the RdRp. The NTP forms a phosphodiester bond with the 3′-terminal nucleotide of the primer strand, which is followed by the generation of a byproduct pyrophosphate ion (PPi). After the active site resets to the open conformation, the PPi is released from the RdRp, and the double-stranded RNA (dsRNA) translocates to empty the active site in preparation for the entry of the next NTP. Consequently, the release of PPi is a critical step in the process of reactivating the NAC for the purposes of viral replication and transcription. The elucidation of its dynamics is the cornerstone of a thorough comprehension of the critical functions of the central molecular machine RdRp.

The structures of viral RdRps with PPi at the active site have been resolved in previous studies, providing a structural foundation for the comprehension of the PPi release. For instance, Yin et al. reported the cryo-EM structure of the SARS-CoV-2 RdRp with PPi at the active site and observed that PPi forms electrostatic interactions with R553 and R555 in motif F. The crystal structures of poliovirus and enterovirus 71 polymerases in the postcatalysis state consistently demonstrated that PPi at the active site interacts with several positively charged residues in motif F. ,, In addition, the cryo-EM structures of influenza A virus polymerase, influenza B virus polymerase, and La Crosse virus polymerase reveal interactions between PPi at the active site and the motifs A, D, and F. − Their roles in the initial stage of the PPi release are implicated by the molecular interactions observed in the RdRp structures.

In addition to the experimental structures, other endeavors have been undertaken to illuminate the dynamics of PPi release in the viral RdRps. Villalba et al. conducted a presteady-state kinetic study and discovered that the release of PPi in hepatitis C virus polymerase NS5B would be slowed down when a specific mismatched substrate is incorporated. This decreased rate of PPi release would enhance fidelity by creating a temporal window for the reverse reaction and decreasing the rates of misincorporation, although the mechanism was not addressed. Furthermore, Bignon et al. characterized the free energy profile of nucleotide incorporation by two-dimensional enhanced sampling quantum mechanics/molecular mechanics (QM/MM) calculations and additionally conducted a 100 ps unbiased QM/MM-MD simulation of the SARS-CoV-2 RdRp product structure after NTP incorporation. They observed interactions between PPi and several positively charged residues (K551, R553, R555, and K798), suggesting that these residues may facilitate the movement of PPi out of the channel. However, this study primarily focused on the chemistry of nucleotide incorporation and did not provide quantitative energetic or kinetic information for the PPi release process. Additionally, Ciardullo et al. also centered their investigation on the catalytic reaction, using QM calculations to investigate CTP and ddhCTP incorporation. They found that ddhCTP incorporation minimally altered the transition-state energetics relative to CTP but substantially stabilized the product state due to a tighter and reoriented PPi, consistent with reduced PPi flexibility observed in their MD simulations. To probe PPi release, they subsequently performed umbrella sampling along a predefined reaction coordinate (distance between PPi and D760), generating a potential of mean force from twenty 1 ns simulations. From this analysis, they proposed that several positively charged residues (K621, K798, K551, K43, and K2) can form a “bucket-brigade” network guiding PPi movement. While Ciardullo et al. provided valuable insights into substrate-dependent product state and PPi positioning, their PPi-release analysis necessarily relied on a chosen reaction coordinate and therefore may restrict accessible pathways. In addition, the resulting free energy profile does not directly yield kinetic information such as transition timescales. Taken together, these studies have provided important clues about the PPi release in viral RdRp. Nonetheless, the overall mechanism of PPi release in the viral RdRp remains incompletely understood. For instance, how many states are involved in the process of PPi diffusing out of the RdRp? What is the function of RdRp in the PPi release? The PPi release adheres to which model in SARS-CoV-2 RdRp as compared to the previous models of other RNA polymerases? Consequently, in order to gain a comprehensive understanding of PPi release in the viral RdRp, it is imperative to simulate the full PPi release process without imposing restrictive reaction coordinates and acquire thermodynamics and kinetics information.

The dynamic information is challenging to obtain by directly simulating the full process of PPi release, as one turnover of the nucleotide addition cycle in viral RNA polymerases usually occurs at over millisecond timescales and thus one single step of conformational change in the cycle can occur at microsecond to millisecond timescale. − The Markov state model (MSM) has become a widely used and cutting-edge tool for bridging the timescale gap between conventional molecular dynamics (MD) simulations and experimental observations, thereby resolving this issue. ,− Based on extensive and relatively short MD simulations, MSM enables the forecasting of long-timescale dynamics via the coarse-graining of time and space. ,,,,,,− MD simulations and MSM have been employed to decode the essential conformational changes of SARS-CoV-2 RNA polymerases and the corresponding acting mechanisms of antiviral drugs. ,− In the present investigation, we developed an MSM based on extensive MD simulations to comprehensively describe the dynamics of PPi release in SARS-CoV-2 RdRp. The metastable states of PPi release were identified, and the overall timescales were estimated. Additionally, the experimental assays further substantiated the identification of the critical protein residues that regulate the release of PPi. Our research has proposed a unique model that could be applied to other coronaviruses, distinct from the PPi release in other RNA polymerases identified previously. In total, our research has elucidated the molecular mechanisms of PPi release in SARS-CoV-2 RdRp, thereby improving our comprehension of the transcription and replication mechanism of viral RdRp.

Results and Discussion

MSM Reveals Three States along the PPi Release Pathway

To investigate the PPi release in SARS-CoV-2 RdRp, a structural model was constructed based on the cryo-EM structure with PPi bound at the active site (PDB ID: 7BV2, see Methods and SI Section 1.1 for details). After generating the preliminary PPi release pathways by caver analysis and steered MD simulations, two rounds of unbiased MD simulations were conducted to ensure the conformations fully relaxed from the preliminary pathways (Figures S1A and S19, see Methods and SI Sections 1.2 and 1.3 for details), laying the foundation for the subsequent MSM construction. To capture the relative movements between RdRp and dsRNA, we featurized the MD conformations using the atom pairs constituted by nucleotides and surrounding protein residues. The Spectral oASIS method , protocol was employed to choose the minimum set of atom pairs that can reproduce the dynamics. The slowest dynamics were determined by the time-lagged independent component analysis (tICA) − to consider the correlation and relaxation of specific motions. The hyperparameters for tICA were optimized by the generalized matrix Rayleigh quotient (GMRQ) − to ensure the robustness of our model. After validating the Markovian property of our model, we obtained three macrostates for elucidating the dynamics of PPi release in SARS-CoV-2 RdRp. Details of MSM construction and validation can be found in Methods and SI Section 2 (Figures S20–S24).

As indicated by our MSM, PPi is released from the active site of the SARS-CoV-2 RdRp with an overall timescale of approximately 3.6 μs (Figures and S2). Three states of PPi were identified along the channel, with the PPi gradually departing from the active site (Figures A and S1B). In particular, the S1 state is the state in which PPi is situated at the active site (Figure A), which is comparable to the experimental structure (PDB ID: 7BV2) (Figure C). The S1 state has the lowest population of approximately 9.2% and would transit to the S2 state with a mean first passage time (MFPT) of 1.08 μs (Figure B). The S2 state is distinguished from the S1 state by the position of PPi inside the channel (Figure A). The RMSDs of PPi conformations in the S2 state, as compared to those in the 7BV2 structure, are consistently greater than those in the S1 state (Figure C). The PPi conformations are compactly surrounded by the channel wall, and this state is also dominantly populated (∼57.5%) (Figure B). A further transition from the S2 to the S3 state occurs with an MFPT of approximately 2.52 μs, and the PPi arrives at the exit of the channel with a reduced population of approximately 33.3% (Figure B). To further corroborate the MSM-derived state transitions, we performed three 1.5 ∼ 2 μs independent simulations. We observed not only the S1→S2 and S2→S3 transitions but also conformational changes that align closely with the MSM-identified pathways (Figure S3), supporting the mechanisms inferred from the MSM.

1.

MSM elucidates three states of PPi release in SARS-CoV-2 RdRp complex. (A) Representative PPi conformations in the three states along the channel. The RdRp is shown in surface representation. The primer and template strands are shown in red and cyan sticks, respectively. (B) Diagram showing the equilibrium populations of three states and their mean first passage times (MFPTs). (C) RMSD values of PPi in the three states, computed using the MD conformations in each state against the experimental structure (PDB ID: 7BV2 ). The mean values and standard deviations were calculated using the bootstrapping algorithm by generating 85 samples. In each sample, 85 trajectories were randomly selected from a total of 85 MD trajectories. (D) Histogram of the distance between MgB and the oxygen atoms of the PPi in each state. The minimum distance in each MD conformation is computed and statistics are made using all the MD conformations in each state.

The three-state model accurately depicts the gradual detachment of PPi from the active site, as evidenced by the increasing RMSDs from the S1 to the S3 state (Figure C). The minimum distances between the oxygen atoms of the PPi and the MgB in the S1, S2, and S3 states of this process are approximately 1.9 Å (Figure D). This indicates that the MgB ion diffuses out of the RdRp in conjunction with the PPi, despite that PPi was parametrized in the absence of MgB. We note that the catalysis in viral RdRps may be facilitated by the other magnesium ion, MgA, in addition to MgB. − Therefore, we also investigated if the presence of MgA would influence the PPi release. In our model, MgA was positioned in the prebound position to simulate the scenario of an open conformation of the active site (Figure S4A). Specifically, MgA is electrostatically interacting with E811 and is coordinated with residues D618 and D761. Such an interaction network is well maintained in the three states of PPi release (Figure S4D–F). In contrast to MgB, the distance between MgA and PPi is consistently greater than 6.8 Å and increases rapidly in the S3 state (Figure S4G), indicating that it would not diffuse in conjunction with the PPi. We note that MgA can also sit at the active site adjacent to the primer strand when PPi is bound (Figure S4B). Accordingly, we also performed MD simulations to examine if MgA in such a location would influence the stability of PPi at the active site (see SI Section 3 for details). We found that the dynamics of PPi release are not altered as compared to the scenario when MgA is at the prebound site (Figure S4A–C), suggesting that the position of MgA in either the prebound site or active site is not critical for the PPi release. Together, our MSM has jointly determined that PPi would continue to coordinate with MgB and diffuse out of the RdRp using a three-state model. Further comparisons across two-, three-, and four-state lumping schemes indicate that the three-state model provides the most appropriate level of coarse-graining, as it captures the dominant PPi release pathway while avoiding unnecessary overpartitioning (Figure S5, see SI Section 4 for details). These findings address the first question raised in the Introduction by showing that PPi release proceeds through three major metastable states that describe the progression from the active site to the exit of the polymerase.

It is also of interest to examine whether active-site opening occurs concurrently with PPi release. Although a universally accepted structural definition of open and closed active-site states in SARS-CoV-2 RdRp is currently lacking, previous studies have suggested that MgB coordination and the hydrogen-bonding network surrounding the hydroxyl group of the substrate or the 3′-terminal nucleotide of the primer strand can serve as geometric indicators of the active-site state , (Figure S6, see SI Section 5 for details). Using these criteria, we analyzed the corresponding distances across the three states identified in this study and compared them with those observed in cryo-EM structures representing open and closed conformations (Figure S7, see SI Section 5 for details). The results suggest that the active site is already in an open conformation prior to PPi release and remains open as PPi release proceeds. Nevertheless, the reliability of this statement requires further validation due to the limited number of available structures proposed to represent open active-site states and by the use of substrate analogs rather than natural substrates, which may perturb local interactions.

As PPi release and the translocation of dsRNA have been suggested as two steps to restart the NAC in viral RdRps, it remains elusive if these two steps are coupled or not. To examine whether dsRNA translocation occurs concurrently with PPi release, we performed principal component analysis (PCA) specifically designed to capture the translocation motion (see SI Section 6 for details). In particular, to construct a principal component that cleanly represents the translocation, we generated structural models of RdRp in both states based on the 7BV2 structure. For each state, we performed a 20 ns MD simulation, in which heavy atoms were position restrained to preserve the characteristic geometry of each state. Snapshots from these two restrained simulations were then combined to perform PCA using the three central base pairs (−2 to −4 positions), which remain structurally stable and thus reliably represent translocation. The resulting PCA shows that PC1 alone accounts for over 99% of the variance and corresponds almost entirely to the translocation motion (Figure S8). We then projected the conformations from 85 × 150 ns unbiased simulations used for MSM construction investigating PPi release onto PC1. We found that the MD conformations are concentrated near the pre-translocation state and distant from the post-translocation state (Figure S9), indicating that the dsRNA retains its pre-translocation conformation throughout the PPi release process. This finding is consistent with our current estimate that PPi release occurs on a substantially shorter timescale (∼3.6 μs) than translocation (∼62.3 μs), implying that the two processes are unlikely to be tightly coupled in SARS-CoV-2 RdRp. This result also aligns with the previous study, which suggests that the translocation of RNA strands in the SARS-CoV-2 RdRp adopts a “Brownian ratchet” model that does not require PPi release to provide extra energy. While our analysis provides insight into the challenging question of whether PPi release and translocation are coupled, further investigations will be valuable to fully establish their mechanistic relationship.

Positively Charged Residues Mediate the PPi Release in SARS-CoV-2 RdRp

Subsequently, we examined the function of the RdRp in the process. Specifically, we conducted a preliminary screening of the protein residues with their heavy atoms within 10 Å of the PPi in the channel (Figure A). We calculated the root-mean-square fluctuations (RMSFs) of the heavy atoms in these residues, as we anticipate that the residues involved in the PPi release will move in tandem with PPi, resulting in a relatively high degree of flexibility. By plotting the distance between the protein residues and PPi against the residues’ RMSFs (Figure B), we discovered that residues K551, R553, R555, K621, and K798 are relatively closer to PPi and exhibit higher flexibility than the other residues. This suggests that they have the potential to participate in the PPi release in SARS-CoV-2 RdRp. This is consistent with a recent study that suggests the PPi release would pass through K551, K621, and K798 using umbrella sampling.

2.

Identifying the protein residues participating in the PPi release in SARS-CoV-2 RdRp. (A) The residues surrounding the PPi release channel. The C_α atoms of the residues with heavy atoms within 10 Å of PPi are shown in white spheres, except that those of K551, R553, R555, K621, and K798 are displayed with individual colors. The motif D and motif F are shown in cartoon representation and colored green and pink, respectively. The primer and template strands are shown in red and cyan sticks, respectively. (B) The minimum distances between the heavy atoms of residues and the PPi group were plotted against the RMSF of the heavy atoms of the residues. The mean values and standard deviations were computed using the bootstrapping method. (C) Correlation coefficient between the side chains of residues and the PPi. (D)–(H) The minimum distance between the nitrogen atoms of the side chains in the protein residues and the oxygen atoms of the PPi. In (B) and (D)–(H), the mean values and standard deviations were calculated using the bootstrapping algorithm.

To further evaluate whether their motions are concomitant with the PPi release, we calculated the Pearson correlation coefficient between each residue and PPi (Figure C, see SI Section 7 for details). Our findings indicate that the movements of K551, R553, and K798 are positively correlated with that of PPi, with correlation coefficients exceeding 0.3. Nevertheless, the correlation coefficients of the R555 and K621 motions are negative and within −0.15, suggesting that these residues unlikely undergo concerted motion with PPi. These results were further solidified after investigating the interactions between the residues and the PPi in each state (Figure D–H). The average distances between PPi and residues K551, R553, and K798 in the three states are within or around 5 Å (Figure D–F). This indicates that the side chains of these residues could adjust their orientations to maintain the interactions with PPi and assist in the movement of PPi along the channel. These results are in agreement with a previous study that hypothesized that residues K551, R553, and K798 are involved in the release of PPi by conducting a 100 ps unbiased QM/MM-MD simulation. Intriguingly, these three residues were also discovered to stabilize the phosphate group of NTP at the active site of SARS-CoV-2 RdRp in several previous studies. ,,,− This suggests that their interactions with the phosphate groups are established prior to the catalysis and are sustained until PPi is released from the RdRp. Notably, K551 and K798 were also identified as key residues in NTP loading in a previous study, suggesting that the PPi release channel spatially overlaps with substrate loading pathways, particularly within the internal region of the polymerase. Conversely, R555 exhibits electrostatic interactions with PPi exclusively in the S1 state, with a distance of approximately 3.1 Å, which weakens or disrupts the interactions in the S2 and S3 states (Figure G). K621 interacts with the PPi on a rare occasion, as its average distance from the PPi is at least 5.8 Å in all states (Figure H). Additional calculations of the interaction energies are consistent with the distance analysis (Figure S10). Together, the correlation coefficients and interaction analysis suggest that K551, R553, and K798 mediate the full PPi release across multiple states, whereas R555 mainly contributes to stabilizing PPi in the initial S1 state. In contrast, K621 plays a minor role due to its weaker interaction and larger distance from PPi. To further assess the robustness of the three-state model, we examined the distances between PPi and the five positively charges residues across the two-, three-, and four-state lumping schemes (Figure S11, see SI Section 4 for details). The three-state model preserves the essential structural distinctions among states without introducing complexity.

In order to elucidate the molecular mechanisms by which the residues K551, R553, and K798 facilitate the gradual release of PPi in SARS-CoV-2 RdRp, we investigated the conformations of these three residues in each state. It is intriguing to observe that the side chain of K551 points to the active site in the S1 state, gradually deviating from it in the S2 state, and ultimately pointing to the channel exit in the S3 state (Figure A). This conformational change of K551 is well described by the stepwise increasing distance between the nitrogen atom in the side chain of K551 and the phosphate group of the 3′-terminal nucleotide in the primer strand (y-axis in Figure B). Also, the swinging motion of K551 occurs concurrently with the diffusion of PPi (Figure B), which implies that K551 aids in pulling PPi out of the active site. We discovered that R553 and K798 could also facilitate the PPi release by the rotation motion of their side chains in the three states (Figure C,E), with the corresponding dihedral angles gradually increasing from the S1 to the S3 state (Figures D,F and S12). The dihedral angles vary in tandem with the release of PPi (Figure D,F), suggesting that the side chains’ rotation of R553 and K798 aids in delivering PPi to the channel exit.

3.

Molecular mechanisms of K551, R553, and K798 facilitating the PPi release. (A) The typical conformations of K551 interacting with PPi in the three states. (B) The x-axis represents the distance between the MgB and oxygen atom in the phosphate group of the 3′-terminal nucleotide of the primer strand (named “3′-OP1”), indicating the location of PPi and MgB along the channel. The y-axis shows the distance between the nitrogen atom “NZ” of K551 and 3′-OP1, describing the swinging motion of K551. (C) The representative conformations of R553 interacting with PPi in the three states. (D) The x-axis represents the distance between the MgB and 3′-OP1. The y-axis shows the dihedral angle defined by 3′-OP1, 3′-OP2, R553:C_α, and R553:CZ atoms (Figure S12A), describing the side chain’s rotation of R553. (E) The representative conformations of K798 interacting with PPi in the three states. (F) The x-axis represents the distance between the MgB and 3′-OP1. The y-axis shows the dihedral angle defined by 3′-OP1, 3′-OP2, K798:C_α, and K798:NZ atoms (Figure S12B), describing the side chain’s rotation of K798. In (B), (D), and (F), the mean values and standard deviations were calculated using the bootstrapping algorithm.

Overall, these results provide insight into the second question by clarifying the role of the polymerase: SARS-CoV-2 RdRp facilitates PPi release primarily through positively charged residues with distinct functional contributions. Specifically, the coordinated side-chain motions of K551, R553, and K798 act together to guide PPi along the exit channel, whereas R555 mainly contributes by stabilizing PPi at the early stage of the release process.

Experimental Assays Validate the Roles of Positively Charged Residues in the PPi Release

To further validate our in silico findings, empirical in vitro experiments were conducted. Nsp12, Nsp7, and Nsp8 were expressed in the baculovirus-insect cell system, and single nucleotide incorporation assays were developed and optimized (Figures S13 and S14, see Methods and SI Sections 8 and 9 for details). NAC is a complex but highly ordered enzymatic process consisting of four major steps initiated by nucleotide binding, phosphodiester bond formation, PPi release, and ended with translocation. To validate our computational findings, we developed a two-step NTP incorporation assay to assess the impact of mutations on the NAC in SARS-CoV-2 RdRp (Figure A,B). This assay compares the incorporation efficiency of the first nucleotide (CTP, corresponding to the +1 band on the gel, Figure C,E) versus subsequent nucleotides (ATP, corresponding to the +2 band, Figure C,E). We assume that the efficiency of incorporating the first nucleotide is primarily influenced by NTP binding and catalysis, while the incorporation of subsequent nucleotides may also depend on additional steps, such as PPi release and translocation after the first nucleotide is added. By analyzing the ratio of band intensities at +2 position relative to the +1 position, we indirectly evaluate the potential effects of mutations on processes beyond initial NTP binding and catalysis, which may include PPi release. We acknowledge that this approach cannot fully isolate PPi release from other steps, such as translocation. However, as shown by the cryo-EM structure with PPi at the active site (PDB ID: 7BV2 ), the minimum distances between the side-chain nitrogen atoms of the five charged residues and the phosphate backbones or ribose hydroxyl groups of the dsRNA are all above 5.0 Å, implicating their minor roles in translocation. Thus, the observed differences in incorporation efficiency primarily reflect impacts on PPi release dynamics. This method provides a practical and scalable alternative to traditional assays, avoiding challenges such as safety concerns or low throughput associated with other techniques (e.g., radioactive or single-molecule fluorescence methods).

4.

Assessment of key residues in PPi release using a two-step NTP incorporation assay. (A) Schematic illustration of the rationale behind the two-step incorporation assay. (B) Sequences of the LS41-GU RNA template and LS2U RNA primer scaffolds used in the two-step NTP incorporation assay. (C) Representative gel showing CTP incorporation (+1) and CTP/ATP incorporation (+2) for K551, R553, and K798 alanine mutants compared to the wild-type. Data shown are representative of three independent biological replicates. (D) Quantification and normalized efficiency analysis of CTP (+1) and CTP/ATP (+2) incorporation for K551A, R553A, and K798A mutants relative to wild-type based on data in (C). (E) Representative gel showing CTP incorporation (+1) and CTP/ATP incorporation (+2) for R555 and K621 alanine mutants compared to the wild-type. Data shown are representative of three independent biological replicates. (F) Quantification and normalized efficiency analysis of CTP (+1) and CTP/ATP (+2) incorporation for alanine mutants of R555 and K621 compared to the wild-type based on data in (E).

Following this protocol, we mutated each one of K551, R553, R555, K621, and K798 individually to alanine. Our results demonstrated that mutations on K551, R553, and K798 would negatively impact the incorporation efficiency of the second over the first nucleotide compared to the wild-type (Figure C,D, Table S1), supporting our hypothesis that these three residues are important for PPi release. R555A also showed a moderate effect in the two-step incorporation assay (Figure E,F, Table S2), which can be attributed to its strong electrostatic interaction with PPi specifically in the S1 state (Figure G). Conversely, the alanine mutation on K621 did not influence PPi release, with the normalized efficiency comparable to the wild-type (Figure E,F, Table S2). These results are consistent with our predictions from the computational simulations, illustrating the essential roles of K551, R553, R555, and K798 in the dynamics of PPi release.

To confirm the capability of our assays to assess whether these residues affect the PPi release, we examined mutations at positions outside the predicted PPi channel. We selected P323L, a naturally occurring variant in the interface domain of several SARS-CoV-2 strains, and K780D, located in the palm domain but distant from the PPi exit pathway. Both mutations produced band intensity ratios at the +2 position relative to the +1 position that were statistically indistinguishable from those of the wild-type (Figure S15A,B, Table S3). This lack of effect reinforces that the reduced relative band intensity ratios observed in K551A, R553A, R555A, and K798A mutants are specific to their roles in PPi release, rather than resulting from nonspecific disruptions to RdRp structure or function.

We acknowledge that the +2 to +1 product ratio does not directly reflect intrinsic PPi release efficiency, as the manual quenching assay lacks the resolution to capture single-nucleotide incorporation rates in processive RdRp complexes and instead provides only a snapshot of product distribution. Moreover, since NTP entry and PPi release share the same pathway, mutations may affect both steps, complicating the isolation of PPi-specific effects. Therefore, our two-step incorporation assay can only indirectly assess the impact of mutations on steps beyond initial NTP binding and catalysis, assuming these steps remain constant between the first and subsequent substrate incorporation events. To complement the limitations of the NTP incorporation assay, we also employed a PPi-mediated cleavage assay (pyrophosphorolysis), the reverse reaction of NTP-driven polymerization. This assay directly probes PPi accessibility to the active site by measuring the ability of RdRp to utilize exogenous PPi to cleave the RNA 3′-terminus. Since this reverse reaction bypasses NTP loading and uses PPi as the sole substrate, it isolates the role of PPi transit more specifically. As PPi must traverse the same structural corridor in reverse to reach the active site for RNA cleavage, the impaired cleavage efficiency in mutant enzymes directly indicates obstruction of the shared PPi transit path. Consequently, we measured and quantified the RNA cleavage efficiency of wild-type and mutant enzymes at 1 mM PPi (Figure A,C). Notably, the normalized RNA cleavage efficiency for K551A, R553A, R555A, and K798A mutants was severely reduced relative to wild-type (Figure B,D). In contrast, the charge-conserving mutants (K551R, R553K, R555K, and K798R) for these residues exhibited cleavage efficiencies comparable to the wild-type (Figure B,D), underscoring that the positive charges at these positions are critical for maintaining an electropositive pathway for PPi movement. Notably, both the K621A mutant and charge-conserving mutant K621R showed efficiency comparable to the wild-type, again supporting that K621 plays a minimal role in the PPi release (Figure C,D).

5.

Assessment of key residues in PPi release using PPi-mediated RNA cleavage assay. (A) Representative gel showing the RNA cleavage products for K551A, R553A, K798A, K551R, R553K, and K798R mutants compared to the wild-type. Reactions were performed with the LS15A/LS2U RNA construct (Figure S14A) in the presence of 1 mM PPi. A representative result from three independent biological replicates is shown. (B) Quantification and analysis of the normalized PPi-mediated RNA cleavage efficiency for K551A, R553A, K798A, K551R, R553K, and K798R mutants compared to the wild-type. (C) Representative gel showing the RNA cleavage products analysis for R555A, K621A, R555K, and K621R mutants compared to the wild-type. A representative result from three independent biological replicates is shown. (D) Quantification and analysis of the normalized PPi-mediated RNA cleavage efficiency for R555A, K621A, R555K, and K621R mutants compared to the wild-type.

Collectively, these complementary approaches provide converging evidence for our proposed mechanism of PPi release in SARS-CoV-2 RdRp. The two-step NTP incorporation assay suggests that alanine mutations at K551, R553, R555, and K798 disrupt processes following initial nucleotide incorporation, likely involving PPi release, and the pyrophosphorolysis assay directly implicates that these residues play roles in PPi release. The consistent results from charge-conserving mutations highlight the importance of electrostatic interactions, and the lack of effect from mutations outside the pathway confirms the specificity of our findings. Together with our computational simulations, these experiments firmly establish K551, R553, R555 and K798 as key residues for PPi release.

The PPi Release in SARS-CoV-2 RdRp Follows a Distinctive Model

To date, distinct models have been proposed to describe the dynamics of PPi release in a variety of RNA polymerases. For instance, the PPi release in the Saccharomyces cerevisiae DNA-dependent RNA polymerase II (Pol II) is characterized by a “hopping model”. In this model, PPi sequentially hops between multiple sites to reach the pore region’s exit. Each site is composed of two or three positively charged residues that are responsible for delivering the PPi (Figure A). A comparable model, the “two-state model”, has been proposed for the release of PPi in bacterial DNA-dependent RNA polymerase (RNAP). In this case, the release channel is evidently shorter than that of Saccharomyces cerevisiae Pol II, and the positively charged residues are concentrated in two distinct regions along the channel (Figure B). The residues in one region would pull the PPi away from the active site, and it would then be transferred to the other region for exclusion from the RNAP. In contrast, the PPi release in the bacteriophage T7 single-subunit RNAP follows the “jump-from-cavity” model (Figure C), in which K472 plays a prominent role due to its pronounced swing motion that facilitates PPi departure. In addition, R627 maintains persistent interactions with PPi throughout the release process, whereas K631 interacts with PPi primarily at the initial stage when PPi remains near the active site.

6.

Comparison of PPi release in different RNA polymerases and sequence alignment of SARS-CoV-2 RdRp with other coronaviruses’ RdRps. (A–D) Illustrative cartoon models depicting the release of PPi in Pol II, bacterial RNAP, T7 RNAP, and SARS-CoV-2 RdRp. (E) The sequence alignment across the viral RdRps. The residues K551, R553, R555, and K798 in the SARS-CoV-2 RdRp and the corresponding residues in other coronaviruses’ RdRps are colored in wheat, green, blue, and purple, respectively. (F) Diagram showing the K551, R553, R555, and K798 in the SARS-CoV-2 RdRp (PDB ID: 7BV2 ). (G) Diagram showing the K551, R553, R555, and K798 in the SARS-CoV RdRp (PDB ID: 6NUR ). (H) Diagram showing the K546, R548, R550, and K793 in the PEDV RdRp (PDB ID: 8G6R ).

In this study we have shown that the PPi release in SARS-CoV-2 RdRp adheres to a “three-state model” (Figure D). In this model, the PPi release is collectively facilitated by positively charged residues. K551, R553, and K798 alter the orientations of their side chains to deliver the PPi out of the channel whereas R555 mainly interacts with PPi at the active site. K551, R553, and R555 are situated on motif F, while K798 is situated in a distinct location near motif D. Intriguingly, we found that the residues K551, R553, R555, and K798 show a high degree of conservation in the coronaviruses (Figure E, see SI Section 10 for details). Besides, the spatial arrangement of the four residues in the PPi release of SARS-CoV-2 RdRp is consistent across various RdRps (Figure F–H). This is supported by a comparison of their locations in the experimental structures of coronavirus RdRps, such as SARS-CoV-2, SARS-CoV, and porcine epidemic diarrhea virus (PEDV). The above observations together imply that the molecular mechanisms of the four residues in the PPi release of SARS-CoV-2 RdRp may be prevalent in coronaviruses. We also extend the conservation analysis to a broader range of viral families. In particular, we examined several positive-strand RNA viral RdRps with available experimental structures, including foot-and-mouth disease virus, enterovirus 71, norovirus, and poliovirus. Following established alignment from previous studies, − we aligned the structures and identified the residues corresponding to K551, R553, R555, and K798 (Figure S16). The sequence comparisons across different viral RdRps show that K798 and R555 in SARS-CoV-2 RdRp are highly conserved. Although R553 is replaced with lysine in other viral RdRps, the positive charge at this position is well maintained. In contrast, K551 is observed only in enterovirus 71 and norovirus, while it is substituted by neutral alanine and glutamine in foot-and-mouth disease virus and poliovirus RdRp, respectively. These results suggest that the positively charged residues at positions equivalent to R553, R555, and K798 likely serve conserved roles in PPi release across positive-strand RNA viruses, whereas the presence and function of K551 may be virus-specific.

After comparing our model of PPi release in SARS-CoV-2 RdRp with those in other polymerases, we found that our mechanism is distinct yet shares certain features with the “jump-from-cavity” model of the T7 RNAP in the sense that charged residues undergo side-chain swings or rotations to assist PPi release (e.g., K551, R553, and K798 in SARS-CoV-2 RdRp and K472 in T7 RNAP). By comparing the side-chain conformations of these charged residues, we observed that the swing motion of K551 in SARS-CoV-2 RdRp shifting from an inward-facing conformation toward channel exit most closely resembles the behavior of K472 in T7 RNAP, even though K798 at the outward region occupies a more analogous spatial position to K472 (Figure S17). In addition, K551, R553, and R555 in SARS-CoV-2 reside within motif F, a structural component equivalent to the O-helix of T7 RNAP, which contains K631 and R627 that participate in PPi interactions during the release process (Figure S17). Interestingly, R627 in T7 RNAP resembles K551 and R553 in SARS-CoV-2 RdRp as they all maintain the interactions with PPi throughout the entire release process, whereas K631 in T7 RNAP plays a similar role as R555 in SARS-CoV-2 RdRp by interacting with PPi primarily when PPi is near the active site (S1 state in T7 RNAP) and weakens its interaction as PPi approaches the exit.

In general, our research suggests that the PPi release in SARS-CoV-2 is facilitated by several positively charged residues: K551, R553, R555, and K798 (Figure D). This mechanism is expected to be conserved among coronavirus RdRps yet distinct from other RNA polymerases largely due to variations in polymerase structural architecture, thereby addressing the third question raised in the Introduction. Nonetheless, it is worthy to note that a unifying feature across polymerases is the involvement of positively charged residues, particularly lysines and arginines, in coordinating and guiding PPi movement.

Conclusion

A critical step in the NAC of viral RdRp is the release of PPi. However, the process’s timescale and the system’s size render it challenging to simulate the process directly. Additionally, the current experimental assays’ resolution presents a challenge in isolating a single step from the entire cycle. To overcome these challenges and elucidate the PPi release in SARS-CoV-2 RdRp, we have incorporated computational simulations and experimental assays to provide a comprehensive description of PPi release in SARS-CoV-2 RdRp, directly addressing the key questions raised in the Introduction. First, we show that PPi release in SARS-CoV-2 RdRp proceeds through three metastable states that collectively describe its progression from the active site to the exit of the channel, with an overall timescale of approximately 3.6 μs. Second, our results clarify the functional role of the polymerase in this process. PPi release is mediated primarily by positively charged residues with distinct yet cooperative contributions. In particular, the coordinated side-chain motions of K551, R553, and K798 guide the PPi through the exit channel, while R555 plays a complementary role by stabilizing PPi at the initial stage of release. These mechanistic insights derived from simulations are strongly supported by two complementary experimental assays: a two-step NTP incorporation assay, which indirectly assessed the impact of mutations on steps following initial nucleotide incorporation (potentially including PPi release), and a PPi-mediated cleavage assay (pyrophosphorolysis), which directly probed PPi accessibility to the active site via the same pathway used in release. Both assays consistently demonstrated that single-point alanine mutations at K551, R553, R555, and K798 impair PPi dynamics, while charge-conserving mutations or mutations outside the pathway have minimal effect, underscoring the specificity and electrostatic importance of these residues. Finally, comparison with established PPi-release models in other RNA polymerases reveals that SARS-CoV-2 RdRp follows a distinct mechanism. While the involvement of positively charged residues in guiding PPi appears to be a shared principle across polymerases, a distinctive feature of SARS-CoV-2 RdRp is that PPi is delivered through the channel by the concerted and simultaneous side-chain motions of three charged residues. Sequence alignment further shows that K551, R553, R555, and K798 are highly conserved among coronaviruses, suggesting that this PPi-release mechanism is likely shared within the coronavirus RdRp family.

It is noted that a previous kinetic study of HCV NS5B has shown that PPi release can be selectively slowed during incorporation of certain mismatched nucleotides, thereby enabling the reverse reaction and enhancing fidelity. To explore whether a similar mechanism may be feasible in SARS-CoV-2 RdRp, we compared the structural environments of the two polymerases by modeling PPi and the MgB ion into the HCV NS5B active site through alignment with SARS-CoV-2 RdRp (Figure S18). In both enzymes, MgB interacts with the nucleotide’s phosphate backbone and PPi can form a hydrogen bond with the primer O3′ group, suggesting that these interactions may “sense” nucleotide conformations and delay PPi release when a mismatch perturbs the active-site geometry. Notably, a recent computational study reported that incorporation of a nucleotide analog in SARS-CoV-2 RdRp leads to a product state with more tightly bound and reoriented PPi, providing indirect support for this proposed mechanism. However, because a high-resolution structure of HCV NS5B with PPi bound is not yet available, these conclusions rely on structural modeling and remain speculative. Direct structural comparison of matched and mismatched states, together with kinetic analyses in both polymerases, will be needed to more firmly establish whether PPi release can serve as a fidelity checkpoint in SARS-CoV-2 RdRp.

Overall, our research has collectively elucidated the mechanism of PPi release in SARS-CoV-2 RdRp at the molecular level and shed light on the PPi release in other viral RdRps. This information would also improve our understanding of the replication and transcription mechanisms of viral RdRps, thereby facilitating the rational design of antivirus inhibitors.

Methods

Modeling the SARS-CoV-2 RdRp Structure with PPi at the Active Site

The cryo-EM structure of the SARS-CoV-2 RdRp complex (PDB ID: 7BV2) was used as the basis to obtain a structural model with PPi bound at the open active site, corresponding to the state after catalysis and active site isomerization from the closed to the open configuration. It has been suggested that the MgA is at the prebound site when the active site adopts an open configuration. Accordingly, we modeled the MgA ion and D761 residue based on the cryo-EM structure with the PDB ID: 7CTT. Besides, we substituted the Remdesivir at the 3′-terminal of the primer strand with the adenosine monophosphate. The system was solvated in a 150.5 Å × 150.5 Å × 150.5 Å dodecahedron box by TIP3P water. After adding sufficient Na⁺ ions to neutralize the system, the whole system comprises approximately 240,000 atoms. Then, three 20 ns MD simulations were conducted in parallel to obtain the equilibrated conformation of PPi at the active site (see SI Section 1.1 for details).

Generating the Preliminary Pathway

First, caver program was utilized to explore the potential pathways by simulating the PPi as a sphere with a radius of 2.5 Å, and residues K160, D164, E167, L41, K43, and K438 around the exit of the top 5 pathways were identified to determine the PPi release directions (Figure S19). To obtain the PPi conformations along the release channel, we conducted steered MD simulations initiated from the last snapshot of the 3 × 20 ns MD simulations and pulled the PPi away from the active site and along the directions defined by the above residues. A bunch of PPi conformations along the pulling pathways were generated. These conformations were split into seven groups by the K-Center clustering algorithm , according to the RMSDs of the MgB and PPi. One random conformation was chosen from each group and used to conduct unbiased MD simulations (see SI Section 1.2 for details).

Unbiased MD Simulations

After 10,000 steps energy minimization and 200 ps position restraint NVT simulation, each conformation was used to seed two 50 ns trajectories under the NVT ensemble (T = 300 K) with different initial velocities. After removing the first 10 ns of each trajectory, 14 × 40 ns MD conformations were split into 50 groups using the K-Center clustering algorithm. , Then, we randomly selected one conformation from each group and a total of 50 conformations were used for the next round of MD simulations. Specifically, two parallel 150 ns simulations were initiated from each conformation with different random seeds under NVT ensemble (T = 300 K). After removing the first 20 ns from each trajectory and the trajectories where PPi was outside the RdRp, we used 85 × 130 ns MD simulations for constructing the MSM (see SI Section 1.3 for details).

Construction and Validation of MSM

MSM is a commonly used kinetic network model to predict the thermodynamics and kinetics of the conformational changes by coarse-graining both time and space, ,,,,,,− and thus serves as a useful tool for our current investigations. In particular, the Spectral oASIS method , was employed to determine the atom pairs to featurize the MD conformations. Then, tICA − was used to identify the slowest dynamics, and the hyper-parameters were selected by the GMRQ method. − Comparison between the MD conformational ensemble and the preliminary pathways generated by steered MD simulations has suggested that our simulations have removed the bias exerted by the pulling force and relaxed away from the preliminary pathways (Figure S1A). After reducing the degree of freedom to the top two tICs, the K-Center algorithm , was used to split the MD conformations into 250 microstates. The implied timescales were computed to determine a lag time of 40 ns to calculate the transition probability matrix of microstate-MSM. In order to elucidate the molecular mechanism of the PPi release in SARS-CoV-2 RdRp, the 250 microstates were further lumped into 3 macrostates by the modified Robust Perron Cluster Analysis (PCCA+) algorithm. , The MFPTs and populations of each macrostate were computed based on the 1 ms Monte Carlo trajectories synthesized using the transition probability matrix of the microstates (see SI Section 2 for details).

Two-Step NTP Incorporation Assay

The PPi release activity of the wild-type and mutant SARS-CoV-2 RdRp Nsp12/7/8 complex was evaluated by comparing the incorporation efficiency of the first and second nucleotides. Briefly, the reaction was set up in a 10 μL mix consisting of 4.5 μM RdRp complex, 0.1 μM fluorescence-labeled RNA primer (LS2U, 5′-FAM-GUCAUUCUCCUAAGAAGCUU-3′), and 0.1 μM RNA template (LS41-GU, 5′-CUAUCCCCAUGUGAUUUCCUGAAGCUUCUUA-GGAGAAUGAC-3′) in the reaction buffer containing 20 mM Tris (pH 7.5), 15 mM NaCl, and 5 mM MgCl₂. The first NTP incorporation step was initiated by adding CTP to a final concentration of 20 μM. After a 10 min incubation at 37 °C to ensure completion of the first incorporation (+1 product), the second NTP incorporation step was performed by adding ATP to a final concentration of 10 μM for 1 min at 37 °C. The shorter pulse for the second incorporation increases sensitivity for detecting differences in postcatalytic steps (PPi release and translocation) that precede the next round of NTP binding and catalysis. Immediately afterward, a 4 μL aliquot of each reaction was mixed with 4 μL of 2 × RNA loading buffer (95% (v/v) formamide, 20 mM EDTA (pH 8.0), 0.02% (w/v) bromophenol blue), and subject to a 20% Acrylamide, 7 M Urea PAGE Gel. Electrophoresis was conducted at 350 V for 60 min, and the FAM-labeled RNA products were visualized using a LI-COR Odyssey M Imager at 488 nm channel. To quantify the PPi release efficiency for first and second NTP incorporation, the ratio of the band intensities of +2 position over the band intensity of +1 position was calculated and normalized to the ratio of wild-type.

PPi-Mediated RNA Cleavage Assay

In a typical 10 μL reaction, 2 μM SARS-CoV-2 RdRp Nsp12/7/8 complex was preincubated with 0.1 μM RNA scaffold LS15A/LS2U (LS2U, 5′-FAM-GUCAUUCUCCUAAGAAGCUU-3′, LS15A, 5′-CUAUCCCCAUGUGAUUUUACAAGCUUCUUA-GGAGAAUGAC-3′) in reaction buffer (as described above) for 15 min at room temperature. Reactions were initiated by adding sodium pyrophosphate to a final concentration of 1 mM. After incubation at 37 °C for 10 min, an equal volume of 2× RNA loading buffer was added to quench the reactions. The cleavage reaction mixtures were resolved by 15% polyacrylamide gel electrophoresis at 300 V for 45 min, and products were visualized using a LI-COR Odyssey M Imager. For each mutant, the uncleaved RNA ratio was quantified as the band intensity of the remaining 20-nt RNA divided by the input RNA intensity. Cleavage efficiency was then calculated and normalized to the wild-type value.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/jacsau.5c00995.

• Supplementary figures, supplementary tables, details of MD simulations, construction and validation of MSM, MD simulations with the MgA in different sites, comparison of the three-state model with two-state and four-state models, investigation of the active-site state in the process of PPi release, investigation of the dsRNA movement, calculation of the Pearson correlation coefficient, constructs, expression, and purification of the Nsp12/7/8 complex with wild-type and mutant Nsp12, RdRp enzymatic activity assay, and sequence and structural alignments across viral RdRps (PDF)

⟁.

T.X., L.Y., X.X., and B.-Y.Z. contributed equally to this work.

The authors declare no competing financial interest.