Mitochondrial ribonuclease P structure provides insight into the evolution of catalytic strategies for precursor-tRNA 5′ processing

Abstract

Ribonuclease P (RNase P) catalyzes the maturation of the 5′ end of tRNA precursors. Typically these enzymes are ribonucleoproteins with a conserved RNA component responsible for catalysis. However, protein-only RNase P (PRORP) enzymes process precursor tRNAs in human mitochondria and in all tRNA-using compartments of Arabidopsis thaliana. PRORP enzymes are nuclear encoded and conserved among many eukaryotes, having evolved recently as yeast mitochondrial genomes encode an RNase P RNA. Here we report the crystal structure of PRORP1 from A. thaliana at 1.75 Å resolution, revealing a prototypical metallonuclease domain tethered to a pentatricopeptide repeat (PPR) domain by a structural zinc-binding domain. The metallonuclease domain is a unique high-resolution structure of a Nedd4-BP1, YacP Nucleases (NYN) domain that is a member of the PIN domain-like fold superfamily, including the FLAP nuclease family. The structural similarity between PRORP1 and the FLAP nuclease family suggests that they evolved from a common ancestor. Biochemical data reveal that conserved aspartate residues in PRORP1 are important for catalytic activity and metal binding and that the PPR domain also enhances activity, likely through an interaction with pre-tRNA. These results provide a foundation for understanding tRNA maturation in organelles. Furthermore, these studies allow for a molecular-level comparison of the catalytic strategies used by the only known naturally evolved protein and RNA-based catalysts that perform the same biological function, pre-tRNA maturation, thereby providing insight into the differences between the prebiotic RNA world and the present protein-dominated world.

According to the RNA world hypothesis, RNA played dual roles as carrier of genetic information and catalyst in a prebiotic world. However, over eons of evolution, proteins with their expanded 20-aa alphabet and greater structural and functional complexity took over many RNA-based functions. Structural insights into this evolutionary transition are limited because of the lack of examples of RNA and protein macromolecules that perform the same biological function in nature. One notable exception is ribonuclease P (RNase P) that catalyzes maturation of the 5′ end of tRNA across all domains of life. Until recently, all known RNase P enzymes included a catalytic RNA component. The discovery of a protein-only RNase P [proteinacous RNase P (PRORP)] from human mitochondria and Arabidopsis thaliana has dramatically shifted this paradigm (1–3). These enzymes represent a unique class of metallonucleases and are conserved among many eukaroytes (1, 2). A. thaliana encodes three PRORP enzymes (PRORP1, -2, and -3) that catalyze pre-tRNA processing (2). PRORP1 localizes to the mitochondria and chloroplast whereas PRORP2 and PRORP3 localize to the nucleus (2), suggesting that protein-based enzymes catalyze pre-tRNA maturation in these cellular locations (2, 3). To gain mechanistic and evolutionary insights into the PRORP enzyme family, we crystallized PRORP1 from A. thaliana.

This structure clearly demonstrates that there is no structural homology between PRORP1 and any of the proteins associated with either the bacterial or the nuclear human RNase P (4, 5), revealing that this protein evolved independently from the ribonucleoprotein RNase P. Furthermore, the combination of structural and biochemical data suggests that the RNA and protein-based catalysts use distinct strategies to bind and cleave pre-tRNA. In PRORP1, the pentatricopeptide repeat (PPR) domain enhances pre-tRNA binding affinity, in contrast to the combination of RNA–tRNA and protein–pre-tRNA leader contacts in the ribonucleoprotein RNase P (5). Whereas both PRORP1 and bacterial RNase P are magnesium-dependent enzymes, the metal ions are coordinated by carboxylate side chains and nonbridging phosphate oxygens/nucleotide carbonyl groups (5), respectively. These studies allow us to begin to compare, at a molecular level, the catalytic strategies and molecular recognition used by the protein-only and RNA-based RNase P enzymes.

Results

Global Architecture of PRORP1.

The crystal structure of a functional recombinant protein-only RNase P from A. thaliana (residues 76–572 with the mitochondrial signal sequence deleted) was solved by single-wavelength anomalous dispersion (SAD) and was refined to a resolution of 1.98 Å (R_work = 18.7%, R_free = 22.0% with good stereochemistry; Table S1). One molecule is present per asymmetric unit and the final refined model includes residues 95–570 (Fig. 1). PRORP1 adopts a conformation that resembles an upside-down “V”, where two arms, each ∼70 Å in length, fold at an angle of ∼56°. Arm 1 is composed of 11 α-helices, whereas arm 2 is composed of a parallel β-sheet flanked by α-helices (Fig. 1). The two arms are attached to a core domain consisting of an antiparallel β-sheet. These discrete structural elements represent three domains: the PPR domain (residues 95–292; red in Fig. 1C), the central domain (residues 328–357 and 534–570; yellow in Fig. 1C), and the metallonuclease domain (residues 358–533; blue in Fig. 1C).

Fig. 1.

Crystal structure of PRORP1. (A) Structure of PRORP1 (helices shown in blue, strands shown in yellow). (B) Topology map of secondary structure elements. (C) Cartoon depiction of the PRORP1 domain arrangements: PPR domain (residues 95–292; red), central domain (residues 328–357 and 534–570; yellow), and metallonuclease domain (residues 358–533; blue). (D) Electrostatic surface potential.

PPR Domain Is Important for Activity and Substrate Affinity.

PPR motifs are often found in tandem and are composed of a helix-turn-helix fold of ∼35 aa (6). These motifs have been implicated in RNA binding and editing activities and are prevalent among proteins targeted to mitochondria and chloroplasts (7–9). The PPR domain in PRORP1 is composed of 11 α-helices forming 5.5 consecutive PPR repeats, each consisting of a helix-turn-helix hairpin (Figs. 1 and 2A). These tandem helical repeats associate to form a right-handed superhelical structure that resembles the arrangement of domains in tetratricopeptide repeat (TPR) motifs. As a result, the PPR domain forms an inner concave surface facing the putative active site of the metallonuclease domain. A comparative search based on the PPR domain using the Dali server (10) reveals structural similarity with a plethora of proteins (or protein domains) that notably include mitochondrial RNA polymerase (mtRNAP) (11), Get4 (12), proteosomal subunit Rpn6 (13), and small glutamine-rich tetratricopeptide repeat protein (hSGT) (14) (Fig. 2B). Except for mtRNAP, which contains the first and only structural data on PPR motifs, all of the homologous domains contain TPR motifs that are involved in protein–protein interactions (6). The PPR domain of mtRNAP is proposed to interact with upstream DNA promoter sequences (11). Similarly, we posit that the PPR domain of PRORP1 facilitates interactions with pre-tRNA.

Fig. 2.

PPR domain architecture and structural Zn-binding site. (A) Tandem repeats of the PPR domain. Domain and profile analysis of PRORP1 using PROSITE predicts three PPR motifs (PPR1, residues 96–130; PPR3, residues 175–209; and PPR4, residues 210–244) on the basis of sequence consensus. The structure reveals two additional helix-turn-helix hairpins that adopt the PPR motif (PPR2, residues 131–174; and PPR5, residues 245–278) despite their divergence from the canonical PPR sequence. The PPR2 hairpin contains an extended loop, ∼11 aa vs. the typically observed 3-aa loop, between α-helices 3 and 4. (B) Structural superposition of PPR/TPR domains: PRORP1 (red; residues 95–293), mtRNAP (yellow; residues 258–331), Get4 (blue; residues 58–236), and the proteasomal subunit Rpn6 (green; residues 38–222). (C) Close-up of the Zn-binding site of PRORP1.

An N-terminal truncation of PRORP1 (Δ245) that removes 8 of the 11 helices in the PPR domain decreases pre-tRNA binding affinity by 34-fold and catalytic activity by ≥2,000-fold (Fig. 3C) with little effect on zinc binding (Table S2). Because the PPR domain is located far away from the metal-binding active site, we propose that this domain is important for properly orientating the leader of pre-tRNA in the active site rather than playing a direct role in catalysis. The electrostatic surface potential for PRORP1 (calculated using Apbs) (15) (Fig. 1D) reveals that the concave surface facing the putative active site has an overall neutral charge, suggesting that the PPR–nucleic acid interaction is not mainly electrostatic. The only positive patch is found in the base of the PPR domain closest to the active site, including side chains from the interior-facing helices and connecting loops. The PPR domain could interact directly with pre-tRNA through contacts between this positive patch and the negative phosphodiester backbone, recognition of the tertiary fold of tRNA, and/or interactions with conserved nucleobases of tRNA, as proposed for other PPR domains (16).

Fig. 3.

Active site, metal dependence, and pre-tRNA binding of PRORP1. (A) Close-up of the Mn(II)-bound PRORP1 active site indicating conserved residues involved in metal binding. (B) Close-up of the Ca(II)-bound PRORP1 active site. (C) Structural alignment of the active sites of the PRORP1 metallonuclease domain (blue) and hExo1 (green bound to DNA) (20). Active-site Mn(II) metals are shown in purple spheres for PRORP1 and in green spheres for hExo1. Black arrows indicate the location of the two aspartates in hExo1 that coordinate the second active-site metal [Mn(II)₂]; there are no equivalent residues in PRORP1 that bind Mn(II)₂. (D) Electron density maps of the active site in the Mn(II)-bound PRORP1. The anomalous difference 2Fo-Fc electron density map, contoured at 3σ, is shown in yellow and is superimposed on the PRORP1 structure. This map was calculated from experimental phases derived from data collected at the Mn edge. The composite omit 2Fo-Fc difference density map for PRORP1 with Mn(II) contoured at 1.5σ is shown in blue. The omit Fo-Fc difference map for the PRORP1 model refined without the active-site Mn atoms and inner sphere water molecules is contoured at 10σ and is shown in green. (E) Representative gels of metal-dependent single-turnover cleavage assays (Upper Left and Lower Left). Reactions containing A. thaliana mitochondrial 5′-³²P-pre-tRNA^Cys, 500 nM Δ76PRORP1, 2.5 mM MgCl₂ or CaCl₂, and 250 μM cobalt(III)hexammine were quenched at specified time points, resolved by denaturing PAGE, and analyzed with a phosphorimager. Mg(II) and Mn(II) activate catalysis of phosphodiester bond hydrolysis whereas Zn(II) and Ca(II) do not (Fig. S2). Fluorescent polarization binding data in 1 mM CaCl₂ (Upper Right) indicate that the binding affinity for fluorescein-labeled mitochondria pre-tRNA^Cys is decreased 34-fold by deletion of four PPR motifs (Δ245PRORP1, blue squares; Δ76PRORP1, black circles). (Lower Right) Cleavage rate constants and K_D values for mutant PRORP1 proteins are summarized. The errors reported for the K_D and k_obs values represent the SD from two and four, respectively, independent experiments (Fig. S3).

Central Domain.

The central domain contains an antiparallel 4-stranded β-sheet that interacts with the PPR domain and two extended loops that connect it to the metallonuclease domain. The central domain houses a conserved structural zinc-binding site (Fig. 2C and Fig. S1). Zinc is coordinated by four invariant residues: C344, C350, H548, and C565. C344 and C350 are found on the 15-aa-long two-turn loop (between β1 and α14) connecting the metallonuclease and the central domains. H548 and C565 are located in β-strands 9 and 10, respectively (Fig. 2C). The zinc structural site appears to be important for both stabilizing the structure of the loop-rich central domain and properly orienting this domain with respect to the metallonuclease domain. The central and metallonuclease domain interface (473 Ų) is mainly stabilized by electrostatic interactions.

The β-sheet of the central domain forms a large hydrophobic interface with helices α9, α10, and α11 of the PPR domain. In addition, 10 hydrogen bonds and two salt bridges (between R335 and D285 and D288) mold the PPR and central domain interface. Furthermore, the PPR domain is linked to the central domain through a 32-residue insertion (shown in pink in Fig. 1C and Fig. S1) found only in plant species. This positively charged insertion contains a loop-helix-loop that interacts with the PPR/central domain interface, forming primarily hydrophobic interactions with the PPR domain (473 Ų) and mainly electrostatic interactions with the central domain (621 Ų), including 10 hydrogen bonds and four salt bridges. The central domain may function as an adaptor, orienting the metallonuclease and PPR domains to optimize catalytic activity and molecular recognition.

Finally, the central domain contains two major positively charged patches (Fig. 1D) that could interact with the tRNA backbone. One of these patches is in close proximity to the zinc site, suggesting that the bound zinc site stabilizes the structure of the central domain such that positively charged residues are oriented to interact with pre-tRNA, similar to the role of structural zinc sites in zinc fingers (17).

Metallonuclease Domain.

The structure of the metallonuclease domain of PRORP1 is a unique high-resolution structure of a Nedd4-BP1, YacP nuclease (NYN) domain, a distinct ribonuclease family member of the PilT N-terminal (PIN) domain-like fold superfamily (18). This structure resembles the nuclease domains of DNA polymerase I and FLAP nucleases; the closest metallonuclease domain structural homologs include Taq polymerase (19) and human exonuclease 1 (20) (hExo1; Fig. 3C). However, the PRORP1 metallonuclease domain lacks hallmark structural features of FLAP nucleases, including the helical arc and the helix-two-turn-helix motifs important for nucleic acid binding. The alteration in the substrate binding site is expected given the differences in the structures of the RNA and DNA substrates of PRORP1 and FLAP nucleases. Nonetheless, these enzymes have a similar architecture around the putative active site.

The FLAP nuclease family members are proposed to use a two-metal-ion catalytic mechanism with the metal ions bound to conserved aspartate residues (20). The metallonuclease domain of PRORP1 contains four invariant aspartate residues (D474, D475, D493, and D399) that are likely important for function based on the structural similarity to hExo1 (Fig. 3) (2). However, PRORP enzymes lack two of the conserved aspartates in FLAP nucleases (D173 and D225 in hExo1, highlighted with arrows in Fig. 3C) that are proposed to chelate a second metal ion. Despite this, structural studies indicate that PRORP1 can bind two metal ions at the active site (Fig. 3). In metal soaking experiments, a single ion is bound to the active site upon addition of Ca(II) or Sr(II) ions, whereas two metal ions are observed in the presence of Mn(II). One metal ion bound to PRORP1 [Ca(II), Sr(II), or Mn₁(II); Fig. 3] forms inner-sphere interactions with D475 and water-mediated contacts with D399, D474, and D493. In the Mn(II)-bound structure, the second metal ion [Mn₂(II)] forms inner-sphere interactions with D475 and D493 and is displaced by 0.5 Å from the position of the second metal ion in hEXo1. The lower metal occupancy (80%) of this site in PRORP1 suggests that Mn₂(II) is more weakly bound. Although structural data cannot prove the function of a metal ion, we propose that the second metal ion activates catalysis by comparison with the FLAP nucleases. Furthermore, the affinity of the second metal ion may be enhanced by interactions with the pre-tRNA substrate rather than aspartate side chains. Given the similarities between the hExo1 and PRORP1 active sites, we predict that the pre-tRNA scissile phosphate will be located between the two metal ions and form inner-sphere interactions with both.

Mutagenesis experiments confirm the functional importance of invariant aspartate residues in PRORP1 (2). Alanine substitution of each of the four aspartate residues (D474, D475, D493, and D399) in Δ76PRORP1 decreases cleavage activity by >1,000-fold without significantly affecting pre-tRNA binding affinity (Fig. 3E), indicating that these side chains are mainly important for catalytic activity. The negatively charged residues located in the active site pocket lead to an overall negative electrostatic potential at the active site (Fig. 1D), perhaps mimicking the negative electrostatic potential at the active site of the bacterial RNA-dependent RNase P.

Metal Activation.

Mg(II) was previously shown to activate PRORP1 catalysis (2). To further analyze the metal dependence, we examined the activation of PRORP1 by other divalent cations in the presence of the magnesium hexahydrate mimic, cobalt(III)hexammine (21), to stabilize the tertiary structure of pre-tRNA. PRORP1 with a stoichiometric zinc ion (Table S2) cannot catalyze pre-tRNA cleavage, indicating that the bound Zn(II) is a structural cofactor (Fig. S2). No cleavage activity is observed upon addition of Ca(II) (Fig. 3E). In contrast, both Mg(II) and Mn(II) activate PRORP1-catalyzed phosphodiester bond hydrolysis to form mature tRNA (Fig. 3E and Figs. S2 and S3). Given the relative abundance of Mg(II) compared with Mn(II) in vivo (22), Mg(II) is likely the PRORP1 cofactor in mitochondria and chloroplasts, similar to other metallonucleases, including bacterial RNase P (23).

Discussion

Here we present a unique crystal structure of a protein-only RNase P, a member of the novel family of PRORP proteins. The metallonuclease domain of PRORP1 provides insight into the active-site architecture, metal dependence, and potential functional roles of active-site side chains in the broad family of NYN domain-containing proteins. The prototypical NYN domain is found across all kingdoms of life, often fused to RNA-binding motifs (18). These domains have evolved to play important roles in RNA processing and they are overrepresented in eukaryotic organisms. The similarity of this class of RNases with the FLAP nucleases is striking, suggesting that they both originated from a common structural fold that diverged from a single ancestor. This common ancestor might have provided an early protein alternative to the RNA-based RNase P (24).

The PRORP1 structure is likely typical of proteins with five in-tandem PPR motifs. A number of proteins containing PPR domains are implicated in diseases that stem from mitochondria dysfunction. For example, leucine-rich pentatricopeptide repeat cassette (LRPPRC) and pentatricopeptide repeat domain 2 (PTCD2) are two human PPR-containing proteins that are important for the processing of mitochondrial transcripts as mutations can lead to a deficient respiratory chain (7, 25).

PRORP enzymes from A. thaliana and Ostreococcus tauri are sufficient to catalyze cleavage of mitochondrial pre-tRNAs in vitro (2, 26). However, human mitochondrial RNase P requires two additional protein components for efficient tRNA maturation, a tRNA methyltransferase termed MRPP1 and its binding partner MRPP2 that has promiscuous alcohol dehydrogenase activity (1). The role of MRPP1 and MRPP2 in pre-tRNA processing remains to be firmly established, but the need for additional components may reflect the noncanonical structure and methyl modification at position 9 of tRNAs encoded by mammalian mitochondria (27). Nonetheless, the structure of PRORP1 provides a foundation for understanding the complexity of human mitochondrial RNase P, serving as a model for the human catalytic subunit of mitochondrial RNase P (MRPP3). The metallonuclease domains of MRPP3 and PRORP1 share all of the conserved active-site residues as well as the side chains that bind zinc in the central domain. Moreover, despite the low sequence identity of their N-terminal domains, the structures of these regions are predicted to be homologous (PPR domains). Thus, in the absence of an MRPP3 structure, PRORP1 provides a roadmap to rationalize how mutations in pre-tRNAs affect substrate recognition by MRPP3 (28).

Comparison with RNA-Based RNase P.

PRORP and RNA-based RNase P enzymes use strikingly different structural scaffolds to catalyze the same biological reaction. Apart from the visually similar V topologies, and the overall negatively charged and relatively flat, open, and accessible active sites, there are no conserved structural features between PRORP1 and the RNase P ribonucleoprotein complex (5). This result is consistent with bioinformatic analyses indicating that PRORP evolved separately and not from their ribonucleoprotein counterparts (2). The protein component of the bacterial RNA-based RNase P is essential for catalysis in vivo but not in vitro at high salt (29). This protein interacts with both the pre-tRNA 5′ leader and the ribozyme to enhance substrate and metal ion affinity and to stabilize the RNA active site (29–31). In contrast, the proteinaceous RNase P has the ability to recognize, orient, and bind pre-tRNA and metal ions without any extraneous assistance. The lack of conserved structural features between the bacterial RNase P proteins and PRORP1 further supports their proposed disparate evolution.

Strikingly, these structural studies suggest that PRORP1 uses a two-metal-ion mechanism to catalyze 5′-end cleavage, as previously suggested for a variety of nucleases, including enzymes composed of either protein or RNA (20, 32). The RNase P ribozyme is similarly a metalloenzyme, requiring at least two cocatalytic metal ions (5, 23). Nevertheless, differential chemical moieties at the active sites of the RNA- and protein-based enzymes likely lead to divergent mechanistic strategies. For example, the side chains of aspartate (D399) and histidine (H498) in PRORP1 are positioned such that they could catalyze deprotonation of the nucleophilic metal water and protonation of the leaving group, respectively (Fig. 4B). In contrast, RNA-based RNase P is proposed to rely on metal activation of bound water to generate the hydroxide nucleophile (33) (Fig. 4A). Additionally, phosphorothioate-substituted pre-tRNA studies suggest that the active-site metal(s) of spinach chloroplast RNase P do not coordinate the pro-R_p oxygen of the scissile phosphate bond (34). This metal coordination is in direct contrast with that of RNA-based RNase P enzymes, but similar to that of other proteinaceous RNases, such as RNase H (35). This direct comparison of the structures of RNA and protein macromolecules that catalyze pre-tRNA cleavage allows insight into the altered catalytic strategies afforded by protein side chains.

Fig. 4.

Mechanistic comparison of the RNA- and protein-based RNase P-catalyzed reaction. (A) Proposed mechanism of cleavage catalyzed by RNA-based RNase P (recreated from ref. 5). The bound metals (M1 and M2) are proposed to be coordinated by nonbridging phosphodiester oxygens and oxygen atoms of nucleotide bases within the catalytic RNA component. The metal (M1)-bound hydroxide is proposed as the catalytic nucleophile. M1 is also proposed to coordinate the pro-R_p oxygen of the scissile phosphate bond. M2 is proposed to position a water molecule for leaving-group stabilization through protonation of the 3′ hydroxyl. (B) Proposed mechanism of cleavage catalyzed by PRORP enzymes. An active-site aspartate is proposed to function as a general base, catalyzing deprotonation of a metal (M1)-bound water to activate the nucleophilic water. On the basis of comparison with hExo1 and lack of a phosphorothioate effect on the pro-R_p oxygen (34), the pro-S_p oxygen of the scissile phosphodiester is predicted to be coordinated by both active-site metal ions to increase electrophilicity and stabilize the transition state. An active-site general acid is proposed to protonate and stabilize the leaving group.

Enhanced catalytic activity is one proposed reason for the evolutionary switch from a prebiotic RNA catalyst to a protein catalyst. However, the efficiency [estimated from k_cat/K_M for PRORP3 (∼6 × 10⁴ M⁻¹⋅s⁻¹) (3) and k_chem/K_1/2 for PRORP1 (∼5 × 10⁴ M⁻¹⋅s⁻¹)] for catalysis of pre-tRNA cleavage under in vitro conditions is slower than that for catalysis by Bacillus subtilis (∼100-fold) or Saccharomyces cerevisiae (∼30-fold) RNase P under similar conditions (3, 36, 37). Nonetheless, PRORPs can complement Escherichia coli RNase P and the large multicomponent yeast nuclear RNase P in vivo (2, 38). This comparison suggests that factors other than catalytic efficiency, such as enhanced stability or regulation of expression and activity, were the main driving forces for the evolutionary switch from RNA-dependent to protein-only RNase P.

Methods

Detailed methods are provided in SI Methods.

The PRORP1 gene from A. thaliana with the mitochondrial signal sequence deleted (residues 76–572) was amplified using PCR and cloned into a pETM-11 vector that adds an N-terminal His₆-tag. The protein was expressed in E. coli in the presence or absence of selenomethione (SeMet) and purified using metal affinity, cation exchange, and gel filtration chromatography. Crystals of the SeMet-derivatized protein were obtained at 4 °C by the vapor diffusion method from 2:1 mixtures of protein solution with reservoir solution. The reservoir solution contained 18% (wt/vol) PEG 3,350 and 0.1 M sodium citrate, pH 5.5. Crystals of SeMet Δ76 PRORP1 and WT Δ76 PRORP1 in the presence of Sr and Ca were obtained by adding 0.02 M SrCl₂ or CaCl₂ into the crystallization solution described above. Crystals of Δ76 PRORP1 in the presence of Mn were obtained through soaking with a solution containing 0.05 M MnSO₄. Diffraction data were collected on beamline GM/CA-CAT 23-ID-D at the Advanced Photon Source, Argonne National Laboratory (Argonne, IL). Data were processed with HKL200 (39). Phenix AutoSol (40) was used to identify the heavy atom sites and calculate density-modified 1.98-Å experimental maps on the basis of a single-wavelength SAD dataset (Se-peak) from one SeMet Δ76 PRORP1 crystal as well as density-modified 2.4-Å experimental maps on the basis of a SAD dataset (Mn-peak) from a Mn-soaked Δ76 PRORP1 crystal (Table S1). Experimental phases were calculated with Phaser, followed by density modification by RESOLVE. COOT (41) was used to manually fix incorrectly modeled residues in all structures, and the final models were built through successive iterative rounds of refinement and manual model building. Refinement was performed using REFMAC5 (42). In the final PRORP1 models residues 76–94 and 571, and 572 were not modeled. The geometric quality of the models was assessed with MolProbity (43). PyMOL (44) was used to create molecular images.