A new crystal structure for JDBD with an ordered loop

The previously published crystal structure of JDBD (24) (PDB-ID: 2XSE) was missing residues 529–537. It has been noticed that the electron density maps from a dataset collected by students following the CSHL 2018 course on X-Ray Methods in Structural Biology, showed rather clear density for that loop. As this loop follows the α4 recognition helix and harbors the Arg532 residue which affects DNA binding (24), the diffraction data from the CSHL 2018 course were reprocessed, and the model was rebuilt and refined. The final structure was validated using MolProbity (32) and it belongs to the 100^th percentile, with a Ramachandran Z-score (33) of −0.98 ± 0.72. The new structure at a resolution of 1.95 Å adopts a highly similar conformation as 2XSE, which had been refined at a nearly identical resolution (1.90 Å). The space group is identical (P6₁22) and the cell dimensions do not differ substantially (a, b, c: 67.34, 67.34, 186.76 Å for 2XSE versus. a, b, c: 68.29, 68.29, 185.84 Å for 8BBM). However, the new crystal structure contains all residues from 392 to 561 (Fig 2A, [omit map in Fig S1]), including the 529–537 loop which is well resolved (Fig 2B and C). The B factors in the loop range between 53.5 and 125.2 (mean 91.6) compared with a mean of 65.3 for the JDBD structure excluding the five α-helices. There are no crystal contacts observed for this loop, and there are no substantial differences in crystal properties overall. Thus, it remains unclear why the 529–537 loop is ordered in the new structure and not in previous experiments.

Investigating the JDBD interaction with J-DNA by computational modeling

We then wanted to use our new crystal structure of JDBD for docking to a J-DNA model to investigate if that would lead to new insights in the possible binding mode of JDBD to J-DNA. Whereas previously we have done such studies manually, here we decided to follow a computational approach. In the manually created model (24), base-J is oriented perpendicular towards the major groove of the DNA and the interactions of the O2 and O3 hydroxyl groups of base-J with the non-bridging phosphoryl oxygen at J-1 in J-DNA are modeled, as first observed by reference 28. To now be able to create a computational model, parameters to model base-J and J-DNA were generated and implemented in HADDOCK2.4 (34). In the docking protocol, JDBD was docked onto a 20-mer J-DNA helix. The Asp525 residue was restrained to be in the proximity of the C4 and C6 hydroxyl groups of the sugar moiety of base-J in both complexes, based on its importance for recognizing base-J.

The docking study using JDBD resulted in three HADDOCK clusters with docked structures, of which the most occupied cluster contains 93% of the generated models. The best scoring model is shown in Fig 3A. The JDBD crystal structure was then used to start MD simulations to explore the behavior and flexibility of this protein domain in solution and obtain additional templates for further docking studies. Two parallel simulations were run, in which both the Cα root-mean-square deviation (RMSD) and root-mean-square fluctuation values were comparable (Fig S2A and B). The overall fold of the protein is maintained as seen by clustering the obtained structures (Fig S3), with, as expected, more flexibility for loops and termini compared with residues located in the helices (Fig S2B). As in both MD simulations, most obtained conformations are represented by the three most occupied clusters (93% coverage in simulation run1, 64% coverage in run2, Fig S4A–D); we assume that the center structures of JDBD represented in these six clusters provide a representative set of structures for the behavior and flexibility of JDBD in solution.

The six selected representative JDBD models from clustering the MD trajectories were used to investigate if protein flexibility would affect the JDBD:J-DNA model proposed by HADDOCK. In the obtained complexes, the overall positioning of JDBD along the J-DNA strand is consistent (Figs 3B and S5A–E and Table S1). The α1 helix is positioned towards the J-DNA, but has a different orientation in Cluster 1, compared with Clusters 2, 3, 4, and 5 (Fig 3C). The model obtained using Cluster 6 as the JDBD template shows a mirrored protein orientation with respect to the J-DNA compared with the models obtained using the centers of Clusters 1–5 as the protein template (Figs 3B and S5F). Although this mirrored orientation is preferred in most of the HADDOCK models (69%), the remaining 31% of the structure models obtained from the docking run with Cluster 6 is similar to the models in Clusters 1–5.

To validate the docking models, the obtained structures were compared with experimental SAXS and HDX-MS data presented in previous work (24). The experimental SAXS curves fit well to the calculated curve for the models obtained from docking, as indicated by the χ² values (Fig S6A–F). Furthermore, the HDX-MS data were mapped onto the docking output (Fig S7A–F), demonstrating that the α4 helix (which contains the Asp525 residue) and the α1 helix (which is positioned towards J-DNA as in the docking models) show the most pronounced HDX-MS reduction.

The interaction of base-J with Asp525 and the known intramolecular interaction with the phosphate group at J-1 that were used as explicit HADDOCK restraints, and are respected in the models (Fig S8A–E). Residues Lys518 and Lys522, which were previously suggested to be involved in complex formation (24), do not form direct interactions with J-DNA in the obtained models. On the other hand, in the complexes with JDBD Clusters 1, 2, and 4, Arg532 seemingly forms hydrogen bonds with the non-bridging phosphoryl oxygen of the nucleic acid two base pairs that are further along the opposite strand (J + 2′ position) (Fig S8A, B, and D). In Clusters 3 and 5, this arginine is positioned such that hydrogen bond formation with the non-bridging phosphoryl oxygen of the nucleic acid base pair at position J + 3′ can be possible (Fig S8C and E). However, the distance between the NH hydrogen of Arg532 and the non-bridging phosphoryl oxygen in the DNA is too long (≥2.5 Å) to form direct hydrogen bonds in all models. Because a docking model is a static representation of the JDBD:J-DNA complex, it is likely that hydrogen bonds or salt bridges are formed with the phosphoryl group at the J + 2′ and/or the J + 3′ position to stabilize the negative charges in the J-DNA backbone.

Computational modeling provides new insight into JDBD mediated J-DNA recognition

Having verified that our previous hypotheses about the JDBD interaction with J-DNA were basically correct, we examined the models for novel insights focusing on charge-conserved residues in the α1–α2 region that are interacting with J-DNA. Based on multiple-sequence alignment of JBP1 in 28 different species (Fig S9), Glu437, His440, Arg448, and Asn455 were selected for site-directed mutagenesis. Arg448 and Asn455 are located in the α1–α2 loop, fully conserved and in proximity to the J-DNA when the α1 helix is oriented towards the DNA. Both residues are also located within the part that shows HDX-MS reduction (9% for peptide 442–463) (Fig 4A and B). The negative charge of residue 437 is conserved in 93% of the species and 86% of residues present at position 440 are positively charged (Fig S8). Both Glu437 and His440 are pointing towards the DNA and are also positioned within the region that shows high HDX-MS reduction (23% for the peptide 434–441) (Fig 4A).

The single-point mutants JDBD-E437A, -H440A, -R448A, and -N455A were produced and purified (Fig S10). For mutant E437A, the expression was substantially reduced compared with wt-JDBD and the other JDBD mutants, suggesting that this mutation could also affect the overall stability of the protein which could hamper binding to DNA. To address this issue, the thermal stability of wt- and mutant JDBD proteins was analyzed at two protein concentrations using nanoDSF. The E437A mutant showed about 9–12°C reduction in melting temperature compared with that of wt-JDBD and the other mutants (Table S2). The other mutants are as stable or slightly more stable compared with the WT, suggesting that the DNA-binding analysis is not affected significantly by altered protein stability. With all mutant proteins in hand, binding to both J-DNA and T-DNA was measured by fluorescence polarization. The wt-JDBD showed an affinity of 17 nM. Substitution of E437 and H440 only resulted in a minor difference in affinity for J-DNA (30 and 57 nM, respectively; Fig 4C and Table S3), indicating that these residues have a negligible contribution to the stabilization of the JDBD:J-DNA complex. In contrast, R448A showed a ∼500-fold reduction in affinity to J-DNA (∼8 μM) and N455A a ∼400-fold reduction (∼7 μM). The affinities of all four mutants and wt-JDBD towards T-DNA were all similar (>100 μΜ, Fig S11 and Table S3), indicating that mutation of Arg448 and Asn455 residues induces a loss in discrimination towards J-DNA (24).

Based on these new experimental insights, residues Arg448 and Asn455 were included as additional active residues in HADDOCK. The results show one clearly most occupied cluster that contains 79% of the generated structure models (Table S4). The best scoring structure in this cluster (cluster1_1.pdb) (Fig 5A and B) fits the SAXS curve obtained previously (24) (Fig 5C). In addition to the interactions found in our initial docking experiments, Arg532 forms hydrogen bonds with the non-bridging phosphoryl oxygens at position J + 2′. Interestingly, Asn455 forms a hydrogen bond with Lys522, which previously was found to be an important residue for the JDBD:J-DNA complex formation (24). This interaction supports Asn455 in forming an additional hydrogen bond with the J-DNA backbone to stabilize the negative charge. The residues Lys518 and Arg447 are also contributing in stabilizing the negative DNA charge distribution by orienting towards the DNA backbone and forming salt bridges with phosphate oxygen atoms. Such interactions are also observed for Arg448 and Arg517, of which, the latter was previously suggested to be relevant for complex formation (24). The residues described in the JDBD:J-DNA interface are in general highly conserved (Fig S12). Asp525 is fully conserved in Leishmania, Trypanosoma, Leptomonas, and Bodo saltans species, so are Lys522, Arg532, and ArgR448. Asn455, which we identify in this study, is not conserved in Leptomonas, and Arg517 is not conserved in Trypanosoma also.

New insights through AI-based modeling of JBP1

The AlphaFold structure prediction model (30) of JBP1 (entry Q9U6M1) (Fig 6A) is a source of new information that became available during the course of this study. In this predicted model, JDBD shows a similar overall fold when compared with our new crystal structure (Fig 6B), with a noticeable difference in the loop between the α4 and α5 helices, that we found ordered in the new JDBD crystal structure. The THD of the AlphaFold model was compared with the hydroxylase domain of the human TET2 homologue (Fig 6C and D), which looks similar based on secondary structure elements (RMS difference of 0.35 over 120 selected Cα atoms). The AlphaFold model also shows that β-sheets are continuing between the N- and the C-terminal parts of the THD (Fig 6D) This observation confirms our previous results describing the THD domain (26), which is formed between the N- and C-terminal parts of JBP1 and retained enzymatic activity in vitro. The AlphaFold model fits the SAXS curve that was obtained previously (26) for the full-length JBP1 protein and the THD only (Fig 6E). The χ² value of the fit is slightly elevated because of the presence of the connecting loops between the THD and the JDBD and the N-terminus in the protein used for measuring the SAXS curve, which were removed from the computational model. Examining the hydroxylation site in the THD domain of the AlphaFold model, the residues involved in Fe²⁺ and oxoglutarate binding are in similar conformations as in the corresponding binding domain TET2 homologue (Fig 6F). These observations suggest that the AlphaFold prediction provides a reliable description of the JDBD and THD that can be used for further investigation.

The predicted aligned error (PAE) plot of the AlphaFold JBP1 model is consistent with the flexibility between the THD and the JDBD (Fig 7A), which we previously suggested based on experimental data (26). Interestingly, the PAEs for the N-terminal ∼40 residues of JBP1 suggest that the N-terminal residues are reliably modeled in proximity to the JDBD. This PAE pattern stands for Leishmania, Crithidia, and Leptomonas species but not for Trypanosoma (Figs 7B and S13). The AlphaFold model for Leishmania tarentolae predicts two short helices in the N-terminal region, connected by a linker. The per-residue confidence scores (pLDDT) range between “low” and “confident” (Fig 7C). In addition, the N-terminal region has positively charged residues that could, in principle, mediate DNA binding (Fig 7D). Finally, we have previously shown that the JDBD binds to J-DNA with an affinity about three times less than full-length JBP1 (24). These observations and previous findings prompted us to investigate if the JBP1 N-terminus plays a role in DNA recognition.

To this end, we decided to study the J-DNA binding affinity of JBP1 truncated after the first predicted N-terminal helix (at position 23, Δ23-JBP1) and after both N-terminal helices (at position 38, Δ38-JBP1). The truncated mutants were expressed and purified (Fig S10), and their affinity for J-DNA was studied by fluorescence polarization (Fig 7E). Whereas wt-JBP1 binds DNA with an affinity of 6.4 nM, Δ23-JBP1 binds J-DNA three times weaker (20.8 nM). Removal of all 38 N-terminal residues results in a further decrease in affinity to 31.2 nM (Table S5). These findings suggest that the N-terminal residues are involved in J-DNA binding. Notably, the affinity of the JDBD alone on J-DNA is 20.7 nM, which is similar to the N-terminally truncated JBP1. The derived free energy of binding (ΔG) of the JDBD alone or Δ23-JBP1 is about −48 kJ mol⁻¹, whereas the ΔG for wt-JBP1 is −45 kJ mol⁻¹. Thus, the difference in binding free energies between JDBD and the truncated mutants is relatively small (3 kJ mol⁻¹ or less). Based on all these data, we attempted to create a molecular model of JBP1 bound to DNA. First, we used the THD of the AlphaFold model and the X-ray model of a TET2 homologue in complex with DNA, to create a model of the THD of JBP1 in complex with J-DNA. The THD active site was bound to a thymidine 13 nucleotides downstream base-J, according to the preferred offset between an existing and a new base-J (27). Then, the JDBD:J-DNA complex we obtained above was added to this model by aligning the corresponding DNA regions. However, several modeling attempts did not yield a model that fits with the SAXS data (Fig S14). During the modeling process, it also became apparent that the N-terminal region in the exact conformation of the AlphaFold model does not allow DNA binding by the JDBD. Based on these findings and as illustrated in Fig 7F, we propose that the combined binding of the THD and JDBD to J-DNA is accompanied by structural rearrangements which likely include DNA bending to allow simultaneous binding of the JDBD and THD-mediated hydroxylation of a thymidine that is located 13 nucleotides downstream in the opposing DNA strand, followed by the N-terminal region wrapping around DNA, and stabilizing the complex.

Macromolecule production and crystallization

Protein expression, crystallization, crystal harvest, and X-ray crystallographic experiment were performed as described before (24). Details are shown in Table 1.

X-ray crystallography procedures

Crystallographic details are shown in Table 2. The images were processed and scaled at the 17-ID-1 (AMX) beamline of the National Synchrotron Light Source II at Brookhaven (36). The integrated data were scaled in CPP4i2 (37, 38) using AIMLESS (39). The space group P6₁22 was selected and diffraction data extended to 1.95 Å. The structure of JDBD (PDB-ID: 2XSE) was used as search model for molecular replacement. Next, the model was refined in multiple iterative cycles between REFMAC (40) and manual modeling in COOT 0.8.9.2 (41). Within REFMAC, a single TLS body was refined for 15 cycles with starting B-factors fixed to a value of 20.0 Ų, jelly body restraints with sigma 0.01, and a maximum distance of 4.2 were set. Typically, 30 cycles were run using riding hydrogens. Optimization of the model based on the PDB-REDO webserver (42) was used to provide additional optimized refinement parameters: simple solvent scaling was used with explicit solvent mask with custom parameters (1.0 increase VDW radius of non-ion atoms, 0.9 increase ionic radius of potential ion atoms, shrink the mask area by 0.9 after calculation). Double conformations were modeled for Ser408, Glu447, Glu488, Met494, and Met546. The final structure was validated using MolProbity (32). The structure has been deposited to the Protein Data Bank (PDB) (43) as entry 8BBM.

Molecular dynamics simulations of JDBD

Docking base-J containing DNA to JBP1 with HADDOCK

Protein assays

Modeling of the JBP1:J-DNA complex using AlphaFold

The AlphaFold2.0 structure of the full-length JBP1 was downloaded (entry Q9U6M1) from the AlphaFold protein structure database (29, 30). To create a model of the THD domain, residues 1–39, and 366–576 (low confidence and JDBD) were removed from the model. Then, the PDB-REDO (42) crystallographic structure of a TET2 homologue of the hydroxylase domain (PDB-ID: 5DEU) was aligned to the THD of this model and the coordinates of loop 113–121 in the AlphaFold model were replaced by the coordinates of the loop 1,291–1,299 of the TET2 homologue. The DNA of the TET2 homologue was used to model THD:J-DNA binding in the AlphaFold structure. The 5HC in the DNA was mutated to hydroxymethyluracil (5MU) and the DNA was adjusted to a 25-mer J-DNA with sequence 5′-TCGATTJGTTCATAGACTAATACGT-3′. This THD model in complex with the 25mer J-DNA was energy minimized in vacuum, followed by energy minimization in water using the Amber99 force field (59) and the TIP3P water model in the YASARA software package (55). The ligands oxoglutarate and Fe²⁺ were added to the THD of the newly generated model based on their respective positioning in the TET2 homologous structure model. Next, the JDBD:JDNA model obtained from docking protocol iii (see the above section) was aligned with the this model in such way that the J-DNA backbone atoms align and the base-J position overlaps. The JDBD was added to the model, and the whole complex was energy minimized in vacuum and water once more. This finally resulted in a JBP1:J-DNA model containing a 25mer J-DNA in complex with the JDBD and THD-JBP1 including the oxoglutarate and Fe²⁺ ligands in the THD.