Neutron diffraction studies of Escherichia coli dihydrofolate reductase complexed with methotrexate

Abstract

Hydrogen atoms play a central role in many biochemical processes yet are difficult to visualize by x-ray crystallography. Spallation neutron sources provide a new arena for protein crystallography with TOF measurements enhancing data collection efficiency and allowing hydrogen atoms to be located in smaller crystals of larger biological macromolecules. Here we report a 2.2-Å resolution neutron structure of Escherichia coli dihydrofolate reductase (DHFR) in complex with methotrexate (MTX). Neutron data were collected on a 0.3-mm³ D₂O-soaked crystal at the Los Alamos Neutron Scattering Center. This study provides an example of using spallation neutrons to study protein dynamics, to identify protonation states directly from nuclear density maps, and to analyze solvent structure. Our structure reveals that the occluded loop conformation [monomer (mon.) A] of the DHFR·MTX complex undergoes greater H/D exchange compared with the closed-loop conformer (mon. B), partly because the Met-20 and β(F-G) loops readily exchange in mon. A. The eight-stranded β sheet of both DHFR molecules resists H/D exchange more than the helices and loops. However, the C-terminal strand, βH, in mon. A is almost fully exchanged. Several D₂Os form hydrogen bonds with exchanged amides. At the active site, the N1 atom of MTX is protonated and thus charged when bound to DHFR. Several D₂Os are observed at hydrophobic surfaces, including two pockets near the MTX-binding site. A previously unidentified D₂O hydrogen bonds with the catalytic D27 in mon. B, stabilizing its negative charge.

Hydrogen (H), the major constituent atom in biological molecules, remains the most difficult to visualize, even with data from ultrahigh-resolution x-ray crystallography (beyond 1.2 Å). Knowledge of hydrogen positions is crucial for understanding enzyme catalysis and molecular recognition and aids drug design. Neutron crystallography (NC) has set several benchmarks, directly revealing hydrogens in proteins and elucidating enzymatic mechanisms ranging from serine proteases (1) to ribonucleases (2) to aspartic proteases (3). Moreover, several other structures have been solved by using NC (4–7).

Since we solved the first neutron structure of a protein (myoglobin) in 1969 (8), there has been immense interest in neutron crystallography. However, its practical use has been limited by weak beam fluxes, the requirement for large crystals (>1 mm³), and a lack of dedicated beamlines at reactor neutron sources. Spallation neutrons provide a new tool for protein crystallographers. Bombarding a metal target with pulses of high-energy protons produces a neutron beam with a range of energies and a pulsed time structure that allows data to be collected with high efficiency and low background using the TOF Laue method. The first protein crystallography station to be built at a spallation source has been operating at Los Alamos Neutron Scattering Center since August 2002 (9). We wanted to investigate the feasibility of solving a protein structure from spallation data. For this study, we used Escherichia coli dihydrofolate reductase (DHFR) in complex with the well-known chemotherapeutic agent methotrexate (MTX; ref. 10) to study protein dynamics, to identify the protonation states at the active site, and to determine solvent structure.

DHFR is essential for most biosynthetic pathways involving one-carbon transfer reactions because of its recycling of THF. It catalyzes the NADPH-dependent reduction of 7,8-dihydrofolate, producing 5,6,7,8-THF. DHFR has been extensively studied by NMR (11), site-directed mutagenesis (12), and molecular dynamics (13) to unravel the potential coupling of motion by distal residues to catalysis (14, 15). In particular, a regulatory loop, the Met-20 loop (residues 9–24), has been observed to adopt open, closed, and occluded conformations by x-ray crystallography (16) and closed and occluded conformations by NMR (17, 18). Conformations of the Met-20 loop induced by ligand binding are stabilized by two accessory loops called the β(F-G) (residues 117–131) and β(G-H) (residues 142–149) loops (Fig. 1A). Also, in stopped-flow kinetics and single-molecule fluorescence experiments, MTX was shown to bind two different conformers of apo DHFR, and the enzyme displayed an additional conformational change attributed to opening and closing of the Met-20 loop (19). Interestingly, the x-ray structure of the DHFR·MTX complex reveals two molecules in the asymmetric unit in which the Met-20 loop adopts closed and occluded conformations (ref. 20; Fig. 1 A and B); these conformational fluctuations are consistent with single-molecule fluorescence results (19, 21). Furthermore, MTX binds in an orientation in which the pterin ring is flipped 180° with respect to substrate, resulting in the MTX N1 atom being placed within H-bonding distance of the catalytic D27 residue (20, 22).

Fig. 1.

Neutron structure of DHFR·MTX complex. (A) Stereo representation of the Cα traces of the two monomers in the asymmetric unit. mon. B (red; closed Met-20 loop) has been superposed onto mon. A (blue; partially occluded Met-20 loop). (B) A close-up view of the two different Met-20 loop conformations present in the structure (shown as Cα traces). Color scheme is as in A. (C) 2.2-Å resolution 2F_o−F_c nuclear density map (blue) (1.5σ) at the interface of the βF and βG strands of mon. A. (D) Backbone amide H/D exchange for mon. A color-coded for refined deuterium occupancy values. Black: Nonexchanged (occupancy = 0–0.15); dark gray: Partially exchanged (occupancy = 0.15–0.70); and, light gray: Fully exchanged (occupancy = 0.70–1.00). (E) The same representation and color-coding as in (D) but for mon. B.

Protein folding and dynamics have been extensively studied by the H/D exchange method (23) and can be coupled with NMR spectroscopy, MS (24), and even Infrared spectroscopy (25). Similarly, analysis of H/D exchange with neutron diffraction (ND) is a powerful tool for investigating dynamics (26, 27). Recently, H/D exchange was used to study the structural fluctuations of various DHFR complexes by MS (28) and NMR (29). In the MS study on E. coli DHFR, the exchange of backbone amides in the Met-20 loop was fast, and ≈75% of the twisted eight-stranded β-sheet of apo DHFR was exchanged. However, ligand binding dampened the exchange rates (28). In the NMR study of Lactobacillus casei DHFR, changes in amide protection induced by ligand binding are not localized to specific binding sites but are extended throughout the entire structure by an intramolecular interaction network (29). With the ND structure of the DHFR·MTX complex, we are able compare the H/D exchange patterns between the occluded and closed conformers to probe loop dynamics. By visualizing H/D exchange, an extensive analysis of water structure and H-bonding patterns can be performed directly.

Here we report the 2.2-Å resolution ND structure of E. coli DHFR bound to MTX from data collected at a protein crystallography station. The present structure provides insights into dynamics, active-site protonation states, and the solvation pattern of DHFR. Moreover, this study demonstrates how neutron structures can be determined by using data collected from relatively small crystals at a modern spallation neutron source.

Results and Discussion

Neutron Structure of E. coli DHFR Bound to MTX.

TOF ND data were collected on a single 0.3-mm³ D₂O-soaked DHFR·MTX crystal. The DHFR·MTX complex crystallized with two molecules in the asymmetric unit, designated A and B. The largest differences between the two monomers occur at the regulatory loop regions: the Met-20, β(F-G), β(G-H), and adenosine-binding (residues 63–72) loops. The most dramatic difference is at the Met-20 loop, which differs in conformation between the monomers from partially occluded in monomer (mon.) A to closed in mon. B (ref. 16; Fig. 1 A and B). The overall Cα rmsd between monomers A and B is 0.52 Å.

The ND structure has been refined to an R_free of 23.3% at 2.2-Å resolution with reasonable geometry (Table 1, which is published as supporting information on the PNAS web site). Upon initial inspection, the neutron maps (Fig. 1C) show significant nuclear density for nearly all main-chain and many side-chain atoms. Cancellation effects due to the negative scattering cross section of hydrogen result in weak nuclear density near hydrocarbons and other atoms bound to nonexchangeable hydrogens. Occupancy refinement of D at exchangeable positions on the amide backbone and at certain side chains was performed. There are 294 exchangeable backbone amide hydrogens in the asymmetric unit. Of these, 176 exchanged for deuterium (D), or ≈59.9%. In mon. A, 95 of 148 amides (≈65.1%) have undergone partial to full H/D exchange, whereas in mon. B, 81 of 148 amides (≈54.7%) have exchanged (Fig. 1 D and E). In our structure, 20 amides are involved in crystal contacts. Of these, 11 have exchanged to D (see Supporting Text, which is published as supporting information in the PNAS web site). Partial or full H/D exchange has occurred for 16 of 18 Arg residues, 11 of 12 Lys residues, and 6 of 10 His residues.

Backbone H/D Exchange and DHFR Dynamics.

Because both the occluded and closed Met-20 loop conformations are observed in our structure (Fig. 1 A and B), the ND method with H/D exchange permits us to compare exchange between the two monomers to gain insight into any differences in dynamics. To prevent overinterpretation of the relevance for small changes in the exchange ratio, we have divided H/D exchange into three classes: nonexchanged (0–15% D), partially exchanged (15–70% D), and fully exchanged (70–100% D), similar to a previous ND study (27) (see Table 2, which is published as supporting information on the PNAS web site). We have compared B factors, solvent accessibility [calculated using AREAIMOL (30)], atom depth, secondary structure, and H-bonding to rationalize the H/D exchange patterns. Monomer A (occluded Met-20 loop) has significantly enhanced amide H/D exchange in comparison to mon. B (closed Met-20 loop; Fig. 1 D and E). Interestingly, mon. A has a 600 Ų larger solvent-accessible surface area compared with mon. B, and the average amide B factors for monomers A and B are 21.3 and 18.2 Ų, respectively. Although there are several differences between monomers A and B in the H/D exchange pattern across the entire DHFR molecule, the greatest of these seem to be at the Met-20, β(F-G), αF-βD (residues 86–89) loops, and β strand H (residues 151–159). We will describe exchange differences specifically for the Met-20, β(F-G) and β strand H. With some notable exceptions, including β strand H, many of the β strands between monomers A and B share similar exchange and H-bonding patterns (Fig. 3 A and B, which is published as supporting information on the PNAS web site).

It is not surprising that the Met-20 and β(F-G) loops in mon. A undergo more H/D exchange than in mon. B, because the behavior of these loops in A is more dynamic in the occluded conformation. In general, the Met-20 loop in A is more solvent-accessible, and the average B factors are slightly higher than in B. In mon. A, 13 of 16 amides are fully exchanged, whereas in mon. B, nine are fully and one is partially exchanged. When comparing the H/D exchange pattern between the two monomers (Fig. 3 C and D) in mon. A, the left side of the loop (residues 12–17) is fully exchanged, whereas in mon. B, residues V13, I14, and M16 are nonexchanged. It is interesting to note that the interloop backbone H-bonds between the Met-20 and β(F-G) loops involving D122 and T123 observed in B are absent in A, because the β(F-G) loop packs more tightly against the Met-20 loop in B than in A (Fig. 3 C and D Inset). At the bottom left of the Met-20 loop, residues 9–11 in mon. B are fully exchanged, whereas in mon. A, V10 is nonexchanged. The main difference is that, in mon. A, the V10 amide forms bifurcated H-bonds to the carbonyl oxygens of I115 and A117, whereas mon. B interacts only with A117. The tip of the Met-20 loop (E17 and N18) in mon. A is fully exchanged, whereas in mon. B, N18 is partially exchanged. On the right side of the Met-20 loop (residues A19-L24) in mon. A, residues A19, W22-L24 exchange, whereas M20 and W22-N23 exchange in mon. B. Note that the M20 amide H-bonds to a D₂O molecule in mon. B, yet this is not observed in mon. A, possibly explaining why exchange has occurred in mon. B. Several D₂O molecules interact with the backbone atoms of Met-20 loop residues, notably D11 and A19 in mon. A and E17, M20 and W22 in mon. B, all of which have undergone exchange.

The β(F-G) loop, which packs against the Met-20 loop, has nine fully and one partially exchanged amides in mon. A, whereas in mon. B there are five amides that have fully exchanged with one partially exchanged (Fig. 3 E and F). For the β(F-G) loop, the average amide temperature factor and surface accessibility are higher in mon. A than in mon. B. The greatest difference in H/D exchange occurs between G121-F125; in B, none of the amides are exchanged, whereas in A, all of the amides except T123 are exchanged. The protection from exchange at these amides in mon. B may result from the interactions made to the Met-20 loop, as mentioned above. In mon. B, the main chain of G121-F125 shares four H-bonds with R12-G15 of the Met-20 loop, whereas the same regions in mon. A form only one H-bond, that between the amide of F125 and carbonyl of R12. Another notable difference occurs at D127; it is exchanged in A but not in B. One possible reason for this can be that in mon. A, D127 makes van der Waals interactions with a cluster of D₂O molecules.

The C-terminal strand, βH (Fig. 3 A and B), is another region that exhibits striking dissimilarity in H/D exchange between the two monomers. In mon. A, six of nine amides have fully exchanged, whereas only two of nine have done so in mon. B. In either monomer, βH is essentially inaccessible to solvent; however, the B factors in mon. A are larger than in mon. B (21.2 vs. 12.4 Ų). Strand βH, the seventh strand from top to bottom in Fig. 3 A and B, lies between strand βF (top) and βG (bottom). Although the H-bonding patterns between the βG and βH strands are conserved in both monomers, they are different between residues 155–157 of βG and βH; in mon. B, this region forms more H-bonds with nonexchanged amides in βH. Interestingly, the same regions between strands βG and βH in mon. A are further apart than in mon. B, permitting intrusion of a D₂O molecule within H-bonding distance of the exchanged amide of E154. It should be noted that second-site revertants found within this strand help to overcome adverse effects to catalytic efficiency in the active site D27S mutant, possibly because of stabilizing long-range interactions with the active site (31). Additionally, sequence epitopes from this strand have been used to generate antibodies against human DHFR (32), and antibody binding at this region distal from the active site significantly affects conformation and catalysis (33).

In an attempt to correlate overall H/D exchange patterns to structural properties including B factors, solvent accessibility, H-bonding, and atomic depth, we plotted these parameters against the D occupancy of backbone amides (Figs. 4 A–F and 5 A–D, which are published as supporting information on the PNAS web site). Although B factors, accessibility, and H-bonding somewhat correlate to exchange in local regions such as the Met-20 loop, Fig. 4 A–F show that, for the entire molecule, there are no obvious patterns. Similar observations were made for trypsin (27). A breakdown of secondary structure vs. H/D exchange shows the following: the cumulative H/D exchange rates for β strands and helices are 57% for mon. A and 47% for mon. B, whereas for loops, they are 62% and 56% for mon. A and B, respectively. However, if helices and loops are pooled together and their exchange rates compared with strands, mon. A has 61% exchange in the helix–loop category and 57% exchange for the strands. In mon. B, 54% exchange has occurred among helices and loops whereas 44.8% exchange has taken place in the strands. This is in agreement with an E. coli DHFR MS study, where the strands were shown to be more protected from H/D exchange as compared with loops and helices (28).

Atomic depth calculations can provide an extra layer of information on buried residues that solvent-accessibility values cannot; in fact, depth seems to correlate better to H/D exchange than accessibility for a number of proteins (34). We calculated depth using DPX (35), defined as the distance a backbone amide is from the nearest solvent accessible atom; this method appears to correlate well with the nearest water-molecule approach (36). Generally, the vast majority of the residues are at the protein surface (those with depths = 0) compared with a few that are buried (Fig. 5 A and B); thus, it is better to compare the percentage of H/D exchange at several depths (Fig. 5 C and D). Interestingly, for mon. A, the H/D exchange pattern seems to correlate with depth (Fig. 5C). For mon. B, this correlation breaks down at depths >4 Å (Fig. 5D). However, in both monomers, the majority (66% and 67%, respectively) of surface amides has undergone partial or full exchange, whereas only 46% and 26.7% of amides that have depths >3 Å in mons. A and B, respectively, are exchanged.

Globally, there may not exist any single parameter to explain the basis for H/D exchange. However, we observe several D₂O molecules that H bond to exchanged amides, even those that are located within the interior of the protein. In the crystalline state, as observed by others (27), solvent permeation (allowing inclusion of ordered D₂O molecules) and local unfolding events (protein “breathing”) are likely to be determinants of H/D exchange, providing a mechanism by which buried amides can come into favorable contact with D₂O to allow exchange (23).

Study of D27·MTX interactions at the DHFR Active Site Using ND.

Examination of the neutron maps revealed both monomers had density for a D atom ≈1.0Å from the N1 atom on MTX (Fig. 2A). Its occupancy was refined to 0.37 in mon. A and 0.73 in mon. B. The distance between the N1 and the D atom in both monomers refined to 0.86 Å, whereas the distance between the D atom and the Oδ2 atom of D27 in both mons. is 2.10 Å. For mon. A, the N1, D, and Oδ2 atoms all are positioned on a near-perfect plane; the angle is 176°. (The same atoms in mon. B deviate from this, at 152.5°.) The occupancy of a D atom placed 1.0 Å from the Oδ2 atom of the D27 carboxylate on both monomers refined to quite low values, suggesting that D27 is unlikely to be neutral in the MTX complex. In addition to the occupancies, the nuclear density around MTX and D27 in both monomers (Fig. 2A) suggests that a D atom is bound to N1. At the same contour level (≥1.5 σ), no nuclear density is observed for a D to be bound to the Oδ2 atom of D27. We have performed full matrix refinement of a 1.0-Å resolution DHFR·MTX x-ray structure to compare D27 carboxylate bond lengths to determine the D27 charge state. This analysis also indicates that D27 is ionized (unpublished data). Taken together, these results indicate that the N1 atom is protonated when MTX is bound to DHFR and D27 is ionized, facilitating the formation of a salt bridge between the MTX N1 and the D27 Oδ2 atoms.

Fig. 2.

Characterization of the DHFR·MTX-binding site and overall solvent structure. (A) Shown in cyan is 2F_o−F_c nuclear density (1.5σ) and in white are F_o − F_c density (+3σ) maps for the D27·MTX interaction site in mon. B. The D₂O shown is 2.6 Å from the D27 Oδ2 atom. (B) Electrostatic surface representation of the crystallographic dimer of DHFR·MTX. Negatively charged surfaces are red, positively charged ones blue and hydrophobic surfaces are white. D₂O molecules are shown as sticks (white for deuterium and red for oxygen). (C) View of the hydrophobic surfaces (1 and 2) shown in D and F. Surface 1 is between the two MTX molecules, whereas surface 2 is near the β(F-G) loop that is rearranged upon MTX binding. The MTX and D₂O molecules are shown as sticks. This orientation is rotated 180° parallel to Y and 90° parallel to Z compared with the stereo image in A1. (D) Closeup view of surface 1. (E) 2F_o−F_c nuclear density (cyan; 1.5σ) for some of the D₂O molecules in surface 1. (F) Closeup view of surface 2.

In mon. B, an F_o−F_c peak (+3σ) existed near the D27 Oδ2 atom; a D₂O was placed in this position and refined (Fig. 2A). The D₂O is within H-bonding distance (2.5 Å) of the D27 Oδ2 atom, and the O-D2–Oδ2 bonding angle is 175.4°. Although this water was not observed in the RT x-ray structure, it is present in our 1.0-Å resolution x-ray structure (unpublished results). It appears the closed Met-20 loop is not so restrictive to disallow D₂O access to the active site. In fact, the MTX N1 atom and the D27 Oδ2 are more solvent-accessible in mon. B than in mon. A. This may explain why D₂O and the fully exchanged D at the N1 atom of MTX occur only in mon. B. Recent theoretical calculations have predicted that Met-20 loop conformational fluctuations affect hydride transfer rates (37) as well as the protonation state of the substrate DHF (38). In our study, the protonation state of MTX does not change; however, the Met-20 loop conformations affect the extent of H/D exchange at the N1 atom.

The ND studies further concur that D27 in both monomers is negatively charged at physiological pH. The ionic interaction between Oδ2 of D27 and the N1 atom of MTX is most likely the driving force behind the several orders-of-magnitude lower K_d of E. coli DHFR for MTX (pM) than for substrates (μM) (39). Although MTX cannot be termed a transition-state analog in the strictest sense, it has been suggested that the conformation of the enzyme closely resembles the transition state when MTX and NADPH are bound in the active site, and the Met-20 loop is closed (40). The inverted conformation of the MTX pteridine ring compared with substrates when bound at the active site reduces overlap between NADPH and MTX, providing optimal binding and less steric clashing at the active site.

Solvent Structure.

ND studies using H/D exchange provide important details of solvent structure (41, 42). Analysis of the nuclear density maps and refinement of D occupancies resulted in 144 D₂O molecules being added to the structure. Additionally, eight solvent molecules have been added as only oxygen atoms, these we define as either orientationally disordered D₂O or nonexchanged H₂O molecules. Most of these eight solvent molecules are near D₂O networks at the protein surface and form H-bonds only with other solvent. A total of 234 H-bonds are formed between the D₂Os and the protein. Of the 144 D₂Os, 37 are conserved between the ND and the RT x-ray structure [3DRC; (43)]. A total of 138 D₂Os comprise the primary solvation shell around both DHFR monomers. Fig. 2B shows how the majority of D₂Os interact with charged surfaces. A few D₂O molecules are found at the slightly hydrophobic dimer interface, and others are observed at the extremity of the protein surface near crystal contact regions. Six D₂Os are >5 Å away from the protein. Of these, only two have B factors >40 Ų, and three form H-bonds with other D₂Os. The 20 D₂Os with the highest B factors (≥45 Ų) are on average 4 Å from the protein. Similarly, other studies have shown that distal D₂Os have higher B factors (27, 42, 44). The majority (13) of this group of 20 is involved in H-bonds to other solvent, and half make apolar O—H-C contacts.

We observe several solvent surfaces at hydrophobic pockets including two main ones that are at the dimer interface and related to each other by a 180° rotation parallel to the z axis (Fig. 2C). The first one is a partially hydrophobic cavity surrounded by two basic surfaces close to the dimer interface where 11 solvent molecules are observed near the MTX pABA-Glu tails (Fig. 2D). Nuclear density maps of D₂Os at this surface are shown in Fig. 2E. Three of the D₂Os H-bond to the protein, and one forms a H-bond with MTX. Four D₂Os make 3.8- to 4-Å polar contacts, and three make apolar contacts with the protein. Of the three D₂Os involved in apolar contacts, two form a H-bond to one another. Five of the 11 solvent molecules are conserved between the RT x-ray and ND structures within this cavity. However, the D₂Os that contact the apolar surface were not observed in the x-ray structure. Of the three apolar D₂Os, 2 have B factors that are ≈30Ų, and the third has a B factor of 46 Ų. The closest protein contact to the latter D₂O is 5.5Å.

A second solvent surface is formed at an intermolecular interface flanked by the Met-20 and β(F-G) loops that rearrange upon ligand binding (Fig. 2C). The rim at each end of the hydrophobic cavity has a negatively charged surface (Fig. 2F). There are eight D₂O molecules that have occupancies near one inside the cavity, of which two make H-bonds to the protein. Only two of the eight D₂Os are conserved in the x-ray structure. Five D₂Os are within 3.7–4.9 Å of polar protein atoms, and one D₂O lies quite far from the protein (5.8 Å). Two of the D₂Os H-bond with each other, whereas one other forms an H-bond with an H₂O. Because each of these cavities is involved in MTX binding directly (or indirectly through loop rearrangement), possibly some of the D₂Os might have been displaced upon ligand entry and relocated to their current position.

Conclusions

Our ND study provides insights into the dynamics of DHFR when bound to the anticancer drug, MTX. We observe MTX binding to the occluded and closed Met-20 loop conformations of DHFR. The occluded conformation (mon. A), which is more solvent accessible and has higher overall B factors, undergoes greater H/D exchange, and its Met-20 and the β(F-G) loop exchange more readily compared with the closed conformation (mon. B). We are able to explain these local differences between monomers in terms of H-bonding, intrusion of D₂O molecules, and molecular packing. However, the H/D exchange rates do not correlate globally to B factors, solvent accessibility, and H-bonding. Yet, some correlation is observed with respect to atom depth. We observe directly a proton at N1 of MTX, imparting it with a positive charge. This result agrees with nearly all available biochemical and complementary structural data for E. coli and L. casei DHFRs: that the MTX·D27 interaction is ionic in nature (45–51). However, it is in disagreement with the theoretical calculations of Cannon et al. (52), who proposed that the MTX N1·D27 pair involves a neutral dipole–dipole interaction. D₂O is observed within several pockets at the protein surface, including two hydrophobic cavities. A previously unidentified D₂O forms a hydrogen bond to the catalytic D27 in mon. B, stabilizing its negative charge. Nearly all of the D₂Os that possess the highest B factors are at the extremity of the primary solvent layer, 4–5 Å from the nearest protein atom. We also observe many D₂Os that form H-bonds with fully exchanged amides, possibly promoting H/D exchange.

Methods

DHFR was purified and crystallized as described (10). ND data were collected at Los Alamos Neutron Scattering Center by using the TOF Laue technique. The structure was determined by difference Fourier and refined by using SHELX (53). Methods are described in detail in Supporting Text.

Abbreviations

DHFR

dihydrofolate reductase

ND

neutron diffraction

MTX

methotrexate

mon.

monomer