The Dynamic Energy Landscape of Dihydrofolate Reductase Catalysis

See allHide authors and affiliations

Science  15 Sep 2006:
Vol. 313, Issue 5793, pp. 1638-1642
DOI: 10.1126/science.1130258


We used nuclear magnetic resonance relaxation dispersion to characterize higher energy conformational substates of Escherichia coli dihydrofolate reductase. Each intermediate in the catalytic cycle samples low-lying excited states whose conformations resemble the ground-state structures of preceding and following intermediates. Substrate and cofactor exchange occurs through these excited substates. The maximum hydride transfer and steady-state turnover rates are governed by the dynamics of transitions between ground and excited states of the intermediates. Thus, the modulation of the energy landscape by the bound ligands funnels the enzyme through its reaction cycle along a preferred kinetic path.

It has long been recognized that dynamic fluctuations in protein conformation play a central role in enzyme catalysis (13). Protein dynamics are implicated in events such as substrate or cofactor binding and product release, and the chemical event itself involves an inherently dynamic process, with changes in atomic coordinates required along the reaction coordinate (4). Although there is considerable evidence from both theory and experiment that many enzymes are inherently flexible, the fundamental mechanisms by which protein fluctuations couple to catalytic function remain poorly understood.

Escherichia coli dihydrofolate reductase (DHFR) has been used extensively as a model enzyme for investigating the relations between structure, dynamics, and function. Theoretical and experimental investigations suggest that protein fluctuations play a direct role in catalysis by DHFR [see (5) for a recent review]. The enzyme catalyzes the reduction of 7,8-dihydrofolate (DHF) to 5,6,7,8-tetrahydrofolate (THF) by using reduced nicotinamide adenine dinucleotide phosphate (NADPH) as a cofactor. The kinetic mechanism involves rebinding of NADPH to assist the release of the THF product. The enzyme (E) cycles through five major intermediates: E:NADPH, E:NADPH:DHF, E:NADP+:THF, E:THF, and E:NADPH:THF (6) (Fig. 1A). The structures of all of the kinetic intermediates, or models of the intermediates, have been determined by x-ray crystallography, and the conformational changes that occur during the catalytic cycle have been delineated (7, 8). The major sites of conformational change include the active-site loop (residues 9 to 24, termed the Met20 loop) and the substrate-binding pocket (7) (Fig. 1B). In the holoenzyme E:NADPH and the Michaelis complex E:NADPH:DHF (modeled by the ternary E:NADP+:folate complex), the Met20 loop adopts a closed conformation, where it packs against the nicotinamide ring of the cofactor bound within the active site. In the three product complexes, the Met20 loop adopts an occluded conformation, where it sterically hinders the nicotinamide ring from binding in the active site; therefore, the nicotinamide ring is outside the pocket in these complexes. The substrate-binding pocket undergoes a similar transition, closing more tightly when both substrate and cofactor are present and opening to release products (7).

Fig. 1.

Conformational changes during the DHFR catalytic cycle. (A) Met20 loop conformations for each complex in the catalytic cycle (8). The complexes are shown in distinctive colors that are used in Fig. 4. (B) Structures of E:NADP+:folate [left, Protein Data Bank (PDB) 1RX2 (7)] and E:NADPH:5,10-dideazaTHF (ddTHF) [right, PDB 1RX6 (7)], illustrating the conformational changes that occur upon hydride transfer. In E:NADP+:folate, a model for the Michaelis complex E:NADPH:DHF, the Met20 loop is in a closed conformation, and the folate-cleft width, measured between the van der Waals contact surfaces of Ile50 and Leu28, is 17.9 Å. In E:NADPH:ddTHF, a model for the product ternary complex E:NADP+:THF, the folate cleft opens by 0.8Å, and the Met20 loop is in an occluded conformation, restricting cofactor access to the active site. Residues that define the active site loop conformation, substrate-/product-binding, and cofactor-binding markers (17) are colored red, blue, and green, respectively. The bound cofactor is colored gold, and folate and ddTHF are shown in magenta. The residues for which dispersion data are shown in Fig. 2 are identified by open circles.

Protein dynamics can be evaluated experimentally by nuclear magnetic resonance (NMR) spin relaxation techniques (9). Carr-Purcell-Meiboom-Gill (CPMG)–based R2 relaxation dispersion experiments monitor motion on the μs to ms time scale that is generally the most relevant for protein conformational change (10). Through these methods, the transverse relaxation rate, R2, can be decomposed into Rex, the contribution from exchange between different conformations, and R0, all other contributions (9). For two-site chemical exchange between a ground state (A) and an excited state (B), R2 relaxation dispersion is a function of the exchange rate constant kex (kex = kA→B + kB→A), the populations of states A and B (p A and pB, respectively), and the chemical shift difference between states A and B (Δω) (11), thus giving information regarding the kinetics and thermodynamics of protein motion (1215) and providing insight into the structure of the higher energy state (13, 15, 16).

The 15N R2 relaxation dispersion measurements for the Michaelis complex model E:NADP+:folate (15) indicated that many of the residues that exhibit exchange contributions to relaxation are directly or indirectly associated with the Met20 loop (Fig. 2A). These residues show characteristic chemical shift differences between closed and occluded complexes, and their resonances have been previously categorized as active site loop conformation markers (17). Likewise, chemical shift perturbation studies identified cofactor-binding and substrate- or product-binding (substrate-/product-binding) marker resonances associated with residues that cluster around the cofactor- and substrate-binding pockets, respectively (17) (Fig. 1B). A comparison of the dynamic chemical shift differences (Δω values) determined from fits of the R2 relaxation dispersion data to the equilibrium chemical shift differences (Δδ values) between the closed complex E:NADP+:folate and the occluded complex E:DHNADPH:folate showed a remarkable linear correlation (15). Thus, the higher energy state contributing to R2 relaxation in the closed E:NADP+:folate complex represents an occluded conformation similar to that found in the E:NADP+:THF product ternary complex.

Fig. 2.

Relaxation dispersion data for each intermediate in the catalytic cycle of DHFR. (Left) Model structures for various intermediates of the DHFR catalytic cycle (7). The backbone is colored red, green, or blue as in Fig. 1B. Residues for which conformational exchange is observed are indicated with spheres, which are colored red, green, and blue for resonances that report on active site loop conformation, cofactor binding, and substrate/product binding, respectively, and gray for resonances that are not identified with any of these categories. Yellow indicates resonances that show broadening, but for which data quality was insufficient to obtain reliable R2 relaxation dispersion results. (Right) Representative 15N R2 relaxation dispersion curves for each complex. A full set of data for all residues that show detectable relaxation dispersion is included (figs. S1 to S5). Error bars indicate estimated uncertainties in R2 (15). (A) E:NADP+:folate (PDB 1RX2) and NMR data at 303 K. (B) E:NADPH (PDB 1RX1) and NMR data at 284 K. (C) E:THF (PDB 1RX5) and NMR data at 300 K. (D) E:NADP+:THF (PDB 1RX4) and NMR data at 300 K. (E) E:NADPH:THF (PDB 1RX6) and NMR data at 300 K. Relaxation dispersion data were collected and analyzed at two external magnetic field strengths (1H 500 MHz and 800 MHz), but only 800 MHz data are shown for clarity. Red curves report on the active site loop conformation marker Gly121; bluecurves, on the substrate-/product-binding marker Asp37; and green curves, on the cofactor-binding marker Ser77 [(A) to (C) and (E)] or Met42 (D). The green curve for Ser77 in (E) (green) has been offset for clarity (right axis). This figure was generated in part by using MOLMOL (27).

A complete set of 15N and 1H R2 relaxation dispersion data have now been obtained for DHFR complexes that represent all of the kinetic intermediates populated in the steady-state catalytic cycle. Dispersion data measured at two frequencies were fitted to the general two-site exchange equations; the methods, dispersion curves, and fitted parameters for all complexes are provided (tables S1 to S4 and figs. S1 to S5). Some of these data are shown in Figs. 2 and 3.

Fig. 3.

Correlation between Δω values obtained from relaxation dispersion measurements and differences (Δδ) between chemical shifts in the ground states of adjacent complexes in the catalytic cycle. (A) Δω for E:NADPH plotted against Δδ [equal to δ(E:NADPH) – δ(E:NADPH:THF), circles; or δ(E:NADPH) – δ(E:NADP+:folate), triangles] (slope = 0.95, R2 = 0.97); (B) Δω for E:THF plotted against Δδ [equal to δ(E:THF) – δ(E:NADP+:THF), circles; or δ(E:THF) – δ(E:NADPH:THF), triangles] (slope = 1.1, R2 = 0.99); (C) Δω for E:NADP+:THF plotted against Δδ [δ(E:NADP+:THF) – δ(NADP+:folate)] by using 10 mM (circles) and 50 mM (triangles) NADP+ (slope = 1.0, R2 = 0.99); and (D) Δω for E:NADP+:THF plotted against Δδ [δ(E:NADP+:THF) – δ(E:THF)] by using 10 mM (circles) or 50 mM (squares) NADP+, or plotted against Δδ [δ(E:NADP+:THF) – δ(E:folate)] by using 10 mM (triangles) or 50 mM (diamonds) NADP+ (slope = 0.91, R2 = 0.94). The data points for the Gly96 amide, which hydrogen-bonds directly to the cofactor, are enclosed in a circle and were not included when determining the line of best fit. (E) Δω for E:NADPH:THF plotted against Δδ [δ(E:NADPH:THF) – δ(E:NADPH)] (slope = 0.97, R2 = 0.98). Residues are colored red, green, and blue to indicate residues reporting on the active site loop conformation, cofactor binding, and substrate/product binding, respectively. Solid symbols indicate that the sign of Δω could be determined from a comparison of HSQC and heteronuclear multiple-quantum coherence spectra at an external magnetic field strength of 1H 500 MHz (28), and open symbols indicate residues where only the absolute values for Δω and Δδ are reported. Error bars indicate uncertainties in Δω estimated by Monte Carlo simulation (15).

Analysis of amide 15N and 1H R2 relaxation dispersion measurements for the holoenzyme E:NADPH revealed exchange processes for many residues located in or around the substrate binding site (Fig. 2B). Dispersive behavior was also observed for several residues in the active site loop and the loop (residues 116 to 132) between β strands F and G (the FG loop), but no relaxation dispersion was seen for residues in the cofactor-binding site. The localization of the residues showing exchange contributions to relaxation around the substrate-binding pocket suggests that the higher energy conformation sampled by E:NADPH plays an important role in capturing the substrate. Indeed, there is a strong linear correlation between the Δω values derived from the relaxation dispersion curves and the Δδ values derived from the chemical shift differences between E:NADPH and E:NADPH:THF, or between E:NADPH and E:NADP+:folate, representing the previous step or the next step in the cycle, respectively (Fig. 3A). This result implies that the E:NADPH complex samples a higher energy substate in which the empty substrate-/product-binding pocket adopts a conformation similar to that of the ligand-bound state. A similar observation has been reported for ribonuclease A (RNaseA): As a result of conformational fluctuations, the free enzyme samples a higher energy state whose structure resembles the ligand-bound form (18). Although many residues in the Met20 and FG loops experience exchange contributions, the derived Δω values do not correlate with the Δδ values between the closed and the occluded conformations (fig. S6); the active site loop conformation in the excited state is currently unknown.

In direct contrast to the E:NADPH complex, residues surrounding the cofactor-binding cleft display exchange contributions to relaxation in the E:THF complex (Fig. 2C). A linear correlation is observed between Δω and Δδ(E:THF – E:NADP+:THF) or Δδ(E:THF – E:NADPH:THF) values (Fig. 3B), suggesting that the higher energy conformation contributing to 15N R2 relaxation in E:THF resembles the product ternary complexes. Any relaxation dispersion observed for residues lining the substrate-binding pocket or in the FG loop can generally be traced to local differences in conformation between E:THF and product ternary complexes (table S2 and fig. S7).

In the E:NADP+:THF complex, conformational changes are observed in the active site loops and the ligand-binding pockets (Fig. 2D). The Δω values for residues surrounding the cofactor-binding cleft and the active site loops correlate with different Δδ values (Fig. 3, C and D). The Δω values for residues in the active site loops correlate to Δδ(E:NADP+:THF – E:NADP+:folate) (Fig. 3C), showing that the occluded E:NADP+:THF complex samples a higher energy state in which the active site loops are in a closed conformation, resembling the conformation of the E:DHF:NADPH Michaelis complex (modeled by E:NADP+:folate) that immediately precedes it in the catalytic cycle. For many of the residues surrounding the cofactor-binding cleft, a linear correlation is observed between Δω and Δδ(E:NADP+:THF – E:folate) and/or Δδ(E:NADP+:THF – E:THF) (Fig. 3D), revealing the presence of an additional excited state in which the conformation of the adenosine-binding site is similar to that in the binary E:THF product complex. The excited protein substates do not reflect physical dissociation of cofactor or chemical changes. The population of E:THF in equilibrium with the ternary product complex E:NADP+:THF is estimated to be 0.4% on the basis of rate constants determined from pre–steady-state analysis (6), whereas the excited state population from relaxation dispersion experiments is much larger (pB > 2.3%). Repeat experiments at fivefold higher NADP+ concentration, where the population of the E:THF complex is estimated to be ∼0.08%, showed identical R2 relaxation dispersion for the residues surrounding the adenosine-binding site (table S3), which rules out cofactor dissociation as the origin of the exchange contributions to the R2 relaxation rates. In addition, the x-ray structures (7) show that the Gly96 amide forms a hydrogen bond to the phosphate group of the cofactor, which leads to a large change in the 15N chemical shift [3.5 to 4.0 parts per million (ppm)] upon binding of NADP+ to the E:THF or E:folate complexes. However, the Δω for Gly96 is much smaller (<1.35 ppm) (Fig. 3D), implying that the hydrogen bond remains largely intact and that conformational exchange is not modulated by cofactor dissociation. The closed excited-state conformation of the active site loops also cannot be a consequence of hydride transfer, because the rate constant of the back reaction is too slow at this pH (∼0.03 s–1) for this process to contribute measurably to R2 relaxation dispersion (6).

The ground-state conformations of E:NADPH:THF and E:NADP+:THF are very similar, as evidenced by their nearly identical 15N heteronuclear single-quantum coherence (HSQC) spectra, yet the two complexes exhibit very different R2 relaxation dispersion. The E:NADPH:THF complex (Fig. 2E) exhibits more pronounced dispersive behavior for residues surrounding the substrate-/product-binding pocket and no conformational exchange for residues in the cofactor-binding cleft. The Δω values for most residues surrounding the substrate-/product-binding pocket correlate strongly with Δδ(E:NADPH:THF – E:NADPH) (Fig. 3E). Again, this cannot be due to the physical dissociation of THF from the complex, because the population of the excited-state (pB > 1.9%) is substantially greater than the population of the binary E:NADPH complex in equilibrium with E:NADPH:THF (population ∼0.12%). Moreover, repeating the experiment at a THF concentration three times higher (estimated population of E:NADPH ∼0.04%) yielded nearly identical results (table S4). Many residues associated with the active site loops also display conformational exchange, yet the derived Δω values for most of these do not correlate with an occluded-to-closed conformational change. This result provides further evidence that we are not observing physical dissociation of product THF to form the closed E:NADPH complex but are monitoring fluctuations into a higher energy conformation of E:NADPH:THF that resembles one without product bound. The higher energy substates sampled by E:NADPH and E:NADPH:THF may be similar, because Δω values for resonances showing dispersion in both the binary and the ternary complex display a linear correlation (fig. S11).

These results can be placed in the context of the catalytic cycle (Fig. 4). The higher energy conformations that we observe in the R2 relaxation dispersion experiments appear to play a direct role in catalysis. In all five intermediates, there is conformational exchange on a μs-ms time scale between the ground-state structure and one or two excited states that resemble the ground state of the preceding and/or the following intermediate in the catalytic cycle. The binary complexes E:NADPH and E:THF sample excited-state conformations that facilitate binding of substrate/product and cofactor, respectively. Thus, binding of ligands to the enzyme appears to occur by a conformational selection (19, 20) or selected-fit (21) mechanism, rather than by the induced-fit mechanism (22) thathas been traditionally invoked to explain substrate-induced conformational change. An underlying tenet of the induced-fit model is conformational homogeneity, with binding occurring by a sequential mechanism; the ligand binds to the enzyme and induces a conformational change that increases the complementarity between ligand and protein. However, most proteins are structurally heterogeneous; their energy landscapes are rugged, and a number of conformational substates lie close in energy to the ground state and are populated through thermal fluctuations (23). In the conformational selection model, a small population of a minor conformational substate resembling the ligand-bound or induced conformation is already present in solution, in a preexisting equilibrium with the major ligand-free state. Ligand binds to the minor substate, causing a shift in the equilibrium such that the ligand-bound conformation becomes the new major substate (19, 20). The experimentally determined bimolecular rate constant for binding of substrate to E:NADPH (4 × 107 M–1 s–1) (6) is consistent with a mechanism that invokes the diffusion-controlled association (∼109 M–1 s–1) of substrate with a small population (pB = 2%) of a binding-competent excited state of the E:NADPH holoenzyme.

Fig. 4.

The dynamic energy landscape of DHFR catalysis. Ground state (larger) and higher energy (smaller) structures of each intermediate in the cycle, modeled on the published x-ray structures (7), are shown color-coded according to the scheme in Fig. 1A, with NADPH and NADP+ shown in gold and substrate, product, and analogs shown in magenta. For each intermediate in the catalytic cycle, the higher energy conformations detected in the relaxation dispersion experiments resemble the ground-state conformations of adjacent intermediates; their interconversion rates, also obtained from the relaxation dispersion experiments, are shown with black arrows. Rate constants for the interconversion between the complexes, measured by pre–steady-state enzyme kinetics at 298 K, pH = 6 (6) are indicated with red arrows. R2 relaxation dispersion measurements were made at pH = 6.8 (E:NADP+:folate) or pH = 7.6 (E:NADPH:THF, E:NADP+:THF, E:NADPH, and E:THF) at 281 K (E:NADPH), 300 K (E:NADPH:THF, E:NADP+:THF, and E:THF), or 303 K (E:NADP+:folate).

Our results suggest that ligand release also occurs through higher energy substates. The excited state structures of E:NADP+:THF and E:NADPH:THF resemble conformations in which the cofactor- or product-binding pocket is empty, even though ligand dissociation has not occurred. Fluctuations that populate these higher energy substates effectively prepare the enzyme for ligand dissociation; this process can be viewed as the opposite to conformational selection and ligand binding. Indeed, Δω values for amides in the substrate-/product- and cofactor-binding pockets of the complementary binary and ternary complexes E:NADPH/E:NADPH:THF and E:(THF or folate)/E:NADP+:THF, respectively, are correlated (fig. S11), suggesting that exchange contributions to relaxation arise from similar, but opposing, processes.

Transitions between the conformational substates occur at rates, determined from the R2 relaxation dispersion experiments, that are directly relevant to DHFR catalysis. Comparison with the rate constants determined from pre–steady-state kinetics (6) provides strong evidence that the rate of progression through the various steps of the reaction cycle is governed by the dynamics of the conformational fluctuations between the ground and the excited states of the kinetic intermediates (Fig. 4). Thus, the first-order rate constant for release of THF from the E:NADPH:THF complex (12.5 s–1 at 298 K), which is the rate-determining step at physiological pH, is very similar to the ground-to-excited state conformational exchange rate constant (12 to 18 s–1 at 300 K) that we measure for the residues surrounding the substrate-/product-binding pocket. This argues strongly that product dissociation occurs from the excited state. Maximum substrate turnover can also be rationalized in the context of this model. Subsequent to hydride transfer, which is effectively instantaneous relative to the rate of protein conformational change (24), the enzyme is converted from the closed E:NADPH:DHF Michaelis complex to a closed E:NADP+:THF complex. The kinetic rate constant for the conformational change from the higher energy closed state to the occluded ground state of E:NADP+:THF (k ∼1200 s–1 at 300 K) is very similar to the pH-independent rate constant for hydride transfer (khyd = 950 s–1 at 298 K) (6). Thus, both the product release and the chemical transformation rate constants are largely determined by the exchange rate constants between substates that are thermally populated within the conformational ensemble; that is, the maximum hydride transfer rate and the steady-state turnover rate are dictated by physical changes within the energy landscape of the enzyme. A correlation between the overall turnover rate and protein motions has also been described for the enzyme cyclophilin (25).

Because R2 relaxation dispersion experiments can generally only characterize higher energy conformations that make up at least 1 to 2% of the ensemble, there may be additional excited states that are inaccessible to the technique. However, the excited-state conformations that we observe, together with the ground-state conformation, will constitute the lowest-energy members of the conformational ensemble of each intermediate. These results imply that the most functionally relevant conformations also possess the lowest energy of all potential conformations. In this view, ligands dictate not only the ground-state conformation but also the most accessible higher energy substates. As ligands change, through binding or dissociation processes or through chemistry, the energy landscape and the populations of the accessible states change in response. Thus, the dynamic energy landscape (26) efficiently funnels the enzyme through its catalytically competent conformations along a preferred kinetic path, where the number and heights of the energetic barriers between consecutive conformations have been minimized.

Supporting Online Material

Materials and Methods

Figs. S1 to S11

Tables S1 to S4

References and Notes

View Abstract

Navigate This Article