Organometallic mechanism of action and inhibition of the 4Fe-4S isoprenoid biosynthesis protein GcpE (IspG)

Abstract

We report the results of a series of chemical, EPR, ENDOR, and HYSCORE spectroscopic investigations of the mechanism of action (and inhibition) of GcpE, E-1-hydroxy-2-methyl-but-2-enyl-4-diphosphate (HMBPP) synthase, also known as IspG, an Fe₄S₄ cluster-containing protein. We find that the epoxide of HMBPP when reduced by GcpE generates the same transient EPR species as observed on addition of the substrate, 2-C-methyl-D-erythritol-2, 4-cyclo-diphosphate. ENDOR and HYSCORE spectra of these transient species (using ²H, ¹³C and ¹⁷O labeled samples) indicate formation of an Fe-C-H containing organometallic intermediate, most likely a ferraoxetane. This is then rapidly reduced to a ferracyclopropane in which the HMBPP product forms an η²-alkenyl π- (or π/σ) complex with the 4th Fe in the Fe₄S₄ cluster, and a similar “metallacycle” also forms between isopentenyl diphosphate (IPP) and GcpE. Based on this metallacycle concept, we show that an alkyne (propargyl) diphosphate is a good (K_i ∼ 300 nM) GcpE inhibitor, and supported again by EPR and ENDOR results (a ¹³C hyperfine coupling of ∼7 MHz), as well as literature precedent, we propose that the alkyne forms another π/σ metallacycle, an η²-alkynyl, or ferracyclopropene. Overall, the results are of broad general interest because they provide new mechanistic insights into GcpE catalysis and inhibition, with organometallic bond formation playing, in both cases, a key role.

Most pathogenic bacteria, plants, as well as malaria parasites (Plasmodium spp.), in contrast to humans, use the Rohmer, nonmevalonate or methyl erythritol phosphate pathway to produce isoprenoids (1, 2), so the development of inhibitors of this pathway is of interest in the context of drug (and herbicide) discovery. The structures of most of the enzymes in the pathway are now known from X-ray crystallography, but the structure of the penultimate enzyme, GcpE: E-1-hydroxy-2-methyl-but-2-enyl-4-diphosphate (1, HMBPP) synthase, EC 1.17.7.1, also known as IspG, has not yet been reported. Its mechanism of action is thus not well understood, and there is only one inhibitor (with an IC₅₀ of ∼1.6 mM) (3). GcpE enzymes catalyze the 2H⁺/2e^- reduction of 2-C-methyl-D-erythritol-2,4-cyclo-diphosphate (2, MEcPP) to HMBPP: All GcpEs contain three highly conserved Cys residues that are essential for catalysis and are thought to bind to an iron-sulfur cluster (4, 5). This cluster is, based on the results of Mössbauer (4) and EPR (5) spectroscopy, thought to have a 4Fe-4S composition. There have been several catalytic mechanisms proposed for GcpE. In one, Kollas et al. (6) proposed ring-opening of the cyclo-diphosphate to form a carbocation, followed by reduction to a radical, which then underwent reduction and dehydration to form the product, HMBPP (Fig. S1A). In a second mechanism, Seemann et al. (7) proposed a similar route, but with subsequent formation of a cation radical (Fig. S1B). In a third mechanism, Brandt et al. (8) proposed a cation → radical → anion mechanism (Fig. S1C). And in a fourth mechanism, Rohdich et al. (9) proposed that MEcPP underwent an OH^--assisted ring opening/ring closing to produce an oxirane 3: that was then reduced to the alkene, 1, via radical intermediates (Fig. S1D). The possible importance of an oxirane intermediate has also been described in recent work by Nyland et al. (10). There are, therefore, numerous mechanistic possibilities that have been proposed, and to try to clarify the GcpE mechanism, we report here the result of a series of spectroscopic observations on the structure of the intermediate reported by Adedeji et al. (5) that forms on addition of 2 to GcpE. This leads to a new GcpE mechanism as well as the discovery of a potent GcpE inhibitor.

Results and Discussion

EPR, ENDOR, and HYSCORE: Clues for Catalysis.

We show in Fig. 1A the 9 GHz EPR spectrum of reduced IspG and in Fig. 1B the spectrum of the transient species that we shall call “X,” formed in the presence of MEcPP. This spectrum could, in principle, arise from bound MEcPP, from a bound epoxide, from bound HMBPP, or from another reactive intermediate. To help distinguish between these possibilities, we prepared HMBPP-epoxide (a mixture of the 2R,3R and 2S,3S epoxides, formed by treating the bromohydrin of HMBPP with NH₃) and investigated its effect on the GcpE EPR spectrum. Remarkably, we find that the same EPR intermediate (X) as that formed on addition of MEcPP (Fig. 1B) forms on addition of HMBPP-epoxide to Escherichia coli GcpE (Fig. 1C), and the same results are obtained with Thermus thermophilus GcpE as well (Fig. S2). This argues against the bound MEcPP possibility for the reactive intermediate, X. In addition, the observation that the spectrum of MEcPP + GcpE (at long incubation times) is identical to that found on HMBPP addition, with both E. coli (Fig. 1 D and E) and T. thermophilus GcpE (Fig. S3), rules out an HMBPP adduct as the origin of species X. The transient must, therefore, be either a bound epoxide, or another, as yet unknown, intermediate.

Fig. 1.

X-band EPR spectra of GcpE ± reactants/products. (A) EPR of E. coli GcpE reduced with 20 equivalents Na₂S₂O₄. (B) EPR of E. coli GcpE + MEcPP, incubated for 1 min. (C) EPR of E. coli GcpE + HMBPP epoxide, incubated for 2 min. (D) EPR of E. coli GcpE + MEcPP, incubated for 45 min. (E) EPR of E. coli GcpE + HMBPP, incubated for 45 min. Frequency = 9.05 GHz, microwave power = 1 mW for A, 0.05 mW for B–E, temperature = 15 K.

To explore this question in more detail, we next obtained the ¹H ENDOR spectra of X from unlabeled MEcPP. As shown in Fig. 2A, a ¹H ENDOR signal with a large hyperfine coupling (A ∼ 11.5 MHz) is observed (the smaller central couplings are due to the Cys-beta protons). This ¹H ENDOR signal does not decrease on exchange (four times, Fig. 2B) but is absent when [u-²H]-MEcPP is used, Fig. 2C, so originates from the ligand. In this deuterated sample, we also find two sets of ²H ENDOR signals in the low frequency region, Fig. 2D. The first has a large hyperfine coupling (A ∼ 1.8 MHz) with a small quadrupole splitting and corresponds to the 11.5 MHz feature found in the ¹H ENDOR spectrum (Fig. 2A). The second set has a smaller coupling (A ∼ 0.5 MHz) and arises from a weaker or long-range interaction with another deuteron in the ligand. We also find that the ¹H ENDOR spectrum of the intermediate X obtained by adding HMBPP epoxide (Fig. 2E) is very similar to that found with the MEcPP intermediate X (Fig. 2A).

Fig. 2.

X-band ENDOR of T. thermophilus GcpE + MEcPP/HMBPP-epoxide reaction intermediate X. (A) ¹H Davies ENDOR of GcpE + MEcPP. (B) ¹H Davies ENDOR of GcpE + MEcPP, in D₂O buffer. (C) ¹H Davies ENDOR of GcpE + [u-²H]-MEcPP. (D) ²H Mims ENDOR of GcpE + [u-²H]-MEcPP, difference spectrum (²H-labeled—unlabeled). (E) ¹H Davies ENDOR of GcpE + HMBPP epoxide. (F) ¹³C Mims ENDOR of GcpE + [u-¹³C]-MEcPP, difference spectrum (¹³C-labeled—unlabeled). The inset is the unsubtracted ENDOR spectrum of the labeled sample showing the ³¹P feature. Frequency = 9.66 GHz; spectra were collected at g₂ (B₀ = 342.2 mT) at 20 K. τ-averaging was used for collecting Mims ENDOR spectra as follows: D , 10 spectra at 8 ns steps with an initial τ = 248 ns; F, 12 spectra at 8 ns steps with an initial τ = 248 ns.

The ¹³C ENDOR spectrum of the reaction intermediate X obtained from [u-¹³C]-MEcPP (Fig. 2F) shows several sets of peaks: One displays a small hyperfine coupling (, A ∼ 0.84 MHz); two sets have medium couplings (, A ∼ 2.3 MHz; , A ∼ 2.3 MHz); while one weak peak (at 6.3 MHz) could originate from the low frequency part of a much larger coupling (, A ∼ 19 MHz). The latter would have |A/2| > ν_L so would be centered at |A/2|, with two peaks separated by 2ν_L (7.3 MHz at 342.2 mT), in which case the high frequency peak should appear at 13.6 MHz and would be obscured by the very strong ¹H peaks.

To help confirm these observations, we next obtained HYSCORE (hyperfine sublevel correlation) spectra (Fig. 3 A–C; expanded spectra are shown in Fig. S4). HYSCORE is a two dimensional form of pulsed EPR spectroscopy (11). In the (+/+) quadrant (on the right) of a HYSCORE spectrum, peaks due to weak interactions (|A/2| < ν_L) are on the antidiagonal, while strong hyperfine interactions (|A/2| > ν_L) are in the (+/-) quadrant (on the left), again on the antidiagonal. With unlabeled MEcPP, we see primarily the peak due to ¹⁴N (most likely from protein backbone nitrogens (12), together with a natural abundance ¹³C background (Fig. 3A), but with a ∼30% ¹³C-enriched [2,3-¹³C]-MEcPP (prepared biosynthetically from [2-¹³C]-glucose and randomly ¹³C-enriched only at C2,3) (13), there are now two additional sets of peaks: at (2.29, 5.14; 5.14, 2.29 MHz) in the (+/+) quadrant, and at (-4, 11; -11, 4 MHz) in the (+/-) quadrant, Fig. 3B. The former correspond to the peaks with medium couplings seen in ENDOR ( or in Fig. 2F), while the latter correspond to the ¹³C ENDOR peak with the large hyperfine coupling ( in Fig. 2F). With a [u-¹³C]-MEcPP labeled sample (having ∼100% ¹³C), the signals are much stronger (Fig. 3C), and we now see that there is a third set of signals having a very small coupling (corresponding to seen in the ENDOR spectrum, Fig. 2F), plus, the cross-peaks with medium couplings in the (+/+) quadrant and the cross-peaks with the large coupling in the (+/-) quadrant are more intense. These results further narrow down the possible structures for the reactive intermediate X. Specifically, the GcpE-bound epoxide 4 is unlikely because nearly equal hyperfine couplings would be expected for C2 and C3, in sharp contrast to the large difference seen experimentally. But can we make other suggestions as to the nature of X, based on these spectroscopic results?

Fig. 3.

HYSCORE spectra and simulation for T. thermophilus GcpE + MEcPP/HMBPP-epoxide reaction intermediate X. (A) GcpE + unlabeled MEcPP. (B) As A but with MEcPP enriched with (∼30%) ¹³C at C2 and/or C3. (C) As A but with [u-¹³C]-HMBPP, 100% ¹³C-labeled. The C1,2’ and 4 assignments are tentative. (D) Simulated ¹³C spectrum with: C3 A_ii = [13,13,20] MHz; C2,4 A_ii = [2.3,2.3,3.8] MHz; and C(1 or 2’) A_ii = [0.25,0.25,0.7] MHz. (E) GcpE + [2,3-¹⁷O]-HMBPP epoxide. Frequency = 9.64 GHz, spectra were collected at ∼g₂ (B₀ = 341.5 mT) at 18 K with τ = 136 ns. Expansions of the experimental spectra are shown in Fig. S4.

Organometallic Intermediates in GcpE Catalysis.

The large (∼16 MHz, Fig. 3D) ¹³C hyperfine coupling observed in the [2,3-¹³C]- and [u-¹³C]-MEcPP samples suggested to us that X might contain an Fe-C bond in which the directly bonded carbon originates from C2 or C3 of MEcPP. This, when taken together with the 11.5 MHz ¹H ENDOR result, suggests that the strongly coupled ¹H is most likely directly bonded to the carbon that also has the large ¹³C hyperfine coupling. The only proton in MEcPP that satisfies this constraint is H3, which then leads to two new candidates for X: the η¹-alkyl complex 5 and the ferraoxetane 6. The difference between these two models is that in one case there is an Fe-O bond, while in the other, this bond is absent. To determine whether Fe-O bonding is present, we obtained the HYSCORE spectrum of X prepared from [2,3-¹⁷O]-HMBPP epoxide (Fig. 3E). Features due to ¹⁷O hyperfine/quadrupole interactions are clearly seen in the (+/-) quadrant, favoring the ferraoxetane 6 as the putative reaction intermediate X, although at present it is not certain whether it forms from 2 and 3 in a consecutive or parallel manner.

The possibility that the ferraoxetane 6 is a reaction intermediate is intriguing because many metallaoxatanes are known as stable species (14, 15), and in the case of Fe interacting with oxirane itself, the 1,2-ferraoxetane has been observed using matrix isolation (16). This species is more stable than is Fe + oxirane (17), and on warming, the ferraoxetane undergoes a [2 + 2] dissociation to ethylene and FeO (16). The involvement of more complex metallaoxetanes in epoxide deoxygenation was proposed early on by Sharpless (18) (and would be essentially the opposite reaction to Sharpless epoxidation), and such species might be involved in oxirane deoxygenation by GcpE as well as by model Fe₄S₄ clusters (19). If the ferraoxetane does represent a reactive intermediate, a possible overall reaction mechanism is as shown at the top of the following Scheme, in which 6 is converted to the π/σ metallacycle 7 and product 1: Initial formation of isomer 6 over the alternative species, 8, seems likely, because as discussed earlier, the large ¹H hyperfine coupling is consistent with a 2-bond rather than a 3-bond interaction of the iron-sulfur cluster with H3. Ring formation might also proceed via radical intermediates, but the EPR and ENDOR results we do see provide no evidence for any stable, carbon-based radicals, which would have g-values at about the free electron g-value. But is there any evidence that the HMBPP metallacycle 7 actually forms? And if it does, can this help us find new inhibitors?

HMBPP (and IPP) Form Metallacycles with GcpE.

We next consider the question as to whether the HMBPP product does in fact bind to the Fe₄S₄ cluster in GcpE. Because HMBPP is, chemically, just a substituted ethylene or allyl alcohol, and because ethylene and allyl alcohol are known to form π (or π/σ) η²-alkenyl “metallacycle” complexes with a nitrogenase FeMo cofactor (which is thought to contain a Fe₃MoS₃X cubane-like structure) (20, 21), we reasoned that HMBPP bound to GcpE might also form a π (or π/σ) complex with the unique fourth Fe in the Fe₄S₄ cluster. As noted above, we find that there are very large changes in the EPR spectra of GcpE on addition of HMBPP (or MEcPP, at ∼45 min) (Fig. 1 A, D, and E), indicating major changes in the cluster’s electronic structure. To test the metallacycle hypothesis further, we obtained the ENDOR spectrum of [u-¹³C]-HMBPP bound to T. thermophilus GcpE, Fig. 4 A and B. As can be seen in Fig. 4A, there are ENDOR resonances due to ¹³C hyperfine coupling (Fig. 4A, A ∼ 1.4 MHz), together with a small ³¹P coupling (Fig. 4B, A ∼ 0.3 MHz), consistent with the idea that HMBPP is bound to GcpE, forming a π (or π/σ) η²-alkenyl metallacycle, as illustrated schematically in Fig. 4C. This type of interaction also occurs with isopentenyl diphosphate (IPP), because as shown in Fig. 4D, a narrow line spectrum (with g = 2.065, 1.995 and 1.975) is obtained for GcpE (from T. thermophilus) in the presence of IPP, and ¹³C and ³¹P ENDOR signals are found on binding of -IPP, Fig. 4 E and F. So, HMBPP and IPP, both of which contain alkene groups, interact with the reduced Fe₄S₄ cluster in GcpE, forming, we propose, η²-alkenyl π (or π/σ) metallacycles, basically the same type of structure as seen with ethylene or allyl alcohol (the HMBPP “parent” molecules) bound to the nitrogenase FeMo cofactor. Notably, this type of π-complex formation is also found with reduced LytB (22), the last enzyme in the nonmevalonate pathway. And because there is no 1-OH group in IPP, alkoxide binding as in LytB (23) is not essential for interaction with the Fe₄S₄ cluster.

Fig. 4.

X-band EPR and ENDOR spectra of GcpE (and IspH) + HMBPP/IPP. (A) ¹³C ENDOR of GcpE + [u-¹³C]-HMBPP, difference spectrum (¹³C-labeled—unlabeled). The bottom inset is the A. aeolicus IspH E126A mutant + [u-¹³C]-HMBPP (from ref. 22). (B) ³¹P ENDOR of GcpE + [u-¹³C]-HMBPP. (C) Schematic illustration of HMBPP bound to an Fe₄S₄ cluster illustrating metallacycle formation. (D) EPR of GcpE + IPP. (E) ¹³C ENDOR of GcpE + [4-¹³C]-IPP, difference spectrum (¹³C-labeled–unlabeled). (F) ³¹P ENDOR of GcpE + [4-¹³C]-IPP. EPR frequency = 9.05 GHz; ENDOR frequency = 9.66 GHz; B₀ was selected as the field where the maximum EPR signal intensity was obtained: A and B, 347.9 mT; E and F, 344.4 mT. τ-averaging (64 spectra at 8 ns step with initial τ = 200 ns) were used for collecting the Mims ENDOR spectra.

Discovery of Potent GcpE Inhibition Involving a Metallacycle (π or π/σ) Complex.

The results described above, in which we find evidence for organometallic (π or π/σ) complexes, led to new ideas for GcpE inhibitors. In their work on alkyne reductions by model Fe₄S₄ clusters, Itoh et al. (24) discovered that diphenylacetylene was reduced to cis-stilbene. This is reminiscent of the reduction of acetylene to ethylene catalyzed by [Fe₄S₄(SPh)₄]^2-/3-/HOAc/Ac₂O reported by McMillan et al. (25), where it was suggested that the reaction involved formation of a π or π/σ complex in which acetylene binds to one of the Fe atoms, with species (resonance hybrids) such as 9a, 9b being formed. Direct evidence for π complex formation between an alkyne and an Fe₄S₄ cluster has been observed by Tanaka et al. (26), who found that the C ≡ C vibrational Raman frequency of acetylene decreased by ≈60 cm^-1 when bound to [Fe₄S₄(SPh)₄]^3- and was accompanied by the loss of one PhS^- ligand (as determined by UV-VIS spectroscopy). Taken together, this literature as well as our ENDOR results suggested to us that because alkenes and alkynes can bind to Fe₄S₄ clusters, species such as 10, propargyl diphosphate, might be GcpE inhibitors, forming η²-alkynyl complexes, just as they do with LytB (22).

The EPR spectrum of propargyl diphosphate 10 bound to E. coli GcpE exhibited a narrow line spectrum (Fig. 5A) clearly distinct from that observed in the absence of the ligand (Fig. 1A). A similar result was obtained for T. thermophilus GcpE + 10 (Fig. 5B). There is, therefore, a major change in the electronic structures of both clusters on binding this acetylenic compound, and the results of an ENDOR experiment using -10 (Fig. 5C) indicates a large (A ∼ 7 MHz) ¹³C hyperfine interaction. As with HMBPP and IPP binding, these results suggest formation of a π (or π/σ) complex, a ferracyclopropene. Notably, the ¹³C hyperfine couplings in the alkyne are much larger than those found in any of the alkene complexes (∼1–3.7 MHz) (20, 22), suggesting stronger binding with the alkyne. This could be due solely to the fact that alkynes are better donors/acceptors than are alkenes, but in this particular system it might also be due to the presence of “resonance” forms that could stabilize alkyne bonding—not dissimilar to the presence of resonance in, e.g., the cyclopropenyl cation [which like some metallacycles (27), is aromatic]. Consistent with the stronger binding affinity suggested by these spectroscopic results, we find that propargyl diphosphate (10) is a competitive GcpE inhibitor with an IC₅₀ ∼ 750 nM (K_i ∼ 330 nM) (Fig. 5D) binding we propose as the π (or π/σ) complex (Fig. 5E) and is ∼1,000× more potent than previously reported GcpE inhibitors (3).

Fig. 5.

Inhibition of GcpE by an alkynyl diphosphate 10 also involves metallacycle formation. (A) 9.05 GHz EPR of E. coli GcpE + 10 equivalents propargyl diphosphate 10 and 20 equivalents sodium dithionite. (B) 9.05 GHz EPR of T. thermophilus GcpE + 10 equivalents propargyl diphosphate 10 and 20 equivalents sodium dithionite. (C) ENDOR of T. thermophilus GcpE + [u-¹³C] propargyl diphosphate 10. The inset is the ENDOR spectrum of GcpE + unlabeled 10, showing only the ³¹P signals. Frequency = 9.66 GHz; Spectra were collected at 15 K at B₀ = 342.4 mT, where the maximum EPR signal intensity was obtained. τ-averaging (64 spectra at 8 ns steps with initial τ = 200 ns) was used. (D) E. coli GcpE inhibition by propargyl diphosphate, IC₅₀ = 750 nM (K_i ∼ 330 nM). (E) Schematic illustration (based on LytB + 10 docking calculation) (22) of how propargyl diphosphate might bind to GcpE, forming an η²-alkynyl complex.

Conclusions

The results we have described above are of interest for several reasons. First, we find that the EPR spectrum of the reactive intermediate X formed on MEcPP addition is the same as that found on HMBPP-epoxide addition, indicating that both MEcPP and HMBPP-epoxide form the same intermediate. Second, we propose a tentative structure for this intermediate: an Fe-C-H containing (organometallic) species, most likely a ferraoxetane, based on the results of ¹H, ²H, ¹³C, and ¹⁷O ENDOR and HYSCORE experiments. Third, we propose that HMBPP, as well as IPP, form π (or π/σ) metallacycle complexes, ferracyclopropanes, based on the EPR and ENDOR results and precedent. Fourth, we find that an alkynyl diphosphate, propargyl diphosphate, is a good (IC₅₀ = 750 nM) GcpE inhibitor lead that also forms a metallacycle (a π or π/σ, η²-alkynyl) complex with the Fe₄S₄ cluster, based on the large hyperfine coupling seen in its ENDOR spectrum and on literature precedent for acetylene/Fe₄S₄ cluster interactions in model systems. These results lead to the idea that, in the future, it may be possible to develop related compounds as novel drugs targeting isoprenoid biosynthesis and that organometallic complex formation may play a role in reactions catalyzed by other Fe₄S₄-containing proteins containing “unique,” fourth Fe atoms.

Methods

Pulsed ENDOR/HYSCORE spectra were obtained on a Bruker ElexSys E-580-10 FT-EPR X-band EPR spectrometer using a Bruker RF amplifier (150 W, 100 kHz–250 MHz, for pulsed ENDOR experiments) and an Oxford Instruments CF935 cryostat. Mims pulsed ENDOR used a three-pulse sequence (π/2_mw-τ-π/2_mw - T - π/2_mw-τ-echo; π/2_mw = 16 ns, with π_RF applied during T). τ-averaging was used to reduce the blind spots that arise from the τ-dependent oscillations. Davies pulsed ENDOR used a three-pulse sequence (; π/2_mw = 48 ns, with π_RF applied during T). HYSCORE used a four-pulse sequence (π/2_mw - τ-π/2_mw - t₁ - π_mw - t₂ - π/2_mw - echo; π/2_mw = 16 ns), 256 points for both t₁ and t₂, each at 20 ns steps. Time-domain data were baseline corrected using a third order polynomial, then Hamming windowed, followed by zero-filling and 2D-Fourier transformation. The HYSCORE spectrum was simulated using the EasySpin program package (28).

Additional details on protein purification and reconstitution, enzyme assays, EPR/ENDOR/HYSCORE sample preparation, and compound syntheses are reported in SI Methods.