Definition of the intermediates and mechanism of the anticancer drug bleomycin using nuclear resonance vibrational spectroscopy and related methods

Abstract

Bleomycin (BLM) is a glycopeptide anticancer drug capable of effecting single- and double-strand DNA cleavage. The last detectable intermediate prior to DNA cleavage is a low spin Fe^III peroxy level species, termed activated bleomycin (ABLM). DNA strand scission is initiated through the abstraction of the C-4′ hydrogen atom of the deoxyribose sugar unit. Nuclear resonance vibrational spectroscopy (NRVS) aided by extended X-ray absorption fine structure spectroscopy and density functional theory (DFT) calculations are applied to define the natures of Fe^IIIBLM and ABLM as (BLM)Fe^III─OH and (BLM)Fe^III(η¹─OOH) species, respectively. The NRVS spectra of Fe^IIIBLM and ABLM are strikingly different because in ABLM the δFe─O─O bending mode mixes with, and energetically splits, the doubly degenerate, intense O─Fe─N_ax transaxial bends. DFT calculations of the reaction of ABLM with DNA, based on the species defined by the NRVS data, show that the direct H-atom abstraction by ABLM is thermodynamically favored over other proposed reaction pathways.

The glycopeptide antibiotic bleomycin (BLM) exhibits high intrinsic anticancer cytotoxicity, and is used in chemotherapy against head, neck, and testicular cancers as well as Hodgkin lymphoma (1, 2). Its cytotoxicity is due to its ability to effect single- and double-strand cleavage of DNA with a number of reduced metal ions and dioxygen (3–5); Fe^II displays the highest in vivo activity (6). BLM provides five nitrogen-based ligands (Fig. 1 A and B) for coordination to the metal ion based on a crystal structure of peroxy Co^IIIBLM (7). The four equatorial ligands are an imidazole nitrogen, a deprotonated amide, a pyrimidine nitrogen, and the secondary amine of the β-aminoalanine. The axial ligand is the primary amine of the β-aminoalanine.

Fig. 1.

(A) Structure of bleomycin (metal ligating atoms are in bold font and underscored); (B) Crystal structure of HOO-Co^III-Bleomycin B2. The boxed DNA binding domain (DBD) and carbohydrate domain (CHD) in A are not shown for clarity. The axial OOH^- ligand was not resolved in the crystal structure. Reproduced from PDB 2R2S (7). Cobalt is in yellow, nitrogen atoms are in blue, carbon atoms are in gray, and oxygen atoms are in red.

Activated BLM (ABLM) is the final intermediate detected before DNA double-strand scission (8). ABLM is proposed to be a low spin (ls) Fe^III─OOH species (9–11), that can be formed via the reaction of Fe^IIIBLM with O₂ and an electron or through a shunt reaction between Fe^IIIBLM and H₂O₂ (8, 12).

A prominent issue has been whether ABLM, a nonheme iron complex, performs heme type chemistry similar to P450 and peroxidase in which heterolytic cleavage of the O─O bond results in an Fe^IV species with a radical on the BLM ligand. Experiments and calculations have argued against this heterolytic mechanism and suggested a mechanism where ABLM reacts directly in H-atom abstraction from the DNA backbone sugar (10, 13, 14). However, a recent theoretical study has postulated that ABLM formation involves reaction of the (BLM)Fe^III─OH₂ complex with H₂O₂ to form a (BLM)Fe^III─O₂H₂ complex (15). This study found the heterolytic cleavage pathway for the (BLM)Fe^III─H₂O₂ complex to be thermoneutral and therefore feasible. Because ABLM is inaccessible to vibrational characterization by resonance Raman (rR) spectroscopy due to photodecay (16), we employed nuclear resonance vibrational spectroscopy (NRVS) (17–20) to obtain vibrational information and define the axial ligands of Fe^IIIBLM and ABLM, thus distinguishing between possible reaction mechanisms.

NRVS is a relatively new synchrotron-based vibrational technique (21, 22), which detects vibrational modes in the inelastic tail of the Mössbauer nuclear transition. NRVS involves selective enhancement of all the vibrational modes involving displacement of the Mössbauer active nucleus (i.e., ⁵⁷Fe) and is not complicated by issues such as photodecay or the need for an intense absorption transition for rR enhancement.

In order to assist with the analysis of these NRVS data, extended X-ray absorption fine structure (EXAFS) spectra of Fe^IIIBLM and ABLM were recollected by using a third generation synchrotron light source thereby achieving higher resolution data (9) over a greater k range for more accurate information about bond distances and the coordination environment of BLM. The combined NRVS and EXAFS data are correlated with DFT calculations of possible structure models to provide clear assignments of the identity of the axial ligands in Fe^IIIBLM and ABLM. Calculations supported by the data are then used to distinguish among the proposed reaction mechanisms.

Results

Fe^IIIBLM.

NRVS.

The partial vibrational density of states (PVDOS) spectra of Fe^IIIBLM prepared in H₂¹⁶O and H₂¹⁸O are shown in Fig. 2A. An isotope sensitive peak is observed at 567 cm^-1, which shifts down to ∼553 cm^-1 with ¹⁸O substitution. Based on the 14 cm^-1 shift, this peak is assigned as the Fe─O stretch of a water-derived ligand consistent with previous rR results (23). One additional intense feature is observed at ∼400 cm^-1, which is isotope insensitive within the 8 cm^-1 resolution of the NRVS experiment, and will be assigned in the DFT calculations section below.

Fig. 2.

(A) NRVS PVDOS spectra for Fe^IIIBLM with ¹⁶O (green) and ¹⁸O (red); (B) Overlay of experimental PVDOS spectrum of ¹⁶O Fe^IIIBLM and the DFT predicted PVDOS spectrum of OH-86-syn-α with ΔFe (Fe displacement in each mode) labeled as blue bars.

Fe^IIIBLM. EXAFS.

The EXAFS spectrum of Fe^IIIBLM has a well defined beat pattern out to k = 15 Å^-1 which gives rise to three main features in the Fourier Transform (FT) (SI Appendix: Fig. S4). The most intense peak arises from the first coordination sphere and has the dominant contribution to the overall EXAFS wave, while the latter two features are a result of the second and outer shell atoms. A small shoulder is also present to low R + Δ (nonphase shift corrected FT) on the main feature indicating that the first shell gives rise to at least two distinct EXAFS waves, indicating the presence of a shorter M-L distance.

Using the model for Fe^IIIBLM in SI Appendix: Fig. S5 (truncated version of Fig. 1A), the first shell EXAFS data were fit by varying the coordination numbers and floating the distances and mean square displacement (σ²) factors. When only a single wave was used to model the first shell, the EXAFS signal is severely underfit. The fit was greatly improved by splitting the first shell, accounting for the majority of the main EXAFS wave (SI Appendix: Fig. S4, fit 1 vs. 2).

Several fits were tested including coordination numbers of 1∶5, 2∶4, and 3∶3 for the split first shell as well as a three-wave first shell fit allowing for the possibility of a longer Fe-X bond. While the data were able to accommodate several coordination environments through the interplay of the σ² factors, the best fit (first shell only) is fit 2 (SI Appendix: Table S1), which employs a 1∶5 split first shell, with a short Fe─O distance of 1.86 Å.

Fe^IIIBLM. DFT calculations.

In order to evaluate the protonation state of the water-derived axial ligand, a variety of (BLM)Fe^III─OH₂ and (BLM)Fe^III─OH models were optimized based on three structural perturbations (orientation of axial ligand, orientation of the amide in methylvalerate, and two conformations of the axial chelate, see Computational Methods) and tested against the observed experimental data. Five (BLM)Fe^III─OH₂ and seven (BLM)Fe^III─OH structures were optimized. A summary of the energetics, Fe─O bond lengths, and Fe─O stretching frequencies can be found in SI Appendix: Table S2. All isomers having a free energy within 10 kcal/mol of the most stable structure were considered as possible structures. This conservative energy criterion was used previously (24). In the DFT computed PVDOS NRVS spectra for all twelve models, intense features are predicted at ∼400 cm^-1 that are not affected by the three structural perturbations. The peak in the 450–650 cm^-1 region of the DFT predicted PVDOS NRVS spectra (SI Appendix: Fig. S2) corresponds to the Fe─O stretch. There are five (BLM)Fe^III─OH₂ models falling within 10 kcal/mol of the lowest energy structure (SI Appendix: Table S2). The Fe─OH₂ stretches and bond lengths of these five models are found in the ranges of 474–553 cm^-1 and 1.919–1.981 Å. There are three (BLM)Fe^III─OH models falling within the 10 kcal/mol criteria (SI Appendix: Table S2). The Fe─OH stretches and bond lengths of these three models are in the ranges of 596–615 cm^-1 and 1.839–1.841 Å. Therefore, the Fe─O bond lengths of (BLM)Fe^III─OH models better match the EXAFS value of 1.86 Å (experimental error of ± 0.02 Å). The DFT calculated Fe─O stretches of the three (BLM)Fe^III─OH models are higher than the NRVS value of 567 cm^-1 (υ₁, in Fig. 2B), and the DFT calculated Fe─O stretches of the five (BLM)Fe^III─OH₂ models are lower than the NRVS value. However, by applying a 0.903 scale factor, found for heme and nonheme Fe^IV = O and Fe^III─OH species (25), the Fe─O stretches of the (BLM)Fe^III─OH better match the experiment value. Based on correlating the DFT calculations to experimental data, the axial ligand of Fe^IIIBLM can be assigned as hydroxide and model OH-88-syn-α (the Fe─O─H plane is 88° relative to the Fe-deprotonated amide bond, the amide in methylvalerate has syn orientation, and the axial chelate has α conformation, see Computational Methods) is used below to assign the NRVS spectrum.

From a correlation of the DFT calculated NRVS PVDOS spectrum of OH-88-syn-α to the data (Fig. 2B), the intense feature at ∼400 cm^-1 of Fe^IIIBLM is assigned as the pair of orthogonal O─Fe─N_ax bends (transaxial bends, ν_2a/b in Fig. 3), which were also observed in mononuclear nonheme Fe^IV = O model complexes (26) as intense features in their NRVS spectra. From the DFT calculations, the two transaxial bends mix with two NRVS inactive, equatorial ligand out of plane ruffling modes (with pyrimidine and deprotonated amide in-phase and out-of-phase) to produce four modes all with significant Fe motion and contributing to the dominant peak ∼400 cm^-1, in Fig. 2B.

Fig. 3.

Major NRVS active vibrational normal modes of Fe^IIIBLM and ABLM, where X = H for Fe^IIIBLM or O for ABLM. ν₁ represents the Fe─O stretching mode; ν_2a/b represents the two orthogonal transaxial bends; and ν₃ represents the Fe─O─X bend.

Using the OH-88-syn-α computational model, a complete set of EXAFS waves were generated and a final EXAFS fit was calculated (SI Appendix: Fig. S6). Various splittings of the first shell coordination environment were tested, however, the 1∶5 first shell split again generates the best fit and maintains the short Fe-O distance of 1.86 Å.

ABLM.

NRVS.

The PVDOS spectrum of ABLM (black) is compared with those of Fe^IIIBLM (starting material, green dashed) and the ABLM decay product (blue dotted) in Fig. 4. Four prominent features are observed at 539, 438, 398, and 328 cm^-1 in the PVDOS spectrum of ABLM. The ABLM decay product shows two intense features in the 200–300 cm^-1 region clearly resolved from the features of ABLM. The most intense feature of Fe^IIIBLM at 402 cm^-1, assigned in the DFT calculations section above as the pair of transaxial bends, overlaps the 398 cm^-1 feature of the ABLM. However, the ABLM sample contains only 4% Fe^IIIBLM from electron paramagnetic resonance (EPR) spin quantitation. Therefore, all four intense peaks can be attributed to ABLM. The highest energy peak is at 539 cm^-1 in ABLM compared to 567 cm^-1 for the highest energy feature in Fe^IIIBLM [the Fe─(OH) stretch], and can be assigned as the Fe─O (peroxide) stretch (vide infra). In summary, ABLM shows two new features relative to Fe^IIIBLM at 328 and 438 cm^-1; the Fe─O stretch has shifted down in energy by 28 cm^-1; and the most intense features of both Fe^IIIBLM and ABLM are at ∼400 cm^-1. These results will be used to test possible ABLM models in a section below.

Fig. 4.

NRVS PVDOS spectrum ABLM (black), ¹⁶O Fe^IIIBLM (green dashed), and ABLM decay product (blue dotted).

ABLM. EXAFS.

As for Fe^IIIBLM, the FT of the EXAFS signal of ABLM in Fig. 5 exhibits three distinct features. The EXAFS wave of ABLM is very similar to that of Fe^IIIBLM in both phase and amplitude, but with slightly greater intensity in the EXAFS beat pattern, resulting in a maximum FT intensity of 0.83 as compared to 0.75 for Fe^IIIBLM. The EXAFS of ABLM is also more intense in the second shell and the distribution of the beat pattern has shifted, potentially reflecting the additional oxygen atom of the peroxide adduct in ABLM (vide infra).

Fig. 5.

ABLM EXAFS spectrum (inset) to k = 15Å^-1 and nonphase-shifted FT of ABLM. From the FT, three distinct shells can been seen in the data at R + Δ ≈ 1.4, 2.2, and 3.0 Å, as well as a redistribution of intensity in the second shell with the presence of an additional oxygen atom in the axial ligand of ABLM relative to Fe^IIIBLM. Dotted lines are the simulated data based on model OOH-68-syn-β.

Starting from the NMR structure of (BLM)Co^III─OOH (PDB ID 1MXK) (27) as an initial structual model of ABLM, the first shell of the EXAFS data was systematically fit to determine the coordination number and shell splitting pattern. Fit values are listed in SI Appendix: Table S3. As for Fe^IIIBLM, fitting the EXAFS wave of ABLM with only a single-component first shell did not reproduce the main EXAFS wave (Fit 1, SI Appendix: Table S3), indicating that at least a dual contribution is required to fit the first shell.

As the splitting pattern is systematically varied, the distances and σ² values of the two paths again adjust to accommodate the changing coordination numbers. Based on the final overall error, the first shell is best fit to a 1∶5 split coordination with a short distance of 1.87 Å, indicating that ABLM also has a short Fe-O/N distance. Fit 1 (single shell) and Fit 2 (split first shell) are shown in SI Appendix: Fig. S7.

ABLM. DFT calculations.

The NRVS and EXAFS data provide the framework on which to evaluate possible structures of ABLM. DFT calculations have been used to generate four possible peroxo level axial ligand structures for ABLM: (BLM)Fe^III(η¹-OOH), (BLM)Fe^III(η¹-O₂H₂) (15), (BLM)Fe^III(η¹-OO), and (BLM)Fe^III(η²-OO) (8) (Scheme 1), where the (BLM)Fe^III(η¹-O₂H₂) model was recently proposed for ABLM in ref. 15. The starting structures for geometry optimizations were built on the NMR structure of (BLM)Co^III─OOH (PDB ID 1MXK) (27), which has an α axial chelate conformation (see SI Appendix: Fig. S1). ABLM experimentally has a ls S = 1/2 ground state (8), thus all geometry optimizations employed this ground state. The ls (BLM)Fe^III(η²-OO) optimization results in only an end-on configuration, thus the η²-OO model can be eliminated. The calculated PVDOS spectrum of ls (BLM)Fe^III(η¹-OO) shows two intense features at 380 and 400 cm^-1 split by 20 cm^-1, compared to the two features at 398 and 438 cm^-1 split by ∼40 cm^-1 in Fig. 4. Additionally, there are two calculated features at ∼320 cm^-1 split by 25 cm^-1, compared to only one feature at 328 cm^-1 in the experimental data (SI Appendix: Fig. S3B). Thus, based on the comparison of the DFT predicted and the experimental PVDOS NRVS data, we can also rule out the ls (BLM)Fe^III(η¹-OO) structure for ABLM.

Scheme 1.

Four possible peroxo axial ligand structures for ABLM.

The DFT predicted PVDOS spectrum of (BLM)Fe^III(η¹-O₂H₂) (SI Appendix: Fig. S3A) empirically matches the experimental PVDOS spectrum. However, the DFT optimized Fe─O bond length 1.970 Å is too long compared to the 1.87 Å EXAFS value of ABLM. Thus, this structure was reoptimized with the Fe─O bond length fixed at the EXAFS value and the PVDOS NRVS spectrum was calculated. The resultant NRVS spectrum (SI Appendix: Fig. S3J) shows only one intense peak at ∼400 cm^-1 in contrast to the two features observed experimentally. The calculated 400 cm^-1 feature corresponds to the transaxial bends while the Fe─O stretch which was at 450 cm^-1 in SI Appendix: Fig. S3A has shifted up in energy to ∼560 cm^-1. These results rule out the (BLM)Fe^III(η¹-O₂H₂) model for ABLM.*

We now consider the (BLM)Fe^III(η¹-OOH) model for ABLM, which had been suggested based on EPR (8, 28, 29), Mössbauer (30), EXAFS (9), electron nuclear double-resonance (ENDOR) spectroscopy (29, 31), mass spectrometry (11), X-ray absorption spectroscopy (XAS) edge structure (9), and magnetic circular dichroism (MCD) spectroscopy (10). A (BLM)Fe^III(η¹-OOH) model, OOH-69-syn-α, was optimized based on the NMR structure of (BLM)Co^III─OOH (PDB ID 1MXK) (27). Four additional models of (BLM)Fe^III(η¹-OOH) were also obtained, based on optimizations of the three structural perturbations in the computational details in SI Appendix. A summary of the relative free energies, Fe─O bond lengths, and Fe─O stretching frequencies is given in SI Appendix: Table S2. Only two models fit the 10 kcal/mol criterion, OOH-69-syn-α and OOH-68-syn-β. OOH-68-syn-β has the same OOH and methylvalerate amide orientations as OOH-69-syn-α, but a different confirmation of the axial chelate (SI Appendix: Fig. S1). The DFT predicted PVDOS spectrum of OOH-69-syn-α (SI Appendix: Fig. S3D) shows three features around 400 cm^-1 (however, two features are observed in the experimental NRVS data in Fig. 4) and the Fe─O stretch is calculated to be 631 cm^-1 (539 cm^-1 in experiment). The DFT predicted PVDOS spectrum of OOH-68-syn-β (SI Appendix: Fig. S3F) agrees well with experiment, with two intense features around 400 cm^-1 split by 37 cm^-1 and the Fe─O stretch at 616 cm^-1. The overestimation of the Fe─O stretch for both models is due to the DFT underestimation of the Fe─O bond length of 1.793 Å (OOH-69-syn-α) and 1.798 Å (OOH-68-syn-β) compared to the experimental EXAFS value of 1.87 Å. We thus reoptimized both structures with the Fe─O bond length fixed at the EXAFS value 1.87 Å and calculated the PVDOS NRVS spectra. The resultant PVDOS spectrum of OOH-69-syn-α (SI Appendix: Fig. S3K) shows two features at 414 and 444 cm^-1 and the Fe─O stretch at ∼540 cm^-1, and the resultant PVDOS spectrum of OOH-68-syn-β (SI Appendix: Fig. S3L) shows two features at 394 and 424 cm^-1 and the Fe─O stretch at ∼530 cm^-1. The β axial chelate conformation better reproduces the experimental intensity pattern of the two features at ∼400 cm^-1.

Based on the DFT calculated spectrum of OOH-68-syn-β with its Fe─O fixed to the EXAFS value (Fig. 6), the peak at 328 cm^-1 is assigned as the δFe─O─O bend (ν₃ in Fig. 3). The peaks at 398 and 438 cm^-1 are assigned as the pair of orthogonal δ O─Fe─N_ax bends (transaxial bends, ν_2a/b in Fig. 3), which are split by 40 cm^-1. In contrast, the δ O─Fe─N_ax bends (ν_2a/b) are effectively degenerate in Fe^IIIBLM (Fig. 2A). This difference is due to the presence of the δFe─O─O bend (ν₃) in ABLM, which selectively mixes with one of the δO─Fe─N_ax transaxial bends (ν_2a) and shifts it up in energy by 40 cm^-1. The other transaxial bend δO─Fe─N_ax (ν_2b) does not interact with ν₃ and remains at approximately the same energy as in Fe^IIIBLM, as reproduced by the DFT calculations on OH-82-syn-β and OOH-68-syn-β. Model OOH-68-syn-β was used to simulate the EXAFS spectrum and its FT of ABLM (Fig. 5).

Fig. 6.

Overlay of experimental PVDOS spectra of ABLM and the DFT predicted PVDOS spectrum of OOH-68-syn-β (Fe-O bond length 1.87 Å) with ΔFe (Fe displacement in each mode) labeled as blue bars.

Note that in ref. 15, (BLM)(H)Fe^III(η¹-OOH) is the last intermediate before O─O cleavage. This proposed species has the C = O group of the threonine of the BLM ligand protonated by the dihydrogen peroxide of the initially formed (BLM)Fe^III(η¹-O₂H₂) species. (BLM)(H)Fe^III(η¹-OOH) is thermally more stable than the preceding intermediates by ∼2 kcal/mol. Thus, this proposed species would be the most populated intermediate before O─O bond cleavage in this model. We have used DFT calculations to predict the PVDOS spectrum of (BLM)(H)Fe^III(η¹-OOH) with both α and β axial chelate conformations and an Fe─O bond fixed to the EXAFS value of 1.87 Å. The PVDOS spectrum of (BLM)(H)Fe^III(η¹-OOH)-β (SI Appendix: Fig. S3O) is inconsistent with experiment. The PVDOS spectrum of (BLM)(H)Fe^III(η¹-OOH)-α (SI Appendix: Fig. S3I) does empirically match the experimental PVDOS spectrum. However, its transaxial bends shift up in energy by ∼20 cm^-1 relative to experiment and its most intense feature does not overlap the PVDOS spectrum of Fe^IIIBLM model OH-88-syn-α (SI Appendix: Fig. S3C), which is inconsistent with experiment (Fig. 4). This calculated shift in the energy of the transaxial bends results from the strong hydrogen bond of the protonated C = O group with the distal oxygen of the hydroperoxo ligand, which hinders the transaxial bend movement. Therefore, (BLM)(H)Fe^III(η¹-OOH) can also be excluded as a model of ABLM.

Discussion

Previous experimental and computational studies have argued that ABLM reacts with substrate via direct H-atom abstraction, and not by a heterolytic or a homolytic reaction pathway (10, 13, 14). However, a recent theoretical study argued for the heterolytic pathway (similar to the peroxidases). This proposed heterolytic mechanism was based on Fe^IIIBLM being an aqua complex where the neutral H₂O ligand could be easily replaced by H₂O₂ to form “ABLM” which was described as (BLM)Fe^III(η¹─O₂H₂) or its isomer with the proton shifted to the carbonyl group of the threonine of BLM (15). From the study presented here, the NRVS data on Fe^IIIBLM show an Fe─O stretch at 567 cm^-1, which is similar to the Fe─OH stretches of the ls ferric alkaline forms of metmyoglobin (551 cm^-1) and the heme–heme oxygenase complex (546 cm^-1) (32). Furthermore, the EXAFS data on Fe^IIIBLM show an Fe─O bond length of 1.86 Å, which is similar to the 1.798 Å Fe─OH bond length of the ls [Fe^III(TMC)(NO)(OH)](ClO₄)₂·CH₃CN complex and of several hs nonheme Fe^III─OH complexes: 1.837 Å for [N4Py(NpNH)Fe^III(OH)][OTf]₂ (33), 1.837 Å for [tpa^MesFe^III(OH)]^- (34), and 1.876 Å for [Fe^III(tnpa)(PhCOO)(OH)]⁺ (35). This EXAFS Fe─O bond length is significantly shorter than the Fe─OH₂ bond length of 2.35 Å for the ls substrate-free Pseudomonas putida cytochrome P-450 (36). Furthermore, the DFT calculated (BLM)Fe^III─OH models have Fe─O bond lengths in the range of 1.839–1.841 Å and scaled υ(Fe─O) in the range of 540–555 cm^-1, which are in agreement with experiment, while the (BLM)Fe^III─OH₂ models have Fe─O bond lengths in the range of 1.919–1.981 Å and scaled υ(Fe─O) in the range of 430–499 cm^-1. Therefore, Fe^IIIBLM can be definitively assigned as a ferric-hydroxide species. Ref. 15 noted that the totally symmetric a_g Fe─O stretch of hs CsFe(SO₄)₂·12H₂O is at 523 cm^-1 (37, 38), which was considered to be close to the 561 cm^-1 value reported for Fe^IIIBLM from rR spectroscopy (23). However, this value is for a normal mode of the six H₂O ligands of the ferrous complex in a crystal, where ref. 39 associated its high value with crystal forces from directional H-bonds.

Comparing the PVDOS NRVS spectra of Fe^IIIBLM and ABLM (Fig. 4), the 567 cm^-1 feature assigned as the Fe─OH stretch of Fe^IIIBLM is not present. However, a feature at 539 cm^-1 is observed for ABLM and assigned as an Fe─O axial stretch. This assignment agrees with the EXAFS that the Fe─O bond of ABLM (1.87 Å) is very similar to that of Fe^IIIBLM (1.86 Å). In addition, the intense feature at 402 cm^-1 in Fe^IIIBLM is replaced by three intense features at 328, 398, and 438 cm^-1 in ABLM. From evaluation of the range of computational models summarized in SI Appendix: Fig. S3, these spectral changes are only consistent with the axial ligand of ABLM being an end-on bound hydroperoxide. The feature at 328 cm^-1 in ABLM is assigned as the δ Fe─O─O bend that mixes with and splits the transaxial bends by 40 cm^-1. Further, from DFT calculations, the β axial chelate conformation (SI Appendix: Fig. S1) better reproduces the ABLM experimental data (SI Appendix: Fig. S3 K and L). This spectral agreement reflects its 0.02 Å longer Fe─N_ax bond which causes the Fe─N_ax stretch to be closer in energy and mix with the transaxial bend, υ_2b, thus enhancing the intensity of this mode, and increasing the υ_2a∶υ_2b intensity ratio, consistent with experiment.

In contrast, when the axial ligand is modeled as an end-on bound H₂O₂ (PVDOS spectrum shown in SI Appendix: Fig. S3A), the transaxial bends do not split and only one feature is observed at 398 cm^-1. There is also a feature at 450 cm^-1 in the calculated NRVS spectrum from the Fe─O stretch. The low value for this vibration reflects the 1.97 Å calculated Fe─O bond length, which is too long compared to the 1.87 Å EXAFS value. Even when the Fe─O bond is constrained to 1.87 Å, the transaxial bends still remain unsplit (SI Appendix: Fig. S3J). Thus, this model is eliminated by the NRVS data. When the axial ligand is modeled as an end-on bound peroxo species (PVDOS spectrum shown in SI Appendix: Fig. S3B), the transaxial bends do split, but also move down in energy in comparison with experiment. This decrease in energy is due to the more negative charge of the (BLM)Fe^III(η¹-OO) model, which significantly elongates the Fe─deprotonated amide bond and thus reduces the force constants of the transaxial bends.

Thus, the dominant differences between the NRVS spectra of Fe^IIIBLM and ABLM are the appearance of a low energy feature at 328 cm^-1 corresponding to δ Fe─O─O bend and the splitting of the transaxial bends by 40 cm^-1 in ABLM (Figs. 4 and 6). In (BLM)Fe^III(OH), the transaxial bends are degenerate within the NRVS resolution. The energy splitting in ABLM is due to the selective mixing of the δFe─O─O bend with one of the δO─Fe─N_ax transaxial bends. The transaxial bend with its iron displacement coplanar with the Fe─O─O plane strongly mixes with the δFe─O─O bend and shifts up in energy. The remaining transaxial bend has its iron displacement orthogonal to the Fe─O─O plane and does not mix with δFe─O─O. The mixing of axial diatomic modes with transaxial bends and the splitting pattern was also observed for a six-coordinate [Fe(TPP)(MI)(NO)] heme model complex (19).

Having unambiguously defined the structures of Fe^IIIBLM and ABLM, we can now evaluate the reaction energies.^† The three possible reaction pathways of ABLM are depicted in Scheme 2B. In the heterolytic cleavage pathway, a solvated proton was used with a energy ΔG_solv(H⁺) = -262.23 kcal/mol (40). Solvation effects were accounted for by use of a polarized continuum model (PCM) with a dielectric constant of ε = 78.4 to model an aqueous environment and one with ε = 4.0 to model a more biological environment. The peroxide shunt reaction is exergonic, ΔG = -12 kcal/mol, when an ε = 4.0 or 78.4. The homolytic cleavage pathway for ABLM is the most unfavorable; endergonic with a ΔG of ∼19 kcal/mol (ε = 4.0 and 78.4). The heterolytic cleavage pathway of ABLM is endergonic, ΔG = 15.1 kcal/mol (ε = 4.0), or slightly exergonic, ΔG = -4.3 kcal/mol (ε = 78.4). The direct H-atom abstraction pathway is the most favored with a ΔG of -9.7 kcal/mol (ε = 4.0) and -10.9 kcal/mol (ε = 78.4). Thus, with an experimentally defined model for ABLM, the direct H-atom abstraction reaction pathway is energetically favored relative to homolytic or heterolytic reaction pathways, in agreement with a previous study using a smaller model (13). Because the metal binding domain of BLM binds to the minor groove of DNA (7), we have also used a dielectric constant of 30 as suggested in ref. 41 to calculate the reaction energetics of the direct H-atom abstraction and heterolytic cleavage pathways, which are -10.4 kcal/mol and -2.2 kcal/mol, respectively; the direct H-atom abstraction from DNA is thus favored by 8.2 kcal/mol. It should be noted that ref. 15 obtained a barrier of 12.5 kcal/mol for the heterolytic cleavage pathway. However, this proposed heterolytic cleavage pathway started from an incorrect description of ABLM as (BLM)Fe^III(η¹─O₂H₂). A free energy of 26.5 kcal/mol (ε = 4.0) or 8.9 kcal/mol (ε = 78.4) is required to protonate the proximal oxygen of ABLM, (BLM)Fe^III(η¹─OOH), to form (BLM)Fe^III(η¹─O₂H₂) (i.e., coordination of the hydrogen peroxide to the low spin Fe^III inductively lowers its pK_a leading to deprotonation at neutral pH). Thus, the (BLM)Fe^III(η¹─OOH) form is present at neutral pH solution. Therefore, evaluating the heterolytic reaction starting from (BLM)Fe^III(η¹─O₂H₂) artificially destabilizes the reactant and lowers the O─O heterolytic cleavage barrier. Taking this protonation energy into consideration, the O─O heterolytic cleavage barriers predicted in refs. 13 and 15 are in the range of 22 to 30 kcal/mol, significantly higher than the 17 kcal/mol activation energy calculated in ref. 13 for the direct H-atom abstraction pathway. The activation barrier of the direct H-atom abstraction pathway is also lower than that for O─O homolytic cleavage pathway as the interaction between the OH^· that would be formed and the DNA substrate stabilizes the transition state along the reaction coordinate of direct H-atom abstraction.

Scheme 2.

(A) Peroxide shunt reaction; (B) Possible reaction pathways for ABLM.

In summary, we have identified the axial ligands of Fe^IIIBLM and ABLM as OH^- and OOH^-, respectively, through NRVS aided by EXAFS spectroscopy and DFT calculations. DFT total energy calculations based on these now well defined structures show the most energetically favored reaction pathway of ABLM (BLM)Fe^III(η¹─OOH) is its direct abstraction of an H-atom from the substrate. This study demonstrates that NRVS is a powerful tool in identifying reactive intermediates in nonheme iron systems.

Materials and Methods

Materials and Spectroscopic Methods.

Sample preparation is in general the same as described previously (9). Blenoxane (a mixture of 60% BLM A2, 30% BLM B2, and 10% other BLMs) was obtained as a gift from Bristol-Meyers Squibb and used as given. ⁵⁷Fe metal (95%) and H₂¹⁸O (97%) was obtained from Cambridge Isotope Laboratories. H₂O₂ (30%) was obtained from Fisher Scientific. NRVS data were collected on beamline 3-ID at the Advance Photon Source. XAS data on Fe^IIIBLM and ABLM were collected by fluorescence detection at the Stanford Synchrotron Radiation Lightsource (SSRL) on beamline 7-3. For experimental details see SI Appendix.

Computational Methods.

Spin-unrestricted DFT calculations were performed with the Gaussian 03 package (42). Geometry optimization and frequency calculations were done using the hybrid density functional B3LYP with the 6-311G* basis set for Fe and the 6-31G* basis set for all other atoms. No imaginary frequency was found for any of the calculated models. Single point calculations were done at the B3LYP/6-311+G** level of theory. Solvation effects were accounted for by the PCM with two limiting conditions, one with a dielectric constant of ε = 78.4 to model an aqueous environment and one with ε = 4.0 for a more biological environment.

The initial Cartesian coordinates were taken from NMR structures [PDB 1GJ2 (43) and 1MXK (27)] of (BLM)Co^III─OOH, and the Co was replaced by Fe and the nature of the axial ligand was varied as described below. The DNA binding domain and carbohydrate domain (Fig. 1A, boxed areas) were truncated to simplify the model. The dangling bonds were saturated by hydrogen atoms. Three geometric perturbations were considered for geometry optimization of the models: 1) variation of the orientation axial ligands, OH^- or OH₂ for Fe^IIIBLM and OOH^- for ABLM models (SI Appendix: Fig. S1). The orientation is defined by θ, which is the deprotonated amide N─Fe─O─X (X = H for Fe^IIIBLM and O for ABLM) dihedral angle; 2) Three different orientations of the amide in methylvalerate, where syn, para, and anti (SI Appendix: Fig. S1) define whether the N─H bond points toward, parallel, and away from the pyrimidine─Fe─O plane independently, before geometry optimization; and 3) two conformations of the transaxial axial chelate, labeled as α and β in SI Appendix: Fig. S1. α conformation was found in one NMR structure of (BLM)Co^III─OOH (43) and the β conformation was found in one NMR structure (27) and a crystal structure (7) of (BLM)Co^III─OOH. The nomenclature used here is such that OH-90-syn-α describes a BLM model with OH^- as the axial ligand with θ = 90°, syn orientation of the amide in methylvalerate, and the α conformation of the axial chelate. DNA, the substrate of ABLM, was model with the entire deoxyribose sugar moiety with the 3′- and 5′-phosphates substituted by hydroxy ligands and the base replaced by a hydrogen atom.