Solution structure of the hydroperoxide of Co(III) phleomycin complexed with d(CCAGGCCTGG)₂: evidence for binding by partial intercalation

Abstract

The bleomycins (BLMs) are natural products that in the presence of iron and oxygen bind to and cause single-strand and double-strand cleavage of DNA. The mode(s) of binding of the FeBLMs that leads to sequence-specific cleavage at pyrimidines 3′ to guanines and chemical-specific cleavage at the C-4′ H of the deoxyribose of the pyrimidine has remained controversial. 2D NMR studies using the hydroperoxide of CoBLM (HOO-CoBLM) have demonstrated that its bithiazole tail binds by partial intercalation to duplex DNA. Studies with ZnBLM demonstrate that the bithiazole tail binds in the minor groove. Phleomycins (PLMs) are BLM analogs in which the penultimate thiazolium ring of the bithiazole tail is reduced. The disruption of planarity of this ring and the similarities between FePLM- and FeBLM-mediated DNA cleavage have led Hecht and co-workers to conclude that a partial intercalative mode of binding is not feasible. The interaction of HOO-CoPLM with d(CCAGGCCTGG)₂ has therefore been investigated. Binding studies indicate a single site with a K_d of 16 µM, 100-fold greater than HOO-CoBLM for the same site. 2D NMR methods and molecular modeling using NMR-derived restraints have led to a structural model of HOO-CoPLM complexed to d(CCAGGCCTGG)₂. The model reveals a partial intercalative mode of binding and the basis for sequence specificity of binding and chemical specificity of cleavage. The importance of the bithiazoles and the partial intercalative mode of binding in the double-strand cleavage of DNA is discussed.

INTRODUCTION

The bleomycins (BLMs) (Fig. 1) are natural products used clinically in combination chemotherapy in the treatment of a number of specific tumor types (1–3). Possible mechanisms of their selective cytotoxicity have not been established. The data thus far support a model in which the BLMs, in the presence of their required cofactors iron and oxygen, bind to and cleave duplex DNA. Hot-spots for double-strand (ds)DNA cleavage exist and appear to be the result of one BLM molecule effecting two cleavage events (4–6). The difficulty of repair of these double-stranded lesions suggests that the cytotoxicity is related to this process. DNA repair enzymes are induced in the presence of BLM, supporting DNA cleavage as a cytotoxic event (7,8). A limited number of sites in RNA can also be efficiently cleaved by BLM, although the phenotypic consequence of such cleavage events and the relationship to cytotoxicity are unclear (9).

Figure 1.

The structure of PLM D1 and the bithiazole and C-terminus of bleomycin A2. Nitrogen ligands coordinating to the metal are underlined. Single letter abbreviations for each moiety are given. (Note that for simplicity B is used for the thiazolinyl-thiazole in PLM throughout the text, even though it is not a bithiazole.)

A number of models by which metallo-BLMs bind to DNA have been presented in the literature based on 2D NMR structures generated from metallo-BLMs bound sequence specifically to oligonucleotides. All of the structures derived using the hydroperoxide of CoBLMs (HOO-CoBLMs) demonstrate that the bithiazole tail is inserted 3′ to the pyrimidine cleavage site and facilitates DNA binding by partial intercalation (10–14). The terminal thiazole ring is well stacked within the duplex DNA, while the penultimate thiazole ring is only partially stacked. The basis of the chemistry at the C-4′ hydrogen of the pyrimidine 3′ to a guanine is also well defined by these model structures. The 4-amino group and N3 of the pyrimidine of the BLM form hydrogen bond interactions with the N-3 of guanine and its 2-amino group 5′ to the cleavage site (15).

Studies with ZnBLMs present several alternative pictures of binding (16–18). Their complexity is in part related to the presence of multiple binding modes in the sequences investigated and the location of the cleavage sites at the ends of the duplexes. In addition, ZnBLM gives an alternative binding mode to the corresponding deglyco-ZnBLM (17,19), in contrast to the HOO-CoBLMs (11,13,14,20). The studies with ZnBLM demonstrate that the bithiazoles can bind in the minor groove; however, the relationship of ZnBLM binding to FeBLM sequence-specific binding and cleavage remains unclear due to imposition of artificial restraints in the molecular modeling. The most compelling evidence for BLM-mediated DNA cleavage via binding modes other than intercalation is provided by recent studies with BLM tethered to control pore glass (CPG) beads or to a trigalactose derivative (21,22). Tethering precludes threading of the bithiazole tail through the minor groove and hence binding by partial intercalation. Studies with these tethered FeBLMs demonstrated DNA cleavage with the same sequence specificity and similar cleavage efficiency as free FeBLM. These results support the hypothesis that FeBLM can bind to DNA by a mode other than intercalation and mediate cleavage. The structural basis, however, for the GpC sequence specificity from a non-intercalative mode of binding remains to be established.

Additional data cited in the literature (17,19), which question the importance of the partial intercalative mode of binding in the chemistry of FeBLM-mediated cleavage, show that phleomycin (PLM) (Fig. 1) cleaves DNA with very similar specificity and efficiency as BLM. PLM has a bithiazole tail in which the penultimate thiazolium ring is partially reduced, a modification disrupting planarity and thus claimed to preclude binding by intercalation. The importance of an intercalative binding mode of BLM to our model for how a single molecule of BLM mediates double-stranded cleavage requires that the relevance of HOO-CoBLM as a model for activated FeBLM (HOO-FeBLM) be further examined (15). The following studies were therefore undertaken to determine whether HOO-CoPLM can bind to oligonucleotide d(CCAGGCCTGG)₂ (1) by the same mode as previously observed for HOO-CoBLM and as our molecular modeling suggested (11,15). Results are presented that establish that while HOO-CoPLM binds more weakly to 1 than BLM, it binds in an almost identical fashion. These results establish that arguments made by comparing similarities between PLM and BLM on the efficiency and specificity of DNA cleavage clearly do not eliminate the partial intercalation mode as being important in single-stranded DNA cleavage events and essential in double-stranded cleavage events. If DNA is really the target of BLMs, then it is likely that double-stranded cleavage is the basis of its cytotoxicity. Our hypothesis is that the unusual choice of bithiazoles for DNA binding may be uniquely related to a double-stranded cleavage model. Studies with PLM and the difference in its propensity to mediate double-stranded cleavage relative to BLM may serve as a probe of the mechanism(s) of their cytotoxicity.

MATERIALS AND METHODS

Preparation of HOO-CoPLM

The copper form of PLM was purchased from CAYLA (France). ApoPLM (50 mg), prepared using a published procedure (23), was dissolved in 2 ml of 100 mM sodium phosphate buffer (pH 6.8). CoCl₂ (1.1 equivalents) was then added to the rapidly stirred solution to ensure oxygenation. The reaction was allowed to proceed for 3 h at room temperature. The mixture of products was separated using a semi-preparative reverse phase Alltech Econosil C-18 column (10 µm) and an elution system of 0.1 M ammonium acetate (pH 6.8) as solvent A and acetonitrile as solvent B. The products were eluted at a flow rate of 3 ml min^–1 using a linear gradient from 12 to 16% A over 60 min (H₂O-CoPLM, retention time 14 min, yield 22 mg, yield 40%; HOO-CoPLM, 21 min, 14 mg, 25%). The lyophilized samples were redissolved in 50 mM sodium phosphate (pH 6.8) and stored at –80°C.

Characterization of HOO-CoPLM

The extinction coefficient of HOO-CoPLM was measured using inductively coupled plasma emission spectroscopy for cobalt quantitation as previously described (12) and was determined to be 0.91 ± 0.2 × 10⁴ M^–1 cm^–1 at 290 nm. The electrospray mass spectrum of HOO-CoPLM was obtained as described previously (11).

Titration and binding studies of HOO-CoPLM with d(CCAGGCCTGG) (1)

Oligonucleotide 1 was prepared and quantitated as previously described (11). Aliquots of HOO-CoPLM (0–1 equivalent) were added to a solution of 1 in D₂O and complex formation was followed by monitoring the changes in the ¹H NMR spectrum. The K_d of HOO-CoPLM with 1 was measured by monitoring fluorescence quenching and Scatchard analysis (11).

NMR experiments

All NMR experiments were performed on 750 or 500 MHz Varian NMR spectrometers or on 600 or 500 MHz custom-built instruments at the Francis Bitter Magnet Laboratory. Data were then transferred to a Silicon Graphics workstation and processed using Felix software version 2.3 (Molecular Simulations Inc., now Accelrys Inc.). ¹H and ¹³C chemical shifts are referenced to an internal standard, sodium 3-(trimethylsilyl)-1-propane sulfonate at 0.00 p.p.m.

DQF-COSY, TOCSY (MLEV-17 spin lock pulse with 70 ms mixing times) and NOESY (50, 200 and 400 ms mixing times) experiments were recorded at 20°C in D₂O or H₂O on HOO-CoPLM and HOO-CoPLM·1. Data sets with 4096 × 512 complex points were acquired with sweep widths of 5000 (500 MHz instrument) or 8000 Hz (750 MHz instrument) in both dimensions and 32 scans per t₁ increment. During the relaxation delay period, a 2.0 s presaturation pulse was used for solvent suppression. For the NOESY experiments in H₂O, a WATERGATE gradient pulse sequence (24) was used, and data sets with 4096 × 512 complex points were acquired with sweep widths of 12 000 (500 MHz instrument) or 15 000 Hz (750 MHz instrument) in both dimensions. The spectra were processed as previously reported (11).

The HMQC (25) spectrum was recorded at 20°C in D₂O with a J_C-H coupling constant of 190 Hz on HOO-CoPLM and HOO-CoPLM·1. Data sets with 2048 × 256 complex points were acquired with 6000 (¹H) and 25 000 Hz (¹³C) sweep widths on a 500 MHz instrument. For every t₁ increment 256 scans were collected. During the relaxation delay period, a 1.5 s presaturation pulse was used for solvent suppression. The spectrum was processed as described previously (11). ROESY (50, 100, 200 and 300 ms mixing times) experiments (26) on HOO-CoPLM·1 at 20°C were acquired in D₂O with 9 kHz rf field strength on a 500 MHz instrument.

Back-calculations

The NOE cross-peak volumes corresponding to the aromatic to H1′ and the aromatic to H2′, H2″ and methyl regions were back-calculated using the matrix doubling method in the FELIX 950 software package (Accelrys Inc.). The average structure produced between 75 and 80 ps at intervals of 1 ps was used in this calculation. The parameters used in this protocol were as follows: a correlation time of 6.0 ns, a Z-leakage rate of 3.0 s^–1 and a distance cut-off of 7.5 Å. The back-calculated or theoretical NOE intensities were then normalized based on the H5/H6 experimental proton peaks of the cytosine residues in 1.

Molecular modeling

All calculations were carried out with CHARMM 24 (27) on a Cray Y-MP or J-90. The calculation of non-bonded van der Waals and electrostatic interactions were truncated at 13 Å, using a force switching function between 8 and 12 Å. The list of non-bonded terms was updated every 20 steps, except in the final molecular dynamics (MD) run, where the list was updated when any atom moved >0.5 Å. The terms for electrostatic interactions and hydrogen bonds were only included in the final 15 ps step of the calculation in the case of HOO-CoPLM·1, and were not included at all when calculating the solution structure of HOO-CoPLM. Following heating, the temperature was maintained by scaling the velocities of the atoms as necessary to keep the temperature at 300 ± 10 K. In the final 15 ps MD phase, scaling was only required one or two times.

Building of initial coordinate files. Phleomycin D1 (Fig. 1) was constructed in Quanta with an overall charge of +2 (Co, +3; HOO, –1; guanidinium, +1; deprotonated amide, –1). Models for HOO-CoPLM were constructed with both an R and S stereochemistry at the B-C4′ position. The bond lengths and bond angles of the thiazolinyl-thiazole rings were based on the X-ray crystal structures of analogous compounds in the Cambridge Structural Database. A and B forms of DNA were constructed in Quanta (28). The program CURVES (29,30) was used to measure DNA conformational parameters.

Distance constraints. For the free HOO-CoPLM, 63 NOEs were used to derive distance constraints for structural calculations (Table S1, Supplementary Material). For the calculations of HOO-CoPLM·1, 352 NOEs were used: 56 intermolecular NOEs (Table 1), 65 intramolecular NOEs from HOO-CoPLM (Table S2), 203 intramolecular NOEs from 1 and 28 Watson–Crick hydrogen bond constraints. These NOEs were classified as strong, medium or weak based on visual inspection of the cross-peak intensities in the 200 ms NOESY spectra. The distance constraints were set at 1.9–3.0, 1.9–4.0 and 3.0–5.0 Å for strong, medium and weak NOEs, respectively. An additional 1 Å was added to the upper limit of constraints for the methyl or methylene hydrogen pseudoatoms.

Table 1. Intermolecular NOEs between HOO-CoPLM and DNA at 20°C with a 300 ms mixing time^a.

5′ End	Strand 1	BLM residues	BLM residues	Strand 2	3′ End
A3			A-CαH /m	H5′b	T18
G4	NH	P-NH2(1) /w	A-CαH /w	H1′	C17
			A-CαH /m	H4′
G5	H4	P-NH2(1) /w			C16
		P-NH2(2) /w
		P-CH3 /w
	H1′	P-CH3 /w
		P-NH2(1) /m
		P-NH2(2) /m
	NH	P-NH2(1) /w
C6	H5″	P-CH3 /s	B-C5H /w	H1′	G15
		V-γCH3 /w
	H5′	P-CH3 /m	B-C5H /w	H2″
		V-γCH3 /w	B-C5H /w	H5′
			B-C5H′, H″ /m,m	NH
			B-C4H′ /m
	H4′	CoOOH /m	B-CαH′ /w
		P-CH3 /w	B-C5H /m	H8
		V-γCH3 /w
	H2″	CoOOH /w
	H2′	CoOOH /w
	H1′	CoOOH /m
		P-CH3 /w
		P-CβH /w
		P-CαH /w
		B-CβH /w
	H5	B-C5H′ /w
C7	H5″	T-CαH /m	B-C5H /m	H1′	G14
	H5′	CoOOH /w	B-C5H /m	H2″
		V-OH /w	B-C5H /m	H2′
		B-NH /w	B-C5H /w	H3′
		T-CαH /m	B-C5H /w	H4′
	H4′	CoOOH /w	B-C5H′, H″ /w,w	NH
		T-CH3 /w	B-C5H /w	H8
		T-CαH /m	D-CγH /w
		B-NH /m	D-CδH′ /w
	H1′	B-NH /w	D-CδH /w
	H5	B-C5H′ /w
3′ End	Strand 1			Strand 2	5′ End

^aP-NH₂ (1) and P-NH₂ (2) are the hydrogens at 10.36 and 7.14 p.p.m., respectively.

^bUsed the H5′ pseudoatom.

Dihedral angle constraints. In the free HOO-CoPLM, vicinal coupling constants were derived from the 1D spectra. The generalized Karplus equation was used to derive dihedral angle constraints (31). The analysis of the coupling constants in HOO-CoPLM·1 was based on visual inspection of the cross-peak sizes in the DQF-COSY spectra collected in D₂O and H₂O. In the DNA dihedral angle constraints on the α and ζ backbone angles (α = 295°, ζ = 240°) were used based on the ³¹P chemical shifts. However, the α and ζ backbone angles of G5∼P∼C6 and G14∼P∼G15 phosphorous residues were constrained to α = 295° and ζ = 180°, consistent with the change in geometry expected with intercalation.

Initial structure. To calculate the solution structures of HOO-CoPLM, starting structures were constructed with either the β-aminoalanine primary amine or the mannose carbamoyl oxygen or amino group as an axial ligand to the Co³⁺. In addition, the thiazolinyl ring of B with both the R and S configurations at C-4′ were constructed. Only the screw sense isomer that places the axial ligand on the opposite face of the equatorial plane as the V and T moieties was used. Twenty initial structures of each of the four models were generated with random conformations for the backbone dihedral angles of the H, V, T and B moieties, and submitted to the restrained molecular dynamics protocol.

The initial structure of HOO-CoPLM·1 was constructed by manually positioning PLM so that only the terminal thiazole ring was partially intercalated between the C6·G15 and C7·G14 base pairs, with the metal-binding region extending out of the minor groove. This preliminary structure was minimized by 200 steps of the steepest descent method, followed by conjugate gradient minimization to a RMS gradient <0.1 and, finally, by the conjugate gradient method using the dihedral and distance constraints, to a RMS gradient <0.1.

MD simulations. Restrained molecular dynamics simulated annealing (MDSA), following the protocol described previously (11,32), was used to generate ensembles of low energy structures that satisfied the experimentally derived constraints. The protocol consisted of heating the system to 1000 K over 8 ps with no distance constraints in the case of free HOO-CoPLM models or over 6 ps with weak distance constraints in the case of HOO-CoPLM·1. Then the force constants applied to the distance constraints were gradually increased to 120 kcal mol^–1 Å^–1 (6.5 ps), followed by a high temperature equilibration step (10 ps). This step was followed by slow cooling to 300 K (7 ps), reduction of the force constants applied to the distance constraints to 60 kcal mol^–1 Å^–1 and a gradual introduction of the dihedral constraints to 60 kcal mol^–1 rad^–1 (10 ps) and the final molecular dynamics stage (15 ps). A final average structure for each iteration was generated by averaging the coordinates of the final 5 ps of the 15 ps molecular dynamics simulation, followed by 2000 steps of conjugate gradient minimization with the distance constraints and HOO-CoPLM dihedral angle constraints. The final average structure of HOO-CoPLM·1 was generated by averaging and minimizing the ensemble of structures generated from 10 separate MDSA calculations. The energy minimized average structure was used for figures, except where noted. The geometries used to describe the structures are reported as the true mean value of the family of structures ± 1 SD.

RESULTS

Preparation and characterization of HOO-CoPLM

HOO-CoPLM was prepared by a modification of the previously published procedure for HOO-CoBLM (32,33). The products, a mixture of hydroperoxide and aquo or hydroxide forms, were separated by HPLC. The extinction coefficient of HOO-CoPLM (ε₂₉₀ = 0.91 × 10⁴ M^–1 cm^–1) was less than HOO-CoBLM (ε₂₉₀ = 2.1 × 10⁴ M^–1 cm^–1). Since A₂₉₀ is largely due to the π to π* transition of the bithiazole, the change from a bithiazole in BLM to the 2-(4′-thiazolinyl)-thiazole in PLM would be expected to decrease the extinction coefficient. HOO-CoPLM was purified by semi-preparative HPLC in 25% yield. The downfield region of the ¹H NMR spectrum of HOO-CoPLM revealed three singlets for H-C2H, H-C4H and B-C5H, indicative of a compound of >95% homogeneity. The NMR spectrum (20°C, pH 6.8) did not change over a period of several months.

To establish that one axial ligand is in fact the hydroperoxide, electrospray mass spectrometry was used. This method has previously been employed to characterize both activated BLM (HOO-FeBLM) and HOO-CoBLM (32,34). The electrospray mass spectrum exhibited a m/z of 759.3 with the predicted isotopic distribution (calculated molecular weight of 1518.4). The total charge of +2 was consistent with prediction.

Characterization of HOO-CoPLM by NMR

Proton chemical shifts and coupling constants of HOO-CoPLM. As a prerequisite to studying the structure of HOO-CoPLM·1, the chemical shift assignments, metal ligand screw sense and overall 3D structure of HOO-CoPLM were established. The chemical shifts of exchangeable and non-exchangeable protons were assigned following the strategy used in previous studies of metallo-BLMs (35–37). The ¹H and ¹³C chemical shift assignments at 20°C and pH 6.8 for HOO-CoPLM are listed in Table S3, as are the assignments from our previous study of HOO-CoBLM at 5°C and pH 6.8. The chemical shifts for HOO-CoPLM and HOO-CoBLM are very similar except for the B region and the sugars.

Surprisingly, chemical shift differences are found for the gulose and mannose groups. Increased chemical shift dispersion of HOO-CoPLM relative to HOO-CoBLM facilitated the unambiguous assignment of all the non-exchangeable protons of both carbohydrates. This dispersion also allowed the assignment of 11 NOEs not observed in HOO-CoBLM (Table S1 and Fig. 1S). This information contributed to a model structure that better defined the position of the sugars relative to the metal-binding region than in our previous studies with CoBLMs.

The vicinal proton coupling constants in HOO-CoPLM were qualitatively similar to those observed in HOO-CoBLM, suggesting a similar overall conformation (Table S4). Torsion angle constraints were derived from the coupling constants for the A, H, V and T groups.

Structure of free HOO-CoPLM. Our previous modeling studies and those of others with HOO-CoBLM favored the screw sense isomer with the hydroperoxide on the same face of the octahedral cobalt as the peptide linker and the bithiazole tail (13,14,32,35). The identity of the second axial ligand (primary amine of β-alanine or the carbamoyl moiety of the mannose) was not unambiguously established, although our laboratory and the Petering laboratory favored the primary amine of β-alanine. A recent crystal structure of Cu(II)BLM bound to the bleomycin resistance protein (BLMA) from Streptomyces verticullus shows that the metal ligation state is superimposable on our HOO-CoBLM model. The screw sense is identical to the one proposed by our modeling studies and the axial ligand is the primary amine of β-alanine (38). Thus our modeling studies with PLM were restricted to the energetically favored screw sense of HOO-CoBLM. MDSA with 63 NOE-derived distance constraints (Table S1) and 11 torsion angle constraints (Table S4) was used to model the structure of HOO-CoPLM by methods previously described (32).

The two preferred screw sense models were constructed with both R and S stereochemistry at the C4′ position of the thiazolinyl ring. Twenty initial structures of each of the four models were generated with random conformations for the backbone dihedral angles of the H, V, T and B moieties and submitted to MDSA (Materials and Methods). The 10 best structures in terms of constraint violations and CHARMM potential energy were selected in each group of structures and analyzed. Tables 2 and S5 provide a summary of how well the calculated structures satisfy the experimental constraints and the CHARMM potential energies. No quantitative distinctions were observed between the structures differing only in the stereochemistry at B-C4′. Consequently, subsequent discussion of the HOO-CoPLM structure will focus on the R stereoisomer supported by the configurational analysis of Hamamichi and Hecht (39).

Table 2. Summary of constraint violations and potential energy terms in the final ensemble of 10 structures from the MDSA with HOO·CoPLM (R configuration at B-C4′).

Axial ligand in model	A-1° amine	M carbamoyl
RMSD from distance restraints (Å)	0.024 ± 0.002	0.019 ± 0.005
Total violations	6 ± 1	6 ± 1
Σ violations (Å)	0.33 ± 0.04	0.27 ± 0.07
Maximum violation (Å)	0.18 ± 0.01	0.13 ± 0.05
Potential energy terms
Total (kcal mol–1)	45 ± 4	74 ± 7
NOE constraint (kcal mol–1 Å–2)	1.1 ± 0.1	0.8 ± 0.4
Dihedral constraint (kcal mol–1 rad–1)	3.1 ± 0.4	1.5 ± 0.6
van der Waals	–73 ± 3	–65 ± 2
Bond	4.8 ± 0.1	7.0 ± 0.3
Angle	40 ± 2	59 ± 4
Dihedral angle	68 ± 1	70 ± 2
Improper angle	1.0 ± 0.1	1.0 ± 0.1
Atomic RMSD for heavy atoms of moieties A-T, Co & -OOH (Å)	0.49 ± 0.17	0.66 ± 0.16

Structures with either the amino group of β-alanine or the carbamoyl moiety of mannose as the axial ligand are able to satisfy the NOE distance constraints equally well. In either case there are very few constraint violations, and none greater than 0.2 Å (Tables 2 and S5). However, when the mannose carbamoyl (either the O or the NH₂) is the axial ligand, the structures are 60% higher in energy than with the primary amine ligand. This suggests that ligation of the mannose carbamoyl to the cobalt creates more strain and less favorable packing in the complex than the primary amine. This potential energy does not include the favorable entropic contribution expected from the β-aminoalanine forming a five member chelate ring upon ligation. These energy differences and the recent structural work discussed below lead us to continue to favor the primary amine of β-alanine as the axial ligand.

The well defined folded conformation of the V and T linker, which results in the placement of B underneath the pyrimidine ring, is nearly identical to the structure previously described for HOO-CoBLM (28,40,41). This is evidenced by the similar coupling constants for the T and V linker and the long range NOE between the P-CH₃ and B-C5H of the thiazole, which was also observed in HOO-CoBLM. Like HOO-CoBLM, the V-NH and T-NH in HOO-CoPLM are ideally positioned to hydrogen bond with the penultimate oxygen of the hydroperoxide ligand (2.2 and 2.5 Å, respectively). Furthermore, the chemical shifts of the V-NH and T-NH protons in HOO-CoPLM (8.74 and 8.72 p.p.m., respectively (Table S3) are similar to the shifts observed in HOO-CoBLM (8.89 and 8.92 p.p.m.) and are ∼1 p.p.m. downfield from the same protons in the Fe- and ZnBLMs (36,37). These downfield shifts are consistent with the protons being deshielded as a result of a hydrogen bond interaction with the hydroperoxide group.

Modeling of the sugars of PLM relative to the metal-binding domain offers an alternative interpretation of the previous studies on OC-FeBLM and HOO-CoPepleomycin (13,37). NOEs between M-C2H and A-CαH and M-C3H and P-CH₃, which were not observed in our previous studies with HOO-CoBLM, have been observed in studies with HOO-CoPepleomycin and OC-FeBLM, respectively. In these latter two cases the NOEs have been cited as evidence that the mannose is held in close proximity to the residues of the metal-binding region as a result of the ligation of the carbamoyl N or O to the metal. Our findings suggest an alternative explanation.

A different interaction that would stabilize the mannose can be proposed based on our present study. In eight out of 10 of the structures with the primary amine as the axial ligand (Fig. S2), the carbamoyl oxygen of the mannose is in position to hydrogen bond with the primary amine of the β-aminoalanine (A) moiety (2.1 Å, 139 ± 5°). It is important to note that no electrostatic or hydrogen bond terms were included in the MDSA calculation of free HOO-CoPLM, so these structures have no bias toward hydrogen bonds that might be an artifact of the CHARMM force field. These structures are defined solely by the most favorable internal geometry, van der Waals interactions and the experimentally derived constraints. The same eight structures also position the mannose carbamoyl oxygen 3.2 Å above the center of the pyrimidine ring, with the C=O double bond nearly co-planar to the ring. This suggests the possibility of a stacking interaction between the carbamoyl carbonyl and the pyrimidine ring. Overall, the structure of HOO-CoPLM is nearly identical to that of HOO-CoBLM (32), with the same axial ligand and screw sense and a similarly defined linker conformation.

NMR of HOO-CoPLM bound to 1

Binding affinity of HOO-CoPLM for 1. Previous fluorescence quenching studies have shown that 1 contains a single binding site for HOO-CoBLM with a K_{d apparent} < 0.17 µM (32). A similar measurement has been made with HOO-CoPLM in an effort to examine the contribution of the thiazolinyl- thiazole domain to DNA binding. Under identical conditions, a K_{d apparent} of 16.7 ± 0.7 µM was obtained.

1D ¹H NMR titration of 1 with HOO-CoPLM. The previous NMR titration of 1 with HOO-CoBLM exhibited the formation of a 1:1 complex in slow exchange on the NMR time scale and allowed determination of its structure (32). A similar experiment with HOO-CoPLM has been carried out to determine the binding mode of the thiazolinyl-thiazole in PLM to 1. The results of titration of 1 with HOO-CoPLM monitored by 1D NMR spectroscopy are shown in Figure 2. The disappearance of A3-H8 in free palindromic DNA upon addition of HOO-CoPLM is accompanied by symmetry disruption and generation of A3-H8 and A13-H8, which are associated with the 1:1 complex in slow exchange on the NMR time scale.

Figure 2.

Titration of 1 with HOO-CoPLM at 20°C. Downfield region of the ¹H NMR (500 MHz): 1 (2.0 mM) in 50 mM sodium phosphate (pH 6.8) with (B) 1.0, (C) 0.5 and (D) 0 equivalents of HOO-CoPLM A2 green added, respectively, and (A) free HOO-CoPLM in 50 mM sodium phosphate (pH 6.8).

Proton assignment of HOO-CoPLM bound to 1. Proton assignments for HOO-CoPLM in the complex (Table S6) have been made following the strategy used previously for HOO-CoBLM·1. In light of the similarity between these complexes, only their differences will be highlighted. In the HOO-CoBLM complex the bithiazole ring protons, B-C5H and B-C5′H, are upfield shifted as a consequence of the bithiazole intercalating between the base pairs C6·G15 and C7·G14. Due to the lack of any COSY connectivities with these protons, their chemical shift assignments were dependent on assignment of carbon chemical shifts associated with each thiazole ring and the fact that these shifts remained unperturbed on complexation with oligonucleotides (11,32). Similarly, in the HOO-CoPLM complex, the ¹³C B-C5 chemical shift at 127.7 p.p.m. (126.8 p.p.m. in the HOO-CoBLM complex) allowed assignment from the HMQC spectrum of B-C5H at 7.15 p.p.m. The unique ¹H and ¹³C chemical shifts of B-C4′ and B-C5′ were straightforward to assign by comparison with the spectra of HOO-CoBLM·1. The changes in chemical shifts observed in the thiazolinyl-thiazole moiety when HOO-CoPLM binds to 1 provide strong evidence for magnetic shielding of these protons due to the intercalation of this region between the base pairs C6·G15 and C7·G14 (Tables D3 and S6). The B-Cβ H/H′ shift upfield from 3.46/3.69 to 2.61/3.52 p.p.m., the B-C5′H/5′′H from 3.84/3.48 to 3.03/2.34 p.p.m., the B-C4′H from 5.79 to 4.55 p.p.m., and the B-C5H shifts from 8.00 to 7.15 p.p.m. Note especially the dramatic shifts of the protons at the two sp³ centers of B-C4′ (5.79 to 4.55) and B-C5′ (3.84 to 3.03).

A second region of interest in characterizing the interaction of HOO-CoPLM with 1 is the 4-NH₂ of the pyrimidinyl propionamide. The changes in the chemical shifts of the NH₂ amino protons observed upon binding to 1 led to the identification of the hydrogen bond interactions that dictate the sequence specificity of BLM-mediated DNA cleavage (11). In HOO-CoPLM the P-NH₂ protons shift from 7.73 and 7.94 p.p.m. in the free form to 7.07 and 10.36 p.p.m. in the complex with 1. These shifts are nearly identical to those observed with HOO-CoBLM. The 2.4 p.p.m. downfield shift for one of these protons is consistent with substantial deshielding due to that proton being involved in a hydrogen bond with N3 of the G5 of 1, as shown below. In addition, the NOEs observed between the P moiety and 1 in the HOO-CoPLM complex are very similar to those in the HOO-CoBLM complex (Table 1).

In the metal-binding region, no COSY cross-peaks could be observed between the A-CαH and A-CβH/H′ protons of HOO-CoPLM, analogous to the ∼4 Hz vicinal coupling constants observed between those protons in the 1D ¹H spectrum of free HOO-CoPLM. This is evidence that the A-Cα and A-Cβ protons maintain a gauche/gauche relationship to one another that does not change upon binding to 1. The same observations were made in our previous studies with HOO-CoBLM and were subsequently shown to be consistent with the participation of the primary amine of the β-aminoalanine in metal binding.

Finally, an exchangeable proton is observed at 8.89 p.p.m. with no through bond couplings. This resonance is identical to the hydroperoxide proton observed in HOO-CoBLM complexed to 1. The 12 intramolecular (Table S2) and six intermolecular NOEs (Table 1) associated with this proton are very similar to those observed in the HOO-CoBLM complex. The fact that the metal-bound hydroperoxide is so well defined in almost exactly the same position is further evidence that the metal-binding domains of HOO-CoPLM and HOO-CoBLM with 1 are very similar.

Proton assignment of 1 in its complex with HOO-CoPLM. The complete assignment of the DNA protons was complicated by the observation of exchange cross-peaks between symmetry related protons on opposite strands of 1 (see Fig. 3 and Fig. S3). Nonetheless, a standard strategy that utilizes NOESY, TOCSY, DQF-COSY and ³¹P-¹H COSY spectra has allowed the assignment of most of the exchangeable and non-exchangeable protons of 1 (Table S7). Some of the protons associated with C-2′ and C-5′ of the deoxyriboses could not be assigned due to spectral overlap. The assignments are very similar to those previously reported for 1 complexed with HOO-CoBLM and thus only the differences between these complexes will be highlighted (Fig. 4S).

Figure 3.

Positive contours of ROESY spectrum (200 ms mixing time) of HOO-CoPLM complexed with 1. Cross-peaks associated with chemical exchange are (A) A3-H8–A13-H8 and (B) A3-H2–A13-H2.

In the ¹H spectrum in H₂O of HOO-CoBLM·1, the chemical shifts of the imino protons of the C6·G15 and C7·G14 base pairs were upfield shifted by ∼0.5–1.0 p.p.m., relative to other G·C base pairs in the decamer (Fig. 4A). In addition, no sequential NOE was observed between these two imino protons. These results provided strong evidence that the bithiazole of BLM was intercalated between these two base pairs. However, in HOO-CoPLM·1 the imino protons associated with the C6·G15 and C7·G14 base pairs are not upfield shifted (Fig. 4B) and there is a weak NOE observed between these imino protons. These results suggest an apparent contradiction to an intercalative binding mode.

Figure 4.

1D imino proton comparisons between (A) DNA/HOO-CoBLM and (B) DNA/HOO-CoPLM. The assignments of the imino protons for the DNA/HOO-CoPLM are given in Table S7. The oval adjacent to G5 and C6 represents the metal-binding domain of BLM (PLM) and the vertical line designates the linker and bithiazole tail and the site of intercalation.

On the other hand, the extensive number and the pattern of NOEs observed between the thiazolinyl-thiazole of HOO-CoPLM and 1 is very similar to the number and pattern of the NOEs observed between the bithiazole of HOO-CoBLM and 1 [table 1 in Wu et al. (11)]. This NOE data is best accommodated by an intercalated structure with the C-terminus threaded 3′ to the cleavage site and into the major groove. The latter interaction is supported by the NOEs between the G14-H8 and the D-Cγ and D-Cδ protons of HOO-CoPLM (Table 1).

The apparent contradictions between these different experiments can be reconciled by invoking differences in the exchange rate between HOO-CoPLM and 1 relative to HOO-CoBLM and 1. This exchange rate (measured in the next section) is in part a consequence of weaker binding of PLM relative to BLM. In addition, the non-planarity of the penultimate thiazoline ring in PLM could alter shielding of the imino protons relative to the planar thiazole ring of BLM, eliminating the basis for the upfield chemical shifts.

Exchange peaks and estimation of k_ex. ROESY spectra provide a mechanism to distinguish between cross-peaks associated with dipole–dipole interactions and those associated with chemical exchange. In the former case the sign is opposite to the diagonal and in the latter it is the same as the diagonal. ROESY spectra of HOO-CoPLM·1 (200 ms mixing time) were therefore acquired to determine whether an exchange rate model is reasonable to accommodate the differences in imino NOE data between PLM and BLM observed. Figure 3 shows cross-peaks between A3-H8 and A13-H8 (A) and A3-H2 and A13-H2 (B) that have positive contours and are thus associated with a chemical exchange process. Furthermore, Figure 3 shows additional examples of exchange peaks in the NOESY spectra (200 ms mixing time) between the aforementioned protons as well as the T8/T18-imino protons and the T8-H6 and T18-H6 protons (26). Thus these cross-peaks that correspond to correlations between protons remote from each other within the 3D structure are reasonably attributed to the chemical exchange phenomenon.

A chemical exchange process of an appropriate magnitude could account for the observed NOE between the imino protons of C6·G15 and C7·G14. If one assumes a symmetrical two-site model with equal populations and equal relaxation times for the two sites, the exchange rate can be determined from the ratio of the peak intensities as

k_ex = ln[(1 + R)/(1 – R)]/2τ_m

where k_ex is the exchange rate (s^–1), R is the ratio of the cross-peak to the diagonal peak and τ_m is the mixing time (42,43). Using this relationship, the k_ex was estimated at ∼50 s^–1 using three pairs of cross-peaks from T8/T18-imino, A3-H2/A13-H2 and A3-H8/A13-H8. A similar analysis of our earlier data for the complex of HOO-CoBLM with 1, at very low contours, estimates the k_ex for HOO-CoBLM to be 1 s^–1. The mechanism of strand interchange is probably associated with dissociation of PLM (BLM) from the duplex followed by a 180° re-orientation and re-binding. Thus the differences in k_ex relative to the exchange rates of the imino protons with water provides an explanation for the weak NOE observed between C6·G15 and C7·G14 imino protons in PLM and not BLM.

Structure of the HOO-CoPLM complex with 1

Molecular modeling. The initial models were constructed by manually docking the terminal thiazole ring of PLM between the G14 and G15 base pairs, in either an A- or B-form DNA model. The penultimate thiazolinyl ring was in the minor groove and the metal-binding region was positioned completely outside the minor groove. Modeling was initially carried out on PLM having B-C4′ in the R and S configurations. Given that no quantitative distinction between these models was possible and the stereochemical assignment supported the R configuration (39), only modeling data using the R configuration is presented (Table S5). Each model was energy minimized and submitted to 10 independent runs of the MDSA protocol using the extensive NMR-derived restraints described in Materials and Methods.

The two families of structures derived from A- and B-form DNA were first evaluated to determine if there are quantitative differences resulting from the different starting models. The RMS differences (0.5–1 Å) between the final averaged structures with different starting DNA conformers are less than or equal to the RMS difference within each of the ensembles themselves (Table 3). Thus the final structures are independent of the DNA conformation in the starting model.

Table 3. Summary of constraint violations and potential energy terms in the final ensemble of 10 structures from the MDSA of the complex of HOO-CoBLM with 1 (R configuration at B-C4′).

DNA conformation of initial structure	B	A
RMSD from experimental distance constraints in the 10 ensemble structures (Å)	0.074 ± 0.003	0.076 ± 0.004
Maximum error (Å)	0.48 ± 0.07	0.52 ± 0.08
Σ distance constraint errors (Å)	8.8 ± 0.4	9.0 ± 0.5
RMSD from experimental distance constraints in the final averaged structures (Å)	0.076	0.074
Maximum error (Å)	0.53	0.54
Σ distance constraint errors (Å)	9.1	9.1
Pairwise RMSD of the final 10 structures from the average structure (Å)
All atoms	1.2 ± 0.3	1.4 ± 0.2
‘Core structure’ (non-hydrogen atoms, excluding DNA ends, and CoPLM D, G and M moietiesa)	0.72 ± 0.24	0.84 ± 0.28
CoPLM, excluding D, G and M	0.29 ± 0.07	0.28 ± 0.10

^aDNA nucleotides 3–8 and 13–18, and the bound CoPLM D1 green without the sugars (G and M), or the C-terminus (D).

As shown in Table 3, the structures satisfy the majority of the distance constraints very well, as indicated by the RMSD of the distance constraints from their assigned limit. However, in each structure between two and four errors in excess of 0.4 Å were observed. These results contrast with our previous studies on HOO-CoBLM·1 where between none and two errors over 0.2 Å were observed, with none greater than 0.25 Å (11). In the present study the NOEs associated with the violated constraints have been assigned with confidence; the basis for the inconsistencies are not understood, but may be due to averaging of two or more closely related conformations.

General description of the complex. The complex between HOO-CoPLM and 1 is shown in Figure S5. The terminal thiazole ring is intercalated between the G14 and G15 bases on the 3′ side of the cleavage site. The metal-binding region is positioned in the minor groove and the positively charged D moiety is threaded into the major groove. The pyrimidine ring of PLM makes hydrogen bonds with G5 on the 5′ side of the cleavage site, positioning the terminal hydroperoxide oxygen 2.3 Å from the C6-H4′. All of the interactions that were shown to be critical in defining the complex between HOO-CoBLM and 1 are also present in this complex. These results validate our earlier predictions from molecular modeling studies with HOO-CoPLM based on the HOO-CoBLM structure (15).

Binding mode and sequence specificity. One of the most notable features of the HOO-CoPLM·1 structure is the partial intercalation of the thiazolinyl-thiazole (B) moiety. The 20 NOEs (Table 1) between B and 1 define its position. Figure 5 shows a comparison of the thiazolinyl-thiazole in HOO-CoPLM (top) and the bithiazole in HOO-CoBLM relative to C6·G15 and C7·G14 (bottom) (11). In both cases, the terminal thiazole ring is well stacked with the G14 and G15 bases. The geometry of the DNA at the intercalation site of HOO-CoPLM differs somewhat from the HOO-CoBLM complex. The C6·G15 base pair has shifted relative to the C7·G14 base pair, positioning the imidazole ring of G15 over the terminal thiazole of the complex. This difference may be due to the positioning of the two bulkier sp³ centers at B-C4′ and B-C5′ between the base pairs or it may simply be the result of the structural averaging presented above.

Figure 5.

Comparison of the partial intercalation observed in HOO-CoBLM (11) and HOO-CoPLM (present study).

DNA conformation in the complex. The intercalation of the thiazolinyl-thiazole of HOO-CoPLM between the C6·G15 and C7·G14 base pairs leads to distortion of the DNA conformation and geometry at the intercalation site similar to that observed with HOO-CoBLM bound to 1. The C6·G15∼C7·G14 step is unwound 12 ± 1°, and the C7·G14∼T8·A13 step is unwound 5 ± 3°, leading to a total unwinding of ∼17°. The rise from C6·G15 to C7·G14 calculated by the CURVES program is 7.1 ± 0.2 Å. This distance is a measure of the separation between these two base pairs, and is smaller than the 8.5 ± 0.2 Å observed as a result of the intercalation of HOO-CoBLM in the same sequence. This result is consistent with HOO-CoPLM spending sufficient time in solution to allow detection of an NOE between the C6 and C7 imino protons and is due to the differences in k_ex for PLM relative to BLM. Finally, the contacts between HOO-CoPLM and the DNA on either side of and directly with the C6 deoxyribose results in the perturbation of the sugar pucker to a C4′-exo conformation (pseudorotation angle = 63 ± 5°). This distortion from a typical B-form conformation is similar to that found in HOO-CoBLM·1.

DISCUSSION

Free HOO-CoPLM, relationship to FeBLM

Overall, HOO-CoPLM adopts the same folded structure as HOO-CoBLM. The modeling reveals in both cases that the axial ligands to Co³⁺ include the primary amine of β-aminoalanine, which is in trans to the hydroperoxide. As in the case of the four previous HOO-Co structural models (BLM, deglycoBLM, pepleomycin and deglycopepleomycin) hydrogen bonds between the linker V-NH and T-NH of PLM to the penultimate oxygen of the –OOH appear to stabilize the structure and pre-organize the linker conformation. The H₂O-CoBLM structure stands in contrast to these five hydroperoxo complexes. H₂O-CoBLM has altered coupling constants in the linker backbone and a lack of NOEs between the metal-binding region and the bithiazole moiety, resulting in much less well defined families of structures with respect to linker pre-organization. We have previously argued that conformational flexibility in the linker region of the H₂O- complex relative to the HOO- complex accounted in part for the reduced binding affinity for 1 (0.17 relative to 2 µM). However, HOO-deglycoBLM and HOO-CoPLM both have pre-organized linkers and both bind much more weakly to 1 (5.6 and 16 µM, respectively). Thus while reduction of the B-C4′=B-C5′ double bond in HOO-CoBLM to a single bond in HOO-CoPLM has no impact on the solution structure, the bithiazole tail clearly effects DNA binding, as do the sugars.

In the PLM structure the enhanced dispersion of sugar chemical shifts and, as a consequence, the increased number of assignable NOEs (Table 1 and Fig. 1S) has allowed better modeling of the carbohydrate domain relative to the metal-binding domain than in any other HOO- complex. The modeling suggests the possibility of a hydrogen bond interaction between protons of the metal-bound primary amine and the carbonyl oxygen of the carbamoyl of mannose. Our modeling results may provide an explanation for recent results using circular dichroism, magnetic circular dichroism and a variety of BLM analogs to investigate the coordination of ferrous BLM (44). In all cases the coordination of the primary amine of β-alanine was concluded to be important. However, the simplest interpretation of their studies was that the 3-O-carbamoyl substituent of the mannose sugar is also directly coordinated to the metal ion. To achieve this coordination, the screw sense of the ligands around the iron must be different from cobalt. The HOO-CoBLM derivatives would therefore not be a good model for activated HOO-FeBLM. They provided a second explanation for their spectroscopic observations: that the carbamoyl moiety interacted through a second coordination sphere via hydrogen bonding with the axial ligand. This explanation would allow for the same screw sense of the ligands around Fe as is observed with Co. Thus our PLM modeling studies provide support for their second interpretation, namely that a hydrogen bonding interaction may exist between the carbamoyl oxygen carbonyl and the protons of the amine axial ligand. Further evidence in support of the chiral organization around the metal observed with the HOO-Co analogs is provided by a recent X-ray structure of Cu(II)BLM bound to a BLM resistance protein (38). In this structure the copper ion is penta-coordinated with the primary amine of β-alanine as an axial ligand, while the carbamoyl group of the mannose forms hydrogen bonds with the protein. Superimposition of the metal-binding domain of our CoBLM model on their CuBLM model shows high similarity, the same chirality and a RMS of 0.7 Å. Thus the recent spectroscopic data on FeBLMs, the X-ray data on CuBLM and NMR data on CoBLMs seem to be converging on one screw sense isomer with the primary amine of β-alanine as an essential axial ligand.

The complex of HOO-CoPLM and 1

The similarities in the efficiency of FeBLM- and FePLM-mediated DNA cleavage (45,46) has been repeatedly invoked to support a non-intercalative model of binding for BLM (17,19). Our modeling data on HOO-CoPLM show, however, that despite the lack of planarity of the thiazolinyl ring of B, a partial intercalative mode of binding strikingly similar to HOO-CoBLM occurs with 1. Our determination of a K_d for HOO-CoPLM binding to 1 and determination of a k_ex with 1 also potentially provide an explanation for previous physical biochemical studies that suggested alternative modes of binding between the BLMs and PLMs (47,48). Studies using linear dichroism (orientation in an electric field) showed differences in DNA lengthening between Cu(II)BLM, Fe(III)BLM and Cu(II)PLM (4.6 Å/molecule, and 3.2 Å/molecule and no lengthening, respectively). The poor binding of Cu(II)BLM and of PLM derivatives relative to BLM suggest that Cu(II)PLM might have such weak binding and consequently spend so much time off the DNA that lengthening is not observed. Our re-analysis of old data in light of our structural model for HOO-CoPLM suggests that it can bind and cleave DNA in a similar fashion to HOO-CoBLM; however, the kinetics and thermodynamics differ.

In addition, our current studies provide an explanation for the results on FeBLM- and FePLM-mediated single-strand to double-strand cleavage ratios under single hit conditions using conversion of supercoiled plasmid DNA to nicked circular and linear DNAs (49). While BLM generates a ratio of single-strand to double-strand cleavage of 10:1 or 20:1, no double-strand cleavage was observed by PLM. Given the more rapid k_ex for PLM relative to BLM, the competition between dissociation from the DNA subsequent to the first cleavage event and reorganization to the second strand for the second cleavage event could be much less favorable for PLM. Unfavorable partitioning would result in reduced double-strand cleavage.

Some in vivo studies comparing PLM and BLM have suggested that PLM is much more toxic than BLM (50). However, problems of differential drug uptake, metabolism, availability of metal ions, differential DNA repair and differential cytotoxicity all add to the complexity of data analysis. Studies with PLM in parallel with studies using BLMs where many of the variables enumerated above are measured simultaneously will be an effective probe of the importance of double-strand cleavage in the toxicity of various BLMs.

In conclusion, the results reported clearly support the ability of PLM to bind by partial intercalation. Therefore our model explaining the sequence specificity and chemical specificity of DNA cleavage for BLM is also valid for PLM. The kinetics of intercalation of PLM also explain the substantially reduced levels of double-strand cleavage observed compared to BLM. If double-strand cleavage is essential to the cytotoxicity of BLM, then the the cytoxicity of PLM would either require conversion to a species with enhanced double-strand cleavage potential or some other mechanism of lethality.

SUPPLEMENTARY MATERIAL

Supplementary Material is available at NAR Online.