DNA Interstrand Cross-links of the Novel Antitumor Trinuclear Platinum Complex BBR3464

The novel phase II antitumor polynuclear platinum drug BBR3464 ([(trans-PtCl(NH3)2)2(μ-trans-Pt(NH3)2(NH2(CH2)6NH2)2)](NO3)4) forms intra- and interstrand cross-links (CLs) on DNA (which is the pharmacological target of platinum drugs). We examined first in our recent work how various intrastrand CLs of BBR3464 affect the conformation of DNA and its recognition by cellular components (Zehnulova, J., Kasparkova, J., Farrell, N., and Brabec, V. (2001) J. Biol. Chem. 276, 22191–22199). In the present work, we have extended the studies on the DNA interstrand CLs of this drug. The results have revealed that the interstrand CLs are preferentially formed between guanine residues separated by 2 base pairs in both the 3′ → 3′ and 5′ → 5′ directions. The major 1,4-interstrand CLs distort DNA, inducing a directional bending of the helix axis and local unwinding of the duplex. Although such distortions represent a potential structural motif for recognition by high mobility group proteins, these proteins do not recognize 1,4-interstrand CLs of BBR3464. On the other hand, in contrast to intrastrand adducts of BBR3464, 1,4-interstrand CLs are not removed from DNA by nucleotide excision repair. It has been suggested that interstrand CLs of BBR3464 could persist considerably longer in cells compared with intrastrand adducts, which would potentiate the toxicity of the interstrand lesions to tumors sensitive to this polynuclear drug.

It is now well established that polynuclear platinum compounds in which two or three platinum coordination units are linked in a linear fashion compose an important new class of anticancer drugs [1,2]. The first clinical compound, currently denoted BBR3464 (see Fig. 1A), is a bifunctional trinuclear DNA-binding agent with an overall 4ϩ charge (3). The phase I trials (4,5) demonstrated a clear pattern of responses in can-cers not normally treatable with cisplatin (cis-diamminedichloroplatinum(II)) (see Fig. 1A), including responses in melanoma and pancreatic and lung cancer. Objective responses in phase II were verified in relapsed ovarian cancer and non-small cell lung cancer (6,7). Pre-clinical studies indicated activity in p53 mutant tumors and a minimum induction of p53 following BBR3464 treatment. The interactions of antitumor polynuclear platinum compounds with target DNA (for reviews, see Refs. 1, 2, and 8) are distinct from those of the mononuclear based cisplatin family and, indeed, unlike those of any DNA-damaging agent in clinical use. Hence, it is important to understand these novel interactions to the greatest extent possible and how they affect DNA structure and function to exploit the full clinical potential of these new agents.
Bifunctional binding of BBR3464 to double-stranded DNA preferentially involves guanine residues and is characterized by the rapid formation of long-range intra-and interstrand cross-links (CLs) 1 in which the platinated sites are separated by 1 or more bp (9). Quantitation of DNA interstrand crosslinking in natural and linear DNAs indicated ϳ20% of the DNA to be interstrand cross-linked. This value is significantly higher than that for cisplatin (10,11); on the other hand, an intriguing aspect of BBR3464 is that long-range delocalized intrastrand adducts are equally or even more probable than interstrand CLs (9). As the [Pt,Pt] intrastrand CLs of BBR3464 are analogs of the major adducts of cisplatin, which forms ϳ90% bifunctional intrastrand adducts between neighboring purine residues on DNA, we examined first in our recent work (12) how the structures of the various types of the intrastrand CLs of BBR3464 affect conformational properties of DNA and how these adducts are recognized by the HMGB1 protein and removed from DNA during in vitro nucleotide excision repair (NER) reactions. These studies were performed because some structures altered by platinum adducts, such as stable directional bending and unwinding, attract various damaged DNAbinding proteins such as those containing HMG domains (13)(14)(15). The binding of these proteins has been postulated to mediate the antitumor properties of the platinum drugs (14,15). In addition, several reports have demonstrated that intrastrand CLs of cisplatin are removed from DNA during NER reactions and that NER is also a major mechanism contributing to cisplatin resistance (16 -18). Recent work has revealed that intrastrand CLs of BBR3464 create a local conformational distortion, but that none of these intrastrand adducts results in a stable curvature such as that observed for the major 1,2-intrastrand CL of cisplatin (12). In addition, we have observed also, in contrast to the 1,2-GG intrastrand CL of cisplatin, no recognition of the intrastrand CLs of BBR3464 by HMGB1 proteins, but we have observed effective removal of these adducts from DNA by NER. We have suggested that the processing of the intrastrand CLs of BBR3464 in tumor cells sensitive to this drug may not be relevant to its antitumor effects because these CLs do not persist to potentiate the anticancer effect of this drug. This also implies that polynuclear platinum compounds apparently represent a novel class of platinum anticancer drugs acting by a different mechanism than cisplatin and its analogs.
On the other hand, BBR3464 also forms interstrand CLs on DNA with a considerably higher frequency compared with cisplatin (9). In general, DNA interstrand CLs could be even more effective lesions than intrastrand adducts in terminating DNA or RNA synthesis in tumor cells and thus could be even more likely candidates for the genotoxic lesion relevant to the antitumor effects of BBR3464. In addition, the interstrand CLs pose a special challenge to repair enzymes because they involve both strands of DNA and would not be repaired so readily as intrastrand lesions because these repair enzymes would require the information in the complementary strand for resynthesis. In this way, the interstrand CLs might persist considerably longer than intrastrand adducts, which would result in a higher cytotoxicity of the interstrand lesions. Interestingly, the interstrand CLs formed by cisplatin, which still remain important candidates for a genotoxic lesion relevant to the antitumor effects of this platinum drug, preferentially form between guanine residues in the 5Ј-GC/5Ј-GC sequence. This adduct induces several irregularities in DNA (19 -21). The guanine residues interstrand cross-linked by cisplatin are not paired with hydrogen bonds to the complementary cytosines, which are located outside the duplex and not stacked with other aromatic rings. All other base residues are paired, but distortion extends over at least 4 bp at the site of the CL. In addition, the cis-diammineplatinum(II) bridge resides in the minor groove, and the double helix is locally reversed to a left-handed, Z-DNA-like form. This adduct induces the helix unwinding by 76 -80°relative to B-DNA and also the bending of 20 -40°of the helix axis at the cross-linked site toward the minor groove. In addition, the interstrand CL of cisplatin is specifically recognized by the full-length HMGB1 protein (22) and its domain B 2 and is not removed by the NER system under conditions in which the 1,2-GG intrastrand CL is readily excised (16).
Cellular pharmacology studies in L1210 and osteosarcoma cells using alkaline elution show the persistence of interstrand CLs with time, consistent with a slower rate of repair (23,24). Hence, data on conformation, recognition by HMGB1 domain proteins, and NER of DNA interstrand CLs of BBR3464 are needed to obtain greater insight into which DNA adduct of BBR3464 is the more likely lesion responsible for the antitumor effects of this polynuclear platinum drug.
In this work, we continue our investigations of the DNA interactions of BBR3464 in cell-free medium to address further fundamental questions about the mechanism of antitumor activity of this novel platinum drug. In a previous report (9) using an indirect assay employing transcription mapping of the DNA fragment globally modified by BBR3464, some sites in DNA involved in the interstrand CLs have been already suggested. However, no systematic study on the sequence specificity of these lesions has been performed. Therefore, in this work, we examined in detail which sites are preferentially involved in the interstrand CLs of BBR3464. In addition, we also studied how interstrand CLs of BBR3464 affect the local conformation of DNA (in particular, bending and unwinding) and how these adducts are recognized by the HMGB1 domain proteins and removed from DNA during in vitro NER reactions.
Platination of Oligonucleotides-The single-stranded oligonucleotides (the top strands of the duplexes in Fig. 1B, except [TGGT] and [GTG(NER)]) were reacted in stoichiometric amounts with the monoaqua derivative of BBR3464. The platinated oligonucleotides were repurified by ion-exchange fast protein liquid chromatography (FPLC). We verified that the modified oligonucleotides contained three platinum atoms by platinum flameless atomic absorption spectrophotometry and by measurement of the optical density. Using DMS footprinting of platinum on DNA (10,29,30), we also verified that in the platinated top strands of all duplexes, N-7 of a single guanine (G) residue was not accessible for reaction with DMS. The oligonucleotides were then treated with hot piperidine and analyzed by denaturing 24% PAGE. For the unmodified oligonucleotides, shortened fragments due to cleavage of the strand at one methylated G residue were observed on the gel. However, no such bands were detected for the oligonucleotides modified by BBR3464. These results indicate that one BBR3464 molecule was coordinated to a single G residue in the top strands of all duplexes. The platinated top strands were allowed to anneal with unplatinated complementary strands (the bottom strands in Fig. 1B) in 0.1 M NaClO 4 . The resulting products were analyzed by FPLC in an alkaline gradient. Using this denaturing gradient, non-interstrand cross-linked strands were eluted as a 20 -23-nucleotide single strand, whereas the interstrand cross-linked strands were eluted later in a single peak as a higher molecular mass species. Only this single peak was collected so that the samples of the interstrand cross-linked duplexes contained no single-stranded molecules. Alternatively, the products were separated on denaturing 8 M urea and 12% polyacrylamide gel; the bands corresponding to interstrand cross-linked duplexes were analyzed by densitometry or were cut off from the gel, eluted, precipitated by ethanol, and dissolved in 50 mM NaCl. Both procedures for the purification of interstrand cross-linked duplexes provided products whose subsequent analysis (see below) gave identical results. FPLC purification and flameless atomic absorption spectrophotometry measurements were carried out in an Amersham Biosciences FPLC system with a Mono Q HR 5/5 column and with a Unicam 939 AA spectrometer equipped with a graphite furnace, respectively. Duplexes [TGGT] and [GTG(NER)] (see Fig. 1B) containing a single 1,2-GG or 1,3-GTG intrastrand CL of BBR3464 or cisplatin in the top strand were prepared as described (10, 12, 29 -32). The unmodified or intrastrand CLs containing duplexes used in the studies of recognition by HMGB1 domain proteins and RPA were purified by electrophoresis on native 15% polyacrylamide gel (monoacrylamide/bisacrylamide ratio of 29:1). Other details have been described previously (10,25,33).
Chemical Modifications-The modifications by KMnO 4 , DEPC, and KBr/KHSO 5 were performed as described previously (33)(34)(35)(36). The strands of the duplexes were 5Ј-end-labeled with [␥-32 P]ATP. In the case of the platinated oligonucleotides, the platinum complex was removed after reaction of the DNA with the probe by incubation with 0.2 M NaCN (pH 11) at 45°C for 10 h in the dark.
Ligation and Electrophoresis of Oligonucleotides-Unplatinated single strands (top strands in Fig. 1B) and the duplexes containing a unique interstrand CL were 5Ј-end-labeled with [␥-32 P]ATP using T4 polynucleotide kinase. The single-stranded, unplatinated oligonucleotides were subsequently annealed with their phosphorylated complementary strands. Unplatinated or interstrand CL-containing duplexes were allowed to react with T4 DNA ligase. The resulting samples along with ligated unplatinated duplexes were subsequently examined on native 8% polyacrylamide gel (monoacrylamide/bisacrylamide ratio of 29: 1). Other details of these experiments were as described in previously (37)(38)(39).
Gel Mobility Shift Assay-The 21-mer oligonucleotides (bottom strands of duplexes [1,4;3Ј-3Ј (21)] and [1,4;5Ј-5Ј (21)] shown in Fig. 1B to which one adenine residue was added to their 3Ј-ends and one adenine residue was removed from their 5Ј-ends) were 5Ј-end-labeled and annealed (see above) to their complementary strands (top strands of duplexes [1,4;3Ј-3Ј (21)] and [1,4;5Ј-5Ј (21)] shown in Fig. 1B), which were either unplatinated (controls) or contained monofunctional adducts of BBR3464 on the central G residue. The platinated duplexes, which had blunt ends, were further incubated for 24 h in 0.1 M NaClO 4 at 37°C to form interstrand CLs. The oligonucleotides containing interstrand CLs were separated by electrophoresis on denaturing 12% polyacrylamide gel, isolated from the gel by elution, extracted by phenol/chloroform and chloroform, precipitated by ethanol, and used as DNA probes in the experiments with HMGB1a and HMGB1b proteins and RPA. The 149-bp probes with blunt ends used in the experiments with the full-length HMGB1 protein were prepared in the same way as those used in the NER assays (see below).
Reactions with the Full-length HMGB1 Protein and Its Domains-The unplatinated and platinated duplexes (0.4 nM) were incubated with increasing concentrations of HMGB1a or HMGB1b protein in 20-l sample volumes containing 10 mM HEPES (pH 7.5), 10 mM MgCl 2 , 50 mM LiCl, 100 mM NaCl, 1 mM spermidine, 0.2 mg/ml bovine serum albumin, and 0.05% Nonidet P40. Samples were incubated on ice for 30 min and then brought to 7% in sucrose and 0.017% in xylene cyanol prior to loading on prerun and precooled (4°C) native 6% polyacrylamide gels (monoacrylamide/bisacrylamide ratio of 29:1). The gels were electrophoresed for 3 h and visualized using an Amersham Biosciences PhosphorImager (Storm 860 system), and the bands were quantitated with ImageQuant software. The experiments with the full-length HMGB1 protein were performed in the same way, but with minor modifications. The labeled 149-bp DNA probes were titrated with the protein in the presence of 0.1 mg/ml sonicated calf thymus DNA in buffer composed of 10 mM HEPES (pH 7.9), 10 mM MgCl 2 , 150 mM NaCl, 0.2 mg/ml bovine serum albumin, 20% (v/v) glycerol, and 1 mM dithiothreitol.
Reactions with RPA-32 P-Labeled DNA substrates (0.5 nM) and the amounts of RPA indicated were incubated at 20°C in 20-l reactions containing 25 mM HEPES-KOH (pH 8.3), 30 mM KCl, 4 mM MgCl 2 , 1 mM EDTA, 0.9 mM dithiothreitol, 45 g/ml bovine serum albumin, and 10% glycerol. To assess binding at equilibrium, reactions were stopped after 30 min by cooling the samples to 0°C. Following addition of gel loading buffer (4 l) containing 100 mM Tris-HCl (pH 8.3), 10% glycerol, and 0.05% orange G, the extent of binding was determined on native 6% polyacrylamide gels. Electrophoresis was performed for 50 min at 4°C; gels were dried and visualized using the Amersham Biosciences Phos-phorImager (Storm 860 system); and the radioactivities associated with bands were quantitated with ImageQuant software.
The intrastrand cross-linked substrates were prepared in the similar following way. The 21-mer oligonucleotide (the top strand of duplex [GTG(NER)] shown in Fig. 1B) was used for preparation of linear 149-bp duplexes with the centrally located 1,3-GTG intrastrand CL of BBR3464 or cisplatin. Uniquely modified 21-mers were end-labeled to introduce a radiolabel at the 11th phosphodiester bond 5Ј to the CL and annealed with the bottom strand of duplex [GTG(NER)]. Two duplexes (arms) with random base pair sequences with overhangs partially overlapping those of the central platinated duplex were ligated to this intrastrand cross-linked duplex (one to each side) by T4 DNA ligase.
Oligonucleotide excision reactions were performed in cell-free extracts (CFEs) prepared from the HeLa S3 and Chinese hamster ovary AA8 cell lines as described (17,42). These extracts were kindly provided by J. T. Reardon and A. Sancar (University of North Carolina, Chapel Hill, NC). In vitro repair of interstrand CLs of BBR3464 was measured in an excision assay using these CFEs and 149-bp linear DNA substrates (see above) as described previously (17) with minor modifications. The reaction mixtures (25 l) contained 10 fmol of radiolabeled DNA, 50 g of CFE, 20 M dATP, 20 M dCTP, 20 M dGTP, and 20 M TTP in reaction buffer (23 mM HEPES (pH 7.9), 44 mM KCl, 4.8 mM MgCl 2 , 0.16 mM EDTA, 0.52 mM dithiothreitol, 1.5 mM ATP, 5 g of bovine serum albumin, and 2.5% glycerol) and were incubated at 30°C for 40 min. DNA was deproteinized and precipitated by ethanol. Reaction products were treated overnight with 0.4 M NaCN (pH 10 -11) at 45°C and precipitated by ethanol prior to resolution on the gels. The excision products were separated on denaturing 10% polyacrylamide gels and visualized using the Amersham Biosciences PhosphorImager (Storm 860 system).

RESULTS
Sequence Specificity of Interstrand Cross-linking-We demonstrated in our previous work (9) that the preferential DNAbinding sites of BBR3464 are G residues and that along with the delocalized intrastrand CLs, the long-range interstrand CLs are the most frequent lesions produced by this drug on DNA. In these CLs, the platinated sites are separated by 1 or more base pairs. Bifunctional BBR3464 must bind to DNA first through one end of the trinuclear unit. In this first step of binding, the kinetic preferences are similar to those of mononuclear species, i.e. they coordinate preferentially to N-7 atoms of G residues. The array of adducts becomes, however, different upon the coordination of the second platinum unit (9). Considering these facts and our intention to investigate, in the present work, interstrand CLs of BBR3464, we have designed a series of synthetic oligodeoxyribonucleotide duplexes, [1,2], [1,4], and [1,6], whose sequences are shown in Fig. 1B. The pyrimidine-rich top strands of these duplexes contained only one G residue in the center (shown in boldface in Fig. 1B). These top strands were modified by BBR3464 so that they contained a single monofunctional adduct of this platinum complex at the central G site. The duplexes were also designed in such a way that their bottom (complementary) strands contained a G residue in different positions symmetrical to the single central cytosine (C) residue (complementary to the platinated G residue in the top strand). In this way, the G residue in the top strand with the monofunctionally attached BBR3464 may close to the 1,2-, 1,4-, or 1,6-GG interstrand CL in duplex [1,2], [1,4], or [1,6], respectively. The 1,2-interstrand CL is formed between G sites in neighboring base pairs, whereas in 1,4-and 1,6-interstrand CLs, the platinated G sites are separated by 2 and 4 base pairs, respectively. G sites in the bottom strands involved in these interstrand CLs are also shown in boldface in Fig. 1B. The nucleotide sequences of the duplexes were also designed in such a way that these interstrand CLs may close to the G residues located on both sides of the central C residue in the bottom strands, i.e. in the 5Ј 3 5Ј or 3Ј 3 3Ј direction. The orientation of the interstrand CL in the 5Ј 3 5Ј or 3Ј 3 3Ј direction can be explained with the aid of the sequence of duplex [1,4] (Fig. 1B). For instance, the 1,4-GG interstrand CL oriented in the 5Ј 3 5Ј direction is that formed in duplex [1,4] between the central G residue in the top strand and G 8 in the bottom strand, whereas the same CL oriented in the 3Ј 3 3Ј direction is that between the central G residue in the top strand and G 14 in the bottom strand.
The monoadducted top strands of duplexes [1,2], [1,4], and [1,6] were hybridized with their complementary strands, and the hybrids were incubated in 0.1 M NaClO 4 at 37°C. The aliquots were withdrawn at various time intervals and analyzed by gel electrophoresis under denaturing conditions. As shown in Fig. 2A for duplex [1,4] modified by BBR3464, only one band was observed for the non-cross-linked duplex. The subsequent incubation resulted in new bands migrating markedly more slowly. Their intensity increased with the incubation time, with a concomitant decrease in the intensity of the band corresponding to the non-cross-linked duplex. This observation can be interpreted to mean that interstrand CLs were formed. From the ratio of the sum of intensities of all bands corresponding to cross-linked duplexes to the sum of intensities of all bands, the percentage of interstrand CLs was calculated (Fig.  2B). The highest rate of this interstrand cross-linking reaction was observed in duplex [1,4]. The CLs in duplex [1,2] were formed with a slightly lower rate, whereas the rate of the cross-linking reaction in duplex [1,6] was markedly lower.
The cross-linking reactions in duplexes [1,2], [1,4], and [1,6] resulted in only one band or single peak in the FPLC profile designated as interstrand (shown for duplex [1,4] in Fig. 2A). After a 24-h reaction period, the bands or FPLC peaks corresponding to the interstrand cross-linked duplexes were cut off from the gel or collected after FPLC separation, respectively; the duplexes were isolated (eluted from the bands and precipitated or only precipitated from the FPLC fractions) and further characterized by Maxam-Gilbert footprinting (10,29,43).
The samples of duplexes [1,2], [1,4], and [1,6] interstrand cross-linked by BBR3464 in which the top strand was only 5Ј-end-labeled with 32 P were reacted with DMS, which does not react with platinated G residues because N-7 is no longer accessible (10,29,43). The adducts were removed by NaCN (29,30), and then the sample was treated with piperidine. In the unplatinated duplexes, the central G residue in the top strands was reactive with DMS (data not shown). It was no longer reactive in all three cross-linked duplexes. This observation confirms that the unique G residue in the top strands remained platinated and was involved in the interstrand CL contained in the single fraction of interstrand cross-linked duplexes (10,29,43).
In additional studies, the 1,2-, 1,4-, and 1,6-interstrand cross-linked duplexes in which the bottom strand was 5Ј-endlabeled with 32 P were examined (Fig. 3). The interstrand crosslinked duplexes were reacted with DMS. These samples were then further treated with NaCN to remove the adducts and finally also with piperidine. The treatment with piperidine of the control unplatinated duplexes resulted in cleavage at all G sites in the bottom strand (Fig. 3, NoPt lanes). If the crosslinked duplexes treated with DMS and subsequently NaCN were cleaved (Fig. 3, Pt/CN Ϫ lanes), bands corresponding to all G residues were observed that had the same intensity as the corresponding bands seen for unplatinated duplexes, except for the G residues marked by arrows in Fig. 3. This result proves that these G residues were platinated and involved in the interstrand CL of BBR3464. Interestingly, the band corresponding to G 12 in the bottom strand of interstrand crosslinked duplex [1,2] entirely disappeared, whereas the bands corresponding to other G residues in the bottom strand had the same intensity as the corresponding bands seen for the unplatinated duplex. This result implies that the interstrand CLs were formed by BBR3464 exclusively in the 3Ј 3 3Ј direction. Somewhat different results were obtained if interstrand crosslinked duplexes [1,4] and [1,6] were analyzed. As shown in  [1,2], [1,4], and [1,6], the numbering of the nucleotide residues in the bottom strand is also shown.

FIG. 2. Kinetics of DNA interstrand CL formation by BBR3464.
Duplexes [1,2], [1,4], and [1,6] were formed by mixing their bottom strands with the complementary top strand uniquely monoadducted by BBR3464 at the central G residue at 37°C. A, autoradiogram of a denaturing 8 M urea and 12% polyacrylamide gel of duplex [1,4] with a 32 P-end-labeled bottom strand. The cross-linking reaction was stopped by adjusting the NaOH concentration to 10 mM and cooling the samples to Ϫ70°C. inter designates bands corresponding to the interstrand cross-linked fraction (see also "Results"); SS indicates the unplatinated strand. one-half of the intensities observed for the corresponding bands seen for the unplatinated duplex. This result can be interpreted to mean that the 1,4-interstrand CLs of BBR3464 are formed with approximately the same preference in both the 3Ј 3 3Ј and 5Ј 3 5Ј directions. Similarly, in the case of duplex [1,6], the intensities of the bands corresponding to the two G residues at positions 7 and 17 of the bottom strand of duplex [1,6] were affected only by formation of the interstrand CL of BBR3464, but to a different extent (Fig. 3C). Whereas the intensity of the band corresponding to G 17 was reduced by only ϳ30%, the intensity of the band corresponding to G 7 was reduced more pronouncedly by ϳ70%. This result indicates that the 1,6interstrand CLs are also formed by BBR3464 in both directions, but with a clear preference for formation of the CL in the 5Ј 3 5Ј direction. Thus, the results shown in Fig. 3 demonstrate that although a short 1,2-interstrand CL of BBR3464 is formed preferentially in the 3Ј 3 3Ј direction, the growing length of these CLs reverses this preference to the opposite 5Ј 3 5Ј direction. Importantly, the sequences of the bottom strands in duplexes [1,2] and [1,4] were chosen so that the 1,7-and 1,8interstrand CLs could be also formed in both directions, but DMS footprinting of the platinated sites in these duplexes interstrand cross-linked by BBR3464 revealed no 1,7-or 1,8interstrand CLs (Fig. 3, A and B). Similarly, no 1,8-interstrand CL was formed in interstrand cross-linked duplex [1,6], in which this type of CL was also possible (Fig. 3C). Thus, the maximum possible length of the long-range interstrand CL of BBR3464 corresponds to the CL in which the platinated sites are separated by 4 base pairs.
Chemical Probes of DNA Conformation-Among the interstrand adducts, 1,4-interstrand CLs represent the DNA lesion formed most readily by BBR3464. Therefore, additional experiments in this work were focused mainly on this type of CL. The oligonucleotide duplexes containing a single 1,4-interstrand CL between G residues oriented in both the 3Ј 3 3Ј and 5Ј 3 5Ј  Fig.  1B, respectively) were further analyzed with chemical probes of DNA conformation. The interstrand cross-linked duplexes were treated with several chemical agents that are used as tools for monitoring the existence of conformations other than canonical B-DNA. These agents include KMnO 4 , DEPC, and bromine. They react preferentially with single-stranded DNA and distorted double-stranded DNA (33)(34)(35)(36)44). As we used for this analysis exactly the same methodology described in detail in our recent work aimed at DNA intrastrand CLs of BBR3464 (12), the results are discussed briefly.
KMnO 4 is hyperreactive with thymine (T) residues in singlestranded nucleic acids and in distorted DNA compared with B-DNA (34,36,45,46). The interstrand cross-linked duplexes showed strong reactivity for the two 5Ј-T residues adjacent to the adduct (shown for duplex [1,4;3Ј-3Ј] in Fig. 4A, ICL lane). A somewhat weaker reactivity was also observed for the 3Ј-T residue adjacent to the platinated G residue involved in the CL.
DEPC carbetoxylates purines at N-7. It is hyperreactive with unpaired and distorted adenine (A) residues in DNA and with left-handed Z-DNA (34,36,47,48). Within the double-stranded oligonucleotides containing the interstrand CL, four base residues in the bottom strand of duplex [1,4;3Ј-3Ј] became reactive (Fig. 4C, ICL lane). The strongly reactive residues were readily identified as the two A residues complementary to the strongly reactive T residues of the top strand. The other two more weakly reactive residues were the platinated G residue and adjacent 3Ј-A residue. A different reactivity of DEPC was observed with duplex [1,4;5Ј-5Ј] containing the interstrand CL formed by BBR3464 in the 5Ј 3 5Ј direction. The 3Ј-A residue adjacent to the platinated G residue in the bottom strand was only reactive (data not shown).
Bromination of C residues and formation of piperidine-labile sites are observed when two simple salts, KBr and KHSO 5 , are allowed to react with single-stranded or distorted doublestranded oligonucleotides (35). No reactivity of these residues was observed within the unplatinated duplexes (shown for the top and bottom strands of duplex [1,4;3Ј-3Ј] in Fig. 4, B and D,  respectively, lanes DS). Within the double-stranded duplexes containing the interstrand CL of BBR3464, no C residue in both strands, including those complementary to the G residues involved in the CL, was reactive (shown for the top and bottom strands of duplex [1,4;3Ј-3Ј] in Fig. 4, B and D, respectively,  ICL lane). The results of the analysis of duplexes [1,4;3Ј-3Ј] and [1,4;5Ј-5Ј] containing 1,4-GG interstrand CLs formed by BBR3464 in the 3Ј 3 3Ј and 5Ј 3 5Ј direction, respectively, by chemical probes are summarized in Fig. 4E.
DNA Unwinding and Bending-Among the alterations of secondary and tertiary structures of DNA to which it may be subject, the role of intrinsic bending and unwinding of DNA is increasingly recognized as of potential importance in regulating replication and transcription functions through specific DNA-protein interactions. For DNA adducts of cisplatin, the structural details responsible for bending and subsequent protein recognition have recently been elucidated (14,15). Given the recent advances in our understanding of the structural basis for the bending of DNA caused by cisplatin CLs, it is of considerable interest to examine how frequent DNA adducts of BBR3464 (interstrand CLs) affect conformational properties of DNA such as bending and unwinding. In this work, we performed studies on the bending and unwinding induced by single, site-specific 1,4-GG interstrand CLs of BBR3464 formed in the oligodeoxyribonucleotide duplexes in the 3Ј 3 3Ј and 5Ј 3 5Ј directions using electrophoretic retardation as a quantitative measure of the extent of planar curvature.
The oligodeoxyribonucleotide duplexes [1,4;3Ј-3Ј] and [1,4; 5Ј-5Ј] (20 -23 bp, whose sequences were identical (21 bp) or similar to those of duplexes [1,4;3Ј-3Ј(21)] and [1,4;5Ј-5Ј(21)] shown in Fig. 1B; the 20-bp duplexes had one marginal C⅐G pair deleted, whereas one and two additional A⅐T pairs were added to the ends in the 22-and 23-bp duplexes, respectively) were used for the bending and unwinding studies in this work. All sequences were designed to leave a one-nucleotide overhang at their 5Ј-ends in double-stranded form (shown for duplexes [1,4;3Ј-3Ј(21)] and [1,4;5Ј-5Ј(21)] in Fig. 1B). These overhangs facilitate polymerization of the monomeric oligonucleotide duplexes by T4 DNA ligase in only one orientation and maintain a constant interadduct distance throughout the resulting multimer. Other experimental details of these studies are given in our recent report describing DNA intrastrand CLs of BBR3464 (12). Autoradiograms of electrophoresis gels revealing resolution of the ligation products of unplatinated 20 -23-bp duplexes [1,4;5Ј-5Ј(20 -23)] or containing a unique 1,4-GG interstrand CL of BBR3464 in the 5Ј 3 5Ј direction are shown in Fig. 5A. A small but significant retardation was observed for the multimers of all platinated duplexes. The K factor is defined as the ratio of calculated to actual length. The variation of the K factor versus sequence length obtained for multimers of the duplexes 20 -23 bp long and containing the unique 1,4-GG interstrand CL of BBR3464 in the 5Ј 3 5Ј direction is shown in Fig. 5B. Maximum retardation was observed for the 22-bp cross-linked duplex. This observation suggests that the natural 10.5-bp repeat of B-DNA and that of DNA perturbed by the interstrand CL of BBR3464 are different as a consequence of DNA unwinding (49). Similar results were also obtained for the 1,4-GG interstrand CL of BBR3464 formed in the 3Ј 3 3Ј direction in duplexes [1,4;3Ј-3Ј(20 -23)] (data not shown).
The unwinding angles were calculated from the exact helical repeat of the 1,4-interstrand cross-linked duplexes (49). The maxima of these curves constructed for the interstrand crosslinked duplexes with a total length of 150 bp (shown for the CL in the 5Ј 3 5Ј direction in Fig. 5C) were determined to be 21.41 Ϯ 0.04 and 21.42 Ϯ 0.04 bp for the CLs in the 3Ј 3 3Ј and 5Ј 3 5Ј directions, respectively. Total sequence lengths other than 150 bp were examined and gave identical results. To convert the interadduct distance in base pairs corresponding to the curve maximum into a duplex unwinding angle in degrees, the value was compared with that of the helical repeat of B-DNA, which is 10.5 Ϯ 0.05 bp (50,51). The difference between the helical repeat of B-DNA and DNA containing the 1,4-interstrand CL of BBR3464 formed in the 3Ј 3 3Ј direction is therefore ((21.41 Ϯ 0.04) Ϫ 2(10.5 Ϯ 0.05)) ϭ 0.41 Ϯ 0.09 bp. There are 360°/10.5 bp, so the DNA unwinding due to one 1,4-interstrand adduct of BBR3464 formed in the 3Ј 3 3Ј direction is 14 Ϯ 3°. The same calculation for the 1,4-interstrand CL formed in the 5Ј 3 5Ј direction yielded the same value for the unwinding angle. This unwinding angle is relatively very small, in contrast to that found, for example, for the 1,2-GG interstrand CL of cisplatin using the same experimental procedure (79° (52) or ϳ90° (19)), but not markedly different from that produced by the 1,3-GG interstrand CLs formed by the dinuclear antitumor platinum compound [(trans-Pt-Cl(NH 3 ) 2 ) 2 (-H 2 N(CH 2 ) n NH 2 )]Cl 2 (n ϭ 2-6) (9°(32)).
The evaluation of the relationship between interadduct distance and phasing for self-ligated multimers composed of the identical number of monomeric duplexes (bend units) resulted in a bell-shaped pattern (shown for the CL in the 5Ј 3 5Ј direction in Fig. 5C)  the bend angles of the 1,4-interstrand CLs of BBR3464 was performed as described previously (19,33,38,53,54) utilizing the following empirical equation (Equation 1), where L represents the length of a particular oligomer with relative mobility K, and RC is the curvature relative to DNA bending induced at the tract of A residues (A tract) (53,55). Application of Equation 1 to the 110-, 132-, and 154-bp multimers of the 22-bp oligomer containing the single 1,4-interstrand CL of BBR3464 leads to a mean curvature of 0.53 or 0.38, respectively, relative to the A tract. The average bend angle per helix turn can be calculated by multiplying the relative curvature by the absolute value of the A tract bend of 20° ( 38,53,55,56). The results indicate that the bends induced by the 1,4-GG interstrand CLs formed by BBR3464 in the 3Ј 3 3Ј and 5Ј 3 5Ј directions are 21°and 15°, respectively. We assigned the bend direction by reference to an A tract, which is bent by ϳ20°toward the minor groove (55), using the same procedure described previously (19). Duplexes [1,4;3Ј-3Јϩ(A/ T) 5 (33)] and [1,4;5Ј-5Јϩ(A/T) 5 (33)] (Fig. 1B) were used, which also contained, besides the single 1,4-interstrand CL of BBR3464 formed in the 3Ј 3 3Ј and 5Ј 3 5Ј directions, respectively, the A tract located "in phase" from the CL (the center of the sequence involved in the CL and the center of the A tract were separated by 11 bp) (Fig. 1B). In the cross-linked multimers, the CLs or the A tracts were separated by 33 bp, corresponding to about three helical turns after incorporation of the estimated 14°of unwinding at the lesion (see above). The cross-linked multimers of duplexes [1,4;3Ј-3Јϩ(A/T) 5 (33)] and [1,4;5Ј-5Јϩ(A/T) 5 (33)] migrated on the gel in all cases more rapidly than their unplatinated counterparts (data not shown). Hence, the effective bend of the helix axis at the center of the 1,4-interstrand CLs of BBR3464 is in the opposite direction from that at the center of the A tract, i.e. the 1,4-GG interstrand CLs formed by BBR3464 in both the 3Ј 3 3Ј and 5Ј 3 5Ј directions bend DNA toward the major groove. Other details of the calculations of the unwinding and bending angles are given in previous studies (19,31,33,38,(52)(53)(54).
Recognition by the Domains of the HMGB1 Protein-The bending of the helix axis induced in DNA by 1,2-intrastrand and 1,2-interstrand CLs of cisplatin and the altered structures attract HMG domain proteins and other proteins (8,13,22,57). This binding of HMG domain proteins to DNA modified by cisplatin has been postulated to mediate its antitumor properties (14,58). As bifunctional BBR3464 and other polynuclear platinum compounds exhibit antitumor activity different from that of cisplatin, it was of considerable interest to examine how the interstrand CLs of BBR3464 are recognized by HMG domain proteins. The interactions of the rat HMGB1 protein, which is the prototypical member of a family of these proteins, and its domains A and B with 1,4-interstrand CLs of BBR3464 were investigated by gel mobility shift experiments (Fig. 6). In the experiments with the domains, the 21-bp duplexes [1,4;3Ј-3Ј(21)] and [1,4;5Ј-5Ј(21)] with blunt ends (see "Experimental Procedures" for their exact sequences) were modified so that they contained a single, site-specific 1,4-GG interstrand CL formed by BBR3464 in the 3Ј 3 3Ј and 5Ј 3 5Ј directions, respectively; or for comparative purposes, also duplex [TGGT] (Fig. 1B) was modified so that it contained a single, site-specific 1,2-GG intrastrand CL of cisplatin. The binding of HMGB1a and HMGB1b to these DNA probes was detected by retardation of the migration of the radiolabeled 21-bp probes on the gel (Fig. 6) (14, 59, 60).
HMGB1a and HMGB1b exhibited negligible binding to the unmodified 21-bp duplexes. As indicated by the presence of a shifted band whose intensity increased with increasing protein concentration (Fig. 6, A and B, lanes 5-8), both HMGB1a and HMGB1b recognized and bound to the duplex containing the 1,2-GG intrastrand CL of cisplatin. The results of the titration of the duplexes containing the 1,4-GG interstrand CL of BBR3464 in the 3Ј 3 3Ј and 5Ј 3 5Ј directions with HMGB1a and HMGB1b are shown in Fig. 6 (A and B, respectively). These titration data indicate that neither HMGB1a nor HMGB1b bound the probes containing the interstrand CL of BBR3464 under conditions in which they exhibited a high affinity for the 1,2-GG intrastrand CL of cisplatin. Hence, the DNA interstrand CLs formed by BBR3464 in both directions are not recognized by HMG domain proteins, or their affinity for these proteins is markedly lower than that of the major 1,2-GG intrastrand CL of cisplatin.
To examine the affinity and specificity of the full-length HMGB1 protein for 1,4-interstrand CLs of BBR3464, 149-bp probes (see "Experimental Procedures"), unmodified or containing a single, site-specific 1,4-GG interstrand CL formed by BBR3464 in the 3Ј 3 3Ј or 5Ј 3 5Ј direction or the 1,2-GG intrastrand CL of cisplatin, were used in electrophoretic mobility shift assays. Not surprisingly, the HMGB1 protein exhibited affinity for the probe containing the 1,2-GG intrastrand CL of cisplatin, as indicated by the presence of a shifted band of the 149-bp probe containing the single 1,2-GG intrastrand CL of cisplatin with the protein (the intensity of which rose with increasing protein concentration) (Fig. 6C, lanes 5-8). The interaction between the protein and the DNA intrastrand cross-linked by cisplatin was not a result of general DNA affinity for the protein because the protein failed to bind the unmodified 149-bp duplexes under identical conditions (Fig.   FIG. 6. Analysis of the binding affinity of 21-bp DNA containing a single, site-specific 1,4-GG interstrand CL of BBR3464 or the 1,2-GG intrastrand CL of cisplatin for HMGB1a (A), HMGB1b (B), and the full-length HMGB1 protein (C) on 6% polyacrylamide gel. Lanes 1-4, unplatinated duplex; lanes 5-8, duplex containing the intrastrand CL of cisplatin; lanes 9 -12 and lanes 13-16, duplex containing the interstrand CL formed by BBR3464 in the 5Ј 3 5Ј or 3Ј 3 3Ј direction, respectively. Lanes 1, 5, 9, and 13, no protein added; lanes 2, 6, 10, and 14, HMGB1a, HMGB1b, and full-length HMGB1 at 0.05, 0.54, and 0.48 M, respectively; lanes 3, 7, 11, and 15, HMGB1a, HMGB1b, and full-length HMGB1 at 0.10, 1.10, and 0.97 M, respectively; lanes 4, 8, 12, and 16, HMGB1a, HMGB1b, and full-length HMGB1 at 0.29, 4.30, and 3.50 M, respectively. 6C, lanes 1-4). On the other hand, under the same experimental conditions, the full-length HMGB1 protein did not bind the 148-bp probes containing either of the 1,4-interstrand CLs of BBR3464 (Fig. 6C, lanes 9 -16). Thus, the results obtained with the full-length HMGB1 protein were entirely consistent with those obtained with its isolated domains.
Nucleotide Excision Repair-NER is a pathway used by human cells for the removal of damaged nucleotides from DNA (61)(62)(63). In mammalian cells, this repair pathway is an important mechanism for the removal of bulky, helix-distorting DNA adducts such as those generated by various chemotherapeutics, including cisplatin (64). Efficient repair of 1,2-GG and 1,3-GTG intrastrand CLs of cisplatin has been reported for various NER systems, including human and rodent excinucleases (16 -18, 65-67). Importantly, intrastrand CLs of BBR3464 and the dinuclear platinum compound [(trans-PtCl(NH 3 ) 2 ) 2 (-H 2 N(CH 2 ) n NH 2 )]Cl 2 are also readily repaired by the NER systems (12,68). The results presented in Fig. 7B (lanes 4 and 5) confirm these reports. Excision repair substrates containing a site-specific 1,4-GG interstrand CL of BBR3464 formed either in the 3Ј 3 3Ј or 5Ј 3 5Ј direction were prepared (Fig. 7A) so that the radioactive label was introduced on both strands 5Ј to the platinated sites to facilitate detection of cleavage sites on either strand. The interstrand cross-linked substrates were incubated with human or rodent CFE. To detect products arising from incision on only one strand, the DNA sample was treated with NaCN after incubation with CFE to remove platinum. The NaCN treatment was included to eliminate the effect of the positively charged platinum complex bound to the excised fragments on their migration on the gel and also to see if the excision would take place only on one strand of the interstrand cross-linked substrate. The products were analyzed by electrophoresis on denaturing polyacrylamide-agarose gel (shown for rodent CFE in Fig. 7B). The substrate containing a single, site-specific 1,3-GTG intrastrand CL of cisplatin was run as a positive control to ensure that the extract was active in NER (Fig. 7B, lane 6). However, in contrast, no excision products could be detected in the NER assay after NaCN treatment if the substrates containing single, site-specific 1,4-GG interstrand CLs formed by BBR3464 in both directions were analyzed using both human and rodent excinucleases (shown for the CLs treated with rodent excinuclease in Fig. 7B, lanes 8  and 10). Varying the incubation times in the range of 10 -60 min and the amount of CFE in the range of 20 -100 g did not alter this result (data not shown).
The mammalian CFEs containing the excinuclease hydrolyze the phosphodiester bond located at a position five nucleotides to the 3Ј-site of the damage and 22-24 nucleotides to the 5Ј-site of the damage (61)(62)(63). The substrates containing the 3Ј 3 3Ј or 5Ј 3 5Ј 1,4-interstrand CL used in this study possessed a radioactive label at the 11th phosphodiester bond to the 5Ј-side of the damage on one strand of the duplex and at the 18th or 12th phosphodiester bond to the 5Ј-side of the damage on the other strand, respectively (Fig. 7A). Thus, if NER of the 1,4-interstrand CLs of BBR3464 occurred in the normal manner, it should be detected by this assay.
The rate-determining step of NER is recognition of the damaged DNA (61,69). This recognition process involves multiple protein components; and for example, RPA is one of these proteins that belongs to the initial damage-sensing factors of human excision nuclease initiating repair (68,70). These recognition proteins preferentially bind to damaged DNA; so, for instance, binding of RPA to damaged DNA may be a sensitive indicator to predict whether mammalian NER could be effective in the removal of damaged nucleotides.
The binding of RPA is thought to proceed via the denaturation of the DNA substrate and subsequent high affinity binding to the single-stranded DNA generated (28,68,71). Hence, it appears particularly interesting to examine whether RPA binds to the interstrand cross-linked substrates because the interstrand CL would prevent denaturation of DNA (which is a prerequisite for RPA binding). To determine the effect of 1,4-GG interstrand CLs of BBR3464 on RPA binding, the 21-bp duplexes were prepared so that they contained a single, sitespecific 1,4-interstrand CL formed in either the 3Ј 3 3Ј or 5Ј 3 5Ј direction (the duplexes identical to those used in the experiments of this work, in which recognition of the interstrand CLs of BBR3464 by HMG domain proteins was examined (see above and Fig. 6)). RPA binding to these duplexes was assessed in a gel mobility shift assay (Fig. 8). The duplexes with blunt ends were purified as described under "Experimental Procedures" so that their samples contained no contaminating single-stranded DNA, as evident from analysis on native polyacrylamide gel (data not shown). The 21-bp duplex [TGGT] containing a single, site-specific 1,2-GG intrastrand CL of cisplatin was run as a positive control (Fig. 8B, lanes 5-8). The results of the gel mobility shift assay demonstrated that increasing RPA concentrations resulted in an increasing amount of this protein bound to the 1,2-GG intrastrand CL of cisplatin. On the other hand, the same analysis performed with the undamaged substrate or the substrate containing a single, site-specific 1,4-interstrand CL formed by BBR3464 in either the 3Ј 3 3Ј or 5Ј 3 5Ј direction revealed that increasing RPA concentrations resulted in a low level of RPA binding to the undamaged duplexes (Fig.  8A, lanes 1-4 and 9 -12) and to the duplexes containing a single, 1,4-interstrand CL of BBR3464 in either direction (lanes 5-8 and 13-16). Quantitation of these results is shown in Fig.  8C. The inability of RPA to bind a substrate containing an interstrand CL of BBR3464 is consistent with RPA binding occurring via denaturation of DNA. A low level of nonspecific RPA binding to undamaged control substrate was explained as being the result of decreased stability of the relatively short duplexes (71). Taken together, these results strongly support the view (see above) that the efficiency of the mammalian NER pathway to remove the interstrand CLs of BBR3464 is very low.

DISCUSSION
The sequence specificity of the interstrand cross-linking in DNA by BBR3464 has been assayed in this work by Maxam- Gilbert footprinting. The experiments presented here (Figs. 2 and 3) clearly demonstrate that the preferential DNA-binding sites in these lesions are G residues. The CLs between G residues separated by 4 bp (1,6-interstrand CLs) were formed at a pronouncedly slower rate compared with 1,2-and 1,4interstrand CLs (Fig. 2B), so it is reasonable to suggest that 1,6-GG interstrand CLs represent less frequent adducts of BBR3464. The CLs between G residues separated by Ͼ4 bp (1,7-and 1,8-interstrand CLs) were not formed under these conditions by BBR3464, so the maximum length of the interstrand CL formed by BBR3464 in DNA is likely to be that in which the cross-linked sites are separated by up to 4 bp. Thus, the formation of long-range CLs by BBR3464 proposed in our earlier reports (9,72) and hence the ability of this trinuclear platinum compound to target larger sequences of DNA are further confirmed. Importantly, the results of this work have also confirmed under competitive conditions that the interstrand CLs of BBR3464 can be formed in both the 3Ј 3 3Ј and 5Ј 3 5Ј directions (Fig. 3). Whereas the 1,2-interstrand CLs between G residues in neighboring base pairs were formed exclusively in the 3Ј 3 3Ј direction, the 1,4-interstrand CLs of BBR3464 were formed with approximately the same preference for both the 3Ј 3 3Ј and 5Ј 3 5Ј directions. The 1,6-interstrand CLs were also formed by BBR3464 in both directions, but with a clear preference for the formation of the CL in the 5Ј 3 5Ј direction. Thus, our results demonstrate that although the shortest 1,2-interstrand CL of BBR3464 is formed with a strong preference for the 3Ј 3 3Ј direction, the growing length of these CLs reverses this preference to the opposite 5Ј 3 5Ј direction. Interestingly, the interstrand CLs of the antitumor dinuclear platinum compounds [(trans-PtCl(NH 3 ) 2 ) 2 (-H 2 N(CH 2 ) n NH 2 )]Cl 2 and [(cis-PtCl(NH 3 ) 2 )H 2 N(CH 2 ) n NH 2 ]Cl 2 (n ϭ 4 and 6; which have leaving chloride ligands cis to the linker) (32, 39) and cisplatin (30,73) are formed exclusively in the 5Ј 3 5Ј direction under the conditions employed here, so the capability of BBR3464 to form interstrand CLs in the 3Ј 3 3Ј direction is one of the unique features of the DNA-binding mode of this novel platinum drug. The reasons for the dependence of the orientation of DNA interstrand CLs of BBR3464 on their length are unknown. In parallel NMR studies, binding of both [(trans-PtCl(NH 3 ) 2 ) 2 (-H 2 N(CH 2 ) 6 NH 2 )]Cl 2 and BBR3464 to the 14-bp duplex sequence 5Ј-d(ATACATGGTA-CATA)-3Ј⅐5Ј-d(TATGTACCATGTAT)-3Ј in both the 5Ј 3 5Ј and 3Ј 3 3Ј CL formations was observed. 3 Pre-association as also observed by NMR studies may influence the direction of the CLs formed, as the central platinum unit lies in the minor groove as a consequence of the pre-association. In the study, we have also described the conformational distortions induced in DNA by the most frequent interstrand CLs of BBR3464, 1,4-GG interstrand CLs formed in both the 3Ј 3 3Ј and 5Ј 3 5Ј directions. The bending experiments were carried out with the double-stranded oligodeoxyribonucleotides [1,4;3Ј-3Ј(20 -23)] and [1,4;5Ј-5Ј(20 -23)] containing the unique interstrand CL in their central sequence. The phasing assay revealed that the 1,4-GG interstrand CLs formed by BBR3464 in the 3Ј 3 3Ј and 5Ј 3 5Ј directions resulted in a directional bending of the helix axis (21°and 15°, respectively, toward the major groove) and duplex unwinding (14°for both types of the interstrand CL) (Fig. 5). In addition to these bending and unwinding effects, the 1,4-interstrand CLs formed by BBR3464 created local conformational distortions revealed by the chemical probes (Fig. 4). These distortions were non-symmetrical and extended mainly over ϳ4 base pairs (Fig. 4E). The patterns of distorted sites were different for the CLs formed in the 3Ј 3 3Ј and 5Ј 3 5Ј directions, confirming a different character of localized conformational distortions due to the different orientation of these lesions. Interestingly, in the case of the 1,4-interstrand CL formed in sequence (5Ј-ATGTACAT-3Ј) 2 , the structure shows unique delocalization of the adduct structure, including the presence of the syn-conformation of deoxyriboadenosines outside the binding site (72).
It has been suggested that HMG domain proteins play a role in sensitizing cells to cisplatin (14,15). It has been shown that HMG domain proteins recognize and bind to DNA CLs formed by cisplatin between bases in neighboring base pairs (14,15,22). The molecular basis for this recognition is still not entirely understood, although several structural details of the 1:1 complex formed between the HMG domain and the duplex containing the 1,2-GG intrastrand CL of cisplatin were recently elucidated (14). The details of how the binding of HMG domain proteins to cisplatin-modified DNA sensitizes tumor cells to cisplatin are also still not completely resolved, but possibilities such as shielding cisplatin-DNA adducts from excision repair or that these proteins could be titrated away from their transcriptional regulatory function have been suggested (8,15,58,74) to explain how these proteins are involved in the antitumor activity.
An important structural motif recognized by HMG domain proteins on DNA containing the major 1,2-GG intrastrand CL of cisplatin is a stable, directional bend of the helix axis toward the major groove (15). As demonstrated in the present work (Fig. 5), the 1,4-GG interstrand CLs of BBR3464 bend the helix axis less efficiently than the CLs of cisplatin (for instance, gel retardation assays revealed that the 1,2-GG intrastrand CL of cisplatin produces a rigid, directed bend 32-34°into the major groove of DNA (38,53)). No recognition of DNA interstrand CLs of BBR3464 by HMGB1 proteins was observed in the present 3  work (Fig. 6). A plausible explanation of this observation may be that the prebending due to the 1,4-interstrand CL of BBR3464 is too small to be recognized by HMG domain proteins. Alternatively, this result may also be associated with some structural features of the 1,4-interstrand CL of BBR3464 revealed recently by the NMR analysis of the 8-bp duplex containing this adduct (72). This analysis has shown that the "central" tetraamine linker trans-H 2 N(CH 2 ) 6 NH 2 Pt-(NH 3 ) 2 (CH 2 ) 6 NH 2 of this CL is situated in or very close to the minor groove of DNA. It is reasonable to expect that this location of the linker could sterically block binding of the HMG domain proteins to DNA (because they also bind to DNA from the minor groove (75,76)). In addition, the location of the linker in the minor groove could restrict the additional DNA bending required for HMG domain binding (14,15). Hence, it is also possible that BBR3464 in the 1,4-interstrand CL restricts the additional DNA bending required for binding of the HMGB1 protein. Thus, from the results of this work, it is clear that the DNA interstrand CLs of the antitumor compound BBR3464 may present a block to DNA or RNA polymerase, but are not a substrate for recognition by HMG domain proteins. From these considerations and from the fact that also intrastrand CLs are not recognized by HMG domain proteins (12), we conclude that the mechanism of antitumor activity of bifunctional BBR3464 does not involve recognition by HMG domain proteins as a crucial step, in contrast to the proposals for cisplatin and its direct analogs.
Several reports have demonstrated that NER is a major mechanism contributing to cisplatin resistance (16 -18). The examination of excision repair of 1,4-interstrand CLs of BBR3464 has revealed that these adducts cannot be removed as readily by excision repair as intrastrand adducts. Hence, the interstrand CLs of bifunctional trinuclear BBR3464 would not have to be shielded by damaged DNA recognition proteins, such as those containing HMG domains, to prevent their repair. Thus, it is reasonable to suggest that interstrand CLs of BBR3464 could persist considerably longer than intrastrand adducts (23,24), which would potentiate the toxicity of BBR3464 to tumor cells sensitive to this drug.
In conclusion, the results of this work provide additional strong support for the hypothesis that platinum drugs that bind to DNA in a fundamentally different manner compared with cisplatin have altered pharmacological properties. Importantly, in contrast to cisplatin, the mediation of antitumor properties of bifunctional trinuclear platinum complexes by HMG domain proteins is unlikely, so polynuclear platinum compounds represent a novel class of platinum anticancer drugs acting by a different mechanism compared with cisplatin and its analogs. The cytotoxic effects of BBR3464 may realistically be due to a cumulative effect of the structurally heterogeneous adducts produced by this drug, but the role of structurally unique interstrand CLs in the antitumor effects of BBR3464 may predominate.