On the structural basis and design guidelines for type II topoisomerase-targeting anticancer drugs

Structural basis of Top2 inhibition by the anticancer drugs m-AMSA and mitoxantrone

The binding of m-AMSA and mitoxantrone between the +1/+4 and −1/+5 DNA base pairs immediately flanking the cleavage sites reveals that, despite significant differences in chemical structure (Figure 1), these two drugs share a similar mechanism of Top2 inhibition with etoposide (Figure 2C–E) (10). The cleavage site-specific drug insertion physically interferes with the stacking between the +1/+4 and −1/+5 base pairs and maintains the separation of the 3′-OH (of the −1 nucleotide) and the enzyme-linked 5′-phosphate (of the +1 nucleotide) by 7 ∼ 8 Å (Figure 3), which effectively traps the Top2cc by preventing the religation reaction. Moreover, compared with the quaternary conformation adopted by two drug-free Top2-DNA binary complexes with a catalytically active DNA cleavage center (24,25), it has been suggested that the etoposide-induced separation of the cleaved DNA ends is accompanied by the two dyad-related Top2 monomers sliding away from each other along their respective H3 helices (10). Similar types of drug-promoted quaternary conformational changes were also observed in the m-AMSA and mitoxantrone-bound structures, which are expected to further discourage the religation reaction by disrupting the in trans-assembled Top2 cleavage center (Supplementary Figure S2).

Open in a separate window

Figure 3.

Detailed views of drug-binding sites. Stereo representations of the binding sites for m-AMSA (A) and mitoxantrone (B). Drugs and DNA are shown as cyan and purple sticks, respectively. Two hTop2β^core monomers are colored differently. Labels belonging to the second monomer are flagged by a prime. Forces that stabilize the R503 side chain in each structure are indicated by green dashed lines. Mg²⁺ and water molecules are shown as green and red spheres, respectively. The distance between the Y821′-linked scissile phosphate and the 3′-OH are indicated.

A notable difference between the parental etoposide-stabilized and the two new Top2cc structures is a switch of deoxyribose ring puckering of the −1 and +5 nucleotides from the 3′-endo to the 2′-endo configuration (Figure 4). The potential significance and validity of this finding are further corroborated by the observation of an identical structural transition in the two crystallographically independent DNA breaks in the asymmetric unit. In addition, three extra crystallographic approaches were used to provide additional support for the change in ring puckering. First, we deliberately modeled the −1 and +5 nucleotides of the m-AMSA- and mitoxantrone-bound structures to adopt the 3′-endo configuration, as observed in the etoposide-bound structure (10); rounds of structural refinement were then applied using the refinement module of PHENIX (22). In both cases, the converged models reverted to the 2′-endo configuration. A reciprocal test also confirmed the robustness of the tendency for these nucleotides to take the 3′-endo configuration in the presence of etoposide. Secondly, we removed the −1 and +5 nucleotides from the three Top2cc structures, and the partial models were subjected to structural refinement using PHENIX. Unbiased mF_o-DF_c electron density maps calculated using the converged partial models clearly show features that agree with the switching in ring puckering for the omitted nucleotides on drug replacement (Figure 4). Finally, structural analyses performed on the drug-free and etoposide-reintroduced crystals revealed that the −1 and +5 nucleotides adopt the 3′-endo configuration in both structures (Supplementary Figure S3), which effectively rules out the post-crystallization handling procedures as the cause for the change in ribose puckering. Taken together, these findings indicated that the formation of 2′-endo riboses is induced specifically by m-AMSA and mitoxantrone; the alternation of R503 rotamer conformation resulted from drug-replacement and the more pronounced stacking between these two drugs and the flanking base pairs are possible causes for the switching in deoxyribose ring puckering (Supplementary Figure S4).

Open in a separate window

Figure 4.

The unbiased difference electron density maps (mF_o-DF_c, contoured at 4.5σ) of the −1 nucleotides are shown to validate the ring puckering change in the deoxyribose in the presence of the respective drugs. The m-AMSA- and mitoxantrone-stabilized axial positions of 3′-OH are indicated by black stars.

Accompanying the switch in ribose puckering from the 3′-endo to the 2′-endo conformation, the equatorially placed 3′-OH of the −1 nucleotide rearranges to take an axial position, which constrains the lone pair electrons on the oxygen of 3′-OH to point away from the 5′-phosphate. Such a structural transition likely perturbs the alignment between the nucleophile (the 3′-OH) and the electrophile (the 5′-phosphotyrosine) and may thus suppress the religation reaction. Collectively, the currently available structures of drug-stabilized Top2cc suggest the insertion between cleaved DNA ends, the decoupling of catalytic groups and the axial repositioning of the 3′-OH of the −1 nucleotide as three possible mechanisms by which the religation reaction can be inhibited.

Cleavage site-specific targeting by m-AMSA and mitoxantrone

Whereas the insertion of their polycyclic aglycone cores between the +1/+4 and −1/+5 base pairs appears to be a major and common mechanistic theme used by drugs targeting the Top2cc, it is well known that the DNA-intercalating potential alone is not sufficient for conferring Top2-poisoning activity. For example, the DNA intercalator ethidium bromide inhibits Top2 activity at higher concentrations but fails to induce Top2cc formation (26), indicating that the intercalation may not be cleavage site-specific and that the inhibitory effect likely results from a reduced association between Top2 and DNA owing to interference from randomly intercalated ethidium bromide. To effectively trap the sparsely populated and transiently formed Top2cc during the catalytic cycle, stabilizing interactions between the protein and the bound drug are required to arrest the drug in the Top2-mediated DNA cleavage site. By revealing the extensive networks of protein–drug interactions, our structures suggest how the cleavage site-specific binding by m-AMSA and mitoxantrone is achieved. As shown in Figures 2D and ​and3A,3A, the planar acridine chromophore of m-AMSA (rings A∼C in Figure 1) inserts into the cleavage site to maintain the separation of the two flanking base pairs, and the bulky methanesulfon-m-anisidide head group protrudes toward the DNA minor groove to mediate specific drug–protein contacts. The anilino moiety packs against the aliphatic part of the R503 side chain, and the methanesulfonamide moiety is anchored by forming a hydrogen bond with E522 and van der Waals interactions with I454, P455, P504, A521 and E522.

Similarly, mitoxantrone uses its polycyclic aromatic dihydroxy-anthraquinone moiety (rings A∼C in Figure 1) to intercalate into the DNA cleavage site and physically blocks the religation reaction (Figures 2E and ​and3B).3B). In contrast to m-AMSA, whose acridine moiety functions exclusively in DNA-intercalation and mediates no direct interactions with the protein, the polycyclic core of mitoxantrone plays dual roles, as the hydroxyl and carbonyl groups of rings A and B are also involved in hydrogen bonds with G504 on the minor groove and Q778 on the major groove side. The two structurally identical hydroxylalkylamino arms that attach symmetrically to the C-ring (via the C5 and C8 carbon atoms) project toward the DNA major and minor grooves, respectively, embracing the +5 guanine base from both sides. Additional interactions that stabilize the bound mitoxantrone are mainly provided by the minor groove-placed hydroxylalkylamino arm, which fits into a crevice located between the protein and the DNA and forms multiple hydrogen bonds with the side chains of N520 and E522 and the main-chain carbonyl group of R503. In contrast, the arm in the major groove is more solvent-exposed and only a single hydrogen bond is present between the drug and the protein. We noted that an earlier nuclear magnetic resonance-based study of the (protein-free) mitoxantrone-RNA binary complex placed both alkylamino arms in the major groove (27); however, our modeling analysis indicates that a similar type of binding is not feasible in the context of the Top2 cleavage complex owing to steric conflicts between the drug and protein residues in both DNA minor and major groove. Comparing the structures of Top2cc stabilized by different drugs clearly revealed that the binding of each drug in the cleavage site is mediated by a distinct and drug-specific set of interactions (Figure 5A–C).

Open in a separate window

Figure 5.

Conformational landscapes of the drug-binding pockets for mammalian Top2-targeting anticancer drugs. (A–C) Surface/stick representations of the etoposide-, m-AMSA- and mitoxantrone-binding sites, respectively. Selected DNA base pairs (purple) and protein (pink) residues are shown as surface representations and are labeled using white letters. Two hTop2β^core monomers are colored differently. Drugs (cyan) and the Y821′-conjugated +1 thymidines are shown as sticks. Labels belonging to the second monomer are flagged by a prime. Cyan letters designate polycyclic ring labeling of drugs. (D and E) Superposition of residues 445–731 and residues 762–821 in hTop2β of the drug-stabilized hTop2cc structures to show structural differences in the minor groove and major groove drug-binding pockets, respectively. Selected residues of hTop2cc structures stabilized by etoposide (blue), m-AMSA (yellow) and mitoxantrone (cyan) are shown as sticks. Side chain and main chain variations are indicated by red and black arrows, respectively. Residues that exhibit no structural change under drug binding are labeled in gray.

Structure–activity relationships and the structural basis of drug resistance

Because the m-AMSA- and mitoxantrone-stabilized Top2cc structures were obtained by post-crystallization ligand replacement rather than direct crystallization, the validity and pharmacological relevance of these new structures must be carefully assessed. Toward this end, we examined whether the structure–activity relationships and drug-resistant mutations known for each drug can be fully accounted for by the observed interaction patterns and drug-binding modes. Among the residues involved in m-AMSA binding (Figure 3A), E522 appears to play a key role with its side-chain carboxylate forming a salt bridge to the amine group of the 1′-methanesulfon substitution, and its Cβ-Cγ-Cδ moiety contacting the 3′-methoxy group and the C2′ atom of the aniline ring (Figure 3A). This set of stabilizing interactions readily explains the impairment of drug activity on the removal of the 1′- and 3′-substitutions from the methanesulfon-m-anisidide head group (19), which is expected to substantially weaken drug binding. The m-AMSA-bound structure also addresses why o-AMSA (Figure 1) is essentially inactive at inducing Top2-mediated DNA breakage compared with the high potency exhibited by m-AMSA (26); steric clashes between the methoxy group and the E522 side chain would be produced when shifting the methoxy group from the meta (3′, as in m-AMSA) to the ortho position (2′, as in o-AMSA). Therefore, despite the fact that the two AMSA stereoisomers display comparable DNA-intercalating activity (26), o-AMSA and other AMSA analogs with ortho-substitutions are incompatible with the minor groove drug-binding pocket and are thus inefficient at stabilizing Top2cc (19). Intriguingly, when detached from the acridine chromophore, the methanesulfon-m-anisidide head group of m-AMSA alone is capable of trapping Top2cc at high concentrations (19). We suspect that by occupying the minor groove binding pocket, the anilino head group may cause the −1 nucleotide to adopt the less-reactive 2′-endo conformation, as noted in the m-AMSA-bound structure (Figure 4), thereby giving rise to DNA intercalation-independent Top2-poisoning activity.

The identification of a glutamate-to-lysine mutation at E522 (E522K) as being resistant to m-AMSA and other acridine derivatives (28) also agrees with our structural observation; most likely, the longer lysine side chain would clash with the anilino ring to disfavor drug binding. Given that E522 does not interact directly with etoposide (Figure 5A) (10), the present study establishes the importance of this residue for the formation of drug-stabilized Top2cc. Interestingly, the E522K mutation displayed hypersensitivity toward etoposide (28), suggesting that the mutation-induced changes in shape and polarity of the minor-groove drug-binding pocket disturb m-AMSA binding but facilitate the binding of etoposide. The observed stabilizing interactions also explain why mutations at R503, another major drug-contacting residue, or the nearby P501 and L502 residues may weaken the protein–drug interactions and thus result in drug resistance (10,29).

Mitoxantrone and ametantrone are structurally closely related derivatives of anthracycline (Figure 1), both being effective anticancer agents with reduced cardiotoxicity compared with their parent compounds (30). While exhibiting similar DNA-intercalating activities (31), however, the Top2-poisoning and antitumor activities of ametantrone are significantly lower than mitoxantrone (32,33), likely because the A ring of ametantrone is not hydroxylated. Using the post-crystallization soaking procedure described above, the structure of ametantrone-bound Top2cc was determined (Table 1 and Supplementary Figure S5). Except for a difference between ring A-mediated interactions and minor adjustments of the nearby protein moieties, these two anthracenedione-stabilized structures are essentially indistinguishable, indicating that a loss of 1- and 4-hydroxyl groups compromises drug function mainly by reducing the number of drug–protein contacts. Additionally, interactions between the hydroxylalkylamino arms and the protein address why modifications of these two branching moieties affect drug efficacy (Figure 5C and Supplementary Figure S5C) (34). Taken together, these findings indicate that the new structures are pharmacologically relevant and that the reported ligand-replacement procedure for the structural characterization of drug-stabilized Top2cc may benefit the structure-based development of new Top2-targeting drugs.

Conformational landscapes of the drug-binding pocket

The three drug-stabilized Top2cc structures reveal a substantial repositioning of drug-contacting residues located on the DNA minor groove side on the binding of various drugs (Figure 5); this finding explains how structurally distinct drugs are accommodated within the protein–DNA interface. The observed structural rearrangements are centered on R503 and are tightly correlated with the location of the minor groove-protruding moiety of each drug. For etoposide, the E-ring faces the cleaved +1 nucleotide and interacts with the nearby K456 and D479 residues, causing the R503 side chain to swing toward the +4 and +5 nucleotide of the opposing strand (Figure 5A). In contrast, both the anilino head group of m-AMSA and the alkylamino arm of mitoxantrone extend into the pocket formed between the +5 nucleotide and flanking residues including E522 (Figure 5B and C), which drives R503 to the vicinity of the +1 nucleotide. Therefore, two distinctive but mutually exclusive regions on the minor groove side of the protein–DNA interface are recognized as crucial for drug binding. Although residues located in the minor groove exhibit notable structural variations on drug replacement (Figure 5D), few changes were detected for the primary drug-interacting residues on the major groove side (Figure 5E), consistent with the notion that the major groove pocket is sufficiently spacious to accommodate drug moieties of different sizes and shapes (10). By demonstrating the magnitude of the structural changes associated with drug-contacting residues, together, these structures reveal a set of conformational landscapes of the drug-binding sites, which may facilitate in silico design of new Top2-targeting agents.

Superimposition of the three drug-stabilized Top2cc structures also shows that structural differences caused by drug exchange are mainly owing to side-chain rearrangements and are confined to the immediate vicinity of the drug-binding pockets, and the approach of P455 toward the anilino head group of m-AMSA is the only evident main-chain displacement (Figure 5D and E). Because P455 is located within a flexible loop that is detached from the core protein domains, this movement should not significantly affect the overall structure. The highly localized nature of these structural changes explains why drug replacement can be achieved without disrupting crystal packing. Despite the usefulness of the soaking procedure in illustrating the binding modes of m-AMSA and mitoxantrone, this technique may not be applicable to intercalating compounds with bulky groups facing the DNA minor groove, such as doxorubicin (Figure 1). The structure of a doxorubicin-DNA binary complex suggests a preference for locating the amino sugar and hydroxymethyl ketone groups in the minor groove on the drug’s intercalation into DNA (35). Assuming that this binding mode of doxorubicin dictates and is not affected by the involvement of Top2, then the presence of doxorubicin in the cleavage site would be strongly discouraged owing to severe steric repulsion from the minor groove-flanking residues when the quaternary structure of Top2cc is constrained by the crystal lattice. Indeed, when soaked into the crystal, doxorubicin was found to bind between the +1/+4 and +2/+3 base pairs rather than the cleavage site (Supplementary Figure S6 and Table S1). Given this potential limitation, we suggest that the reported soaking procedure for drug exchange would be most suitable for those Top2-targeting agents whose binding (at the cleavage site) can be accommodated by local side-chain rearrangements.

Portions of this research were carried out at beamline 13B1 of the National Synchrotron Radiation Research Center (Taiwan). The authors thank the PX team for assistance during data collection, C. Austin for sharing unpublished results and T.-s. Hsieh and the members of Chan lab for helpful discussion.