Sequence Specificity and DNA Interactions of a Series of Polybenzamide Mustards

Abstract

A series of minor groove targeted polybenzamide mustards (see figure 1.13 a and b) has been investigated as potential antitumour agents. These compounds were previously found to span a wide range of cytotoxicities as measured by the P388 IC50 assay, suggesting they possess important biological differences and structure-activity relationships. The non-covalent binding of compound 1 was found to be specific for sequences containing three or more consecutive adenines while compound 10 showed a less specific binding to adenine/ thymine rich sequences in general. The binding of compounds 1 and 10 to DNA required a higher concentration than equivalent binding with Hoechst 33258, a previously characterised minor groove binding ligand, which suggests compounds 1 and 10 are not as DNA affinic as Hoechst 33258. The sequence specificity of covalent bonding was quite diverse across the different compounds examined. Compounds which have the same para-substituted benzene backbone structure as compound 1 showed similar bonding specificities both to compound 1 and each other. These compounds mostly alkylated adenines in runs of three or more. There were however subtle differences between these compounds which were related to the number of mustard units and the number and location of cationic dimethylamino groups. The similarities in sequence specificity of non-covalent binding of compound 1 and covalent binding of compounds 2-6 suggests the backbone para-substituted benzene structure is the major determinant of sequence specificity. Modelling studies suggest that two multicentered hydrogen bonds are formed between the inward facing NH groups of the carboxamide linkers and the adenine N3/ thymine O2 atoms on DNA, as well as several favourable van der-Waals interactions between the adenine C1 hydrogens and the benzene hydrogens of the drugs. Electrostatic potential of the minor groove of DNA at A/ T rich sequences probably also contributes to sequence selectivity. Compound 8 had a similar sequence specificity to compounds 2-6 but alkylated adenines less sequence specifically, probably due to stabilising van der-Waals interactions at the central bicyclo[2.2.2]octane unit. Compounds 7 (meta substituted central benzene) and 9 (2, 4 substituted 1-methyl pyrazole central unit) had a high specificity for alkylating the underlined adenine in the consensus sequence 5'-A/TAG/CA/TN. Modelling studies suggest both compounds can form a H-bond between an inward facing carbonyl oxygen on the carboxamide linker and the 2-amino goup of guanine. Van der-Waals interactions are not as prominent with compounds 7 and 9 as they are with compounds 2-6 and 8, but electrostatic interactions probably still participate. Compounds 11, 12 and chlorambucil show a high degree of guanine N7 alkylation compared to adenine alkylation and are probably not as well targeted to the minor groove as the other compounds, although compounds 11 and 12 alkylate DNA more readily than chlorambucil. The single central cationic charge of compounds 11 and 12 may be buried in the minor groove with a mustard group protruding outwards or vice versa. Compound 12 which has structural similarities to compound 7 shows some sequence specific similarities also. The higher alkylation efficiency of compounds 11 and 12 compared with chlorambucil suggests the presence of a cationic group indirectly targets these compounds to DNA. In addition to adenine alkylation all compounds showed a lower level of guanine N7 alkylation as determined by piperidine cleavage and some bisalkylators also alkylated guanines strongly at unique sequences at a position other than N7 (probably N3). All of the bisalkylating compounds showed a similar strong ability to crosslink plasmid DNA in vitro compared with chlorambucil. Compounds 5-9 crosslinked plasmid DNA 25-50 times better, on a molar basis, than chlorambucil while compounds 11 and 12 were less efficient than compounds 5-9 but still more efficient than chlorambucil. Crosslinking ability in vitro was not correlated with cytotoxicity suggesting other factors such as sequence selectivity or DNA repair are more relevant, or that crosslinks are not readily formed in vivo. Crosslinking by compounds 6 and 7 is not inhibited by Hoechst 33258, despite Hoechst 33258 inhibiting adenine alkylation as assayed by strand cleavage methodology. This strongly suggests that crosslinking assayed in plasmid DNA occurs at sequences not strongly alkylated in the minor groove or more likely at guanine N7 sites in the major groove of DNA. If minor groove crosslinks are formed they form much more slowly than major groove crosslinks and cannot be readily detected in the plasmid assay. Compounds that were tested on the NCI's panel of human tumour cell lines showed large differences in cytotoxicity within cell lines and between compounds. These differences may be due to several factors such as sequence specificity, crosslinking ability, drug resistance phenotype and DNA repair phenotype. All compounds were tested for the ability to induce mitotic recombination, other mutations and mitochondrial toxicity in Saccharomyces cerevisiae. Structure-activity relationships were not obvious but compound 5 induced high levels of mitotic recombination and other mutations while compound 6 induced high levels of other mutations only. Only compound 4 showed significant mitochondrial toxicity which may be related to its high degree of cationic character, its A/T rich sequence selectivity and it's ability to only form monoadducts. Mutagenesis in Chinese hamster ovary AS52 cells revealed striking differences between chlorambucil and compounds 6 and 7. The mutation frequency to 6-thioguanine resistance increased in a dose dependent manner with chlorambucil but not with compounds 6 and 7. The mutation spectra of the gpt gene in AS52 cells shows chlorambucil induces mostly large deletions. The mutation frequency of compounds 6 and 7 at the l0% survival level were the same as control levels, however gpt gene analyses showed the mutational spectra differed. Only one mutant clone derived from compound 6, which possessed the gpt protein coding sequence, showed a DNA sequence change in this region, suggesting mutations were induced specifically in the promoter region of the gene, a theory backed up by preliminary Southern hybridisation analyses. Recombinational mechanisms may also result in lack of gene function and homologous gpt sequences exist downstream of the functional gene which may mediate these processes. The one compound 6 mutant which possessed a mutated gpt protein coding sequence had a 33 bp deletion directly 5' to a potential hairpin secondary structure and compound 6 bonding site. This clone had remarkable similarities to one observed following treatment with the minor groove alkylator CC-1065. Mutants induced by compound 7 also had a high degree of wild type gpt protein coding sequences, but some clones had point mutations including a frameshift single base pair deletion. One possible hypothesis to explain these results is that compound 6 adducts are less easily recognised by repair enzymes when they are located in the promoter region of the gpt gene than when they are in the protein coding region. Concommitors with this unrepaired adducts may lead to mutations via DNA polymerase error, in a similar manner to CC-1065. Compound 7 adducts may be more easily recognised by repair systems but may be less accurately repaired once recognised, leading to mutations. The results obtained in this work suggest future research can take several directions, and further compound structures designed to target longer DNA sequences are proposed.