Generation of Escherichia Coli Nitroreductase Mutants Conferring Improved Cell Sensitization to the Prodrug CB1954¹

Gene delivery to tumor cells using DNA complexes or viral vectors allows the production of prodrug-activating enzymes, thereby sensitizing the tumors to prodrugs. This approach can achieve a greater therapeutic index than systemic administration of the corresponding active drug, with which toxicity to vital organs limits the dose that can be used (1). This strategy for cancer gene therapy is commonly called GDEPT³ or virus-directed enzyme prodrug therapy (VDEPT).

Several enzyme/prodrug combinations for GDEPT have been described; the most widely used are HSV-TK with GCV (2) and Escherichia coli cytosine deaminase with 5-FC (1). We have focused on another prodrug, CB1954 (5-[aziridin-1-yl]-2,4-dinitrobenzamide; Ref. 3). This is a weak, monofunctional alkylating agent that can be reduced by E. coli NTR at either of the nitro groups, to the corresponding hydroxylamino derivatives (4). The 4-hydroxylamino product (produced in 50% yield) is particularly cytotoxic, because it leads to the formation of interstrand DNA cross-links, which are poorly repaired by the cell (5). Activated GCV and 5-fluorouracil (generated from 5-FC) inhibit DNA synthesis, and, thus, their toxicity is largely restricted to cells in S phase. In contrast, the toxicity of activated CB1954 is seen also in nonreplicating cells (6, 7); this may be advantageous because, in human tumors, the fraction of cells replicating their DNA at any time can be quite low. As with the other enzyme/prodrug combinations, bystander cell killing is seen with NTR/CB1954 because of local spread of the activated prodrug (8, 9, 10). This allows an antitumor effect to be obtained even when only a minority of cells produce the enzyme (11).

In animal models, CB1954 treatment can result in complete regression of a high proportion of tumors that stably express NTR (10, 11). Significant therapeutic responses to CB1954 have also been observed after in vivo delivery of the NTR gene to established tumors (7, 11, 12). Following from these studies, the system is currently in Phase I clinical trials (13, 14).

The efficacy of NTR/CB1954 correlates with the dose of vector and level of NTR expression, both in vitro and in vivo (7, 12). As with other therapeutic systems, difficulties in achieving an effective level and distribution of transgene expression after in vivo gene delivery lower its efficacy. Increasing the specific activity of NTR for CB1954 would increase the yield of activated prodrug and improve the efficacy of the system when gene delivery is limiting. Others have shown that mutants of HSV-TK can significantly improve cell sensitization to GCV and acyclovir, relative to the WT (15, 16, 17). The aim of our study was to generate improved mutants of NTR.

E. coli NTR is a homodimeric flavoprotein in which the FMN cofactor is reduced by NADH or NADPH and which, in turn, reduces a variety of nitroaromatic and quinone substrates. Its structure has been determined by X-ray crystallography, both alone (18) and complexed with the NAD analogue, nicotinic acid (19). On the basis of the structure, we now report the systematic generation of NTR mutants with amino acid substitutions around the active site. This has led to the identification and characterization of mutants that increase the sensitivity of E. coli to CB1954. We go on to show that the F124K mutant is ∼5-fold more efficient than the parental, WT, NTR in sensitizing human ovarian carcinoma cells to CB1954, an improvement that has potential to translate to greater clinical efficacy.

A DNA fragment, containing the tac promoter and lac operator from pDR540 (Pharmacia) with two SfiI sites inserted downstream, was inserted into the unique HindIII site of the bacteriophage λ vector λNM1151 (20). A 1746-bp DNA fragment containing the kanamycin resistance gene from pACYC177 (New England Biolabs) was then cloned into the unique EcoRI site of this vector, to produce λJG3J1. The WT NTR gene was PCR amplified from E. coli strain DH5α and inserted between the SfiI sites of λJG3J1 to produce λJG16C2 (Fig. 2 A).

A pUC19-based plasmid (pJG12B1), containing a tac promoter-NTR cassette identical to that in λJG16C2, was used as the template for PCR-based site-directed mutagenesis. For each site to be mutated, partially overlapping PCR products were generated spanning the 5′ and 3′ ends of the gene. For the 5′ fragments, the upstream primer was within the tac promoter and the downstream primers spanned the target site, with NNN (all 4 mixed deoxynucleotides) at this codon. For the 3′ PCR fragments, the upstream primer commenced immediately after the mutated codon, and the downstream primer was in vector sequence. After generating the 5′ and 3′ PCR fragments for mutants at each position, we combined them by mixing and PCR amplification with the flanking primers. The final products were digested with SfiI and directionally cloned between the SfiI sites of λJG3J1. After packaging the DNA, the phage particles were used to infect the NTR-deficient E. coli strain UT5600 (21), and colonies of lysogens were selected on agar plates containing kanamycin.

Lysogens were grown overnight in LB/kanamycin, in 96-well plates, then replica-plated onto freshly prepared LB/kanamycin agar plates containing 50 mm Tris (pH 7.5), 0.1 mm IPTG, and CB1954 (0–400 μm). After 24-h incubation at 37°C, the growth of each clone was evaluated. The NTR genes of clones that appeared more sensitive to CB1954 than did WT genes were amplified by PCR and were sequenced.

To determine the 50% inhibitory concentration (IC₅₀), lysogens were grown to mid log phase in liquid culture and diluted to ∼1000 cells/ml; 100 μl were spread onto a series of Tris-buffered LB/kanamycin plates containing 0.1 mm IPTG and 0–400 μm CB1954. After 36-h growth, the colonies on each plate were counted. The IC₅₀ was determined as the concentration of CB1954 that caused a 50% reduction in the number of surviving colonies.

The vector vPS1233 is based on an E1, E3-deleted adenovirus similar to that described previously (7), except the E1 deletion extended beyond E1B coding sequences, removing nucleotides 360-3524 of WT Ad5. The expression cassette incorporates the cytomegalovirus promoter derived from pLNCX (22); the WT or F124K NTR genes amplified from the appropriate phage clones; the poliovirus internal ribosome entry site from pLNPOZ (23); the GFP gene from pEGFP-1 (Clontech); and the RNA splicing and termination signals from human β-globin and complement C2 genes, respectively (7). The full-length viral genomes were constructed as plasmids; virus was generated by transfection into HEK293 cells (24) and was cloned by two rounds of plaque purification before expansion and purification by CsCl density-gradient centrifugation. The titer of virus stocks was determined by plaque assay on HEK293 or 911 cells (25).

SK-OV-3 cells were grown in HEPES-buffered DMEM with 10% FCS. Cells were harvested by trypsinization, resuspended in culture medium, and allowed to recover for 1.5 h in suspension at 37°C. The cells were then infected (at 1.5 × 10⁶ cells/ml) with adenoviruses expressing WT or F124K NTR, at the specified MOI. The suspension was incubated at 37°C for 1.5 h with occasional gentle mixing before plating 1.5 × 10⁴ cells in 150 μl of medium per well of a 96-well plate. After 2 days, the medium was replaced with medium containing the appropriate concentration of CB1954. This was replaced with fresh medium without prodrug after 24 h. Cell survival was determined by MTT (3-[4,5-dimethylthiazol-2-yl]-2,5-diphenyltetrazolium bromide) assay (26) 3 days after addition of the prodrug.

Fig. 1 shows two views of the active site of NTR, derived from our previously determined structure of the enzyme complexed with the substrate analogue, nicotinic acid (19). The nicotinic acid (and by implication, the headgroup of the substrate NADH) stacks in a hydrophobic sandwich between the aromatic ring of residue F124 and the FMN cofactor involved in hydride transfer, within a pocket at the dimer interface. Because the enzyme follows a substituted enzyme (“ping-pong”) mechanism (27), CB1954 must bind in a similar position to nicotinic acid. The nine amino acid residues indicated in Fig. 1 (S40, T41, Y68, F70, N71, G120, F124, E165, and G166) surround the substrate-binding pocket and were, therefore, prioritized for mutagenesis. For each of these target residues, a pool of all possible codon variants was generated and cloned into the bacteriophage λ vector, λJG3J1 (Fig. 2,A). The resulting λ libraries of NTR mutants were used to generate bacterial lysogens containing single, chromosomally integrated, copies of the λ prophage in E. coli UT5600, a strain in which the endogenous NTR gene (nfsB) is deleted (21). Clones were replica-plated onto agar containing a range of concentrations of CB1954; representative examples are shown in Fig. 2 B. Each clone was given a numerical score from 0 to 9, based on visual evaluation of its growth on the different concentrations of prodrug. The vector control with no inserted NTR gene (Vec) showed strong growth at all prodrug concentrations tested (up to 400 μm), and scored 0. Lysogens expressing WT NTR were able to grow at 100 but not at 200 μm CB1954, and scored 4. Of the mutants, some conferred less sensitivity to CB1954 than WT (e.g., clone 227, score 2), whereas others conferred greater sensitivity (e.g., clone 199, score 9).

Table 1 lists the total number of clones screened for each site and their prodrug sensitivity scores. Although, as expected, most mutants conferred less sensitivity to prodrug than WT NTR, suggesting impairment of CB1954 activation by the NTR variants, mutants showing increased sensitivity to CB1954 were obtained at seven of the nine positions. Several clones (2–8%) with improved prodrug sensitivity compared with WT were detected among the mutants at sites S40, T41, Y68, N71, and G120. Most strikingly, 32% of mutants at F70 and 43% at F124 conferred greater prodrug sensitivity than did WT. The NTR genes of 547 clones with CB1954 sensitivity similar to or greater than WT (score ≥4) were sequenced. Fifty different amino acid substitutions were identified that resulted in increased prodrug sensitivity. Fig. 2 C shows the substitutions identified at each position and their mean sensitivity scores.

A selection of higher-scoring mutants at each position were assayed more quantitatively in a colony-forming assay by individually plating the bacterial lysogens at low density over a range of prodrug concentrations. The resulting colonies were counted and plotted as percentage survival versus prodrug concentration; representative examples are shown in Fig. 3,A. For each mutant assayed, the IC₅₀, (i.e., the prodrug concentration that resulted in 50% reduction in colony number), is shown in Fig. 3 B. The average IC₅₀ determined for clones with WT NTR was 143 μm CB1954. Specific amino acid substitutions at NTR residues S40, T41, Y68, F70, N71, and F124 resulted in lower IC₅₀s, i.e., increased sensitivity of the bacteria to CB1954. Eight different amino acid substitutions at residue F124 resulted in IC₅₀s below 50 μm CB1954. The lowest IC₅₀s (25–30 μm) were obtained for the mutants F124N and F124K.

To compare their efficacy in human tumor cells, the F124K mutant and WT NTR were inserted separately into the replication-defective adenovirus vector vPS1233. Human SK-OV-3 ovarian carcinoma cells infected with 10 or 100 pfu of either virus per cell were treated with a range of concentrations of CB1954. As shown in Fig. 4 A, uninfected cells were not affected by CB1954 even at 300 μm, whereas those infected with virus expressing NTR were all killed at this prodrug concentration. At the MOI of 100 pfu/ cell, the IC₅₀ for cells expressing WT NTR was ∼1.5 μm CB1954, whereas that for cells expressing the F124K mutant was <0.5 μm (3.2-fold reduction). When used at just 10 pfu/cell, the IC₅₀ obtained with WT NTR was 14 μm CB1954, whereas that with F124K NTR was 2.5 μm (5.6-fold reduction).

In a similar experiment, the multiplicity of virus infection was varied, and the cells were exposed to 1 μm CB1954 (Fig. 4 B). To achieve 50% cell killing at this prodrug concentration required 100 pfu of the virus expressing WT NTR per cell, but only ∼20 pfu of the virus expressing F124K NTR per cell. The experiments were repeated with a second batch of each virus and showed similar improvement in IC₅₀ for CB1954 (3.7-fold at MOI 10 or 100) and in virus dose required to achieve comparable efficacy at 1 μm CB1954 (5.7-fold reduction) for the F124K NTR. These results are typical of at least five separate experiments.

NTR, in combination with the prodrug CB1954, has entered clinical trials for cancer gene therapy. However CB1954 is a relatively poor substrate for NTR, with a slow turnover (catalytic rate constant k_cat ∼ 360 min⁻¹) and high K_m (∼860 μm; Ref. 27). The peak serum concentrations of CB1954 that can be achieved clinically are well below this K_m, in the range 1–10 μm (13). Thus, although the clinical efficacy is still unknown, it is clear that enhancing the efficiency of CB1954 activation by NTR, particularly at low substrate concentration, would increase the yield of activated prodrug and should improve the potential for clinical success.

From the crystal structure of NTR bound to nicotinic acid, we identified nine residues that could directly affect substrate binding and/or catalysis, and mutated these individually. Libraries of random mutants at each position were screened for their ability to sensitize a NTR-deficient strain of E. coli to CB1954. For each site mutated, at least 300 clones, and in most cases >500, were screened to minimize the likelihood that any of the 64 possible codons had not been sampled. In the 547 clones sequenced, the base frequencies at the NNN positions of the randomized codons were similar (32% A, 26% C, 21% G, 21% T). Thus, it is likely that all possible codons were represented within the libraries screened.

Of the nine targeted residues, the only WT amino acid residue found to be essential for CB1954 reduction is G166. Additionally, no improved variants were found at the neighboring residue E165, and all of those with WT levels of activity retained glutamate. At G120, substitutions with A, T, and S were found but conferred no significant benefit. At the remaining six sites, several amino acid substitutions resulted in clearly improved sensitization of the bacteria to CB1954. At S40, the same range of small amino acids were found as at G120, giving in this case a ∼30% reduction in IC₅₀ for CB1954. The adjacent T41 can be substituted for by the long hydrophobic residues leucine or isoleucine, by the polar residues asparagine or serine, and also by glycine, producing up to a ∼60% reduction in IC₅₀. The sidechain of N71 can be substituted by the polar residues serine, threonine, aspartate, and glutamine. The shortest of these, serine, gave >60% reduction in IC₅₀ for CB1954 in the bacterial assay.

The three residues Y68, F70, and F124 yielded a diverse range of amino acid substitutions. At position 68, glycine and asparagine gave the most activity with CB1954 (∼50% reduction in IC₅₀). At F70, the IC₅₀s for the six substitutions assayed (alanine, glutamate, leucine, proline, serine, and valine) were all similar, below 50% that of WT. The similar IC₅₀s of such diverse substitutions suggests that the improvement results from the removal of an inhibitory interaction of F70, allowing better substrate access or product release.

Phenylalanine F124 is the residue at which amino acid substitutions resulted in the greatest improvements over WT. In the structure of the enzyme complex with nicotinic acid, the phenyl ring of F124 stacks with the aromatic ring of the ligand. This may increase the affinity for aromatic substrates such as CB1954 but may impede the reaction if it sterically restricts prodrug access to the site, positions the prodrug suboptimally, or delays the release of reaction products. Fifteen amino acid substitutions at F124 resulted in increased prodrug sensitivity; only aspartate, glutamate, proline, and arginine substitutions were not recovered. However in contrast to the situation at F70, the degree of improvement with the CB1954 substrate varied markedly depending on the amino acid substitution. Zenno et al. (28) found that, whereas, unlike WT, several F124 NTR mutants could reduce flavin substrates, only one of these (F124H) had increased activity (2-fold) toward nitrofurazone and nitrofurantoin. In our study, we have found that several substitutions (F124N, -K, -A, -M, -V, -L, -H, and -G) gave IC₅₀s below 50 μm CB1954 in the bacterial assay, ∼3–5-fold lower than WT. Clearly, F124 imposes a major restriction on substrate binding, and different residues at this position are optimal for different ligands.

Our strategy was based on the expectation that assays in E. coli would be predictive of the relative efficacy of NTR mutants in sensitizing human tumors to CB1954. We tested this assumption for the mutant F124K, for which there was ∼5-fold reduction in IC₅₀ relative to WT in E. coli. The WT and F124K NTR genes were inserted into adenovirus vectors, and tested for their ability to sensitize SK-OV-3 human ovarian carcinoma cells to CB1954. When tested at fixed MOI, the IC₅₀ for CB1954 was, on average, ∼4-fold lower for F124K than for WT NTR. At a fixed concentration of 1 μm CB1954, ∼5-fold less adenovirus expressing the F124K NTR was required than with WT NTR to achieve 50% cell kill. Similar data were obtained with two independent batches of the viruses and also when the relative virus doses were compared on the basis of expression of the internal GFP reporter gene rather than by titer (results not shown). These results demonstrate a good correlation between the bacterial and tumor cell assays and demonstrate directly that the F124K NTR is about 5-fold more efficient than the WT at sensitizing human tumor cells to CB1954.

The enzyme HSV-TK has also been subject to mutagenesis, to obtain variants more suitable for GDEPT (15, 16). One of these, mutant 30, conferred about 1000-fold greater sensitivity to GCV than did WT in transfected rat glioma cells (17). However, in prostate cancer models, this mutant was less effective than WT (29). When characterized kinetically, it was found that GCV was actually a worse substrate for mutant 30 than for WT, but the mutation was even more detrimental to thymidine metabolism. This led to the suggestion that reduced competition for the active site by intracellular thymidine accounted for the enhanced cell sensitization to GCV in the glioma cells (17). The variation in efficacy of this mutant in different model systems could be attributable to differences in the intracellular thymidine pool. For NTR, in vitro kinetic analysis of the F124K mutant shows that the K_m for CB1954 has been reduced relative to WT; hence, this mutation directly improves the activity for this substrate.⁴

Other groups have sought to improve the efficacy of the NTR enzyme/prodrug system by investigating alternative prodrug substrates, including several nitrogen mustard analogues of CB1954. Of these, SN23862 {5-(N,N-bis[2-chlorethyl]amino)-2,4-dinitrobenzamide} has been studied most extensively. Although the k_cat for this substrate is ∼3–4-fold higher than for CB1954, the K_m is also increased by a similar factor (30). In cell killing assays in vitro, CB1954 shows a comparable or lower IC₅₀ than does SN23862 in most published reports (30, 31) and in our own data.⁵ A recent report found improved bystander killing with SN23862 in vitro; however, it was less effective than CB1954 in a mouse tumor model (32). The bisbromoethyl analogue appears more promising (31, 32), but kinetic data with this substrate have not yet been reported. Clearly, these two approaches are not incompatible, and the most effective therapy may ultimately involve improved, alternative prodrugs in combination with engineered enzyme variants.

In this article, we have described the generation of a number of NTR mutants that provide improved sensitization of E. coli to CB1954. Using an adenovirus vector to express the F124K mutant in human ovarian tumor cells, we demonstrated a ∼5-fold improvement over WT NTR at clinically achievable prodrug concentrations. Furthermore, because substitutions resulting in enhanced prodrug sensitization of E. coli were found at six positions within NTR, this raises the possibility that further improvement could result from combinations of the single-site mutants described here. Enhanced mutant NTRs exemplified by the mutant F124K offer a significant improvement over WT NTR for use with CB1954 in cancer gene therapy and form part of a “second generation” of therapeutic genes in which natural functions are being specifically adapted to their clinical applications.

We thank Dr. F. Bernard for assistance with the analysis of F70 mutants; Dr. D. Cronin for assistance producing Fig. 1; and Dr. V. Mautner, Prof. L. S. Young, and Prof. D. J. Kerr for helpful discussions.