Nitro-chloromethylbenzindolines: hypoxia-activated prodrugs of potent adenine N3 DNA minor groove alkylators

Severe hypoxia arises in tumors as a result of a structurally and functionally abnormal microvascular system and plays a major role in tumor progression (1). Hypoxia also represents an obstacle for all major forms of cancer therapy. For example, the higher frequency of metastasis from more severely hypoxic tumors potentially compromises outcomes of surgery (2) and radiotherapy (3). Hypoxia also confers resistance of the primary tumor to ionizing radiation (4) and compromises chemotherapy through multiple mechanisms (5). These include inaccessibility of hypoxic cells to blood-borne drugs, low proliferative activity in hypoxic regions of tumors, selection of cells with reduced sensitivity to apoptosis, and the induction of molecular responses to hypoxic stress that exacerbates drug resistance. These features make hypoxic tumor cells an important therapeutic target.

In addition, because hypoxia is more extensive and severe in tumors than in normal tissues (6), it represents a selective tumor target. These considerations have resulted in much interest in designing prodrugs that are activated selectively by metabolic reduction within hypoxic regions of tumors (5, 7, 8). These hypoxia-activated prodrugs (HAP) act as direct oxygen sensors through the ability of O₂ to inhibit the first step in their reductive metabolism, usually by reoxidizing the initial radical intermediate (9). Several HAPs are currently in clinical trial (N-oxides tirapazamine and banoxantrone, nitro compounds PR-104 and TH-302, and quinones apaziquone and RH1), and the approved drug mitomycin C has weak selectivity for hypoxic cells (10).

Preclinical and clinical evaluation of HAP has identified several critical requirements, which, arguably, are not present optimally in any of the above agents. In particular, some are substrates for competing activation by oxygen-independent two-electron reductases, as shown for apaziquone (10), RH-1 (11), and PR-104 (12), which limits their selectivity for hypoxia. Some penetrate poorly into hypoxic regions of tumors, as shown for tirapazamine (13) and apaziquone (14). In addition, tirapazamine is activated at relatively high oxygen concentrations (15), which may contribute to its toxicity in normal tissues (16, 17). Further, with the exception of PR-104 and related dinitrobenzamide mustards (18–20), there is little evidence for bystander effects resulting from diffusion of active (reduced) metabolites from hypoxic zones. This feature is likely to be important for successful exploitation of hypoxia by HAP because it removes the requirement for metabolic activation in every hypoxic cell and, in particular, allows killing of hypoxic cells at oxygen concentrations too high to activate HAPs but low enough to be resistant to conventional cytotoxics (18, 20). A further limitation with existing HAP is that the resulting active metabolites generally fall within chemical classes that are subject to multidrug resistance (e.g., nitrogen mustards and topoisomerase II poisons), which may limit their utility.

These considerations have led us to explore a novel class of HAP with the potential to address these limitations. The cyclopropylindoline antitumor antibiotics, exemplified by CC-1065 and duocarmycin SA, alkylate N3 of adenine in a sequence-selective manner in the minor groove of DNA and are some of the most potent cytotoxic agents known (21). Several analogues have shown antitumor activity in preclinical models (22), and four (adozelesin, carzelesin, bizelesin, and KW-2189) were evaluated clinically. All lacked antitumor activity at their maximum tolerated doses (MTD), which were limited by myelotoxicity to which humans seemed to be considerably more sensitive than mice (23). We have developed synthetic analogues of the cyclopropylindolines in which the key phenol of the alkylating subunit is replaced by an amino functional group (amino-chloromethylindolines; refs. 24–26). This structural change generates agents that share many of the same properties as the phenolic compounds (26–28) but allows the possibility that the corresponding nitro compounds might act as HAP through bioreduction via an initial oxygen-sensitive one-electron step to the amino-chloromethylindolines. We have shown that one particular example of a nitro-cyclopropylindoline [termed nitro-CI-5,6,7-trimethoxyindole (TMI) in the present study] becomes 400-fold more cytotoxic in vitro when activated by the Escherichia coli nfsB nitroreductase (NTR; ref. 29), but have not previously examined whether these compounds can be activated by mammalian enzymes under hypoxic conditions.

Here, we evaluate a series of nitro-cyclopropylindoline derivatives to identify candidates with potential as HAP and explore the mechanism of action and therapeutic utility of a preferred subclass (nitro-CBIs) in preclinical models.

The amino and nitro compounds of Fig. 1A were synthesized as reported previously (25, 26, 30, 31). Stock solutions in DMSO were stored at −80°C. Chlorambucil was purchased from Sigma. Tirapazamine and the NTR substrates CB 1954 and SN 27686 (32) were synthesized in this laboratory. The adenine adduct of amino-CBI-TMI was prepared by reaction with calf thymus DNA followed by thermal depurination, essentially as described for amino-CI-TMI (27); details of adduct isolation and characterization are given in Supplementary Materials and Methods.

Origins of the repair-competent Chinese hamster ovary cell line AA8 and its repair-defective mutants are given in Supplementary Table S1. Sources of the human tumor cell lines (HT29, SKOV3, HCT8, and HCT116) are as previously described (19). RIF-1 and EMT6 murine tumor lines were from Dr. J. Martin Brown (Stanford University). The origin of the HCT116-NTR^puro line, which stably expresses NTR, has been described (32). All cell lines were maintained in α-MEM containing 5% to 10% of fetal bovine serum without antibiotics and reestablished every 3 mo from frozen stocks confirmed to be free of Mycoplasma by PCR-ELISA (Roche Diagnostics).

Antiproliferative potency was determined using 4-h exposure of cells in 96-well plates under aerobic conditions at pH 7.4 as previously described (19) or at pH 6.5 by acidification acutely during drug exposure or chronically for 18 h before and during drug exposure. Details of adjustment of pH are given in Supplementary Materials and Methods. Clonogenic assays for cell sterilization under aerobic and hypoxic conditions were done using stirred and continuously gassed (5% CO₂ in air or N₂, respectively) cell suspensions (10⁶ cells/mL) as previously described (20). The hypoxic cytotoxicity ratio (HCR) was estimated as the ratio of concentrations for 10% surviving fraction (C₁₀) under aerobic and hypoxic conditions.

Three-dimensional multicellular layer (MCL) cultures were grown from HCT116 cells (targets) or from 1:1 mixtures of targets and activators (HCT116-NTR^puro) as previously described (32). MCLs were exposed to nitro-CBI-5-[(dimethylamino)ethoxy]indole (DEI) for 5 h in stirred vessels under 5% CO₂/95% O₂ (to suppress hypoxia and thus endogenous metabolism by one-electron reductases), dissociated with trypsin, and plated with and without 3 μmol/L puromycin to determine clonogenic survival (32). The interpolated C₁₀ values were calculated for targets alone (T), targets in cocultures (T_C), and activators in cocultures (A_C).

Cell suspensions (2.5 × 10⁶ or 5 × 10⁶ cells/mL) were treated with nitro-CBI-DEI as for clonogenic assays. At various times, samples of 100 μL were withdrawn and precipitated with 400 μL acetonitrile (MeCN) containing 25 μmol/L bis(2-chloroethyl)-N-methyl-N-(4-methyl-2-nitrobenzyl)ammonium chloride (33) as internal standard (IS). Alternatively, cell suspensions were rapidly centrifuged, 100 μL of extracellular medium were extracted with 400 μL MeCN/IS, and the cell pellets were extracted with 100 μL MeCN/IS. Samples were stored at −80°C, thawed, and centrifuged (12,000 × g for 5 min), and the supernatant was reduced to∼50 μL under vacuum. After addition of 100 μL of 0.45mol/L ammonium formate buffer (pH 4.5)/MeCN (1:3, v/v), 100 μL of samples were analyzed by liquid chromatography/mass spectrometry (LC/MS) using an Agilent LC/MSD with a Zorbax 2.1 × 150 mm column (5 μm C8 packing) and a 45 mmol/L aqueous ammonium formate buffer (pH 4.5)/MeCN gradient. Detection was by diode array absorbance (nitro-CBI-DEI) and electrospray (positive mode) MS in scan mode or by selected ion monitoring at m/z 305 (IS) and 463 (amino-CBI-DEI), with a drying gas flow of 12 L/min, nebulizer pressure of 40 psig, drying gas temperature of 350°C, and capillary voltage of 4,000 V. Calibration curves (analyte/IS ratio versus analyte concentration) were constructed by spiking nitro-CBI-DEI or amino-CBI-DEI into each matrix (medium, cell pellets, and whole culture) and extracting with MeCN/IS in the same manner; for cell pellets, the analytes were added to the MeCN/IS solution.

Single-cell suspensions (10⁷ cells/mL) were prepared from RIF-1 tumors by enzymatic digestion using a cocktail of Pronase (0.5 mg/mL), collagenase (0.2 mg/mL), and DNase (0.2 mg/mL). Cells were exposed to amino-CBI-TMI or nitro-CBI-TMI as for clonogenic assays for 3 h. Cells were collected by centrifugation and pellets were lysed in 1 mL water containing 0.5 μmol/L of the previously reported (27) amino-CI-TMI adenine adduct as IS and heated to 70°C for 1 h to depurinate alkylated adenine bases. After cooling, samples were extracted with 1 mL dichloromethane in glass tubes and the organic layer was evaporated to dryness and reconstituted in 100 μL of DMSO and 30 μL was analyzed by LC/MS essentially as above by using an Alltima 3.2 × 150 mm column (C8 5 μm packing). Positive mode electrospray ionization was used for selective ion monitoring of nitro-CBI-TMI (m/z 496), amino-CBI-TMI (m/z 466), and the amino-CI-TMI and amino-CBI-TMI adenine adducts (m/z 515 and 565, respectively).

Animal studies were approved by the University of Auckland Animal Ethics Committee (approval R256) and used specific pathogen-free animals bred and housed in the University of Auckland. Compound formulation and dosing is described in Supplementary Table S4. Initially, groups of three male C₃H/HeN mice were observed for 60 d, and the dose was adjusted using 1.33-fold (1/8th log₁₀) dose increments as required to determine the MTD, which was defined as the highest dose causing no deaths or morbidity requiring euthanasia (e.g., >15% body weight loss) in a group of six mice.

RIF-1 tumors were passaged alternately in vitro and in vivo, in female C₃H/HeN mice, using Twentyman's protocol (34). HCT8 and HT29 tumors were grown in female homozygous nude (CD1-Foxn1^nu) mice (Charles River Laboratories) by s.c. inoculation of 10⁷ cells from tissue culture. Tumor growth was monitored thrice weekly with electronic calipers, and mice were randomized to treatment (specified in the figure legends) as tumors reached a mean diameter of 8 to 10 mm. Tumors were collected 18 h after treatment and dissociated enzymatically to determine clonogenic survivors per gram tumor as previously described (35). For tumor growth delay studies, RIF-1 tumors were grown in the gastrocnemius muscle and monitored by measurement of leg diameter. Mice were treated at a leg diameter of 9.5 to 10.5 mm, corresponding to a tumor weight of 0.4 to 0.6 g, and monitored until the end point (leg diameter of 13 mm, corresponding to 1.4 g tumors). Tumor growth delay was calculated as the difference in time to end point for control and treated tumors. Significance of treatment effects (growth delay or log cell kill for clonogens) was tested using ANOVA with Dunnett's multiple comparison test.

Three previously reported (26, 31, 36) synthetic variants of the alkylating subunit were evaluated (CI, DSA, and CBI) in combination with either a TMI or DEI sidechain; the structures are illustrated in Fig. 1A. Consistent with previous studies (26, 31, 36), the amino compounds (Fig. 1A, R = NH₂) potently inhibited cell proliferation in aerobic cultures (Fig. 1B). Potency increased in the order CI-TMI < DSA-TMI < CBI-DEI < CBI-TMI, with the latter providing subnanomolar IC₅₀ values following 4-hour drug exposure. With the exception of the less potent amino-CI-TMI, the murine cell lines (RIF-1 and EMT6) were consistently more sensitive than the Chinese hamster (AA8) or human (HT29 and SKOV3) cell lines.

The cytotoxic potency of DNA minor groove alkylators may in part reflect the difficulty in repairing the resulting DNA lesions (37). We compared the sensitivity of Chinese hamster ovary cells with defects in specific DNA repair pathways to amino-CBI-DEI and the reference alkylator chlorambucil, which primarily cross-links guanine in the major groove. Cells with defects in the ERCC1-XPF endonuclease, which has a critical role in both nucleotide excision repair and homologous recombination repair (HRR; refs. 38, 39), or the Rad51 paralog Rad51D, which is essential for HRR (40), were markedly hypersensitive to chlorambucil (Fig. 1C; absolute IC₅₀ values are shown in Supplementary Table S1) as recently reported (41). In contrast, the cytotoxicity of amino-CBI-DEI was less influenced by repair phenotype. Even in the Rad51D^−/− cell line (51D1), which showed the greatest amino-CBI-DEI sensitivity relative to the corresponding isogenic wild-type (51D1.3), the differential was only one third of that for chlorambucil.

The nitro-chloromethylindolines were much less toxic than the corresponding amino compounds in all cases (Fig. 1B), with a pattern of sensitivity similar to that of the matching amino compound. Nitro/amino IC₅₀ ratios were ∼3,000 for CBI-TMI and in the range 90 to 480 for DSA-TMI and CBI-DEI, with the CI-TMI ratio indeterminate because of limited solubility of the nitro compound. These large nitro/amino differentials suggested the nitro compounds as candidate HAPs.

Because hypoxic cells often reside in an acidic microenvironment (42), we also tested whether lowering of extracellular pH from 7.4 to 6.5, either chronically for 18 hours before and during drug exposure or acutely during drug exposure, affected IC₅₀ values for EMT6 cells (Fig. 1D; absolute IC₅₀ values are shown in Supplementary Table S2). Chlorambucil, with a weakly acidic sidechain, showed a 5.5- and 6.5-fold increase in potency when pH was lowered to pH 6.5 acutely or chronically, respectively, as expected because of enhanced cellular uptake of weak acids at low extracellular pH (43). No significant effects of pH were observed for the neutral (TMI sidechain) compounds, but the two examples with a basic DEI sidechain showed a significant 2.8- to 5-fold reduction in potency at pH 6.5.

To test whether the nitro compounds are selectively toxic under hypoxic conditions, we exposed UV4, RIF-1, and HT29 cell lines as stirred single-cell suspensions under aerobic and hypoxic conditions and assessed cell killing using a clonogenic survival assay (Fig. 2). The illustrated dose response for nitro-CBI-TMI against UV4 cells (Fig. 2A) shows a large HCR of 300-fold. In contrast, amino-CBI-TMI showed no hypoxic selectivity (Fig. 2A). The ∼20-fold lower potency of nitro-CBI-TMI than amino-CBI-TMI under hypoxia suggests that it is only partially metabolized to the amine under these conditions. Comparison of the four nitro compounds (Fig. 2B) showed a similarly large HCR for nitro-CBI-TMI in RIF-1 cells (∼200-fold) but lower potency and selectivity against HT29. This same pattern was seen with nitro-DSA-TMI and nitro-CBI-DEI, although with lower HCR values (Fig. 2B).

To determine the mechanism of this hypoxia-selective cytotoxicity, we used stirred cell suspensions at high density (5 × 10⁶/mL) to test whether UV4, RIF-1, and HT29 cells reduce nitro-CBI-DEI. Cellular uptake of the nitro compound (20 μmol/L) by HT29 cells was rapid (<5 minutes) and extensive, with intracellular concentrations ∼5,000-fold higher than extracellular under both aerobic and hypoxic conditions (Supplementary Table S3). There was no significant change in total (intracellular plus extracellular) nitro-CBI-DEI over at least 4 hours in all three cell lines under either aerobic or hypoxic conditions (data not shown). However, amino-CBI-DEI was detected as a metabolite in hypoxic cultures of all cell lines by LC/MS, its structure being confirmed by comparison of retention time, absorbance spectrum, and mass spectrum with a synthetic standard (Supplementary Fig. S1). Quantitation of amino-CBI-DEI in HT29 suspensions during hypoxic incubation with 10 μmol/L nitro-CBI-DEI showed an approximately linear increase in the cells but no increase with time under aerobic conditions (Fig. 3A). By 2 hours, low concentrations of amino-CBI-DEI could also be detected in the extracellular medium of the hypoxic cultures (Fig. 3B). Summing the metabolite in both the intracellular and extracellular compartments (data not shown) gave an initial rate of metabolism of ∼30 pmol/mL culture/h corresponding to 0.3% conversion of the prodrug per hour at this cell density (5 × 10⁶ cells/mL), which is too slow a conversion to detect loss of the parent nitro compound. At half this cell density, all three cell lines (UV4, RIF-1, and HT29) showed broadly similar rates of hypoxic metabolism of nitro-CBI-DEI to amino-CBI-DEI (Fig. 3C). Reduction of nitro-CBI-DEI to amino-CBI-DEI was also detected in hypoxic S9 preparations from HT29 cells and, at higher rates, from mouse liver (data not shown), whereas under aerobic conditions the sole metabolite identified (by MS² using ion trap MS) in mouse liver S9 was the tertiary amine N-oxide resulting from oxidation of the (dimethylamino)ethoxy sidechain (data not shown).

We showed reaction of amino-CBI-TMI with N3 of adenine in DNA by incubating the compound with calf thymus DNA at a 1:25 compound/base pair ratio followed by thermal depurination (see Supplementary Data for details). The major dichloromethane soluble product (89% yield) was the adenine N3 adduct 1 (Fig. 4A), which was fully characterized by nuclear magnetic resonance spectroscopy and MS. As shown in Fig. 4B, the same adenine adduct (1) was detected in RIF-1 cells following hypoxic incubation with nitro-CBI-TMI and heating to induce depurination, along with amino-CBI-TMI and an elimination product (loss of HCl) from nitro-CBI-TMI (2; Fig. 4A). The elimination product 2 was also seen in aerobic RIF-1 cells subject to the same heat treatment, but the reduction products (amino-CBI-TMI and adduct 1) were not (Fig. 4B). Quantifying the adenine adduct by MS showed a concentration-dependent increase in the amino-CBI-TMI adduct (1) when RIF-1 cells were incubated under hypoxia with either amino- or nitro-CBI-TMI (Fig. 4C).

The observation that hypoxic cells exposed to nitro-CBI-DEI release amino-CBI-DEI into the extracellular medium (Fig. 3B) raised the question as to whether diffusion of this metabolite out of hypoxic zones might provide a bystander effect. To test this possibility in an experimentally accessible manner, we sought a model system in which cells capable of reductively activating nitro-CBI-DEI could be cocultured with (but distinguished from) a second population of nonmetabolizers. The E. coli nfsB nitroreductase (NTR) is well characterized as a broad-spectrum two-electron oxygen-insensitive nitroreductase and is known to activate nitro-CI-TMI (29). HCT116 cells that stably express NTR (HCT116-NTR^puro cell line) showed the expected hypersensitivity to the dinitrobenzamides CB 1954 and SN 27686, which are known NTR prodrugs (Fig. 5A; ref. 32). Nitro-CBI-DEI showed a more modest (4.6-fold) but highly significant differential for the NTR-positive cells (Fig. 5A), suggesting this to be a poorer NTR substrate than the dinitrobenzamide prodrugs. However, NTR generated the same amino-CBI-DEI metabolite as endogenous reductases in hypoxic tumor cells (Supplementary Fig. S2), showing this to be a valid model for investigating bystander effects resulting from hypoxic metabolism of nitro-CBI-DEI.

We used the MCL three-dimensional cell culture model (18, 32) to test whether activation of nitro-CBI-DEI by HCT116-NTR^puro cells (“activators”) is capable of killing NTR-null HCT116 cells (“targets”) in these cocultures. MCLs were grown for 3 days from targets alone or 1:1 mixtures of activators and targets, the latter providing MCLs comprising 34.5 ± 2.5% (SE) activators as assessed by clonogenic assay of control MCLs in the presence and absence of puromycin. Determination of clonogenic cell killing (Fig. 5B) showed that cells in MCLs grown from targets alone were 12-fold less sensitive to nitro-CBI-DEI than NTR-positive activator cells in cocultures, as measured by the C₁₀ ratio T/A_c. However, killing of targets was greatly increased in the cocultures, with target cells now only 2-fold less sensitive than activators themselves (T_c/A_c ratio). This shows an efficient bystander effect presumably resulting from reduction of nitro-CBI-DEI to its diffusible amino metabolite in a tissue-like three-dimensional cell culture system.

Given the promising characteristics of the nitro-CBI compounds as HAPs in culture, we compared their toxicity to that of the corresponding amino metabolites following administration of single doses to mice. The relatively insoluble CBI-TMI compounds were formulated in 10% DMSO/40% polyethylene glycol/50% water (v/v/v) for i.p. administration. The MTD of amino-CBI-TMI (0.042 μmol/kg) was at least 500-fold lower than nitro-CBI-TMI, which was nontoxic at the highest dose that could be administered in solution (23.7 μmol/kg; Supplementary Table S4). Nitro-CBI-DEI could be formulated at higher concentrations, using 5% DMSO/20% polyethylene glycol/75% lactate buffer (pH 4.0; v/v/v), and gave MTD values of 31.6, 31.6, and 42.1 μmol/kg following i.p., i.v., and p.o. administration, respectively. Toxicity of amino-CBI-DEI varied markedly with route of administration, with MTD values of 0.0562, 0.562, and 7.5 μmol/kg for i.p., i.v., and p.o. dosing, respectively. Thus, the nitro-CBI-DEI was 560- to 5.6-fold less toxic than amino-CBI-DEI depending on the route of administration. Both amino compounds, and to a lesser extent nitro-CBI-DEI, at doses above the MTD caused late deaths (Supplementary Table S4). Deaths occurring >3 weeks after i.p. dosing were associated with i.p. organ adhesions, gastrointestinal tract obstruction, and histopathologic signs of chemical peritonitis at necropsy. The very low MTD of amino-CBI by the i.p. route presumably reflects this local toxicity, whereas its very high MTD by oral dosing suggests low oral bioavailability.

Given its superior formulation characteristics, evaluation of therapeutic activity focused on the basic analogue nitro-CBI-DEI. To test whether this kills hypoxic cells in tumors, we combined single i.p. doses of the drug with a dose of ionizing radiation sufficient to sterilize aerobic cells and assessed surviving clonogens 18 hours after treatment (Fig. 6A–C). At its MTD, nitro-CBI-DEI by itself was inactive against RIF-1 tumors; however, it showed significant activity when dosed 5 minutes after irradiation (Fig. 6A). This activity was at least as great as the reference bioreductive drug tirapazamine at its MTD (the difference between nitro-CBI-DEI and TPZ was not statistically significant; Fig. 6A). Against HCT8 xenografts, nitro-CBI-DEI after irradiation again increased overall tumor cell killing, and the trend was similar in a third tumor model (HT29 xenografts), although in this case the effect of nitro-CBI-DEI after irradiation was not statistically significant.

Investigation of the time course of interaction with radiation against RIF-1 tumors (Fig. 6B) showed an apparent minimum in activity for drug 60 minutes before irradiation. In this experiment (with just three mice per group), the combination of nitro-CBI-DEI with radiation was not statistically different from radiation alone at any individual time point, but pooling the values for drug after (but not before) irradiation showed highly significant activity (P = 0.005). The dose response for nitro-CBI-DEI against RIF-1 tumors showed weak monotherapy activity, which was only statistically significant above the MTD (Fig. 6C). Dosing with nitro-CBI-DEI 5 minutes after irradiation showed greater activity than radiation alone, both at and above the MTD. At each dose level, the log cell kill additional to radiation alone was significantly (P ≤ 0.05 by t test) greater than for drug only. Thus, nitro-CBI-DEI shows selective toxicity for hypoxic cells in RIF-1 tumors.

Given that the combination of TPZ 2 to 3 hours before cisplatin has been reported to be synergistic against RIF-1 tumors (44), we also evaluated nitro-CBI-DEI with cisplatin in this model using tumor growth delay as the end point (Fig. 6D). TPZ at MTD (270 μmol/kg) administered 2.5 hours before cisplatin provided a nonsignificant (P = 0.08) 3.3-day increase in growth delay over cisplatin alone. Nitro-CBI at its MTD (31.6 μmol/kg) gave modest but significant (P = 0.016) antitumor activity. Combination of nitro-CBI with cisplatin showed maximal activity when the compounds were coadministered, with addition of the nitro-CBI providing a highly significant (P = 0.0017) 5.6-day increase in tumor growth delay relative to cisplatin only.

This study shows that highly potent amino-chloromethylindoline DNA minor groove alkylating agents can be generated by reductive metabolism of the corresponding nitro compounds in hypoxic tumor cells. The nitro-chloromethylindolines, and especially the chloromethylbenzindoline subclass (nitro-CBIs), therefore represent a novel HAP design with potential for targeting hypoxic cells in tumors.

The striking (approximately 90- to 3,000-fold) difference in cytotoxic potency between nitro- and amino-CBI compounds in aerobic cell cultures indicates that the nitro group powerfully suppresses DNA alkylation. This is consistent with the proposal that a strongly electron-withdrawing nitro group will inhibit spirocyclization to the active form of the alkylating subunit, whereas bioreduction to an electron-donating group can flip this electronic switch and promote DNA alkylation (29). In the small panel of cell lines investigated (Fig. 1B), there was a ∼10-fold difference in aerobic sensitivity between the most sensitive (RIF-1) and resistant (HT29 and SKOV3) cell lines. The similar pattern for the nitro and amino compounds suggests that the differences in aerobic sensitivity to the nitro prodrugs are not a result of their aerobic activation in a subset of cell lines, unlike the situation with quinone or dinitrobenzamide HAPs, which are activated by variably expressed two-electron reductases (10, 12). Consistent with this, the lower hypoxic selectivity of the nitro prodrugs in HT29 relative to UV4 and RIF-1 reflects resistance under hypoxia rather than sensitivity under aerobic conditions [as would occur if, for example, the high activity of DT-diaphorase in HT29 cells (10) resulted in two-electron activation]. Thus, the HCR seems to be determined more by sensitivity to the amino metabolite than by differences in rates of reductive metabolism. The similarity in rates of hypoxic metabolism of nitro-CBI-DEI was confirmed by assay of the amino metabolite in these cell lines (Fig. 3C). Of note, the greater sensitivity of RIF-1 than HT29 cells to amino-CBI-DEI in vitro is paralleled by the greater sensitivity of RIF-1 than HT29 tumor xenografts (Fig. 6), although a larger series of cell lines needs to be investigated to determine what part the intrinsic sensitivity to the amino metabolites might play in determining antitumor activity.

The amino-CBIs represent attractive cytotoxic effectors for HAP design for reasons other than their high cytotoxic potency. One of their notable features is their insensitivity to DNA repair phenotype compared with the representative DNA interstrand cross-linking agent chlorambucil (Fig. 1C), suggesting that they represent “stealthy” lesions that are not readily recognized by repair pathways. Hypersensitivity of ERCC1 or XPF mutant cells to amino-CBI-DEI was even lower than reported for the achiral amino-CBI compound AS-I-145, for which mutation of either gene results in 7- to 8-fold hypersensitivity of Chinese hamster ovary cells (45). This insensitivity to ERCC1/XPF removes an important mechanism of resistance to cross-linking agents (46). HRR-defective Rad51D^−/− cells show greater hypersensitivity, as expected, given that the cyclopropylindolines such as adozelesin are known to cause replication fork arrest (47) and that HRR has a key role in resolving such lesions (48), although dependence on HRR again seemed less than for chlorambucil.

Perhaps of even greater significance for HAP design, the amino-CBI-DEI metabolite is clearly able to elicit a bystander effect through diffusion from the cell in which it is formed. Not only is the amine observed as an extracellular metabolite in single cell cultures (Fig. 3B), but experiments using MCLs show the ability of this metabolite to kill surrounding cells in a tissue-like environment (Fig. 5B). These experiments exploited the finding that the E. coli nfsB nitroreductase (NTR) reduces nitro-CBI-DEI to the same amino metabolite as hypoxic reductases in tumor cells (Supplementary Fig. S2), and showed very efficient killing of NTR-negative bystander cells. This feature is considered to be a critically important requirement for maximally exploiting tumor hypoxia (5, 18) and particularly for avoiding the potential problem that many cells in tumors may not be sufficiently hypoxic to activate nitro or quinone HAPs but may still be hypoxic enough to be resistant to standard chemotherapy or radiotherapy (20, 49).

The nitro-CBI compounds evaluated in mice were markedly less toxic than the corresponding amines (Supplementary Table S4), suggesting that they do not undergo extensive systemic activation. We focused on the nitro-CBI-DEI prodrug for evaluation of antitumor activity because of its higher aqueous solubility, recognizing, however, that its basic sidechain has the potential to compromise potency against hypoxic cells if they are in acidic microenvironments. Nitro-CBI-DEI showed selectivity for hypoxic cells in RIF-1 tumors as indicated by greater activity in combination with radiation than for the prodrug alone. This differential (∼2-fold; Fig. 6C) was much smaller than the hypoxic differential seen for RIF-1 cells in cell culture (50-fold; Fig. 2B). It is not yet clear whether this reflects the operation of a bystander effect (leading to hypoxia-dependent killing of oxic cells), inefficient killing of hypoxic cells because of poor extravascular diffusion of the prodrug, suppression of potency because of acidosis in hypoxic regions, or some other limitation. Although the antitumor activity of nitro-CBI-DEI against RIF-1 tumors is encouraging, significant activity (in combination with radiation and in combination with cisplatin) is only seen at or above the MTD (Fig. 6C and D). This inability to harness the full cytotoxic potential of the amino-CBI metabolite may reflect, in part, the relatively slow rates of cellular reduction of nitro-CBI-DEI by hypoxic cells in vitro (Fig. 3C). In summary, although systematic lead optimization is obviously required, nitro-CBI-DEI represents a promising proof of principle for a new class of HAPs.

W.R. Wilson is a stockholder; W.R. Wilson and A.V. Patterson are consultants; and W.R. Wilson, F.B. Pruijn, S.P. Syddall, H.D.S. Liyanage, and M. Tercel have stock options in Proacta, Inc., which holds licenses related to the title compounds.

We thank Alan E. Lee for assistance with the animal studies, Sonia Alix for MS, Dr. Maruta Boyd for nuclear magnetic resonance spectroscopy, Dr. L. Jonathan Zwi (all University of Auckland) for evaluation of histopathology, and Dr. Larry H. Thompson (Brookhaven National Laboratory) who generously provided the repair-defective cell lines.