Antibody-Drug Conjugates Designed to Eradicate Tumors with Homogeneous and Heterogeneous Expression of the Target Antigen

Conjugates of the anti-CanAg humanized monoclonal antibody huC242 with the microtubule-formation inhibitor DM1 (a maytansinoid), or with the DNA alkylator DC1 (a CC1065 analogue), have been evaluated for their ability to eradicate mixed cell populations formed from CanAg-positive and CanAg-negative cells in culture and in xenograft tumors in mice. We found that in culture, conjugates of either drug killed not only the target antigen-positive cells but also the neighboring antigen-negative cells. Furthermore, we showed that, in vivo, these conjugates were effective in eradicating tumors containing both antigen-positive and antigen-negative cells. The presence of antigen-positive cells was required for this killing of bystander cells. This target cell–activated killing of bystander cells was dependent on the nature of the linker between the antibody and the drug. Conjugates linked via a reducible disulfide bond were capable of exerting the bystander effect whereas equally potent conjugates linked via a nonreducible thioether bond were not. Our data offer a rationale for developing optimally constructed antibody-drug conjugates for treating tumors that express the target antigen either in a homogeneous or heterogeneous manner. (Cancer Res 2006; 66(6): 3214-21)

Antibody-drug conjugates, also called immunoconjugates, have been developed as anticancer agents with increased tumor selectivity through binding of their monoclonal antibody moiety to tumor-associated antigens (1). Antibodies that target different tumor cell-surface antigens have been linked to various types of cytotoxic compounds (2). Antibody conjugates with highly cytotoxic compounds, such as calicheamicins, maytansinoids, auristatins, CC1065 analogues (DC drugs), or novel taxoids, exhibit potent and selective killing of target tumor cells in vitro and in animal models (3, 4). Several maytansinoid conjugates are currently in clinical development (5, 6) and a conjugate of an anti-CD33 antibody with calicheamicin, gemtuzumab ozogamicin (Mylotarg), has been approved for the treatment of relapsed acute myeloid leukemia (7).

Antibody-cytotoxic drug conjugates are designed to bind selectively to, and then kill, antigen-positive cells. Consequently, the antitumor efficacy of such conjugates might be impaired if the target antigen is expressed in tumors in a heterogeneous fashion. It has been reported that several types of antibody-targeted cytotoxic agents, such as radioimmunoconjugates, immunoliposomes, antibody-directed prodrug therapy (ADEPT), and gene-directed enzyme prodrug therapy (GDEPT), can kill not only antigen-positive tumor cells but also proximally located antigen-negative tumor cells (8–10). Such killing was termed “bystander effect.” Little, however, is known about the efficacy of antibody-cytotoxic drug conjugates against tumors that express the target antigen in a heterogeneous manner. A conjugate of the anti-Le^y antibody BR96 with doxorubicin (BR96-Dox) was tested in rats for its ability to eradicate human tumor xenografts with heterogeneous Le^y antigen expression and found to be ineffective against antigen-negative or low-expressing cells (9, 10). In contrast, a conjugate of the anti-CanAg monoclonal antibody C242 with the cytotoxic maytansinoid DM1 was effective in mice against human tumor xenografts that expressed the antigen heterogeneously (11). It has been hypothesized (11, 12) that target cells efflux the drug, which then kills the neighboring cells independently of antigen expression.

Here we report studies elucidating the mechanism of this target cell–activated killing of bystander cells. We observed that killing of bystander antigen-negative cells through targeting of antigen-positive cells with a conjugate of the humanized C242 antibody (huC242), covalently linked either to DM1 or DC1 drugs, was dependent on the nature of the linker used in the conjugate. Conjugates with linkers that contained a disulfide bond, but not those linked through a thioether bond, showed extensive proximal cell killing both in vitro and in vivo. We found that antigen-positive cells process the disulfide-linked conjugate and release a maytansinoid drug that is highly cytotoxic to antigen-negative cells. Meanwhile, a processing of the thioether-linked conjugate by antigen-positive cells yields a poorly cytotoxic maytansinoid drug. Therefore, conjugates with limited or prominent bystander cytotoxicity can be designed through manipulation of their linker composition.

Immunoconjugates. For the preparation of immunoconjugates, the maytansinoid DM1 (N^2′-deacetyl-N^2′-(3-mercapto-1-oxopropyl)-maytansine) was synthesized from the microbial fermentation product ansamitocin P-3 as previously described (13). The synthesis of the analogue of CC-1065, 5-[(3-mercapto-1-oxopropyl)amino]-bis-indolyl-(seco)-1,2,9,9a-tetrahydrocyclopropa[c]benz[e]indol-4-one compound, DC1, has been reported elsewhere (14). The humanization of the C242 antibody (huC242) was done by the resurfacing method that has previously been described (15). Antibody-drug conjugates were prepared using N-succinimidyl-4-(2-pyridyldithio)pentanoate for disulfide linkage or N-succinimidyl-4-(N-maleimidomethyl)cyclohexane-1-carboxylate (SMCC) for thioether linkage as described elsewhere (11, 13). The immunoconjugates used in this study contained, on the average, 3.5 cytotoxic drug molecules, DM1 or DC1, per antibody molecule (Fig. 1).

Cell lines. COLO 205 (human colon adenocarcinoma, ATCC CCL-222), HL-60 (human acute promyelocytic leukemia, ATCC CCL-240), and Namalwa (human Burkitt's lymphoma, ATCC CRL-1432) cultures were maintained in RPMI 1640 supplemented with 10% heat-inactivated fetal bovine serum (FBS) and 50 μg/mL gentamicin sulfate. HT-29 (human colon adenocarcinoma, ATCC HTB-38), A375 (human malignant melanoma, ATCC CRL-1619), and HepG2 (human hepatocellular carcinoma, ATCC HB-8065) cultures were maintained in DMEM supplemented with 10% heat-inactivated FBS and 50 μg/mL gentamicin sulfate. SNU-16 (gastric carcinoma, ATCC CRL-5974) culture was maintained in modified RPMI 1640 (ATCC, 30-2003) supplemented with 10% heat-inactivated FBS and 50 μg/mL gentamicin sulfate. All cell lines were cultured in a humidified incubator at 37°C, 6% CO₂.

Binding of the huC242 antibody to CanAg. The binding of the huC242 antibody to CanAg-positive cells was evaluated by an indirect immunofluorescence assay using flow cytometry. Cells (5 × 10⁴ per well) were plated in a round-bottomed 96-well plate and incubated at 4°C for 3 hours with serial dilutions of huC242 antibody in 0.2 mL of α-MEM supplemented with 2% (v/v) normal goat serum (Sigma, St. Louis, MO). Each sample was assayed in triplicate. Control wells lacked huC242. The cells were then washed with 0.2-mL cold (4°C) medium and stained with fluorescein-labeled goat anti-human immunoglobulin G (IgG) antibody for 1 hour at 4°C. The cells were again washed with medium, fixed in 1% formaldehyde/PBS solution, and analyzed using a FACSCalibur flow cytometer (BD Biosciences, San Jose, CA).

Three-dimensional collagen cytotoxicity assay. COLO 205 cells (2 × 10⁵/mL) were treated with huC242-DM1 (1 × 10⁻⁸ mol/L of conjugated antibody, 24 hours incubation at 37°C, 6% CO₂), 1% formaldehyde (20 minutes incubation), or 5 Gy UV irradiation (UV Stratalinker 1800, Stratagene, La Jolla, CA). Treated cells were washed thrice with 20 to 30 mL of fresh culture medium and resuspended in culture medium at 1 × 10⁸/mL. Untreated COLO 205, Namalwa, A375, and HepG2 cells were washed and resuspended in a similar manner. One microliter of each of the untreated cell suspensions was then mixed with 0, 1, 2, 4, or 8 μL of the treated COLO 205 cells. One hundred microliters of ice-cold collagen gel solution (three-dimensional collagen cell culture system, Chemicon International, Temecula, CA) were added to the cell mixtures and the cell-collagen mixtures were dispensed immediately (5 μL/well) in round-bottomed 96-well plates. The plates were incubated at 37°C for 1 hour to allow polymerization of the collagen. Fresh culture medium (200 μL) was then gently added to each well. Five to six days later, the number of viable cells was determined using a colorimetric cell proliferation assay measuring the reduction of 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) to formazan by live cells. Fifty microliters of the MTT stock solution (5 mg/mL in PBS) were added to each well. The plates were incubated at 37°C for 3 hours and then centrifuged for 10 minutes at 860 × g. Culture medium was aspirated from each well, the formazan crystals were solubilized with DMSO (100 μL per well), and A_540nm was measured. The relative number of cells (cell survival) in each well was calculated by correcting for the medium background reading and then dividing each value by the average of the values in the control (untreated cells) wells.

Liquid culture cytotoxicity assay. Dilutions of immunoconjugates in the appropriate medium were added to wells of 96-well round-bottomed plates containing 2 × 10³ COLO 205, 2 × 10³ SNU-16, or 2 × 10³ Namalwa cells, or a mixture of 2 × 10³ COLO 205 with 2 × 10³ Namalwa cells, or a mixture of 2 × 10³ SNU-16 with 2 × 10³ Namalwa cells. The plates were incubated for 5 days at 37°C, 6% CO₂. Pictures of the cells in each well were then taken with a Nikon Diaphot 300 inverted microscope.

ELISA assays for the quantitation of huC242-DM1, huC242-SMCC-DM1, and maytansinoid drug released from the conjugate. The methods have previously been described (16). Briefly, to measure the concentrations of huC242-DM1, the conjugate from either the standard or the test samples was captured on ELISA plates coated with a murine anti-maytansinoid monoclonal antibody (ImmunoGen, Cambridge, MA) and detected with horseradish peroxidase–labeled donkey anti-human IgG (Jackson ImmunoResearch, West Grove, PA). To determine the concentration of the maytansinoid drug released from the conjugate, protein was precipitated from samples by addition of 5 volumes of ice-cold acetone. Precipitated protein was sedimented by centrifugation at 16,000 × g and the concentrations of released drug in the supernatant were determined using a competition ELISA as follows. Plates were coated with DM1 that had been conjugated to bovine serum albumin (BSA-DM1; ImmunoGen). Dilution samples of a maytansine standard solution and test samples were mixed with biotinylated murine anti-maytansinoid monoclonal antibody (ImmunoGen) and then incubated in the BSA-DM1–coated plate. Finally, the bound biotinylated anti-maytansinoid antibody was detected with streptavidin-horseradish peroxidase (Jackson ImmunoResearch).

Human xenograft tumor models in mice. Five-week-old, female CB-17 severe combined immunodeficient (SCID) mice were obtained from Taconic (Germantown, NY). One week later, mice were inoculated s.c. with one of the following cell lines: HT-29 (2 × 10⁶ cells per mouse), COLO 205 (1 × 10⁷), Namalwa (2 × 10⁶), or a mixture of COLO 205 and Namalwa cells (1 × 10⁷ and 2 × 10⁶ cells per mouse, respectively). Nine to thirteen days later, each mouse received daily i.v. bolus injections of PBS (control), huC242-SMCC-DM1, or huC242-DM1 containing 25, 75, or 150 μg/kg of conjugated DM1 on 5 consecutive days (qd × 5). Tumor dimensions were measured twice per week and the volume was calculated as 1/2 (L × W × H), where L is the length, W the width, and H the height of the tumor.

Immunohistochemistry. Tumor tissues were fixed in 10% formalin and embedded in paraffin. Immunohistochemical examination of the paraffin sections was done using murine C242 antibody to detect COLO 205 cells and murine anti-CD38 antibody (RDI-CD38abm-290, Research Diagnostic, Flanders, NJ) to detect Namalwa cells using the avidin-biotin-peroxidase (ABC) technique.

HuC242-DM1 kills nontarget cells proximal to its target cells in vitro. The immunoconjugate, cantuzumab mertansine or huC242-DM1, is effective in eradicating CanAg-expressing human tumor xenografts in mice, including those that express the antigen in a heterogeneous manner (11), but is not effective against CanAg-negative xenograft tumors.²

To determine if the huC242-DM1 conjugate associated with the CanAg-positive cells has the ability to kill neighboring nontargeted cells, we studied its behavior in vitro using antigen-positive and antigen-negative mixed cell cultures in a three-dimensional collagen matrix. Cells can be embedded into the matrix at high densities simulating the in vivo tumor environment.

In a first series of experiments, treated and untreated antigen-positive COLO 205 cells (Fig. 2) were used. One set of cells was incubated in culture medium with huC242-DM1 at 37°C for 16 hours. The unbound conjugate was then removed by extensive washing and increasing numbers of treated cells were mixed with samples of untreated cells to give ratios of treated to untreated cells of 0:1 (control), 1:1, 4:1, and 8:1, respectively. The cell mixtures were then embedded into collagen, incubated for 5 days, and the number of viable cells in each sample was determined in an MTT assay. In further control cultures, the immunoconjugate-treated COLO 205 cells were replaced by formaldehyde-treated or UV light–treated COLO 205 cells. The results are shown in Fig. 3A. The surviving fraction of cells progressively decreased in the cultures containing increasing numbers of conjugate-treated cells (Fig. 3A,, black columns). Thus, untreated cells were killed in the mixed cultures. This killing was conjugate dependent and not due to any inhibitory effect of dying COLO 205 cells on the untreated cell population because COLO 205 cells killed by either formaldehyde treatment or UV irradiation did not significantly affect the proliferation of the untreated cells (Fig. 3A , gray and white columns, respectively).

In a second series of experiments, COLO 205 cells treated with huC242-DM1, as described above, were mixed with antigen-negative cell lines at a ratio of 1:1, embedded in collagen, incubated for 5 days, and then assayed for cell growth in an MTT assay. The antigen-negative cell lines were of diverse origins: a Burkitt's lymphoma (Namalwa), a melanoma (A375), and a hepatocarcinoma (HepG2). As shown in Fig. 3B, huC242-DM1-treated cells inhibited the growth of all cocultured, antigen-negative cell lines. Namalwa cells were the most sensitive to the cytotoxic effect generated by conjugate pretreated COLO 205 cells, as indicated by a decrease in the surviving fraction to <0.2; HepG2 cells were the least sensitive with a surviving fraction of ∼0.7; and A375 and COLO 205 cells showed intermediate sensitivity.

We next examined whether the target cell–dependent killing of nontarget cells also occurs in liquid cell cultures. We used antigen-negative, nonadherent Namalwa cells as reporter cells in mixed cultures and semiadherent COLO 205 as target cells. The two cell types were plated together in a 96-well round-bottomed plate. Within a few hours, all cells settled together in close proximity in the middle of the well. The experiment included wells that contained 2 × 10³ Namalwa cells alone, 2 × 10³ COLO 205 cells alone, and a mixture of 2 × 10³ cells of each of the two cell lines, all in 0.2-mL liquid culture medium containing 1 nmol/L huC242-DM1. After incubation for 5 days at 37°C, we took photographs of the cell populations (Fig. 3C). In the absence of the conjugate (Fig. 3C^,, left column), the number of cells in all wells increased some 20-fold as determined by cell counting. The mixed cell populations were also analyzed by flow cytometry to confirm that both the Namalwa and COLO 205 cells had proliferated. Alexa-labeled huC242 served to stain CanAg-positive COLO 205 cells whereas the B cell–specific Alexa-labeled anti-CD19 antibody was used for detection of the Namalwa cells. This analysis revealed that after 5 days of proliferation, conjugate-free mixed cultures consisted of 40% COLO 205 cells and 60% Namalwa cells,³

showing that both cell types had proliferated at similar rates. As expected, huC242-DM1 killed most cells in wells containing CanAg-positive COLO 205 cells (Fig. 3C,, top row) and did not significantly affect the growth in wells containing antigen-negative Namalwa cells (Fig. 3C,, middle row). In wells with a mixed population of cells, the conjugate killed both COLO 205 and Namalwa cells (Fig. 3C , bottom row). Thus, huC242-DM1-treated COLO 205 cells can eradicate proximal antigen-negative cells in liquid culture as well as in a three-dimensional collagen cell culture.

To test if target cell lines other than COLO 205 can exert this cytotoxic effect onto bystander cells, we did an analogous experiment using another CanAg-positive cell line, the gastric carcinoma SNU-16 (Fig. 2). As shown in Fig. 3D, huC242-DM1 (1 nmol/L) killed SNU-16 cells, did not affect the proliferation of Namalwa cells, and killed the mixed Namalwa/SNU-16 cell population. Together, these experiments show that the in vitro bystander effect of huC242-DM1 is not limited either to a particular type of CanAg-positive target cell or a particular type of CanAg-negative bystander cells.

Bystander effect of immunoconjugates is dependent on the nature of the antibody-drug linker. HuC242-DM1 consists of DM1 molecules conjugated to the huC242 antibody through a disulfide-containing linker. This linker is readily reduced by thiols in vitro in a cell-free environment (data not shown), and therefore is likely to be cleavable by abundant cell-associated thiols. To test whether the presence of the disulfide bond in the linker is required for generation of the bystander effect, we synthesized huC242-SMCC-DM1, a conjugate in which DM1 is linked to the antibody via a nonreducible thioether bond (Fig. 1), and tested its in vitro cytotoxicity on COLO 205 cells, Namalwa cells, and mixed COLO 205/Namalwa cell populations. HuC242-SMCC-DM1 was as potent as huC242-DM1 in killing COLO 205 target cells with an IC₅₀ value of 4 × 10⁻¹¹ mol/L (data not shown). This cytotoxicity was CanAg selective because neither of the conjugates was cytotoxic for the antigen-negative Namalwa cells in the entire concentration range tested (up to 1 × 10⁻⁹ mol/L). Unlike huC242-DM1, huC242-SMCC-DM1 displayed only marginal, if any, cytotoxicity on bystander cells in either the three-dimensional collagen matrix assay or liquid culture assay (Fig. 3B and C, respectively), suggesting that the mechanism of the bystander effect for huC242-DM1 includes a disulfide bond cleavage step.

Bystander effect is not limited to antibody-maytansinoid conjugates. To determine if the bystander effect is a unique property of antibody-maytansinoid conjugates, we examined conjugates of huC242 with DC1. DC1 is an analogue of the minor groove–binding DNA alkylator CC-1065 and differs from DM1 in both its structure (Fig. 1) and mechanism of action. Previously, we had reported potent antigen-selective conjugates of DC1 with an anti-CD19 antibody and an anti-CD56 antibody (14). We constructed analogous conjugates of DC1 with huC242, in which the drug and the antibody were conjugated via either a disulfide-containing linker (huC242-DC1) or a thioether-containing linker (huC242-SMCC-DC1) as shown in Fig. 1, and tested their ability to kill antigen–positive and bystander cells in vitro. At concentrations between 2 × 10⁻¹⁰ and 2 × 10⁻⁹ mol/L, either conjugate killed most of the cells in antigen-positive COLO 205 cultures (Fig. 3C,, top row, results for the 1 × 10⁻⁹ mol/L concentration) but not the cells in antigen-negative Namalwa cultures (Fig. 3C,, middle row), confirming that their cytotoxicity is antigen dependent. When tested on mixed target/nontarget cell populations, huC242-DC1 was able to kill most cells of both kinds whereas huC242-SMCC-DC1 was not (Fig. 3C , bottom row). These experiments show that the ability to kill bystander cells is not an exclusive property of DM1 conjugates but applies to other antibody-drug conjugates linked by disulfide-containing bonds.

Bystander effect of huC242-DM1 is generated through processing of the conjugate by target cells and release of a cytotoxic maytansinoid into the medium. To define the mechanism of the bystander effect, we determined if CanAg-expressing cells treated with huC242-DM1 release a cytotoxic compound into the medium, which then diffuses to neighboring cells. Target COLO 205 or antigen-negative Namalwa cells were incubated with huC242-DM1 or huC242-SMCC-DM1 (10⁻⁷ mol/L) at 37°C for 0, 6, 24, or 48 hours. The cells were removed from the medium by centrifugation and the supernatants were assayed for the presence of huC242-DM1 and free maytansinoids using two different ELISA methods (see Materials and Methods). The results are shown in Fig. 4A, to C. After a 48-hour incubation with the target cells, the concentration of either conjugate in the supernatant decreased ∼3-fold and, concomitantly, the concentration of a free maytansinoid species increased ∼4-fold (Fig. 4A and C). A similar incubation of huC242-DM1 with CanAg-negative Namalwa cells did not lead to either disappearance of conjugate or appearance of a free maytansinoid in the medium (Fig. 4B), indicating that binding to the cell-surface antigen is necessary for the processing of these conjugates and release of maytansinoid drugs into the medium.

The cytotoxic activity of the released maytansinoid drugs was then assayed by exposing CanAg-negative Namalwa cells to the conjugate-containing medium conditioned with COLO 205 cells. The nonspecific cytotoxicity of huC242-DM1-containing medium gradually increased ∼10-fold over 48 hours, as shown by the change from an initial IC₅₀ value of 2 × 10⁻⁸ mol/L (supernatant harvested immediately after mixing, without incubation) to 3 × 10⁻⁹ and 1.7 × 10⁻⁹ mol/L after incubation for 24 and 48 hours, respectively (Fig. 4D). In contrast, no detectable increase in the cytotoxicity of huC242-DM1-containing medium occurred when it was conditioned by the presence of Namalwa cells (Fig. 4E). The enhanced cytotoxicity of conjugate-containing, COLO 205-conditioned medium parallels the increase in the concentration of the free maytansinoid in Fig. 4A. These data suggest a mechanism for the bystander cytotoxicity of huC242-DM1 in which the conjugate binds to antigen-expressing cells, followed by cell-mediated processing of the conjugate and progressive release of a cytotoxic maytansinoid species capable of killing neighboring cells.

Incubation of huC242-SMCC-DM1 with COLO 205 cells led to only a modest 2-fold increase in nonspecific cytotoxicity of the culture medium (Fig. 4F). Because both the disulfide-linked conjugate and the thioether linked conjugate are processed by COLO 205 cells and release a comparable amount of free maytansinoid species into the medium (Fig. 4A and C), distinct metabolites are likely formed from huC242-DM1 and huC242-SMCC-DM1. We have recently identified the chemical structures of the metabolites and confirmed that some of the products of the huC242-DM1 processing are 100- to 10,000-fold more potent than the metabolite of huC242-SMCC-SM1 (17).

Bystander effect of huC242-DM1 in human xenograft tumor models. By analogy with the in vitro mixed cell culture systems, we developed mixed xenograft tumor models consisting of CanAg-expressing COLO 205 cells and CanAg-negative Namalwa cells. A mixture of COLO 205 and Namalwa cells was injected s.c. into SCID mice (a total of 1.2 × 10⁷ cells per mouse). Every animal (a total of 23 mice) developed a measurable tumor and, 9 days after the cell implanting, the mean tumor volume reached ∼100 mm³. Two 9-day-old tumors were removed from sacrificed animals and analyzed for the presence of COLO 205 cells (CanAg+/CD38−) and Namalwa cells (CD38+/CanAg−) by immunohistochemical staining of the tissues with the murine C242 antibody and a murine anti-CD38 antibody, respectively. The marker for COLO 205 cells (Fig. 5A,-a) and that for Namalwa cells (Fig. 5A,-b) stained areas of approximately equal size, confirming the mixed nature of the tumors. The vast majority of both COLO 205 cells and Namalwa cells appeared viable, with little necrosis apparent. Groups of mice bearing such mixed tumors of ∼100 mm³ in size were then treated with five i.v. injections, given on 5 consecutive days (qd × 5), of PBS (control group), huC242-SMCC-DM1, or huC242-DM1, at daily doses of the conjugates containing 150 μg/kg of linked DM1. One day after the last injection, two mice from each group were sacrificed and their tumors were analyzed by immunohistochemistry as described above. In the PBS-treated control group, the tumor sections looked similar to those obtained from untreated tumors (Fig. 5A, compare c with a, and d with b), with similarly-sized areas of COLO 205 and Namalwa tissues, both displaying only a minimal degree of necrosis. Sections from tumors treated with huC242-SMCC-DM1 showed strongly necrotic COLO 205 tissue (50-60% necrotic, Fig. 5A,-e) and healthy Namalwa tissue (Fig. 5A,-f). Finally, sections from tumors treated with huC242-DM1 showed a high degree of necrosis (70-80%) in both COLO 205 and Namalwa tissues (Fig. 5A , g and h). These in vivo data mirror our in vitro results using mixed cell populations. The activity of the conjugate with the nondisulfide linker was restricted to the cells that express the target antigen whereas the conjugate with the disulfide-containing linker killed both the target and the nontarget tumor cells in the same tumor.

These immunohistochemistry results were in agreement with the effects of huC242-DM1 and huC242-SMCC-DM1 on the tumor growth. Treatment with either conjugate was equally effective against CanAg-positive COLO 205 tumors (Fig. 5B,-a): tumors in four of five mice completely regressed and did not relapse until day 90, when the experiment was terminated, and in each group, one mouse relapsed leading to a 28-day delay in the tumor growth. Neither conjugate was active on nontarget Namalwa tumors: huC242-SMCC-DM1 treatment did not delay tumor growth whereas huC242-DM1 effected only a modest 7-day delay in tumor growth (Fig. 5B,-b), confirming that the antitumor activities are CanAg selective. The two conjugates differed markedly in their antitumor activity against mixed COLO 205-Namalwa tumors. HuC242-SMCC-DM1 did not delay the progression of mixed tumors whereas huC242-DM1 caused complete regressions of tumors within 2 weeks (Fig. 5B -c) with four of five mice remaining tumor-free until the end of the study (150 days, i.e., ∼34 tumor doubling times). Immunohistochemical analysis of the only relapsed tumor in this group revealed that the tumor consisted entirely of Namalwa cells (data not shown). In conclusion, only the conjugate with the disulfide linker showed an ability to kill bystander cells in vivo.

We further assessed the in vivo bystander effect of huC242-DM1 and huC242-SMCC-DM1 in the HT-29 human colon cancer xenograft model in SCID mice. HT-29 tumors are examples of special cases of “mixed” tumors that are generated spontaneously from a single cell line yet are heterogeneous with respect to antigen expression. HT-29 cells express the huC242 antibody target, CanAg, only on a minority of cells (20-40%) in vitro (Fig. 2) or in xenograft tumors (11). Mice bearing established (∼130 mm³) HT-29 tumors were treated with either huC242-DM1 or huC242-SMCC-DM1 at various doses (25, 75, or 100 μg/kg of linked DM1, qd × 5), and tumor growth was monitored. Whereas the disulfide-linked conjugate huC242-DM1 induced marked tumor growth delays at each dose (5-25 days depending on the dose; Fig. 6A), the huC242-SMCC-DM1 conjugate produced only marginal antitumor effects (2.5-3.5 days of growth delay; Fig. 6B). Taken together, these data provide evidence that huC242-SMCC-DM1 conjugate containing a nondisulfide linker is efficacious only against tumors in which all proliferating cells express the target antigen. In contrast, the bystander effect associated with the disulfide-containing conjugate renders this conjugate also effective against tumors in which only a fraction of cells expresses the target antigen.

Antibody-drug conjugates show targeted antigen-dependent killing of cells in vitro and antigen-selective antitumor activity in cancer models in vivo (2). In this study, we report that two antibody-drug conjugates containing different drugs, linked via a disulfide bond, can kill antigen–negative cells in mixed antigen-positive and antigen-negative cell populations in vitro. Furthermore, studies in vivo with mixed s.c. xenograft tumors in SCID mice produced similar results. These results support a mechanism of cytotoxicity for these antibody-drug conjugates, which includes binding of the conjugate to target cells, cleavage of the conjugate disulfide bond (presumably through a disulfide exchange reaction), and release of a drug capable of killing nearby nontarget cells. Similar cytotoxicity in a vicinity of target cells was previously described for radioimmunoconjugates, immunoliposomes, ADEPT, and GDEPT, and referred as bystander effect (8). This report is a demonstration that antibody-drug conjugates can be engineered either to produce a bystander effect (disulfide-linked conjugates) or exert a precise killing of antigen-presenting cells without damaging proximal antigen-negative tissues (thioether-linked conjugates).

Immunohistochemistry studies with antibodies binding to tumor-associated antigens revealed that many solid tumors express the target antigen in a heterogeneous fashion and are populated with both antigen-positive and antigen negative cells (18, 19). For example, tumors in about half of patients in the cantuzumab mertansine phase I study expressed CanAg either heterogeneously or focally (6). This heterogeneity of antigen expression was considered a barrier to the effective treatment of such tumors with antibody-drug conjugates because of the presumed target-restricted cell killing by immunoconjugates (20). The in vitro and in vivo bystander cytotoxicity associated with disulfide linker–containing conjugates establishes a rationale for immunoconjugate treatment of tumors even if they exhibit heterogeneous antigen expression.

The bystander effect adds a degree of nonselective killing activity to the target cell–restricted cytotoxicity of antibody-drug conjugates. Potentially, this could be a drawback if normal cells in tissues surrounding the cancer tissue are affected. However, this collateral toxicity might be well tolerated if it is limited to a small number of cells in intimate proximity to the tumor tissues. Furthermore, normal tissue barriers might prevent the cytotoxic effect and the potential toxicities contributed by the bystander effect to normal tissues might be mitigated by the inherent resistance of nondividing cells to the antimitotic action of maytansinoids (21, 22). As a potential advantage, the bystander cytotoxicity may damage tissues intricately involved in supporting the tumor growth, such as endothelial cells and pericytes of tumor neovasculature, or tumor stromal cells, resulting in enhanced antitumor activity of the conjugate against tumors expressing the antigen either homogeneously or heterogeneously.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank Dr. Dorfman, Brigham and Women's Hospital (Boston, MA), for his expert evaluation of immunohistochemistry slides; Sharon Wilhelm, Kate Lai, Robert Zhao, and Hans Erickson for preparation of the antibody-drug conjugates; Kathleen Whiteman for her help with tissue culture experiments; and Hans Erickson, John Lambert, Robert Lutz, Rita Steeves, and Wayne Widdison for numerous enlightening discussions during the project.