Confirmation and prevention of targeted toxicity by a recombinant fusion toxin

To target malignant cells by exploiting their display of limited numbers of tumor associated antigens, ligands for these antigens have been connected to protein toxins capable of killing a cell with a single molecule. Toxins that show this high potency include ricin, diphtheria toxin,and Pseudomonas exotoxin (1–3). A variety of antibodies and growth factors have been chemically conjugated or genetically fused to mutant forms of these toxins, and found to produce major responses in clinical trials (4–14). One such agent, denileukin diftitox, contains human interleukin-2 fused with amino acids 1 to 388 of diphtheria toxin and is approved for the treatment of a subset of patients with cutaneous T-cell lymphoma (14). Another agent under clinical development, BL22, contains an Fv fragment of an anti-CD22 monoclonal antibody fused to truncated Pseudomonas exotoxin and produces complete remissions in a high percentage of patients with chemoresistant hairy cell leukemia (7).

Early in the development of protein toxins, it was feared that these agents would be excessively toxic due to nonspecific cross-reaction with vital normal tissues. For example, even after replacing the binding domain of Pseudomonas exotoxin with ligands for human but not murine antigens, dose-limiting hepatotoxicity is commonly observed in mice (15). Further, nearly all toxin-containing agents are reported to cause hypoalbuminemia after i.v. administration in humans, possibly due to nonspecific uptake into endothelial cells.

On the contrary, results from several clinical trials strongly suggest that targeting via ligand-antigen binding can cause dose-limiting toxicity. For example, the immunotoxin LMB-1, which caused dose-limiting vascular leak syndrome in solid tumor patients, binds to Le^ϒ antigen on endothelial cells and such binding can be competed by free antibody (16). The anti-erbB2 recombinant immunotoxin erb38 showed dose-limiting hepatotoxicity at (very?) high dose levels, attributed to low levels of erbB2 expression in the liver (17). Several immunotoxins produced dose-limiting neurotoxicity and were found to bind to antigens on neural tissues (18, 19).

In the phase I clinical trial of human granulocyte-macrophage colony-stimulating factor (GM-CSF) fused to amino acids 1 to 388 of diphtheria toxin (DT₃₈₈GMCSF; ref. 20), high fevers and transaminase elevations at low-dose levels were found to be dose limiting (11). It was hypothesized that monocytes and macrophages, which contain significant levels of GM-CSF receptors, were being targeted, resulting in liberation of cytokines. Hepatic Kupffer cells, known to display GM-CSF receptors (21, 22), were thought to liberate cytokines after binding DT₃₈₈GMCSF and thereby damage neighboring hepatocytes.

To test this hypothesis, Westcott et al. (23) present an elegant rodent model of this human toxicity. Because human GM-CSF does not react with the rodent GM-CSF receptor, a construct containing murine GM-CSF and diphtheria toxin amino acids 1 to 390 (DT_390mGMCSF) was used (24, 25). They found that rats had significantly higher transaminase elevations after treatment with DT₃₉₀mGMCSF than after treatment with DT₃₈₈GMCSF. Liver histology in DT₃₉₀mGMCSF-treated rats showed hepatocyte swelling but not death. Staining liver sections for caspase-3 showed sinusoidal cells but not hepatocytes to be the targets of attack by DT₃₉₀mGMCSF. By staining with monoclonal antibodies with varying specificities for hepatic Kupffer cells, the authors showed depletion of such cells in livers of DT₃₉₀mGMCSF-treated rats. Culture of primary hepatocytes showed potent killing of Kupffer cell–enriched nonparenchymal cells but not parenchymal hepatocytes. Immunofluoresence studies ruled out an effect on sinusoidal endothelial cells.

The Frankel group developed this model not only to investigate the clinical toxicity but also to test several novel recombinant toxins, which might retain the efficacy but not the toxicity of DT₃₈₈GMCSF. The first of these was DT₃₈₈IL3, which binds to the interleukin-3 receptor that is abundant on myeloid leukemia cells but not on macrophages (26–28). The second is DTU2GMCSF, which contains a urokinase-type plasminogen activator cleavage site instead of the normal furin cleavage site of diphtheria toxin. Cleavage at the normal furin site in the disulfide loop between amino acids Arg¹⁹³ and Ser¹⁹⁴ is required for the enzymatic domain of the toxin to enter the cytosol and cause cytotoxicity (29). Because myeloid leukemia cells but not monocytes or macrophages contain high levels of urokinase-type plasminogen activator, DTU2GMCSF offers intracellular specificity for the malignant target cells. In the rat model presented in this issue, the authors showed that, unlike DT₃₉₀mGMCSF, murine forms of both DT₃₈₈IL3 and DTU2GMCSF did not cause transaminase elevations in rats and did not deplete hepatic Kupffer cells in hepatic sections. Moreover, the ex vivo cytotoxicity of DTU2GMCSF toward Kupffer cell–enriched nonparenchymal cells was three orders of magnitude lower than that of DT₃₉₀mGMCSF. These data are encouraging, although conclusions regarding DTU2GMCSF must await missing data on its efficacy toward malignant target cells.

The recombinant immunotoxin LMB-2, containing an anti-CD25 Fv fused to truncated Pseudomonas exotoxin, is similar to DT₃₈₈GMCSF in that, in phase I clinical testing, it caused fever and transaminase elevations (8, 9). LMB-2 induced major responses in several different hematologic malignancies and it was not clear whether these toxicities were mediated in part by death of malignant cells. Lethal doses of LMB-2 injected into mice cause hepatic necrosis and LMB-2 was found to accumulate into Kupffer cells, suggesting an indirect mechanism for the hepatic necrosis (30). Evidence that LMB-2 could induce tumor necrosis factor (TNF)-α production in peritoneal macrophages, together with demonstration of high intrahepatic TNF-α levels in LMB-2-treated mice, pointed to TNF-α as a mediator of toxin-induced hepatotoxicity. Unlike DT₃₈₈GMCSF, LMB-2 in phase I testing did not cause life-threatening hepatotoxicity but rather reversible transaminase elevations unassociated with hepatic insufficiency (8). Thus, the authors have applied protein engineering and careful in vivo studies to approach a toxicity that is serious in both animals and patients. Although the severity of the hepatotoxicity problem is especially significant with DT₃₈₈GMCSF, it is hoped and anticipated that their efforts will facilitate clinical development of protein toxins from both diphtheria and other organisms.