The efficacy of cancer gene therapy depends critically on "bystander effects" by which genetic modification of tumor cells results in killing of unmodified cells in the local microenvironment. In gene-dependent enzyme-prodrug therapy, expression of a prodrug-activating suicide gene is used to generate a cytotoxic metabolite that diffuses to nontransduced cells. The objective of this study was to develop a physiologically relevant tissue culture model for quantifying bystander effects and to validate the model using as an example the activation of dinitrobenzamide prodrugs (e.g., CB 1954) by Escherichia coli aerobic nitroreductase (NTR). Bystander effects were measured in three-dimensional multilayer cocultures of NTR+ and NTR- cells by determining clonogenic survival curves for both cell types using V79, Skov3, or WiDr as parental cell lines. Bystander killing by CB 1954 was much more efficient in multilayers than monolayers at equivalent cell:medium ratios, whereas the chloromustard analogue of CB 1954 showed even greater efficiency. For a series of dinitrobenzamides, bystander killing in multilayers showed a positive correlation with prodrug lipophilicity and also correlated with the bystander effect in mixed tumor xenografts grown from the same NTR+ and NTR- WiDr cell lines (r(2) = 0.84; P < 0.001). The multilayer model identified a bromomustard prodrug (SN 24927) with superior therapeutic activity to CB 1954 that provided curative activity against WiDr tumors comprising 1:1 mixtures of NTR+ and NTR- cells. This study demonstrates the utility of the multilayer tissue culture model for quantifying and optimizing bystander effects in tumors and identifies a new lead prodrug for NTR gene-dependent enzyme-prodrug therapy.