We have analyzed transgene (lacZ) expression from a first-generation adenovirus (Ad) vector in comparison to helper-dependent (hd) Ads deleted for various portions of the viral coding sequences and generated by using the Cre/loxP helper-dependent system (R. J. Parks et al., Proc. Natl. Acad. Sci. USA 93:13565-13570, 1996). An hd vector deleted for approximately 70% of the Ad genome (AdRP1001) provided levels and durations of transgene expression similar to those of a control first generation Ad vector containing an identical expression cassette. Deletion of all Ad sequences from the hdAd and replacement with a approximately 22-kb fragment of lambda DNA resulted in a decrease in the level and duration of lacZ expression which could not be reversed by the inclusion of a matrix attachment region. However, substitution of the lambda stuffer in the fully deleted hdAd with sequences from the human hypoxanthine-guanine phosphoribosyltransferase gene resulted in significantly improved transgene expression. In vitro assays for cytotoxic T lymphocytes (CTL) directed against putative peptides encoded by the vector backbone showed that, although CTL were generated against the vector containing the lambda DNA, no such CTL were generated against the vector containing the hypoxanthine-guanine phosphoribosyltransferase (HPRT) sequences. Surprisingly, the rate of loss of the HPRT- and lambda-containing vectors from mouse liver was similar, despite the differences in expression kinetics, indicating that the lambda stuffer-directed CTL were inefficient at eliminating the transduced cells. Thus, the nature of the DNA backbone of hdAds can have important effects on the functioning of the vector. Since most fully deleted vectors require "stuffer" DNA as part of the vector backbone to maintain optimum vector size, these observations must be taken into account in the design of hdAd vectors.