Comparative mutagenesis of gamma- or X-ray-induced tandem DNA lesions G[8,5-Me]T and T[5-Me,8]G intrastrand cross-links was investigated in simian (COS-7) and human embryonic (293T) kidney cells. For G[8,5-Me]T in 293T cells, 5.8% of progeny contained targeted base substitutions, whereas 10.0% showed semitargeted single-base substitutions. Of the targeted mutations, the G --> T mutation occurred with the highest frequency. The semitargeted mutations were detected up to two bases 5' and three bases 3' to the cross-link. The most prevalent semitargeted mutation was a C --> T transition immediately 5' to the G[8,5-Me]T cross-link. Frameshifts (4.6%) (mostly small deletions) and multiple-base substitutions (2.7%) also were detected. For the T[5-Me,8]G cross-link, a similar pattern of mutations was noted, but the mutational frequency was significantly higher than that of G[8,5-Me]T. Both targeted and semitargeted mutations occurred with a frequency of approximately 16%, and both included a dominant G --> T transversion. As in 293T cells, more than twice as many targeted mutations in COS cells occurred in T[5-Me,8]G (11.4%) as in G[8,5-Me]T (4.7%). Also, the level of semitargeted single-base substitutions 5' to the lesion was increased and 3' to the lesion decreased in T[5-Me,8]G relative to G[8,5-Me]T in COS cells. It appeared that the majority of the base substitutions at or near the cross-links resulted from incorporation of dAMP opposite the template base, in agreement with the so-called "A-rule". To determine if human polymerase eta (hpol eta) might be involved in the mutagenic bypass, an in vitro bypass study of G[8,5-Me]T in the same sequence was carried out, which showed that hpol eta can bypass the cross-link incorporating the correct dNMP opposite each cross-linked base. For G[8,5-Me]T, nucleotide incorporation by hpol eta was significantly different from that by yeast pol eta in that the latter was more error-prone opposite the cross-linked Gua. The incorporation of the correct nucleotide, dAMP, by hpol eta opposite cross-linked T was 3-5-fold more efficient than that of a wrong nucleotide, whereas incorporation of dCMP opposite the cross-linked G was 10-fold more efficient than that with dTMP. Therefore, the nucleotide incorporation pattern by hpol eta was not consistent with the observed cellular mutations. Nevertheless, at and near the lesion, hpol eta was more error-prone compared to a control template. The in vitro data suggest that translesion synthesis by another Y-family DNA polymerase and/or flawed participation of an accessory protein is a more likely scenario in the mutagenesis of these lesions in mammalian cells. However, hpol eta may play a role in correct bypass of the cross-links.