The development of malignancy has been associated with both the activation o f oncogenes and the inactivation of tumor suppressor genes. Whereas recent data implicate tumor suppressor genes as cell-cycle checkpoints, the nature and timing of tumor suppressor gene inactivation during multistage carcinogenesis is still largely uncharacterized. To address this issue, we used a syngeneic mouse epidermal model system. By creating somatic-cell hybrids between nontumorigenic x benign (291 x 291.09RAT), nontumorigenic x malignant (291 x 291.05RAT and 291 x 291.03RAT). benign x malignant (291.09RAT x 291.03RAT) and malignant x malignant (291.03RAT x 291.05RAT) clones, multiple tumor suppressor activities were detected. Most importantly, we demonstrated the first example of the complete suppression of benign papillomas in vivo, thus implicating tumor suppressor gene activity loss as an early event in skin carcinogenesis. In addition, the carcinoma phenotype was suppressed in vivo by nontumorigenic, benign, and heterologous malignant keratinocytes. The so- matic-cell hybrids expressed the differentiation-specific keratins, K1 and K10, in response to high extracellular calcium concentrations (1.4 mM) in vitro. All of the hybrids had fewer local metastases than did the parental lines, and when tumor formation was not suppressed, the resulting tumors were highly differentiated. Polymerase chain reaction analysis of the neomycin-resistance gene at nontumorigenic injection sites indicated an absence o f injected hybrids, and subsequent analyses failed to detect nontumorigenic 291 cells 1 wk after transplantation. These data demonstrate that distinct tumor suppressor gene activities are lost at discrete stages during multistage carcinogenesis and are consistent with the hypothesis that tumor suppression can occur through induction of terminal differentiation.