Although several abnormalities in gap junction (GJ) structure and/or function have been described in neoplasms, the molecular mechanisms responsible for many of the alterations remain unknown. The identification of a family of GJ proteins, termed connexins, prompted this study of connexin32 (Cx32), connexin26 (Cx26) and connexin43 (Cx43) expression during rat hepatocarcinogenesis. Using antibody, cDNA and cRNA probes, we investigated connexin mRNA and protein expression in preneoplastic and neoplastic rat livers. In normal liver, Cx32 is expressed in hepatocytes throughout the hepatic acinus, Cx26 is restricted to periportal hepatocytes, and Cx43 is expressed by mesothelial cells forming Glisson's capsule. Most preneoplastic altered hepatic foci generated by diethylnitrosamine (DEN) initiation and either phenobarbital (PB) or 2,3,7,8-dichlorodibenzo-p-dioxin (TCDD) promotion exhibited decreased Cx32 or increased Cx26 staining. Foci from either protocol failed to display Cx43 immunoreactivity. In the majority of PB-promoted foci, Cx32 immunoreactivity decreased independently of changes in mRNA abundance. Continuous thymidine labeling, following cessation of PB promotion, showed that downregulation of Cx32 staining is reversible in foci that are promoter-dependent for growth, but irreversible in lesions that are promoter-independent for growth. Hepatic neoplasms from rats initiated with DEN and promoted with PB or TCDD also displayed modified connexin expression. While all 24 neoplasms studied were deficient in normal punctate Cx32 and Cx26 staining, altered cellular localization of these proteins was apparent in some tumors. Immunoblotting of crude tissue extracts revealed that neoplasms with disordered Cx32 staining showed immunoreactive bands with altered electrophoretic mobility. These observations show that hepatomas may downregulate Cx32 expression through changes in the primary structure of Cx32 or by post-translational modifications. Northern blotting of total tumor mRNAs failed to demonstrate consistent changes in the abundance of Cx32, Cx26 or Cx43 transcripts. Some tumors expressed steady-state transcripts without observable immunoreactivity, indicating that some hepatomas downregulate connexin immunoreactivity independently of mRNA abundance. Increased levels of Cx43 mRNA and protein were found in several neoplasms, but immunostaining was always localized to nonparenchymal cells. Areas of bile duct proliferation and cholangiomas displayed Cx43 staining, whereas, cholangiocarcinomas were deficient in immunoreactivity. These findings show that alterations in the expression of connexins, by either downregulation or differential induction, represent common modifications during hepatocarcinogenesis. Although our results imply that connexins represent useful markers for the boundary between tumor promotion and progression, preneoplastic and neoplastic rat hepatocytes fail to use a common mechanism to modify connexin expression.