Recent studies have shown that, even with a minimal content of carcinoembryonic antigen (CEA), normal human colonic epithelial cells express substantial amounts of CEA mRNA and colonic mucosal fragments cultured in vitro produce CEA quite actively, indicating that CEA should no longer be considered to be of an oncofetal nature. To understand the basis of the usefulness of CEA as a tumor marker, we analyzed the release of CEA, a glycosyl-phosphatidylinositol (GPI)-anchored protein, from colonic epithelial cells, by culturing isolated colonic crypts in collagen gel. The crypts appeared to preserve their morphological and biochemical integrity in the gel for at least 16 hr, and released CEA spontaneously. Three forms of CEA--spontaneously released CEA, CEA liberated with phosphatidylinositol-specific phospholipase C (PI-PLC) and CEA in cell lysates--were indistinguishable on SDS-PAGE. This is in contrast to recombinant CEA spontaneously released from CHO transfectants, which showed a smaller molecular mass than that of PI-PLC-cleaved recombinant CEA. By phase separation using Triton X-114, CEA in the cell lysates of crypts was separated mostly into the detergent phase, while the spontaneously released and the PI-PLC-cleaved CEA were separated into the aqueous phase. When the cells were metabolically labeled with the precursors of the GPI-anchor, 3H-ethanolamine but not 3H-palmitic acid was found in the spontaneously released CEA. These findings suggest that, in contrast to the proteolysis-like release of the recombinant CEA from CHO cells, CEA in normal colonic epithelial cells is released by a non-proteolytic cleavage, which probably occurs through the action of some endogenous phospholipase.