Evidence of dynamic development of cytokeratin (CK) 18-positive disseminated tumor cells in bone marrow of curatively resected cancer patients has implicated a subclinical minimal residual disease as a biologically relevant component in solid cancer. However, differentiation between irrelevant shed cells and those cells potentially capable of causing later recurrence has not yet been made. In parallel, accumulating data show functional association of the urokinase plasminogen activator (uPA) system and the membranous uPA receptor (uPA-R) with the capacity of a tumor cell for invasion and metastasis. The present study was designed to find descriptive evidence in vivo concerning whether uPA-R could be one potential characteristic for metastatically relevant phenotypes of disseminated tumor cells. An immunocytochemical double staining for uPA-R and CK18 (immunogold/alkaline phosphatase anti-alkaline phosphatase) was performed on perioperative and follow-up bone marrow aspirations of 78 curatively resected gastric cancer patients, if positive tumor cell status had been shown previously with the single alkaline phosphatase anti-alkaline phosphatase method. Bone marrow cells (10(6)) were examined in each assay. Postoperative qualitative and quantitative development of uPA-R-expressing disseminated tumor cells was followed in relation to uPA-R-negative cells and correlated with later clinical relapse. Double staining could be performed perioperatively or in follow-up, or both, in 58 of 78 patients. Expression of uPA-R on perioperatively disseminated tumor cells significantly correlated with later quantitative increases of tumor cells (P = 0.0009). Overall median tumor cell numbers with uPA-R expression significantly increased during follow-up from a median value of 5.5 to 10.0 in 10(6) cells (P = 0.008), and the mean relative percentage of uPA-R-positive, compared with uPA-R-negative, disseminated tumor cells also increased, from 47.9% at surgery to 68.6% in follow-up (P < 0.001). This was mainly due to patients with later tumor relapse (increase from 63.9 to 80.7%, P = 0.001). Patients without relapse showed slight increases at lower percentage levels (5.7% at surgery, 7.4% in follow-up). Differences for relapsing patients were significant (surgery, P = 0.006; follow-up, P < 0.001). Our results suggest from an in vivo model that uPA-R may be one antigen that enables identification and follow-up observations of metastatically relevant phenotypes of disseminated tumor cells, differentiating their individual potential for causing relapse.