Complete remissions of the androgen-dependent Shionogi mouse mammary carcinoma are observed after androgen withdrawal but invariably the disease recurs and is refractory to further hormonal manipulations. To determine the proportions of androgen-dependent (AD) and -independent (AI) tumorigenic stem cells in parent and recurrent tumors an in vivo limiting dilution assay was developed. There was a marked enrichment of stem cells in the recurrent tumors (1/200 tumor cells) relative to the parent tumors (1/4000 tumor cells) when assayed in male hosts. By assaying tumor takes in female mice, the proportion of AI stem cells was found to be 1/370,000 tumor cells in the parent vs 1/800 tumor cells in the recurrent carcinoma; a 500-fold increase in AI stem cells resulting from androgen-withdrawal. Unexpectedly, no enrichment of AI stem cells was evident in regressing parent tumors; rather, the proportion of such cells was very small (1/2,200,000 tumor cells). This finding implies that the AI cells which survive androgen withdrawal may result from the ability of small number of initially AD stem cells to adapt to an altered hormonal environment. This adaptive process was further defined in terms of the disappearance of androgen receptors from the nucleus and the expression of androgen-repressed genes including the proto-oncogenes, c-fos and c-myc, and the cell death gene, TRPM-2; all of which are constitutively active in recurrent AI tumor cells. Overall, our results indicate: (1) the tumor mass consists mainly of differentiated cells; (2) stem cells initially are AD but at most the killing effect of androgen-withdrawal will be limited to 2-3 logarithms before compensatory adaptive mechanisms supervene; and (3) progression of stem cells to an AI state, in which they are resistant to the killing effects of cell death genes, might be prevented by the inhibition of androgen-repressed adaptive mechanisms which come into play when androgens are withdrawn.