Although histone deacetylase inhibitors (HDACis) are emerging as a new class of anticancer agents, the mechanism of tumorselective killing by HDACi is not well understood. We used suppression of mortality by antisense rescue technique (SMART) to screen the key genes responsible for the tumor-selective killing by trichostatin A (TSA). Twenty-four genes were identified, the most significant of which was ubiquitin B (UbB). The expression of UbB was selectively upregulated by TSA in tumor cells, but not non-malignant cells. Further observation indicated that TSA induced a substantial dissipation of mitochondrial transmembrane potential, release of cytochrome c into the cytosol, and proteolytic cleavage of caspases-3/9 in HeLa cells, which was apparently mediated by ubiquitylation and the subsequent degradation of mitochondrial membrane proteins including BCL-2 and MCL-1. In contrast, knockdown of UbB expression inhibited the TSA-induced apoptotic cascade by abolishing TSA-induced ubiquitylation and the subsequent degradation of mitochondrial membrane proteins. Furthermore, apicidine, another HDACi, exhibited activity similar to that of TSA. Interestingly, TSA induced UbB-dependent proteasomal degradation of BCR-ABL fusion protein in K562 leukemic cells. Thus, our findings highlight the essential role of UbB and UbBdependent proteasomal protein degradation in HDACi-induced tumor selectivity. The mechanism provides a novel starting point for dissecting the molecular mechanism underlying the tumor selectivity of HDACi. 1-3 The clinical potential of HDACis has been well exemplified by the successful development of Vorinostat, which was recently approved by the US Food and Drug Administration for treatment of cutaneous T-cell lymphoma. 4 Despite the rapid clinical progress achieved, the mechanism of action of HDACis is not yet well understood. One of the central questions is how these agents selectively kill tumor cells while sparing normal cells. Identification of the critical intracellular targets responsible for the tumor selectivity of HDACis will further improve the design of optimized clinical protocols. More attractively, unraveling the potential 'death programs' selectively activated in tumor but not in normal cells will have broader implications for the understanding of tumorigenesis and the design of targeted therapies.Identification of the target genes essential for the selective induction of apoptosis in tumor cells has proven to be very difficult. It is well established that HDACis could affect up to 10% of all known genes at the transcriptional level. 5 In addition to HDACs, many non-histone proteins are also regulated by HDACis by influencing the molecular events of acetylation, protein-protein interactions and stability, and so on. Previously, a significant number of apoptotic and cell-cycle regulatory genes have been identified by different groups and proposed to be effectors responsible for the tumorselective action of HDACis. 6 In acute promyelocytic leukemia, however, preferential induction of tumor-n...