In evolution strategies aimed at isolating molecules with new functions, screening for the desired phenotype is generally performed in vitro or in bacteria. When the final goal of the strategy is the modification of the human cell, the mutants selected with these preliminary screenings may fail to confer the desired phenotype, due to the complex networks that regulate gene expression in higher eukaryotes. We developed a system where, by mimicking successive infection cycles with HIV-1 derived vectors containing the gene target of the evolution in their genome, libraries of gene mutants are generated in the human cell, where they can be directly screened. As a proof of concept we created a library of mutants of the human deoxycytidine kinase (dCK) gene, involved in the activation of nucleoside analogues used in cancer treatment, with the aim of isolating a variant sensitizing cancer cells to the chemotherapy compound Gemcitabine, to be used in gene therapy for anti-cancer approaches or as a poorly immunogenic negative selection marker for cell transplantation approaches. We describe the isolation of a dCK mutant, G12, inducing a 300-fold sensitization to Gemcitabine in cells originally resistant to the prodrug (Messa 10K), an effect 60 times stronger than the one induced by the wt enzyme. The phenotype is observed in different tumour cell lines irrespective of the insertion site of the transgene and is due to a change in specificity of the mutated kinase in favour of the nucleoside analogue. The mutations characterizing G12 are distant from the active site of the enzyme and are unpredictable on a rational basis, fully validating the pragmatic approach followed. Besides the potential interest of the G12 dCK variant for therapeutic purposes, the methodology developed is of interest for a large panel of applications in biotechnology and basic research.