NK cells are part of the innate immune system, and are important players in the body’s first defence line against virus-infected and malignantly transformed cells. While T cells recognize neoplastic cells in an MHC-restricted fashion, NK cells do not require prior sensitization and education about the target. In leukemia and lymphoma patients undergoing allogeneic hematopoietic stem cell transplantation not only T cells but also NK cells have been found to mediate potent graft-versus-tumor effects. Hence, autologous or donor-derived NK cells hold great promise for cancer immunotherapy. Since the generation of highly purified NK cell products for clinical applications is labor-intensive and time consuming, established human NK cell lines such as NK-92 are also being considered for clinical protocols. NK-92 cells display phenotypic and functional characteristics similar to activated primary NK cells. While NK-92 cells are highly cytotoxic towards malignant cells of hematologic origin, they do not affect healthy human tissues. NK-92 cells can be expanded under GMP-compliant conditions, and can therefore be provided in sufficient numbers with defined phenotypic characteristics for clinical applications. Safety of NK-92 cells for adoptive immunotherapy was already shown in two phase I/II clinical trials. In contrast to malignant cells of hematologic origin, most solid tumor cells are not sensitive to unmodified NK-92 cells. Hence, to overcome resistance mechanisms of tumor cells and to broaden the target spectrum of NK-92 cells, gene-modified variants have been generated which express chimeric antigen receptors (CARs) that specifically target tumor surface antigens. The expression of these CARs is sufficient to redirect their cytotoxic activity towards otherwise NK cell-resistant target cells. Extending these earlier approaches, in the framework of this work optimized CAR constructs that target the pancarcinoma antigen epithelial cell adhesion molecule (EpCAM) were derived and functionally characterized. In collaboration with Heike Daldrup-Link’s laboratory (University of California San Francisco, USA) non-invasive imaging modalities to analyze biodistribution and tumor homing properties of retargeted NK-92 cells were evaluated. To enhance the persistence of adoptively transferred NK-92 cells in vivo, means to overcome NK-92 cells’ dependence on exogenous IL-2 for survival and cytolytic activity were investigated. EpCAM is expressed on a variety of tumors of epithelial origin including ovarian, gastric, colorectal, pancreatic, breast, lung and endometrial cancers. In epithelial cells EpCAM is mainly expressed at basolateral membranes, and EpCAM is involved in calcium-independent homotypic cell-cell adhesions. In tumor cells high and de novo EpCAM expression is not only restricted to basolateral membranes but can also be found on apical membranes. Tumor cells retain EpCAM expression throughout tumorigenesis and metastasis formation. Due to its surface expression and immunogenicity EpCAM has been exploited as target for immunotherapy. In earlier work in our group a prototypic, first generation EpCAM-specific CAR construct (31.z) harboring a murine flexible hinge region and murine CD3 ζ as signaling domain was derived and functionally characterized in NK-92 cells. To reduce the immunogenicity for their potential clinical application, this CAR construct was humanized by exchanging the hinge region and the intracellular signaling domain with corresponding sequences of human origin. In T cells incorporation of additional co-stimulatory domains derived from CD28 and 4-1BB significantly enhanced persistence and anti-tumor effects of adoptively transferred cells. Based on these findings a modified, second generation CAR construct encompassing transmembrane and intracellular regions of CD28 in addition to CD3 ζ intracellular signaling domains was derived (31.28.z). Both CAR constructs were stably expressed in NK-92 cells, and furthermore, expression of both CAR variants promoted antigen-specific lysis of antigen-expressing prostate and breast cancer cell lines. In competition experiments the cytotoxic activity of NK-92/31.z and NK-92/31.28.z cells towards antigen-expressing tumor cells was significantly reduced in the presence of parental MOC31 monoclonal antibody, indicating that binding of the EpCAM-specific CAR to its antigen on tumor cells is necessary to trigger antigen-specific cytotoxicity. At high effector to target ratios NK-92/31.28.z cells displayed slightly higher cytotoxic activity towards EpCAM-expressing target cell lines than NK-92/31.z cells, suggesting that incorporation of co-stimulatory domains had beneficial effects on the cytotoxic activity. For clinical applications the development of non-invasive imaging methods is necessary to follow the biodistribution of adoptively transferred cells and guide the identification of responders and non-responders at an early time point. In collaboration with Heike Daldrup-Link’s laboratory the homing properties of EpCAM-specific NK-92 cells to prostate tumor xenografts in rodent models was analyzed (University of California San Francisco, USA). At that time NK-92 cells expressing the second generation EpCAM-specific CAR 31.28.z were not yet available, and thus homing experiments were performed with NK-92 cells expressing the first generation CAR 31.z. For magnetic resonance imaging studies parental and EpCAM-specific NK-92 cells were labeled with clinical applicable ferumoxide particles. Labeled, gene-modified NK-92 cells displayed reduced CAR expression and reduced cytotoxic activity towards EpCAM-expressing DU145 prostate cancer cells in vitro. Nevertheless, MRI revealed specific accumulation of ferumoxide labeled EpCAM-specific NK-92 cells in DU145 tumor xenografts in athymic rats. In tumor sections of treated animals the presence of EpCAM-specific NK-92 cells was verified by Prussian blue and CD57 staining of tumor sections. In another study homing of DiD-labeled EpCAM-specific NK-92 cells to DU145 tumor xenografts was shown by optical imaging. These findings imply that specific targeting of NK-92 cells is retained in vivo, and that non-invasive imaging strategies can be employed to analyze biodistribution of NK-92 cells. Enhanced persistence of adoptively transferred cytotoxic effector cells has a major impact on the effectiveness of immunotherapy. Primary cytotoxic effector cells as well as NK-92 cells require IL-2 for their proliferation and to gain full activity of their effector functions. To bypass the need of exogenously supplied cytokines, the expression of chimeric cytokine receptors (CCR) harboring IL-2R β and IL-2R γ chains instead of CD3 ζ as signaling domains might initiate cytokine-like signals upon contact with the respective antigen. These interactions might support growth and survival of NK-92 cells in the absence of exogenous IL-2. As a starting point, a codon-optimized ErbB2-specific CAR consisting of the scFv(FRP5) single chain antibody fragment, a human CD8 α hinge region and human CD3 ζ transmembrane and intracellular domains was used. Transmembrane and intracellular domains of IL-2R β and IL-2R γ chains were amplified from NK-92 cell-derived cDNA, and were used to exchange the CD3 ζ domain in the ErbB2-specific construct. In human primary tumors EpCAM and ErbB2 overexpression are frequently found, and often correlate with poor prognosis. Hence, co-expression of ErbB2-specific CCRs with an EpCAM-specific CAR may provide NK cells with antigen-specific killing via EpCAM recognition and with antigen-dependent growth via binding to ErbB2. However, attempts to activate CCRs in NK-92 cells via co-incubation with antigen-expressing cells or cross-linking of the CCRs with recombinant antigen did not result in cytokine-independent but antigen-dependent growth. Likewise, no triggering of signal transducer and activator of transcription 5 (STAT5) was observed, which is a hallmark of IL-2 mediated signal transduction. The interactions between CCRs and their antigen might not be strong enough to trigger cytokine-like signals supporting the growth of cells in the absence of exogenous cytokines, and furthermore, might not lead to a significant up-regulation of STAT5-mediated signal transduction. An alternative approach to circumvent the need of exogenous cytokines is ectopic expression of homeostatic cytokines IL-2 and IL-15 in lymphocytes. In T cells expression of these cytokines is sufficient to render cells independent from exogenously supplied cytokines. In this work a lentiviral expression vector encoding IL-15 (SIEW-IL15) was generated, and used for transduction of NK-92 cells. This resulted in ectopic expression of IL-15 and cellular proliferation in the absence of exogenously supplied cytokines. Even after prolonged culture without exogenous IL-2, NK-92/IL15 cells retained their cytotoxic activity towards NK-sensitive target cells. Although expression of IL-15 in HC11 and COS-7 cells using the same vector led to secretion of bioactive IL-15 into culture supernatants, neither secreted nor surface-bound IL-15 was detected in NK-92/IL15 cells, implying that IL-15 promotes survival of gene-modified cells in a strictly autocrine fashion. In addition, NK-92 cells that were freshly transduced with SIEW-IL15 could be efficiently enriched by cytokine withdrawal. NK-92/IL15 cells that were co-transduced with an EpCAM-specific CAR retained their ability to grow in the absence of exogenously supplied cytokines and their antigen-specific cytotoxic activity. Based on these results, a bicistronic vector construct was generated allowing the simultaneous expression of a CAR construct and IL-15 as selection marker. EpCAM-specific CAR constructs (31.28.z and 31.TM) were inserted into the bicistronic expression cassette. NK-92 cells were transduced with these bicistronic expression constructs and selected by cytokine withdrawal. After 14 to 21 days of culture in the absence of IL-2 transduced cells grew out from which CAR-expressing NK-92 cells with high and homogenous surface expression were further enriched by FACS sorting. NK-92/31.28.z.IL15 cells displayed high cytotoxic activity towards EpCAM-expressing breast cancer cell lines, while EpCAM-negative melanoma cells were not lysed. The results of this work demonstrate that the expression of first (31.z) and second (31.28.z) generation CARs in NK-92 cells is sufficient to induce antigen-specific cytotoxicity. Furthermore, a specific accumulation of NK-92/31.z cells but not unmodified NK-92 cells was detected in EpCAM-expressing prostate carcinoma xenografts in athymic rats, indicating that specific targeting of these cells is retained in vivo. Ectopic expression of IL-15 renders the cells independent from exogenous cytokines, while they retain their cytotoxic activity even after prolonged culture without IL-2. Furthermore, ectopic expression of IL-15 in NK-92 cells can be used for selective enrichment of gene-modified cells by cytokine withdrawal. Subsequently, bicistronic expression constructs that allow simultaneous expression of a CAR construct and IL-15 as selection marker were generated. Expression of these bicistronic expression vectors in NK-92 cells is feasible, and might facilitate enrichment of gene-modified cells for clinical applications.