In vivo or in vitro selection for resistance to natural cytotoxic cell lysis selects for variants with increased tumorigenicity

Experiments were designed to test the hypothesis that transformed cells that are NC sensitive must escape NC activity if they are to grow as tumors in normal individuals. NC-resistant variants were selected either in vivo or in vitro from NC-sensitive cell lines that grow as tumors in immunodeficient mice but not in syngeneic normal mice. The tumorigenicity of cloned NC-resistant variants was compared with the parental cell lines and to cell lines that went through the selection procedure, but after cloning remained NC sensitive. Cloned NC-resistant cell lines derived from tumors that developed in x-irradiated nude mice after the injection of an NC-sensitive cell line are tumorigenic in normal mice, whereas cloned NC-sensitive cell lines derived from the same tumors are unable to grow as tumors in normal mice. Similarly, six of seven NC-resistant cloned cell lines independently isolated after in vitro selection for NC-resistance are tumorigenic in normal mice, whereas cloned NC-sensitive cell lines isolated from the same in vitro selected populations are not tumorigenic in normal mice. Thus, either the in vivo or in vitro selection of NC-resistant cells selects for cells tumorigenic in normal mice; these findings, along with our previous observations that selection for cells tumorigenic in normal mice selects for NC resistance, provide compelling evidence that escape from NC activity is required before some transformed cells can grow as tumors in normal mice.     

Sequences within the gag gene of mouse mammary tumor virus needed for mammary gland cell transformation

Previously, we identified a group of replication-competent exogenous mouse mammary tumor viruses that failed to induce mammary tumors in susceptible mice. Sequence comparison of tumorigenic and tumor-attenuated virus variants has linked the ability of virus to cause high-frequency mammary tumors to the gag gene. To determine the specific sequences within the gag gene that contribute to tumor induction, we constructed five distinct chimeric viruses that have various amino acid coding sequences of gag derived from a tumor-attenuated virus replaced by those of highly tumorigenic virus and tested these viruses for tumorigenic capacities in virus-susceptible C3H/HeN mice. Comparing the tumorigenic potentials of these viruses has allowed us to map the region responsible for tumorigenesis to a 253-amino-acid region within the CA and NC regions of the Gag protein. Unlike C3H/HeN mice, BALB/cJ mice develop tumors when infected with all viral variants, irrespective of the gag gene sequences. Using genetic crosses between BALB/cJ and C3H/HeN mice, we were able to determine that the mechanism that confers susceptibility to Gag-independent mammary tumors in BALB/cJ mice is inherited as a dominant trait and is controlled by a single gene, called mammary tumor susceptibility (mts), that maps to chromosome 14.     

CD8 T cell help for innate antitumor immunity

Innate immunity is considered to initiate adaptive antitumor responses. We demonstrate that monoclonal CD8 T lymphocytes reactive to tumor Ag P1A on P815 mastocytoma cells provide essential "help" to NK cells for rejection of P1A-deficient tumors. RAG-deficient mice have normal NK cells but do not reject either tumor. Reconstitution of these mice with P1A-specific T cells conferred resistance to both P1A-expressing and -deficient tumor cells provided they were present at the same site. Elimination of Ag-negative tumor variants required both activated T and NK cells. Gene expression profiling of NK cells infiltrating P1A-positive tumors in mice with specific CD8 T cells demonstrated an activated effector phenotype. However, CD8 T cell help to NK cells appeared ineffective for P1A-negative variants separated from the P1A-positive tumor. Local tumor Ag-specific T cell-NK cell collaboration results in the elimination of tumor cells whether they express or not the T cell tumor Ag epitope, thus containing the emergence of tumor escape variants before metastasis.     

The tumor-promoting effect of TNF-alpha involves the induction of secretory leukocyte protease inhibitor

According to the cancer immunoediting concept, inflammatory mediators play not only a critical role in promoting host protection against cancer but also contribute to cancer cell growth and survival. TNF-alpha is a critical factor in this network. However, the mechanisms underlying the tumor-promoting effect of TNF-alpha have not been fully elucidated yet. We previously reported that in vitro culture of Lewis lung carcinoma 3LL cells with TNF-alpha-producing macrophages resulted in enhanced resistance toward TNF-alpha-mediated lysis and increased malignancy of the 3LL cells. In this study, we analyzed the effects of endogenous TNF-alpha on TNF-alpha resistance and malignant behavior in vivo of low-malignant/TNF-alpha-sensitive 3LL-S cells and cancer cells derived from 3LL-S tumors that developed in wild-type or TNF-alpha(-/-) mice. Interestingly, 3LL-S cells acquired a malignant phenotype in vivo depending on the presence of host TNF-alpha, whereas acquisition of TNF-alpha resistance was TNF-alpha-independent. This result suggested that malignancy-promoting characteristics of 3LL-S cells other than TNF-alpha resistance are influenced in vivo by TNF-alpha. We previously identified the malignancy-promoting genes, secretory leukocyte protease inhibitor (SLPI) and S100A4, as being up-regulated in 3LL-S cells upon their s.c. growth in wild-type mice. In this study, we show that SLPI, but not S100A4, was induced in 3LL-S cells both in vitro and in vivo by TNF-alpha, and that silencing of in vivo induced 3LL-S SLPI expression using RNA interference abrogated in vivo progression but did not influence TNF-alpha resistance. These data indicate that SLPI induction may be one mechanism whereby TNF-alpha acts as an endogenous tumor promoter.     

Induction by mycoplasmas of tumor necrosis factor-alpha from macrophages

Several species of mycoplasmas including M. pneumoniae, the causative agent of human respiratory infection, were investigated for tumor necrosis factor-alpha (TNF-alpha) induction. The cytotoxic activity to Meth A cells of peritoneal macrophages purified from BALB/c mice was enhanced markedly when cultured with either viable or nonviable mycoplasmas. The supernatant of macrophage culture mixed with mycoplasmas, M. pneumoniae or A. laidlawii, showed a potent cytotoxic activity to TNF-alpha-sensitive but not to TNA-alpha-insensitive L cells. Addition of anti-TNA-alpha antiserum inhibited completely the cytotoxic activity of the supernatant, indicating that the cytotoxic activity is due mostly to TNF-alpha. These results strongly suggest that mycoplasmas possess an activity to induce TNF-alpha, which enhances the cytotoxic activity of macrophages and prevent infection with mycoplasmas in vivo.     

In vitro effects of 2-methoxyestradiol on MCF-12A and MCF-7 cell growth, morphology and mitotic spindle formation

The influence of 2-methoxyestradiol (2ME) was investigated on cell growth, morphology and spindle formation in a tumorigenic (MCF-7) and non-tumorigenic (MCF-12A) epithelial breast cell line. Inhibition of cell growth was more pronounced in the MCF-7 cells compared to the MCF-12A cells following 2ME treatment. Dose-dependent studies (10(-5)-10(-9) M) revealed that 10(-6) M 2ME inhibited cell growth by 44% in MCF-12A cells and by 84% in MCF-7 cells (p-value < 0.05). 2ME-treated MCF-7 cells showed abnormal metaphase cells, membrane blebbing, apoptotic cells and disrupted spindle formation. These observations were either absent or less prominent in MCF-12A cells. 2ME had no effect on the length of the cell cycle between S-phase and the time a mitotic peak was reached in either cell line but MCF-7 cells were blocked in mitosis with no statistically significant alterations in the phosphorylation status of Cdc25C. Nevertheless, Cdc2 activity was significantly increased in MCF-7 cells compared to MCF-12A cells (p-value < 0.05). The results indicate that 2ME disrupts mitotic spindle formation and enhances Cdc2 kinase activity, leading to persistence of the spindle checkpoint and thus prolonged metaphase arrest that may result in the induction of apoptosis. The tumorigenic MCF-7 cells were especially sensitive to 2ME treatment compared to the normal MCF-12A cells. Therefore, differential mechanism(s) of growth inhibition are evident between the normal and tumorigenic cells.     

Tumor growth in vivo selects for resistance to tumor necrosis factor

The relationship between in vivo tumor growth and resistance to TNF in WEHI-164 cells has been examined. When a highly TNF-sensitive clone of WEHI-164 was grown in vivo in syngeneic mice it became resistant to rTNF such that a 4 to 5 log higher concentration of TNF was required to produce tumor lysis in vitro. When compared with an in vitro selected TNF-resistant variant, the in vivo selected line was significantly more tumorigenic. The resistant phenotype of both the in vivo and in vitro selected variants was stable in culture and both selected lines were also resistant to lysis by syngeneic spleen cells with natural cytotoxic activity. The parental clone and the two variants were equally sensitive to lysis by allo-CTL and expressed similar levels of MHC class I Ag. Resistance to TNF in the two variants was not a function of de novo production of TNF measured as supernatant TNF activity or TNF mRNA expression. These studies are the first to demonstrate that in vivo tumor growth results in resistance to TNF and therefore may have direct relevance to the efficacy of TNF in the treatment of human neoplasms.     

Natural cytotoxic cells and tumor necrosis factor activate similar lytic mechanisms

The lytic activity of natural cytotoxic (NC) cells has several characteristics which clearly distinguish it from other cell-mediated lytic activities and from most soluble cytolytic factors. An exception is the lytic activity mediated by tumor necrosis factor (TNF). In this paper, we report a detailed comparison of NC and TNF lysis of target cells which are used as prototype NC targets or TNF targets, and show that the two cytolytic activities have very similar, if not identical, lytic mechanisms. We present data showing that target cells which are NC-sensitive are also TNF-sensitive and that target cells which are NC-resistant are also TNF-resistant. Moreover, cells selected either in vivo or in vitro for NC resistance are selected for TNF resistance, and cells selected for TNF resistance are selected for NC resistance. The analysis of the kinetics of 51Cr release mediated by NC cells or by TNF show that both activities affect similar kinetics, in that there is no cell lysis for several hours after targets and effectors first interact. However, NC and TNF lytic activities can be distinguished. By using the cell lines 10ME or B/C-N as targets, it can be shown that whereas NC-mediated lysis is dependent on protein synthesis, TNF-mediated lysis is not. We also show that targets which are resistant to NC-mediated lysis because they express a protein synthesis-dependent resistance mechanism also require protein synthesis to resist TNF-mediated lysis, suggesting that the same resistance mechanism protects cells against both NC cells and TNF. Together, these data strongly support the hypothesis that NC cells and TNF activate the same lytic mechanism within target cells and that TNF may mediate the lytic activity of NC effector cells.     

Stepwise immunologic selection of antigenic variants during tumor growth

Using tumor-specific effector cells as probes, we have studied the immunologic changes that occur in tumor cells during continuous growth in a host. As a model, we used a highly immunogenic ultraviolet light (UV)-induced tumor that is rejected regularly by normal mice but grows progressively when transplanted into UV-irradiated mice. The immunogenic tumor growing continuously in these partially immunocompromised mice gave rise to genetically stable progressor variants that were poorly immunogenic. A sequence of changes in susceptibility to activated macrophages and tumor-specific cytolytic T cells was observed when serial reisolates from the continuously growing tumors were analyzed. First, the tumor cells developed resistance to the cytocidal effects of activated macrophages. This was followed by the loss of one and then a second tumor-specific antigen defined by syngeneic cytolytic T cells. The phenotypes of the developing antigen loss variants and their sequence of appearance were the same in several independent experiments, and the process was apparently determined by a hierarchy of the host's immune response to multiple independent tumor-specific antigens expressed by a single malignant cell. Our ability to generate the predicted variants in vitro before they actually appear in vivo suggests a possible approach to preventing the outgrowth of such immunoselected variants from a tumor.     

Functional evidence for a second tumor suppressor gene on human chromosome 17.

The development and progression of human tumors often involves inactivation of tumor suppressor gene function. Observations that specific chromosome deletions correlate with distinct groups of cancer suggest that some types of tumors may share common defective tumor suppressor genes. In support of this notion, our initial studies showed that four human carcinoma cell lines belong to the same complementation group for tumorigenic potential. In this investigation, we have extended these studies to six human soft tissue sarcoma cell lines. Our data showed that hybrid cells between a peripheral neuroepithelioma (PNET) cell line and normal human fibroblasts or HeLa cells were nontumorigenic. However, hybrid cells between the PNET cell line and five other soft tissue sarcoma cell lines remained highly tumorigenic, suggesting at least one common genetic defect in the control of tumorigenic potential in these cells. To determine the location of this common tumor suppressor gene, we examined biochemical and molecular polymorphic markers in matched pairs of tumorigenic and nontumorigenic hybrid cells between the PNET cell line and a normal human fibroblast. The data showed that loss of the fibroblast-derived chromosome 17 correlated with the conversion from nontumorigenic to tumorigenic cells. Transfer of two different chromosome 17s containing a mutant form of the p53 gene into the PNET cell line caused suppression of tumorigenic potential, implying the presence of a second tumor suppressor gene on chromosome 17.     

Cell lines cultured at high density are resistant to lysis by tumor necrosis factor and natural cytotoxic cells

It has been suggested that natural cytotoxic (NC) cell activity and tumor necrosis factor (TNF), the molecular mediator of NC activity, are capable of protecting individuals against the progression of incipient tumors or could be useful in cancer therapy regimens. Much of this speculation arises as a result of in vitro studies, on a variety of tumor cells, demonstrating the cytolytic and cytostatic properties of NC and TNF activities. Here, evidence is presented showing that certain mouse fibroblast cell lines, generally considered sensitive to NC and TNF lysis, are quite resistant to these lytic activities when cultured at high cell density. Although a soluble factor that renders these same target cells resistant to NC and TNF lysis has been described, no such factor is involved in this high density-induced resistance. Rather, it appears that cell to cell contact of the targets is critical. Moreover, the induced resistance to NC and TNF lysis does not result from loss of either NC recognition determinants or TNF receptors by the target cells, but is the consequence of increased expression of a protein synthesis-dependent resistance mechanism. These observations raise the issue of the in vivo phenotype of cells characterized in vitro as sensitive to NC and TNF lysis. It is entirely possible that certain cells which are considered sensitive to NC and TNF activities are, in fact, resistant to these cytolytic activities when growing as tumors (i.e., at high cell density). Should this be the so, NC and TNF cytolytic activities may not function in vivo or may function only via some indirect means.     

Cellular tumorigenicity in nude mice. Role of susceptibility to natural killer cells

The athymic nude mouse is a useful animal model for assaying the neoplastic growth potential in vivo of animal cells transformed in vitro. Despite the demonstrated absence of thymus-dependent immunological functions, however, the nude mouse has now been shown to reject transplants of certain highly malignant heterologous tumors. In addition, a few transformed mammalian cell lines that exhibit all or most of the cellular phenotypes usually associated with malignancy fail to grow as tumors when injected into nude mice. In a continuing study to identify the in vitro phenotypes associated with tumor-forming ability in vivo, we investigated the role of cellular susceptibility to the naturally occurring, thymus-independent lymphocytes (natural killer or NK cells) in determining tumor induction by animal cells in nude mice. A representative collection of animal cells (ranging from normal human diploid cell strains to highly tumorigenic clonal cell lines, either transformed in vitro or derived from experimental tumors) was tested to see if the ability of cells to form tumors is consistently correlated with their susceptibility to NK cell-mediated lysis measured in vitro with splenic leukocytes from nude mice. If the physiological role of the NK cells in vivo were to recognize, and possibly to destroy, incipient tumor cells in situ, a direct association between cellular tumorigenicity and susceptibility to NK activity, might be expected. If, on the other hand, the formation of growing tumors by animal cells in nude mice depended on their ability to escape the cytolytic activity of NK cells, cellular tumorigenicity would be associated with cellular resistance to NK cells. Results obtained in this study failed to confirm either of these associations. Thus, cellular suscepbibility to NK cells, at least as determined by direct cytotoxicity assay in vitro, is not a useful predictive indicator of cellular tumorigenicity in nude mice.     

Malignant mouse melanoma cells do not form tumors when mixed with cells of a non-malignant subclone: relationships between plasminogen activator expression by the tumor cells and the host's immune response.

Mouse melanoma clones B559 and B78 are highly tumorigenic when injected into C57BL/6J mice. Tmor formation by these cells is suppressed when they are mixed with nonmalignant bromodeoxyuridine-grown clone C3471 before injection. C3471 cells suppress tumor formation only in immunocompetent hosts; mixtures of B559 and C3471 cells or C3471 cells alone form tumors in antithymocyte serum (ATS)-treated mice. Explants of C3471 tumors grown in ATS-treated mice form tumors in immunocompetent mice, most of which regress. Inability of C3471 or mixtures of C3471 with malignant cells to grow in normal mice, as contrasted with ability to grow in immunosuppressed mice, indicates that host response is involved. Both tumorigenic clones have high plasminogen activator activity, whereas nontumorigenic clone C3471 has none. Mixture of either tumorigenic clone with C3471 cells decreases plasminogen activator in vitro. C3471 tumor explants from ATS-treated mice initially express plasminogen activator, but lose the capacity to express this activity upon prolonged cultivation in vitro. Explants from B559 tumors retain plasminogen activator in long term culture. Close physical contact between C3471 and B559 cells appears essential both for inhibiton of plasminogen activator expression by B559 cells in vitro, and for tumor suppression in vivo. These findings suggest that production of plasminogen activators by tumor cells may play an important role in suppressing the host's immune response locally to an inoculum of syngeneic tumor cells.     

Karyotypic changes associated with spontaneous acquisition and loss of tumorigenicity in a human transformed bronchial epithelial cell line: Evidence for In vivo selection of transformed clones

Summary In this study, we describe the karyotypic changes associated with the spontaneous acquisition of tumorigenicity in an immortalized tumor bronchial cell line. Neoplastic transformation of the NL20 human bronchial epithelial cell line occurred after 3 yr in culture, and was associated with loss of chromosome 18 together with acquisition of multiple copies of 9q21.2→34. The nontumorigenic NL20 cell line had been established by transfection of human bronchial epithelial cells with the SV40 T antigen, and had retained a relatively stable karyotype after the first 32 passages in vitro. However, when cells from p184 were inoculated into nude mice, a transplantable tumor was obtained that was derived from a minor clone present in this otherwise stable line. Subsequent passaging of the NL20 cells in vitro did not yield further tumors, and the minor clone from which the tumorigenic NL20T cell line derived was no longer evident in NL20 cells by Passage 205. Furthermore, the original tumorigenic NL20T cells lost the neoplastic phenotype after 25 passages in vitro and reverted to the nontumorigenic karyotype observed at p189. In contrast to the loss of the tumorigenic phenotype and karyotype, which occurred with in vitro passaging of the original tumor, when the NL20T cells were passaged in other nude mice, they continued to give rise to tumors with sevenfold amplifications of 9q sequences and loss of chromosome 18, and cells from the secondary tumors (NL20T-A cells) have maintained a stable karyotype and remain tumorigenic even after 64 passages in vitro. A mixture of 10% tumorigenic NL20T-A and 90% nontumorigenic NL20 cells formed tumors in athymic nude mice when cultured in vitro on fibronectin, but not on plastic; cytogenetic analysis demonstrated that the tumors and cell cultures were composed of tumorigenic NL20T-A cells, whereas cytogenetic analysis of cells cultured on plastic were identical to the nontumorigenic NL20 cells. These data support the hypothesis that neoplastic transformation in our original cell line arose from in vivo selection of a small mutant clone, which had arisen in culture and was subsequently selected in vivo but was lost with in vitro culture.     

Tumor formation by SV40-transformed human cells in nude mice: the role of SV40 T antigens

Human cells transformed in vitro by SV40 rarely form tumors in nude mice. We examined whether these cells as a group are inherently nontumorigenic or whether they are potentially tumorigenic but rejected by the athymic host, possibly by nonspecific immune mechanisms. SV80 and NG8 are SV40-transformed human cell lines that express all of the transformed properties, including anchorage-independent growth, but do not form tumors in adult nude mice after injection of as many as 10(8) cells. Both the SV80 and NG8 cell lines have SV40-specific transplantation antigens which crossreact with those present on SV40-transformed (but tumorigenic) rodent cells. We found that SV80 cells, though not NG8 cells, induced progressively growing lethal tumors if the cells are injected repeatedly into neonatal nude mice. Somatic cell hybrids between SV80 or NG8 cells and a highly tumorigenic cell line derived from a human tumor continue to express the virus-induced antigens and fail to form tumors in adult nude mice. These results strongly suggest that at least for some SV40-transformed human cells, the failure to form tumors in nude mice may be due to their expression of virus-induced transplantation antigens rather than the absence of tumorigenic potential.     

A membrane form of TNF-alpha presented by exosomes delays T cell activation-induced cell death

In common with many other cell types, synovial fibroblasts produce exosomes. In this study, we show that the exosomes produced by synovial fibroblasts obtained from individuals with rheumatoid arthritis (RASF), but not exosomes produced by synovial fibroblasts obtained from individuals with osteoarthritis, contain a membrane bound form of TNF-alpha as demonstrated by colloidal gold immunostaining of TNF-alpha and confirmed by both Western blot and mass spectrometry. The RASF-derived exosomes, but not exosomes derived from fibroblasts obtained from individuals with osteoarthritis, are cytotoxic for the L929 cell, a TNF-alpha-sensitive cell line, and stimulate activation of NF-kappaB and induction of collagenase-1 in RASF. These effects are blocked by addition of soluble TNFR1 (sTNFbp), suggesting that a TNF-alpha-signaling pathway mediates these biological activities. sTNFbp also reduced the production of exosomes by RASF, suggesting the interruption of a positive amplification loop. Exosomes can transmit signals between cells, and RASF exosomes, effectively taken up by anti-CD3-activated T cells, activated AKT and NF-kappaB and rendered these activated T cells resistant to apoptosis. Neutralization of exosomal membrane TNF-alpha by sTNFbp partially reversed this resistance, suggesting that not only TNF-alpha but also additional exosomal proteins may contribute to the development of apoptosis resistance.     

Mice deficient in tumor necrosis factor-alpha are resistant to skin carcinogenesis.

Given the associations between chronic inflammation and epithelial cancer, we studied susceptibility to skin carcinogenesis in mice deficient for the pro-inflammatory cytokine TNF-alpha (refs. 5,6). TNF-alpha(-/-) mice were resistant to development of benign and malignant skin tumors, whether induced by initiation with DMBA and promotion with TPA or by repeated dosing with DMBA. TNF-alpha(-/-) mice developed 5-10% the number of tumors developed by wild-type mice during initiation/promotion and 25% of those in wild-type mice after repeated carcinogen treatment. TNF-alpha could influence tumor and stromal cells during tumor development. The early stages of TPA promotion are characterized by keratinocyte hyperproliferation and inflammation. These were diminished in TNF-alpha(-/-) mice. TNF-alpha was extensively induced in the epidermis, but not the dermis, in TPA-treated wild-type skin, indicating that dermal inflammation is controlled by keratinocyte TNF-alpha production. Deletion of a TNF-alpha inducible chemokine also conferred some resistance to skin tumor development. TNF-alpha has little influence on later stages of carcinogenesis, as tumors in wild-type and TNF-alpha(-/-) mice had similar rates of malignant progression. These data provide evidence that a pro-inflammatory cytokine is required for de novo carcinogenesis and that TNF-alpha is important to the early stages of tumor promotion. Strategies that neutralize TNF-alpha production may be useful in cancer treatment and prevention.     

Crucial role of TNF-alpha in CD8 T cell-mediated elimination of 3LL-A9 Lewis lung carcinoma cells in vivo

The role of perforin, IFN-gamma, and TNF-alpha in anti-tumor CD8 T cell immunity was examined in a new tumor model using a CD8 T cell epitope (GP33) derived from lymphocytic choriomeningitis virus as a tumor-associated Ag. In contrast with parental 3LL-A9 (A9) Lewis lung carcinoma cells that progressively grow in C57BL/6 mice, s.c. injection of GP33-transfected A9GP33 tumor cells induced a protective GP33-specific CD8 T cell response that led to complete tumor cell elimination. Tumor regression was dependent on perforin, IFN-gamma, or TNF-alpha, because A9GP33 tumors developed in mice deficient in one of these genes. A9GP33 tumors arising in perforin- and IFN-gamma-deficient mice represented GP33 Ag-loss variants, demonstrating that GP33-specific CD8 T cells from these mice were able to exert an Ag selection pressure. In contrast, tumor cells growing in TNF-alpha knock-out mice still expressed the tumor-associated GP33 peptide despite the presence of activated GP33-specific CD8 T cells. These findings provide evidence for a crucial role of TNF-alpha in A9 tumor cell elimination by CD8 T cells in vivo.     

Up-Regulation of Imp3 Confers In Vivo Tumorigenicity on Murine Osteosarcoma Cells

Osteosarcoma is a high-grade malignant bone tumor that manifests ingravescent clinical behavior. The intrinsic events that confer malignant properties on osteosarcoma cells have remained unclear, however. We previously established two lines of mouse osteosarcoma cells: AX cells, which are able to form tumors in syngeneic mice, and AXT cells, which were derived from such tumors and acquired an increased tumorigenic capacity during tumor development. We have now identified Igf2 mRNA-binding protein3 (Imp3) as a key molecule responsible for this increased tumorigenicity of AXT cells in vivo. Imp3 is consistently up-regulated in tumors formed by AX cells, and its expression in these cells was found to confer malignant properties such as anchorage-independent growth, loss of contact inhibition, and escape from anoikis in vitro. The expression level of Imp3 also appeared directly related to tumorigenic ability in vivo which is the critical determination for tumor-initiating cells. The effect of Imp3 on tumorigenicity of osteosarcoma cells did not appear to be mediated through Igf2-dependent mechanism. Our results implicate Imp3 as a key regulator of stem-like tumorigenic characteristics in osteosarcoma cells and as a potential therapeutic target for this malignancy.