Immune Selection and Emergence of Aggressive Tumor Variants as Negative Consequences of Fas-Mediated Cytotoxicity and Altered IFN-γ-Regulated Gene Expression

Introduction

Tumor-specific immune responses often develop or can be induced in cancer-bearing hosts via active or adoptive immunotherapy ( 1– 6), yet complete tumor eradication occurs infrequently ( 7– 12). The failure of the immune system to more consistently and effectively eradicate neoplastic disease in immune competent hosts is not fully understood and has remained a fundamental paradox in tumor immunology and immunotherapy. It has been proposed that cancerous cells use diverse mechanisms to counterattack the immune response, reflecting virtually all phases of antigen-specific T-cell development and activation ( 13). Therefore, although a variety of molecular alterations have been observed in tumors as they become more progressive and better equipped to evade or inhibit host defenses ( 14– 17), it remains to be fully understood how such changes in cancer cells are thought to occur initially and whether immunologically driven events may also contribute to the generation of tumor escape variants expressing those more aggressive phenotypes. One interesting proposal is that tumor cells expressing these more aggressive genetic or epigenetic traits emerge as a result of an endogenous immune-selective process termed “cancer immunoediting” ( 18). This phenomenon is conceptually akin to the generation of radioresistant or chemoresistant neoplastic clones.

Elements of both innate and adaptive immune responses, including natural killer cells and CD8⁺ T lymphocytes, respectively, engage the perforin/granzyme and Fas pathways as the principal effector mechanisms to mediate cellular cytotoxicity ( 19– 21). Furthermore, the production of IFN-γ by these activated lymphocytes, for example, has been shown to contribute significantly to antitumor activity via several mechanisms, including phenotypic or functional modification of neoplastic cells, rendering them more amenable to immune recognition and attack via Fas-dependent and Fas-independent pathways ( 22– 27). Indeed, earlier studies had pointed toward the perforin pathway as a major force regulating tumor development and progression ( 27– 32). Recent studies have now shown that the Fas pathway also plays a critical role against tumor growth or progression in mouse models, including those reflecting spontaneous or experimental metastasis ( 19, 23, 33– 38). Therefore, both the perforin and the Fas pathways constitute significant barriers against tumor growth.

The fact that both pathways exert positive antitumor properties raises the opposing hypothesis that if neoplastic subpopulations develop resistance to either one or both pathways, this may facilitate tumor escape, which in turn may influence metastatic formation. Thus, despite extensive efforts dedicated to the field of tumor escape, little is known about the relative contributions of these two lymphocyte-mediated effector mechanisms as potential selective pressures favoring the emergence of aggressive tumor variants (ATV). Consequently, this study addressed the relative contributions of the perforin- and Fas-based effector mechanisms in immune selection and emergence of ATV. To do so, we made use of a mouse model of experimental lung metastasis and isolated tumor sublines from mice under conditions of innate or adaptive immune-selective pressures. Genome-scale expression approaches were used to assess similarities and differences in gene expression profiles between the parental tumor and the corresponding sublines derived thereof. Overall, we showed that Fas-dependent, but not perforin-dependent, interactions were a dominant force driving the biological selection of more ATV. We further showed that such ATV displayed a diminished responsiveness in their expression of IFN-γ-regulated genes, a proinflammatory cytokine pathway important for the regulation of tumor-associated biological properties, such as immunogenicity, antigenicity, and even responsiveness to apoptotic induction ( 22– 27, 39). Collectively, our data indicated for the first time that Fas-mediated cytotoxicity, in cooperation with the actions of IFN-γ, constituted significant forces in the immune selection and emergence of more malignant tumor variants in vivo. The involvement of these two pathways in the generation of such ATV from preexisting malignancies may represent at least in part a type of cancer immunoediting facilitating tumor progression. Overall, these studies may have important implications for understanding the possible negative consequences of an antitumor immune response in the generation of ATV that may contribute at least in part to the failure of a given immunotherapy.

Materials and Methods

Mice. Female BALB/c (H-2^d) mice were obtained from the National Cancer Institute-Frederick Cancer Research Animal Facility (Frederick, MD). Female FasL-deficient CPt.C3-Tnfsf6gld mice on a BALB/c background (henceforth termed gld) were obtained from The Jackson Laboratory (Bar Harbor, ME). Female perforin-deficient (pfp) mice on a BALB/c background were kindly provided from M. Smyth (Peter MacCallum Cancer Institute, East Melbourne, Victoria, Australia) via R. Wiltrout (LEI, National Cancer Institute). All mice were housed, maintained, and studied in accordance with approved NIH guidelines for animal use and handling.

Tumor cells. The CMS4 sarcoma ( 40) cell line was kindly provided by A. DeLeo (University of Pittsburgh, Pittsburgh, PA). The CMS4-met subline was produced from the parental CMS4 population by one in vivo passage in the lungs of normal BALB/c mice as described ( 41). Tumor cells that emerged from these lung digests, termed CMS4-met, were then maintained in culture. The CMS4.sel subline was selected from the parental line in vitro following six successive passages in the presence of anti-Fas stimuli ( 34). Briefly, CMS4 cells were first treated with recombinant mouse IFN-γ (100 units/mL, R&D Systems, Minneapolis, MN) and tumor necrosis factor-α (TNF-α; 100 units/mL, R&D Systems) overnight followed by culture with anti-mouse Fas monoclonal antibody (mAb; 10 μg/mL, clone Jo2; PharMingen, San Diego, CA) and protein G (10 μg/mL, Sigma Chemical, St. Louis, MO), to maximize cross-linking of anti-Fas at approximately weekly intervals for a total of four cycles. These cells then underwent two additional cycles of IFN-γ/TNF-α exposure plus recombinant human soluble FasL (sFasL; 100 ng/mL, Alexis, San Diego, CA).

CMS4-met.cntl and CMS4-met.sel were derived from CMS4-met. CMS4-met cells (2.5 × 10⁵ cells per mouse) were injected i.v. into the lateral tail vein of normal BALB/c mice; 10 days later, wild-type (WT)-CTL or saline (HBSS) was injected i.v. as described below in Adoptive Transfer. Control or CTL-treated mice were euthanized and the tumor sublines, designated CMS4-met.cntl and CMS4-met.sel, respectively, were isolated from lung digests of independent mice as described above. Figure 1 delineates the ontogeny of the five different lines/sublines used in this study.

Immunostaining and flow cytometry. For Fas staining, tumor cells were stained with fluorescence-conjugated anti-Fas mAb (clone Jo2) or an isotype-matched hamster IgG and analyzed by flow cytometry. For IFN-γ receptor and TNF-α receptor staining, tumor cells were incubated with purified hamster anti-mouse IFN-γ receptor α-chain mAb (clone 2E2, PharMingen) or purified hamster anti-mouse TNF-α receptor p55 (clone 55R-286, PharMingen) followed by incubation with fluorescence-conjugated anti-hamster IgG (Kirkegaard & Perry, Gaithersburg, MD). Cell preparations were then analyzed by flow cytometry. Caspase-1 and STAT1 protein levels were determined by flow cytometric analysis of intracellular-stained cell preparations. Briefly, tumor cells were first fixed and permeabilized using the Fix & Perm kit (Caltag, Burlingame, CA). Cell preparations were then incubated with rat anti-mouse caspase-1 mAb (clone 1H11, Alexis) or mouse anti-STAT1 mAb (clone 1; cross-reactive with human and mouse cells, BD Biosciences/Transduction Laboratories, San Diego, CA) as described ( 42) followed by incubation with rhodamine-conjugated anti-rat IgG (Kirkegaard & Perry) or fluorescence-conjugated anti-mouse IgG (Kirkegaard & Perry), respectively. After staining, cells were analyzed by flow cytometry.

Measurement of cell death. Cell death was measured by propidium iodide (PI) staining; however, similar results were observed by terminal deoxynucleotidyl transferase–mediated dUTP nick end labeling assays. Briefly, cytokine-treated cells were incubated with or without recombinant human sFasL (20 ng/mL) for 20 to 24 hours. Cells were then collected and stained with PI for 10 minutes at room temperature according to the manufacturer's instructions (R&D Systems). After staining, the cells were washed and immediately analyzed by flow cytometry. The percentage of cell death was calculated by the formula: (% PI-positive cells with cytokine) − (% PI-positive cells without cytokine) or (% PI-positive cells with sFasL) − (% PI-positive cells without sFasL). In regard to the caspase-1 inhibition experiments, IFN-γ-treated tumor cells were cultured in the absence or presence of two different peptide-based caspase inhibitors with specificity for caspase-1, Z-YVAD-FMK, or Z-LEVD-FMK (ICN Biomedical, Aurora, OH) at a final concentration of 20 μmol/L as directed by the manufacturer. A negative control peptide, Z-FA-FMK (ICN Biomedical), was included in a parallel set of cultures. The tumor cells were preincubated with the various peptides at 37°C for 30 minutes followed by the addition of sFasL. The cultures were then incubated at 37°C for 24 hours, collected, stained with PI, and analyzed by flow cytometry.

Production of tumor-specific CD8⁺ CTL lines. CD8⁺ CTL lines reactive against the CMS4 sarcoma were established from either WT BALB/c, BALB/c-gld, or BALB/c-pfp mice using an anti-CTLA-4 mAb-based immunotherapy as described previously ( 37). Splenic-derived CD8⁺ CTL lines were maintained and propagated in 24-well plates (1 × 10⁵-2 × 10⁵ per well) by weekly stimulation with irradiated (20 Gy) syngeneic normal BALB/c splenocytes (5 × 10⁶ per well) as APC and irradiated (200 Gy) CMS4-met cells (1 × 10⁵ per well) as a source of cognate antigen and interleukin (IL)-2 (60 IU/mL, Tecin, Hoffmann-La Roche, Nutley, NJ).

Cytotoxicity assays. CTL activity was assessed by ⁵¹Cr release assays as described ( 37). Target cells were labeled with Na₂ ⁵¹CrO₄ (ICN Biomedical). CTL were recovered from culture by centrifugation over a Ficoll-Hypaque gradient. CTL and radiolabeled target cells were then coincubated in 96-well, U-bottomed plates at various effector/target ratios. After incubation for 18 hours, supernatants were collected using a Supernatant Collection System (Skatron Co., Sterling, VA). Radioactivity was quantitated using a gamma counter. Percentage specific ⁵¹Cr release was calculated according to the following formula: % specific lysis = [(experimental cpm − spontaneous cpm) / (total cpm − spontaneous cpm)] × 100%, where cpm is counts per minute. Total ⁵¹Cr release was obtained by adding 0.2% Triton X-100 (final concentration) to the wells. Data are reported as the mean ± SD of triplicate wells and representative of one of two separate experiments.

Adoptive transfer. Treatment of tumor-bearing mice by CTL adoptive transfer was conducted as described previously ( 41). Briefly, the indicated CMS4 line/subline was injected i.v. into the lateral tail vein (2.5 × 10⁵ cells in 100 μL HBSS). Ten days later, CTL (3 × 10⁶ cells in 100 μL HBSS, 4-5 days following in vitro stimulation) or saline (HBSS) was also injected into the lateral tail vein. Mice receiving CTL were euthanized 14 days after the adoptive transfer (or 24 days post-transplant), whereas control mice were typically euthanized 17 days post-tumor transplant because of disease burden. Lungs were processed for presence of tumor nodules by India ink staining as described ( 41).

Reverse transcription-PCR analysis. Total RNA was isolated from tumor cells using RNA STAT-60 reagent (Tel-Test, Friendswood, TX) according to the manufacturer's instruction and used for the first-strand cDNA synthesis using the ThermoScript Reverse Transcription-PCR (RT-PCR) System (Invitrogen, San Diego, CA). The cDNA was then used as templates for PCR amplification of mouse Fas, mouse STAT1, mouse caspase-1, and mouse β-actin. The following variables were used: 30 seconds at 94°C, 30 seconds at 60°C, and 1 minute at 72°C for various cycles (Fas, 30 cycles; all other genes, 26 cycles). The PCR primers for mouse Fas were as follows: forward primer 5′-ATGCTGTGGATCTGGGCT-3′ and reverse primer 5′-TCACTCCAGACATTGTCC-3′. The PCR primers for mouse β-actin were as follows: forward primer 5′-ATTGTTACCAACTGGGACGACATG-3′ and reverse primer 5′-CTTCATGAGGTAGTCTGTCAGGTC-3′. The PCR primers for mouse STAT1 were as follows: forward primer 5′-CTTCTTCCTGAACCCCCCG-3′ and reverse primer 5′-CCCATCATTCCAGAGGCACAG-3′. The PCR primers for mouse caspase-1 were as follows: forward primer 5′-AACATCTTTCTCCGAGGGTTGG-3′ and reverse primer 5′-TCAGCAGTGGGCATCTGTAGC-3′. To quantify PCR band intensities, gel images were first captured with an Epi ChemiII Digital Image System (UVP, Upland, CA). The individual PCR-amplified DNA fragment intensities were then analyzed with NIH-Image software (NIH, Bethesda, MD).

Western blot analysis. Tumor cells were solubilized in lysis buffer consisting of 1% Triton X-100 in 20 mmol/L Tris-HCl (pH 7.4), 150 mmol/L NaCl, 1 mmol/L EDTA, and a proteinase inhibitor cocktail (Sigma Chemical) for 60 minutes on ice. Proteins were separated on a 4% to 20% SDS-polyacrylamide gradient gel and transferred to Immobilon-P membrane (Millipore, Bedford, MA) using the Xcell II Blot Module (Invitrogen). Anti-STAT1 mAb (clone c111, Santa Cruz, San Diego, CA) or anti-β-actin mAb (clone AAC-15, Sigma Chemical) was used as the primary mAb followed by incubation with a horseradish peroxidase–conjugated anti-mouse IgG (Amersham-Pharmacia, Sunnyvale, CA). Immunodetection was done using the Enhanced Chemiluminescence Plus kit (Amersham-Pharmacia).

Global gene expression profiling. Total RNA was isolated from cells using RNA STAT-60 reagent and used for cDNA probe preparation. cDNA probes were synthesized using the FairPlay microarray labeling kit (Stratagene, La Jolla, CA). The cDNA probes were then labeled with Cy3 or Cy5 monofunctional reactive dye (Amersham Biosciences, Piscataway, NJ). The appropriate Cy3- and Cy5-labeled probes were combined along with 10 μg mouse Cot-1 DNA (Invitrogen, Carlsbad, CA) and 4 μg yeast tRNA in a final volume of 15 μL and incubated at 98°C for 1 minute. The denatured probes were mixed with 15 μL of 2× hybridization buffer (50% formamide, 10× SSC, 0.1% SDS). The hybridization solution and cDNA probe mixtures were added to the processed National Cancer Institute mouse oligomicroarray slides, which were then placed in hybridization chambers and incubated at 43°C for 16 hours. The slides were then washed for 5 minutes in 2× SSC and 0.1% SDS, for 5 minutes in 1× SSC, and for 5 minutes in 0.2× SSC and then spin dried. Fluorescence images were captured using a Genepix 4000 (Axon Instruments, Union City, CA). Both image and signal intensity data were loaded into a database supported by the Center for Information Technology of NIH. Cy3/Cy5 intensity ratios from each gene were calculated and subsequently normalized to ratios of overall signal intensity from the corresponding channel in each hybridization. The normalized data were then extracted from the database as text files and analyzed using computer software JMP (SAS Institute, Cary, NC) to compare the gene expression profiles quantitatively. For clustering analysis, Cluster and TreeView programs ( 43) were used to analyze the gene expression patterns in a one-dimensional hierarchical clustering to generate gene dendrograms based on the pair-wise calculation of the Pearson coefficient of normalized fluorescence ratios as measurements of similarity and linkage clustering. The clustered data were loaded into TreeView and displayed by the graded color scheme.

Statistical analysis. Where indicated, data were reported as the mean ± SD. Statistical analysis was determined using an unpaired, two-sided t test, with Ps < 0.05 considered statistically significant.

Results

An in vivo–derived CMS4 metastatic subpopulation genetically resembled a neoplastic subline generated in vitro by anti-Fas selection. The CMS4-met and CMS4.sel sublines served as a potentially important pair of malignant populations to examine the relationship among biological selection, Fas expression, and malignant phenotype ( Fig. 1). Because CMS4.sel was produced under defined in vitro conditions whereby anti-Fas was the only selective force ( 34), we used this subline as a reference population to compare with CMS4-met and to determine whether Fas-dependent interactions influenced the generation of CMS4-met in vivo. We found previously that these three groups of CMS4 cells exhibited different levels of cell surface Fas, which paralleled their sensitivity to Fas-mediated death after treatment with the proinflammatory cytokines IFN-γ and TNF-α, cytokines shown previously to maximize the apoptotic response ( 44). The parental CMS4 population was most sensitive to Fas-mediated death followed by CMS4-met and CMS4.sel ( 34). When these three tumor cell populations were injected into BALB/c mice, mice receiving CMS4 cells formed only a few detectable lesions (typically <20 foci per entire lung), whereas a larger number of nodules were visible in mice receiving CMS4-met and CMS4.sel (typically >150 foci per entire lung; ref. 34). Therefore, CMS4-met was more similar to CMS4.sel compared with the parental CMS4 line both in terms of expressing lower levels of Fas and higher metastatic tumor growth in the lung.

To better determine the relationship of the two metastatic sublines, we used genome-scale gene expression profiling to compare the gene expression patterns of these two metastatic sublines to the parental CMS4 line. Our DNA array chips contained 21,997-long oligonucleotides representing approximately 21,997 genes. Using an arbitrary cutoff of 1.6-fold (log₂ value, 0.47), any gene that had a log₂ ratio value of >0.47 in all three replicate experiments was considered to be overexpressed in CMS4-met and/or CMS4.sel. Conversely, any gene that had a log₂ ratio value of less than −0.47 in all three replicate experiments was considered underexpressed in CMS4-met and/or CMS4.sel. This cutoff was chosen for this analysis at least in part to generate a larger pool of genes and therefore a better representation of the general expression patterns for comparison at the genome-scale. Of the 21,997 genes analyzed, collectively 315 genes were identified based on this cutoff criterion. (The complete data are deposited in the ArrayExpress database, accession # 1E-MEXP-286.) Figure 2A illustrates the complete linkage hierarchical clustering of the expression patterns of the 315 genes.

To examine similarities or differences in the gene expression patterns between CMS4-met and CMS4.sel, we reanalyzed the 315 genes. Genes that had log₂ ratio values of >0.47 in one subline as well as a log₂ ratio value of >0 (overexpression direction) in the other subline were classified as an overexpressed pattern. Conversely, genes that had log₂ ratio values of less than −0.47 in one subline as well as a log₂ ratio value of <0 (underexpression direction) in the other subline were categorized as an underexpressed pattern. Genes that fell into the overexpression pattern or underexpression pattern were then considered to be expressed in a “similar expression pattern” between CMS4-met and CMS4.sel. The remainder of the genes was categorized as a “distinct expression pattern” between CMS4-met and CMS4.sel. By these criteria, 70.8% or 223 of these 315 genes displayed a similar expression pattern between CMS4-met and CMS4.sel, whereas 29.2% or 92 genes illustrated a distinct expression pattern ( Fig. 2B). Next, we plotted the genes in the “similar pattern” group and analyzed the expression levels of individual genes ( Fig. 2C). These data revealed that the expression patterns of both overexpressed and underexpressed genes were qualitatively similar between CMS4-met and CMS4.sel ( Fig. 2C); however, the expression intensities of these genes were different between these two sublines. The degrees of overexpression (P = 1.4 × 10⁻¹²) and underexpression (P = 1.2 × 10⁻⁸) were greater in CMS4.sel than in CMS4-met ( Fig. 2C). The average intensity ratios of the overexpressed genes for CMS4-met and CMS4.sel were 0.38 and 0.65, respectively, and the average intensity ratios of the underexpressed genes for CMS4-met and CMS4.sel were −0.4 and −0.63, respectively. Therefore, these data indicated that CMS4-met qualitatively resembled CMS4.sel. Because CMS4.sel was selected in vitro under defined conditions whereby anti-Fas was the only selective force, the fact that CMS4-met was similar to CMS4.sel suggested that host-derived, Fas-dependent events played a meaningful role in the generation of CMS4-met in vivo. However, the observation that nearly 30% of the genes displayed a distinct expression pattern implied that Fas-independent factors in vivo also helped to ultimately shape the CMS4-met tumorigenic phenotype.

Additional control experiments were done to assess whether IFN-γ and/or TNF-α played direct cytotoxic roles in the generation of CMS4.sel from the parental CMS4 population. To do so, we examined the lytic sensitivity of CMS4 cells to IFN-γ, TNF-α, or both cytokines together. As shown in Fig. 2D, essentially no cell death was detected in any of these cytokine-treated groups. In contrast, recombinant sFasL induced substantial death of the dual cytokine-treated CMS4 cells. Therefore, IFN-γ or TNF-α either separately or in combination did not induce death of CMS4 cells but rather served to sensitize these cells to Fas-mediated lysis, which is consistent with earlier observations ( 34). To further determine whether IFN-γ and/or TNF-α played a direct role in the generation of CMS4.sel, we examined these cells for changes in cell surface IFN-γ and TNF-α receptor expression levels ( Fig. 2E and F). Although all three groups of cells expressed low levels of both receptors, particularly for the TNF-α receptor, we found that CMS4.sel expressed the same degree of IFN-γ receptor and TNF-α receptor levels compared with CMS4 or CMS4-met. These data suggested that both of these receptors were not down-regulated by CMS4.sel as a result of the selection.

CMS4-met and CMS4.sel displayed altered responses to Fas-mediated, but not to perforin-mediated, cellular cytotoxicity. Perforin- and Fas-mediated cytotoxicity are two major antitumor effector mechanisms used by CTL. Next, we tested the three CMS4 populations for lytic sensitivity toward tumor antigen-specific CTL expressing or lacking a particular effector mechanism ( Fig. 3 ). To do so, we made use of three distinct CTL populations: WT-CTL (FasL-competent and perforin-competent), gld-CTL (FasL-deficient but perforin-competent), and pfp-CTL (FasL-competent but perforin-deficient). P815 was included as an antigen-negative control target ( 41) to illustrate lytic specificity. Previous experiments have shown that lysis of CMS4 targets by these WT-CTL occurred essentially through both perforin and Fas/FasL pathways ( 37).

Although pfp-CTL mediated lysis of all three CMS4 targets, lytic activity was highest against CMS4 followed by CMS4-met and CMS4.sel in descending order ( Fig. 3). This hierarchical pattern of lytic efficiency thus paralleled Fas expression levels and mirrored the cell death patterns seen with anti-Fas stimuli against these three cell lines ( 34). It is also important to note that such tumor-specific differences in lytic activity by pfp-CTL were unveiled in an effector/target titratable fashion ( Fig. 3). Because CMS4-met and CMS4.sel were not completely Fas resistant, Fas-mediated lysis by pfp-CTL may still occur during conditions of effector cell excess, resulting from the higher effector/target ratios. In contrast to pfp-CTL, WT-CTL and gld-CTL mediated lysis of all three targets with comparable lytic efficiency, suggesting that perforin sensitivity of these targets was not significantly altered ( Fig. 3). These latter data also illustrated that the Fas/FasL pathway became unmasked under conditions in which the perforin pathway became compromised, such as in the case with pfp-CTL or with pharmacologic inhibition ( 37). Therefore, the metastatic sublines (CMS4-met and CMS4.sel) retained lytic sensitivity toward perforin but displayed diminished sensitivity to Fas-mediated cytotoxicity (as determined by pfp-CTL).

Tumor-specific CTL exerted a selective pressure that contributed further to the generation of aggressive tumor variants in vivo. Our results thus far suggested that if Fas-dependent interactions were involved in the generation of CMS4-met in vivo, such interactions likely were initiated by resident cell types within the host. Because the kinetics of tumor growth in the lung occur rather quickly over the 14-day period, it was likely that tissue-associated events or innate immunity, as opposed to an endogenous adaptive immune response, contributed more significantly to such an antitumor selective pressure. To examine the potential contribution of an adaptive immune response toward immune selection, we made use of an adoptive immunotherapy paradigm consisting of CMS4-specific CD8⁺ CTL (i.e., WT-CTL). In this model, mice received CMS4-met; 10 days later, when tumor burden was well established, these same mice received treatment with CTL via adoptive transfer. Under such conditions of a preexisting extensive disease burden, CTL adoptive transfer led to significant but incomplete tumor regression ( 37). This model thus allowed us to recover surviving tumor cells, which were then examined for potential alterations in Fas expression/function and metastatic ability on reinjection into fresh groups of syngeneic mice. In fact, we were able to establish several CMS4-met sublines from mice after interaction with WT-CTL in vivo. As controls, CMS4-met sublines were produced from mice receiving saline instead of CTL. The following studies were then conducted using representative sublines from CTL-treated mice and control mice, henceforth termed CMS4-met.sel and CMS4-met.cntl, respectively.

In mice receiving equal numbers of the three tumor sublines, CMS4-met.sel displayed the highest level of metastatic lung tumor growth followed by CMS4-met.cntl and CMS4-met in descending order ( Fig. 4A ). (In general, because all three sublines were aggressive, it was difficult to precisely quantitate the number of foci/lung; therefore, these observations were depicted as photomicrographs.) In fact, CMS4-met.sel formed so many nodules that it essentially covered the entire surface of the lung. Additionally, both CMS4-met.cntl treated with IFN-γ alone (P = 0.02), TNF-α alone (P = 0.05), or both cytokines (P = 0.012) and CMS4-met.sel treated with IFN-γ alone (P = 0.005), TNF-α alone (P = 0.02), or both cytokines (P = 0.005) were significantly less responsive to Fas-mediated death compared with CMS4-met under these same single or paired cytokine treatment conditions ( Fig. 4B). Analysis of TNF-α receptor expression by flow cytometry also revealed that both in vivo–derived sublines displayed low but comparable levels (data not shown) with those of CMS4-met ( Fig. 2E). Thus, under innate or adaptive immune-selective pressures, these observations indicated that more malignantly proficient tumor cell subpopulations could be recovered, which also displayed lower levels of Fas sensitivity.

To better assess the relationship between CMS4-met.cntl and CMS4-met.sel sublines, we used genome-scale gene expression profiling to compare the gene expression patterns with CMS4-met using the approach described in Fig. 2. Of the 21,997 genes analyzed, collectively 1,590 genes were selected based on that cutoff criterion. (The complete data are deposited in the ArrayExpress database, accession # 2E-MEXP-287.) Fig. 4C illustrated the complete linkage hierarchical clustering of the expression patterns of these 1,590 genes. Overall, the genome-scale gene expression patterns between CMS4-met.cntl and CMS4-met.sel were very similar, and only small fractions were distinct ( Fig. 4C). To better compare the similarity between these two sublines, we then grouped these 1,590 genes into overexpression and underexpression patterns, similarly as described in Fig. 2. By this selection criteria, 83.4% or 1,326 of these 1,590 genes displayed a similar expression pattern between CMS4-met.cntl and CMS4-met.sel. In contrast, 16.6% or 264 of these 1,590 genes possessed a distinct expression pattern ( Fig. 4D).

Next, we plotted the overexpressed and underexpressed genes separately in a dot plot format ( Fig. 4E). These data revealed that the expression patterns of both overexpressed and underexpressed genes were clearly similar in CMS4-met.cntl and CMS4-met.sel. However, differences between these two variant sublines were again noted, which resided in their gene expression intensities. The degrees of overexpression and underexpression were greater in CMS4-met.sel compared with CMS4-met.cntl. The average intensity ratios of the overexpressed genes for CMS4-met.cntl and CMS4-met.sel were 0.61 and 0.93, respectively (P = 9.7 × 10⁻³⁵).The average intensity ratios of the underexpressed genes for CMS4-met.cntl and CMS4-met.sel were −0.44 and −0.83, respectively (P = 7.7 × 10⁻⁶³). Taken together, we concluded that CMS4-met.sel qualitatively, but not necessarily quantitatively, resembled CMS4-met.cntl in their global gene expression profiles.

Identification of genes overexpressed and underexpressed in the CMS4-met.sel subline. To identify individual genes that were differentially expressed in the CMS4-met.sel subline compared with the CMS4-met subline, we selected genes shown in Fig. 4 that were either overexpressed or underexpressed by at least 2-fold in all three replicate experiments. Such genes were then functionally classified according to their ontology. Five groups were defined and known genes in these five groups are illustrated in Table 1 : (a) immune response, (b) cell cycle and death control, (c) cell adhesion and communication, (d) transcription regulation, and (e) signal transduction. Because numerous genes are shown, their precise functional roles in the processes of immune selection, Fas resistance, and/or malignant proficiency warrant further study. Nonetheless, in regard to genes potentially pertinent to or correlated with Fas resistance, two genes were noted: Fas death domain-associated protein and STAT1. Because the Fas death domain-associated protein has been described to bind the Fas receptor ( 45), its down-regulation in CMS4-met.sel might affect its sensitivity toward Fas-mediated death. Although STAT1 is both a transcription factor and a signaling molecule downstream of IFN-γ receptor engagement, it has also been linked to the regulation of Fas-mediated death ( 25). Therefore, it is possible that as tumors become more aggressive, resultant subpopulations or clones may become much less responsive to the actions of IFN-γ perhaps at the level of IFN-γ-regulated gene expression, which is explored below.

Genome-scale analysis of IFN-γ-induced alterations in gene expression of the CMS4 and CMS4-met populations. Next, we examined the expression profiles of IFN-γ-regulated genes in the CMS4-met subline, for example, compared with its less aggressive parental CMS4 population. CMS4 and CMS4-met were pretreated with IFN-γ for either 4 or 24 hours and then collected for gene expression kinetics by DNA microarray analysis. Three independent experiments under each condition were carried out. Again, an arbitrary cutoff of 1.6-fold (log₂ value, 0.47) was chosen to select genes whose expression levels were either increased (larger than 0.47 in all three replicate experiments) or decreased (less than −0.47 in all three replicate experiments) by IFN-γ treatment. Based on these selection criteria, we collectively identified 689 genes whose expression levels were altered in at least one time point in CMS4 and/or CMS4-met. (The complete data are deposited in the ArrayExpress database, accession # 3E-MEXP-288.) Next, we clustered these 689 genes to analyze their expression patterns in CMS4 and CMS4-met as described in Materials and Methods ( Fig. 5A ).

In general, the gene expression kinetics between CMS4 and CMS4-met were similar ( Fig. 5A), indicating that in vivo selection (of CMS4-met) did not qualitatively cause significant changes in the expression patterns of IFN-γ-regulated genes at least as defined by this genome-scale approach. To analyze the expression profiles of these IFN-γ-regulated genes in further detail, we grouped these genes into up-regulated and down-regulated patterns as described in Figs. 2 and 4. By this criteria, 96.6% of these genes displayed a similar expression pattern between CMS4 and CMS4-met, whereas 3.4% of these genes harbored a distinct expression pattern at the 4-hour time point ( Fig. 5B). At 24 hours, however, the gene expression patterns between these two cell populations were even more similar, with 97.4% displaying a similar gene expression pattern ( Fig. 5B).

Next, we plotted the up-regulated and down-regulated genes separately in a dot plot format ( Fig. 5C). The overall expression patterns of both up-regulated and down-regulated genes were very similar between CMS4 and CMS4-met. However, differences were noted between these two populations at a quantitative level in that the increase or decrease in mRNA levels of IFN-γ-regulated genes was greater in CMS4 compared with CMS4-met at both time points ( Fig. 5C). The average log ratios of the up-regulated genes were 0.62 and 0.87 at 4 and 24 hours, respectively, in CMS4 cells and 0.50 and 0.67 at 4 and 24 hours, respectively, in CMS4-met cells. Statistical analysis indicated that the differences in the up-regulated genes were not significant between CMS4 and CMS4-met at 4 hours but reached significance at 24 hours (P = 5.3 × 10⁻⁷). The average log ratios of the down-regulated genes were −0.24 and −0.78 at 4 and 24 hours, respectively, in CMS4 cells and −0.20 and −0.55 at 4 and 24 hours, respectively, in CMS4-met cells ( Fig. 5C). Statistical analysis indicated that the difference in the down-regulated genes also was not significant between CMS4 and CMS4-met at 4 hours but reached significance at 24 hours (P = 5.3 × 10⁻¹⁰). Taken together, these data indicated that CMS4-met expressed an overall lower responsiveness to IFN-γ, consistent with the notion that a more malignant phenotype at least in this model also harbored a comparatively reduced responsiveness to IFN-γ signaling.

Differential up-regulation of Fas, STAT1, and caspase-1 in CMS4 versus CMS4-met. We showed that the expression intensities of the IFN-γ-regulated genes were altered in CMS4-met compared with CMS4. To determine whether changes in IFN-γ-regulated gene expression were functionally linked to the Fas pathway, we focused on three genes, Fas, STAT1, and caspase-1, whose expression levels were differentially altered after IFN-γ treatment as revealed by DNA microarray analysis. RT-PCR analysis confirmed that CMS4-met compared with CMS4 was less responsive quantitatively to IFN-γ in terms of up-regulation of Fas, STAT1, and caspase-1 ( Fig. 6A ). Quantification of the PCR band intensities revealed that the Fas transcript level was significantly higher in CMS4 compared with CMS4-met at both 4 hours (P = 0.004) and 24 hours (P = 0.008) after IFN-γ treatment ( Fig. 6B). STAT1 transcript level was also significantly higher in CMS4 compared with CMS4-met at both 4 hours (P = 0.009) and 24 hours (P = 0.03) after IFN-γ treatment ( Fig. 6B). Whereas caspase-1 transcript level was not significantly different between CMS4 and CMS4-met cells 4 hours after IFN-γ treatment ( Fig. 6B), it was significantly different 24 hours (P = 0.05) after IFN-γ treatment ( Fig. 6B).

Next, we examined STAT1 and caspase-1 at the protein level in CMS4 and CMS4-met after IFN-γ treatment for 24 hours. Intracellular staining with a caspase-1-specific mAb revealed weak but reproducible caspase-1 protein expression in both cell lines ( Fig. 6C). IFN-γ treatment increased caspase-1 protein levels in both CMS4 and CMS4-met ( Fig. 6C). The difference in caspase-1 protein levels between CMS4 and CMS4-met was significant for both untreated (P = 0.007) and IFN-γ-treated cells (P = 0.00006; Fig. 6C). The STAT1 protein levels were low in both CMS4 and CMS4-met ( Fig. 6D). IFN-γ treatment increased total STAT1 protein levels in both CMS4 and CMS4-met. However, the increase was greater in CMS4 compared with CMS4-met for both untreated (P = 0.07) and IFN-γ-treated cells (P = 0.009; Fig. 6D).

We also examined changes in STAT1 levels by Western blot analysis. Engagement of the IFN-γ receptor by IFN-γ activates Janus-activated kinase that phosphorylates STAT1. The phosphorylated STAT1 then translocates to the nucleus and functions as a transcription factor to regulate IFN-γ-regulated gene expression ( 46). Therefore, STAT1 exists in both unphosphorylated and phosphorylated forms. Here, we showed that IFN-γ treatment differentially increased the total STAT1 protein levels in CMS4 versus CMS4-met ( Fig. 6D). To determine whether IFN-γ treatment increased both forms of STAT1 in CMS4 and CMS4-met, we examined the STAT1 protein levels in CMS4 and CMS4-met before and after IFN-γ treatment by Western blot analysis. The Western blots were probed with a STAT1-specific mAb that cross-reacts with both forms of STAT1. We found that untreated CMS4 and CMS4-met contained only unphosphorylated STAT1 ( Fig. 6D, bottom right). IFN-γ treatment not only increased unphosphorylated STAT1 protein levels but also induced STAT1 phosphorylation in both cell lines. However, the levels of both unphosphorylated and phosphorylated STAT1 seemed greater in CMS4 compared with CMS4-met ( Fig. 6D, bottom right).

Previously, we have shown that IFN-γ sensitized human colon carcinoma cells to Fas-mediated death through induction of caspase-1 ( 23). Similarly, we now showed that IFN-γ regulated caspase-1 in this mouse tumor model ( Fig. 6A-C). To determine whether caspase-1 was linked to IFN-γ sensitization in this study, we examined the functional role of caspase-1 in Fas-mediated death in IFN-γ-treated CMS4 cells ( Fig. 6E and F). The IFN-γ-treated CMS4 cells were incubated with recombinant sFasL in the absence or presence of two different peptide-based caspase-1 inhibitors or a negative control peptide. In the presence of the caspase-1 inhibitors, sFasL-induced cell death was strongly reduced, whereas the negative control peptide had no significant inhibitory effect ( Fig. 6E and F). Therefore, a more malignant tumor phenotype, such as CMS4-met, may also possess a reduced capacity to respond to IFN-γ signaling, including genes involved in or required for Fas-induced death.

Discussion

Tumor escape mechanisms have been generally associated with the down-regulation of MHC-antigen complexes, cellular adhesion molecules, or costimulation capabilities as well as aspects related to tolerance induction or immune suppression of T-cell activation or proliferation within the host-tumor microenvironment ( 14– 17, 47– 49). What remains to be further understood, however, is whether host innate or adaptive immune interactions influence the emergence of more ATV harboring alterations in Fas-mediated death as well as Fas-associated and IFN-γ-regulated gene expression, which was the basis of this study. If the immune response against a neoplasm can mediate selection of Fas/IFN-γ-refractory ATV, for example, this may have profound consequences in the field of cancer immunotherapy.

Because of extensive similarities in genome-scale gene expression patterns between CMS4-met and CMS4.sel, our findings supported the notion that anti-Fas selection likely played an important role in the biological selection and generation of CMS4-met in vivo ( Fig. 2). CMS4.sel was generated in vitro with defined anti-Fas stimuli ( 34). Therefore, the selection pressure applied in the generation of CMS4.sel was Fas directed. Under these conditions, the Fas-sensitive cells were gradually eliminated, whereas the Fas-resistant cells were reciprocally enriched. The observations that CMS4-met was similar to CMS4.sel in terms of Fas expression, metastatic capability, and genome-scale gene expression patterns strongly suggested that endogenous Fas/FasL interactions regulated the types of tumor variants that proliferated and colonized in the lung. This notion is further supported by the finding that CMS4-met retained sensitivity to perforin-mediated lysis (using gld-CTL) but displayed less sensitivity to Fas-mediated lysis (using pfp-CTL) compared with the parental CMS4 population, which was lysed equally well by both gld-CTL and pfp-CTL ( Fig. 3). Therefore, it was reasonable to postulate that resident Fas-based interactions represented a dominant but not necessarily the exclusive determinant of the biological selection of the CMS4-met variant.

We then sought to examine the influence of an adaptive T cell response on the potential generation of ATV in vivo, which may have important implications for active and/or adoptive immunotherapy. Scientifically, the use of tumor-specific T cells as an extrinsic biological selective pressure in vivo also represented a more specific, targeted, and defined approach to evaluate the consequences of host-tumor interactions. The CMS4-specific CTL employed here as an adaptive immune-selective pressure is lytically efficient both in vitro and in vivo ( 37, 41) and uses both perforin- and Fas-dependent pathways to lyse their targets ( 37). We made use of this model to examine whether ATV can emerge from CTL-tumor interactions in vivo. In fact, we were able to isolate stable CMS4 sublines from mice treated with antigen-specific CTL, which expressed deficiencies in Fas function and possessed enhanced metastatic ability ( Fig. 4). Although it remains to be fully determined whether deficiencies in Fas function were causally related to the selection process, it is noteworthy that increased Fas resistance was a characteristic of such ATV.

Although CMS4-met.sel exhibited a genome-scale gene expression profile that qualitatively resembled CMS4-met.cntl, we also observed quantitative differences in gene expression intensities between these two sublines. Such quantitative differences in gene intensity profiles between these sublines may have significant bearing on their biological behavior, which warrants further investigation. Furthermore, it is important to point out that CMS4-met.sel sublines were produced from mice only after a 2-week CTL-tumor interaction in vivo. Consequently, it remains to be fully understood whether additional tumor escape mechanisms, besides those described in this study, may be operative which either were not yet observed or in fact may occur with a longer, more persistent adaptive immune (CTL-induced) selective pressure.

It is also interesting to note that CMS4-met.cntl was even more Fas insensitive compared with CMS4-met ( Fig. 4). This observation suggested that host-derived anti-Fas interactions (via a successive in vivo passage) played an important role in the nature of the consequential tumorigenic phenotypes, which may reflect a form of “cancer immunoediting” ( 18). Moreover, the observations that the expression intensities of genes overexpressed and underexpressed were greater in CMS4-met.sel compared with CMS4-met.cntl and that malignant proficiency was improved in CMS4-met.sel compared with CMS4.met.cntl ( Fig. 4) indicated that CTL adoptive transfer exerted an additional selective pressure over and above that of the host response alone on the nature of the resultant tumor variants. The exact functional significance of these genes, however, remains to be elucidated.

Lastly, we analyzed gene expression profiles between a matched pair of poorly and highly metastatic CMS4 variants in response to IFN-γ exposure ( Figs. 5 and 6). IFN-γ production is essential for protecting a host against tumor development and growth, perhaps at least in part through altering immunogenicity as well as regulating or sensitizing developing or progressing malignancies to Fas-mediated death ( 22– 25, 39, 50). Here, we found that the genome-scale gene expression kinetics was almost identical between these two populations. However, the degree of these up-regulated and down-regulated genes was greater in CMS4 than in CMS4-met, suggesting that an altered expression level of IFN-γ-regulated genes at the genome scale, and not just a few unique genes, potentially underscored a molecular basis for the differential sensitivity of these populations to Fas-mediated death and perhaps their distinct metastatic behavior.

Some of the antineoplastic effects of IFN-γ have been attributed to the direct modulation of apoptotic properties of tumor cells. Tumor cells that develop an altered responsiveness in IFN-γ-regulated gene expression may have a unique biological advantage that favors tumor escape from immune interactions. In this mouse model, the sensitivity of CMS4 cells to Fas-mediated death in vitro was potentiated by IFN-γ via a caspase-1-dependent or caspase-1-regulated mechanism. The observations that CMS4-met was less responsive to IFN-γ in genome-scale gene expression ( Fig. 5) as well as in their expression levels of several specific IFN-γ- or Fas-associated genes compared with CMS4 ( Fig. 6) strongly suggested that IFN-γ also served as a potentially integral determinant in the immune selection process.

The results from these studies showed for the first time that both intrinsic (host-derived or innate immune factors) and extrinsic (adaptive immune elements) factors likely influenced the generation of the various CMS4-met tumorigenic phenotypes, reflecting at least in part the outgrowth of tumor variants bearing a progressive down-regulation of Fas expression/function. It is important to point out, however, that because these studies were conducted with an already existing tumor cell population, our observations defined changes to a malignancy that had previously likely developed effective escape mechanisms. Thus, in the face of additional intrinsic or extrinsic selective pressures, as shown in this study, preexisting tumors may be further “edited” or reshaped resulting in variants with progressively heightened malignant properties.

In this model, experimental metastatic tumor formation occurs rapidly during these studies. Therefore, in the absence of any extrinsic, tumor antigen-specific immune pressure, the CMS4-met or CMS4-met.cntl sublines likely emerged as a consequence of a host (resident) and/or innate immune-based selective mechanism. For example, these findings do not exclude the possibility that host-derived, FasL-expressing lung cells ( 51) might play a role in CMS4-met generation, which remains to be determined. In contrast, the CMS4-met.sel subline likely emerged as a result of both intrinsic and extrinsic elements, which may have important considerations for immunotherapy applications. Both CMS4-met.cntl and CMS4-met.sel sublines expressed reduced sensitivity to Fas-mediated death and enhanced metastatic capability compared with CMS4-met, suggesting that Fas-dependent interactions was an important component or characteristic of the immune selection process and thus a possible outcome of cancer immunoediting. These findings also suggest that how a malignant population responds to IFN-γ signaling underlie a molecular basis of tumor regression or tumor progression (escape/immune selection), converging at least in part at the level of cytokine (IFN-γ) regulation of Fas expression and function.