Long-Term Suppression of Tumor Growth by TNF Requires a Stat1- and IFN Regulatory Factor 1-Dependent IFN-γ Pathway but Not IL-12 or IL-18¹

Tumornecrosis factor released locally by immune effector cells such as lymphocytes and macrophages can play an essential role in the rejection of cancer cells (1). The local effects of TNF on tumor development have been modeled extensively in mice using tumor cells engineered to secrete TNF (2, 3, 4, 5, 6, 7). TNF secretion consistently causes a strong long-term suppression of tumor formation in vivo regardless of the genetic background of the host and origin of the tumor cells. As in natural immune responses, the suppression is localized near the TNF-secreting cells and does not extend to tumor cells implanted at distant sites because the TNF concentration rarely increases systemically. The powerful antitumor effect of TNF is thought to be triggered through TNF-RI even though almost all cell types express two TNFRs, TNF-RI and TNF-RII, because murine TNF and human TNF, which binds only murine TNF-RI (8), have similar antitumor effects in mice (5). The mechanism also appears to be indirect (4, 9, 10) and requires radiation-sensitive host cells (2, 3) using the B₂ integrins ICAM-1 (7) and/or MAC-1 (4). Granulocytes (6) and T cells (2, 4, 7) have been implicated in suppression. CD4 and CD8 T cells may initiate suppression by releasing TNF at the tumor site, but once TNF is produced, they are clearly not essential for suppression (3, 4, 5) and only contribute to a stronger suppressive effect and occasionally the complete rejection of antigenic tumor cells (1, 2, 7). The contribution of radiation-resistant, non-bone marrow (BM)³-derived cells constituting the tumor bed that supports tumor growth has largely been neglected even though they also express both TNFRs. Moreover, the involvement of additional cytokines that may help TNF to orchestrate the appropriate immune response has not been examined.

The aim of this study was to determine whether TNFR expression by non-BM-derived host cells and IFN-γ, a cytokine known to enhance synergistically the antitumor activities of TNF, are required for locally released TNF to suppress tumor development. We report herein that tumor suppression required TNFR expression by both BM- and non-BM-derived host cells. It also required that the BM-derived cells have intact IFN-γ, IFN-γR, Stat1, and IFN regulatory factor 1 (IRF-1) genes, but the active production of IFN-γ and its activity could not be detected. These results point to a novel function of the endogenous IFN-γ pathway that affects the BM cells essential for locally released TNF to suppress tumor growth.

B6.129-Tnfrsf1a^tm1Mak (TNF-RI^−/−), B6.129S7-Ifng^tm1Ts (IFN-γ^−/−), B6.129S2-Irf1^tm1Mak (IRF-1^−/−), B6.129S1-Il12b^tm1Jm (IL-12^−/−), B6.129S7-Rag1^tm1Mom (Rag1^−/−), and the 129-Ifngr^tm1Agt (IFN-γR^−/−) mice originally generated in the 129 background (11) were bred from breeder pairs initially obtained from The Jackson Laboratory (Bar Harbor, ME). B6.129P2-Il18^tm1Aki (IL-18^−/−) mice, backcrossed to C57BL/6 for eight generations, were kindly provided by Dr. A. Zychlinsky (New York University School of Medicine, New York, NY) and were intercrossed with IL-12^−/− mice to generate IL-12^−/−IL-18^−/− mice. C57BL/6 mice were purchased from the National Cancer Institute, Frederick Cancer Research Facility (Frederick, MD). 129S6/SvEv and 129S6/SvEv-Stat1^tm1 (Stat1^−/−) mice were obtained from Taconic Farms (Germantown, NY). All mice were maintained in a specific pathogen-free barrier facility at the University of Chicago (Chicago, IL). The 8101 and the 9203 cell lines were established from UV-induced tumors in a C57BL/6 mouse and a 129S6/SvEv-Stat1^tm1 mouse, respectively. The PRO4L.7 progressor tumor cell line was derived from a UV-induced regressor tumor in a C3H mouse (12). The Mc51-9 tumor cell line was established from a tumor induced by methylcholanthrene in a 129-Ifngr^tm1Agt mouse (13). The J558L-IFN-γ, 6132A-IFN-γ cell lines and the XMG1.2 hybridoma secreting anti-IFN-γ Ab have been described previously (14, 15, 16). All tumor cell lines caused progressive tumors when 5 × 10⁶ cells were injected s.c. into wild-type (Wt) syngeneic mice. The tumor cell lines were cultured in DMEM with 5% FCS (cDMEM).

Transfection of PRO4L.7 cells with the human TNF gene has been described elsewhere (5). 8101 was transfected with the pCMVIEAK1-DHFR plasmid encoding human TNF (17) using Superfect (Qiagen, Valencia, CA). A clone, 8101-TNF, that secreted up to 12,000 U (216 ng)/10⁵ cells/day into 1 ml of medium and proliferated at a rate similar to the parental 8101 cells was expanded from G418-resistant transfectants. 9203 and Mc51-9 tumor cells and XMG1.2 hybridoma cells were made to secrete TNF by retroviral transduction. Infectious retroviruses were generated by transfecting the Phoenix ecotropic or amphotropic packaging cell lines (Dr. G. Nolan; Stanford University, Palo Alto, CA) with the SFG-hTNF construct (Dr. G. Dranoff; Dana-Farber Cancer Institute, Boston, MA). Culture supernatants were collected 48 h after transfection, filtered through a 0.45-μm Millex-HA syringe filter (Millipore, Bedford, MA), and incubated with 9203, Mc51-9, and XMG1.2 cells in the presence of 8 μg/ml Polybrene (Sigma-Aldrich, St. Louis, MO) for 24 h. Flow cytometric analyses for intracellular TNF in the bulk infectants 9203-TNF, Mc51-9-TNF, and XMG-TNF using PE-conjugated anti-Hu-TNF-α (BD Immunocytometry Systems, San Jose, CA) showed that the majority of infectants produced TNF. The high infection efficiency of this system allowed the use of bulk infectants for experiments. 9203-TNF and Mc51-9-TNF secreted up to 10,600 U (189 ng) and 7000 U (125 ng)/10⁵ cells/day into 1 ml of medium, respectively.

Tumor cell proliferation was measured over several days using the MTT assay. Briefly, 10³ tumor cells in 100 μl of cDMEM were plated into the wells of a 96-well tissue culture plate (Costar, Corning, NY) and maintained in a 10% CO₂ humidified incubator. At specified times, 20 μl of 5 mg/ml MTT (Sigma-Aldrich) in PBS was added to each well and the cultures were incubated for 4–24 h at 37°C. After addition of 100 μl of acidified SDS (10% SDS and 0.02 N HCl), plates were incubated overnight at 37°C. The absorbance at 570 and 650 nm was measured in a VERSAmax 96-well plate reader (Molecular Devices, Sunnyvale, CA). The value obtained by subtracting the absorbance at 650 nm from 570 nm reflects tumor cell density and was used as a measure of tumor cell proliferation over time. For growth in vivo, mice were injected s.c. in the flanks with 5 × 10⁶ tumor cells. Tumor volume was calculated using the formula V = πabc/6, where a, b, and c are the three orthogonal diameters of the tumor.

The TNF assay is based on the cytotoxicity of TNF against the TNF-sensitive cell line 1591-RE3.5. Briefly, 10⁵ TNF-secreting tumor cells were cultured in 1 ml of cDMEM in a 24-well tissue culture plate (Costar). Culture supernatants were collected after 24 h and serially diluted 2-fold with cDMEM in 96-well plates. Recombinant human TNF (Genentech, South San Francisco, CA) was serially diluted 10-fold, starting at 5 × 10⁵ U/ml. Pipette tips were changed with every dilution to prevent carryover. Five × 10³ 1591-RE3.5 cells were then added to each well and incubated at 37°C for 48 h. The cytotoxicity of TNF against 1591-RE3.5 cells was measured using the MTT assay described in the previous section. A standard curve was generated by plotting the percent cytotoxicity, calculated using the formula (1 − Abs_TNF/Abs_medium) × 100 against the concentration of rTNF, and the amount of TNF secreted by the tumor cells was extrapolated from the amount of rTNF required to achieve 50% cytotoxicity.

Thirty milligrams of lyophilized gammaglobin fraction of rabbit anti-asialo GM1 antiserum precipitated with 50% ammonium sulfate (Wako Chemicals, Dallas, TX) was reconstituted with 1 ml of PBS. Rag-1^−/− mice were depleted of NK cells by i.p. injection of 22 μl of the reconstituted antiserum 3 days before tumor challenge and every 3 days thereafter until the end of the experiment. Depletion was confirmed at the end of the experiment by flow cytometric analyses of peripheral blood cells stained for the pan-NK-reactive DX5 Ab (BD PharMingen, San Diego, CA). Ascites used for neutralizing IFN-γ in vivo was generated in athymic nude mice primed with pristane (Sigma-Aldrich) and inoculated with the XMG1.2 hybridoma cells. Endogenous IFN-γ was neutralized by i.p. injection of ascites fluid at indicated times before tumor inoculation and weekly thereafter until the end of the experiment.

Tumor-bearing mice were killed by overexposure to CO₂. Mice were sprayed with 70% ethanol before the skin was peeled away to expose the tumors. Fragments of the tumors were removed and minced using sterilized razor blades. The minced tumor fragments were transferred to 25-cm² tissue culture flasks containing 5 ml of cDMEM supplemented with 50 μg/ml gentamicin (Life Technologies, Rockville, MD) and 50 U/ml penicillin/streptomycin (Life Technologies) and cultured at 37°C in a humidified incubator. The culture medium was refreshed when tumor cells from the fragments had adhered to the flask and begun to proliferate.

Total RNA was isolated from tumor fragments using an RNeasy kit according to the manufacturer’s instructions (Qiagen). Total RNA was treated with DNase I (Life Technologies) and then reverse transcribed into cDNA. First-round PCR was performed with the primers 5′-AAC GCT ACA CAC TGC ATC TTG G-3′ and 5′-GAC TCC TTT TCC GCT TCC TG-3′ and nested PCR was performed with the primers 5′-GAC TTC AAA GAG TCT GAG G-3′ and 5′-GCT GTT TCT GGC TGT TAC TG-3′. The primers were purchased from Life Technologies.

Briefly, 2 × 10⁵ BM cells or 5 × 10⁵ splenocytes from Wt C57BL/6 mice were incubated with 100, 500, or 2500 tumor cells in 1 ml of cDMEM in the wells of 24-well tissue culture plates. After 24 or 48 h, culture supernatants were collected and measured for IFN-γ production by ELISA (Endogen, Woburn, MA). As a positive control for IFN-γ release, BM cells and splenocytes were primed with 20 U/ml recombinant human IL-2 for 4 h and then stimulated with 10 μg/ml LPS (18).

Recipient mice were medicated with Bactrim (Hoffman-LaRoche, Nutley, NJ) in drinking water (5 ml per 500-ml drinking bottle) 3 days before BM transfer. Recipient mice were irradiated on the day of transfer. In preliminary experiments, mice were irradiated with 10 Gy; however, because the 8101 tumor grew significantly slower or not at all in some recipient mice, the radiation dose was reduced to 3 and 6 Gy. Since the lower radiation doses did not significantly affect the growth rate of the 8101 tumor when it was transplanted 3–4 wk after irradiation, these radiation doses were used in all of the presented experiments. The BM from the femurs of one donor mouse was used for four recipient mice. To isolate the BM, the ends of the femurs were cut off and the BM was flushed from the femur with serum-free DMEM using a 25-gauge needle attached to a 1-ml syringe. The BM was dispersed into a single cell suspension, and debris and cell clumps were removed by filtration through a Nytex membrane (TETKO, Kansas City, MO). Typical yield from two femurs was between 1 and 1.5 × 10⁷ cells. The BM cells were resuspended at 10⁷ cells/ml and 0.2 ml was injected either in the tail vein or through the retro-orbital capillary plexus. The extent of reconstitution was determined 3–4 wk after reconstitution by flow cytometric analyses of peripheral blood from chimeric mice for donor-derived cells.

Mean tumor volume at the specified time after tumor challenge was compared between groups using the Wilcoxon rank sum test. Time to tumor distribution was computed using the Kaplan-Meier estimator and compared between groups by the log rank test.

When injected s.c., tumor cells secreting TNF caused a poorly vascularized, pale, flat, and granular lesion that persisted for weeks without changing or only very slowly increasing in size (Fig. 1, E and F). In contrast, tumor cells not secreting TNF rapidly caused a well-vascularized tumor (Fig. 1, I and J). Histologically, individual TNF-secreting tumor cells were interspersed among a dense mononuclear and polymorphonuclear infiltrate (Fig. 1, G and H) while non-TNF-secreting tumor cells formed a densely packed mass with occasional cellular infiltrate (Fig. 1, K and L). Similar results were obtained with TNF-secreting tumor cells from different strains (data not shown). This histologic appearance agrees with previous results demonstrating that suppression of TNF-secreting tumor cells requires inflammatory cells (3, 4).

To take advantage of the various knockout strains available in the C57BL/6 genetic background, a C57BL/6 tumor model was established. The C57BL/6 tumor cell line 8101 caused progressively growing tumors in Wt C57BL/6 mice (Fig. 2, left). 8101 was transfected to secrete human TNF; the resulting tumor cell line was referred to as 8101-TNF. The decision to use human TNF in a murine model was based on the previous finding that secretion of human TNF suppressed the growth of a murine tumor in nude mice (5). In addition, since human TNF binds only murine TNF-RI (8), TNF-RI^−/− mice can be used to show that suppression of tumor growth was caused by human TNF secreted by the tumor cells. The growth of 8101-TNF tumor cells was significantly suppressed in Wt mice (Fig. 2, left). The suppression was caused by TNF secreted by the tumor cells since it was abrogated in TNF-RI^−/− mice (Fig. 2, left) but not due to a direct effect of TNF on the tumor cells because 8101-TNF and 8101 grew at a similar rate in culture (Fig. 2, right). Growing tumors from TNF-RI^−/− mice were reisolated and found to secrete as much TNF as cells that had not been injected into mice (data not shown). Thus, TNF secretion suppressed the growth of the otherwise progressively growing C57BL/6 tumor 8101.

The fact that TNF-secreting tumor cells are resistant to the direct cytotoxic effects of TNF (9, 10) suggested that TNF must act on host cells. To determine whether suppression requires TNF-RI expression on non-BM-derived cells, BM chimeric mice were generated by transferring Wt BM cells into sublethally (3 or 6 Gy) irradiated TNF-RI^−/− mice. Control chimeric mice were generated by reconstituting irradiated Wt mice with Wt BM cells (Wt → Wt) and irradiated TNF-RI^−/− mice with TNF-RI^−/− BM cells (TNF-RI^−/− → TNF-RI^−/−). One month after reconstitution, 40% of the peripheral blood cells in the chimeric mice were found to have originated from donor BM cells (data not shown). Fig. 3 shows that 8101-TNF caused tumors with similar kinetics in Wt → TNF-RI^−/− and TNF-RI^−/− → TNF-RI^−/− mice and that the Wt → Wt mice were significantly more resistant to 8101-TNF, developing tumors after prolonged delay. These results show that suppression required TNF-RI expression on non-BM-derived host cells.

IFN-γ^−/− mice were used to determine whether IFN-γ was required for the suppression of TNF-secreting tumor cells. 8101-TNF tumor cells were suppressed in Wt mice as previously observed but grew progressively and at similar rates in IFN-γ^−/− mice and TNF-RI^−/− mice (Fig. 4,A), indicating a strict dependence on IFN-γ for TNF-induced suppression. To confirm the requirement for IFN-γ, IFN-γR^−/− mice were also tested for the ability to suppress TNF-secreting tumor cells. Since the IFN-γR^−/− mice were maintained in the 129 instead of the C57BL/6 genetic background, the Mc51-9 tumor cell line, derived from a syngeneic IFN-γR^−/− mouse, was infected to secrete TNF. Fig. 4 B shows that the resulting tumor cell line Mc51-9-TNF was suppressed in Wt mice but grew progressively in IFN-γR^−/− mice.

The IFN-γ signaling pathway diverges at the transcription factor Stat1 (19) and again at the downstream transcription factor IRF-1. Differential activation of these transcription factors leads to induction of different subsets of IFN-γ-regulated genes. To determine whether pathways involving these two transcription factors were required for TNF-induced suppression, Stat1^−/− mice and IRF-1^−/− mice were challenged with TNF-secreting tumor cells. The 9203 tumor cell line, derived from a Stat1^−/− mouse, was infected to secrete TNF, and the resulting 9203-TNF cells were suppressed in Wt mice but grew progressively in Stat1^−/− mice (Fig. 4,C). Similarly, 8101-TNF cells were suppressed in Wt mice but grew progressively in IRF-1^−/− mice (Fig. 4 D).

These results indicated that an intact host IFN-γ pathway was required for TNF to suppress tumor growth. The fact that Wt mice suppressed both the IFN-γR-deficient Mc51-9-TNF cells (Fig. 4,B) and the Stat1-deficient 9203-TNF cells (Fig. 4 C) showed that the suppression was due to IFN-γ acting on host cells but not on the TNF-secreting tumor cells. Moreover, the effects of IFN-γ must be mediated through Stat1 and IRF-1 and the subset of IFN-γ-inducible genes they regulate.

It is possible that the only function for TNF was to induce IFN-γ and that once induced IFN-γ may cause suppression without requiring the TNF pathway. To address this possibility, mice lacking both TNFRs were challenged with J558L tumor cells secreting IFN-γ (J558L-IFN-γ). In Wt mice, J558L grew progressively (Fig. 5,A) while J558L-IFN-γ was rejected (Fig. 5,B). In TNF-RI/II^−/− mice, J558L grew progressively in four of eight mice (Fig. 5,C) while J558L-IFN-γ was rejected by all seven mice (Fig. 5 D). The reason that some TNF-RI/II^−/− but not Wt mice rejected J558L is not clear, but this difference was also observed with three other independently generated tumor cell lines (data now shown). These results showed that locally released IFN-γ can suppress tumor growth in the absence of the TNF pathway.

IFN-γ is known to be produced only by hemopoietic cells such as NK cells, T cells, macrophages, B cells, and CD8α⁺ dendritic cells. To verify that BM cells are the source of IFN-γ, sublethally irradiated IFN-γ^−/− mice were reconstituted with Wt BM cells (Wt → IFN-γ^−/−) and tested for the ability to suppress TNF-secreting tumor cells. One month after reconstitution, flow cytometric analyses showed that 45–60% of peripheral blood cells were derived from donor BM cells (data not shown). Wt BM cells enabled IFN-γ^−/− mice to suppress the growth of 8101-TNF, delaying tumor development like the Wt → Wt mice (Fig. 6 A). Since 8101-TNF grew progressively in the IFN-γ^−/− → IFN-γ^−/− mice, it is unlikely that the suppression in the Wt → IFN-γ^−/− mice resulted from radiation damage, also called “tumor bed effect” (20).

To determine whether TNF directly activated BM cells to produce IFN-γ, sublethally irradiated IFN-γ^−/− mice were reconstituted with BM cells from TNF-RI^−/− mice (TNF-RI^−/− → IFN-γ^−/−) and challenged with 8101-TNF. BM-derived cells in these chimeric mice have the ability to produce IFN-γ but cannot be activated by TNF. The TNF-RI^−/− → IFN-γ^−/− mice developed tumors at the same rate as the IFN-γ^−/− → IFN-γ^−/− mice and significantly faster than the Wt → IFN-γ^−/− mice (Fig. 6 B), indicating that TNF acted directly on the BM cells. However, it appears that TNF did not induce the BM cells to secrete IFN-γ because when BM cells or splenocytes were cultured with 8101-TNF as a source of TNF for 2 days, there was no detectable IFN-γ in the culture supernatant (data not shown).

To identify the target of IFN-γ, sublethally irradiated IFN-γR^−/− mice were reconstituted with Wt BM cells (Wt → IFN-γR^−/−) and then challenged with Mc51-9-TNF cells. Flow cytometric analyses showed that up to 70% of the peripheral blood cells in the chimeric mice expressed IFN-γR (data not shown). Tumor development was significantly delayed in the Wt → IFN-γR^−/− mice compared with the IFN-γR^−/− → IFN-γR^−/− mice (Fig. 6,C). However, unlike the Wt → Wt mice that all rejected the tumor challenge (Fig. 6 C), five of eight Wt → IFN-γR^−/− mice developed tumors after a significant delay. This difference suggested that the number of Wt BM cells in the reconstituted IFN-γR^−/− mice may not have been enough to reject the tumor challenge and that a higher degree of reconstitution would have led to tumor rejection. It is also possible that IFN-γ acting on BM cells was sufficient to reduce the rate of tumor growth, but IFN-γ acting on both BM and non-BM cells was required for tumor rejection.

Rag-1^−/− and NK-depleted Rag-1^−/− mice were used to determine whether T, B, or NK cells produced the IFN-γ required for growth suppression. Rag-1^−/− mice were depleted of NK cells by treatment with rabbit anti-asialo GM1 antiserum 3 days before tumor challenge and every 3 days thereafter. Flow cytometric analyses of peripheral blood from treated mice indicated that the number of cells expressing the pan-NK marker DX-5 was reduced from 50 to 60% to 2 to 10% (data not shown). The cytolytic activity of splenocytes from the treated mice against the NK-sensitive target YAC-1 cell line was also significantly reduced (data not shown). Fig. 7 shows that mice lacking T cells, B cells, and the majority of NK cells remained capable of suppressing the growth of 8101-TNF. This result suggested that another cell type, such as macrophages or CD8α⁺ dendritic cells, may be the source of IFN-γ required for the suppression.

Earlier results suggested that TNF did not directly induce BM cells or splenocytes to produce IFN-γ. TNF may have instead activated stromal cells or inflammatory cells to produce secondary factors such as IL-12 and IL-18 that in turn induced IFN-γ production. To determine whether these two major IFN-γ-inducing cytokines were necessary for TNF to suppress tumor growth, IL-12^−/− (21) and IL-18^−/− (22) mice were challenged with 8101-TNF. Since IFN-γ production is further reduced in the absence of both IL-12 and IL-18 (22), IL-12^−/−IL-18^−/− mice were also generated and then challenged with 8101-TNF. Fig. 8 shows that mice lacking IL-12, IL-18, or both IL-12 and IL-18 remained fully capable of suppressing the TNF-secreting tumor cells.

The fact that TNF-mediated suppression was abrogated in mice lacking IFN-γ, IFN-γR, Stat1, or IRF-1suggested that IFN-γ was produced in response to the TNF-secreting tumor cells. To determine whether the IFN-γ message was produced within the lesion caused by 8101-TNF, total RNA was isolated from the small lesion produced at the site of tumor injection in Rag-1^−/− mice and from the large tumor in TNF-RI^−/− mice 26 days after tumor injection. The IFN-γ message was readily detected by RT-PCR in the 8101-TNF tumor isolated from TNF-RI^−/− mice and in an IFN-γ-secreting tumor isolated from Rag-1^−/− mice treated with anti-IFN-γ Ab (Fig. 9,A). Surprisingly, there was no detectable IFN-γ message in the nodule of the 8101-TNF tumor remaining in the Rag-1^−/− mice (Fig. 9 A); in some samples, no product was generated even after an additional 35 cycles with nested PCR primers (data not shown).

To confirm that TNF-mediated suppression did not require IFN-γ protein within the lesion, the XMG1.2 hybridoma cells producing neutralizing anti-IFN-γ Ab were transduced to secrete TNF (XMG-TNF). This provided the means to focus a high concentration of IFN-γ-neutralizing Ab in the vicinity of the TNF-secreting tumor cells. The amount of anti-IFN-γ Ab produced in 1 ml of culture medium by 10⁶ XMG-TNF hybridoma cells within 24 h was sufficient to neutralize 62 ng of recombinant murine IFN-γ (data not shown). Fig. 9 B shows that while the XMG1.2 hybridoma cells developed large tumors, the XMG-TNF hybridoma cells were strongly inhibited. These results along with the absence of the IFN-γ message in the suppressed tumor lesion firmly established that IFN-γ was not produced within the tumor.

To determine whether IFN-γ produced away from the tumor cells is essential for tumor inhibition, athymic nude mice were treated systemically with anti-IFN-γ Ab 3 days before tumor inoculation and weekly thereafter. We expected that systemic neutralization of IFN-γ would reproduce the phenotype of the IFN-γ^−/− mice and eliminate growth suppression, but treatment with IFN-γ-neutralizing Ab had no measurable effect over PBS treatment on tumor suppression 2 wk after tumor challenge (Fig. 9,C, upper right). The Ab treatment effectively neutralized IFN-γ because in the same mouse it enabled the progressive growth of IFN-γ-secreting 6132A tumor cells (Fig. 9,C, upper left), which is usually suppressed unless the secreted IFN-γ was neutralized (16). IFN-γ neutralization also enabled 8101 tumor cells engineered to secrete IFN-γ to grow as progressively as the parental 8101 tumor cells (data not shown). Interestingly, with continued treatment, 8101-TNF began to grow considerably faster in anti-IFN-γ Ab-treated mice than in PBS-treated mice, but still much slower than in TNF-RI^−/− mice. Since it was possible that extended Ab treatment was required to completely neutralize IFN-γ, the anti-IFN-γ Ab treatment was initiated 30 days before tumor challenge. However, this treatment failed to accelerate the growth rate of 8101-TNF in nude mice to that in the TNF-RI^−/− mice (Fig. 9 C, lower right).

In this study, the local effects of TNF on cancer cells were modeled using cancer cells secreting TNF. TNF-secreting cancer cells were insensitive to the cytotoxicity of TNF in vitro (10) and therefore unlikely to be the direct targets for TNF in vivo. Instead, TNF arrested tumor development through an indirect mechanism involving both BM- and non-BM-derived nonmalignant host cells. We propose that TNF stimulated non-BM-derived cells to release chemokines (23, 24) that attracted BM-derived cells to the tumor cells (Fig. 10). When the recruited BM cells, most likely monocytes, were then exposed to TNF in situ, an inflammatory environment was created that arrested tumor cell growth (25). BM-derived cells may also have been directly recruited by TNF to the tumor cells (26) and then activated in situ by factors produced by TNF-stimulated non-BM-derived cells.

TNFR expression on non-BM-derived stroma was necessary for TNF-induced tumor suppression, but the type of stromal cells targeted by TNF was not identified in this study. Because of the known antiangiogenic effects of TNF, it is tempting to suggest that TNF directly targeted endothelial cells and prevented tumor development by inhibiting tumor angiogenesis (27). Indeed, lesions containing TNF-producing tumor cells showed much less prominent vessel formation than tumors arising from Wt tumor cells (Fig. 1, E and F compared with Fig. 1, I and J). However, abundant patent capillaries were clearly visible in areas of dense cellular infiltrates surrounding the growth-inhibited TNF-secreting tumor cells (arrows in Fig. 1, G and H), suggesting that destruction of capillaries may not be the primary event preventing tumor development. It is important to consider the alternative possibility that the absence of a proangiogenic environment suppressed tumor development. Inhibition of tumor cells that produce large amounts of angiogenic factors such as vascular endothelial growth factor may limit the extent of angiogenesis by reducing the availability of angiogenic factors. The inflammatory environment associated with the TNF-secreting tumor cells may also prevent macrophage/monocytes from performing their essential role in normal and pathological angiogenesis (27, 28). Thus, TNF-induced tumor suppression may have resulted from an inability to induce tumor angiogenesis rather than active vessel destruction. In addition to the mechanisms outlined above, suppression also required an endogenous IFN-γ pathway involving IFN-γ, IFN-γR, Stat1, and IRF-1. We showed specifically that the BM-derived cells must originate from mice that have the IFN-γ and IFN-γR genes and most likely the Stat1 and IRF-1 genes. Interestingly, a connection between IFN-γ production or activity with TNF-induced tumor suppression could not be established by the various assays that were performed. First, TNF-mediated suppression occurred in the absence of detectable IFN-γ message in the growth-arrested lesions. Second, systemic treatment with IFN-γ-neutralizing Abs failed to block TNF-mediated suppression while effectively abrogating the suppression of three independently generated IFN-γ-producing cancer cell lines 6123A-IFN-γ, J558L-IFN-γ, and 8101-IFN-γ. It could be argued that the anti-IFN-γ Abs had not penetrated into the diffusion-limited space between TNF-secreting tumor cells, but this argument becomes untenable for explaining the continued growth arrest of TNF-producing hybridoma cells that also produced large amounts of neutralizing anti-IFN-γ Abs. However, we cannot exclude the possibility that a more complete neutralization of IFN-γ produced and functioning outside of the tumor site is required to block the inhibitory effects of TNF. Third, there was no detectable IFN-γ protein when 8101-TNF was cultured with BM cells and splenocytes as potential sources of IFN-γ (data not shown), suggesting that TNF cannot directly stimulate BM cells to produce IFN-γ. Fourth, tumor suppression occurred in the absence of T cells and B cells and the majority of NK cells that are potential sources of IFN-γ as well as the two major IFN-γ-inducing cytokines IL-12 and IL-18. Other cytokines such as IFNαβ (29) and IL-15 (30) can also induce IFN-γ, but were not examined in this study. Finally, TNF-mediated suppression did not require IFN-γ acting on or through the tumor cells (31) since tumor cells lacking IFN-γR or Stat1 were nevertheless suppressed. These results show that despite the requirement for an intact IFN-γ pathway, IFN-γ production or activity could not be measured. This is clearly different from tumor suppression resulting from measurable IFN-γ production at the tumor site that can be effectively blocked by treatment with anti-IFN-γ Ab, as modeled in this study by the IFN-γ-secreting tumor cells (Figs. 5 and 9). Thus, different mechanisms must be involved in the growth inhibition of the IFN-γ-secreting tumor cells and the TNF-secreting tumor cells. It appears that TNF does not mediate all of its antitumor activities through IFN-γ but rather utilize alternative mechanisms which facilitate the essential effects of IFN-γ.

This is the only study that we are aware of that definitively shows that an intact endogenous IFN-γ pathway is required for TNF to suppress tumor growth. Previously, Keller et al. (32) had shown that treatment with anti-IFN-γ antiserum abrogated the increased resistance of rats inoculated with heat-killed Corynebacterium parvum or Listeria monocytogenes against transplantable tumors. Although C. parvum and L. monocytogenes can prime macrophages to produce TNF, this study did not show that the enhanced resistance was strictly due to TNF production or that IFN-γ neutralization blocked an antitumor response induced by TNF. Doherty et al. (33) had shown that treatment with anti-IFN-γ antiserum blocked the ability of TNF to induce necrosis of established tumors in T cell-competent mice. However, IFN-γ neutralization also accelerated tumor growth when mice were not treated with exogenous TNF, thereby raising the possibility that IFN-γ neutralization did not block the antitumor effects of TNF but only increased the rate of tumor growth by interfering with a T cell-dependent antitumor response. It is also important to emphasize that the necrosis of established tumors studied by Doherty et al. (33) is quite different from the arrest of tumor development in our study. Although central hemorrhagic necrosis can be induced by TNF only when established tumors are at least 1 cm in average diameter (34, 35, 36) and the effectiveness of repeated injection is limited by development of tolerance (37), the persistent tumor suppression caused by sustained TNF release is not affected by these limitations.

The exact timing when the IFN-γ pathway was utilized could not be established in this study. Nevertheless, the results may be reconciled by proposing that the IFN-γ pathway is important long before tumor challenge for the differentiation/development of a BM-derived cell that later responds to TNF or factors induced by TNF and mediates growth arrest. IFN-γ neutralization may have had little or no effect on growth suppression in Wt mice because the putative cell had already differentiated and/or developed and become fully capable of mediating tumor growth arrest. That IFN-γ may play such a role is consistent with the recent observation that IFN-γ drives human monocyte differentiation into macrophages instead of dendritic cells (38). The putative cell is likely to have a long life span because continuous treatment with IFN-γ-neutralizing Ab for nearly 2 mo did not completely abrogate TNF-induced suppression. Macrophages are BM-derived cells that have powerful antitumor activities (39). So far, however, there has been no reported defect in macrophage development in IFN-γ^−/−, IFN-γR^−/−, Stat1^−/−, and IRF-1^−/− mice. IFN-γ may direct macrophages to polarize into specific functional subsets analogous to IFN-γ directing CD4⁺ T cell polarization to the Th1 subset. Mills et al. (40) recently showed that mice with a tendency to develop Th1 or Th2 T cells also develop macrophages with a corresponding profile characterized by NO and TGF-β1 production. A transcription factor that may be critical for determining the macrophage phenotype is IFN consensus sequence binding protein (ICSBP), an IFN-γ-inducible transcription factor specifically expressed in hemopoietic cells. ICSBP is required for the differentiation of myeloid progenitors into mature macrophages (41) and, in combination with IRF-1 and PU.1, regulates a number of IFN-γ-responsive genes. It is tempting to speculate that ICSBP may direct mature macrophages to differentiate further into a cell type that is required for TNF to suppress tumor growth. Additional studies using the ICSBP^−/− mice and other knockout mice with defects in the endogenous IFN-γ pathway may elucidate the role of the endogenous IFN-γ pathway and possibly help to define a novel type or characteristic of effector cells needed for TNF to suppress tumor development.

We thank Dr. Arturo Zychlinsky for providing the IL-18-deficient mice and Drs. Donald A. Rowley and Yang-Xin Fu for critical reading of this manuscript.