An Anti–Transforming Growth Factor β Antibody Suppresses Metastasis via Cooperative Effects on Multiple Cell Compartments

The past decade has seen a dramatic increase in the number of molecularly targeted therapeutics entering the clinic for the treatment of cancer. Such therapeutics generally either target driving oncogenic lesions within the tumor (e.g., Herceptin) or interfere with the establishment or maintenance of a tumor-permissive stroma (e.g., Avastin). Transforming growth factor β (TGF-β)s are pleiotropic growth factors whose overexpression by many advanced tumors correlates with metastasis and poor prognosis (1–4). Emerging evidence suggests that TGF-βs are important mediators of the dynamic and reciprocal interactions between the tumor parenchyma and cells of the tumor microenvironment (5), so targeting this pathway offers the attractive possibility of affecting both compartments simultaneously.

The complex biology of the TGF-βs poses challenges for the development of effective therapeutic strategies however. In the normal adult animal, TGF-βs play important roles in maintenance of homeostasis, particularly in the immune system, and in orchestration of the organismal response to injury (6, 7). In epithelial carcinogenesis, there is a wealth of clinical and preclinical data supporting the hypothesis that TGF-βs generally play a tumor suppressor role in the early stages of the process and only switch to a tumor-promoting role later in disease progression (1, 2). Tumor cells and stromal cells are critical targets at both stages. Early in the process, loss of TGF-β response in an initiated epithelial cell can promote the early stages of tumorigenesis through impairment of tumor-cell autonomous suppressor mechanisms, such as growth inhibition, differentiation, apoptosis, and maintenance of genomic stability (2). However, loss of TGF-β response in fibroblasts or T cells can also promote tumorigenesis in overlying epithelia, due at least in part to ectopic secretion of tumor-promoting growth factors and cytokines (8, 9). In contrast in the later stages of the carcinogenic process, overexpression of TGF-β can promote tumor progression by enhancing migration, invasion, and survival of the tumor cells and by generating a tumor-promoting stroma, primarily through enhanced angiogenesis and suppression of immune surveillance (1, 2). The therapeutic challenge is to selectively ablate the tumor promoting effects of TGF-β while sparing desirable effects on normal homeostasis and tumor suppression.

Despite this complex biology, initial preclinical studies showed considerable efficacy of antibody-like TGF-β antagonists in suppressing metastasis, with few of the anticipated toxicities such as increased spontaneous tumorigenesis or autoimmune manifestations (10, 11). Anti–TGF-β antibodies also completely inhibited tumor recurrence in a fibrosarcoma model and reduced metastases in a colon cancer model (12). As a result of these and other studies, a number of TGF-β pathway antagonists are in late preclinical or early clinical development for treatment of patients with advanced cancer (13, 14). Rational development of these agents will require a detailed understanding of which cell compartments and which biological processes are affected by TGF-β antagonism in vivo. Using the 4T1 transplantable mouse model of metastatic breast cancer (15), here, we show that treatment with an anti–TGF-β antibody suppresses metastasis through a combination of effects on multiple interacting cell types. Most of these effects occur locally at the tumor site and cooperate to enhance CD8+ T-cell activity. We propose that this locally distributed mechanism of action may underlie the surprising lack of toxicity of this class of TGF-β antagonist (10, 16).

Cell culture and reagents. The 4T1 cell line was provided by Dr. Fred Miller at the Barbara Ann Karmanos Cancer Institute, Detroit, MI, and cultured as previously described (17). TGF-β1 and tumor necrosis factor (TNF)-α were purchased from R&D Systems. Staurosporine was purchased from Sigma-Aldrich. Anti-CD8 monoclonal antibody (clone 2.43 rat IgG2b; American Type Culture Collection TIB210) was generated by Harlan Bioproducts. Antiasialo GM1 was purchased from Wako. The anti–TGF-β murine monoclonal antibody, 1D11, which neutralizes all three isoforms of TGF-β (18), and an isotype-matched IgG1 monoclonal antibody, 13C4, which was raised against Shigella toxin and serves as a control, were provided by Genzyme Corp.

In vivo tumorigenicity and metastasis assay. All animals were maintained according to the National Cancer Institute's Animal Care and Use Committee guidelines, under approved animal study protocols. For the spontaneous metastasis format, 4 × 10⁴ 4T1 cells in 40 μL of PBS were inoculated into the surgically exposed left thoracic mammary fat pad of anesthetized 7-week-old female BALB/cANCr mice. After inoculation, the mice were randomized into two treatment groups, with 10 to 15 animals per group. Anti–TGF-β antibody (1D11) or control antibody (13C4) were administered thrice per week i.p at 5 mg/Kg, starting 1 d after cell inoculation. Primary tumors were surgically excised on day 10. Mice were euthanized by carbon dioxide narcosis on day 28. Macroscopic quantitation of metastases was performed by counting the number of nodules on the surface of the inflated lung. For microscopic quantitation of lung metastases, each lobe of the lung was processed for H&E staining and evaluated by a board-certified veterinary pathologist (MRA). For the experimental metastasis format, 4,500 4T1 cells were injected into the tail vein of 7-week-old female BALB/c mice. In select experiments, mice were injected i.p. with rat anti-CD8 antibody (clone 2.43; 0.5 mg per mouse) or rabbit anti–asialo-GM1 antibody (30 μL per mouse) on days −4, −3, −2, +3, +10, and +17 relative to the tumor injection (on day 0) to deplete CD8+ or natural killer (NK) cells. Depletion was confirmed to be >90% by fluorescence-activated cell sorting (FACS) analysis of blood samples from the treated mice. Lungs were harvested on day 21 and analyzed as above.

Quantitative reverse-transcription PCR. The real-time quantitative PCR was performed using the iCycler iQ Real-time PCR Detection System (Bio-Rad) with SYBR green dye (Stratagene). First-strand cDNA was prepared from total RNA using a SuperScript III first strand synthesis kit (Invitrogen). The quantitative reverse transcription-PCR (RTQ-PCR) was done in triplicate. Primer details are given in Supplementary Data.

Transfection of tumor cells. A COOH terminally Myc-tagged dominant-negative type II TGF-β receptor (DNR), comprising nucleotides 1 to 656 of the human sequence, was ligated into pB vector (a kind gift from Dr. Jan Pinkas at Genzyme, Framingham, MA) and transfected into 4T1 cells using Lipofectamine 2000 transfection reagent (Invitrogen). Stable transfectants were selected and DNR expression was confirmed by promoter-reporter assay and immunoblotting analysis. To minimize the effect of clonal variation, we pooled populations of transfected clones.

In vitro apoptosis assay. Apoptosis of 4T1 cells in vitro was quantitated using the Cell Death Detection ELISA assay kit (Roche), or by FACS detection using the Annexin V-FITC apoptosis kit (BD Biosciences PharMingen). Briefly, 4T1 cells were treated with or without TGF-β1 (5 ng/mL) for 4 d in DMEM containing 3.5% fetal bovine serum (FBS), and then TNF-α (25 ng/mL) was added to select wells for the final 2 d before harvesting for assessment of apoptosis as above.

Conditioned medium collection, tissue extraction, and ELISA assays. Serum-free conditioned medium was prepared from near confluent cell cultures and assayed for TGF-βs, after acid activation of the medium. TGF-β levels in mammary glands and tumors were determined after acid ethanol extraction as previously described (19). Quantitation of TGF-β1, TGF-β2, and TNFRSF1 was performed by using Quantikine ELISA kits (R&D Systems), whereas TGF-β3 was quantitated by ELISA as described (19).

Immunocytochemistry and immunofluorescence. Active and total TGF-β in sections from metastasis-bearing lungs were visualized by immunofluorescence on fresh frozen tissue cryosections essentially as described (20). Assessment of proliferation, apoptosis, blood vessel density, and T-cell infiltration was done on formalin-fixed paraffin-embedded sections, immunostained with anti–phospho-histone H3 (06-570; Upstate), ApopTag Peroxidase In situ Apoptosis kit (Chemicon), anti-CD3 (A0452; Dako), or anti-CD31 (sc1506; Santa Cruz). Images of the entire section were acquired using an Aperio Scanscope (model T3). Individual images were manually adjusted for luminosity to compensate for variable counterstain intensity, and image analysis was conducted using Image-Pro Plus v 5.1 (Media Cybernetics). A set of standardized color segmentation, morphologic filters, and area calibration files was used to quantitate the number of metastases, the individual metastasis area, and the number of positive cells (marker dependent) per metastasis. A minimum of one randomly selected high power field (×200) was assessed for each metastasis within a given lung section. For primary tumors, a minimum of 10 random high power fields were assessed for each tumor. All data were exported into Excel for further analysis.

Immunophenotyping and FACS analysis. Primary tumors or tumor-bearing lungs were excised, minced with scissors to a fine slurry in 15 mL per lung of digestion buffer (RPMI 1640, 5% FBS, 1 mg/mL collagenase, and 30 μg/mL DNase), and digested at 37°C for 30 min. The suspension was pelleted, resuspended, and spun through a discontinuous 40%:70% Percoll gradient to enrich for leukocytes. Single-cell suspensions were stained and analyzed on FACSCalibur (Becton Dickinson) using CellQuest (BD BioSciences) software. For direct staining to determine the phenotype of lymphocyte populations, the following conjugated antibodies were purchased from BD BioSciences or eBiosciences: CD45PerCP, CD4 APC, CD8APC, CD3PerCP, DX-5 APC, NKG2D PE, CD25 PE, CD69 FITC, and CD122PE. For determination of Rae-1γ levels on tumor cells, 4T1 cells were harvested after 4 d incubation with or without TGF-β1 (5 ng/mL), then fixed and stained with phycoerythrin rat anti-mouse Rae-1γ (BD Biosciences PharMingen) and analyzed by flow cytometry.

Statistical analysis. Unpaired parametric Student′s t test or nonparametric Mann-Whitney U tests were used to analyze the data, unless otherwise indicated in the text.

4T1 cells retain some responsiveness to TGF-β in vitro and make moderately elevated levels of TGF-β in vivo. Like many aggressive cancer cell lines, 4T1 cells in culture were, at best, only weakly growth inhibited by TGF-β, but they did respond to TGF-β by phosphorylating Smad2, and inducing Smad-dependent reporter gene expression (Supplementary Fig. S1). 4T1 cells in culture secreted predominantly TGF-β1 (28 ng per 10⁶ cells per 48 h), some TGF-β3 (6.9 ng per 10⁶ cells per 48 h), and low to no TGF-β2 (<0.4 ng per 10⁶ cells per 48 h). In vivo, 4T1 tumors had TGF-β1 protein levels that were ∼3-fold elevated in the primary tumor when compared with the normal mammary gland [220 ± 24 versus 79 ± 16 ng TGF-β1 per gram wet weight of tissue (n = 5); P < 0.001]. Immunostaining also showed mildly elevated total TGF-β in 4T1 lung metastases when compared with the surrounding lung parenchyma, although without any major increases in the active TGF-β fraction (Supplementary Fig. S2). TGF-β pathway activation within metastases, as assessed by immunostaining for phosphoSmad2, was evident but heterogeneous (Supplementary Fig. S2). Finally, in common with some studies on human breast cancers (21, 22), we did not see significantly elevated circulating levels of TGF-β in the plasma of tumor-bearing mice when compared with age- and sex-matched normal mice [2.5 ± 0.8 ng/mL (n = 5) versus 1.7 ± 0.1 ng/mL (n = 5); P = not significant (NS)]. Thus, the 4T1 model has many of the TGF-β–related properties that we would expect to find in a human breast tumor caught before end stage, and does not have a particularly exaggerated TGF-β production or response.

Suppression of 4T1 cell metastasis by anti–TGF-β antibody treatment is dependent on both the innate and the adaptive arms of the immune system. In the 4T1 model, after orthotopic implantation of tumor into the #2 mammary fat pad, large numbers of small lung metastases were seen (typically 30–80 per mouse), although if metastasis was initiated by injection of 4T1 cells into the tail vein, we observed smaller numbers of metastases (6–12 per mouse) that were of larger size. However, regardless of experimental format, in multiple independent experiments, we found that treatment with the 1D11 anti–TGF-β antibody caused a statistically significant 50% to 60% reduction in the number of macroscopic and of histologically confirmed lung metastases (Fig. 1A–B). Mice treated with the 1D11 antibody were healthier than mice treated with the 13C4 control antibody, as judged by their higher body weight, and they showed less infiltration of tumor cells into the thoracic lymph nodes (Supplementary Fig. S3). Histologically, there were no obvious differences between tumors from the two treatment groups (data not shown).

Because the tail vein injection format introduces tumor cells directly into the vasculature, the similarity of the results in the two experimental formats suggests that 1D11 acts primarily on a step in the metastasic cascade that lies downstream of invasion and intravasation. TGF-β has previously been shown to promote the extravasation step in the mouse mammary tumor virus–Neu transgenic model of metastatic breast cancer (23). In the 4T1 model, we found that the majority of metastases were extravascular regardless of treatment [85% ± 7% in control antibody-treated mice (n = 11) versus 90% ± 11% in 1D11-treated mice (n = 12); P = NS], suggesting that if extravasation is a TGF-β–dependent step in this model, it is either not rate limiting or not accessible to the antibody.

Depending on the model system used, studies with other TGF-β antagonists have implicated effects either on angiogenesis or on immune surveillance as mechanisms that may underlie efficacy (12, 24–29). By immunohistochemical approaches, we saw evidence in support of both mechanisms in the 4T1 model. Treatment with 1D11 was associated with a statistically significant decrease in microvessel density in the primary tumor, although not in the lung metastases, and with significantly increased T-cell infiltration at both sites (Fig. 1C and D).

TGF-β has potent suppressive effects on the generation and/or effector function of most immune cells (30). To address the potential role of immune-mediated mechanisms in the current model, we depleted mice of CD8+ cells or NK cells. Depletion of CD8+ cells had no effect on the number of lung metastases in the absence of 1D11 treatment, suggesting that there is no effective T-cell mediated immune surveillance in this model in the untreated state (Fig. 2A,, compare columns 1 and 3). In contrast, depletion of NK cells resulted in a dramatic increase in the number of lung metastases (Fig. 2B,, compare columns 1 and 3), suggesting that the innate arm of the immune surveillance system is quite active. Most of the efficacy of 1D11 treatment was lost on depletion of CD8+ cells (Fig. 2A; compare columns 1 versus 2 with 3 versus 4), whereas treatment with 1D11 was still capable of suppressing metastasis in the NK-depleted mice, although to a lesser extent than in intact mice (Fig. 2B). Thus 1D11 treatment seems to unmask effective immune surveillance. Interestingly, blockade of TGF-β response in the 4T1 tumor cells using a DNR also ablated most of the metastasis-suppressing effect of 1D11 (Fig. 2C). Combining data from multiple experiments, we can calculate that ∼75% of the antibody efficacy is dependent on the presence of CD8+ cells, whereas ∼35% is dependent on the presence of NK cells, and ∼60% requires an intact TGF-β response in the tumor cell itself (Fig. 2D). Because the sum of these effects is >100%, the data suggest that treatment with 1D11 enhances antitumor immune responses by a cooperative mechanism that involves several cellular compartments, including the CD8+ T cells, NK cells, and the tumor itself.

Effects of anti–TGF-β antibody treatment on immune cell number and function are only seen locally at the primary tumor or metastatic site. Enhanced immune surveillance is frequently associated with increased numbers of tumor-specific cytotoxic T-cells (CTL) in the spleen. However, we were unable to show any increase in splenic CTL activity against 4T1 cells after treatment of the mice with 1D11 in repeated experiments, using multiple different experimental formats (data not shown). In addition, there were no significant effects of 1D11 treatment on relative numbers or activation status of immune cells in the spleen (data not shown). In respect to the lack of splenic CTL activity, our results differed from those of others, using different antagonists in this and other models (27–29). We reasoned that 1D11 might be primarily affecting immune cell infiltration and function specifically at the primary tumor or metastatic sites. We were unable to recover sufficient viable immune cells from the primary tumor for functional analysis. However, by RTQ-PCR of tumor-derived RNA, we showed that 1D11 treatment was associated with a 2-fold increase in CD8b mRNA, consistent with an increased infiltration of the tumor by CD8+ T cells. Furthermore, the ratio of perforin to CD8b and granzyme B to CD8b mRNAs in the primary tumor was also increased ∼2-fold, consistent with enhanced expression of cytotoxic effector molecules in the infiltrating CD8+ cells (Fig. 3A). These data are consistent with localized increases in CTL number and activity at the primary tumor site.

To address the nature and activation status of immune cells at the metastatic site, we prepared cell suspensions from the lungs of tumor-bearing mice treated with 1D11 or control antibody 13C4, and analyzed the infiltrating immune cells by FACS. We found no statistically significant differences in the percentage of B cells (CD19+) or myeloid cells (Gr-1+ and CD11b+) in the leukocyte gate, and no change in the ratio of CD4+ to CD8+ T cells (CD3+ gate). However, there was a significant increase in the percentage of NK cells (CD3-DX5+) in 1D11-treated lungs (Fig. 3B). We also examined expression of a number of markers associated with immune cell activation or function. Although there was no increase in the CD4+CD25+ fraction that includes the FoxP3+ regulatory T cells (Supplementary Fig. S4) and no change in expression of CD25 or CD69 early activation markers in either CD4+ or CD8+ T-cell subsets (data not shown), there was a mild but significant increase in the fraction of CD8+ cells expressing CD122 [interleukin (IL)-2/IL-15Rβ] in the lungs, and a trend toward increased expression of TNF-related apoptosis-inducing ligand in the CD8+ compartment (Supplementary Fig. S4). NKG2D, which is an activating receptor on NK cells and a coactivator on CD8+ T cells, has previously been shown to be down-regulated on NK cells in patients whose tumors overexpress TGF-β (31). We found that 1D11 treatment significantly increased NKG2D expression in the CD8+ T-cell compartment, although having no effect on NKG2D expression on NK cells where it was already high (Fig. 3C).

TGF-β antagonism increases tumor cell susceptibility to recognition and killing by the immune system. Because effects of 1D11 treatment on the immune cell compartment were relatively mild and seemed unlikely alone to account for the therapeutic efficacy, we next asked whether 1D11 might affect the tumor cell itself in such a way as to further enhance immune surveillance. Blocking the response to TGF-β specifically in the tumor cell compartment using a DNR caused a decrease in metastatic efficiency (Fig. 4A,, compare columns 1 and 3), suggesting that endogenous TGF-β acts on the tumor cell to promote metastasis. Part of this suppressive effect was lost if the mice were depleted of CD8+ T cells (Fig. 4A , compare columns 3 and 4), suggesting that a direct effect of TGF-β on the tumor cell itself must contribute to the evasion of effective immune surveillance.

TGF-β has been shown to down-regulate expression of MHC components in other systems (32), which should render the tumor cell less visible to the immune system. However, we did not find a significant effect of TGF-β or 1D11 on basal or IFN-γ–induced expression of the MHC class I component H2Kd in vitro or in vivo as assessed by RTQ-PCR and FACS analysis (data not shown) in the 4T1 model. Because we had shown that 1D11 treatment up-regulates expression of the coactivating receptor NKG2D on CD8+ T cells, we looked for effects of 1D11 on the expression of Rae-1γ, the cognate NKG2D ligand, on the tumor cells. Rae-1γ is a stress-induced molecule that alerts the immune system to the presence of an incipient tumor (33). TGF-β treatment significantly down-regulated Rae-1γ mRNA and protein in 4T1 cells in vitro (Fig. 4B), and conversely, 1D11 treatment up-regulated Rae-1γ mRNA in 4T1 primary tumors in vivo (Fig. 4C). Rae-1γ expression was also enhanced on the surface of tumor cells recovered from 1D11-treated mice as determined by FACS analysis (Fig. 4C). This increased expression of an NKG2D ligand on the tumor would be expected to enhance immune cell activation, particularly in the context of up-regulated NKG2D expression on the immune cells. Indeed, such an effect has been shown in a glioma model system using antisense strategies for TGF-β blockade (34).

TGF-β has complex effects on cell survival, acting either as a proapoptotic factor or as a prosurvival factor, depending on the cell type and biological context (35). TGF-β treatment had no effect on survival of 4T1 cells in response to serum deprivation (data not shown), suggesting that there was no generalized prosurvival effect of TGF-β on the intrinsic cell death pathway and downstream apoptotic machinery in this cell type. However, immune cells kill in part by activation of death receptors on the target cell, so we next looked for effects on the extrinsic death pathway. Expression of Fas and Fas ligand on 4T1 tumor cells was unaffected by TGF-β in vitro or 1D11 in vivo (data not shown). However, we found that TGF-β treatment significantly down-regulated expression of the TNF-α death receptor, TNFRSF1a, on 4T1 cells in vitro at the mRNA and protein level (Fig. 5A). Conversely, 1D11 treatment up-regulated TNFRSF1a mRNA in 4T1 tumors in vivo (Fig. 5B). Down-regulation of TNFRSR1a by TGF-β in vitro was associated with enhanced resistance to cell killing by TNF-α (Fig. 5C), consistent with the possibility that 1D11 treatment makes tumor cells more susceptible to death receptor–mediated cell killing in vivo. The inhibitory effect of TGF-β on expression of both TNFRSF1a and Rae-1γ was not restricted to the 4T1 line but was also seen in EMT6 (mouse), MDA MB231 (human), and Hs578T (human) breast cancer cell lines (Supplementary Fig. S5).

Most preclinical studies using TGF-β antagonists to suppress tumor progression have found surprisingly little toxicity (13, 14), so it is of considerable interest to understand how such agents can selectively antagonize the undesirable effects of TGF-β on tumor progression. Our data in the 4T1 breast cancer model suggest that anti–TGF-β antibody therapy reduces the number of lung metastases primarily through the integration of many biological effects on multiple cellular compartments. Quantitatively, the most important outcome of these effects is the unmasking of effective CD8+ T-cell activity against the tumor. Local recruitment and/or activation of cells of both the innate and adaptive arms of the immune system were increased by 1D11 treatment, and tumor cell visibility to the immune system and susceptibility to cell killing were also enhanced. Additionally, angiogenesis was mildly suppressed in the treated tumors, as has also been observed with small molecule receptor kinase antagonists (27). Because the effects of the anti–TGF-β antibody are distributed over multiple cellular compartments and molecular effectors, we refer to this therapeutic mechanism as “death by a thousand cuts” (see model in Fig. 6). The distributed aspect of this mechanism may be critical for a good therapeutic index and has a number of important implications as discussed below.

The overall importance of immune mechanisms to the therapeutic efficacy of TGF-β antagonists is highlighted by a number of recent studies (12, 26–29, 36, 37). However, the relative contributions of the different immune compartments to the therapeutic efficacy of TGF-β antagonism seem to vary from tumor to tumor. For example, in the 4T1 model, we found that anti–TGF-β treatment primarily activated CD8-mediated antitumor immunity, showing only a relatively small enhancement of NK activity. However, in a syngeneic rat glioma model, the therapeutic efficacy of TGF-β antagonism using a soluble TGF-β receptor showed a much greater dependency on NK cells (37), and nude mice, which have higher than normal levels of NK cells, may show a particularly exaggerated contribution from the innate arm (25). Similarly, in models of renal and prostate cancer, the 1D11 antibody was recently shown to decrease the number of FoxP3+ regulatory T cells in lung metastases (38), but there was no evidence for 1D11-induced down-regulation of FoxP3+ tumor-infiltrating lymphocytes in the 4T1 breast model.⁸

TGF-βs have long been known to have potent suppressive effects on the immune system (7). However, integrating our data with published studies, it is clear that the extent or the nature of TGF-β antagonism can dramatically affect the immunologic outcome. Near total blockade of TGF-β response specifically in T cells caused increased resistance to tumorigenesis and metastasis, associated with robust antitumor immune responses (39–42). However, this approach also results in the development of widespread autoimmune manifestations (43), such as are seen in the TGF-β1 null mouse (44), suggesting that such profound TGF-β blockade in the T-cell compartment may not be desirable. In contrast, the enhanced antitumor immunity that we saw in the current study seemed to be the local result of synergism between many smaller magnitude effects of anti–TGF-β antibodies on multiple cellular compartments, including the tumor cells themselves, and treatment was not associated with a generalized immune activation. Unlike investigators who have used small molecule TGF-β receptor kinase inhibitors (26, 27, 29), we did not detect increases in antitumor CTL activity or in activation status of splenocytes from mice treated with 1D11. Because one such study also used the 4T1 model (27), the data suggest there may be differences in the mode of action of these two types of agent, possibly reflecting a more extensive TGF-β blockade with the kinase inhibitors.

Although the locally distributed mode of action of anti–TGF-β antibodies in the vicinity of the tumor may be critical to their lack of immunotoxicity, it also poses practical problems. The development of useful biomarkers of effective TGF-β antagonism in vivo by this class of agent may be difficult because there is no major change in any one molecular marker in a readily accessible cell compartment. Furthermore, although the general principle of distributed action is likely to be broadly applicable, the specific cellular target and detailed nature of the molecular changes may vary between tumor types. For example, tumor-derived TGF-β was implicated in the down-regulation of NKG2D expression on NK cells in colon cancer and glioblastoma patients (31, 34). In contrast, in the 4T1 model, we found that NKG2D expression on NK cells at the metastatic site was unaffected by 1D11 treatment, whereas expression of NKG2D on CD8+ T cells was significantly enhanced. TGF-β antibodies may therefore represent a class of therapeutics for which a very broad net will have to be cast to capture a constellation of potential biomarkers on several different cell types, only a fraction of which may be affected in any given patient. Emerging technologies for efficient assay multiplexing will be useful in this regard.

In the present work, we have focused on mechanisms underlying the effect of anti–TGF-β treatment on metastasis number. In a companion study, we show that anti–TGF-β treatment decreases the size of lung metastases by a distinct immune-mediated mechanism, in which anti–TGF-β suppresses the ability of the tumor to subvert CD8+ T cells into making IL-17, a tumor cell survival factor (45). Both effects of anti–TGF-β combine to reduce overall metastatic efficiency. It is of interest that TGF-β antibodies were as efficacious as the conventional chemotherapeutic paclitaxel in suppressing metastasis in the 4T1 model (46), despite the fact that this model does not show a particularly high level of TGF-β production or activation in the tumor or circulation. The data suggest that even patients with relatively low levels of TGF-β expression in their tumors may have compromised antitumor immune responses that could be at least partially restored by TGF-β antagonism. Combining TGF-β antagonists with other therapeutic approaches, such as nanomicellar drug delivery (47), is likely to give further benefit.

The TGF-βs arose relatively late in evolution at a time when organismal complexity was considerably increased (48). Because essentially every cell type in the body can respond to TGF-β in some way, TGF-βs may play particularly important roles in higher order functional integration between different cellular compartments. Tonic, low-level TGF-β signaling between organ parenchyma and stroma is clearly important for maintenance of homeostasis in normal tissues, as local inactivation of TGF-β pathway components in fibroblasts or immune cells results in the generation of a procarcinogenic microenvironment (8, 9). However, local overexpression of TGF-β, as is seen in many advanced human tumors, also generates a tumor-promoting microenvironment, albeit of a different form (1, 2). Thus, this regulatory system seems to be very delicately poised. We propose that the main effect of anti–TGF-β antibody treatment is to normalize the distortions in this complex intercellular communication network that are caused by tumor-induced overexpression of TGF-β, thereby restoring the microenvironment to a more tumor suppressive state.

No potential conflicts of interest were disclosed.

Grant support: Intramural Research Program of the NIH, National Cancer Institute, Center for Cancer Research, Z01 BC 005785 (L.M. Wakefield), and with federal funds from the National Cancer Institute, NIH, under contract N01-CO-12400 (MRA). The National Cancer Institute has a Cooperative Research and Development Agreement with Genzyme Corporation, Framingham, MA.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.