Targeted Bioactivity of Membrane-Anchored TNF by an Antibody-Derived TNF Fusion Protein¹

Tumor necrosis factor is expressed by numerous cell types as a 26-kDa, type II transmembrane TNF (memTNF)³ and as a soluble 17-kDa form (sTNF), which results from proteolytic cleavage of memTNF (1, 2). TNF signals are transduced by two different plasma membrane receptors, TNFRSF1A (TNF receptor superfamily, member 1A; p60; TNF-R1) and TNFRSF1B (TNF receptor superfamily, member 1B; p80; TNF-R2) (3). Although most cell lines and primary tissues coexpress both receptor types, cellular responses to the soluble 17-kDa component are mediated by the interaction with TNF-R1 (4). In addition to its role as a major mediator of inflammatory responses and immune cell proliferation, sTNF is also involved in the lysis of infected or transformed cells (5). In contrast to the systemic activity of sTNF, memTNF exerts its cytotoxic effects in a locally restricted manner, because it depends on direct cell-to-cell contact (6). After appropriate stimulation, human monocytes, macrophages, and CD8⁺ T cells present membrane-anchored TNF as a key component in their cytolytic pathways (7, 8). Although memTNF acts predominantly via TNF-R2, it can potently trigger both receptor types (9), leading to a positive cooperative receptor cross-talk (10). This agonistic triggering of TNF-R2 is not only a simple additive effect to TNF-R1 signaling, but can even cause a shift in the cellular response pattern to TNF. Tumor cells that are fully resistant to sTNF-TNF-R1-mediated cytotoxicity become highly vulnerable and are killed through TNF-R2-dependent activation upon presentation of memTNF (9). However, this superior capacity of full TNF-R2-activation to initiate cellular responses is not restricted to direct cytotoxicity against tumor cells. TNF-R2 also confers typical TNF responses in local inflammatory processes (6). TNF-R2 binding leads to significantly higher expression levels of T lymphocyte HLA-DR proteins or endothelial tissue factor (11), stimulates macrophages more efficiently to defend invading microorganisms (12), and optimizes activation of the respiratory burst of polymorphonuclear leukocytes (PMN) (13).

Considering the broad range of the multiple memTNF-dependent effector mechanisms, the use of its bioactivity as an anticancer agent is a promising therapeutic approach. However, minimizing TNF toxicity to normal tissue would require both the prevention of systemic soluble TNF-like action and the locally restricted enrichment of memTNF activity to the tumor compartment. Concerning epithelial cancers, the tumor compartment consists of two discrete, but interdependent, components, the malignant cells and the tumor stroma. The predominant cells in the tumor stroma are nontransformed fibroblasts, expressing the genetically stable fibroblast-activating protein (FAP) in >90% of all cases (14). The FAP-positive stromal cells commonly contribute 50–90% of the tumor mass and are located in close vicinity to the endothelial cells of the tumor capillaries and surround the tumor nodules. Focusing the bioactivity of memTNF to the FAP-positive tumor stroma should therefore promote the antitumor efficacy of the cytokine at several levels. 1) Reflecting the fact that endothelial cells are the major target for TNF-mediated hemorrhagic tumor necrosis, the FAP-dependent enrichment of TNF in a memTNF-like manner should initiate enhancement of endothelial tissue factor production by cooperative TNF-R signaling. 2) The TNF-initiated induction of adhesion molecules on the luminal surface of the tumor vessels should favor Ag-neighbored adherence of PMNs and trigger the release of reactive oxygen species, causing stromal tissue damage. 3) Furthermore, not only FAP-positive cells, but also the 4) malignant cells in their proximity, could activate autoproteolytic programs, even under conditions where a phenotypic switch of the cellular response pattern should have to be induced to become susceptible.

To direct these memTNF-induced cellular response programs to the FAP-positive tumor stroma, we established an Ab fusion protein that consists of the humanized anti-FAP Ab (15) and human TNF. The TNF sequence started at aa 79 of the precursor protein and was blocked through direct linkage to the IgG1 hinge region. The CH2/CH3 region of the human IgG1 Fc portion was replaced by the human TNF sequence to prevent nonspecific binding to FcRs. This type of linkage should provide a certain rigidity of the fusion protein, thus mimicking the membrane-anchored presentation of TNF via Ab-mediated specific binding to FAP. Furthermore, the design of the TNF compound as a dimer should minimize any FAP-independent systemic activity of sTNF, which physiologically signals as a compact trimer.

Our in vitro and in vivo data demonstrate that the fusion protein exhibits an effector profile identical with that of memTNF associated with up to 10-fold reduced toxicity compared with sTNF, proving that the Ab-linked TNF dimer is an optimal transmitter of memTNF bioactivity.

RPMI 1640 medium (supplemented with 10% (v/v) heat-inactivated FCS, penicillin (100 U/ml), streptomycin (0.1 mg/ml), and glutamine (0.3 mg/ml); all from Life Technologies, Karlsruhe, Germany) was used as standard medium. HT1080 FAP-transfected (HT1080 FAP) and mock-transfected (HT1080 par) cells were supplied by Boehringer Ingelheim (Biberach, Germany). WEHI-164 S (TNF-sensitive) and WEHI-164 R (TNF-resistant) cells were cultured in standard medium as previously described (16). Colo205 was obtained from American Type Culture Collection (Manassas, VA). HUVEC-c endothelial cell-supplemented growth medium was obtained from PromoCell (Heidelberg, Germany). Human recombinant TNF and soluble TNF-R1 were purchased from Genzyme (Neu-Isenburg, Germany). Rat anti-mouse CD11b (clone M1/70) and rat anti-mouse platelet endothelial cell adhesion molecule-1 (PECAM-1; clone MEC13.3) were purchased from BD PharMingen (San Diego, CA). Biotinylated goat anti-human IgG was a component of the Vectastain ABC kit (Vector Laboratories, Burlingame, CA), biotinylated rabbit anti-mouse F(ab)₂ and goat anti-rat IgG F(ab)₂ were obtained from Dianova (Hamburg, Germany). Infliximab was purchased from Centocor (Leiden, Netherlands), Etanercept from Whyeth Pharma (Munster, Germany), and murine anti-idiotypic anti-FAP from Boehringer Ingelheim. The anti-G250 mAb was provided by Prof. S. Warnaar (Pathology Department, University of Leiden, Leiden, The Netherlands).

RT-PCR of mRNA isolated from PBMC was performed as previously described (7). Amplification of the mature human TNF cDNA sequence was performed using the following primer pairs: 5′-ATCACCATGGTCTCATCTTCTCGAACCCCGAGTG-3′ and 5′-AGTTTCTAGATCACAGGGCAATGATCCCAAAGTAG-3′.

The cDNA sequence coding for the variable H (HC) and L (LC) chain sequences of the humanized anti-FAP Ab was amplified as previously described (7). The respective cDNA template of the humanized anti-FAP Ab was provided by Boehringer Ingelheim, and the following primer pairs were used: HC reverse, 5′-CAGGTTTAAACGCCGCCACCATGGACTGGACCTGGCGCGTGTTTTGC-3′; HC forward, 5′-CAGGGATCCACTCACCTGAGGAGACGGTGAC-3′; LC reverse, 5′-CAGGAAATTTCGCCGCCACCATGGAGACAGACACACTCCTGCTATGG-3′; and LC forward, 5′-CAGGGATCCACTCACGTTTTATTTCCACCTTGGTCCCCTG-3′.

A eukaryotic expression vector system (pREN) with either human constant IgG1 HC or κ LC sequences under the control of the elongation factor α promoter was used for recombinant protein expression. Variable HC and LC sequences were inserted by restriction digest (Pme/BamHI) into the corresponding pREN vector. The pREN HC vector was modified by inserting an NcoI site in-frame after the hinge region to insert the TNF cDNA fragment (anti-FAP-TNF HC). Methotrexate resistance was incorporated into the HC vector, and neomycin resistance was incorporated into the LC vector.

Stable transfected CHO cell lines were established by electroporation using a GenPulser (Bio-Rad, Munich, Germany). In brief, 400 μl of cells (1 × 10⁷/ml) were mixed with 10 μg of circular DNA, electroporated (270 V, 975 μF), and plated out in a volume of 150 ml of medium. G418 selection (0.5 mg/ml; Life Technologies, Karlsruhe, Germany) commenced after 2 days, methotrexate (5 nM; Wyeth, Münster, Germany) was added on day 7 and increased to 100 nM over 2 wk. Stable cell lines producing high levels of Abs were expanded, weaned off FCS, and transferred into a Technomouse system (Integra Biosciences/Technomara, Fernwald, Germany) for large scale production.

Technomouse-derived Ab supernatant was dialyzed against PBS (pH 7.2) overnight (4°C) and passed over a protein L-Sepharose column (fast protein liquid chromatography system; Amersham Pharmacia Biotech, Freiburg, Germany). Binding Abs were eluted stepwise by adding 0.1 M glycine/HCl (pH 3.5), and samples were neutralized rapidly by the addition of 1 M Tris buffer (pH 8.0). Molecular mass determination of native immunocytokine was performed with a Superdex 200 prep grade HiLoad 16/60 using low and high gel filtration chromatography calibration kits (ferritin, 460 kDa; catalase, 206 kDa; aldolase, 107 kDa; albumin, 67 kDa; OVA, 43 kDa; chymotrypsinogen A, 25 kDa; Amersham Pharmacia Biotech). The elution volumes were determined from the chromatogram, and the calibration curve of calculated K_av values vs the log m.w. was generated. The m.w. of anti-FAP-IgG1 TNF dimer was obtained from analysis of the calibration curve. SDS-PAGE analysis was performed as described previously (17), and protein bands were visualized by staining with Coomassie Brilliant Blue (Sigma-Aldrich, Munich, Germany). Proteins were transferred to a polyvinylidene difluoride membrane (Millipore, Bedford, MA) for Western blotting analysis, and membranes were blocked with 10% nonfat dry milk before incubating with rabbit anti-human TNF (1/1000), followed by goat anti-rabbit IgG conjugated to HRP (1/3000; Bio-Rad, Munich, Germany), and were developed by chemiluminescence (ECL kit; Amersham Pharmacia Biotech).

Flow cytometry was performed as described previously (17). In brief, 1 × 10⁶ tumor cells were incubated with purified Abs with the indicated specificity and concentration (30 min, 4°C). Cells were washed twice with PBS and incubated with rabbit anti-TNF (1/1000 dilution, 30 min, 4°C; DAKO, Hamburg, Germany), and finally, the complex was visualized by adding PE-conjugated goat anti-rabbit serum (1/100 dilution; DAKO). Ten thousand cells of each sample were counted and analyzed.

Binding of anti-FAP-IgG1-TNF was also assessed by ELISA. Ninety-six-well, flat-bottom microtiter plates (MaxiSorp Immuno microwell plates; Nunc, Roskilde, Denmark) were coated (overnight, 4°C) with Infliximab (1 μg/ml), Etanercept (1 μg/ml), or recombinant human TNF-R1 (rhTNF-R1; 1 μg/ml) in 50 μl of coating buffer/well. Plates were blocked with 1.5% gelatin in PBS, and then the indicated reagents resolved in PBS were added in serial dilutions (1 h, room temperature). After incubation with murine anti-idiotypic anti-FAP Ab (1 μg/ml, 1 h, room temperature), biotinylated rabbit anti-mouse serum (1/2000, 1 h, room temperature) and HRP (1/50,000, 15 min, room temperature; Roche, Mannheim, Germany) were added. Plates were developed by the addition of o-phenylenediamine substrate (Sigma-Aldrich, Deisenhofen, Germany). The reaction was stopped with 3 M HCl, and plates were read on a Fluorometer (model 1420, Victor 2; Wallac, Turku, Finland) at 490 nm.

PMN were purified from venous blood of healthy donors as previously described (18). All H₂O₂ measurements were performed in flat-bottom microtiter plates (MaxiSorp Immuno microwell plates; Nunc) using the plate-reading fluorometer (18). Plates were precoated with 1 μg/ml human plasma fibronectin (Harbor Bio-Products, Norwood, MA) in 50 μl of Krebs-Ringer phosphate with glucose (KRPG)/well (37°C, 2 h) in a humidified incubator (5% CO₂). Before use, plates were washed three times with PBS and prewarmed to 37°C, and then 100 μl of KRPG containing 50 μM A6550 (Molecular Probes, Eugene, OR) and 1 U/ml HRP (Sigma-Aldrich; reaction mixture) were added. Triggering agents and Abs were added thereafter in 10 μl of KRPG/well. Plates were incubated (37°C, 10 min), and the reaction was started by the addition of 3 × 10⁴ PMN in 20 μl of KRPG to each well. Measurements were made at selected time points (37°C over 120 min in air). H₂O₂ standard curves with A6550 were generated in parallel, and fluorescence changes from the activated PMN were converted to nanomolar concentrations of H₂O₂ using these standard curves.

Different methods were used for the detection of cell death. One way of assessing dead cells was the annexin V-biotin method according to the manufacturer’s recommendation (Roche). In brief, tumor cells (5 × 10⁵) were cultured in 24-well plates overnight in the presence of the indicated amount of Abs or human TNF, washed in PBS, and then incubated with annexin V-biotin (10 μl) in reaction buffer (10 mM HEPES/NaOH (pH 7.4), 140 mM NaCl, and 5 mM CaCl₂). After 15 min, cells were washed, and the percentage of annexin V-FITC-positive cells was determined by flow cytometry as described above. Assays were repeated at least three times. For analysis of memTNF-mediated cytotoxicity, Colo205 target cells were cocultured with FAP-transfected or mock-transfected FAP-negative HT1080 cells (2 × 10⁴/well). Experiments with the Colo205 clone were performed as previously described (9). In brief, the HT1080 clones were seeded in microtiter plates overnight and incubated with the indicated amounts of Abs (30 min, 4°C). After washing with PBS, cells were fixed with 2% formaldehyde (10 min, 37°C) and washed three times with PBS before the Colo205 cells were added (2 × 10⁴/well). The viability of the target cells was measured after 48 h using the EZ4U assay (Biomedica, Wien, Austria) according to the manufacturer’s instructions at 492 nm. Absorbance of untreated cells defined 100% viability.

The HUVEC endothelial cells were prepared as previously described (19) and cultured in supplemented growth medium. Cells were detached from the culture dish by EDTA treatment (0.5 mM) and scraping. The assay was conducted with whole cells in suspension at second or third passage. Tissue factor (TF) activity was assessed by monitoring the hydrolysis of chromogenic substrate S-2222 (Chromogenix, Milan, Italy) as previously described (9, 20). In brief, 1 × 10⁵ endothelial cells were seeded in triplicate with 2 × 10⁵ FAP-positive or -negative HT1080 cells and preincubated (30 min, 4°C) with the indicated amounts of reagents. After coincubation (4 h, 37°C), HUVECs were washed, and reaction buffer (50 mM HEPES, 100 mM NaCl, 5 mM CaCl₂, 0.5% BSA, 10 mM FVIIa (American Diagnostica, Greenwich, CT), and 100 mM FX (American Diagnostica), pH 7.4) was added. FXa formation was assessed using 400 μM S-2222. Hydrolysis was monitored at 37°C by measuring the change in absorbance at 405 nm. A standard curve with upper and lower detection limits of 25 and 0.395 ng/ml was generated using lipidated human recombinant TF.

Mice were sacrificed, and xenotransplanted solid tumors together with a margin of surrounding tissue were harvested. Excised samples were snap-frozen in liquid nitrogen and stored at −80°C. Sections were cut and placed on SuperFrost Plus microscope slides (Menzel-Glaser, Braunschweig, Germany). For staining, slides were fixed in cold acetone, air-dried, rehydrated in PBS, and then blocked using a Biotin Blocking System (DAKO, Carpinteria, CA). Before adding the primary Ab (anti-mouse CD11b or anti-mouse PECAM-1, 1/100), slides were blocked with healthy human donor serum (Vectastain ABC Kit; Vector Laboratories, Burlingame, CA). For staining with humanized/chimeric Abs (anti-human FAP, anti-human G250, or irrelevant isotype-matched human IgG; 1/5000), slides were blocked with FCS (1/100). The biotinylated secondary Ab (goat anti-human IgG (1/200) or goat anti-rat IgG (1:1000)) and the preformed avidin-biotinylated HRP complex (ABC reagent) were used according to the manufacturer’s instructions. The Ab-ABC complex was visualized with a 3-amino-9 ethylcarbazole (Sigma-Aldrich, St. Louis, MO)-based chromagen, resulting in a pink-brown coloration of Ag-positive cells. All slides were counterstained with Meyer’s hematoxylin.

Adult male and female BALB/c nu/nu mice (6–8 wk old; Charles River Laboratories, Wilmington, MA) were used for toxicity and efficacy assessment. Tumor engraftment to analyze FAP expression on tumor cells in vivo was achieved by s.c. injection of 2 × 10⁶ FAP-transfected HT1080 cells in the right flank. This cell number was known to achieve a 100% take rate (data not shown). FAP Ag expression on xenografts was confirmed every 8 days over 4 wk (two mice per time point). In the first experimental setting, 15 animals were divided randomly into three experimental groups (five mice per group), and treatment was initiated when solid tumors had reached ∼5 mm³. Two treatment groups received equal doses (100 μg suspended in 150 μl of PBS) of anti-FAP-IgG1-TNF dimer or of the parental anti-FAP Ab administered i.v. by tail vein injection every 3 days for five consecutive injections. Control mice received the same volume of PBS alone. Health and body weight were monitored daily, whereas tumor size measurements were assessed on treatment days. Animals were sacrificed when tumor volume reached >1 cm³. The second model relied on the simultaneous establishment of HT1080-FAP-positive and parental, mock-transfected Ag-negative (HT1080 par) xenografts on opposing flanks in the same mice. Cells of each cell line (2 × 10⁶; suspended in 150 μl of PBS) were injected s.c. into the right or left flank, respectively. Fifteen simultaneous FAP-positive and -negative tumor-bearing animals were then divided into five groups of three and treated in the same manner. Tumor growth and body weight were again monitored closely. The final in vivo experiment was performed with an identical protocol design compared with the second model. Treatment, however, was extended by the application of rhTNF and anti-G250-TNF dimer (our unpublished observations) to characterize the specificity of effector recruitment and to define maximum tolerated doses. All studies were performed under an existing animal ethics approval as required by federal law.

In the present immunocytokine, the CH2/CH3 of the human Fc region has been replaced by the mature human TNF molecule, resulting in the formation of a TNF dimer stabilized by the natural disulfide bond IgG1 hinge region. The variable domains of the humanized anti-FAP Ab were inserted through specific restriction sites, and the final fully human vector construct was confirmed by DNA sequencing. High level expression of the recombinant immunocytokine in CHO cells was achieved with a final yield of 15 mg/liter of culture. The immunocytokine was purified in the current study using a one-step procedure on protein L columns. Analysis of purified anti-FAP-IgG1-TNF dimer by size exclusion fast protein liquid chromatography showed that the immunocytokine migrated at ∼150 kDa (Fig. 1,A). Furthermore, the purity and integrity of the eluted material were analyzed by gel electrophoresis under reducing and nonreducing conditions and Western blotting (Fig. 1 B). The recombinant H chain product (anti-FAP-HC) is slightly smaller than a natural IgG1 H chain and runs at 47 kDa. The human L chain is not altered and has a size of 28 kDa.

Target Ag binding of the immunocytokine was assessed by flow cytometry and compared with its parental anti-FAP Ab counterpart. At all Ab concentrations ranging from 10 μg/ml to 10 ng/ml, comparable binding properties regarding affinity and specificity were observed without any significant difference (Fig. 2,A). Binding of the anti-FAP-TNF construct was demonstrated on HT1080-FAP⁺ cells using an anti-human TNF-α Ab. Furthermore, correct biochemical assembly of the TNF moiety was analyzed by ELISA, demonstrating proper binding to Infliximab (chimeric anti-human TNF mAb), rhTNF-R1, or Etanercept (rhTNF-R2-Fc fusion protein), respectively. Binding of the anti-FAP-TNF dimer was visualized using a murine anti-idiotypic anti-FAP Ab (Fig. 2, B–D) to prove the integrity of the entire construct.

In the first instance, WEHI-164 cells were used to investigate the sTNF-like activity of the anti-FAP-TNF dimer in comparison with that of the parental FAP Ab and trimerized rhTNF. Using annexin V as a very sensitive indicator of cell death, a strong correlation between the concentration of the anti-FAP-TNF construct and the level of cell killing was observed. Specific cell lysis was detected down to concentrations of 1 pg/ml for the anti-FAP-TNF dimer. However, a significantly stronger cell lysis was seen for commercially available recombinant rhTNF (∼10-fold). Neither medium alone nor the parental anti-FAP Ab at the highest concentration (10 μg/ml) induced a significant level of cell death. A WEHI-164 variant cloned under the exposure to TNF was partially resistant to both rhTNF and the anti-FAP-TNF construct (Fig. 3,A). Coincubation with Etanercept inhibited the anti-FAP-TNF-mediated lysis of TNF-sensitive WEHI-164 cells, whereas an irrelevant IgG control Ab failed to block TNF-activity (Fig. 3 B).

It has been widely demonstrated that human PMN release large quantities of reactive oxygen species in response to TNF when the cells are adherent to surfaces coated with extracellular matrix proteins. Fig. 4 shows the respiratory burst of human PMN adherent on fibronectin-coated plates measured as H₂O₂ production. Addition of the anti-FAP-TNF construct resulted in high superoxide production, whereas the parental anti-FAP Ab induced an effect comparable to that of KRPG buffer. Responses induced by anti-FAP-TNF dimer were similar to those to rhTNF.

The assays described above proved the potency of the dimerized cytokine to act like sTNF. However, the anti-FAP-TNF dimer was designed to achieve targeted bioactivity of memTNF via both TNF-R types after binding to the FAP Ag. For that purpose the anti-FAP-TNF construct was absorbed to FAP-positive or -negative cells, and the unbound Ab was washed off and then added to WEHI-164 cells. As expected, only the immunocytokine attached to FAP-positive HT1080 cells was able to mediate a strong and concentration-dependent lysis of WEHI target cells (Fig. 5,A). No such activity was detected when the parental HT1080 cell line (mock-transfected, FAP Ag-negative) or parental anti-FAP Ab were used. To investigate the ability of the immunocytokine to induce additive cytotoxicity via TNF-R2-dependent signaling by mimicking the memTNF message, a second experiment was performed. In contrast to the WEHI-164 S clone, Colo205 cells are completely resistant to the cytotoxic activity of sTNF, even at high concentrations. Only by memTNF-like signaling did Colo205 cells become highly susceptible to TNF-mediated autoproteolysis (9). The anti-FAP-TNF dimer exhibited memTNF-like activity via presentation by FAP-expressing HT1080 cells to the Colo205 line, resulting in a dose-dependent measurable death of the target cells. The control samples including coincubation of FAP-negative HT1080 cells with the fusion protein, the unbound anti-FAP-TNF dimer alone, or the FAP-bound parental Ab failed to induce lysis of the Colo205 cells, respectively (Fig. 5 B).

One of the main tumoricidal mechanisms of TNF represents the alteration of tumor vasculature leading to subsequent tumor necrosis. The central step to initiate this thrombotic process involves enhanced synthesis and surface expression of endothelial TF. Stimulation of HUVEC with memTNF leads to a significantly higher TF expression compared with sTNF (9, 11). We determined the induction of TF expression on HUVECs by the anti-FAP-TNF dimer acting either in solution or presented in cell-bound status via FAP Ag by a two-step chromogenic assay. The saturated concentration of sTNF leading to highest TF expression was 35 ng/ml, in accordance with previously reported data (9). Under these conditions, coincubation of HUVECs with FAP-expressing HT1080 cells alone or with FAP-negative HT1080 cells in the presence or the absence of anti-FAP-TNF dimer failed to induce increased TF synthesis. However, increasing amounts of TF activity, despite saturated sTNF concentration, were detectable after coincubation of HUVECs with FAP-expressing HT1080 cells preincubated with anti-FAP-TNF dimer (Fig. 6).

BALB/c nu/nu mice bearing FAP⁺ solid human fibrosarcoma xenografts (∼5 mm³) were treated with a total of five consecutive injections, each 3 days apart, to analyze the in vivo toxicity and activity of the anti-FAP-TNF fusion protein. To compare equimolar concentrations, 30 μg of human TNF was compared with 100 μg of anti-FAP-TNF. The group of tumor-bearing mice treated with 30 μg of rhTNF all died within 24 h and could not be challenged with a second injection. Mice treated with the immunocytokine, however, showed some signs of stress, but all completed treatment without any lethal side effects. The anti-FAP-TNF immunocytokine administered five times at 100 μg/injection did not cause significant weight loss when only one solid tumor had been established per mouse (Fig. 7). When animals were challenged simultaneously with the FAP Ag-positive and -negative cell lines, however, all animals lost weight over time regardless of their treatment. The effect on weight loss was most pronounced in the group of mice treated with repetitive doses of the anti-FAP-TNF immunocytokine, but was still minor (Fig. 7).

Previous experiments had shown that the HT1080-FAP⁺ cell line is, per se, not TNF sensitive and can only be lysed in vitro by our construct or human TNF, respectively, in the presence of cycloheximide (data not shown). In addition, established FAP⁺ fibrosarcomas were only poorly vascularized (Fig. 9,B), thus limiting both the intratumoral uptake of the fusion protein and the antitumor effects mediated through TNF activity on endothelial cells. Therefore, the major focus of this animal model was on toxicity induced by our construct in comparison with equimolar concentrations of rhTNF. Nevertheless, the repeated i.v. injection of 100 μg of anti-FAP-TNF dimer caused a significant delay of tumor growth (Fig. 8,A). Control animals treated with the parental anti-FAP Ab, a control immunocytokine with no FAP specificity, or PBS alone, respectively, showed no growth inhibition. To demonstrate that the antitumor effect of the anti-FAP-TNF dimer is Ag dependent, we established simultaneously HT1080-FAP⁺ (FAP-positive) and HT1080 par (FAP-negative) tumor xenografts on opposite flanks of the same mice (Fig. 8,B). Tumor-bearing animals underwent the same regimen as in the first experiment. Application of the anti-FAP-TNF dimer again caused regression of the FAP-positive fibrosarcomas in the same manner, whereas mock-transfected tumor cells lacking FAP expression were not affected in their growth pattern. Immunohistological analysis of anti-FAP-TNF-treated tumors demonstrated a strong accumulation of murine PMNs in FAP-expressing tumors (Fig. 9,A). Staining of tumor sections showed a solid mass of tumor cells without significant vessel structures assessed by immunohistochemistry using the endothelial cell marker PECAM-1. Vessel structures that stained positively for PECAM-1 were restricted to the outer connective tissue capsule surrounding the tumor (Fig. 9,B). To clarify that the induction and extent of PMN enrichment were triggered by targeted delivery of the TNF fusion protein, we performed an additional xenograft study. Animals bearing FAP-positive and -negative tumors were challenged with an increasing dose regimen. In an analogy to former treatment, equal doses (100 μg), but increasing numbers of injections, were administered every 3 days. Tumors were harvested 24 h after the last injection for immunohistochemical analysis. The enrichment of anti-CD11b-stained murine leukocytes in the targeted tissue corresponded to the injected cumulative dose of the anti-FAP TNF dimer (Fig. 9, C and E). Again, fusion protein-mediated PMN recruitment was strictly Ag dependent (Fig. 9, D and F), as detection of leukocytes surrounding FAP-negative tissue failed (Fig. 9,G), and treatment of animals with the parental anti-FAP Ab did not result in leukocyte homing (Fig. 9,H). Furthermore, effector cell recruitment to FAP⁺ tissue could not be achieved by application of dimerized TNF coupled to an Ab with irrelevant Ag-specificity (anti-G250-TNF dimer; Fig. 9 I).

In the present study we describe a new human immunocytokine targeting tumor stroma by binding to the FAP Ag and delivering membrane-anchored TNF-α bioactivity. We have deliberately chosen the IgG1-hinge-TNF Ab format because a CH2/CH3-deleted fusion protein has some advantages compared with similar approaches where the cytokine moiety is attached behind the CH3 domain (21). The latter construct resembles an intact IgG1 Ab with the cytokine at the C-terminal end. The resulting tetravalent Ab constructs maintain most of their Fc region-associated effector functions, such as long half-life and FcR binding (22). Accordingly, one would expect that a TNF fusion protein derived from an intact IgG1 Ab would bind to unprimed macrophages and induce the typical signal pattern of sTNF (23). As a consequence, systemic side effects and unfavorable tumor-targeting properties should limit the clinical use of such a compound. The CH2/CH3-deleted fusion protein described by us has no known FcR-binding properties (data not shown). The truncation left the hinge region unaffected to ensure the stability of the fusion protein. Blocking the N-terminal TNF domain through direct linkage to the hinge region even increases rigidity and reinforces the final fusion protein to retain the character of a dimeric protein, because mature TNF monomers associate intimately to form a solid state. However, this raises the question of the general activity of the TNF portion of the immunocytokine. Data from the literature dealing with deletion of the first amino acid residues of native TNF either reported no influence on the activity or even a 4-fold increase (24). Another important question concerns the dimeric structure of the TNF component, because a 3-fold symmetry of TNF and TNF-R superfamilies physiologically defines the essential signaling stoichiometry and structure (25).

In addition to recent data demonstrating the superior capacity of memTNF when acting in a positive cooperative manner through both TNF-R forms, a TNF-R2-specific mAb was developed that was able to mimic the activity of memTNF in the presence of sTNF (9). Adjacent kinetic studies revealed that this Ab stabilized the receptor/ligand binding, leading to a prolonged half-life of the complex. The authors raised the hypothesis that the intracellular signaling strength could depend on the kinetics of receptor-ligand complex formation, because the rapid dissociation of sTNF from TNF-R2 failed to induce intracellular signals (9). Subsequent studies showed that the agonistic activity of this Ab was dependent on its dimeric IgG1 structure, as Fab were inactive (26). The fact that dimerization of TNF-R2 is an obligatory step for the initiation of the signal cascades (27) also suggests that the secondary clustering of the TNF-TNF-R2 complexes by this Ab could reflect the responsible mechanism for enhanced memTNF-TNF-R2 signaling capacity. Consequently, the extent of TNF-R2 signaling is not solely determined by the affinity of the receptor-ligand interaction (28). This is in accordance with our finding that the anti-FAP-TNF dimer binds to Etanercept (Fig. 2,B), a fusion protein made up of the extracellular domain of TNF-R2 and the hinge and Fc domain of human IgG1 (29). Complexes of an anti-FAP-TNF dimer and Etanercept were more stable over time compared with an anti-FAP single-chain Ab genetically fused to rhTNF monomer (data not shown). Together this might explain why the anti-FAP-IgG1-TNF dimer in the current study exerts the bioactivity of memTNF, although the dimerized structure causes a reduced affinity for TNF-R1 (30) compared with recombinant soluble TNF (Fig. 3). Furthermore, the induction of intracellular death domain signaling pathways via TNF-R1, including those leading to apoptosis, is based on internalization of the ligated receptor complex (31). The 150-kDa size of our fusion protein could sterically hamper this endocytosis. In contrast to TNF-R1-induced apoptotic pathways, the responses of PMNs to TNF-triggered respiratory burst are probably not dependent upon receptor complex internalization, because the proline-rich tyrosine kinase (pyk2) that plays a critical role in this activation process is not linked to the death domains of TNF-R1. Therefore, known inhibitors of TNF-R1 endocytosis could not block the kinase activity (31, 32). The enzyme acts by integrating the TNF and β₂ integrin signals, whereas the cooperation of TNF-R2 in the respiratory burst-inducing process is more important at low TNF concentrations, suggesting a modulatory role at physiologically relevant concentrations of the cytokine (6). This is in accordance with our data and explains the equal extent of H₂O₂ release after stimulation with rhTNF or in response to our fusion protein, as all reagents acted in solution on adherent human PMNs (Fig. 4). Again, this argues for the importance of the CH2/CH3 deletion in our IgG1 fusion protein to prevent TNF-induced triggering of peripheral blood macrophages subsequent to FAP-independent FcR binding.

To date, our results confirmed the expected behavior of the IgG1 fusion protein when acting in solution. The challenged activity profile, however, included induction of the memTNF-TNF-R2-mediated signal capacity. In several coculture assays, the FAP Ag-bound fusion protein showed the juxtatropic cytotoxicity of cell surface TNF that signals by cell-to-cell contact (Figs. 5 and 6). We chose the Colo205 assay to demonstrate that our construct could be responsible for a strong signal enhancement by inducing stabilization of TNF-TNF-R2 complexes. The induction of cell death by TNF-R2 signaling in Colo205 cells reflected the memTNF-like activity of the anti-FAP-IgG1-TNF dimer, resulting in IgG1-dependent receptor clustering (Fig. 5,B). Further evidence for the potency of the construct to initiate TNF-R2-dependent pathways was shown by TF induction on HUVEC. Both TNF-Rs are involved in TNF responses resulting in the expression of this endothelial prothrombotic factor (11). However, under conditions of saturated concentrations of sTNF acting through TNF-R1, only additional signaling through TNF-R2 yields superior rates of TF expression (9). Again, presentation of the anti-FAP-TNF construct in its Ag-bound status to the human endothelial cells resulted in memTNF bioactivity leading to enhanced cell surface expression of TF (Fig. 6). This is the key event for the activation of the extrinsic coagulation cascade resulting in thrombotic infarction of tumors. Destruction of tumor vessels is the major effect of TNF when used in isolated limb perfusion as the only clinically accepted use for rhTNF (33). To overcome this restriction of TNF application for tumor immunotherapy, the advantage of our construct with regard to reduced toxicity in vivo had to be demonstrated, because after i.v. application the TNF-portion of the fusion protein could potentially interact with the type I TNF-R of nontargeted cells, above all endothelial cells, despite Ab linkage. This stimulation could increase vascular permeability and finally lead to capillary leak syndrome (34, 35). As FAP expression of transfectants was detectable over 4 wk of solid growth in BALB/c nu/nu mice (data not shown), in vivo targeting of FAP-transfected HT1080 cells provided an appropriate test system to examine the toxicity profile of the fusion protein. Human TNF in its trimeric structure is ∼30-fold less active in BALB/c nu/nu mice than its murine homologue (36) with a calculated 50% lethal dose of 1.2 μg for recombinant mouse TNF and 20–30 μg for rhTNF, respectively (37). Therefore, we decided to use single injections of 30 μg of human TNF to achieve proposed 50% lethal dose level. As expected, all animals treated died within 24 h. However, mice treated with 100 μg of anti-FAP-TNF dimer/injection did not die and could be treated repetitively. Only mice bearing xenografts of the FAP Ag-negative and -positive HT1080 cell line simultaneously experienced significant weight loss and cachexia. Nevertheless, it is well known from the literature that tumor-bearing animals are more susceptible to TNF-mediated kachexia (38).

The animal model was not designed primarily to demonstrate antitumor efficacy, because rhTNF fails to induce murine TNF-R2 signaling (39), but the cooperative signaling through both TNF-Rs represents the therapeutic principle of the anti-FAP-TNF dimer, exerting its antitumor activity in a membrane TNF-like manner. Therefore, drug-related therapeutic events should be expected via TNF-R1 (35). As stated above, the HT1080 FAP⁺ cell line is not TNF sensitive, and we could not expect to induce tumor cell apoptotis by our construct. Furthermore, the growth pattern of xenotransplanted fibrosarcoma was characterized by poor vascularization, thus excluding fusion protein-mediated TNF signaling leading to endothelial TF up-regulation causing hemorrhagic tumor necrosis. Finally, the respiratory burst of mouse neutrophils in response to soluble TNF is 10-fold lower than that for their human counterparts (40, 41). However, when established xenografts were treated from day 9 after tumor inoculation, a significant delay in tumor growth was observed. Immunohistochemical analysis revealed FAP Ag-dependent recruitment of murine leukocytes as the main effector mechanism, exclusively induced by anti-FAP-TNF dimer. This was a reliable and dose-dependent inducible effect. In addition, we confirmed the fusion protein-mediated growth delay in a second experiment in which mice had mock-transfected FAP Ag-negative and -positive tumor xenografts established simultaneously. Again, the FAP-positive tumors revealed delayed growth after anti-FAP-TNF treatment, with no change in the growth pattern for the FAP-negative tumors established on the opposite flank of the mouse.

In conclusion, we present data on a TNF-based immunocytokine that exerts its Ag-restricted therapeutic activity after i.v. injection with tolerable systemic toxicity. The anti-FAP-TNF dimer gains the potent bioactivity of memTNF exclusively after binding to the target Ag and initiates various cellular response programs. Focusing the described synergistic antitumor activities to the FAP-positive tumors minimized in vivo side effects. Two previous studies have described the concept of CH2/CH3 truncation to design Ab hybrid molecules similar to ours, but never achieved production levels that allowed them to perform relevant assays (42, 43). Our genetically engineered construct reached production yields of 15 mg/liter. A transition into good manufacturing practice-conforming desired production processes should be feasible and guarantee the initiation of clinical trials (44).

We thank Anja Reich and Nicole Jordan for excellent technical assistance. We thank the Department of Clinical and Experimental Surgery for housing and handling of animals.