Novel strategies to mimic transmembrane tumor necrosis factor-dependent activation of tumor necrosis factor receptor 2

Tumor necrosis factor receptor 2 (TNFR2) is known to mediate immune suppression and tissue regeneration. Interestingly, the transmembrane form of tumor necrosis factor (tmTNF) is necessary to robustly activate TNFR2. To characterize the stoichiometry and composition of tmTNF during TNFR2 activation, we constructed differently oligomerized single chain TNF ligands (scTNF) comprised of three TNF homology domain (THD) protomers that mimic tmTNF. Using a variety of cellular and in vivo assays, we can show that higher oligomerization of the scTNF trimers results in more efficient TNF/TNFR2 clustering and subsequent signal transduction. Importantly, the three-dimensional orientation of the scTNF trimers impacts the bioactivity of the oligomerized scTNF ligands. Our data unravel the organization of tmTNF-mimetic scTNF ligands capable of robustly activating TNFR2 and introduce novel TNFR2 agonists that hold promise as therapeutics to treat a variety of diseases.

of sTNF for TNFR1 is mainly caused by stabilization of ligand/receptor complexes, while transient binding of sTNF to TNFR2 results in short-lived signal incompetent complexes 15,16 .
A probable reason for the tmTNF-dependence for TNFR2 activation is the higher demand of TNFR2 for ligand-mediated crosslinking to allow signaling cluster formation 16,17 . In this line, oligomerization of soluble ligand trimers, e.g. via antibody-mediated crosslinking, does not increase activation of TNFR1 18 , whereas secondary oligomerization of soluble TNF trimers converts this molecule into an active TNFR2 agonist 19,20 .
To characterize the stoichiometry and composition of tmTNF necessary to efficiently activate TNFR2, we genetically engineered differently oligomerized TNFR2-selective scTNF ligands. The activity of these fusion proteins was then compared in regard to TNFR2 binding, TNF/TNFR2 complex formation and induction of specific cellular responses.
TNFR2 selectivity of the oligomerized TNF muteins was confirmed using HeLa and L929 cells. In contrast to wild-type TNF used as a positive control, none of the TNFR2-selective fusion proteins induced TNFR1-dependent IL-6 secretion in HeLa cells (Fig. 3A). Furthermore, none of the fusion proteins activated TNFR1-dependent cell death in L929 (Fig. 3B), verifying that also the higher oligomerized p53-sc-mTNF R2 and GCN4-sc-mTNF R2 had lost affinity for TNFR1 and the capacity to activate this receptor.
Oligomerization of covalently stabilized scTNF is necessary for formation of higher order TNF/TNFR2 clusters. Receptors of the TNF family are activated by ligand-mediated oligomerization 29 and efficient signal initiation requires the formation of larger ligand/receptor complexes 16,30 . We, therefore, investigated the differences between the oligomeric TNF muteins to induce TNF/TNFR2 signaling cluster formation. Mouse BV-2 microglia cells were incubated with different concentrations of the fusion proteins and TNF/ TNFR2 cluster formation was visualized via immune fluorescence (Fig. 4A,B). At low concentrations (0.1 nM), GCN4-sc-mTNF R2 induced the highest amount of TNF/TNFR2 signaling complexes, indicated by yellow spots due to colocalization of TNF (green) and TNFR2 (red) signals. Around 50% less TNF/TNFR2 signaling complexes were induced with p53-sc-mTNF R2 . Using a high concentration of 1 nM, similar efficient TNF/TNFR2 cluster formation (90-100%) was observed after incubation with p53-sc-mTNF R2 and GCN4-sc-mTNF R2 . In contrast, both scTNF R2 and EHD2-sc-mTNF R2 were less capable of inducing TNF/TNFR2 cluster formation, with 30-40% clusters formed at the highest concentration. Notably, at low concentrations (0.1 nM), almost no cluster formation was observed with sc-mTNF R2 , whereas EHD2-sc-mTNF R2 induced approximately 20% cluster formation.
Bioactivity of the fusions proteins was further tested using Kym-1 cells that are highly sensitive for TNFR2-induced cell death 31 . Therefore, cells were incubated with the fusion proteins for 24 hours and cell viability was measured using the crystal violet assay. Again, bioactivity was increased with higher oligomerization of the TNF muteins ( Fig. 5B and Table 3). No significant TNFR2 signal induction was observed in this cellular system for sc-mTNF R2 at the used concentrations. In contrast, p53-sc-mTNF R2 induced an approximately 10-times stronger signal than EHD2-sc-mTNF R2 , whereas GCN4-sc-mTNF R2 was 10-times more potent than p53-sc-mTNF R2 . Next, we investigated TNFR2-induced proliferation of mouse thymocytes. Therefore, cells were incubated in presence of the different fusion proteins for 4 days and proliferation was determined by measuring their metabolic activity. All oligomerized fusion proteins induced proliferation. Whereas sc-mTNF R2 only promoted proliferation at higher concentrations, all oligomerized fusion proteins efficiently induced proliferation ( Fig. 5C and Table 3). Whereas p53-sc-mTNF R2 (EC 50 value 40 pM) was just slightly more potent than EHD2-sc-mTNF R2 (EC 50 value 55 pM), GCN4-sc-mTNFR 2 induced a more than 3-times increased response (EC 50 value 16 pM) compared to EHD2-sc-mTNF R2 .
We next investigated TNFR2 signal induction in different immune cells (activated human and mouse T cells, regulatory T cells), known to express TNFR2 [32][33][34] . We choose to use a concentration of 0.3 nM of the fusion proteins, due to the different effects observed in the previous assays for this concentration. Similar results were obtained for both, TNFR2-induced expansion of human CD25 + HLA-DR + (Fig. 6A and Table S1) and mouse CD25 + TNFR2 + (Fig. 6B) activated T cells. Thus, GCN4-sc-mTNF R2 proved to be the strongest inducer of T cell proliferation, whereas no significant activity was recorded for sc-mTNF R2 compared to control cultures. For both, human and mouse T cells, p53-sc-mTNF R2 showed an apparent stronger activity than EHD2-sc-mTNF R2 , although the difference was statistically not significant. Since TNFR2 is a key marker of regulatory T cells   Table 2. QCM affinity measurements of oligomeric TNF muteins.
(Tregs) 32,34 and TNF is known to expand and stabilize Tregs via TNFR2 32, 35-37 , we then investigated the potential of the different sc-mTNF R2 fusion proteins to expand Tregs. Therefore, CD3 + mouse T cells were activated using αCD3, cultivated in presence of the tmTNF mimetics and the number of CD4 + CD25 + FoxP3 + Tregs was quantified by flow cytometry (Fig. 6C). Similarly to the previous experiments, the tetrameric GCN4-sc-mTNF R2 molecule was the most potent activator, inducing the strongest expansion of Tregs.

Dodecavalent covalently stabilized scTNF show improved bioactivity in vivo. Recently,
it was shown that in vivo application of a TNFR2 agonist leads to the expansion of Tregs and protects mice from acute graft versus host disease 38 . Therefore, we administered low concentrations (1 mg/kg) of the TNF muteins EHD2-sc-mTNF R2 , p53-sc-mTNF R2 and GCN4-sc-mTNF R2 in mice and determined the number of CD4 + CD25 + FoxP3 + Tregs in the spleen after three days (Fig. 7A). Interestingly no significant activity of EHD2-sc-mTNF R2 compared with saline control was observed. In contrast, both p53-sc-mTNF R2 and GCN4-sc-mTNF R2 efficiently expanded Tregs at the applied dose of 1 mg/kg in vivo. No significant difference was observed between p53-sc-mTNF R2 and GCN4-sc-mTNF R2 at this dose. However, at a tenfold higher concentration (10 mg/kg) EHD2-sc-mTNF R2 induced an efficient expansion of Tregs, too (Fig. S2). The systemic tolerance of the oligomeric TNF fusion proteins was assessed by measuring the levels of C-reactive protein (CRP), a key inflammatory marker, in the blood (Fig. 7B). Neither 24 nor 72 hours after administration of the TNF muteins, altered CRP levels compared to saline-treated animals were observed.

Discussion
Several receptors of the TNF receptor superfamily depend on a transmembrane ligand to become robustly activated. Examples are the ligand/receptor pairs of CD95L and CD95, APRIL and TACI, OX40 L and OX40, TRAIL and TRAILR2, and TNF and TNFR2 14,18,[39][40][41] . For mouse CD95L and TRAIL, soluble trimeric ligands solely containing the THD failed to bind to their cognate receptors and consequently were not able to activate receptor signaling. However, binding of these ligand variants to CD95 and TRAILR2 was restored by stabilization of the trimers via fusion to a trimerization domain. Although trimer stabilization restored receptor binding, this ligand still was not capable to efficiently activate its receptor 4 . Similar, covalent stabilization of TNF trimers improved the affinity to TNFR2, but did not improve the bioactivity 10 .
Previously, different approaches have been exploited to generate soluble TNF ligands mimicking the activity of the transmembrane form, e.g. by artificially increasing the avidity of TNF ligands through mutagenesis or by immobilizing them on a cell surface or the extracellular matrix (ECM). Antibody-mediated multimerization of Flag-tag-containing variants of soluble TNF, CD95L, TRAIL, and CD40L with Flag-specific antibodies resulted in a dramatic increase in avidity and efficient activation of their cognate receptors TNFR2, CD95, TRAILR2, and CD40, respectively 30,42,43 . Similar, antibody-mediated cross-linking of TNFR2 via the monoclonal non-activating antibody 80M2 leads to pre-activation of TNFR2 on the cell surface, thereby mimicking receptor activation by the membrane form of TNF, if soluble recombinant TNF is added 14 . Genetic fusion of soluble CD95L, OX40L trimers to an oligomerization domain resulted in a dramatic increase in avidity and efficiently activated CD95 and OX40 18,30,40 . Alternatively, immobilization of soluble ligands can elevate their activity. For instance, binding of soluble CD95L or soluble APRIL to components of the extracellular matrix converts these ligands into highly active cytokines [44][45][46] . Similar data show that the activity of soluble trimeric variants of CD27L, CD40L, and 41BBL can be strongly increased by oligomerization or cell surface immobilization 7 . The dependence of the naturally occurring transmembrane ligand or membrane-mimetic variants for robust activation is especially obvious for TNFR2. Whereas it was shown that trimer stabilization both improves activation of TNFR1 10 and TNFR2 47 , oligomerization of trimeric TNF ligands is highly superior in activating TNFR2 [19][20][21] . In line with this, additional data indicate that mere mechanical fixation of TNF to an immobilized surface is insufficient to efficiently activate TNFR2. Rather, a certain density of the ligand and an additional stabilization of TNFR2 by cluster formation seems to be mandatory for efficient activation 48 . Whereas data exist that oligomerization of a trimeric scTNF ligand improves its bioactivity, no data were available on the possible stoichiometry and composition of oligomerized TNF ligands necessary to robustly activate TNFR2.
Receptor binding studies revealed that the dodecavalent variants p53-sc-mTNF R2 and GCN4-sc-mTNF R2 both bind more stable to TNFR2 compared to the hexavalent EHD2-sc-mTNF R2 . Furthermore, using different cellular response assays, GCN4-sc-mTNF R2 consistently exerted the highest bioactivity. In addition, in all assays a slightly increased bioactivity of p53-sc-mTNF R2 compared to EHD2-sc-mTNF R2 was observed. A first important step for efficient signal induction is the formation of higher order TNF/TNFR2 signaling clusters. Our microscopical analyses showed that GCN4-sc-mTNF R2 efficiently promoted the cluster formation even at low sub-nanomolar concentrations. Likewise, p53-sc-mTNF R2 was superior in cluster-formation compared to EHD2-sc-mTNF R2 , which translated into a 3.5-fold higher bioactivity in a Cxcl-2 chemokine secretion assay with BV-2 cells.
Altogether, our data show that the three-dimensional orientation of tmTNF-mimetic TNF ligands is important and determines the efficiency of TNFR2 signal induction. Our data suggest that the configuration of GCN4-sc-mTNF R2 shares the highest similarity with the assumed natural tmTNF configuration, since all TNF trimers in the oligomer are arranged in a way that they point into the same direction. In contrast, p53-sc-mTNF R2 seems to have more a spider-like shape in which the binding sites point into the directions of a tetrahedron 24 . We, therefore, hypothesize that these different 3D-structures mainly impact the distance between the scTNF units, with the highly potent GCN4-sc-mTNF R2 configuration displaying the least distance between the scTNF units. Therefore, it is likely that the distance between TNF trimers in tmTNF or tmTNF mimetic ligands also plays a role for the bioactivity. Interestingly, the in vivo application of the tetramers p53-sc-mTNF R2 and GCN4-sc-mTNF R2 is highly superior to dimeric EHD2-sc-mTNF R2 , whereas no significant differences between p53-sc-mTNF R2 and GCN4-sc-mTNF R2 were observed. This superior bioactivity of p53-sc-mTNF R2 in vivo compared to the in vitro assays might be due to possible cis and trans activation of TNFR2 on the same and on neighboring cells, respectively. We propose that a p53-sc-mTNF R2 molecule that binds to TNFR2s with a minimum of two scTNFs is capable of activating TNFR2, similar to EHD2-sc-mTNF R2 , but still possesses two free scTNFs capable of activating TNFR2 in trans on other cells. Keeping in mind that TNF/TNFR2 clusters are formed, such a higher order p53-sc-mTNF R2 /TNFR2 cluster with free scTNF molecules could mimic natural tmTNF signaling, i.e. an additional transactivation of a juxtaposed cell.
Since TNFR2 plays an important role in immune regulation and tissue regeneration, ligands that effectively activate TNFR2 hold great promise as novel therapeutics to treat a variety of diseases, including inflammatory and neurodegenerative diseases 11,13,49,50 . Drug potency and efficacy is of particular interest for therapeutic applications, as lower doses of the particular drug are needed to induce maximal therapeutic effects. Therefore the high and specific bioactivity of tmTNF-mimetic tetrameric scTNF oligomers represent a novel approach towards TNFR2-selective drugs.
In conclusion, we show that controlled oligomerization of scTNFs is a prerequisite to efficiently activate TNFR2. By introducing various oligomerization domains to generate dimeric or tetrameric scTNF ligands with different three-dimensional orientation of the scTNF units, we identified formats suitable for efficient TNF/   TNFR2 complex formation and signal induction and high bioactivity in in vitro and in vivo model systems. Our findings, therefore, are of principal relevance for development of TNFR2 agonists for potential therapeutic application. For the use in humans completely human molecules are needed. This can be readily achieved by exchanging the sc-mTNF R2 part to its human counterpart scTNF R2 which was previously described by our group 19,21 . The yeast GCN4 tetramerization domain was used here as a model of a coiled-coil structure. To circumvent potential immunogenicity of foreign protein domains, this domain could be exchanged to similar coiled-coil tetramerization domains of human proteins. In addition, other modules and formats resulting in a GCN4-sc-mTNF R2 -like configuration are possible to design highly potent TNFR2 agonists. In summary, our data show that TNFR2 activation by dodecavalent ligands is clearly superior to trivalent or hexavalent ligands. We also demonstrate that the predicted 3D-structure of a dodecavalent TNFR2-selective TNF mutein appears to be important for the bioactivity of the ligand. In addition, because of conserved structural features of the members of the TNF/TNFR superfamily, our findings suggest that similar conditions exist for other members of this biologically important cytokine family.
HPLC. Approx. 20 µg protein was applied to a BioSep-SEC-S2000 column (Phenomenex, Aschaffenburg, Germany) equilibrated with PBS and eluted at a flow rate of 0.5 ml/min. For determining the size of recombinant proteins, standard proteins were run under the same conditions. TNFR binding assay. ELISA plates (Greiner, Frickenhausen, Germany) were coated with TNFR1-Fc or TNFR2-Fc fusion proteins at 1 µg/ml in PBS and incubated at 4°C overnight. Residual binding sites were blocked with 2% skim milk powder in PBS at RT for 2 hours. TNF muteins were diluted in 2% skim milk powder in PBS and incubated for 1 hour at RT. Bound proteins were detected with mouse monoclonal antibodies to TNF (clone F6C5; 1 µg/ml; incubation for 1 hour at RT) and HRP-conjugated anti-mouse IgG antibodies (diluted 1:10000; incubation for 1 hour at RT), followed by incubation with TMB substrate solution. Reaction was stopped by addition of 1 M H 2 SO 4 and the absorbance at 450 nm was determined with an absorbance reader (Multiskan FC, Thermo Scientific, Karlsruhe, Germany) and data were analyzed using the software Microsoft Excel and GraphPad Prism 4 (GraphPad, La Jolla, CA). Between each step, non-bound proteins were removed by washing 4 times with 0.005% Tween-20 in PBS. Enzyme-linked immunosorbent assay. BV-2 or HeLa cells were stimulated as indicated, supernatants were collected after 24 hours and analyzed by an ELISA specific for Cxcl-2 (BV-2, R&D Systems, Minneapolis MN) or IL-6 (HeLa, Biolegend, San Diego CA) according to the instructions of the manufacturer. For determination of inflammatory marker CRP in the blood, whole blood was withdrawn and analyzed by an ELISA specific for mouse CRP (R&D Systems, Minneapolis MN) according to the instructions of the manufacturer. The absorbance at 450 nm was determined and the amount of released Cxcl-2 or IL-6 was determined with the provided standard and calculated using the software GraphPad Prism.
Biological samples. Blood was obtained from volunteer donors, from whom informed consent was obtained. Blood withdrawal was carried out in accordance with relevant guidelines and regulations and all experimental protocols were approved by Ethik-Kommision Universitätsklinikum Tübingen (Project number 283/2014B02).
Human T cell isolation and culture. Blood of volunteer human donors was diluted 1:2 with RPMI medium. Then 30 ml diluted blood was layered over 10 ml Histopaque-1077 (Sigma-Aldrich) and centrifuged for 20 min at 800 g without brake. Interphase, including mononuclear cells of peripheral blood (PBMCs) was removed and washed with 30 ml RPMI (300 g, 5 min). For removal of platelets, cells were resuspended in 40 ml RPMI and centrifuged for 5 min at 200 g. Then, CD3 + T cells were isolated by magnetic separation using the Pan T Cell Isolation Kit (Miltenyi Biotech). Purified T cells were plated in X-Vivo 15 medium (Lonza) in αCD3 coated (6 h at 4 °C) 96-well (U form) plates for T cell activation. Cells were incubated in presence of IL-2 and the oligomerized fusion proteins for 4 days. Then surface expression of CD25 and HLA-DR was determined by flow cytometry according to manufacturer's instructions (Miltenyi Biotech, Bergisch-Gladbach). At least 20,000 cells per sample were acquired. Data were acquired using a MACSQuant Analyzer 10 (Miltenyi) and analyzed with FlowJo (FlowJo, LLC, Ashland, Oregon).
Mouse T cell isolation and culture. Spleens from C57BL/6 wildtype mice were dissociated through a 40 µm cell strainer and collected in 10 ml MACS buffer (PBS, 0.5% BSA, 2 mM EDTA). Splenocytes were centrifuged (300 g, 5 min) and washed once with 10 ml MACS buffer. Then CD3 + T cells were isolated using the FACS Aria III and plated in αCD3 coated (6 h at 4°C) 96-well (U form) plates for T cell activation. T cells were cultivated in presence of IL-2 and the oligomerized fusion proteins for 4 days. Then expression of CD25, TNFR2 and/or FoxP3 was determined by flow cytometry according to manufacturer's instructions (Miltenyi Biotech, Bergisch-Gladbach). Data were acquired using a MACSQuant Analyzer 10 (Miltenyi) and analyzed with FlowJo (FlowJo, LLC).
In vivo Treg assay. Animal care and treatment were carried out in accordance with Committee Stuttgart (Regierungspräsidium Stuttgart) guidelines on the use of experimental animals at the University of Stuttgart. Animal experiments were approved by Regierungspräsidium Stuttgart (permit no. 35-9185.81/0422). Oligomerized fusion proteins (1 mg/kg) were administered intraperitoneal (i.p.) to C57BL/6 wild type mice. After 3 days a second injection (1 mg/kg) was applied. After 7 days, spleens were extracted and splenocytes were isolated. Therefore, spleens were dissociated through a 40 µm cell strainer and collected in 10 ml MACS buffer (PBS, 0.5% BSA, 2 mM EDTA). Splenocytes were centrifuged (300 g, 5 min) and incubated in 3 ml RBC buffer (0.15 M NH 4 Cl, 10 mM KHCO 3 , 0.1 M EDTA) per spleen for 5 minutes at room temperature to lyse red blood cells. Then 10 ml MACS buffer was added and splenocytes were centrifuged for 5 min at 300 g. Cells were washed once with 10 ml MACS buffer (5 min, 300 g) and afterwards collected in MACS buffer. Expression of markers CD25 and FoxP3, within the subpopulation of CD4 + T cells, was determined by flow cytometry according to manufacturer's instructions (Miltenyi Biotech, Bergisch-Gladbach). Data were acquired using a MACSQuant Analyzer 10 (Miltenyi) and analyzed with FlowJo (FlowJo, LLC).
Quantification of TNF/TNFR2 complex clustering. BV-2 cells were stimulated with TNF muteins for 15 minutes at 37 °C. Then cells were immediately washed two times with ice cold PBS and fixed with 4% PFA in PBS solution. Then unspecific binding sites were blocked with 4% BSA in PBS and cells were incubated with antibodies against TNF (HP8001, Hbt) and TNFR2 (AF-426-PB, R&D systems), followed by detection with appropriate fluorescence labeled antibodies. Nuclei were counterstained with DAPI. Stainings were analyzed on a Zeiss Axio Observer Spinning Disc microscope equipped with a Plan-Apochromat 63x/1.4 Oil DIC objective and an Axiocam 503 mono CCD camera. The following excitation lasers and emission filters were used: DAPI: 405 diode laser, 450/50 nm filter; GFP, 488 nm diode laser, 525/50 nm filter; RFP, 561 nm (RFP) diode laser, 600/50 nm filter. Z-stacks of tile regions containing 6 × 6 images were acquired and maximum intensity projections were calculated. Image processing was done in Zen blue 2.1 software (Zeiss, Germany).
Quantitative image analysis was done with CellProfiler version 2.2 51 . Nuclei were segmented via the DAPI staining and a 120 pixel wide ring mask was drawn around each nucleus representing the cell mask. TNF Vesicles were segmented under the cell mask using the A488 staining and unified for each cell. Mean intensity of the TNFR2 signal using the A546 staining was measured under the unified vesicles representing the grade of co-localized vesicles between TNFR2 (red) and TNF (green) per cell.