IL15-Based Trifunctional Antibody-Fusion Proteins with Costimulatory TNF-Superfamily Ligands in the Single-Chain Format for Cancer Immunotherapy

Strategies interfering with the regulatory network of the immune response are intensively pursued for cancer therapy (1). Beside the development of checkpoint inhibitors, antagonizing immune inhibition, increasing attention is now being paid to the development of agonists that directly promote and drive the immune response. This involves in particular cytokines of the common-gamma chain cytokine family (e.g., IL2, IL15) and costimulatory members of the TNFR superfamily (e.g., 4-1BB, OX40, GITR; refs. 2, 3). IL15 is next to IL2 one of the most promising members of the common-gamma chain cytokine family for cancer therapy (4). It induces in particular the proliferation and differentiation of CD8⁺ T cells and supports the survival of CD8⁺ memory T cells. Furthermore it is involved in the generation, proliferation, and activation of natural killer (NK) cells. Unlike IL2, IL15 rather inhibits the activation-induced cell death (AICD) and has no essential influence on regulatory T cells (5). Under physiologic conditions, IL15 is bound to the IL15Rα chain and presented in trans to cells expressing the IL15Rβγ, like T cells and NK cells (6). Although IL15 is active in solution, binding to the IL15Rα chain enhances significantly its activity (7). Preclinical studies in mice showed that the antitumor effect of IL15 could be strongly increased when applied with an IL15Rα chain fragment as complex (IL15/IL15Rα-Fc) or fusion protein (IL15-IL15Rαsushi+; IL15(N72D)-IL15Rαsushi-Fc; refs. 8–11). In addition, tumor-directed antibody-fusion proteins with IL15 in the IL15Rα context resulted in enhanced antitumor effects in different mouse models (12–14). Furthermore, improved therapeutic effects could be achieved by combining IL15 with other approaches like chemotherapeutic agents, vaccines, adoptive cell therapy, checkpoint inhibitors (anti-PD-L1 anti-CTLA-4), and other cytokines (e.g., IL21, IL12), revealing its combinatory potential (15). Current clinical trials with IL15 involve mainly ALT-803 [IL15(N72D)-IL15Rαsushi-Fc] in phase I/II in patients with hematologic and solid cancers (16).

4-1BB, OX40, and GITR are among the most investigated costimulatory receptors of the TNFSF. Expressed on activated T and NK cells, all of them have shown antitumor potential in preclinical studies in mice (17). Costimulation by 4-1BB is fundamentally involved in the proliferation, differentiation, and survival of CD8⁺ T cells, potentiating the cytotoxic T-cell immune response. Its positive influence on NK-cell activity is also getting attention by enhancing the ADCC of tumor-directed therapeutic monoclonal antibodies (18). Other than 4-1BB, costimulation by OX40 appears less pronounced on CD8⁺ T cells and rather distinctive in enhancing the proliferation of CD4⁺ T cells and counteracting the function of regulatory T cells (19). In the tumor therapeutic context, also GITR was reported to play an important role in promoting the proliferation and function of effector T cells and to confer resistance to regulatory T-cell inhibition (20). Thus, these costimulatory TNFSF receptors constitute attractive immunotherapeutic targets, although there are still many unknowns how they compare mechanistically and which are the terms for their most effective application (21). Currently, agonistic antibodies (anti-4-1BB, anti-OX40, anti-GITR) and ligand-Fc fusion proteins (OX40L-Fc and GITRL-Fc) are being evaluated in phase I/II clinical trials in patients with solid tumors (18–20). So far, exploration of combinatory approaches with costimulatory TNFSF agonists included conventional strategies like chemotherapy, radiotherapy, and vaccination, as well as checkpoint inhibitors and other immunostimulatory monoclonal antibodies (17, 22, 23). Thus, there is evidence for the synergistic potential of combined immunostimulation, but also an increased risk of immune-related adverse events as result of the systemic administration of such agents (22).

The costimulatory ligands of the TNFSF are normally expressed as transmembrane proteins on antigen-presenting cells (APC). The functional unit is a homotrimer that forms by noncovalent interaction of the extracellular domain (24). In the case of 4-1BBL, OX40L, and GITRL, if the extracellular domain is shed, the soluble homotrimeric ligand loses receptor activation capacity. Taking advantage of this fact, tumor-directed antibody-fusion proteins with the extracellular domain of these TNFSF ligands have been generated that showed targeting-dependent activity, that is only antibody-mediated binding to the tumor cell led to cell surface presentation of the TNFSF ligand, mimicking the physiologically active state of the ligand (25–27). Furthermore, a trifunctional antibody-fusion protein with 4-1BBL and RD_IL15 was generated, where the effect of RD_IL15 on T-cell proliferation and IFNγ release was enhanced in targeting-dependent manner by the costimulation of the 4-1BBL component in vitro (28). In addition, this trifunctional fusion protein showed a stronger antitumor effect in a syngeneic lung tumor mouse model than the corresponding bifunctional antibody-fusion proteins. Since the trifunctional fusion protein presented as a homotrimeric, hence rather bulky molecule, with 3 antibody units, 3 IL15 units and 1 functional 4-1BBL unit per molecule, further improvement of the format was indicated.

Here, we report a further development of the trifunctional antibody-fusion protein with RD-IL15 and 4-1BBL. Introducing 4-1BBL in the single-chain format (3 consecutively fused 4-1BBL ectodomains; ref. 29) leads to a monomeric molecule presenting only 1 functional antibody, cytokine, and ligand unit, respectively. Nevertheless, comparison of the first and second format showed similar activity in soluble form and despite reduced antibody avidity, increased stimulatory capacity in target-bound form for the second generation trifunctional molecule. In addition, we incorporated scOX40L and scGITRL into this novel format and showed that the concept of the trifunctional antibody-fusion protein with RD_IL15 and sc4-1BBL could also be extended to other costimulatory members of the TNFSF. Trifunctional fusion proteins were in general more effective enhancing proliferation of T cells than the combination of corresponding bifunctional antibody-fusion proteins. Strongest combinatory effect increasing the degranulation of CD8⁺ T cells and the proliferation of CD4⁺ T cells was demonstrated for the trifunctional fusion protein with scGITRL. Further, in vivo experiments in a syngeneic lung tumor mouse model confirmed a stronger therapeutic effect for the tumor-directed trifunctional fusion protein in comparison to a nontargeted version or the combination of respective bifunctional antibody-fusion proteins.

Antibodies and recombinant proteins were purchased from Biolegends (anti-human CD3-PE, 317308; anti-human CD3-PerCP/Cy5.5, 317336; anti-human CD56-APC, 362504; anti-human CD107a-FITC, 328606; anti-human CCR7-PE, 353204; anti-human CD45RA-APC; 304112), Miltenyi Biotech (anti-human CD4-VioBlue, 130-097-333; anti-human CD8-PE/Vio770, 130-096-556; anti-His-PE, 130-092-691), KPL (anti-mouse IgG (H+L), 01-10-06), and R&D Systems (anti-human CD3ϵ, MAB100). Human IFNγ DuoSet ELISA Kit (DY285) and CellTrace CFSE (C34554) were obtained from R&D Systems and Life Technologies, respectively. Mouse melanoma B16-FAP cells (transfectants with human FAP) and B16wt cells were kindly provided by Prof. Klaus Pfizenmaier (Institute of Cell Biology and Immunology, University of Stuttgart) and were cultured in RPMI1640, 5% FBS and in case of B16-FAP supplemented with 200 μg/mL zeocin and passaged for 4 weeks after initiation of culture from stocks. CTLL-2 cells (IL2/IL15 growth-dependent mouse lymphocyte cytotoxic T lymphocyte), provided by Prof. Peter Scheurich (Institute of Cell Biology and Immunology, University of Stuttgart), were cultured for 2 weeks in RPMI1640, supplemented with 20% FBS, 10 mmol/l HEPES, 0.05 mmol/l β-mercaptoethanol, 1 mmol/l sodium pyruvate, nonessential amino acids, and 400 IU/mL rhIL2 (Immunotools). Cells were tested for mycoplasmas (Lonza; LT07-705) and their morphologic appearance was monitored by microscopy. Cell lines were not re-authenticated. Human peripheral blood mononuclear cells (PBMC) were isolated from buffy coats of healthy donors (Blood Bank) by density gradient centrifugation (LSM-1077; Promocell) and cultivated in RPMI1640, 10% FBS. C57BL/6N mice were purchased from Charles River.

The generation of scFv_4-1BBL (25), scFv_sc4-1BBL, scFv_scOX40L, scFv_scGITRL (29), scDbFAPxCD3 (26), RD_IL15_scFv, RD_IL15_scFv_4-1BBL, and RD_IL15_scFv_m4-1BBL (28) have been described previously. Trifunctional fusion proteins in the single-chain format were generated by replacing the individual extracellular domain of 4-1BBL in the RD_IL15_scFv_4-1BBL construct by the single-chain sc4-1BBL [aa 71-254, linkers (GGGGS)₄], scOX40L (aa 51–183, linkers GGGSGGG), scGITRL (aa 72–199, linkers GGSGGGGSGG), mscGITRL [aa 49–173, linker (GGGS)₅] in the backbone vector pSecTagA (Life Technologies). ScTNFSF ligands were synthesized by GeneArt (Life Technologies). Sequence of RD_IL15_scFv_FAP_scGITRL is indicated in Supplementary Fig. S1. RD_IL15_scFv_CEA_mscGITRL was obtained by replacing the FAP-directed scFv36 (30) by the CEA-directed scFvMFE-23 (31). All recombinant proteins were produced in transiently transfected HEK293-6E cells (NRC Biotechnology Research Institute, Canada) according to the standard protocol of the cell line provider. Supernatants were harvested 96 hours posttransfection, dialyzed, and recombinant proteins purified by immobilized metal ion affinity chromatography as described previously (26).

Purified protein was analyzed by size-exclusion chromatography on a Yarra SEC-3000 column (Phenomenex) and for thermostability by dynamic light scattering with the ZetaSizer Nano ZS (Malvern) as described in ref. 29.

A total of 2 × 10⁵ B16-FAP cells were incubated with the respective fusion protein for 1 hour at 4°C. Bound protein was detected by PE-conjugated anti-hexahistidyl-tag or anti-4-1BBL antibody, respectively. Fluorescence was measured by MACSQuant Analyzer10/VYB (Miltenyi Biotech) and data were analyzed using FlowJo (Tree Star).

A total of 2 × 10⁴ CTLL-2 cells/well were seeded in 96-well plates and starved in IL2-free medium for 4 hours before addition of the respective fusion proteins. After 3 days, cell proliferation was determined by measuring cell viability by MTT assay (12).

PBMCs were thawed and monocytes discarded by plastic adherence. The next day PBMCs were stained with carboxyfluorescein diacetate succinimidyl ester (CFSE) at 625 nmol/L/1 × 10⁶ cells/mL following the manufacturer's instructions. For analysis in nontargeted form, fusion proteins were incubated with 2 × 10⁵ PBMCs/well for 6 days. In targeted-form, at first, 2 × 10⁴ B16-FAP cells/well were seeded and arrested the next day by incubation with mitomycin C (10 μg/mL; Sigma) for 2 hours at 37°C. After washing, B16-FAP cells were incubated for 1 hour at room temperature with the respective fusion protein followed by an additional washing step before addition of suboptimal concentrations (5–50 ng/mL) of anti-CD3 mAb [crosslinked with goat anti-mouse IgG (H+L) in the ratio of 1:3] and 2 × 10⁵ CFSE-labeled PBMCs/well. After 6 days, immune cells of interest were labeled with fluorescence-conjugated antibodies directed against respective cell-surface markers and proliferation of defined cell populations measured by multicolor flow cytometry analysis.

A total of 2 × 10⁴ B16-FAP cells/well were seeded in 96-well plates. The next day cells were arrested with mitomycin C, washed, and incubate with the respective fusion proteins for 1 hour at RT before washing and addition of suboptimal concentration (50 ng/mL) of cross-linked anti-CD3 mAb and 2 × 10⁵ PBMCs/well. After 2 days, cell-free supernatants were harvested and IFNγ concentration was determined by sandwich ELISA following the instruction of the manufacturer's protocol.

B16-FAP cells targeted with fusion protein were preincubated with a suboptimal concentration (8 ng/mL) of crosslinked anti-CD3 mAb and PBMCs as described previously for the IFNγ release assay. After 5 days, prestimulated PBMCs were transferred to freshly seeded B16-FAP cells and incubated for additional 6 hours in presence of 30 pmol/L bispecific antibody scDbFAPxCD3. Next, PBMCs were harvested and degranulation of T cells determined via CD107a measurement by flow cytometry (CD107a-FITC/CD4-VioBlue/CD8-PEVio770/CD3-PE).

Animal care and in vivo experiments were performed in accordance with Federal and European guidelines and have been approved by Institutional Animal Care and Use Committee and state authorities. Therapeutic potential of the fusion proteins was determined in a syngeneic B16-FAP lung tumor mouse model. Female C57BL/6N mice (34 weeks, 23–34 g) were injected intravenously with 8.5 × 10⁵ B16-FAP cells/mouse on day 0. Treatment of the study groups (n = 6) with the respective fusion proteins (0.02 or 0.2 nmol/day) or PBS as a control was administered intraperitoneally on day 1, 2, and 10. Mice were sacrificed on day 21, lungs removed, fixed in Fekete solution and tumor foci counted.

Unless stated otherwise, all data are represented as mean ± SD of 3 independent experiments. Blockshift correction (bar graph) was performed as described in ref. 29. One-way ANOVA followed by Tukey post hoc test was used to determine statistical significance (Graphpad Prism). P < 0.05 was considered statistically significant.

The trifunctional antibody-fusion protein RD_IL15_scFv_FAP_4-1BBL had been described previously (28). In the initial format, the FAP-directed antibody in the single-chain Fv (scFv) format is fused N-terminally to human IL15 connected to part of the human IL15Rα (aa 31–107; RD) and C-terminally to the extracellular domain of 4-1BBL (aa 71–254). Noncovalent intermolecular trimerization mediated by the TNF-superfamily member 4-1BBL leads to the formation of a homotrimeric molecule presenting 3 antibody units, 3 RD_IL15 units, and 1 trimeric 4-1BBL unit (Fig. 1A). To reduce the complexity and size of the molecule, a second format was generated, where the single extracellular domain of 4-1BBL was replaced by 3 consecutive extracellular domains of 4-1BBL connected by short amino acid linkers. Here, intramolecular trimerization of 4-1BBL can take place in the configuration of a monomeric molecule composed of 1 antibody unit, 1 RD_IL15 unit, and 1 trimeric 4-1BBL unit (Fig. 1A). SDS-PAGE analysis of the purified fusion proteins showed the corresponding bands correlating to the calculated molecular mass of RD_IL15_scFv_FAP_4-1BBL monomer (71.1 kDa) and RD_IL15_scFv_FAP_sc4-1BBL (112.6 kDa), respectively (Fig. 1B). Size-exclusion chromatography indicated for RD_IL15_scFv_FAP_4-1BBL, a single peak at approximately 233 kDa, and for RD_IL15_scFv_FAP_sc4-1BBL, a main peak at approximately 172 kDa accompanied by a minor peak at 374 kDa (Fig. 1C). Here, the N-glycosylation predicted for IL15 and the heterogeneous domain composition of the fusion proteins can be expected to influence the elution pattern, accounting for a general increase in apparent molecular size. Analysis of thermostability by dynamic light scattering showed for both formats a melting point at 47°C (Fig. 1D). However, subsequent aggregation was delayed for the single-chain format (RD_IL15_scFv_FAP_sc4-1BBL), pointing to an increased stability by the sc4-1BBL component. Functional analysis showed that antibody-mediated binding was stronger for RD_IL15_scFv_FAP_4-1BBL (EC₅₀ = 0.4 ± 0.1 nmol/L) compared with RD_IL15_scFv_FAP_sc4-1BBL (EC₅₀ = 1.7 ± 0.1 nmol/L) as expected by the difference in antibody units/molecule (Fig. 2A; Supplementary Fig. S2A). On the other hand, IL15 activity was rather similar for both formats in soluble form, stimulating the proliferation of the cytokine growth-dependent mouse CTLL-2 cell line (IL15Rαβγ; Fig. 2B) and resting human T cells (IL15Rβγ) in concentration-dependent manner (Fig. 2C; Supplementary Fig. S2B). For the latter, both formats showed to be slightly less effective (approximately by factor 3.5) than RD_IL15_scFv, indicating that fusion to 4-1BBL in either format hindered to some extent the cytokine activity in solution. In contrast, in target-bound form, the trifunctional antibody-fusion proteins displayed equal or even stronger activity than the targeted bifunctional RD_IL15_scFv. Here, the RD_IL15_scFv_FAP_sc4-1BBL showed clearly better properties than RD_IL15_scFv_FAP_4-1BBL in enhancing the proliferation (Fig. 2D; Supplementary Fig. S2C) and IFNγ release (Fig. 2E) of CD3-stimulated T cells. Thus, in the new format antibody-mediated binding with reduced avidity led to cytokine and single-chain ligand presentation that showed to be not only sufficient, but even of improved efficacy in vitro. Furthermore, the antitumor effect of both formats was analyzed in vivo in an immunocompetent lung tumor mouse model. For this purpose, corresponding mouse compatible trifunctional fusion proteins were generated by replacing human with mouse 4-1BBL. B16-FAP tumor cells were injected intravenously, followed by fusion protein treatment intraperitoneally on day 1, 2, and 10. On day 21, lungs were removed and tumor lesions counted. Because this is an aggressive, fast growing tumor model, the treatment schedule was expected to promote the antitumor immune response at an early time point after tumor cell inoculation and reinforce it to a later time point to achieve maximal effect in the 3-week time frame. Here, a strong reduction in lesions of over 50% was achieved by the fusion proteins in both formats, confirming their antitumor potential (Fig. 2F).

Next, the concept of trifunctional antibody-fusion proteins with RD_IL15 and 4-1BBL was extended to other costimulatory members of the TNF superfamily. OX40L and GITRL were selected because of their well-documented antitumor potential (20, 32) and the fact that they had already been successfully converted into the single-chain format and incorporated into bifunctional antibody-fusion proteins (29). Considering the favorable in vitro data obtained for RD_IL15_scFv_FAP_sc4-1BBL, the single-chain format was maintained and in analogy the trifunctional antibody-fusion proteins RD_IL15_scFv_FAP_scOX40L and RD_IL15_scFv_FAP_scGITRL generated. All fusion proteins were produced in HEK293-6E cells and purified by IMAC. SDS-PAGE analysis under reducing conditions revealed bands of approximately 124 kDa for RD_IL15_scFv_FAP_scOX40L (MW 99.1 kDa) and 105 and 93 kDa for RD_IL15_scFv_FAP_scGITRL (MW 96.7 kDa) consistent with the fusion proteins being glycosylated (Fig. 3A). Size-exclusion chromatography showed similar patterns for both fusion proteins with a main peak at 197 kDa and a minor peak at 361 kDa for RD_IL15_scFv_FAP_scOX40L and a main peak at 174 kDa and a minor peak at 319 kDa for RD_IL15_scFv_FAP_scGITRL (Fig. 3B). Thus, a tendency to dimerization was observed for all trifunctional fusion proteins to a comparable degree. Antibody-mediated binding was demonstrated on B16-FAP cells for all fusion proteins with EC₅₀ values of 3.3 ± 1.2 nmol/L (RD_IL15_scFv), 4.0 ± 0.5 nmol/L (RD_IL15_scFv_scGITRL) and 11.6 ± 4.3 nmol/L (RD_IL15 _scFv_scOX40L) (Fig. 3C; Supplementary Fig. S2D). No binding was detected on FAP-negative B16 cells, confirming target-binding specificity. Also adequate cell surface presentation of the costimulatory ligand was demonstrated for the target-bound form of the fusion protein by flow cytometry analysis, detecting bound fusion protein via the corresponding recombinant costimulatory receptor-Fc protein (Supplementary Fig. S3). IL15 activity was shown for the fusion proteins in solution, inducing the proliferation of the cytokine growth-dependent mouse cell line CTLL-2 (human IL15 is cross-reactive, whereas human OX40L and GITRL are not cross-reactive with the respective mouse receptors). Here, RD_IL15_scFv_scGITRL showed to be similar and RD_IL15_scFv_scOX40L 7-fold less active than the bifunctional RD_IL15_scFv (Fig. 3D). Also, a reduced effect on PBMC proliferation was observed for nontargeted RD_IL15_scFv_scGITRL and RD_IL15_scFv_scOX40L in comparison to RD_IL15_scFv (Fig. 3E; Supplementary Fig. S2E). However, in target-bound form, similar to what we observed for RD_IL15_scFv_sc4-1BBL, trifunctional RD_IL15_scFv_scOX40L and RD_IL15_scFv_scGITRL were more efficient at higher concentrations enhancing the proliferation of CD3-stimulated T cells than the bifunctional RD_IL15_scFv. No proliferative effect of the fusion proteins was observed in absence of CD3 stimulation (Fig. 3F; Supplementary Fig. S2F). Thus, not only 4-1BBL, but also GITRL and OX40L could be successfully incorporated and combined with RD_IL15 in a trifunctional antibody-fusion protein in the single-chain format. The functional properties of the individual components were conserved, emphasizing the advantages of the target-bound form in enhancing the immune stimulation of T cells.

The immunomodulatory potential of combining RD_IL15 with either 4-1BBL, OX40L, or GITRL was further evaluated by comparing the activity of the trifunctional antibody-fusion protein with co-applied bifunctional antibody-fusion proteins. For the latter, RD_IL15_scFv, scFv_sc4-1BBL, scFv_scOX40L, and scFv_scGITRL were used, where the antibody-moiety and the orientation of the antibody-fusion were the same as in the trifunctional antibody-fusion protein. Equimolar amounts of cytokine and costimulatory ligand, respectively, were applied. Fusion proteins were targeted to B16-FAP cells and cocultured with PBMCs in presence of suboptimal concentrations of anti-CD3 mAb. Proliferation of CD4⁺ T cells, CD8⁺ T cells, and NK cells was measured after 6 days by flow cytometry (Fig. 4; Supplementary Fig. S2G). In general, especially at lower concentrations, trifunctional antibody-fusion proteins achieved stronger effects on T-cell proliferation than the combination of corresponding bifunctional antibody-fusion proteins, independent of the combination partner. This was particularly evident on CD4⁺ T cells were strongest signal enhancement was obtained by the trifunctional fusion proteins with OX40L (Fig. 4E) and GITRL (Fig. 4H). CD8⁺ T cells on the other hand are already highly responsive to RD_IL15, thus further enhancement by additional costimulation in form of trifunctional molecules resulted in only minor improvements of the effect (Fig. 4A, D, G). NK-cell proliferation was induced by RD_IL15, either as bifunctional or trifunctional molecule, where the combination with 4-1BBL or GITRL in a single molecule showed some favorable potential at low concentration (10 nmol/L; Fig. 4C, I), whereas the combination with OX40L showed no benefit (Fig. 4F). Further T-cell subset analysis showed that naïve, central memory, effector memory, and effector CD4⁺ and CD8⁺ T cells were responsive to the trifunctional fusion proteins, enhancing their proliferation and inducing changes in the composition increasing the proportion of effector memory cells (Supplementary Fig. S4). Enhanced activation of regulatory T cells was not detected (Supplementary Fig. S5).

Analyzing the degranulation of CD8⁺ T cells (Fig. 5A–C), trifunctional fusion proteins appeared as strong stimulators, where the combinatorial effect was most pronounced for RD_IL15_scFv_scGITRL (Fig. 5C). A similar strong effect was achieved also by RD_IL15_scFv_sc4-1BBL, nevertheless here the degranulation potential of 4-1BBL was clearly dominant in the configuration, because comparable effects were obtained by the bifunctional scFv_sc4-1BBL only (Fig. 5A; Supplementary Fig. S2H). Interestingly, RD_IL15_scFv_sc4-1BBL was also effective in enhancing the degranulation of CD4⁺ T cells (Fig. 5D), a phenomenon that was not observed for RD_IL15_scFv_scGITRL or RD_IL15_scFv_scOX40L (Fig. 5E, F). Thus, the combination of RD_IL15 with different costimulatory members of the TNF-superfamily ligands resulted in different immune cell stimulation pattern. Trifunctional antibody-fusion proteins showed here the advantage of producing the strongest effects by reflecting not less than the individual dominant cytokine/ligand activity and further enhancing the overall activity in most cases. On the contrary, combination of bifunctional antibody-fusion proteins were clearly less effective in enforcing the stimulatory signal and the effect could even fall behind of that achieved by the individual dominant cytokine/ligand.

RD_IL15_scFv_scGITRL emerged as interesting candidate, because it did not only show strong stimulatory potential on CD8⁺ T-cell and NK-cell proliferation, but also distinguished in its trifunctional format enhancing the degranulation of CD8⁺ T cells and the proliferation of CD4⁺ T cells. Thus, a mouse compatible version of the fusion protein was generated for further in vivo studies (Supplementary Fig. S6). Because mouse GITRL is reported to present as dimer (33), human scGITRL was replaced by a mouse scGITRL composed of only 2 extracellular GITRL domains. In the B16-FAP lung tumor mouse model, administration of the trifunctional antibody-fusion protein led to a stronger antitumor effect than the application of the combination of the corresponding bifunctional antibody-fusion proteins (Fig. 6). Furthermore, targeting demonstrated to be an important factor to achieve the antitumor effect of the antibody-fusion protein, because a trifunctional antibody-fusion protein directed against the carcinoembryonic antigen (CEA) which is not expressed in mice, was clearly less effective in reducing the number of lung tumors. Thus, further proof of concept for the trifunctional antibody-fusion protein approach was provided in vivo, identifying RD_IL15_scFv_scGITRL as promising candidates for further development.

In this study, we present the further development of IL15-based trifunctional fusion proteins with costimulatory TNF-superfamily members. In comparison with the initial format, we observed that the reduction of functional antibody units and RD_IL15 units from 3 to 1 did not translate into reduced cytokine/ligand activity. On the contrary, the activity was either similar in soluble form (Fig. 2B, C) or even improved in target-bound form (Fig. 2D, E). In solution, where the cytokine activity of the fusion protein is attributed to the RD_IL15 component, the effect on CTLL-2 (IL15Rαβγ) and PBMCs (IL15Rβγ) was not further influenced by the RD_IL15 valence. This is in accordance with reports of other antibody-fusion protein formats, where bivalent IgG_IL15_RD and monovalent IL15_RD were similar effective inducing the proliferation of Kit225 cells (IL15Rαβγ/IL15Rβγ; refs. 13, 14) and also monovalent scFv_RD_IL15 and bivalent Db_RD_IL15 demonstrated comparable capacity to induce proliferation of PBMCs (Supplementary Fig. S7). Thus, localizing more than 1 RD_IL15 unit into a single fusion protein seems not to be required or advantageous to increase the apparent cytokine activity of the molecule. Fusing the costimulatory ligand (4-1BBL or sc4-1BBL) to RD_IL15_scFv did not further enhance, but rather inhibit the cytokine activity of the soluble trifunctional fusion protein, pointing to an inactive form of the costimulatory ligand in solution. Only after antibody-mediated binding and thereby cell surface presentation of the ligand, its activity was recovered and could converge with the activity of RD_IL15. Previously, we have reported targeting-dependent activity of bifunctional antibody-fusion proteins with costimulatory members of the TNF superfamily in both formats (27, 29). Introducing the single-chain format into TNFSF ligands resulted not only in fully functional homotrimeric units, but also contributed to stabilize them. Importantly, cell surface presentation of the ligand via a single antigen-binding unit was sufficient for activity display. Here, we showed for the first time that the single-chain TNFSF format can also be effectively introduced into the design of trifunctional IL15-based antibody-fusion proteins thereby conserving its particular functional properties in soluble and target-bound form. It could be inferred that high antibody valency is not required and does not directly relate to improved ligand activity. The presence of redundant antigen-binding units might here rather sterically interfere with the optimal presentation of the ligand, which could explain the improved efficacy of the antibody-fusion proteins with the ligand in the single-chain format.

Other than RD_IL15, TNFSF members are known to gain activity by oligomerization (34, 35). All trifunctional and bifunctional antibody-fusion proteins with scTNFSF showed a similar degree of dimerization that could be attributed to the antibody component, because the scFv format is prone to partial dimer formation (36). Thus, the presence of a fraction of hexameric ligands could eventually contribute to enhance the costimulatory effect, especially in solution. Analysis of the soluble bifunctional fusion proteins (scFv_scTNFSF) showed that 4-1BBL was prone to certain activity in solution, whereas OX40L and GITRL were rather not (29). In addition, there is evidence from other studies that a hexameric configuration of OX40L and GITRL in solution might not be per se sufficient to induce relevant receptor activation. Recent studies with Fc-OX40L (MEDI6383) and Fc-GITRL (MEDI1873), where the TNFSF ligand is fused to the C-terminus of an Fc fragment leading to an hexameric ligand configuration covalently enforced by the incorporation of a trimerization domain, showed only a minimal costimulatory activity on T cells for the soluble form. Strong activity was only achieved upon Fc receptor-dependent cell surface presentation or cross-linking (37, 38). Indeed, we have isolated the monomeric fraction of RD_IL15_scFv_scGITRL (trimeric ligand) by FPLC and observed that the activity on T-cell proliferation in soluble and target-bound form was rather enhanced in comparison to the isolated dimer fraction (hexameric ligand) or the protein preparation containing the additional dimer fraction, respectively (Supplementary Fig. S8). Thus, partial dimer formation is here not expected to enhance the targeting-independent activity and the associated risk of systemic toxicity.

In general, in target-bound form, the trifunctional fusion proteins induced the strongest immune stimulation and showed combinatory benefit. Combination of respective bifunctional fusion proteins was less effective, apparently hampered by antibody-mediated competition for the same target on the cell surface. The effect of the single bifunctional fusion proteins could be outranged to different degree depending on the composition of the particular trifunctional molecule. Thus, simultaneous targeting of IL15 and TNFSF ligand to the tumor cell offers different quality of stimulation/costimulation at the same time and location, where the impact of such individual or cooperative activity varies on different immune cell subsets. Here, IL15 and 4-1BBL showed both strong impact on the proliferation and the cytotoxic potential of CD8⁺ T cells, expanding in particular the effector memory population (Supplementary Fig. S4). Under physiologic conditions, cooperation of IL15 and 4-1BBL are assumed to play an important role in the homeostasis of CD8⁺ memory T cells, in particular of the effector memory phenotype, where IL15 induces the expression of 4-1BB on memory CD8⁺ T cells which then are maintained by 4-1BBL stimulation (39). In addition, this population has been shown to be the predominant phenotype of tumor-infiltrating T cells in diverse primary tumors and metastases (40–42). Also, 4-1BB was shown to be upregulated on tumor-infiltrating lymphocytes (TIL) in hypoxic conditions of the tumor microenvironment (43) and has been proposed as biomarker to identify and enrich naturally occurring tumor-reactive T cells for adoptive transfer with the support of IL15/IL7 for ex vivo expansion (44). Furthermore, the antitumor effects achieved by the trifunctional antibody-fusion proteins with RD_IL15/4-1BBL of the first and second generation in the syngeneic lung tumor mouse model (Fig. 2F; ref. 28) point to 4-1BBL and IL15 as promising combination partners for cancer immunotherapy. A strong individual record as immunotherapeutic target has also been reported for GITR and OX40 (19, 20), but comparative analysis to elucidate corresponding mechanisms and application advantages are still rare. Here, in the trifunctional antibody-fusion protein approach, the combination of GITRL with IL15 was especially advantageous demonstrating synergistic activity enhancing the degranulation of CD8⁺ T cells and the proliferation of CD4⁺ T cells in vitro. This is consistent with the observation that IL15 and GITR, but not OX40, are related in a homeostatic context were IL15 induces the upregulation of GITR on CD8⁺ memory T cells (39). Furthermore, both molecules have shown independently the capacity to induce primarily the proliferation and cytotoxic function of CD8⁺ T cells and NK cells, leading to enhanced antitumor response (16, 45). In addition, the therapeutic effect of GITR agonists is also partly related to their effect on CD4⁺ T cells, including expansion of effector helper cells and prevention of Treg suppression (46, 47). Hence, a comprehensive and versatile promotion of an antitumor immune response could be expected by this combination. Interestingly, colocalization appears here advantageous, because stronger antitumor effects were achieved by the RD_IL15_scFv_scGITRL fusion protein in comparison to the combination of respective bifunctional antibody-fusion proteins in the mouse model (Fig. 6). Also, tumor-directed targeting of the fusion protein was required for maximal antitumor effect confirming the relevance of antibody-mediated targeting for this approach. Thus, the second generation of trifunctional antibody-fusion proteins constitutes a promising platform for the development of novel potential immunotherapeutics like RD_IL15_scFv_scGITRL.

No potential conflicts of interest were disclosed.

Conception and design: D. Müller

Development of methodology: N. Beha

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Harder, S. Ring

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N. Beha, M. Harder, S. Ring, D. Müller

Writing, review, and/or revision of the manuscript: N. Beha, M. Harder, R.E. Kontermann, D. Müller

Study supervision: D. Müller

We thank Dr. Oliver Seifert for the assistance in the animal experiments and Doris Göttsch for FPLC performance. We thank Dr. Beate Luz (Katharinenhospital) for providing buffy coats. This work was supported by a grant of the Deutsche Forschungsgemeinschaft: MU2956/4-1 (to D. Müller).

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.