A Multispecific Anti-CD40 DARPin Construct Induces Tumor-Selective CD40 Activation and Tumor Regression

Malignant tumors harness a variety of immunosuppressive mechanisms to evade the immune system, allowing for their growth (1). A broad range of monoclonal antibodies (mAb) have been designed to impair this tumor immune escape process by modulating immunoregulatory molecules expressed by innate and adaptive immune cells. In particular, mAb inhibiting the negative T-cell checkpoint regulators CTLA-4 and PD-1 have had impressive clinical success in the treatment of cancer patients (2–4). Other targets of investigational drugs for cancer immunotherapy include members of the tumor necrosis factor receptor superfamily, such as CD137 (4-1BB), GITR, and CD40. The CD40 receptor is expressed by dendritic cells (DC), myeloid cells, and B cells. Binding to the receptor by its natural ligand, CD40L, expressed by helper T cells, results in activation of CD40⁺ cells. Activation of CD40 by agonistic mAb has a potential to convert immunologically “cold” tumors into “hot,” sensitizing them to subsequent therapy with T-cell checkpoint inhibitors (5).

Previous clinical studies with systemic administration of low doses of agonistic CD40 mAb have shown some pharmacodynamic effects, but so far, the pharmacologic activity has not yet translated to a meaningful clinical response, most likely due to the limited therapeutic window (6).

The dose-limiting toxicity, grade 3 and 4 adverse events at the 0.2 mg/kg dose level, seen for some agonist CD40 mAb (7, 8), might be due to systemic activation of CD40 signaling. It is likely that most CD40 mAb bind to the Fc receptor (FcR) expressed on immune cells, endothelial cells, and platelets, resulting in a high degree of clustering of CD40 receptors, leading to systemic activation (9, 10) and possibly to systemic toxicity. To this day, no CD40-agonistic antibody has been approved for cancer therapy, but several are in development, such as selicrelumab (8), sotigalimab (11), and mitazalimab (12).

To overcome this potential issue and substantially improve the therapeutic index of CD40-targeting agents, we designed a construct capable of activating CD40 independently of FcR-mediated cross-linking. We developed a multispecific designed ankyrin repeat protein (DARPin) therapeutic candidate, termed α-FAPxCD40, designed to activate the CD40 receptor in the tumor microenvironment (TME), through cross-linking via stromally expressed fibroblast activation protein (FAP), rather than systemically. FAP is abundantly expressed by cancer-associated fibroblasts of many types of solid tumors, as well as by malignant cells in a subset of human sarcomas, but has very limited expression in healthy tissues (13, 14). With this tumor-focused approach, the α-FAPxCD40 DARPin construct could also overcome other limitations connected to Fc dependency of targeted mAb therapies. For instance, patients carrying certain allelic variants of FcR with relatively low affinities for the Fc domain show lower responses to various therapeutic antibodies (15, 16).

The design and the characterization of the α-FAPxCD40 DARPin therapeutic candidate is described here. These studies explore in vitro the FAP-dependent activation of different immune cell populations with the α-FAPxCD40 construct and in turn show how this translates in vivo into α-mFAPxCD40-induced rejection of FAP-positive tumors without signs of systemic toxicity.

The preclinical findings presented here support the development of the α-FAPxCD40 DARPin molecule as a therapeutic candidate, acting selectively in TME with an Fc-independent mechanism of action against FAP-positive tumors.

All antibodies used in the study are listed in Supplementary Table S1.

α-FAPxCD40 is a chain of four sequentially positioned covalently linked DARPin domains composed of a previously described N-terminal anti-human serum albumin (HSA) binder (H) for half-life extension (17), an anti-human FAP domain (F) for tumor localization and clustering and two identical anti-human CD40 domains (C–C) for immune activation, H–F–C–C. α-mFAPxCD40 is a chain of four covalently linked DARPin domains with the same molecular geometry as α-FAPxCD40, but specific for the respective mouse targets. Amino acid sequences of both human and mouse molecules can be found in Supplementary Sequence Data Files S1 and S2. Different controls were generated: (i) the negative control (neg-CTRL), which binds HSA but with both the FAP- and CD40-binding domains replaced with DARPin domains, which do not bind any antigen (consensus DARPin Ni2C, H–X–X–X; (ii) the FAP-targeting control (FAP-CTRL) where the anti-FAP DARPin is replaced with a nonbinding DARPin, H–X–C–C; and (iii) the CD40-targeting control (CD40-CTRL) where the two anti-CD40 DARPin domains are replaced with two nonbinding DARPins, H–F–X–X.

The α-CD40 mAb is a human CD40 agonist IgG2 monoclonal antibody produced by Evitria using the sequence of selicrelumab adopted from the US 7,338,660 B2 patent (sequence 21.4.1). The α-mCD40 mAb is a mouse CD40 agonist monoclonal rat IgG2a antibody, clone FGK45, obtained from Bio X Cell.

CHO (CCL-61), WM266-4 (CRL-1676), WI-38 (CCL-75), Panc02 (CRL-2553), and U-87 MG (HTB-14) were obtained from ATCC. All the cell lines were cultured according to the supplier instructions. MC38 and MB49 cells were obtained from the University of Basel and cultured in DMEM (Gibco) supplemented with 10% FBS (Gibco) in 5% CO₂. Cell lines were not authenticated. FAP-expressing CHO cell lines were generated using a plasmid containing the ORF of human FAP (OriGene Technologies; RG204692). Expression of FAP was analyzed by flow cytometry using the anti-FAP ESC11 (Evitrea). Two different CHO-FAP clones (c), CHO-FAP c1.9 and CHO-FAP c1.7, were obtained showing different FAP expression levels. If not indicated otherwise, the data shown in this article were generated using CHO-FAP c1.7. MC38 cells were transfected with a plasmid encoding murine FAP (MC38-FAP). The sequence of the Fap gene can be found in Supplementary Sequence Data File S3. Mouse tumor cells were tested to be free of murine viruses at IDEXX BioAnalytics, Germany. All cell lines were typically passaged 5 to 10 times between thawing and experimental use. All cells were cultured between March 2016 and February 2021. All cells were tested to be free of Mycoplasma by the PlasmoTest Mycoplasma Detection Kit (InvivoGen) and from September 2020 by the MycoAlert Mycoplasma Detection Kit (Lonza).

DARPin modules were raised against HSA, FAP, and CD40 starting from naïve DARPin libraries using four to six ribosome display selection rounds (18). The individual DARPin modules described in the article were screened from the ribosome display pools using high-throughput homogeneous time-resolved fluorescence assays for their binding to HSA, FAP, and CD40, respectively. α-FAPxCD40 and α-mFAPxCD40 were generated by DNA synthesis or standard cloning methods. Single-domain and multidomain DARPin molecules carrying an N-terminal 6xHistidine tag were expressed in Escherichia coli and purified by affinity chromatography followed by a size-exclusion chromatography (Aekta Xpress purification system, His Trap FF Crude from Cytiva and HiLoad 26/600 Superdex 200 from GE Healthcare; ref. 19).

FAP-binding kinetic determination: Human FAP ECD was produced in insect cells at the CRO Reliatech, based on the sequence from UniprotKB Q12884, and immobilized on a GLC chip (Bio-Rad 176-5011, 1200 RU). α-FAPxCD40 was applied as analyte at concentrations of 50, 25, 12.5, 6.25, and 3.13 nmol/L with an association period of 180 seconds and a dissociation period of 1,800 seconds using a constant flow of 100 µL/min. The signals were double-referenced against the PBS plus Tween 20 (PBST)-treated control lanes. CD40-binding kinetic determination: Biotinylated human CD40 was immobilized on an NLC chip (Bio-Rad) to levels of 300 RU. α-FAPxCD40 was applied as analyte at concentrations of 3, 1.5, 0.75, 0.38 and 0.19 nmol/L with an association period of 180 seconds and a dissociation period of 900 seconds using a constant flow of 100 µL/min. The signals were double-referenced against the PBST-treated control lanes. HSA-binding kinetic determination: α-FAPxCD40 was immobilized on an FAP-coated GLC chip to a level of 120 RU, and HSA was applied as analyte at concentrations of 100, 33, 11, and 3.2 nmol/L with an association of 120 seconds and dissociation of 600 seconds using a constant flow of 100 µL/min. The signals were double-referenced against the PBST running buffer and nonrelevant target to which α-FAPxCD40 was not binding. The kinetics were calculated on the first 300 seconds of dissociation.

Size-exclusion chromatography–multiangle light scattering (SEC- MALS): α-FAPxCD40 was run at a concentration of 2 mg/mL on an Agilent 1200 HPLC system. The MALS signal of the separated sample was recorded using a Wyatt miniDAWN TREOS online detector. UV signals were detected at 280 nm with 320 nm as reference wavelength.

For the thermal denaturation experiment, α-FAPxCD40 was diluted to 0.25 µmol/L in PBS pH 7.4 in 117.100F-QS cuvettes with in-solution temperature control sensors. The circular dichroism (CD) signal was recorded at 222 nm (nitrogen flow 3 L/min) during a heating phase followed by a cooling phase. The CD signal was normalized to mean residual ellipticity with the unit deg cm²/dmol.

α-FAPxCD40 aggregation onset was measured by dynamic light scattering at a concentration of 1 mg/mL in PBS (triplicate) using a Wyatt DynaPro Plate Reader instrument. Paraffin oil was added to prevent evaporation. The applied heat ramp ranged from 25°C to 85°C with a rate of 0.5°C/min. Every well was assessed using four single light scattering acquisitions of 5 seconds each.

For crystallization, the complex of the untagged and deglycosylated CD40 (aa 21–193) together with the untagged anti-CD40 DARPin domain was produced to a final concentration of 36.7 mg/mL in 10 mmol/L HEPES/NaOH pH 7.0 and 150 mmol/L NaCl.

Crystals of the DARPin–CD40 complex were obtained at 277 K by the sitting drop vapor diffusion method. Data collection was performed with 20% (v/v) ethylene glycol as cryo-protectant at the Swiss Light Source (SLS) using cryogenic conditions. The diffraction data were processed with the programs autoPROC (20), XDS (21), and AIMLESS of the CCP4 program suite (22). The phases were obtained by molecular replacement using the solved CD40 crystal structure (PDB 3QD6) as search model. The model building and refinement were performed with COOT (23) and REFMAC5 (CCP4 program suite, applying TLS refinement and NCS restraints). Water molecules were added with COOT in the peaks of the Fo-Fc map contoured at 3.0 followed by refinement with REFMAC5 and use of the validation tool of COOT and visual inspection. Two DARPin-CD40 complexes were located in the asymmetric unit: chains A(CD40):B(DARPin) and chains C(CD40):D(DARPin). The electron density of complex C:D was of lower quality compared with complex A:B, likely due to different crystallography contacts. As the interface of the two complexes displayed a very similar structure (RMSD: 0.258 Å, calculated on the Cα pairs of the DARPin domain and the CRD1 domain, aa 25–59, of CD40), the complex A:B was used as representative. All figures were generated with PyMol (Version 2.0 Schrödinger, LLC). Information regarding the data collection parameters and refinement statistics are indicated in Supplementary Table S2.

For crystallization, the N-terminal 6xHis-TEV sequence of the anti-CD40 DARPin domain was removed after purification using a 100-fold excess of His-tagged TEV protease (Sigma, T4455-10KU) in HBSG buffer (20 mmol/L HEPES, 10% glycerol, 300 mmol/L NaCl pH 7.5) with 1 mmol/L DTT and 5 mmol/L EDTA at 4°C (buffer prepared at Molecular Partners using chemicals from Roth AG, Thermo Fisher Scientific AG, and Promega AG). After the addition of 10 mmol/L MgCl₂, the cleaved DARPin molecule was recovered via negative Ni-NTA followed by SEC (Aekta Xpress purification system using the columns His Trap FF Crude from Cytiva and HiLoad 26/600 Superdex 200 from GE Healthcare) in HBSG. CD40(aa 21–193)-THB-8xHis was expressed in cell culture in the presence of tunicamycin at Proteros Biostructures GmbH. Protein from culture supernatant was purified via Ni-NTA in 50 mmol/L Tris-HCl pH 7.5, 375 mmol/L NaCl, and 10% glycerol and digested with thrombin in 50 mmol/L Tris-HCl pH 7.5, 375 mmol/L NaCl, 20 mmol/L imidazole, and 10% glycerol. The untagged protein was further purified via negative Ni-NTA as above followed by SEC in 10 mmol/L HEPES/NaOH pH 7.0 and 150 mmol/L NaCl. The purified protein was mixed at a 1:1.2 ratio with the anti-CD40 DARPin domain, the complex was purified via SEC equilibrated in 10 mmol/L HEPES/NaOH pH 7.0 and 150 mmol/L NaCl, and concentrated to 36.7 mg/mL.

Cells were resuspended in FcR-blocking solution (BD Bioscience, 564220), stained with antibodies (see Supplementary Table S1), resuspended in Live/Dead staining (ThermaFisher, L34957), fixed using BD Cell Fix solution, and analyzed by an Attune NxT flow cytometer. Raw data were analyzed using FlowJo software (version 10.0.3). Cells were gated on live and single cells (FSC-A vs. FSC-H) using Live-Dead discriminating dye followed by gating on lineage markers. The median fluorescence intensity (MFI) and percentage of cells positive for specific markers were exported and plotted using GraphPad prism software, version 8.1.2. Fluorescence minus one controls were used as reference for setting the gates. FAP quantification was performed with the flow cytometer methodology using the QIFKIT (Dako) bead assay according to the manufacturer's recommendations.

At the termination of in vivo efficacy experiments, tumors were cut into small pieces (about 10 mm³) and resuspended in dissociation buffer [RPMI-1640 with 1/50 Liberase TL (Sigma-Aldrich, 0.08–0.28 Wunsch units/mL, 05401020001), 1/250 DNase I (Sigma-Aldrich, ≥400 Kunitz units/mg protein, #9003-98-9, DNA25-1)], and processed in a gentleMACS Octo Dissociator (Miltenyi Biotec) to obtain single-cell suspensions. The cells were washed and filtered through 70-µm and 40-µm filters and stained as previously described. Samples were stained for live/dead, CD45, CD11b, CD11c, F4/80, CD206, and CD80 in order to analyze the myeloid population. For the T-cell panel, the cells were stained for CD45, TCR β, CD8, and CD4.

Buffy coats were obtained from the Zurich blood donation center, and peripheral blood mononuclear cells (PBMC) were isolated by density centrifugation using Leucosep tubes (VWR International AG, GREI227288_250). Human CD19⁺ B cells were enriched from PBMC using positive selection (MicroBeads Kit, 30-050-301) according to the manufacturer's recommendations. CD19⁺ B cells and FAP⁺ cells were seeded together at a 2:1 ratio with or without 600 µmol/L HSA in the presence of α-FAPxCD40 or α-CD40 mAb at a range of indicated concentrations. Cell cocultures were incubated for 24 hours at 37°C in 5% CO₂, and expression of activation markers on CD20⁺ B cells was assessed by flow cytometry.

Human CD14⁺ monocytes were isolated from PBMC using negative selection (Miltenyi Biotec, 130-091-153) and differentiated into DC using human IL4 (Miltenyi Biotec, 130-093-922; 100 ng/mL) and granulocyte macrophage–colony stimulating factor (GM-CSF; Miltenyi Biotec, 130-093-866; 200 ng/mL) for 5 days. Monocyte-derived DC (MDDC) and irradiated (3,000 Rad for 30 minutes with the irradiator Faxitron, Precision CellRad, 2328A50148) CHO-FAP or CHO-WT cells were seeded together at a 2:1 ratio in the presence of 600 µmol/L HSA (CSL Behring, 150570) and indicated amounts of α-FAPxCD40 or α-CD40 mAb. Cell cocultures were incubated for 48 hours at 37°C in 5% CO₂, and expression of costimulatory molecules on CD14^– and CD209⁺ MDDC was assessed by flow cytometry, while secretion of human IL12 in culture supernatants was assessed by ELISA (PeproTech, 900-K96).

Human CD14⁺ monocytes were isolated as described above and differentiated into macrophages (MΦ) in Macrophage Base Medium (PromoCell, C-28057) plus 100 ng/mL GM-CSF for 5 days. Monocyte-derived MΦ (MDMΦ) and irradiated (3,000 Rad for 30 minutes with the irradiator Faxitron, Precision CellRad, 2328A50148) CHO-FAP or CHO-WT cells were cocultured at a 2:1 ratio in the presence of 600 µmol/L HSA in Macrophage Base Medium plus 1% FBS (Corning, 36-C016-CV) and 100 ng/mL GM-CSF (Miltenyi Biotec, 130-093-866) with dose titration of α-FAPxCD40 or α-CD40 mAb. Cell cocultures were incubated for 48 hours at 37°C in 5% CO₂, and the secretion of different cytokines was assessed using a cytokine array kit (BioLegend, 740503).

Spleens from female C57BL/6JRj mice (Janvier, 8–12 weeks old) were gently smashed using a 70-µm filter (VWR International AG, 734-0003) to obtain single-cell suspensions. B cells were isolated from the single-cell suspension using a negative selection isolation kit (Miltenyi Biotec, 130-090-862) according to the manufacturer's instructions. B cells were seeded together with CHO-FAP or CHO-WT cells at a 10:1 ratio in the presence of indicated amounts of α-mFAPxCD40, α-mCD40 mAb, or appropriate controls. Cell cocultures were incubated for 48 hours at 37°C in 5% CO₂, after which the upregulation of activation markers on CD20⁺ B cells was assessed by flow cytometry.

DC were obtained by flushing the femur bone marrow from female C57BL/6JR mice with PBS. The cells were cultivated for 7 days in the presence of 25 ng/mL GM-CSF and 5 ng/mL IL4. Bone marrow–derived DC (BMDC) were then purified using a mouse pan dendritic cell isolation kit (Miltenyi Biotec, 130-100-875) according to the manufacturer's instructions. DC and irradiated CHO-FAP or CHO-WT were seeded together at a 10:1 ratio in the presence of α-mFAPxCD40, α-mCD40 mAb, or appropriate controls at indicated concentrations. Cell cocultures were incubated for 48 hours at 37°C in 5% CO₂, after which expression of activation markers on CD45⁺CD11c⁺ DC was analyzed by flow cytometry.

Bone marrow cells were cultivated for 6 days in the presence of macrophage-colony stimulating factor (M-CSF; R&D Systems 216-MC-025; 100 ng/mL). After 6 days, differentiated MΦ were cocultured in new media in the presence of irradiated CHO-FAP or CHO-WT cells at a 2:1 ratio, together with 100 nmol/L of α-mFAPxCD40, α-mCD40 mAb, or appropriate controls. The cocultures were incubated for 48 hours at 37°C in 5% CO₂, after which (i) expression of M1 and M2 markers on CD45⁺CD11b⁺F4/80⁺ MΦ was analyzed by flow cytometry and (ii) secretion of different cytokines was analyzed using a cytokine array kit (BioLegend, 740845).

Eight- to 10-week-old female C57BL/6JRj mice were obtained from Janvier and acclimatized for 1 week. Housing conditions and all the in vivo experiments of this article were performed according to the protocols approved by the Veterinary Authorities of Kanton Zurich according to Swiss law (protocols: Kantonale Tierversuchsnummer: 230-15, Federale Tierversuchsnummer: ZH102/16 27842, ZH174/19 31654, and 75641 33310). Mice were inoculated with tumor cells subcutaneously into the right hind flank/back region at day 0. Mice were randomized into treatment group when the tumors reached a mean volume above 300 mm³ for MC38-FAP tumors and about 100 mm³ for wild-type MC38 tumors (MC38-WT), and treated with the indicated reagents. The test compounds were dosed intraperitoneally (i.p.) and adjusted according to the body weight at 10 µL/G body weight. Tumor-bearing mice received three doses, every 3 to 4 days, of α-mFAPxCD40 (2.5 mg/kg), FAP-CTRL (2.5 mg/kg), CD40-CTRL (2.5 mg/kg), and α-mCD40 Ab (5 mg/kg). These doses were chosen based on previous preclinical studies using the same α-mCD40 Ab (clone FGK45), where the dose of 5 mg/kg, equal to 100 µg per injection, was used (24–27). With this regimen, α-mCD40 Ab (clone FGK45) has been shown to be effective, but toxic. DARPin doses were calculated to be equimolar to the antibody for an unbiased comparison. In the dose escalation experiment, doses between 0.005 and 50 mg/kg were used. Tumor growth was monitored every 3 to 4 days where length and width of the tumor were measured with calipers. The tumor volume was calculated with the formula: (length × (width)² × π)/6. Mice were carefully monitored twice per week for the presence of signs of stress. During routine monitoring, mice were checked for any alterations of their behavior, such as hunched posture, reduced grooming and ruffled fur, reduced level of spontaneous activity, reduced food and water intake, and any other abnormalities. At termination, mice were euthanized, and tumors and organs were removed for analyses.

Mice were subcutaneously inoculated into the right flank with 9×10⁶ of MC38-FAP cells. When tumors reached the size of about 500 mm³, mice were injected with approximately 6 MBq of indium-111–conjugated DARPin molecules, corresponding to 100 kBq per µg of DARPin molecules, into the tail vein.

Single-photon emission computed tomography (SPECT)/X-ray computed tomography (CT) images were performed with the NanoSPECT/CTPlus camera (version 1.2, Bioscan). Acquisitions of SPECT and CT were performed with the software Nucline (version 1.02). The CT was also reconstructed with the software Nucline, whereas for SPECT the software HISPECT was used (version 1.4.3049, Scivis GmbH). Fusions of SPECT and CT data were analyzed with the VivoQuant postprocessing software (Version 3.5, Invicro). Whole-body activity was measured in a gamma counter tube prior to imaging acquisition. SPECT/CT in vivo images were taken from anesthetized mice at time points 4, 24, 48, 72 and 96 hours after injection of ¹¹¹In-DARPin constructs. The 3D segmentation tool of the VivoQuant postprocessing software was used for quantification of the measured radioactivity in the tumors.

MC38-FAP tumor–bearing mice were treated as described above, injected with approximately 200 KBq of indium-111–conjugated DARPin molecules, corresponding to 4 kBq per µg of DARPin molecules, into the tail vein, and euthanized 4, 24, 48, 72 or 96 hours after injection. Organs of interest were dissected, weighed, and the radioactivity was determined as counts per minute (CPM) with a γ-counter (Packard Cobra II Gamma D5010). CPM values per analyzed organ were converted to μg DARPin molecule per organ (μg DARPin molecule/organ) using the total amount of DARPin molecule per mouse as reference value. The amount μg DARPin molecule/organ was converted to μg DARPin molecule per gram tissue (μg DARPin molecule/g tissue) and plotted in the graphs as a percentage of the injected dose per gram of tissue (% ID/g) and expressed as mean ± SD, using 4 mice per time point.

Tissue sections were stained using automated IHC Staining Platform Ventana Discovery Ultra (Roche). Briefly, deparaffined formalin-fixed paraffin-embedded (FFPE)-tumor slides were incubated for 2 hours at 37°C with anti-DARPin (for the antibody clones, see Supplementary Table S1). Anti-DARPin was labeled by HRP-conjugated anti-rabbit secondary antibody and visualized using ChromoMap DAB kit (Roche). Finally, the slides were counterstained with hematoxylin, dehydrated in a gradient of ethanol, and coverslips were applied to the slides. Stained slides were visually analyzed by a research pathologist using Leica DM1000 Microscope at 200× magnification. For the quantitative assessment of the DARPin molecule in the tumor tissue, the H-score method was used, considering both the intensity of the staining (0, negative; 1, weak; 2, moderate; 3, strong) and the percentage of positive cells (0–100%). Final H-score was calculated using the following formula: H-score = (% stained cells at 0) × 0 + (% stained cells at 1+) × 1 + (% stained cells at 2+) × 2 + (% stained cells at 3+) × 3. The H-score value ranges from 0 to 300.

For the FAP expression analysis, the FAP–antibody complexes were detected by the HRP system using the Omni Rabbit HRP Auto Dispenser (Roche Diagnostics) for 20 minutes at 37°C. The slides were scanned with Vectra Polaris and assessed qualitatively for the FAP expression. No quantitative assessment was done.

Mice from the antitumor efficacy experiments were bled from vena saphena 24 hours after the first injection into Multivette 600Z Gel (Sarstedt 15.1674). The blood was centrifugated, and the serum was frozen to −80°C until further analysis. All samples were thawed only once. Alanine aminotransferase (ALT) and aspartate aminotransferase (AST) analyses were performed using the MAK052 and MAK055 kits (Sigma-Aldrich), respectively, by following the manufacturer's guidelines.

Blood from tumor-bearing mice was taken via vena saphena 24 hours after the first injection as above. The indicated cytokines were then analyzed in serum using Luminex assays (R&D Systems) according to the manufacturer's recommendations.

For in vitro titration curves, area under the curve (AUC) was calculated and used for statistical comparison. For in vivo tumor experiments, Kruskal–Wallis nonparametric statistical analysis was performed. The statistical tests used are indicated in the figure legends for each experiment. P values are indicated by asterisks as *, P < 0.05; **, P < 0.01; ***, P < 0.001. GraphPad prism was used for statistical analysis.

The data generated in this study are available within the article and its supplementary data files. Data related to the biological structure shown in this study are deposited in Worldwide Protein Data Bank (wwPDB) with ID 7P3I.

The α-FAPxCD40 DARPin molecule binding human CD40 and FAP was designed on a DARPin scaffold (19) and lacks an Fc domain, avoiding systemic CD40 activation through FcR cross-linking. The α-FAPxCD40 construct consists of a chain of four covalently linked DARPin domains: an N-terminal anti-HSA domain for half-life extension (17), an anti-human FAP domain for tumor localization and clustering, and two identical anti-human CD40 domains for immune activation (Fig. 1A). To select the optimal format for α-FAPxCD40 molecule, different valences of the CD40 domains were explored. Molecules with one (α-FC), two (α-FCC), or three (α-FCCC) DARPin CD40 domain were compared in in vitro assays (Supplementary Fig. S1A). Two α-CD40 DARPin (α-FCC), compared with one α-CD40 DARPin (α-FC), increased the biological activity in vitro. Three α-CD40 DARPin (α-FCCC) did not significantly improve activity compared with α-FCC, but slightly decreased FAP specificity (Supplementary Fig. S1A). Based on these results, bivalent CD40 was selected for the final format.

Binding to the respective targets was measured both in solution and on cells with surface plasmon resonance (SPR) and flow cytometry techniques. SPR experiments demonstrated that α-FAPxCD40 binds to human CD40, human FAP, and HSA with an affinity of 0.103 ± 0.012 nmol/L, 0.307 ± 0.019 nmol/L and 50 nmol/L, respectively (Supplementary Fig. S1B). Flow cytometry analysis confirmed binding to CD40 on B cells and FAP expressed on CHO cells (Fig. 1B). Detailed biophysical characterization of α-FAPxCD40 demonstrated high thermal stability with a melting temperature (Tm) onset higher than 80°C and an aggregation temperature onset higher than 84°C (Supplementary Fig. S1C). Moreover, SEC-MALS analysis showed that α-FAPxCD40 was monodisperse and monomeric (Supplementary Fig. S1D). Structure analyses using X-ray crystallography revealed that the CD40 domain of α-FAPxCD40 binds to cysteine-rich domain (CRD) 1 (aa 23–59) of the CD40 receptor on the opposite side of the CD40L binding region, suggesting the absence of direct binding site competition between the construct and the CD40L (Fig. 1C and D). We observed a similar scenario in both binding and activation competition assays performed in the presence of CD40L. In these assays, α-FAPxCD40 retained its capacity to bind and activate CD40 receptor (Fig. 1E and F). These data are in line with the X-ray crystallography studies, suggesting that α-FAPxCD40 will not block binding of CD40L.

As α-FAPxCD40 does not bind to either mouse CD40 or mouse FAP, a similar surrogate molecule specific for murine CD40 and FAP (α-mFAPxCD40) with matching format was designed and produced (Supplementary Table S3), enabling in vivo studies in immunocompetent mouse models.

The biological activity and the FAP-dependent mechanism of action of α-FAPxCD40 were initially tested in in vitro cellular assays using human antigen-presenting cells (APC): B cells, DC, and MΦ, purified or differentiated from PBMC of healthy donors. The various APC were cocultured with FAP-positive or FAP–negative cells. Upregulation of costimulatory molecules and secretion of cytokines into the media were measured as markers of cell activation (28). The publicly available sequence of selicrelumab was used to produce α-CD40 mAb that activates APC independently of FAP-mediated cross-linking. Upregulation of CD54, CD69, CD70, and CD86 on B cells was observed for both α-FAPxCD40 and the α-CD40 mAb; however, in the case of α-FAPxCD40, the activation occurred only in the presence of FAP-positive cells, confirming a mechanism of action strictly dependent on FAP-mediated cross-linking (Fig. 2A; Supplementary Fig. S2A). In contrast, the α-CD40 mAb activated human B cells independently of the presence of FAP. Stringent FAP-dependent activity of α-FAPxCD40, as determined by IL12 secretion and upregulation of CD80, CD83, and CD86, was also observed in MDDC (Fig. 2B; Supplementary Fig. S2B). Monocyte-derived macrophages (MDMΦ) likewise showed FAP-dependent activation upon treatment with α-FAPxCD40, producing various cytokines only in the presence of FAP-positive cells, but not of FAP-negative cells (Fig. 2C; Supplementary Fig. S2C).

Similar to human α-FAPxCD40, the mouse-specific molecule α-mFAPxCD40 showed FAP-dependent activation of CD40 on murine APC, B cells and DC, also confirming a mechanism of action strictly dependent on the presence of FAP in the mouse system (Supplementary Fig. S3).

The capacity of the α-mFAPxCD40 molecule to promote M1-like macrophages was evaluated in an in vitro assay. Bone marrow–derived macrophages (BMMΦ) were cocultured with FAP-positive or FAP-negative CHO cells, treated with CD40 agonists, and phenotypically characterized by flow cytometry using CD206 and CD80 as polarization markers. α-mFAPxCD40 decreased the M2-like (CD206⁺)/M1-like (CD80⁺) ratio only in the presence of FAP, suggesting that the molecule was able to augment the presence of M1-like MΦ in the in vitro culture (Supplementary Fig. S4A and S4B). In contrast, an agonist mAb specific for mouse CD40 (α-mCD40 mAb) increased the proportion of M1-like MΦ independent of FAP expression (Supplementary Fig. S4A and S4B).

In parallel, MΦ were investigated for their ability to secrete different cytokines upon CD40 stimulation. IL12p40, IL18, CCL22, and CCL7 were upregulated (compared with the control IgG group) in the presence of α-mFAPxCD40 with a mechanism of action dependent on FAP expression (Supplementary Fig. S4C).

The correlation between FAP density and the activity of α-FAPxCD40 was investigated in an in vitro B-cell coculture assay, as previously described in Fig. 2. Specifically, human B cells were cocultured with six different cell lines with varying FAP surface densities (Fig. 3A). α-FAPxCD40 induced the upregulation of CD69, in terms of percentage of cells expressing both the receptor and MFI (Fig. 3B and C). Notably, in the presence of cells with higher FAP expression (Fig. 3A), the α-FAPxCD40 construct produced a similar maximal level of activation for both B cells (with CHO-FAP c1.7, CHO-FAP c1.9 and WM266-4) and MDDC (with CHO-FAP c1.7) as that seen with the α-CD40 mAb (Fig. 3B and C; Supplementary Fig. S5). However, in the presence of cells with low FAP expression, the maximal level of activation observed for α-FAPxCD40 was substantially lower than that seen for the α-CD40 mAb. These data support the FAP-dependent mechanism of activation of α-FAPxCD40 and suggest that maximal activation will occur in those tissues, such as tumors, where FAP is expressed at higher density.

The low FAP expression seen in syngeneic mouse tumor models does not mimic the abundant presence of FAP in human solid tumors. Therefore, a mouse FAP-transfected MC38 colorectal cancer model (MC38-FAP) was developed (Supplementary Fig. S6). To confirm that α-mFAPxCD40 construct localizes to the FAP-expressing tumors, localization studies were performed using SPECT/CT analysis of MC38-FAP tumor–bearing mice injected with the α-mFAPxCD40 construct, described previously in Supplementary Figs. S3 and S4. The SPECT/CT imaging experiment revealed localization of α-mFAPxCD40 in the tumor (Fig. 4A) and a maximal tumor uptake was seen from 24 to 72 hours after injection of the indium-111–conjugated α-mFAPxCD40. SPECT/CT results were extended in an in vivo tissue distribution study, in which radiolabeled α-mFAPxCD40 or Neg-CTRL were injected into MC38-FAP tumor–bearing mice and quantified from 4 to 96 hours after injection. α-mFAPxCD40 accumulates over time in tumor tissue, but not in other organs (muscle, bone marrow, liver or kidney), compared with the background signal defined by the neg-CTRL (Fig. 4B). Additional control constructs that bind HSA and only one target, either FAP or CD40, were evaluated. Only α-mFAPxCD40 demonstrated significantly enhanced uptake in the tumor, indicating that tumor disposition of the construct requires both FAP and CD40 expression (Fig. 4C).

In additional studies, mice injected with α-mFAPxCD40 were sacrificed, and large tumor sections were scanned and analyzed quantitively for the presence of DARPin molecules by IHC. As observed in the biodistribution study (Fig. 4B and C), α-mFAPxCD40 accumulates in the tumor, achieves peak accumulation after 72 hours, and requires both FAP and CD40 expression (Fig. 4D and E).

Some off-tumor signal was observed in the spleen (Fig. 4B and C), likely because of on-target distribution mediated by the CD40-specific domain of α-mFAPxCD40.

These results provide evidence that α-mFAPxCD40 can localize and accumulate in vivo in a syngeneic mouse tumor model expressing FAP.

Large tumors are characterized by reduced vascularization and a profoundly immunosuppressive microenvironment, limiting the therapeutic activity of immuno-based approaches (29). To mimic this more aggressive yet more clinically relevant tumor stage, we investigated the antitumor activity of α-mFAPxCD40 in immunocompetent mice bearing well-established (≈300 mm³) colorectal MC38-FAP tumors.

The antitumor activity of α-mFAPxCD40 and α-mCD40 mAb was evaluated against wild-type (MC38-WT; Fig. 5A) and FAP-transfected MC38 (MC38-FAP) tumors (Fig. 5B). The α-mCD40 mAb produced robust antitumor activity against MC38-WT, whereas the α-mFAPxCD40 construct was not active in this model that does not significantly express FAP (Fig. 5A). In contrast, both the α-mFAPxCD40 and α-mCD40 mAb demonstrated antitumor activity against FAP-expressing (MC38-FAP) tumors (Fig. 5B). Importantly, the FAP-CTRL that still retained the capacity to bind CD40, but not FAP, did not produce therapeutic activity, confirming that FAP binding of α-mFAPxCD40 is required for its in vivo activity.

At doses of 5 and 50 mg/kg, α-mFAPxCD40 resulted in regression of MC38-FAP tumors in eight of eight and in seven of eight mice, respectively (Fig. 5C). In contrast, the non–FAP-binding control (FAP-CTRL) even at 50 mg/kg did not show activity. At doses of 0.5 mg/kg or less, α-mFAPxCD40 did not impact tumor progression.

Altogether, these data provide evidence that α-mFAPxCD40 induces strong and durable in vivo antitumor activity in a FAP-dependent manner in a mouse model of colorectal cancer.

To demonstrate tumor memory immune response, mice with tumors that regressed following treatment with α-mFAPxCD40 were rechallenged with MC38-FAP or MC38-WT tumor cells. All mice previously treated with α-mFAPxCD40 completely rejected the second inoculum of MC38-FAP tumors (Fig. 5D), indicating the presence of a memory immune response. The same outcome was observed in mice rechallenged with MC38-WT tumor cells (Fig. 5D), suggesting that the immunologic memory was not limited to the FAP-related antigen, but was more broadly directed to additional tumor antigens. Both MC38-WT and MC38-FAP tumors grew progressively when inoculated into naïve mice (Fig. 5E). Cured mice were also rechallenged with different tumor cell lines, Panc02 and MB49. Most of these tumors were able to grow, but with a longer lag and doubling time (Fig. 5F and G). These data suggest that Panc02 and MB49 might share some common antigens with MC38. Alternatively, α-mFAPxCD40–mediated activation of the innate immune system could restrain the growth of these different tumor cell lines in a nonantigen-specific manner. Experimental evidence suggests that the multispecific α-mFAPxCD40 has the capacity to induce protective antitumor immunologic memory and, furthermore, that this memory is not limited to the transfected FAP but extends to the antigens on the tumor cells.

α-mFAPxCD40 is intended to activate CD40 locally in the TME, where FAP is expressed in order to minimize systemic toxicity. We therefore conducted in vivo studies to assess the safety profile of α-mFAPxCD40, in comparison with α-mCD40 mAb known to induce liver toxicity and elevated proinflammatory cytokines in blood (26, 30, 31). MC38-FAP tumor–bearing mice were treated with either α-mFAPxCD40 or α-mCD40 mAb, and loss of body weight, signs of stress, serum cytokine and transaminase elevation, and liver tissue damage were assessed as parameters of toxicity. Treatment with α-mFAPxCD40, in contrast to the α-mCD40 mAb, did not result in elevation of IL6, TNFα, IFNγ, or IL12p70 (Fig. 6A) nor increases in transaminases (Fig. 6B). No signs of stress or changes in the body weight (Supplementary Fig. S7) were observed following the administration of α-mFAPxCD40.

In addition, liver toxicity was assessed by IHC. Histological analysis of the livers of mice treated with α-mCD40 mAb revealed diffuse tissue damage characterized by vascular-centered multifocal inflammation with aggregates of mononuclear leukocytes, thrombosis and necrosis. In contrast, livers of mice treated with α-mFAPxCD40 did not show any evidence of tissue damage and had a similar histologic profile to that observed in vehicle-treated livers (Fig. 6C). In conclusion, α-mFAPxCD40 was active in vivo and inhibited tumor growth similarly to the α-mCD40 mAb (Fig. 5), but, in sharp contrast to the α-mCD40 mAb, did not show signs of systemic toxicity. Altogether, these data support the concept that the multispecific DARPin molecule exerts therapeutic activity selectively at the site of disease, leading to a strong antitumor activity at well-tolerated doses.

To better elucidate the specific mechanism of the construct, we performed flow cytometry analysis of MC38-FAP tumor–bearing mice enrolled in a short-term experiment described in Fig. 5B, where treated animals still had a small tumor mass that could be analyzed at termination. One of the most important hallmarks of CD40 agonists is the activation of professional APC, in particular DC (32–34). Flow cytometry analysis of tumors revealed an increase of the frequency of DC expressing the activation marker CD80 upon treatment with α-mFAPxCD40, similarly to the result obtained with α-mCD40 mAb (Fig. 7A), confirming in vivo the data generated in the in vitro coculture system (Supplementary Fig. S3B).

Kashyap and colleagues showed in the MC38 model that CD40 agonist antibody alone is able to promote myeloid cell skewing in tumors (34). In agreement with this study, we observed that both α-mFAPxCD40 and α-mCD40 mAb reduced the number of CD11b⁺F480⁺ tumor-associated macrophages (TAM) in the leukocyte infiltrate of MC38-FAP tumor, but significantly increased the M1-like/M2-like TAM ratio, thus promoting a more inflamed TME (Fig. 7B). These data suggest that α-mFAPxCD40, similarly to the α-mCD40 mAb, can promote myeloid cell activation in tumors by improving DC activation and M1 skewing. Importantly, these pharmacodynamic changes in the TME were not observed with the non–FAP-binding control (FAP-CTRL), binding CD40 but not FAP, confirming one more time the FAP-mediated mechanism of action of α-mFAPxCD40.

In addition, tumors treated with CD40 agonists displayed an increased percentage of CD8⁺ T cells and a higher absolute number and frequency of CD4⁺ T cells, likely as a consequence of the reshaping of the myeloid compartment (Fig. 7C and D). These mechanistic studies suggest that α-mFAPxCD40 induces a more inflamed TME phenotype characterized by (i) a higher presence of activated CD80⁺ DC, (ii) an increased presence of M1-like MΦ, and (iii) an enhanced presence of T cells, providing a possible mechanism of the therapeutic activity induced by α-mFAPxCD40.

Immunotherapy targeting the PD-1/PDL-1 axis has revolutionized treatment for a subset of cancer patients, but the majority of tumors are still resistant to immune-checkpoint inhibition (35, 36). It is therefore important to study the mechanisms of anti–PD-1/PD-L1 resistance and to identify other modalities that can either reverse or circumvent it. One approach is to suppress immunosuppression driven by the TME in order to restore T-cell infiltration and functionality. Pharmacologic activation of APC by inducing CD40 receptor signaling has been considered a promising approach. Agonistic anti-CD40 antibodies induce systemic immune activation and have shown antitumor activity, but have substantial safety concerns and as such they have not yet been approved for the treatment of cancer (8, 37).

We reasoned that a therapeutic strategy designed to activate CD40 selectively in the TME could reduce systemic toxicity while maintaining or increasing activity. The versatile DARPin technology platform was used to design a conditional agonist, intended to induce immune activation selectively in the TME upon simultaneous binding to CD40 and a tumor-specific anchor target. This construct, which lacks an Fc, cannot be cross-linked by FcR binding and, therefore, should result in reduced systemic activation. The well-characterized FAP protease was selected as a tumor anchor to generate the therapeutic compound α-FAPxCD40, which is specific for the human FAP and CD40 target proteins. A surrogate molecule (α-mFAPxCD40) selective for the murine orthologs was used for in vivo pharmacologic studies in mice.

In vitro, α-FAPxCD40 induced APC activation in the presence of FAP-expressing cells at low nanomolar concentrations, whereas activation was not observed with FAP-negative control cells even at concentrations as high as 400 nmol/L, the highest tested concentration (Fig. 3), suggesting that α-FAPxCD40 was highly potent and highly selective.

In a B-cell coculture activation assay using a panel of cell lines expressing different amounts of FAP, a correlation between the activity of α-FAPxCD40 and the FAP expression level was observed, with maximal activation similar to that of the α-CD40 mAb only seen at a high density of FAP expression. Numerous imaging studies have confirmed that FAP expression in normal tissues is absent or low compared with expression in tumors (13, 14, 38, 39). A whole-body PET imaging study of cancer patients with FAP-specific tracers confirmed high uptake into tumors, and much lower, if any, uptake into normal tissues (14). The differential expression of FAP expression between normal and neoplastic tissues, the high level of FAP expression required for optimal activity, and the high selectivity of α-FAPxCD40 suggests that in patients, α-FAPxCD40 may be active within the FAP-expressing tumor parenchyma but not outside the tumor where the physiologic expression of FAP is limited. This tumor-localized immune activation may both reduce systemic toxicity and improve activity relative to α-CD40 mAbs.

In vivo, SPECT/CT and biodistribution studies confirmed increased uptake of α-mFAPxCD40 into the FAP-expressing MC38 murine colon carcinoma model. However, in the interpretation of these data, it is important to consider that the pattern of FAP expression in FAP-transfected tumors is different compared with human tumors, in which FAP is typically found in the tumor stroma.

Similar studies (40–42) using FAP-targeted small molecules have shown an impressive tumor-specific accumulation, reaching tumor:organ ratios greater than 20:1 as early as 1 hour after systemic administration. These studies, together with our data, support the relevance of FAP as a good target for localizing various immunomodulatory molecules.

We observed α-mFAPxCD40 reaching the peak of accumulation in the tumor mass after around 72 hours. α-mFAPxCD40–mediated immune activation might increase the expression of CD40 on APC in the TME for further increasing the retention of the DARPin construct. This dynamic process could explain the incremental accumulation of the molecule in the tumor over an extended period, 72 to 96 hours. Importantly, in line with this activation hypothesis, the intratumoral increase of the constructs requires both CD40 and FAP-binding domain.

Tumor accumulation of α-mFAPxCD40 led to complete eradication of the majority of tumors. Specifically, the antitumor efficacy of α-mFAPxCD40 was FAP dependent, as shown by the lack of or minimal activity (i) in the FAP-CTRL group and (ii) in FAP^Neg/Low MC38-WT tumor-bearing mice. The conditional activity of α-mFAPxCD40 led to a tumor-localized immune activation without signs of systemic toxicity, differently than the α-mCD40 mAb whose therapeutic efficacy was associated with hepatic toxicity. The good safety profile of α-FAPxCD40 could be especially advantageous in combination cancer therapy.

Although the immune activation induced by α-mFAPxCD40 was limited to the TME, the treatment was able to evoke a memory immune response against tumor antigens as shown in rechallenge experiments. Macroscopically cured mice were able to completely reject MC38-FAP or MC38-WT tumors, indicating that treatment had induced and/or amplified immune responses to antigens that were not sufficiently immunogenic in the absence of pharmacologic CD40 activation. Our observation is compatible with preclinical findings that CD40 stimulation results in a long-term antitumor protection by supporting the generation of antigen-specific stem-like memory CD8⁺ T cells (43).

Finally, the preclinical findings reported in this article have supported the start of the clinical development of α-FAPxCD40 (MP0317) in a first-in-human safety and tolerability study in patients with relapsed/refractory advanced solid tumors (clinical trial NCT05098405).

N. Rigamonti reports a patent for WO 2021/229067 pending; he was an employee at Molecular Partners AG during the conduct of the study. N. Veitonmäki reports a patent for WO 2021/229067 pending; and employee of Molecular Partners during the conduct of the study. C. Domke reports a patent for WO 2021/229067 pending. S. Barsin reports a patent for WO 2021/229067 pending. S. Jetzer reports a patent for WO 2021/229067 pending. T. Lekishvili reports she was an employee at Molecular Partners AG during the conduct of the study. F. Malvezzi reports a patent for WO 2021/229067 pending. M. Gachechiladze reports a patent for WO 2021/229067 pending. M. Behe reports other support from Molecular Partners during the conduct of the study. V. Levitsky reports a patent for WO2021/229067 pending. P.A. Trail reports a patent for WO 2021/229067 pending; and consultant for a number of companies in the general area of oncology; however, there is no conflict of interest relative to this paper; chair of the scientific advisory board of Tubulis GmbH, a company focused on antibody drug conjugates, and there is no conflict of interest. No disclosures were reported by the other authors.

N. Rigamonti: Conceptualization, supervision, validation, visualization, methodology, writing–original draft, project administration, writing–review and editing. N. Veitonmäki: Conceptualization, resources, supervision, writing–review and editing. C. Domke: Data curation, formal analysis, validation, investigation, visualization. S. Barsin: Data curation, formal analysis, investigation, visualization. S. Jetzer: Data curation, formal analysis, investigation, visualization. O. Abdelmotaleb: Data curation, formal analysis, investigation, visualization. R. Bessey: Data curation, formal analysis, investigation, visualization. T. Lekishvili: Data curation, formal analysis, investigation, visualization. F. Malvezzi: Data curation, formal analysis, visualization. M. Gachechiladze: Data curation. M. Behe: Data curation, formal analysis, validation, investigation, visualization. V. Levitsky: Conceptualization, validation. P.A. Trail: Conceptualization, resources, supervision, validation, writing–original draft, writing–review and editing.

We thank A. Goubier, G. Adolf, H. Ji, R. Kenefeck, S. Plyte, S. Wullschleger, U. Fiedler, V. Calabro, and V. Kirkin for the fruitful discussions on the scientific data; D. Gassert for the suggestions on the manuscript; A. Valeri, C. Zitt, D. Lang, H. Poulet, M. Franchini, M. Matzner, and W. Ali for help with the in vivo experiments; A. Schlegel, J. Schwestermann, L. Juglair, M. Paladino, N. Velden, P. Latha, S. Mangold, and T. Hospordasch for help with some in vitro experiments; A. Blanc and S. Imobersteg for help with the in vivo imaging studies; the Flow Cytometry Core Facility at Eidgenössische Technische Hochschule (ETH) Zurich for cell sorting.

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.