Fibroblast Activation Protein α-Targeted CD40 Agonism Abrogates Systemic Toxicity and Enables Administration of High Doses to Induce Effective Antitumor Immunity

The CD40 receptor on antigen-presenting cells (APC) plays a key role in regulating immune responses and triggering the induction of adaptive immunity. Binding of its trimeric ligand CD40L, which is mainly expressed by Th cells, leads to CD40 receptor multimerization, inducing a plethora of downstream signaling pathways (1, 2). As a result, APCs become activated and mature, a process commonly referred to as APC licensing. CD40L-mediated APC activation increases their expression of costimulatory molecules as well as production of proinflammatory cytokines and enhances their capability of antigen processing and presentation (3, 4). Employing agonistic anti-CD40 antibodies to mimic this licensing of APCs for tumor antigen presentation is, therefore, an attractive strategy in the context of cancer immunotherapy (CIT). In particular, the ability of anti-CD40 molecules to activate dendritic cells (DC) and subsequently increase (cross-) priming of tumor-specific T cells (5, 6), has fueled the notion that CD40 agonists could be crucial CIT combination partners for eliciting long-lasting and sustainable antitumor responses (7–9). It is thus not surprising that a number of agonistic anti-CD40 antibodies have been developed over the last 20 years, of which several have undergone clinical evaluation in patients with cancer (10–15). Although some of the tested antibodies triggered clinical responses, major challenges of anti-CD40 agonist treatments were revealed. Across different clinical studies, dose-limiting side effects such as cytokine release syndrome (CRS) and hepatotoxicity were observed (12–15). At the same time, agonistic anti-CD40 molecules displayed a very short serum half-life, indicative of a substantial sink effect due to the broad expression of the CD40 receptor on hematopoietic and nonhematopoietic cells (8, 14). Therefore, target-mediated drug disposition by the peripheral CD40 sink, as well as dose limitations due to on-target systemic toxicity, restrict the amount of systemically administered, first-generation CD40 agonists that can be delivered to the tumor microenvironment to activate APCs locally.

In contrast, various preclinical studies employing intratumoral or peritumoral injection of CD40-activating agents have demonstrated that a stimulation of CD40 limited to the tumor area is well tolerated and induces effective antitumor responses. Despite the confinement of CD40 agonism to a primary tumor, systemic tumor-specific T-cell responses were raised, which enabled the eradication of secondary distant tumors and protected from tumor rechallenge (16, 17).

These data provide a solid rationale for targeting anti-CD40 agonism to the tumor. As intratumoral administration of anti-CD40 agonists is technically challenging and limited to few accessible tumor lesions, we have instead developed a tumor-targeted anti-CD40 agonistic antibody, the bispecific fibroblast activation protein α (FAP)-CD40 molecule, for systemic administration. Unlike nontargeted anti-CD40 antibodies, FAP-CD40 does not induce systemic CD40 stimulation. Instead, FAP-CD40 exclusively triggers APC activation in the presence of FAP, a serine protease that is highly expressed on cancer-associated fibroblasts in the tumor stroma (18). Of note, while a majority of carcinomas display high FAP levels, only low amounts of FAP are expressed in few healthy adult tissues such as the lymph nodes (LN; refs. 19, 20). The highly tumor-specific expression of FAP combined with its presence across many tumor types and recently published evidence describing a preferential localization of APC niches within the stromal area of human tumors (21), account for our choice of tumor target.

Our preclinical data demonstrate that high doses of FAP-CD40 can be administered safely to overcome sink effects, allowing accumulation of our molecule in the tumor. We show that FAP-CD40–mediated DC activation is limited to the tumor and the LNs. This locally restricted CD40 agonism can induce potent immune responses resulting in tumor eradication and durable immunologic memory.

Recombinant biotinylated murine FAP, the human CD40 agonists FAP-huCD40 and its corresponding nontargeted control DP47-huCD40, SGN40 huIgG1, SGN40 moIgG1, selicrelumab (22), DP47, the murine surrogate molecule FAP-moCD40, as well as FGK4.5 moIgG1 were produced at Roche Innovation Center Zurich (RICZ) and Munich (RICM).

The human FAP-targeted CD40 agonist FAP-huCD40 consists of a huIgG1 containing two CD40 binding Fv regions (0817 domain, recognizing human and cynomolgus CD40). Mutations (Pro329Gly, Leu234Ala, and Leu235Ala; PGLALA mutation; ref. 23) to prevent fragment crystallizable γ receptor (FcγR) binding were introduced within the Fc region of the antibody. To ensure correct assembly of light and heavy chains, the knob-into-hole strategy (24) as well as charged mutations in the C_H1 domains (EE) and C_L domains (RK) were employed. For monovalent FAP binding, one anti-FAP Fab fragment (recognizing human, cynomolgus, and murine FAP) was fused via its V_L-C_H1 chain to the huIgG1 Fc-knob chain by a 4G4S linker. To prevent mispairings, the light- and heavy-chain variable regions of the anti-FAP Fab were exchanged (25). For the nontargeted DP47-huCD40 molecule, the anti-FAP Fab domain of FAP-huCD40 was replaced by the nonbinding DP47 Fab (described previously in ref. 26). For the conventional human CD40 agonist control molecules, the V_H and V_L sequences of SGN40 (27) were inserted in an Fc-competent huIgG1 (SGN40 huIgG1) or moIgG1 (SGN40 moIgG1) framework. The control antibody DP47 is based on a PGLALA-mutated huIgG1 with two nonbinding DP47 Fv domains. For the murine surrogate molecule FAP-moCD40, the anti-human CD40 Fv domains of the FAP-huCD40 molecule were replaced by the V_H and V_L regions of the murine anti-CD40 agonist FGK4.5 (28). FGK4.5 moIgG1 was created by inserting the V_H and V_L regions of original FGK4.5 molecule, which is a rat IgG2a antibody, into an Fc-functional moIgG1 backbone (see Fig. 1A; Supplementary Fig. S1C for a schematic overview of all anti-CD40 antibody constructs used).

Antibodies were expressed by transient transfection of human embryonic kidney (HEK 293) cells. Antibodies were purified from cell culture supernatants by affinity chromatography using MabSelectSure-Sepharose (GE Healthcare) chromatography. Purified antibodies were characterized by capillary electrophoresis SDS, size-exclusion chromatography, mass spectrometry, and endotoxin determination.

Cells were maintained under sterile conditions at 37°C in a humidified incubator (5% CO₂) and passaged regularly upon reaching 80% confluency. MC38-huCEA cells were generated by City of Hope (29) and engineered to express mRFP and ovalbumin (OVA) at RICZ. The plasmid encoding the OVA transgene was kindly provided by ProQinase. MC38-huCEA-mRFP-OVA cells were kept under selection with 1.5 μg/mL puromyin (Invivogen, ant-pr), 100 μg/mL hygromycin (Invivogen, ant-hg), and 300 μg/mL geneticin (Gibco, 10131035).

MC38 cells were obtained from City of Hope and transfected with murine Fap at RICZ. MC38-FAP cells were selected with 6 μg/mL puromycin (Invivogen, ant-pr). To establish a cell line with robust in vivo growth characteristics, MC38-FAP cells were subcutaneously injected in C57BL/6J mice. Tumors that showed good growth were resected, dissociated, and tumor cells were grown and expanded in vitro under puromycin selection for cell line establishment. KPC-4662 cells were received from University of Pennsylvania (Philadelphia, PA; ref. 30) and engineered to express huCEA at RICZ. KPC-4662-huCEA cells were selected for huCEA expression with 500 μg/mL hygromycin. All cell lines used were routinely checked for Mycoplasma using the MycoAlert Mycoplasma Detection Kit (Lonza, LT07–118). Cells were used for experiments one to three weeks after thawing.

Female mice were used for all studies. Mice were maintained under specific-pathogen-free condition with daily cycles of 12 hours light and 12 hours darkness according to committed guidelines (GV-Solas; Felasa; TierschG). Experimental study protocols were reviewed and approved by the Cantonal Veterinary Office in Zürich (license ZH223/2017, ZH224/2017, and ZH225/2017) and by the Institutional Animal Care and Use Committee of RICZ. After arrival, animals were maintained for one week to get accustomed to the new environment and for observation. They were afterwards implanted with a subcutaneous transponder for identification and maintained one more week for recovery. Continuous health monitoring was carried out on a regular basis. Mice were between 6 to 9 weeks of age at the start of the experiments. In case of tumor-bearing mice, tumors were measured at least two to three times per week. Mice were randomized into different treatment groups according to tumor volume. OT-1 [C57BL/6-Tg(TcraTcrb)1100Mjb/Crl] and C57BL/6J mice were acquired from Charles River France. HuCD40tg mice (background C57BL/6NT) expressing both the human and murine CD40 receptor, were generated by Taconic by oocyte pronuclear injection of a bacterial artificial chromosome construct encoding the human CD40 receptor under control of the human CD40 promoter. The huCD40tg mice were provided by the Baylor research institute. Fap^−/− mice were kindly provided by Dr. Junichiro Sonoda (Genentech; ref. 31). HuCEAtg mice [C57BL/6JTgN(CEAGe) 18FJP] were obtained under license agreement from Beckman Research Institute of City of Hope (32). HuCEAtg, huCD40tg and Fap^−/− mice were bred by Charles River.

The cynomolgus monkey study was ethically reviewed and approved by the Animal Welfare and Ethics Committee of Charles River Laboratories Edinburgh. The project license application was granted by the British Home Office. Further study details can be found within the Supplementary Material and Methods section.

For flow cytometry staining, cells were incubated with 3 μg/mL of Fc receptor blocking mouse IgG isotype control (Thermo Fisher Scientific, 10400C) for 10 minutes at 4°C. Cells were subsequently stained with a mixture of fluorescently labeled antibodies and fixable viability dye (Invitrogen) in Brilliant Stain buffer (BD, 563794; see Supplementary Table S1 for a list of antibodies used) for 30 minutes at 4°C. Afterwards, cells were washed twice with PBS. For intracellular stainings, extracellularly labeled cells were incubated with fixation/permeabilization solution (eBioscience, 00–5523–00) for 30 minutes at 4°C, washed twice with permeabilization buffer, and then stained for 30 minutes at room temperature with intracellular staining antibodies diluted in permeabilization buffer. Cells were washed two times with permeabilization buffer and analyzed the same day using a 5-laser LSR-Fortessa (BD Bioscience with DIVA software). Data analysis was performed using the FlowJo version 10 software (FlowJo, LLC).

To isolate splenic immune cells, murine spleens were cut into small pieces and digested with 0.1 mg/mL collagenase D (Sigma-Aldrich, 11088866001) and 0.05 mg/mL DNase (Sigma-Aldrich, D5025–150KU) for 10 minutes at 37°C. EDTA (Applichem, A4892.1000) was added at a concentration of 0.01 mol/L, followed by a second incubation step at 37°C for 5 minutes. The splenocytes solution was smashed through a 70-μm cell strainer (Corning, 431751). Red blood cells were lysed with cell lysis buffer (BD, 555899). Mouse CD11c UltraPure microbeads (Miltenyi Biotec, 130–108–338), mouse CD8a⁺ T Cell Isolation Kit (Miltenyi Biotec, 130–104–075) or the Mouse B Cell Isolation Kit (Miltenyi Biotec, 130–090–862) were used according to the manufacturer's instructions to isolate different immune cell populations from the splenocytes suspension. For subsequent in vitro assays with splenic immune cells, cells were resuspended in R10 (RPMI1640; Gibco, 31870–025) supplied with 10% FBS (Gibco, 16140), 1% penicillin–streptomycin (P/S; Gibco, 11548876), 1% l-glutamine (Gibco, 25030–024), 1% sodium-pyruvate (Gibco, 11360–039), 1% nonessential amino acids (Gibco, 11140–035), and 50 μmol/L β-Mercaptoethanol (Gibco, 31350–010).

For binding assays, MC38-FAP, MC38, or B cells from huCD40tg mice were resuspended in flow cytometry staining buffer (eBioscience, 00–4222–57) containing the different antibodies at the indicated range of concentrations and incubated for 120 minutes at 4°C. Afterwards, the cells were washed three times. Cells were further stained with secondary antibody solution containing R-Phycoerythrin (PE) conjugated AffiniPure F(ab')₂ Fragment Goat Anti-Human IgG, Fcγ Fragment Specific (1:50 dilution; Jackson ImmunoResearch, 109–116–098) and incubated for 60 minutes at 4°C. Cells were washed twice and resuspended in flow cytometry staining buffer containing 0.2 μg/mL 4′,6-diamidino-2-phenylindole (Roche, 10236276001) and acquired the same day.

1.0 × 10⁵ splenic B cells or 0.25 × 10⁵ splenic DCs from huCD40tg mice were seeded per well. Streptavidin Dynabeads (Thermo Fisher Scientific, 11205D) were coated with biotinylated mouse FAP according to the manufacturer's instructions and added to the APCs in a bead:cell ratio of 2:1. As control, noncoated beads were used. For APC activation with cellular FAP, MC38-FAP or parental MC38 cells were irradiated at 50 Gy using a RS 2000 irradiator (Rad Source Technologies) and 0.25 × 10⁵ MC38 or MC38-FAP cells were seeded per well. 0.25 × 10⁵ splenic DCs or 0.5 × 10⁵ splenic B cells from huCD40tg mice were added. SGN40 moIgG1 was crosslinked by preincubation for 30 minutes at room temperature with anti-mouse Fc F(ab')₂ (in-house endotoxin-purified, obtained from Jackson, 115–006–071) in a 1:1 ratio. CD40 agonistic antibodies were added at the indicated concentrations. DC activation was measured by flow cytometry after 24 hours, while B-cell activation was assessed after 48 hours.

For APC activation assays with nodal FRCs, FRCs were isolated from inguinal, axial, and brachial LNs of Fap^−/− or C57BL/6J mice and in vitro expanded as described by Fletcher and colleagues (33). 0.14 × 10⁵ FRCs per well were seeded in DMEM + 10% FBS + 1% P/S. 0.3 × 10⁵ splenic B cells or 0.25 × 10⁵ splenic DCs from C57BL/6J mice were added one day later to the FRCs. FAP-moCD40 was added at the indicated concentrations. DC activation was measured after one day, while B-cell activation was measured after 3 days by flow cytometry.

DCs were loaded with OVA protein via the DEC-205 receptor using the OVA Antigen Delivery Reagent (Miltenyi Biotec, 130–094–663) in combination with a biotinylated anti-mouse DEC-205 antibody (Miltenyi Biotec, 130–101–854) according to the manufacturer's protocol. 0.25 × 10⁵ DCs from huCD40tg mice were seeded per well. Murine FAP-coated or noncoated Dynabeads were added to the DCs at a 2:1 bead:cell ratio. Next, different agonistic anti-CD40 antibodies were added at the indicated concentrations. Twenty-four hours later, CD8⁺ T cells were isolated from the spleens of OT-1 mice added to the OVA-loaded DCs at a DC:T-cell ratio of 1:2. After three days, T cells were either used for flow cytometry analysis of T-cell numbers and activation or T cells were used for coculture with OVA⁺ target cells. For the killing assay, 5 × 10³ MC38-huCEA-mRFP-OVA cells were seeded in R10 and allowed to adhere before DC-primed T cells from the 5 nmol/L FAP-huCD40 plus FAP-beads conditions were transferred. The number of transferred T cells was determined to be 1 × 10⁴ based on flow cytometry analysis. As control, 2.5 × 10⁴ freshly isolated OT-1 T cells were added to the OVA-expressing tumor cells. MC38-huCEA-mRFP-OVA cell killing was measured by determining the red target cell area over time by Incucyte (Sartorius) analysis.

Fap^−/− mice were injected subcutaneously in the flank with 12.5 mg/kg OVA alone or in combination with either 13.3 mg/kg FAP-moCD40 or 10 mg/kg FGK4.5 moIgG1. C57BL/6J wild-type mice were injected with the combination of OVA and FAP-moCD40. Animals were sacrificed on day 3 or day 8 after vaccination and spleens as well as inguinal LNs were harvested in PBS.

HuCD40tg mice were injected subcutaneously in the flank with 2 × 10⁶ MC38-FAP cells in RPMI with 50% Matrigel (Corning, 354234). When tumors reached a mean size of 220 to 275 mm³, anti-CD40 agonists FAP-huCD40, SGN40 moIgG1 or selicrelumab were injected intraperitoneally at the indicated concentrations. Mice of the vehicle group received equal volumes of histidine buffer intraperitoneally. Animals were sacrificed at the indicated time points and liver, spleen, inguinal LNs as well as subcutaneous tumor were either harvested in PBS, paraformaldehyde (PFA), or 10% buffered formalin solution or were immediately frozen in liquid nitrogen. Blood was collected in serum tubes (BD, 365967), spun down at 15,000 × g for 10 minutes at 4°C to collect the serum, and serum samples were frozen at −20°C until analyzed.

For the tumor rechallenge experiment, tumor-free mice treated with FAP-huCD40 (9 out of 9 mice) were injected subcutaneously with 2 × 10⁶ MC38-FAP cells into the flank contralateral to the primary injection site 93 days after the first tumor cell injection. Naïve huCD40tg mice were injected simultaneously with tumor cells as control. Animals were sacrificed 54 days after the rechallenge when tumors in the vehicle group reached a mean size of 1,800 mm³.

For the in vivo studies using the KPC-4662-huCEA tumor model, huCEAtg mice received subcutaneous 0.3 × 10⁵ KPC-4662-huCEA pancreatic cells in RPMI with 50% Matrigel into the flank. Randomized mice (average tumor size of around 240 mm³/group) were treated intraperitoneally with vehicle or 13.3 mg/kg moFAP-CD40. Some animals were sacrificed eight days after therapy injection and tumors were resected for immunofluorescence staining analysis.

HuCD40tg mice were injected subcutaneously in the flank with 2 × 10⁶ MC38-FAP cells in RPMI with 50% Matrigel. When the tumors were between a size of 250 to 350 mm³, mice were intravenously injected with the indicated concentrations of AF647 labeled FAP-huCD40 and selicrelumab or histidine buffer (vehicle mice).

Mice were anesthetized at the indicated time points with 2% isoflurane and the blood collected retro-orbitally in heparin-coated tubes. The animals were sacrificed and organs were extracted, weighed, and placed into 4% PFA. Organs were washed in PBS and placed in the IVIS Spectrum in vivo imaging system (PerkinElmer) to acquire fluorescent images. Fifty microliters of blood was put in an Eppendorf tube for FLI imaging. In vivo imaging software (PerkinElmer) was used for imaging data analysis. Signal intensities for each group of organs were normalized together. A region of interest was drawn to capture the area around the signal. Total flux was captured as the number of photons per second and normalized against the organ weight (total flux at time point t/organ weight = drug concentration at time point t = C_t). No normalization to weight was done for the blood as the same volumes were used for all samples and for the LNs due to their very low weight. The background for each organ (C₀) was determined by measuring the total flux normalized to organ weight of the respective organ from vehicle animals at the 2-hour time point. To determine the normalized drug concentrations at each given time point, C_t values were divided by the corresponding C₀ values.

To analyze the binding of labeled anti-CD40 agonists on different cell subsets within different organs, mice treated with AF647-labeled molecules were sacrificed 24 hours after injection of the antibodies. Spleen, inguinal LNs, and tumors were harvested in PBS, processed as described below, and analyzed by flow cytometry.

IHC staining of MC38-FAP tumors for murine FAP and CD3 was done as described previously (26).

For flow cytometry analysis spleens, tumors and LNs were cut into small pieces and digested with 0.05 mg/mL DNase and 0.1 mg/mL collagenase D (spleens and LNs) or 1 mg/mL collagenase D (tumors) for 10 minutes (spleens and LNs) or 25 minutes (tumors) at 37°C. EDTA (0.01 mol/L) was added for 5 more minutes of incubation to spleens and LNs. Cell suspensions were smashed through a 70-μm cell strainer. Red blood cell lysis was performed for spleen samples. Cells were subsequently stained for flow cytometry analysis.

Snap frozen tumor pieces were lysed with the Bio-Plex Cell Lysis Kit (Bio-Rad) using Lysing Matrix S tubes (MP Biomedicals) and the Precellys Evolution tissue homogenizer (Bertin Technologies) according to the manufacturers' instructions. Protein content of tumor lysates was determined with the Pierce bicinchoninic acid assay kit. Forty micrograms of tumor protein was used of each sample for multiplexed cytokine analysis. The Bio-Plex Pro Mouse Cytokine 23-plex Assay Kit (Bio-Rad, M60009RDPD) or the Bio-Plex Pro Mouse Chemokine Panel 31-Plex Assay kit (Bio-Rad, 12009159) and the FLEXMAP 3D platform (LUMINEX) were used according to the manufacturers' instructions for analysis of the concentration of cytokines in serum and tumor.

For histopathologic assessment, formalin fixed liver samples were trimmed and processed for paraffin embedding. Four-micron–thick slices were stained with hematoxylin and eosin. Stained slides were used for histopathologic evaluation by a board-certified pathologist. Serum samples collected at necropsy were used for analysis of liver enzymes with ADVIA 1800 chemistry system (Siemens) according to the manufacturer's instructions.

Tumors were bisected and fixed in Cytofix solution (BD Biosciences, 554655) for 21 hours at 4°C. Half tumors were embedded in 4% low gelling temperature Agarose (Sigma, A0169) and sectioned into 70-μm–thick sections using a Vibratome (Leica 1200s) and common razorblades. Tumor sections were blocked and permeabilized at the same time using home-made “BlockPerm” buffer [Tris 0.1 mol/L (VWR, 89500–584) + 0.3% Triton X (Sigma, T8787) + 1% mouse serum (Jackson, 015–00–120) + 1% BSA (Sigma, A9576)] for 15 minutes. Fluorescently labeled antibodies were diluted in BlockPerm and added to the tumor sections for 15 hours at room temperature. The following antibodies were used: CD8 BV421 (clone 53–6.7, BioLegend, 100738), CD4 BV570 (clone RM4–5, BioLegend, 100542), FoxP3 (clone D6O8R, Cell Signaling Technologies, 12653) conjugated in-house to CF633 via a kit from Biotium (92257), E-cadherin (polyclonal, R&D Systems, AF648) conjugated in-house to CF555 via a kit from Biotium (92254), and FAP (clone 28H1, produced in-house) conjugated in-house to CF647 via a kit from Biotium (92259).

Sections were then washed with Tris 0.1 mol/L two times for 20 minutes and subsequently mounted in Fluoromount G (Thermo Fisher Scientific, 00–4985–02) using #1 coverslips (ROTH). Sections were imaged on a Leica SP8 inverted confocal microscope using a 40× oil immersion objective at 512 by 512 pixels and 1-μm thick virtual z-sections. Individual tiles were stitched together in the microscope software (LAS X, Leica Microsystems) and further processed using Imaris 9.6 (Bitplane). Briefly, in Imaris CD8 and CD4 channels were segmented with the following parameters (Smoothing: 1.52, Background subtraction: Default, Cell splitting: Yes; Quality for splitting: Default, Filter out small objects: below 250 Voxels) and statistics were exported to .csv files. FoxP3 intensity within CD4 segmented objects, as well as positional information (XY coordinates) of CD4 and CD8 objects were assessed in Matlab 2020a (Mathworks) and resulting images and data were exported for either direct use (images of segmented object localization) or subsequent plotting and statistical analysis.

Statistical significance was evaluated using the GraphPad Prism Version 8.4.2 (GraphPad Software) with the statistical test indicated. P values of or below 0.05 were considered statistically significant. Significance levels are indicated as: ns, not significant = P > 0.05; *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001; ****, P ≤ 0.0001.

Illustrations in Fig. 4B and Supplementary Fig. S1B were created with BioRender.com.

Our goal was to generate an anti-CD40 agonist that induces a potent CD40 stimulation selectively in the tumor area but not in the periphery. To achieve this, we created a bispecific FAP-huCD40 antibody containing two CD40-binding domains (0817 domain, recognizing human and cynomolgus CD40) and one C-terminally fused FAP binding site (recognizing human, cynomolgus, and murine FAP; Fig. 1A). For both, the anti-FAP and the anti-CD40 moieties, high-affinity binders were selected (Supplementary Fig. S1A). The activation of CD40 by bivalent agonistic anti-CD40 antibodies commonly depends on FcγR crosslinking (17, 34, 35). To prevent FAP-huCD40–mediated CD40 activation by FcγRs in the periphery, mutations in the FAP-huCD40 antibody Fc region were introduced, that abrogate FcγR binding (PGLALA mutations). In addition, the PGLALA mutations ensure that cells bound by FAP-huCD40 are not depleted due to antibody-dependent cellular or complement-dependent cytotoxicity (23).

By introducing the functionally “silent” Fc region in combination with a C-terminal anti-FAP domain into our CD40 agonist, we aimed at designing a molecule that can solely be crosslinked and activated through FAP binding. We envisioned FAP⁺ cells to function as scaffolds that allow clustering of FAP-huCD40 antibodies, leading to CD40 receptor multimerization and thus to a FAP-restricted induction of CD40 downstream signaling (Supplementary Fig. S1B).

We first tested FAP-huCD40 binding to both its targets CD40 and FAP. To evaluate CD40 binding, we used splenic B cells of mice transgenic for the human CD40 receptor (huCD40tg mice), as the anti-human CD40 moiety of the bispecific molecule is not cross-reactive to murine CD40. The human CD40 binding properties of FAP-huCD40 were comparable with those of the nontargeted human CD40 agonist SGN40 (Fig. 1A and B). Therefore, we considered SGN40 to be a suitable FAP-independent anti-CD40 agonist for comparison in our subsequent assays.

When evaluating FAP binding, we found that FAP-huCD40 but not the corresponding control molecule DP47-huCD40 (DP47 is a non-binding germline control; Fig. 1A) bound to murine FAP recombinantly expressed on MC38 murine colon adenocarcinoma cells (MC38-FAP). In accordance with these observations, no binding of FAP-huCD40 to non-FAP–expressing MC38 cells was detected (Fig. 1C and D).

Next, we assessed FAP-huCD40's potential to activate CD40-expressing APCs in vitro. Splenic DCs or B cells from huCD40tg mice were coincubated with FAP-huCD40 and either FAP-coated or noncoated polystyrene beads. Beads instead of cells were chosen as source of FAP to avoid potential cell-derived APC-activating factors that could affect the potency of our antibody in vitro.

With FAP-huCD40 in combination with FAP-coated beads, DCs and B cells exhibited a strong dose-dependent increase of the activation markers CD70, CD80, CD86, and murine CD40 (huCD40tg mice maintain murine CD40 expression; human CD40 could not be measured by flow cytometry, as FAP-huCD40 competed with labeled anti-human CD40 antibodies). In contrast, when using noncoated beads, FAP-huCD40 did not induce any APC-activating effects. No stimulation of DCs or B cells was observed with DP47-huCD40 in either the presence or absence of FAP (Fig. 2A and B; Supplementary Fig. S2A and S2B).

To compare FAP-huCD40's effects to those of a nontargeted human anti-CD40 agonist, we used the FcγR crosslinking-dependent antibody SGN40. To ensure maximum functionality of the SGN40 molecule in the huCD40tg mouse model, in which human CD40 receptors but murine FcγRs are expressed, the variable regions of the original SGN40 huIgG1 molecule were inserted into an Fc-competent murine IgG1 framework (SGN40 moIgG1; Fig. 1A). The murine IgG1 was selected, as murine CD40 agonists of this subclass are known to be most efficiently activated by FcγR crosslinking (35). SGN40 moIgG1 induced a moderate B-cell activation, while DC activation markers increased only slightly. Although DCs and B cells express FcγRs, this expression might not be sufficient to induce effective hyperclustering of anti-CD40 agonists. We therefore enhanced crosslinking of SGN40 moIgG1 by pre-incubating it with an anti-Fc F(ab')₂ (SGN40 moIgG1 XL), resulting in a more pronounced DC as well as B-cell activation. As expected, in contrast to FAP-huCD40, SGN40's activity was independent of the presence of FAP (Fig. 2A and B; Supplementary Fig. S2A and S2B). To confirm FAP-CD40's FAP specificity and to ensure that cellular FAP triggers similar effects as FAP coated to beads, we further tested FAP-huCD40 in presence of MC38-FAP versus MC38 cells. FAP-huCD40 induced APC activation in cocultures of DCs or B cells with MC38-FAP but not MC38 cells (Supplementary Fig. S2C and S2D). Together, our data highlight that FAP-huCD40 mediates a potent and exclusively FAP-dependent activation of DCs and B cells.

We next asked whether the observed FAP-huCD40–induced DC activation results in an enhanced capability of DCs to cross-prime CD8⁺ T cells. To answer this question, OVA was delivered via the antigen uptake receptor DEC-205 to splenic DCs isolated from huCD40tg mice. Targeting of antigen to the endocytic DEC-205 receptor facilitates efficient MHC class I antigen presentation (36). The OVA-loaded DCs were incubated with FAP-huCD40 in the presence of FAP-coated or noncoated beads. Subsequently OT-1 T cells (specific for the OVA-derived SIINFEKL peptide in the context of H-2k^b; ref. 37) were added to the DC cultures. OVA-pulsed DCs incubated with FAP-huCD40 in the presence of FAP induced a potent T-cell proliferation and activation. This effect was entirely FAP dependent, as no increase in T-cell numbers or activation markers were observed in conditions with FAP-huCD40 in combination with noncoated beads. Consistent with our previous results and irrespective of the presence of FAP, SGN40 moIgG1 XL induced activation and proliferation of OT-1 T cells, while DP47-huCD40 displayed no signs of activity (Fig. 2C). Of note, activated OT-1 T cells derived from the FAP-huCD40 plus FAP-beads conditions exceeded fresh OT-1 T cells in their capability of killing OVA-expressing MC38 cells (Fig. 2D). These data demonstrate that DC activation by FAP-CD40 in a FAP-specific manner can promote effective T-cell cross-priming.

We demonstrated FAP-CD40's effects upon crosslinking by FAP coated on beads or recombinantly expressed by MC38 cells. Because in both cases abundant amounts of FAP are provided, we wanted to investigate whether FAP-CD40 crosslinking and activity can as well be observed in the presence of limited FAP levels. This question seemed in particular relevant as low amounts of FAP are expressed by fibroblastic reticular cells (FRC), a population of myofibroblasts, in the LNs (20).

To validate whether FRCs might trigger FAP-CD40 crosslinking, we isolated them from murine LNs followed by in vitro expansion. Subsequently, these FRCs were coincubated with murine B cells or DCs and the murine surrogate molecule FAP-moCD40, which has equivalent molecular properties as FAP-huCD40, but binds the murine CD40 receptor (Supplementary Fig. S1C and S2E). FAP-moCD40 significantly enhanced the expression of different APC activation markers when added to APC/FRC cocultures. No effect of FAP-moCD40 was observed with in vitro expanded FRCs from FAP knockout (Fap^−/−) mice (Fig. 3A; Supplementary Fig. S3A). This indicates that the nodal FRC FAP expression, although being more than 50-fold lower than that of MC38-FAP cells (Supplementary Fig. S3B), is sufficient to induce hyperclustering of FAP-CD40 leading to APC activation.

Our results were substantiated by an in vivo vaccination study in nontumor-bearing wild-type versus Fap^−/− mice using OVA as immunogen. Upon subcutaneous injection of OVA and FAP-moCD40, DCs were activated in the LNs (inguinal, adjacent, and contralateral to the injection site) of wild-type but not Fap^−/− mice. In the spleen, no DC activation was detected in wild-type mice, in line with the fact that murine splenic FRCs lack FAP expression (Fig. 3B; ref. 20). Surprisingly, FAP-moCD40 failed to activate B cells in the LNs of wild-type mice. A marginal B-cell activation could be detected in the spleen of wild-type but not Fap^−/− mice (Supplementary Fig. S3C). OVA combined with a nontargeted murine CD40 agonist (Fc-functional FGK4.5 moIgG1) enhanced DC as well as B-cell activation in the LNs and spleens of Fap^−/− mice (Fig. 3B; Supplementary Fig. S3C). In line with the APC activation results, the nontargeted anti-CD40 agonist triggered CD8⁺ T-cell proliferation in Fap^−/− mice. In contrast, T-cell proliferation was induced with FAP-moCD40 in wild-type mice but not in Fap^−/− mice (Fig. 3C).

Thus, besides confirming the strict FAP dependency of the FAP-CD40 molecule, our data suggest that even low levels of FAP can induce activity of FAP-CD40. In that context, in particular FAP-expressing FRCs might play a role in crosslinking FAP-CD40 in the LNs, resulting in an in situ nodal DC activation.

After verifying FAP-CD40's functionality and FAP specificity, we investigated the antitumor effects of FAP-dependent CD40 stimulation in the syngeneic subcutaneous MC38-FAP tumor model in huCD40tg mice. Compared with human tumors, subcutaneous MC38 tumors show decreased desmoplasia and contain lower amounts of stromal FAP (38). As these reduced murine FAP levels might be a limiting factor for our in vivo studies, we chose to use MC38 cells with recombinant FAP expression. MC38-FAP tumors are characterized by an accumulation of T cells at the tumor border and moderate T-cell infiltration as well as high FAP expression levels at baseline (Fig. 4A; Supplementary Fig. S4A).

We first compared the effects of FAP-huCD40 and the FcγR-dependent CD40 agonist SGN40 moIgG1 on tumor growth inhibition and immune activation (Fig. 4B). Strikingly, a single dose of 13.3 mg/kg FAP-huCD40 induced a complete tumor remission in all of the treated mice while an equivalent dose of 10 mg/kg (doses matched based on molar concentrations) SGN40 moIgG1 had no impact on tumor growth. When tumor-free mice in the FAP-huCD40 group were rechallenged with MC38-FAP tumor cells, the tumors were rejected in all the animals (Fig. 4C).

Consistent with observations from other groups using high doses of systemically administered nontargeted anti-CD40 agonists (16, 17), substantial side effects occurred in the SGN40 moIgG1-treated group. Mice receiving SGN40 moIgG1 experienced an average body weight loss of more than 15% four days after therapy injection (Fig. 4D). Of note, 2 out of 9 mice treated with SGN40 moIgG1 had to be sacrificed due to a reduction in body weight of more than 20%. The rest of the mice recovered six days after the weight drop. Moreover, SGN40 moIgG1 administration induced liver damage evidenced by an increase in liver enzymes and hepatocellular degeneration observed by histopathology (Fig. 4E). In contrast, none of the FAP-huCD40–treated animals displayed any signs of toxicity (Fig. 4D and E).

Thus, a single high dose of FAP-huCD40 induced efficient long-term protective antitumor immunity without causing the side effects typically observed upon treatment with systemic CD40 agonists.

We next aimed at obtaining a thorough understanding of the differences in immune cell stimulation triggered by FAP-huCD40 compared with SGN40 moIgG1.

The analysis of the levels of 23 inflammatory cytokines in serum versus tumor of treated mice three days after therapy injection revealed that FAP-huCD40 mediated an immune activation largely confined to the tumor site. Among the cytokines that displayed the strongest tumoral increase in response to FAP-huCD40 were the Th cell type 1 response-inducing IL12, as well as C-C motif chemokine 5 (CCL5), which plays an important role in attracting DCs to the tumor (Fig. 4F; Supplementary Fig. S4B; ref. 39). Unlike FAP-huCD40 and indicative of its systemic immunostimulatory effects, SGN40 moIgG1 failed to evoke a strong cytokine response in the tumor but induced a broad range of cytokines in the serum (Fig. 4F).

Furthermore, FAP-huCD40 significantly enhanced DC activation in the tumor-draining and nondraining LNs as well as in the tumor three days posttreatment. In the spleen, only minor DC activating effects of FAP-huCD40 were observed. In contrast, SGN40 moIgG1 induced a superior nodal and splenic but no tumoral DC activation (Fig. 4G). A strong B-cell activation triggered by SGN40 moIgG1 was detected in the LNs and to a lesser extent in the spleen and tumor. Compared with SGN40 moIgG1, FAP-huCD40 induced no or an inferior activation of B cells in LNs and spleen. Tumoral B-cell activation in FAP-huCD40 and SGN40 moIgG1–treated mice was, depending on the investigated activation marker, similar or more pronounced with SGN40 moIgG1 (Fig. 4H; Supplementary Fig. S4C).

Interestingly, only three days after therapy administration we observed a strongly enhanced expression of the early activation marker CD69 on cytotoxic T cells and an increased CD8⁺ T cell to regulatory T cell (Treg) ratio in tumors of FAP-huCD40- but not SGN40 moIgG1–treated animals (Fig. 4I).

To substantiate our findings, we compared FAP-huCD40's in vivo effects to those of another nontargeted agonistic anti-CD40 antibody. Selicrelumab is classified as one of the human CD40 agonists with highest potency, mediating CD40 activation in a partially FcγR crosslinking-independent fashion (7, 34, 40, 41).

While a high dose of 13.3 mg/kg FAP-huCD40 induced rapid tumor elimination in 8 out of 10 MC38-FAP tumor-bearing huCD40tg mice, an equivalent dose of selicrelumab merely conferred low growth-inhibitory effects (Fig. 5A; Supplementary Fig. S5A for individual tumor growth curves).

Coinciding with our previous observations, only the nontargeted anti-CD40 agonist selicrelumab, but not FAP-huCD40 induced a severe body weight loss (Supplementary Fig. S5B). Multiplexed cytokine analysis and investigation of DC activation in different organs confirmed the enhanced intratumoral effects of FAP-huCD40 as opposed to the strong systemic immunostimulatory effects of selicrelumab, similar to those of SGN40 moIgG1 (Supplementary Fig. S5C and S5D). Moreover, analysis of the expression of 518 immune response–related genes in the tumor one day after treatment revealed an increased tumoral effect of FAP-huCD40 compared with selicrelumab. Immunologic pathways such as IFN, toll-like receptor, and cytokine signaling were among the ones particularly impacted by FAP-huCD40 treatment (Supplementary Fig. S5E and Supplementary Table S2).

In accordance with these results, investigation of the biodistribution of the highly dosed fluorescently labeled antibodies over 72 hours showed increased accumulation of FAP-huCD40 versus selicrelumab in the tumor tissue. Both molecules showed accumulation in the tumor-draining and nondraining LNs, with selicrelumab displaying higher nodal levels over time. Clearing of FAP-huCD40 as well as selicrelumab from blood and organs of perfusion (spleen, lung) as well as excretory organs (liver, kidney) started after 2 to 24 hours postinjection. It is noteworthy that selicrelumab showed a delayed clearance in the spleen associated with a higher plateau concentration compared to FAP-huCD40 (Fig. 5B and C).

FAP-huCD40 outperformed the nontargeted anti-CD40 agonists SGN40 moIgG1 and selicrelumab in vivo and conferred remarkable efficacy with a single high dose. On the basis of these findings, we next asked whether a lower concentration would diminish our molecule's efficacy. Indeed, potent antitumor responses were only mediated by a FAP-huCD40 dose of 13.3 mg/kg. Lowering the dose to 4 mg/kg resulted in merely a slight tumor growth inhibition and did not induce significantly better anti-tumor efficacy compared with the corresponding low dose of selicrelumab (3 mg/kg; Fig. 5D).

Comparing the 13.3 mg/kg versus the less efficacious 4 mg/kg FAP-huCD40 treatment revealed a major impact of the dosing on intratumoral DCs. A threefold dose increase from 4 mg/kg to 13.3 mg/kg translated to threefold more labled FAP-huCD40 molecule bound to DCs in the tumor (Fig. 5E). These pronounced dose-related differences could not be seen for tumoral B cells (Supplementary Fig. S5F). In the LNs and spleen, dose-dependent effects of FAP-huCD40 on DCs were only moderate or absent. In contrast, high dosing of selicrelumab led to a strong increase of labeled antibody bound to DCs in LNs and spleen but not in the tumor (Fig. 5E). Hence, a high dose of FAP-huCD40 seems in particular crucial for the bispecific antibody to reach intratumoral DCs.

Highly dosed FAP-huCD40 showed a remarkable efficacy in MC38-FAP tumors. To corroborate FAP-CD40's in vivo effects, we tested our bispecific molecule in a second, T-cell–excluded pancreatic cancer model. Human CEA transgenic (huCEAtg) mice were subcutaneously implanted with murine pancreatic KPC-4662 tumor cells recombinantly expressing human CEA (KPC-4662-huCEA). Unlike subcutaneous MC38 tumors, subcutaneous KPC-4662 tumors develop thick stromal banding patterns that can also be frequently observed in human tumors (30, 38). Consequently, we detected a high, stroma-restricted FAP expression in subcutaneous KPC-4662-huCEA tumors (Fig. 6A). FAP-moCD40 mediated distinct tumor growth–inhibitory effects in KPC-4662-huCEA tumor-bearing mice (Fig. 6B). Tumor regressions ranging from 20% to 90% were observed in 6 out of 10 FAP-moCD40-treated animals (Fig. 6C). Strikingly, the FAP-moCD40–induced tumor growth inhibition was preceded by a strong CD4⁺FoxP3⁻ and CD8⁺ T-cell tumor infiltration detected eight days after FAP-moCD40 administration (Fig. 6D–F). Thus, our data suggest that in presence of abundant amounts of stromal FAP, a high dose of FAP-CD40 induced inflammation of KPC-4662-huCEA tumors, thereby effectively inhibiting tumor growth.

On the basis of the finding that high doses of FAP-huCD40 are required to mediate potent antitumor responses, we tested whether these high doses are tolerated in a nonrodent toxicology species. Two groups of cynomolgus monkeys were consecutively dose-escalated weekly with two different dose increments. The first group of animals received three doses from 10 to 100 mg/kg FAP-huCD40 and the second group three doses from 200 to 400 mg/kg (Supplementary Fig. S6A). All treated animals were exposed to high levels of FAP-huCD40 (Supplementary Fig. S6B). High concentrations of systemically administered FAP-huCD40 did not result in mortalities or FAP-huCD40–related effects on clinical observations or body weight (Supplementary Table S3A). Multiple phase I studies with agonistic anti-CD40 antibodies encountered dose-limiting toxicities including CRS, liver function test abnormalities, and a decrease in platelet counts (12–15). Therefore, we examined the FAP-huCD40–treated animals for secretion of eleven different cytokines, and analyzed hematology, clinical chemistry, and histopathology. For most of the cytokines measured, the concentrations were below the limit of quantification. For IL8, CCL2 and CCL5 results were measurable, but showed no obvious dosing effects (Supplementary Table S3B). No changes in liver enzymes (Supplementary Fig. S6C) were observed. Coagulation parameters such as platelet counts, D-dimer, activated partial thromboplastin time (APTT), prothrombin time (PT), and fibrinogen remained similar to untreated controls (Supplementary Fig. S6D). Taken together, administration of FAP-huCD40 at very high doses up to 400 mg/kg was well tolerated and did not result in any CD40-related toxicities in cynomolgus monkeys.

In this study, we introduce for the first time a CD40 agonistic antibody whose activity is strictly dependent on the presence of FAP. By creating a FAP-restricted anti-CD40 molecule, we aimed to enable safe administration of high doses to overcome peripheral CD40 sink effects. We envisioned that the increased therapeutic window but also the targeting of our molecule to the tumor area via FAP could lead to a more efficient and selective activation of tumoral APCs, thereby enhancing tumor antigen (cross-) presentation and consequently antitumor immunity.

The results of our in vitro APC activation and T-cell priming assays clearly indicated that FAP-CD40 provides a strong CD40-stimulating signal in a FAP-specific manner. These findings were recapitulated by in vivo studies in huCD40tg mice, in which FAP-huCD40 predominantly triggered potent immune activation in the tumor in contrast to the nontargeted CD40 agonists SGN40 moIgG1 and selicrelumab, which induced strong peripheral but minor intratumoral effects. Restriction of CD40 stimulation to FAP⁺ areas abolished the pronounced side effects caused by systemic CD40 activation. A high dose of FAP-huCD40 could therefore be safely administered. The good tolerability of FAP-CD40 was confirmed in cynomolgus monkeys, indicating a favorable safety profile of our therapy for clinical development. Of note, our in vivo efficacy data demonstrated that the administration of a high dose of FAP-targeted anti-CD40 agonist is indispensable for the induction of successful antitumor immunity; only when high amounts of FAP-huCD40 were injected, complete MC38-FAP tumor regression and long-term protection of treated mice were induced. In contrast, minor to no effects on tumor growth inhibition were observed with highly dosed SGN40 moIgG1 and selicrelumab.

With SGN40 moIgG1 and selicrelumab, we selected two nontargeted CD40 agonists that possess different FcγR crosslinking requirements as well as potencies. The huIgG2 selicrelumab is one of the strongest CD40 agonists and acts partially independent of crosslinking (7, 40, 41). It is noteworthy that the affinities of huIgG2s are similar for both the murine and human FcγRIIB (42, 43), the FcγR subclass considered most crucial for Fc-mediated crosslinking of anti-CD40 agonists (35, 44). Hence, in the huCD40tg mouse model, both FcγR-independent and FcγR-dependent activation of selicrelumab is possible.

For SGN40 moIgG1, the Fv domains of the crosslinking-dependent human anti-CD40 agonist SGN40 were combined with a murine IgG1 backbone to enable a good functionality in huCD40tg mice. Of note, murine IgG1s have a higher affinity to the murine FcγRIIB compared to the affinity of human IgG1s for the human FcγRIIB (42, 43). Thus, crosslinking and consequently biological activity of SGN40 moIgG1 in huCD40tg mice is presumably enhanced relative to the activity of SGN40 huIgG1 in humans.

Therefore, it seems unlikely that insufficient crosslinking could account for the lack of efficacy of the nontargeted CD40 agonists in our huCD40tg mouse model.

This was supported by the analysis of immune cell activation induced by FAP-huCD40 versus SGN40 moIgG1 and selicrelumab: the overall immuno-pharmacodynamic effects conferred by the non-targeted anti-CD40 antibodies were not less potent per se. On the contrary, they triggered a stronger activation of B cells, DCs, as well as T cells in the LNs and the spleen, compared with FAP-huCD40. The nontargeted CD40 agonists, however, failed to induce DC or T-cell responses in the tumor, whereas with FAP-huCD40 tumoral DC and, presumably concomitant, tumoral CD8⁺ T-cell activation was observed. Hence, we hypothesize that stimulation of DCs in the tumor is essential for FAP-huCD40's efficacy. Two factors could account for the superiority of FAP-huCD40 in activating tumor DCs: FAP-mediated crosslinking might be more effective in the tumor than FcγR-crosslinking and, as evidenced by our biodistribution data, targeting to FAP enhanced the intratumoral accumulation of our antibody.

Indeed, our dose–response study supported the hypotheses that (i) the amount of CD40 agonist in the tumor is a limiting factor for efficient tumoral DC activation and that (ii) this tumoral DC activation could be crucial for FAP-CD40's efficacy: decreasing the FAP-huCD40 dose led to a significantly lower amount of antibody reaching tumor DCs, accompanied by a loss of efficacy. In addition, mechanistic studies of the accompanying paper by Labiano and colleagues, entailing the use of cDC1-deficient Batf3^−/− mice and T-cell depletion, indicate the requirement of an activation of the cDC1-CD8⁺ T-cell axis for FAP-CD40–mediated antitumor immunity (45).

While proposing an instrumental role of tumoral DC activation for FAP-CD40's efficacy, we found evidence that low FAP expression by nodal FRCs might as well contribute to FAP-CD40's mode of action. In FAP-huCD40–treated animals, an accumulation of the therapeutic antibody as well as DC activation was not only observed in the tumor, but also, albeit to a lower degree, in tumor-draining and nondraining LNs. In contrast, in FAP-negative spleens no antibody accumulation or substantial effects of FAP-huCD40 on DCs were detected. It is possible that activated nodal DCs result from intratumoral stimulation and subsequent homing to the LNs. However, taken together, our results suggest that in situ DC activation in the LNs is initiated by FAP-CD40. First, DC activation was similar in the tumor-draining versus nondraining LNs of FAP-huCD40–treated animals. This would likely not be the case if only trafficking of tumoral DCs accounted for the observed LN DC activation. Second, our in vitro studies showed that FAP expression on nodal FRCs is sufficient to trigger FAP-moCD40 crosslinking. Third, nodal DC activation was also observed in non–tumor-bearing mice when FAP-moCD40 was administered in a vaccination setting. Thus, it seems likely that FAP-CD40 acts directly in LNs. This might be a beneficial feature, given the importance of, in particular, the tumor-draining LNs for DC-mediated cross-presentation of tumor antigen (46). Further investigations are required to determine to which extent this nodal activity contributes to the efficacy of our molecule.

Surprisingly, in the LNs of both tumor-bearing and non–tumor-bearing mice, FAP-CD40 activated DCs, whereas B-cell activation levels remained unchanged. Moreover, FAP-huCD40 was superior to nontargeted CD40 agonists in activating DCs but not B cells in the tumor. This is in opposition to our in vitro data, which clearly demonstrated the ability of FAP-CD40 to trigger potent B-cell responses. This discrepancy might be partially explained by a lack of colocalization of FAP⁺ cells and B cells in vivo. FAP-expressing FRCs are known to directly interact with DCs in the nodes via the C-type-lectin-like-2-podoplanin axis (47), which might allow the close contact needed for FAP-CD40–mediated activation. In contrast, the quality or frequency of nodal B cell–FRC interactions might be insufficient. Nonetheless, given the broad FAP expression in our MC38-FAP tumor model, it seems unlikely that a dearth of colocalization alone accounts for the diminished tumoral B-cell activation in FAP-CD40–treated animals.

The overall superior B-cell–activating effects of non-targeted CD40 agonists compared with FAP-huCD40 suggest that in the MC38-FAP tumor model B-cell activation might not play a major role for the potent antitumor activity of FAP-huCD40.

It is worth mentioning that we detected a slight FAP-dependent B-cell activation in the spleens of FAP-CD40–treated mice. As splenic FRCs do not express FAP (20), this effect might be caused by migrating B cells from FAP⁺ areas rather than by an in situ splenic activation. Low frequencies of FAP-expressing cells are not only reported for the LNs but also other organs, including the bone marrow (48), which were not the subject of our studies but might play a role for the observed splenic B-cell activation.

Our investigations primarily focused on the impact of FAP-CD40 on DCs and B cells; however, numerous studies provided experimental evidence for a beneficial effect of CD40 agonists on macrophages. Activation of this cell subset and reprogramming toward a proinflammatory tumor-associated macrophage phenotype, as well as induction of tumor stroma degradation, are among the described mechanisms by which anti-CD40 antibodies can drive macrophage-mediated antitumor immunity (49, 50). Potential effects of FAP-CD40 on macrophages were not addressed within the scope of this study but are under investigation.

Utilizing the MC38-FAP and KPC-4662-huCEA murine tumor models, we demonstrated that a single high dose of FAP-huCD40 conferred remarkable antitumor efficacy. With MC38-FAP, we used a tumor cell line with recombinant FAP expression, while our studies in KPC-4662-huCEA tumor-bearing mice allowed us to confirm FAP-CD40's antitumor activity in a tumor model in which FAP expression was restricted to the tumor stroma. Regarding the clinical development of our bispecific antibody, it will be pivotal to explore the effects of FAP-CD40 in tumors with varying FAP expression levels and localization patterns. Importantly, the accompanying article by Labiano and colleagues demonstrates a notable antitumor efficacy of FAP-CD40 in combination with radiotherapy in a murine orthotopic head and neck tumor model, which recapitulates the tumor stromal FAP expression patterns of human head and neck squamous cell carcinomas (45).

In addition, as FAP expression can be induced in fibroblasts during wound healing (18), it will be important to evaluate whether patients' surgical wounds could pose a potential risk for side-effects during FAP-CD40 treatment.

Another vital aspect to be considered regarding clinical development of FAP-CD40 will be the selection of combination partners. An attractive combination strategy might be pairing of CD40 agonists with checkpoint inhibitors, which has been shown to enable effective priming and durable T-cell responses (9). Moreover, chemo- or radiotherapy entailing immunogenic cell death and spilling of tumor antigen is believed to function as potent in situ vaccination together with CD40 agonists (8, 51, 52). Indeed, the promising preclinical results that Labiano and colleagues demonstrated by applying hypofractionated radiotherapy plus FAP-CD40 treatment provide a strong rationale for this combinatorial strategy (45).

In summary, our data show that agonizing CD40 and thereby bridging the gap between innate and adaptive immunity, is unequivocally a powerful tool for CIT. Yet, the full therapeutic capacity of the CD40 pathway has not been leveraged by previously tested nontargeted anti-CD40 antibodies. On the basis of our preclinical evidence, we propose that a FAP-targeted CD40 agonist can be safely administered at high doses to unleash its full potential in the tumor area and, with the right combinatorial approach, holds great promise for clinical applications.

E. Sum reports a patent for WO 2018/185045 A1 pending to F. Hoffmann-La Roche AG and ownership of Roche stock (options). M. Rapp reports a patent for WO 2018/185045 A1 pending to F. Hoffmann-La Roche AG, a patent for WO 2020/070041 A1 pending to F. Hoffmann-La Roche AG, and ownership of Roche stock (options). P. Fröbel reports ownership of Roche stock options. M. Le Clech reports a patent for WO 2020/070041 A1 pending to F. Hoffmann-La Roche AG and ownership of Roche stock (options). H. Dürr reports a patent for WO 2018/185045 A1 pending to F. Hoffmann-La Roche AG and a patent for WO 2020/070041 A1 pending to F. Hoffmann-La Roche AG. A.M. Giusti reports ownership of Roche stock options. M. Perro is an employee of Roche at the time of the work described in the manuscript. L. Kunz reports personal fees from Roche Glycart AG during the conduct of the study, as well as personal fees from Roche Glycart AG outside the submitted work. E. Menietti reports personal fees from Roche during the conduct of the study and personal fees from Roche outside the submitted work. P. Brünker reports a patent for WO 2018/185045 A1 pending to F. Hoffmann-La Roche AG, a patent for WO 2020/070041 A1 pending to F. Hoffmann-La Roche AG, and participation in Roche Connect program (stock options). U. Hopfer reports ownership of Roche non-voting equity securities. M. Lechmann reports participation in Roche Connect and ownership of Roche stocks. V. Nicolini reports other from Roche during the conduct of the study and other from Roche outside the submitted work; in addition, V. Nicolini has a patent 10577429 issued and is a Roche employee. S. Labiano reports grants and personal fees from Roche Glycart during the conduct of the study; in addition, S. Labiano has a patent for EP20207768.1 pending to F. Hoffmann-La Roche AG. F. Weber is an employee of Roche and participates in the "Roche Connect" program (owns stock of Roche). T. Emrich is an employee of Roche Diagnostics GmbH, Germany and reports Roche bonus shares and profit certificates. P. Romero reports grants from Roche pRED during the conduct of the study and personal fees from MaxiVax outside the submitted work; in addition, P. Romero has a patent for AD2838 EP BS pending, has a patent for AC1689 PCT BS pending, and is editor-in-chief of Journal for Immunotherapy of Cancer. C. Trumpfheller reports a patent for EP20207768.1 pending to F. Hoffmann-La Roche AG, a patent for WO 2018/185045 A1 pending to F. Hoffmann-La Roche AG, and a patent for WO 2020/070041 A1 pending to F. Hoffmann-La Roche AG; C. Trumpfheller also reports ownership of Roche stock (options). P. Umaña reports a patent for WO 2018/185045 A1 pending to F. Hoffmann-La Roche AG, a patent for WO 2020/070041 A1 pending to F. Hoffmann-La Roche AG, and a patent for EP20207768.1 pending to F. Hoffmann-La Roche AG; P. Umaña also reports ownership of Roche shares. No disclosures were reported by the other authors.

E. Sum: Conceptualization, formal analysis, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. M. Rapp: Conceptualization, formal analysis, supervision, validation, investigation, visualization, methodology, project administration, writing–review and editing. P. Fröbel: Formal analysis, investigation, methodology. M. Le Clech: Conceptualization, formal analysis, investigation, visualization, methodology. H. Dürr: Conceptualization, resources. A.M. Giusti: Formal analysis, visualization. M. Perro: Conceptualization, visualization. D. Speziale: Formal analysis, investigation, visualization, writing–original draft. L. Kunz: Formal analysis, investigation, visualization, methodology, writing–original draft. E. Menietti: Formal analysis, investigation, visualization, methodology, writing–original draft. P. Brünker: Conceptualization. U. Hopfer: Conceptualization, formal analysis, visualization, writing–original draft. M. Lechmann: Conceptualization, formal analysis, visualization. A. Sobieniecki: Investigation. B. Appelt: Investigation. R. Adelfio: Investigation. V. Nicolini: Investigation, visualization. A. Freimoser-Grundschober: Resources, investigation. W. Jordaan: Investigation. S. Labiano: Conceptualization. F. Weber: Formal analysis. T. Emrich: Formal analysis. F. Christen: Investigation. B. Essig: Investigation. P. Romero: Conceptualization, supervision. C. Trumpfheller: Conceptualization, supervision, validation, project administration, writing–review and editing. P. Umaña: Conceptualization, supervision, writing–review and editing.

We thank Maximiliane König, Klaus Neff, and Anton Jochner (RICM) for their contributions to design, production, and supply of the bispecific antibody constructs used in this study. We thank Christophe Chabbert and Said Aktas (RICZ) for their help with the statistical and NanoString data analyses. We also thank Junichiro Sonoda (Genentech) for kindly providing the Fap^−/− mice. We gratefully acknowledge the RICZ Cell Technologies group for the generation of recombinant cell lines used for in vitro and in vivo studies. Moreover, we would like to thank the members of the RICZ in vivo pharmacology team for their support of the mouse studies. Thanks to Claudia Haftmann (RICZ), Olivera Cirovic (RICB), and Dörte Klostermeyer Rauber (Hoffmann-La Roche Patent Department) for the discussions and critical reading of the manuscript. We are also grateful to the RICZ Biochemistry and Downstream Processing teams for the supply, fluorescent labeling, biotinylation, and endotoxin purification of reagents used in this study. We thank Florian Kast for his support in generating BioRender graphs. This work was funded by F. Hoffmann-La Roche AG. Pedro Romero was supported by grants from Swiss Cancer League (KFS-4404–02–2018) and the Swiss National Science Foundation (310030_182735).

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.