T Cell–Mediated Targeted Delivery of Anti–PD-L1 Nanobody Overcomes Poor Antibody Penetration and Improves PD-L1 Blocking at the Tumor Site

During the last decade, mAbs targeting receptors and ligands regulating the adaptive immune system have emerged as a powerful therapy against cancer. Among the so-called immune checkpoint inhibitors (ICI), antibodies targeting the receptor programmed death 1 (PD-1) and its ligand (PD-L1) are currently the mainstays of cancer immunotherapy. PD-L1 is expressed by some tumor cells and some immune cells. Its interaction with PD-1 expressed on T cells negatively regulates T-cell activity (1). PD-L1–specific antibodies have been used to restore CD8⁺ T-cell activation resulting in unprecedented survival improvements in patients with non–small cell lung cancer or urothelial carcinoma (2, 3). The clinical activity of PD-L1–specific antibodies has also been shown in breast cancer, renal cell carcinoma, and melanoma (4–6). Still, a substantial fraction of patients receiving ICI do not respond to the treatment or undergo disease progression after an initial response phase (7). Different resistance mechanisms have been identified, including a lack of tumor-specific T-cell generation (8), dysfunction of T cells (9), and the presence of immunosuppressive cells in the tumor microenvironment (TME; ref. 10). Discovery of novel strategies to overcome resistance and increase the number of responding patients remains one of the main challenges in cancer immunotherapy.

A major drawback of ICI treatment is the development of immune-related adverse events (irAE) triggered by the activation of T cells recognizing healthy tissue (11). These events can lead to death in a fraction of patients (12). Different strategies have therefore been developed to reduce the systemic toxicity of immune checkpoint blockade. Among these, intra or peritumoral injection of ICI reduced toxicity while preserving or even increasing therapeutic efficacy compared with systemically injected antibody in mouse models (13, 14). Although these approaches represent an interesting proof of concept, their clinical value is limited since patients with advanced disease typically present with a high number of metastases mostly located in organs such as the lung or liver.

It has been proposed that tumor-targeting T cells could be used as delivery vehicles to bring ICI to the tumor site. Although promising results have been reported in the context of adoptive transfer of chimeric antigen receptor (CAR) T cells, there have been only a few studies so far (15–17). Further research and development are needed to demonstrate the therapeutic potential of local delivery strategies and bring them closer to clinical use.

Nanobodies are single variable domain (VHH) fragments derived from camelid heavy-chain antibodies. They represent the smallest antigen-binding antibody fragments, with a molecular size of around 15 kDa, and are known for their excellent stability and solubility (18, 19). Their good tumor penetration ability is counterbalanced by a very short half-life in the blood stream, due to renal filtration, and these characteristics have rather supported their development as imaging tools (20). In this study, we investigate using tumor-specific CD8⁺ T cells to bring PD-L1–blocking nanobodies to the tumor site, so that we capitalize on the properties of nanobodies to ensure both good tumor penetration and minimal systemic exposure.

Using a mouse colon tumor model, we compared the therapeutic efficacy as well as the intratumoral delivery of PD-L1–blocking therapy based on local secretion of nanobodies or systemic injection of antibody. We demonstrated that our local delivery approach improved tumor control compared with classical antibody injection. We also observed limited tumor penetration of systemically injected PD-L1–specific antibody, preventing it from reaching PD-L1–expressing cells in the TME. This barrier was overcome by local secretion of PD-L1–blocking nanobodies.

MSCV-Thy1.1-DEST retroviral vector was obtained from Addgene (plasmid #17442). The IRES-Thy1.1 cassette was removed by double digestion with BglII–MluI restriction enzymes. The plasmid was then ligated with self-designed sticky end paired oligos ordered from IDT (https://www.idtdna.com; Supplementary Data S1A) to obtain the “MSCV empty vector”. Amino acid sequence for anti–PD-L1 nanobody C7 is described in Broos and colleagues (21), and amino acid sequence of the anti–PD-L1 nanobody 5DXW is available in the RCSB open access protein database (22). Amino acid sequences were reverse translated using Mac Vector (Version 17.5.6). For each nanobody, point mutations were introduced to obtain the C7-T57A and 5DXW-T61V nonglycosylated mutants. Each nanobody was flanked by two “G-G-G-G-S” linkers. An IL2 signaling peptide was added in 5′ and an HA Tag was added in 3′, resulting in the four constructs displayed in Supplementary Data S1B. The four nanobody-encoding DNA constructs were ordered as gBlocks from IDT and cloned into MSCV empty vector (using BglII and SalI restriction enzymes sites). The 5DXW-T61V DNA construct was cloned in the pET-21b+ vector (Sigma-Aldrich, #69740) using NcoI and SalI restriction enzymes sites. Sequence for truncated ovalbumin missing 49 amino acids in the N-terminal part to avoid secretion (23) was ordered as a gblock from IDT and cloned into the lentiviral vector pTM898 provided by Professor Thomas Michiels (De Duve Institute, Brussels, Belgium). DNA fragments encoding full-length Cd274 (gene ref:60533) or Pdcd1lg2 (gene ref: 58205) genes were ordered as gblocks from IDT and cloned into pTM899 lentiviral vector, provided by Professor Thomas Michiels (De Duve Institute, Brussels, Belgium), to obtain pTM899–mPD-L1 and pTM899–mPD-L2 vectors. Supplementary Data S2 describes the different vectors used in this study.

Escherichia coli ClearColi BL21 (DE3) chemically competent cells (Immunosource, #60810-1) were transformed by heat-shock with the pET-21b+ plasmid expressing the C-terminally His-tagged PD-L1 nanobody. They were then grown in Lysogeny Broth medium (LB Miller, Sigma-Aldrich, #L3522) supplemented with 50 μg/mL of kanamycin (Merck Life Science, #K-4000) at 37°C. When cells reached an OD600 of 0.6, expression of the nanobody was induced with 1 mmol/L isopropyl β-d-1-thiogalactopyranoside (IPTG, Promega, #V3953) for 3 hours at 37°C. After harvest at 4,000 × g, for 15 minutes, at 4°C, the bacterial pellets were frozen and stored at −20°C. Cells were resuspended in buffer A (100 mmol/L Tris pH8, 300 mmol/L NaCl) supplemented with a protease inhibitor cocktail (cOmplete EDTA-free, Merck Life Science, #11836170001) and lysed by two passages through a French press at 1,500 psi. After clearing the lysate at 40,000 × g for 30 minutes, at 4°C, the soluble protein extract was filtered through a steritop Milipore Express PLUS 0.22 μm PES filter (Merck Millipore, #S2GPT05RE) and applied on a 1 mL His-Trap HP column (Cytiva, #17524801), using the AKTÄ pure system (Cytiva, #17524801). After washing the column with buffer A, the protein was eluted with a linear gradient up to 100% buffer B (100 mmol/L Tris pH8, 300 mmol/L NaCl, 300 mmol/L imidazole) over 25 column volumes. To complete the purification, the protein was then applied on a HiLoad 16/60 Superdex 75 size-exclusion chromatography column (GE Healthcare, #28989333), using PBS (MP Biomedicals, #092810307) as running buffer.

B16F10-OVA cell line was a gift from Professor Muriel Moser (Université Libre de Bruxelles, Brussels, Belgium) in 2015. B16F10-OVA cells were transfected with pTM899-mPD-L1 lentiviral vector using TransIT-LT1 (Mirus, #MIR 2306) and kept under puromycin (Invivogen, #ant-pr-1) selection. MC38 cell line was a gift from Professor Mark Hull (University of Leeds, Leeds, United Kingdom) in 2017. MC38 cells were transduced by adding fresh pTM898-OVA lentivirus on the cell culture for 24 hours (see section below for virus production). Cells were then selected with geneticin (G-418, Roche, #04727878001) to obtain an in house MC38-OVA cell line, no clonal selection was performed. MC38-OVA cells were transduced by adding fresh pTM899-mPD-L1 or pTM899-mPD-L2 lentivirus on the cell culture for 24 hours (see section below for virus production). Cells were then selected with puromycin to obtain MC38-OVA PD-L1⁺ and MC38-OVA PD-L2⁺ cell lines, respectively. T429.18 clone was derived from an induced Amela TiRP tumor referred to as T429, as previously described (24). HEK 293T cell line was a gift from Professor Kris Thielemans (Vrije Universiteit Brussel, Brussels, Belgium) in 2006. Authentication was performed in November 2019 by short tandem repeat profiling (Promega Powerplex hs 16). Murine tumor cells (B16F10-OVA, T429.18 and all MC38 cell lines) and HEK 293T cells were cultured in Iscove's modified Eagle medium (IMDM; Gibco, #12440053) supplemented with L-arginine (0.55 mmol/L, Merck, #181003), L-asparagine (0.24 mmol/L, Merck, #101566), glutamine (1.5 mmol/L, Merck, #100289), β-mercaptoethanol (50 μmol/L, Sigma, #M3148), 10% FBS (Sigma-Aldrich, #F7524), 100 IU/mL penicillin, and 100 μg/mL streptomycin (Pen Strep, Gibco, #15140148). OT-I and TCRP1A CD8⁺ T cells were cultured in the same medium supplemented with 25 U/ml IL2 (Proleukin, Novartis). L1210.P1A.B7-1 cells expressing the antigen P1A and the surface protein B7-1 have been described previously (24) and were cultured in DMEM (Gibco, #11966025) supplemented with 10% FBS, 100 IU/mL penicillin, and 100 μg/mL streptomycin. HEK 293T and tumor adherent cells were detached with PBS buffer supplemented with 2 mmol/L EDTA. Cells were routinely tested for Mycoplasma using MycoAlert Detection Kit (Lonza, #LT07-118). The last test was performed in September 2020. Cells underwent maximum 20 passages after thawing.

OT-I transgenic mice (C57BL/6-Tg(TcraTcrb)1100Mjb/Crl, Charles River Laboratory) and C57BL/6J Ola Hsd mice (Envigo) were purchased and bred at the animal facility of the Ludwig Institute for Cancer Research (Brussels, Belgium). B10.D2;Ink4a/Arf^flox/flox (referred as TiRP-10B^−/−) mice were produced in-house (24) and were used as recipients for T429.18 tumor transplantation experiments. TCRP1A mice heterozygous for the H-2L^d/P1A_35–43-specific TCR transgene were produced in house and were kept on the B10.D2;Rag1^−/− background (24, 25). All animal procedures were performed in accordance with national and institutional guidelines for animal care, and with the approval of the Comité d'Ethique pour l'Expérimentation Animale from the Secteur des Sciences de la Santé, UCLouvain (2019/UCL/MD/24).

PD-L1 knock-out in MC38-OVA cells (MC38-OVA PD-L1^KO) was performed by electroporation of Cas9 ribonucleoprotein (RNP) complexes. Alt-R crRNA for murine PD-L1 (Mm.Cas9.CD274.1.AQ) was obtained from IDT. Alt-R crRNA and Alt-tracrRNA (IDT) were annealed in a thermocycler and then mixed with Alt-R S.p. Cas9 Nuclease V3 (IDT, #1081058). Tumor cells were nucleofected with Cas9-RNP complex in supplemented SF Cell Line Nucleofector Solution (Lonza, #V4XC-2032), using a Lonza 4D-Nucleofector (HEK-293 program). After 12 days, 100 ng/mL recombinant mouse IFNγ (Gibco, #PMC4031) was added to the culture medium for 48 hours. Cells were then collected, and surface stained for PD-L1 (PE anti-mouse PD-L1 clone 10F.9G2, BioLegend, #124308). PD-L1–negative cells were sorted with a FACSAria III.

For HEK 293T transfection, cells were plated the day before, at a concentration of 300,000 cells/mL in full supplemented IMDM but without penicillin–streptomycin. On the day of transfection, a mixture of DNA plasmid and TransIT-LT1 (Mirus, #MIR 2306) was prepared according to manufacturer's protocol in Opti-MEM medium (Gibco, #11058-021) and added to the cell culture. The day after transfection, medium was collected and replaced by full supplemented IMDM. For nanobody expression in HEK 293T cells, cells were transiently transfected with the different MSCV retroviral vectors and cells were collected after 6 days. For viral production, HEK 293T cells were transfected with packaging and expression plasmids. For retrovirus, we used a mixture of PCL-Eco plasmid (Addgene, #12371) and MSCV-5DXW-T61V plasmid or MSCV empty plasmid; for lentivirus, we used a mixture of MD2.G-pSPAX2 plasmids (Addgene, #12259 and 12260) and pTM898-Ova, pTM899-mPD-L1 or pTM899-mPD-L2 lentiviral vectors. Virus-containing medium was collected after 2 and 3 days and filtered through 0.45-μm filters (Merck Millipore, #SLHVR33RS).

Spleen and lymph nodes from OT-I and TCRP1A transgenic female mice aged between 8 and 12 weeks were smashed using the plunger of a syringe. After red blood cells lysis (RBC lysis buffer, eBioscience, #00-4300-54), CD8⁺ T cells were isolated using CD8a (Ly-2) mouse Microbeads (Miltenyi Biotec, #130-117-044). OT-I T cells were stimulated with CD3/CD28 Dynabeads Mouse T activator (Gibco, #11453D). TCRP1A CD8⁺ T cells were stimulated with irradiated L1210.P1A.B7-1 cells. Two or three days after activation, T cells were transduced with MSCV-5DXW-T61V vector or MSCV empty vector in the presence of 50 μg/mL Protamine Sulfate (Sigma, #P3369-10G) and centrifuged for 4 hours at 32°C at 1,200 × g. T cells were collected 6 days after activation for adoptive-cell transfer (ACT) to mice. Before injection, for OT-I T cells, stimulation beads were removed with a magnet. For TCRP1A CD8⁺ T cells, dead cells were removed by performing Ficoll gradient centrifugation using the Lymphoprep medium (Stemcell, #07851). For in vitro assays and flow cytometry analysis, OT-I T cells were used 7 days after activation.

For nanobody binding assays, B16F10-OVA PD-L1⁺ cells were incubated with supernatant from nanobody-expressing HEK 293T cells and then stained with anti-HA Tag (PE anti-HA.11 Epitope Tag, clone 16B12, BioLegend, #901518). Amount of supernatant was normalized according to nanobody quantity detected in the supernatant by Western blot analysis (quantification was performed by ImageJ). For recombinant PD-1 blocking assays, B16F10-OVA PD-L1⁺ or MC38-OVA PD-L1⁺ were used as indicated. Cells were first incubated either with supernatant from transfected HEK 293T cells, PD-L1–specific mAb (purified anti-mouse PD-L1, clone 10F.9G2, BioLegend, #124318), or sera from treated mice as indicated. Cells were then washed and stained with recombinant PD-1–PE (rPD-1 PE) obtained by coupling recombinant mouse PD-1 Fc chimera (R&D Systems, #1021-PD) to R-phycoerythrin using Lightning-Link R-PE Labeling Kit (Abcam, #ab102918). Percentage of rPD-1 blocking activity in sera from treated mice was calculated as [1-(rPD-1 median fluorescence intensity with serum from treated mouse/mean rPD-1 median fluorescence intensity with sera from PBS-treated mice)] × 100.

To validate that the binding of PD-L1–specific antibody clone MIH7 to PD-L1 is not impaired by the presence of the nanobody or the PD-L1–specific injected mAb, MC38-OVA PD-L1⁺ were first incubated with increasing amounts of supernatant from 5DXW-T61V–secreting OT-I T cells or increasing concentrations of injected mAb (purified anti-mouse PD-L1, clone 10F.9G2, BioLegend, #124318). The cells were then washed and stained with PD-L1–specific antibody clone MIH7 (PE anti-mouse PD-L1, Biolegend, #1554404). PBS buffer supplemented with 1 mmol/L EDTA and 1% FBS was used for incubations and washings through all experiments. Incubations for binding, blocking, and staining were performed for 20 minutes at 4°C.

Culture medium from transfected HEK 293T cells was collected, centrifuged to pellet cell debris, and supernatant was used for western blot analysis. Transduced OT-I T cells at day 7 after activation were centrifuged, supernatant was collected and the cell pellet was lysed in Pierce RIPA buffer (Thermo Fisher, #89901) supplemented with Halt Protease-phosphatase inhibitor cocktail (Thermo Fisher, #78429). Protein concentration of OT-I T cells lysates was evaluated by Pierce BCA Protein assay (Thermo Fisher, #23225). Lysates, supernatants, or mice sera were added with Pierce Lane Marker Reducing Sample Buffer (Thermo Fisher, #39000), heated (95°C, 10 minutes) and loaded on polyacrylamide gel (Bolt 4%–12%, Invitrogen, #NW04122). After migration, proteins were transferred to iBlot NC stacks (Invitrogen, #IB23002). Membrane was blocked with 5% milk and stained with anti-HA Tag (Invitrogen, clone 2-2.2.14, #26183, 1:5,000), or anti-vinculin (Cell Signaling Technology, clone E1E9V, #13901S, 1:5,000). Secondary antibodies used were anti-mouse IgG HRP (Cell Signaling Technology, #7076, 1:2,500) and anti-rabbit IgG HRP (Cell signaling, #7074, 1:2500), respectively. Protein detection was performed with the chemiluminescent SuperSignal WestPico substrate (Thermo Scientific, #34578). Pictures were captured with Fusion FX camera (Vilbert Lourmat).

OT-I T cells transduced with MSCV-5DXW-T61V plasmid or MSCV empty plasmid were collected 7 days after activation and plated with different numbers of MC38-OVA PD-L1⁺ or MC38 tumor cells in full supplemented IMDM. After 24 hours, cells were detached with PBS supplemented with 2 mmol/L EDTA, stained with Viability dye eFluor 780 (eBioScience, #65-0865-14) and blocked with Trustain FcX (BioLegend, #101320), and then stained with anti-CD8 BV421 (BioLegend) and rPD1-PE. Tumor cells were gated as eFluor780^–CD8^– cells and percentage of killing was calculated as [(Tumor cells alone-Tumor cells with OT-I)/Tumor cells alone] × 100.

C57BL/6 female mice aged between 8 and 12 weeks were inoculated subcutaneously with 1.5 × 10⁶ MC38 or MC38-OVA tumor cells as indicated. After 10 days, mice were randomized according to tumor size. Mice received either one single injection of 10 × 10⁶ OT-I cells (transduced with MSCV-5DXW-T61V vector or MSCV empty vector) by retro-orbital intravenous (i.v.) injection; and/or PD-L1–specific mAb (BioXCell, clone 10F.9G2, #BE0101) or 100 μL of PBS by intraperitoneal (i.p.) injection, every 3 to 4 days for a total of 4 injections. Mice from different treatment groups were cohoused in same cages. Tumor dimensions were measured every 2 to 3 days with a caliper until mice reached humane endpoint. Tumor volume was calculated with the formula: (Length × width × width/2). Last recorded tumor volume of dead or sacrificed mice was considered for the calculation of mean tumor volumes at subsequent time points. For the T429.18 tumor model, TiRP-10B^−/− mice were inoculated subcutaneously with 1.5 × 10⁶ T429.18 tumor cells. After 14 days, mice received one single injection of 10 × 10⁶ TCRP1A CD8⁺ T cells (transduced with MSCV-5DXW-T61V vector or MSCV empty vector) by retro-orbital intravenous injection. To evaluate nanobody or antibody distribution during treatment, mice were sacrificed 7 days after treatment was started for sample collection as indicated. Parts of tumors and spleens were embedded in TissueTek O.C.T. compound (Sakura, #94-4583) in a mold and frozen on carbonic ice. Blood was collected by heart puncture, one part was collected in IMDM with 40 U/mL heparin (Heparine Leo) and used for flow cytometry, the rest remained at room temperature for 1 hour to clot. Blood clot was then centrifuged (15 minutes, 1,500 × g) and serum supernatant was collected. For flow cytometry analysis, tumors, spleens and distant lymph nodes (contralateral from tumor bed) were smashed with a syringe plunger to obtain a cell suspension. Cell suspension of splenocytes and blood cells in heparin underwent red blood cell lysis (eBioscience, #00-4300-54) for 5 minutes at room temperature. For the systemic injection of anti–PD-L1 nanobody, mice bearing MC38-OVA tumors received 10 μg or 50 μg of purified anti–PD-L1 nanobody intraperitoneally, 1 hour, 14 hours, or 24 hours before sacrifice. Tumors, non–tumor draining lymph nodes, spleens, and serum were collected and processed as described above.

PBS supplemented with 1 mmol/L EDTA and 1% FBS was used in all experiments. Samples were first incubated with Fixable Viability Dye eFluor 780 (eBioscience, #65-0865-14) and Trustain FcX (BioLegend, #101320). Samples were then surface stained with different antibodies obtained from BioLegend: anti-HA.11 Epitope Tag (16B12, PE), anti-CD45 (30F11, Alexa700), anti-CD11b (M1/70, BV711), anti-CD11c (N418, BV421), anti-F4/80 (BM8, BV510), anti-Gr1 (RB6-8C5, APC), anti-CD3e (145-2C11, FITC, BV510, AF700), anti-CD8a (56-6.7, PE-Cy7, BV421), anti–PD-L1 (10F.9G2, PE), anti–PD-L1 (MIH7, PE), anti–PD-L2 (TY25, APC), rat IgG2bk Isotype (RTK 4530, PE), rat IgG2ak Isotype (RTK 2758, APC), and mouse IgG1 isotype (MOPC-21, PE). For intracellular staining, cells were fixed using Fixation Buffer (BioLegend, #420801,) and permeabilized with Perm/wash Permeabilization buffer (BioLegend, #421002). Staining was performed with anti-HA Tag antibody or isotype control diluted in permeabilization buffer. Samples were acquired with LSR Fortessa Cytometer (BD). Data were analyzed with FlowJo software, version 10.7.2.

MC38-OVA PD-L1⁺ or PD-L1^KO tumor cells were incubated with PD-L1–specific antibody (purified anti-mouse PD-L1, clone 10F.9G2, BioLegend, #124318), the supernatant of 5DXW-T61V–secreting OT-I T cells, or PBS for 25 minutes at 4°C. Cells were washed twice and cell pellets were resuspended in TissueTek O.C.T. compound (Sakura, #94-4583) in a mold, then frozen on carbonic ice.

Frozen cell pellets and frozen samples from treated mice were cut in 6-μm slices with a cryostat (CryoStar NX70, Thermo Fisher Scientific). Sections were thawed and directly fixed for 5 minutes in 4% formaldehyde. All the following steps were performed at room temperature. Endogenous peroxidases were blocked with IHC/ISG Peroxidase Block (Enzo, #ACC107-0100) for 15 minutes. Protein blocking was performed for 1 hour with a solution of TBS-Tween containing 2% milk, 5% biotin-free BSA and 1% human immunoglobulins (Privigen, CSL Behring). For sections to be stained with anti-rat IgG-HRP secondary, 1.25% goat serum was added to the mixture (VectorLab, #MP-7444). Primary antibodies were diluted in TBS-Tween containing 1% BSA and incubated for 1 hour at room temperature. Rabbit anti-HA Tag antibody (Cell Signaling Technology, clone C29F4, #3724, 1:500) was used to detect the nanobody. Rabbit PD-L1–specific antibody (R&D, MAB90781, 1:250) or rat PD-L1–specific antibody (eBioscience, clone MIH5, #14-5982-82, 1:250) were used for detecting PD-L1. Rat anti-CD8 (BioLegend, clone 53-6.7, #100701, 1:50) was used to detect CD8. These primary antibodies were followed by Dako Envision+ Polymer HRP anti-rabbit (Dako, #K4003) or ImmPRESS HRP goat anti-rat IgG, mouse adsorbed (Vector Laboratories, #MP-7444) secondary antibody. For the detection of PD-L1–specific antibody clone 10F.9G2 (injected mAb), staining with ImmPRESS HRP goat anti-rat IgG was performed directly after blocking. For cell pellets, staining was revealed with DAB (Abcam, #64238), nuclei were counterstained with hematoxylin and slides were mounted with HIGHDEF IHC mount (Enzo). Images were acquired with PANNORAMIC confocal (3DHISTECH). For mouse samples, staining was revealed with the Tyramide Signal Amplification system (TSA). Tyramide hydrochloride (Sigma, #T2879-5G) was conjugated with FITC-NHS (Sigma, #21878) or Sulfo-Cyanine3 NHS ester (Lumiprobe, #21320). Fluorochrome-coupled tyramide was diluted in a buffer containing 0.1 mol/L boric acid, 3 mol/L NaCl, 0.1% Tween 20 (pH 7.8), 0.003% H₂O₂ and applied directly for 10 minutes. For multiplex staining on mouse samples, the whole procedure was performed for each staining. Finally, anti-CD146 staining was added as a one-step incubation with Alexa Fluor 647 anti-mouse CD146 10 μg/mL (BioLegend, #134718). Nuclei were counterstained with Hoechst 33342 (Invitrogen, #H3570) 10 μg/mL in TBS-Tween containing 10% BSA. Slides were mounted with HIGHDEF IHC Fluoromount (Enzo). Images were acquired with Oyster PANNORAMIC 250 Flash III (3DHISTECH).

For the calculation of colocalized areas, Indica labs module Area Quantification FL v1.2 was used. Staining threshold was defined based on negative controls: samples from mice treated with PBS (for injected mAb analysis) or samples from mice treated with ACT of control OT-I (for nanobody analysis). Samples stained with secondary alone (anti-rabbit or anti-rat IgG HRP) were used to set threshold for anti–PD-L1 staining using rabbit or rat PD-L1–specific mAb, respectively. Single stained surface areas, colocalized surface, and total surface areas were recorded for each sample and were used for the calculation of percentage of PD-L1 area covered by treatment. To calculate the correlation of PD-L1–specific injected mAb/nanobody versus PD-L1, Indica labs module CytoNuclear FL v2.0 was used to recognize each cell in the sample based on DAPI signal and the raw signal intensities for each staining were recorded.

MC38-OVA PD-L1⁺ tumor cells incubated with or without PD-L1–specific antibody clone 10F.9G2 (purified anti-mouse PD-L1, BioLegend, #124318) were stained with anti-PD-L1-PE (PE anti-mouse PD-L1, clone 10F.9G2, BioLegend, #124308). Average PE mean fluorescence intensity (MFI) on tumor cells not incubated with PD-L1–specific antibody was considered as the control condition with 100% free PD-L1. Free PD-L1 on cells incubated with the antibody was then calculated as: MFI of anti-PD-L1(10F.9G2)-PE staining of cells pre-incubated with PD-L1–specific antibody clone 10F.9G2 (‘injected mAb’) as a percentage of MFI from control tumor cells. For cell populations from mice treated with PD-L1–specific mAb or PBS, the calculation was based on MFI of anti–PD-L1 (clone 10F.9G2)–PE staining of cells from mAb-injected mice as a percentage of the mean MFI from PBS-injected mice.

Statistical analyses were performed using Prism 6 (GraphPad Software) or Python 3.7. Comparison between two groups was performed using the paired or unpaired Student t test as indicated. Two-way ANOVA was used to compare tumor growth curves. Density plots were visualized using Seaborn 0.11.1. Pearson correlations were calculated with scipy 1.6.1. Pairwise comparisons indicating the significance of differences in boxplots were calculated using Welch two-sided t test. ANCOVA was used to compared linear regression slopes. P value significance in figures is marked as: ns, P  ≥ 0.05, * 0.01 < P < 0.05, ** 0.001 < P < 0.01, *** 0.0001 < P < 0.001, **** P < 0.0001, unless explicitly stated otherwise.

The data generated in this study are available within the article and its Supplementary Data files. Code used for IF staining intensities analysis is available on Code Ocean (https://codeocean.com/capsule/2303411).

For the targeted delivery of PD-L1 blockade at the tumor site, we sought to engineer tumor-targeting CD8⁺ T cells for the secretion of an anti–PD-L1 nanobody. We identified two nanobodies targeting murine PD-L1 from the literature: the C7 nanobody, which was described as an imaging tool by Broos and colleagues (21), and the nanobody 5DXW from the RCSB open access protein database (22). Amino acid sequences were reverse translated and DNA sequences were flanked by IL2 signaling peptide (to ensure nanobody secretion) and HA Tag (to allow detection) coding sequences. The constructs were then inserted into a retroviral vector (Supplementary Fig. S1). The secretion of nanobodies upon vector expression was first confirmed in HEK 293T cells (Supplementary Fig. S1B). However, the molecular weight of the expressed nanobodies was higher than expected, related to N-linked glycosylation. Selective point mutations were introduced in both 5DXW and C7 nanobody sequences to modify the NXT amino acid motif, and upon expression these mutant nanobodies reached the expected 15 kDa size (Supplementary Fig. S1A and S1B). We confirmed by flow cytometry that all secreted nanobodies bound PD-L1 (Supplementary Fig. S1C). Only 5DXW nanobody and its mutant version were able to block PD-1/PD-L1 interaction (Supplementary Fig. S1D). The non-glycosylated 5DXW-T61V nanobody was chosen for further use, as its smaller size might maximize its penetration ability. Finally, we validated that the 5DXW-T61V nanobody was specific for PD-L1 and did not bind murine PD-L2 (Supplementary Fig. S1E).

We then engineered tumor-specific CD8⁺ T cells to secrete the selected nanobody. CD8⁺ T cells transgenic for an ovalbumin peptide–specific TCR (OT-I T cells) were stimulated and transduced with MSCV-5DXW-T61V vector or the empty MSCV vector as a control (Fig. 1A). As shown in Fig. 1B, MSCV-5DXW-T61V was successfully introduced into OT-I T cells with a transduction rate of around eighty percent, as detected by intracellular anti-HA Tag staining (Supplementary Fig. S2). The production and secretion of 5DXW-T61V nanobody were confirmed by western blot analysis (Fig. 1C). When cultivating transduced OT-I T cells with MC38-OVA PD-L1⁺ tumor cells, secreted 5DXW-T61V impaired the binding of recombinant PD-1 labeled with PE fluorochrome (rPD-1 PE) to tumor cells, confirming its PD-L1 blocking capacity (Fig. 1D). Secretion of 5DXW-T61V did not improve the killing capacity of OT-I T cells (Supplementary Fig. S1F).

We then evaluated the ability of transduced OT-I T cells to deliver anti–PD-L1 nanobody in vivo. 5DXW-T61V–secreting OT-I or control OT-I T cells were adoptively transferred into mice bearing MC38-OVA tumors and tumor samples were collected 7 days later. As tumor macrophages are known to represent the major PD-L1⁺ immune cell population in the TME (26), we assessed the abundance of secreted 5DXW-T61V nanobody on their surface by flow cytometry. The whole population of tumor macrophages, identified as CD45⁺CD11b⁺CD11c⁺F4/80⁺, were stained by anti-HA Tag in mice treated with 5DXW-T61V-secreting OT-I T cells (Fig. 1E; Supplementary Fig. S2). 5DXW-T61V nanobody was also found on tumor cells (Fig. 1E).

These results confirmed that adoptively transferred engineered tumor-specific T cells efficiently secreted anti–PD-L1 nanobody in the TME.

We next evaluated the therapeutic efficacy of 5DXW-T61V–secreting OT-I T cells in the MC38-OVA model and compared it to the standard i.p. injection of PD-L1–specific antibody. Treatment was initiated ten days after tumor implantation, when tumors were well established (Fig. 2A). As shown in Fig. 2B and C, ACT of 5DXW-T61V–secreting OT-I T cells strongly improved tumor control as compared to ACT of control OT-I T cells (P < 0.0001). In contrast, the addition of i.p. administration of an PD-L1–specific antibody to ACT of control OT-I T cells only moderately improved tumor control as compared to ACT alone (P = 0.0080). ACT of 5DXW-T61V–secreting OT-I T cells showed better tumor inhibition as compared to the combination of ACT of control OT-I T cells plus PD-L1–specific antibody (P = 0.023). These results demonstrate that PD-L1 blocking therapy achieved by targeted delivery of an anti–PD-L1 nanobody is superior to the standard i.p. injection of PD-L1–specific antibody.

We next evaluated the intratumoral delivery of anti–PD-L1 treatment by performing IF analysis on tumor samples from mice bearing MC38-OVA tumors that had been treated with i.p. injection of PD-L1–specific antibody or treated with ACT using 5DXW-T61V–secreting OT-I T cells. Antibody validations for IF are depicted in Supplementary Fig. S3A. PD-L1–specific mAb (‘Injected mAb’) or anti–PD-L1 nanobody were co-stained with PD-L1 and IF images were processed further with HALO software for quantification. As shown in Fig. 3A, secreted nanobody was found to colocalize with PD-L1 in wide areas (in yellow, right upper panel), representing on-target distribution of secreted anti–PD-L1 nanobody. In sharp contrast, tumor samples from mice treated with PD-L1–specific antibody showed large areas (in red, left upper panel) where PD-L1 was expressed but no injected mAb was detected. In these samples, only few colocalizing areas representing on-target antibody distribution were found. This was repeatedly observed in multiple tumor samples (Fig. 3B) and was not related to differential expression of PD-L1 across samples (Fig. 3C). Increasing the dose of the injected antibody did not significantly improve intratumoral antibody penetration (Supplementary Fig. S3B–S3D).

We also noticed that injected mAb was mostly detected in filiform structures suggestive of blood vessels. Costaining for CD146 confirmed that these structures were delineated by endothelial cells, suggesting that injected PD-L1–specific antibody was trapped in the vicinity of blood vessels (Fig. 3A, left bottom panels). In contrast, in mice treated with ACT using 5DXW-T61V–secreting OT-I T cells, nanobody detection was not restricted to CD146⁺ structures (Fig. 3A, right bottom panels).

To further quantify the enrichment of injected mAb or secreted nanobody on PD-L1–expressing cells in the tumor, signal intensity for PD-L1 staining and either injected mAb or secreted nanobody staining on all nonendothelial cells of multiple samples were computed using HALO software. As depicted in Fig. 3D (left panel), injected mAb was detected at higher levels on cells expressing low levels of PD-L1. In contrast, secreted nanobody was detected in a proportionate manner on PD-L1–expressing cells (right panel). Pearson correlation calculation confirmed that the distribution of secreted nanobody on PD-L1–expressing cells in the TME was correlated to PD-L1 expression whereas the distribution of injected mAb was not (Fig. 3E). Poor antibody distribution was restricted to the tumor, as the same analysis on spleen samples from mice injected with PD-L1–specific antibody revealed a correlation between injected mAb and PD-L1 detection (Supplementary Fig. S4A and S4B). These analyses demonstrate that besides being able to widely diffuse inside the tumor, secreted nanobody can also accumulate on cells with higher levels of PD-L1 expression.

Although injected mAb was mostly detected on CD146⁺ blood vessels structures, correlation between injected mAb and PD-L1 detection remained poor on these cells (Supplementary Fig. S4C left panel and S4D). This suggests that injected mAb is not enriched in the vicinity of blood vessels due to higher level of PD-L1 expression but rather because there is a barrier preventing deeper diffusion. In contrast, secreted nanobody detection correlated with PD-L1 expression on CD146⁺ cells, as it did for all other cells in the tumor (Supplementary Fig. S4C right panel and S4D).

Finally, we assessed whether tumor penetration would impact PD-L1 blocking capacity of PD-L1–specific injected antibody or secreted nanobody at the tumor site. As shown previously, binding of rPD-1-PE to PD-L1–expressing cells is blocked in vitro by PD-L1–specific injected mAb and secreted nanobody (Fig. 1D). Recombinant PD-1-PE staining was therefore applied to single cell suspensions from tumor samples of treated mice. Since tumor-infiltrating myeloid cells seem to be crucial PD-L1–expressing cells in the TME, we quantified rPD-1-PE binding on CD11b⁺F4/80⁺ and CD11b⁺Gr1⁺ cells. To normalize rPD-1-PE signal to the level of PD-L1 expression, each sample was also stained with an anti–PD-L1 (clone MIH7) whose binding ability was not hampered by the injected mAb or the nanobody (Supplementary Fig. S5A and S5B). As shown in Fig. 3F, a positive correlation between PD-L1 expression and rPD-1-PE signal was observed in all groups. However, in mice treated with ACT using 5DXW-T61V–secreting OT-I T cells, the slope angle of the correlation was decreased as compared to mice treated with ACT using control OT-I T cells, indicating a blockade of PD-L1 by the secreted nanobody. In contrast, the correlation remained unchanged in tumor samples from mice that received ACT using control OT-I T cells combined with injection of PD-L1–specific antibody. These results suggest that injected mAb was not able to block PD-L1 in the TME, likely because of its inability to penetrate the tumor, as discussed above. The same observation was made in mice that did not receive ACT and were treated with PBS or PD-L1–specific antibody alone (Supplementary Fig. S5C).

Nanobodies are known for their excellent tissue penetration ability, thanks to their small size allowing diffusion in the TME (27). When we injected tumor-bearing mice systemically with anti–PD-L1 nanobody at the same molar quantity as the PD-L1–specific mAb, we observed penetration of the injected nanobody into the tumor tissue (Supplementary Fig. S6A). Secreted nanobody was also found binding to PD-L1–expressing cells at a distance from the 5DXW-T61V–secreting OT-1 T cells, confirming that secreted nanobody can indeed diffuse well within the TME (Supplementary Fig. S6B and S6C).

Altogether, delivery of anti–PD-L1 nanobody using tumor-specific T cells overcame the barrier of intratumoral penetration and improved PD-L1 blocking at the tumor site as compared to injected PD-L1–specific antibody. This presumably underlies the better therapeutic activity of our targeted delivery approach.

Since we observed a low intratumoral penetration of injected PD-L1–specific antibody, we compared PD-L1 occupancy achieved by the antibody in the periphery and in the tumor. Recombinant PD-1-PE staining could not be used for this because no binding of r-PD-1 was found on cells from secondary lymphoid organs, possibly because of the lower level of PD-L1 expression in those organs as compared to the tumor site (Supplementary Fig. S7).

We used flow cytometry to evaluate the proportion of antibody-bound and free PD-L1 molecules upon injection of PD-L1–specific antibody by staining with the same antibody clone labeled with PE fluorochrome. In vitro assays confirmed that staining with anti-PD-L1 clone 10F.9G2 coupled to PE (‘anti–PD-L1-PE’) was prevented upon preincubation with unlabeled antibody, in a dose-dependent manner. This allowed us to estimate the proportion of PD-L1 molecules free from antibody binding (Fig. 4A and B). We translated this concept in vivo and estimated the proportion of “free PD-L1” on different cell populations in mice bearing MC38-OVA tumors by comparing anti–PD-L1-PE staining between mice treated with PD-L1–specific injected mAb and mice that did not receive the antibody (Fig. 4C and D). We focused our analysis on myeloid cells from the tumor and from the secondary lymphoid organs. As shown in Fig. 4C and D, the percentage of free PD-L1 on myeloid cells in secondary lymphoid organs was low, suggesting a high coverage by the injected PD-L1–specific antibody. In contrast, the majority of PD-L1 molecules appeared free from antibody binding at the tumor site (Fig. 4D). When evaluating the presence of PD-L1 blocking therapy in the sera of treated mice as the ability to block rPD-1 binding on PD-L1–expressing tumor cells, we found that the sera from mice treated with PD-L1–specific antibody prevented the binding of rPD-1-PE, revealing the presence of the blocking antibody (Fig. 4E). Systemically injected PD-L1–specific antibody exhibits therefore a detrimental distribution pattern, leading to massive PD-L1 occupancy in the periphery but not in the TME. This phenomenon might underlie the suboptimal therapeutic efficacy of PD-L1–specific antibody as well as the triggering of irAEs.

We next evaluated whether targeted delivery of anti–PD-L1 nanobody would reduce systemic exposure to PD-L1 blocking therapy. In that aim, we first analyzed the distribution of transferred T cells in MC38-OVA or MC38 tumor–bearing mice treated with ACT using 5DXW-T61V–secreting OT-I T cells. By staining for intracellular HA Tag in CD8⁺ T cells, we observed that 5DXW-T61V–secreting OT-I T cells were highly enriched in MC38-OVA tumors, but not in MC38 tumors (Fig. 5A; Supplementary Fig. S8). We next evaluated the percentage of myeloid cells bound by the nanobody in the tumor and in the secondary lymphoid organs. As shown in Fig. 5B, the secreted nanobody was detected on different PD-L1⁺ immune cells at the tumor site, but not in the secondary lymphoid organs in mice bearing MC38-OVA tumors. Furthermore, in contrast to MC38-OVA tumors, the secreted nanobody was barely detectable in MC38 tumors by flow cytometry (Fig. 5B) or by IF (Fig. 5C). These results confirmed the antigen-specificity of T cell–mediated nanobody delivery to the tumor site. This was not dependent on the high-affinity OVA antigen recognized by OT-I T cells, as we also observed tumor-specific delivery of the anti–PD-L1 nanobody in mice bearing melanoma T429.18, which expresses the natural MAGE-type antigen P1A, after adoptive transfer of TCRP1A CD8⁺ T cells transduced with the 5DXW-T61V nanobody expression construct (Fig. 5D and E).

Altogether, these results demonstrate that nanobody-secreting T cells migrate preferentially to the tumor and thereby allow specific intratumoral delivery of anti–PD-L1 nanobody. Similarly, when using IF to estimate nanobody distribution in the spleens of mice treated with ACT using 5DXW-T61V–secreting OT-I T cells, nanobody detection could not be differentiated from background staining, suggesting that nanobody amounts were very low in the spleen as compared to the tumor in these mice (Supplementary Fig. S9A). In sharp contrast, we observed a wide distribution of injected mAb on PD-L1⁺ cells in the spleen of antibody-treated mice (Supplementary Fig. S9B), strongly correlating with PD-L1 expression (Supplementary Fig. S4A and S4B) and in accordance with our flow cytometry results. Furthermore, no PD-1 blocking activity was observed in the sera of mice treated with ACT using 5DXW-T61V–secreting OT-I T cells (Fig. 4E), indicating that the secreted anti–PD-L1 nanobody was not circulating in the blood.

Besides targeted delivery, the short half-life of nanobodies in the blood may further contribute to the reduced systemic exposure to secreted nanobody. As shown in Supplementary Fig. S10, after systemic injection of 5DXW-T61V nanobody into mice, a wide distribution of injected nanobody was found after 1 hour in the tumor, lymph nodes, spleen, and serum. However, after 24 hours, the injected nanobody could no longer be detected (Supplementary Fig. S10A–S10D).

Collectively, these data confirmed that targeted delivery of anti–PD-L1 nanobody using tumor-specific T cells leads mostly to nanobody secretion at the tumor site. This strategy therefore reduces systemic exposure to PD-L1 blockade as compared to the systemic injection of PD-L1–specific antibody and might minimize the risk of developing irAEs.

The current report demonstrates that local delivery of anti–PD-L1 nanobodies using tumor-specific T cells enhances the potential of immune checkpoint blockade therapy by deeply reaching PD-L1 at the tumor site. The use of engineered T cells to deliver immunotherapeutic agents has mainly been described in the context of CAR T cells (15–17). Secreting CAR T cells have proven to be effective, but these studies mostly focused on improving CAR T–cell function and persistence in the tumor. Whether engineered T cells could be more potent than systemic treatment to deliver immune checkpoint blockade at the tumor site has not been evaluated so far. Furthermore, in these reports, CAR T cells were engineered to secrete PD-L1–specific antibody (15) or anti–PD-1 single-chain fragment variable (scFv) (16, 17), which might not be ideal tools for local secretion. ScFv represents the association of the variable light chain and variable heavy chain of an antibody, resulting in a molecular weight of around 30 kDa. However, it has been shown that monomeric nanobodies (15 kDa) have better tumor penetration capacities than dimeric nanobodies (30 kDa) (27). We therefore proposed to secrete nanobodies to obtain a better intratumoral penetration. We chose the nonglycosylated mutant of the anti–PD-L1 blocking nanobody 5DXW, as this nanobody could be produced as a 15 kDa protein. Another advantage of nanobody secretion is the modularity of nanobodies: since coding sequences are short, it is possible to combine multiple VHH sequences in the same viral vector (28). Application of CAR T-cell technology to solid tumors is currently limited by numerous factors (29). We would favor the engineering of tumor-infiltrating lymphocytes (TILs) for secreting anti–PD-L1 nanobodies. ACT of TILs has proven efficacious in melanoma and lung cancer patients (30, 31), and viral transduction of human TILs is feasible (32, 33).

We observed a wide distribution of secreted nanobody in the tumor bed. This was likely related to the ability of the nanobody to diffuse in the TME, as good intratumoral distribution was also observed after systemic injection of the nanobody. It has been shown that nanobodies have superior tumor penetration ability compared to full antibodies (27). Whether nanobody penetration was further increased by the cytotoxic activity of secreting T cells is a question deserving further investigations.

Although the use of nanobodies raises the concern of their immunogenicity in humans, camelid and human immunoglobulin heavy chain variable genes display a high sequence identity, and nanobodies can be further humanized to reduce their immunogenicity (34, 35). From the current experience accumulated in early phase clinical trials, occurrence of anti-drug antibodies (ADA) is comparable between humanized nanobodies and humanized or fully human immunoglobulins (35). PD-L1–specific antibodies used in the clinic are not exempt from immunogenicity, and cancer patients treated with the humanized PD-L1–specific antibody atezolizumab develop ADA at high frequency, ranging from 13% to 54% of patients across trials (36). In conclusion, although experience of nanobody use in human remains scarce as compared to antibodies, it does not appear that immunogenicity will be a major limitation for their clinical application.

Our observations shed light on the poor tumor penetration of PD-L1–specific antibody. It has been demonstrated that PD-L1–specific antibody distributes mostly to non-tumor tissues (37). Deng and colleagues further suggested that the antibody could not accumulate in the tumor parenchyma and was primarily associated with vasculature at the tumor site (38). This poor tumor uptake seems particularly detrimental since PD-L1 itself is expressed at higher levels on immune cells, such as macrophages or dendritic cells, in the tumor compared to non-tumor tissue (39). Importantly, these observations have also been confirmed in humans by Bensch and colleagues, who tracked radiolabeled atezolizumab in cancer patients. They observed antibody uptake in several organs and target-specific binding on PD-L1 in the spleen and in lymph nodes. On a few biopsies, they also noticed a poor colocalization between radiolabeled antibody and PD-L1 expression in the tumor at the microscopic level (40). Collectively, and further suggested by this report, it seems very likely that the penetration of PD-L1–specific antibodies in the tumors of patients is heterogeneous. Poor intratumoral penetration has also been observed in mice injected with an antibody targeting epidermal growth factor receptor (EGFR) on cancer cells. These findings have recently been confirmed in patients with head and neck cancer (41, 42). Furthermore, it was shown in mice that poor antibody penetration limited efficacy when targeting EGFR or HER2 (43, 44). Antibody penetration should therefore be regarded as a key barrier to efficacy of PD-L1 blocking therapy and this deserves further evaluation in cancer patients.

Anti–PD-L1 systemic therapy is also limited by the emergence of irAEs that are most of the time manageable but still lead to death in a small fraction of patients. It was demonstrated in patients with cancer treated with the CTLA-4–specific antibody ipilimumab that systemic concentration of ipilimumab correlated with the rate of irAEs (45). Concerning anti–PD-1/L1 agents, a link between drug exposure and toxicity is less clear, but the rate of irAEs is globally lower than with CTLA-4–specific antibody (46, 47). Nevertheless, it is generally accepted that reducing systemic exposure to immune checkpoint blockade might decrease the risk of developing irAEs. In mouse models, intratumoral or peri-tumoral administration of ICI antibodies was suggested as a strategy to reduce toxicity (13, 14). We demonstrated that injected PD-L1–specific antibody was massively occupying PD-L1 in secondary lymphoid organs, which might trigger the activation of T cells directed to self-antigens. In contrast, the use of nanobody-producing tumor-specific T cells allowed tumor-specific nanobody delivery, thanks to the accumulation of transferred T cells at the tumor site. This favorable distribution pattern was not observed with systemic nanobody injection. Furthermore, while the short half-life of nanobodies in the blood stream limits sustainable tumor exposure after systemic injection, it likely contributes to the weak systemic exposure with our strategy. Indeed, we observed no PD-L1 blocking activity in the serum of mice treated with ACT using 5DXW-T61V–secreting T cells. Our strategy therefore minimizes the risk of irAEs as compared to the injection of PD-L1–specific antibody.

It has been reported that PD-L1 is expressed by endothelial cells in the tumor and that it suppresses the activity of antigen-specific T cells in mice (48) and correlates with poor CD8⁺ T-cell infiltration in humans (49). By IF, we also observed PD-L1 expression on endothelial cells in the tumor. Both injected PD-L1–specific antibody and secreted nanobody were found on endothelial cells. These two strategies must therefore be able to block the potential immunosuppressive role of PD-L1 on endothelial cells in the tumor. However, although secreted nanobody bound to PD-L1 on endothelial cells in a proportionate way, injected antibody accumulated on these cells independently of the level of PD-L1 expression, suggesting that its accumulation on these cells is related to its inability to penetrate deeper into the tumor.

In conclusion, we have shown that targeted delivery of anti–PD-L1 nanobody could improve tumor control and reduce systemic exposure to PD-L1 blockade compared to injected PD-L1–specific antibody in vivo. We demonstrated that locally secreted nanobody efficiently reached PD-L1 at the tumor site and left peripheral PD-L1 untouched, while PD-L1–specific antibody had a detrimental distribution pattern, massively occupying PD-L1 in the periphery but not in the tumor. Our report suggests that tumor penetration is a limiting factor for the efficacy of ICI antibodies and we propose the use of nanobody-secreting T cells as a promising strategy to overcome this limitation.

B.J. Van den Eynde reports grants and personal fees from iTeos Therapeutics, personal fees from Amgen, personal fees from Vaccitech, personal fees from Oncorus, personal fees from SRIW Life Science, and grants and nonfinancial support from Apogenix outside the submitted work; in addition, B.J. Van den Eynde has a patent for cancer vaccines pending and a patent for immunomodulatory agents pending. No disclosures were reported by the other authors.

P.-F. Petit: Conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, writing–original draft. R. Bombart: Data curation, methodology, writing–review and editing. P.-H. Desimpel: Data curation, validation. S. Naulaerts: Software, formal analysis, writing–review and editing. L. Thouvenel: Resources, methodology. J.-F. Collet: Resources, writing–review and editing. B.J. Van den Eynde: Supervision, funding acquisition, methodology, writing–review and editing. J. Zhu: Conceptualization, supervision, funding acquisition, methodology, writing–review and editing.

We thank Minh Phuong Le and Loubna Boudhan for excellent technical help, Vincent Stroobant for TSA reagents production, the LAF animal facility for mice production, Professor T. Michiels and Professor M. Moser for kindly providing plasmids and cell lines, and Isabelle Grisse for editorial assistance. This work was supported by Ludwig Cancer Research, de Duve Institute (Belgium), and Université catholique de Louvain (Belgium). This work was also supported by grants from the following: Le Fonds de la Recherche Fondamentale Stratégique–WELBIO (Walloon Excellence in Life Sciences and Biotechnology), Belgium (Grant WELBIO-CR-2019C-05); Fonds pour la Recherche Scientifique (FNRS), Belgium (Grants EOS O000518F and PDR T.0091.18); Fondation contre le Cancer, Belgium (Grant 2018-090). European Union’s Horizon 2020 Research and Innovation Programme (Grant Agreement No. 754688, MESI-STRAT [Systems Medicine of Metabolic-Signaling Networks].

P-F. Petit was supported by FNRS-aspirant (Grant number 1.A.818.18F).

R. Bombart was supported by FRIA (Grant number 1.E.002.21F).

P-H. Desimpel was supported by Televie (Grant number 7.4598.21).

S. Naulaerts was supported by FNRS-EOS DECODE (Research program O000518F).

J. Zhu and B.J. Van den Eynde were supported by Ludwig Institute for Cancer Research.

J. Zhu was supported by Fondation Contre le Cancer (Grant number 2019-094).

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.