An Engineered IL15 Cytokine Mutein Fused to an Anti-PD1 Improves Intratumoral T-cell Function and Antitumor Immunity

Cytokines that activate the immune response often display robust efficacy in preclinical murine tumor models (1). However, systemic cytokine administration is often associated with severe toxicities, preventing cytokine delivery to the tumor at efficacious concentrations (2, 3). The therapeutic index for cytokines could be improved by localizing their effects to the tumor microenvironment. A potential strategy to address these concerns is to generate cytokines fused to antibodies that recognize T-cell markers that are overexpressed in the tumor microenvironment.

IL15 and the related cytokine IL2 have similar functions and both bind to the IL2 receptor β and γ subunits, but they differ in their private receptor IL15Rα and IL2Rα, respectively (4, 5). Due to these differences in α-subunit binding, IL2 uniquely functions to maintain regulatory T (Treg) cells and participates in activation-induced T-cell death, whereas IL15 maintains natural killer (NK) cells and CD8⁺ memory T cells (5, 6). In the clinic, IL15 and IL2 therapies have been limited due to the short half-life of the molecules and toxicities associated with systemic immune activation (7–10). Several approaches have aimed to improve the efficacy and pharmacokinetics of these cytokines while reducing their inherent toxicity. Many of these approaches are showing promise as they advance to the clinic, including promising preliminary results as monotherapy or in combination with programmed death receptor 1 (PD1) blockade (11–13). These data have rejuvenated interest in the potential of IL2- and IL15-based therapeutics (14).

PD1 is an effective target on tumor-infiltrating lymphocytes (TIL) for cancer therapy (15, 16). Antibodies that block PD1 or its ligand PD-L1 and activate these antitumor T cells are approved in multiple cancer indications (17). Despite encouraging and sometimes durable responses in a subset of patients with cancer, most patients do not respond; therefore, there is a large unmet need to improve upon these therapies. The observation that IL15 treatment enhances the proliferation of PD1 blockade–unresponsive human CD8⁺ TILs provides mechanistic support for combining IL15 with anti-PD1 to overcome anti-PD1 resistance (14).

Of the IL2Rβ/γ-expressing cell types, NK cells express the highest level of IL2Rβ, both in mouse and human, and are highly sensitive to systemic delivery of IL2 and IL15 (18). Due to the preferential binding and resultant uptake of IL15 therapeutic agents by peripheral and/or liver-resident NK cells, untargeted IL15 could remain limited in its ability to deliver sufficient exposure to activate intratumoral CD8⁺ T cells. This is important as intratumoral CD8⁺ T-cell activation and expansion are associated with the efficacy of cancer immunotherapies in animal models and in the clinic (19–21). In our present study, we developed an anti-PD1–IL15 fusion (anti-PD1–IL15m) that optimally activated intratumoral antigen-experienced CD8⁺ T cells, which are enriched for PD1 in both mouse (22) and human tumors (15, 23). This molecule yielded efficient targeting of IL15 bioactivity to PD1-expressing cells and reduced the natural preference of IL15 to engage PD1-negative NK cells. Overall, our data demonstrated that PD1 anchoring of an IL15 mutein could potentially allow for the systemic administration of anti-PD1–IL15m to potentiate tumor-specific T-cell proliferation and function to maximize antitumor efficacy while ameliorating adverse events.

All animal studies were conducted according to the NIH animal care guidelines and following protocols approved by the Institutional Animal Care and Use Committee (IACUC) from Pfizer Inc. C57BL/6 and Balb/c mice, ages 5 to 7 weeks, were obtained from the Jackson Laboratory.

B16F10 cells, MC38 cells, and 32D clone 3 (CRL-11346) were obtained from ATCC in 2017 and cultured according to the instructions. B16-F10 cells were cultured in DMEM (Gibco, Catalog No. 11574486) supplemented with 10% FBS (Gibco, Catalog No. 10437028), 1% penicillin/streptomycin (Pen-strep; Thermo Fisher Scientific, Catalog No. 15140122), and 1% sodium pyruvate (Gibco, Catalog No. 11360070). MC38 cells were culture in RPMI 1640 (Gibco, Catalog No. 11875-093) supplemented with 10% FBS and 1% Pen-strep. The 32D cells were cultured in DMEM supplemented with 10% FBS and 10 ng/mL IL3 (Beckton Dickinson, Catalog No. 354040). HepG2 cells were obtained from ATCC in 2017 and cultured in DMEM supplemented with 10% FCS (Sigma, Catalog No. F7524). HepG2 cells were transduced with retroviral vectors containing H2B-RFP plasmid (AddGene, Catalog No. 26001); lentiviral RFP-H2B plasmids were a kind gift from Lisanne Mout (Urology, Erasmus MC, Rotterdam, the Netherlands). All cell lines were regularly tested using the mycoAlert Mycoplasma Detection Kit (Lonza, Catalog No. LT07-218) and were used for less than 10 passages. We used Expi293F (Thermo Fisher Scientific, Catalog No. A14528) cells for protein production.

The original 32D cells were retrovirally transduced with the following plasmids: (i) pRF768: doxycycline-inducible plasmid containing human PD1 sequence with mCherry, and (ii) pRF791: constitutive expression plasmid containing human IL2Rβ sequence with GFP. Doxycycline chloride (Sigma, Catalog No. T7660–5G) was added to cells the night before experiments to induce mCherry-hPD1 expression. For the primers used to generate these plasmids and the plasmid maps used in the 32D cell line engineering, see Supplementary Table S1 and Supplementary Figs. S1–S4.

The lentiviral vector pLVX-SFFVp-Puro-P2A-tTA was generated as follows: DNA comprising the Spleen Focus-Forming Virus promoter (SFFVp), followed by the coding sequence for Puromycin N-acetyltransferase (Puro), a 2A peptide (P2A), and the Tet-On 3G tetracycline-controlled transactivator (tTA), and flanked with BstXI and MluI restriction enzyme sites was synthesized by GenScript. This DNA was then subcloned into the pLVX-TetOne vector (Takara, Catalog No. 631846) after digestion of both with BstXI and MluI. A derivative of this plasmid, pRF713, was constructed by PCR amplifying DNA encoding the mCherry fluorescent protein, followed by the viral T2A sequence (amino acid sequence: GSGEGRGSLLTCGDVEENPGP), followed by an EcoRI site and a BamHI site, all as an MfeI–BglII-flanked amplicon, which was then digested with MfeI plus BglII restriction enzymes. Following digestion of pLVX-SFFVp-Puro-P2A-tTa with EcoRI+BamHI, the mCherry-T2A fragment was subcloned in to yield pRF713.

To generate plasmid pRF768, DNA encoding the human PD1 open reading frame was amplified by PCR, digested with EcoRI plus BamHI, and subcloned into the EcoRI–BamHI sites of pRF713. Human PD1 sequence (NM_005018) was amplified from a pCMV6-based plasmid (Origene, Catalog No. RC210364) with primers RF764 and RF765, yielding pRF768.

To generate plasmid pRF791, we first generated lentiviral transfer vector pRF779, which contains an EF1alpha promoter upstream of a polylinker, as well as the SFFV promoter-driven Blasticidin resistance cassette. We then generated an EcoRI(5′)–BamHI(3′)-flanked PCR product with the coding region for TurboGFP, a linker plus the viral P2A sequence (amino acids, GGSGATNFSLLKQAGDVEENPGP), and the coding region for human CD122 aka IL15 receptor beta (the CD122 gene was synthesized by DNA2.0, based on Uniprot entry P14784 amino acid sequence information). Digestion of the PCR production and pRF779 with EcoRI plus BamHI followed by ligation generated pRF791. All PCR-amplified regions of all constructs were confirmed by DNA sequencing. The lentiviral helper plasmids psPAX2, an HIV-1 gag-pol packaging plasmid, and pMD2.G, a VSV-G expression plasmid, were generous gifts from Dr. Didier Trono [Swiss Federal Institute of Technology in Lausanne (EPFL)].

Lentivirus (LV) was produced by transient transfection of sub-confluent HEK-293T/17 cells (ATCC, Catalog No. CRL-11268) in 6-well plates. Briefly, psPAX2, pMD2.G, and either pRF768 or pRF791, were cotransfected at a 3:1:4 mass ratio, respectively, using Lipofectamine 2000 (Thermo Fisher Scientific, Catalog No. 11668030) and following the manufacturer's instructions. The following day, the media was replaced, and 48 hours later the LV supernatant was harvested and filtered through a 0.45 μm syringe filter (Millipore, SLHV004SL). Fresh LV supernatant was used immediately to transduce subconfluent 32D target cells by diluting LV supernatant in an equal volume of 32D growth medium. Three days later, transduced cells were selected by addition of 20 μg/mL Blasticidin (pRF791) and/or 1 μg/mL (final conc) Puromycin (pRF768) to the culture medium.

A total of 58 patients, 27 with colorectal cancer and 31 with hepatocellular carcinoma (HCC), were enrolled between March 2017 and October 2019 (see Supplementary Tables S2 and S3, respectively). The study was conducted according to the Declaration of Helsinki and all human tissues and bloods were obtained through protocols approved by the local ethics committee (Medische Ethische Toetsings Commissie Erasmus MC Rotterdam). Written informed consent was obtained from all donors. Enrolled patients did not receive chemotherapy or immunosuppressive therapy for at least 4 weeks prior to surgery.

The 31 patients with HCC that were enrolled were eligible for resection (n = 29) or liver transplantation (n=2). The 31 patients with colorectal cancer included in this study were eligible for resection of primary tumor (n = 15) or liver metastasis (LM; n = 12). Matched fresh tissue samples from tumor tissue and surrounding tumor-free tissue (>2 cm from tumor margin) were collected and subsequently intrahepatic or intracolonic lymphocytes as well as TILs were isolated. Peripheral blood was obtained perioperatively. How samples were collected and stored until processing is described below in the section “Preparation of human TILs, tumor-free tissue lymphocytes, and peripheral blood mononuclear cells.”

Anti-human PD1 clone 10 and anti-mouse PD1 clone F12–3 were discovered in-house at Pfizer. The amino acid sequences of heavy chains and light chains were codon optimized and synthesized at GeneArt (Thermo Fisher Scientific) before being cloned into a proprietary mammalian expression IgG vector driven by a CMV promoter, with effector-null mutations (L234A, L235A, G237A) in the heavy chain constant region. Similarly, for the cytokine receptors, the coding sequences of extracellular domains of human IL15Rα (accession number: Q13261, residues 31–205), the sushi domain of human IL15Rα (IL15Rα-su, residues 31–95), and human IL2Rβ (accession number: P14784, residues 27–240) were fused with C-terminal 8xHis tags and cloned into a CMV-driven plasmid backbone with a signal peptide as secretion leader. For antibody-IL15 fusion, human IL15 harboring corresponding attenuated receptor-binding mutations were fused to the C-terminus of the heavy chain containing the Hole mutations (S354C, T366S, L368A, Y407V) whereas the Knob mutations (Y349C, T366W) was introduced to the other heavy chain to facilitate Fc heterodimerization during coexpression. The human IL15 superagonist was a homodimeric, noncovalent complex between human IL15Rα-su fused to human IgG1 Fc and human IL15 with N72D mutation to enhance its agonistic activities, produced via coexpressing of the two corresponding plasmids.

All constructs were sequence verified before going into transient transfection in Expi293 cells (Thermo Fisher Scientific, Catalog No. A14527). Briefly, transfection-grade DNA of corresponding chains were mixed with 25 kDa linear PEI (prepared as 1 mg/mL solution, Polysciences, Catalog No. 23966) and incubated for 30 minutes before added to the Expi293 cells (at density of 3 × 10⁶) at 1 mg DNA per liter ratio. Four to 16 hours later, valproic acid (sodium salt; Sigma, Catalog No. P4543) and CHO CD Efficient Feed B (Invitrogen, Catalog No. A10240-01) were added as boost. Secreted recombinant proteins in the conditioned medium were harvested after day 4 or 5 after transfection by centrifugation. For any Fc fusions, target proteins were captured using Protein A affinity chromatography on a MabSelectSuRe column (Cytiva, formerly GE Life Sciences, Catalog No. 28-4064-10), followed by ion-exchange chromatography on a Mono S 5/50 GL or Mono S 10/100 GL column (Cytiva, formerly GE Life Sciences, Catalog No. 17516801 or 17516901) and size exclusion chromatography on a HiLoad 16/600 Superdex 200 prep grade or HiLoad 26/600 Superdex 200 prep grade column (Cytiva, formerly GE Life Sciences, Catalog No. 28989335 or 28989336). His-tagged molecules were purified using nickel affinity chromatography on a HisTrap excel column (Cytiva, formerly GE Life Sciences, Catalog No. 17371206), followed by size exclusion chromatography on a Superdex 200 and/or Superdex 75 GL 10/300 column (Cytiva, formerly GE Life Sciences, Catalog No. 17517501 or 17517401). Monomeric species were pooled and formulated into 1× PBS pH 7.4. Reagents used during purification include 1× PBS pH 7.4 (Gibco, Catalog No. 10010031), 0.1 M sodium citrate pH 3.6 (Teknova, Catalog No. S8440), 0.5 M MES pH 6.0 (Alfa Aesar, Catalog No. J62574AE), 5 M NaCl (Invitrogen, Catalog No. AM9759), 0.5 M sodium phosphate pH 8.0 (Alfa Aesar, Catalog No. J60825AK), and 2 M Imidazole pH 8.0 (Teknova, Catalog No. I7080).

The binding affinities of the anti-PD1–IL15 fusion proteins and hIL15 (R&D, Catalog No. 247-ILB-005) to human IL15Rα and human IL2Rβ were measured on a Biacore T200 Surface Plasmon Resonance–based biosensor (GE Lifesciences). An anti-human Fc reagent (Goat anti-human IgG Fcγ-specific, SouthernBiotech, Catalog No. 2014–01) was injected into a Biacore CM4 sensor chip (GE Lifesciences) that had been pre-activated with NHS/EDC. Anti-PD1–IL15 fusion proteins were captured via the anti-human Fc on the surface of the sensor chip. Different concentrations of purified in house human IL15Rα or human IL2Rβ were injected as analytes. Buffer cycles were collected for each captured fusion protein for double-referencing purposes using the protocol described in (24). For steady-state affinity analysis, the double-referenced equilibrium binding responses were fit with a 1:1 Langmuir steady-state model using Biacore T200 Evaluation Software version 2.0.

Tumor or surrounding tumor-free tissue were stored in Belzer University of Wisconsin solution (Bridge to Life, Catalog No. 100520) at 4°C overnight or immediately processed upon receipt. Single-cell suspensions were obtained by enzymatic tissue digestion followed by filtration. In short, colon tissue was cut into small pieces and stirred in the presence of EDTA (10% FCS, 15 mmol/L HEPES, 1 mmol/L EDTA in PBS) four times 15 minutes at 37°C. Then, colon as well as cut primary colorectal cancer tissue were stirred in the presence of 400 U/mL collagenase VIII (Sigma-Aldrich, Catalog No. C2139) and 0.2 mg/mL DNAse I (Roche, Catalog No. 4716728001) in HBSS (Sigma-Aldrich, Catalog No. H9394) for 60 minutes at 37°C. Liver and HCC tissues were cut into small pieces and then stirred in a digestion mix containing 0.125 mg/mL collagenase IV (Sigma-Aldrich, Catalog No. C4-22-1G) and 0.2 mg/mL DNAse I (Roche, Catalog No. 4716728001) in HBSS for 40 minutes at 37°C. Subsequently, digested tissue pieces were filtered through a 100 μm pore filter followed by 70 μm pore filter and loaded onto a ficoll (Cytiva, Catalog No. GE17-1440-02) layer. Mononuclear leukocytes were isolated by centrifugation at 840 × g for 20 minutes without brakes. Peripheral blood mononuclear cells (PBMC) were isolated using ficoll density centrifugation.

Nivolumab (Selleckchem, Catalog No. A2002) and pembrolizumab (Selleckchem, Catalog No. A2005) were labeled with a PE fluorophore (Thermo Fisher Scientific, Catalog No. Z25455) and anti-hPD1–IL15m was labeled with APC according to the manufacturer's protocol (Thermo Fisher Scientific, Catalog No. A20181). hIL2Rβ/hPD1 32D cells were induced overnight with doxycycline at 1 μg/mL. Competition experiments were done by incubating doxycycline treated hIL2Rβ/hPD1 32D cells at 4°C for 1 hour with nivolumab or pembrolizumab at the following concentrations: 0, 0.01, 0.04, 0.1,0.5, 1, and 5 μg/mL. The treated cells were then stained at the same temperature with 1 μg/mL APC-conjugated anti-hPD1–IL15m for another hour. Cells were washed and resuspended in 150 μL staining buffer (PBS supplemented with 3% FBS) and analyzed on the flow cytometry the same day.

Single-cell suspensions from tissues were prepared as described above. Antibodies against the following molecules were used throughout the paper if not otherwise indicated: anti-mouse CD45 (30-F11; Biolend, Catalog No. 103116), anti-mouse CD4 (RM4–5; Millipore Sigma, Catalog No. MABF577), anti-mouse TCRγδ (GL3; BD BioSciences, Catalog No. 553178), anti-mouse TCRβ (H57–597; BD BioSciences, Catalog No. 553171), anti-mouse CD8α (53–6.7; BioLegend, Catalog No. 100712), anti-mouse NK1.1 (PK136; BioLegend, Catalog No. 108708), anti-mouse CD19 (6D5; BioLegend, Catalog No. 115512), anti-mouse CD11c (N418; BioLegend, Catalog No. 117311), anti-mouse CD86 (GL-1; BioLegend, Catalog No. 105014), anti-mouse MHC-II (M5; BioLegend, Catalog No. 107620), anti-mouse CD103 (M290; BD BioSciences, Catalog No. 563087), anti-mouse/human FoxP3 (FJK-16S; Thermo Fisher Scientific, Catalog No. 17-5773-82), anti-mouse Nur77 (12.14; Thermo Fisher Scientific, Catalog No. 12-5965-82), anti-human Fc (H2; SouthernBiotech, Catalog No. 9042-04), anti-mouse/human CD11b (M1/70; BioLegend, Catalog No. 101208), anti-mouse CD69 (H1.2F3; BioLegend, Catalog No. 104502), anti-mouse/human pSTAT5 (47; BD BioSciences, Catalog No. 612599), anti-human CD4 (RPA-T4; BioLegend, Catalog No. 317416), anti-human CD8 (SK1; BioLegend, Catalog No. 344704), anti-human CD62 L (DREG-56; BioLegend, Catalog No. 304840), anti-human CD45RO (UCHL1; BioLegend, Catalog No. 304222), anti-human CD25 (2A3, BD BioSciences, Catalog No. 340907), anti-human CD45 (HI30; BD BioSciences, Catalog No. 562279), anti-human CD3 (UCHT1; BD BioSciences, Catalog No. 565119), anti-human CD16 (CB16; Invitrogen, Catalog No. 25-0168-42), anti-human PD1 (ebioJ105 and MIH4; eBioSciences, Catalog Nos. 25-2799-42 and 11-9969-42, respectively), anti-human perforin (dG9; BioLegend, Catalog No. 308126). Fixable live/dead cell discrimination was performed using LIVE/DEAD Fixable Aqua stain (Thermo Fisher Scientific, Catalog No. L34966) according to the manufacturer's instructions. Surface staining was carried out on ice for 30 minutes if not indicated otherwise, and intracellular staining was performed using the Foxp3 staining kit (eBioScience, Catalog No. 00-5523-00) according to the manufacturer's instructions. Flow cytometry data from each sample was collected from the Cytoflex Flow Cytometer or a FACSCanto Flow Cytometer. Analyses were done using FlowJo 10.4.

Fresh PBMCs, TILs, or doxycycline hydrochloride treated 32D cell lines were resuspended at 20 to 30 × 10⁶ cells/mL in serum-free RPMI medium and aliquoted at 50 μL per well into 96-well U-bottom plates and rested at 37°C for 10 to 30 minutes. Various IL15 agents (2× stock) were added to cells to desired dilutions as indicated. The treatment was stopped by fixing cells at room temperature (RT) immediately with 4% PFA (2% final volume). Cells were stained with antibodies according to manufacturer protocol against cell surface markers and live/dead dye. After fixation with 4% PFA for 10 minutes, cells were washed and resuspended in 150 μL of prechilled Perm buffer III according to manufacturer's protocol (BD Biosciences, Catalog No. 558050). Cells were then stained with antibodies against intracellular markers (pSTAT5 and perforin; 45 minutes to 1 hour at RT) and analyzed on the flow cytometry the same day.

Fresh PBMC were resuspended at 2,000,000 cells/mL in 1640 RPMI medium supplemented with 10% hAB serum (Corning, Catalog No. 35060CI), GlutaMax (Gibco, Catalog No. 35050061), MEM NEAA (Gibco, Catalog No. 11140050), sodium pyruvate (Gibco, Catalog No. 11360070), and βME (Gibco, Catalog No. 21985023). Cells were plated at 100 μL with 200,000/well in a 96-well plate. Anti-hPD1–IL15m or wild-type (WT) IL15 were added at 2× desired concentration to reach the final concentration at 0.0001, 0.001, 0.01, 0.1, 1, 10, 100, 1,000 and 5,000 nmol/L. A hundred microliters of culture medium was collected on day 3 and cytokines IFNγ, IL6, TNFα, GM-CSF, and IL2 from medium were analyzed using human cytokines MSD V-Plex platforms from Mesoscale according to manufacturing protocols (Meso scale discovery, Catalog No. K15049D).

B16-F10 melanoma cells and MC38 were cultured to sufficient quantities for subcutaneous implantation of 50,000 cells per mouse using a 28-gauge insulin syringe. Groups were treated with either PBS or therapeutic compounds. Length and width of tumors were recorded using caliper measurements, and tumor volume was calculated using the formula Volume = Width² × 0.5 (Length). Body weight and tumor volume were measured every 3 or 4 days (as noted) throughout the study. Where indicated, approximately 75 to 100 μL of whole blood was collected by submandibular bleeds in tubes containing 5 μL of heparin (Sargent Pharmaceuticals, Catalog No. 25021-400-10) and used to identify lymphocyte subsets by flow cytometry.

For pharmacokinetic assessment, anti-mPD1, anti-mPD1-IL15 agonist, anti-mPD1–IL15m-NQ, or anti-mPD1–IL15m was administered to B16-F10 tumor–bearing mice (n = 3 per group) or B16-F10 WT mice at 0.1, 0.3, 1, and 5 mg/kg subcutaneously. Blood was collected at 12, 36, 60, 84, 108, 132, 156, and 180 hours after dosing and processed to serum. The concentrations of anti-mPD1 or the fusion proteins were quantitated using a ligand-binding assay utilizing the Gyrolab lmmunoassay (Gyrolab Technology). The anti-PD1 or anti-mPD1–IL15 various fusion proteins were captured onto the Gyrolab Bioaffy CD using a biotinylated goat anti-human IgG (H+L; Southern Biotech, Catalog No. 2049-08). Bound anti-mPD1–IL15m fusion was detected using a mouse anti-human IL15 (Thermo Fisher Scientific, Catalog No. MA1-24810) labeled with Alexa Fluor 647. The concentrations of anti-PD1 or fusion proteins were determined by interpolation from a calibration curve that was fitted using a 5-parameter logistic regression model. The range of quantitation of the assay was 0.1 to 1.5 μg/mL in 100% serum. Cytokines IFNγ, IL6, TNFα, and GM-CSF from serum were analyzed using mouse cytokines MSD U-Plex platforms from Mesoscale according to manufacturing protocols (Meso scale discovery, Catalog No. K152A0H).

To deplete CD8⁺ cells, 350 μg of anti-CD8β (BioXcell, Catalog No. BE0223) were administered intraperitoneally to mice on day −3, day −1, and once weekly thereafter upon treatment (day 1). NK⁺ cells were depleted by intraperitoneal administration of anti-NK1.1 (200 μg; BioXcell, Catalog No. BE0036) beginning on day −2 and once weekly thereafter upon therapeutic compound administration. To prevent immune-cell egress from lymph nodes to the circulation, the inhibitor FTY720 (Sigma-Aldrich, Catalog No. SML0700) was administered intraperitoneally at day −3, day −1, and every 3 days thereafter during the study.

Each tumor was added to a Miltenyi C tube (Miltenyi, Catalog No. 130-093-237) containing tumor dissociation buffer (Miltenyi, Catalog No. 130-096-730, for each sample, 2.35 mL of DMEM, 10 μL of Reagent D, 5 μL of reagent R, and 1.25 μL of reagent A). Samples are placed on the gentleMACS OCTO Dissociator with the setting M-37_mTDK1 protocol. Once completed, the cell paste was transferred on top of a 70 μm MACS SmartStrainer over a 15 mL conical vial. Ten milliliters of DMEM containing 5% FBS was used to rinse the tumors through the strainer. The cells were then pelleted by centrifugation at 1,200 rcf for 5 minutes. Cell viability and number were measured on the Vi-Cell XR.

Human TILs were cultured in RPMI1640 (Gibco, Catalog No. 21875034) supplemented with 10% human AB serum (Sigma, Catalog No. H4522), 2 mmol/L L‐glutamine (Invitrogen, Catalog No. 25030149), 50 mM Hepes Buffer (Lonza, Catalog No. 15630080), 1% penicillin–streptomycin (Life Technologies, Catalog No. 15140122), 5 mmol/L Sodium Pyruvate (Gibco, Catalog No. 11360070), and 1% minimum essential medium nonessential amino acids (MEM NEAA; Gibco, Catalog No. 11140050) at 37°C in 96-well flat-bottom plates (5 × 10⁵ CD45⁺ cells per well). Cell were treated with 1 to 10,000 pmol/L IL15 agonist, anti-hPD1-IL15m, Iso-IL15m, and/or 10 μg/mL anti-PD1 for 4 to 9 days.

Human TILs and HepG2-RFP were cocultured in effector:target ratios of 3:1 or 10:1 in RMPI complete with 10% human serum and CellEvent Caspase-3/7 Green Detection Reagent (1:500; Thermo Fisher Scientific, Catalog No. C10723;). Alexa Fluor 568 (HepG2-RFP) and Alexa Fluor 488 (CellEvent Caspase-3/7 Green Detection Reagent) were detected in 16 fields per well using a 20× air objective for 30 hours at 2 hours intervals. Alexa Fluor 488 median fluorescence intensity within Alexa Fluor 568–positive HepG2-RFP cells was used to determine Caspase 3/7 activity.

Tumors were harvested on day 7 and day 10 and viable CD45⁺ cells were FACS sorted. Single-cell emulsions were obtained using the 10× Genomics Controller and the v2 library and Gel Bead Kit (10× Genomics, Catalog No. 1000265). RNA-sequencing (RNA-seq) libraries were prepared as instructed by the 10 × 3′ v2 Kit protocol. Resulting libraries were sequenced on an Illumina NextSeq using a NextSeq 500/550 v2.5 High Output Kit (Illumina, Catalog No. 20024907). Single-cell RNA-seq (scRNA-seq) data have been deposited in NCBI's Gene Expression Omnibus under the accession code GSE160579.

Human TILs were cultured for 9 days in the absence or presence of 100 pmol/L anti-hPD1–IL15m or the combination of 10 μg/mL anti-hPD1 and 100 pmol/L IL15 agonist. Then, CD8⁺ (CD45⁺CD3⁺CD56⁻CD8⁺) and CD4⁺ TILs (CD45⁺CD3⁺CD56⁻CD4⁺CD127⁺CD25⁻) were sorted into 500 μL RLTplus buffer (Qiagen, Catalog No. 1053393) using a FACSAria cell sorter (BD Biosciences). RNA was isolated using RNeasy Plus Micro Kit (Qiagen, Catalog No. 74034) according to the manufacturer's instructions. Barcoded mRNA-seq cDNA libraries were prepared using TruSeq RNA Access Libraryprep Kit (Illumina, Catalog No. 20020189) according to the manual. Barcoded RNA-seq libraries were onboard clustered using HiSeq X Kit v2 using 2.5 nmol/L and PE 150 bps and were sequenced on the Illumina HiSeqX using the HiSeq X Kit v2 (PE 150 cycle). RNA-seq data have been deposited in NCBI's Gene Expression Omnibus under the accession code GSE157893.

The statistical significance of differences between experimental groups was assessed by two-way ANOVA using GraphPad Prism 7.04. For single comparisons, a two tailed Student t test was used. Patient-derived samples were analyzed using nonparametric Wilcoxon paired test, Spearman correlation analysis, or Kruskal–Wallis test (*, P < 0.05; **, P < 0.01; ***, P < 0.001).

A prerequisite for PD1-targeted IL15 cytokine therapy is that there is enrichment in expression of PD1 at the tumor site and expression of IL15 receptor by PD1⁺ TILs. We first examined PD1 and IL2Rβ subunit expression levels on CD8⁺, CD4⁺, and NK cells in PBMCs, tumor-free tissues, and tumor tissues from patients with HCC and colorectal cancer. Expression of IL2Rβ was significantly higher on NK cells than T cells in HCC tumors (Fig. 1A and B), whereas tumor-infiltrating NK cells expressed lower IL2Rβ levels than NK cells from tumor-free liver tissue (Fig. 1B). Furthermore, significantly higher PD1 [both PD1⁺ percentage and PD1 mean fluorescence intensity (MFI)] expression was detected on CD8⁺ and CD4⁺ cells in the tumor tissue of patients with HCC and colorectal cancer compared with peripheral blood, whereas NK cells showed low PD1 levels in all tissues (Fig. 1C and D; Supplementary Fig. S5A–S5C). These data suggest that PD1 can serve as a suitable target to deliver immunomodulatory agents, including cytokines, to intratumoral T cells.

To optimally activate intratumoral CD8⁺ T cells that are enriched for the inhibitory marker PD1 in human tumors (Fig. 1C and D), we generated a fusion protein designated in this report as anti-PD1–IL15 mutein (anti-hPD1–IL15m). This molecule consists of a human PD1-binding human IgG1 antibody with bispecific knob-into-hole mutations on CH3 (Y349C, T366W, S354C, T366S, L368A, and Y407V) that is linked monovalently to an IL15 variant molecule with mutations (N1G-D30N-E46G-V49R-E64Q) at the C-terminus through a flexible “GGGGSGGGGSGGGG” linker (Fig. 2A). This molecule utilizes a modified (L234A, L235A, and G237A) human IgG1 Fc region to abrogate Fcγ receptor and C1q binding (25, 26). The five mutations (N1G, D30N, E46G, V49R, and E64Q) were introduced in IL15 to eliminate IL15Rα binding and reduce IL2Rβ/γ affinity (Fig. 2B and C) from a K_D of 21.6 nmol/L (hIL15) to K_D over 1,500 (mutein) nmol/L. Abolishing binding to IL15Rα could increase the concentration of soluble IL15 by preventing its clearance from the circulation through binding to IL15Rα⁺ cells such as endothelial cells (27). When fused to an anti-PD1, the decreased IL2Rβ/γ binding affinity of the IL15 mutein should result in weak activity on PD1 cells but higher, avidity-driven activity on PD1-expressing immune cells (Fig. 2D). The PD1 specificity also should function to not only block the PD1/PD-L1 axis and prevent PD1-mediated T-cell suppression, but also to preferentially deliver the IL15 bioactivity to PD1-expressing intratumoral CD8⁺ T cells. To confirm that anti-hPD1–IL15m can block the PD1/PD-L1 axis, we generated competitive binding profiles of anti-hPD1–IL15m against nivolumab and pembrolizumab, which target epitopes on the PD1 molecule with high affinity and specificity to block PD1/PD-L1 interaction (28). We engineered a 32D cell line to stably express a doxycycline-inducible mCherry-tagged hPD1 protein, enabling a surrogate marker for hPD1 expression after doxycycline treatment (Fig. 2E; Supplementary Fig. S6). This 32D cell line was also transduced to express hIL2Rβ to permit IL15 responsiveness because the parental 32D cells lack this critical component of the functional IL15 receptor (Fig. 2E). Anti-hPD1–IL15m started to lose binding to hPD1 on 32D cell line when nivolumab and pembrolizumab concentrations increased (Fig. 2F). At the concentration of 1 μg/mL, nivolumab and pembrolizumab completely blocked the binding of anti-hPD1–IL15m to hPD1-expressing cells, suggesting anti-hPD1–IL15m binds to similar epitopes of nivolumab and pembrolizumab and likely block PD1/PD-L1 interactions.

To confirm that anti-hPD1–IL15m preferentially activated cells that coexpress IL2Rβ/γ and PD1, we compared the ability of anti-hPD1–IL15m to induce IL2R-dependent STAT5 (pSTAT5) signaling in doxycycline-induced PD1⁺ 32D cells as compared with cells in which PD1 was not induced. Anti-hPD1–IL15m exhibited superior pSTAT5 induction on PD1⁺ 32D cells, whereas recombinant hIL15 or an IL15 superagonist—a fusion protein containing human IL15 with the N72D mutation and linked to the IL15Rα Sushi domain for increased potency and biased IL2Rβ/γ activity (29)—had no preference in pSTAT5 induction in the PD1⁺ and PD1⁻ cells (Fig. 2G and H). Therefore, anti-hPD1–IL15m allowed for preferential targeting of PD1-expressing cells over those not expressing PD1.

Because of affinity differences in human PD1 and mouse PD1, we generated a mouse surrogate molecule anti-mPD1–IL15m, consisting of an anti-mouse PD1 with a modified (L234A, L235A, and G237A) human IgG1 Fc region fused with an IL15 mutein (N1A, D30N, E46G, and V49R) that has similar binding properties for mouse IL15 receptors (eliminated IL15Rα binding and similar binding affinity to IL2Rβ) anti-hPD1–IL15m with human IL15 receptors (Supplementary Table S1). First, we performed an ex vivo biodistribution analysis to determine if anti-mPD1–IL15m localized to the tumor microenvironment by binding PD1 on intratumoral CD8⁺ T cells in B16-F10 tumor–bearing mice. By staining with anti-hFc, we detected anti-mPD1–IL15m on T cells in the tumors of treated mice at day 2 (Fig. 3A), and both parent anti-mPD1 and anti-mPD1–IL15m bound to >90% of the CD8⁺ TILs, which was significantly higher than CD8⁺ T cells from peripheral blood and spleen (Fig. 3A). We also observed much lower binding of anti-mPD1 and anti-mPD1–IL15m to NK cells in the peripheral blood, spleen, and tumors (Fig. 3B). The binding pattern of anti-mPD1 resembled that of the anti-mPD1–IL15m in this analysis, suggesting that anti-mPD1–IL15m binding was driven by PD1. We believe the loss of anti-mPD1–IL15m detection in the tumor at day 5 (Fig. 3C) is because the fusion protein had significantly lower exposure and shortened half-life compared with the anti-mPD1 (Fig. 3D). Loss of anti-mPD1–IL15m detection correlated with anti-mPD1–IL15m clearance in the serum after 132 h (Fig. 3A, C, and D), whereas the anti-mPD1 still bound to intratumoral CD8⁺ T cells (Fig. 3C) in accordance with longer serum exposure. These data suggest that the anti-mPD1–IL15m was cleared more rapidly than the anti-mPD1 and that anti-mPD1–IL15m targeted the tumor microenvironment through PD1-expressing TILs.

The antitumor efficacy of anti-mPD1–IL15m was determined in two syngeneic mouse tumor models, B16-F10 melanoma and MC38 colon cancer. Anti-mPD1–IL15m was administered as a single subcutaneous injection resulting in robust dose-dependent efficacy in both tumor models (Fig. 3E and F). In addition, we compared the antitumor efficacy of anti-mPD1–IL15m to an IL15 superagonist. This superagonist has enhanced biologic activity in vivo and a substantially longer serum half-life than recombinant IL15 (30). The molecular weight of the IL15 superagonist is half that of the anti-mPD1–IL15m molecule and the IL15 superagonist has two IL15 moieties per molecule. Hence, we had to dose the anti-mPD1–IL15m 4 times higher than the superagonist in order to reach dosing equimolarity. Comparison between anti-mPD1–IL15m and IL15 superagonist in the B16-F10 melanoma model at equimolar IL15-matched doses demonstrated that anti-mPD1–IL15m was more efficacious than IL15 superagonist, anti-mPD1, or IL15 superagonist combined with anti-mPD1 (Fig. 3G). In addition, doses higher than 0.25 mg/kg of IL15 superagonist (1 mg/kg), although reaching similar antitumor efficacy as anti-mPD1–IL15m (Fig. 3H), led to significant transient body weight loss (BWL) at much lower doses than anti-mPD1–IL15m (Fig. 3I and J). These data indicate that at non-toxic dose levels, anti-mPD1–IL15m is a superior immunomodulatory agent when compared with combined treatment of IL15 superagonist and PD1 blockade.

Although anti-mPD1–IL15m treatment at 5 mg/kg induced the best antitumor efficacy (Fig. 3E and F), it also led to concomitant toxicity as BWL (Fig. 3K and L). To gain insight into immune mechanisms underlying the toxicity of anti-mPD1–IL15m, we depleted NK or CD8⁺ T cells to examine how they impacted BWL. We found depletion of NK1.1⁺ but not CD8⁺ T cells prevented BWL following anti-mPD1–IL15m treatment (Fig. 3M). Moreover, anti-mPD1–IL15m treatment at 5 mg/kg induced significant upregulation of NK-cell numbers both in the tumor and peripheral blood (Fig. 3N and O). These data suggest our approach is effective at low doses to target IL15 away from NK cells and toward CD8⁺ T cells to achieve better antitumor responses and fewer toxicities. However, NK cells will be activated at high doses and ultimately contributes to therapeutic efficacy and toxicities.

In patients treated with IL15, cytokine production correlates with the development of immune-related toxicities (31). This is called cytokine release syndrome and it is thought to result from the unchecked activation and proliferation of T-cell and NK-cell subsets. In line with this, we observed dose-dependent cytokine release by human PBMCs following in vitro stimulation with WT IL15 agonist (fusion of WT IL15 and IL15Rα Sushi domain) or anti-hPD1–IL15m (Supplementary Fig. S7). However, in vitro treatment with anti-hPD1–IL15m elicited lower levels of IFNγ, IL2, GM-CSF, IL6, and TNFα (Supplementary Fig. S7) compared with WT IL15 agonist, suggesting anti-hPD1–IL15m may lead to reduced risk of severe cytokine-mediated toxicities in vivo compared with WT IL15 agonist.

In vivo, compared with anti-mPD1–IL15m-NQ (Supplementary Table S4), anti-mPD1–IL15 WT agonist (anti-mPD1 fused to WT IL15 agonist with an IL2Rβ binding K_D = 37.7 nmol/L; Supplementary Table S4) treatment substantially increased systemic cytokine release (Supplementary Fig. S8), like we observed in hPBMC in vitro (Supplementary Fig. S7). IL15m_NQ is another mouse surrogate similar to IL15m with an IL2Rβ binding K_D = 518 nmol/L (Supplementary Fig. S8A). Anti-mPD1–IL15m_NQ also consists of the same mouse PD1 binding human IgG1 antibody and monovalent linker as anti-mPD1–IL15m, but the IL15 variant molecule has different mutations, it is six mutations (D22N-Y26F-E64Q-E53Q-E89Q-E93Q) at the C-terminus to eliminate IL15Rα binding and reduce IL2Rβ/γ affinity compared with WT IL15. Its binding affinity for IL2Rβ/γ affinity is similar to that of anti-mPD1–IL15m (Supplementary Fig. S8A). This suggested that the binding affinity to IL2Rβ contributed to the cytokine-mediated toxicity and the use of more attenuated IL15 molecules could potentially help to avoid treatment-related side effects. Moreover, it is well established that IL2Rβ can mediate internalization and degradation of IL2 (32–34). Here, we observed that the binding affinity to IL2Rβ also contributed to the distinct pharmacokinetic profiles in the IL15 fusion proteins. Compared with anti-mPD1, the fusion proteins had significantly lower exposures and shortened half-life (Supplementary Fig. S8B). The anti-mPD1–IL15 WT agonist cleared faster than anti-mPD1–IL15m_NQ in the serum, suggesting IL2Rβ-mediated endocytosis played a major role in these fusion proteins' pharmacokinetic properties.

To determine the mechanisms by which anti-mPD1–IL15m treatment showed enhanced efficacy of tumor control compared with IL15 superagonist and anti-mPD1 combinatorial treatment, we compared the immune subset profile within B16-F10 tumors after treatments with vehicle, anti-mPD1, IL15 superagonist, anti-mPD1 + IL15 superagonist, or anti-mPD1–IL15m using flow cytometry. Although there was some heterogeneity, anti-mPD1–IL15m significantly increased the number of CD8⁺ TILs compared with all other treatments while leaving the number of peripheral blood CD8⁺ T cells unaffected (Fig. 4A). B16-F10 tumor cells are known to be sensitive to NK-cell and CTL cytotoxicity and both cells can be activated by IL15. However, the major functional cell types that induced antitumor responses after anti-mPD1–IL15m treatment were unclear, as was whether the efficacy was dependent on lymph node egress. Therefore, we evaluated the roles of NK cells, CD8⁺ T cells, and lymph node egress in B16-F10–bearing C57BL/6 mice treated with 1 mg/kg anti-mPD1–IL15m. The therapeutic effect of anti-mPD1–IL15m was abolished in mice depleted of CD8⁺ cells (anti-CD8, Fig. 4B) and less affected by NK-cell depletion (anti-NK1.1; Fig. 4C) or treatment with FTY720 (Fig. 4D), an inhibitor of T-cell egress from lymph nodes. These data suggest that the therapeutic effect of anti-mPD1–IL15m was mainly driven by CD8⁺ cells, whereas NK cells and the influx of newly recruited T cells into the tumor microenvironment were less important for antitumor immunity induced by anti-mPD1–IL15m. In contrast, IL15 superagonist efficacy was compromised when NK cells or CD8⁺ T cells were depleted, and also when lymph node egress was inhibited (Fig. 4E–G). This indicated that IL15 superagonist–mediated antitumor efficacy was dependent on CD8⁺ T cells, NK cells, and immigration of recently primed T cells into tumor tissue. Taken together, these results suggest that although NK cells and T cells from lymph nodes contribute to the therapeutic efficacy of anti-mPD1–IL15m, intratumoral CD8⁺ T cells are the major population that mediated the efficacy of this agent.

To obtain an unbiased view of the transcriptome and cellular composition of tumors from B16-F10–bearing mice after treatment with anti-mPD1–IL15m, we analyzed intratumoral CD45⁺ immune cells by scRNA-seq from mice treated with PBS, anti-mPD1 and/or IL15 superagonist or anti-mPD1–IL15m. We identified 21 clusters and assigned cell identities to the clusters (Fig. 5A; Supplementary Fig. S9A; Supplementary Data File S1). The dominant effect of anti-mPD1–IL15m treatment was a dramatic increase in CD8 frequencies and cell numbers, and a reduction in myeloid populations [tumor-associated neutrophils (TAN), dendritic cells (DC), and tumor-associated macrophages (TAM)] compared with the other treatment groups. In contrast, IL15 superagonist, with or without anti-mPD1, treatment induced the most dramatic upregulation of NK cells (Fig. 5B).

Intratumoral CD8⁺ T cells typically bear markers of exhaustion and reside in either a progenitor-exhausted or terminally exhausted subset (35). To investigate how treatment with anti-mPD1–IL15m affected transcriptional signatures of exhaustion in CD8⁺ TILs, we analyzed the five CD8 clusters identified by unsupervised clustering (Fig. 5C). This analysis revealed that Mki67, a marker of cellular proliferation and T-cell reinvigoration in mouse models upon checkpoint blockade treatment (36), was upregulated in C20 (Fig. 5D; Supplementary Data File S2). C20 exhibited strong proliferation- and prosurvival-associated signatures (Fig. 5E) and was increased in response to anti-mPD1–IL15m treatment (Fig. 5F). However, in the group treated with anti-mPD1 or IL15 superagonist alone, or the combination of anti-mPD1 and IL15 superagonist, C20 was not enriched. Instead, a moderately enhanced frequency of C17 and a reduction of C14 was observed (Fig. 5F). C17 likely represents stem-like/progenitor CD8⁺ T cells (Tex cells) as C17 showed high expression of Ly6c2, Tcf7, and Gzmk (Fig. 5D; Supplementary Fig. S9B; Supplementary Data File S2). In addition, expression of homing receptors Cxcr3 and Cxcr4, which are associated with stem-like/progenitor exhausted CD8⁺ Tex cells (37, 38), were enriched in cluster 17 (Supplementary Fig. S9B), accompanied by enriched T-cell migration pathways (Fig. 5E). Previously well-established exhaustion markers Pdcd1, Lag3, Tox, and Havcr2 (39) were highly expressed in both C14 and C21 clusters (Fig. 5D). C14 expressed Cd244 (Supplementary Fig. S9B), which is reported as a more terminally exhausted CD8⁺ T-cell phenotype, and was enriched for exhaustion and senescence pathways, suggesting that C14 represented a terminally exhausted CD8⁺ T-cell population (Fig. 5E). C21 highly expressed Ifng, Ccl3, and Ccl4 as well as Il2ra and thus resembled effector-like CD8⁺ TILs (Fig. 5D; Supplementary Fig. S9B; Supplementary Data File S2). C15 was enriched with IFN-associated genes such as Ifit1 and Ifit3 (Fig. 5D; Supplementary Fig. S9B) and associated with antiviral response pathways (Fig. 5E). We observed no major differences in C15 and C21 across treatments (Fig. 5F). Although the integrated transcriptional analysis established that the intratumoral CD8⁺ T-cell compartment comprised five major subsets and demonstrated the expansion of these subsets upon treatment, the relationship and differentiation states within these clusters remained to be elucidated. To understand this further, we constructed single-cell trajectory analyses composed of pseudo-temporally ordered CD8⁺ T cells across the distinct T-cell states (Supplementary Fig. S9C). C17 cells were located at one edge, whereas C20, enriched in tumors after anti-mPD1–IL15m treatment, occupied the opposite branch, pointing toward a possible conversion of Tex cells (C17) into proliferating cells displaying an early/intermediate exhausted state (C20), in accordance with previous reports (Supplementary Fig. S9C; ref. 40). Consistent with this model, we found key stem-like markers within C17, Tcf7, downregulated in C20, whereas effector marker Granzyme B expression was markedly increased (Fig. 5D). These results suggest that the loss or downregulation of genes associated with a Tex-cell phenotype (C17) was associated with the acquisition of an effector phenotype (C20) in response to anti-mPD1–IL15m treatment.

Throughout all clusters, anti-mPD1–IL15m treatment resulted in a large shift from high Eomes/Tox toward a high Tbet, low Tox/Eomes CD8⁺ TIL signature (Fig. 5G). Coexpression of Tox and T-bet identifies distinct exhausted CD8⁺ T-cell populations in chronic infection and cancer, and lower Tox expression was associated with T-bet^hi Tex cells with regained “effector-like functions” (41). Indeed, accompanying the increase of T-bet in the anti-mPD1–IL15m treated group is a marked upregulation of Granzyme (Gzmb) and Bcl2, and a downregulation of Tox and Eomes (Fig. 5G). Differential gene expression (DEG) analysis revealed that anti-mPD1–IL15m treatment induced a transcriptomic landscape that was distinct from that of cells treated with IL15 superagonist plus anti-mPD1 or PBS (Supplementary Fig. S10A and S10B). Compared with stimulation with IL15 superagonist plus anti-mPD1, anti-mPD1–IL15m boosted the expression of several granzyme genes in CD8⁺ TILs in most clusters (Supplementary Fig. S10B; Supplementary Data File S3). Consistent with this, “Granzyme-mediated signaling” and “Cytolysis” were among the top enriched pathways in anti-mPD1–IL15m–treated CD8⁺ TILs compared with either PBS treated (Supplementary Fig. S10C) or IL15 superagonist plus anti-mPD1 treatment (Fig. 5H) in clusters C14, 15, 20, and 21. Collectively, those data point toward an anti-mPD1–IL15m–induced shift in the tumor immune landscape in favor of CD8⁺ Tex TILs that regained proliferative and effector functions.

To evaluate the translational relevance of our mouse data, we performed bulk RNA-seq on in vitro IL15 agents treated HCC-derived human CD8⁺ TILs. Patient and basic tumor-immune characteristics for the samples used in this assay are provided in Supplementary Table S5. Principal component analysis revealed that TILs treated with anti-hPD1–IL15m differed greatly from TILs treated with IL15 agonist and anti-hPD1 (Fig. 6A; Supplementary Data File S4). Pathway analysis of DEGs demonstrated that CD8⁺ TILs treated with anti-hPD1–IL15m showed enhanced expression of markers associated with cell-cycle progression and cell survival, whereas cell death and senescence pathways were downregulated compared with CD8⁺ TILs treated with IL15 agonist and anti-hPD1 (Fig. 6B; Supplementary Fig. S11A and S11B). Collectively, human CD8⁺ TILs treated in vitro with anti-hPD1–IL15m closely resembled cluster 20 of the scRNA-seq analysis of murine CD8⁺ TILs (Fig. 6B), pointing toward a shared reprogramming of mouse (Fig. 5E) and human CD8⁺ TILs toward a prosurvival and proliferation signature. In particular, genes associated with T-cell activation (IL2RA, CD38), cytotoxic potential (IFNG, GZMB, FASLG), cell survival (BCL2, MCL1), and cell cycle (MKI67, CDC20) were upregulated in response to anti-hPD1–IL15m treatment, whereas the combination of IL15 agonist and anti-hPD1 did not upregulate expression of these genes (Fig. 6C).

To verify that the observed transcriptomic modifications affected expansion and function of human TILs, we tested the effects of anti-hPD1–IL15m on human HCC- and LM from colorectal cancer (LM-CRC)–derived TILs in vitro. Frequencies of pSTAT5⁺CD8+ TILs were upregulated by anti-hPD1–IL15m in a dose-dependent manner, whereas treatment with the nontargeting Iso-IL15m did not result in STAT5 phosphorylation (Supplementary Fig. S11C). In accordance with PD1 levels, anti-hPD1–IL15m treatment enhanced the percentage of pSTAT5⁺CD8⁺ TILs to a greater extent than in CD8⁺ PBMCs (Supplementary Fig. S11D). Although WT IL15 agonist treatment strongly upregulated Ki67 expression in both tumor-derived NK cells and CD8⁺ T cells, anti-hPD1–IL15m preferentially enhanced Ki67 expression in CD8⁺ TILs but not tumor-derived NK cells at concentrations of 1 nmol/L or less (Fig. 6D).

To analyze the specificity of anti-hPD1–IL15m targeting, CD8⁺ TILs were gated as PD1⁺ and PD1⁻ subsets according to PD1 expression after in vitro culture (Supplementary Fig. S11E). Upon anti-hPD1–IL15m treatment, PD1⁺CD8⁺ TILs were most sensitive to Ki67 induction (Supplementary Fig. S11F). In contrast, WT IL15 agonist in combination with anti-hPD1 enhanced Ki67 levels to a similar degree in the absence or presence of PD1 expression, and comparably with anti-hPD1–IL15m in PD1⁺CD8⁺ TILs (Supplementary Fig. S11F). Consistent with this, baseline PD1 but not IL2Rβ expression on CD8⁺ TILs correlated significantly with Ki67 expression upon anti-hPD1–IL15m treatment (Fig. 6E; Supplementary Fig. S11G), demonstrating that the activity of anti-hPD1–IL15m on proliferation of human tumor–derived CD8⁺ T cells is mainly driven by PD1 levels. Consistent with elevated Ki67 expression, anti-hPD1–IL15m treatment enhanced absolute cell counts of CD8⁺ TILs, whereas NK-cell numbers remained largely unaffected (Fig. 6F).

In line with the transcriptomic data, after anti-hPD1–IL15m treatment, the frequency of GzmB-expressing cells as well as Bcl2 and CD137 levels were elevated in the majority of PD1⁺CD8⁺ TILs compared with untreated TILs (Fig. 6G–I). In addition, we further evaluated the cytotoxic capacity of CD3⁺ TILs treated with anti-hPD1–IL15m in an allogeneic HepG2 cytotoxicity assay. HCC-derived CD3⁺ TILs were stimulated with anti-hPD1–IL15m and subsequently cocultured in the presence of Caspase 3/7 substrate with RFP-H2B–transduced HepG2 cells. In contrast to nonstimulated TILs or TILs treated with Iso-IL15m, TILs treated with anti-hPD1–IL15m decreased HepG2 numbers (Fig. 6J) and induced HepG2 apoptosis, as measured by caspase 3/7 upregulation (Fig. 6K). Taken together, these data show that anti-hPD1–IL15m preferentially induced PD1⁺CD8⁺ TIL expansion and activation, and boosted the cytotoxic potential of HCC-derived TILs.

One of the limiting factors for effective immunotherapy is the expansion of immunosuppressive CD4⁺FoxP3⁺ Treg cells. Given the similarities between IL2 and IL15, and that PD1 expression is observed on Tregs (42), anti-hPD1–IL15m may also engage Treg cells and promote their proliferation. In B16-F10 tumors, we identified four clusters of CD4⁺ TILs, which were annotated as Treg cells (C1–C3) and conventional T (Tconv) cells according to Foxp3 and Il2ra levels (Fig. 7A; Supplementary Data File S5). Treatment with anti-mPD1–IL15m or IL15 superagonist with or without anti-mPD1 preferentially expanded Tconv cells over Treg cells, whereas anti-mPD1 alone did not alter the frequency of CD4⁺ T-cell subsets compared with PBS-treated controls (Fig. 7B).

Next, we analyzed the effect of anti-hPD1–IL15m on CD4⁺ TILs from patients with cancer. Tumor-infiltrating activated Treg (aTreg) cells were gated as CD4⁺CD45RA⁻Foxp3^hi and were present in the majority of patient samples at higher frequencies than in blood or surrounding tumor-free tissue, and expressed PD1 at slightly higher densities compared with Tconv (T helper) cell subsets (Supplementary Fig. S12A–S12D). Nevertheless, Tconv-cell subsets were more sensitive to pSTAT5 induction by anti-hPD1–IL15m treatment than CD4⁺Foxp3⁺ Treg cells (Fig. 7C). IL2 plays a critical role in the maintenance of FoxP3 expression and Treg-cell functions. We observed that anti-hPD1–IL15m was less effective in maintaining Foxp3 expression compared with IL2 in purified tumor-derived CD25^hiCD127⁻ Treg cells (Supplementary Fig. S11E and S11F), whereas treatment with anti-hPD1–IL15m enhanced expression of genes associated with T-cell activation and effector function in CD4⁺ TILs (Fig. 7D), suggesting that anti-hPD1–IL15m had more impact on CD4⁺ Tconv cells than Treg cells. Taken together, these data suggest that anti-PD1–IL15m selectively activates intratumoral CD4⁺ Tconv cells over Treg cells in both mice and humans.

Various approaches have been used to modify cytokines for cancer immunotherapy (43) with a goal of increasing the therapeutic index. One approach is to attach cytokines to tumor-targeting antibodies or antibody fragments (44, 45) to increase local concentration and the pharmacologic activities of the cytokines at tumor sites, thereby decreasing the systemic effective dose and improving the therapeutic window. However, tumor-targeting efficacy is often limited by the high-affinity cytokine receptors that are ubiquitously expressed on normal cells and tissues. These receptors may not only provide “sinks” that bind the immunocytokines and impair their ability to reach the tumor cells, but also contribute to toxicity in peripheral tissues (3, 46).

To address the persisting systemic toxicity issue, numerous mutational or chemical modification efforts have been employed to modulate cytokine–receptor interactions, although most of these studies focused on IFNα, IL21, and IL2 (47–56). Here, we engineered an anti-hPD1–IL15m antibody–cytokine fusion that was designed to deliver PD1-mediated avidity-driven IL15 receptor stimulation preferentially to intratumoral T cells while reducing the natural preference of IL15 for peripheral NK cells and other PD1⁻ lymphocytes. Indeed, anti-hPD1–IL15m treatment of human CD8⁺ TILs led to upregulation of pSTAT5, proliferation, and cytotoxic activities, and tumor-derived NK cells and peripheral CD8⁺ T cells remained largely unaffected. Similar results were obtained in in vivo mouse tumor models using the mouse surrogate anti-mPD1–IL15m. These differences between peripheral and intratumoral activation could potentially allow systemically administered anti-hPD1–IL15m to reach the intratumoral activity required for inducing tumor regression in patients, without causing overt systemic toxicity, and represent the basis of differentiation from other untargeted IL15 agonists.

Although the therapeutic effects of IL2 and IL15 in the clinic are encouraging, a number of patients have experienced serious treatment-related adverse events caused by systemic proinflammatory cytokine production. In this study, we showed that cytokine release from human PBMCs or mouse serum was significantly lower following treatment with anti-hPD1–IL15m or anti-mPD1–IL15m_NQ as compared with IL15 WT agonist or anti-mPD1–IL15 WT agonist, respectively. In addition, we observed that the anti-mPD1–IL15m mouse surrogate achieved remarkable efficacy without measurable weight loss in B16-F10 tumor–bearing mice. Although the mechanism of the toxicity may be multifactorial by different immuno-oncology agents, immunotoxicity in recombinant cytokine–treated patients has been shown to correlate with the level of induced circulating inflammatory cytokines and weight loss (3). Our data suggest that the anti-hPD1–IL15m may reduce cytokine release–related toxicities by PD1 targeting and IL15 attenuation.

T-cell infiltration of the tumor has been associated with patient survival and improved immunotherapy responses (57), and a major factor to checkpoint-inhibitor resistance (anti-PD1 or anti-CTLA4) is lack of tumor-infiltrating T cells. Therefore, the expansion of TILs with anti-PD1–IL15m could potentially provide a therapeutic advantage in checkpoint inhibitor–resistant tumors. In this study, we have demonstrated that anti-PD1–IL15m treatment resulted in much better efficacy than that mediated by anti-PD1 treatment. Because PD1⁺ T cells are enriched in tumors, anti-PD1–IL15m has the capacity to both block the PD1/PDL1 axis and efficiently target IL15 bioactivity to intratumoral T-cell antitumor responses. This two-pronged effect likely explains why anti-PD1–IL15m provided superior antitumor efficacy compared with IL15 cytokines or anti-PD1 single agents. In addition, we have also observed enhanced antitumor efficacy of anti-PD1–IL15m compared with the combination of IL15 superagonist with anti-mPD1 in B16-F10 tumor–bearing mice. This IL15 superagonist, also called ALT803, has advanced to the clinic and shown promising preliminary results as a monotherapy, or in combination with PD1 blockade, both in anti-PD1–naïve and anti-PD1–relapsed/refractory patients (11, 58). The fact that anti-mPD1–IL15m alone demonstrated a more effective therapeutic effect compared with the IL15 superagonist or the combination of IL15 superagonist and anti-mPD1, suggests a high translational promise for clinical use. Moreover, mice bearing B16-F10 or MC38 exhibited dose-dependent efficacy with the treatment of anti-mPD1–IL15m, suggesting that the approach of anti-hPD1–IL15m may be generally applicable to multiple types of solid tumors.

T-cell exhaustion is a major outcome of PD1 blockade and is manifest by a spectrum of phenotypic and functional properties of tumoral CD8⁺ T-cell subpopulations (35, 59, 60). Recent advances in scRNA-seq technologies have helped resolve this heterogeneity and the distribution of certain Tex-cell subsets in tumors could be related to response to checkpoint blockade (39). PD-1 blockade efficacy has been associated with an increase of progenitor exhausted T cells (35), as well as an increase of effector functions in early and terminal CD8⁺ Tex cells (59). In accordance with previous reports, in the tumors of mice treated with anti-mPD-1 (with or without IL15 superagonist), we observed an enrichment of cells belonging to the C17 progenitor Tex-cell subset. The group treated with anti-mPD1–IL15m greatly expanded C20. Trajectory analyses demonstrated that C17 and C20 represented two distinct developmental CD8⁺ T-cell states, whereby Tex cells (C17) may progress toward an intermediate exhausted (C20) state that is highly proliferative and “effector like,” as recently described by Beltra and colleagues (41). When we compared the transcriptome of patient-derived CD8⁺ TILs treated in vitro with anti-hPD1–IL15m to cells treated with IL15 agonist plus anti-hPD1, we found an expression signature that closely resembled C20, pointing toward the development of a proliferating, effector-like exhausted human TIL population in response to anti-hPD1–IL15m treatment. This conclusion was confirmed at the protein level by the observation that human CD8⁺ TILs cultured in vitro with anti-hPD1–IL15m showed induction of granzyme B and Bcl2, and at the functional level by their enhanced cytotoxic killing capacity. Further studies are required to understand why anti-mPD1–IL15m, but not the combined treatment of IL15 superagonist plus anti-mPD1 might induce the conversion of Tex cells (C17) toward granzyme-expressing proliferating CD8⁺ Tex cells (C20), and if the antitumor functions of these proliferating, effector-like exhausted TILs are the primary contributors to the increased efficacy observed after treatment with anti-mPD1–IL15m.

In conclusion, we have developed an anti-PD1–fused IL15 cytokine mutein for cancer immunotherapy that delivers the benefit of IL2/15 to tumoral CD8⁺ T cells. We showed that this agent enhances the expansion and cytotoxic capacity of patient-derived TILs and antitumor efficacy in mouse tumor models, accompanied by reduced adverse effects. Altogether, our study holds promise that administration of the engineered anti-hPD1–IL15m can overcome the current limitations of cytokine therapies and provide a strong rationale for the clinical evaluation of this fusion protein in cancer patients.

L. Campos Carrascosa reports grants from Pfizer Inc. during the conduct of the study. Y.A. Yeung reports a patent for US20190263877 pending to Pfizer Inc. M.L.-H. Chu reports a patent for IL15 variants and uses thereof (11059876) issued. V. de Ruiter reports grants from Pfizer Inc. during the conduct of the study. E. Kraynov reports other support from Pfizer Inc. during the conduct of the study. P.P. Boor reports grants from Pfizer Inc. during the conduct of the study. R. Feldman reports a patent for US20190263877 pending to Pfizer Inc. L. Mosyak reports a patent for US62/636,371 pending (USAPP Not Published IL15 variants and uses thereof) and a patent for US62/636,362 pending (USAPP Not Published IL15 variant and uses thereof). L. Lin reports a patent for IL15 variants and uses thereof pending. K.A. Ching reports other support from Pfizer Inc. during the conduct of the study, as well as other support from Pfizer Inc. outside the submitted work. J. Kwekkeboom reports grants from Pfizer Inc. during the conduct of the study. J. Chaparro-Riggers reports a patent for US62/636,371 pending and a patent for US62/636,362 pending. No disclosures were reported by the other authors.

Y. Xu: Data curation, formal analysis, writing–original draft. L. Campos Carrascosa: Data curation, formal analysis, writing–original draft. Y.A. Yeung: Conceptualization, data curation, supervision. M.L.-H. Chu: Data curation, methodology, writing–review and editing. W. Yang: Data curation, formal analysis. I. Djuretic: Conceptualization, supervision, methodology. D.C. Pappas: Data curation, methodology. J. Zeytounian: Data curation. Z. Ge: Resources. V. de Ruiter: Data curation. G.R. Starbeck-Miller: Writing–review and editing. J. Patterson: Writing–review and editing. D. Rigas: Data curation. S.-H. Chen: Data curation, writing–review and editing. E. Kraynov: Data curation, writing–review and editing. P.P. Boor: Data curation. L. Noordam: Resources. M. Doukas: Data curation. D. Tsao: Data curation. J.N. Ijzermans: Resources. J. Guo: Data curation. D.J. Grünhagen: Resources. J. Erdmann: Resources. J. Verheij: Resources. M.E. van Royen: Data curation. P.G. Doornebosch: Resources. R. Feldman: Conceptualization, supervision, investigation. T. Park: Data curation. S. Mahmoudi: Resources. M. Dorywalska: Resources. I. Ni: Resources. S.M. Chin: Data curation. T. Mistry: Data curation. L. Mosyak: Data curation. L. Lin: Resources. K.A. Ching: Supervision, methodology. K.C. Lindquist: Data curation, methodology. C. Ji: Methodology. L.M. Londono: Data curation. B. Kuang: Resources. R. Rickert: Supervision, writing–review and editing. J. Kwekkeboom: Conceptualization, supervision, writing–review and editing. D. Sprengers: Conceptualization, supervision, writing–review and editing. T.-H. Huang: Supervision, writing–review and editing. J. Chaparro-Riggers: Conceptualization, supervision, writing–review and editing.

The authors thank German Vergara and Teresa Radcliffe for supporting the animal studies. They thank Barbra Sasu, Samantha Bucktrout, and Will Somers for support of this project. The design, study conduct, and financial support for this research were provided by Pfizer Inc. Pfizer Inc. participated in the interpretation of data, review, and approval of the publication. Financial support for this research was provided by Pfizer Inc.

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.