Antibody Mediated Inhibition of HLA/LILR Interactions Breaks Innate Immune Tolerance and Induces Antitumor Immunity

Abstract

Immune check-point blockade for the treatment of malignancies has been focused on reversing inhibitory pathways in T lymphocytes. Natural killer (NK) cells are a potent innate defense against tumors and virally infected cells, but their therapeutic manipulation for anti-cancer immunity has been inadequately explored. Considerable attention has been focused on approaches to blocking inhibitory receptors on NK and myeloid cells. Most effort has been directed to the killer immunoglobulin-like receptors (KIR) and CD94/NKG2A on NK cells. Another set of receptors with similar function in both NK cells and myeloid cells is the leukocyte immunoglobulin like receptors (LILR) that interact with a wide variety of HLA molecules. Using pan-anti-HLA mAbs that recognize a conserved epitopic region on HLA also seen by LILR, we investigated their functional effects in several models of tumor immunity. The pan-anti-HLA-mAbs blocked the binding of most LILRs, did not block killer cell immunoglobulin-like receptors (KIR) or CD94/NKG2A/C or TCR recognition. They also activated dysfunctional NK cells explanted from a variety of human cancers, and resulted in enhancement of tumor immunity in humanized mice. The mAbs also exert direct anti-tumor effects. These results suggest that activation of innate immunity via disruption of HLA/LILR interactions is a potent approach for control of both primary tumors and potentially tumor metastases.

Introduction

NK cell mediated innate immunity has evolved as a rapid, first line of defense against multiple pathogens and tumors (1-4). NK cells recognize classical MHC-I (HLA-A, -B, -C in humans or H2-K, -D, -L in the mouse) via differentially expressed inhibitory and activating receptors such as members of KIR (CD158) and LILR (CD85), or of the murine Ly49, families (5, 6). A dynamic balance of inhibitory and stimulatory signals thus maintains immune homeostasis and regulates NK cell activation. Although both KIR and LILR have both activating and inhibitory allelomorphs, the inhibitory forms consistently dominate the resting, homeostatic situation. Complete loss of self-MHC-I expression on tumor cells (“missing self”) abrogates KIR/LILR or Ly49 dependent inhibition resulting in immediate NK cell activation and lysis of tumor cell (7). Structural studies have demonstrated that the murine Ly49 molecules interact with amino acid residues of the membrane proximal (α2 and α3) domains of the H2 heavy chain as well as with the β₂-microglobulin (β₂m) light chain but do not interfere with the TCR binding site (8). We previously demonstrated that global inhibition of Ly49/MHC-I interactions by administration of a pan-anti-MHC-I mAb, M1/42, to unmanipulated mice resulted in marked activation and proliferation of NK cells, myeloid cells, and T cells (9).

The dramatic in vivo effects of blocking H2/Ly49 interactions raised the possibility of translating these findings to humans by blocking HLA/KIR or HLA/LILR interactions. Crystal structures of HLA/KIR complexes reveal that the binding sites of human KIR on HLA antigens, overlap with the TCR binding site and would therefore likely inhibit CD8⁺ T cell activation and suppress the adaptive immune response (10). In contrast, as illustrated by crystal structures of HLA-A2/LILRB1, -HLA-F/LILRB1, and -HLA-G/LILRB1/LILRB2 complexes (11-14), the LILRs interact with conserved regions of the MHC-I membrane proximal α3 domain and the nonpolymorphic β₂m subunit and are therefore unlikely to interfere directly with T cell recognition which is focused on the α1 and α2 helices and the bound peptide. Finally, unlike KIRs which exclusively interact with HLA, LILRs can bind to non-HLA ligands offering additional possibilities for therapeutic manipulation (15).

The strong similarity between the binding interactions of H2/Ly49 and HLA/LILR raised the possibility that a pan-anti-HLA mAb having reactivity similar to the anti-mouse H2 mAb M1/42 would block the binding of the LILRBs to HLA, fail to block both the KIR and TCR binding sites on HLA, and result in activation of both the human innate and adaptive immune responses. In this report, we demonstrated that two previously well characterized anti-pan-HLA mAbs, W6/32 (16) and DX17 (17) met all three criteria. Crytallographic analysis as well as binding and functional studies with human PBMC (hPBMC) demonstrated that both DX17 Fab and W6/32 Fab blocked the interaction of multiple LILR multimers with HLA and resulted in enhanced NK cell proliferation and cytokine release. Blockade of HLA/LILR interaction in an ex vivo perfusion system and in humanized mouse tumor models revealed potent anti-tumor immunity. Overall, these results demonstrate that HLA/LILR interactions represent both an important pathway regulating innate immune system homeostasis and a potential target for check-point inhibition for the treatment of cancers in humans.

Materials & Methods

Mice

NSG and NSG mice expressing transgenic human IL-15 (NSG-IL-15) mice were purchased from Taconic or Jackson Laboratories (Taconic, Cat# 13683; Jackson Laboratories, Cat#030890) and housed under specific pathogen-free conditions. All mice were sex and age matched and used between 8 and 12 weeks of age. All animal protocols used in this study were approved by the National Institute of Allergy and Infectious Diseases Animal Care and Use Committee (Protocol # LISB 15E).

Cell Lines

All cell lines were maintained in RPMI supplemented with 10% FCS, 2 mM L-glutamine, 1X NEAA, 10 mM HEPES pH 7.2, 100 U/ml penicillin, 100 ug/ml streptomycin, and 50 ug/ml gentamicin in a humidified incubator at 37°C and 5% CO_2. The HeLa cell line was obtained from ATCC (CCL-2), the human pancreatic cell line KLM-1 was kindly provided by Dr Serguei Kozlov (National Cancer Institute, MD, USA)), Jurkat cells expressing the T4H2 T cell receptor specific for a gp100-derived peptide presented by HLA-A2 was a kind gift of Dr Brian Baker (University of Notre Dame, IN, USA), and DX17 hybridoma was a kind gift of Dr Lewis Lanier (University of California, San Francisco, CA, USA).

Isolation of hPBMC from healthy individuals and cancer patients

Human peripheral blood collected from 20 healthy volunteers from NIH Blood Bank or 25 cancer patients (adenocystic-2, cholangiocarcinoma-4, colorectal-4, gastric-1, gastrointestinal stromal tumors (GIST)-4, mesothelioma-2, neuroendocrine pancreatic-5, ovarian-1, small bowel adenocarcinoma-1, and metastatic melanoma-1) from the National Cancer Institute-Center for Cancer Research. Human PBMC were centrifuged at 2000 rpm for 15 minutes (acceleration-1, deceleration-1) over Ficoll-Hypaque to obtain total leukocytes. After removing RBC by ACK lysis buffer, a single cell suspension of leukocyte buffy coat layers was washed with FACS buffer and live cells were counted using a hemocytometer and trypan-blue exclusion staining. Written informed consent according to U.S. Common Rule was obtained as detailed by the Institutional Review Board of the National Institutes of Health (#NCT-01915225).

Tumor infiltrating lymphocytes (TIL) Isolation

TILs were obtained from 25 cancer patients (adenocystic-2, cholangiocarcinoma-4, colorectal-5, gastric-1, gastrointestinal stromal tumors (GIST)-4, mesothelioma-2, neuroendocrine pancreatic-5, ovarian-1, and small bowel ddenocarcinoma-1) undergoing surgical resection after written informed consent at the National Cancer Institute (NCI) Clinical Center according to U.S. Common Rule under protocol #13C0176, approved by the Institutional Review Board of the National Institutes of Health (NIH). The freshly harvested tumor biopsy samples obtained from the operating room ranging from 1-2 cm thickness and 1-2 cm diameter were used. Necrotic tumor lesions were excluded from these samples and then dissected into small pieces in complete RPMI medium (10 ml) and placed in Gentle MACS C-tubes. The tubes were then subjected to program 37C_H_TDK_1X3 on the Gentle MACS Octo Dissociator (Miltenyi Biotec). Red blood cells from the dissociated tissues were ACK-lysed. A single-cell suspension was obtained by passing the cells twice through a 70mm-pore size strainer (BD Falcon) and finally washed with FACS-staining buffer (PBS containing 10% FBS).

hPBMC culture with Fab or Fc-silenced anti-pan-MHC-I antibodies

hPBMC from healthy individuals or cancer patients was suspended in RPMI 1640 complete medium supplemented with 10% FBS, 2 mM L-glutamine, 1 mM sodium pyruvate, 1 mM HEPES, 0.1 mM nonessential amino acids, 50 mM 2-mercaptoethanol, and 100 U/ml penicillin and streptomycin. 10 μg of DX17, W6/32 Fab, DX17LALAPG or negative control hIgG1LALAPG was added to 0.5x10⁶ mononuclear cells in 500 μL medium in 48-well plates (Falcon) and incubated for 48 to 96 hours.

Flow cytometry

hPBMC from healthy donors or cancer patients and their corresponding TILs (n=25) were suspended in complete RPMI-1640 medium. For surface staining, the cells were initially blocked using FCX Trustain (Biolegend) for 5 minutes, followed by staining with the surface conjugate for 30 minutes in FACS staining medium (PBS, 10% heat-inactivated FBS, and 0.05% sodium azide). Intracellular Ki-67 staining was done by fixing and permeabilizing the surface-stained cells with Foxp3 staining buffer set (ThermoFisher) as per the manufacturer's guidelines and then staining with anti-Ki67. To perform intracellular IFNγ staining, PBMC in complete medium were stimulated with a cocktail of cell stimulation and protein transport inhibitors (Thermo Fisher) containing PMA, ionomycin, Brefeldin A, and monensin for 3-4 hours at 37°C. Cells were washed, fixed, permeabilized, and stained with anti-IFN-γ (BD Biosciences) for 1 hour at 4°C. Finally, the cells were washed and analyzed using BD LSR Fortessa X-20 cell analyzer and BD FACSDiva Software (version 8.0.) and data further analyzed by Flowjo^™ V10 (Treestar).

CellTrace Violet proliferation assay

hPBMCs were stained with CellTrace violet according to manufacturer’s instruction (ThermoFisher) and cultured (0.5X10⁶ cells in 500μl complete RPMI medium) for 48 hours with 10μg hIgG1LALAPG or DX17LALAPG mAb. After 48 hours Cell-Trace Violet (CTV) dilution was determined on live CD3⁻CD16⁺ NK cell by flowcytometry using BD Fortessa and analyzed by Flowjo^™ V10 (Treestar).

HLA-A2 GP100 tetramer staining

HLA-A2⁺T4H2 Jurkat cells (0.1 x 10⁶ cells) were incubated with 10 μg of anti-HLA-specific Fab, F(ab’)₂, or mAb (DX17 or W6/32) at 4°C. After 10 minutes, PE-labeled HLA-A2 GP100 or TAX tetramer (5 μg/sample) as a negative control was added to the antibody-conjugated single-cell suspension and incubated for another 20 minutes. After 30 minutes of total incubation time, the cells were washed with FACS staining buffer and analyzed using the BD LSR Fortessa X-20 cell analyzer and BD FACS Diva software version 8.0.

LILR-Fc, KIR-Fc, CD94/NKG2A-Fc and CD94/NKG2C-Fc protein staining

hPBMC (0.1 x 10⁶) or single HLA transfected cell lines were incubated with specific anti-HLA monoclonal antibodies (10μg/ml per sample) at 4°C. After 10 minutes, a multimeric staining reagent consisting of strepavidin-conjugated-biotinylated Fc-dimers specific to LILR, KIR, CD94/NKG2A and CD94/NKG2C (5 μg/sample; MedChem Express, R&D systems, & Acro Biosystems) was added to the single-cell suspension that had been conjugated with the respective antibody. The cells were then incubated for an additional 20 minutes. After a total incubation time of 30 minutes, the cells were washed with FACS staining buffer and analyzed using the BD LSR Fortessa X-20 cell analyzer and BD FACS Diva software version 8.0.

Antibodies

The VH and VL sequences (sequence available in PDB 8TQ5) encoding the DX17 antibody were identified by RT-PCR and sequencing (Creative Biolabs, Shirley, NY, USA). The H and L DNA sequences were optimized for human codon usage and subcloned into the EcoRI/BamHI sites of pCDNA3.1(−) for secreted expression. DNA encoding the mouse Fc segment of the DX17 H chain was replaced with human IgG1 Fc by subcloning the VH gene into pFUSE-CHIg-hG1 (Invivogen, San Diego, CA). This is designated DX17-huIgG1. To abolish FcgR binding, the mutations L234A, L235A, and P329G were introduced in the H chain constant region (18) by site-directed mutagenesis with QuikChange Lightning Multi Site-Directed Kit Agilent, Santa Clara, CA, USA) following manufacturer’s instructions. This Fc-silenced antibody is designated DX17LALAPG. For preparation of Fab fragments, a 6X HIS tag followed by a stop codon was inserted after the second cysteine in the IgG1 hinge region using the Q5 Site-Directed Mutagenesis Kit (New England Biolabs, Ipswich, MA, USA). All DNA constructs were verified by sequencing.

Plasmids encoding the DX17 H and L chains were mixed together at a 1:2 ratio, respectively, and transfected into Expi293F cells with Expifectamine following manufacturer’s instructions (ThermoFisher Gibco, USA). Culture medium was harvested 6 days later and purified either by affinity chromatography on Protein A Sepharose (Cytiva, Uppsala, Sweden) for the full antibody or by metal affinity chromatography on a TALON column (Takara, Mountain View, CA, USA) for Fab, followed by size exclusion chromatography on a Superdex200Increase column (Cytiva, Uppsala, Sweden).

Purified W6/32 was purchased from BioXcell (#BE0079). W6/32 Fab was prepared by adding 100 mg of W6/32 (BioXCell, Lebanon, NH, USA) in a volume of 9 ml to 1ml of washed immobilized papain agarose beads (ThermoScientific) in phosphate buffered saline pH 7.2 containing 20 mM cysteine and 10 mM EDTA. The mixture was incubated at 37 C for 5 hours in an Eppendorf ThermoMixer. The digest was dialyzed against 1 liter of PBS overnight and passed through a 10 ml bed volume Protein G (Cytiva) column to remove undigested antibody. The flow through was concentrated and purified by size exclusion chromatography on a Superdex75 HiLoad 16/600 column in PBS. Fab and F(ab’)₂ preparations of DX17 were prepared with the Pierce Mouse IgG1 Fab and F(ab’)₂ preparation kit according to manufacturer’s instructions. We did not measure endotoxin in studies using the Fab fragments of the antibodies. All studies with the DX17LALAPG reagent were performed with an endotoxin low (<2 EU/mg of protein) preparation of the mAb prepared by BioXcell.

Single MHC-I expressing HeLa transfectants

HeLa cells were grown in RPMI supplemented witb 10% FCS, 2mM L-glutamine, 1X NEAA, 10 mM HEPES pH 7.2, and 50 ug/ml gentamicin. Plasmid encoding full length HLA-A02:01 was obtained from Genecopoiea (Cat # EX-Z5803-M02-10) and inserted into the Sleeping Beauty vector pSBi-Pur (Addgene plasmid 60523 deposited by Eric Kowarz) by InFusion cloning (Takara, San Jose, CA, USA) ) at the SfiI site downstream of the EF-1-alpha promoter. To express other HLA allelomorphs, the extracellular domain of HLA-A02:01 from Gly1 – Trp274 (numbering is of the mature protein and excludes the first 24 residues which are from the signal peptide) was replaced with the corresponding extracellular domains of HLA-B44:05 (sequence as in PDB 1SYV), HLA-C03:04 (IMGT/HLA Accession HLA00413, gift of John Altman, NIAID Tetramer Facility), HLA-E (IMGT/HLA Accession HLA00934, synthesized by Genscript, NJ, USA), and HLA-G (IMGT/HLA Accession HLA00939, synthesized by Genscript, NJ, USA), by InFusion cloning (Takara, San Jose, CA, USA). MHC-I deficient HeLa cell line was generated using the Alt-R^™ CRISPR-Cas9 systm (IDT, Coralville, IA, USA) following manufacturer’s instructions. Two pre-designed gRNAs from IDT were used to target each HLA-A, HLA-B, and HLA-C loci for a total of 6 gRNAs in the transfection mix. The sequences of the 6 gRNAs are: for HLA-A, gRNA1: CGTAGCCCACGGCGATGAAG, gRNA2: GATAATGTATGGCTGCGACG; for HLA-B, gRNA1: CGCTGTCGAACCTCACGAAC, gRNA2: CATGACCAGTACGCCTACGA; for HLA-C, gRNA1: TATGACCAGTCCGCCTACGA, gRNA2: GATCACCCAGCGCAAGTTGG. One week following transfection cells were stained with W6/32 to confirm the presence of a W6/32 negative population. The culture was expanded and cells negative for W6/32 staining were FACS sorted and subsequently expanded. 1 μg of HLA-A2, B4405, E, and G cloned into the Sleeping Beauty plasmid as described above was co-transfected with 100 ng of the pSB100X plasmid encoding the Sleeping Beauty transposase (Addgene 34789, deposited by Izsvak Zsuzsanna) in 50% confluent cultures in a 6-well plate. 48 hours after transfection the cells were transferred to medium containing 10 μg/ml puromycin. Five days later cells were stained with W6/32 or allelomorph-specific antibodies to confirm expression. W6/32 positive cells were then FACS sorted.

Recombinant soluble MHC-I engineering.

Bacterial expression, denaturation, refolding, and purification of MHC-I were carried out essentially as described earlier (19). Briefly, HLA-A2:01 and HLA-B44:05 heavy chains, and human β₂-microglobulin (β₂-m) were separately expressed in E. coli as insoluble inclusion bodies following IPTG induction and dissolved in 6M guanidine-HCl (EMD Millipore #369080-1KG), 100 mM TRIS pH8, 2 mM EDTA, 0.1mM DTT. HLA heavy chain and β₂-m were renatured at 1:4 molar ratio in refolding buffer containing 0.4M arginine-HCl (Sigma-Aldrich #A5131-1KG), 5 mM reduced glutathione (Sigma-Aldrich #G4251-100G), 0.5 mM oxidized glutathione (Sigma-Aldrich #G4376-100G), and 10-fold molar excess of the influenza peptide GILGFVFTL for HLA-A2 or the HLA-DPα derived self-peptide EEFGRAFSF for HLA-B44:05 (peptides were syntheized by Biopeptek, Malvern, PA, USA). Following incubation for four days at 4 °C, the refolding mixture was dialyzed against 25 mM TRIS pH 8, 150 mM NaCl (TBS), concentrated, and purified by size-exclusion chromatography on Superdex 200 (Cytiva, Uppsala, Sweden) and ion-exchange chromatography on mono Q (Cytiva, Uppsala, Sweden).

Surface Plasmon Resonance (SPR)

SPR experiments were carried out on a BiaCore T200 (Cytiva, Uppsala, Sweden) at 25°C in 10 mM TRIS pH 7.4, 150 mM NaCl, 3 mM EDTA, and 0.05% Surfactant P20 at a flowrate of 30 mL/min. DX17-huIgG1 and DX17LALAPG were covalently coupled to the surface of a CM5 chip to 600 RU. Graded concentrations of MHC-I ranging from 31.2 nM to 1 mM were sequentially injected over the antibody surfaces interspersed with a regeneration cycle with 0.1M glycine pH 2.3 between each concentration. Binding experiments were repeated three times. Sensorgrams were globally fit to a 1:1 binding model with BiaCore T200 Evaluation Software 3.1 and plotted with Prism (GraphPad Software, San Diego, CA, USA).

Crystallization of the DX17/B*44:05 complex, data collection, and refinement

Purified HLA B*44:05 and DX17 Fab were mixed together at a 1:1.5 molar ratio at a final protein concentration of 10 mg/ml and incubated on ice for 2 hrs. Crystallization conditions were identified by screening hanging drops at 18 °C. Crystals of DX17/HLA B*44:05 were grown in 12% PEG 20,000, 0.1 M MES pH 6.5, cryoprotected in mother liquor containing 10% ethylene glycol and flash-frozen in liquid nitrogen. Diffraction data were collected at wavelength 1.033 Å at Southeast Regional Collaborative Access Team (SER-CAT) beamline 22ID at the Advanced Photon Source, Argonne National Laboratory. The data was processed with XDS to 2.3 Å resolution (see Supplementary Table S1). The structure was solved by molecular replacement with Phaser using HLA B*44:05-T73C (7TUC) as a search model. For the DX17 Fab search, we used a modified model of 4M7J (both heavy and light chains) with the CDR loops omitted as the initial search model. Molecular replacement solutions were readily found and were subjected to several rounds of refinement with Phenix interspersed with manual building in Coot. R_work/R_free (%) values for final refined model of DX17/HLA B*44:05 are 19.4/23.6 respectively. Data collection and refinement statistics are summarized in Supplementary Table S1. The structure is deposited in the PDB as 8TQ5.

Structure analysis

The surface area of the interaction between MHC-I and antibody was calculated with AreaImol ( as implemented in the CCP4 program suite, with the option “area differences due to ligand/subunit”, specifying either the whole antibody or individual H and L chains as ligands. The interface residues were identified using NCONT, with a 4 Å cutoff distance, as implemented in the CCP4 program suite (Supplementary Table S2). All structure figures were constructed in ChimeraX.

Ex vivo perfusion system

For ex vivo perfusion studies, only small tumor lesions ranging between 1mm-2mm in thickness and 2 mm to 5 mm in diameter were used. The tumor cores from large tumor lesions tend to be necrotic and, therefore, were excluded from the perfusion assay systems. Six different cancer patients' TILs (gastric cancer, colorectal cancer, mesothelial cancer, ovarian cancer, and small bowel adenocarcinoma) were analyzed independently by flow cytometry after hIgG1LALAPG and DX17LALAPG mAb incubations for 48 hours in ex vivo perfusion systems.

Tumor tissue from eligible patients was procured directly from the operating room at the beginning of each operation. Tumor bearing mesothelium or core needle biopsies were mounted loosely on the platform and secured in place with a silk suture. The tumor-bearing platforms were then immediately transferred into the perfusion system within a sterile incubator (Thermo Heracell VIOS 160i CO2 incubator) at 37° C and 5% CO2 for the duration of experimentation. The duration of the tissue preparation on to the platform was limited to an average of five minutes.

The perfusion system consisted of a chamber connected to an oxygenator (PermSelect silicone hollow fiber membrane module with 1000 cm² surface area) and a peristaltic pump (Ismatec ISM828 Reglo Analog Variable Speed Pump) using laboratory grade tubing (Ismatec Tygon LMT-55). The platform was 3D-printed with autoclavable photopolymer resin (Formlabs Form 2, Dental LT Clear resin). The Platform consisted of two rings (base and tissue-mount) connected with two posts. The base ring had a diameter of 18.6 mm and was designed to fit snuggly into holes of the chamber lid in order to suspend the tissue in the perfusate. The tissue-mount ring was a smaller ring of 9.56 mm with a 0.5 mm indentation for suture fixation. The two rings were connected with two 16.75 mm posts containing of four 1 mm holes to permit perfusate circulation. The chamber consisted of 3D-printed 60mm dish and lid from autoclavable photopolymer resin (Formlabs Form 2, Dental LT Clear resin). The lid was constructed with four 14 mm apertures to suspend the perfusion platforms. The chamber was designed with inflow and outflow ports positioned opposite of each other, at offset heights of 5 mm and 2 mm from the base of the dish, respectively. The system was optimized before introduction of the tissue with autologous plasma perfusate allowed to equilibrate to 37 degrees Celsius with optimal gas dissolved using humidified 93% O₂:7% CO₂ at 0.5-1.5 liters/minute. Physiologic parameters (pH, sodium, etc) were maintained using point of care instruments. To compensate for evaporative losses over time, continuous introduction of sterile water and insulin was infused at a rate of 0.25-0.75 mL/hour using a metered syringe injector. Experimental drugs were introduced into the perfusate as indicated.

scRNA-seq

Colorectal Cancer patient derived TILs isolated from the ex vivo perfusion system for 48 hours were enriched for CD45⁺ using REAlease CD45 (TIL) MicroBead Kit according to manufacturer’s protocol (Miltenyi Biotec). After removing dead cells with the Dead Cells Removal Kit (Miltenyi), the CD45⁺ TILs treated with oligonucleotide-tagged antibodies for cellular indexing of transcriptomes and epitopes by sequencing using Universal Total Cell Sequencing Kit according to manufacturer’s protocol. CD45⁺ TILs were Fc blocked (Human TruStain FcX, BioLegend) and labeled with oligonucleotide-tagged antibodies for cellular indexing of transcriptomes and epitopes by sequencing with TotalSeq-B Human Universal Cocktails B, v1.0.

Library preparation was performed using the Chromium NextGEM Single Cell 3’ Reagent Kits v3.1 (Dual Index) with Feature Barcode technology for Cell Surface Protein (10X Genomics), according to the manufacturer’s instructions. Briefly, antibody-labelled cells were resuspended in a Master Mix and were loaded onto a microfluidics chip together with partitioning oil and barcoded gel beads to generate the gel bead-in-emulsion (GEM). The 10X Genomics Chromium pressurized the chip compartments to deliver cells and gelbeads into a common microfluidics channel at a controlled rate, where they were isolated into emulsion droplets by oil such that only one cell and only one unique barcoded gelbead were isolated together in each oil droplet, called a GEM. The poly-A RNA from the cell lysate contained in every single GEM was retrotranscribed to cDNA, which contains an Ilumina R1 primer sequence, Unique Molecular Identifier (UMI) and the 10X Genomics Barcode. The pooled barcoded cDNA was then cleaned up with Silane DynaBeads, amplified by PCR and the appropriate sized fragments were selected with SPRI select reagent for subsequent library construction. The high molecular weight fraction containing the cDNA derived from poly-adenylated mRNA was used for 3’ Gene expression library preparation, while the low molecular weight fraction containing DNA derived from cell surface protein Feature Barcode was used for preparation of cell surface protein libraries. During the library construction, Ilumina R2 primer sequence, paired-end constructs with P5 and P7 sequences and a sample index were added. Final libraries were then sequenced on an Illumina NextSeq2000 Sequencing System using 2 NextSeq 2000 P3 Reagent Kits (100 Cycles) (Illumina) with the following cycles of sequencing chemistry: Read1=28 cycles; Index1=10 cycles; Index2=10 cycles; Read2=90 cycles.

scRNA-seq data analysis

The fastq files underwent 10x Genomics Cell Ranger count (version 7.1.0) for gene expression and antibody-derived tag transcript counts per cell. Seurat (version 4.3.0) was used to analyze the raw counts from Cell Ranger. In the hIgG1LALAPG TIL sample, cells with the number of features below 500 or above 6000 or with more than 38000 UMIs were filtered out. Similarly, in the DX17LALAPG TIL sample, cells with the number of features below 500 or above 7500 features or with more than 50000 UMIs were removed. In both samples cells that had a mitochondrial gene expression of greater than 10% were filtered out. The typical Seurat workflow with default parameters (unless otherwise specified) was then applied to all samples, including NormalizeData, FindVariableFeatures, ScaleData using all features, RunPCA, RunUMAP (using PCA dimensions 1-30), FindNeighbors (using PCA dimensions 1-30), and FindClusters (using resolution 0.1). Doublets were identified and removed using the scDblFinder (version 1.14) R package. All objects were integrated using the Seurat integration methods with default parameters, including SelectIntegrationFeatures, FindIntegrationAnchors, and IntegrateData.

The integrated object was then processed through NormalizeData, FindVariableFeatures, ScaleData, RunPCA, RunUMAP (using PCA dimensions 1-30), FindNeighbors (using PCA dimensions 1-30), and FindClusters (using resolution 0.1); default parameters were used unless specified. The SingleR (version 2.2.0) R package was used for cell typing, and the Novershtern eference dataset was accessed from the celldex (version 1.10) R package. Differential expression was done after sub-setting the integrated object to only hIgG1LALAPG TIL and DX17LALAPG TIL samples. Differential expression in TILs was done on immune cell types, including Monocytes, NK cells, memory CD4⁺ T cells, and memory CD8⁺ T cells, comparing DX17LALAPG to hIgG1LALAPG cells in gene expressions. The Seurat implementation of Wilcoxon Rank Sum test was used for differential expression. Mitochondrial, ribosomal, and mitochondrial ribosomal genes were filtered from the differential expression results. Single cell RNA data has been uploaded to the Geo Database (accession ID GSE262381).

Lymphocyte isolation

Spleen, liver and lungs were harvested from NSG or NSG-IL15 transgenic mice and homogenized the tissues using cell strainer (70 μm, BD Falcon USA). After homogenization Liver and lungs lymphocytes were isolated by 32% percoll mediated centrifugation. Red blood cells were lysed using sterile ACK lysing buffer [NH₄Cl (0.15 M), KHCO3 (10 mM) and Na₂EDTA(0.1 mM), pH7.3]. by removal of RBC and followed by Lymphocytes were washed, suspended in sterile complete medium [RPMI medium supplemented with 10% heat-inactivated fetal bovine serum (FBS), L-glutamine (2 mM), sodium pyruvate (1 mM), HEPES (1 mM), non-essential amino acids (0.1 mM), 2-mercaptoethanol (50 μM), and penicillin and streptomycin (100 U/ml)].

Humanized mouse tumor model

KLM-1 cancer cells were grown in DMEM medium supplemented with 10% heat-inactivated FBS, L-glutamine (2mM), sodium pyruvate (1mM), HEPES (10mM), non-essential amino acids (0.1mM), 2-mercaptoethanol (50μM), and penicillin and streptomycin (100U/ml). Sex- and age-matched hIL15 transgenic mice were injected subcutaneously (right flank region) with 1.5 x 10⁵ KLM-1 cells suspended in 100 mL of sterile PBS. After 12-14 days of tumor cell engraftment, 10 x 10⁶ CD3⁻ hPBMC, isolated from healthy individuals by negative selection on magnetic beads (Miltenyi Biotec), were retro-orbitally injected into the tumor-bearing mice. Isotype control and Fab or Fc silenced anti-pan-HLA (W6/32 and DX17) mAb were administered retro-orbitally (i.v.) or intraperitonially (i.p.) twice a week for two weeks. Tumor size was measured in three dimensions regularly using digital calipers, and tumor volumes were calculated by multiplying length x width x depth. On day 30 post tumor implant, tumors were excised under sterile conditions, and TILs were prepared after mincing the tumor into tiny pieces and further dissociated by Gentle MACS tissue Dissociator (Program:m_impTumor_03).The dissociated tissue was passed through a 70mm strainer (BD Falcon) and then treated with ACK lysing buffer to lyse RBC. The single-cell suspension of TILs was washed with FACS staining buffer and purified by 30% percoll density gradient centrifugation.

Ingenuity Pathway Analysis

Differentially expressed genes of NK cells, monocytes, and memory CD8 harvested from ex vivo perfusion systems derived TILs were imported into the Ingenuity Pathways Analysis (IPA) software (Ingenuity Systems; Qiagen et al., USA www.ingenuity.com/) for functional canonical pathway and upstream regulatory analysis. The canonical pathway was adjusted using Fisher’s exact t-test with a default p-value of >0.05 or a threshold of -log(p-value)>1.3 for significance. A threshold of -log(P-value) >2 was used for canonical pathway, and a Z-score >2 was defined as the threshold of significant activation. Conversely, the threshold of significant inhibition was defined as a Z-score <−2. The canonical pathways generated from differentially expressed genes of immune cell subsets were sorted by positive (orange) to negative (blue) Z scores.

Statistical Analysis

Statistical comparisons between groups were analyzed using a two-tailed unpaired t-test or one-way ANOVA with Prism 10 (GraphPad Software). A p-value of less than 0.05 was considered statistically significant.

Data availability

Macromolecular biologic structural data has been deposited in wwPDB. Single cell RNA data has been uploaded to the Geo Database (accession ID GSE262381). ll other materials, data, protocols, resources, and reagents described in this paper (Supplementary Table S3) are available upon request to the corresponding author, Ethan M. Shevach.

Results

HLA interactions with LILR but not with KIR, CD94/NKG2A/C, and TCR are blocked by DX17 and W6/32 mAbs

We focused our investigations on two pan-anti-human mAbs, DX17 and W6/32 (20, 21). Reasoning that their broad reactivity is due to recognition of the most conserved features of HLA molecules, i.e. the β₂m and α3 domains, we asked which HLA-reactive innate immune receptors recognize β₂m and α3 and are thus likely to be blocked by DX17 and W6/32. The LILR family of receptors satisfied these criteria (6). We examined the ability of DX17 and W6/32 to block the binding of recombinant LILR-Fc multimers to cell surface HLA on monocytes. (The gating strategies for lymphocytes and monocytes of healthy hPBMC are shown in Supplementary Fig. S1A and S1B). DX17 and W6/32 inhibited binding of all LILRAs and LILRBs, except LILRB4, to cell surface HLA (Fig. 1A). LILRA4 was not tested. To further confirm the staining and blocking results, we employed HeLa cell lines expressing single HLA alleles, designated HLA*A2:01(designated A02:01), B44:05, C03:04, E01:01, and G01:01 and examined the binding of LILRB1 and LILRB2. LILRB1 (Supplementary Fig. S2A) and LILRB2 (Supplementary Fig. S2B) bound to the single MHC-I transfectants, and their binding was blocked by DX17 but not by the B1.23.2 mAb which recognizes an epitope distinct from that of DX17 and W6/32 on HLA-B/C (20, 21). Binding of LILRB1 or LILRB2 to the E01:01 transfectant could not be detected (Supplementary Fig. S2C). DX17 and W6/32 cross block each other completely but they do not block another anti-HLA*A02 mAb, BB7.2 (Supplementary Fig. S2D).

Figure 1. DX17 and W6/32 block binding of LILRs, but not KIRs, CD94/NKG2A/C and TCR, to MHC-I.

(A) Staining of monocytes by individual biotin-labeled LILR-Fc constructs after preincubation of the cells with the indicated anti-HLA antibodies. (B) Staining of monocytes by individual biotin-labeled KIR2D-Fc constructs after preincubation of the cells with the indicated anti-HLA antibodies. (C) Staining of PBMC monocytes by individual unlabeled KIR3DL1-Fc constructs after preincubation of the cells with the indicated anti-HLA antibodies. All samples were stained with anti-Fc-APC. (D) Staining of PBMC monocytes or HLA-E01:01/HeLa by biotin-labeled CD94/NKG2A and CD94/NKG2C after preincubation of the cells with the indicated anti-HLA antibodies. (E) Staining of T4H2 Jurkat cells with the PE-labeled gp100/HLA-A2 tetramer or negative control tetramer (anti-TAX) following preincubation of the cells with the indicated anti-HLA antibodies.

We then examined the ability of DX17 to block HLA binding of three inhibitory KIR2D isoforms and one KIR3D isoform on monocytes (Fig. 1B and C) and to the B44:05 and C03:04 transfectant. B1.23 blocked KIR binding (21) but DX17 did not inhibit KIR binding (Supplementary Fig. S2E and S2F). DX17 or B1.23.2 also failed to block the binding of CD94/NKG2A and CD94/NKG2C to their cognate ligand, HLA-E, on both monocytes and to the transfected E:01:01 line (Fig.1D). Finally, we examined whether DX17 would block TCR recognition of cognate peptide/MHC-I complexes using a Jurkat line transfected with a TCR (T4H2) specific for a gp100 peptide presented by HLA-A2 (22). Staining of this Jurkat line by the HLA-A2gp100 tetramer was not blocked by DX17 and W6/32 (Fig.1E). Blocking was observed with an HLA-A2 specific antibody, clone BB7.2 (23).

In contrast to Ly49 antigens in the mouse and KIRs in humans, LILRs have been reported to have a much broader tissue distribution (6). We performed a comprehensive study of the expression of LILRs on freshly isolated healthy hPBMC (Supplementary Fig. S1C and S1D) using the indicted gating strategy (Supplementary Fig. S2A and S1B). Monocytes (CD3⁻CD14⁺) uniformly expressed all members of the family with the exception of LILRA4 which we did not detect with the available mAb. Varying percentages of B cells (CD3⁻CD19⁺) expressed LILRB1, B5, and A3. Among NK cells (CD16⁺CD56^dim), ~50% expressed LILRB1, while LILRA3 could be detected on ~20% of NK cells. Approximately 25% of CD8⁺ T cells express LILRB1 with little or no expression of the other LILRs. Thus, monocytes express the full array of LILRB and LILRA molecules, while T cells, NK cells, and B cells reflect varied surface display.

Structural basis for the pan-anti-MHC-I reactivity and blocking of LILRs by DX17 and W6/32

We investigated the binding of DX17 and W6/32 to HLA by surface plasmon resonance (SPR) and by X-ray crystallography. DX17 and W6/32 have similar binding kinetics and near identical contact surfaces on the HLA. Both antibodies bind to HLA with similar affinities (K_D 30-50 nM) and kinetic on and off rates (Fig. 2, A and B). DX17 and W6/32 prefer both a human MHC-I heavy chain and human β₂m for binding (Supplementary Fig. S3A and S3B) indicating that their epitopes encompass both subunits. Reduced binding of HLA molecules complexed with murine β₂m was observed as was a detectable interaction with some mouse H2 molecules when complexed with human β₂m (24)(Supplementary Fig. S3A and S3B).

Figure 2. Structure of the DX17/B44:05 complex.

Kinetics and affinity of the interaction of HLA-A2:01 and HLA-B44:05 with the MAbs DX17 (A) and W6/32 (B) as measured by SPR. The data are expressed as mean ± standard deviation of three independent experiments. (C) Overall structure of the DX17/B44:05 complex shown as ribbon diagram. (D) Footprint of DX17 on the B44:05 heavy chain. The heavy chain (purple) and β₂m (yellow) are displayed as surfaces. Residues contacting the DX17 VH are shown in cyan and those contacting the DX17 VL are shown in salmon color. Residues in white are contacted by both VH and VL of DX17. The contacting residues on B44:05 are indicated. (E) Footprint of DX17 on the β₂m subunit of B44:05, colored as in (D). (F) Structure based sequence alignment of the heavy chains of selected MHC-I molecules indicating residues that interact with DX17, W6/32 (from PDB 7T0L), LILRB1 (from PDB 1P7Q), and LILRB2 (from PDB 2DYP). The key to the symbols is shown adjacent to the figure. The MHC-I sequences are the extracellular regions of HLA-A2:01 (IMGT/HLA Accession HLA00006), HLA-B44:05 (IMGT/HLA Accession HLA00322), HLA-C05:01 (IMGT/HLA Accession HLA00427), HLA-E (IMGT/HLA Accession HLA00934), HLA-F (IMGT/HLA Accession HLA01096), HLA-G (IMGT/HLA Accession HLA00939), and H2-D^d (PDB ID 1DDH). (G) Structure based sequence alignment of the mature forms of human (Uniprot P61769) and mouse β₂m (Uniprot P01887) with the interacting residues depicted as in (E).

We determined the crystal structure of the DX17 Fab/ B44:05 complex to 2.3 Å resolution and compared it to the crystal structure of the W6/32 Fab/B27:05 complex previously reported (25, PDB 7T0L). Data collection and refinement statistics are shown in Supplementary Table S1 and a list of intermolecular contacts in Supplementary Table S2. As seen in the overall structure of the complex (Fig. 2C), DX17 engages B44:05 beneath the peptide binding platform and interacts with conserved residues on both the MHC heavy chain and β₂m subunits. The interaction with DX17 buries an area of 1,159 Ų on MHC-I, over half of which, 635 Ų, is on the B44:05 heavy chain and the remaining 524 Ų on β₂m. Contacts to the B44:05 heavy chain include residues Tyr85 and Gln87 in the α1 domain, Gly121-Asp123 and Ala136-Thr138 in the α2 domain, and residues Thr225-Glu233 in a strand of the α3 domain that includes the CD8-binding Thr225-Asp227 loop (Fig. 2D and E). Of the 17 contacts on the HLA heavy chain, 13 are conserved in HLA-A, -B, -C, -E, -F, and -G (Fig. 2F) explaining the broad reactivity of DX17 for both classical and nonclassical HLA molecules. DX17 contacts β₂m amino terminal residues Ile1, Gln2, Arg3, Thr4, and Lys6, and residues Lys58-Ser61 in a loop that abuts the floor of the MHC-I platform domain (Fig. 2D and E). The footprint of W6/32 on B27:05 is remarkably similar to that of DX17 on B44:05 (Supplementary Fig. S3C-S3E), although W6/32 buries a slightly larger area on MHC-I (1,363 Ų vs 1,159 Ų). W6/32 shares all contacts with DX17 on the HLA heavy chain with the addition of the conserved Asn86 in the α1 domain, the partially conserved Lys223, and the conserved Lys243 in the α3 domain (Fig. 2F). Similarly, W6/32 shares all contacts with DX17 on β₂m, with additional contacts to Ser56 and Thr86 (Fig. 2G). Superposition of the DX17 and W6/32 MHC-I complexes demonstrates the strong similarities in their docking modes (Supplementary Fig. S3F).

The structures of DX17 and W6/32 complexes with MHC-I also provide insight into the steric clashes that prevent simultaneous engagement by both LILR and antibody. The footprint of LILRBs on HLA has been previously illuminated by the structures of the D1/D2 domains of LILRB1 in complex with HLA-A2 (1P7Q¹⁰, 4NO0²⁵, 6EWA, 6EWC, 6EWO²⁶) and HLA-F (5KNM²⁷), and of the D1-D4 domains of LILRB1 with HLA-G1 (6AEE¹²). Similar to DX17 and W6/32, the LILRs focus primarily on the MHC α3 domain and β₂m. However, because the LILRs approach MHC-I at a different angle than DX17 and W6/32, their interaction with α3 is directed toward the membrane proximal 193-200 loop, a site that is not seen by either DX17 or W6/32. Blocking of LILRs by DX17 and W6/32 is thus largely due to steric hindrance by their VH/VL focused on the amino-terminal residues 1-4 of β_2-m where the footprints of LILRs and the antibodies coincide. Competition in favor of the antibodies is assured by their significantly higher affinities for MHC-I (K_D 30-50 nM, (Fig. 2A and B)) compared to LILR affinities for MHC-I which are only in the micromolar range (12).

mAb mediated blocking of HLA/LILR interactions induces activation and proliferation of human NK cells in vitro

To determine if inhibition of HLA/LILR interactions by DX17 and W6/32 leads to immune cell activation in vitro, we cultured PBMCs for four days in the presence of intact DX17, DX17 Fab, or W6/32 Fab. (The gating strategy on the four-day cultures is shown in Supplementary Fig. S4A-E). Ninety to 95% of the cells were viable after four days of culture with or without antibody treatment. Inhibition of HLA/LILR interactions by DX17 Fab and W6/32 Fab resulted in enhancement of CD16 expression, while NKp46 receptor expression on NK cells (CD16⁺NKp46⁺) was maintained in the presence or absence of DX17 Fab and W6/32 Fab (Fig. 3A). DX17 Fab and W6/32 Fab also markedly enhanced NK cell proliferation as measured by increased Ki-67 staining (Fig. 3B) and activation as indicated by expression of IFNγ (Fig. 3C) and IL-15Rα (Fig. 3D). It is therefore possible that treatment with DX17 might induce a cytokine storm in vivo. No enhancement was seen with full length DX17. KIR expression increased in the Fab-treated cultures (Fig. 3E and F). NK cells cultured in medium alone lost the expression of LILRB1 (Fig. 3E and F), while cells cultured with DX17 Fab or W6/32 Fab maintained the expression of LILRB1 at levels observed on freshly explanted NK cells (Supplementary Fig. S1). Monocytes did not proliferate under any conditions tested (Supplementary Fig. S5A), upregulated IL-15Rα expression on treatment with DX17 Fab, or W6/32 Fab (Supplementary Fig. S5B), and maintained LILRB1 expression with or without the addition of the antibodies (Supplementary Fig. S5C). No proliferation of CD4⁺ or CD8⁺ T cells was observed during culture with DX17 Fab (Supplementary Fig. S5D and S5E).

Figure 3. DX17 markedly activates NK cells and monocytes in the hPBMC.

Representative flow cytometry analysis of the expression of NKp46 (A), Ki-67 (B), IFNγ (C), IL-15Rα (D), KIR2DL1/L2(E, F) and LILRB1(E, F) on CD16⁺CD56^dim NK cells following incubation of healthy donor hPBMC for 96 hours with the indicated antibodies. The measurement of Ki-67 and IL-15Rα expression are done in the absence of PMA/ionomycin while IFNγ production requires treatment with PMA/ionomycin. The percentage of CD16⁺CD56^dim, CD16⁺NKp46⁺ NK cells (G) and CD16⁻CD56⁺ and NKp46⁻LILRB1⁺ NK cells (H) in hPBMC and TILs from patients with the indicated different types of cancer is shown. (I) Representative example of the expression of NKp46 on NK cells from PBMC from a patient with metastatic melanoma cultured for 48 hours in vitro in the presence of DX17LALAPG or control hIgG1LALAPG. (J) Summary of the percentage of NKp46⁺ NK cells from patients with five different cancers cultured for 48 hours in vitro in the presence of DX17LALAPG or control hIgG1LALAPG. (K) Representative example of the expression of IL15Rα on CD14⁺ monocytes from PBMC from a patient with metastatic melanoma cultured for 48 hours in vitro in the presence of DX17LALAPG or control hIgG1LALAPG. (L) Summary of the percentage of CD14⁺IL15Rα⁺ monocytes from patients with five different cancers cultured for 48 hours in vitro in the presence of DX17LALAPG or control hIgG1LALAPG. (M) Representative example of the expression of CD86 on CD14⁺monocytes from PBMC from a patient with colorectal cancer cultured for 48 hours in vitro in the presence of DX17LALAPG or control hIgG1LALAPG. (N) Summary of the percentage of CD14⁺CD86⁺ Monocytes, from patients with three different cancers cultured for 48 hours in vitro in the presence of DX17LALAPG or control hIgG1LALAPG. (O) Representative example of the expression of CD163 on CD14⁺monocytes from PBMC from a patient with colorectal cancer cultured for 48 hours in vitro in the presence of DX17LALAPG or control hIgG1LALAPG. (P) Summary of the percentage of CD14⁺CD163⁺ Monocytes, from patients with three different cancers cultured for 48 hours in vitro in the presence of DX17LALAPG or control hIgG1LALAPG.

It seemed likely that the failure of the intact DX17 mAb to induce NK cell proliferation (Fig. 3A) and IL-15Rα upregulation on NK cells (Fig. 3C) and monocytes (Supplementary Fig. S5B) was due to the interaction of the Fc region of the mAb with inhibitory Fc receptors on myeloid cells as these assays were performed with unseparated hPBMC. In contrast to W6/32 Fab and DX17 Fab, both intact W6/32 and DX17 inhibited the binding of anti-CD16 to NK cells and anti-CD32 to monocytes (Supplementary Fig. S5F and S5G). To abrogate interactions with Fc receptors, we first mutated W6/32 and DX17 Fc residues Leu234 and Leu235 to alanine (W6/32LALA and DX17LALA), and also generated double mutants (Pro329Gly, W6/32LALAPG and DX17LALAPG) (26). While W6/32LALA and DX17LALA partially blocked the binding of CD16 to NK cells and CD32 to monocytes, both W6/32LALAPG and DX17LALAPG did not inhbit anti-FcR binding. The DX17LALAPG mutation did not alter the binding properties of the mAb as DX17LALAPG cross blocked the binding of W6/32 to lymphocytes (Supplementary Fig. S6A), displayed similar affinities for A02:01 and B44:05 as DX17 (Supplementary Fig. S6B and S6C), induced NK cell proliferation in 3-day PBMC cultures (Supplementary Fig. S6D and S6E), enhanced NK cell production of IFNγ, and enhanced expression of both LILRB1 and IL-15Rα (Supplementary Figs S6F-H).

Inhibition of HLA/LILR interactions enhances expression of the natural cytoxicity receptor NKp46 on NK cells and TILs from multiple cancer patients

Human peripheral blood NK cells can be divided into two subsets based on the expression of CD16 and CD56 (27, 28). Mature KIR⁺ NK cells (80-90%, CD16⁺CD56^dim) play critical roles in tumor immunity, whereas the immature KIR^dim/− NK cell subset (5-15%, CD16⁻CD56^bright) plays a role in immunoregulation. Both subsets express multiple natural cytotoxicity receptors (NCRs) including NKp46, NKp30, NKG2D and DNAM1 which recognize diverse stress-induced ligands or altered proteoglycans expressed on cancer cells (29, 30) Several studies have suggested that the number of NK cells in the peripheral blood of patients with malignancies is lower than normal and that their NK cells express fewer NCRs (30, 31). We therefore examined NK cell frequency and NCR expression on both the peripheral blood and TILs of patients with advanced cancers. We first excluded monocytes based on cell size and gated on CD3⁻CD19⁻ lymphocytes (Supplementary Fig. S7) and quantitated the percentages of live CD16⁺CD56^dim cells, CD16⁺CD56^dimNKp46⁺cells, CD16⁻CD56⁺, and NKp46⁻LILRB1⁺ cells in freshly isolated PBMCs and corresponding tumor biopsy samples (Fig. 3G and H). In PBMCs, modest decreases were found in the percentages of CD16⁺CD56^dim and CD16⁺CD56^dimNKp46⁺ populations accompanied by modest increases in the CD16⁻CD56^bright cells and NKp46⁻CD56^brightLILRB1⁺ cells. Among TILs, CD16⁺CD56^dim cells and CD16⁺CD56^dimNKp46⁺ cells were almost undetectable, while CD16⁻CD56^bright cells and NKp46⁻LILRB1⁺ cells were markedly increased (Fig. 3H).

We then explored the effects of DX17LALAPG on NKp46 expression by PBMC in five cancer patients. The gating strategies for lymphocytes and monocytes from a a patient with metastatic melanoma cancer are shown in Supplementary Fig. S7). In contrast to the results obtained with healthy donor PBMC where NKp46 expression was maintained in control cultures (Fig. 3A), NKp46 expression was lost when PBMC from cancer patients were cultured in the absence of stimulation (Fig. 3I). Addition of DX17LALAPG treatment during the culture restorated NKp46 expression on 95% - 99% of NK. Similar results were seen when we cultured hPBMC from patients with other cancers in the presence of DX17LALAPG (Fig. 3J). DX17LALAPG also upregulated IL-15Rα and CD86 expression, and downregulated CD163 expression on monocytes from patients with different cancers (Fig. 3K-P).

To determine whether DX17 LALAPG was capable of activating lymphocytes within the tumor microenvironment (TME), we utilized an oxygenated ex vivo perfusion circuit (32) that maintains explanted tumors in near-physiologic conditions with the use of autologous patient plasma (Supplementary Fig. S8A). Fresh intact tumor tissue from an advanced colorectal cancer patient was perfused with control hIgG1LALAPG or DX17LALAPG (5μg/ml) for 48 hours. Following perfusion, the tumor explants were removed and dissociated into single cell suspensions. To confirm that DX17LALAPG could penetrate the tumor sample, we stained the cell suspension with the cross-blocking W6/32 mAb. (Gating strategy is shown in Supplementary Fig. S9). Marked reduction of W6/32 staining was observed on NK cells (and T cells) derived from DX17LALAPG treated tumors indicating that the antibody had penetrated the intact tumor sample and bound to MHC-I (Supplementary Fig. S8B). Similar to what was observed in PBMC cultures from cancer patients, DX17LALAPG markedly upregulated NKp46 expression on NK cells, upregulated IL-15Rα and CD86 expression, and downregulated CD163 expression on monocytes derived from the explanted tumor tissue from five different advanced cancer patients (Fig. 4A-H).

Figure 4. DX17 markedly avctivates NK cells and monocytes in TILs.

(A) The expression of NKp46 on NK cells in the indicated TIL subpopulations from a patient with colorectal cancer following culture for 48 hrs in the ex vivo perfusion system in the presence of DX17LALAPG (5μg/ml perfusate) or control human IgG1LALAPG antibody. (B) Summary of the expression of NKp46 on NK cells in TILs from patients with five various cancers cultured for 48 hours in the ex vivo perfusion system in the presence of DX17LALAPG or control hIgG1LALAPG. (C) Representative example of the expression of IL15Rα on CD14⁺monocytes from TIL from a patient with colorectal cancer cultured for 48 hours in vitro in the presence of DX17LALAPG or control hIgG1LALAPG. (D) Summary of the expression of IL-15Rα on monocytes in TILs from patients with various cancers cultured for 48 hours in the ex vivo perfusion system in the presence of DX17LALAPG or control hIgG1LALAPG. (E) Representative example of the expression of CD86 on CD14⁺monocytes from TILs from a patient with colorectal cancer cultured for 48 hours in the ex vivo perfusion system in the presence of DX17LALAPG or control hIgG1LALAPG. (F) Summary of the percentage of CD14⁺CD86⁺ monocytes from patients with colorectal and gastric cancer cultured in the ex vivo perfusion system for 48 hours in vitro in the presence of DX17LALAPG or control hIgG1LALAPG. (G) Representative example of the expression of CD163 on CD14⁺monocytes from TILs from a patient with colorectal cancer cultured for 48 hours in the ex vivo perfusion system with DX17LALAPG or control hIgG1LALAPG. (H) Summary of the percentage of CD14⁺CD163⁺monocytes from patients with colorectal and gastric cancers cultured for 48 hours in the ex vivo perfusion system with DX17LALAPG or control hIgG1LALAPG.

Molecular analysis of leukocyte activation of TME cells ex vivo with HLA/LILR blockade

A primary colorectal carcinoma biopsy sample was placed in the ex vivo perfusion system and cultured for 48 hours in the presence of DX17LALAPG (5μg/ml) or control hIgG1LALAPG. The biopsy was then dissociated and human CD45⁺ TILs were purified from both cultures by magnetic cell sorting and processed using single-cell RNA-seq. Although more cells were recovered from the DX17LALAPG sample (13,800) than the hIgG1LALAPG sample (7892), the relative cellular compositions were similar (Fig. 5A). Monocytes accounted for ~20-25% of the total cells, while the frequencies of dendritic cells (1%) and B cells (~3.5%) were low. Predominant among the T lymphocytes were memory CD4⁺ (41-43%) and memory CD8⁺ (18-25%) subsets compared to naive CD4⁺ (4-6%) and naive CD8⁺ (2-3%) T cells. Very few NK cells (<1%) were recovered from either group explored by UMAP analysis (Fig. 5B). We then performed a detailed differential gene expression and pathway analysis of monocytes, memory CD4⁺ and CD8⁺ T lymphocytes, and NK cell clusters on a single-cell basis as well as by volcano plot and heat map analysis (Fig. 5C-F, Supplementary Fig. S10 and S11).

Figure 5. Differential gene expression analysis of monocytes and memory CD8⁺ T cells from TILs of a single colorectal cancer patient after culture in the ex vivo perfusion system in the presence of DX17LALAPG or control hIgG1.

(A) Pie chart representation of the percentages of the indicated cell subsets as determined by scRNA-seq. (B) UMAP analysis depicted clusters of immune cell subsets. (C) Heat map analysis of differentially expressed genes in monocytes from scRNAseq data. (D) Volcano plot analysis of differentially expressed genes in monocytes. (E) Heat map analysis of differentially expressed genes in memory CD8⁺ T cells from scRNAseq data. (F) Volcano plot analysis of differentially expressed genes in memory CD8⁺ T cells.

Treatment of cultures with DX17LALAPG enhanced the expression of innate and inflammatory receptors (CLEC5A, S100A8/9), cytokines, and chemokines (IL1b, IL8, CCL3, CCL4, CXCL7, CXCL8, MMP9, etc.), as well as a complement gene, C3, on monocytes. In contrast, immunosuppressive molecules TREM2, APOE, EGR1, and GPNMB were downregulated (Fig. 5C and D, Supplementary Fig. S10A). Pathway analysis of differentially expressed genes demonstrated that MHC Class I mediated antigen processing pathway, IL-8, and IL-12 signaling pathways were upregulated, while the alternative activation signaling pathway and IL-4/IL-13 signaling pathways were down-regulated after DX17LALAPG treatment (Supplementary Fig. S10B).

Marked changes in gene expression by memory CD8⁺ T lymphocytes were also induced by culture in the presence of DX17LALAPG (Fig. 5E and F, Supplementary Fig. S10C) including enhancement of genes encoding Th1 type cytokine (IFNG), cytotoxic granules (GZMA/B/H/K, PERF1), chemokines (CCL4, CCL5, CXCL5), and transcription factors (RUNX3, Eomes, KLRD1) that play critical roles in sustained long term effector or memory CD8⁺ T cells responses. Pathway analysis of activated memory CD8⁺ T lymphocytes indicated upregulated TCR and costimulation signaling, IFNγ signaling, Th1 pathway, DAP12 interactions, IFN α/β and CGAS-STING signaling pathway, and down-regulated Th2 pathway, PD1/PD-L1- and CTLA4-dependent checkpoint inhibition signaling (Supplementary Fig. S10D). Despite the paucity of NK cells in either sample, the addition of DX17LALAPG to the ex vivo perfusion system enhanced NK cell expression of chemokines (CCL3/4, CXCL3/5, IL8), transcription factors (ETS2), and kinases involved in pAKT-mTOR signaling (PDPK1, IPPK, Raptor) (Supplementary Fig. S10 E and S10F). Pathway analysis of differentially expressed genes revealed that DX17LALAPG significantly enhanced multiple pro-inflammatory cytokine pathways (IL-1/−7/−8/−9/−10/−12 as well as IL-15), PI3K/pAKT, mTOR, JAK/STAT and ERK/MAPK signaling and down-regulated PTEN, RIPK1-mediated necrosis, and apoptosis signaling (Supplementary Fig. S11A). Only modest cytokine and chemokine gene expression activation by memory CD4⁺ T cells was observed (Supplementary Fig. S11 B-D). Thus, blockade of HLA/LILR interactions in the TME as reflected in the ex vivo perfusion system culture resulted not only in marked activation of the innate immune response but also a dramatic increase in the adaptive anti-tumor response, particularly in memory CD8⁺ T lymphocytes

DX17LALAPG augments NK and T cell activation in vivo in humanized mice

To assess the potential of HLA/LILR blockade to induce cellular activation in vivo, we reconstituted NSG mice that lack mouse T, B, and NK cells with hPBMC followed by treatment with W6/32 Fab (250μg/mouse, iv) (Fig. 6A). (Gating strategy is shown in Supplemtary Fig. S12A-C). W6/32 Fab markedly enhanced human NK and T cell proliferation reflected by elevated Ki-67 staining of lymphocytes from both spleen, liver and lung (Fig. 6B-D).

Figure 6. Administration of W6/32 Fab to NSG or NSG-IL15 animals enhances human NK and T cell proliferation in lymphoid and non-lymphoid organs.

Outline of the experimental protocol for engraftment of NSG mice with PBMC followed by antibody treatment (A). Flow cytometric expression of Ki-67 in NK and T cell populations from spleen (B), lung (C) and liver (D) on day 6. Statistical analysis (n= 3) is shown on the right. Outline of experimental protocol for engraftment of NSG-IL15 mice with NK-enriched PBMC followed by antibody treatment (E). Expression of Ki-67 in NK and T cell populations from spleen (F), lung (G) and liver (H) on day 16. Statistical analysis (n= 3) is shown on the right.

W6/32 Fab-mediated immune cell activation was also explored in human IL-15 transgenic NSG animals that support engraftment and long-term survival of human NK cells (33). Human NK cell-enriched (T cell-depleted) PBMC were transferred into IL-15 transgenic NSG mice (Fig. 6E). W6/32 Fab was then administered and mice were euthanized 48 hours later. A marked enhancement of NK cell proliferation in spleen, liver and lung was observed (Fig. 6F-H). The engrafted population contained a small percentage of CD4⁺ and CD8⁺ T cells and an enhancement of their proliferation was also observed. Thus, W6/32 Fab treatment induces proliferation of human NK and T cells in both lymphoid and non-lymphoid organs in NSG mice.

Pan-anti-MHC-I triggers tumor immunity in vivo in humanized mice

To determine whether HLA/LILR blockade by anti-MHC-I would also augment antitumor immunity, we used the KRAS and p53 mutated human pancreatic tumor cell line, KLM-1 (34) which expresses all classical and non-classical HLA class I antigens, and multiple NCR specific ligands (Calr [35], MICA/B, ULBP, FcγRII (CD32), LILRBs and LILRAs (Supplementary Fig. S13A-C). IL-15 transgenic NSG mice were injected with KLM-1 tumor cells, engrafted with hPBMC on day 14, and three days after reconstitution were treated with DX17LALAPG or control hIgG1LALAPG (Fig. 7A). Tumor growth was markedly reduced in the DX17LALAPG treated mice (Fig. 7B and C). Similar results were seen when mice were injected with KLM-1 cells and treated with W6/32 Fab (Supplementary Fig. S13D-E). NK cells isolated from the TIL of treated mice expressed higher levels of Ki-67 and NKp46 (Fig. 7D). (Gating strategy is shown in Supplementary Fig. S12D).

Figure 7. DX17 LALAPG augments antitumor immunity in NSG-IL-15 mice.

(A) Outline of the experimental protocol for the establishment of the KLM-1 pancreatic cell line in NSG-IL15 mice followed by engraftment with CD3⁻ PBMC (Day 14), administration of the antibodies (100μg DX17LALAPG or isotype control mAb, i.v.; every 3-4 days intervals, from day 17 to day 28) and analyses of TILs on day 30. (B) Growth kinetics of KLM-1 tumor and (C) tumor sizes in DX17 LALAPG or control hIgG1 LALAPG treated mice. (D) Percentage of CD16⁺NKp46⁺ NK cells and expression of Ki-67 on CD16⁺ NK cells in TILs on day 30 in hIgG1 LALAPG and DX17 LALAPG treated mice. (E) Outline of the experimental protocol for treatment of the KLM-1 cells in NSG-IL-15 mice engrafted or non-engrafted with CD3⁻ PBMC. (F) Comparison of KLM-1 growth kinetics and (G) tumor size in CD3⁻ PBMC grafted versus ungrafted NSG-IL15 mice. (H) Selected LILRA and (I) LILRB expression on freshly harvested Epcam⁺ cancer cells isolated from biopsies of the indicated six different cancers.

Several studies have suggested that expression of members of the LILRB, but not LILRA, family on tumors positively correlated with tumor progression (36-40). Analysis of an colorectal and pancreatic tumor data set also indicated that multiple LILRs are expressed by colorectal and pancreaatic carcinomas at the mRNA level (Supplementary Fig. S14). As KLM-1 cells express multiple LILRs it remains possible that anti-MHC-I mediated inhibition of tumor growth might not only result from blocking HLA/LILRB interaction between tumor cells and effector cells, but may be secondary to blockade of cis or trans interactions of LILRs with HLA on the tumor cells themselves. We repeated the studies described above in IL-15 transgenic mice that had not been reconstituted with hPBMC. In this setting significant inhibition of tumor growth was also seen in mice treated with DX17LALAPG (Supplementary Fig. S13F and S13G). We then compared directly the anti-tumor effects of DX17LALAPG treatment on tumor bearing mice that had been reconstituted with hPBMC or non-reconstituted controls (Fig. 7E-G). While DX17 LALAPG alone inhibited tumor growth to the same extent as mice reconstituted with CD3-negative cells alone, much greater inhibition of growth was observed in mice reconstituted with CD3-negative hPBMC and treated with DX17LALAPG (Fig. 7F and G). Lastly, these studies raised the possibility that primary tumors or tumor metastases might also express members of the LILR family. We readily detected the expression of multiple LILRA and LILRB members on Epcam⁺ tumor cells of diverse tumor origin (Fig. 7H and I).

Discussion

Immune check point blockade has revolutionized the treatment of many cancers (41). However, only a small number of inhibitory pathways have been identified, and blocking these pathways has proven to be useful in a minority of patients. The field has been focused on reversing the inhibitory and exhaustion pathways manifest by anti-tumor effector T cells. Over 100 cell surface receptors with inhibitory (ITIM)-containing motifs have been identified (42). Many of these inhibitory receptor families are expressed by cells of the innate immune system including NK cells and myeloid cells. Here, we demonstrate that members of the LILR family play important roles in the control of the innate response against human tumors and function as a new class of checkpoint inhibitors.

Previous studies in a mouse model demonstrated that an pan-anti-H2 mAb which inhibited the binding of members of the Ly49 antigen family to their target H2 ligand resulted in a potent anti-tumor response initiated by NK cell activation, and subsequent activation of the adaptive immune response (9). While KIRs in humans play a role similar to the murine Ly49 antigens and bind HLA, the binding sites on HLA of many of the KIR family members overlap with the TCR footprint and blocking KIR/HLA interactions could potentially inhibit T cell function (10). We initially identified two pan-anti-HLA mAbs which blocked HLA/LILR interactions, did not block HLA/TCR interactions, and blocked the binding of dimeric constructs of LILRB1, B2, B3, B5, but not LILRB4, and the binding of LILRA1, A2, A3, A5, and A6 (A4 not tested) to all classical and some non-classical HLA molecules. Crystallographic structures confirmed that both DX17 and W6/32 have overlapping binding sites and that their footprints overlap with the binding site of LILRB1/B2 on HLA. While HLA/LILR interactions were blocked by DX17 and W6/32, these mAbs had no effect on the binding of several KIRs or CD94/NKG2A/NKG2C. In general, MHC-I molecules display a number of different faces that are seen by their ligands. TCRs engage their complementarity determining regions with the membrane distal α1α2 peptide structure to recognize MHC-I bound peptides (43). The CD94/NKG2A (inhibitory) and the CD94/NKG2C (activating) heterodimers bind p/α1α2 of HLA-E similarly (44). KIR2 and KIR3 molecules bind the carboylterminal region of α1α2 peptide, in a MHC-dependent and peptide influenced manner (45), while LILR focus on the membrane proximal α3/β₂m domain unit (46). Among each of these MHC-I surfaces, the pan-anti-HLA mAb, DX17 and W6/32 only compete with the LILR and the CD8 sites, leaving potential recognition by TCR, KIR, or CD94/NKG2A/NKG2C available.

The addition of Fab fragments of both mAbs (DX17 or W6/32) to cultures of PBMC in the absence of other stimuli resullted in NK cell proliferation, enhanced capacity to produce IFN-γ, and upregulated expression of IL-15Rα. While monocytes did not proliferate in culture, enhanced expression of the IL-15Rα and CD86 were observed as well as downregulation of CD163. We speculaate that the failure of intact DX17 or W6/32 to induce NK cell activation was likely due to engagement of their Fc regions with an inhibitory FcR on myeloid cells which resulted in inhibition IL-15Rα upregulation since Fab fragments and a chimeric version of DX17 containing human IgG1 with a disabled Fc region were stimulatory. The capacity of these mAbs to activate human NK cells and T cells in vivo was demonstrated by treatment of immunodeficient NSG mice that had been reconstituted with hPBMC.

NK cells, both in the peripheral blood of patients with advanced cancers and in the TME, are dysfunctional and a decreased percentage of NK cells in patients with tumors has been correlated with tumor progression and poor patient survival (30). Dysfunctionality is characterized not only by a decrease in number but also with decreased proliferation, production of IFNγ and TNFα, as well as impaired cytotoxic activity. Most importantly, in many cancers, NK cells express low levels of NCRs including NKp46, NKp30, CD16, and others (47). We have confirmed these findings and extended them by demonstrating that NK cells in the TME of several solid tumors have an abnormal phenotype characterized by the presence of a high percentage of CD16⁻CD56⁺NKp46⁻ (so-called immature) NK cells which also express high levels of LILRBs. Culture of hPBMC in vitro or TILs derived from perfusion of intact tumors ex vivo in the presence of DX17 resulted in increased expression of NKp46 on NK cells. NK cells and monocytes from a primary colon carcinoma perfused ex vivo in the presence of DX17 upregulated a pro-inflammatory gene expression profile and memory CD8⁺ T cells in the TILs and monocytes demonstrated marked upregulation of a Th1-like response and downregulation of Th2-like responses and checkpoint inhibitory pathways.

When we inoculated NSG mice with KLM-1 cells, a human pancreatic tumor cell line, and then reconstituted the mice with PBMC, mice treated with DX17 had a marked reduction of tumor growth, accompanied by NK cell proliferation and NKp46 expression in the TIL. KLM-1 cells expressed many members of the LILR family raising the possibility that the therapeutic effects of HLA/LILR blockade were due to direct effects on the tumor cells. Marked reduction of tumor sizes was observed when we treated NSG mice that had been inoculated with the KLM-1 cells but not reconstituted with PBMC. We then directly compared treatment of tumor inoculated mice that had or had not been reconstituted with PBMC. While some reduction of tumor size was seen in the non-reconstituted cohort, the greatest degree of reduction of tumor sizes was seen in the PBMC reconstituted group.

Multiple studies have demonstrated expression of LILRB family members on tumor cells but not adjacent tissue (48, 49). In our studies, expression of both LILRB and LILRA members was readily demonstrable by flow cytometry on several freshly explanted solid tumors of diverse origin. As mAb treatment alone can enhance anti-tumor effects, these results raise the possibility that interaction of LILRs with HLA either on the tumor cells themselves (in cis) or with various lymphoid cells (in trans) contributes to tumorigenicity (50). The possible tumor promoting effects of engagement of either LILRB or LILRA alone or together is an important area for future investigation.

Several clinical studies have been reported targeting inhibitory receptors on NK cells including KIRs, NKG2A, and LILRs. One study used IPH2101 (Lirilumab) which blocks interactions between KIR2DL-1, −2, −3 and HLA. Unfortunately, phase II studies with this mAb in several cancers did not demonstrate efficacy (51). Other studies have utilized Monalizumab, a humanized mAb to NKG2A, an inhibitory receptor distinct from KIRs and the LILRB family, which blocks the binding of NKG2A to its ligand HLA-E and prolongs progression-free survival of patients with non-small cell lung cancer when given together with anti-PD-L1 (52). Several early clinical studies have used antagonistic mAbs to LILRB1, LILRB2, either alone or as bispecific reagents (52-55) or as components of chimeric antigen receptor (CAR) T cells (56), but clinical effectiveness has not yet been reported. While reversing the LILRB-mediated inhibitory signals on NK cells is the proposed primary mechanism of action, one study has shown that blocking LILRB1 engagement on macrophages with either anti-LILRB1 or anti-HLA enhanced the capacity of macrophages to kill some tumor cell lines in an Fc-dependent manner in the presence of an anti-CD47 mAb (57). However, in contrast to our data, this study did not observe enhanced killing with anti-LILRB1 or anti-HLA alone in vitro or in vivo. Recent studies (58-60) have demonstrated that antagonistic mAbs to LILRB1 and LILRB2 can reprogram macrophages in the TME from an immunosuppressive M2 phenotype to a stimulatory M1 phenotype. In one study (59), an anti-LILRB2 mAb appeared to function in the presence of cis or trans engagement similar to what we observed in our experiments treating mice that had not been reconstituted with human PBMC.

One major problem with blocking one or several members of an inhibitory receptor family is the presence of unblocked members of the same family. For example, Lirilumab blocks KIR2DL-1-2-3, but does not block KIR3DL family members. Our results suggest that blocking both the engagement of LILRB and LILRA families by a mAb to their common ligand, HLA, potentially represents the most efficient method of neutralizing the inhibitory effects of members of this pathway on innate immune cells in the TME as well as the potential tumor promoting function of LILR signaling in the tumor cells themselves. This approach implies that LILRB signaling dominates over LILRA and that blocking releases the dominant effect of LILRB. Selective targeting of individual LILRBs may potentially result in altered signaling (either enhancement or suppression) by LILRA family members. Complete neutralization of LILR signaling results not only in potent activation of NK cells but appears to generate a downstream pathway involving monocyte/macrophage activation, IL-15 production, and activation of memory T lymphocytes. Our studies raise the possibility of the potential therapeutic use of pan-anti-HLA monoclonal antibodies as checkpoint inhibitors in the treatment of localized tumors and tumor metastases (61). In addition, blocking HLA/LILR interactions may also have a role in the treatment of chronic infectious diseases, and as an adjuvant for augmentation of responses to weak vaccines (9).

SYNOPSIS.

The authors report that blocking interactions between Leukocyte Ig-like Receptors (LILRs), which are expressed on human NK and myeloid cells, and HLA using a monoclonal antibody to HLA induces NK and myeloid cell activation and antitumor immunity.