Female NOD-SCID interleukin (IL)2Rγ-null (NSG) mice, aged 6–8 weeks (weight about 20 g), were purchased from the animal core facility of UT Southwestern. Mice were kept in a specific pathogen free room with a 12 hours light/dark cycle, controlled room temperature and ab libitum food and water. Mice were randomly allocated to each treatment group for experiments.

Expi293F (Cat#A14528) was obtained from Life Technologies (Carlsbad). Hematological cancer cell lines 697, MHH-CALL-2, and OPM2 were purchased from DSMZ (Braunschweig, Germany). KMS27, KMS26, KMS12PE and KMS20 were purchased from Health Sciences Research Resources Bank, Japan Health Sciences Foundation. LILRB1 reporter cells were described previously.18 All other cell lines were purchased from ATCC except as noted. Hematological cancer cell lines were maintained in Roswell Park Memorial Institute (RPMI) 1640 supplemented with 10% heat-inactivated fetal bovine serum (FBS) (Sigma Aldrich) (R10). Solid tumor cell lines were maintained in Dulbecco's Modified Eagle Medium (DMEM) with 10% heat-inactivated FBS except for H460 and H1299， which were maintained in R10. NKL cells were cultured as previously described.19 All cell culture medium were supplemented with 1% penicillin and streptomycin.

Peripheral blood mononuclear cells (PBMC) were separated from the buffy coats of healthy donors (Interstate Blood Bank) by gradient centrifugation using Ficoll media (GE Lifesciences). To isolate LILRB1 positive NK cells, PBMC were incubated with antihuman CD56 microbeads (Miltenyi Biotech) and separated using an AutoMACS Pro Separation System. Isolated CD56⁺ cells were then stained with anti-CD3-PE (clone: HIT3a, BioLegend)), anti-CD56-FITC (clone: TULY56, eBioscience) and anti-LILRB1-APC (clone: HP-F1, eBioscience) or Mouse IgG1 kappa Isotype Control-APC (eBioscience). LILRB1⁺ NK cells (CD56⁺CD3^-) were sorted using a fluorescence-activated cell sorting (FACSAria) I system. Sorted LILRB1⁺ NK cells were maintained in the same medium as NKL cells for 2–3 days. Cancer patient samples were obtained from UT Southwestern Medical Center (UTSW) Tissue Bank and Hematologic Malignancies Tissue Bank of UTSW. Patient NK cells were isolated using the same protocol except that AutoMACS-separated NK cells were cultured, activated, and used for cytotoxicity assays without FACS.

Two New Zealand white rabbits were immunized subcutaneously (sc.) with 0.5 mg recombinant human LILRB1 protein (untagged ECD, Sino Biological). Four boosters were given after the initial immunization in a 3-week interval. Serum anti-LILRB1 titers were evaluated by indirect ELISA. Single LILRB1 positive memory B cells were cultured in 96-well plates for 14 days (RevMAb Biosciences USA). The culture supernatants were analyzed by ELISA for antibody binding to LILRB1 protein (untagged ECD, Sino Biological). Antibody variable region genes were cloned from those positive B cells using methods described previously.20

Full length IgG was produced in human embryonic kidney Expi293F cells using an ExpiFectamine 293 Transfection Kit (Catalog #: A14528, Gibco) by cotransfection HEK293 cells with both heavy and light chain expression constructs.20 After 7 days of fed-batch culture, the culture supernatants were harvested and antibodies were purified by affinity chromatography using protein A resin (Catalog #: 10–2001-XM, Repligen) as described previously.20

Classical sandwich epitope binning experiments were performed on an 8-channel Octet RED96 System as described.20 First, a baseline was established in kinetics buffer (Cat#18–1105, ForteBio) for 3 min. Then antibodies (30 µg/mL) were captured by protein A biosensors (Cat#18–5010, ForteBio) for 4 min. The remaining Fc-binding sites on the biosensors were then blocked with an irrelevant rabbit antibody (200 µg/mL) for 4 min, followed by soaking the biosensors in kinetics buffer for 10 s. The biosensors were exposed to recombinant LILRB1 (25 µg/mL) for 4 min, and then exposed to the secondary antibodies (40 µg/mL). The surfaces were regenerated for 45 s in 100 mM glycine (pH 2.6). The antibody pairs were assessed for competitive binding. Additional binding signal observed by the secondary antibody indicates an unoccupied epitope (non-competitor). A lack of additional binding by the secondary antibody indicates epitope blocking (competitor).

For antibody affinity measurement, antibodies (analyte, 30 µg/mL) were loaded onto the protein G biosensors (Cat#18–5082, ForteBio) for 4 min. Following a short baseline in kinetics buffer for 20 s, the loaded biosensors were exposed to solutions of recombinant LILRB1 in a range of concentrations (0.1–133 nM). After background was subtracted, data were fit to a 1:1 binding model to extract an association rate (k_on) and dissociation rate (k_off). The KD was calculated using the ratio k_off/k_on. Kinetic constant ranges for Octet RED96 systems is between 1 mM to 10 pM. Thus, affinity higher than 10 pM will be shown as KD <10 pM. All experiments were performed with shaking at 1000 rpm. All raw data were processed using ForteBio Octet Data Analysis Software V.9.0.

Ig-like C2-type domains, which including D1, D2, D3, D4 of LILRB1, were defined as described previously.21 The motif between D4 and transmembrane domain is considered as the stalk region or attached region. Different mutants of LILRB1 were generated using QuikChange Lightning Site-Directed Mutagenesis Kits (Agilent, Cat #210519) with LILRB1 wild type DNA construct as the template. LILRBs ECD, LILRB1 mutants and different domains were cloned into the Fc fusions expression vectors in frame with an antibody heavy chain signal sequence and the Fc portion of human IgG₁ (hIgG₁). Expression, purification and quantification of the Fc fusion proteins are the same as those for the antibodies described above.

LILRB1 recombinant proteins (50 ng/well, 100 µL per well) were coated on Corning 96-well ELISA plates at 37°C for 4 hours. Plates were blocked for 2 hours at 37°C with 5% non-fat milk. After washing with phosphate buffered saline with tween (PBST), 100 µL of serial diluted anti-LILRB1 antibodies were added and incubated for 60 min at 37°C. Subsequently, the plates were washed with PBST for five times and incubated for 30 min with anti-rabbit or anti-human F(ab')2 HRP-conjugated antibody (Jackson ImmunoResearch, Cat #111-036-003 and #109-036-003). Plates were washed again with PBST for another five times, then the immunoreactions were developed with 3,3',5,5'-Tetramethylbenzidine (TMB) substrates (Sigma, Cat #T0440) and stopped by the addition of 2 M H₂SO₄ before the plate was read at 450 nm.

Anti-LILRB1 rabbit antibodies were humanized by CDR-grafting as described previously.20 Briefly, CDRs in the heavy and light chains of the rabbit antibody were defined by a combination of three methods: Kabat, IMGT and Paratome. The parental rabbit mAb and the most homologous human germline sequences were aligned and residues that are known not to be structurally critical or subjected to change during the in vivo maturation process were identified in the mutational lineage guided analysis and humanized. DNA sequences encoding humanized light chain variable (VL) and heavy chain variable (VH) were synthesized (GENEWIZ) and a hIgG signal peptide and a Kozak sequence were engineered at the 5′ ends of the VL and VH sequences.22 The humanized VL and VH fragments were then cloned into expression vector in fusion with human constant regions for mammalian expression.

HLA-G-1 cDNA with a signal peptide mutant23 was cloned into pLentiLox3.7-PuroR plasmid. Lentivirus plasmids pLentiLox3.7-luciferase-PuroR, pLentiLox3.7-HLA-G-PuroR and pLVX-MICA-ZsGreen plasmids were used to overexpress target proteins in cell lines as described previously.24 The infected cells were enriched by selection with 1 µg/mL puromycin for luciferase or HLA-G overexpression. MICA-positive cells were enriched by FACS for three times (clone: 6D4, BioLegend). LILRB2-5 reporter cells and LILRA1-6 reporter cells were made by retrovirus infection, according to previous reported.18 25

Carboxyfluorescein succinimidyl ester (CFSE, Thermo Fisher Scientific) labeling of target cells combined with propidium iodide (PI) staining of dead cells is a reliable method to measure in vitro cell killing by flow cytometry.26–28 Briefly, NK cells were cocultured with CFSE-labeled leukemia cells in U-bottom 96-well plates for 4 hours. Then, each sample was mixed with PI and analyzed by FACS Calibur. Cell lysis was calculated by the percentage of PI-positive leukemia cells among total leukemia cells. Spontaneous cancer cell death, in the absence of NK cells, was less than 5% and subtracted from total killing in the presence of NK cells.

A total of 5×10⁴ NKL cells were cocultured with 5×10⁴ cancer cells in U-bottom 96-well plates for 24 hours. Interferon-γ (IFN-γ) release was detected in culture supernatants by ELISA (BioLegend) following the manual provided by the vendor.

A total of 5×10⁶ (CFSE)-labeled 697 (NKL resistant) and 697-MICA (NKL sensitive) cells were mixed and injected in combination with 5×10⁷ NKL into NSG mice intraperitoneally. Anti-LILRB1 antibody (10 mg/kg) or control hIgG was administrated retro-orbitally. After 24 hours, cells from mice peritoneal cavities were harvested and stained with anti-MICA antibody and analyzed by FACS Calibur1 (BD Biosciences). In vivo cytotoxic activities of NKL were calculated as:

NK=Ratio of 697-MICA/697 in mice receiving 697, 697-MICA cells and NKL cells.

CN=Ratio of 697-MICA/697 in mice receiving 697 and 697-MICA cells.

% in vivo cytotoxic activity= (1- NK /CN)*100.

In vivo cytotoxic activity of NKL cells against 697-MICA was also tested in NSG mice sc. A total of 1×10⁶ luciferase-expressing 697-MICA (697 MICA-luci) cell was mixed with 5×10⁶ NKL cells and injected into mice sc. Anti-LILRB1 antibody (10 mg/kg) or control hIgG was administrated to each mouse retro-orbitally. Bioluminescence imaging (BLI) was conducted 48 hours later to monitor the remaining 697-MICA cells in mice.

Age 6–8 weeks NSG mice were sublethally irradiated with 200 cGy X-ray on day −1. On day 0, each mouse was given 5×10⁵ KMS27-luci together with 5×10⁶ NKL cells via tail vein injection. Anti-LILRB1 antibody (10 mg/kg) or control hIgG was administrated retro-orbitally on day 0, day 3 and day 7, then once a week for 1 month. Another 5×10⁶ NKL cells were injected on day 14. A total of 10 000 IU human IL2 were administrated to each mouse through IP injection every other day. Tumor burden was measured by BLI on day 28 and day 35.

Data are presented as mean±SE. Statistical significance between two groups is calculated by two-tailed unpaired t-test except as noted. Kaplan-Meier survival curves were analyzed using a log-rank test. Differences are considered statistically significant if p<0.05.

We have characterized the function of immune inhibitory receptors of the LILRB family and immune activating receptor NKG2D in cancer development.8 19 20 24 25 29–31 To investigate if LILRB1 can be a molecular target for cancer immunotherapy, we determined LILRB1 expression on NK cell from patients with cancer in comparison with that from health donors. Results showed that LILRB1 is mainly expressed on CD56^dim NK cells, rather than CD56^bright NK cells, from both healthy donors and patients with cancer (figure 1A). We then evaluated the percentage of LILRB1 expressing NK cells among CD56^dim NK cells from peripheral blood of patients with prostate cancer or MM. We found that percentages of LILRB1⁺ NK cells (among CD56^dim NK) were significantly higher in blood from patients with late-stage prostate cancer (3b and 3 c) than in blood from healthy donors (figure 1B, online supplementary table 1). Moreover, NK cells in peripheral blood from patients with MM who had persistent disease while on treatment had greater percentage of LILRB1⁺ NK cells (among CD56^dim NK) than that from health donors or patients with minimal disease to complete response (figure 1C, online supplementary table 2). These results suggest that the percentage of LILRB1⁺ NK cells is significantly higher in patients with late-stage cancer or poor prognosis, and LILRB1 may be a molecular target for immunotherapy for these patients.

To evaluate therapeutic potential of anti-LILRB1 mAbs in activating NK cells, we generated a panel of rabbit anti-LILRB1 mAbs by screening single memory B clones, cloning antibody genes and evaluating the blocking activity of recombinant mAbs (online supplementary figure S1A,B).32–34 In total, 229 anti-LILRB1 specific single B cell clones were generated from 384 LILRB1 B cell culture wells and assessed by ELISA (online supplementary table 3). Among them, we identified 174 single B cell clones can produce mAbs binding to LILRB1 expressed on cell surface of LILRB1 NFAT-green fluorescent protein (GFP) reporter cell (online supplementary table 4), which express GFP when the cell surface LILRB1 is cross-linked.18 We selected 60 binders for further cloning and expression, finding 44 antibodies with EC50 of low nanomolar range (0.05–3 nM) (online supplementary figure S1C). To group the 44 mAbs by their binding epitopes, classic sandwich epitope binning assay was performed by using Octet RED96, and twelve epitope bins were identified for these 44 high binders (online supplementary figure S1D).

To screen for antagonistic anti-LILRB1 mAbs, LILRB1 receptor cells were cocultured with K562 cells overexpressing HLA-G1 (K562-HLA-G), the MHC class I molecule with the highest affinity for LILRB1 (figure 2A). The LILRB1 reporter cells were activated by K562-HLA-G cells, but not by parental K562 cells (MHC class I negative) (figure 2A). We screened all 44 antibodies using this reporter assay. Antibodies in bin 1, bin 7, bin 8 and bin 9 showed antagonistic effects (online supplementary figure S1E). Antibodies from bin 1 showed the strongest blocking efficacy (figure 2B). B1-176 in bin 1 showed specific binding to LILRB1 as assessed by flow cytometry (figure 2C), ELISA (figure 2D) and Octet RED96 (online supplementary figure 1F). B1-176 showed high affinity to LILRB1 with a KD value of about 10 pM (figure 2E), significantly higher than that of two commercial anti-LILRB1 mAbs HP-F1 and GHI/75 (online supplementary figure 1G). Therefore, we chose B1-176 to further evaluate its binding epitope and function in modulation of NK cell activation.

To understand mechanism of B1-176 in blocking LILRB1 activation, we determined its binding epitope. B1-176 showed specific binding to the D1D2 region of LILRB1 (online supplementary figure 2A) not to the other domains, as assessed by ELISA (figure 2F). We made a series of amino acid mutations in the D1D2 region of LILRB1 to determine key amino acids for B1-176 binding (online supplementary figure 2B). Two amino acids (R84 and Y76) were found to contribute to LILRB1 interacting with B1-176 (figure 2G). Molecular docking of B1-176 was performed using Discovery Studio 2017. Based on the PDB structure of LILRB1 D1D2 region (PDBID:1UFU) and the modeling structure of B1-176 using the two key amino acids R84 and Y76, we identified the most possible binding model. In this model, the long HCDR3 of B1-176 inserts into the narrow ligand-binding pocket on LILRB1 D1 domain. Arg-84 of LILRB1 may interact with Tyr-102, Glu-103 and Asp-109 from HCDR3 loop by hydrogen bonds and electrostatic interactions. Tyr-76 of LILRB1 may be involved in hydrogen bonds with Asp-104 from HCDR3 (figure 2H).

To minimize the immunogenicity of B1-176 for potential clinical applications, we humanized B1-176 and conducted molecular engineering for optimization of the antibody sequences. The Kabat/IMGR/Paratome combined-CDRs in the heavy and light chain of B1-176 were identified as described previously.20 The CDR sequences here are showed as heavy chain complementarity-determining regions (HCDR) and light chain CDR (LCDR). Human germline IGHV3-53*04 and IGKV1-9*01 were selected as the backbones. Heavy chain Hu-176 VH-1 and light chain Hu-176 VL were generated from the CDR grafting (figure 3A). However, antibodies with humanized heavy chain sequence VH-1 showed reduced binding to LILRB1 (figure 3B). After multiple rounds of single back-mutations in VH-1, mutation evaluation showed that amino acid L48 in FR2 was important to restore the binding activity (figure 3A,B). Molecular docking showed that Fabs of humanized and rabbit B1-176 shared similar superposition structures (figure 3C).

Considering the wide expression of LILRB1 on normal immune cells, rabbit B1-176 was expressed as hIgG₁ with N297A mutation (B1-176-N297A) and humanized B1-176 was expressed as hIgG₁ with N297A mutation (Hu B1-176-N297A) or with L234A, L235A, and P329G mutations (Hu B1-176-LALAPG) to abolish Fc mediated immune effector function.35 36 All B1-176 with Fc mutants maintained strong affinity to LILRB1 (figure 3D).

We screened for cancer cell lines that could activate LILRB1 by coculturing cancer cells with LILRB1 reporter cells (figure 4A). Flow cytometry was used to assess the expression of MHC class I molecules (antibody clone:HP-1F7, Santa Cruz), the ligands of LILRB1, on the cell surface of hematological and solid cancer cell lines (figure 4A). We found that the majority of cancer cell lines expressed MHC class I molecules and stimulated the activation of LILRB1 reporter cells. Exceptions were K562, DLD1 and Daudi cells, which did not express MHC class I molecules on cell surface. These results suggest that the ability of cancer cells to activate the LILRB1 reporter cells is positively correlated with their MHC class I expression levels. B1-176 and Hu B1-176-N297A blocked the activation of LILRB1 reporter cells stimulated by hematological cancer cell lines (figure 4B) or solid tumor cell lines (figure 4C). These results suggest that rabbit and humanized versions of B1-176 can block LILRB1 activation induced by HLA-G and other MHC class I molecules. It is noteworthy that LILRB1 binds to the conserved alpha3 domain of MHC class I.37 This may explain why our antagonistic anti-LILRB1 antibody screened from K562-HLA-G cells can block LILRB1 activation by other cancer cell lines.

We assessed the in vitro efficacy of anti-LILRB1 mAbs using the NK cell line NKL. NKL cells express high levels of LILRB1 on surface (online supplementary figure 3A). Treatment with different versions of anti-LILRB1 mAb: B1-176-N297A, Hu B1-176-N297A, and Hu B1-176-LALAPG, increased the cytotoxic activities of NKL cells against the MM cell line KMS27 at similar levels (figure 5A). Anti-LILRB1 antibody also increased the cytotoxic activity of NKL against other MM cell lines OPM2 and RPMI8226, and the T cell leukemia line Jurkat (figure 5B). By flow cytometry, we found that LILRB1 was expressed on the cell surface of certain leukemia and lymphoma cell lines, such as 697 and Raji cells, but not on MM cell lines, Jurkat cells, or the majority of malignant plasma cells from a patient with MM (online supplementary figure 3B-D). Since the N297A mutation or LALAPG on Fc would abolish ADCC function of the antibody, our results indicate that the antagonistic anti-LILRB1 antibodies increased the natural cytotoxic activity of NKL cells against MM cell lines. Interestingly, we found that pre-B leukemia cell line 697 and Burkitt lymphoma cell line Raji were resistant to NKL cytotoxicity even when treated with anti-LILRB1 antibody (figure 5C). A previous study reported that MHC class I chain-related gene AB (MICA/B), a ligand for the NK cell activating receptor NKG2D, was expressed on NK-sensitive acute lymphoblastic leukemia (ALL) cells but was absent on NK-resistant ALL cells.38 To sensitize the 697 cells and Raji, we overexpressed MICA on 697 and Raji cells. As expected, MICA on the cancer cell surface stimulated the cytotoxic activities of NKL cells against these cells and anti-LILRB1 antibody further increased the cytotoxic activities of NKL cells (figure 5C,D). These results suggest that LILRB1 blockade and NKG2D activation on NKL cells synergistically increase their cytotoxic activity. Besides hematologic cancer cells, anti-LILRB1 antibody stimulated NKL cells to kill solid tumor cancer cell lines, such as breast cancer cell line MDA-MB-231 and melanoma cell line MALME-3M (figure 5E).

NK cells can also improve the response of other immune cells against cancer cells by secreting cytokines. Anti-LILRB1 antibody increased IFN-γ secretion from NKL cells stimulated by T cell leukemia Jurkat cells and MM cell lines RPMI8226, OPM2 cells and KMS27 (figure 5F). We observed that anti-LILRB1 antibody increased the IFN-γ secretion from NKL cells induced by 697-MICA but not 697. This result suggests that LILRB1 blockade and NKG2D activation synergistically stimulated IFN-γ from NKL cells (figure 5F).

Moreover, we confirmed the function of anti-LILRB1 antibodies using primary NK cells. Primary NK cells from healthy donors’ buffy coats were isolated and LILRB1⁺ NK cells were enriched by FACS. Anti-LILRB1 antibody increased the cytotoxic activities of primary LILRB1⁺ NK cells from healthy donor against MM cell lines (KMS27 and OPM2), 697 cells or Raji cells, but not against Daudi cells, which do not express the LILRB1 ligand MHC class I (figure 6A). Since NK cells also express MHC class I, this result suggests that MHC class I on neighboring NK cells or cis-MHC class I does not inhibit the cytotoxic activity of NK cells against cancer cells. Here, 697 and Raji cells were sensitive to the cytotoxic activity of primary NK cells, likely due to the more potent cytotoxic activity of primary NK cells than that of NKL cells. One previous publication also reported that activated primary NK cells showed stronger cytotoxic activity than NKL cells, possibly due to the relative higher level of cytotoxicity receptors and NK94 on primary NK cells.39 We also isolated NK cells from the peripheral blood of one patient with MM whose NK cells was about 80% LILRB1 positive, and used the patient’s NK cells for cytotoxic assay. Anti-LILRB1 mAb increased cytotoxic activity of patient’s NK cells against MM cell line KMS27 (figure 6B).

We tested the in vivo function of anti-LILRB1 antibody using NSG mice. NKL-resistant 697 cell and NKL-sensitive 697 MICA mixtures were injected together with NKL cells into the peritoneal cavity of NSG mice. After 24 hours, the ratio of 697-MICA/697 was analyzed and calculated as the cytotoxic activity of NKL cells in vivo (figure 7A). Our results showed that anti-LILRB1 antibody increased the in vivo cytotoxic activity of NKL cells against 697-MICA cells. The cytotoxic activity of NKL cells was also tested in vivo by injecting the NKL and luciferase-overexpressed 697-MICA cell mixtures into NSG mice sc. After 48 hours, BLI showed that mice receiving anti-LILRB1 antibody had significantly lower tumor burdens compared with control hIgG-treated mice (figure 7B). Together, the anti-LILRB1 antibody can improve the in vivo cytotoxic function of NKL cells.

We further assessed the efficacy of anti-LILRB1 antibody to prevent MM development in a xenograft mouse model. NSG mice were engrafted with KMS27 cells and NKL cells through their tail vein and were administrated with control IgG or anti-LILRB1 antibody, with cancer development monitored by BLI. The mice that received NKL cells and anti-LILRB1 antibody had significantly lower disease burden and longer survival than did mice that received NKL cells and control IgG or mice that received only PBS (figure 7C). Taken together, our results suggest the antagonistic anti-LILRB1 antibody can improve the function of NKL cells against MM development in vivo.