B7H6 is the predominant activating ligand driving natural killer cell-mediated killing in patients with liquid tumours: evidence from clinical, in silico, in vitro, and in vivo studies Authors: Sunyoung Lee, Seok Joo Chae, In-Hwan Jang, Se-Chan Oh, Seok-Min Kim, Soo Yun Lee, Ji Hyun Kim, Jesang Ko, Hang J Kim, Ik-Chan Song, Jae Kyoung Kim, Tae-Don Kim Published: 2024 ABSTRACT SummaryBackgroundNatural killer (NK) cells are a subset of innate lymphoid cells that are inherently capable of recognizing and killing infected or tumour cells. This has positioned NK cells as a promising live drug for tumour immunotherapy, but limited success suggests incomplete knowledge of their killing mechanism. NK cell-mediated killing involves a complex decision-making process based on integrating activating and inhibitory signals from various ligand-receptor repertoires. However, the relative importance of the different activating ligand-receptor interactions in triggering NK killing remains unclear.MethodsWe employed a systematic approach combining clinical, in silico, in vitro, and in vivo data analysis to quantify the impact of various activating ligands. Clinical data analysis was conducted using massive pan-cancer data (n = 10,595), where patients with high NK cell levels were stratified using CIBERSORT. Subsequently, multivariate Cox regression and Kaplan–Meier (KM) survival analysis were performed based on activating ligand expression. To examine the impact of ligand expression on NK killing at the cellular level, we assessed surface expression of five major activating ligands (B7H6, MICA/B, ULBP1, ULBP2/5/6, and ULBP3) of human tumour cell lines of diverse origins (n = 33) via flow cytometry (FACs) and their NK cell-mediated cytotoxicity on by calcein-AM assay using human primary NK cells and NK-92 cell lines. Based on this data, we quantified the contribution of each activating ligand to the NK killing activity using mathematical models and Bayesian statistics. To further validate the results, we performed calcein-AM assays upon ligand knockdown and overexpression, conjugation assays, and co-culture assays in activating ligand-downregulated/overexpressed in liquid tumour (LT) cell lines. Moreover, we established LT-xenograft mouse models to assess the efficacy of NK cell targeting toward tumours with dominant ligands.FindingsThrough the clinical analysis, we discovered that among nearly all 18 activating ligands, only patients with LT who were NK cell-rich and specifically had higher B7H6 level exhibited a favorable survival outcome (p = 0.0069). This unexpected dominant role of B7H6 was further confirmed by the analysis of datasets encompassing multiple ligands and a variety of tumours, which showed that B7H6 exhibited the highest contribution to NK killing among five representative ligands. Furthermore, LT cell lines (acute myeloid leukemia (AML), B cell lymphoma, and T-acute lymphocytic leukemia (ALL)) with lowered B7H6 demonstrated decreased susceptibility to NK cell-mediated cytotoxicity compared to those with higher levels. Even within the same cell line, NK cells selectively targeted cells with higher B7H6 levels. Finally, LT-xenograft mouse models (n = 24) confirmed that higher B7H6 results in less tumour burden and longer survival in NK cell-treated LT mice (p = 0.0022).InterpretationWhile NK cells have gained attention for their potent anti-tumour effects without causing graft-versus-host disease (GvHD), thus making them a promising off-the-shelf therapy, our limited understanding of NK killing mechanisms has hindered their clinical application. This study illuminates the crucial role of the activating ligand B7H6 in driving NK cell killing, particularly in the context of LT. Therefore, the expression level of B7H6 could serve as a prognostic marker for patients with LT. Moreover, for the development of NK cell-based immunotherapy, focusing on increasing the level of B7H6 on its cognate receptor, NKp30, could be the most effective strategy.FundingThis work was supported by the National Research Council of Science & Technology (NST) grant (CAP-18-02-KRIBB, GTL24021-000), a National Research Foundation grant (2710012258, 2710004815), and an Institute for Basic Science grant (IBS-R029-C3).Keywords: B7H6, Natural killer cell, NK killing, Activating ligand, Major activating ligand, Liquid tumour Research in context Evidence before this study We searched PubMed on April 22, 2024, for relevant articles published from database inception using the search terms “NK cells” AND “activating ligands”. Although over 100 studies have evaluated individual activating ligands, none have yet analyzed the comparative potency across various activating ligands, which reflects the physiological state. We also searched for relevant articles using the search terms “NK cells”, “Clinical”, “in vitro”, and “in vivo” with proper Boolean operators in the same databases, on the same dates. While over 10,000 studies employed a single-level approach, no studies utilized a combined approach incorporating clinical data analysis alongside in silico/in vitro/in vivo experiments. Added value of this study Our study utilized a clinical analysis to assess the significance of all identified activating ligands. Among these ligands, B7H6 emerged as the sole correlate with survival in patients with LT. To support this finding, we quantified the contribution of each activating ligand to the NK killing activity with in silico analysis. This analysis also predicted that B7H6 is the most potent activating ligand inducing NK cell-mediated killing against LT. This was validated by high-quality evidence across multiple scales. This evidence included calcein-AM assays, conjugation assays, and co-culture assays in B7H6-downregulated/overexpressed LT cell lines. Finally, the established LT xenograft mouse models confirmed that higher B7H6 levels enhance NK cell killing efficiency against LT. Implications of all the available evidence Our data demonstrate that the B7H6-NKp30 interaction is a dominant inducer of NK cell-mediated cytotoxicity. Unlike other activating ligands, high B7H6 expression correlated with improved prognosis in patients with LT. Furthermore, in vitro NK-mediated cytotoxicity against LT cell lines and in vivo survival of LT-xenograft models significantly increased with higher B7H6 expression. Collectively, these findings suggest that the B7H6-NKp30 pathway is the most promising for further development of NK-based immunotherapies. Introduction Natural killer (NK) cells are innate lymphoid cells that can recognize and kill transformed cells, including virus-infected cells and tumour cells. Due to their high efficacy while causing minimal side effects,1, 2, 3, 4, 5, 6 NK cells represent a promising form of tumour immunotherapy. However, using NK cells in their natural state as a living drug has had limited success in various clinical trials due to the difficulty of controlling NK killing activity.7, 8, 9, 10, 11 To overcome this challenge and develop effective anti-tumour therapies utilizing NK cells, it is crucial to understand how various activating and inhibitory signals regulate NK killing activity.8,11,12 The exploration of the NK killing mechanism commenced over half a century ago with the identification of “missing-self” by Karre and Ljunggen.13 They observed that NK cells target cells lacking expression of major histocompatibility complex (MHC)-Class 1 molecules. In cases where cells express MHC-Class 1 and are recognized as “self,” activating signals from various activating ligands such as MICA/B, ULBP families, and B7H6 induce NK killing if they outweigh inhibitory signals.9,10,14, 15, 16, 17, 18 Despite the fact that activating ligand-receptor pairs are not equally potent, no studies have investigated which pairs are the most potent (Fig. 1a).10 This is because existing studies have regulated only one ligand at a time, and focused solely on the qualitative effect (i.e., either activating or inhibitory) on NK killing activity. Thus, a fundamental question about NK killing remains unanswered: Among the various types of activating signals, which signal is the major determinant of NK killing activity? A systematic approach is required to identify key activating ligands in NK killing. (a) NK cells recognize tumours via specific and varied ligand-receptor interactions (denoted by different shapes and colors). NK cells make the decision to kill tumours or not by integrating signals from various receptors. However, the specific contribution (impact) of each activating ligand on this NK cell killing decision remains unclear, impeding the identification of key activating ligands (red question mark). (b) To address this, our study utilizes a systematic approach combining in clinical, in silico, in vitro, and in vivo data analyses. To address this gap, we conducted a quantitative and systematic investigation into the impact of activating signals on NK cell killing (Fig. 1b). Our clinical analysis (n = 10,595) showed that only B7H6 among nearly all identified activating ligands (n = 18) is correlated with a better prognosis. To further validate this unexpected result, we integrated the data on surface ligand expressions and NK-mediated cytotoxicity from various tumour cell lines (n = 33), and then quantified the contribution of each ligand to NK killing using mathematical models and Bayesian statistics. Our in silico analysis revealed that B7H6 predominates in NK killing among five major activating ligands, especially in liquid tumour (LT) cell lines. This discovery was further validated through in vitro and in vivo experiments, where regulating B7H6 most strongly affected NK-mediated cytotoxicity. Our approach suggests that B7H6 is the most promising therapeutic target in NK cell-based therapy for LT and may also be a high value predictive marker of response to NK cell-based therapy. Methods Study design and data sources for clinical analysis RNA-sequencing expression data and clinical information of the 10,595 patients with 37 types of tumours were downloaded from the genomic data commons (GDC) data portal (https://portal.gdc.cancer.gov). The patients with solid tumours (ST; n = 9960) were derived from The Cancer Genome Atlas Program (TCGA) dataset for the various tumour types (n = 33) (See Table 1, Table 2 for details). The patients with LT (n = 2406) were used from 2 types of TCGA datasets and 5 independent datasets: The Therapeutically Applicable Research to Generate Effective Treatments (TARGET)- acute myeloid leukemia (AML) as AML1 (n = 187), TCGA as AML2 (n = 151), TARGET-acute lymphocytic leukemia (ALL)-P2 as B-ALL (n = 267) and T-ALL (n = 265), Oregon Health and Science University (OHSU)- Chronic myeloproliferative disorders (CNL) as CNL (n = 51), BEATAML1.0 (OHSU-AML) as AML3 (n = 450), diffuse large B cell lymphoma (DLBC) from TCGA (n = 47), the Cancer Genome Characterization Initiative-Burkitt Lymphoma Genome Sequencing Project (CGCI-BLGSP) as mature B-cell lymphoma (MBL) (n = 112), and The Multiple Myeloma Research Foundation (MMRF)-COMMPASS as plasma cell tumours (PCL) (n = 787). We classified all kinds of lymphoma as LT because of their cellular origin. In TARGET-ALL-P2, because T-and B-ALL are phenotypically and molecularly different, we analyzed them separately. The fragments per kilobase of transcript per million (FPKM) values transformed to a log 2 scale were used for analysis. Table 1Clinical features of each data set of patients with LT investigated.TARGETOHSUCGCI-BLGSPMMRFTCGA(n = 719)(n = 590)(n = 112)(n = 787)(n = 198)Diagnosis age <20 years719–112–– >20 years and <60 years–––31689 >60 years–––471109Sex Female2902184132594 Male42930271462104 Unknown–70–––Type of tumour Acute Myeloid Leukemia (AML)187450––151 B cell Acute Lymphocytic Leukemia (B-ALL)267–112–– T cell Acute Lymphocytic Leukemia (T-ALL)265––––Chronic Myeloproliferative Disorders (CNL)–51––– Diffuse Large B cell Lymphoma (DLBC)––––47 Mature B cell Lymphoma (MBL)––––– Plasma Cell Tumour (PCL)–––787–Vital status Alive4451216062496 Dead24728052163102 Unknown27100–––Site of resection or biopsy Blood175265––– Bone marrow5443133787170 Lymph nodes or other organs––109–10 Unknown–12––18Average of NK cell proportion Percentage (%)0.540.060.0470.0230.054 IQR(0.033–0.082)(0.04–0.963)(0.03–0.06)(0.007–0.046)(0.028–0.088)Average of CD8 T cell proportion Percentage (%)0.0130.0120.0480.0070.018 IQR(0–0.051)(0–0.042)(0.013–0.085)(0–0.016)(0–0.074)Average of CD4 T cell proportion Percentage (%)0.1570.0990.2660.0520.148 IQR(0.98–0.157)(0.537–0.165)(0.21–0.306)(0.03–0.08)(0.1–0.241)N/A, not available.IQR, Inter Quartile Range. Clinical features of each data set of patients with LT investigated. N/A, not available. IQR, Inter Quartile Range. Table 2Clinical features of each dataset of patients with ST investigated.TCGA(n = 9960)Diagnosis age >60 years4956 <60 years5004Sex Female5165 Male4795Type of tumour Adrenocortical Carcinoma (ACC)79 Breast Invasive Carcinoma (BRCA)1096 Bladder Urothelial Carcinoma (BLCA)406 Cervical Squamous Cell Carcinoma (CESC)293 Cholangiocarcinoma (CHOL)36 Colon Adenocarcinoma (COAD)448 Esophageal Carcinoma (ESCA)161 Glioblastoma Multiforme (GBM)167 Head and Neck Squamous (HNSC)501 Kidney Chromophobe (KICH)64 Kidney Renal Clear Cell (KIRC)531 Kidney Renal Papillary Cell (KIRP)286 Brain Lower Grade Glioma (LGG)524 Liver Hepatocellular Carcinoma (LIHC)368 Lung Adenocarcinoma (LUAD)514 Lung Squamous Cell Carcinoma (LUSC)493 Mesothelioma (MESO)84 Ovarian Serous Cystadenocarcinoma (OV)378 Pancreatic Adenocarcinoma (PAAD)177 Pheochromocytoma Paraganglioma (PCPG)183 Prostate Adenocarcinoma (PRAD)500 Rectum Adenocarcinoma (READ)158 Sarcoma (SARC)260 Skin Cutaneous Melanoma (SKCM)457 Stomach Adenocarcinoma (STAD)350 Testicular Germ Cell Tumours (TGCT)139 Thyroid Carcinoma (THCA)511 Thymoma (THYM)118 Uterine Corpus Carcinoma (UCEC)544 Uterine Carcinosarcoma (UCS)54 Uveal Melanoma (UVM)80AJCC T stage Ta or Tis9 T11808 T22552 T32358 T4775 Tx92 N/A2366AJCC N stage N03975 N11662 N2736 N3235 Nx953 N/A2399AJCC M stage M05728 M1361 Mx1462 N/A2409Gradea G1288 G21490 G31738 G494 Gx62 Low Grade20 High Grade394 N/A5874Vital status Alive7831 Dead2129 Unknown–Average of NK cell proportion Percentage (%)0.053 IQR(0.031–0.08)Average of CD8 T cell proportion Percentage (%)0.044 IQR(0.006–0.091)Average of CD4 T cell proportion Percentage (%)0.272 IQR(0.209–0.338)aThe criteria for high grade may vary depending on the tumour type. Clinical features of each dataset of patients with ST investigated. The criteria for high grade may vary depending on the tumour type. The workflow for clinical analysis is presented in Fig. 2. It involves the use of TCGA pan-cancer data for ST (n = 9289) and five datasets for LT (n = 1306), all of which are primary tumour samples with available vital records in biospecimen or clinical files. These datasets were used to estimate hazard ratios and conduct survival analysis. Among the nine LT datasets, four datasets (T-ALL in TARGET-ALL-P2, OHSU-AML, OHSU-CNL, and TCGA-DLBC) were excluded from risk ratio and survival analysis due to challenges in assessing survival. Specifically, the T-ALL dataset in TARGET-ALL-P2 was excluded because the survival rate for the NK cell-rich T-ALL group was 100%. The OHSU-AML was excluded from clinical analysis due to the absence of duration of survival, and the OHSU-CNL and TCGA-DLBC datasets were excluded due to the low number of NK cell-rich samples (n < 10). To avoid data bias from different datasets, each tumour type or dataset was normalized using standardization separately before integrating datasets. Workflow of clinical data analysis. RNA-sequencing expression data of the 10,595 patients with 37 types of tumours were downloaded from the genomic data commons (GDC) data portal (https://portal.gdc.cancer.gov). Using the computational quantification of immune cells via CIBERSORT, the samples were divided into two groups: an NK cell-rich group, with NK cell ratios ranging from 50-80%–100% (statistically significant correlation, p < 0.05), and an NK cell-poor group, with NK cell ratios up to 50–80%. TCGA pan-cancer for ST (n = 9289) and 5 datasets for LT (n = 1306), which are samples of primary tumours and existing information of vital records in biospecimens or clinical files, were used to perform survival analysis. In detail, T-ALL was excluded because the survival rate for NK cell-rich patients was 100%, and TCGA-DLBC and OHSU-CNL were excluded due to an insufficient number of NK cell-rich patients for meaningful survival analysis (n < 10). The association between NK cell activating ligands and CD8 T cell or CD4 T cell-rich patients was also investigated to confirm whether the effect was dependent on NK cells or all tumour-infiltrated lymphocytes. Sex information was obtained from the publicly available datasets in the GDC Data Portal, which aggregates clinical data from multiple studies. The data used in this study primarily rely on medical records rather than self-reported information. Procedures for clinical analysis To construct the t-distributed stochastic neighbor embedding (t-SNE) plot of expression profiles for pan-cancer, the genes of the NK cell activating ligands involved in NK killing (n = 18) were used as input data.19,20 The expression level of each sample was standardized before merging the expression profile of pan-cancer. The perplexity parameter was set to 30 and the plot was embedded in two dimensions. Expression patterns were shown in dot plots from the expression of ligands. CIBERSORT was used for the computational deconvolution of immune cells for the 10,595 patients from gene expression data (https://cibersort.stanford.edu).21 Among the expression values available through the GDC portal, we employed log2-scaled FPKM values for the deconvolution method.22 The created gene expression file with the 10,595 cases was uploaded to CIBERSORT as a mixture file, and CIBERSORT was run with the following options: LM22 reference file, 500 permutations, and quantile normalization disabled. The frequency of NK cells included activated NK cells and resting NK cells in an LM22 reference file. The samples, which were the distribution of NK cell ratios from 50 to 80%–100% with statistically significant correlation (p < 0.05), were called the NK cell-rich group, and the others (up to 50–80%) were called the NK cell-poor group. To avoid data bias, we performed the division of groups based on the relative proportion of NK cells separately for each dataset. Multivariate Cox regression models (95% confidence intervals) were performed using 18 genes of known activating ligands to estimate the significance of overall survival in patients. For detailed information on the activating ligands, refer to Appendix 1 p 1. Survival curves were visualized using Kaplan–Meier (KM) plots and log-rank tests, comparing the ligands-high and ligands-low groups divided by the median cut-off for each activating ligand. Study design for in vitro/in vivo analysis For the quantitative study of different activating ligands (n = 5), we performed in silico analysis based on the surface ligand expression data and NK cell-mediated cytotoxicity data. We utilized flow cytometry (FACs) to investigate the surface expression of activating ligands (n = 5) on various human tumour cell lines (n = 33) from various origins. We also measured NK cell-mediated cytotoxicity against each tumour cell line using human primary NK (pNK) cells and the NK-92 cell line. With the obtained experimental data, the NK cell-mediated cytotoxicity prediction model was constructed and analyzed to confirm which ligand was dominant. We then performed in vitro experiments including calcein-AM assay, conjugation assay, co-culture assay, and western blotting, and in vivo experiments including blood analysis, bioluminescence imaging (BLI), and survival analysis in LT-xenograft mice (n = 24). Mathematical modeling and Gibbs variable selection (GVS) To quantify the contribution of five representative activating ligands to NK killing, we developed a mathematical model predicting cytotoxicity with activating ligand expression, specifically, the NK cell-mediated cytotoxicity of a cell line assuming additive impact of individual ligands, following previous studies.23,24 Then, we fitted the model to incorporate the relation between the NK-mediated cytotoxicity and surface ligand expression levels. In addition to the model, we also constructed 25 models as its variants. Then, each model was fitted to account for the experimentally measured pNK (or NK-92)-mediated cytotoxicity of LT and ST groups. Subsequently, we utilized the Akaike information criterion (AIC)25, 26, 27 to assess each model and identify the key combinations of activating ligands in each group. Finally, the statistical significance of each activating ligand in predicting pNK-mediated cytotoxicity was assessed using the GVS with the model.28 Construction of mathematical models predicting NK cell-mediated cytotoxicity with surface expression of tumour ligands To assess the contribution of representative activating ligands to NK cell-mediated cytotoxicity, we developed a mathematical model predicting cytotoxicity with activating ligand expression. Since NK cells express a sufficient amount of activating receptors, including NKG2D and NCR families, we assumed that receptor expressions are saturated, and thus the NK cell-mediated cytotoxicity is mainly determined by ligand expression. Specifically, the NK cell-mediated cytotoxicity of a cell line is described by a sum of the contribution of each ligand, as in previous studies.23,24 The contribution of each ligand was described by a Hill function following a previous study,29 with the Hill exponent of 2. We assumed that the Hill function describing the contribution of each ligand has the same normalized expression at half of the maximum contribution to reduce the number of parameters used in the fitting. For example, the predicted pNK-mediated cytotoxicity of the i-th cell line at E:T ratio of 0.5:1, Cpredi,0.5:1, is described as follows: Here, Basal is the basal killing of pNK, and Lj,i is the normalized expression of the ligand j of the i-th cell line (the minimum of expression across cell lines of 0 and the maximum of 10). wj is the weight of the ligand j, which quantifies the extent to which ligand j can contribute to the pNK-mediated cytotoxicity. That is, wj is used to quantify the killing weight of each ligand by calculating how much increase in the pNK-mediated cytotoxicity can be achieved by increasing the expression of the ligand j from the minimum to the maximum value. K is the threshold for activating ligands, which is the normalized ligand expression at half of the maximal contribution of each ligand. Eq. (1) is also used to predict the NK-92-mediated cytotoxicity. Specifically, the predicted NK-92-mediated cytotoxicity of i-th cell line at E:T ratio of 5:1, Cpredi,5:1, is described by Cpredi,5:1(LULBP1,i,LULBP2/5/6,i,LULBP3,i,LMICA/B,i,LB7H6,i)=Basal+∑j={MICA/B,ULBP1,ULBP2/5/6,ULBP3,B7H6}wjLj,i2Lj,i2+K2, but with different parameters. To describe the pNK-mediated cytotoxicity at various E:T ratios, we multiplied the scaling factor with the cytotoxicity prediction at E:T ratio of 0.5:1. Specifically, to describe the pNK-mediated cytotoxicity at E:T ratios (0.25:1 and 1:1) in LT, we multiplied two scaling factors, A0.25:1 and A1:1 to the Cpredi,0.5:1 (i.e., A0.25:1·Cpredi,0.5:1 and A1:1·Cpredi,0.5:1). For ST, the scaling factor A2:1 is also multiplied to the predicted pNK-mediated cytotoxicity with E:T ratio 0.5:1. Similarly, for the model predicting the NK-92-mediated cytotoxicity, A1:1 and A10:1 were multiplied to the Cpredi,5:1 (i.e., A1:1·Cpredi,5:1 and A10:1·Cpredi,5:1) to describe NK-92-mediated cytotoxicity at E:T ratio of 1:1 in LT and 10:1 in ST, respectively. Finally, when the predicted cytotoxicity exceeds 100%, we capped the value at 100% to ensure the cytotoxicity does not exceed this maximum limit. To account for the experimental data, we fitted the models predicting either the pNK (or NK-92)-mediated cytotoxicity for the LT and the ST, respectively, by minimizing the following squared error: Here, Cexpi,E:T is the pNK (or NK-92)-mediated cytotoxicity of the i-th cell at the given E:T ratio, which is measured in the experiment. For the minimization, we utilized the MATLAB built-in lsqcurvefit function. We performed the fitting with not only the full model (Eq. (1)) but also its modifications, where the cytotoxicity is determined by the subset of the five ligands rather than the whole five ligands: Cpredi,E:T=AE:T(Basal+∑j∈XwjLj,i2Lj,i2+K2), where X is the subset of {MICAB,ULBP1,ULBP2/5/6,ULBP3,B7H6}. Since the original model (Eq. (1)) has five activating ligands, which were measured from the experiment, 25 models were constructed in total, which is equal to the number of all the subsets of {MICA/B,ULBP1,ULBP2/5/6,ULBP3,B7H6}. Then, each model was fitted to the experimentally measured pNK (or NK-92)-mediated cytotoxicity of the given group (LT or ST) by minimizing Eq. (2). The fitted results were evaluated by the AIC25, 26, 27 to identify the key combinations of activating ligands in each group. Since the number of data points used in the fitting was small, we calculated the corrected AIC of each model by the following formula: where n is the number of data points used in the fitting, k is the number of parameters used in the model, and Error is the squared error obtained from Eq. (2). The AIC evaluates the accuracy of the model while it penalizes models for complexity, and thus the model with low AIC predicts the NK cell-mediated cytotoxicity accurately with a small number of variables. By investigating models with low AIC values, key ligand combinations in NK killing decisions can be obtained. Furthermore, the statistical significance of each ligand in predicting pNK-mediated cytotoxicity was tested using the GVS with the model.28 The GVS introduces a latent variable γj for each ligand, which has the value of one if the ligand j affects the cytotoxicity and zero otherwise, resulting in the following form of the variable selection model: where the error in predicting the cytotoxicity εi was assumed to be normally distributed (εi∼N(0,σ2)). The significance of each ligand in explaining the pNK-mediated cytotoxicity can be expressed by the posterior probability that the ligand j is included in the model, i.e., P[γj=1] after obtaining the posterior samples of all unknown parameters from the Markov chain Monte Carlo (MCMC). If the inclusion probability of a ligand is large (i.e., P[γj=1]≈1), this indicates the ligand j affects the pNK-mediated cytotoxicity. For the Bayesian parameter inference with MCMC, we complete the model by introducing the following prior distributions: Hyperparameters used in the prior distributions are as follows: τ=100 (for LT) or 10 (for ST), s0=1, Kmax=100, Amax=10, ηw=0.2, ηA=0.003. The likelihood function is written by L(γ,w,A,K,Basal)=∏E:T∏i1σ2πexp(−12·(Cexpi,E:T−Cpredi,E:T)2σ2), where γ=(γMICA/B,γULBP1,γULBP2/5/6,γULBP3,γB7H6), w=(wMICA/B,wULBP1,wULBP2/5/6,wULBP3,wB7H6), and A=(A0.25:1,A1:1) for the LT group and A=A2:1 for the ST group with setting A0.5:1=1. With this Bayesian model, the posterior samples of parameters are drawn from MCMC by iterating following steps: 1) Update of Basala)Generate a proposed parameter Basal˜ from N(Basal,ηW2).b)Calculate the likelihood L(γ,w,A,K,Basal˜).c)Accept Basal˜ with probability min(L(γ,w,A,K,Basal˜)L(γ,w,A,K,Basal),1) and reject otherwise. ) Update of Basala)Generate a proposed parameter Basal˜ from N(Basal,ηW2).b)Calculate the likelihood L(γ,w,A,K,Basal˜).c)Accept Basal˜ with probability min(L(γ,w,A,K,Basal˜)L(γ,w,A,K,Basal),1) and reject otherwise. a)Generate a proposed parameter Basal˜ from N(Basal,ηW2). Generate a proposed parameter Basal˜ from N(Basal,ηW2). b)Calculate the likelihood L(γ,w,A,K,Basal˜). Calculate the likelihood L(γ,w,A,K,Basal˜). c)Accept Basal˜ with probability min(L(γ,w,A,K,Basal˜)L(γ,w,A,K,Basal),1) and reject otherwise. Accept Basal˜ with probability min(L(γ,w,A,K,Basal˜)L(γ,w,A,K,Basal),1) and reject otherwise. 2)Update of Wja)For an activating ligand j, generate a proposed parameter w˜j from N(wj,ηw2).b)Calculate the likelihood L(γ,w˜,A,K,Basal), where w˜ is a vector equal to w in all elements but one by replacing wj in w by w˜j.c)Accept w˜j with probability min(L(γ,w˜,A,K,Basal)L(γ,w,A,K,Basal),1) and reject otherwise.d)Repeat for all the activating ligands. )Update of Wja)For an activating ligand j, generate a proposed parameter w˜j from N(wj,ηw2).b)Calculate the likelihood L(γ,w˜,A,K,Basal), where w˜ is a vector equal to w in all elements but one by replacing wj in w by w˜j.c)Accept w˜j with probability min(L(γ,w˜,A,K,Basal)L(γ,w,A,K,Basal),1) and reject otherwise.d)Repeat for all the activating ligands. a)For an activating ligand j, generate a proposed parameter w˜j from N(wj,ηw2). For an activating ligand j, generate a proposed parameter w˜j from N(wj,ηw2). b)Calculate the likelihood L(γ,w˜,A,K,Basal), where w˜ is a vector equal to w in all elements but one by replacing wj in w by w˜j. Calculate the likelihood L(γ,w˜,A,K,Basal), where w˜ is a vector equal to w in all elements but one by replacing wj in w by w˜j. c)Accept w˜j with probability min(L(γ,w˜,A,K,Basal)L(γ,w,A,K,Basal),1) and reject otherwise. Accept w˜j with probability min(L(γ,w˜,A,K,Basal)L(γ,w,A,K,Basal),1) and reject otherwise. d)Repeat for all the activating ligands. Repeat for all the activating ligands. 3)Update of AE:Ta)For a given E:T ratio, generate a proposed parameter A˜E:T from N(AE:T,ηA2).b)Calculate the likelihood L(γ,w,A˜,K,Basal), where A˜ is a vector equal to A in all elements but one by replacing AE:T in A by A˜E:T.c)Accept A˜E:T with probability min(L(γ,w,A˜,K,Basal)L(γ,w,A,K,Basal),1) and reject otherwise.d)Repeat for all the E:T ratios. )Update of AE:Ta)For a given E:T ratio, generate a proposed parameter A˜E:T from N(AE:T,ηA2).b)Calculate the likelihood L(γ,w,A˜,K,Basal), where A˜ is a vector equal to A in all elements but one by replacing AE:T in A by A˜E:T.c)Accept A˜E:T with probability min(L(γ,w,A˜,K,Basal)L(γ,w,A,K,Basal),1) and reject otherwise.d)Repeat for all the E:T ratios. a)For a given E:T ratio, generate a proposed parameter A˜E:T from N(AE:T,ηA2). For a given E:T ratio, generate a proposed parameter A˜E:T from N(AE:T,ηA2). b)Calculate the likelihood L(γ,w,A˜,K,Basal), where A˜ is a vector equal to A in all elements but one by replacing AE:T in A by A˜E:T. Calculate the likelihood L(γ,w,A˜,K,Basal), where A˜ is a vector equal to A in all elements but one by replacing AE:T in A by A˜E:T. c)Accept A˜E:T with probability min(L(γ,w,A˜,K,Basal)L(γ,w,A,K,Basal),1) and reject otherwise. Accept A˜E:T with probability min(L(γ,w,A˜,K,Basal)L(γ,w,A,K,Basal),1) and reject otherwise. d)Repeat for all the E:T ratios. Repeat for all the E:T ratios. 4)Update of Ka)Generate a proposed parameter K˜ from K+2·Bernoulli(0.5)−1b)Calculate the likelihood L(γ,w,A,K˜,Basal).c)Accept K˜ with probability min(L(γ,w,A,K˜,Basal)L(γ,w,A,K,Basal),1) and reject otherwise. Update of Ka)Generate a proposed parameter K˜ from K+2·Bernoulli(0.5)−1b)Calculate the likelihood L(γ,w,A,K˜,Basal).c)Accept K˜ with probability min(L(γ,w,A,K˜,Basal)L(γ,w,A,K,Basal),1) and reject otherwise. a)Generate a proposed parameter K˜ from K+2·Bernoulli(0.5)−1 Generate a proposed parameter K˜ from K+2·Bernoulli(0.5)−1 b)Calculate the likelihood L(γ,w,A,K˜,Basal). Calculate the likelihood L(γ,w,A,K˜,Basal). c)Accept K˜ with probability min(L(γ,w,A,K˜,Basal)L(γ,w,A,K,Basal),1) and reject otherwise. Accept K˜ with probability min(L(γ,w,A,K˜,Basal)L(γ,w,A,K,Basal),1) and reject otherwise. 5)Update of γja)For a given ligand, set γ˜j=1 if γj=0 and γ˜j=0 if γj=1.b)Calculate the likelihood L(γ˜,w,A,K,Basal) where γ˜ is a vector equal to γ in all elements but one by replacing γj in γ by γ˜j.c)Accept γ˜j with probability min(L(γ˜,w,A,K,Basal)L(γ,w,A,K,Basal),1) and reject otherwise.d)Repeat for all the ligands. Update of γja)For a given ligand, set γ˜j=1 if γj=0 and γ˜j=0 if γj=1.b)Calculate the likelihood L(γ˜,w,A,K,Basal) where γ˜ is a vector equal to γ in all elements but one by replacing γj in γ by γ˜j.c)Accept γ˜j with probability min(L(γ˜,w,A,K,Basal)L(γ,w,A,K,Basal),1) and reject otherwise.d)Repeat for all the ligands. a)For a given ligand, set γ˜j=1 if γj=0 and γ˜j=0 if γj=1. For a given ligand, set γ˜j=1 if γj=0 and γ˜j=0 if γj=1. b)Calculate the likelihood L(γ˜,w,A,K,Basal) where γ˜ is a vector equal to γ in all elements but one by replacing γj in γ by γ˜j. Calculate the likelihood L(γ˜,w,A,K,Basal) where γ˜ is a vector equal to γ in all elements but one by replacing γj in γ by γ˜j. c)Accept γ˜j with probability min(L(γ˜,w,A,K,Basal)L(γ,w,A,K,Basal),1) and reject otherwise. Accept γ˜j with probability min(L(γ˜,w,A,K,Basal)L(γ,w,A,K,Basal),1) and reject otherwise. d)Repeat for all the ligands. Repeat for all the ligands. 6)Update of σa)Sample σ2 from IG (1+n2,1+12∑E:T∑i(Cexpi,E:T−Cpredi,E:T)2).b)Calculate σ from the sampled σ2. Update of σa)Sample σ2 from IG (1+n2,1+12∑E:T∑i(Cexpi,E:T−Cpredi,E:T)2).b)Calculate σ from the sampled σ2. a)Sample σ2 from IG (1+n2,1+12∑E:T∑i(Cexpi,E:T−Cpredi,E:T)2). Sample σ2 from IG (1+n2,1+12∑E:T∑i(Cexpi,E:T−Cpredi,E:T)2). b)Calculate σ from the sampled σ2. Calculate σ from the sampled σ2. The above steps were iterated 106 times to generate a chain of parameters with a sufficient length. Then, the first 2 · 105 steps were excluded from the chain as the burn-in stage. The resulting distributions of parameters are shown. By utilizing the model, we explored the extent to which NK cell-mediated cytotoxicity can be attributed to a single ligand. Specifically, we employed a mathematical model based on the expression of a single ligand to predict NK cell-mediated cytotoxicity, as described by the following equation: where Basal is the basal killing of pNK, and Li is the normalized expression of the ligand of the i-th cell line (the minimum of expression across cell lines of 0 and the maximum of 10). w is the weight of the ligand, which quantifies the contribution of the ligand to the pNK-mediated cytotoxicity. In our in silico analysis, this model (Eq. (5)) was applied to analyze pNK-mediated cytotoxicity at different levels of B7H6 expression. The model was fitted using surface B7H6 expression and pNK-mediated cytotoxicity data across various E:T ratios (0.25:1, 0.5:1, and 1:1). Furthermore, the model was fitted using pNK-mediated cytotoxicity data along with the various surface expression levels of MICA/B to compare the dependence of NK cell cytotoxicity on B7H6 and MICAB. Cell culture Raji, Ramos, Jurkat, U937, CCRF-CEM, MOLT-4, MDA-MB-231, SK-OV-3, THP-1, AU565, Kasumi-1, Kasumi-3, MV4-11, HL-60, Huh7, HepG2, SK-BR-3, MCF-7, HEK293T, U87-MG, and NK-92 cell lines were purchased from American Type Culture Collection. KG-1, RPMI8226, RPMI6666, CCRF-SB, Jiyoye, and H9 cell lines were purchased from Korean Cell Line Bank. NCI–H460, PC-9, and A498 cell lines were provided by Dr. Haiyoung Jung. BCPAP, 850-5C, Ca ski, 786-0, and ACHN cell lines were provided by Dr. Suk Ran Yoon. DLD-1, Sw480, HCT116, HT29, and Sw620 cell lines were provided by Dr. Ji Yoon Noh. Further information for cell lines is attached in the reagent validation file. Additional information about the cell lines is presented below. Cell lineMediumSerumComponents for complete mediumRRIDRajiRPMI10% FBS–CVCL_0511RamosRPMI10% FBS–CVCL_0597JurkatRPMI10% FBS–CVCL_0065U937RPMI10% FBS–CVCL_0007CCRF-CEMRPMI10% FBS–CVCL_0207MOLT-4RPMI10% FBS–CVCL_0013MDA-MB-231RPMI10% FBS–CVCL_0062SK-OV-3RPMI10% FBS–CVCL_0532BCPAPRPMI10% FBS–CVCL_0153850-5CRPMI10% FBS–CVCL_1054Ca skiRPMI10% FBS–CVCL_1100NCI-H460RPMI10% FBS–CVCL_0459DLD-1RPMI10% FBS–CVCL_0248Sw480RPMI10% FBS–CVCL_0546PC-9RPMI10% FBS–CVCL_B260786-0RPMI10% FBS–CVCL_1051HCT116RPMI10% FBS–CVCL_0291HT29RPMI10% FBS–CVCL_A8EZSw620RPMI10% FBS–CVCL_0547CCRF-SBRPMI10% FBS–CVCL_1860RPMI 8226RPMI10% FBS–CVCL_0014THP-1RPMI10% FBS25 mM BMECVCL_0006AU565RPMI10% FBS25 mM HEPESCVCL_1074Kasumi-1RPMI20% FBS–CVCL_0589Kasumi-3RPMI20% FBS–CVCL_0612JiyoyeRPMI20% FBS–CVCL_1317H9RPMI20% FBS–CVCL_1240RPMI 6666RPMI20% FBS–CVCL_1665KG-1IMDM20% FBS–CVCL_0374MV4-11IMDM20% FBS–CVCL_0064HL-60IMDM20% FBS–CVCL_0002Huh7DMEM10% FBS–CVCL_0336HepG2DMEM10% FBS–CVCL_0027SK-BR-3DMEM10% FBS–CVCL_0033MCF-7DMEM10% FBS–CVCL_0031A498DMEM10% FBS–CVCL_1056HEK293TDMEM10% FBS–CVCL_0063U87-MGMEM10% FBS–CVCL_0022ACHNMEM10% FBS–CVCL_1067NK-92a-MEM12.5% FBS + 12.5%HS0.2 mM MI + 0.1 mM BME+ 0.02 mM FA + 200 U/ml IL-2CVCL_2142RPMI; RPMI-1640 (Welgene, LM011-01), IMDM; Iscove's Modified Dulbecco's Medium (Welgene, LM004-01), DMEM; Dulbecco's Modified Eagle's Medium (Welgene, LM-001-05), MEM; Minimum Essential Medium Eagle (Welgene, LM007-07), a-MEM; alpha minimum essential medium (Welgene, LM008-02), FBS; Fetal bovine serum (Avantor Seradigm, 97068-085), HS; Horse Serum (Gibco, 16050-122), BME; 2-mercaptoethanol (Sigma, M3148), HEPES; Hydroxyethyl piperazine Ethane Sulfonic acid (Gibco, 15630-080), MI; Myo-inositol (Sigma, I7508), FA; Folic acid (Sigma, F8758), IL-2; recombinant Interleukin-2 (Peprotech, Cat. 200–02). RPMI; RPMI-1640 (Welgene, LM011-01), IMDM; Iscove's Modified Dulbecco's Medium (Welgene, LM004-01), DMEM; Dulbecco's Modified Eagle's Medium (Welgene, LM-001-05), MEM; Minimum Essential Medium Eagle (Welgene, LM007-07), a-MEM; alpha minimum essential medium (Welgene, LM008-02), FBS; Fetal bovine serum (Avantor Seradigm, 97068-085), HS; Horse Serum (Gibco, 16050-122), BME; 2-mercaptoethanol (Sigma, M3148), HEPES; Hydroxyethyl piperazine Ethane Sulfonic acid (Gibco, 15630-080), MI; Myo-inositol (Sigma, I7508), FA; Folic acid (Sigma, F8758), IL-2; recombinant Interleukin-2 (Peprotech, Cat. 200–02). All complete culture mediums included 1% Antibiotic-Antimycotic (Gibco, 15240-062). Mycoplasma contamination was tested with a commercially available mycoplasma detection kit (iNtRON, Cat. 25239). Primary NK cell isolation/culture/differentiation from cord blood (CB) The procurement and use of primary cells from CB mononuclear cells were approved by the Public Institutional Review Board designated by the Ministry of Health and Welfare (IRB approval number: P01-201610-31-002). RosetteSep™ Human CD3 Depletion Cocktail (STEMCELL, 15661) was added to CB to deplete the CD3 population, and the buffy coat was isolated using the Ficoll-PaqueTM PREMIUM (GE Healthcare, 17-5442-03) gradient method. After isolation of the buffy coat, ACK was added for 5 min at 37 °C to remove red blood cells (RBC). To obtain fresh CD56+ primary NK cells, CD56+ cells were separated with CD56 microbeads (miltenyi biotec, 130-050-401) after the isolation step. The obtained CD56+ cells were cultured for 24 h for priming in RPMI 1640 containing 10% FBS, 1% anti–anti, and 200 U/ml interleukin (IL)-2. To obtain differentiated and exhausted primary human CD56+CD3- NK cells, we treated samples with IL-15 (Peprotech, 200-15), IL-21 (Peprotech, 200-21), and hydrocortisone (HC) (STEMCELL, ST07925) until mean day 9∼12 as previously described.30 Isolated MNC were cultured in MEM-alpha (Gibco, 12561-056) containing 10% FBS and 1% antibiotic-antimycotic. The CD56+CD3- population was checked periodically during differentiation and NK cells were used in CD56-high (>85%) and CD3-negative (<1%) conditions. Patient sample The sample was approved by the Public Institutional Review Board designated by the Ministry of Health and Welfare (IRB approval number: CNUH 2018-08-013-019). Consent was obtained from all participants. FACs and antibodies Target cells were stained with the following antibodies. TargetDyeBrandCatRRIDULBP1PER&D systemsFAB1380PAB_2687471ULBP2/5/6PER&D systemsFAB1298PAB_2214693ULBP3PER&D systemsFAB1517PAB_10719122MICA/BPER&D systemsFAB13001PAB_663947B7H6PER&D systemsFAB7144PAB_2636810NKp46PER&D systemsFAB1850PAB_2149279NKp44PER&D systemsFAB22491PAB_2149305KIR2DL1PER&D systemsFAB1844PAB_2130401KIR2DL2/3PER&D systemsFAB2014PAB_2296488KIR3DL1PER&D systemsFAB12251PAB_416835B7H6APCR&D systemsFAB7144AAB_3451988NKp30PEBD Biosciences558407AB_647240NKG2DPEBD Biosciences557940AB_396951CD70PEBD Biosciences561935AB_396158VimentinPEBD Biosciences562337AB_10897167CD45PEBD Biosciences555483AB_395875CD56FITCBD Biosciences562794AB_2737799CD3APCBD Biosciences557597AB_398626CD45APCBD Biosciences555485AB_398600CD56BV421BD Biosciences562751AB_2732054CD34PE-cy7BD Biosciences348801AB_400389HSPGPE-cy7BD Biosciences565781AB_27393517-AADPerCP-Cy-5.5Biogem62700–80AB_3451991Goat anti-mouse IgGAPCBiolegend405308AB_315011PE; Phycoerythrin, APC; Allophycocyanin, BV; Brilliant Violet, FITC; Fluorescein, PerCP-Cy-5.5; PerCP-Cyanine5.5. Fc block (BD Biosciences Cat 564220) was treated in each sample to prevent nonspecific antibody staining. Further information for material is attached in the reagent validation file. The Mean Fluorescence Intensity (MFI) ratio was obtained by dividing the mean value of the target-stained sample by the mean value of the target-unstained sample. Most of the measurements were performed using PE dye to eliminate dye differences. PE; Phycoerythrin, APC; Allophycocyanin, BV; Brilliant Violet, FITC; Fluorescein, PerCP-Cy-5.5; PerCP-Cyanine5.5. Fc block (BD Biosciences Cat 564220) was treated in each sample to prevent nonspecific antibody staining. Further information for material is attached in the reagent validation file. The Mean Fluorescence Intensity (MFI) ratio was obtained by dividing the mean value of the target-stained sample by the mean value of the target-unstained sample. Most of the measurements were performed using PE dye to eliminate dye differences. Generation of lentiviral vectors pLVX-AcGFP1-C1 (Takara, 632155) was used to construct a permanent B7H6 (NCBI accession number: NM_001202439.3) overexpressing cell line. pLKO.1–TRC cloning vector (Addgene, 10878) was used to construct a permanent B7H6 KD cell line (shRNA sequence: 5′-CCACAAAGUCUGAGAAACA-3′). pCAG-SpCas9-GFP-U6-gRNA (Addgene, 79145) was used to construct a permanent B2m knockout cell line (gRNA sequence: 5′- AGATGTCTCGCTCCGGTGGCC-3′). Constructing permanent cell lines Permanent cell lines were constructed through lentivirus infection as described previously.31 Viruses containing each vector were generated by a 3rd lentivirus generation system using TransIT2020 (Mirus, MIR5406) and OPTI-MEM (Gibco, 31985-070). The viruses were infected into each cell line using Protamine sulfate (Sigma, P3369) after infection. Infected cells were selected by Puromycin dihydrochloride (Sigma, P8833) or Green Fluorescent Protein (GFP). After selection, single clones were generated by an automated high-speed flow cytometry sorter (BD bioscience) after 24 hr. Individual clones were expanded and examined for target expression by FACs. In the case of B7H6-downregulated U937 cells, the luciferase-expressing vector was additionally infected for BLI. Transient knockdown of B7H6/ULBP1/ULBP2/MICAB/NKp30 B7H6 was transiently downregulated in Raji and Jurkat cell lines as previously described.32 Cells were washed two times with OPTI-MEM. siRNA, which targets B7H6 (siRNA sequence: 5′ CCACAAAGUCUGAGAAACA-3′), ULBP2 (siRNA sequence: 5′-GGUUAAAGCUGAUACCAAA-3′), MICB (siRNA sequence: 5′-GGAGAGGAGCAGAGGUUCA-3′), or NKp30 (siRNA sequence: 5′-GGCUGGUGGUGGAGAAAGA-3′) was transfected at a final concentration of 10uM with electroporation using the NEPA21 Electroporator (Nepagene). AccuTarget™ Negative Control siRNA (Bioneer, SN-1003) was transfected as a control. Measurements of mRNA levels, surface expression of B7H6, and NK cell-mediated cytotoxicity were conducted 24 hrs after transfection. In the experiments with progressive concentration changes of B7H6, the siRNA concentration was varied to 0, 0.5, 1, 2, 2.5, 5, 10, and 20 μM. Progressive overexpression of B7H6 and MICB B7H6 (NCBI accession number: NM_001202439.3) and MICB (NCBI accession number: NM_001289160.2 (transcript variant 2)) mRNA was synthesized through in vitro transcription (invitrogen, AM1345) and transfected into cell lines via electroporation to induce variable expression of activating ligands. Ligand expression and cytotoxicity were measured after 18 hr in mRNA transfection samples. Measuring NK cell-mediated cytotoxicity NK cell-mediated cytotoxicity was measured with the Calcein-AM (Thermofisher, C1430) release assay as previously described.32 Target tumour cells were stained with Calcein-AM, a cell-permeant dye, for 1 hr at 37 °C. The labeled cells (1 × 104 cells) and serially diluted effector cells were co-cultured in 96 well round-bottom culture plates (SPL, 34096) for 4 hr at 37 °C after staining. Released Calcein-AM into the supernatant was measured using a multimode microplate reader (Molecular Devices, San Jose, CA, USA) after removing the supernatant to 96 well black plates (SPL, 30496). The maximum level of released Calcein-AM was estimated based on the addition of 2% Triton X-100 (Merk, 108643), final volume to the target cells. The spontaneous level of released Calcein-AM was estimated by adding culture medium to the target cells. The percentage of specific lysis was calculated according to the following formula: (sample release—spontaneous release)/(maximum release—spontaneous release) × 100 (%). In addition, when measuring the cytotoxicity of 33 cell lines, we set the effector:target (E:T) ratio that can best show the difference of cytotoxicity between cell lines. In particular, since pNK kills target cells more sensitively than NK-92, we used lower ratios (0.25:1, 0.5:1, and 1:1) than NK-92's ratios (1:1, 5:1, and 10:1) in the assay. Conjugation assay Target cells (1 × 105 cells) were stained with Calcein-AM, and NK cells (1 × 105 cells) were stained with CellTrace™ Violet Cell Proliferation Kit (Thermofisher, C34571), for flow cytometry, for 1 hr at 37 °C. After co-incubation of the target cells and NK cells by time point, samples were fixed with 4% paraformaldehyde (PFA) and analyzed with FACs. The percentage of conjugation was calculated according to the following formula: (NK cell-bound target/total target) × 100 (%). Coculture assay Target cells (1 × 106 cells) were stained with CellTrace™ Violet Cell Proliferation Kit and effector cells (1 × 10ˆ6 cells) were stained with Calcein-AM. After staining, target and effector cells were cocultured for each time point (0/1/2/4 hr). After cocultivation, 7-AAD was stained for 15 min at room temperature. The 7-AAD+ population was measured by FACs. Western blot analysis NK cells cocultured with target cells were collected and lysed for Western blotting analysis as previously described.33 Briefly, cells were collected by centrifugation and resuspended in Roche complete lysis M (Merk, 719956001) to extract protein. Protein extracts were resolved on a 10% SDS-PAGE under constant voltage conditions. After electrophoresis, the protein was transferred to a transfer membrane (Merck Millipore, IPVH00010). Blots were incubated at 4 °C overnight with the following antibodies. TargetBrandCatRRIDanti-ppI3KCell Signalling TechnologyCat. 4228SAB_659940anti-pI3KCell Signalling TechnologyCat. 4292SAB_329869anti-pERKCell Signalling TechnologyCat. 9101SAB_331646anti-ERKCell Signalling TechnologyCat. 9102SAB_330744anti-GAPDHCell Signalling TechnologyCat. 5174AB_10622025After staining with primary antibodies, blots were washed three times with PBS containing 0.1% tween 20 (Duchefa biochemie, OH, USA), and incubated for 1 hr at room temperature with Goat anti-Rabbit IgG Secondary Antibody, HRP (Thermofisher scientific, 31460). Immunoreactivity was determined using Super signal west pico plus (Thermo scientific, 34580) or Immobilon Western Chemiluminescent HRP Substrate (Millipore, WBKLS0500). Results were analyzed by CS analyzer4. Further information for material is attached in the reagent validation file. After staining with primary antibodies, blots were washed three times with PBS containing 0.1% tween 20 (Duchefa biochemie, OH, USA), and incubated for 1 hr at room temperature with Goat anti-Rabbit IgG Secondary Antibody, HRP (Thermofisher scientific, 31460). Immunoreactivity was determined using Super signal west pico plus (Thermo scientific, 34580) or Immobilon Western Chemiluminescent HRP Substrate (Millipore, WBKLS0500). Results were analyzed by CS analyzer4. Further information for material is attached in the reagent validation file. Mice Animal experiments were approved by the Institutional Animal Care and Use Committee (IACUC) guidelines of the Korea Research Institute of Bioscience and Biotechnology (KRIBB) (Approval number: KRIBB-AEC-20174) and performed following institutional guidelines for animal care (National Institutes of Health, Bethesda, MD, USA). All mice were housed in a pathogen-free animal facility under a 12 hr light–dark cycle. Male mice were used for the overall study. No more than three mice were kept per cage to minimize stress during the experiment. In vivo experiments Since B7H6 is expressed specifically in humans, a humanized mouse model was required. Therefore, a humanized mouse was constructed by injecting a human tumour cell line expressing B7H6 into NSIG (NOD.Cg-Prkdcscid II2rgtm1Sug/Jic, 5–8 weeks) mice. The humanized mouse model allowed us to investigate the killing of human NK cells against human tumour cells under in vivo conditions, but it was limited in investigating the unique response of NK cells in the mouse body. Before the experiment, we established all the protocols and proceeded as follows. Due to the nature of the in vivo experiment, six mice per group were used to account for the number of outliers. Specifically, a total of NSIG mice were purchased from Saeronbio, and mice were randomly divided into two groups before tumour injection. Random numbers were generated using the standard = RAND () function in Microsoft Excel. U937-shControl-Luc (n = 6) or U937-shB7H6-Luc cells (n = 6) (1 × 106 cells) were intravenously (i.v) injected into mice (6 weeks) on day 0. pNK (1 × 107 cells) or PBS were injected i.v. on days 0, 3, 5, and 8. Successful differentiation of pNK cells was confirmed by FACs before injection. Tumour burden was estimated every 2–3 days with BLI. XenoLight D-luciferin, potassium (Perkin Elmer, 122799) was applied by intraperitoneal (i.p) injection and incubated 15 min before capturing. No mice were excluded from the experiment. Mice with BLI values that differed too much were excluded from representing BLI data, but no mice were excluded in analyzing survival rate. Blood analysis was proceeded on day 20. Blood was extracted from the tail vein. After treating ACK, 5 min at 37 °C to remove RBC, human CD45 was stained and analyzed by FACs. Survival rate analysis, BLI measurements, and blood analysis were analyzed by individuals who did not undergo tumour and NK cell injections, and the individual didn't know which group received U937_ctrl or U937_shB7H6. The study did not have a humane endpoint because of the survival analysis. To minimize potential confounders, injection and tumour growth measurements were always performed in the same order and cages for each group were always located in the same place. Quantification and statistical analyses In vitro experimental samples were tested in duplicate or triplicate, and experiments were performed three to eight times independently. Statistical significance was assessed with Pearson correlation test and Student's t-tests. The Log-rank (Mantel–Cox) Test was used to assess the significance of the in vivo survival rate. All experimental data were analyzed by GraphPad Prism (Version 5.01, GraphPad Software), and clinical data were analyzed by the R software package (ver. 3.6.3). Values of p < 0.05 were statistically significant. Roles of funding source The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. Results Only B7H6 levels among numerous activating ligands correlate with the prognosis of patients with LT To explore the impact of activating ligands on NK cell activity for patients with various types of tumour, we used the RNA expression profiles of the ligands from patients with 37 types of tumour (n = 10,595) using data from the GDC data portal (Table 1, Table 2). We observed distinct gene expression profiles between LT and ST (Fig. 3a; Appendix 1 p 2). This prompted us to perform survival analysis for the LT and ST groups separately, to assess which ligands are most sensitively associated with patient survival. For each group (LT and ST), we categorized patients based on their relative NK cell levels to confirm whether the clinical relevance of activating ligands is attributed to NK cell activity (Fig. 3b). Specifically, for each group, we quantified the proportion of NK cells in each sample via a computational quantification method, CIBERSORT. Patients were further subdivided into two categories: those with a relatively high proportion of NK cells among immune cells (referred to as NK cell-rich) and those with a relatively low proportion of NK cells (referred to as NK cell-poor) (Fig. 3b), employing various thresholds (top 50%, 40%, 30%, and 20%). After dividing patients into these subgroups, we conducted multivariate cox regression for each activating ligand to identify ligands associated with patient survival (Fig. 3b). To avoid data bias from different datasets, we divided patients into NK cell-rich and NK cell-poor groups based on the relative proportion of NK cells within each dataset. We then normalized the ligand expression levels before splitting patients into ligand-high and ligand-low groups (Appendix 1 p 3). Clinical data from patients with LT confirms the dominance of B7H6. (a) A t-SNE projection of NK cell activating ligands from patients' expression profiles (dark scale; LT, light scale; ST) (n = 10,595). (b) Schematics of the NK cell-dependent survival analysis in patients. (c-d) Dot plot of Cox hazard ratio by NK cell activating ligands in (c) NK cell, (d) CD8 T cell, and CD4 T cell-rich patients. Left: LT; Right: ST. Samples with relatively high proportions of NK cells, which were defined by the distribution of the proportion (the top 50, 40, 30, and 20%), were categorized as rich groups. The size and red color indicate low-risk value. Statistical significance was displayed only for low-risk values. The protein symbol is written in parentheses next to the gene symbol. For a detailed display of the Cox hazard ratio, see Appendix 1 pp 4–7. (e–g) KM survival curves of the NK cell-rich (the top 30%) or NK cell-poor groups (the bottom 70%) of total patients with LT and each dataset: (e) total LT; (f) PCL (MMRF-COMMPASS); (g) B-ALL (TARGET-B-ALL); (h) MBL (CGCI-BLGSP); (i) AML1 (TARGET-AML); (j) AML2 (TCGA-AML). Patients were stratified using a cutoff value of median B7H6 expression level. The NCR3LG1 gene was named the protein symbol, B7H6. ∗P ≤ 0.05, ∗∗P ≤ 0.01, ∗∗∗P ≤ 0.001, calculated by Wald test (c) and (d) and Log-rank test (e–j). Among nearly all identified activating ligands (Appendix 1 p 1, Supplemental Table S1), B7H6 was the only activating ligand whose increased expression significantly reduced the risk ratio for LT across various thresholds (Fig. 3c; Appendix 1 p 2, panel a–b). Notably, as the threshold for the NK cell-rich group increased from the top 50% to the top 20%, the risk ratio decreased more drastically (Appendix 1 p 4). This indicates that B7H6, uniquely among the 18 activating ligands examined, significantly influences the survival of patients with LT. In contrast, for NK cell-rich ST groups, no activating ligand demonstrated clear relevance to survival (Appendix 1 p 5). While MICB and B7H6 were correlated with increased survival of patients with ST, the association was weaker compared to the correlation between B7H6 and survival of patients with LT (Appendix 1 p 5). In addition, B7H6 expression was not associated with improved survival when analyzed in relation to CD4 or CD8 T cell proportions (Fig. 3c–d; Appendix 1 pp 6–7). This indicates that the prolonged survival in patients with LT who have higher B7H6 levels is primarily due to NK killing activity. To further characterize the observed association between B7H6 expression and survival, we generated KM curves for patients with LT. We divided the LT group into B7H6-high and B7H6-low groups based on the median B7H6 level. Consistent with our earlier findings (Fig. 3c), patients with high B7H6 expression in the NK cell-rich group (especially the top 30%) showed significantly improved survival (Fig. 3e, p = 0.0069, Log-rank test). This association became weaker as the threshold for the NK cell-rich group decreased from 20% to 50% (Appendix 1 p 5, panel c). Finally, for the NK cell-poor group, there was no difference in survival depending on the B7H6 expression level (Fig. 3e). This further supports a model in which higher B7H6 levels lead to prolonged patient survival via NK killing. We next explored whether this survival benefit of high B7H6 expression applies across all LT types. We analyzed survival benefits by each dataset: PCL, B-ALL, MBL, AML1, and AML2. The PCL, B-ALL, and MBL types displayed a similar trend, with patients in the B7H6-high group showing better prognoses (Fig. 3f–h, p = 0.0111, 0.0041, and 0.0332, respectively, Log-rank test). While the AML types did not exhibit a statistically significant effect, the B7H6-high group within the AML types still exhibited a trend toward better prognosis compared to the B7H6-high group in the NK cell-poor patients (Fig. 3i–j). Mathematical analysis predicts that B7H6 is a representative activating ligand inducing the NK killing response in LT Although only B7H6 was found to be significant for survival, our clinical analysis could be influenced by potential dataset bias. To determine whether the significant role of B7H6 is independent of such biases, we conducted additional in vitro, in silico, and in vivo studies. As a first step, we investigated how B7H6 expression affects the susceptibility of tumour cell lines to NK cell-mediated killing. We selected five well-investigated activating ligands, including B7H6 (B7H6, MICA/B, ULBP1, ULBP2/5/6, and ULBP3),34, 35, 36, 37, 38, 39 from those considered in the clinical study (Appendix 1 p 1, Supplemental Table S1).9,10,14, 15, 16, 17, 18 We then measured the levels of these five major NK cell-activating ligands and the NK cell-mediated cytotoxicity of 33 tumour cell lines using two types of NK cells: primary NK cells differentiated from human CB (pNK) (Appendix 1 p 8, panel a–c) and the NK-92 cell line (Appendix 1 p 8, panel d–e), with varying effector-to-target (E:T) ratios (E:T = 0.25:1, 0.5:1 and 1:1 for pNK, E:T = 1:1, 5:1 and 10:1 for NK-92, see Appendix 1 p 9 for details) (Fig. 4a–b). A mathematical modeling analysis based on data from 33 tumour cell lines predicts that of the five major NK cell-activating ligands, surface expression of B7H6 has the greatest impact on NK killing. (a) Cartoon illustrating the process for quantifying surface expression of five major NK cell activating ligands in 33 tumour cell lines and their NK-mediated cytotoxicity. (b) Activating ligand expression (MIC; MICA/B, BP1; ULBP1, BP2; ULBP2/5/6, BP3; ULBP3, B7; B7H6) and NK cell-mediated cytotoxicity (pNK; E:T ratio = 0.5:1, NK-92; E:T ratio = 5:1) across cell lines (Black: LT, Gray: ST) (n = 3∼6). Cell lines are classified by hierarchical clustering using ligand expression. (c) Pearson correlation coefficient of pNK-mediated cytotoxicity with activating ligand expression in LT and ST. (d) Schematics for model construction to capture the contribution of each ligand. (e) Killing weight of five activating ligands in each model. (f) Probability of activating ligands affecting pNK-mediated cytotoxicity estimated by GVS. (g) Measured (dots) and predicted (line) pNK-mediated cytotoxicity of LT (E:T ratio = 1:1). (h) Left: Activating ligand expression and pNK-mediated cytotoxicity of additional LT cell lines. Right: pNK-mediated cytotoxicity prediction using only B7H6 (E:T ratio = 0.5:1). ∗P ≤ 0.05, ∗∗P ≤ 0.01, ∗∗∗P ≤ 0.001, ∗∗∗∗P ≤ 0.0001 calculated by Pearson correlation (c). Consistent with the bulk mRNA sequencing data (Fig. 3a), hierarchical clustering40 based on surface ligand expression levels separated the cell lines into two subgroups, LT and ST (Fig. 4b). To understand how these ligands contribute to NK killing activity, we analyzed the relationship between the expression levels of the five major NK cell-activating ligands and NK cell-mediated cytotoxicity in each group separately (LT and ST). For the LT group, B7H6 emerged as the sole activating ligand whose surface expression strongly correlated with cytotoxicity mediated by both pNK and NK-92 cells (Fig. 4c; Appendix 1 p 9, panel a–c) across varying E:T ratios, which is consistent with the survival analysis (Fig. 3). As other ligands (MICA/B, ULBP1, ULBP2/5/6, and ULBP3) share a common receptor, NKG2D, we also investigated the correlation between the combined expression of NKG2D ligands and NK-mediated cytotoxic value. However, their combined expression did not correlate significantly with cytotoxicity (p > 0.8 for every E:T ratio, Pearson's correlation test). Additionally, we also examined three other activating ligands that tend to work effectively as activating ligands in our analysis (Fig. 3). These ligands did not exhibit any significant correlation, further underscoring the importance of B7H6 (Appendix 1 p 9, panel d). To determine whether this dominance is restricted to short-term cytotoxicity (less than 4 h), we extended our investigation to longer-term cytotoxicity. The dominance of B7H6 remained robust over the extended time period of 24 hr (Appendix 1 p 9, panel e). In contrast, no such strong correlation was observed in ST (Fig. 4c; Appendix 1 p 9, panel f–h). We further investigated the nonlinear combined contribution of the five activating ligands to NK cell-mediated cytotoxicity, which the simple correlation cannot detect, by developing mathematical models (n = 25) based on all possible combinations of the five activating ligands (Fig. 4d). Across all combinations, models that accurately reproduced NK cell-mediated cytotoxicity for LT consistently showed B7H6 as a highly effective ligand in influencing cytotoxicity, while MICA/B was identified as only weakly effective (Fig. 4e). Notably, models including B7H6 demonstrated up to a 25% reduction in error, as measured by AIC, compared to those that did not include B7H6 as an effective ligand. In contrast, the other three ligands did not show a consistent pattern of effectiveness (Fig. 4e). The strong role of B7H6 in LT was also observed when the models were fitted to the NK-92-mediated cytotoxicity (Appendix 1 p 10, panel a). On the other hand, in ST, no ligand, including B7H6, showed a significant contribution to NK killing (Fig. 4e; Appendix 1 p 10, panel a). The dominant role of B7H6 in LT but not in ST was also supported by the GVS, which quantifies the statistical significance of each ligand to predict cytotoxicity (Fig. 4f; Appendix 1 p 10, panel b–c).28 Moreover, 77% of the variance in pNK-mediated cytotoxicity for LT was well explained by a model focusing solely on the relationship between B7H6 expression and NK cell-mediated cytotoxicity in LT (Fig. 4g). To test whether the dominant role of B7H6 is due to a bias in the dataset we used (Fig. 4b), we utilized additional LT cell lines (n = 5) for the validation. The pNK-mediated cytotoxicity of these additional cell lines was also accurately predicted by the model (Fig. 4h). The dominant effect of B7H6 was confirmed in various LT cell lines To support the clinical and in silico analyses that B7H6 plays a predominant role in the NK killing decision, we knocked down (KD) B7H6 in various LT cell lines with high expression of B7H6 (U937 in AML, Raji in B-cell lymphoma, and Jurkat in T-ALL) and assessed its effect on the NK killing response, including the formation of cellular conjugates between NK cells and tumours, an indicator of the initial stage of the killing process where the NK cell forms an immune synapse with the target, and the phosphorylation of NK cell activation signals pI3K and ERK (Fig. 5a; Appendix 1 pp 11–12, panel a). B7H6 KD resulted in a significant decrease in pNK-mediated (U937; p < 0.0001, Raji; p < 0.0001 Jurkat; p < 0.001, two-tailed Student's t-test) and NK-92-mediated cytotoxicity (U937; p < 0.001, Raji; p < 0.001, Jurkat; p < 0.0001, two-tailed Student's t-test) (Fig. 5b; Appendix 1 pp 11–12, panel b), the formation of cellular conjugates between NK cells and tumours (U937; p = 0.005 for 2 hr, p = 0.004 for 3 hrs, p = 0.014 for 4 hr, Raji; p = 0.0068 for 1 hr, p = 0.0017 for 2 hr, p = 0.00625 for 4 hr, two-tailed Student's t-test) (Fig. 5c; Appendix 1 p 11–12, panel c), and phosphorylation levels of the NK cell activation signaling molecules,41,42 pI3K and ERK (Appendix 1 pp 11–12, panel d). Consistently, when we overexpressed (OE) B7H6 in the low-B7H6 LT cell line THP-1 (Appendix 1 pp 11–12, panel e), we observed both an increase in NK cell-mediated cytotoxicity (pNK; p < 0.01, NK-92; p < 0.001, two-tailed Student's t-test) (Appendix 1 pp 11–12, panel f–g) and cellular conjugation (p = 0.0013 for 4 hr, two-tailed Student's t-test) (Appendix 1 pp 11–12, panel h). In addition, when co-cultured with pNK cells, B7H6 KD U937 cells showed low dead cell populations (7-AAD+) compared to controls (p = 0.0064 for 1 hr, p = 0.0122 for 2 hr, p < 0.0001 for 4 hr, two-tailed Student's t-test) (Appendix 1 pp 11–12, panel i–j). Consequently, we conclude that NK cells target and kill specifically B7H6-high LT, which is consistent with the clinical analysis (Fig. 3c–j). B7H6 is a representative activating ligand in LT cell lines. (a–c) NK killing response of LT cell lines after B7H6 KD (red) or control (blue). (a) Surface expression of B7H6. Left: U937 (n = 3). Middle: Raji (n = 3). Right: Jurkat (n = 3). (b) pNK-mediated cytotoxicity. Left: U937 (n = 4). Middle: Raji (n = 4), Right: Jurkat (n = 6). (c) NK-bound tumour population of U937 cells at the indicated time points after co-culture (n = 5). The data represent mean ± s.e.m. (d) Surface expression of B7H6 in patients with AML (n = 6). (e) Gating strategy of dead cells (7-AAD+) in AML blasts (pNK−CD45dimCD34+). (f) Population of 7-AAD+ in AML cells co-cultured with pNK_scramble or pNK_siNKp30 (n = 6). (g) Experimental scheme for (h) and (i). (h) NK cell-mediated cytotoxicity in U937 (n = 20) and Raji (n = 20) cells expressing different levels of B7H6 and the predicted cytotoxicity based on the mathematical model. To express B7H6 at different levels, B7H6 was KD with B7H6-siRNA. X-axis; surface expression of B7H6 (E:T ratio = 1:1). Y-axis; pNK-mediated cytotoxicity. (i) NK-mediated cytotoxicity in THP-1 cells expressing different levels of B7H6 (n = 5) and MICA/B (n = 7). To express B7H6 and MICA/B at different levels, B7H6 or MICB was overexpressed through mRNA transfection. X-axis; surface expression of ligands (B7H6 or MICB) (E:T ratio = 1:1). Y-axis; pNK-mediated cytotoxicity. ∗P ≤ 0.05, ∗∗P ≤ 0.01, ∗∗∗P ≤ 0.001, and ∗∗∗∗P ≤ 0.0001, calculated by two-tailed Student's t-test (a), (b) and (c) or paired t-test (f). For detailed description of the mathematical model in (h) and (i), see methods. NK killing is related not only to the expression of B7H6 but also to its interaction with its cognate receptor, NKp30. To determine whether the B7H6-NKp30 axis effectively regulates NK killing, we examined the NKp30 KD pNK cells (Appendix 1 p 13, panel a)-mediated cytotoxicity changes on B7H6 KD LT cells (Raji and U937) (Appendix 1 p 13, panel b). At this time, NKp30 KD pNK cells showed cytotoxicity values that were independent of the expression level of B7H6 (Raji; p = 0.69, U937; p = 0.49, two-tailed Student's t-test) (Appendix 1 p 13, panel c), different from control pNK cells (Raji; p = 0.015, U937; p = 0.002, two-tailed Student's t-test), indicating that the B7H6-NKp30 axis effectively influences NK killing. To confirm the influence of the B7H6-NKp30 axis on NK killing in patient cells, we examined the population of dead cells (7-AAD+) after coculture of NKp30 KD pNK cells with patient-derived AML cells (Fig. 5d–f, Appendix 1 p 13, panels d–e). These AML cells exhibited a significantly lower percentage of dead cells compared to the control group (p = 0.04, paired t-test) (Fig. 5f), supporting the effect of the B7H6-NKp30 axis in NK killing against LT cells. We used cytokine-differentiated human pNK cells and NK-92 cell lines to assess the importance of B7H6. To explore whether pNK cells in their natural state, without cytokine differentiation, have a distinct receptor subset compared to these two types of NK cells, we examined their NK cell receptor expression levels. We found that fresh pNK cells without cytokines (Appendix 1 p 14, panel a) had lower levels of activating receptors and higher levels of inhibitory receptors than differentiated NK cells (Appendix 1 p 8, panels a–c). To determine whether these differences in receptor subsets influence the impact of B7H6 on NK cell-mediated killing, we measured the cytotoxicity of fresh pNK cells against B7H6 KD Raji cells. The results showed that fresh pNK cells also experienced a significant decrease in cytotoxicity when B7H6 was downregulated (Appendix 1, p. 14, panel b; Pearson's correlation = 0.90, p-value = 0.04), indicating that all three types of NK cells with different receptor subsets rely on B7H6-dependent killing. We further investigated the amount of B7H6 expression needed to achieve optimal NK killing capacity. To do so, we conducted experiments with varying levels of B7H6 expression in two different LT cell lines (U937 and Raji) and assessed NK killing at each B7H6 level (Fig. 5g–h). Surprisingly, both U937 and Raji exhibited a sharp increase in NK killing when B7H6 expression was between MFI ratios of 2.5 and 4.5 (Fig. 5h). This suggests that despite the diversity among LT subtypes, a similar amount of B7H6 is needed to trigger optimal NK cell killing. Finally, we conducted a comparative analysis to assess the impact of modulating B7H6 on NK killing efficiency compared to MICB, one of the well-known activating ligands. Specifically, we modulated differential expressions of B7H6 and MICB in THP-1 and evaluated their sensitivity to NK killing (Fig. 5g). Compared to B7H6, varying MICB expression had a lesser effect on NK killing against THP-1 cells, suggesting a less pronounced influence on NK cells (Fig. 5i). In both Jurkat and Raji cells, downregulation of ULBP2 and MICAB led to a reduction in NK cell killing, but this effect was less pronounced than when B7H6 was downregulated. In Jurkat cells, the reduction in killing was significant for B7H6 (p = 0.003, two-tailed Student's t-test) and MICAB (p = 0.0038, two-tailed Student's t-test), but not for ULBP2 (p = 0.157, two-tailed Student's t-test) (Appendix 1 p 15, panels a–b). Similarly, in Raji cells, downregulation of B7H6 significantly reduced NK killing (p = 0.02, two-tailed Student's t-test), while ULBP2 had a weaker, non-significant effect (p = 0.108, two-tailed Student's t-test) (Appendix 1 p 15, panels c–d). This observation aligns with the clinical data analysis (Fig. 3c–d) and the predictions derived from the mathematical analysis that B7H6 has a greater effect on NK killing than other activating ligands including MICA/B and ULBP 1, 2, and 3 (Fig. 4d–f). NK cells sensitively target B7H6-high tumours in LT-xenograft mice To explore the anti-tumour activity of NK cells against B7H6-high LT in vivo, we established LT-xenograft mouse models by transplanting either control (shControl-Luc) or B7H6-KD (shB7H6-Luc) U937 cells (Fig. 6a). In addition, we intravenously injected human pNK cells, which were expanded in vitro with cytokines (IL-15, IL-21, and HC) over four doses across 8 days. After tumour transplantation, tumour growth was monitored by blood analysis (on day 20 after post-transplant), BLI every 2–4 days, and survival analysis. Treatment of mice with pNK resulted in a reduction of CD45+ (human leukocyte common antigen) cells in the blood (Fig. 6b) of control U937-transplanted mice (p < 0.0001, two-tailed Student's t-test). In addition, the tumour burden, as measured by BLI, decreased in pNK-treated mice compared to those treated with PBS (Fig. 6c). The median survival of mice transplanted with shControl-Luc cells increased from 20.5 days to 24 days following pNK treatment (p = 0.0022, ordinary one-way ANOVA) (Fig. 6d), highlighting the efficacy of NK cells. However, the changes in tumour growth in shB7H6-Luc-transplanted mice following pNK treatment were minimal (Fig. 6e–g). Treatment of shB7H6-Luc-transplanted mice with pNK cells reduced the CD45+ cells in the blood (p = 0.035, two-tailed Student's t-test), but the reduction was not significant compared to control mice (Fig. 6e). Furthermore, there was no significant difference in tumour burden, as assessed by BLI (Fig. 6f), or in survival (Fig. 6g). These results suggest that NK cells can effectively target B7H6-high LT cells also in vivo. High-B7H6 tumours sensitively targeted by NK cells in leukemia xenograft mice. (a) Leukemia-xenograft models were constructed by intravenously injecting U937 into NSG mice (n = 6). pNK or PBS were further injected into mice at D0, D3, D5 and D8. (b–g) Comparison between the impact of NK cells in U937 control injected mice (b–d) and B7H6-KD injected mice (e–g). (b and e) Population of hCD45 in mouse blood (n = 3∼5) after pNK (closed circle) and PBS injection (open circle). The data represent mean ± s.e.m. (c and f) BLI (four representative mice). (d and g) KM survival curves (n = 6) of pNK-injected mice (closed circle) and PBS-injected mice (open circle). ∗P ≤ 0.05, ∗∗P ≤ 0.01, and ∗∗∗∗P ≤ 0.0001 calculated by two-tailed Student's t-test (b) and (e) or ordinary one-way ANOVA (d) and (g). Discussion Our study reveals B7H6 as a predominant factor in promoting NK killing of LT cells. Through our comprehensive analysis of pan-cancer data, we have elucidated that individuals with elevated levels of B7H6 manifest a more favorable prognosis compared to those with lower levels. The dominance of B7H636,43, 44, 45 in patients with LT is further supported by previous studies where expression levels of NKp30, the specific receptor for B7H6, were found to correlate with the survival of patients with LT.46 This was further supported by in silico analysis based on the data from 33 tumour cell lines of various origins, which predicted that B7H6 has a dominant effect on NK killing against LT, which we further confirmed with in vitro and in vivo experiments. This suggests that B7H6 is the sole feasible target candidate, at least among the 18 activating ligands we examined, for treating patients with LT using NK cell-based therapies. In addition, because B7H6 is expressed specifically in tumours, targeting B7H6 has the added benefit of providing a safe and effective treatment that targets and kills tumour cells without killing normal cells. This study highlights the potential role of B7H6 expression in selecting patients for NK cell-based immunotherapy in LT (Fig. 3). Previous studies have shown that B7H6 is expressed in many different tumour types, including lymphoma and leukemia.47 Our findings suggest that patients with high B7H6 expression in their tumour cells may benefit more significantly from this therapy (Fig. 3c). This aligns with recent research highlighting B7H6's role in promoting NK cell activation and tumour cell death, particularly in hematological malignancies.48 Therefore, B7H6 expression could serve as a potential biomarker to predict a patient's response to NK cell-based therapy for LT. However, some AML datasets show bad prognoses with B7H6 expression, which is likely due to B7H6's intrinsic cell signaling (Fig. 3i). B7H6 promotes cell proliferation and migration via various signaling pathways such as STAT3 and RAS in hematological tumour cells.49,50 It is also associated with immune cell activation through genes like MYC, IFNG, TNF, JUN, and FOS.48,51 Given these findings, while B7H6 generally promotes favorable outcomes through NK cell activation, its role in some AML cases may be more complex, potentially driving tumour progression through intrinsic signaling pathways. This complexity underscores the need for further investigation into the context-dependent effects of B7H6 in AML. To identify and confirm a major ligand, we primarily utilized two types of NK cells: differentiated primary NK (pNK) cells from umbilical CB and the NK-92 cell line. Despite differences in NK receptor subsets (Appendix 1 p 8, panels a–e), results from both NK cell types consistently indicated that B7H6 strongly induces NK cell-mediated killing. Notably, freshly isolated pNK cells from CB exhibited B7H6-dependent cytotoxicity even with low expression levels of activating receptors, such as NKp30. There was a strong correlation between B7H6 expression and the cytotoxicity levels mediated by fresh pNK cells (Appendix 1 p 14, panels a–b). In patient analyses, those with higher NK cell counts in their peripheral blood mononuclear cells (PBMCs) demonstrated a lower hazard ratio when B7H6 expression was high (Fig. 3e–i). This suggests that B7H6 may similarly influence NK cell-mediated killing in PBMCs, regardless of differences in NK cell subsets. This suggests that B7H6 may influence NK cell-mediated killing across various NK cell subsets, but it remains possible that it preferentially binds different ligands in different subsets. In contrast, the efficacy of NK cells for treating patients with low B7H6 levels may be limited (Fig. 3e). A potential direct therapy could involve increasing B7H6 expression in these patients. This could be achieved by utilizing an oncolytic virus to deliver B7H6 to tumour cells, thereby enabling NK cells to effectively target the tumours. However, since B7H6 signaling itself can enhance tumour cell proliferation,47,52 it would be necessary to use a mutant form of B7H6 that prevents this intrinsic signaling and avoids promoting tumour growth. Alternatively, to increase the therapeutic effect without modulating the expression of B7H6, it is recommended to use CAR-NK cells targeting B7H6. Our in vitro experiments (Fig. 5h) indicate that the killing activity of natural NK cells requires a certain threshold of B7H6 expression (MFI<3.5) to effectively target LT cell lines. CAR-NK cells that specifically target B7H6 will be able to sensitively detect even tumour cells that express less B7H6 than a certain threshold of B7H6 expression (MFI < 3.5). Another way of exploiting the potential of the B7H6-NKp30 pair for CAR-NK is to apply the structure of NKp30 to CAR. The robust NK cell response to B7H6-high LT cells suggests that the activation mechanism via its receptor, NKp30, is highly potent. Therefore, our findings suggest that alongside B7H6, NKp30 could serve as a valuable target for CAR design in NK cells. This approach may also address a current challenge in CAR-NK development by identifying the most potent target for CAR construction. Unlike LT, no ligand could be identified that dominantly affected NK killing in ST because activating ligands showed a correlation with NK killing that varied depending on the conditions (Fig. 4c; Appendix 1 p 9, panel d). Even the simple summation of activating ligand expression failed to accurately predict pNK-mediated killing in ST (p > 0.2 for every E:T ratio, Pearson's correlation test). This indicates that a complex synergistic effect of ligands may regulate NK cell activity in ST. Indeed, a properly weighted sum of the expressions of MICA/B, B7H6, and ULBP1 showed a higher correlation than the individual values in ST (Appendix 1 p 9, panels i–j). Analysis of the clinical data from patients with ST also showed that both MICB and B7H6 affected survival, which was different in patients with LT (Appendix 1 p 5). Therefore, unlike LT, to understand the mechanism of NK killing in ST, it is essential to investigate both the expression of multiple activating ligands and their synergistic effects. Investigating synergistic effects of activating ligands in ST might be possible by applying scaled-up data with a sufficient number of tumours to our extendable framework in the in silico analysis. The strength of our study is that we analyzed the effects of several activating ligands using a variety of methods, including clinical data analysis and in vitro/in vivo/in silico analysis. By using multiple methods, we were able to quantify and compare the relative strengths of the ligands and validate our results in several ways. Although our study focused on five major ligands out of the 18 known activating ligands, our mathematical model is scalable, allowing for the analysis of a larger number of ligands across a broader range of cell lines. Expanding the mathematical analysis to incorporate a more extensive set of ligands would be a valuable direction for future research. Our study has certain limitations. We utilized bulk RNA sequencing data from multiple datasets which we analyzed with CIBERSORT software. However, CIBERSORT is unable to distinguish NKp30-expressing NK cells within cell types. This limitation prevents us from fully confirming the dominant effect of B7H6 in the clinical analysis of patients with LT. Additionally, due to the characteristics of RNA data, it is not possible to determine the absolute amount of B7H6 expression and the number of NK cells required for B7H6 to affect the survival of patients with LT. This is because activating ligands are likely regulated at various levels, from post-transcription to post-translation, including RNA splicing, transport to the cell surface, retention in the ER/Golgi, or secretion through shedding or exosomes. Therefore, it is necessary to utilize immunohistochemistry data to confirm the relationship between the protein amount of B7H6 and the killing of NK cells. In addition, it is necessary to investigate the surface expression of B7H6 in blasts from patients with LT to establish the expression level of B7H6 as a criterion for NK cell therapy in clinical practice. Furthermore, since we do not have a dataset of patients with hematologic malignancy who received NK cells and concurrently confirmed the expression of B7H6, we were unable to investigate the function of B7H6 under direct NK cell therapy conditions. Finally, the activation of NK cells is not determined by activating signals alone, but by the interaction of activating and inhibitory signals. In this paper, we only addressed the relative strength of activating signals, but addressing the interaction of activating and inhibitory signals would be a good future study. In conclusion, our systematic approach showed that only B7H6 strongly triggers NK killing in patients with LT among nearly all activating ligands (n = 18). The significance of this study is that, unlike most other studies that analyze activating ligands separately, we took a systematic approach and analyzed most of the activating ligands together to confirm the predominance of B7H6. We propose B7H6 as the sole feasible therapeutic target and a predictive marker for response to NK cell-based therapy. Contributors T.D.K. and J.K.K. conceptualized the study and acquired funding. S.L. and S.J.C. developed the methodology. S.J.C and I.H.J. developed the software. I.H.J. curated, formally analyzed, accessed, verified, and visualized clinical data. S.J.C. curated, formally analyzed, accessed, verified, and visualized mathematical modeling data. S.L. curated, formally analyzed, accessed, verified, and visualized in vitro and in vivo experimental data. S.L., S.J.C., J.K.K., and T.D.K. wrote the original draft. T.D.K. and J.K.K. were project administrators and supervised the study. T.D.K. provided study resources. All authors contributed to investigation. All authors revised the manuscript. All authors had full access to all the data in the study and had final responsibility for the decision to submit for publication. All authors read and approved the final version of the manuscript. Data sharing statement All data used for mathematical models, statistical analyses, and clinical data analysis from this study is available in the appendix and supplementary materials. Computational codes used in this study are available in the following database: https://github.com/Mathbiomed/NK_ActL. Declaration of interests All authors declare no competing interests. Acknowledgements We especially thank the late S. H. Ha for helpful comments and for discussing mathematical modeling by using multi-dimensional biological data. We thank J. H. Lee for providing the pCAG-SpCas9-GFP-U6-gRNA vector. We thank Sunghwan Bae (Bstar Artwork) for scientific illustration. We also thank Life Science Editors for editorial assistance. This work was supported by the National Research Council of Science & Technology (NST) grant (CAP-18-02-KRIBB, GTL24021-000), a National Research Foundation grant (2710012258, 2710004815), and an Institute for Basic Science grant (IBS-R029-C3). Footnotes Supplementary data related to this article can be found at https://doi.org/10.1016/j.ebiom.2024.105459. Contributor Information Jae Kyoung Kim, Email: jaekkim@kaist.ac.kr. Tae-Don Kim, Email: tdkim@kribb.re.kr. Supplementary data Supplemental Appendix 1 Supplemental Appendix 2 Reagent validation file Data_S1_revision Data_S2_revision Supplemental western blots References Associated Data Supplementary Materials Supplemental Appendix 1 Supplemental Appendix 2 Reagent validation file Data_S1_revision Data_S2_revision Supplemental western blots