Purpose:

PD-(L)1–blocking antibodies have clinical activity in metastatic non–small cell lung cancer (NSCLC) and mediate durable tumor remissions. However, the majority of patients are resistant to PD-(L)1 blockade. Understanding mechanisms of primary resistance may allow prediction of clinical response and identification of new targetable pathways.

Experimental Design:

Peripheral blood mononuclear cells were collected from 35 patients with NSCLC receiving nivolumab monotherapy. Cellular changes, cytokine levels, gene expression, and polymorphisms were compared between responders and nonresponders to treatment. Findings were confirmed in additional cohorts of patients with NSCLC receiving immune checkpoint blockade.

Results:

We identified a genetic variant of a killer cell immunoglobulin-like receptor (KIR) KIR3DS1 that is associated with primary resistance to PD-1 blockade in patients with NSCLC. This association could be confirmed in independent cohorts of patients with NSCLC. In a multivariate analysis of the pooled cohort of 135 patients, the progression-free survival was significantly associated with presence of the KIR3DS1 allele (HR, 1.72; 95% confidence interval, 1.10–2.68; P = 0.017). No relationship was seen in cohorts of patients with NSCLC who did not receive immunotherapy. Cellular assays from patients before and during PD-1 blockade showed that resistance may be due to NK-cell dysfunction.

Conclusions:

We identified an association of the KIR3DS1 allelic variant with response to PD-1–targeted immunotherapy in patients with NSCLC. This finding links NK cells with response to PD-1 therapy. Although the findings are interesting, a larger analysis in a randomized trial will be needed to confirm KIRs as predictive markers for response to PD-1–targeted immunotherapy.

This article is featured in Highlights of This Issue, p. 2941

Translational Relevance

Only a minority of patients with NSCLC respond to anti–PD-(L)1 therapy. The identification of resistance mechanisms to anti–PD-(L)1 therapy can identify biomarkers of response and potential new targets for cancer immunotherapy. Here, we identify an allelic variant of an NK-cell receptor to be associated with resistance to PD-1–targeted immunotherapy in NSCLC. On the basis of this finding, further exploratory analyses show that NK cells might be potential mediators of resistance to PD-1–blocking antibodies. The described KIR variant is not associated with outcome differences in patients undergoing chemotherapy for stage IV or in the potentially curative setting of stage I–III disease.

Adaptive immune escape of tumor cells by upregulation of ligands for programmed death 1 (PD-1) has been identified as an important and targetable pathway in many cancer types (1, 2). Many clinical trials have demonstrated that targeting PD-1 or its ligand PD-L1 with immune checkpoint inhibitors (ICI) can restore antitumor immunity and induce durable remission (1, 2). In patients with non–small-cell lung cancer (NSCLC), several trials have demonstrated a survival benefit with PD-(L)1–blocking antibodies in patients with metastatic (3–7) and locally advanced NSCLC (8). Yet, most patients have primary resistance to PD-(L)1 blockade: approximately 20% to 25% of unselected patients with NSCLC achieve an objective response while 40% to 50% have progressive disease. Understanding mechanisms of primary resistance to PD-(L)1 blockade is important in the search for potentially targetable pathways and may ultimately allow discrimination between patients that derive benefit from PD-(L)1 blockade and those requiring combination strategies or alternative treatment approaches.

Different mechanisms of primary and acquired resistance to PD-(L)1 blockade have been described, including defects in human leukocyte antigen (HLA) class I presentation, cellular adaptation of the tumor microenvironment, low mutational burden, and mutations or defects in the IFNγ pathway (9). To date, most of the focus of investigations has been on how T cells interact with tumor cells, while other immune cells may be critical as well. Recently, cross-talk between natural killer (NK) and dendritic cells has been associated with response to PD-1 blockade in melanoma (10). In this study, we aimed to identify mechanisms of resistance and potential new targetable pathways in patients with metastatic NSCLC that received PD-1 blockade using a comprehensive analysis of peripheral blood from responding and nonresponding patients. We found that the allelic variant KIR3DS1 is associated with primary resistance to PD-1 blockade and is mainly expressed on NK cells. KIR3DS1 is an alternative allele at the KIR3DL1 locus coding for an activating receptor unlike the KIR3DL1 alleles that encode for inhibitory receptors. KIR3DS1 mainly differs from KIR3DL1 in its signaling binding domains at the cytoplasmic tail (11). In addition, peripheral NK-cell reactivity is enhanced in patients responding to PD-1 blockade, suggesting that NK cells are involved in response to immunotherapy.

Patients and sample preparation

Serum and peripheral blood mononuclear cells (PBMC) were prospectively collected from 35 patients with metastatic NSCLC (discovery cohort) immediately before the first administration and after two cycles (week 4) of treatment with PD-1 blockade at the University Hospital Basel (Basel, Switzerland). All patients that were planned for immunotherapy after progression on chemotherapy were asked to participate. Ethics approval was obtained from the local ethical committee to analyze the tissue and blood samples (Ethikkomission Nordwestschweiz, EK321/10) and written informed consent was obtained from all patients prior to sample collection in accordance with the Declaration of Helsinki. Objective radiologic response was determined by RECIST 1.1 by a thoracic radiologist, with scans performed every 6–8 weeks. Single-cell suspensions of primary NSCLC tumor samples were prepared by digestion after surgical removal.

For validation and comparison, we examined whole-exome sequencing data from a previously published cohort of patients with metastatic NSCLC treated with pembrolizumab (12). In addition, we examined a third independent cohort of patients from St. Gallen with NSCLC with PD-(L)1 inhibitors by PCR (approved by the local ethical committee, Ethikkommission Ostschweiz, EKOS). Finally, early-stage NSCLCs in the TracerX study cohort not treated with immunotherapy were analyzed by whole-exome sequencing (13).

Cell lines

The K562 cell line (ATCC CCL-243) was received from the ATCC in 2010. Cells were confirmed to be negative for Mycoplasma by PCR as described (14) after every freeze–thaw cycle and then passaged every 2–3 days for a maximum of 10 passages.

Multicolor flow cytometry

PBMCs were isolated as published previously (15). Multicolor flow cytometry was performed after staining dead cells with Live/Dead Zombie UV and various panels of antibodies (see Appendix p 2). Corresponding isotype antibodies were used as controls. Tumor samples were analyzed by flow cytometry with a Fortessa LSR II flow cytometer (BD Biosciences). For intracellular cytokine staining, cells were fixed and permeabilized using IC-Fixation or Fix/Perm Kit (eBioscience). For RNA sample collection, peripheral CD8+ T cells were sorted by fluorescence-activated cell sorting using a BD FACS Aria III.

Molecular analysis of KIR3DS1/L1 and HLA locus

The KIR3DS1/L1 locus was analyzed from germline DNA isolated from peripheral blood by PCR as described previously (16). The HLA-A, -B, and -C were typed by molecular methods (FluoGene;Inno-train) and the presence of Bw4 and Bw6 within the HLA-A and HLA-B determined.

Bioinformatic analysis of KIR3DS1/L1 and HLA status from whole-exome sequencing data

To identify polymorphisms and genetic variants in KIR alleles from whole-exome sequencing data, germline whole-exome sequencing data from the TracerX dataset EGAD00001003206 was obtained under a data access agreement through the European Genome Archive (https://ega-archive.org). Sequence alignments were first sorted by read name using SAMtools 1.3.1 (17) and separated into paired FASTQ files using picard 2.9.2 (http://broadinstitute.github.io/picard). Parts of the KIR genome region (GRCh37; KIR3DL1 ± 100bp chr19:55327793-55342333, KIR3DS1 ± 100bp chr19_gl000209_random:69997-84758) were extracted from the provided sequence alignments using SAMtools 1.3.1. We used BLAST to identify three unique sequence regions distinguishing KIR3DS1/L1 (Supplementary Data). We then used featureCounts (18) to quantify the number of sequencing reads from each individual normal sample aligning to these three distinguished regions (primary, concordant paired-end alignments, having >90% region overlap, and no CIGAR operations). We calculated the LogR ratio for the individual counts and used these measures to assess the KIR3DS1/L1 genotype (the script for the analysis is deposited at https://github.com/mui-icbi/KIRcaller). We used the “Pushing Immunogenetics to the Next Generation” (PING) method developed by Norman and colleagues (19) in an independent cohort (TracerX) to cross-validate the genotyping of the KIR alleles.

RNA sequencing and analysis

Library was constructed with the Illumina TruSeq Nano Kit and sequenced with the Illumina HiSeq 3000. Reads were aligned to the UCSC hg19 version of the human genome using the spliced read aligner STAR (2.5.1). The R/Bioconductor package QuasR (version 1.10.1, R version 2.3.4) was used to assess read and alignment quality. Gene expression was quantified by the qCount function of QuasR using an exon-union model of hg19 RefSeq genes (downloaded from UCSC 2015-02-09). The R package edgeR was used for differential gene expression analysis. For within time-point comparisons, a linear model including the combined factor “response” and “time” was fitted to the raw counts using function glmFit and differential expression between responders [i.e., partial response (PR) according to RECIST 1.1 criteria] and nonresponders [i.e., progressive disease (PD) by RECIST 1.1] at time points before and during treatment were evaluated by a likelihood ratio test using function glmLRT from edgeR. P values were adjusted by controlling the FDR. For within-response group comparisons, a second model including the factor “patient id” was fitted and the differential temporal effects between both groups were evaluated by likelihood ratio tests. The results of the RNA sequencing have been deposited on Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo) under the number GSE111414.

TML NGS library preparation and sequencing

The Oncomine Tumor Mutation Load Assay (TML assay) is a PCR-based next-generation sequencing assay. The panel covers 1.7 megabases (Mb) of 409 genes that are associated with cancer. The panel consists of 15,513 primer pairs that are evenly distributed in two pools. The genomic regions covered by the panel are divided into 1.2 Mb of exonic and 0.45 Mb of intronic sequences.

Utilizing Ion AmpliSeq library preparation technology, library preparation requires only 20 ng of input DNA (10 ng/primer pool) extracted from formalin-fixed, paraffin-embedded cancer specimens. Briefly, a multiplex PCR is performed to amplify the target regions in the genome. The primer sequences are then digested from the ends of the PCR products using a proprietary enzyme (Thermo Fisher Scientific). Finally, barcoded sequencing adapters are ligated to the enriched targets, and the final libraries are cleaned up using AmpureXP beads (Agencourt). Template preparation was performed on an Ion Chef instrument and the loaded 540 chips were sequenced on an Ion S5 Semiconductor Sequencer (Thermo Fisher Scientific). Minimal coverage (>500×) and uniformity (>90%) values were used to ensure that only high-quality data were used for downstream analysis.

Reads were aligned to the hg19 reference genome using Torrent Suite Software 5.8 and BAM files were transferred to Ion Reporter Software 5.6 for variant calling and secondary analysis including tumor mutation burden (TMB) calculation.

TML score calculation

TMB was calculated using TML analysis workflow “Oncomine Tumor Mutation Load – w1.1” on the Ion Reporter Software v5.6. Mutation load using this workflow is defined as nonsynonymous and synonymous somatic mutations per covered megabases, including missense and nonsense point mutations.

Briefly, after variant calling, variants were filtered by a minimal allelic frequency of 10% and germline variants were removed: variant alleles present from the 1,000 Genome Project, NHLBI GO Exome Sequencing Project, and Exome Aggregation Consortium were filtered out. The remaining somatic variants were the used to calculate the TMB value by dividing the number of variants by the number of covered megabases.

NK cell degranulation and cytotoxicity assay

Whole-bulk PBMCs containing NK cells from patients with NSCLC before and during PD-1 blockade treatment were thawed. A total of 1 × 104 K562 target cells were cocultured with 2 × 105 PBMCs (effector to target ratio 20:1) in the presence of fluorochrome-coupled anti-CD107a antibody. After 1 hour of coincubation, Monensin and Brefeldin were added to the culture to accumulate cytokines intracellularly. After 5 hours of additional coincubation, the cells were fixed and stained for IFNγ and TNFα. The samples were then analyzed by flow cytometry gated on CD56+ CD3 NK cells. NK effector functions were further assessed by a cytotoxicity assay performed using whole-bulk PBMCs at an effector to target ratio of 20:1 using K562 cells previously labeled by Vybrant DiO Dye (Thermo Fisher Scientific). The cytotoxic capacity of the NK cells was determined after 6 hours coincubation by measuring the percentage of PI+/DiO dye+ K562 target cells by flow cytometry taking into account nonspecific K562 cell death.

Multiplex cytokine measurements

Serum samples before and during treatment were used. NK and T-cell cytokines were measured by a flow cytometry bead assay that allows measurement of 13 cytokines (LEGENDplex; BioLegend). Binding of cytokines to the beads was measured using a Fortessa II analyzer (BD Biosciences).

HLA-F IHC

Tissue sections from patients were analyzed by IHC for HLA-F expression. Paraffin sections were incubated with a monoclonal rabbit anti-human HLA-F IgG (Abcam, clone EPR6803).

Data availability

Raw and processed RNA sequencing data from our discovery cohort are deposited on GEO under accession number GSE111414. The other datasets used were accessed upon request through the responsible institutions: TracerX cohort through EGA accession number EGAD00001003206; Rizvi confirmation cohort through dbGaP accession number phs000980.v1.p1.

Statistical analysis

We used multivariable Cox regression models to investigate the association between KIR3DS1/L1 status and progression-free survival (PFS) adjusted for age and smoking status in the pooled analysis (discovery and confirmation cohort). An additional sensitivity analysis was conducted on the discovery cohort with PD-L1 in the model (considering expression levels of 0%, 1%–50%, and >50%). All analyses were conducted on complete cases, no techniques for multiple imputations were applied. We used the statistical program R (version 3.4.3) for all multivariable analyses and creating the Kaplan–Meier plots. GraphPad Prism version 6.05 was used for statistical analysis of flow cytometry data. All reported P values are two-sided and considered exploratory.

Minimal cellular changes in peripheral blood of patients with NSCLC undergoing PD-1–blocking antibody therapy

PBMCs and serum samples were collected from 35 patients with NSCLC immediately before and after two cycles (week 4) of treatment with nivolumab (Table 1). All patients had progressed on or after at least one prior platinum-containing chemotherapy regimen. Median age was 62 years, 42.9% of patients were female, 14.3% had squamous cell histology, and 74.3% adenocarcinoma histology (Table 1). Eastern Cooperative Oncology Group (ECOG) performance status (PS) was 0 or 1 in 77.1% of patients.

Table 1.

Characteristics of patients with NSCLC in the discovery cohorta

CharacteristicsDiscovery cohort (N = 35)
Median age, year (range) 62 (35–83) 
Male sex, n (%) 20 (57.1) 
Histology, n (%) 
 Adenocarcinoma 26 (74.3) 
 Squamous cell carcinoma 5 (14.3) 
 Large cell carcinoma 1 (2.9) 
 NOS 3 (8.5) 
Smoking history, n (%) 
 Never 7 (20) 
 Former 14 (40) 
 Current 14 (40) 
Molecular markers, n (%) 
 No markers detected 22 (62.8) 
KRAS mutation 9 (25.6) 
EGFR mutation 1 (2.9) 
ALK translocation 1 (2.9) 
MET exon 14 deletion 1 (2.9) 
HER2 mutation 1 (2.9) 
Previous therapy lines, n (%) 
 1 33 (84.6) 
 2 3 (7.7) 
 >2 3 (7.7) 
ECOG PS, n (%) 
 0 9 (25.7) 
 1 18 (51.4) 
 2 7 (20) 
 3 1 (3) 
CharacteristicsDiscovery cohort (N = 35)
Median age, year (range) 62 (35–83) 
Male sex, n (%) 20 (57.1) 
Histology, n (%) 
 Adenocarcinoma 26 (74.3) 
 Squamous cell carcinoma 5 (14.3) 
 Large cell carcinoma 1 (2.9) 
 NOS 3 (8.5) 
Smoking history, n (%) 
 Never 7 (20) 
 Former 14 (40) 
 Current 14 (40) 
Molecular markers, n (%) 
 No markers detected 22 (62.8) 
KRAS mutation 9 (25.6) 
EGFR mutation 1 (2.9) 
ALK translocation 1 (2.9) 
MET exon 14 deletion 1 (2.9) 
HER2 mutation 1 (2.9) 
Previous therapy lines, n (%) 
 1 33 (84.6) 
 2 3 (7.7) 
 >2 3 (7.7) 
ECOG PS, n (%) 
 0 9 (25.7) 
 1 18 (51.4) 
 2 7 (20) 
 3 1 (3) 

Abbreviations: ALK, anaplastic lymphoma kinase; NOS, not otherwise specified.

aPatients all received nivolumab monotherapy for at least 3 cycles.

Changes in immune cells in the peripheral blood were analyzed in 19 patients with available serial samples. After two cycles of treatment, there were no significant changes in numbers of CD8+ and CD4+ T cells during treatment (Fig. 1A), but a slight increase in CD4+ CD25+ FoxP3+ regulatory T cells was seen (Fig. 1A). No changes were observed in CD56+ CD3 NK-cell numbers or naïve T cells (CD45RA+ CCR7+ T cells; Fig. 1A). Because CD8+ T cells are considered the main effectors of checkpoint blockade and cancer immunotherapies (1, 2), we analyzed differentially expressed genes in peripheral CD8+ T cells before and during treatment in 5 patients with PR and 5 patients with primary progressive disease to identify mechanisms of resistance to PD-1 therapy (Supplementary Table S1). Messenger-RNA sequencing revealed alterations in very few genes in total peripheral CD8+ T cells during treatment (Supplementary Table S2). Only one gene, the gene for the killer immunoglobulin-like receptor KIR3DS1, was clearly differently expressed between responders and nonresponders (Fig. 1B).

Figure 1.

Presence of KIR3DS1 polymorphism is associated with resistance to PD-1 blockade. A, Frequency of peripheral CD8+ T cells, CD4+ T cells, regulatory CD4+ CD25+ FoxP3+ T cells, NK (CD56+ CD3) cells, naïve CD8+ T cells, and naïve CD4+ T cells in the peripheral blood of patients with NSCLC undergoing PD-1–blocking treatment (Tx) with nivolumab determined by flow cytometry, N = 19, *, P < 0.05 by paired Student t test. Data represented as mean ± SD. B, MA plot of the expression analysis in patients responding to PD-1 blockade versus patients not responding. C, Best radiographic response determined by RECIST in patients with NSCLC treated with nivolumab. Response assessment was performed in homozygous KIR3DL1 (L/L) and heterozygous KIR3DS1/L1 (S/L) patients. D, Number of patients with NSCLC in the discovery cohort with the KIR3DS1 allele present or absent that have progressive disease (PD) or not progressive disease (no PD) in the discovery cohort. Fisher exact test was used for statistical analysis. E, PFS of discovery cohort (N = 35), according to KIR3DS1 status. Statistical analysis by Wilcoxon signed-rank test.

Figure 1.

Presence of KIR3DS1 polymorphism is associated with resistance to PD-1 blockade. A, Frequency of peripheral CD8+ T cells, CD4+ T cells, regulatory CD4+ CD25+ FoxP3+ T cells, NK (CD56+ CD3) cells, naïve CD8+ T cells, and naïve CD4+ T cells in the peripheral blood of patients with NSCLC undergoing PD-1–blocking treatment (Tx) with nivolumab determined by flow cytometry, N = 19, *, P < 0.05 by paired Student t test. Data represented as mean ± SD. B, MA plot of the expression analysis in patients responding to PD-1 blockade versus patients not responding. C, Best radiographic response determined by RECIST in patients with NSCLC treated with nivolumab. Response assessment was performed in homozygous KIR3DL1 (L/L) and heterozygous KIR3DS1/L1 (S/L) patients. D, Number of patients with NSCLC in the discovery cohort with the KIR3DS1 allele present or absent that have progressive disease (PD) or not progressive disease (no PD) in the discovery cohort. Fisher exact test was used for statistical analysis. E, PFS of discovery cohort (N = 35), according to KIR3DS1 status. Statistical analysis by Wilcoxon signed-rank test.

Close modal

Association of KIR3DS1 genetic variant with resistance to PD-1 blockade

KIRs are a polymorphic family of NK-cell receptors that bind to HLAs (20, 21). The gene product of KIR3DS1 is an activating variant of the inhibitory KIR3DL1 receptor that binds to Bw4 epitope–containing HLAs (20, 21). The KIR3DS1 allele was present in the germline of 4 of 5 patients with primary progressive disease and 0 of 5 patients with PR, prompting us to study the allelic variant in the full cohort.

Genotyping of the discovery cohort (n = 35) revealed that 15 (42.9%) patients carried the KIR3DS1 variant while 20 (57.1%) were homozygous for KIR3DL1. The whole cohort had an overall response rate of 22.8% (8/35) and a PFS of 2.4 months (Supplementary Fig. S1A). No patient was homozygous for KIR3DS1 in our cohort. Clinical characteristics were similar between these two groups (Table 2). Thirteen of 17 (76.5%) patients with primary progression carried the activating receptor KIR3DS1 compared with 2 of 10 (20%) of patients with PR (P = 0.007). Patients with the KIR3DS1 allele present had a significantly higher chance for disease progression (Fig. 1C). The overall response rate (ORR) in patients homozygous for the KIR3DL1 allele was 40% [7/20 PR, 35%; 1/20 complete response (CR), 5%]. In contrast, the response rate in patients with the KIR3DS1 allele was 0% (P = 0.01). PFS to PD-1 blockade was also worse in patients carrying the KIR3DS1 allele [Fig. 1E; HR, 2.68; 95% confidence interval (CI), 1.58–4.61; P = 0.02]. Patients carrying either the KIR3DS1 allele or carrying no allele for Bw4-containing HLAs showed a lower risk of progression compared with patients carrying KIR3DS1 alone, but the consideration of Bw4-containing HLAs (ligands for KIR3DL1) did not better discriminate between progressing and nonprogressing patients (Supplementary Fig. S1B and S1C). In addition, the PFS of prior chemotherapy regimen did not differ between patients with or without the KIR3DS1 allele (Supplementary Fig. S2A; HR, 0.80; 95% CI, 0.41–1.55; P = 0.50).

Table 2.

Patient characteristics of the discovery cohort, according to KIR3DS1/L1 status

CharacteristicKIR3DL1/L1 (N = 20)KIR3DS1/L1 (N = 15)P
Median age, year (range) 61 (49–83) 61 (35–75) 0.43a 
Male sex, n (%) 10 (50) 11 (73.3) 0.30b 
Histology, n (%) 
 Nonsquamous 17 (85) 13 (86.6) >0.99b 
 Squamous 3 (15) 2 (13.4)  
Smoking history, n (%) 
 Yes 17 (85) 11 (73.3) 0.43b 
 No 3 (15) 4 (26.7)  
ECOG status, n (%) 
 0–1 15 (75) 12 (80) >0.99b 
 2–3 5 (15) 3 (20)  
PD-L1 status, n (%) 
 Negative 6 (30) 8 (53.3) 0.11c 
 1–50% 5 (25) 2 (13.4)  
 >50% 5 (25) 0 (0)  
 Missing 4 (20) 5 (33.3)  
CharacteristicKIR3DL1/L1 (N = 20)KIR3DS1/L1 (N = 15)P
Median age, year (range) 61 (49–83) 61 (35–75) 0.43a 
Male sex, n (%) 10 (50) 11 (73.3) 0.30b 
Histology, n (%) 
 Nonsquamous 17 (85) 13 (86.6) >0.99b 
 Squamous 3 (15) 2 (13.4)  
Smoking history, n (%) 
 Yes 17 (85) 11 (73.3) 0.43b 
 No 3 (15) 4 (26.7)  
ECOG status, n (%) 
 0–1 15 (75) 12 (80) >0.99b 
 2–3 5 (15) 3 (20)  
PD-L1 status, n (%) 
 Negative 6 (30) 8 (53.3) 0.11c 
 1–50% 5 (25) 2 (13.4)  
 >50% 5 (25) 0 (0)  
 Missing 4 (20) 5 (33.3)  

aStudent t test.

bFisher exact test.

cχ2 test.

We next explored two cohorts of patients with metastatic NSCLC treated with PD-(L)1–blocking antibodies as confirmation cohorts. First, we analyzed a cohort of 34 patients with metastatic NSCLC treated with pembrolizumab (Rizvi confirmation cohort). We selected this cohort because whole-exome sequencing data were available (12). Seven of 13 patients with primary progression carried the allele for the activating receptor KIR3DS1 compared with 1 of 12 of patients with PR (P = 0.03). All other responders (11 of 12) were homozygous for KIR3DL1. In this cohort, the response rate in the patients with KIR3DS1 was 9.1% (1/11) and without KIR3DS1 47.8% (11/23). Similarly, PFS to PD-1 blockade was worse in patients carrying the KIR3DS1 allele (Fig. 2A; HR, 2.64; 95% CI, 1.58–4.61; P = 0.01). KIR types in this cohort were determined by a new bioinformatics algorithm that has been developed by us and successfully validated in an independent cohort (TracerX) against a well-characterized published method (19). In addition, we examined by PCR the presence for KIR3DS1 of another new cohort of 66 patients with metastatic NSCLC that received a PD-(L)1 inhibitor (St. Gallen confirmation cohort). There was a trend with a median time-to-progression of 5.1 months in the KIR3DS1-positive group versus 10.2 months in the KIR3DS1-negative group. Yet, we did not observe a statistically significant difference of the PFS in this cohort (Fig. 2B; HR, 1.37; 95% CI, 0.69–2.70; P = 0.36). In this cohort, the response rate in the patients with KIR3DS1 was 7.1% (2/28) and 21.1% (8/38) without KIR3DS1. The ORR of this cohort is lower than would be expected in such a treatment group (15.1%). This could be a possible explanation for the lack effect in this independent group alone. Unlike in the discovery cohort, KIR3DS1 homozygous patients were found in the confirmation cohorts (2 in the Rizvi cohort, 4 in the St. Gallen cohort).

Figure 2.

KIR3DS1 association with PFS in NSCLC checkpoint inhibitor confirmation cohorts. A, PFS of confirmation cohort according to KIR3DS1 status (N = 34) from Rizvi and colleagues. Statistical analysis by Wilcoxon signed-rank test. B, PFS of confirmation cohort according to KIR3DS1 status (N = 66) of an additional cohort from the cantonal hospital in St. Gallen. Statistical analysis by Wilcoxon signed-rank test.

Figure 2.

KIR3DS1 association with PFS in NSCLC checkpoint inhibitor confirmation cohorts. A, PFS of confirmation cohort according to KIR3DS1 status (N = 34) from Rizvi and colleagues. Statistical analysis by Wilcoxon signed-rank test. B, PFS of confirmation cohort according to KIR3DS1 status (N = 66) of an additional cohort from the cantonal hospital in St. Gallen. Statistical analysis by Wilcoxon signed-rank test.

Close modal

Univariate analysis of PFS of the combined cohorts (discovery cohort and the two confirmation cohorts) confirms a significantly worse outcome when KIR3DS1 was present with a median PFS of only 2.6 months compared with 6.3 months (Supplementary Fig. S3; Supplementary Table S5; HR, 1.89; 95% CI, 1.24–2.87; P = 0.0029). Multivariate analysis considering age and smoking status demonstrated a significantly increased risk of progression in patients with the KIR3DS1 allele (Supplementary Table S6; HR, 1.91; 95% CI, 1.25–2.91; P = 0.0027). When a multivariate analysis was performed that included tumor PD-L1 status considering three groups of expression (0%, 1%–50%, and >50%), the significant effect of the KIR3DS1 status on the PFS remained (Supplementary Table S7; HR, 1.72; 95% CI, 1.10–2.68; P = 0.017). In contrast to this finding, among patients (n = 99) with early-stage NSCLCs who were not treated with immunotherapy (TracerX cohort), there was no difference in disease-free survival according to KIR3DS1 status (Supplementary Fig. S2B; HR, 1.41; 95% CI, 0.58–3.24; P = 0.44). We could not observe an association between PD-L1 expression and the KIR3DS1 status (Table 3). We could also not observe an association with the TMB of patients that had TMB data available (Table 3).

Table 3.

Association of PD-L1 status and TMB of pooled cohort

KIR3DS1 statusL/LS/L or S/SP
PD-L1 status, n (%) 
 0% 23 (36) 17 (39.5) 0.92a 
 1–50% 18 (28) 12 (27.9)  
 >50% 23 (36) 14 (32.6)  
TMB, n (%) 
 Low (0–11 mutations/Mb) 13 (39.4) 9 (60) 0.22b 
 High (>11 mutations/Mb) 20 (60.6) 6 (40)  
KIR3DS1 statusL/LS/L or S/SP
PD-L1 status, n (%) 
 0% 23 (36) 17 (39.5) 0.92a 
 1–50% 18 (28) 12 (27.9)  
 >50% 23 (36) 14 (32.6)  
TMB, n (%) 
 Low (0–11 mutations/Mb) 13 (39.4) 9 (60) 0.22b 
 High (>11 mutations/Mb) 20 (60.6) 6 (40)  

aχ2 test.

bFisher exact test.

The KIR3DS1 receptor associates NK cells with resistance to immune checkpoint blockade

To provide mechanistic insights into the association of KIR3DS1 expression and immune checkpoint blockade resistance, we determined the expression of KIR3DS1 on peripheral immune cells by flow cytometry. KIR3DS1/L1 was mainly expressed on NK cells compared with CD8+ T cells both in the peripheral blood and in the tumor (Fig. 3A). Both KIR3DL1/L1 and KIR3DS1/L1 patients showed similar NK-cell numbers and expression of PD-1, CD56 (NCAM1), TBET (TBX21), EOMES, and KI67 in NK cells (Supplementary Fig. S4). PD-1 expression on peripheral NK cells was very low, but higher and significantly decreased upon treatment in patients homozygous for KIR3DL1. TIM-3 (HAVcr-2) was higher before treatment in KIR3DL1/L1 patients compared with KIR3DS1/L1 patients (Supplementary Fig. S4F). Expression of other inhibitory and activating receptors on NK cells remained unchanged during treatment (Supplementary Fig. S4).

Figure 3.

KIR3DS1 on NK cells is involved in the association with resistance to PD-1 blockade. A, Two representative flow cytometric analyses of peripheral and intratumoral NK cells and CD8+ T cells for KIR3DL1 and KIR3DS1 (y-axis, antibody that recognizes both KIR3DS1 and KIR3DL1; x-axis, antibody that recognizes only KIR3DL1). B, Functional analysis of peripheral NK cells before and during treatment with PD-1 blockade. Peripheral NK cells were incubated with K562 target cells, and intracellular cytokine production was determined by flow cytometry. N = 7; **, P < 0.01 by paired two-way ANOVA with Sidak multiple comparison testing. C, Changes of IFNγ and IL6 in the peripheral sera of patients with NSCLC during PD-1 blockade therapy. N = 7; *, P < 0.05 for the interaction between genotype and treatment by paired two-way ANOVA with Sidak multiple comparison testing. D, Representative images of HLA-F staining for NSCLC tissue by IHC in red coloring (bars, 100 μm).

Figure 3.

KIR3DS1 on NK cells is involved in the association with resistance to PD-1 blockade. A, Two representative flow cytometric analyses of peripheral and intratumoral NK cells and CD8+ T cells for KIR3DL1 and KIR3DS1 (y-axis, antibody that recognizes both KIR3DS1 and KIR3DL1; x-axis, antibody that recognizes only KIR3DL1). B, Functional analysis of peripheral NK cells before and during treatment with PD-1 blockade. Peripheral NK cells were incubated with K562 target cells, and intracellular cytokine production was determined by flow cytometry. N = 7; **, P < 0.01 by paired two-way ANOVA with Sidak multiple comparison testing. C, Changes of IFNγ and IL6 in the peripheral sera of patients with NSCLC during PD-1 blockade therapy. N = 7; *, P < 0.05 for the interaction between genotype and treatment by paired two-way ANOVA with Sidak multiple comparison testing. D, Representative images of HLA-F staining for NSCLC tissue by IHC in red coloring (bars, 100 μm).

Close modal

To test the functional potential of NK cells in patients treated with PD-1 blockade, we incubated peripheral NK cells with K562 target cells and analyzed degranulation by CD107a surface staining and production of intracellular cytokines. Patients homozygous for KIR3DL1 showed increased cytokine production after PD-1 blockade monotherapy compared with those carrying the KIR3DS1 allele (Fig. 3B; Supplementary Fig. S5A). Similarly, more KIR3DL1/L1 patients had an increase in tumor cell killing (63%) during treatment compared with patients carrying the KIR3DS1 allele (28%) (Supplementary Fig. S5B). Multiplex analysis of NK and T-cell cytokines in the peripheral blood showed a tendency toward reduced IFNγ and significantly increased IL6 levels in patients with NSCLC with KIR3DS1/L1 heterozygosity (Fig. 3C). There were no significant changes of peripheral T cells depending on the KIR3DS1 status (Supplementary Fig. S5C–S5H). The presence of activating ligands for KIR3DS1 could lead to continuous stimulation of NK cells and potential NK-cell exhaustion. We therefore stained available tissue for the KIR3DS1 ligand HLA-F. Indeed, several tumors showed a clear positivity for HLA-F IHC staining (Fig. 3D; Supplementary Fig. S6; Supplementary Table S8). In general, most NSCLC samples demonstrated an intermediate to high staining score suggesting that HLA-F upregulation in lung tumors is a common phenomenon (Supplementary Fig. S6; Supplementary Table S8).

We identified a receptor gene variation, which is known to modulate NK-cell activation, being associated with worse clinical outcome in patients with metastatic NSCLC treated with PD-1 blockade. We confirmed our findings in a combined analysis of cohorts of patients with metastatic NSCLC treated with PD-(L)1–blocking antibodies. We also show that the genetic variation is not associated with prognosis in patients with localized NSCLC after resection (TracerX) cohort. In general, KIR3DS1 seems to be rather a predictive marker for PD-(L)1 therapy than a prognostic marker, although further studies including prospective trials would be needed to definitively answer this question.

The KIR3DS1/L1 genetic variation has been previously linked to the outcome of treated patients with cancer (22–25). In one study, the presence of the KIR3DS1 allele predicted increased risk of progression in patients receiving autologous stem cell transplantation for multiple myeloma (24). In another study, the KIR3DL1/S1 status was associated with outcomes in patients with neuroblastoma treated with anti-GD2 mAb (23). Recent mechanistic studies in a humanized mouse model have shown that the presence of inhibitory KIR3DL1 and Bw4-containing HLA ligands can educate NK cells to be more responsive to the loss of MHC-I molecules on tumor cells (26).

We have identified the KIR3DS1 allele to be associated with resistance to PD-1 blockade by expression analysis in peripheral CD8+ T cells. Although expression was different between responders and nonresponders on a transcript level, only minimal protein expression was found by flow cytometry on CD8+ T cells compared with NK cells. We therefore focused further analysis on NK cells. NK-cell education describes a process by which the chronic interaction of inhibitory KIRs (e.g., KIR3DL1) with their MHC ligands (e.g., Bw4) promotes NK-cell responsiveness through yet unknown mechanisms (27–29). Our correlative analysis suggests that NK cells are involved in the mechanism of KIR3DS1-associated resistance to PD-1 blockade, because KIR3DL1 or KIR3DS1 were only minimally expressed on peripheral and intratumoral T cells. NK-cell education mediated by KIR3DL1-Bw4 ligation seems unlikely to be the primary cause of resistance, because the predictive value of KIR3DS1 was reduced when the Bw4 ligands were considered. Thus, the presence or absence of KIR3DL1-binding Bw4-containing HLAs does not improve the discrimination between responders and nonresponders in patients homozygous for KIR3DL1 (Supplementary Fig S1B). In addition, differences in peripheral NK-cell functionality were only observed after initiation of PD-1 blockade and not prior treatment, which should also be different if education was involved.

Recent analysis of HLA types and response to immune checkpoint inhibitors suggests an association with certain HLA-B types, mainly linked to presentation capacities of specific HLA-B types (30). It is possible that these HLA types also associate with differences in NK functionality, which would need further investigation in larger cohorts given the vast numbers of possible combinations. Another possible underlying mechanism might be the induction of dysfunctional or “exhausted” NK cells by continuous KIR3DS1 stimulation by its ligand HLA-F, which was found to be highly expressed in lung cancer (31). Indeed, we confirmed strong expression of HLA-F in selected patients, although this could not be performed in all patients due to unavailability of tumor tissue. Yet, we currently lack a deeper mechanistic understanding of the cellular network that mediates resistance in patients expressing KIR3DS1. Complex human in vitro systems or humanized mouse models would be required for functional studies, as KIR receptors and HLAs both are very weakly conserved between mice and humans. In addition, larger prospective trials are needed to validate the presence of KIR3DS1 as predictive marker for primary resistance in tumors other than lung cancer and combination treatments such as PD-1/CTLA-4 blockade.

Our results support the hypothesis that NK cells are involved in resistance to PD-1 blockade. NK cells have been known to be important players in the immunosurveillance of cancers and recent efforts have led to the development of checkpoint inhibitors that target inhibitory receptors on NK cells for cancer immunotherapy (21, 32). Lirilumab, for example, targets the immunoglobulin-like domain 2 (2D) of KIR2DL1, 2 and 3, which is engaged by HLA-C epitopes (21, 32). Our findings suggest that NK cells play an important role in mediating the effect of PD-1 blocking immunotherapy. It is conceivable that cross-talk between NK cells and the tumor microenvironment, rather than an independent effect of NK cells, could be responsible for our observations. Indeed, a recent publication described the interaction of NK cells with dendritic cells within the tumor to induce a PD-1 checkpoint inhibitor–responsive microenvironment (10). A recent analysis in a mouse model has additionally provided evidence that NK cells dictate the infiltration of classical dendritic cells into tumors and thereby play an important role in antitumor immunity (33).

In summary, we provide evidence that the KIR3DS1 genetic variant is associated with the outcome in patients with NSCLC treated with PD-1–directed immunotherapy (summarized in Supplementary Fig. S7). Our findings suggest an important role of NK cells to modulate the therapeutic efficacy of PD-(L)1–targeted therapies. As NK-cell–targeted therapies are available, they might be particularly useful in patients with the KIR3DS1 allele present. Thus, it will be of utmost clinical relevance to dissect the cellular mechanisms how NK cells interact with other immune effectors in the tumor microenvironment.

S.I. Rothschild is a consultant/advisory board member for Bristol-Myers Squibb, MSD, AstraZeneca, and Roche. P. Jermann reports receiving commercial research grants from Bristol-Myers Squibb, and reports receiving other commercial research support from Thermo Fisher Scientific. I. Alborelli reports receiving commercial research grants from Bristol-Myers Squibb, and reports receiving other commercial research support from Thermo Fisher Scientific. F. Stenner is a consultant/advisory board member for Bristol-Myers Squibb. Z. Trajanoski reports receiving speakers bureau honoraria from Merck and Roche. H. Läubli reports receiving commercial research grants from Bristol-Myers Squibb and is a consultant/advisory board member for Palleon Pharmaceuticals. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S.I. Rothschild, A. Zippelius, H. Läubli

Development of methodology: M.P. Trefny, F. Uhlenbrock, D. Rieder, A. Zippelius, H. Läubli

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.P. Trefny, S.I. Rothschild, F. Uhlenbrock, M. Kaiser, P. Herzig, S. Poechtrager, D.S. Thommen, S. Savic, P. Jermann, I. Alborelli, S. Schaub, F. Stenner, M. Früh, K.D. Mertz, H. Läubli

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.P. Trefny, S.I. Rothschild, F. Uhlenbrock, D. Rieder, B. Kasenda, M.A. Stanczak, F. Berner, A.S. Kashyap, M. Kaiser, P. Herzig, D.S. Thommen, F. Geier, P. Jermann, I. Alborelli, M. Früh, Z. Trajanoski, A. Zippelius, H. Läubli

Writing, review, and/or revision of the manuscript: M.P. Trefny, S.I. Rothschild, D. Rieder, B. Kasenda, M.A. Stanczak, F. Berner, A.S. Kashyap, S. Savic, F. Stenner, M. Früh, Z. Trajanoski, K.D. Mertz, A. Zippelius, H. Läubli

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D. Rieder, P. Herzig, S. Poechtrager, S. Savic, L. Flatz, K.D. Mertz, A. Zippelius,

Study supervision: S.I. Rothschild, A. Zippelius, H. Läubli

The authors thank Priska Auf der Maur (University and University Hospital of Basel, Basel, Switzerland) for designing the graphical summary. The authors thank the sciCORE (http://scicore.unibas.ch/) scientific computing core facility at University of Basel (Basel, Switzerland) for helping to perform calculations. The authors acknowledge the CRUK TracerX consortium for providing access to the sequencing data of patients undergoing chemotherapy, and Jamie Ashman (Prism Ideas) for editorial assistance in the submission of this manuscript. Most importantly, the authors thank all the patients that allowed the use of their material and made this work possible. H. Läubli was supported by grants from Lichtenstein Foundation, Schoenemakers Foundation, Krebsliga beider Basel, and the Goldschmidt-Jacobson Foundation. A. Zippelius was supported by Krebsforschung Schweiz (KFS-3394-02-2014) and Swiss National Science Foundation (320030_162575).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Chen
DS
,
Mellman
I
. 
Elements of cancer immunity and the cancer-immune set point
.
Nature
2017
;
541
:
321
30
.
2.
Topalian
SL
,
Drake
CG
,
Pardoll
DM
. 
Immune checkpoint blockade: a common denominator approach to cancer therapy
.
Cancer Cell
2015
;
27
:
450
61
.
3.
Borghaei
H
,
Paz-Ares
L
,
Horn
L
,
Spigel
DR
,
Steins
M
,
Ready
NE
, et al
Nivolumab versus docetaxel in advanced nonsquamous non-small-cell lung cancer
.
N Engl J Med
2015
;
373
:
1627
39
.
4.
Brahmer
J
,
Reckamp
KL
,
Baas
P
,
Crino
L
,
Eberhardt
WE
,
Poddubskaya
E
, et al
Nivolumab versus docetaxel in advanced squamous-cell non-small-cell lung cancer
.
N Engl J Med
2015
;
373
:
123
35
.
5.
Fehrenbacher
L
,
Spira
A
,
Ballinger
M
,
Kowanetz
M
,
Vansteenkiste
J
,
Mazieres
J
, et al
Atezolizumab versus docetaxel for patients with previously treated non-small-cell lung cancer (POPLAR): a multicentre, open-label, phase 2 randomised controlled trial
.
Lancet
2016
;
387
:
1837
46
.
6.
Herbst
RS
,
Baas
P
,
Kim
DW
,
Felip
E
,
Perez-Gracia
JL
,
Han
JY
, et al
Pembrolizumab versus docetaxel for previously treated, PD-L1-positive, advanced non-small-cell lung cancer (KEYNOTE-010): a randomised controlled trial
.
Lancet
2016
;
387
:
1540
50
.
7.
Reck
M
,
Rodriguez-Abreu
D
,
Robinson
AG
,
Hui
R
,
Csoszi
T
,
Fulop
A
, et al
Pembrolizumab versus chemotherapy for PD-L1-positive non-small-cell lung cancer
.
N Engl J Med
2016
;
375
:
1823
33
.
8.
Antonia
SJ
,
Villegas
A
,
Daniel
D
,
Vicente
D
,
Murakami
S
,
Hui
R
, et al
Durvalumab after chemoradiotherapy in stage III non-small-cell lung cancer
.
N Engl J Med
2017
;
377
:
1919
29
.
9.
Syn
NL
,
Teng
MWL
,
Mok
TSK
,
Soo
RA
. 
De-novo and acquired resistance to immune checkpoint targeting
.
Lancet Oncol
2017
;
18
:
e731
e41
.
10.
Barry
KC
,
Hsu
J
,
Broz
ML
,
Cueto
FJ
,
Binnewies
M
,
Combes
AJ
, et al
A natural killer-dendritic cell axis defines checkpoint therapy-responsive tumor microenvironments
.
Nat Med
2018
;
24
:
1178
91
.
11.
Parham
P
,
Norman
PJ
,
Abi-Rached
L
,
Guethlein
LA
. 
Variable NK cell receptors exemplified by human KIR3DL1/S1
.
J Immunol
2011
;
187
:
11
9
.
12.
Rizvi
NA
,
Hellmann
MD
,
Snyder
A
,
Kvistborg
P
,
Makarov
V
,
Havel
JJ
, et al
Cancer immunology. Mutational landscape determines sensitivity to PD-1 blockade in non-small cell lung cancer
.
Science
2015
;
348
:
124
8
.
13.
Jamal-Hanjani
M
,
Wilson
GA
,
McGranahan
N
,
Birkbak
NJ
,
Watkins
TBK
,
Veeriah
S
, et al
Tracking the evolution of non-small-cell lung cancer
.
N Engl J Med
2017
;
376
:
2109
21
.
14.
Choppa
PC
,
Vojdani
A
,
Tagle
C
,
Andrin
R
,
Magtoto
L
. 
Multiplex PCR for the detection of Mycoplasma fermentans, M. hominis and M. penetrans in cell cultures and blood samples of patients with chronic fatigue syndrome
.
Mol Cell Probes
1998
;
12
:
301
8
.
15.
Thommen
DS
,
Schreiner
J
,
Muller
P
,
Herzig
P
,
Roller
A
,
Belousov
A
, et al
Progression of lung cancer is associated with increased dysfunction of T cells defined by coexpression of multiple inhibitory receptors
.
Cancer Immunol Res
2015
;
3
:
1344
55
.
16.
Uhrberg
M
,
Valiante
NM
,
Shum
BP
,
Shilling
HG
,
Lienert-Weidenbach
K
,
Corliss
B
, et al
Human diversity in killer cell inhibitory receptor genes
.
Immunity
1997
;
7
:
753
63
.
17.
Li
H
,
Handsaker
B
,
Wysoker
A
,
Fennell
T
,
Ruan
J
,
Homer
N
, et al
The sequence alignment/map format and SAMtools
.
Bioinformatics
2009
;
25
:
2078
9
.
18.
Liao
Y
,
Smyth
GK
,
Shi
W
. 
featureCounts: an efficient general purpose program for assigning sequence reads to genomic features
.
Bioinformatics
2014
;
30
:
923
30
.
19.
Norman
PJ
,
Hollenbach
JA
,
Nemat-Gorgani
N
,
Marin
WM
,
Norberg
SJ
,
Ashouri
E
, et al
Defining KIR and HLA class I genotypes at highest resolution via high-throughput sequencing
.
Am J Hum Genet
2016
;
99
:
375
91
.
20.
Malmberg
KJ
,
Sohlberg
E
,
Goodridge
JP
,
Ljunggren
HG
. 
Immune selection during tumor checkpoint inhibition therapy paves way for NK-cell "missing self" recognition
.
Immunogenetics
2017
;
69
:
547
56
.
21.
Chiossone
L
,
Vienne
M
,
Kerdiles
YM
,
Vivier
E
. 
Natural killer cell immunotherapies against cancer: checkpoint inhibitors and more
.
Semin Immunol
2017
;
31
:
55
63
.
22.
Boudreau
JE
,
Giglio
F
,
Gooley
TA
,
Stevenson
PA
,
Le Luduec
JB
,
Shaffer
BC
, et al
KIR3DL1/HL A-B subtypes govern acute myelogenous leukemia relapse after hematopoietic cell transplantation
.
J Clin Oncol
2017
;
35
:
2268
78
.
23.
Forlenza
CJ
,
Boudreau
JE
,
Zheng
J
,
Le Luduec
JB
,
Chamberlain
E
,
Heller
G
, et al
KIR3DL1 allelic polymorphism and HLA-B epitopes modulate response to Anti-GD2 monoclonal antibody in patients with neuroblastoma
.
J Clin Oncol
2016
;
34
:
2443
51
.
24.
Gabriel
IH
,
Sergeant
R
,
Szydlo
R
,
Apperley
JF
,
DeLavallade
H
,
Alsuliman
A
, et al
Interaction between KIR3DS1 and HLA-Bw4 predicts for progression-free survival after autologous stem cell transplantation in patients with multiple myeloma
.
Blood
2010
;
116
:
2033
9
.
25.
Venstrom
JM
,
Pittari
G
,
Gooley
TA
,
Chewning
JH
,
Spellman
S
,
Haagenson
M
, et al
HLA-C-dependent prevention of leukemia relapse by donor activating KIR2DS1
.
N Engl J Med
2012
;
367
:
805
16
.
26.
Boudreau
JE
,
Liu
XR
,
Zhao
Z
,
Zhang
A
,
Shultz
LD
,
Greiner
DL
, et al
Cell-extrinsic MHC class I molecule engagement augments human NK cell education programmed by cell-intrinsic MHC class I
.
Immunity
2016
;
45
:
280
91
.
27.
Shifrin
N
,
Raulet
DH
,
Ardolino
M
. 
NK cell self tolerance, responsiveness and missing self recognition
.
Semin Immunol
2014
;
26
:
138
44
.
28.
Raulet
DH
,
Vance
RE
. 
Self-tolerance of natural killer cells
.
Nat Rev Immunol
2006
;
6
:
520
31
.
29.
Elliott
JM
,
Yokoyama
WM
. 
Unifying concepts of MHC-dependent natural killer cell education
.
Trends Immunol
2011
;
32
:
364
72
.
30.
Chowell
D
,
Morris
LGT
,
Grigg
CM
,
Weber
JK
,
Samstein
RM
,
Makarov
V
, et al
Patient HLA class I genotype influences cancer response to checkpoint blockade immunotherapy
.
Science
2018
;
359
:
582
7
.
31.
Lin
A
,
Zhang
X
,
Ruan
YY
,
Wang
Q
,
Zhou
WJ
,
Yan
WH
. 
HLA-F expression is a prognostic factor in patients with non-small-cell lung cancer
.
Lung Cancer
2011
;
74
:
504
9
.
32.
Guillerey
C
,
Huntington
ND
,
Smyth
MJ
. 
Targeting natural killer cells in cancer immunotherapy
.
Nat Immunol
2016
;
17
:
1025
36
.
33.
Bottcher
JP
,
Bonavita
E
,
Chakravarty
P
,
Blees
H
,
Cabeza-Cabrerizo
M
,
Sammicheli
S
, et al
NK cells stimulate recruitment of cDC1 into the tumor microenvironment promoting cancer immune control
.
Cell
2018
;
172
:
1022
37
.