NKG7 Is a T-cell–Intrinsic Therapeutic Target for Improving Antitumor Cytotoxicity and Cancer Immunotherapy

Cytotoxic CD8⁺ T cells (CTL) are a crucial component of the immune system. They are capable of specifically targeting and eliminating malignant tumor cells throughout the body (1). To accomplish this, CTLs are first primed by antigen-presenting cells in secondary lymphoid organs and then traffic to the tumor site where they encounter their cognate antigen on the tumor cell itself. This triggers the initiation of an immune synapse followed by exocytosis of cytolytic granules containing perforin and granzymes, which results in tumor-cell destruction. The coordination of the cytolytic killing machinery within CD8⁺ T cells is multi-faceted and complex, involving numerous proteins as well as the filamentous actin (F-actin) and microtubule cytoskeletons to ensure directed delivery of lytic granules to the contact site formed between the CD8⁺ T cell and the tumor target (2, 3). Although extrinsic factors within the tumor microenvironment that affect T-cell killing have been well studied in the context of immune checkpoints or regulatory cells (4, 5), the T-cell intrinsic factors that are required for human CTLs to achieve and maintain highly efficient antitumor cytotoxicity are still not well defined.

This gap in knowledge has become particularly apparent now that there are a variety of immune-modulating agents used therapeutically, all of which are designed to enhance CTL responses against tumors. The ultimate goal of tumor vaccines, chimeric antigen receptor (CAR) T-cell methodologies, and immune checkpoint inhibitors is to produce CTLs that can efficiently eliminate tumor cells. However, no consensus exists as to what defines the specific subset of effector CD8⁺ T cells responsible for the “best” tumor-cell killing or what components need to be present within CTLs to get the most robust antitumor response.

PD-1/PD-L1–binding mAbs are among the most widely used of these cancer immunotherapies and treatment with these agents can lead to complete remission in a small proportion of patients. Unfortunately, the majority of patients treated experience no detectable clinical response (6, 7). Although the initial mechanism of action of PD-1/PD-L1 blockade was thought to be the rescue of exhausted T cells within the tumor microenvironment, more recent studies point to the importance of T-cell proliferation in the periphery after treatment (8–10). These peripheral T-cell clones then infiltrate the tumor and are likely responsible for tumor killing. It has been shown that in both responders and nonresponders, distinct modules of genes associated with therapy are induced within peripheral CD8⁺ T cells, including those involved in cell division. Still, in nonresponders, this does not lead to a measurable decrease in tumor burden (11, 12). This cannot be fully explained by a failure of T cells to infiltrate tumors, as even some so called “hot” tumors with molecular signatures that belie significant T-cell infiltration do not respond to anti–PD-1 therapy (13, 14). Thus, we hypothesized that there are additional T-cell intrinsic factors necessary for effective tumor-cell killing that are deficient in the T cells of nonresponders. If these factors could be identified, they would not only offer insight into how to optimize CTL function for adoptive cellular therapy (ACT) but could also potentially be restored in CD8⁺ T cells to overcome resistance to PD-1/PD-L1 checkpoint inhibition.

Herein, we employed single-cell RNA-sequencing (scRNA-seq) technology to analyze CD8⁺ T cells from the peripheral blood of complete responders (no tumor measurable at the time of posttreatment sampling) and nonresponders to anti–PD-1 therapy, comparing differential gene expression across time [baseline (BL) vs. postinitiation of treatment (PT)] and, in a second analysis, gene expression within the largest and most expanded T-cell clones. This investigation highlighted that appropriate expression of a cytolytic granule associated molecule, natural killer cell granule protein-7 (NKG7), was a feature of the cytotoxic CD8⁺ T cells within complete responders to anti–PD-1 therapy. We further characterized NKG7 within human primary CD8⁺ T cells and uncovered its role in cytolytic granule trafficking and release. To begin translating this knowledge to a potential therapy, we used mRNA transfection to modulate NKG7 expression within the CD8⁺ T cells of nonresponders. This was sufficient to improve cytolytic T-cell killing of tumor cells in vitro and increase their response to PD-1/PD-L1 blockade. NKG7 mRNA therapy also improved the antitumor activity of murine tumor antigen–specific CD8⁺ T cells in vivo in a model of ACT. Finally, we considered how NKG7 expression may be modulated at the transcription level. Together, our results have identified NKG7 as a necessary T-cell intrinsic factor for the cytotoxic function of CD8⁺ T cells and established NKG7 as a new therapeutic target for cancer immunotherapy.

This study was performed to identify gene expression differences within the CD8⁺ T cells of responders versus nonresponders to anti–PD-1 therapy with the goal of modulating the identified targets to improve response. This objective was addressed by (i) performing scRNA-seq analysis of 16 paired patient samples, taken from BL and posttreatment timepoints (before and after initiation of therapy), (ii) determining that reduced NKG7 was a common feature among nonresponders and validating this finding in separate sets of patient data (protein-based and primary tumor-based studies), (iii) characterizing the function of NKG7 in human CD8⁺ T cells by knockdown of NKG7 in primary cells followed by cytotoxicity assays, time-lapse imaging studies, electron microscopy, and calcium-flux measurements, (iv) studying the effect of overexpression of NKG7 through transfection with NKG7 mRNA on both human patient samples taken from nonresponding patients and on antigen-specific CD8⁺ T cells from mice (in vivo modeling), and (v) screening for transcription factors that modulate NKG7 gene expression through siRNA-mediated knockdown of transcription factors followed by chromatin immunoprecipitation qPCR (ChIP-qPCR), qRT-PCR, Western blot, and functional cytotoxicity assays. Sample size was determined by the investigators according to previous experimental experience. The exact n numbers used in each experiment are indicated in the respective figure legends. Samples were allocated randomly whenever possible. For mouse experiments, the mice were randomized at day 5 posttumor injection, upon first tumor measurement, such that all groups had similar averaged tumor sizes prior to treatment. Lab members performing caliper measurements of mouse tumors were blinded to treatment group. No data was excluded from the analysis. Outliers were included in statistical analyses and are represented in the graphs (all formatted as dot plots so that variations between individual patients, healthy donors, or mice can be clearly evaluated).

All human participants provided signed informed written consent; the study was approved by the Mayo Clinic Rochester institutional review board (IRB) and was conducted according to Declaration of Helsinki principles. Peripheral blood and tumor tissue samples from patients were collected at either BL (prior to therapy) or PT after written consent was obtained from each participant (adults of both genders and all races). Clinical course, treatment information, and outcomes in patients treated with anti–PD-1/PD-L1 therapy at Mayo Clinic were retrospectively collected. Response to treatment was evaluated according to standard clinical practice guidelines using RECIST (15). An overview of the patient samples used in the study is provided in Supplementary Table S1 and details for the 8 patients whose samples were analyzed via scRNA-seq are available in Supplementary Table S2. The response listed [complete response (CR) vs. progressive disease (PD)] was apparent (using RECIST) at the 12-week PT timepoint for all patients, except for sample R-3, who was still categorized as PD at 12 weeks PT but had a CR overall (noted at 24 weeks PT). “Healthy Donor” human blood leukocytes were acquired from anonymous donors who had consented for blood donation at the Blood Transfusion Center at Mayo Clinic.

Peripheral blood mononuclear cells (PBMC) were isolated from patient blood samples via centrifugation with lymphoprep (catalog no. 07851, StemCell Technologies) and SepMate conical tubes (catalog no. 85450, StemCell Technologies). Samples were then stored in liquid nitrogen for later use and thawed on the day of the experiments. PBMCs were isolated from healthy donors using the same protocol as above but from buffy coat blood. Once PBMCs were obtained from healthy donor blood, CD8⁺ T cells were isolated using a magnet-based CD8⁺ T-cell isolation kit according to the manufacturer's protocol (catalog no. 19053, StemCell Technologies), and the cells were used immediately for experiments.

PBMCs from patient blood were thawed and stained with anti-CD45, anti-CD3, and anti-CD8 (antibody details are in Supplementary Table S3). CD45⁺CD3⁺CD8⁺ cells were then sorted using a FACSAria II sorter (BD Biosciences) running FACSDiva v8.0.1 software. After sorting, cells were washed twice in 1 × PBS + 0.05% BSA and immediately submitted to the Medical Genome Facility Genome Analysis Core for further processing.

The cells were first counted and measured for viability using the Vi-Cell XR Cell Viability Analyzer (Beckman-Coulter) and a basic hemocytometer and light microscope. Barcoded Gel Beads were thawed from −80°C to room temperature and the cDNA master mix was prepared according to the manufacturer's instructions for Chromium Single Cell 3′ v2 library & Gel bead kit or A-Chip kit (catalog no. PN-120237 or PN-120236, 10x Genomics). The target capture number of cells was 2,000 to 3,000, so a volume containing more than 5,000 live cells was combined with cDNA master mix. The cell suspension and master mix, thawed Gel Beads and partitioning oil were added to a Chromium Single Cell A chip. The filled chip was loaded into the Chromium Controller, where each sample was processed and the individual cells within the sample were captured into uniquely labeled Gel Beads-In-Emulsion (GEM). The GEMs were collected from the chip and taken to the bench for reverse transcription, GEM dissolution, and cDNA clean-up. The resulting cDNA was a pool of uniquely barcoded molecules. Cleaned and measured pooled cDNA continued on to library construction, where standard Illumina sequencing primers and a unique i7 Sample index were added to each cDNA pool.

All cDNA pools and resulting libraries were measured using Qubit High Sensitivity assays (Thermo Fisher Scientific, catalog no. Q32851), Agilent Bioanalyzer High Sensitivity chips (Agilent, catalog no. 5067–4626), and Kapa DNA Quantification reagents (Kapa Biosystems, catalog no. 07960298001). Libraries were sequenced at 60,000 fragment reads per cell following Illumina's standard protocol using the Illumina cBot and HiSeq 3000/4000 PE Cluster Kit (Illumina, catalog no. PE-410–1001). The flow cells were sequenced as 100 × 2 paired-end reads on an Illumina HiSeq 4000 using HiSeq 3000/4000 sequencing kit (Illumina, catalog no. FC-410–1002 and FC-410–1001) and HCS v3.3.52 collection software. Base-calling was performed using Illumina's RTA version 2.7.3.

Initial processing was as described in 3′ sequencing methods. The resulting cDNA created a pool of uniquely barcoded molecules that were used to generate 5′ gene expression libraries and enriched T-cell receptor (TCR) libraries. Not more than 50 ng mass of cDNA sample was used to create a gene expression library. A much lower amount of cDNA was required to build a TCR enriched library. During library construction, standard Illumina sequencing primers and a unique i7 Sample index (10× Genomics, catalog no. PN-120262) were added to each cDNA pool (gene expression and TCR). As above, all cDNA pools and resulting libraries were measured using Qubit High Sensitivity assays (Thermo Fisher Scientific, catalog no. Q32851), Agilent Bioanalyzer High Sensitivity chips (Agilent, catalog no. 5067–4626), and Kapa qPCR DNA Quantification reagents (Kapa Biosystems, catalog no. 07960298001). Gene expression libraries were sequenced at 50,000 fragment reads per cell and TCR libraries were sequenced at 5,000 fragment reads per cell. Sequencing steps followed Illumina's standard protocol using the Illumina cBot and HiSeq 3000/4000 PE Cluster Kit (Illumina, catalog no. PE-410–1001). The flow cells were sequenced as 100 × 2 paired end reads on an Illumina HiSeq 4000 using HiSeq 3000/4000 sequencing kit (Illumina, catalog no. FC-410–1002 and FC-410–1001) and HCS v3.3.52 collection software. Base-calling was performed using Illumina's RTA version 2.7.3.

10X Genomics Cell Ranger Single Cell Software Suite (v3.1.0) was used to demultiplex raw base call (BCL) files generated from the sequencer into FASTQ files. Cell scRNA-seq data and TCR sequencing data were aligned and quantified using Ranger count and Cell Ranger vdj protocols, respectively, against their corresponding GRCh38 human reference genome downloaded from 10X Genomics website.

We used the Seurat package (v3.1; RRID:SCR_007322; refs. 16, 17) to perform integrated analyses of single cells. Genes expressed in less than 3 cells and cells that expressed less than 200 genes and more than 40% mitochondria genes were excluded for downstream analysis in each sample. For scRNA-seq analysis, we followed the Seurat SCTransform integration workflow and the comparative analysis workflow. Each dataset was SCTransform-normalized and the top 3,000 Highly Variable Genes (HVG) across cells were selected. The datasets were integrated based on “anchors” identified between datasets before principal component analysis (PCA) was performed to do linear dimensional reduction. Shared Nearest Neighbor (SNN) Graph was constructed to identify clusters on the low-dimensional space [top 30 statistically significant principal components (PC)]. Enriched marker genes in each cluster conserved across all samples were identified, and differentially expressed genes between two selected conditions were detected using the default Wilcoxon rank-sum test at cluster level or patient sample level. A resolution of 0.06 was selected for R-1, R-2, NR-1, and NR-2 datasets (8-sample integration). Paired integration analysis (PT vs. BL) was also performed individually for each patient sample (R-1, R-2, NR-1, and NR-2) with a common resolution of 0.1. A resolution of 0.08 was used for R-3, R-4, NR-3, and NR-4 datasets (8-sample integration analysis). Integration analyses of scRNA-seq data and single-cell TCR data were performed first based on cell barcodes from the top five clones (by clone size) mapped to those in gene expression data. A second analysis was completed to consider expanded clones. Clones were considered “expanded” if the clone was detected in more cells in the PT sample versus the BL sample for an individual patient. Cells with expanded clones (size ≥ 10 in PT samples) were selected for differential gene expression analysis (responders vs. nonresponders). Total number of expanded clones in each sample: NR-3 = 8; NR-4 = 21; R-3 = 18; R-4 = 3.

Velocyto.py (v0.17.16; ref. 18) was used to generate a loom file for each sample with expressed repetitive elements masked. RNA velocity analysis was performed using the scVelo package (v0.2.3; ref. 19) by the dynamic modeling approach, and the dynamic RNA velocities were projected onto the Seurat integrated Uniform Manifold Approximation and Projection (UMAP) embedding.

Primary tumor biopsies (N = 47; n = 25 responders, n = 22 nonresponders) were collected prior to immunotherapy treatment as part of an ongoing biorepository effort at Mayo Clinic (IMPRESS). RNA-sequencing (RNA-seq) libraries for tumor biopsy samples [melanoma and non–small cell lung cancer (NSCLC)] were prepared by a third party (TEMPUS) according to their platform. The RNA-seq paired-end sequence data was then rerun using MAP-RSeq version 3.0.0 (20), an integrated RNA-seq bioinformatics pipeline developed at the Mayo Clinic for comprehensive analysis of raw RNA-seq paired-end reads. Normalized gene expression was calculated as explained previously (21). The abundance of NKG7 mRNA was extracted from the normalized RNA-seq data and then subjected to ROC analysis, which was performed using the R package “Epi” version 2.40.

PBMCs from patient samples were thawed, washed in complete RPMI (catalog no. 11875093, Gibco), and then resuspended in FACS buffer (1X PBS, 2 mmol/L EDTA, and 3% FBS; catalog no. 10437028, Gibco) at a concentration of 1 × 10⁶ cells/100 μL (see Supplementary Table S4 for reagent details). Cells were then stained with antibodies for surface molecules for 30 minutes at 4 °C. Antibodies used were specific for CD3, CD8, PD-1, and CX3CR1 (see Supplementary Table S3). Cells were then washed once in FACS buffer and resuspended in Fixation Buffer (catalog no. 420801, Biolegend) for 20 minutes at 4°C. After fixation, cells were washed once with ICS Perm Wash Buffer (catalog no. 421002, BioLegend), and then stained for the intracellular molecule NKG7 (catalog no. 30-69AA, Epigentek) for 1 hour at 4°C in 100 μL of ICS Perm Wash Buffer. Following intracellular staining, cells were washed once with ICS Perm Wash Buffer and then resuspended in FACS buffer for flow cytometry analysis, which was completed on a Cytoflex LX (Beckman Coulter; DAQ Version: V2.233, MCB Version: V3.01) running CytExpert software (RRID:SCR_017217).

MCF-7, A375, Hut78, P815, and EL-4 cell lines were purchased as validated cell lines from ATCC (from 2010 to 2020) and cultured in supplemented DMEM. Aliquots of each cell line were cultured a maximum 30 days after being thawed. They have not been reauthenticated in the past year. B16-OVA cells were a kind gift from Dr. Richard Vile, Mayo Clinic, and were cultured in complete RPMI under selection with G4-18 (0.8 mg/mL; Gibco, catalog no. 11811-031). Further details regarding cell culture reagents can be found in Supplementary Table S4. All cell lines were assessed for Mycoplasma contamination via PCR and were negative.

The rabbit polyclonal antibody to NKG7 (UniProtKB – Q16617) was obtained by immunizing a rabbit with keyhole limpet hemocyanin-conjugated NKG7 peptide DFWFEAVGPTHSAHSGLWPTGHGDI (Cocalico Biologicals). The anti-NKG7 polyclonal rabbit serum was affinity purified using Sulfolink (catalog no. 44895, Pierce Chemical) as per manufacturer's instructions. Supplementary Figure S1 shows that the NKG7 polyclonal antibody detects a single band at 17 kDa in wild-type (WT) CD8⁺ primary T cells but does not detect the same band in NKG7 knockout (KO) cells, validation of its specificity. The full membrane of this blot can be found in Data File S1.

Expanded CD8⁺ T cells were washed once in serum-free (SF) RPMI, resuspended with SF-RPMI at 1.5 million cells/mL, and rested in a cell culture incubator (37°C, 5% CO₂) for 10 minutes. 180,000 cells were plated directly on 18-mm round coverslips (catalog no. 1.5; Electron Microscopy Sciences) precoated with poly-L-lysine [0.1% in H₂O (w/v); MilliporeSigma, catalog no. P8920) and incubated in a cell culture incubator for 2 minutes. Cells on coverslip were then fixed with 4% paraformaldehyde (PFA) in PBS for 18 minutes and permeabilized with 0.15% Triton-X 100 in PBS for 4 minutes at room temperature. Coverslips were stained with antibodies specific for NKG7, LAMP1 (Santa Cruz; clone H4A3), or perforin (BD Biosciences; clone δG9) antibodies, followed by donkey anti-mouse IgG or donkey anti-rabbit IgG antibodies conjugated with Alexa Fluor-488 (Thermo Fisher, catalog no. A21202; A21206) or Alexa Fluor-568 (Thermo Fisher Scientific, catalog no. A10037; A10042). Polymerized F-actin was detected using Alexa Fluor-647–conjugated phalloidin (Thermo Fisher Scientific, catalog no. A22287). Coverslips were mounted on glass slides using Antifade mounting reagent (VectorLabs, catalog no. 1000-10). NKG7 colocalization with LAMP1 and perforin was examined using an Airyscan-equipped LSM800 confocal microscope with a 63X oil immersion objective of NA 1.4 (Zeiss). Z-stack images containing 6 to 10 single T cells were taken in each experiment, and experiments were independently repeated twice. Acquired raw images were processed with Airyscan on Zen Blue software (version 2.6; Zeiss) and exported to tiff format. Colocalization analysis among NKG7, LAMP1, and perforin in expanded CD8⁺ T cells was performed and automated using a customized macro program within ImageJ (version 1.45s; NIH; RRID:SCR_003070). For each channel, maximum intensity projection images were generated from z-stack images. A whole cell region was then defined as the area enclosed by NKG7 fluorescence. Pearson correlation coefficient between two fluorescence channels within a cell was calculated using a Coloc2 plug-in within ImageJ with the Costes automatic threshold setting (22).

Human CD8⁺ T cells were isolated from healthy donors (see “Processing of peripheral blood samples”) and then centrifuged at 200 × g for 10 minutes prior to nucleofection (4D nucleofector system, Lonza). 10 million cell aliquots were combined with 500 pmol/L siRNA (siControl or siNKG7) in 100 μL P3 Buffer (Lonza Kit, catalog no. V4XP-3024) per nucleocuvette. Program FI-115 was used. Following transfection, cells were transferred to RPMI media (no FBS, no cytokines) for recovery. After 4 hours, warmed media was added such that each well contained RPMI with 10% FBS, rhIL2: 10 U/mL, rhIL15: 10 ng/mL, and rhIL7: 10 ng/mL (reagent details in Supplementary Table S3). The next morning, cells were stimulated as indicated in each experiment.

• Control siRNA: ON-TARGETplus Nontargeting pool (catalog no. D-001810-10-20, Dharmacon)

• NKG7 siRNA: SMARTpool ON-TARGETplus (catalog no. L-016047-00-0020, Dharmacon)

To measure the efficiency of NKG7 knockdown, cells were transfected as described above, activated with anti-CD3/anti-CD28 beads (catalog no. 11132D, Gibco), and collected at 48 hours postactivation in RLT buffer (Qiagen, catalog no. 79216). RNA was then isolated using Qiagen RNeasy Plus Mini Kit (catalog no. 74134). Reverse transcription-reaction was completed using SuperScript III Reverse Transcriptase (catalog no. 18080–400, Invitrogen) and 200 ng of RNA template per sample. qRT-PCR analysis was completed using a Quant-Studio 3 Real-Time PCR System (Applied Biosystems) and TaqMan Fast Advanced Master Mix (catalog no. 444455, Applied Biosystems). Taqman assays included: NKG7 (Hs01120688-gl) and 18S as a control (Hs99999901-Sl), both from Applied Biosystems. All samples were run in triplicate. Relative expression was calculated using the 2^–ΔCt method.

CD8⁺ T cells from healthy donors were pelleted via centrifugation at 3,000 rpm for 15 minutes and then lysed in NP-40 buffer (20 mmol/L Tris-CL, pH 8.0, 137 mmol/L NaCl, 10% glycerol, 1% NP-40, 2 mmol/L EDTA) and concentrations measured via protein assay using Bio-Rad reagent (catalog no. 500-0006). Once diluted to equal concentrations, samples were combined with Laemmli sample buffer (Bio-Rad, catalog no. 1610747) and run using a Mini-PROTEAN Electrophoresis and Transfer system (Bio-Rad, catalog no. 1658004). Membranes were incubated with primary antibody overnight at 4°C (antibodies used were specific for NKG7, β-actin, ETS-1, or GAPDH; details listed in Supplementary Table S3). Secondary antibodies were added the next morning for 1 hour [horseradish peroxidase (HRP)–conjugated anti-rabbit and anti-mouse; details in Supplementary Table S3]. Signal was visualized using SuperSignal West Pico PLUS Chemiluminescence Substrate Kit (catalog no. 34577, Thermo Fisher Scientific) on a Syngene G:Box Chemi XX6 system running GeneSys software (V1.6.1.0; RRID:SCR_015770). Images were formatted for figures using Microsoft PowerPoint (Microsoft Office 2016). Full images are included in the Supplementary Data File S1.

CD8⁺ T cells isolated from healthy donors were activated with anti-CD3/CD28 beads (catalog no. 11132D, Gibco) for 48 hours. After 48 hours, beads were removed using a magnet. CD8⁺ T cells were then combined with MCF7 tumor cells at a ratio of 1:5 (Target:T cell), spun at 20 × g for 1 minute and then cultured for 1 hour at 37°C. The cocultures were then gently spun at 20 × g for 1 minute at 4°C with no brake. The cells were fixed in 1% osmium tetroxide (catalog no. 251755, Millipore Sigma) for 1 hour or at 4°C overnight, then postfixed for 1 hour in 1% osmium tetroxide. For transmission electron microscopy (TEM), the samples were dehydrated, embedded in Spurrs resin (catalog no. 14300; Low Viscosity Embedding Kit, Electronic Microscopy Sciences), sectioned at 90 nm, and observed using a Joel 1400 electron microscope (Joel USA Inc.). For quantification, images of individual tumor cell:T-cell interactions were printed and the number of granule structures within the T cells along the tumor cell:T-cell interface were counted manually by two different readers.

Tissue samples from human tonsils, metastatic melanoma in the small bowel, and squamous cell carcinoma of the tongue were originally fixed with 10% formalin for 48 hours at room temperature followed with ethanol tissue dehydration and replacement with Xylene before embedding tissues into paraffin blocks. The paraffin tissue sections were cut at 5-μm and deparaffinized with xylene and rehydrated with a graded series of ethanol. Slides were rinsed in distilled water and placed in Diva Decloaker 10X (catalog no. DV2004LX, Biocare Medical) antigen retrieval solution diluted to 1X. Slides were placed in Biocare Medicals pressure cooking Decloaking Chamber and heated to 121°C for 30 seconds then heated at 90°C for 10 seconds. Slides were removed from the chamber rinsed with running distilled water and incubated for 5 minutes in Dako TBS wash buffer (catalog no. S3006). All of the following steps were conducted at room temperature. Tissue sections were incubated for 10 minutes in PeroxiAbolish (catalog no. PXA969L, Biocare Medical), washed 2 × 5 minutes in wash buffer, incubated for 10 minutes in Background Sniper (catalog no. BS966L, Biocare Medical), and then washed 2 × 5 minutes in wash buffer. Sections were incubated for 2 hours in anti-human NKG7 rabbit polyclonal antibody diluted 1:500 in Renaissance Background Reducing Antibody Diluent (catalog no. PD905L, Biocare Medical; final concentration 1.56 μg/mL) and washed 3 × 5 minutes in wash buffer. Subsequently, slides were incubated for 10 minutes in rabbit probe (catalog no. M3R513H, Biocare Medical), washed 2 × 5 minutes and incubated 10 minutes in rabbit HRP polymer (Mach 3 anti-rabbit HRP polymer kit; catalog no. M3R513H, Biocare Medical) and washed 2 × 5 minutes. Staining was visualized with DAB (catalog no. BDB2004L, Biocare Medical) for three minutes. Slides were counterstained with hematoxylin (catalog no. 7211, Richard Allen-Scientific). Sections were blued in running tap water and dehydrated in a graded series of ethanol and cleared in xylene. Coverslips were mounted with Permount (catalog no. SP15-100, Thermo Fisher Scientific). An EVOS imaging system (Invitrogen) was used to capture images of slides with representative NKG7 staining.

CD8⁺ T cells from healthy donors were isolated and transfected with siRNA, as described above, then cultured with anti-CD3/anti-CD28 beads. After 48-hour stimulation, CD8⁺ T cells were collected (beads removed) and labeled with lysotracker (Red 555, Invitrogen, catalog no. L7528) for 1 hour and WGA (Far Red 647, Invitrogen, catalog no. W32466) for 10 minutes in the dark. MCF-7 cells were stained with Calcein (Green 488) for 30 minutes, tumor cells and CD8⁺ T cells were suspended at 12,500 × 10 μL and 25,000 × 10 μL in serum-free RPMI and put on ice for 10 minutes. Target cells and effector cells were then mixed together at a ratio of 1:2 and spun at 200 rpm for 5 minutes, then put on a poly-d-lysine coated glass bottom microwell dishes (catalog no. P35GC-1.0-14-C, MatTek Corporation). Images were acquired every minute for 30 minutes using an LSM 780 laser scanning confocal microscope (Zeiss) equipped with an EC-Plan-Neoflouar 40×/0.75 DIC (air). Particle count was performed using Fiji software (RRID:SCR_002285). Segmentation and manual thresholding were used to distinguish lysotracker-positive particles from background. Only cells with individual vesicles were selected to avoid false quantification created by vesicle aggregation. An average of 10 to 12 cells per image were used for particle counting. Four images used for each condition (siControl vs. siNKG7). The open-source Fiji plugin Trackmate was used to perform single particle tracking (23). Track length and velocity (distance per time) were determined for lysotracker-positive vesicles in CD8⁺ T cells in contact with tumor cells or contact-free. Two independent experiments were performed for each condition and three regions were randomly selected for time-lapse imaging (approximately 10 CD8⁺ T cells per field of view).

CD8⁺ T cells were isolated from healthy donors, expanded by placing them on a 6-well plate precoated with 1 μg/mL anti-CD3 (clone OKT3, catalog no. 317326, BioLegend) and anti-CD28 (clone CD28.2, catalog no. 302934, BioLegend), and cultured in RPMI with 10% FBS, 100 IU/mL penicillin, 100 mg/mL streptomycin, and 2 mmol/L L-glutamine supplemented with 100 IU/mL recombinant human IL2 (see Supplementary Table S4 for reagent details).

Cas9/RNP complex was prepared and transfected into primary human CD8⁺T cells as previously described (24). Equimolar amount of NKG7-specific CRISPR RNA (crRNA; GGGACCGGCAGAGCTCCATG; IDT) and trans-activating crispr RNA (tracrRNA; IDT) were mixed, incubated at 95°C for 5 minutes, and cooled down to room temperature. Four hundred and fifty picomoles of annealed crRNA-tracrRNA was mixed with 180 pmol of purified Cas9 protein (Alt-R s.p. Cas9 Nuclease V3, IDT, catalog no. 1081059) and incubated at room temperature for 10 minutes, creating the Cas9/RNP complex. Prepared Cas9/RNP complex was transfected into human CD8⁺ T cells using a 4D nucleofector system (Lonza). 10 × 10⁶ purified CD8⁺ T cells were resuspended in 40 μL of P2 nucleofection solution (catalog no. V4XP-2032, Lonza) and mixed with Cas9/RNP complex. Cells with mixture were then nucleofected using program EH-100. Following transfection, cells were transferred into prewarmed complete RPMI with 100 IU/mL IL2 and rested. Seven days posttransfection, cells were stimulated and expanded on a tissue-culture plate precoated with 1 μg/mL of anti-CD3 (clone OKT3) and 1 μg/mL of anti-CD28 (clone CD28.2). All experiments were performed 14 to 21 days postactivation.

PBMC's from patients with melanoma were preserved with CryoStor (catalog no. 07952, StemCell) and thawed in complete RPMI containing 10% FBS, rhIL2: 10 U/mL, rhIL15: 10 ng/mL, and rhIL7: 10 ng/mL, and kept at 37°C overnight. The next morning, viable cells were transfected with control mRNA or NKG7 mRNA. NKG7 mRNA was prepared by Tri-Link based on NCBI reference sequence NM_005601. Control mRNA was CleanCap OVA mRNA, also from Tri-Link (catalog no. L-7610). Final mRNA concentration for transfection was 200 μg/mL in a 20-μL volume of P3 Buffer (Lonza Kit, catalog no. V4XP-3024) and program FI-115 on the 4D Nucleofector (Lonza) was used. CD8⁺ T cells from healthy donors were freshly isolated and transfected the same day with either Control mRNA or NKG7 mRNA (final mRNA concentration of 100 μg/mL in a 100-μL volume of P3 Buffer and program FI-115 on the 4D Nucleofector. Following transfection, cells were transferred to RPMI media (no FBS, no cytokines) for recovery. After 4 hours, warmed media was added such that each well contained RPMI with 10% FBS, rhIL2: 10 U/mL, rhIL15: 10 ng/mL, and rhIL7: 10 ng/mL. The next morning, soluble anti-CD3 (5 μg/mL; catalog no. 555336, BD Pharmingen) and anti-CD28 (5 μg/mL; catalog no. 555725, BD Pharmingen) were added for activation, with or without anti–PD-1/PD-L1 blockade (atezolizumab or pembrolizumab at 20 μg/mL) for 72 hours. Calcein assay was used to measure cytotoxicity as described above. For the NKG7 mRNA rescue experiment, CD8⁺ T cells transfected with Cas9/RNP were expanded for 19 to 21 days before transfection with NKG7 mRNA. After mRNA transfection, cells were rested for 2 days before degranulation assay.

Expanded CD8⁺ T cells and P815 cells were washed and resuspended in complete RPMI at 2.5 to 5 × 10⁶ cells/mL. Anti-human CD3 (clone OKT3) was added to P815 cells at 2.5 to 5 μg/mL (target). Hundred microliters of CD8⁺ T cells (0.25–0.5 × 10⁶ cells) was then added to same number of target and cells were incubated at 37°C for multiple times. At each time, cells were collected, labeled with Zombie Aqua viability dyes in PBS according to the manufacturer's instruction (BioLegend, catalog no. 423101), and stained using the following antibodies in FACS buffer (1% FBS, 1 mmol/L EDTA, 0.09% NaN₃ in PBS): FITC antimouse CD16/32, PerCP anti-human CD8, and PE anti-human CD107a (antibody details in Supplementary Table S3). After staining, cells were washed in PBS and fixed with 1% PFA in PBS. All samples were acquired on a BD FACSCanto II instrument with FACSDiva software. Percentage of CD107a⁺ population among CD8⁺CD16/32⁻ population was examined using a FlowJo software (RRID:SCR_008520; version 9.96).

CD8⁺ T cells from healthy donors were isolated and transfected with siRNA or mRNA as described above. Cells were harvested at a density of 1 × 10⁶/mL and labeled with Indo-1 AM (5 μmol/L; catalog no. 565879, BD Pharmingen) at 37 °C for 30 minutes. Cells were washed with complete RPMI and incubated in complete RPMI for another 30 minutes at 37 °C. Finally, cells were suspended and kept in 37°C water bath before acquisition on an LSR II cytometer (BD Biosciences) running FACSDiva v6.1.3 software (RRID:SCR_001456). After baseline acquisition for 1 minute, the analogue of nicotinic acid, adenine dinucleotide phosphate (NAADP-AM; 10 μmol/L; catalog no. 21000, AAT Bioquest) was added and sample collection immediately resumed. In addition, we used trans-Ned-19 (an inhibitor of NAADP pathway; 10 μmol/L, Enzo Life Sciences, catalog no. ALX-270–503-M001) or Nigericin (an ionophore agent that depletes acidic Ca²⁺ stores; 10 μmol/L, Sigma, catalog no. SML1779) in culture medium 30 minutes before the addition of NAADP-AM to test effects on calcium release at basal level and induced by NAADP-AM. Flow data was analyzed by FlowJo, using the kinetic analyzer function (FlowJo, LLC).

The Mayo Clinic Institutional Animal Care and Use Committee (IACUC) approved all animal experiments and mice were maintained under pathogen-free conditions in the animal facility at Mayo Clinic.

Spleens were harvested from OT-1 transgenic mice [C57BL/6-Tg (TcraTcrb)1100Mjb/J, Jackson Laboratory] and processed to single cell. Cells were then divided into aliquots of 10 × 10⁶ cells/15 mL tube and spun down at 200 × g for 10 minutes. Each pellet was then resuspended in 100 μL of nucleofection media (P3 Lonza kit, catalog no. V4XP-3024) containing either control mRNA (EGFP) or murine NKG7 mRNA (8.28 ug/10 × 10⁶ cells) and nucleofected using the 4D nucleofector system (Lonza) on program DN-100. NKG7 mRNA was prepared by Tri-Link from NCBI reference sequence: NM_024253. Control mRNA was Tri-Link CleanCap EGFP mRNA (L7601). Following transfection, cells were transferred to RPMI media (no FBS, no cytokines) for recovery. After 4 hours, warmed RMPI was added such that each well contained 10% FBS, rhIL2: 10 U/mL, rhIL15: 10 ng/mL, and rhIL7: 10 ng/mL. The next morning, 1 μg/mL OVA_257–264 peptide (catalog no. A5503, Sigma) was added for activation. Forty-eight hours after stimulation, cells were collected, counted and prepared for in vitro and in vivo assays.

To measure NKG7 mRNA levels after 48 hours of activation, cells from three separate OT-1 mice were transfected as described above, activated, then collected in RLT buffer (Qiagen, catalog no. 79216) and processed using the Qiagen RNeasy Plus Mini Kit according to the manufacturer's instructions (catalog no. 74134). The RT reaction was completed using SuperScript III Reverse Transcriptase (catalog no. 18080–400, Invitrogen) with 125 ng of RNA template per sample. qRT-PCR analysis was completed using a Quant-Studio 3 Real-Time PCR System (Applied Biosystems) and TaqMan Fast Advanced Master Mix (catalog no. 444455, Applied Biosystems). Taqman assays included β-actin (Mm00607939–s1) and NKG7 (Mm01203147_g1), both from Applied Biosystems. All samples were run in triplicate. Relative expression was calculated using the 2^–ΔCt method.

Cells from whole spleen of OT-1 transgenic mice were transfected with control or NKG7 mRNA, as described above. EL-4 mouse T-cell lymphoma cells were incubated with 1 μg/mL of OVA_257–264 peptide in RPMI media (no FBS) for 1 hour at 37°C. EL-4 cells were then washed twice with HBSS and labeled with Calcein-AM (5 μmol/L; incubation at 37°C in dark, 30 minutes, shaking occasionally). Following calcein-labeling, cells were again washed twice with HBSS and then resuspended at 1 × 10⁵/mL in RPMI (no FBS). EL-4 cells and activated OT-1 cells were then mixed together at indicated target to effector ratios in RPMI without FBS, and seeded into the wells of a 96-well U-bottom plate. Remaining steps of calcein assay protocol and percent cytotoxicity calculations were identical to those listed above (see “Calcein release cytotoxicity assay”).

0.5 × 10⁶ B16-OVA cells suspended in 100 μL of sterile PBS were injected SubQ into C57BL/6 WT mice (The Jackson Laboratory, catalog no. 000664). Cells from whole spleen of OT-1 transgenic mice were transfected with control or NKG7 mRNA, as described above, then activated for 48 hours with OVA_257–264 peptide prior to injection. On day 6 and day 13 of B16-OVA tumor growth, 1 × 10⁶ of the activated OT-1 cells were injected into each mouse (peritumoral injection). Perpendicular tumor measurements were taken twice per week by calipers. Mice were euthanized if tumors became ulcerated or grew beyond 200 mm² in size, in compliance with animal care guidelines. For experiments testing combination antibody and OT-1 treatment, mice and cells were prepared as above, but mice were treated with either anti–PD-1 alone (clone G4, 100 μg/mouse; Millipore Sigma, catalog no. mabc1132), or anti–PD-1 in combination with OT-1 T cells transfected with either control or NKG7 mRNA. Antibody was given on day 5 and day 11 of tumor growth. 1 × 10⁶ OT-1 cells were given on days 6 and 12 of tumor growth.

The DNA sequence located 500 bps upstream of NKG7 (chr19:51,371,620-51,372,654) was extracted from UCSC Genome Browser (RRID:SCR_005780) from Human Dec. 2013 (GRCh38/hg38) reference genome. The virtual laboratory PROMO's (http://alggen.lsi.upc.es/cgi-bin/promo_v3/promo/promoinit.cgi?dirDB=TF_8.3) predictive software was used to generate a list of potential transcription factors that could bind the promoter sequence (dissimilarity margin of ≤ 5%). An additional search was conducted using P-Match software (http://gene-regulation.com) using 1,000 bps upstream of NKG7 and the same reference genome. All hits were then cross-referenced against the scRNA-seq data sets to determine if the transcripts for these candidate transcription factors could be detected in the CD8⁺ T cells. CEBPB, ETS1, REL, STAT4, and YY1 had detectable expression and were therefore selected for initial knockdown in Hut78 via siRNA nucleofection.

siRNA knockdown was completed in Hut78 or primary CD8⁺ T cells using Lonza P3 Primary 4D Nucleofection Kit (catalog no. V4XP-3024) and Lonza 4D Nucleofector (see Supplementary Table S5 for siRNA catalog numbers). A pulse code of DN100 was used for the Hut78 cell line and code FI-115 was used for primary cells. Following transfection, cells were transferred to RPMI (no FBS, no cytokines) for recovery. After 4 hours, warmed media was added such that each well contained RPMI with 10% FBS, rh IL2: 10 U/mL, rh IL15: 10 ng/mL, and rh IL7: 10 ng/mL. The cells were then collected 48 hours after nucleofection in RLT buffer from the Qiagen RNeasy Plus Mini Kit (catalog no. 74134), followed by RNA isolation using the kit and reverse transcription using SuperScript III Reverse Transcriptase kit (catalog no. 18080-400, Invitrogen). Two hundred and twenty nanograms of RNA template was used per sample for the Hut78 reverse transcription reactions and 80 to 450 ng of RNA template was used per sample for the CD8⁺ primary T-cell reactions due to varying RNA yields from primary T cells, but all samples were normalized to a housekeeping gene. qRT-PCR analysis was then completed using a Quant-Studio 3 Real-Time PCR System (Applied Biosystems) and SYBR green reagents (Invitrogen, catalog no. A25742). qRT-PCR primer sequences used are listed in Supplementary Table S6. All samples were run in triplicate. Relative expression was calculated using the 2^–ΔCt method.

Thermo Fisher Scientific Pierce Magnetic ChIP Kit (catalog no. 26157) was used per the manufacturer's protocol. MNase digestion timing was 15 minutes at 37°C for 4 × 10⁶ cells. Sonication was completed using a Diagenode Biorupter with for 12 cycles, 30 seconds on/30 seconds off, at 4°C. Chromatin was incubated with Ets1 antibody (catalog no. 14069, Cell Signaling Technology) or 2-μL rabbit IgG (provided in the Magnetic ChIP Kit) overnight at 4 °C. After DNA recovery, quantification was completed with the QuantStudio 3 Quant-Studio 3 Real-Time PCR System and SYBR green reagents. Primer sequences used are listed in Supplementary Table S6.

ETS1 siRNA or control siRNA was transfected into primary CD8⁺ T as described above. Cells were stimulated with anti-CD3/anti-CD28 beads 24 hours after nucleofection. Following 48 hours of activation, cells were washed, counted, and combined with calcein-labeled MCF-7 target cells. Calcein assay and percent cytotoxicity calculations were completed as described above (see “Calcein release cytotoxicity assay”).

The statistical tests performed are stated within the figure legends. Analysis of scRNA-seq data was performed as stated above. The statistical analysis for all other data was performed using Graphpad Prism (RRID:SCR_002798) v8.4.3 or later. Bar height represents mean, and error bars are SE of the mean, unless otherwise stated.

scRNA-seq data has been deposited at the Gene Expression Omnibus (GEO; RRID:SCR_005012) under accession number GSE164237. Raw data for all other graphs shown have been submitted in Data File S1. Any other data can be obtained upon request.

To identify gene expression levels linked to effective CTL killing, we isolated CD8⁺ T cells from the peripheral blood of patients with advanced stages of melanoma who received anti–PD-1 therapy (Fig. 1A; Supplementary Table S2). Matched samples from BL and 12 weeks PT from 2 complete responders (R-1 and R-2) and two nonresponders (NR-1 and NR-2) were compared using scRNA-seq analysis. UMAP clustering of gene expression profiles from all eight samples identified four cell subsets: naïve-like, effector, effector exhausted (KLRB1), and a cluster high in mitochondrial gene expression (Fig. 1B and C). The first three of these clusters exhibited substantial overlap with those reported by Fairfax and colleagues (11), indicating these broad phenotypes are consistent across studies of human peripheral CD8⁺ T cells.

The highest expression levels of genes encoding known cytolytic molecules [i.e., granulysin (GNLY); perforin-1(PRF1); granzyme H (GZMH); granzyme B (GZMB)] were all associated with the effector cluster (Fig. 1D), which was chosen for further analysis. We hypothesized that these are likely the peripheral CTLs shown to replenish the tumor-infiltrating CD8⁺ T cells upon treatment with immune checkpoint inhibitors (8, 9). Thus, any differences between responders and nonresponders in this cell subset may ultimately affect clinical outcome. We found that the effector population was represented in all patients and its relative proportion in comparison with the total CD8⁺ T-cell profile did not change within each individual between BL and pos-treatment timepoints (Fig. 1E). We then completed differential gene expression analyses and looked for gene expression changes that occurred posttreatment relative to baseline within each individual, i.e., a paired (PT vs. BL) integration analysis for each patient. Among those changes, we asked what differentiated responders from nonresponders (Fig. 1F). We found 46 genes differentially expressed by both nonresponder samples and 30 genes differentially expressed by both responder samples (Supplementary Table S7). Among the genes upregulated in both nonresponders were TCF7, CCR7, and NELL2, all of which are associated with naive and early-differentiated memory T cells. Genes upregulated in both responders included effector-related genes GZMH, GZMM, GZMA, and CCL5, as well as KLRG1, which is known to be expressed on antigen-experienced, fully differentiated effector CD8⁺ T cells (25). Both responders also upregulated TIGIT, an immune checkpoint with an ITIM domain and suppressive function that upregulated in peripheral CD8⁺ T cells of responders to anti–PD-1 therapy (26). However, we were most interested in those genes showing opposite trends in responders versus nonresponders and found one gene satisfying this criterion: NKG7 was upregulated in the effector cells of the two responders and downregulated in the two nonresponders by 12 weeks PT (Fig. 1G–I). Thus, we found that although all patients harbored peripheral effector T cells, those T cells were not all equally effective in eliminating tumors as evidenced by the clinical response. Our initial dataset suggested that appropriate expression of NKG7 in cytotoxic CD8⁺ T cells could be important for antitumor T-cell activity, particularly in the context of anti–PD-1 therapy.

Expansion of CD8⁺ T-cell clones is indicative of antigen recognition, and the number of large CD8⁺ T-cell clones in peripheral blood samples correlates with response to anti–PD-1-therapy (11). Whether molecular differences exist between these large clones in responders versus nonresponders has not been shown. Using an additional set of patient samples (n = 2 responders, n = 2 nonresponders; paired samples from BL and PT) and scRNA-seq analyses, we found that unsupervised clustering yielded 4 clusters identical to the first set of patient samples, shown in Fig. 1B, along with one small additional cluster, “LYZ,” which may represent a group of T cells with capacity for antimicrobial function, or possibly a small number of natural killer (NK) cells/macrophages (Fig. 2A; refs. 27–29). Because this set of scRNA-seq included both gene expression and TCR α/β sequencing, we were able to compare the top five largest CD8⁺ T-cell clones in each sample (“largest” meaning clones with the highest number of identical TCR sequences represented among all T cells sequenced for each individual at each timepoint) (Fig. 2A–C). In both responders and nonresponders, these large clones were predominantly composed of the effector phenotype with a cytotoxic profile (GNLY, effector cluster; Fig. 2D). The large clones from responders, however, exhibited higher expression of genes associated with T-cell cytotoxicity, including NKG7 (Fig. 2E and F). We did additional analyses to determine which specific clones were expanded after therapy as opposed to baseline. These were not necessarily the largest clones, but they did increase in size upon anti–PD-1 treatment. Again, we found that the most expanded clone (Fig. 2G) and the top five most expanded clones (Fig. 2H) in responders exhibited higher expression of NKG7 versus the expanded clones in nonresponders. Thus, both scRNA-seq cohorts revealed a correlation between NKG7 expression in cytotoxic CD8⁺ T cells and response to anti–PD-1 therapy.

We next sought to corroborate the NKG7 gene expression differences between responders and nonresponders at the protein level. The chemokine receptor CX3CR1 marks a population of peripheral CD8⁺ T cells of the effector memory phenotype that are responsive to anti–PD-1/L1 therapy (30–32). We found a significant increase in the frequency of CX3CR1⁺CD8⁺ T cells that expressed NKG7 as compared with CX3CR1⁻CD8⁺ T cells (Fig. 3A). In addition, we observed a decreased frequency of CX3CR1⁺CD8⁺ cells expressing NKG7 in nonresponders to anti–PD-1 therapy, as compared with responders, at both the baseline and posttreatment timepoints (Fig. 3B).

As our initial studies had been in peripheral blood, we next considered whether NKG7 expression in tumor biopsies was associated with treatment response. Although NKG7 has been measured via RNA-based methods in bulk tumor samples, NKG7 protein expression has never been visualized within human tissues. Thus, we used IHC staining to confirm that NKG7 was expressed within lymphocytes in both normal lymphoid and tumor tissues (Fig. 3C). We then examined NKG7 expression in a set of primary tumor biopsies (n = 47; melanoma and NSCLC) in which bulk RNA-seq was performed. We found that prior to anti–PD-1/L1 treatment, nonresponders had lower levels of NKG7 mRNA expression when compared with responders (Fig. 3D). Further, the level of NKG7 expression within the tumor at BL was reasonably predictive of response to immune checkpoint blockade (AUC = 0.86; Fig. 3E). Taken together, our RNA- and protein-based analyses suggest that the lack of appropriate expression of NKG7 in cytotoxic CD8⁺ T cells may contribute to nonresponsiveness to anti–PD-1/L1 immunotherapy.

The efficacy of cancer immunotherapy is ultimately dependent upon whether T cells can effectively kill tumor cells. Since lower levels of NKG7 were found in the effector CD8⁺ T cells of nonresponders to anti–PD1 treatment, we next considered the possibility that NKG7 may play a functional role in T-cell cytolytic activity. This was plausible given the initial characterization of NKG7 (also named GMP-17) as a molecule associated with cytolytic granules (33, 34). In addition, a recent study using murine CD8⁺ T cells found that NKG7 colocalizes with the known cytolytic granule-associated protein LAMP1 (CD107a) upon induction of cytolytic granule exocytosis and that knock-out of NKG7 decreased NK cell–mediated cytotoxicity in mouse metastatic tumor models (35). We sought to corroborate these findings in primary human CD8⁺ T cells.

Immunofluorescent staining confirmed that NKG7 is maintained in close proximity to CD107a and is somewhat associated with perforin in human cells (Fig. 4A and B). Notably, we found that depletion of NKG7 by siRNA knockdown resulted in a significant reduction of T cell–mediated cytotoxicity (Fig. 4C and D; Supplementary Fig. S1A and S1B). To determine the mechanism by which decreased NKG7 expression results in reduced CD8⁺ T cell–mediated cytotoxicity, we next quantified cytolytic granule release after NKG7 depletion. We found that CRISPR/Cas9-mediated KO of NKG7 in primary human CD8⁺ T cells resulted in a reduction in degranulation following T-cell stimulation (Fig. 4E; Supplementary Fig. S1C). We further confirmed that this was an NKG7-specific effect by restoring NKG7 expression to the NKG7 KO cells (via NKG7 mRNA transfection), which rescued the phenotype and restored degranulation to normal levels (Fig. 4F and G).

We next examined the interaction between tumor cells and CD8⁺ T cells via electron microscopy. We observed direct cell–cell contact between T cells and tumor cells in both control and NKG7 siRNA conditions (Fig. 4H). However, we noted a decreased number of cell membrane extensions between T cells and tumor cells when T cells were transfected with NKG7 siRNA. In addition, we observed small granules present within the presumed immune synapse between control siRNA–transfected T cells and the tumor cells, but fewer of these granules were present at the synapse within T cells that were deficient in NKG7 (Fig. 4I). This agreed with our CD107a results. We next considered whether there were fewer granules overall present (per cell) when NKG7 was depleted. A broader examination of the electron microscopy images revealed that the morphology and densities of cytolytic granules were similar in control and NKG7 knockdown CD8⁺ T cells (Supplementary Fig. S2A), but the overall number of cytolytic granules was decreased in NKG7 knockdown CD8⁺ T cells compared with control cells (Supplementary Fig. S2B). In addition, we found the granules were dispersed with certain distances in control CD8⁺ T cells, but there were more aggerated granules in NKG7 KO CD8⁺ T cells (Supplementary Fig. S2C).

To better characterize the dynamics of granule trafficking, we prepared cocultures of tumor cells and CD8⁺ T cells and monitored them using real time confocal microscopy. CD8⁺ T cells were labeled with lysotracker to visualize lysosomal vesicles and wheat germ agglutinin (WGA), which binds to glycoproteins of the cell membrane (Fig. 5A). Labeled T cells were then cocultured with calcein-stained MCF-7 tumor cells. We observed fewer cytotoxic granules per cell in the NKG7 siRNA cells versus control siRNA cells (Fig. 5B). In terms of cytotoxicity, control CD8⁺ T cells began killing tumor cells within 10 minutes of coculture, with high speed and long-distance movement of granules toward the tumor–immune cell interface within those T cells as they attacked their targets. In contrast, NKG7 siRNA–transfected CD8⁺ T cells engaged tumor cells, but their granules showed reduced speed and shorter distance movements within the cytoplasm (Fig. 5C and D). Thus, we conclude that reduced NKG7 affects the overall number of granules within T cells, and that the granules that remain display decreased velocity and trafficking distance upon tumor cell target engagement.

Exocytosis and trafficking of cytolytic granules is Ca²⁺-dependent (36), and the four membrane-spanning domains of NKG7 are structurally similar to the γ subunit of an L-type voltage-gated calcium channel (34). Thus, we next measured and compared the intracellular Ca²⁺ levels of CD8⁺ T cells with or without NKG7 siRNA transfection using the calcium-sensitive fluorescent indicator Indo-1 AM. NAADP-AM, induces Ca²⁺ release from acidic organelles, such as cytolytic granules (37, 38). For this assay, T cells from healthy donors were labeled with Indo-1 AM and baseline fluorescence acquisition was performed for 1 minute via flow cytometry. Then, NAADP-AM was added, and acquisition immediately resumed to visualize peak calcium flux. When NKG7 was depleted from CD8⁺ T cells, we found that NAADP-AM–induced Ca²⁺ release was significantly decreased (Fig. 5E). Conversely, increased expression of NKG7 resulted in an increased peak calcium flux after NAADP-AM stimulation (Fig. 5F). We found in the NKG7 mRNA–transfected WT T cells there was already more calcium detected in the cytoplasm at baseline before addition of NAADP-AM, although it was not a significant increase compared with control mRNA–transfected T cells (Fig. 5F, BL). This raises the possibility that NKG7 overexpression may cause a release of calcium from other, nongranule, sources of calcium into the cytoplasm.

To test this possibility, we treated control or NKG7 mRNA transduced CD8⁺ T cells with Nigericin, an ionophore agent that depletes acidic Ca²⁺ stores and dissipates Ca²⁺ release–initiated by NAADP-AM. Treatment with Nigericin suppressed the Ca²⁺ release–initiated by NAADP-AM in both control cells and NKG7 overexpressing cells (Fig. 5G), but the decrease of Ca²⁺ release was not significant in CD8⁺ T cells with NKG7 overexpression (Fig. 5H). Similarly, when we treated those CD8⁺ T cells with Trans-Ned-19, an inhibitor of the NAADP pathway, the Ca²⁺ release initiated by NAADP-AM was inhibited in both control cells and NKG7 overexpressing cells, but was not significant in NKG7 overexpressing cells (Fig. 5G and H). The two inhibitors did not significantly decrease the basal level of calcium in the NKG7 overexpressing cells compared with control cells. Taken together, our results suggest that NKG7 may promote Ca²⁺ release from both NAADP-affected acidic organelles (like cytolytic granules) and NAADP-unaffected sources (perhaps the endoplasmic reticulum).

Collectively, these data point to a role for NKG7 in regulating calcium release from lysosomal vesicles that impacts cytolytic granule trafficking and release within cytotoxic CD8⁺ T cells.

Because reduced levels of NKG7 compromised the cytolytic efficiency of CD8⁺ T cells, we next hypothesized that by increasing expression of NKG7 we could enhance the cytolytic activity of effector CD8⁺ T cells, resulting in better response to immune checkpoint inhibitors. In the last several decades, technology has advanced such that immune cells can be effectively modified via transfection with mRNA that is produced in vitro (39), with the particular advantage that this mRNA can be transfected into nonproliferative primary T cells and presents low risk of genome mutagenesis. This opens the door for innovative therapeutic strategies designed to modulate therapeutic targets within T cells directly, through the addition of mRNA.

We found that transfection of primary human CD8⁺ T cells with NKG7 mRNA resulted in increased levels of NKG7 protein (Supplementary Fig. S3A). To test the functional impact of NKG7 overexpression in the most clinically relevant samples, we next transfected PBMCs from patients with melanoma who failed to respond to anti–PD-1 treatment (Fig. 6A). The addition of NKG7 mRNA significantly increased T-cell killing of melanoma tumor cells, and this effect was even more robust when combined with anti–PD-1 or anti–PD-L1 pretreatment (Fig. 6B and C). To confirm this was a T cell–specific effect, we also transfected isolated CD8⁺ T cells from healthy donors with NKG7 mRNA. We found that increasing NKG7 expression led to increased T-cell cytotoxicity (Supplementary Fig. S3B). Although the T-cell cytotoxicity of NKG7 mRNA–transduced CD8⁺ T cells increased significantly in culture with anti–PD-L1 compared with control mRNA cells, this increase was not significant in culture with media only or with anti–PD-1, suggesting that NKG7 mRNA transfection may not further benefit T-cell responses to anti–PD-1 therapy in healthy donors who may already have sufficient basal levels of NKG7.

To demonstrate that overexpression of NKG7 could improve CD8⁺ T-cell killing in an antigen-specific manner, we transfected OT-1 transgenic T cells with murine NKG7 mRNA or control (GFP) mRNA (Fig. 6D). OT-1 cells with increased NKG7 expression exhibited enhanced cytolytic ability when challenged with EL4 cells pulsed with OVA peptide in vitro (Fig. 6E). In addition, NKG7 mRNA–transfected OT-1 cells suppressed the aggressive growth of B16-OVA cells in vivo significantly better than control OT-1 cells (Fig. 6F–H). Using a similar ACT model, we also tested combination treatment with anti–PD-1 antibody and OT-1 cells. Although combination anti–PD-1 and NKG7 mRNA–transfected OT-1 treatment significantly decreased tumor growth and increased overall survival versus no treatment controls, the B16-OVA model was already responsive to anti–PD-1 therapy alone and the addition of NKG7 mRNA–transfected OT-1s did not add further benefit (Supplementary Fig. S4). This indicates that the endogenous T cells already had sufficient NKG7, as they were able to respond to anti–PD-1. Thus, the addition of NKG7 mRNA is likely to be therapeutically indicated only in settings where NKG7 levels are reduced or depleted, as in nonresponders to checkpoint blockade.

Little is known regarding upstream mechanisms that could be responsible for regulating NKG7 transcription. Thus, we performed in silico analysis of binding motifs contained within the NKG7 promoter sequence to identify putative transcriptional regulators (Fig. 7A; Supplementary Fig. S5). We then used siRNA to knock down these transcription factor candidates in Hut78 T cells and found that knockdown of the transcription factor ETS1 resulted in an increase in NKG7 mRNA expression (Fig. 7B). ETS1 is known to regulate genes involved in both T- and NK-cell differentiation and activation (40–42). Further, members of the ETS1 transcription factor family bind to the promoters of effector T cell–relevant genes such as PERFORIN and CFOS to regulate their expression (43–45), so we next considered whether ETS1 might bind to the promoter of NKG7. Using ChIP-qPCR, we confirmed that ETS1 was indeed associated with the NKG7 promoter in CD8⁺ T cells isolated from healthy human donors (Fig. 7C). We then knocked down ETS1 in additional healthy donor primary CD8⁺ T-cell samples and observed an increase in NKG7 expression at both the mRNA and protein levels (Fig. 7D–F). Finally, to determine whether this modulation had functional significance, we transfected CD8⁺ T cells from healthy donors with ETS1 siRNA or control siRNA followed by activation with anti-CD3/CD28 and found that ETS1 siRNA–transfected cells were better able to kill MCF-7 target cells in the calcein-based cytotoxicity assay (Fig. 7G), as we hypothesized, given the higher levels of NKG7 that result from ETS1 knockdown.

Using our scRNA-seq dataset (from Fig. 1), we next completed analysis of “RNA velocity.” RNA velocity leverages the relative ratio between intronic (unspliced) and exonic (spliced) mRNAs in scRNA-seq data to infer the rate of change in transcript abundance and thus predict the future transcriptional state for a cell (18). The underlying assumption in this analysis was that genes are initially transcribed in an unspliced manner and then spliced, such that observed intronic reads can be interpreted as corresponding to nascently transcribed mRNAs. Using the UMAP projection from Fig. 1B, we found that the overall flow of cell velocity started from the naive cells and continued over to effector cells (black arrows in Fig. 7H). More specifically, we found the velocity of NKG7 transcription gradually increases in effector cells (Fig. 7I). In contrast, EST1 gene expression demonstrated positive velocity of ETS1 in naive cells, but decreased velocity in effector cells (Fig. 7I). Thus, our scRNA-seq data from patient samples further supports our findings that NKG7 is subject to regulation by ETS1 in peripheral CD8⁺ T cells. We conclude that ETS1 is a transcriptional repressor of NKG7 expression and suggest that interventions that modulate ETS1 may represent another opportunity for optimizing CD8⁺ T-cell cytotoxicity.

This work establishes a causal connection between appropriate expression of NKG7 and the cytolytic capacity of CD8⁺ T cells, particularly in the setting of immune checkpoint inhibitor therapy. Through the use of scRNA-seq technology, we were able to assess gene expression differences in the effector T cells of responders versus nonresponders to anti–PD-1 treatment across multiple time points within individual patients. This revealed that NKG7 expression was reduced in the cells of nonresponding patients, which was confirmed at the protein level. Our subsequent studies highlighted the functional role of NKG7 in cytolytic CD8⁺ T-cell activity, as decreased NKG7 levels resulted in significantly less tumor-cell killing. This is in agreement with a recent report by Ng and colleagues, which showed that KO of NKG7 can compromise antitumor immunity in mouse models of metastatic tumor growth (35). Our study goes further to assess the role of NKG7 in human CD8⁺ T cells and provides evidence that NKG7 mRNA therapy enhances the cytolytic function of T cells from patients who may not have sufficient NKG7.

Insufficient neoantigen expression (mutation burden) by the tumor or lack of PD-L1 on the tumor cell surface (46–50) are among the explanations proposed for why anti–PD-1 therapy may fail. Our study concludes that a deficiency in an intrinsic component of cytolytic T cells can limit the effectiveness of immunotherapy. In agreement with our finding, NKG7 was among the list of transcripts found to be differentially expressed in the peripheral CD8⁺ T cells of responders versus nonresponders to anti–PD-1 therapy by Fairfax and colleagues (11). However, their bulk RNA-seq analysis generated a list of thousands of differentially expressed transcripts, making it difficult to pinpoint NKG7 as a critical component among so many options in the context of effector T cells. Our scRNA-seq and approach to analysis yielded fewer hits, directing us early-on to NKG7 as a correlate with response. However, we were surprised when we completed the initial functional studies and found that rather than just serving as a biomarker for treatment outcome, NKG7 was necessary to drive optimal CD8⁺ T-cell cytotoxicity. Why and how NKG7 plays such a dominant role in modulating cytolytic activity is still not fully clear, but we showed that modulation of this one protein was able to alter granule number, trafficking, and release. It may be that NKG7 is not only working locally at the level of each granule, but also is working more globally to affect broader cytoplasmic or nuclear signals. Our finding that the level of NKG7 within CD8⁺ T cells was directly linked to the amount of calcium flux at baseline and after NAADP-stimulation supports this model, given the importance of calcium as a second messenger in a variety of T-cell processes.

We observed that NKG7 expression declined at the protein level upon activation of CD8⁺ T cells (Supplementary Fig. S1B and Supplementary Fig. S3A). This suggests a potential self-restraining mechanism inherent to cytotoxic T cells meant to prevent extensive killing activity that could damage normal tissues. This is supported by findings that NKG7 may be upregulated in lymphocytes in the setting of autoimmune conditions, organ transplant rejection, and other pathologies (51–53). However, within the tumor microenvironment, where effector CD8⁺ T cells are confronted with rapidly dividing tumor cells, T cells may need more persistent expression of NKG7 to maintain highly efficient tumor-cell killing capacity and effectively overcome the exponential growth of the tumor.

We show that at the transcription level, ETS1 negatively regulates NKG7 expression, but there are likely a number of other factors that influence NKG7 expression that remain to be determined. In the current study, to bypass the need to control transcription itself, we piloted an mRNA transfection approach. Translationally, this would allow T cells to be modified and then adoptively transferred to improve immune therapy (54, 55). As mRNA does not require a nuclear phase for efficient expression, it is an attractive means for achieving rapid and sufficient expression of NKG7 in nondividing cells (like human primary T cells). The use of mRNA instead of DNA as a therapeutic substance is also advantageous due to the absent risk of insertional mutagenesis. Though the short half-life of mRNA could be a potential therapeutic disadvantage in some cases, in the current context, the transient nature of mRNA may avoid the unwanted effects of permanent overexpression of NKG7 within cytotoxic T cells that could lead to normal tissue damage.

We endeavored to use human patient samples whenever possible to ensure our findings were relevant to the clinical setting. However, the paucity of live CD8⁺ T cells that can be recovered from frozen patient PBMC samples prevented us from being able to specifically isolate patient CD8⁺ T cells prior to transfection with NKG7 mRNA. This limitation was addressed with the use of healthy donors' purified CD8⁺ T cells as a proof of concept, but we plan to include further testing in freshly isolated patient samples in future studies. To test our hypothesis in vivo, we used a preclinical mouse model. However, the OT-1 T cells used in this assay had normal levels of NKG7 expression prior to mRNA transfection. This may not reflect the state of CD8⁺ T cells in clinical nonresponders, which our studies suggest have a reduced amount of NKG7 both at BL and posttherapy. This limitation prevented us from determining if there is a dose-dependent effect of the NKG7 mRNA in vivo. In addition, we found that the B16-OVA tumor model was responsive to anti–PD-1 therapy at BL. Although the adoptive transfer of NKG7 overexpressing OT-1 T cells combined with anti–PD-1 was just as effective, we were unable to detect any additional benefit from the combination treatment. With regard to our proposed therapeutic approach, pretreatment assessment of NKG7 levels would further direct clinicians to a subset of patients that would likely fail immune checkpoint inhibitor therapy alone, but would benefit from our proposed gene therapy, i.e., transduction of NKG7 mRNA in combination with immune checkpoint inhibitors. To that end, our future studies will provide additional detection tools to quantify the expression of NKG7 in peripheral T cells and in tumor tissues to further validate NKG7 as a novel biomarker that can direct rational combination therapy.

In summary, our results indicate that anti–PD-1 immune checkpoint blockade fails to optimize CD8⁺ cytotoxic T-cell response if the intrinsic, cytolytic granule-associated factor NKG7 is not sufficiently expressed. We further show that NKG7 is critical for cytotoxic CD8⁺ T-cell antitumor activity due to its role in cytolytic granule trafficking and release. Finally, we demonstrate that NKG7 mRNA transfection can improve cytotoxic CD8⁺ T-cell killing of tumor cells, representing a new therapeutic target that could increase response rates to cancer immunotherapy, including immune checkpoint inhibitor and adoptive T-cell therapy.

D.D. Monie reports grants from NCI during the conduct of the study. A.S. Mansfield reports other support from Genentech, AbbVie, Bristol Myers Squibb, Janssen, and Shanghai Roche Pharmaceuticals and grants from Mark Foundation, NIH, and DoD outside the submitted work, and is a nonremunerated director of the Mesothelioma Applied Research Foundation. S.N. Markovic reports grants from NIH during the conduct of the study, as well as grants from Bristol Myers Squibb and Sorrento Therapeutics outside the submitted work. S.S. Park reports grants from MacroGenics and NCI, and other support from AstraZeneca outside the submitted work. M.C. Liu reports other support from Eisai, Genentech, Merck, Novartis, Seattle Genetics, Tesaro, AstraZeneca, Pfizer, and Genomic Health outside the submitted work. G. Vasmatzis reports other support from WholeGenome LLC outside the submitted work. No disclosures were reported by the other authors.

T. Wen: Conceptualization, formal analysis, investigation, methodology, writing–review and editing. W. Barham: Conceptualization, formal analysis, investigation, methodology, writing–original draft, writing–review and editing. Y. Li: Conceptualization, software, formal analysis, methodology, writing–review and editing. H. Zhang: Investigation, writing–review and editing. J.K. Gicobi: Investigation, writing–review and editing. J.B. Hirdler: Investigation, writing–review and editing. X. Liu: Investigation, methodology, writing–review and editing. H. Ham: Resources, formal analysis, investigation, writing–review and editing. K.E. Peterson Martinez: Investigation, writing–review and editing. F. Lucien: Investigation, writing–review and editing. R.R. Lavoie: Investigation, writing–review and editing. H. Li: Software, formal analysis, writing–review and editing. C. Correia: Software, formal analysis, writing–review and editing. D.D. Monie: Software, formal analysis, writing–review and editing. Z. An: Investigation, writing–review and editing. S.M. Harrington: Investigation, writing–review and editing. X. Wu: Resources, investigation, writing–review and editing. R. Guo: Resources, writing–review and editing. R.S. Dronca: Resources, writing–review and editing. A.S. Mansfield: Resources, writing–review and editing. Y. Yan: Resources, writing–review and editing. S.N. Markovic: Resources, writing–review and editing. S.S. Park: Resources, writing–review and editing. J. Sun: Resources, writing–review and editing. H. Qin: Resources, writing–review and editing. M.C. Liu: Resources, writing–review and editing. G. Vasmatzis: Software, investigation, writing–review and editing. D.D. Billadeau: Resources, funding acquisition, investigation, writing–review and editing. H. Dong: Conceptualization, supervision, funding acquisition, methodology, writing–original draft, writing–review and editing.

We thank all the patients who generously contributed samples and participated in the study. We also acknowledge the Division of Medical Oncology at the Mayo Clinic for providing excellent patient care. This study was supported in part by the generosity of Mayo Clinic benefactors. We acknowledge the Genome Analysis Core at Mayo Clinic, Rochester, including technologists Vernadette A. Simon and Fariborz Rakhshan-Rohakhtar as well as supervisors Julie S. Lau, Samantha J. McDonough, and Mark Mutawe. We are grateful for the electron microscopy imaging service provided by Bing Huang and Scott I. Gamb at the Mayo Clinic Microscope and Cell Analysis Core. BioRender.com was used to create the graphics throughout the manuscript.

We acknowledge support from the Mayo Clinic Center for Individualized Medicine's IMPRESS program and High-Definition Therapeutics program (to H. Dong), the Mayo Clinic Center for Biomedical Discovery (to H. Dong), the Mayo Clinic Cancer Center's David F. and Margaret T. Grohne Cancer Immunology and Immunotherapy program (H. Dong, D.D. Billadeau), the Richard M. Schulze Family Foundation (to H. Dong), the Innovation Accelerator Awards program of the Office of Translation to Practice (OTP) at the Mayo Clinic Center for Clinical and Translational Science (CCaTS), NIH grant F30 CA250326 (to W. Barham), NIH grant F30 CA250122 (to D.D. Monie), National Institute of General Medical Sciences T32 GM65841 (to W. Barham and D.D. Monie), NIH grant R21 CA251923 (to A.S. Mansfield), NIH grant K12 CA 090628 (to Y. Yan), NIH grant R21 CA197878 (to H. Dong), NIH grant R01 AI095239 (to H. Dong), NIH grant R01 CA200551 (to H. Dong and S.S Park), and NIH grant R01 CA256927 (to H. Dong).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.