Enhancing cancer killing and natural killer cell persistence by targeting NOXA, a predictor of poor patient survival

TL;DR

Deleting a protein called NOXA in natural killer cells improves their survival and cancer-killing ability, potentially boosting cancer immunotherapy.

Contribution

This study identifies NOXA as a key target for improving NK cell persistence and function in cancer immunotherapy.

Findings

NOXA deletion in NK cells enhances their serial cancer-killing efficacy and post-cryopreservation cytotoxicity.

NOXA-KO NK cells show improved metabolic fitness and proliferative capacity.

NOXA deletion rescues NK cell survival in cytokine-rich tumor microenvironments.

Abstract

Natural killer (NK)-cell-based therapeutics have emerged as promising modalities in cancer immunotherapy due to their potent ability to target and kill cancer cells. However, their clinical efficacy is often constrained by the limited in vivo persistence of NK, which hinders sustained therapeutic effects. This study aimed to enhance NK cell survival and functionality by inhibiting apoptosis, thereby boosting the long-term efficacy of NK-cell-mediated therapeutics. Through univariate and multivariate Cox proportional hazards regression analyses, we found that the high expression of Bcl-2-interacting mediator of cell death (BCL2L11 [BIM]) and phorbol-12-myristate-13-acetate-induced protein 1 (PMAIP1 [NOXA]) in NK cells is associated with poorer survival across various cancer types. Based on potential clinical relevance, we employed CRISPR-Cas9 technology with single-guide RNA to knock out NOXA in NK cells. This modification enhanced the serial cancer-killing efficacy in repeated assays and patient-derived organoid models. In contrast, BIM-knockout (KO) did not confer any additional benefits. Additionally, NOXA-KO NK cells exhibited enhanced post-cryopreservation cytotoxicity, superior metabolic fitness, and extended proliferative capacity. These findings highlight the potential of NOXA-KO NK cells to advance the efficacy of NK-cell-based cancer therapies.

Survival analyses by Kang and colleagues linked elevated NOXA expression in NK cells to poor cancer outcomes. We show that TME-enriched IL-6 and transforming growth factor β1 (TGF-β1) strongly induce NOXA, driving NK cell apoptosis and functional exhaustion. NOXA deletion rescued NK cell survival and cytotoxicity in cytokine-rich TMEs, identifying NOXA as a target to enhance NK cell immunotherapy.

Figures

Prognostic analysis of tumor patients based on NK cell density and mRNA expression of *PMAIP1* and *BCL2L11*(A) Kaplan-Meier survival analysis of patients with tumor from The Cancer Genome Atlas (TCGA) dataset (*n* = 8,469) stratified by natural killer (NK) cell score quartiles. Quartiles are displayed in ascending order: Q1 (black), Q2 (blue), Q3 (yellow), and Q4 (red). The log rank *p* value indicates the overall difference in survival across the four quartile groups (Q1–Q4). (B and C) Kaplan-Meier survival curves for patients with pancreatic adenocarcinoma (PAAD) with low NK cell scores (*n

Cancer-derived conditioned media induces apoptosis-related gene expression and dysfunction in NK cellsHuman primary NK cells were incubated with control media (RPMI), AsPC-1-CM, or Capan-2-CM, followed by functional and molecular analyses. (A) NK cells were treated with CM for 72 h and stained with Annexin V/7-aminoactinomycin D (7-AAD) to assess apoptosis. (B) Metabolic activity in NK cells after 24 h of CM exposure was measured by oxygen consumption rate (OCR) and extracellular acidification rate (ECAR). (C) CM-pretreated NK cells were co-cultured with AsPC-1-Luc or Capan-2-Luc cancer cells

Generation of *BIM*- or *NOXA*-knockout NK cells via electroporation-mediated CRISPR/Cas9 delivery(A) Schematic diagram of the genomic loci of Bcl-2 interacting mediator of cell death (*BIM*) and phorbol-12-myristate-13-acetate-induced protein 1 (*PMAIP1* [*NOXA*]) targeted by CRISPR/Cas9 using single-guide RNA (sgRNA) ribonucleoprotein (RNP) complexes. (B) NK cell viability was assessed 2 days post-electroporation using flow cytometry. Viability was calculated as the proportion of live cells within the total cell population. Data are shown as mean ± standard error of mean (SEM). (C) Expressio

Evaluation of 3D organoid killing efficacy by NK cells(A) Representative bright-field image of cultured organoids (stitched image, left, scale bar, 500 μm) and hematoxylin and eosin (H&E)-stained section of fixed organoids (right, scale bar, 100 μm).(B) Visualization of co-cultured organoids and NK cells; organoids and NKs were labeled using CellTracker Deep Red and carboxyfluorescein diacetate succinimidyl ester (CFSE), respectively (scale bars, 100 μm). (C–E) NK-cell-mediated killing in tumor organoids. Engineered NK cells (control-, *BIM*-KO, *NOXA*-KO, *BIM-NOXA*-DKO NK) were co-cultured w

Prolonged proliferation and enhanced metabolic activity in NOXA-KO NK cells(A) The effect of *NOXA* KO on proliferation marker Ki67 expression in NK cells was assessed via flow cytometry. (B) Proliferation assay of CFSE-labeled NK cells was analyzed by flow cytometry, showing enhanced cell proliferation in *NOXA*-KO NK cells. (C) Expression of the NK cell activation marker CD69 and the degranulation marker CD107a was measured after co-culture with AsPC-1 or Capan-2 cells for 6 h. (D) Expression of the NK-cell-activating receptors was analyzed using flow cytometry. (E) Cytokine secretion by NK

Full text

Introduction

Natural killer (NK) cells are promising candidates for cancer and infection therapies due to their human leukocyte antigen (HLA)-independent target recognition, enabling “off-the-shelf” manufacturing with minimal toxicity.1^,^2 This feature supports scalable and cost-effective allogeneic applications.3 Recent advances in chimeric antigen receptor (CAR)-NK cells have demonstrated clinical efficacy in hematologic cancers. For instance, anti-CD19 CAR-NK cells expressing interleukin (IL)-15, derived from allogeneic umbilical cord blood, achieved a one-year overall survival rate of 68% and progression-free survival of 32% in CD19-positive malignancies.4^,^5 CAR-NK therapies also exhibited a favorable safety profile, with minimal cytokine release syndrome and neurotoxicity compared to CAR-T therapies, supporting their development as safer allogeneic options.

Despite potent cytotoxicity, NK-cell-based therapies often yield transient responses due to limited in vivo persistence.6 Unlike T cells, NK cells typically do not generate durable antigen-specific memory populations, and this limited capacity for memory formation contributes to their reduced in vivo persistence. Enhancing persistence is thus critical. Memory-like NK cells can be induced by haptens, viral infections, or cytokines.6^,^7^,^8^,^9 In humans, infections with human cytomegalovirus expands NKG2C^+^CD57^+^ NK cells via interleukin-12 (IL-12)/signal transducer and activator of transcription 4 signaling, enhancing proliferation.10^,^11^,^12 Similarly, cytokine-induced memory-like (CIML) NK cells, generated through stimulation with IL-12, IL-15, and, IL-18, exhibit enhanced proliferation, cytotoxicity, and interferon-gamma (IFN-γ) secretion,13 along with epigenetic and metabolic adaptations promoting extended persistence.14^,^15 These attributes have led to clinical trials of CIML NK cells for hematologic and solid tumors.13 Moreover, engineering NK cells to overexpress IL-15 or IL-21 enhances their expansion and long-term persistence.5^,^16^,^17 To further improve NK cell persistence, we focused on apoptosis-regulating genes.

Apoptosis in NK cells is regulated by the B cell lymphoma 2 (BCL-2) protein family, which governs cell survival and death through a balance of pro- and anti-apoptotic pathways, and whose members are classified into three subfamilies based on their BCL-2 homology (BH) domains: anti-apoptotic proteins (containing four BH domains), multi-domain pro-apoptotic proteins (three BH domains), and BH3-only proteins.18^,^19 The BH3-only subfamily, including BAD, Bcl-2-interacting mediator of cell death (BIM), BID, phorbol-12-myristate-13-acetate-induced protein 1 (PMAIP1 [NOXA]), PUMA, BIK/BLK, BMF, HRK/DP5, and BECLIN-1, comprises essential initiators of apoptosis, acting either as direct activators or as de-repressors of anti-apoptotic functions.20 For instance, BIM and BID activate mitochondrial outer membrane permeabilization by binding to pro-apoptotic effecters BAK and BAX, leading to caspase activation and apoptosis.21^,^22 Meanwhile, sensitizer BH3-only proteins like NOXA and BAD promote apoptosis by neutralizing anti-apoptotic family members.19^,^23

Given their crucial role in apoptotic regulation, targeting BH3-only proteins represents a potential strategy to enhance NK cell survival and cytotoxic function. Notably, IL-15 has been shown to suppress apoptosis in NK cells by modulating the expression of regulators such as Bim, Noxa, and Mcl1 in murine models.21 To overcome the limitation of short-lived NK cell activity, the present study aimed to engineer apoptosis-related genes in human NK cells to improve their persistence and cytotoxic potential, thereby advancing the therapeutic utility of NK-cell-based cancer treatments.

Results

High NK cell density in PDAC tissues is associated with increased immune cell infiltration

To investigate immune cell populations in pancreatic ductal adenocarcinoma (PDAC), we performed multiplexed immunofluorescence (mIF) analysis on tissue sections from 17 patients with PDAC. This technique enabled the simultaneous detection of multiple biomarkers, allowing detailed characterization of tumor cells and immune cell populations, including NK cells, T cells, macrophages, and dendritic cells (DCs) (Figure S1). Notably, NK cell density in PDAC tissues positively correlated with the numbers of CD3-positive T cells and CD11c-positive DCs (Figure S1). Using inForm software for mIF image analysis, we stratified the tissues into two groups: NK^high^ and NK^low^ (Figure S1B). Tissues with high NK cell density exhibited a “hot” immune microenvironment, characterized by increased immune cell infiltration and a lower proportion of tumor cells, in contrast to NK^low^ tissues. These findings suggest that increased NK cell density contributes to a more favorable immune landscape and may be associated with improved therapeutic responsiveness. These relationships were further supported by a correlation matrix analysis (Figure S2), which revealed selective positive associations between NK cells and other immune subsets, including T cells and DCs, reinforcing the link between NK cell abundance and an immunologically active tumor microenvironment (TME).

NK-cell-associated transcriptional programs and prognostic implications across cancers

To evaluate the role of apoptosis-related genes in NK cells across different cancers, we analyzed the relationship between survival probability and NK cell score using The Cancer Genome Atlas (TCGA) dataset (n = 8,469) (Figures 1A, S3, and S4). Based on prior biological evidence implicating NOXA and BIM in the regulation of NK cell survival and functional fitness, we focused our analysis on these apoptosis-related genes and examined their prognostic relevance in relation to NK cell infiltration. As shown in Figure S3, the NK signature score was validated using single-nucleus RNA-sequencing data, confirming that it accurately reflects NK-cell-specific transcriptional signals. Patients in the lowest NK cell density quartile (Q1) exhibited significantly poorer overall survival (OS) compared to those in the highest quartile (Q4) (Figure 1A). Intriguingly, high mRNA expression of PMAIP1 (NOXA) and BCL2L11 (BIM) was associated with worse survival outcomes, particularly in NK^low^ groups, compared to NK^high^ groups (Figures 1B, 1C, and S4). To further assess prognostic implications, we conducted univariate and multivariate Cox proportional hazards regression analyses for PMAIP1 (NOXA) and BCL2L11 (BIM) expression across pan-cancer types (Figure S5; Table 1). In univariate analysis, high expression levels of PMAIP1 (NOXA) and BCL2L11 (BIM) were significantly associated with poor OS in both NK^high^ and NK^low^ groups across multiple tumor types (Figures 1D–1G and S5). Multivariate analysis confirmed that the elevated expression of these genes was consistently linked to worse OS in both pan-cancer and tumor-specific contexts (Table 1). These findings suggest that the higher expression of PMAIP1 (NOXA) in NK cells is associated with lower survival probabilities across multiple cancer types, exhibiting a more consistent and robust prognostic pattern. In contrast, the prognostic impact of BCL2L11 (BIM) appears more variable and context-dependent across tumor types. To further validate the robustness of PMAIP1 (NOXA) as a prognostic marker, we performed a univariate Cox regression analysis across all 25 TCGA cancer types without NK content stratification (Figure S5). This unified analysis confirmed that PMAIP1 (NOXA) expression is significantly associated with poor overall survival in the pan-cancer cohort (hazard ratio [HR] = 1.06, p < 0.001), with particularly strong effects observed in pancreatic adenocarcinoma (PAAD) (HR = 1.43, p = 0.002), kidney renal papillary cell carcinoma (KIRP) (HR = 1.66, p < 0.001), and kidney renal clear cell carcinoma (KIRC) (HR = 1.39, p < 0.001) (Figure S5E). In contrast*, BCL2L11 (BIM)* showed no significant pan-cancer association (HR = 0.98, p = 0.438) and exhibited highly variable, tumor-specific effects (Figure S5D). This highlights NOXA as potential molecular targets for NK cell engineering strategies aimed at enhancing anti-tumor immunity and improving clinical outcomes.Figure 1. Prognostic analysis of tumor patients based on NK cell density and mRNA expression of PMAIP1 and BCL2L11(A) Kaplan-Meier survival analysis of patients with tumor from The Cancer Genome Atlas (TCGA) dataset (n = 8,469) stratified by natural killer (NK) cell score quartiles. Quartiles are displayed in ascending order: Q1 (black), Q2 (blue), Q3 (yellow), and Q4 (red). The log rank p value indicates the overall difference in survival across the four quartile groups (Q1–Q4). (B and C) Kaplan-Meier survival curves for patients with pancreatic adenocarcinoma (PAAD) with low NK cell scores (n = 89), stratified by the median levels of PMAIP1 (NOXA) (B) into high (PMAIP1 (NOXA) ^high^) (n = 48) and low (PMAIP1 (NOXA) ^low^) (n = 41) groups and BCL2L11 (BIM) (C) into high (BCL2L11 (BIM) ^high^) (n = 31) and low (BCL2L11 (BIM) ^low^) (n = 58) groups. (D–G) Univariate Cox regression analysis evaluating prognostic outcomes based on NK cell score and the expression of PMAIP1 (D, E) and BCL2L11 (F, G) across various tumor types. Tumor types with hazard ratios (HRs) > 1 and p < 0.05 are highlighted in red. The analysis includes HRs, 95% confidence intervals (CIs), and statistical significance values.Table 1. Multivariate Cox regression analysis in pan-cancer and various tumor typesTumor typesVariablesHR (95% CI)p valueVariablesHR (95% CI)p valuePancancerPMAIP11.08 (1.05–1.1)<0.001PMAIP1 (scaled)1.15 (1.1–1.2)<0.001BCL2L110.94 (0.89–1)0.0417BCL2L11 (scaled)0.96 (0.92–1)0.0417NK score0.4 (0.24–0.66)<0.001NK score (scaled)0.93 (0.89–0.97)<0.001PAADPMAIP11.46 (1.16–1.84)****0.0014PMAIP1 (scaled)1.46 (1.16–1.85)**0.0014BCL2L110.95 (0.57–1.58)0.8412BCL2L11 (scaled)0.97 (0.76–1.25)0.8412NK score6.48 (0.05–794.54)0.4465NK score (scaled)1.09 (0.87–1.38)0.4465GBMPMAIP11.15 (0.99–1.35)0.0745PMAIP1 (scaled)1.22 (0.98–1.51)0.0745BCL2L111.19 (0.81–1.73)0.3712BCL2L11 (scaled)1.09 (0.9–1.34)0.3712NK score0.4 (0–116.63)0.7516NK score (scaled)0.96 (0.77–1.2)0.7516OVPMAIP10.96 (0.84–1.11)0.6004PMAIP1 (scaled)0.96 (0.83–1.12)0.6004BCL2L111.31 (1–1.72)0.0524BCL2L11 (scaled)1.17 (1–1.36)0.0524NK score0.43 (0.04–4.19)0.4656NK score (scaled)0.94 (0.8–1.11)0.4656LUADPMAIP11.13 (1.01–1.26)****0.0259PMAIP1 (scaled)1.19 (1.02–1.38)**0.0259BCL2L111.05 (0.81–1.37)0.7099BCL2L11 (scaled)1.03 (0.89–1.19)0.7099NK score0.15 (0.01–1.57)0.113NK score (scaled)0.88 (0.75–1.03)0.113LUSCPMAIP10.98 (0.84–1.13)0.7554PMAIP1 (scaled0.98 (0.84–1.13)0.7554BCL2L110.87 (0.67–1.13)0.2989BCL2L11 (scaled)0.93 (0.8–1.07)0.2989NK score0.63 (0.09–4.57)0.6494NK score (scaled)0.97 (0.84–1.11)0.6494PRADPMAIP11.12 (0.64–1.94)0.6922PMAIP1 (scaled)1.13 (0.62–2.05)0.6922BCL2L110.9 (0.28–2.9)0.8577BCL2L11 (scaled)0.95 (0.52–1.72)0.8577NK score0 (0–53831.61)0.4278NK score (scaled)0.75 (0.36–1.54)0.4278BLCAPMAIP11.08 (0.96–1.2)0.1952PMAIP1 (scaled)1.11 (0.95–1.31)0.1952BCL2L110.96 (0.77–1.2)0.7247BCL2L11 (scaled)0.97 (0.83–1.14)0.7247NK score0.07 (0.01–0.46)0.0057NK score (scaled)0.8 (0.68–0.94)0.0057TGCTPMAIP11.78 (0.5–6.31)0.3715PMAIP1 (scaled)1.93 (0.46–8.1)0.3715BCL2L113.35 (0.46–24.62)0.2339BCL2L11 (scaled)1.74 (0.7–4.3)0.2339NK score29.9 (0–8562001.46)0.5961NK score (scaled)1.37 (0.43–4.42)0.5961ESCAPMAIP11.19 (0.97–1.46)0.0924PMAIP1 (scaled)1.21 (0.97–1.5)0.0924BCL2L111.02 (0.69–1.49)0.938BCL2L11 (scaled)1.01 (0.8–1.28)0.938NK score4.04 (0.13–122.84)0.4234NK score (scaled)1.1 (0.88–1.37)0.4234KIRPPMAIP1*****1.52 (1.2–1.91)<0.001PMAIP1* (scaled)1.85 (1.31–2.6)<0.001BCL2L111.94 (1.2–3.13)****0.0069BCL2L11 (scaled)****1.6 (1.14–2.24)****0.0069NK score2.82 (0.03–290.75)0.6617NK score (scaled)1.06 (0.8–1.41)0.6617LIHCPMAIP11.2 (1.09–1.34)****<0.001PMAIP1 (scaled)1.39 (1.15–1.66)<0.001BCL2L110.84 (0.69–1.02)0.0866BCL2L11 (scaled)0.86 (0.72–1.02)0.0866NK score0 (0–0.11)<0.001NK score (scaled)0.7 (0.57–0.87)<0.001CESCPMAIP10.94 (0.75–1.16)0.5524PMAIP1 (scaled)0.93 (0.73–1.18)0.5524BCL2L111.01 (0.72–1.42)0.9563BCL2L11 (scaled)1.01 (0.79–1.28)0.9563NK score0.03 (0–0.45)0.0117NK score (scaled)0.72 (0.56–0.93)0.0117SARCPMAIP11.03 (0.94–1.12)0.5753PMAIP1 (scaled)1.06 (0.87–1.28)0.5753BCL2L111.37 (1.05–1.79)****0.0217BCL2L11 (scaled)****1.3 (1.04–1.62)****0.0217NK score0.06 (0.01–0.78)0.0313NK score (scaled)0.78 (0.62–0.98)0.0313BRCAPMAIP10.84 (0.75–0.94)0.0017PMAIP1 (scaled)0.78 (0.67–0.91)0.0017BCL2L111.08 (0.84–1.39)0.5262BCL2L11 (scaled)1.05 (0.9–1.24)0.5262NK score0.03 (0–0.28)0.0026NK score (scaled)0.77 (0.65–0.91)0.0026THYMPMAIP10.54 (0.26–1.1)0.0877PMAIP1 (scaled)0.46 (0.19–1.12)0.0877BCL2L112.6 (0.86–7.87)0.0898BCL2L11 (scaled)1.99 (0.9–4.43)0.0898NK score18385733.8 (591.16–571815137282.54)0.0015NK score (scaled)2.98 (1.52–5.86)0.0015COADPMAIP11.06 (0.73–1.53)0.7694PMAIP1 (scaled)1.05 (0.77–1.42)0.7694BCL2L111.31 (0.85–2)0.2186BCL2L11 (scaled)1.17 (0.91–1.51)0.2186NK score0.07 (0–9.03)0.283NK score (scaled)0.85 (0.63–1.14)0.283STADPMAIP10.99 (0.86–1.14)0.9118PMAIP1 (scaled)0.99 (0.85–1.16)0.9118BCL2L111.2 (0.95–1.53)0.122BCL2L11 (scaled)1.13 (0.97–1.32)0.122NK score0.31 (0.03–2.73)0.2889NK score (scaled)0.92 (0.78–1.08)0.2889KIRCPMAIP11.39 (1.24–1.56)****<0.001PMAIP1 (scaled)1.51 (1.31–1.74)<0.001BCL2L110.79 (0.6–1.04)0.09BCL2L11 (scaled)0.88 (0.75–1.02)0.09NK score2.57 (0.3–22.15)0.3891NK score (scaled)1.06 (0.93–1.22)0.3891THCAPMAIP11.21 (0.87–1.68)0.2661PMAIP1 (scaled)1.34 (0.8–2.23)0.2661BCL2L112.98 (1.11–8.01)****0.0303BCL2L11 (scaled)****1.83 (1.06–3.14)****0.0303NK score0 (0–0.12)0.0172NK score (scaled)0.49 (0.27–0.88)0.0172HNSCPMAIP11.11 (0.96–1.28)0.1578PMAIP1 (scaled)1.11 (0.96–1.28)0.1578BCL2L110.82 (0.66–1.01)0.0596BCL2L11 (scaled)0.87 (0.76–1.01)0.0596NK score0.11 (0.02–0.73)0.022NK score (scaled)0.84 (0.72–0.98)0.022UCECPMAIP11.01 (0.84–1.23)0.8954PMAIP1 (scaled)1.02 (0.79–1.31)0.8954BCL2L111.72 (1.13–2.61)****0.0113BCL2L11 (scaled)****1.42 (1.08–1.86)****0.0113NK score0.01 (0–0.51)0.0205NK score (scaled)0.71 (0.53–0.95)0.0205READPMAIP10.94 (0.47–1.89)0.8671PMAIP1 (scaled)0.95 (0.55–1.66)0.8671BCL2L110.89 (0.29–2.72)0.8363BCL2L11 (scaled)0.94 (0.54–1.65)0.8363NK score0.2 (0–122973.19)0.8111NK score (scaled)0.93 (0.51–1.69)0.8111SKCMPMAIP11.2 (0.9–1.62)0.2193PMAIP1 (scaled)1.31 (0.85–2)0.2193BCL2L110.84 (0.57–1.24)0.3791BCL2L11 (scaled)0.84 (0.56–1.25)0.3791NK score0.02 (0–2.33)0.1093NK score (scaled)0.72 (0.49–1.07)0.1093LGGPMAIP11.1 (0.92–1.32)0.3059PMAIP1 (scaled)1.09 (0.92–1.29)0.3059BCL2L110.78 (0.63–0.97)0.0272BCL2L11 (scaled)0.84 (0.71–0.98)0.0272NK score18.82 (0.25–1394.41)0.1815NK score (scaled)1.13 (0.94–1.35)0.1815The hazard ratio (HR), 95% confidence intervals (CIs), and p values were calculated using multivariate Cox regression analysis. Factors included in the analysis were NK cell scores, BCL2L11, and PMAIP1 mRNA levels. Tumor types with HR > 1 and p value <0.05 are indicated in bold, indicating a significant association with poorer prognosis. p values were not adjusted for multiple comparisons across tumor types.

Cancer-derived conditioned media impairs NK-mediated cancer killing and induces pro-apoptotic proteins NOXA and BIM

To investigate the impact of tumor-derived factors on NK-cell-mediated cytotoxicity, we assessed apoptosis and metabolic dysfunction in NK cells exposed to cancer-derived conditioned media (CM). Tumor-derived growth factors, cytokines, and other soluble mediators are known to contribute to NK cell dysfunction within the TME. To mimic this environment, human primary NK cells were incubated with CM from AsPC-1 and Capan-2 pancreatic tumor cells, while Roswell Park Memorial Institute (RPMI) served as a control media. Apoptotic populations were quantified using Annexin V/7-AAD staining (Figures 2A and S6). As expected, exposure to both AsPC-1- and Capan-2-derived CM significantly increased early and late apoptosis in NK cells, indicating that apoptosis is a key mechanism underlying CM-induced NK cell dysfunction. We next assessed the metabolic activity of CM-treated NK cells as a measure of NK cell activation.24 Notably, CM exposure reduced both glycolysis and oxidative phosphorylation (Figure 2B), indicating a suppressed metabolic state. This suggests that tumor-derived CM impairs NK cell metabolic fitness.Figure 2. Cancer-derived conditioned media induces apoptosis-related gene expression and dysfunction in NK cellsHuman primary NK cells were incubated with control media (RPMI), AsPC-1-CM, or Capan-2-CM, followed by functional and molecular analyses. (A) NK cells were treated with CM for 72 h and stained with Annexin V/7-aminoactinomycin D (7-AAD) to assess apoptosis. (B) Metabolic activity in NK cells after 24 h of CM exposure was measured by oxygen consumption rate (OCR) and extracellular acidification rate (ECAR). (C) CM-pretreated NK cells were co-cultured with AsPC-1-Luc or Capan-2-Luc cancer cells for 24 h, and cancer-cell-killing activity was measured. (D) Expression of apoptosis-related proteins BIM and NOXA in CM-treated NK cells was analyzed using western blotting. Representative blots show NOXA and BIM levels following 24 h exposure to CM derived from normal pancreatic epithelial cells (HPDE) or pancreatic cancer cell lines (AsPC-1 and Capan-2). BIM and NOXA protein levels were quantified from three independent experiments and normalized to β-actin. (E) Western blots showing the expression of NOXA and BIM in NK cells following 20 h exposure to IL-6, IL-10, or TGF-β1. All data are presented as mean ± standard error of the mean (SEM). p values were calculated using one-way ANOVA with multiple comparison tests (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001).

To assess the functional impact of CM on NK cell-mediated cytotoxicity, CM-pretreated NK cells were co-cultured with luciferase-expressing pancreatic cancer cells (AsPC-1-Luc and Capan-2-Luc). Cytotoxicity was quantified using a luminescence-based assay at a 1:1 effector-to-target (E:T) ratio over 24 h. CM preconditioning significantly reduced NK-cell-mediated cancer killing against both target cell types (Figure 2C). Furthermore, western blot analysis revealed significant upregulation of the pro-apoptotic proteins BIM and NOXA in CM-treated NK cells, implicating apoptosis-inducing pathways in the observed functional impairment (Figures 2D and S7). Consistent with these findings, CM generated from the non-malignant pancreatic epithelial cell line (HPDE) induced only a minor increase in BIM and NOXA (Figure 2D), indicating that cancer-cell-derived factors are substantially stronger drivers of this apoptotic response. In line with these observations, NK cells exposed to TME-enriched cytokines (IL-6, TGF-β, and IL-10) exhibited differential induction of apoptotic mediators (Figure 2E). Notably, both IL-6 and TGF-β induced NOXA expression in NK cells, indicating that multiple TME-associated cytokines contribute to pro-apoptotic signaling.

Generation of BIM- or NOXA-knockout NK cells via Cas9/sgRNA ribonucleoprotein complexes

To determine the functional roles of BIM and NOXA in NK cells, we employed clustered regularly interspaced short palindromic repeats (CRISPR)/Cas9-mediated gene editing to generate BIM- and NOXA-knockout (KO) NK cells. CRISPR/Cas9 ribonucleoprotein (RNP) complexes were introduced into NK cells by electroporation, with single-guide (sg) RNAs targeting exon 3 of BIM and exon 2 of NOXA (Figure 3A). Post-electroporation, NK cell viability remained high, with over 90% of cells viable, indicating minimal cytotoxic effects of the procedure (Figures 3B and S8). Western blot analysis confirmed successful knockdown of BIM and NOXA in single-KO and double-KO (DKO) NK cells, with protein levels effectively depleted in the sgRNA amounts tested (Figures 3C and S9). Editing efficiency was further validated using T7 endonuclease I (T7E1) assay and Inference of CRISPR Edits (ICE) analysis, both of which revealed high indel frequencies (Figures 3D and 3E). ICE analysis was performed using 2 μg of Cas9 and 100 pmol of sgRNA per sample. Together, these results confirm the successful generation of BIM- and NOXA-deficient NK cells using CRISPR/Cas9 technology, providing a robust platform for dissecting the roles of apoptosis-related genes in NK cell survival and anti-tumor function.Figure 3. Generation of BIM- or NOXA-knockout NK cells via electroporation-mediated CRISPR/Cas9 delivery(A) Schematic diagram of the genomic loci of Bcl-2 interacting mediator of cell death (BIM) and phorbol-12-myristate-13-acetate-induced protein 1 (PMAIP1 [NOXA]) targeted by CRISPR/Cas9 using single-guide RNA (sgRNA) ribonucleoprotein (RNP) complexes. (B) NK cell viability was assessed 2 days post-electroporation using flow cytometry. Viability was calculated as the proportion of live cells within the total cell population. Data are shown as mean ± standard error of mean (SEM). (C) Expression levels of BIM and NOXA were evaluated by immunoblotting 5 days post-electroporation. The blots shown are one of three independent replicates (all replicates are shown in Figure S5). (D) Knockout (KO) efficacy for BIM and NOXA was evaluated in NK cells 48 h post-electroporation using the T7 endonuclease I (T7E1) assay. (E) Indel frequencies in NK cells edited with Cas9 RNPs were analyzed using Inference of CRISPR Edits (ICE).

NOXA-KO NK cells exhibit enhanced cancer killing in both 2D and 3D organoid models

We evaluated the cytotoxic activity of NK cells following gene editing with Cas9/sgRNA RNP complexes. Engineered NK cells were co-cultured with luciferase-expressing pancreatic cancer cells (AsPC-1-Luc and Capan-2-Luc) at an E:T ratio of 1:1 for 24 h, and cytotoxicity was quantified using a luminescence-based assay (Figure 4A). Intriguingly, NOXA-KO NK cells demonstrated a significant increase in cancer cell killing, whereas BIM-KO NK cells and DKO NK cells did not show any improvement in cytotoxic activity. Importantly, cryopreserved and thawed NOXA-KO NK cells retained their enhanced killing capacity, highlighting their potential for “off-the-shelf” therapeutic applications (Figure 4B). Furthermore, NOXA-KO NK cells pretreated with tumor-derived CM for 24 h exhibited enhanced cytotoxicity against both AsPC-1 and Capan-2 cancer cells, suggesting that inhibition of apoptosis pathways may boost NK cell function even under immunosuppressive conditions (Figure 4C). We next assessed whether NOXA contributes to NK cell dysfunction under TME-enriched cytokine conditions (Figure 4D). Strikingly*, NOXA*-KO NK cells maintained high cytotoxic activity even in the presence of inhibitory cytokines such as IL-6 and TGF-β1. These findings indicate that NOXA deletion confers functional resilience to NK cells within immunosuppressive, cytokine-rich TMEs.Figure 4In vitro evaluation of NK-mediated anti-tumor activities under standard and suppressive conditions(A) Single KO or DKO NK cells were co-cultured with AsPC-1-Luc and Capan-2-Luc cells. After 24 h of incubation, luminescence-based assays were performed to assess cancer-killing activity. (B) Cryopreserved NK cells were thawed and analyzed for their cancer killing activity using luminescence-based assays. (C) Cytotoxicity of NK cells pretreated with conditioned media (CM) was evaluated after 24 h of co-culture with cancer cells. (D) Effect of cytokines on NK cell cytotoxicity. NK cells were co-cultured with luciferase-expressing tumor cells at 0.5:1 E:T ratio for 24 h in the presence of IL-6 or TGF-β1 (20 ng/mL). Cytokines were added directly to the NK-tumor co-cultures and maintained throughout the assay. (E) Schematic illustration of a serial cancer killing assay. NK cells were co-cultured with luciferase-expressing cancer cells, and surviving NK cells were sequentially transferred into freshly cultured cancer cells at 24-h intervals. (F and G) Single-KO and DKO NK cells were sequentially challenged three times with freshly cultured tumor cells. Cancer-killing activity was evaluated using a luminescence-based assay with AsPC-1-Luc cells (F) and a cell viability assay with SNU-3947-TO-2D cells (G), both in the presence of IL-2. (H) Serial cancer killing assays were also conducted in the absence of IL-2 to assess NK cell functionality under cytokine-limited conditions. Error bars represent the mean ± standard SEM of three independent experiments. Statistical significance was calculated as follows: ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001.

To further assess the durability of cytotoxic activity, we performed a sequential cancer-killing assay (Figures 4E–4H). NK cells were serially transferred to freshly seeded tumor cells every 24 h for three consecutive rounds. Notably, only NOXA-KO NK cells exhibited significantly enhanced sequential killing activity against both AsPC-1-Luc and SNU-3947-TO_2D cells. Moreover, NOXA-KO NK cells maintained their enhanced cytotoxicity even in the absence of cytokine supplementation (Figure 4H), underscoring their intrinsic cytotoxic potential.

Next, we tested the cytotoxic effects of engineered NK cells using a patient-derived pancreatic cancer organoid model. Organoid formation was confirmed through bright-field microscopy and hematoxylin and eosin (H&E) staining of fixed sections (Figure 5A). To visualize the co-culture setup, organoids were labeled with CellTracker Deep Red (red), and NK cells were labeled with carboxyfluorescein diacetate succinimidyl ester (CFSE, green), confirming successful co-localization within the 3D culture environment (Figure 5B). NK-cell-mediated organoid killing was monitored over a 13-day period, with imaging performed every 3 days (Figure 5C). Consistent with 2D culture results, NOXA-KO NK cells caused a marked reduction in organoid size. Live organoids were visualized using the cyto3D detection kit (green), and organoid areas were quantitatively analyzed using SPIP image analysis software, allowing clear segmentation between organoids and NK cells (Figures 5D, 5E, and S10). Quantitative analysis showed that NOXA-KO NK cells induced a significantly greater reduction in organoid area compared to other NK groups. Additionally, apoptotic organoids were visualized using caspase-3/7 staining (red), which marks regions undergoing apoptosis. A substantial increase in caspase-3/7-positive apoptotic areas was observed in organoids treated with NOXA-KO NK cells (Figures 5F and 5G). To further validate these findings using organoid-specific measurements independent of NK cell contributions, we employed patient-derived pancreatic tumor organoids engineered to stably express both firefly luciferase and tdTomato (Figure S11). This dual-reporter system enables precise quantification of organoid viability through luciferase activity and direct visualization of live organoid structures via tdTomato fluorescence—neither of which are expressed by NK cells. Consistent with the cyto3D live/caspase-3/7 staining results, luciferase assays confirmed that NOXA-KO NK cells induced a significantly greater loss of organoid viability compared with control NK cells. Furthermore, by analyzing caspase-3/7 signal specifically within tdTomato-positive organoid regions to exclude potential NK-cell-derived fluorescence, NOXA-KO NK cells exhibited significantly increased organoid-restricted apoptosis (Figure S11). Taken together, these complementary measurements demonstrate that the enhanced cytotoxic activity observed in NOXA-KO NK cells reflects true organoid killing, rather than differences in NK cell survival or artifacts arising from mixed cell populations.Figure 5. Evaluation of 3D organoid killing efficacy by NK cells(A) Representative bright-field image of cultured organoids (stitched image, left, scale bar, 500 μm) and hematoxylin and eosin (H&E)-stained section of fixed organoids (right, scale bar, 100 μm).(B) Visualization of co-cultured organoids and NK cells; organoids and NKs were labeled using CellTracker Deep Red and carboxyfluorescein diacetate succinimidyl ester (CFSE), respectively (scale bars, 100 μm). (C–E) NK-cell-mediated killing in tumor organoids. Engineered NK cells (control-, BIM-KO, NOXA-KO, BIM-NOXA-DKO NK) were co-cultured with organoids for 13 days (n = 3 per group). (C) Bright field images were captured every 3 days. On day 13, live organoids were confirmed using the cyto3D kit (scale bars, 200 μm). Representative images are shown as follows: organoids only, control-NK, BIM-KO-NK, NOXA-KO-NK, and BIM-NOXA-DKO-NK. (D) Organoid area was quantified using image analysis software (SPIP, Image Metrology, v.6.7.3; scale bars, 200 μm). Changes in organoid area over time are presented in the graph (two-way analysis of variance [ANOVA]; ∗p < 0.05, ∗∗∗∗p < 0.0001). (E) The measured organoid area on day 13 is shown in the graph (one-way ANOVA; ∗p < 0 .05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001). (F and G) Co-culture with NOXA-KO NK cells resulted in a significant increase in apoptotic areas compared to other groups. (F) Representative images show bright field images (scale bars, 200 μm), live cells (green), and dead cells (red). (G) Quantification of live and apoptotic areas was performed using the FIJI software (Mann-Whitney test; ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001).

Depletion of NOXA enhances NK cell proliferation and drives metabolic reprogramming

To investigate the role of the pro-apoptotic gene NOXA in NK-cell-mediated cytotoxicity, we performed proliferation assays using NOXA-KO NK cells (Figures 6A and 6B). Remarkably, NOXA deficiency led to prolonged NK cell proliferation, as evidenced by elevated expression of the proliferation marker Ki67 and increased cell division, assessed via a CFSE-based assay. Consistently, NOXA-KO NK cells exhibited heightened activation when co-cultured with pancreatic cancer cell lines AsPC-1 and Capan-2, demonstrating enhanced cytotoxic functionality (Figure 6C). Given that NOXA antagonizes the pro-survival factor Mcl-1, its induction by TME-associated cytokines such as IL-6 and TGF-β1, likely amplifies apoptotic pressure by destabilizing mitochondrial integrity (Figures 2E, 4D, and S12). In this context, the preserved cytotoxicity and improved persistence of NOXA-KO NK cells are consistent with partial stabilization of Mcl-1,21 supporting a model in which dysregulation of the NOXA-Mcl-1 axis constitutes a key bottleneck for NK cell survival within suppressive TME. This mechanistic link also aligns with our clinical findings showing that higher NOXA expression correlates with poorer patient survival, collectively highlighting the NOXA-Mcl-1 axis as a functionally relevant and therapeutically actionable vulnerability in NK cell biology.Figure 6. Prolonged proliferation and enhanced metabolic activity in NOXA-KO NK cells(A) The effect of NOXA KO on proliferation marker Ki67 expression in NK cells was assessed via flow cytometry. (B) Proliferation assay of CFSE-labeled NK cells was analyzed by flow cytometry, showing enhanced cell proliferation in NOXA-KO NK cells. (C) Expression of the NK cell activation marker CD69 and the degranulation marker CD107a was measured after co-culture with AsPC-1 or Capan-2 cells for 6 h. (D) Expression of the NK-cell-activating receptors was analyzed using flow cytometry. (E) Cytokine secretion by NK cells was quantified using ELISA. NK cells were co-cultured with AsPC-1-luc or Capan-2-luc cells at 1:1 E:T ratio for 24 h, and supernatants were analyzed. (F and G) Mitochondrial metabolic activity in NK cells was evaluated using the Seahorse XF Cell Mito Stress Test, showing enhanced oxidative metabolism in NOXA-KO NK cells. Metabolic activities in NOXA-KO NK cells was assessed under normal conditions (F) and after 24 h of treatment with tumor-derived conditioned media (CM) (G). Statistical significance was determined using two-way analysis of variance (ANOVA) followed by multiple comparison test. ∗p < 0.05, ∗∗∗p < 0.001. (H and I) Metabolite analysis was conducted to assess pathways related to glycolysis, the tricarboxylic acid (TCA) cycle (H), and the pentose phosphate pathway (I). Bar graphs summarizing levels of each metabolite represent three independent experiments and are displayed as mean ± standard SEM (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001).

In addition to enhanced proliferation, NOXA-KO NK cells increased the expression of key activation and cytotoxicity-associated surface receptors. Specifically, NOXA-KO NK cells showed significantly higher levels of CD16, DNAM-1, NKG2D, and NKp30 compared with control NK cells (Figure 6D). These activation markers are closely linked to NK cell effector functions, indicating that NOXA deletion promotes a more activated and functionally competent NK cell phenotype. Moreover, cytokine secretion assays revealed that NOXA-KO NK cells produced markedly greater amounts of IFN-γ and TNF-α when co-cultured with AsPC-1-Luc or Capan-2-Luc pancreatic cancer cells (Figure 6E). Together, these findings indicate that a loss of NOXA extends NK cell longevity while concurrently elevating their activation profile and effector cytokine output in response to tumor targets. NOXA is a known regulator of p53-induced apoptosis, primarily functioning through mitochondrial disruption and modulation of metabolic pathways25 Given the crucial role of metabolic reprogramming in sustaining NK cell persistence and effector functions—such as long-term survival, cytokine production, and tumor cell killing—we evaluated metabolic changes in NOXA-KO NK cells.26 We performed comprehensive metabolic profiling by measuring the oxygen consumption rate (OCR), extracellular acidification rate (ECAR), and proton efflux rate (PER) (Figure 6F). Strikingly, NOXA-KO NK cells demonstrated significantly increased OCR, ECAR, and PER, indicating acute enhancement of both mitochondrial oxidative phosphorylation (OXPHOS) and glycolysis. Furthermore, NOXA depletion enhanced metabolic activity even under tumor-derived CM treatment, mimicking the TME (Figure 6G). These findings suggest that the absence of NOXA increases metabolic capacity, enabling NK cells to meet elevated energy demands under immunosuppressive conditions. To further elucidate the metabolic landscape of NOXA-deficient NK cells, we analyzed intracellular metabolites associated with glycolysis, the tricarboxylic acid (TCA) cycle, the pentose phosphate pathway, and lipid metabolism (Figures 6H, 6I, and S13). Notably, levels of glucose-6-phosphate—a key regulator of glycolytic flux and energy production—were significantly elevated in NOXA-KO NK cells (Figure 6H). Additionally, the level of AMP, a crucial energy sensor, was significantly increased in NOXA-KO NK cells, suggesting the activation of AMP-activated protein kinase (AMPK) (Figure 6I). This AMPK activation likely supports the increased metabolic demands by promoting both glycolysis and mitochondrial respiration.27 Collectively, these results demonstrate that NOXA deletion triggers acute metabolic activation, supporting improved NK cell function under metabolic stress.

Discussion

Poor efficacy of NK-based therapeutics is often linked to short-term persistence and survival of NK cells. To address this, we targeted pro-apoptotic genes to enhance NK cell proliferation and effector functions. Among several candidates, high expression of BIM and NOXA in NK cells was strongly associated with poorer OS in patients with PDAC and other cancers. In patients with PDAC, NK^low^ groups demonstrated a negative correlation between BIM and NOXA expression and survival (Figure 1). Furthermore, univariate and multivariate Cox regression analyses identified BIM and NOXA as key prognostic factors across diverse cancer types (Figures 1D–1G; Table 1). These findings suggest that apoptosis-mediated signaling in NK cells significantly impacts their clinical efficacy, contributing to poor clinical outcomes. However, the two genes exhibited distinct prognostic behavior. BIM showed variable and context-dependent associations and did not reach significance at the pan-cancer level. In contrast, NOXA displayed a more consistent survival pattern across NK^low^ and NK^high^ groups and demonstrated reproducible prognostic associations in PAAD, KIRP, and KIRC. This cross-consistency supports NOXA as a more robust and biologically plausible prognostic factor compared with the more variable patterns observed for BIM. Among the various cancer types, we selected pancreatic cancer as the primary disease model because it represents one of the most immunosuppressive solid tumors, characterized by minimal NK infiltration and profoundly impaired NK cytotoxicity.28^,^29 This severe degree of dysfunction provides a biologically meaningful context in which NOXA-mediated pathways may exert stronger or more detectable effects. In addition, pancreatic tumors display high intrinsic apoptotic resistance driven by dysregulated BCL-2 family signaling, making the balance between pro-apoptotic factors such as NOXA and BIM particularly relevant. Finally, our pan-cancer analysis revealed a reproducible association between NOXA expression and patient outcomes specifically in PAAD, further justifying the use of pancreatic cancer as a focused and mechanistically informative model for evaluating NOXA-dependent NK biology.

We hypothesized that apoptosis impairs NK cell function and survival upon tumor interaction and thus targeted the reduction of BIM and NOXA expression to enhance NK-mediated tumor responses. Interestingly, while BIM is widely recognized as a key negative regulator of IL-15-dependent NK cell survival in mice, our study highlights a distinct role for NOXA in human NK cells. In mouse NK cells, IL-15 promotes survival by suppressing apoptosis via the PI3K-Foxo3a pathway, with BIM-KO enhancing cytotoxicity21^,^30 Conversely, NOXA is upregulated following IL-15 withdrawal and works in cooperation with BIM to mediate NK cell apoptosis, thereby limiting persistence and cytotoxic potential. However, accumulating evidence suggests that human NK cells experience more pronounced apoptotic pressure within solid tumors,31^,^32^,^33 where inflammatory cytokines, hypoxia, and stromal interactions converge to drive BCL-2-family-mediated stress responses.34^,^35 In line with this, we found that NOXA expression is strongly induced when human NK cells encounter tumor cells or TME-associated cytokines such as IL-6 and TGF-β1, whereas BIM showed a less consistent pattern of regulation (Figure 2). These observations support the notion that NOXA plays a more prominent role in dictating NK cell vulnerability to TME-derived apoptotic cues—a role that becomes particularly relevant in highly immunosuppressive tumors such as pancreatic cancer. Consistent with this model, NOXA-deficient NK cells maintained cytotoxic effector function even under exposure to immunosuppressive, cytokine-rich conditions, indicating that a loss of NOXA reduces NK cell sensitivity to TME-driven functional suppression (Figure 4).

Although IL-15 is a potent enhancer of NK cell survival, our data indicate that NOXA-deficient NK cells retain additional proliferative and cytotoxic benefits even under IL-15 stimulation (Figure S12). Because IL-15 signaling and NOXA depletion act on distinct regulatory axes—cytokine-mediated survival versus intrinsic apoptotic restraint—their combination may provide complementary advantages for sustaining NK cell fitness in suppressive TMEs. This suggests a potential translational opportunity in integrating NOXA depletion with IL-15-based cytokine engineering to further enhance NK cell therapeutic durability. To functionally validate these mechanistic and translational insights, we next evaluated how NOXA or BIM deletion directly affects NK cell proliferation, metabolism, and cytotoxicity. Using CRISPR/Cas9 to knock out BIM or NOXA in ex vivo-expanded human primary NK cells, we observed that NOXA depletion significantly enhanced NK-cell-mediated cancer killing, even in the absence of cytokines. Surprisingly, BIM-KO NK cells showed minimal improvement in cytotoxicity, despite the upregulation of BIM in response to tumor-derived CM. In contrast, NOXA-KO NK cells demonstrated prolonged proliferation, as evidenced by CFSE-dilution assays and increased Ki67 expression (Figures 6A and 6B). This sustained proliferation was accompanied by enhanced metabolic activity, measured through Seahorse metabolic analysis and lysate-based metabolite profiling (Figures 6F–6I and S12). Acute increases in both OXPHOS and glycolysis were observed, reflecting elevated mitochondrial respiration and energy production that support NK cell functionality and persistence. Together, these results suggest that the pro-apoptotic genes BIM and NOXA have distinct roles in regulating human NK cell function, particularly in terms of longevity and cytotoxicity. While BIM appears to play a more prominent role in murine NK cells, NOXA emerges as a critical regulator in human NK cells, particularly under TME conditions.

In this study, deletion of pro-apoptotic genes using Cas9/sgRNA RNP complexes led to marked improvements in proliferation, metabolic fitness, and cancer-cell-killing capacity of human NK cells. These improvements indicate that targeting NOXA can advance the development of next-generation NK cell therapeutics with prolonged persistence and sustained functionality. Such advancements are likely to translate into superior clinical outcomes, particularly when using engineered NK cells to efficiently target tumors. This work highlights the therapeutic potential of modulating pro-apoptotic signaling to overcome key limitations in current NK-cell-based immunotherapies. By mitigating apoptosis, such strategies may support longer NK cell survival, greater resilience under hostile conditions, and improved cytotoxic efficacy—even within the immunosuppressive TME. Beyond boosting intrinsic cytotoxicity, enhancing the antigen-presenting capabilities of NK cells may also help initiate and amplify adaptive immune responses, including T cell activation and expansion, and coordination with other immune effectors, ultimately generating a broader and more effective anti-tumor immune response.

In conclusion, our findings support the potential of NOXA-KO NK cells as a promising new modality in cancer immunotherapy, offering sustained cytotoxicity, improved metabolic adaptability, and enhanced therapeutic efficacy. Further studies are warranted to fully explore their clinical applicability across various tumor types and therapeutic settings.

Materials and methods

Multiplexed mIF in patient-derived tumor sections

This study was conducted in accordance with the Declaration of Helsinki and approved by the appropriate Institutional Review Board (IRB) approval at the Asan Medical Center (Approval numbers 2018-0745 and 2019-0631). The patients were recruited from the Hepato-Biliary and Pancreatic Surgery Division at the affiliated institution. All participants who underwent surgery for pancreatic cancer were enrolled in the study following written informed consent. Seventeen tumor tissue samples from patients diagnosed with PDAC were analyzed. mIF staining and analysis were performed by prismCDX Co., Ltd (Gyeonggi-do, Korea). Primary antibodies against T cells, NK cells, macrophages, DCs, and cancer cells were used, and antigen visualization was performed using Astra-dye reagents (TheraNovis, San Francisco, CA, USA). A detailed list of the antibodies and their corresponding Astra-dyes is available in Table S1. Slides were scanned at 20× magnification using the PhenoImager HT system (Akoya Biosciences, Marlborough, MA, USA). Representative training images were selected using the Phenochart Whole Slide Viewer (v.1.1.0, Akoya Biosciences). Analysis was performed using inForm Tissue Analysis Software (v.2.6, Akoya Biosciences), which generated unmixed multispectral images using a spectral library.36^,^37 Single-cell segmentation and phenotyping were performed using 4′,6-diamidino-2-phenylindole (DAPI) staining, with phenotype assignment based on expression intensity and localization of target markers. Cells were classified as NK cells (CD56^+^ CD3^−^ CD68^−^ CD11c^−^ CK^−^), T cells (CD3^+^ CD56^−^ CD68^−^ CD11c^−^ CK^−^), macrophages (CD68^+^ CD3^−^ CD56^−^ CD11c^−^ CK^−^), and DCs (CD11c^+^ CD3^−^ CD56^−^ CD68^−^ CK^−^). Data exported from inForm were further processed in RStudio (v.4.2.1) using the phenoptr and phenoptrReport packages (Akoya Biosciences). This comprehensive pipeline enabled detailed characterization of immune and tumor cell populations within the TME.

NK cell signature scoring

To validate the findings, upper quartile normalized gene expression data (Illumina HiSeq_RNA sequencing v2-RSEM_gene_normalized) and corresponding clinical metadata for 8,469 primary tumor samples were retrieved from the Broad GDAC Firehose database (https://gdac.broadinstitute.org). NK cell signature scoring was performed using a previously validated 20-gene NK cell signature.38^,^39 For the PAAD cohort (n = 177) from The Cancer Genome Atlas (TCGA), signature scores were calculated using the singscore R package (v.1.20.0).40 Patients were dichotomized into NK score-low (n = 89) and NK score-high (n = 88) groups based on the median score. This stratification facilitated comparative analyses of clinical and molecular traits associated with NK cell activity.

Prognosis analysis of tumor patients

Survival analysis was conducted using the log rank test and visualized with Kaplan-Meier plots. Univariate and multivariate Cox proportional hazards regression analyses were performed to assess the prognostic relevance of selected factors. All survival analyses were conducted using the survival R package (v.3.7–0), with statistical significance defined as p < 0.05. All statistical computations were performed in R (v.4.2.2).

Data sources

This study utilized publicly available, fully anonymized data and thus did not require additional ethical approval. The datasets were retrieved from the Broad Genome Data Analysis Center (GDAC) Firehose portal (https://gdac.broadinstitute.org). Specifically, upper quartile normalized gene expression data (Illumina HiSeq RNA-sequencing v.2, RSEM gene normalized) and matched clinical information for 8,469 primary cancer tissue samples were used for analysis.

Cell culture

The pancreatic cancer cell lines Capan-2 and AsPC-1 were procured from the Korean Cell Line Bank (KCLB, Seoul, Korea). Cells were cultured in RPMI1640 medium (Welgene, Gyeongsan, Korea) supplemented with 10% fetal bovine serum (FBS, Welgene) and 1% antibiotic–antimycotic (AA) solution (Welgene). The human pancreatic duct epithelial cell line (H6c7) (HPDE) was purchased from Kerafast (Boston, MA, USA) and cultured using Keratinocyte SFM (1X) kit (#17005042; Thermo Fisher Scientific, Waltham, MA, USA) with 1% AA solution according to the manufacturer’s growth conditions and subculturing method. Cell cultures were maintained under standard conditions at 37°C in a humidified incubator with 5% CO_2_.

Organoid and organoid-derived 2D cells

The organoid line SNU-3947-TO, established from the tissue of a patient with PDAC, was obtained from KCLB. To enable organoid-specific detection in co-culture experiments, SNU-3947-TO were transduced with a lentiviral vector encoding tdTomato and luciferase (pDRM18-LTN, Addgene #174721). Transduced organoid was expanded in Cultrex Reduced Growth Factor Basement Membrane Extract (BME; R&D Systems, Systems, Minneapolis, MN, USA). For experimental assays, organoids were resuspended in 100 μL BME and seeded at 20 μL per well in low-attachment CellStar 48-well plates (Greiner Bio-one, Monroe, NC, USA) under standard organoid growth conditions. To generate organoid-derived 2D cells, organoids were transferred to gelatin-coated dishes and cultured for 7 days. Adherent cells were sub-cultured and maintained in Advanced Dulbecco’s Modified Eagle Medium/Nutrient Mixture F-12 (DMEM/F12) supplemented with 10% FBS and 1% AA solution.

Ex vivo expansion and isolation of human primary NK cells

Peripheral blood mononuclear cells were obtained from Lonza (Switzerland). Cells were cultured in NK MACS medium (Miltenyi Biotec, Bergisch Gladbach, Germany). Primary NK cells were expanded by co-culturing with 100 Gy-irradiated K562 feeder cells (American Type Culture Collection, Manassas, VA, USA) in the presence of recombinant human IL-2 (200 IU/mL) and IL-15 (10 ng/mL; PeproTech, Rocky Hill, NJ, USA), as described previously.41 After a 2-week expansion period, NK cells were isolated using the NK Cell Isolation Kit (Miltenyi Biotec) according to the manufacturer’s protocol.

2D cancer-killing assay

NK-cell-mediated cytotoxicity was evaluated using a luciferase-based 2D killing assay. Cancer cells were genetically modified to express NanoLuc luciferase, following established protocols.42 Luciferase-expressing cancer cells were co-cultured with NK cells at specified effector-to-target (E:T) ratios for defined durations. To investigate the factors affecting NK cell cytotoxicity, NK cells were cryopreserved in a freezing medium consisting of 50% FBS, 40% RPMI 1640, and 10% DMSO, and subsequently thawed for the assay. These NK cells were either pre-incubated in cancer cell culture media for 24 h prior to the assay or subjected to the killing assay in the presence of human recombinant IL-6 (200-06-20UG; PeproTech), human recombinant TGF-β1 (#781804; BioLegend, San Diego, CA, USA), or human recombinant IL-15 (200-15-250UG; PeproTech). Luminescence was measured using the Nano-Glo Luciferase Assay System (N1120; Promega, Madison, WI, USA) per the manufacturer’s instructions. Luminescence intensity was used as a quantitative measure of cancer cell death induced by NK cells. For 2D assays involving organoid-derived cells, SNU-3947-TO_2D cells were co-cultured with NK cells under comparable conditions. Following co-culture, cell viability was assessed using a viability assay kit (MediFab, Seoul, Korea), and absorbance at 600 nm was measured using a GloMax Discover Microplate Reader (Promega).

3D organoid-killing assay

For cytotoxicity assays, organoids were transferred to BME-free Corning 96-well Spheroid microplates (Corning, Corning NY, USA) and co-cultured with NK cells at an E:T ratio of 2:1 in organoid culture medium supplemented with 200 IU/mL recombinant human IL-2. Fluorescence images of organoids were acquired every 3 days throughout the co-culture period using a Nikon Ts2 fluorescence microscope (Nikon, Tokyo, Japan). On day 13, organoid viability and apoptosis were assessed using complementary methods.

For viability assessment, tdTomato fluorescence intensity was quantified using ImageJ software (National Institutes of Health, Bethesda, MD, USA), and the fluorescence intensity was normalized to organoid-only controls. Additionally, D-luciferin (150 μg/mL; MedChemExpress, Monmouth Junction, NJ, USA) was added to each well, and luminescence was measured using a Victor X3 multilabel plate reader (PerkinElmer, Waltham, MA, USA). Relative viability was calculated based on the reduction in luminescence signal compared to organoid-only controls.

To assess apoptosis, CellEvent Caspase-3/7 Red Detection Reagent (5 μM; Thermo Fisher Scientific) was added to the co-cultures according to the manufacturer’s protocol. Following 60 min of incubation at 37°C, fluorescence images were acquired using a Nikon Ts2 fluorescence microscope. Apoptotic areas positive for caspase-3/7 signal were quantified using ImageJ software and expressed as a percentage of total organoid area.

For comparison with 2D models, some organoids were transferred to gelatin-coated dishes and cultured for 7 days to generate adherent cells. Adherent cells were then sub-cultured and maintained in Advanced DMEM/F12 containing 10% FBS and 1% AA solution.

Western blot analysis

Cells were lysed using radioimmunoprecipitation assay buffer (Thermo Fisher Scientific) supplemented with a protease inhibitor cocktail (Thermo Fisher Scientific). Lysates were incubated on ice for 20 min, followed by centrifugation at 14,000 rpm for 20 min to collect the supernatant containing total proteins. Proteins were separated by SDS-PAGE and transferred onto a polyvinylidene fluoride membranes (1620177; Bio-Rad, Hercules, CA, USA). A molecular weight protein ladder (26626; Thermo Fisher Scientific) was used to confirm protein separation. Membranes were blocked with 5% skim milk in 0.2% Tween 20/Tris-buffered saline and incubated with primary antibodies at the indicated dilutions. After washing, membranes were incubated with species-specific secondary antibodies conjugated to horseradish peroxidase (HRP). Protein bands were visualized using either Pierce ECL western blotting substrate (#32106; Thermo Fisher Scientific) or SuperSignal West Femto Maximum Sensitivity Substrate (#34095; Thermo Fisher Scientific) and imaged with the ChemiDoc system (Bio-Rad). The following antibodies were used in this study: anti-BIM antibody (1:1,000, ALX-804-527-C100; Enzo life sciences, Farmingdale, NY, USA), anti-NOXA antibody (1:1,000, ab13654; Abcam, Cambridge, United Kingdom), anti-MCL-1 antibody (1:1,000, A0250; ABclonal, Woburn, Massachusetts, USA), anti-beta actin antibody (1:2,000, sc47778; Santa Cruz, USA), goat anti-rabbit IgG HRP-linked antibody (1:5,000, #31470; Invitrogen, Carlsbad, CA, USA), and goat anti-mouse IgG and HRP-linked antibody (1:5,000, #31430; Invitrogen).

Generation of BIM- and/or NOXA-KO NK cells

To generate NK cells with BIM- and/or NOXA KO NK cell, cells were treated ex vivo with CRISPR/Cas9 RNPs. Sequence-specific CRISPR RNAs (crRNAs) were hybridized with trans-activating CRISPR RNAs (tracerRNAs) in duplex buffer (#11-01-03-01, Integrated DNA Technologies [IDT], Coralville, IA, USA) according to the manufacturer’s instructions. The crRNA-tracrRNA duplexes were complexed with Streptococcus pyogenes Cas9 (spCas9) protein (IDT) at a 2:1 w/w ratio and incubated for 15 min at room temperature to form RNP complexes. Five million NK cells were electroporated with 12.2 pmol of Cas9 RNPs using the Neon Transfection System (Invitrogen) with a single 20-ms pulse at 1,800 mV. Cells were incubated for 48 h post-electroporation and assessed for gene editing efficiency using the T7E1 assay to confirm the KO of BIM or NOXA.

Genomic DNA was extracted using the QIAamp DNA Mini Kit (QIAGEN, Hilden, Germany). Target sequences were amplified with Platinum SuperFi II PCR Master Mix (Invitrogen), and PCR products were digested with T7E1 (New England biolabs, Ipswich, MA, USA) according to the manufacturer’s instructions. The primer sequences used were BIM (Forward (F): 5′- GGAAATGGAAGTGTGTATGAATGG -3′; Reverse (R): 5′-GGGGAGTACAGAAACACACTAC -3′); and NOXA (F: 5′- GGCTGGTGACTTATGCTACTC -3′; R: 5′- CCAGCGGTAATCTTCGGC -3′).

ICE analysis

To verify guide RNA specificity and evaluate KO efficiency, PCR amplicons of BIM and NOXA (as described above) were subjected to Sanger sequencing (commercial provider). KO efficiency was analyzed using the ICE analysis tool (Synthego, Redwood City, CA, USA). The primers used for sequencing were as follows: BIM (5′- ATGTAGAAGACTCTGCCACTC -3′) and NOXA (5′- AAGGTAACACTTGCCTCATCC -3′).

Flow cytometry

Flow cytometric analysis was performed using the CytoFLEX system (Beckman Coulter Life Sciences, Indianapolis, IN, USA), and data were analyzed with FlowJo software (BD Biosciences, San Jose, CA, USA). The following antibodies were used: PE anti-human CD16 (360704; BioLegend), PE/Cyanine7 anti-human CD16 Antibody(360704; BioLegend), PE/Cyanine7 anti-human CD69 (328620; BioLegend), Brilliant Violet 421 anti-human CD226 (DNAM-1) antibody (338332; BioLegend), APC anti-human CD107a (LAMP-1) (328620; BioLegend), human NKG2C/CD159c Alexa Fluor 700-conjugated antibody (FAB138N; R&D Systems), APC/Cyanine7 anti-human CD314 (NKG2D) antibody (#320824; BioLegend), APC anti-human CD337 (NKp30) antibody (# 325210; BioLegend), APC anti-human CD336 (NKp44) antibody (#325110; BioLegend), Pacific Blue anti-human CD335 (NKp46) antibody (#331912; BioLegend), and Pacific Blue anti-human Ki-67 (350512; BioLegend). Dead cells were identified using Fixable Viability Dye eFluor 780 (65-0865-14; Invitrogen).

CFSE-based proliferation assay

Cell proliferation was assessed using the CellTrace CFSE Cell Proliferation Kit (C34554; Invitrogen). Control and NOXA-KO NK cells were stained with 2.5 μM CFSE in phosphate-buffered saline (PBS) 37° for 15 min. Following staining, cells were washed by centrifugation at 1,500 rpm for 4 min. Proliferation was assessed on days 5 and 8 post-electroporation using flow cytometry.

Apoptosis assay (AnnexinV/7-AAD)

To evaluate apoptosis induction by cancer-cell-conditioned media, NK cells were incubated with media collected from AsPC-1 or Capan-2 cultures for 3 days. After incubation, cells were harvested and washed with Annexin V binding buffer (51-66121E; BD Biosciences). Cells were stained with Annexin V (BD Biosciences) and 7-aminoactinomycin D (7-AAD; BioLegend) and analyzed using the CytoFLEX system (Beckman Coulter Life Sciences).

Metabolic activity assessment

NK cell metabolic activity was evaluated using the Seahorse XF Mito Stress Test (Agilent Technologies, Waldbronn, Germany) according to the manufacturer’s protocol. One day prior to the assay, the sensor cartridge was hydrated with XF Calibrant (103792-100) and incubated at 37°C in a non-CO_2_ incubator. Cell culture microplates were coated with poly-D-lysine (Gibco, Carlsbad, CA, USA) for 1 h at room temperature. NK cells (4 × 10^5^ cells per well) were seeded into XF DMEM base medium (103575-100) supplemented with 4.5 mM glucose (103577-100), 1 mM sodium pyruvate (103578-100), and 2 mM L-glutamine (103579-100), all from Agilent Technologies. The plate was centrifuged at 200 × g for 1 min. Inhibitors including oligomycin, carbonyl cyanide-p-trifluoromethoxyphenylhydrazone (FCCP), and a rotenone/antimycin A mixture (103015-100) were prepared in assay medium and loaded into the appropriate ports of the hydrated cartridge. The assay was performed using the Seahorse XFe96 Analyzer (Agilent Technologies) following a 30-min calibration period. Data were collected and analyzed using Wave software (Agilent Technologies).

Metabolite analysis from lysed NK cells

NK cells were harvested using 1.4 mL of cold methanol/water (80:20, v/v) following rapid and sequential washes with PBS and water. Cell lysis was achieved by vigorous vortexing. Internal standard solutions were then added: 50 μL of 5 μM ^13^C_5_-glutamine (for energy metabolism), 1 μM of 18:0 D_70_-phosphatidylcholine, and 1 μM of 16:0 D_31_-18:1-phosphatidylethanolamine (for phospholipid profiling). Metabolites were extracted by liquid-liquid extraction after the addition of chloroform and water, separating the aqueous and organic phases. Both fractions were dried under vacuum centrifugation and stored at −20°C until further analysis by liquid chromatography-tandem mass spectrometry (LC-MS/MS). Prior to LC-MS/MS, the dried aqueous phase (energy metabolism) was reconstituted in 50% methanol, while the organic phase (phospholipids) was reconstituted in 100% methanol. Standard metabolites and internal standards were purchased from Avanti-Polar Lipids and Sigma-Aldrich. All solvents including water were purchased from J. T. Baker, Phillipsburg, NJ, USA.

LC-MS/MS analysis

Energy metabolism metabolites were analyzed using LC-MS/MS 1290 high-performance liquid chromatography (HPLC; Agilent Technologies) and QTRAP 5500 mass spectrometer (AB Sciex, Toronto, Canada) in negative ion mode. Phospholipid profiling was performed using a 1290 HPLC (Agilent, Waldbronn, Germany) coupled to a Triple Quad 6500 system (AB Sciex) in positive ion mode. Metabolites related to energy metabolism were separated using a Synergi fusion RP column (50 × 2 mm). The mobile phases were as follows: A: 5 mM ammonium acetate in H_2_O and B: 5 mM ammonium acetate in MeOH. The gradient elution was programmed as follows: 0% B for 5 min, 0%–90% B for 2 min, hold at 90% for 8 min, return to 0% B for 1 min, and then hold at 0% B for 9 min. The LC flow rate was maintained at 70 μL/min, except from 7 to 15 min when it was increased to 140 μL/min. The column temperature was maintained at 23°C. For phospholipids profiling, Zorbax Eclipse Plus C18 column (2.1 × 50 mm) was used with mobile phase A (10 mM ammonium acetate in MeOH/isopropyl alcohol (IPA)/H2O (900/50/50, v/v/v) and mobile phase B (10 mM ammonium acetate in MeOH/IPA/H2O (940/50/10, v/v/v). The separation gradient was as follows: hold at 60% B for 10 min, 60%–90% B for 0.1 min, hold at 90% for 7.9 min, 90%–60% B for 0.1 min, and finally hold at 60% B for 6.9 min. The LC flow was 300–400 μL/min, and the column temperature was maintained at 23°C. Metabolites were quantified using extracted ion chromatograms (EICs) corresponding to specific multiple reaction monitoring (MRM) transitions. For energy metabolism, peak areas were normalized to internal standards and total protein concentration in each sample. Phospholipids were quantified using external standard calibration curves (0.1–5,000 nM; R^2^ > 0.99), and the values were similarly normalized to total protein. All data analysis was performed using Analyst v.1.7.1 software (AB Sciex).

Data and code availability

The data that support the findings of this study are available from the corresponding author, upon reasonable request.

Acknowledgments

This study complied with the Declaration of Helsinki and was reviewed and approved by the Institutional Review Board (IRB) of Asan Medical Center (IRB No. 2018-0745, 2019-0631). Patients who underwent surgery for cancer were enrolled in this study, and informed consent was obtained before taking part in the study. We thank the Bioinformatics core lab and Metabolomics core lab at the ConveRgence mEDIcine research cenTer (CREDIT) and Asan Medical Center for support and instrumentation. This work was supported by the 10.13039/501100003725National Research Foundation of Korea Grant funded by the Korean Government (grant numbers: 2020M3A9I4038667, 2022R1A2C1008751, RS-2023-NR076986, RS-2024-00509114, and RS-2025-02216943) and by the 10.13039/501100003693KIST Institutional Program (grant number: 2E33761).

Author contributions

S.-B.K. was responsible for curation, formal analysis, investigation, validation, visualization, resources, writing—original draft, and writing—review & editing. J.H.J. was responsible for data curation, formal analysis, investigation, validation, visualization, and writing—original draft. S.W.K. and H.Y. were responsible for investigation and methodology. S.L. was responsible for methodology and software. J.-H.O. and C.O.S. were responsible for methodology, formal analysis, and software. S.-H.L. was responsible for validation. E.J. was responsible for conceptualization, funding acquisition, supervision, and writing—review & editing. M.J. is responsible for conceptualization, funding acquisition, validation, supervision, project administration, writing—original draft, and writing—review & editing. All authors reviewed the manuscript.

Declaration of interests

The authors declare no competing interests.