Enhanced PD-L1 targeting boosts the cytotoxic activity of FOLR1- CAR NK92 cells against ovarian cancer

Abstract

Chimeric antigen receptor (CAR)-T cells have achieved remarkable success against hematologic malignancies; however, their application is associated with risks such as cytokine release syndrome (CRS) and neurotoxicity. In contrast, CAR-NK cells not only avoid these toxicities but also retain the natural cytotoxic activity of NK cells, demonstrating great potential for cancer immunotherapy.

Our previous study demonstrated the potent efficacy of folate receptor alpha (FOLR1)-targeted CAR-NK92 cells against ovarian cancer (OC). Nevertheless, the application of CAR-NK therapy in solid tumors, including OC, still faces challenges such as tumor antigen heterogeneity and the immunosuppressive tumor microenvironment (ITME). A key mediator of ITME is programmed death-ligand 1 (PD-L1), which not only correlates with poor prognosis in OC but also drives T cell exhaustion via the PD-1/PD-L1 axis. Moreover, its inducible upregulation under NK cell-based therapy supports PD-L1 as a viable therapeutic target in OC.

To enhance the therapeutic potential of CAR technology for OC, we engineered two novel third-generation bispecific CAR-NK92 cells based on our prior FOLR1-CAR design. These constructs—Tandem PD-L1/FOLR1-CAR (Tan-CAR1) NK92 and Tandem FOLR1/PD-L1-CAR (Tan-CAR2) NK92—were designed to simultaneously target FOLR1 and PD-L1. Notably, we confirmed that PD-L1 expression was significantly upregulated in the co-culture supernatant of effector and target cells. In vitro, Tan-CAR2 NK92 cells exhibited markedly superior cytotoxicity against FOLR1^highPD-L1^low OC cells and enhanced degranulation compared with FOLR1-CAR NK92 cells. In xenograft mouse models, Tan-CAR2 NK92 cells showed effective tumor infiltration and induced significantly greater tumor regression than both FOLR1-CAR NK92 and PD-L1-CAR NK92 groups. In conclusion, the tandem bispecific CAR-NK92 cells developed in this study represent a meaningful advance in immunotherapy. Their dual mechanism—broader antigen recognition and enhanced immune infiltration—effectively addresses two major therapeutic barriers in OC, offering a promising strategy for treating refractory cases.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00262-026-04443-7.

Introduction

Ovarian cancer (OC) has the highest mortality rate among gynecological malignancies. Due to non-specific symptoms, over 70% of patients are diagnosed at an advanced stage, and despite initial treatment with current clinical modalities, up to 75% of patients experience recurrence within two years, resulting in a five-year overall survival rate of less than 40% [1]. Therefore, in addition to exploring effective early screening methods, developing novel and efficient treatment strategies remains an urgent priority in clinical practice and research.

CAR-T cell therapy genetically engineers T cells to express CARs that specifically recognize tumor-associated antigens (TAAs), enabling them to destroy cancer cells. This approach has demonstrated remarkable efficacy against hematologic malignancies [2]. However, nearly all patients receiving CAR-T therapy experience varying degrees of CRS, which may progress to severe complications such as heart failure, multi-organ dysfunction, and even death, significantly limiting its clinical application. CAR-NK therapy, utilizing natural killer (NK) cells as a cellular platform, has emerged as a highly promising alternative. NK cells retain their innate immune mechanisms to eliminate tumor cells, including receptor-mediated recognition of stress-induced ligands and antibody-dependent cell-mediated cytotoxicity (ADCC). These features help reduce the risk of tumor recurrence caused by antigen escape. Furthermore, CAR-NK therapy rarely induces cytokine storms or neurotoxicity and does not cause graft-versus-host disease (GVHD). In the field of solid tumors, preclinical research on CAR-NK cells has made significant progress in OC, with main targets including FOLR1 [3], mesothelin [4], and MUC16 [5, 6]. Our previous research [7] confirmed that FOLR1-CAR NK92 cells significantly suppressed OC progression in murine models.

However, these studies face common challenges during treatment [8, 9], including tumor immune escape, the ITME, and antigen heterogeneity. All these factors restrict the antitumor efficacy of CAR technology and may lead to disease recurrence. PD-L1, a type I transmembrane protein of the B7 family, exhibits heterogeneous expression (28–80%) in OC tissues but remains at low levels in normal tissues [10, 11]. PD-L1 expression is closely associated with OC metastasis and poor patient prognosis [12, 13]. Moreover, the PD-1/PD-L1 signaling pathway is a primary mechanism of adaptive immune resistance in tumors. It triggers tyrosine phosphorylation of the ITIM and ITSM motifs within the intracellular domain of PD-1, leading to T cell exhaustion and regulatory T cell differentiation [14–16], ultimately forming an ITME that facilitates immune escape. Experimental evidence [17, 18] indicates that blocking this pathway restores T cell function, enhances their activity, and increases the number of tumor-infiltrating lymphocytes (TILs), thereby promoting tumor clearance.

In this study, we integrated PD-L1 targeting into CAR-NK92 design to overcome therapeutic barriers and improve outcomes. Building on our prior FOLR1-CAR research, we engineered two bispecific tandem CAR-NK92 cell lines with different single-chain variable fragment (scFv) arrangements in the extracellular domain: Tandem PD-L1/FOLR1-CAR (Tan-CAR1) and Tandem FOLR1/PD-L1-CAR (Tan-CAR2). The cells were assessed on three OC lines: SKOV3 and OVCAR3 (both FOLR1^highPD-L1^low) and A2780 (FOLR1^lowPD-L1^low). Tan-CAR2 NK92 cells exerted superior antitumor activity compared with FOLR1-CAR NK92 cells both in vitro and in vivo. Moreover, the PD-L1-targeting capacity of Tan-CAR2 significantly promoted TIL infiltration. Additionally, due to the inducible expression of PD-L1 in OC cells, PD-L1-CAR NK92 cells retained significant cytotoxic effects even against SKOV3 and OVCAR3 (both FOLR1^highPD-L1^low).

In summary, this study reports the first tandem CAR design targeting both FOLR1 and PD-L1. We not only validate the effectiveness of the FOLR1‑scFv–PD‑L1‑scFv tandem configuration in bispecific CAR‑NK92 cells but also show that enhancing PD-L1-targeting capability significantly improves therapeutic outcomes against OC. These results offer a novel strategy to optimize CAR‑NK cell therapy and provide a foundation for advancing tumor immunobiology‑based treatments.

Materials and methods

Patients and tissue samples

Tumor tissues were obtained from OC patients who underwent surgical resection between June 2020 and June 2025 at Chongqing Daping Hospital. Ethical approval for the use of human subjects was obtained from the Research Ethics Committee of Daping Hospital (Chongqing, China), which granted a waiver of informed consent (2025–414).

Cell lines

Human OC cell lines SKOV3, OVCAR3, A2780, and the human NK92 cell line were purchased from Fenghui (Hunan, China). SKOV3 was cultured in McCoy’s 5A (Vivacell, China) supplemented with 10% fetal bovine serum (FBS, Vivacell, China). OVCAR3 was cultured in RPMI-1640 (Vivacell, China) containing 20% FBS and 0.01 mg/mL insulin (Vivacell, China). A2780 was grown in RPMI-1640 containing 10% FBS. NK92 cells were maintained in Minimum Essential Media complete medium (Procell, China). All the cell lines were maintained in a humidified incubator at 37 °C with 5% CO₂.

Immunohistochemistry (IHC) Staining

Human OC samples and mouse OC xenograft tissues were prepared into paraffin sections. After dewaxing, rehydrating, and antigen retrieval, sections were blocked at room temperature for 30 min. Primary antibodies were incubated overnight at 4 °C, including anti-PD-L1 (1:800, Abcam, 13684 T), anti-FOLR1 (1:250, Invitrogen, MA5-23917), and anti-CD56 (1:1000, Servicebio, GB112671). After washing, either anti-rabbit HRP (1:200, Servicebio, GB23303) or anti-mouse HRP-conjugated (1:200, Servicebio, GB23301) secondary antibodies were applied at 37 °C for 1 h. Staining was visualized with 3, 3′-diaminobenzidine (DAB; Thermo Fisher Scientific, USA). Images were obtained using a Nikon E100 microscope or a BestScope fully automated digital slide scanner. The expression levels of PD-L1 and FOLR1 were evaluated in a blinded fashion by two experienced pathologists using a 4-point scale based on staining intensity: 0 (negative, no staining), 1 (weak, pale yellow), 2 (moderate, yellowish-brown or light brown), and 3 (strong, dark brown or tan). A score of ≥ 2 was defined as membrane positivity.

Flow Cytometry

OC cell lines were pre-plated in two 24-well plates. Each well was seeded with 1 mL of cell suspension at a density of 1.5 × 10^6 cells/mL with three replicate wells per group. After complete adhesion, one plate was taken, and 1 × 10^5 NK92 cells in 1 mL of medium were added to the experimental group wells, while 1 mL of medium only was added to the control group wells. After co-culture for 4 h, 2 mL of supernatant was collected and added to the corresponding wells in the other plate. Following incubation for an additional 6 h, tumor cells were collected, equally aliquoted into five tubes, and stained separately with the following antibodies: blank tube, APC-iso-PD-L1 (1:100, Abcam, ab199093), PE-iso-FOLR1 (5uL/10^6 cells, Biolegend, 400214), APC-PD-L1 (1:100, Abcam, ab209960), PE-FOLR1 (5uL/10^6 cells, Biolegend, 908304). After incubation at 4 °C in the dark for 30 min, samples were loaded onto a CytoFlex flow cytometer (Beckman Coulter, USA) for detection.

Overlapping extension PCR and genomic DNA PCR

Overlapping extension PCR was used to generate four CAR sequences: Tan-CAR1, Tan-CAR2, FOLR1-CAR, and PD-L1-CAR. Detailed information on PCR templates, primers, products, and sequence verification is provided in the Supplementary Materials. Genomic DNA was extracted from CAR-NK92 cells using a Genomic DNA Extraction Kit (TIANGEN, DP304) and used as the PCR template. The primer sequences were forward 5′-ATGGCCCTCCCAGTTACC-3′ and reverse 5′-CGGGATCCTCAGCGGGGGGGCAGGGC-3′. The 20 μL PCR reaction mixture contained 10 μL of DNA polymerase (Vazyme, P212), 0.8 μL each of primers, 2 μL of genomic DNA template, and 6.4 μL of ddH₂O. The PCR protocol was as follows: initial denaturation at 95 °C for 360 s; 32 cycles of denaturation at 95 °C for 15 s and extension at 72 °C for 235 s; and a final extension at 72 °C for 300 s.

Lentivirus production and transduction

The synthetic CARs were cloned into the lentiviral vector pRRLSIN-cPPT-SFFV-EGFP-MCS-SV40-puromycin (GenScript, China). Lentiviral particles were generated by transiently transfecting HEK 293T cells. The virus-containing supernatant was collected, filtered through a 0.45 μm membrane, and concentrated by ultracentrifugation (25,000 rpm, 2 h, 4 °C). NK92 cells were transduced with the concentrated lentivirus at an MOI of 300 and cultured at 37 °C in 5% CO₂.

Flow sorting

At 96 h post-transduction, GFP expression in NK92 cells was observed under a fluorescence microscope (LEICA, Germany). A single-cell suspension of lentivirus-infected NK92 cells was prepared at a concentration of 8 × 10^6 cells/mL. After filtration, sorting was performed using a flow cytometer (Beckman Coulter, USA). Sorted NK92 cells were cultured at 37 °C, 5% CO₂ for subsequent experiments.

Quantitative real-time PCR(qPCR)

Total cellular RNA was isolated using Trizol reagent (Invitrogen, 15596018CN). Extracted RNA was reverse-transcribed into cDNA, which was used for qPCR with a Genestar SYBR Green kit (Takara, Japan), using GAPDH as the endogenous reference gene. Relative expression levels were calculated using the 2⁻ΔΔCT method. The primers used are shown in Table 1.

Table 1.

qPCR primers

Primer	Primer sequence (5’-3’)	TM/℃
CD3ξ-F	TCCCAGTTACCGCCCTTCTCCT	63.0
CD3ξ-R	ATGCTCTGTCCAGGAGACCCAGAC	63.2
GAPDH-F	GCACCGTCAAGGCTGAGAAC	59.5
GAPDH-R	TGGTGAAGACGCCAGTGGA	59.4

Cellular immunofluorescence

Confocal dishes pre-coated with poly-lysine (Gibco, A3890401) were seeded with NK92 cell suspension to achieve final cell density of 80–90%. After attachment, cells were fixed, permeabilized, and blocked. Subsequently, cells were incubated with an anti-Flag primary antibody (1:1000, Sigma, F1804) overnight at 4 °C and then with a CoraLite594-conjugated goat anti-mouse antibody (1:200, Proteintech, SA00013-3) for 1 h in the dark. Images were acquired using a fluorescence microscope.

Cell counting kit-8 (CCK-8) assay

Wild-type NK92 cells and five sorted NK92 cell lines were seeded at 1 × 10^4 cells per well in 96-well plates (100 μL per well). Unmodified NK92 cells served as the control group. Cells were continuously cultured at 37 °C, 5% CO₂. At 0, 24, 48, and 72 h of culture, 10 μL of CCK-8 solution (MCE, HY-K0301) was added to each well and co-cultured for 3 h. Absorbance was measured at 450 nm using a SpectraMax M2 microplate reader (Molecular Devices, USA).

Lactate dehydrogenase (LDH) cytotoxicity assay

To assess cytotoxicity, effector cells were co-cultured with target cells (SKOV3, OVCAR3, A2780) at various effector-to-target (E/T) ratios (1:1, 2:1, 4:1, 8:1). After 8 h, LDH assay reagent (DOJINDO, CK12-500) was added to the wells according to the manufacturer’s instructions. Absorbance was measured at 490 nm using a SpectraMax M2 microplate reader. Cytotoxicity (%) was calculated using the formula: [(ER—ESR—TSR + CMB) / (TMR—VCC—TSR + CMB)] × 100%. ER: experimental release, ESR: effector spontaneous release, TSR: target spontaneous release, CMB: culture medium background, TMR: target maximum release, VCC: volume correction control.

CD107a expression

Target cells were incubated with effector cells for 4 h at an E/T ratio of 4:1. Effector cells were then collected and aliquoted into three tubes. Antibodies were added and labeled as unstained control, APC-isotype control (5uL/10^6 cells, Biolegend, 400122), and APC-CD107a (5uL/10^6 cells, Biolegend, 328620). Cells were incubated at 4 °C in the dark for 30 min and analyzed for CD107a expression using a CytoFLEX flow cytometer.

In vivo antitumor experiments

SKOV3 was selected for xenograft due to its high engraftment efficiency and appropriate growth rate. A total of 30 five-week-old female BALB/c nu/nu mice (purchased from Beijing Huafukang Biotechnology) were randomly divided into five groups: Tan-CAR1 NK92, Tan-CAR2 NK92, PD-L1-CAR NK92, FOLR1-CAR NK92, and NK92 control. After one week of acclimation, each mouse was subcutaneously inoculated with 3.5 × 10^6 SKOV3 cells in the right dorsal shoulder region. When tumor volume reached approximately 100 mm³, 3.5 × 10^6 effector cells were injected peritumorally. Body weight and tumor volume were monitored every 48 h. Mice were euthanized immediately if they exhibited more than 20% loss of initial body weight, or if ulceration or infection occurred at the tumor site. At the end of the experiment, tumors were completely excised. Photographed, weighed, and fixed tumors in 4% paraformaldehyde (BIOSSCI, BP003) for preservation.

Hematoxylin–eosin (HE) staining

Paraffin-embedded mouse xenografts were sectioned into 4-µm-thick paraffin sections. Sections were baked, dewaxed, and hydrated in ethanol, followed by sequential staining with hematoxylin and eosin. Finally, they were dehydrated with ethanol, cleared with xylene, and promptly mounted. Sections were examined for pathological changes under a light microscope (LEICA, Germany).

Statistical analysis

Data are presented as mean ± standard deviation (SD) from at least three independent biological replicates. Biological replicates were defined as experiments performed on separate days with independently cultured cells; technical replicates were averaged prior to analysis. Statistical analyses were performed using GraphPad Prism 8 (GraphPad Software Inc, USA), and flow cytometry data were analyzed using FlowJo V10 (FlowJo LLC, USA). For statistical comparisons, unpaired two-tailed Student’s t test was used for two groups, and one-way or two-way ANOVA with Tukey’s post-hoc test was used for multiple comparisons. Survival was evaluated by the Kaplan–Meier test. Significance levels: ns, p > 0.05; * or #, p < 0.05; ** or ##, p < 0.01; *** or ###, p < 0.001; **** or ####, p < 0.0001.

Results

Expression of PD-L1 and FOLR1 in human OC tumor tissues

Tumor targeting by CAR-NK cells is closely related to the expression of TAAs on the tumor cells surface. To evaluate PD-L1 and FOLR1 expression in OC, we performed IHC on 34 paraffin-embedded human OC samples. Both antigens were detected on the cell membrane. Analysis showed that 38.24% (13/34) of OC tissues were PD-L1-positive, with 14.71% (5/34) showing strong positivity. Meanwhile, 58.82% (20/34) were FOLR1-positive, with 32.35% (11/34) showing strong positivity (Fig. 1A, B).

Fig. 1.

FOLR1 and PD-L1 are highly expressed in a subset of OC tissues and predicted poor prognosis. A Representative immunohistochemical (IHC) staining of PD-L1 and FOLR1 in human OC tissues. Staining intensity was evaluated independently by two experienced pathologists using a 4-point scale (0–3). A score of ≥ 2 was defined as membrane positivity. Scale bars: 20 µm. B The percentage of PD-L1 and FOLR1 positive staining with different scores in 34 OC samples. C Kaplan–Meier curves for progression-free survival (PFS) based on PD-L1 and FOLR1 expression

Furthermore, Kaplan–Meier survival analysis revealed that high PD-L1 expression was significantly associated with shorter progression-free survival (PFS) and an increased risk of disease progression or death (HR = 1.51, P = 2e-05). Patients with high FOLR1 expression showed a comparable result (HR = 1.24, P = 0.0026) (Fig. 1C). These findings indicate that high expression of PD-L1 and FOLR1 are both poor prognostic factors in OC patients. In summary, PD-L1 and FOLR1 can serve as potential ideal therapeutic targets for OC.

Induced PD-L1 expression in OC cell lines

We investigated FOLR1 and PD-L1 expression in human OC cell lines (SKOV3, OVCAR3, and A2780) by flow cytometry. Considering that cytokines such as IFN-γ and TNF-α secreted by immune effector cells (including T cells and NK cells) in the immune microenvironment can upregulate PD-L1 expression in tumor cells [19, 20], we established an induction group, in which tumor cells were co-cultured with supernatant from effector-target cell co-culture system to assess changes in PD-L1 and FOLR1 expression.

The positive percentage assessment showed that parental SKOV3, OVCAR3, and A2780 cells exhibited < 10% PD-L1 expression, and FOLR1 expression was 85.89 ± 3.42%, 99.54 ± 0.55%, and 2.57 ± 0.43%, respectively (Fig. 2A). After induction, PD-L1 expression increased to 30.29 ± 2.41% (P < 0.0001), 23.73 ± 1.82% (P < 0.0001), and 7.77 ± 1.32% (P > 0.05), respectively, while FOLR1 expression was 93.62 ± 0.46%, 99.79 ± 0.15%, and 1.80 ± 0.83%, with no significant change (P > 0.05) (Fig. 2B, C). These findings indicated that factors derived from NK92 cells could induce upregulation of PD-L1 in OC cells (SKOV3 and OVCAR3), whereas FOLR1 expression was not significantly affected.

Fig. 2.

PD-L1 and FOLR1 surface antigen expression in parental vs. supernatant-induced OC cell lines. A Expression of PD-L1 and FOLR1 on three parental OC cell lines. B Expression of PD-L1 and FOLR1 on three supernatant-induced OC cell lines. C Quantification of antigen-positive rate and mean fluorescence intensity (MFI) for all OC cell lines. Data are presented as mean ± SD from three independent experiments (n = 3 per group). P values were calculated by two-way ANOVA followed by Tukey’s multiple comparisons test. Statistical analysis is shown for comparisons between parental and corresponding induced tumor cells (*) or among induction groups (#). n.s.: not significant (P > 0.05); **P < 0.01; ****P < 0.0001; ##P < 0.01; ####P < 0.0001

Thus, both NK92 and NK92-derived CAR-NK92 effector cells can upregulate PD-L1 expression in OC cells. We conclude that this finding is generalizable to CAR-NK92 cells, given that NK92 is the parental line used in our study.

Results based on mean fluorescence intensity (MFI) were consistent with this trend. Notably, the MFI of the target antigen in OVCAR3 was significantly higher than that in SKOV3, with A2780 showing the lowest MFI, thus providing three antigen density levels: high (OVCAR3), medium (SKOV3), and low (A2780) for functional evaluation of CAR-NK cells. Overall, SKOV3 and OVCAR3 were selected as target cells for Tan-CAR NK92 functional assays, while A2780 was used as a negative control.

Tan-CAR NK92 cell generation and proliferation

The two tandem CARs differ only in the order of their scFv domains (Fig. 3A), and four lentiviral vectors encoding various CARs are represented in Fig. 3B. At 96 h post-lentivirus transduction, NK92 cells expressed green fluorescence (Fig. 3C). As CAR and GFP are driven by the same promoter, we sorted GFP-positive cells by fluorescence-activated cell sorting (FACS) to purify CAR-expressing engineered NK92 cells. Following expansion, all five engineered NK92 cell lines exhibited GFP positivity rates above 85%: NK92-EV (NK92 cells transduced with empty vector), Tan-CAR1 NK92, Tan-CAR2 NK92, PD-L1-CAR NK92, and FOLR1-CAR NK92 (Fig. 3D).

Fig. 3.

Lentiviral transduction and purification of Tan-CAR NK92 cells. A Schematic diagram of two tandem CARs. The two CARs differ only in the order of their single-chain variable fragment (scFv) domains. B Schematic diagram of Tan-CAR and single-target CAR structures. All CARs are third-generation designs, structurally composed of CD8α signal peptide, extracellular scFv domain, IgG1 Fc fragment, CD8α hinge region, CD28 transmembrane region, CD28-CD137-CD3ζ intracellular signaling domain, and a 3FLAG tag. Each CAR is linked to EGFP via a T2A peptide under a single promoter. C After lentiviral transduction, fluorescence expression in NK92 cells from each group was observed under a fluorescence microscope. Untransduced wild-type NK92 cells (control group) showed no green fluorescence, while NK92 cells transduced with empty vector or various CAR constructs (experimental groups) all exhibited distinct green fluorescence. D Fluorescence expression rate of experimental group cells detected by flow cytometry following FACS enrichment. Using untransduced NK92 cells as a blank control, the GFP-positive rate exceeded 85% in all experimental groups. NK92-EV: NK92 cells transduced with empty vector. Data are representative of three independent experiments

We confirmed CAR expression at different levels through genomic DNA PCR, qPCR and immunofluorescence assays. The results demonstrated that Tan-CAR1, Tan-CAR2, PD-L1-CAR, and FOLR1-CAR genes were successfully integrated into the NK92 cell genome (Fig. 4A). All four CAR genes showed effective expression at both transcriptional (Fig. 4C) and protein levels, with the latter detected via their FLAG tag (Fig. 4B). As shown in Fig. 4B, no specific fluorescence signal was detected in parental NK92 cells or NK92-EV controls, whereas strong red fluorescence signals were observed on the membrane of NK92 cells transduced with Tan-CAR1, Tan-CAR2, FOLR1-CAR, or PD-L1-CAR. Quantitative analysis showed that the proportion of FLAG-positive (CAR-positive) cells exceeded 70% in all CAR-transduced groups. In addition, CCK-8 assay results indicated that the proliferative capacity of all engineered CAR-NK92 cells was comparable to that of the NK92 control group (Fig. 4D).

Fig. 4.

Analysis of CAR expression and cell proliferation in Tan-CAR NK92 cells. A Genome DNA-based PCR identification. The results confirmed successful integration of Tan-CAR1/2, PD-L1-CAR, and FOLR1-CAR into the respective NK92 cell genomes (lanes 2–5). Band sizes matched those of corresponding positive controls (Tan-CAR1 NK92 plasmid, PD-L1-CAR NK92 plasmid) (lanes 6–7). The empty vector control (lane 1) showed no specific band. B Representative immunofluorescence images of CAR expression on engineered NK92 cells. CARs were detected via FLAG tag (red); nuclei were stained with DAPI (blue). Scale bar, 50 µm. Quantification of mean fluorescence intensity from three independent experiments (n = 3 per group). Data were analyzed by one-way ANOVA with Tukey’s post-hoc test, using the NK92 group as control. C qRT-PCR analysis of CAR transcription levels. High levels of CAR-specific mRNA were detected in all CAR-NK92 cell lines. Data are from three independent experiments (n = 3 per group) and were analyzed by unpaired two-tailed Student’s t test, with the NK92-EV group as the control. D CCK-8 assay. Two-way ANOVA with Tukey’s post-hoc test was used to compare groups at each time point. Data are presented as mean ± SD from three independent experiments (n = 3 per group); n.s.: not significant (P > 0.05)

Cytotoxicity of tan-CAR NK92 cells in vitro

LDH cytotoxicity assay was employed to assess the cytotoxic effects of CAR-NK92 cells at different E/T ratios. Against SKOV3 and OVCAR3 cells, the lytic abilities of the five effector cells were as follows: Tan-CAR2 NK92 > FOLR1-CAR NK92 > PD-L1-CAR NK92 > Tan-CAR1 NK92 > NK92. Among these, Tan-CAR2 NK92 cells demonstrated the strongest specific lysis, which was significantly greater than that of FOLR1-CAR NK92 cells (p < 0.05). Moreover, Tan-CAR2 exhibited significantly superior antitumor activity compared with Tan-CAR1 (p < 0.05). Both FOLR1-CAR NK92 and PD-L1-CAR NK92 also showed potent killing relative to NK92 cells (p < 0.05). In contrast, against A2780 cells, all CAR-NK92 groups displayed similar cytotoxicity with no statistical difference (p > 0.05) (Fig. 5), indicating that CAR-NK92 cells do not induce CAR-mediated specific cytotoxicity toward antigen-negative cells.

Fig. 5.

Cytotoxic killing activity of Tan-CAR NK92 cells against FOLR1-positive OC cells. Lactate dehydrogenase (LDH) cytotoxicity assay detected the lysis rates of SKOV3, OVCAR3, and A2780 by NK92, PD-L1-CAR NK92, FOLR1-CAR NK92, Tan-CAR1/2 NK92 cells under different effector-to-target (E/T) ratios. All the data are presented as means ± SD from three independent experiments (n = 3 per group). Statistical analysis was performed by two-way ANOVA followed by Tukey’s post-hoc test. * indicate significant differences between CAR-NK92 groups and the NK92 control group at each E/T ratio. # indicate significant differences among CAR-NK92 groups specifically at the 8:1 E/T ratio. n.s.: not significant (P > 0.05); *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001; #P < 0.05; ##P < 0.01; ####P < 0.0001

Cytotoxic effects of CAR-NK92 cells against OVCAR3 cells were generally stronger (p < 0.0001 vs. NK92 control) than those against SKOV3 cells at low E/T ratios (1:1, 2:1) (ns or p < 0.01 vs. NK92 control). Tan-CAR1 NK92 displayed differential cytotoxicity toward the two target cell lines: Its lysis of SKOV3 cells was comparable to that of NK92 cells (p > 0.05), whereas its lysis of OVCAR3 cells was more robust (p < 0.05 vs. NK92 control). These differences may be attributed to higher target antigen expression in OVCAR3 cells than in SKOV3 cells (Fig. 2C), suggesting that higher antigen density facilitates more effective scFv–antigen binding and thereby enhances CAR-mediated cytotoxicity. As the E/T ratio increased, tumor lysis rates correspondingly rose. At the maximum E/T ratio of 8:1 used in this study, all groups exhibited their peak level of target cell lysis. In summary, the cytotoxicity of Tan- and single-target CAR NK92 was antigen density- and E/T ratio-dependent.

Activation and degranulation of tan-CAR NK92 Cells

We assessed NK cell activation and degranulation by detecting CD107a expression on the NK cell surface. As reported [21, 22], the activation and degranulation levels of NK cells correlate positively with their cytotoxic function. This also applies to NK92 cells, the specific NK cell line. After co-culture with SKOV3 (Fig. 6A) or OVCAR3 (Fig. 6B) cells, CD107a expression was significantly upregulated on Tan-CAR2 NK92 (p < 0.0001 vs. NK92 control) and FOLR1-CAR NK92 cells (p < 0.0001 vs. NK92 control). Among these, Tan-CAR2 NK92 cells showed significantly higher CD107a expression (p < 0.001 vs. FOLR1-CAR NK92 group) (Fig. 6D). In contrast, upon co-culture with A2780 cells (Fig. 6C), all effector groups showed similarly low levels of CD107a expression with no significant difference (p > 0.05). Tan-CAR1 NK92 cells showed minimal upregulation of CD107a, with no significance (p > 0.05 vs. NK92 control) when co-cultured with SKOV3 or OVCAR3 cells (Fig. 6D), and failed to show any CAR-mediated tumor-killing activity. These results were consistent with the trend observed in the LDH cytotoxicity assay.

Fig. 6.

Activation of Tan-CAR NK92 Cells and Antigen-Associated Degranulation. Flow cytometry analysis of CD107a protein expression in NK92, PD-L1-CAR NK92, FOLR1-CAR NK92, Tan-CAR1 NK92, and Tan-CAR2 NK92 cells after co-culture with SKOV3 (A), OVCAR3 (B), and A2780 (C) cells. D Data statistics on the positive expression rate of CD107a in CAR-NK92 cells. Data are presented as mean ± SD from three independent experiments (n = 3 per group). One-way ANOVA with Tukey’s post-hoc test was used for multiple comparisons. * vs. NK92 control; # among CAR-NK92 groups. n.s.: not significant (P > 0.05); ****P < 0.0001; ###P < 0.001

Mouse xenograft tumor model and tan-CAR NK92 treatment

By subcutaneously injecting SKOV3 cells into BALB/c nu/nu mice, we successfully established tumor xenograft models. Following peritumoral administration of engineered CAR-NK92 cells, we assessed tumor progression by monitoring body weight and measuring tumor dimensions (Fig. 7A). During the initial phase of the study, body weights remained stable in all groups. On day 12 post-treatment, one mouse in the PD-L1-CAR NK92 group exhibited a body weight loss of more than 20% (Fig. 7C); therefore, we terminated the experiment immediately.

Fig. 7.

In vivo cytotoxic activity of Tan-CAR NK92 cells. A Schematic diagram of mouse ovarian cancer xenograft model establishment and treatment/monitoring workflow. B The photograph of transplanted tumors. C Mice body weight changes. D Tumor growth curves during treatment. Tumor volume expressed as relative tumor volume (RTV = Vt/V0, where Vt is tumor volume on day t; V0 is tumor volume on the day of effector cell injection (day 0, measured before the injection)). Data in (C) and (D) were analyzed by two-way ANOVA followed by Tukey’s post-hoc test. E Tumor weights at the end of the experiment. Statistical significance was determined by one-way ANOVA followed by Tukey’s post-hoc test for multiple comparisons. * vs. NK92 control; # among CAR-NK92 groups. All data in (C-E) are presented as means ± SD (n = 6 per group). n.s.: not significant (P > 0.05); **P < 0.01; ****P < 0.0001; ##P < 0.01

The results showed that mice treated with PD-L1-CAR NK92, FOLR1-CAR NK92, and Tan-CAR2 NK92 cells exhibited significantly smaller relative tumor volumes than the NK92 control group (p < 0.05) (Fig. 7B, D). Among them, Tan-CAR2 NK92 cells showed more pronounced inhibition of tumor growth (p < 0.05 vs. FOLR1-CAR NK92 or PD-L1-CAR NK92 cells) (Fig. 7B, E). Notably, mice treated with PD-L1-CAR NK92 cells exhibited significant (> 20%) body weight loss between days 8 and 12 post-injection, along with behavioral changes (hunched posture, reduced activity), suggesting potential toxic side effects.

At the experimental endpoint, histological analysis of xenograft tumors showed heterogeneous PD-L1 expression across all groups. Compared with the pale yellow staining in the NK92 control group, the CAR-NK92 treatment groups exhibited more cells with dark brown staining in the same field of view. Conversely, FOLR1 expression was markedly reduced in the FOLR1-CAR NK92 group and nearly undetectable in the Tan-CAR2 NK92 group (Fig. 8A), demonstrating potent and specific cytotoxicity of CAR-NK92 cells against FOLR1-positive OC cells.

Fig. 8.

Histological analysis of xenograft tumors via HE and IHC staining. A IHC analysis of PD-L1 and FOLR1 expression in xenograft tumors. B Representative HE staining photographs of tumor sections in each group. Red arrows indicate the abnormal apoptotic and necrotic area (nuclear pyknosis and fragmentation, loosely arranged tumor cells and loss of cellular morphology), and black arrows indicate lymphocyte infiltration. C IHC analysis of the CD56 expression on NK92 cells in xenograft tumors. Purple arrows indicate NK92 infiltration. Data are presented as means ± SD (n = 6 per group). * vs. NK92 control; # among CAR-NK92 groups. n.s.: not significant (P > 0.05); **P < 0.01; ****P < 0.0001; ##P < 0.01. Magnification: 10 × (scale bar: 200 µm); Magnification: 40 × (scale bar: 50 µm)

HE staining further confirmed tumor cell death (Fig. 8B). Sections from the PD-L1-CAR NK92, FOLR1-CAR NK92, and Tan-CAR2 NK92 groups revealed typical features of tumor cell death, including loosely arranged tumor cells, nuclear pyknosis and fragmentation, loss of cellular morphology, and pronounced lymphocyte infiltration. Among them, the most extensive infiltration was observed in the Tan-CAR2 NK92 group. In contrast, Tan-CAR1 NK92 and unmodified NK92 groups showed dense tumor growth with only scattered lymphocyte infiltration.

NK92 cell infiltration was detected using its specific marker CD56 (Fig. 8C). The results showed that the number of engineered NK92 cells in tumors was significantly higher in the Tan-CAR2 NK92 group than in the FOLR1-CAR NK92 group (P < 0.01).

To conclude, Tan-CAR2 NK92 and both single-target CAR NK92 showed significant antitumor effects, with Tan-CAR2 NK92 demonstrating consistently greater lymphocytic infiltration. PD-L1 expression in tumors may be induced by targeted immunotherapy, making it a promising antigen for tumor-targeted therapy.

Discussion

CAR-based immune cell therapy has achieved breakthroughs in preclinical studies for solid tumors such as OC. Nevertheless, drawbacks such as the immunosuppressive microenvironment and tumor antigen heterogeneity severely limit the sustained antitumor effects of CAR technology.

PD-L1 is expressed on tumor cells and various stromal components (such as tumor-associated macrophages, myeloid-derived suppressor cells, and cancer-associated fibroblasts), collectively forming a complex immunosuppressive network that suppresses antitumor immunity. Blocking the PD-1/PD-L1 axis is considered to enhance CAR therapy. Currently, PD-1/PD-L1 antibody monotherapy has limited efficacy in OC, with objective response rates below 15% and no survival benefit when combined with platinum-based chemotherapy; it is recommended as first-line treatment only in specific biomarker-selected subgroups. However, emerging evidence indicates that combining the antibodies with targeted therapies alleviates treatment-induced immunosuppression and enhances tumor regression. Thus, although PD-1/PD-L1 antibodies are clinically limited as therapeutics, PD-L1 itself remains a promising antigenic target. Unlike antibodies that transiently block the PD-1/PD-L1 axis, CAR-based approaches enable permanent elimination of PD-L1-positive cells within the tumor microenvironment (TEM).

In this study, we designed and constructed tandem CAR NK92 cells targeting FOLR1 and PD-L1 for the first time to overcome tumor immunosuppression and antigen heterogeneity. Compared with FOLR1-CAR NK92 cells, bispecific CAR-NK92 (Tan-CAR2 NK92) cells displayed significantly superior tumor cell lysis and degranulation against SKOV3 and OVCAR3 (FOLR1^highPD-L1^low) cells and markedly suppressed SKOV3 xenograft tumor growth in mice. Collectively, these data demonstrate that enhanced PD-L1-targeted therapy augments the cytotoxicity of FOLR1-CAR NK92 cells toward OC.

Combining PD-L1 blockade with CAR therapy holds promise for significantly enhancing antitumor efficacy [23, 24]. In this study, Tan-CAR2 NK92 cells showed more potent antitumor effects against FOLR1^highPD-L1^low OC cells than single-target FOLR1-CAR NK92 cells. Notably, PD-L1-CAR NK92 cells exhibited superior antitumor activity against PD-L1-low OC cells both in vitro and in vivo relative to the NK92 control group. We speculate that the possible mechanisms underlying these observations are as follows:

1. Dual-target recognition, synergistic activation and a self-amplifying effect [25]. The anti-PD-L1 scFv in CAR-NK92 cells may efficiently recognize PD-L1 antigen that is either inherently expressed or upregulated on tumor cells. When co-cultured with OC cells, the anti-FOLR1 scFv in the tandem CAR preferentially binds highly expressed FOLR1 on tumor cells, while the anti-PD-L1 scFv recognizes the naturally low level of PD-L1, triggering CAR signaling and initiating direct killing of tumor cells. Meanwhile, this process also induces the release of a large number of cytokines such as IFN-γ, which subsequently upregulates PD-L1 expression on OC cells. The newly synthesized PD-L1 is then recognized by the anti-PD-L1 scFv, creating a positive feedback loop that amplifies CAR-NK92 cell signaling and antitumor activity.

2. Alleviation of the immunosuppressive TME (ITME). Compared with the FOLR1-CAR NK92 treatment group, the Tan-CAR2 NK92 group exhibited greater infiltration of lymphocytes and CAR-NK92 cells. We suggest that the anti-PD-L1 scFv within the CAR competitively binds to PD-L1 on tumor cells, thereby blocking the PD-1/PD-L1 signaling axis between TILs and tumor cells. This relieves immunosuppression, enhances lymphocyte infiltration and cytotoxic activity, and ultimately promotes the endogenous antitumor immune response.

PD-L1 is not only overexpressed in OC but also serves as a key regulator of the ITME. In the above inferences, integrating PD-L1 targeting into CAR-NK92 design has two advantages. On one hand, it broadens the antigen recognition spectrum and alleviates treatment limitations caused by antigen heterogeneity in solid tumors. On the other hand, it blocks the PD-1/PD-L1 signaling axis, reversing immunosuppression and inhibiting tumor immune escape. However, this proposed mechanism requires further experimental validation. Notably, CAR therapy-induced PD-L1 upregulation may further broaden the targetable antigen range, establishing a self-amplifying positive feedback loop of “antigen recognition-cytokine release-antigen re-upregulation.” Given this functional expansion and PD-L1 expression in various normal tissues, there is a potential risk of off-target toxicity. Therefore, during PD-L1-targeted CAR-based immunotherapy, safety should be ensured through strategies such as locoregional administration and routine safety monitoring (e.g., serum biochemistry, inflammatory cytokines).

The efficacy of a CAR is closely related to its structural configurations including scFv affinity, the length and flexibility of the hinge region, and composition of co-stimulatory domains. Critically, the scFv arrangement of a tandem CAR is identified a key factor determining its function [26]. To validate this, we constructed two tandem CARs targeting PD-L1 and FOLR1: Tan-CAR1 (PD-L1-scFv-FOLR1-scFv) and Tan-CAR2 (FOLR1-scFv-PD-L1-scFv). Functional assays revealed that the killing efficacy of Tan-CAR2 was significantly better than that of Tan-CAR1, and the latter one showed even lower potency than single-target CARs. Given that the two CAR constructs share similarities in scFv affinity, hinge region, and co-stimulatory domain composition, we attribute the functional advantage of Tan-CAR2 to the scFv arrangement, which is more conducive to effective binding to the corresponding antigens on tumor cells. We speculate that the difference may stem from the distinct membrane topologies of the two target antigens: FOLR1 is a GPI-anchored cell surface molecule [27, 28], whereas PD-L1 is a type I transmembrane protein with an extracellular domain [29, 30]. This topological difference may allow Tan-CAR2 to achieve more efficient antigen recognition and binding while keeping smoother mechanical force distribution. Validation of the above hypothesis is beyond the scope of this study and warrants further investigation. Consequently, during the design of tandem CARs, scFv arrangement should be considered as a core design consideration.

NK92 is a representative NK cell line. Its advantages include multi-pathway cytotoxicity, low toxicity [31, 32], and low-level expression of inhibitory receptors, making it an ideal effector cell for CAR-based immunotherapy. Furthermore, its MHC-independent target recognition enables rapid early killing. NK92 exerts multidimensional antitumor effects through ATCC, activation receptor-mediated killing, and “missing-self” recognition. Chovatiya N et al. [33] demonstrated that poorly differentiated, highly invasive OC cells fail to upregulate MHC-I, allowing them to evade T cell attack, whereas NK cells can eliminate such tumors via the “missing-self” pathway. Consequently, CAR-engineered NK92 cells represent a particularly promising therapeutic strategy for refractory OC.

The antitumor activity of Tan-CAR NK92 depends not only on antigen expression and E/T ratio but also on the inherent sensitivity of cancer cells to NK cell-mediated killing. According to the results, SKOV3 cells displayed greater resistance to non-specific lysis by NK92 cells than did OVCAR3 or A2780 cells. CD107a measurements of NK cell activation further corroborated this observation. Similarly, Klapdor et al. [34] suggested that SKOV3 cells are more resistant to NK cells than A2780 cells. Therefore, in clinical practice, it is crucial to consider differences in tumor resistance to NK cells between different patients or between different disease stages of the same patient; relying solely on antigen expression levels to predict therapeutic efficacy may be risky. Additionally, NK92 cells exhibit inherent antitumor cytotoxicity against all OC cell lines tested.

To address tumor antigen heterogeneity and the risk of immune escape, various dual-targeting strategies can be employed for CAR-engineered effector cells. These include the simultaneous or sequential infusion of distinct CARs, co-expression of two CARs within a single effector cell, or design of a bispecific CAR comprising two antigen-recognition domains. This study constructed a bispecific CAR that includes both PD-L1 and FOLR1 recognition domains. This single construct can simultaneously target a TAA and the immunosuppressive microenvironment, demonstrating outstanding functional advantages, specifically stronger persistence, cytotoxicity, and activation.

In this study, Tan-CAR2 NK92 cells hold potential as a targeted immunotherapy for FOLR1-positive OC, representing a promising alternative to traditional therapies. However, several limitations exist: 1) We did not identify or establish a suitable dual-positive (FOLR1^high PD-L1^high) OC cell line for Tan-CAR2 functional validation. 2) The in vivo evaluation of Tan-CAR was constrained by a suboptimal effector cell dose (3.5 × 10^6 cells per mouse, single injection), offering no numerical advantage over tumor cells. 3) Mice treated with PD-L1-CAR NK92 cells experienced weight loss (> 20%), potentially due to “on-target, off-tumor” toxicity resulting from basal PD-L1 expression in healthy tissues. Although no significant toxicity was observed in the Tan-CAR2 NK92 group, long-term safety and persistence evaluations of Tan-CAR2 NK92 cells are warranted in future studies. 4) Although the subcutaneous xenograft model used in this study facilitates monitoring of tumor growth and assessment of cellular effects, it cannot fully replicate the immunosuppressive microenvironment of intraperitoneal OC dissemination. Hence, further validation of the key findings using intraperitoneally implanted OC models is warranted. 5) The implanted OC cells did not fully represent the antigenic heterogeneity of patient tumors. 6) Although NK92 cells have been extensively studied in clinical research, their inherent properties restrict widespread application [35, 36]. First, being derived from lymphoma, NK92 cells require pre-infusion irradiation, which severely limits their long-term efficacy. Second, they are highly dependent on exogenous IL-2, whose repeated injections may cause serious adverse reactions. Moreover, due to the lack of CD16, NK92 cells cannot mediate ADCC, a key antitumor mechanism. In contrast, primary NK cells have no tumorigenic risk and exhibit superior safety and prolonged in vivo persistence. Additionally, umbilical cord blood is a common source for CAR-NK cell production, as it can be abundantly provided by public cord blood banks [37, 38]. Importantly, primary CAR-NK cells have achieved landmark progress in clinical research [39–41], validating their therapeutic potential against relapsed or refractory diseases. Therefore, validation of our findings in primary NK cells is essential.

In summary, our results proved that FOLR1 is highly expressed in OC tissues and cell lines, while PD-L1 exhibits inducible overexpression in the context of CAR-mediated therapy. Both FOLR1 and PD-L1 are promising targets for OC-targeted therapy. Meanwhile, tandem FOLR1/PD-L1-CAR NK92 cells demonstrate robust anti-ovarian cancer activity both in vitro and in vivo, suggesting a potential alternative therapeutic strategy for ovarian cancer patients with high or inducible expression of FOLR1 and/or PD-L1.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Jianxin Guo, Jian Han and Xuena Chen contributed to study conception and design. Xuena Chen, Jie Huang, Ge Diao, Min Tian, Yu Yang, Dan Liu and Yunhe Ma performed the experiments, collected experimental samples and analyzed the data. Xuena Chen and Jianxin Guo wrote and revised the manuscript. All authors have read and agreed to the published version of the manuscript.

Funding

The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.

Data Availability

All data supporting the findings of this study are available within the paper and its Supplementary Information. Detailed information regarding overlapping extension PCR and sequence verification is provided in the Supplementary Materials. The cited patent information is provided in the reference list and can be publicly accessed via at the following URL: https://www.lens.org/lens/patent/143-261-701-932-814/frontpage

Declarations

Conflict of interest

The authors declare no competing interests.

Ethical approval

Ethical approval for the use of human subjects was obtained from the Research Ethics Committee of Daping Hospital (Chongqing, China), which granted a waiver of informed consent (2025–414). The animal study was approved by the Ethics Committee of the Army Medical University (AMUWEC20257044).