Improving immune attack on sarcoma by NK cells modified to target EphA2

Sarcoma cell lines A673 (CRL‐1598), RD (CCL‐136), RH30 (CRL‐2061) and K‐562 lymphoblast cells (CCL‐243) were purchased from the American Type Culture Collection (ATCC), while MG63 (ATCC CRL‐1427), SaOS (ATCC HTB‐85) and U‐2 OS (U2OS, HTB‐96) were provided by A/Prof. Orazio Vittorio (Children's Cancer Institute & UNSW). Sarcoma cell lines are maintained in the following reagents purchased from Gibco: Dulbecco's modified eagle medium (DMEM) with 1× GlutaMAX supplement, 10% heat inactivated foetal bovine serum (HI‐FBS), 1% penicillin–streptomycin (PenStrep) and 1% non‐essential amino acids (NEAA). NK cell lines NK92 and Khyg1 cells were provided by Prof. Nicholas Huntington (Monash University) and maintained in Roswell Park Memorial Institute medium (RPMI) media (Gibco) with 1× GlutaMAX supplement, 10% HI‐FBS, 1% PenStrep, 10 mM HEPES and 300 IU/mL hIL‐2, NK MACS media (Miltenyi Biotec, Bergisch Gladbach) containing 10% human AB serum (Sigma‒Aldrich), 300 IU/mL hIL‐2 (Peprotech) and 5 ng/mL hIL‐15 (Peprotech) and RMPI media with 1× GlutaMAX, 10% HI‐FBS, 1% PenStrep and 1× NEAA, respectively. HEK293F cells were purchased from Thermo Fisher Scientific and cultured in Freestyle 293F Expression Media (Gibco) as per manufacturer's protocol. For in vivo tracking, stable green fluorescent protein (GFP)‐luciferase expressing MG63 and RD cell lines (RD‐/MG63‐GFP‐Luc⁺) were generated by transducing the cells with a lentiviral vector encoding firefly luciferase and EGFP—pLentipuro3/TO/V5‐GW/EGFP‐Firefly Luciferase (plasmid #119816, Addgene). GFP‐positive transduced cells were enriched by fluorescence‐activated cell sorting using a FACSAria II instrument (BD Biosciences).

Human primary NK cells were isolated from healthy donor buffy coats obtained from the Australian Red Cross Australia (agreement no. 23‐10QLD‐10). PBMCs were isolated using Ficoll Paque (Cytiva) and Leucosep (Greiner Bio‐One) centrifugation. Primary NK cells were purified from PBMCs using the MojoSort Human NK Cell Isolation Kit (BioLegend) according to manufacturer's protocol. To culture and expand human NK cells, 1 million/mL cells were cultured in complete NK cell culture media consisting of supplemented NK MACS media (Miltenyi Biotec), 5% heat inactivated human AB serum (Sigma‒Aldrich), 500 IU/mL hIL‐2 and 5 ng/mL hIL‐15 in G‐Rex 24 multi‐well cell culture plates (Wilson Wolf) in the presence of NK cell activation beads (Miltenyi Biotec) or in presence of CD86/4‐1BBL/mbIL‐15/mbIL‐21‐transfected K562 feeder cells48 at a 1:1 ratio (NK:feeder) and plates were incubated at 37°C with 5% CO₂. Similarly, mononuclear cells from cord blood (CB) were plated in G‐Rex 24‐well plates (#80192 M—Wilson Wolf) at 2 million cells per well in 6 mL of complete NK MACS medium in presence of K562 feeder cells. Cells were maintained in complete NK cell culture media, with a half media change every 3–4 days and full media change every 7 days until utilised in experiments.

Feeder cells were obtained by transfecting K562 cells (ATCC) at the Genome Engineering Facility (Children's Medical Research Institute) with Sleeping Beauty plasmids encoding CD86, 4‐1BBL, mbIL‐21 and mbIL‐15. Plasmids were kindly gifted by Professor Alexander McLellan (University of Otago).48 The engineered K562 feeder cells were expanded in RPMI media, 10% HI‐FBS and GlutaMAX, irradiated (100 Gy) and cryopreserved in 10% dimethyl sulfoxide (DMSO) and 90% FBS for future utilisation.

EphA2‐CAR constructs were designed to include a T7 promoter at the 5′ end, followed by a single‐chain variable fragment (scFv) specific to human EphA2 (clones 4H5,49 D250 or D2‐1A751), a flexible and commonly used CD8α hinge,52 a CD28‐CD3ζ signalling domain,53 with the inclusion of GFP sequence separated from the CAR construct by a 2A self‐cleaving peptide at the 3′ end. EphA2‐CAR‐GFP constructs were cloned into the plasmid pcDNA3.1 backbone and assembled using NEBuilder HiFi DNA Assembly kit (New England Biolabs [NEB]) according to manufacturer's instructions. Colony clones selected and purified using Wizard Plus SV Minipreps DNA Purification Systems kit (Promega) and all cloning steps were validated by restriction enzyme digestion and sequencing. HiScribe T7 ARCA mRNA Kit (NEB) and Monarch RNA Clean Up Kit (NEB) were used to in vitro transcribe mRNA from linearised DNA template and purify mRNA, respectively. Synthetic codon‐optimised and uridine‐modified EphA2‐CAR (4H5 clone) without GFP sequence and eGFP mRNA sequences were additionally produced by the University of Queensland BASE mRNA Facility. mRNA produced at the facility uses a pBASE vector for in vitro transcription (IVT) of mRNA that features the codon optimised coding sequence flanked by the human alpha globin 5′ UTR and AES‐mtRNR1 3′ UTR sequences, a 120 nt segmented polyA tail with adjacent BsaI for linearisation. IVT mRNA is purified and routinely verified using RNA ScreenTape on the Agilent 2200 TapeStation system (Agilent Technologies). BASE mRNA uses standard rNTPs (ATP, CTP, GTP), with UTP substituted with equimolar concentrations of the modified counterpart for N1‐methylpseudouridine (m1Ψ) or further modified with 10%, 25% or 50% adenosine‐5′‐(α‐thio)‐triphosphate (ATPαS; Jena Bioscience). Capping is performed co‐transcriptionally with the CleanCapAG reagent. For stability testing, mRNA samples with various nucleotide modifications were stored at 4°C or room temperature (RT) for 1 week. Concentration was measured using an Implen NanoPhotometer (Implen). mRNA integrity was measured using an Agilent TapeStation 4200 with the RNA ScreenTape assay (Agilent). Briefly, mRNA samples were diluted to 200 ng/µL, then 1 µL of sample was added to 5 µL of RNA ScreenTape Sample Buffer. Samples were then vortexed for 1 min using the IKA MS3 vortex (IKA) and denatured at 72°C for 3 min before running the RNA ScreenTape assay.

The design, manufacture and formulation of lipid nanoparticle (LNP)‐containing mRNA was performed as per routine protocols based in Moderna Therapeutics formulations,54 used by the UQ BASE mRNA Facility.55 Cholesterol was obtained from Avanti Lipids. SM‐102, 1,2‐distearoyl‐sn‐glycero‐3‐PC (DSPC) and DMG‐PEG 2000 were obtained from Cayman Chemical. mRNA‐LNPs were prepared using the NanoAssemblr Ignite (Precision NanoSystems). mRNA was diluted in .1 M sodium acetate buffer at pH 4.0 (Fisher Bioreagents) to a final concentration of approximately 130 µg/mL. Lipids were prepared to at a total lipid concentration of 10 mg/mL in 200 proof ethanol (Sigma‒Aldrich) at a molar ratio of 50:38.5:10:1.5, SM‐102, cholesterol, DSPC and DMG‐PEG 2000, respectively. LNPs were formulated at a flow rate ratio of 3:1, aqueous to organic phase, and a total flow rate of 12 mL/min. Immediately post‐formulation, LNPs were dialysed overnight at 4°C in 1× tris‐buffered saline (TBS) using a 10 K MWCO Slide‐a‐Lyzer dialysis cassette (Thermo Fisher Scientific). Following dialysis, LNPs were concentrated to 1 mL using a 10 K MWCO Amicon Ultra‐15 centrifugal filter (Sigma‒Aldrich). LNPs were filtered using a .22 µM filter (FilterBio) and 10% sucrose was added before storage at ‒20°C. Size, polydispersity and zeta potential analysis was completed using a Zetasizer Ultra (Malvern Panalytical) on an LNP sample diluted 1:20 in UltraPure water (Thermo Fisher Scientific). Encapsulation efficiency and encapsulated concentration was determined using the Quant‐it RiboGreen RNA assay kit (Invitrogen Life Technologies).

For electroporation experiments, NK cell media was refreshed 24 h before transfection. NK cells from culture were resuspended in MaxCyte electroporation buffer (MaxCyte) containing RNAse inhibitor (Promega), and CAR mRNA (15 pmol/million cells or 200 µg/mL). For in vivo studies, controls with no mRNA added to the cells during transfections were used as the mock group. The MaxCyte STx system (MaxCyte) was used to electroporate cells using either NK‐2 or NK‐3 programs. Following electroporation, the cells were recovered for 20 min in a 37°C cell culture incubator in warm complete media. NK cells were then harvested at 24–72 h post‐electroporation for setting up cytotoxicity assays, flow cytometry analysis and for i.v. injections. For LNP‐based expression, NK cells were cultured in media containing LNP carrying mRNA at the same concentration used for electroporation experiments (15 pmol/million cells or 200 µg/mL).

For flow analysis, EphA2‐CAR‐GFP or GFP expression was determined 24–48 h post‐electroporation by staining 1 × 10⁵ cells with 1:1000 dilution of Zombie Aqua (BioLegend) for 10 min in phosphate‐buffered saline (PBS), followed by washing with 1× flow buffer (1× PBS containing 2% FBS and 2 mM ethylenediaminetetraacetic acid (EDTA)) once before staining with or without 1:50 dilution of Alexa Fluor 647‐conjugated anti‐G4S linker (E7O2V) rabbit monoclonal antibody (Cell Signaling Technology, Inc.) for 45 min at RT. For detection using Protein L, cells were stained either with .2 ng/µL of biotinylated Protein L (Genscript) for 45 min followed by a 1:400 dilution of V450 Streptavidin (BioLegend) for 45 min or 1:50 dilution of Protein L‐PE (Acrobiosystems) for 30 min at RT. For detection using goat anti‐human IgG‐PE (Jackson ImmunoResearch Laboratory), cells were stained with a 1:20 dilution of goat anti‐human IgG‐PE for 30 min at RT and washed twice with 1× flow buffer before flow analysis. For cell surface expression of EphA2 on paediatric sarcoma cell lines, the cells were washed in 1× flow buffer at 4°C before blocking with human Fc Block (1:100) for 15 min. Cells were then incubated with anti‐EphA2 antibody (phycoerythrin [PE] conjugate) clone SHM16 (BioLegend) for 1 h at RT and washed twice before flow analysis. For detection of functional activation in NK cells, electroporated NK cells were rested for 1 h at 37°C in complete NK MACS media containing 5 ng/mL of rhIL‐15. NK cells were then incubated for another 4 h in a cocktail containing 1:1000 dilution of Monensin, 1:1000 dilution of Brefeldin A with or without 1:500 dilution of Cell Activation Cocktail (BioLegend) and 1:100 dilution of anti‐human CD107a (BioLegend) in NK MACS media containing rhIL‐15 (5 ng/mL). Cells were stained for viability using Live/Dead Fixable Near‐IR (Invitrogen Life Technologies) and intracellular interferon‐gamma (IFN‐γ; BioLegend) and granzyme B (Becton Dickinson) production. All flow experiments were acquired on a BD LSRFortessa X‐20 (Becton Dickinson) or a CytoFLEX V2‐B5‐R3 (Beckman Coulter) flow cytometers and analysed with either FlowJo v10.8.1 (TreeStar) or FCS Express 7 Research Edition (De Novo Software) programs in a classical supervised manner (exclusion of debris/dead cells, gating of singlet, analysis of expression for markers) unless otherwise stated. For microscopy analysis of GFP expression after electroporation of EphA2‐CAR‐GFP or eGFP mRNA, cells were visualised using the Olympus IX73 fluorescent microscope and images were taken using the Olympus cellSens imaging software (Olympus). Images were processed using ImageJ (National Institutes of Health) system software. For analysis of GFP expression in lipofected HEK293F cells, flow cytometry was carried out on the Cytoflex V2‐B5‐R3 (Beckman Coulter) flow cytometer with 2–10 × 10⁴ single, viable cells/sample.

Calcein acetoxymethyl (AM) release‐based killing assay was adapted as previously described.56, 57, 58 In vitro Calcein AM cytotoxicity assays were performed to evaluate the cytotoxic capacity of the CAR‐modified NK cells over 4 h. Target cells were resuspended in NK cell cytotoxicity medium (phenol red‐free RPMI medium [Gibco] with 1 mM sodium pyruvate [Gibco], 1× NEAA [Gibco], 1× GlutaMAX [Gibco], 55 µM 2‐mercaptoethanol [Gibco] and 1% HEPES [Gibco]), at a final concentration of 10⁶ cells/mL and incubated with 15 µM Calcein‐AM (Invitrogen Life Technologies) for 30 min at 37°C with occasional shaking. Stained target cells were then washed twice with 10 mL PBS. NK cells (effector cells) were then serial diluted in NK cell cytotoxicity media (+10% FBS for NK92 cells or 5% AB serum for peripheral blood (PB)‐ or cord‐blood‐derived NK cells) in a 96‐well U‐bottom plate in various effector‐to‐target (E:T) ratios in triplicates. Target cells were then added to the plate at a concentration of 3 × 10³ cells per well and the plate was incubated at 37°C for 4 h. To determine the maximum and spontaneous Calcein release values, target cells alone were incubated with either NK cell cytotoxicity media containing 5% Triton X‐100 or NK cell cytotoxicity media alone, respectively. After 4 h, 100 µL supernatant was transferred to a 96‐well white flat bottom plate (Nunc, Thomas Scientific) and fluorescence was read using a CLARIOstar plate reader (BMG Labtech) (excitation 483‐10, emission 530‐10). Specific target lysis was calculated using the following formula: [(NK cell‐induced Calcein AM release − spontaneous Calcein AM release)/(maximum Calcein AM release − spontaneous Calcein AM release) × 100].

RNA‐seq expression data and clinical characteristics for the primary tumour samples collected at the diagnosis stage of paediatric sarcoma patients were collected from the Therapeutically Applicable Research to Generate Effective Treatments (TARGET) database (Target‐OS database; https://www↗.cancer.gov/ccg/research/genome‐sequencing/target/usingtarget‐data; accession study: phs000468.v18.p7) and ArrayExpress database (https://www.ebi.ac.uk/biostudies/arrayexpress↗; accession ID E‐TABM‐1202 RMS dataset).59 The optimal cut‐off for the single factor in the survival analysis was determined using the data‐driven approach using the Conditional Inference Tree (CTREE) algorithm from the ‘party’ package (version 1.3.17) in R to create decision tree that optimises the separation between high and low groups for survival analysis.60 The survival analysis was performed using the ‘survival’ package (version 3.7.0), and Kaplan‒Meier (KM) curves were plotted using the ‘survminer’ package (version 0.4.9). Gene set scoring was performed to calculate glutathione metabolism score using the singscore approach.61 In short, genes are sorted in ascending order based on their transcript abundance. NK cell infiltration scoring method was obtained from ref.62 TARGET OS datasets are available without restrictions on their use in publications since 2019. We further processed single‐cell (sc)RNA‐seq data of infiltrating immune cells in OS from,63, 64 and healthy human femoral head65 as control, accessible from GEO database under the accession number GSE147390↗. Apart from the originally published quality control of these datasets, no additional quality controls were applied. The scRNA‐seq data were integrated using bbknn.66 Selected genes associated with NK cells exhaustion were visualised as violin plots using scanpy's67 sc.pl.violin function.

To support human NK cell survival and function, a human IL‐7 (hIL‐7) and IL‐15 (hL‐15) double knock‐in (KI) immunodeficient NOD SCID gamma (NSG) mice68, 69 backcrossed with an HLA‐A2 NSG mice70 were used (NSG/A2/hIL‐7/hIL‐15‐KI). Mice were bred and maintained onsite at the Translational Research Institute Biological Research Facility and housed under specific pathogen‐free conditions. Mice aged 6–8 weeks were used for stable engraftment of the human OS or RMS cell lines. All experiments were conducted following the animal ethics guidelines provided by the National Health and Medical Research Council of Australia and approved by the University of Queensland—Health Sciences Animal Ethics Committee (ethics number 2021/AE000294↗).

To assess therapeutic efficacy of EphA2‐CAR‐NK cells against sarcoma in vivo, RMS (RD) and OS (MG63) cell lines were engrafted into the NSG/A2/hIL‐7/hIL‐15‐KI mice. For orthotopic RMS tumour inoculation, male and female mice were restrained with the hind limb in flexion and 2 × 10⁶ RD GFP⁺ Luc⁺ cell suspension were inoculated intra‐muscularly using a 30G insulin syringe needle, as described previously,71 after buprenorphine subcutaneous (s.c.) injection for analgesia (10 µg per 10 g of body weight). Mice were either left untreated (PBS only) or intravenously (i.v.) treated with 1 × 10⁶ mock or EphA2‐CAR transfected NK92 cells or primary NK cells when indicated. The mice lungs were inspected for metastasis ex vivo using the in vivo imaging system (IVIS) optical imager as previously described.72 For the s.c. OS model, 1 × 10⁶ MG63 GFP⁺ Luc⁺ cells were injected s.c. in the right flank of male mice. After 250 cGy of whole body γ‐irradiation on day ‒1, mice were treated with 3–5 × 10⁶ cells EphA2‐CAR mRNA or no mRNA (mock group) NK cells resuspended in PBS, or with PBS only (untreated) as indicated. Tumour engraftment and progression were evaluated by external caliper measurements of s.c tumours greatest longitudinal (length) and transverse (width) diameters 2–3 times a week, and on weekly IVIS imaging beginning 10 min after intraperitoneal injection of an aqueous solution of d‐luciferin potassium salt (150 mg/kg). Photons emitted from luciferase expressing cells were quantified using the Living Image 3.0 software program. Tumour volume was calculated by the ellipsoidal formula: V = 1/2 (length × width²). Mice were regularly monitored and sacrificed once tumours reached a size of 1 cm³ or if reaching a preestablished animal ethics wellbeing score of 3 in any category or a cumulative score of 15, and an overall survival was calculated as mice were sacrificed according to this criterion.

All other results are presented as mean ± SE. KM plots were used to analyse OS and RMS datasets from the TARGET program and E‐TABM‐1202 RMS dataset. For all other data, unpaired t‐test or Kruskal‒Wallis test was used to assess differences between two groups, one‐way or two‐way ANOVA and Tukey's multiple comparison test between two or more groups, and Gerhan‒Breslow‒Wilcoxon test for survival analysis (GraphPad Prism 10 Software). Statistical analysis in scRNAseq datasets (Wilcoxon rank sum test) of marker gene expression difference between NK cells was performed using scanpy's sc.tl.rank_genes_group function, where Benjamini‒Hochberg73 adjusted p < .05 was considered statistically significant.

To validate the prognostic impact of NK cell infiltration and EphA2 expression in sarcoma patients, we utilised data from the TARGET‐OS program and transcriptional data obtained from the E‐TABM‐1202 dataset59 for RMS. These analyses included 84 OS patients with available gene expression profiles and clinical outcome data as well as 101 RMS patient samples with transcriptional gene expression profiles stratified by PAX3/FOXO1 Fusion Gene Status. Our analyses indicate that higher EphA2 expression are associated with significantly worse 5‐year survival rates in OS (Figure 1A) and RMS patients (Figure 1B) as compared to patients with lower EphA2 expression, suggesting that this receptor is a promising target for therapeutic intervention in these cancers. We further explored NK cell infiltration using a curated scoring method,62 and observed that OS patients with higher NK cell scores also have the best survival rates (Figure 1C). This led us to further analyse intratumoural NK cells using scRNAseq datasets, where we compared NK cells infiltrating OS and healthy femoral head as control (Figure 1D) and verified that OS‐infiltrating NK cells had elevated expression of the transcriptional factor TOX, and inhibitory markers such as CD96, KLRG1, KLRC1 and LAG3 as compared to control bone‐infiltrating cells (Figure 1E), indicating an exhausted phenotype. This exhaustion could potentially be reversed by equipping tumour‐infiltrating NK cells with enhancement‐of‐function genes such as a EphA2‐CAR and underscore the potential for developing a novel immunotherapeutic strategy for paediatric sarcomas.

Numerous studies have highlighted EphA2's role in tumour cell proliferation, migration and survival, making it a prime candidate for targeted therapy in solid tumours. The development of anti‐EphA2 humanised antibodies has led to the generation of multiple antibody clones, each designed with slightly different mode of action.74 From the array of available clones, we selected three specific anti‐EphA2 clones—4H5, D2 and D2‐1A7 due to availability of sequences in literature. Notably, 4H5 and D2‐1A7 clones were also previously demonstrated to have high specificity against EphA2 in preclinical studies utilising CAR‐T cells42, 46, 75 and antibody‒drug conjugates.51 We designed a second‐generation CAR sequence consisting of an scFv sequence derived from the clones selected, linked with a flexible CD8 hinge and CD28 and CD3ζ intracellular signaling domains, followed by a P2A sequence and a GFP sequence (EphA2‐CAR‐GFP) (Figure 2A). To test the delivery of mRNA into NK cells, we generated mRNA of the EphA2‐CAR‐GFP sequence using IVT and transfected it into an NK cell line Khgy1 using the MaxCyte STx super‐electroporator equipment and its respective processing assemblies. EphA2‐CAR‐GFP mRNA generated from all three clones of 4H5, D2 and D2‐1A7 could all be successfully expressed in Khyg1 NK cells (Figure 2B). To validate whether each of the CAR clones can be expressed on the cell surface, we tested various commonly used CAR detection methods. Using Protein L, goat anti‐IgG (H+L) and the newly available anti‐G4S linker antibodies and flow cytometry, we observed different sensitivities of each CAR detection method for the various clones when expressed in Khyg1 cells (Figure 2C) or in NK92 and primary human NK cells (Figure 2D). In our hands, we further observed varying levels of non‐specific staining of mock transfected primary human NK cells using protein L and goat anti‐IgG (H+L) antibody (data not shown). In contrast, little to no non‐specific staining were observed in primary human NK cells using the anti‐G4S linker detection antibody by flow cytometry. We also studied different mRNA deliveries by comparing the MaxCyte STx‐based electroporation with mRNA‐LNP delivery (Figure S1). While the LNP approach effectively delivered mRNA and facilitated CAR expression, cells transfected using the MaxCyte STx system demonstrated significantly higher expression levels. Consequently, this method was selected for the subsequent experiments in this study.

We validated EphA2 expression in sarcoma cell lines available in our laboratory (Figure 3A) and tested EphA2‐CAR‐NK cells against these cell lines using a Calcein AM cytotoxicity assay. Cytotoxicity assays were carried out using NK92 cells due to their increased functional activity and clinical relevance as compared to Khyg1 cells (Figure S2A). A comparison of the different clones used to generate EphA2‐CAR mRNA showed that while all three EphA2‐CAR constructs expressed in NK92 cells were highly sensitive against the OS cell line MG63 (Figure 3B), only EphA2‐CAR sequence generated from the 4H5 clone could significantly enhance killing activity of NK92 cells against another OS cell line, U‐2 OS (Figure 3B). These observations indicate that among the three clones tested, EphA2‐CAR generated from the scFv of the 4H5 antibody may have the highest specificity against EphA2 expressed in OS cells. To further validate the efficacy of the EphA2‐CAR generated from the 4H5 clone against other sarcoma cell lines, we transfected NK92 cells with EphA2‐CAR mRNA at a lower and MaxCyte‐recommended concentration of 200 µg/mL, as we were observing higher incidences of cell death after electroporation with 15 pmol of mRNA delivered per million NK cells (data not shown). Here, we observed that EphA2‐CAR‐NK92 cells retained enhanced specificity against OS cell lines MG63 and U‐2 OS, as well as SaOS (Figure 3C). EphA2‐CAR‐NK92 cells also had enhanced killing activity against an EWS cell line A673 and two RMS cell lines, RD and RH30 as compared to mock transfected NK92 cells in vitro. The specificity of the EphA2‐CAR‐NK cells was validated in EphA2‐knockout (KO) RD and RH30 cell lines, where little killing of EphA2‐KO cells was observed (Figure 3D). We further show that EphA2‐CAR could also be effectively delivered into freshly isolated PB‐derived primary human NK cells, which enhanced their killing specificities against EphA2‐positive RD and RH30 cell lines in vitro (Figure S2B). In summary, these results indicate the potential of developing an EphA2‐CAR mRNA sequence based on the 4H5 clone for therapeutic use against sarcoma.

The accelerated development of a COVID‐19 vaccine has allowed for mRNA technologies to progress at unprecedented speed and has shed light on the various potential modifications that could be made at the mRNA level to greatly enhance mRNA stability and reduce innate immunogenicity for therapeutic uses. This was particularly relevant in both COVID‐19 vaccines produced by BioNTech and Moderna,76 of which efficacy were shown to be attributed to the incorporation of modified ribonucleotide pseudouridine (Ψ) or N¹‐pseudomethyluridine (m1Ψ). We have validated the efficacy of delivering and generating EphA2‐CAR mRNA with m1ψ modification to NK92 and primary human NK cells, demonstrating enhanced target specificity against the EphA2 antigen in sarcoma cells. In Figure 4A, we confirm that the use of m1ψ‐modified mRNA (green) does not induce increased immunogenic responses in primary NK cells, as indicated by the comparable levels of key activation markers (granzyme B, CD107a and IFN‐γ) between m1ψ mRNA‐treated cells and control conditions (UTP 100%, ATP; red). The flow cytometry profiles appear similar and quantitative analysis revealed no statistically significant differences between the conditions (p > .05), indicating that the m1ψ‐modified mRNA does not elicit an exaggerated immune activation response (Figure 4A). Additionally, this m1ψ modification further enhances protein expression in vitro, as compared to IVT protocols incorporating uridine (data not shown), confirming the suitability of modified mRNA for electroporation into NK cells for transient CAR engineering.

We also tested whether other chemical modifications could be applied to uridine or m1ψ‐modified mRNA for further enhancement of stability of EphA2‐targeting CAR‐NK cells, such as the incorporation of ATPαS in the IVT reaction which introduces phosphorothioate bonds into the mRNA and increases protein yield and reduces RNase‐induced degradation.77, 78 We first tested the stability of different mRNA formulations such as mRNA chemical compositions incorporating normal uridine, methoxyuridine, m1Ψ, as well as m1Ψ with varying concentrations of ATPαS. All mRNA produced with chemical modifications were found to be stable when stored either at 4°C or RT by achieving satisfactory RNA Integrity Number and concentrations at all conditions (Figure S4). As a starting point, eGFP m1Ψ mRNA were first synthesised containing either 0% or 10% ATPαS. Our results demonstrate that incorporation of 10% ATPαS into eGFP m1Ψ mRNA significantly increases eGFP intensity across 168 h post‐lipofection in HEK293F cells in vitro (Figure 4B). We then synthesised EphA2‐CAR m1ψ mRNA containing 0% or 10% ATPαS and demonstrate a similar trend to eGFP in HEK293F cells across 96 h post‐transfection in NK92 cells (Figures 4C and S3), indicating increased mRNA stability and therefore enhanced and prolonged protein expression with the incorporation of 10% ATPαS. We then extended our comparisons of EphA2‐CAR mRNA transfection (Figure S5) across different sources of clinically relevant NK cells,79, 80 including PB‐sorted NK cells, the NK92 cell line, and CB‐derived NK cells. These comparisons utilised MaxCyte STx‐based electroporation with various mRNA chemical compositions incorporating the following: normal uridine, methoxyuridine, m1Ψ, as well as m1Ψ with 10%, 25%, or 50% ATPαS. For comparison, LNP‐based transfection using normal uridine mRNA was also tested. The LNP approach consistently demonstrated lower transfection efficiency compared to MaxCyte STx‐based electroporation in PB and CB NK cells. Notably, LNPs failed to transfect NK92 cells, whereas all tested chemical modifications and normal uridine mRNA achieved efficient EphA2‐CAR expression in NK cells using MaxCyte STx (Figure S5A). Furthermore, cytotoxicity assays were performed on PB and CB NK cells, as well as NK92 cells, following transfection. These assays revealed no significant differences in the enhancement of killing capacity against the target sarcoma cell line between the various mRNA chemical modifications (Figure S5B) in each cell type. In a time point‐kinetic experiment, we observed optimal CAR expression after transfection with m1Ψ mRNA at the 24‐h mark (Figure S6). Notably, NK92 cells equipped with EphA2‐CAR translated using m1Ψ + 10% ATPαS had higher killing activity against RD cells, albeit not statistically significant, as compared to EphA2‐CAR‐NK92 cells generated using m1Ψ mRNA 72 h post‐electroporation (Figure 4D). These observations indicate that ATPαS modifications, among other possible modifications, are suitable to generate more stable expression of CARs in NK cells for future therapeutic purposes against sarcoma.

Finally, we evaluated the anti‐tumour efficacy of EphA2‐CAR‐NK cells in distinct in vivo sarcoma models. In an RMS model, where RD cells were orthotopically inoculated into mice (Figure 5A), i.v. weekly administration of EphA2‐CAR‐NK92 or fortnightly pNK cells significantly suppressed local tumour progression (Figure 5B‒D). In the same setting, a subsequent experiment was performed to analyse metastatic burden in the lungs, which was observed to be reduced in mice treated with EphA2‐CAR‐NK cells (Figure 5E). In a separate OS model using MG63 cells (s.c. inoculated) and PB‐derived EphA2‐CAR‐NK cells for treatment (Figure 5F), primary tumour growth was markedly inhibited in the EphA2‐CAR‐NK cell group (Figure 5G), accompanied by a notable survival benefit (Figure 5H). These findings demonstrate that EphA2‐CAR expression, in both NK92 cells and primary blood‐derived NK cells, enhances their anti‐tumour activity and survival outcomes in different sarcoma models.

The data that support the findings of this study are openly available in the TARGET‐OS database at https://www.cancer.gov/ccg/research/genome‐sequencing/target/usingtarget‐data↗ (accession study phs000468.v18.p7), ArrayExpress database at https://www.ebi.ac.uk/biostudies/arrayexpress↗ (accession ID E‐TABM‐1202 RMS dataset, Refs.63, 64 and GSE147390↗).