Umbilical cord blood (CB) samples were collected from donors at AP-HP Hôpital Saint-Louis (Unité de Thérapie Cellulaire, CRB-Banque de Sang de Cordon, Paris, France) after written informed consent and in compliance with ethical guidelines. Enriched CB CD34+ cells for specific experiments were purchased from Lymphobank (Besançon, France). Peripheral blood from healthy adult donors was obtained with written informed consent from the French Blood Transfusion Institute (Etablissement Français du Sang) (Paris, France).

Cord blood mononuclear cells (MNCs) were isolated by density gradient centrifugation using Lymphocyte Separation Medium (LSM) (Eurobio Scientific, Les Ulis, France). CD34⁺ cells were then enriched immunomagnetically from MNCs using Indirect human CD34⁺ cell isolation kit (Miltenyi Biotech, Bergisch Gladbach, Germany) following manufacturers’ instructions. The purity of isolated CD34⁺ cells exceeded 90%.

Peripheral blood mononuclear cells (PBMCs) were isolated from adult healthy donor peripheral blood using density gradient centrifugation with LSM. PB-NK cells were then immunoselected negatively from PBMCs in a magnetic separator using NK cell isolation kit (Miltenyi Biotech, Bergisch Gladbach, Germany) according to manufacturers’ instructions. The purity of isolated CD3^-CD56⁺ cells exceeded 90%.

K562 myelogenous leukemia cell line was purchased from ATCC (Manassas, VA, USA). K562-GFP/Luc cell line was purchased from Creative Biogene (CSC-RR0374, New York, NY, USA). Both cell lines were cultured in Roswell Park Memorial Institute 1640 Medium, with GlutaMAX™ Supplement (RPMI) (Gibco, Courtaboeuf, France) containing 50 µg/ml gentamycin and 10% Fetal Bovine Serum (FBS Gibco) (Gibco, Courtaboeuf, France). Puromycin (0.8 µg/mL) (InvivoGen, Toulouse, France) was added for the selection of stably transduced K562-GFP/Luc cells. They were passaged every 2-3 days.

Isolated PB-NK cells were cultured and expanded for 10 to 20 days in NK MACS medium (Miltenyi Biotech, Bergisch Gladbach, Germany) supplemented with 1% NK MACS Supplement, 5% human serum albumin (HSA) (BioWest, Nuaillé, France), 50 µg/ml of gentamycin (Gibco, Courtaboeuf, France) and human cytokines: 20 ng/mL hIL-15 (PeproTech, Neuilly-sur-Seine, France) and 500 IU/mL recombinant hIL-2 (Proleukin, Novartis, Liverpool, UK), following manufacturers’ instructions.

CB CD34⁺ HSPCs were preactivated overnight on Fc-hDLL4 fusion protein and RetroNectin^®-coated wells in X-vivo 20 medium (Lonza, Walkersville, MD, USA) containing the same cytokines (hSCF, hFLT3-L, hTPO, hIL-7, hTNFα) described above (hDLL4 culture step). The cells were then transduced under the preactivation condition with a lentiviral vector encoding an anti-CD19 chimeric antigen receptor (CAR) (FMC63) (19-scr, Flash Therapeutics, Toulouse, France) at a multiplicity of infection (MOI) of 100, in presence of 4 µg/ml protamine sulfate (Sigma-Aldrich, St. Louis, MO, USA). After 6 hours of transduction, the cells were washed with α-MEM (Gibco, Courtaboeuf, France) supplemented with 20% FBS HyClone and cultured for a total of 7 days in Fc-hDLL4 culture conditions to produce transduced lymphoid progenitors.

The antibodies listed in Table 1 were used for flow cytometry (FC) analyses. For surface staining, cells were incubated with 7-AAD (Miltenyi Biotech, Bergisch Gladbach, Germany) and the appropriate antibodies for 15 to 45 min at 4°C, washed and resuspended in FACS buffer (0.5% BSA Fraction V (EuroBio Scientific, Les Ulis, France) and 2mM EDTA (ThermoFisher Scientific, Dardilly, France) in 1X PBS (Eurobio Scientific, Les Ulis, France). For intracellular stainings, cells were first surface stained with Viobility™ Fixable Dye (Miltenyi Biotech, Bergisch Gladbach, Germany) and the appropriate antibodies for 30min in ice and washed with FACS buffer. The surface-stained cells were fixed and permeabilized using either BD Cytofix/Cytoperm Fixation/Permeabilization Kit (BD Biosciences, Belgium) or eBioscience Foxp3/Transcription Factor Staining Buffer Set (Thermo Fisher Scientific, Dardilly, France) according to manufacturer’s instructions and stained with the appropriate antibodies for 30min in ice, and then washed and resuspended in FACS buffer. Data acquisition was performed using Novocyte^® flow cytometer (Agilent Technologies, Santa Clara, CA, USA), and the acquired data were analyzed using Kaluza Analysis Software (version 2.1, Beckman Coulter, Krefeld, Germany). All gatings were performed on live cells which were determined by exclusion of 7-AAD or Viobility dyes.

ProT-NK D14, ProT-NK D21 or expanded PB-NK cells were stimulated with K562 cells by co-incubating them at a ratio of 1:2 (NK to K562) for 6 hours in RPMI supplemented with 10% FBS HyClone inside an incubator at 37°C, 5% CO₂. CD107a antibody (Miltenyi Biotech, Bergisch Gladbach, Germany) was added at the beginning of co-incubation. After 1 hour of incubation, protein transport inhibitors BD GolgiPlug and GolgiStop (BD Biosciences, Belgium) were added and further incubated for 5 hours. Cells were analyzed by FC after a total of 6 hours of co-incubation.

FC-based cytotoxicity assay was performed using K562 cells as target cells. Target cells were, labelled with CellTrace Violet (CTV) dye (Invitrogen, ThermoFisher Scientific, Dardilly, France) according to manufacturer’s instructions, to distinguish them (CTV+) from effector NK cells (CTV-). Labelled target cells were incubated with effector NK cells (ProT-NK D14, ProT-NK D21 or expanded PB-NK) at different effector-to-target ratios in RPMI supplemented with 10% FBS HyClone and 100 IU/mL of recombinant hIL-2 (Proleukin, Novartis, Liverpool, UK) for 5 hours at 37°C, 5% CO₂ inside an incubator. Cells were then stained with 7-AAD and analyzed by FC. The specific killings by effector NK cells were calculated as:

Mass cytometry analysis was performed using specifically designed antibody panels for phenotypic ( Supplementary Table S1 ) and functional analyses ( Supplementary Table S2 ). Metal isotope labelled antibodies were purchased from Fluidigm (San Francisco, CA, USA). Antibodies listed in Supplementary Table S3 were labelled manually using Maxpar X8 Antibody Labelling Kit for Lanthanides and Maxpar MCP9 Antibody Labelling Kit for Cadmium (Fluidigm, San Francisco, CA, USA), following manufacturer’s instructions. For surface staining, cells were incubated with Cisplatine Cell-IDTM (Fluidigm, San Francisco, CA, USA) at 2.5μM for 5 minutes at room temperature (RT) to label dead cells. and the appropriate antibodies ( Supplementary Tables S1, S2 ) for 30 min at 4°C, washed and resuspended in FACS buffer and fixed with 1 mL of staining solution containing 1.6% paraformaldehyde (Sigma-Aldrich, St. Louis, MO, USA) for 10 minutes at 4°C. For intracellular stainings, cells were first incubated with Cisplatine Cell-IDTM (Fluidigm, San Francisco, CA, USA) as described above and surface-stained with the appropriate antibodies for 30 min in ice and washed with FACS buffer. The surface-stained cells were fixed and permeabilized using Foxp3/Transcription Factor Staining Buffer Kit (Tonbo Biosciences, San Diego, CA, USA) and stained with the appropriate intracellular antibodies for 1 hr at 4°C, and then washed and resuspended in FACS buffer. Following extracellular or intracellular staining, the cells were washed and resuspended in 1mL of Fix and Perm Buffer containing 1:4000 of Iridium intercalator (Fluidigm, San Francisco, CA, USA) and incubated overnight at 4°C. The stained cells were stored at -80°C until analysis. The data was acquired on Helios mass cytometer with CyTOF software version 6.7.1014 (Fluidigm, San Francisco, CA, USA) at the Cytometry Platform of La Pitié-Salpétrière Hospital. Mass cytometry standard files were normalized using CyTOF Software v. 6.7.1014. To identify cell clusters from CyTOF data, either Uniform Manifold Approximation and Projection [UMAP (42)] or cluster identification [FlowSOM (43)] were run. For both analyses, samples were run on equal numbers of events per sample. To visualize each cell on a two-dimensional map, a UMAP algorithm was applied with the following setting: 200 epochs, minimum distance of 0.4, learning rate of 1, number of nearest neighbors of 15. Clustered heatmap used Euclidian distance and standard normalization in Omiq. To automatically identify differing cell clusters among groups, FlowSOM algorithm was run with 7 nearest neighbors (k = 7). The distinct clusters were manually annotated based on the selected markers: CD56, CD161, CD94, NKG2D, CD57, CD16, KIRs, NCRs, and CXCR4.

NOD.Cg-Prkdcscid Il2rgtm1Wjl Tg(IL-15)1Sz/SzJ (NSG-Tg(hIL-15) mice were purchased from Jackson Laboratories and bred in-house at LEAT Institut Necker Enfants Malades animal facility (registration number A75-15-34). The studies conducted adhered to French ethical regulations and all the procedures performed were in accordance with protocols approved by the “Services Vétérinaires de la Préfecture de Police de Paris” (Veterinary Services of the Paris Police Department) and the “Comité d’Ethique en matière d’Expérimentation Animale Paris Descartes” (Paris Descartes Animal Experimentation Ethics Committee) (CEEA-034) under the number APAFIS#38814- 202202011338747_v8, Université Paris Descartes, Paris, France.

NSG-Tg(hIL-15) mice (8 -12 weeks old) were conditioned with a low dose of busulfan (30 mg/kg), administered intraperitoneally in 2 split doses of 15 mg/kg in two consecutive days at an interval of 24h to improve human cell engraftment. One day post-conditioning, mice were anesthetized with isoflurane (Vetflurane, Virbac, Carros, France), and injected intravenously with 20 x 10⁶ of either PB-NK, ProT-NK D14 or ProT-NK D21 cells through retro-orbital plexus. Bone marrow, liver, spleen and blood were analyzed using FC 5 and 9 days post-injection to assess NK cell engraftment.

To assess the in vivo anti-tumor activity of ProT-NK cells, human acute myeloid leukemia (AML) tumor xenograft mouse model was established in NSG-Tg(hIL-15) mice (8 -12 weeks old). Mice were conditioned and anesthetized as described in the previous section. One day after conditioning, they were injected subcutaneously with K562-Luc cells either alone (0.5 x 10⁶ cells/mouse) or mixed with PB-NK or ProT-NK D14 cells (20 x 10⁶ cells/mouse) under the skin overlying the lower back. Tumor progression was monitored via bioluminescent imaging (BLI) every 3-4 days for 50 days using IVIS Spectrum CT (Revvity, Waltham, MA, USA). D-luciferin (150 mg/kg, Revvity, Waltham, MA, USA) was intraperitoneally injected, and images were captured under isoflurane anesthesia within 20 min of injection. Data were analyzed using Living Image software (IVIS Imaging Systems), with BLI and X-ray superimposition signals quantified after 3D tomography. A region of interest was determined by delineating the bioluminescent zone, and the absolute number of cells was obtained using in vitro calibration with corresponding K562-GFP/Luc cells. Bioluminescence data is expressed as photons/s. Mice were sacrificed by cervical dislocation when tumors reached 1.7 cm in diameter or if specific criteria were met (e.g. severe weight loss, poor coat and skin condition, static activity or paraplegia).

Mice blood samples were collected in EDTA tubes by orbital plexus bleeding with capillary tubes from anesthetized mice. The mice were then sacrificed by cervical dislocation and right femurs, spleen and liver were collected and processed for FC analysis. Bone marrow cells were flushed out of femur bones with 25G needles and 10mL syringes in RPMI supplemented with 10% FBS Gibco and filtered through 40 µM cell strainers. Spleens were mechanically dissociated using 1ml syringe plungers in RPMI supplemented with 10% FBS Gibco and passed through 40 µM cell strainers. Livers were dissociated mechanically by crushing between two glass slides for mild dissociation. The liver cell suspensions were collected and filtered through 70 µM cell strainers. Hepatocytes were separated from liver cell suspensions with low-speed centrifugation (50 rcf) and discarded. Lymphocytes were then isolated from the remaining cell suspension by density gradient centrifugation using 33% Percoll solution (OptiPrep, Stemcell Technologies, Vancouver, Canada) and resuspended in RPMI supplemented with 10% FBS Gibco.

Data are presented as mean ± SD for frequency variables and mean ± SEM for numerical variables. Statistical analyses were conducted using GraphPad Prism version 10. Statistical differences between groups for in vitro studies were evaluated using either one-way or two-way ANOVA. For in vivo study results, non-parametric tests were performed using (Kruskal-Wallis Test). Kaplan Meier plots were used to calculate survival probabilities and P-values were computed using log-rank (Mantel-Cox) test. Linear regression with the Wald test was used for CD161 expression and NK cell purity correlation. P-values ≤0.05 were considered as statistically significant: *p ≤ 0.05; **p ≤ 0.01; ***p ≤ 0.001, ****p ≤ 0.0001. ns: non-significant.

We have shown that CB CD34⁺ HSPCs cultured during 7 days in a feeder-cell free system containing immobilized hDLL4 and a specific set of cytokines resulted in the production of >90% of CD7 expressing T-cell progenitors (ProTcells) that express intracellular CD3ϵ and transcription factor GATA3, known to be associated with T-cell commitment (41, 44). To exclude any potential other than T-cell differentiation, we cultured ProTcells in conditions promoting differentiation into other lymphoid lineages. Unexpectedly, we observed efficient and rapid differentiation of ProTcells into NK cells (ProT-NK) (CD3^-CD56⁺), after culturing them in a hDLL4-free and feeder-cell-free system with a specific cytokine cocktail for NK cell differentiation for a total of 14 or 21 days. The purity of ProT-NK cell product reached an average range of 80 to 95% at both day 14 (ProT-NK D14) and day 21 (ProT-NK D21), with minimal detection of CD3⁺ cells ( Figure 1A ).

This NK cell production process had been tested on small-scale, with the first step performed in 96- to 24-well plates and the second step in up to 6-well plates. We aimed to evaluate the feasibility of scaling up this culture process and to assess whether the scale-up would affect the quality of the generated NK cells. In the large-scale process, the first step was performed in T25/T75 flasks, and the second step was carried out T150 flasks ( Figure 1B ). No significant differences were observed in NK cell (CD3^-CD56⁺) frequencies between small-scale and large-scale production for both day 14 and day 21 ProT-NK products with their frequencies ranging from 87.7 ± 6.1% (D14) to 90.8 ± 5.1% (D21) (mean ± SD) for small-scale and from 82.4 ± 10.8% (D14) to 88.4 ± 8.9% (D21) (mean ± SD) for large-scale ( Figure 1B ), suggesting that NK cell purity was not affected by the scale-up. The analysis of NK cell yields from different CD34⁺ HSPC donors ( Figure 1C ) showed that a single CD34⁺ CB HSPC produced an average of 2037 ± 406 (D14) and 4981 ± 1087 (D21) (mean ± SEM) ProT-NK cells for small-scale and 2906 ± 633 (D14) and 5063 ± 681 (D21) ProT-NK cells for large-scale productions ( Figure 1C ). Importantly, no significant difference was observed in NK cell yields of small-scale and large-scale productions ( Figure 1C ), indicating that scale-up did not impact ProT-NK cell yield.

Since CD161 has already been reported to be highly expressed early during NK cell development (45), we investigated if ProTcells express CD161. Flow cytometry analysis revealed that a fraction of CD7⁺ ProTcells express CD161 ( Supplementary Figure S1A ). Then, we used a linear regression analysis to investigate whether there is any correlation between the percentage of CD7⁺CD161⁺ cells in ProTcell and the purity of the generated NK cell product ( Figure 1D ). The analysis revealed a strong correlation between initial proportions of CD7⁺CD161⁺ cells present at ProTcell stage and CD3^-CD56⁺ NK cell purity in the resulting ProT-NK cell product, with an R² value of 0.72 (p = 0.0162) at day 14 and an even stronger correlation at day 21, with an R² of 0.85 (p = 0.0083) ( Figure 1D ). These findings suggest that initial percentage of CD7⁺CD161⁺ cells in ProTcell may predict NK cell purity of our ex vivo generated ProT-NK cell product and that more than 50% of CD7+CD161+ in ProTcells may be necessary to achieve over 90% of CD3^-CD56⁺ ProT-NK cell purity. However, the CD3^-CD56⁺ cell yield didn’t correlate with the CD7⁺CD161⁺ initial cell proportion in ProTcells ( Supplementary Figure S1B ). Collectively, these data demonstrate that our ex vivo feeder cell-free culture process reliably produces pure NK derived from CB ProTcells (ProT-NK) in large-scale without significant T cell contamination, highlighting the efficiency and scalability of our system.

Next, we characterized our ProT-NK cell products and compared them with expanded PB-NK cells to assess if our ex vivo generated ProT-NK cells are similar to PB-NKs, as PB-NK cells represent standard physiologically differentiated NK cells. Flow cytometry analysis revealed that, in comparison to PB-NK cells, both ProT-NK D14 and D21 expressed similar percentages of NK cell activation receptors NKp30, NKp44, NKp46, and NKG2C ( Table 2 , Figure 2A , Supplementary Figures S2, S3A ). However, ProT-NK D14 and D21 cells showed significantly higher expressions of DNAM-1 and ProT-NK D21 showed significantly higher NKG2D expression compared to PB-NK cells ( Table 2 , Figure 2A , Supplementary Figures S2, S3A ). Conversely, both ProT-NK D14 and D21 display lower CD16 expressions than PB-NK ( Table 2 , Figure 2A , Supplementary Figures S2, S3A ). Regarding inhibitory receptors, both ProT-NK D14 and D21 showed lower expression of KLRG1, KIR2DL2/3 and KIR3DL1/2 compared to PB-NK ( Table 3 , Figure 2B , Supplementary Figure S3B ). CD94 and NKG2A were highly expressed in PB-NK, while their expressions were comparatively much lower in ProT-NK D14 and D21 ( Table 3 , Figure 2B , Supplementary Figure S3B ). Overall, the ProT-NK cell phenotype is characterized by high levels of activation receptors and low levels of inhibitory markers.

We then analyzed the expression of key cytotoxic molecules, perforin and granzyme B, which are critical for NK cell cytotoxicity (46) ( Figure 2C , Supplementary Figure S3C ). Both ProT-NK D14 and D21 cells showed high perforin and granzyme B expressions, comparable to expanded PB-NK cells (PB-NK: 85.4 ± 23.5%, ProT-NK D14: 75.7 ± 15.7%, ProT-NK D21: 77.3 ± 14.9%).

NK ontogeny relies on the expression of transcription factors ID2, EOMES and T-bet (47, 48). Therefore, to confirm NK identity, the expression of these transcription factors was assessed in ProT-NK cells and compared to expanded PB-NK cells ( Figure 2D ). ProT-NK cells expressed significantly lower EOMES than PB-NK (PB-NK: 80.7 ± 13.3%, ProT-NK D14: 2 ± 0.3%, ProT-NK D21: 11 ± 3%). T-bet expression was comparable between PB-NK and ProT-NK D21 (PB-NK: 89 ± 6%, ProT-NK D21: 81 ± 7%) but was lower in ProT-NK D14 cells (45 ± 1%). ID2 was highly expressed in both ProT-NK D14 and D21 cells, comparable to PB-NK (PB-NK: 99 ± 0.1%, ProT-NK D14: 99 ± 0.1%, ProT-NK D21: 99 ± 0.2%) ( Figure 2D ). In summary, ProT-NK D14 are ID2^hiEOMES^loT-bet^int and ProT-NK D21 are ID2^hiEOMES^loT-bet^hi compared to PB-NK cells, which are ID2^hiEOMES^hiT-bet^hi. Altogether, these findings indicate that ProT-NK cells express key activation receptors (except CD16), cytotoxic molecules (perforin, granzyme B), and transcription factors ID2 and T-bet, confirming their NK identity. However, they remain less mature than PB-NK cells, as reflected by their lack of CD16, KIR molecules, and EOMES, and lower NKG2A/CD94 expression, highlighting how the unique phenotype of ProT-NK cells differs significantly from that of PB-NK cells.

Given the distinct phenotype of ProT-NK cells revealed by flow cytometry, we further explored their characteristics in-depth through cytometry by time of flight (CyTOF), using two extensive NK-related panels: one focused on phenotype and other on function. We analyzed three donors per condition (PB-NK, ProT-NK D14, and ProT-NK D21) and used UMAP (42) to visualize cell populations based on protein expressions ( Figure 3 , Supplementary Figure S4 ). Most of the ProT-NK D14 and D21 cells expressed high levels of CD56 and CD161, with no contamination of other cell lineages (as indicated by lack of expressions of CD3, CD14, CD19 and CD33) ( Figure 3A ), confirming their NK identity and purity, consistent with flow cytometry results. PB-NK cells expressed lower levels of CD161 compared to ProT-NK D14 and D21 ( Figure 3A ). IL-2Rβ and CD57 were lower in ProT-NK cells compared to PB-NK cells ( Figure 3A ). KIR2DS4, an activating KIR, was expressed in both ProT-NK D14 and D21, with ProT-NK D21 showing similar levels of expression to PB-NK cells ( Figure 3B ). KIR2DL1S1 was expressed at low levels across all conditions ( Figure 3B ). Other activation markers showed the following: NKG2D expressions in ProT-NK cells were comparable to PB-NK. CD16 expression was absent in ProT-NK, contrary to PB-NK, which exhibited a distinct subpopulation expressing CD16 ( Supplementary Figures S4A, B ). DNAM-1 was comparable with ProT-NK D21 and PB-NK but lower in ProT-NK D14 ( Supplementary Figures S4A, B ). NCRs (NKp30, NKp44, NKp46) expression patterns were similar across ProT-NK and PB-NK ( Supplementary Figures S4A, B ). KLRG1 and NKG2A/CD94 heterodimer were detected in all samples but at higher levels in PB-NK ( Supplementary Figures S4A, C ). Inhibitory KIRs (KIR2DL2L3, KIR3DL1) were low or absent in ProT-NK ( Supplementary Figures S4A, C ). Importantly, chemokine receptor expression analysis revealed high CCR6 and CXCR3 expressions in ProT-NK D14 cells, comparable to PB-NK, while the expressions of other chemokine receptors (CCR5, CXCR2, CXCR4) were higher in PB-NK. Notably, CXCR4 expression was almost absent in all conditions ( Figure 3C ). The transferrin receptor (CD71), which is known to be related to enhanced cell metabolism and growth (49, 50), was strongly expressed by ProT-NK D14, while PB-NK and ProT-NK D21 showed lower relative expression, suggesting enhanced metabolism and growth of ProT-NK D14 cells ( Figure 3D ). Moreover, comparable to PB-NK, ProT-NK D14 and D21 showed high expressions of CD98 ( Figure 3D ), which is already shown to be upregulated in metabolically active NK cells (51–53). Interestingly, CD38, reported to be associated with fratricide and exhaustion (54, 55) was not as higly expressed by ProT-NK cells than by PB-NK. CD69 and Fas-L were expressed both by ProT-NK and PB-NK cells. Fas-L was expressed at higher levels by PB-NK than by ProT-NK cells ( Figure 3D ). Finally, when CD8 expression was not detected in ProT-NK cells, PD-1 was minimally expressed in ProT-NK D21 likely due to the extended culture time ( Figure 3D ).

NK cell development progresses through defined stages that impact their function (56). To elucidate these stages within our ProT-NK cell product, we applied FlowSOM clustering on merged PB-NK, ProT-NK D14, and ProT-NK D21 data ( Figure 3E ). FlowSOM generated seven clusters ( Figure 3E ), and based on the expressions of CD56, CD161, CD94, NKG2D, CD57, CD16, KIRs, natural cytotoxicity receptors (NCRs), and CXCR4 we correlated them with already established NK cell developmental stages (63–65). Analysis of the clusters by manually annotating them revealed the presence of all NK cell developmental stages, from pre-NKP to terminally mature NK cells. Cluster abundance in PB-NK, ProT-NK D14, and ProT-NK D21 was then assessed ( Figure 3F ). ProT-NK D14 was enriched in stage 1 (pre-NKP/non-NK: CD56^loCD161⁺CD94^-NKG2D^-) and stage 3 (iNK: CD56⁺CD161^brightCD94⁺NKG2D⁺CD16^-), while ProT-NK D21 showed an increased presence of stage 4b cells (CD56^bright: CD161⁺CD94⁺NKG2D⁺NCRs⁺⁺) in addition to stage 1 and 3, suggesting an activated state after long cytokine exposure. PB-NK cells displayed higher abundance of mature clusters, including stages- 2 (NKP: CD56⁺CD161^-CD94⁺NKG2D⁺), 4b (CD56^bright: CD161⁺CD94⁺NKG2D⁺NCRs⁺), 5 (CD56^dim: CD161⁺CD94⁺NKG2D⁺NCRs⁺⁺CD16⁺KIR⁺⁺CD57^-) and 6 (terminally mature: CD56^dimCD161^-CD94⁺NKG2D⁺NCRs⁺⁺CD16⁺KIR⁺⁺CD57^hi). Overall, these results show distinct maturation stages in ProT-NK cells, with early markers in ProT-NK D14 and more mature features in ProT-NK D21, highlighting their unique developmental profile compared to PB-NK cells.

Subsequently, we assessed the functional and cytotoxic potential of ProT-NK cells. First, we focused on their degranulation ability, a key function of NK cells that involves releasing their cytolytic granules (e.g. perforin and granzyme B) into the extracellular space (57). The NK cells were stimulated with K562 cells for 6 hours and the degranulation was detected by surface expression of CD107a. Without stimulation, PB-NK cells showed spontaneous degranulation at low percentages (13.1 ± 9.8% CD107a⁺ (mean ± SD)), and so did ProT-NK (ProT-NK D14: 5.5 ± 3.1% CD107a⁺, ProT-NK D21: 6 ± 3.2% CD107a⁺) ( Figure 4A , Supplementary Figure S5A ). When stimulated with K562, ProT-NK cells degranulated efficiently (ProT-NK D14: 56.5 ± 10.5% CD107a⁺, ProT-NK D21: 47.5 ± 20.7% CD107a⁺ (mean ± SD)), at levels comparable to PB-NK cells (56.6 ± 21.6%) ( Figure 4A , Supplementary Figure S5A ).

Next, we analyzed cytokine production in response to K562 cell stimulation and showed that ProT-NK cells produced low levels of IFNγ (ProT-NK D14: 1.9 ± 2.2%, ProT-NK D21: 2.5 ± 2.5%), significantly lower than produced by PB-NK cells (9.4 ± 7.3%) ( Figure 4B , Supplementary Figure S5B ). Interestingly, ProT-NK cells showed higher TNFα production compared to PB-NK cells (PB-NK: 16.9 ± 12.4%, ProT-NK D14: 29.7 ± 13.1%, ProT-NK D21: 25 ± 19.6%), although this difference was not statistically significant ( Figure 4C , Supplementary Figure S5C ).

Mass cytometry analysis of ProT-NK and PB-NK cells after stimulation with K562 cells demonstrated their high expressions of the cytotoxic molecules Granzyme A, Granzyme B and Perforin, with PB-NK showing higher levels of Granzyme A, Granzyme B, Perforin ( Figure 4D , Supplementary Figures S5D, E ). They produced both TNFα and IFNγ with much lower IFNγ levels in ProT-NK cells than in PB-NK, confirming the FC results ( Supplementary Figures S5D, E ). Bcl-2, an anti-apoptotic protein, was significantly higher in both ProT-NK D14 and D21 compared to PB-NK ( Figure 4D , Supplementary Figure S5D ). LAG-3, which is known to be associated with immune cell exhaustion (58), was expressed at similar levels in PB-NK and ProT-NK D21, while ProT-NK D14 exhibited lower levels ( Figure 4D , Supplementary Figure S5D ), indicating a less exhausted phenotype.

The cytotoxic activity of ProT-NK cells was further studied using K562 cells as target cells. After co-incubation for 5 hours at various effector-to-target ratios ( Figure 4E , Supplementary Figure S5E ), ProT-NK D14 and PB-NK showed similar level of target cell killing at 1.25:1 ratio (PB-NK: 45.3 ± 15.8%; ProT-NK D14: 43.8 ± 1%, mean ± SD), while ProT-NK D21 showed a comparatively low level of killing (26.4 ± 0.1%). At 2.5:1 ratio, although not significant, target cell killing by ProT-NK D14 was higher (66.6 ± 8.7%) than PB-NK (53.3 ± 20.5%) and significantly greater than ProT-NK D21 (37.5 ± 17.4%). ProT-NK D14 consistently showed significantly superior killing than PB-NK and ProT-NK D21 when effector-to-target ratios were increased from 5:1 to 20:1 reaching a plateau of killing at 10:1. These results showed that both ProT-NK D14 and D21 could kill K562 cells with ProT-NK D14 displaying greater potency than that of ProT-NK D21. Remarkably, ProT-NK D14 was as cytotoxic as or more cytotoxic than PB-NK cells. In sum, ProT-NK cells, especially at day 14, possess functional and cytotoxic potential, efficiently killing target cells in vitro.

To further explore the homing potential and persistence of ProT-NK cells, we administered them in a murine model. Murine IL-15 cross-reacts poorly with human IL-15 receptors. To better replicate human NK cell biology, we used NSG-Tg(hIL-15) mice, which express human IL-15 at physiological levels and support NK cell differentiation after CD34⁺ HSCT (59, 60). This model is pertinent to evaluate both the homing ability and in vivo differentiation and persistence of ProT-NK cells. Adult NSG-Tg(hIL-15) mice were conditioned with busulfan (30 mg/kg). One day after conditioning, NK cells (PB-NK, ProT-NK D14 or ProT-NK D21) were injected intravenously at a single dose of 20 x 10⁶ CD3^-CD56⁺ cells per mouse. Bone marrow, liver and spleen were analyzed at 5 and 9 days post-injection ( Figure 5A ), applying the gating strategy detailed in Supplementary Figure S6A .

Human CD45⁺ and CD3^-CD56⁺ cells were detected in the bone marrow (BM), liver, and spleen in all groups (PB-NK, ProT-NK D14, ProT-NK D21), on both days, confirming that ProT-NK cells can home to both hematopoietic and non-hematopoietic organs in NSG-Tg(hIL-15) mice ( Figures 5B-D ). In the BM, hCD56⁺ cells were detected at comparable frequencies in ProT-NK D14 and PB-NK injected mice on day 5 (93.8 ± 5.3% for PB-NK and 76.1 ± 13.5% for ProT-NK D14 (mean ± SD)) ( Figure 5B , Supplementary Figures S6B, C ). By day 9, PB-NK levels decreased (67.3 ± 14.2%), whereas ProT-NK D14 levels remained stable (72.6 ± 15.5%) suggesting that ProT-NK D14 cells persist in the BM up to 9 days after intravenous injection. ProT-NK D21 levels in BM were significantly lower (38.8 ± 7.5%) than those of ProT-NK D14 on day 5 (76.1 ± 13.5%) and remained stable through day 9 (38.2 ± 21.7%) ( Figure 5B ). A similar pattern was observed in liver ( Figure 5C ) and spleen ( Figure 5D ), where on day 5, PB-NK cells showed significantly higher hCD56⁺ frequencies (91.2 ± 4.9% liver, 98.6 ± 1.1% spleen) compared to ProT-NK D14 cells (66.1 ± 8.3% liver, 52.7 ± 25.1% spleen). By day 9, ProT-NK D14 levels remained lower than those of PB-NK but the difference was no longer significant ( Figures 5C, D ). ProT-NK D21 cells showed consistently lower frequencies on day 5 in both the liver (48.3 ± 11.6%) and spleen (33.2 ± 14.2%) compared to both PB-NK and ProT-NK D14 cells, and these levels remained stable by day 9 ( Figures 5C, D ). When analyzing hCD3^-CD56⁺ cell numbers in BM ( Figure 5E ) we observed that NK cell mean number was 3-fold higher in PB-NK cell injected mice on day 5 post transplantation (2.2 ± 0.4 x 10⁴ cells, mean ± SEM) compared to mice injected with ProT-NK D14 (7428 ± 877). By day 9, the numbers were similar between PB-NK and ProT-NK due to a decrease in PB-NK, (7067 ± 1051) and an increase in ProT-NK D14 numbers (9144 ± 641). This suggests greater persistence and expansion capacity for ProT-NK D14 cells than for PB-NK cells in the BM. NK cell numbers in ProT-NK D21 condition slightly decreased from day 5 (1280 ± 332) to day 9 (877 ± 137), indicating their limited persistence ( Figure 5E ). A similar trend was observed in cell numbers for both liver ( Figure 5F ) and spleen ( Figure 5G ). On day 5, PB-NK cell injected mice exhibited a significantly higher numbers–by 10-fold–of hCD3^-CD56⁺ cells (3.5 ± 1.2 x 10⁶ cells in the liver and 4.4 ± 1.3 x 10⁵ cells in the spleen) compared to ProT-NK D14 group (3.3 ± 0.8 x 10⁵ cells in the liver and 2.6 ± 0.6 x 10⁴ cells in the spleen). By day 9, this difference was no longer statistically significant in either organ, though NK cell numbers in ProT-NK D14 group remained lower than those of PB-NK group ( Figures 5F, G ). ProT-NK D21 cell injected mice consistently showed lower hCD3^-CD56⁺ numbers on day 5 in both liver (1.9 ± 0.6 x 10⁵ cells) and spleen (1 ± 0.2 x 10⁴ cells) compared to PB-NK and ProT-NK D14 conditions, with the numbers remaining stable or lowering through day 9 ( Figures 5F, G ).

We evaluated the maturation of ProT-NK cells in vivo by assessing the kinetics of CD16 and NKG2A expression in BM ( Figures 5H, I ). CD16 is expressed mainly by CD56^dim mature NK cells (61, 62), while NKG2A/CD94 heterodimer is expressed on CD56^bright immature NK cells (56). Interestingly, while hCD16 expression of NK cells in BM remained stable for PB-NK cell injected mice from day 5 (mean ± SD, 46.5 ± 12.5%) to day 9 (39.6 ± 20.1%), its expression was increased significantly for ProT-NK D14 group (15.7 ± 6.4% for day 5 and 25.4 ± 11.1% for day 9). However, CD16 expression showed no significant change for ProT-NK D21 group from day 5 (12.4 ± 6.7%) to day 9 (12.6 ± 16.7%) ( Figure 5H ). NKG2A expression in NK cells spiked significantly for ProT-NK D14 and D21 conditions from day 0 (5.6 ± 4.7%, 17.6 ± 17.2% respectively) to day 5 (87 ± 11.7%, 73.5 ± 23.2% respectively) and remained high until day 9 (93.9 ± 6.5%, 85.7 ± 21% respectively), reaching the expression level of PB-NK cells, which had already high levels ex vivo ( Figure 5I ). This data indicates that ProT-NK cells, which were immature before injection, can mature in vivo in the BM of NSG-Tg(hIL-15) mice. These findings demonstrate that the homing, persistence, and maturation capacities of ProT-NK cell, particularly D14, enable it to outperform the ProT-NK D21 product. ProT-NK D14 cells were thus chosen for further investigation of their functional potential in vivo.

We established a human tumor xenograft model in NSG-Tg(hIL-15) mice by subcutaneous injection of K562 luciferase expressing cells (K562-Luc) to study the in vivo anti-tumor activity of ProT-NK D14 cells. NSG-Tg(hIL-15) mice (8 -12 weeks old) were conditioned with low dose busulfan for two consecutive days before NK injection. On day 0, mice received a subcutaneous (s.c.) injection of K562-Luc alone (0.5 x 10⁶ cells/mouse), or a mix (before injection) of K562-Luc and PB-NK or ProT-NK D14 (20 x 10⁶ cells/mouse). The tumor and NK cells were mixed immediately before injection under cold conditions, ensuring minimal contact time (3 minutes) to prevent pre-injection cytotoxicity. Another group received ProT-NK D14 (20 x 10⁶ cells/mouse) one day after K562-Luc injection (0.5 x 10⁶ cells/mouse), around the tumor site ( Figure 6A ). Tumor growth was monitored over time using bioluminescence imaging (BLI) with the IVIS^® Spectrum system.

Remarkably, while tumor growth, indicated by increased bioluminescence, was observed in mice injected with K562-Luc alone, no tumor growth (or bioluminescence increase) was detected in mice co-injected with K562-Luc and either PB-NK cells or ProT-NK cells at all analyzed time points (up to 62 days) ( Figures 6B, C ). Both PB-NK and ProT-NK cells eliminated K562-Luc cells subcutaneously within 24 hours, as shown by bioluminescence acquisition at day 1. Interestingly, delayed ProT-NK injection slowed down tumor ( Figures 6B, C , violet lines), suggesting that ProT-NK cells can kill tumor cells in vivo. Additionally, all mice injected with K562-Luc cells alone reached the endpoint (tumor size ≥ 1.7 cm in diameter) and were sacrificed between day 32 and day 46. In contrast, 100% of the mice co-injected with PB- or ProT-NK cells survived until the last analysis point (day 62) ( Figure 6D ). Notably, delayed ProT-NK injection significantly extended survival compared to the untreated group (p=0.0295) and were sacrificed between day 35 and day 60. This data implies that ProT-NK cells possess antitumor potential in vivo.

To determine whether genetically modified transduced ProT-NK could be generated in our ex-vivo feeder-cell-free culture system, CB HSPCs were pre-activated on immobilized hDLL4 culture supplemented with a cytokine cocktail ( Figure 7A ), transduced with a lentiviral vector encoding an anti-CD19 chimeric antigen receptor (CAR) (FMC63), and cultured for 7 days in the hDLL4 culture phase. After 7 days, the resulting ProTcells (>85% CD7⁺) were cultured in a second feeder cell-free NK differentiation and expansion phase with cytokines and were analyzed on day 14 ( Figure 7A ). The ProTcells from CAR-transduced condition efficiently differentiated into ProT-NK cells (CD3^-CD56⁺), achieving purity levels over 80% in the final product ( Figures 7A, B ) with an average of 30.4 ± 16.7% of CAR expressing ProT-NK cells ( Figure 7C ). No significant differences were observed in the NK cell frequencies between non-transduced (NT) (mean ± SD, 89.4 ± 11.6% of CD3^-CD56⁺), and CAR-transduced condition (mean 89.4 ± 10.8% of CD3^-CD56⁺) ( Figure 7B ). Additionally, the NK cell yields ( Figure 7D ) for NT (mean ± SEM, 959 ± 395) and CAR-transduced conditions (738 ± 236) showed no significant difference, indicating that the transduction step does not impair NK differentiation and expansion in this culture system. Surface marker characterization ( Figure 7E ) revealed that ProT-NK cells derived from CAR-transduced condition expressed similar levels of activation receptors (NKG2D, NKp44, NKp46, DNAM-1 and CD16) as NT ProT-NK cells. Likewise, both cell types expressed comparable levels of inhibitory receptors (KLRG1, CD94/NKG2A and KIR3DL1/2). Collectively, these data demonstrate that our ex vivo culture process can generate CAR ProT-NK cells.