K562 cells were maintained in RPMI 1640 medium (Gibco, #1875093) supplemented with 10% fetal bovine serum (FBS) (Gibco, #26140079) and penicillin/streptomycin (Peni-Strep) (1%) (Gibco, #15140130) in a 5% CO₂ incubator. OP9-DLL4 cells were maintained in α-MEM (Gibco, #12571063) supplemented with 20% FBS and penicillin/streptomycin (1%) in a 5% CO₂ incubator. Anonymous human donor PBMCs were obtained by Ficoll centrifugation. NK cells were purified using RosetteSep (StemCell Technologies, #15025). The Lenti-X 293T cell line were cultivated according to Imeri et al., 2024 (37). Briefly, The Lenti-X 293T cell line was cultured in Dulbecco’s Modified Eagle’s Medium (DMEM) supplemented with 10% FBS.

iPSC lines were generated by the Center for IPSC Therapies (CiTHERA, Genopole, Evry, France). The iPSC line used for this study (PB68) was derived from PBMCs of a healthy donor using the Sendai virus (ThermoFisher, #A16517). A Master Bank of one clone (PB68.6) was produced and qualified. Pluripotency was assessed by FACS analysis of the expression of the markers Oct4, Nanog, Tra1-60, and SSEA4. Pluripotency was also assessed by teratoma formation assay after intra-muscular injection of 3x10⁶ of iPSCs into 6-week-old NOD/SCID mice (Charles River Laboratories). After ten weeks, teratoma was dissected and fixed in 4% paraformaldehyde, embedded in paraffin and stained with hematoxylin and eosin in order to assess the presence of ectodermic, endodermic and mesodermic tissues. Genomic integrity was confirmed by conventional karyotype (46 XY). The PB68.6 line was cultured in Essential 8 Flex medium (Gibco, #A1517001) on Geltrex (Gibco, #A1413202). Passages were performed at confluence using 0.5 mM ethylenediaminetetraacetic acid (EDTA) (Invitrogen, #15575020) in phosphate-buffered saline (DPBS) solution (Gibco, #14190144).

For the feeder-free derivation of iNKs from iPSCs (feeder-free iNKs), cells were differentiated according to the protocol of H. Zhu et al., 2019 (36). Briefly, a single-cell suspension of iPSCs was seeded at 8000 cells per well in a 96-well plate containing APEL2 medium (Stemcell Tech, #78062) supplemented with cytokines (40 ng/mL SCF, Stemcell Tech, #78062; 20 ng/mL BMP-4, Stemcell Tech, #78211.1; and 20 ng/mL VEGF-165, Stemcell Tech, #78073.1) and 10 μM ROCKi (Y-27632) (Stemcell Tech, #72304). The plate was then centrifuged at 300g for 5 min and incubated at 37°C in 5% CO₂ for 6 days. The resulting embryoid bodies (EBs) were transferred to a 6-well plate containing NK cell differentiation medium (56.6% DMEM and 28.3% F12 (Gibco, # 11765054), 15% heat-inactivated human AB serum (Sigma Aldrich, # H3667), 1% P/S, 1% GlutaMAXTM (Gibco, #35050038), 1 μM β-mercaptoethanol (Sigma Aldrich, # M3148), 1% ITS supplement (Sigma-Aldrich, I3146), 20 mg/L ascorbic acid (Sigma-Aldrich A4403)) supplemented with cytokines: 20 ng/mL SCF, 20 ng/mL IL-7 (Stemcell Tech, #78196.1), 10 ng/mL IL-15 (Stemcell Tech, #78031), 10 ng/m FLT3 ligand (FLT3L) (Stemcell Tech, #78009), and 5 ng/mL IL-3 (Stemcell Tech, #78040) (1st week) for 3 weeks.

In the feeder-based protocol, iPSC-derived iNK (OP9-DLL4 iNK) cells were differentiated using an adapted version of the protocol of Euchner et al., 2021 (32). iPSCs were detached using EDTA, transferred to low-attachment plates, and incubated overnight in APEL2 medium containing 10 µM ROCKi (Y-27632) to allow EB formation. EBs were collected the next day and transferred into APEL2 medium supplemented with 40 ng/mL BMP-4, 10 ng/mL bFGF (Stemcell Tech, #78188.1) and 50 ng/mL VEGF. On day 4, the EBs were collected again and transferred into APEL2 medium supplemented with 50 ng/mL SCF, 20 ng/mL Flt3L, 20 ng/mL IL-3, and 30 ng/mL TPO (Stemcell Tech, #78210.1). On day 14, EBs were collected and transferred onto overgrown OP9-DLL4 stromal cells in α-MEM supplemented with 20% FBS, 20 ng/mL SCF, 5 ng/mL FLT3l, 5 ng/mL IL-7, and 10 ng/mL IL-15. iNK cells were harvested and transferred every week onto new OP9-DLL4 cells.

For extracellular staining, cells were washed with DPBS and incubated with antibodies in a volume of 100 mL of DPBS with an adequate amount of antibody ( Supplemental Table 1 ). For intracellular staining, cells were then fixed with BD Cytofix™ (BD, 554655) for 10 minutes at room temperature, protected from light, washed, and permeabilized with BD Perm/Wash™ (BD, 555028) together with intracellular antibodies for 20 minutes at 4°C. Flow cytometry was performed with a BD FACS LSRFortessa™ analyzer and Miltenyi MACSQuant 10™ data were examined using FlowJo software version 10.1 (Tree Star, Inc.).

NK cells were incubated in NK containing RPMI 1640 with 10% FBS (Gibco, #A5256701), 1% GlutaMAX™ Supplement (Gibco, #35050038), and 50 UI/mL of human IL-2 (Miltenyi, #130-097-743). NK cells were cultivated overnight and then harvested for intracellular staining of Ki67.

NK cells were tested for their expansion ability using artificial antigen presenting cells (aAPCs) K562-mbIL21. aAPCs were generated using lentivirus expression cassette containing membrane bound IL-21 fusion protein described by Denman et al., 2012 (38). In order to produce mbIL21 expressing lentiviruses, we used Lenti-X 293T as a packaging cell line and ps-PAX2.2, and pMD2.G as packaging vector and envelope vector, respectively. Briefly, the Lentix-293T cells were co-transfected by lipofectamine 3000 reagent (ThermoFisher, L3000015) with 20µg packaging vector of ps-PAX2.2 (Addgene), 10µg envelope vector of pMD2.G (Addgene), and 30µg of mbIL21 vector, synthetized by VectorBuilder. iNK were coculture at a ratio of 1∶2 (iNK∶aAPC) in NK expansion medium. Cultures were refreshed with half-volume media changes every two to three days, and re-stimulated with aAPCs at ratio of 1∶1 every seven days.

Primary NK cells were harvested by negative selection from healthy donor blood using RosetteSep™ Human NK Cell Enrichment Cocktail (Stemcell Tech, #15025). Total RNA was extracted using the RNeasy Mini Kit (Qiagen, #74104). A total RNA library was prepared with TrueSeq technology (Illumina) and paired-end reads of 75 pb were sequenced on an Illumina apparatus (Illumina). Fastq files were aligned using the STAR algorithm (version 2.7.6a) (39) with two-pass basic mode and the Ensembl reference genome release 101. Reads were then counted using RSEM (version 1.3.1) (40) and statistical analyses were performed on the read counts with the DESeq2 package (DESeq2_1.26.0) of R (version 3.6.3) using the standard DESeq2 normalization method (median of ratios) and a pre-filter of reads and genes (reads uniquely mapped on the genome, or up to 10 different loci with a count adjustment, and genes with at least 10 rads in at least 3 different samples) (41). Supervised differential gene expression analysis was performed with EdgeR version 3.40.2 to compare experimental conditions (42). Geneset enrichment analysis was performed with the GSEA standalone application (43) version 2-2.2.2.4, based on the Hallmarks MSigDb database version 2023 (44). Expression heatmaps were drawn with the Pheatmap R-package version 1.0.12.

Based on the FANTOM consortium database of known interactions between human transcription factors (45), an R-package was designed to collect and test interactions between the transcription factors that were found to be differentially expressed in the human transcriptome. These analyses produce a combination score and an enrichment network depicting the combinations between transcription factors. This package is available at the following address: https://github.com/cdesterke/tfcombined (accessed on January 15th, 2024).

To characterize the developmental signature of NK cells, single-cell transcriptome data were obtained from an immune development cell atlas (https://developmental.cellatlas.io/fetal-immune) and used to analyze the genes that were differentially expressed between human NKs from fetal bone marrow and from yolk sac (24). The top 100 differentially expressed genes between NK cells in these developmental conditions were then employed to distinguish between transcriptome conditions from feeder-free iNKs and OP9-DLL4 iNKs. Volcano plots and expression heatmaps were created with the transpipe 1.4.0 R-package (https://github.com/cdesterke/transpipe14, accessed on April 2^nd, 2024) following limma analysis (46).

NK cells stained with CD56-BV421 were incubated with or without cancer targets (K562 cells) at a 1:2 ratio of effector to target. CD107a-PEVio770 antibody was added to each well and incubated for 1 hour. Then, Monensin (BD GolgiStop™ Protein Transport Inhibitor, BD Biosciences, #554724) and Brefeldin A (BD GolgiPlug™ Protein Transport Inhibitor, BD Biosciences, #555029) were added and incubated for 2 more hours. CD107a expression was evaluated following data normalization of NK cells without target cell co-culture and a PMA/ionomycin positive control.

NK cells were incubated overnight with or without cancer target cells (K562) at a 1:2 ratio of effector to target. Supernatants were tested according to the manufacturer’s instructions with the IFN-γ ELISA kit (ThermoFisher, #KAC1231) and the Granzyme B ELISA kit (ThermoFisher, #BMS2027-2). Results that were below the detection limits were considered to be 0 for statistical purposes.

Cytotoxicity induced by NK cells was assessed by staining the surface of target cells first with CellTrace Yellow (ThermoFisher Scientific, #C34567) and then Annexin V. For CellTrace staining, cells were incubated with 1 μL of the dye per 10⁶ cells/mL in PBS for 20 min at 37°C. Stained target cells were then co-cultured at different ratios with iNK cells for 2 h and subsequently stained with Annexin V – APC (ThermoFisher Scientific, #88-8007-74) according to the manufacturer’s instructions. Briefly, cells were washed with 1X binding buffer, centrifuged, and resuspended in binding buffer containing Annexin V – APC. After 15 minutes of incubation, cells were washed again and analyzed by FACS.

Two-tailed Student’s t-tests were performed using GraphPad Prism (GraphPad Software, Boston, Massachusetts, USA). Means and SD are depicted on the graphs. All experiments were performed three times. ns, not significant; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

All experiments were conducted using the same iPSC line ( Supplementary Figure 1 ). The differentiation of hematopoietic progenitor cells (HPCs) was induced via two distinct methods: (i) the OP9-DLL4 protocol, in which 3D embryoid bodies (EBs) formed through self-aggregation in ultra-low-attachment (ULA) plates, and (ii) the feeder-free protocol, which employed spin aggregation of single-cell suspensions ( Figure 1A ). The profiles of the HPCs were then assessed at D14 (OP9-DLL4) or D6 (feeder-free) based on the expression of the CD34, CD45, and CD43 surface markers. Both protocols were able to produce CD34⁺ cells: 50.3 ± 5.8% for the OP9-DLL4 protocol and 27.9 ± 15.2% for the feeder-free protocol (P-value < 0.05). However, CD45 and CD43 were only found in cells derived using the OP9-DLL4 protocol (31.1 ± 6.1% CD34⁺CD45⁺ and 28.8 ± 7.1% CD34⁺CD43⁺; Figures 1B, C ). Because enrichment of HPCs was not required, undigested D14 EBs (OP9-DLL4) and undigested D6 EBs (feeder-free) were then placed in NK differentiation conditions. To obtain a better understanding of the differentiation kinetics, we monitored the expression of CD34, CD45, and the NK lineage marker CD56 weekly. As expected, in both protocols, a decrease was observed during differentiation in the hematopoietic progenitor marker CD34 while expression of the pan-mature hematopoietic marker CD45 and the NK marker CD56 increased proportionally. We also monitored the emergence of lymphoid progenitors weekly throughout the differentiation process. A small population of [CD34⁺]CD7⁺CD45RA^+/- cells emerged 7 days after NK differentiation with the OP9-DLL4 protocol. This population appears to disappear by the second week of NK differentiation. However, we were unable to detect this population in the Feeder-free protocol. After 3 to 5 weeks of differentiation, almost 90% of floating cells in both protocols expressed CD56 ( Supplementary Figures 2A, B ).

The iNK cells generated from both protocols expressed the typical NK markers CD56⁺CD16^- and CD56⁺CD16⁺ and were negative for CD3 ( Figure 2A ). CD16 (FcγRIII) is responsible for the antibody-dependent cell-mediated cytotoxicity (ADCC) characteristic of NK cells. This marker is typically expressed in mature stage 5 blood-circulating NK cells in adults. Interestingly, it was found here to be expressed at similar levels in iNKs derived from both protocols (18.3 ± 9.1% for OP9-DLL4 iNKs and 17.9 ± 6.1% for feeder-free iNKs) in the CD56⁺ fraction ( Figure 2B ). Instead, CD7, which is a marker of lymphoid commitment, was found only in OP9-DLL4 iNKs; it appeared gradually in CD56⁺ cells at the endpoint of differentiation (27.5% in OP9-DLL4 iNKs and 1.6% in feeder-free iNKs cells) ( Supplementary Figure 2C ). NKp80, which characterizes stage 4b mature and cytotoxic NK cells, was also detectable in up to 61.6% of OP9-DLL4 iNKs, but was undetectable in terminally differentiated feeder-free iNKs ( Figure 2A ).

Compared to feeder-free cells, OP9-DLL4 iNKs displayed a distinct phenotypic profile, with CD56⁺ populations exhibiting a significantly higher proportion of cells positive for DNAM-1 (69.8 ± 28.9% for OP9-DLL4 iNK cells versus 21.5 ± 17.6% for feeder-free iNK cells, P-value < 0.05) and NKp46 (72.9 ± 17.9% for OP9-DLL4 iNK cells versus 23.7 ± 24.6% for feeder-free iNK cells, P-value < 0.05). Instead, the expression of the receptors NKP30, NKG2A, and NKG2D was similar in both protocols. Both sets of iNK cells also expressed the KIR receptor 2DL2-3 (CD158b) at low levels (2.6 ± 0.7% for OP9-DLL4 and 1.1 ± 1.1% for feeder-free) ( Figure 2B ). To better assess the profile of the CD56⁺ cell population at the endpoint of iNK differentiation, we looked at the relative expression of markers as measured by mean fluorescence intensity (MFI) ( Figure 2C ). The feeder-free iNK cells exhibited a generally muted expression profile compared to OP9-DLL4 iNKs, with a reduced expression of DNAM-1 (P-value < 0.05) and a relatively lower (though not significantly so) expression of NKP46 (P-value < 0.08).

To characterize their proliferation capacity, we stimulated iNK cells overnight with IL-2 (50 UI/mL) and analyzed the proportion of Ki67⁺ cells. The proportion of CD56⁺Ki67⁺cells was higher in OP9-DLL4 iNKs (33.1 ± 1.0%, compared to 8.8 ± 0.6% for feeder-free iNK cells), suggesting that they have an enhanced proliferation capacity ( Supplementary Figure 2D ). Taken together, these results suggest that, compared to feeder-free iNKs, OP9-DLL4 iNK cells exhibit a more differentiated, lymphoid-primed phenotype and express higher levels of activation receptors such as NKP46 and DNAM-1.

To examine the generated NK cells in greater detail, we conducted a transcriptomic analysis of both OP9-DLL4 and feeder-free iNKs. Between these two sets of cells, we identified 2148 genes with significant differences in expression (differentially expressed genes, DEGs) ( Supplementary Figure 3 ). Among these, we identified chemotactic markers such as CCL17 and CCL1, which were upregulated in OP9-DLL4 iNK cells, as well as transcription factors like GATA2. We also detected the transcription factor NFATC4 and the immune checkpoint LAG3 as having higher expression in feeder-free iNK cells. Using the normalized transcriptome data, we conducted a gene set enrichment analysis (GSEA) comparing the two distinct sets of iNK cells using the Hallmark gene set database. OP9-DLL4 iNK cells showed significant enrichment in immune-related gene sets, including those involved in the IFN-γ response, TNF signaling via NF-κB, complement pathways, and inflammatory responses ( Figure 3A ). In contrast, GSEA revealed that feeder-free iNK cells exhibited enrichment in gene sets related to transcriptional regulation and cell cycle progression ( Figure 3B ).

The transcriptomes of our iNKs were also compared with those of peripheral blood NKs cells. Among the genes that were differentially expressed between the three conditions, we identified 164 human transcription factors ( Supplementary Table 1 ). Moreover, unsupervised principal component analysis of the expression patterns of these 164 transcription factors was able to effectively discriminate among the three groups of cells: 70.22% of the variance distinguished primary NK cells from iNK cells along the first principal axis. In contrast, 16.39% separated OP9-DLL4 iNKs from feeder-free iNKs along the second principal axis ( Figure 3C ).

This pattern of segregation was further confirmed by unsupervised clustering based on the top 75 differentially regulated transcription factors ( Figure 3D ). Notably, OP9-DLL4 iNK cells exhibited major upregulation of GATA2, FOS, FOSB, JUN, ZEB2, and STAT1 compared to feeder-free iNKs, while feeder-free iNK cells showed increased expression of HOXB5, MYBL2, HC1, and E2F2 compared to OP9-DLL4 iNKs. Additionally, analysis of the transcription factors highlighted in OP9-DLL4 iNK cells revealed the enrichment of a comprehensive cistrome comprising interactions among IRF1, FOS, STAT1, STAT2, STAT3, HES1, and GATA2, all of which are known to be involved in NK cell differentiation and maturation. In contrast, feeder-free iNK cells were characterized by a major cistrome enriched in MCM3, MCM7, MCM6, and MCM8, which are implicated in replication and mitosis ( Supplementary Figure 4 ).

To assess the generated iNK cells in an embryonic context, we analyzed the profile of NK cells derived from different anatomical sites using an immune development cell atlas (https://developmental.cellatlas.io/fetal-immune) (24). Specifically, we compared single-cell transcriptome data from the human yolk sac and fetal bone marrow using UMAP dimension reduction. This analysis revealed significant differences in gene expression profiles between these two cell populations. Effectively, after dimension reduction, bone marrow NK cells were found to be concentrated at the top left of the UMAP visualization of the human NK cell cluster ( Figure 4A ) and yolk sac NKs were found to be concentrated at the bottom of the heatmap ( Figure 4B ). Differential expression analysis was performed between the two NK identities and the top 50 differentially expressed genes ( Supplementary Table 2 ) were identified in each cell population (histograms in Figures 4A, B ). This NK developmental signature (100 genes) was then employed to distinguish between the iNK cell transcriptomes ( Figures 4C, D ). OP9-DLL4 iNK cells were found to be more enriched in HLA, B2M, CD74, GRZB, and PRF1, which are characteristic of fetal bone marrow NK cells. These results suggest that, compared to feeder-free iNKs, OP9-DLL4 iNK cells present a transcriptional profile closer to that of bone marrow NK cells.

To assess the functionality of the iNK cells generated using feeder-free and feeder-based protocols, we investigated the degranulation capacity of these cells after co-culture with leukemic K562. As expected, iNK cells from both protocols were able to efficiently degranulate (more than 80% of cells in the positive control after PMA stimulation) (data not shown). After stimulation with K562 cells, OP9-DLL4 iNK cells showed a much higher degree of degranulation (1388.7 ± 35.2 MFI versus 657.3 ± 29.2 MFI for feeder-free iNKs, P-value < 0.0001) ( Figure 5A ). We corroborated this result by measuring, by ELISA assay, the concentration of granzyme B in the supernatant of overnight-stimulated iNK cells. As expected, OP9-DLL4 iNK cells showed a much higher concentration of granzyme B in the supernatant than feeder-free iNKs (23.1 ± 0.5 ng/mL and 0.5 ± 0.4 ng/mL, respectively, P-value < 0.0001) ( Figure 5B ). Similarly, we used ELISA assay to measure IFN-γ secretion in the supernatant after overnight stimulation with either PMA or K562 cells. Both types of iNK cells displayed a strong IFN-γ response to stimulation with PMA (2914.4 ± 140.0 pg/mL for OP9-DLL4 iNKs and 1929.9 ± 1104.3 pg/mL for feeder-free iNKs). However, IFN-γ secretion was much higher in OP9-DLL4 iNKs than in feeder-free iNKs after stimulation with K562 cells (234.1 ± 193.9 pg/ml versus 0.1 ± 0.03 pg/mL, respectively, P-value < 0.05) ( Figure 5C ). Finally, to evaluate the short term cytotoxic potential of these iNK cells, we evaluated the proportion of apoptotic cells resulting from co-culture with K562 cells at different ratios of effector to target (0.25:1, 0.5:1, 1:1, and 2:1). Interestingly, we observed a much higher percentage of apoptotic cells in all ratios with the OP9-DLL4 iNK co-culture ( Figure 5D ). Altogether, our data suggest that OP9-DLL4 iNKs are superior in terms of functionality in in vitro assays against the K562 cancerous cell line.

To further evaluate the cytotoxic potential of iNK cells, we generated artificial antigen-presenting cells (aAPCs) expressing mbIL21 (K562-mbIL21) to promote their expansion and maturation ( Supplementary Figure 5A ). This cell line appeared to activate iNK cells, evidenced by an increased expression of certain activating receptors such as DNAM-1 and NKp46, particularly in feeder-free iNK cells where expression levels were initially low ( Supplementary Figure 5B ). However, this cell line did not induce logarithmic expansion of iNK cells. After two weeks of co-culture, the cells exhibited unfortunately apoptotic morphology and showed signs of anergy in cytotoxicity assays against K562 cells ( Supplementary Figure 5C ).