Abstract

   Human induced pluripotent stem cells (iPSCs) offer a promising source
   for chimeric antigen receptor (CAR)–engineered natural killer (NK)
   products. However, complex iPSC-NK (iNK) manufacturing challenges
   clinical use. Here, we identified LiPSC-GR1.1 as a superior iPSC line
   for iNK production. By engineering LiPSC-GR1.1 with a mesothelin
   (MSLN)–targeting CAR and interleukin-15 (IL-15), we achieved robust
   differentiation of iPSCs into mature activated iNK cells with enhanced
   tumor killing efficacy, superior tumor homing, and vigorous
   proliferation. Single-cell transcriptomic analysis revealed that
   transforming growth factor–β (TGF-β)–producing tumor cells up-regulated
   major histocompatibility complex molecules and down-regulated MSLN
   post–CAR-IL-15 iNK treatment. Tumor-infiltrating CAR-IL-15 iNK cells
   exhibited high levels of CAR, IL-15, and NK-activating receptors,
   negligible checkpoint exhaustion markers, and extremely low levels of
   NK suppressive factors CISH, TGFBR2, and BATF, enabling them to sustain
   activation, metabolic fitness, and effective tumor killing within
   TGF-β–rich hypoxic tumor microenvironment. Overall, we developed
   MSLN.CAR-IL-15–engineered GR1.1-iNK therapy with enhanced antitumor
   efficacy for solid tumor treatment.
     __________________________________________________________________

   MSLN.CAR-IL-15 engineering boosted iPSC-NK production and improved iNK
   antitumor efficacy.

INTRODUCTION

   Chimeric antigen receptor (CAR) T cell therapy has been highly
   effective for treating blood cancers but shows limited success against
   solid tumors ([40]1). CAR-engineered natural killer (NK) cells provide
   both CAR-dependent and CAR-independent antitumor effects with lower
   toxicity and offer scalable off-the-shelf production, making them a
   promising alternative cell therapy ([41]2, [42]3). Induced pluripotent
   stem cell (iPSC)–based NK (iNK) cell therapies have several advantages,
   such as a nearly limitless source, efficient and stable CAR expression
   with one-time genetic modification, and a relatively homogenous cell
   product with a more standardized production pipeline ([43]4). A recent
   clinical report on FT596, an iPSC-derived CD19 CAR NK cell therapy,
   demonstrated good tolerance and encouraging outcomes in patients with
   relapsed or refractory B cell lymphoma, achieving an complete response
   rate of 85% in follicular lymphoma, 64% in aggressive large-cell
   lymphomas excluding de novo diffuse large B cell lymphoma, and 30%
   among 20 patients who had previously failed CD19 CAR T cell therapy
   ([44]5). This study supports iPSC-derived, gene-modified NK cells as
   highly promising allogeneic cell therapy for cancer and other diseases.
   However, CAR iNK manufacturing is a complex and challenging process
   that includes CAR engineering in iPSCs, differentiating iPSCs into
   functional iNK cells and expanding these cells at a large scale
   ([45]6). iPSCs derived from different donors exhibit genetic
   variability, which can influence the differentiation potential,
   functionality, and safety profile of the resulting iNK cells ([46]7).
   Thus, iPSC line diversity also poses challenges for standardizing cell
   manufacturing processes for iPSC-based therapies ([47]8). The
   Regenerative Medicine Program was launched by the US National
   Institutes of Health (NIH) to develop resources to catalyze therapeutic
   use of iPSCs. To date, this program has generated 1 clinical-grade
   current Good Manufacturing Practices (cGMP) iPSC line and 14
   research-grade iPSC lines, all of which are accessible to the research
   community. Among them, six iPSC lines were generated from healthy
   donor–derived cord blood CD34^+ cells, including the cGMP-manufactured
   standardized line LiPSC-GR1.1, which has been well characterized
   ([48]9, [49]10). Ideally, using well-characterized cGMP-manufactured
   human iPSC lines to produce CAR iNK products may result in time and
   cost reduction in translation into the clinic because of elimination of
   the time-consuming and resource-intensive quality control measures
   needed for characterization of newly reprogrammed iPSCs.

   Mesothelin (MSLN) is a glycosylphosphatidylinositol-anchored membrane
   glycoprotein highly expressed in many solid tumors but with limited
   expression on normal tissues, making it an attractive target antigen
   for a cell therapy ([50]11). We previously generated a highly effective
   CAR T product using the sequence of humanized hYP218 scFv targeting the
   MSLN membrane–proximal region. This CAR T product showed superior tumor
   killing and intratumoral infiltration as compared to CAR T generated
   using the SS1 scFv targeting the MSLN membrane–distal region ([51]12).
   Thus, we hypothesized that hYP218 scFv would also be promising for
   creating iPSC-derived CAR NK for treatment of MSLN-positive solid
   tumors. A very recent study also showed that YP218.CAR-engineered NK
   derived from peripheral blood (PBNK) had superior antitumor efficacy as
   compared to SS1.CAR-engineered PBNK in MSLN-positive ovarian cancer
   ([52]13).

   NK cells generally have a short lifespan in vivo, especially without
   support from cytokines like interleukin (IL-2) or IL-15 ([53]14), which
   plays a critical role in NK cell development and homeostasis ([54]15,
   [55]16). IL-15 has been included in CAR constructs for genetically
   modifying primary NK products to locally enhance NK cell survival,
   expansion, and antitumor activity ([56]17). Most of the research in
   this area has focused on hematopoietic cancers, such as
   IL-15–expressing CD19 CAR NK therapies ([57]18–[58]20). Studies
   investigating IL-15–expressing CAR NK cells in solid tumors are still
   relatively limited, and even fewer have explored the use of
   IL-15-CAR–engineered iPSC-NK in this setting. Consequently, our
   understanding of the molecular features of IL-15–expressing CAR iNK in
   the context of local solid tumors remains unclear.

   In this study, we identified LiPSC-GR1.1 as a superior iPSC line for
   iNK production. We genetically engineered LiPSC-GR1.1 using a nonviral
   PiggyBac transposon system expressing MSLN-targeting hYP218.CAR plus
   IL-15 and achieved robust differentiation of MSLN.CAR-IL-15 GR1.1-iNK
   cells. We then explored NK biomarker and transcriptional profiles of
   the GR1.1-iNK product and evaluated the tumor killing efficacy of the
   GR1.1-iNK cells using in vitro and animal model studies. In addition,
   we demonstrated tumor infiltration and proliferation of the
   MSLN.CAR-IL-15 GR1.1-iNK cells in a xenograft tumor model and
   investigated the transcriptional signatures of iNK-treated tumor as
   well as tumor-infiltrating MSLN.CAR-IL-15 GR1.1-iNK cells using
   single-cell sequencing.

RESULTS

LiPSC-GR1.1 identified as a superior human iPSC line for NK cell
differentiation

   We selected the well-characterized LiPSC-GR1.1 line ([59]9, [60]10) and
   five NIH Center for Regenerative Medicine (NCRM) iPSC lines with a
   normal karyotype for NK differentiation study. The NCRM iPSC lines
   exhibited a similar SSEA-4^+TRA-1-60^+ pluripotency phenotype ([61]Fig.
   1B) and were validated for their trilineage differentiation potential
   through scorecard assay (fig. S1). Using a standardized, simple, and
   efficient “spin embryoid body (EB)” iPSC-NK differentiation protocol
   ([62]Fig. 1A) ([63]21), we compared LiPSC-GR1.1 to the five NCRM iPSC
   lines for their iNK differentiation potential. A total of 8000
   TrypLE-adapted iPSCs was seeded in each well of 96-well round-bottom
   plates in APEL culture medium containing a Rho-associated coiled-coil
   kinase (ROCK) inhibitor, stem cell factor (SCF), vascular endothelial
   growth factor (VEGF), and bone morphogenetic protein 4 (BMP-4) and spun
   to form EBs. All the cell lines initially aggregated and formed EBs on
   day 6 postcell seeding ([64]Fig. 1C). After day 6, EBs were then
   directly transferred into six-well plates with NK cell differentiation
   medium for 4 weeks. During weeks 2 to 4, we observed substantial
   hematopoietic cell differentiation from EBs derived from LiPSC-GR1.1
   and NCRM5. The EBs of these two lines began to disappear, with more
   free-floating, multicellular aggregates observed during the third week
   of NK differentiation. A low number of floating hematopoietic cells
   were observed with the EBs derived from the NCRM6 line. The EBs derived
   from NCRM1, NCRM2, and NCRM4 failed to further differentiate. The
   average iNK differentiation yields were 1.3 × 10^5 per EB, 8.8 × 10^4
   per EB, and 9.0 × 10^3 per EB from iPSC lines LiPSC-GR1.1, NCRM5, and
   NCRM6, respectively ([65]Fig. 1D). After 4 weeks of NK differentiation,
   >98% of cells harvested from the supernatant of LiPSC-GR1.1 and NCRM5,
   as well as >90% of cells harvested from the supernatant of NCRM6, were
   CD45^+CD56^+ NK cells ([66]Fig. 1E) with higher CD56 expression as
   compared to the PBNK cells ([67]Fig. 1F).

Fig. 1. Differentiation of iNK cells from six human iPSC lines.

   [68]Fig. 1.
   [69]Open in a new tab

   (A) Scheme of NK differentiation from iPSCs. EBs were generated using
   the spin method from a single-cell dissociated human iPSCs seeded on
   day −6 and maintained under feeder- and serum-free condition in a
   U-bottom 96-well plate. Six days later (after hematopoietic progenitor
   cells generated), EBs were transferred into NK cell differentiation
   conditions on day 0. NK cells were then harvested 4 weeks
   postdifferentiation. (B) The expression of pluripotency markers SSEA-4
   and TRA-1-60 was detected in the iPSC lines via flow cytometry. (C)
   Left panel: morphology of day 0 EBs and 4 weeks of NK differentiation
   from six different iPSC cell lines. Scale bars represent 1000 μm. Right
   panel: representative magnified pictures of differentiated iNK cells 4
   weeks post-NK differentiation. Scale bars represent 200 μm. (D) iNK
   differentiation yield per EB from different iPSC lines (n = 2 to 3
   independent differentiation experiments performed for each iPSC line).
   ND means that no differentiation was observed in these lines. *P < 0.05
   and ***P < 0.001, analyzed using unpaired Student’s t test. (E) NK
   differentiation from the three iPSC lines LiPSC-GR1.1, NCRM5, and NCRM6
   was validated by staining for CD45 and CD56. (F) The median
   fluorescence intensity (MFI) of CD56 was measured by flow cytometry on
   PBNK cells (from three healthy donors) and iNK cells (derived from
   LiPSC-GR1.1, NCRM5, and NCRM6).

Robust differentiation of NK cells from MSLN.CAR-IL-15–engineered LiPSC-GR1.1
cells

   To target MSLN-expressing solid tumors, the DNA construct MSLN.CAR that
   encodes the hYP218 scFv, followed by a CD8a hinge spacer, the NK group
   2 member D (NKG2D) transmembrane domain (TM), the 2B4 intracellular
   domain (ICD), and the CD3ζ intracellular signaling domain, along with
   an independently translated truncated human epidermal growth factor
   receptor (EGFR) polypeptide (EGFRt) sequence for tracking purpose, was
   subcloned into a PiggyBac transposon vector. IL-15 was also included in
   the construct to support NK persistence in vivo ([70]Fig. 2A) ([71]9).
   We then genetically engineered LiPSC-GR1.1 cells with the PiggyBac
   transposon system carrying MSLN.CAR or MSLN.CAR-IL-15 and enriched the
   iPSCs stably expressing the components of the constructs by
   fluorescence-activated cell sorting (FACS) of EGFRt^+ iPSCs. CAR
   expression by the expanded CAR or CAR-IL-15–engineered LiPSC-GR1.1
   cells was validated using flow cytometry ([72]Fig. 2B). IL-15
   production was detected only in the supernatant of the cultures of
   CAR-IL-15–engineered iPSCs using enzyme-linked immunosorbent assay
   (ELISA) ([73]Fig. 2C). We next tested the NK cell differentiation
   potential of these iPSCs. It is observed that a robust yield of iNK
   cells was attained from MSLN.CAR-IL-15 iPSCs (~1.6 × 10^6 per EB)
   compared to the limited iNK yield from mock iPSCs (~2.3 × 10^5 per EB)
   or MSLN.CAR-engineered iPSCs (~4.5 × 10^5 per EB) at 4 weeks
   postdifferentiation initiation ([74]Fig. 2D and fig. S2). In contrast,
   we only observed a slightly increased NK cell yield from the
   LiPSC-GR1.1 cells when we increased the IL-15 concentrations to 20 and
   40 ng/ml in the differentiation medium ([75]Fig. 2E), suggesting that
   genetically modifying iPSCs to constantly express endogenous IL-15
   ensures more robust and sustained signaling for hematopoietic
   differentiation within the mesoderm. We then examined NK markers and
   CAR expression by these differentiated iNK cells using flow cytometry.
   The percentage of CD45^+CD56^+ NK cells in culture supernatants was
   >96% in all three groups ([76]Fig. 2F). CAR expression was 66.4% in
   MSLN.CAR iNK cells and 85.8% in MSLN.CAR-IL-15 iNK cells. In addition
   to the significant increase in differentiation yield, the recovery and
   expansion capacity of differentiated MSLN.CAR-IL-15 iNK cells
   post–cryopreservation was also observed to be superior compared to both
   mock iNK cells and MSLN.CAR iNK cells ([77]Fig. 2G), suggesting that
   IL-15 expression could also benefit the viability and functionality of
   iNK cells after preservation.

Fig. 2. Robust iNK differentiation yield from MSLN.CAR-IL-15–transduced
LiPSC-GR1.1.

   [78]Fig. 2.
   [79]Open in a new tab

   (A) MSLN.CAR-IL-15 piggyBac transposon vector construct encoding the
   CAG promoter, MSLN.CAR (consisted of MSLN-targeted hYP218 scFv, CD8a
   hinge spacer, NKG2D transmembrane domain, 2B4 ICD, and CD3ζ chain
   stimulatory domain), EGFRt, and human IL-15. GMCSFRss, GM-CSF
   receptor-α chain signal sequence directing cell surface expression. (B)
   Genetic modification of LiPSC-GR1.1 cells with MSLN.CAR and
   MSLN.CAR-IL-15 piggyBac transposon systems. MSLN.CAR expression was
   examined by measuring EGFRt expression in transfected iPSCs. (C) Human
   IL-15 production in the supernatant of iPSC culture. ****P < 0.0001.
   (D) NK differentiation yield per EB from MSLN.CAR-IL-15–modified iPSCs
   and control iPSCs. ***P < 0.001. (E) iNK differentiation yield achieved
   with different concentrations of IL-15 (10, 20, 40, and 80 ng/ml) added
   to the mock LiPSC-GR1.1-iNK differentiation media. *P < 0.05. (F) NK
   differentiation and CAR expression in NK were validated by detecting
   CD45, CD56, and EGFRt expression in the cells. (G) Cryopreserved
   differentiated iNK cells were thawed and expanded for 7 days. The fold
   increase of expansion is shown. **P < 0.01. The statistics were
   analyzed using unpaired Student’s t test.

Phenotypic and transcriptome profiling of MSLN.CAR-IL-15–engineered GR1.1-iNK
cells

   We then used flow cytometry to detect NK biomarkers on iNK cells
   freshly harvested at 4 weeks post–initiation of differentiation, as
   compared to the PBNK control. We observed high levels of NKG2D, DNAM-1,
   NKp46, NKp44, CD94, Fas-L, and TRAIL, intermediate level of NKG2A, and
   low levels of CD16 and CD158 (KIR2DL1/S1/S3/S5) on mock iNK, MSLN.CAR
   iNK, and MSLN.CAR-IL-15 iNK cells ([80]Fig. 3A). To compare
   transcriptional profiling of freshly differentiated iNK cells to mature
   PBNK cells, we performed RNA sequencing (RNA-seq) on the RNA extracted
   from healthy donor PBNK, mock iNK, and MSLN.CAR-IL-15–modified iNK
   cells. Principal components analysis (PCA) ([81]Fig. 3B) and
   unsupervised hierarchical clustering of gene expression (fig. S3A)
   showed that three NK populations formed three distinct clusters. PBNK
   cells were strongly separated from mock iNK and MSLN.CAR-IL-15 iNK
   cells along the PC1 axis; mock iNK and MSLN.CAR-IL-15 iNK cells shared
   a similar PC1 value but were clearly separated along the PC2 axis. We
   then analyzed the differentially expressed genes (DEGs) and the
   corresponding gene-enriched pathways. Overall, 4724 DEGs (2373
   up-regulated and 2351 down-regulated) were identified when comparing
   mock iNK cells versus PBNK cells; 1371 DEGs (649 up-regulated and 722
   down-regulated) were identified when comparing MSLN.CAR-IL-15 iNK cells
   versus mock iNK cells (fig. S3B). Kyoto Encyclopedia of Genes and
   Genomes (KEGG) and Hallmark pathway gene set enrichment analysis (GSEA)
   of the DEGs showed a significant increase in pathways including cell
   division, DNA replication, p53 signaling pathway, AGE-RAGE (advanced
   glycation end product-receptor for advanced glycation end product)
   signaling pathway, and cholesterol homeostasis in iNK cells as compared
   to PBNK cells ([82]Fig. 3C). Compared to mock iNK cells, MSLN.CAR-IL-15
   iNK cells showed increased expression of genes associated with cell
   division, DNA replication, nuclear factor κB (NF-κB) signaling,
   transient receptor potential channel calcium signaling, and Wnt
   signaling pathways. We then profiled the expression of NK
   activity–associated genes in different categories ([83]Fig. 3D).
   Although GR1.1-iNK cells showed similar expression of many NK markers
   in common with PBNK cells, they exhibited distinct features. The genes
   that were increased in GR1.1-iNK cells included NCAM1 (CD56), NCR2
   (NKp44), NCR3 (NKp30), KLRC1 (NKG2A), CD96, HAVCR2 (Tim3), CD276
   (B7-H3), TNF, TNFSF14 (LIGHT), CCR1, CCR5, CCR6, CCR8, CXCR3, and
   TNFSF10 (TRAIL), while the genes that were decreased in GR1.1-iNK cells
   included FCGR3A (CD16A), B3GAT1 (CD57) KLRF1 (NKp80), CD160, TIGIT,
   LAG3, CXCR1, CXCR2, CXCR4, and CX3CR1. MSLN.CAR-IL-15 iNK cells
   expressed a similar level of most markers and cytokines, except for
   higher IL15 and lower CD200R1 and CEACAM1, as compared to mock iNK
   cells. In addition, we observed a high level expression of some
   “adaptive NK” feature genes IL32, S100A4, and ITGA1 (CD49A) in
   GR1.1-iNK cells (fig. S3C) ([84]22).

Fig. 3. Characterization of freshly differentiated GR1.1-iNK cells.

   [85]Fig. 3.
   [86]Open in a new tab

   GR1.1-iNK cells were harvested 4 weeks post-EB differentiation. (A) The
   cell surface receptor expression in NK cells was detected via flow
   cytometry. CD56^+CD3^− PBNK and CD56^+ GR1.1-iNK cells were gated for
   the indicated biomarker analyses. The dark gray area represents
   biomarker staining. The light gray area represents isotype control. (B
   to D) RNA-seq of PBNK, mock iNK, and MSLN.CAR-IL-15 iNK cells. Data are
   from one RNA-seq analysis with three biological replicates per group.
   (B) PCA of DEGs. Each dot represents one sample. Each color represents
   an NK population. (C) Bubble plot of GSEA analysis showing the
   highest-ranked top 10 KEGG and Hallmark gene sets significantly
   up-regulated (P < 0.05, adjusted P < 0.1) in mock iNK versus PBNK (top
   panel) and MSLN.CAR-IL-15 iNK cells versus mock iNK cells (bottom
   panel). The x axis represents the enrichment score, and the size of the
   bubble represents the number of genes in the gene set. (D) The heatmap
   shows the gene expression [log[2](TPM + 1)] (where TPM means
   transcripts per million) of NK phenotypic markers in PBNK, mock iNK,
   and MSLN.CAR-IL-15 iNK cells, including NK signature, activating
   receptors, inhibitory KIR, checkpoint, cytokine and receptors,
   chemokine receptors, and cytolytic molecule–related genes.

Enhanced cytotoxicity of MSLN.CAR-IL-15 iNK cells against human solid tumor
cells

   To evaluate the tumor cell killing ability of the GR1.1-iNK cells, we
   performed cytotoxicity assays by coculturing different modified
   GR1.1-iNK cells with target cancer cell lines for 24 hours.
   MSLN.CAR-IL-15 iNK displayed the highest killing ability against all
   the tumor cells we tested (pancreatic cancer KLM-1, mesothelioma
   NCI-meso29, NCI-meso21, NCI-meso63, ovarian cancer OVCAR8, and gastric
   carcinoma cell line NCI-N87), as compared to mock iNK and MSLN.CAR iNK
   ([87]Fig. 4, A and B). MSLN.CAR iNK and MSLN.CAR-IL-15 iNK killed more
   MSLN^+ KLM1-WT than MSLN^− KLM1-KO target cells, suggesting that MSLN
   recognition by CAR increased specific killing ([88]Fig. 4A). On the
   other hand, MSLN.CAR-IL-15 iNK showed a higher killing ability than
   mock iNK and MSLN.CAR iNK when cocultured with MSLN^− KLM1-KO target
   cells, indicating that chronic stimulation by IL-15 led to NK
   activation that promoted cytotoxicity independent of the CAR targeting
   specificity ([89]Fig. 4A). We then tested CD107a expression (granule
   release) of iNK cells. Without target stimulation, we observed slightly
   higher CD107a expression in MSLN.CAR-IL-15 iNK cells than in mock iNK
   and MSLN.CAR iNK cells; when stimulated with KLM-1 tumor cells,
   MSLN.CAR iNK and MSLN.CAR-IL-15 iNK cells demonstrated higher CD107a
   expression, while mock iNK cells had limited CD107a expression
   ([90]Fig. 4C). These results demonstrate that MSLN.CAR-IL15
   modification of the LiPSC-GR1.1 cells not only significantly increased
   the iNK differentiation yield but also markedly enhanced the killing
   function of the differentiated iNK against various tumor cells.

Fig. 4. In vitro tumor killing function of MSLN.CAR-IL-15 GR1.1-iNK cells.

   [91]Fig. 4.
   [92]Open in a new tab

   (A) Cytotoxicity of mock iNK, MSLN.CAR iNK, and MSLN.CAR-IL-15 iNK
   cells against MSLN^+ KLM1-WT and MSLN^− KLM1-KO cells. (B) Cytotoxicity
   of all the iNK cells against multiple MSLN^+ solid tumor cell lines.
   (C) CD107a degranulation signature induced by the coculture of iNK
   cells with KLM-1 tumor cells.

Tumor regression and improved survival in MSLN.CAR-IL15 iNK–treated MSLN^+
tumor–bearing mice

   To evaluate the antitumor efficacy of MSLN.CAR-IL15 iNK cells in vivo,
   we first used the patient mesothelioma-derived xenograft NCI-meso63
   mouse model established previously ([93]23). The luciferase-expressing
   NCI-meso63 tumor cells were injected intraperitoneally 7 days before
   iNK treatment. The mice were then administered two doses of freshly
   differentiated iNK cells on days 0 and 7 and monitored weekly using
   bioluminescence imaging (BLI) ([94]Fig. 5A). Tumor growth was markedly
   decreased in mice receiving MSLN.CAR-IL-15 iNK cells as compared to
   mice receiving mock iNK cells, although the mice receiving mock iNK
   cells did show some antitumor effects when compared to the untreated
   animals because of the CAR-independent activity of NK cells ([95]Fig.
   5, B and C). The median overall survival was 48 days without treatment,
   57 days with mock iNK treatment, and 79 days with MSLN.CAR-IL-15 iNK
   treatment ([96]Fig. 5D). In addition, we also tested the iNK in a more
   aggressive pancreatic tumor KLM-1 mouse model (fig. S4A). Only
   MSLN.CAR-IL-15 iNK treatment showed significant antitumor effects and
   prolonged survival, while mock iNK treatment had a minimal effect in
   this tumor model (fig. S4, B to D). In both mouse models we tested, we
   did not observe any toxicity in the mice treated with CAR-IL-15
   GR1.1-iNK cells. Considering the logistical advantages of
   cryopreservation for large-scale production and distribution of an
   off-the-shelf iNK product, we also evaluated the efficacy of iNK
   expanded from cryopreserved differentiated iNK stocks using the
   NCI-meso63 tumor model (fig. S5, A to C). These expanded MSLN.CAR-IL-15
   iNK cells also exhibited enhanced tumor killing and improved overall
   survival of these mice compared to the expanded mock iNK treatment. The
   antitumor activity was similar to the antitumor effects achieved with
   freshly differentiated MSLN.CAR-IL-15 iNK cells. Given the advantages
   of scalability and logistical ease, we used the expanded iNK cells from
   frozen stocks for our following mechanistic mouse model studies.

Fig. 5. Antitumor efficacy of MSLN.CAR-IL-15 GR1.1-iNK cells in the
NCI-meso63 mouse model.

   [97]Fig. 5.
   [98]Open in a new tab

   (A) Schematic of tumor inoculation and iNK treatments in the NCI-meso63
   tumor model. ip, intraperitoneally; D, day. (B) Tumor growth monitored
   through BLI. P values were analyzed using unpaired Student’s t test.
   (C) Tumor growth based on BLI measurements. The statistics were
   analyzed using unpaired Student’s t test. (D) The median overall
   survival of different groups is shown [P = 0.07, mock iNK versus
   untreated; P < 0.005, CAR-IL-15 iNK versus mock iNK, analyzed using the
   log-rank (Mantel-Cox) test].

Increased infiltration of MSLN.CAR-IL-15 iNK cells in NCI-meso63 tumors

   Treatment of solid tumors is partially constrained by the limited tumor
   homing and persistence of NK cells. To investigate the solid tumor
   infiltration ability of iNK cells, we adjusted the experimental design
   by increasing the NCI-meso63 tumor cell number, treating the tumor with
   a lower number of iNK cells, and doing the iNK transfer on day 16
   posttumor inoculation when the tumor was substantially larger than in
   the preceding experiment ([99]Fig. 6A). With only one low-dose iNK
   treatment, we did not observe significant tumor reduction on day 6
   post-iNK treatment ([100]Fig. 6B). We harvested tumor tissue on day 7
   after iNK treatment and observed a substantial increase in
   hCD45^+hCD56^+mCD45^− cells in the MSLN.CAR-IL-15 iNK–treated tumors
   compared to tumors from mock iNK–treated mice and CAR iNK–treated mice
   ([101]Fig. 6, C and D). Approximately 80% of these hCD45^+hCD56^+ cells
   expressed the MSLN.CAR. There was a small number of hCD45^+mCD45^−
   cells in the spleens of mice given MSLN.CAR-IL-15 iNK cells, and this
   small number was higher than what was observed in spleens from mock
   iNK–treated mice ([102]Fig. 6D). Notably, the percentage of CAR^+ iNK
   cells was much higher in tumor than in the spleen. We next visualized
   iNK tumor infiltration in situ using multiplex immunofluorescence
   staining for MSLN (tumor marker), hCD45/hCD56 (NK marker), and Ki67
   (proliferation marker). hCD45^+ and hCD56^+ cells were significantly
   increased in the MSLN.CAR-IL-15 iNK–treated MSLN^+ tumor as compared to
   tumors from mice receiving mock iNK cells ([103]Fig. 6, E and F).
   hCD45^+ cells colocalized with MSLN^+ tumor cells, suggesting possible
   direct interaction ([104]Fig. 6G). Some of these hCD45^+ cells
   coexpressed Ki67 ([105]Fig. 6, G and H), indicating recent or ongoing
   proliferation of MSLN.CAR-IL-15 iNK cells within the tumors. These
   findings suggest that the MSLN.CAR-IL-15 modification significantly
   improved iNK cell tumor infiltration and persistence.

Fig. 6. Increased tumor infiltration of MSLN.CAR-IL-15 iNK cells in
NCI-meso63 tumor.

   [106]Fig. 6.
   [107]Open in a new tab

   (A) Schematic of iNK treatments and tissue harvest in the NCI-meso63
   tumor model. (B) Tumor growth monitored using BLI. (C) Flow cytometric
   analysis of hCD45^+mCD45^− cells in the tumors on day 7 post-iNK
   treatment (top panel). The hCD45^+mCD45^− cells were further gated to
   analyze hCD56 and hEGFRt expression (bottom panel). (D) Percentage of
   hCD45^+mCD45^− cells among total live single cells isolated from
   spleens and tumors on day 7 (left panels); percentage of
   hEGFRt^+hCD56^+ cells among hCD45^+mCD45^− cells in spleens and tumors
   harvested from MSLN.CAR-IL-15 iNK–treated mice on day 7 (right panel).
   n.s., not significant. (E) Multiplex immunofluorescence imaging of
   tumors harvested from mice treated with mock iNK and MSLN.CAR-IL-15 iNK
   cells on day 7 post-iNK treatment. Scale bars represent 150 μm. (F)
   hCD45^+ cell density in iNK-treated tumor. Three independent whole
   tissue images were analyzed. (G) High-magnification images showing
   Ki67^+ hCD45^+ NK cells colocalized with hMSLN^+ tumor cells in mice
   treated with MSLN.CAR-IL-15 iNK. Scale bars represent 40 μm. (H)
   Percentage of Ki67^+ cells among total hCD45^+ cells in iNK-treated
   tumor. Three independent whole tissue images were analyzed. P values
   were analyzed using unpaired Student’s t test.

Single-cell transcriptional profiling of MSLN.CAR-IL-15 iNK–treated
NCI-meso63 tumors

   NCI-meso63 is a highly metastatic patient mesothelioma-derived
   xenograft cell line established without single-cell cloning, which
   retained its heterogeneity ([108]23). To characterize the
   transcriptional state of the NCI-meso63 tumors during therapy with
   MSLN.CAR-IL-15 iNK cells, as well as that of tumor-infiltrating
   MSLN.CAR-IL-15 iNK cells, we performed single-cell RNA-seq (scRNA-seq)
   on total cells isolated from untreated tumors (three samples),
   iNK-treated tumors (three samples) (same harvest on day 7
   post–MSLN.CAR-IL-15 GR1.1-iNK treatment as shown in [109]Fig. 6A), and
   MSLN.CAR-IL-15 GR1.1-iNK preinfusion product (two samples) ([110]Fig.
   7A). To achieve an unbiased comparison among groups, scRNA-seq data
   from the three groups were pooled for downstream analyses. After
   quality control filtering, we collected a total of 23,287 human cells
   across eight samples (table S1). Through unsupervised clustering, seven
   tumor clusters (11,501 cells) and seven NK clusters (11,786 cells) were
   classified ([111]Fig. 7A) on the basis of the respective canonical
   marker genes of mesothelioma (MSLN, KRT8, and KRT18) and NK cells
   [PTPRC, NCAM1, NKG7, and hYP218 (CAR)] ([112]Fig. 7B), as well as top
   marker genes (up to 100) expressed in each cluster (data S1).

Fig. 7. scRNA-seq analysis of MSLN.CAR-IL-15 iNK cell–treated NCI-meso63
tumors.

   [113]Fig. 7.
   [114]Open in a new tab

   (A) UMAP plots visualizing transcriptome-defined clusters of NK cells
   and tumor cells. (B) Representative signature genes across NK and tumor
   cells. The size of the dots indicates the percent of cells expressing
   the gene, while the color of the dots indicates the average gene
   expression level. (C) UMAP for total cells from untreated and treated
   groups. Cell types are annotated with the same color scheme as in (A).
   (D) Fractions of defined tumor subclusters among tumor cells in treated
   and untreated groups. (E) UMAP of identified aneuploid and diploid
   cells using copyKat analysis (top) and their fractions in each tumor
   cluster (bottom). (F) Expression of representative tumor signature
   genes across tumor subclusters. (G) MSLN expression in tumor
   subclusters in untreated and treated groups. (H) Top up-regulated and
   down-regulated Hallmark pathways in treated versus untreated tumor
   subclusters (adjusted P < 0.25). The x axis represents the enrichment
   scores, and the size of the bubble represents the number of genes in
   the gene set. JAK, Janus kinase; UV, ultraviolet. (I) Expression of HLA
   type I and HLA type II genes in untreated and iNK-treated tumor
   subclusters.

Up-regulation of MHC molecules and reduction of target MSLN in TGF-β–rich
tumors post–MSLN.CAR-IL-15 iNK cell treatment

   We next investigated the features of tumor subclusters. Consistent with
   the flow cytometry and immunofluorescence findings, we observed both
   tumor clusters and NK clusters as the major populations (57% NK cells
   versus 43% tumor cells) in the MSLN.CAR-IL-15 GR1.1-iNK–treated tumors
   as compared to the untreated tumors ([115]Fig. 7C). Using copyKAT copy
   number variant analysis ([116]24), we identified T2, T3, T4, and T6 as
   the malignant tumor clusters ([117]Fig. 7D) with high expression of
   MSLN ([118]Fig. 7B). We then focused on these clusters for the
   downstream analysis. The proportion of T3, which expressed the highest
   level of MSLN among all the tumor subclusters, was significantly
   reduced after MSLN.CAR-IL-15 GR1.1-iNK treatment ([119]Fig. 7E).
   Instead, clusters T2 and T6 emerged as the dominant populations after
   treatment. The four malignant tumor clusters highly expressed
   mesenchymal genes and TGFB1, along with various genes linked to cell
   growth, metabolism, inflammation, and stress responses, highlighting
   tumor heterogeneity and the critical role of transforming growth
   factor–β (TGF-β) in creating an immunosuppressive microenvironment. In
   addition, we observed significant reduction of MSLN expression in all
   the four clusters in the treated group ([120]Fig. 7G), indicating
   effective targeted specific killing mediated by MSLN.CAR-IL-15 iNK.
   Further analysis of Hallmark pathway enrichment revealed that allograft
   rejection, interferon-γ response, and interferon-α response pathways
   were enriched in all the tumor clusters, while tumor necrosis factor–α
   (TNF-α) signaling via NF-κB and epithelial mesenchymal transition
   pathways were distinctly enriched in different subclusters ([121]Fig.
   7H). The expression of both human lymphocyte antigen (HLA) class I– and
   HLA class II–related genes was significantly up-regulated in tumor
   clusters post-iNK treatment ([122]Fig. 7I), which contributed to the
   allograft rejection pathway enrichment score, suggesting that tumor
   cells may regulate iNK cell activity through the major
   histocompatibility complex (MHC) molecules ([123]25, [124]26).

Robust activation, increased cytotoxicity, and cycling ability of
tumor-infiltrating MSLN.CAR-IL-15 iNK cells

   Last, we sought to elucidate the alterations in the transcriptional
   profile of the iNK cells after they encounter the tumor
   microenvironment. The preinfusion iNK cells were mainly composed of
   32.1% NK2, 32.7% NK3, 21.9% NK4, and 12.6% NK6, as compared to the
   tumor-infiltrating (post-infusion) iNK cells primarily composed of
   75.4% NK1, 15% NK5, and 8.5% NK7 ([125]Fig. 8A). The hYP218-scFv
   sequence (CAR) expression was observed to be higher in
   tumor-infiltrating NK1, NK5, and NK7 compared to preinfusion NK
   clusters, suggesting that the high expression of CAR may be essential
   for directing iNK cells into the tumor ([126]Fig. 7B). We observed
   distinct characteristics of preinfusion and postinfusion iNK cells
   based on the expression of cluster-defining marker genes ([127]Fig.
   8B). In the preinfusion NK subclusters, NK2 consisted of highly
   proliferative iNK cells in culture, marked by high expression of cell
   division genes like MKI67 and AURKB; NK3 represented activated NK
   cells, with high expression of IFNG and NK signaling genes such as CLNK
   and FYB1; the NK4 cluster showed elevated expression of inhibitory
   receptors KLRC1/KLRD1 (NKG2A/CD94) and TNFRSF4 (OX40); and NK6
   displayed high levels of mitochondrial genes, indicating cell stress
   and cellular damage. In the postinfusion NK cells, NK1 was the dominant
   subcluster, expressing high levels of cytotoxic effector genes (CST7,
   SRGN, and GZMA), indicating its key role in tumor cell killing. The NK5
   cluster was characterized by high expression of cell division genes,
   including STMN1, RRM2, and ZWINT, along with the tumor-infiltrating NK
   marker RGS1 ([128]27), suggesting that they may correspond to the
   Ki67^+ NK population observed in immunofluorescence imaging. In
   addition, NK5 coexpressed marker genes of both the NK1 and NK2
   clusters, which are associated with proliferation and cytotoxicity,
   highlighting its strong cytolytic activity and ability to persist in
   the tumor environment. NK7, marked by high expression of SEMA4D
   (CD100), PRKCH (PKCθ), and CBLB, exhibited the lowest expression of NK
   function–related genes among the postinfusion clusters. Notably, RGS1,
   CBLB, and FYN were expressed at higher levels, while IFNG was expressed
   at lower levels across all tumor-infiltrating NK clusters compared to
   the preinfusion NK clusters.

Fig. 8. Single-cell transcriptional profiles of NCI-meso63 tumor–infiltrating
CAR-IL-15 iNK cells.

   [129]Fig. 8.
   [130]Open in a new tab

   (A) Fractions of defined NK subclusters in NK cells in preinfusion and
   postinfusion groups. (B) Expression of representative signature genes
   across NK subclusters. (C) Expression of NK-associated biomarker genes
   (activation, inhibitory receptors, cytokines, chemokines, and
   cytolytic-related categories). (D) Up-regulated and down-regulated
   Hallmark pathways in total postinfusion versus preinfusion iNK cells
   (adjusted P < 0.1). (E to G) Heatmap indicating the expression of
   selected gene sets in NK subtypes.

   We next analyzed the expression of NK-associated biomarkers in the NK
   subclusters ([131]Fig. 8C). Tumor-infiltrating clusters NK1 and NK5
   showed high expression of NK-activating receptors, including
   NKG2D-DAP10 (KLRK1/HCST), NKG2C (KLRC2), CD94 (KLRD1), and NKp30
   (NCR3), suggesting that they were robustly activated. Among the
   inhibitory killer cell immunoglobulin-like receptor (KIR) and
   immunosuppressive checkpoint genes, we observed high expression of
   KIR2DL4, intermediate levels of NKG2A (KLRC1), and notable reduction of
   CD96 in the tumor-infiltrating iNK cells, along with negligible
   expression of other checkpoint markers. The protein expression of cell
   surface biomarkers of NKG2D, NKG2C, CD94, NKp30, NKG2A, and KIR2DL4 in
   tumor-infiltrating CAR-IL-15 iNK cells was validated through flow
   cytometry (fig. S6). The chemokine and receptor profiles of
   tumor-infiltrating NK1 and NK5 showed increased expression of CCL5,
   CCL4, XCL1, XCL2, and CXCR3, suggesting that they may be involved in NK
   tumor homing and activation ([132]28, [133]29). In contrast, CCR1 and
   CCR5 were decreased in tumor-infiltrating NK cells, which promote NK
   liver homing ([134]30, [135]31). Regarding cytotoxicity-related genes,
   GZMB and SYTL3 were up-regulated in the NK1 cluster, while PFR1
   (perforin) and TNFSF10 (TRAIL) were down-regulated in
   tumor-infiltrating iNK cells as compared to preinfusion iNK cells.

Reduction of CISH, TGFBR2, and BATF in IL-15–producing tumor-infiltrating iNK
cells to adapt to the hypoxic, TGF-β–rich tumor microenvironment

   The pathway enrichment analysis revealed the pathways significantly
   enriched in tumor-infiltrating NK cells versus preinfusion NK cells,
   including TNF-α signaling via NF-κB, hypoxia, interferon-γ response,
   interferon-α response, apoptosis, heme metabolism, IL-2-STAT5 (signal
   transducers and activators of transcription 5) signaling, estrogen
   response, and p53 pathway ([136]Fig. 8D). In contrast, the cell
   division–related pathway E2F target and G[2]-M checkpoint were
   down-regulated in the tumor-infiltrating NK cells compared to the
   highly proliferative preinfusion NK product. These findings indicate
   that tumor-infiltrating NK cells were actively engaged in complex
   biological processes to mount an effective immune response against the
   tumor and tune their own function and metabolism to adapt to the tumor
   microenvironment. We then analyzed the expression of different gene
   sets to understand the biological processes across different NK
   subclusters.

   We observed significant up-regulation of IL-15 signaling genes in
   tumor-infiltrating NK clusters, including IL15, IL2RB, IL2RG, STAT5B,
   and JAK1 ([137]Fig. 8E). Notably, these genes exhibited higher
   expression levels in the NK1 subcluster compared to NK5. Unexpectedly,
   JAK3 was down-regulated in postinfusion NK cells compared to
   preinfusion iNK cells. In addition, the negative regulators of IL-15
   signaling CISH were markedly reduced, while SOCS1 was markedly elevated
   in postinfusion iNK cells as compared to preinfusion iNK cells. In the
   tumor microenvironment characterized by high levels of TGF-β, NK1 and
   NK5 clusters exhibited low expression of TGFBR2 (with particularly low
   levels in the NK5 cluster) and simultaneously up-regulated negative
   regulators of TGF-β signaling, SKI, SKIL, and TGIF1, indicating that
   these NK clusters may be adapting to mitigate the suppressive effects
   of TGF-β. In contrast, the cluster NK7 with an impaired NK function
   showed high expression of genes involved in TGF-β signaling. We also
   noted that activator protein 1 family transcription factors were
   enriched in postinfusion NK clusters except for the gene BATF, which
   exhibited lower expression in postinfusion NK cells compared to
   preinfusion NK cells.

   A series of genes related to cell cycle arrest, DNA repair, and cell
   survival processes was enriched in postinfusion NK1 and NK5 clusters at
   similar levels ([138]Fig. 8F), contributing to NK response to hypoxia
   and collectively enabling NK cells to thrive in challenging tumor
   microenvironments. The pro-apoptotic genes BCL2L11, BTG1, and DUSP22
   were expressed at lower levels in NK5 compared to NK1. As compared to
   preinfusion NK cells, postinfusion NK cells showed up-regulation of
   genes involved in nutrient uptake and mitophagy, lower levels of
   oxidative phosphorylation (OXPHOS), and similar levels of glycolysis
   ([139]Fig. 8G). Enhanced nutrient uptake in tumor-infiltrating NK
   suggests that NK cells are adapting to their environment by increasing
   the availability of amino acids and other substrates, which can support
   cell survival and function despite other metabolic impairments
   ([140]32). Up-regulation of mitophagy may help remove dysfunctional
   mitochondria, thereby maintaining mitochondrial health and function,
   even when overall oxidative metabolism is compromised ([141]33). This
   may help NK cells maintain their effector functions. We also observed
   high expression of METRNL in postinfusion NK1 and NK5, which may be
   involved in disrupting the mitochondrial function in NK cells
   ([142]34). In addition, NK5 displayed higher levels of OXPHOS and
   glycolysis compared to NK1, aligning with the increased proliferation
   observed in NK5. This metabolic fitness advantage in NK5 may enable
   these NK cells to meet the metabolic demands necessary for both
   cytotoxicity and proliferation, thereby enhancing their sustained
   functionality in the hypoxia-stressed tumor microenvironment.

DISCUSSION

   In this study, we identified LiPSC-GR1.1 as a superior iPSC cell line
   for producing an iNK product using a standardized simple
   differentiation method. Through genetic modification of LiPSC-GR1.1
   with an MSLN-targeting CAR-IL-15 construct, we achieved robust iNK
   differentiation yield. We also demonstrated the characteristics and
   functionality of these MSLN.CAR-IL-15 iNK cells through comprehensive
   in vitro and in vivo experiments, including validating CAR expression
   and profiling iNK cell biomarkers, and demonstrated the tumor killing
   efficacy of these engineered iNK cells against MSLN-positive solid
   tumors. In addition, we observed a significant infiltration of
   MSLN.CAR-IL-15–engineered NK cells into MSLN-positive tumor,
   demonstrating their tumor homing ability and proliferation within the
   tumor microenvironment. Last, we detailed the single-cell
   transcriptional profiles of iNK-treated and untreated tumors, as well
   as of tumor-infiltrating MSLN.CAR-IL-15 iNK cells, which uncovered the
   early responses of MSLN-positive tumor cells to iNK cell–mediated
   killing and provided a comprehensive analysis of characteristics of
   MSLN.CAR-IL-15 iNK cells adapting to the solid tumor context.

   iPSC-NK differentiation is a prolonged and intricate process ([143]35).
   To address the challenges of large-scale production, enrichment and
   expansion of CD34^+ hematopoietic progenitor cells before iNK
   differentiation are frequently used, with further postdifferentiation
   NK cell expansion using feeder cells being a necessary step ([144]36).
   The spin EB method developed by Zhu and Kaufman ([145]21) to generate
   iPSC-NK under feeder-free conditions without CD34^+ enrichment
   significantly simplifies the operational workflow. The drawback is that
   only certain iPSC lines can be efficiently differentiated into NK cells
   using this method and the iNK yield is relatively low. Here, we
   identified LiPSC-GR1.1 as a superior source for producing iNK using
   this simple method. By introducing IL-15 in LiPSC-GR1.1, we increased
   the iPSC-NK differentiation yield by nearly 10-fold, which could
   significantly shorten the expansion and simplify the manufacturing
   process. While IL-15 engineering alone is expected to increase iNK
   yield and enhance NK cell activity, CAR engineering likely further
   augments the selective and sustained cytotoxicity of IL-15–expressing
   iNK cells against tumor cells, as demonstrated by the superior
   antitumor efficacy of CAR-IL-15–expressing cord blood NK cells compared
   to those expressing IL-15 alone ([146]37). IL-15 plays a multifaceted
   role in the regulation of NK cells. While it is essential for NK cell
   development and homeostasis, prolonged exposure to IL-15 can result in
   NK cell exhaustion ([147]38). Of note, our dosing experiment indicated
   that higher IL-15 concentrations did not necessarily result in better
   differentiation outcomes. We observed that increasing IL-15 to a high
   level (e.g., 80 ng/ml) had a detrimental impact on NK cell
   differentiation. Therefore, picking iPSC colonies highly expressing
   IL-15 for NK production may negatively affect subsequent NK cell
   differentiation. We also demonstrate that freshly differentiated
   MSLN.CAR-IL-15 GR1.1-iNK cells can be cryopreserved as a stock and
   subsequently expanded after thawing to an even larger scale without
   loss of functionality. This approach offers a standardized method for
   streamlined iNK production using the LiPSC-GR1.1 cell line. In
   addition, this protocol can be adapted for various CAR genetic
   modifications or strategies to enhance CAR NK therapy.

   The MSLN.CAR-IL-15 iNK product expresses most activating NK receptors
   as well as the inhibitory receptors NKG2A and CD96. Upon entry into
   tumors, MSLN.CAR-IL-15 iNK cells showed increased NKG2D and NKp30
   expression, together with decreased CD96 and NKG2A expression, but
   maintained a high level of KIR2DL4. Since HLA-G expression was also
   observed to be up-regulated in tumors, it will be important to test
   whether blockade of HLA-G/KIR2DL4 signaling can further improve the
   vulnerability of the tumors to iNK treatment. Other well-known
   checkpoint molecules that suppress the NK cell function, such as TIGIT,
   CTLA-4, KLRG1, and PD-1 ([148]39), were found to be negative in
   MSLN.CAR-IL-15 iNK cells, even after infiltrating and interacting with
   tumors, suggesting that the GR1.1-NK product has a superior ability to
   counteract these suppressive mechanisms developed by tumors.

   Strategies targeting CISH and TGFBR2 have been explored to enhance NK
   cell therapies. CISH knockout reprograms iPSC-derived NK cell
   metabolism, increases their sensitivity to IL-15 signaling, and
   improves in vivo persistence in tumor models ([149]40). In addition,
   disrupting the TGF-β pathway through TGFBR2 knockout has been shown to
   enhance NK cell–mediated tumor killing ([150]41, [151]42). While
   targeting the TGF-β pathway can help prevent NK cell dysfunction, once
   the dysfunction has been established, it becomes persistent because of
   significant epigenetic reprogramming. It is reported that IL-15 could
   impair TGF-β signaling in human T lymphocytes ([152]43). It is likely
   that the endogenous IL-15 expression in CAR iNK cells helped counteract
   TGF-β signaling–induced suppression on iNK cells in the tumor. A recent
   study revealed that the activator protein 1 transcription factor BATF
   plays a crucial role in driving epigenetic reprogramming triggered by
   the TGF-β pathway. This reprogramming leads to the up-regulation of
   checkpoint molecules such as CTLA-4, LAG3, PD-1, and NT5E, resulting in
   NK exhaustion ([153]44). We found that MSLN.CAR-IL-15 iNK cells express
   very low level of CISH, TGFBR2, and BATF when infiltrating tumors. This
   unique profile may favor NK cells to effectively navigate the tumor
   microenvironment, thus optimizing their antitumor responses while
   mitigating the risks of exhaustion.

   It is observed that BTG1 was highly expressed in the tumor-infiltrating
   cluster NK1, with an intermediate level expressed in NK5. Recently, a
   phase 1 trial study of GD2-CAR-IL-15 NKT in refractory neuroblastoma
   showed that BTG1 is a key driver of hyporesponsiveness in exhausted NKT
   and T cells ([154]45). Knockdown of BTG1 in GD2-CAR-IL-15 NKT cells
   enhanced tumor elimination in a neuroblastoma mouse mode. Thus, we
   hypothesize that deletion of BTG1 may also enhance iNK antitumor
   efficacy. In addition, the DUSP22 level was found to be significantly
   lower in the cycling NK5 cluster compared to NK1, with the lowest
   expression also observed in the highly proliferative preinfusion NK2
   cluster. Previous studies have shown that DUSP22 inhibits T cell
   expansion by inducing apoptosis, whereas its knockout promotes T cell
   activation ([155]46, [156]47). Therefore, DUSP22 could serve as another
   potential target for future research aimed at improving iNK
   persistence.

   Despite the enhanced antitumor activity of MSLN.CAR-IL-15 iNK cells
   observed in different tumor models, tumor relapse eventually occurred
   even with multiple administrations of the product. MSLN.CAR-IL-15 iNK
   mediated specific killing against MSLN-positive tumor cells and led to
   the selective expansion of MSLN^low tumor cells, highlighting that
   antigen loss is always a concern for CAR-directed cellular therapy
   ([157]48). In addition, tumors responded to NK treatment by expressing
   TGF-β, up-regulating MHC I/II molecules and genes involved in the
   interferon-γ response and TNF-α signaling via NF-κB pathways. While MHC
   I can directly limit NK cell activation via inhibitory receptors, MHC
   II can create a regulatory network that suppresses NK cell activity and
   promotes tumor immune evasion ([158]49, [159]50). At the same time,
   such MHC molecule up-regulation on tumor cells may enhance tumor
   recognition by T cells. A combination therapy study has shown that
   iPSC-NK helped recruit T cells to the tumor site and cooperated with T
   cells to exert enhanced tumor killing ([160]36). NK cells may regulate
   a range of other immune cells, including dendritic cells and
   macrophages ([161]51). IL-15 produced by MSLN.CAR-IL-15 iNK cells may
   also enhance the survival and functional capabilities of tumor-resident
   NK cells and CD8^+ T cells ([162]52). Future experiments using the
   constantly improving humanized mouse models may help address the
   cross-talk between NK cells and other immune cells in the tumor
   microenvironment.

   In summary, we report a proof-of-concept approach for robustly
   differentiating iNK from the LiPSC-GR1.1 iPSC line genetically modified
   to stably express MSLN.CAR-IL-15. We demonstrate the substantial
   antitumor effects of these iNK cells against solid tumors, conduct a
   deep characterization of their phenotypic and transcriptional states,
   and report how the interplay between tumor cells and MSLN.CAR-IL-15 iNK
   cells affects both populations during therapy. Our work supports the
   clinical translation of MSLN.CAR-IL-15–engineered GR1.1-iNK for the
   treatment of patients with advanced treatment refractory solid tumors
   that highly express MSLN.

MATERIALS AND METHODS

Study design

   The goals of this study included the development of a robustly
   differentiated MSLN-targeting CAR-IL-15 iNK product using the
   cGMP-manufactured iPSC line LiPSC-GR1.1 to phenotypically characterize
   MSLN.CAR-IL-15 GR1.1-iNK and to investigate the antitumor efficacy and
   the mechanism of action of MSLN.CAR-IL-15 GR1.1-iNK cells in
   MSLN-positive solid tumors. We tested six iPSC lines for iNK
   differentiation using the spin EB method. LiPSC-GR1.1 was identified as
   a superior iPSC line for iNK differentiation and selected for genetic
   modification using an MSLN.CAR-IL-15–expressing PiggyBac
   transposon/transposase system we created. LiPSC-GR1.1 iPSCs stably
   expressing MSLN.CAR-IL-15 were differentiated to GR1.1-iNK cells. We
   then determined NK cell receptor expression levels on GR1.1-iNK cells
   by flow cytometry and analyzed the transcriptome of GR1.1-iNK cells
   compared to PBNK cells isolated from healthy donors using bulk RNA-seq.
   The antitumor efficacy of MSLN.CAR-IL-15 GR1.1-iNK against
   MSLN-positive solid tumors was evaluated using in vitro and mouse tumor
   models. In addition, we investigated the infiltration of MSLN.CAR-IL-15
   GR1.1-iNK in solid tumor tissues by flow cytometry and multiplex
   immunofluorescence. Last, to better understand tumor responses to
   MSLN.CAR-IL-15 GR1.1-iNK, as well as NK signatures associated with
   tumor homing and antitumor effects, we performed scRNA-seq analysis on
   the total cells isolated from MSLN.CAR-IL-15 GR1.1-iNK–treated tumor
   tissues, untreated tumor tissues, and preinfusion MSLN.CAR-IL-15
   GR1.1-iNK product.

Mice, tumor cell lines, iPSC lines, and reagents

   NOD/SCID/γc^−/− (NSG) mice were obtained from the Jackson Laboratory
   (Bar Harbor, ME). All mice were maintained in a dedicated pathogen-free
   environment following NIH guidelines. Animal procedures reported in
   this study were conducted under an animal study protocol (no. TGOB-001)
   approved by the by the Animal Care and Use Committee of the National
   Cancer Institute and in accordance with federal regulatory requirements
   and standards. Established tumor cell lines KLM-1 MSLN-WT/KO [received
   from C. Alewine, CCR, National Cancer Institute (NCI), NIH, Bethesda,
   MD], OVCAR-8 (received from H. Kobayashi, CCR, NCI, NIH, Bethesda, MD),
   N87 (obtained from American Type Culture Collection, Manassas, VA), and
   the EBV-transformed B cell line SMI-LCL (received from R. W. Childs,
   Cellular and Molecular Therapeutics Branch, National Heart Lung and
   Blood Institute, NIH, Bethesda, MD) ([163]53) were maintained in RPMI
   1640 medium with 10% fetal bovine serum (FBS), 2 mM l-glutamine, and
   100 U penicillin-streptomycin. Early-passage mesothelioma cell lines
   (NCI-meso21, NCI-meso29, and NCI-meso63) were maintained in RPMI 1640
   medium with 20% FBS, 2 mM l-glutamine, and 100 U
   penicillin-streptomycin. The mesothelioma cell lines were established
   from ascites or pleural fluid obtained from patients with mesothelioma
   treated at the NCI (Bethesda, MD) under an Institutional Review
   Board–approved protocol ([164]ClinicalTrials.gov [165]NCT 01950572).
   The methods for the establishment of primary culture cell lines have
   been described previously ([166]54). KLM-1 MSLN-WT/KO, OVCAR-8,
   NCI-meso29, and NCI-meso63 were stable luciferase-expressing cell lines
   ([167]12). The iPSC line LiPSC-GR1.1 was generated at Lonza
   Walkersville Inc., as previously described ([168]9, [169]10), and
   distributed through RUCDR Infinite Biologics at Rutgers University
   (https://stemcells.nindsgenetics.org/). The iPSC lines NCRM1, NCRM2,
   NCRM4, NCRM5, and NCRM6 were obtained from the iPSC core of the
   National Heart, Lung, and Blood Institute (NHLBI). These lines are
   available to the research community. All the iPSC lines were generated
   from healthy donor–derived cord blood CD34^+ cells and cryopreserved in
   CryoStor CS10 (STEMCELL Technologies). Recombinant human cytokines used
   in NK differentiation and culture included VEGF and BMP-4 (both
   obtained from R&D Systems Inc.), as well as SCF, IL-3, IL-15, IL-7,
   Flt3 ligand, IL-2, and IL-21 (all obtained from PeproTech). The
   antibodies used for flow cytometry and immunofluorescence are included
   in table S2.

CAR construct cloning

   Lentiviral packaging plasmid psPAX2 expressing CD8.BB^ICD.Z.CAR
   (composed of MSLN-targeted hYP218 scFv, CD8a hinge spacer, CD8 TM,
   4-1BB costimulatory ICD, and stimulatory domain CD3ζ chain), as well as
   a truncated human EGFR polypeptide (EGFRt), was created previously
   ([170]12). The fragment of IL-15 plus T2A between EGFRt and IL-15 was
   synthesized as a gene fragment (by GeneWiz) and subcloned into the
   psPAX2 vector through seamless cloning using the NEBuilder HiFi DNA
   Assembly Cloning Kit (New England Biolabs) according to the
   manufacturer’s instructions. CD8 TM and 4-1BB-ICD were then replaced
   with a gene fragment of NKG2D TM and 2B4-ICD (synthesized by
   Genescript) through seamless cloning to generate plasmid psPAX2
   expressing NKG2D/2B4^ICD.Z.CAR as well as NKG2D/2B4^ICD.Z.CAR-IL15. The
   PiggyBac transposon vector PBCAG-eGFP with a CAG promoter was a gift
   from J. Loturco (Addgene plasmid no. 40973) ([171]55). To create
   CAR-expressing PBCAG plasmids, the expression cassette encoding
   NKG2D/2B4^ICD.Z.CAR-EGFRt or NKG2D/2B4^ICD.Z.CAR-EGFRt-IL-15 was
   amplified from the template plasmid psPAX2 through polymerase chain
   reaction and then subcloned into the Xma I and Not I sites of the PBCAG
   vector through seamless cloning.

Cell culture of human iPSCs

   iPSC lines were maintained feeder free in complete Essential 8 Flex
   medium (Thermo Fisher Scientific) in six-well plates coated with
   truncated vitronectin recombinant human protein (Thermo Fisher
   Scientific, A14700) at a concentration of 0.5 μg/cm^2. They were
   routinely passaged as small clumps using 0.5 mM EDTA in
   phosphate-buffered saline (PBS) at a split ratio of 1:6 to 1:10 every 3
   to 4 days after reaching 60 to 80% confluence. After EDTA treatment,
   hiPSCs were transferred to new vitronectin-coated plates in fresh
   medium supplemented with the ROCK inhibitor (ROCKi) Y-27632 (10 μM, R&D
   Systems Inc.). The next day, the medium was changed to hiPSC medium
   without ROCKi.

NK cell differentiation from human iPSCs

   The derivation of NK cells from iPSCs has been previously described
   ([172]21). Briefly, 8000 TrypLE-adapted iPSCs were seeded in 96-well
   round-bottom plates with APEL medium containing SCF (40 ng/ml), VEGF
   (20 ng/ml), BMP-4 (20 ng/ml), and Y-27632 (10 μM). After 6 days of
   hematopoietic differentiation, spin EBs were washed once with NK
   differentiation medium and then transferred into six-well plates with
   NK differentiation medium containing IL-3 (5 ng/ml; first week only),
   IL-15 (10 ng/ml), IL-7 (20 ng/ml), SCF (20 ng/ml), and Flt3 ligand (10
   ng/ml) for 4 weeks. Half-medium changes were performed twice a week. NK
   cells were harvested 4 weeks post-NK differentiation. For the IL-15
   concentration test, different concentrations of IL-15 (10, 20, 40, and
   80 ng/ml) were used during 4 weeks of NK differentiation. Three
   independent experiments were performed.

Stable CAR constructs expression in the LiPSC-GR1.1 line

   LiPSC-GR1.1 cells were dissociated as a single-cell suspension using
   TrypLE and then washed with fresh E8 medium supplemented with ROCKi. A
   total of 0.2 million iPSCs was harvested through centrifugation,
   resuspended in 20 μl of P3 primary solution containing 1 μg of
   CAR-expressing PBCAG transposon vector plus 0.25 μg of Super PiggyBac
   Transposase (cat. no. PB210PA-1, System Bioscience), and transferred
   into nucleocuvette strips. Nucleofection was performed using Lonza
   4D-Nucleofector with pulse setting CA-137 according to the
   manufacturer’s guidelines. The iPSCs nucleofected without PBCAG
   transposon were used as mock control. CAR-expressing iPSCs were
   enriched through FACS of EGFR^+ iPSCs during days 7 to 9
   postnucleofection.

IL-15 ELISA

   iPSCs were seeded at 3 × 10^5 cells per well in a six-well plate and
   cultured for 72 hours without a medium change. Then, the supernatants
   of the iPSC culture were collected. IL-15 was analyzed using the ELISA
   MAX Deluxe Set Human (BioLegend) according to the manufacturer’s
   guidelines. Three independent experiments were performed.

iNK cryopreservation and expansion

   Freshly differentiated iNK cells were harvested 4 weeks
   postdifferentiation and cryopreserved in CryoStor CS10 at 1.0 × 10^7
   cells/ml. For expansion testing, iNK cells were initially thawed and
   cultured in complete Xpander media supplemented with 10%
   heat-inactivated human AB serum, IL-2 (100 U/ml), IL-21 (100 ng/ml),
   and irradiated [100 gray (Gy)] SMI-LCL feeder cells (7.5:1 feeder-to-NK
   cell ratio) with a starting cell concentration of 1.0 × 10^6 cells/ml.
   Expanded iNK cells were harvested on day 7 postexpansion for the
   following studies.

PBNK isolation

   Human peripheral blood mononuclear cells (PBMCs) from healthy donors
   were obtained from the NIH Clinical Center Department of Transfusion
   Medicine under the NIH Institutional Review Board–approved and NIH
   Institutional Review Board–consented healthy donor program. PBMCs were
   isolated by density gradient centrifugation on Histopaque 1077
   (Sigma-Aldrich). Briefly, 15 ml of the collected blood sample was
   diluted in an equal volume of PBS and layered over onto an equal volume
   of Histopaque 1077 (Sigma-Aldrich). Gradients were centrifuged at 400g
   for 30 min at room temperature. The PBMC interface was carefully
   removed by pipetting and washed twice with PBS by centrifugation at
   300g for 10 min. For the cell isolation, the cell number was determined
   using a Countess Automated Cell Counter (Invitrogen). Primary NK cells
   were purified from PBMCs using the human NK Cell Isolation Kit
   (Miltenyi Biotec). The NK purity was confirmed as 80 to 95% by flow
   cytometry. Three independent experiments were performed.

In vitro NK cell cytotoxicity assays

   NK cell cytotoxicity was determined in vitro in a coculture of NK cells
   and tumor cells by measuring direct killing of MSLN-expressing tumor
   cells by the NK cells. Briefly, 100 μl of complete RPMI 1640 media
   containing 5000 tumor cells was seeded in a 96-well U-bottom plate.
   Four hours after seeding the tumor cells, 100 μl of media containing NK
   cells was added into the coculture across a range of effector-to-target
   (E:T) ratios. The luciferase signal from the remaining live tumor cells
   was determined 24 hours postcoculture using the Luciferase Assay System
   (Promega, catalog no. E1501). Percent killing was calculated using the
   following formula: % Killing = 100 × [1 − Relative light units (RLUs)
   from coculture wells/RLU from target-alone wells]. Two or three
   independent experiments were performed.

Mouse tumor model study

   To evaluate iNK cell function against NCI-meso63 tumor in vivo, 5- to
   6-week-old female NSG mice (the Jackson Laboratory) were
   intraperitoneally injected with 3 × 10^6 luciferase-expressing
   NCI-meso63 tumor cells on day −7. BLI (PerkinElmer, IVIS Lumina III In
   vivo Imaging System) was performed on day −1. The mice were then
   conditioned with 225-cGy radiation and divided into groups with similar
   starting BLI values. The two doses of 8 × 10^6 iNK cells (harvested
   either on day 28 post-NK differentiation or 1 week postexpansion from
   frozen iNK stock) were intraperitoneally injected into tumor-bearing
   mice on days 0 and 7. Tumor growth was monitored weekly via BLI. To
   evaluate the iNK cell function against KLM-1 tumor cells in vivo, NSG
   mice were intraperitoneally injected with 5 × 10^5
   luciferase-expressing KLM-1 tumor cells on day −5. BLI imaging was
   performed on day −1. The mice were then conditioned with 225-cGy
   radiation and randomly divided into groups with similar starting BLI
   values. Two doses of 1 × 10^7 and one dose of 5 × 10^6 iNK cells
   harvested during days 28 to 35 post-NK differentiation were
   intraperitoneally injected into tumor-bearing mice on days 0 and 7, and
   tumor growth was monitored weekly via BLI (n = 5 to 7 mice per group
   for the tumor model efficacy study). To investigate the tumor response
   and iNK cell tumor infiltration in the NCI-meso63 tumor model, we
   injected the mice with 6 × 10^6 NCI-meso63 tumor cells on day −16 and
   treated the mice with 6 × 10^6 iNK cells (product of 1-week expansion
   from frozen iNK stock) on day 0. BLI was performed to monitor the tumor
   growth on days −1 and 6. All the tissues and blood were harvested on
   day 7 post-iNK treatment. The tumors were harvested for flow cytometry,
   immunofluorescence, and scRNA-seq analysis. The spleens were harvested
   for flow cytometry analysis (n = 3 mice per group). Two or three
   independent experiments were performed.

Bulk mRNA sequencing

   Total RNA was isolated from freshly isolated PBNK cells (from healthy
   donors) and freshly differentiated iNK cells harvested 4 weeks post-NK
   differentiation using TRIzol reagent according to the manufacturer’s
   guide (Invitrogen). The quantity and quality of RNA were analyzed using
   Agilent TapeStation. mRNA sequencing samples were pooled and sequenced
   on NextSeq 2000 P2 using the Illumina stranded mRNA ligation kit having
   77 to 88 million pass filter reads with more than 92% of bases above
   the quality score of Q30. Reads of the samples were trimmed for adapter
   quality bases using Cutadapt before alignment with the reference genome
   (hg38) and the annotated transcripts using STAR. The average mapping
   rate of all samples was 95%. Unique alignment was above 90%. There were
   3.33 to 4.82% unmapped reads. The mapping statistics were calculated
   using Picard software. The samples had 0.47% ribosomal bases. Percent
   coding bases were between 42 and 51%. Percent untranslated region bases
   were 31 to 33%, and mRNA bases were between 74 and 83% for all the
   samples. Library complexity was measured in terms of unique fragments
   in the mapped reads using Picard’s MarkDuplicate utility. The samples
   had 70 to 75% nonduplicate reads. In addition, gene expression
   quantification analysis was performed for all samples using STAR/RSEM
   tools. The normalization of the RNA counts and DEG analysis were
   performed using the Limma Voom R package. Transcripts with
   significantly differential expression (P < 0.05, log[2] fold change >1)
   were used for subsequent pre-ranked GSEA using the EasyGSEA tool
   ([173]56) for Hallmark and KEGG gene set scoring.

Single-cell isolation from mouse tissues

   From NCI-meso63 tumor–bearing mice, spleens were harvested in cold PBS,
   gently mashed, and filtered through a two-chamber sterile filter bag
   (Thermo Fisher Scientific), and the resulting single-cell suspension
   was collected from the other side of the bag. The mouse peritoneal
   tumors were harvested in cold serum-free RPMI 1640 medium, and total
   cells were isolated using a tumor dissociation kit (Miltenyi Biotec)
   according to the manufacturer’s instructions. Briefly, tumor tissues
   were digested in serum-free RPMI 1640 medium supplemented with human
   tumor dissociation kit enzymes (Miltenyi Biotec) at 37°C under
   continuous rotation for 90 min. Dissociated single-cell suspensions
   were then filtered through a 70-μm mesh filter and pelleted at 300g for
   5 min. For flow cytometry analysis, all the single cells isolated from
   the spleens and tumors were washed with FACS buffer (PBS + 0.5% FBS + 1
   mM EDTA). For scRNA-seq study, the single cells isolated from the
   tumors were washed with PBS + 0.04% bovine serum albumin (BSA). The
   single cells prepared were then used for the following study.

Flow cytometry

   For human iPSCs, iPSCs were dissociated into a single-cell suspension
   with TrypLE, harvested, washed, then resuspended in E8 medium
   containing Essential 8 Flex medium, and incubated with fluorescent
   conjugated antibodies at 4°C for 30 min. For iNK cells, cells were
   harvested from the cultured supernatant, washed, then resuspended in
   PBS, and incubated with fluorescent conjugated antibodies at 4°C for 30
   min. For the cells harvested from NK/tumor coculture or isolated from
   peritoneal tumors and spleens, cells were first stained with the Zombie
   Aqua Fixable Viability Kit (BioLegend, cat. no. 423102) at room
   temperature for 30 min and then incubated with 1% FBS at 4°C for 15 min
   to block nonspecific binding sites before incubating with fluorescent
   conjugated antibodies at 4°C for 30 min. For CD107a degranulation
   assay, NK cells were incubated with or without KLM-1 cells at 2:1 E:T
   ratios in the presence of an anti-CD107a antibody for 1 hour, followed
   by addition of GolgiStop (BD Biosciences, cat. no. 554724) for an
   additional 2 hours of incubation. At the completion of incubation,
   cells were washed with PBS, then first stained with the Zombie Aqua
   Fixable Viability Kit at room temperature for 30 min, and then
   incubated with 1% FBS at 4°C for 15 min to block nonspecific binding
   sites before incubating with fluorescent conjugated antibodies at 4°C
   for 30 min. Flow cytometry detection was performed on CytoFLEX LX and
   analyzed using FlowJo software. Two or three independent experiments
   were performed.

Multiplex immunofluorescence

   Fresh isolated tumor tissues were fixed in BD Cytofix/Cytoperm solution
   (BD Bioscience, cat. no. 554722) diluted 1:4 in PBS for 2 days at 4°C,
   washed three times in PBS, and maintained in PBS for 10 hours at 4°C.
   Then, the tissue blocks were transferred into PBS containing 30%
   sucrose for 2 days until the tissue blocks sank to the bottom of the
   30% sucrose solutions. The tissues were then embedded in Scigen
   Tissue-Plus O.C.T. Compound (Thermo Fisher Scientific, cat. no.
   23-730-571) and stored at −80°C before use. Cross sections
   (12-μm-thick) of the tumor tissue blocks were cut by using a Leica
   CM1950 cryostat. The O.C.T. compound of tissue sections was removed by
   washing the tissue sections with PBS for 30 min before being blocked in
   1% BSA and 0.1 M tris buffer (pH 7.4) containing 0.3% Triton X-100 for
   30 min at room temperature. A microwave-assisted immunofluorescence
   staining protocol was adapted for MSLN detection by using a mouse
   anti-human MSLN antibody amplified by a fluorescent dye–directly
   conjugated donkey anti-mouse secondary antibody ([174]57). The sections
   were then blocked in 1% BSA and 1% mouse serum in 0.1 M tris buffer (pH
   7.4) containing 0.3% Triton X-100 and stained with the following
   fluorescent dye–directly conjugated primary antibodies: CD45-AF532,
   CD56-PE/CF594, and Ki67-AF700 together with nuclear dye Helix NP-NIR
   (BioLegend, cat. no. 425301). Then, tissue sections were washed in 0.1
   M tris buffer (pH 7.4) and mounted with SlowFade Gold Antifade Mounting
   solution (Thermo Fisher Scientific, cat. no. S36937) using no. 1.5
   cover glass (VWR, cat. no. 48393-241). Digital scan of stained tumor
   sections was performed using an inverted Leica TCS SP8 X confocal
   system equipped with an 80-MHz pulsed white light laser, four
   gallium-arsenide Hybrid Detectors, and one multialkali photomultiplier
   tube with spectral detection capability. A 40× (numerical aperture,
   1.3) oil emersion objective lens was used for scanning the sections
   with a pixel size of 568.74 mm by 568.74 mm and a pixel dwell time of
   1.2 ms. Digital images acquired by the Leica confocal system were tile
   stitched and then processed to correct signal spillovers from nearby
   channels and the autofluorescence signal within the tumor sections by
   using the manual unmixing method of the Automatic Dye Separation
   function within the Leica Application Suite X (LAS X) software package
   (version 4.4.0.24861). The output images (.lif files, Leica file
   format) were converted into Imaris version 5.5 files by Imaris software
   (version 9.5.0, Bitplane) and imported into Imaris. Images of
   individual channels were passing through a Gaussian filter to remove
   random noise before pseudocolor assignment and visualization. In all
   image processing steps, the image size and aspect ratio were maintained
   identical by keeping the pixel size unchanged. Two or three independent
   experiments were performed.

Single-cell RNA sequencing

   Eight single-cell samples were prepared for scRNA-seq, including two
   samples of preinfusion product, three samples of total cells isolated
   from the untreated tumors (individual mouse as one sample), and three
   samples of total cells isolated from the MSLN.CAR-IL-15 iNK–treated
   tumors (four mice pooled as one sample). Dead cells and debris were
   removed using the Dead Cell Removal Kit (Miltenyi Biotec, 130-090-101)
   following the manufacturer’s instructions. Single-cell suspensions were
   washed once with ice-cold PBS with 0.04% BSA by centrifugation at 350g
   and gently resuspended in fresh buffer. The cellular samples were
   loaded in the lane according to the 10x Genomics Chromium Next GEM
   Single Cell 5′ v2 (Dual Index) user guide with a single capture lane
   per sample. The recovery was 6000 cells per lane. Cell partitioning
   completed with a uniform emulsion consistency, and the reverse
   transcription polymerase chain reaction was run. All subsequent steps
   of library preparation and quality control were performed as described
   in the 10x Genomics user guide. Then, the sequencing was performed on
   the NovaSeq 6000 system with the NovaSeq S2 Reagent kit version 1.5
   (100 cycles) following the cycling parameters: 26 base pairs (bp;
   read1), 10 bp (index 1), 10 bp (index2), and 90 bp (read2).

scRNA-seq data processing

   Base calling was performed using RTA 3.4.4, demultiplexing was
   performed using CellRanger version 8.0.0 (Bcl2fastq 2.20.0), and
   alignment was performed using cellranger version 8.0.0 (STAR 2.7.2a).
   Sequenced reads were aligned to a custom human GRCh38 and mouse GRCm39
   reference sequence made by adding the hYP218-scFv sequence to the
   refdata-gex-GRCh38_and_GRCm39-2024-A reference. Unique molecular
   identifier (UMI)–adjusted aligned reads were used to generate a
   single-cell barcode and gene expression matrix for downstream analysis.

   The preprocessed gene expression matrix generated by the CellRanger
   (10x Genomics) pipeline was imported into Seurat (version 5.1.0).
   Putative droplet doublets were detected using scDblFinder (version
   1.16.0) with default parameters. Doublet cells were removed from each
   sample separately. Samples were merged. As a quality control step, all
   cells expressing <200 genes were removed, as well as cells that
   contained <500 UMIs and >20% mitochondrial counts. The filtered gene
   expression matrix was normalized using the NormalizeData function with
   default parameters. A total of 3000 variable genes was identified using
   the FindVariableFeatures function with the vst method. PCA was applied
   with the RunPCA function to reduce dimensionality after regressing for
   the number of UMIs (counts) and percentage mitochondrial genes with the
   ScaleData function. The 15 most informative principal components were
   used for clustering and uniform manifold approximation and projection
   (UMAP) for dimension reduction. Shared nearest neighbors were computed
   with the FindNeighbors function, and cells were then clustered using
   the FindClusters function with the resolution set to 0.8. A UMAP was
   generated with the RunUMAP function. After the first round of
   unsupervised clustering, we calculated the percentages of human and
   mouse genes in a cell in each cell cluster. The mouse cell clusters
   were removed from downstream analysis, which focused on human cells.
   Genes of the mouse genome were also removed from human cells. Then, we
   performed a second round of unsupervised clustering on human cells. The
   second-round clustering procedure was performed the same as the
   first-round clustering with a 0.7 resolution and dimensionality
   reduction with top 20 principal components for visualization. Human NK
   cells and tumor clusters were annotated using curated in-house
   signatures. The cells annotated as B cells (LCL NK feeder cells) were
   removed from the analysis. R (version 4.3.1) and RStudio (2023.09.1
   Build 494, “Desert Sunflower” Release) were used for analysis. DEGs
   were performed using the FindMarkers function with the Wilcoxon rank
   sum test, with the logfc.threshold, min.pct, and only.pos parameters
   set to log[2](1.5), 0.1, and F, respectively. DEGs were filtered with a
   cutoff of either pct.1 or min pct.2 > 0.25 and adjusted P < 0.05 for
   GSEA. Gene set variation analysis implemented in the fgsea R package
   (version 1.28.0) was used for GSEA with 10,000 permutations performed.
   The Hallmark gene sets were retrieved from the msigdbr R package
   (version 7.5.1). The input genes were calculated by FindMarkers and
   sorted by avg_log2FC, and enriched pathways with less than three genes
   in leadingEdge or adjusted P > 0.25 were filtered out. To recognize
   malignant cell subsets, we used copyKAT (version 1.1.0) to determine
   human normal cells and malignant cells with the hg20 genome and other
   parameters set as default ([175]24). The identified aneuploid cells
   were extracted as malignant cells.

Statistical analysis

   Unpaired Student’s t test was used to compare the differences between
   groups for in vitro and in vivo studies. Mouse survival differences
   were compared using the log-rank (Mantel-Cox) test. Statistical
   analysis was performed using GraphPad Prism 6.0. P < 0.05 was
   considered statistically significant.

Acknowledgments