The NOTCH–FOXM1 Axis Plays a Key Role in Mitochondrial Biogenesis in the Induction of Human Stem Cell Memory–like CAR-T Cells

Chimeric antigen receptor–engineered T cells (CAR-T) transfer has been developed as a potent therapeutic approach for hematologic cancers (1–3). A recent clinical trial revealed that CAR-T therapy for leukemia achieved high rates of short-time complete remission, but was not able to sustain 1-year relapse-free survival in a proportion of patients (4). To manufacture CAR-T cells, T-cell receptors (TCR) and cytokines continuously stimulate T cells, resulting in T-cell exhaustion and dysfunction (5). These dysfunctional or exhausted CAR-T cells are easily expired and elevate the risk of relapse (6). It has been reported that CAR-T cells from completely responding patients with chronic lymphocytic leukemia (CLL) were enriched in memory-related genes (7) and sustained remission was associated with an elevated frequency of stem cell–like memory T (T_SCM) phenotypes of CAR-T cells in patients (7, 8).

To improve the efficacy of CAR-T cells, various methods have been tried for tuning T-cell differentiation and memory formation (9–12). Less-differentiated memory phenotypes such as central memory T (T_CM) and T_SCM phenotypes govern high proliferative capacity, long-term survival, and in vivo durability (13, 14). Especially, T_SCM cells provide long-term antitumor response and are considered to be fit for CAR-T therapy (15).

In general, differentiated T cells use glycolysis for proliferation in response to antigens, whereas memory T cells prefer to use fatty acid oxidation (FAO)–dependent oxidative phosphorylation (OXPHOS; refs. 16, 17), and OXPHOS-mediated ATP production contributes to a favorable and durable antitumor response in tumor microenvironments (18–20). Peroxisome proliferator activated receptor gamma coactivator 1α (PGC1A)–driven mitochondrial biogenesis and OXPHOS support a strong antitumor effect (21). Furthermore, a strong link between FAO-dependent OXPHOS and memory T-cell development has been reported. For example, cell-intrinsic lipolysis following intrinsic fatty acid synthesis is required for memory formation (17). Mitochondrial biogenesis and morphological changes by CD28 signaling promote memory T-cell generation (22), and selective agents modifying mitochondrial metabolism conserve less-differentiated phenotypes of T cells that enhances the antitumor effect (23–25).

Although various in vitro methods for T_SCM generation have been reported, most of them generate T_SCM cells from naïve T cells (12, 14). We previously reported that the coculture with NOTCH ligand-expressing feeder cells converts fully activated or more differentiated memory cells into induced T_SCM-like (iT_SCM) cells (26, 27). These “functional” iT_SCM cells expanded more efficiently in response to antigen re-stimulation than any other T-cell subsets. iT_SCM cells also showed a long-lived and self-renewing potential, were resistant to cell-cycle arrest and apoptosis after TCR stimulation, and exhibited a superior antitumor ability in comparison with naïve and other memory T-cell subsets. However, why NOTCH signaling redirects fully activated cells toward T_SCM-like cells remains unclear. Here, we show that NOTCH and its downstream gene, forkhead box M1 (FOXM1) facilitate mitochondria biogenesis, fatty acid synthesis and OXPHOS, resulting in iT_SCM formation. CAR-iT_SCM induction by FOXM1 gene overexpression enhanced antitumor effects compared with conventional CAR-T. We also propose that FOXM1-mediated metabolic fitness can provide a novel CAR-T strategy.

Authenticated and Mycoplasma-tested OP9 (RCB1124), NALM6 (RCB1933), and K562 (RCB0027) cells were purchased from RIKEN Bio Resource Center (Tsukuba, Ibaraki, Japan). We cloned human Delta-like 1 (hDLL1) and produced hDLL1 expression vector with lentiviral vector CS-II Human Delta-like 1–expressing-OP9 cells (OP9-hDLL1) were cultured in alpha-minimum essential medium (α-MEM; Thermo Fisher Scientific) supplemented with 20% FBS and 1% penicillin/streptomycin. Lymphoblastoid cell lines (LCL) were established from peripheral blood mononuclear cells (PBMC) provided from three healthy donors, as described previously (28). Authentication and Mycoplasma testing of LCLs were not conducted. All cell lines were frozen down at early passages (<7) and used in the experiments within 5 passages after thawing.

Human PBMCs were prepared by specific gravity centrifugal methods from peripheral blood. Peripheral blood was provided by three healthy donors who were EBV-seropositive [VCA-IgG (+), EBNA (+)]. We prepared CD8⁺ T cells from PBMCs with depletion of non-CD8α⁺ T cells using a human CD8⁺ T-cell isolation kit (#130-096-495, Miltenyi Biotec). To induce EBV-specific CD8⁺ T cells, CD8⁺ T cells were labeled with carboxyfluorescein diacetate succinimidyl ester (CFSE) or Cell Trace Far Red and were subsequently cocultured with 40 Gy-irradiated autologous LCLs for 7 days. To induce anti-CD3/CD28 Abs-activated CD8⁺ T cells, CD8⁺ T cells were stimulated with Dynabeads Human T-Activator CD3/CD28. This study was approved by the Ethics Committee of Keio University (Tokyo, Japan; approved number 20120039).

We previously described a method for human iT_SCM generation (26). In brief, to activate Notch signaling, activated T cells were cocultured with OP9-hDLL1 cells. Human T cells and OP9-hDLL1 cells were cocultured with human IL7 (PeproTech; 10 ng/mL) in α-MEM for 11 days.

Human Notch1 intracellular domain (NICD) was synthesized, and human FOXM1 gene was PCR amplified from the human cDNA library. The cDNA was subcloned into pMEI-5 vector containing the MCS-IRES-Venus or the MCS-P2A sequence. A method for generating the CAR expression vector was described previously (29). Retroviral transduction was performed, as described previously (30). Briefly, human CD8⁺ T cells were activated by anti-CD3/28 microbeads starting on day 0. Twenty-four hours after activation, retroviral solution was added, and the cells were centrifuged at 2,500 rpm. for 2 hours at 35°C. Eighteen hours after spin infection, the cells were transferred into 10% AB serum containing fresh RPMI1640 media to remove retrovirus.

Measurement of the extracellular acidification rate (ECAR) and the oxygen consumption rate (OCR) was performed using an XF24 Extracellular Flux Analyzer (Seahorse Bioscience). Isolated naïve, T_EM, T_CM, iT_SCM, and both the NICD- and FOXM1ΔN-transduced T cells were resuspended in the Seahorse XF base medium and plated (5 × 10⁵ cells per well) into Seahorse 24-well plates coated with Cell-Tak (20 μg/mL, Corning). The ECAR and OCR were measured with a Mitostress set (Seahorse Bioscience). Oligomycin inhibits ATP synthase (complex V), which decreases electron flow through the electron transport chain (ETC), resulting in a reduction of the OCR. This decrease in OCR is linked to cellular ATP production. FCCP is an uncoupler that collapses the proton gradient and disrupts the mitochondrial membrane potential. As a result, electron flow through the ETC is uninhibited, and oxygen consumption by complex IV reaches the maximum. The third injection is a mixture of rotenone and antimycin A, a complex I inhibitor and a complex III inhibitor, respectively. The treatment completely shuts down mitochondrial respiration. Oligomycin (1.5 μmol/L), carbonyl cyanide-p-trifluoromethoxyphenylhydrazone (FCCP; 1.5 μmol/L), and the mixture of antimycin A/rotenone (1 μmol/L each) and etomoxir (200 μmol/L) were added at the indicated time points. The OCR/ECAR ratio was calculated as the basal OCR value per basal ECAR value. We measure the amount of intracellular ATP in T cells using an ATP Bioluminescence Assay Kit (Roche) following the manufacturer's instructions.

Before capillary electrophoresis-mass spectrometry (CE-MS)-based metabolome analysis was performed, human CD8⁺ T cells were harvested and washed with Dulbecco PBS. These cells were washed in 10% mannitol (Wako) and then plunged into methanol that contained internal standards of methionine sulfone for cations and 2-morpholinoethanesulfonic acid for anions (300 μmol/L each). Cells were collected for CE-MS using an Agilent capillary electrophoresis system equipped with an air pressure pump, an Agilent 1100 series mass selective detector mass spectrometer, and an Agilent 1100 series isocratic high-performance liquid-chromatography pump with a G1603A adaptor kit and a G1607A sprayer kit (Agilent Technologies) as described previously (31). Values (%) were corrected for the amount of total protein.

The recombinant SpCas9 protein bearing a nuclear localization was prepared as described previously (32). To prepare SpCas9/sgRNA RNP complexes, 30 μg of crRNA and 30 μg of tracRNA were hybridized for 5 minutes at 95°C. The 60 μg of hybridized crRNA/tracRNA complex and 30 μg of purified recombinant SpCas9 protein were preincubated with a P2 Primary Cell 4D-NucleofectorTM X Kit S (P2 buffer; Lonza) for 15 minutes at room temperature. Human CD8⁺ T cells were activated by anti-CD3/28 microbeads starting on day 0. Forty-eight hours after activation, CD8⁺ T cells were resuspended in P2 buffer (Lonza).The cell suspensions were then mixed with P2 buffer containing Cas9 RNPs, to a final volume of 20 μL. The 20 μL of cell suspension was electroporated by a 4D-Nucleofector X unit (program code: EH100, Lonza). After electroporation, the cells were cultured in serum-free RPMI1640 medium for 3 hours and replated in RPMI1640 medium with 10% AB serum without CD3/28 microbeads. Twenty-four hours after electroporation, T cells were restimulated by anti-CD3/28 microbeads. The cells were cultured over 4 days until further experiments.

Antibody dilution factors are 1:100 for antibodies purchased from BioLegend or eBioscience. To measure mitochondria mass, cells were stained with 100 nmol/L MitoTracker Green (Thermo Fisher Scientific) or MitoTracker Far Red (Thermo Fisher Scientific) for 15 minutes in RPMI1640 medium with 10% AB serum then washed twice before flow cytometry. To measure glucose uptake ability, T cells were incubated with 100 μmol/L 2-NBDG (Invitrogen) for 2 hours in RPMI1640 medium with 10% AB serum then washed twice before flow cytometry. To measure neutral lipids, cells were stained with 500 ng/mL Bodipy 493/503 (Thermo Fisher Scientific) for 10 minutes in PBS then washed twice before flow cytometry or fluorescent microscopy cell imaging. To identify fatty acid uptake ability, cells were stained with 1 μmol/L Bodipy FL C₁₆ (Thermo Fisher Scientific) in PBS with 20 μmol/L fatty acid-free BSA then quenched by adding ice-cold PBS with 2% FBS and washed twice before flow cytometry. For PGC1A and CPT1A staining, cells were fixed and permeabilized with 4% paraformaldehyde and ice-cold 90% methanol and incubated with antibodies for 1 hour at room temperature. We performed flow cytometry acquisition on a FACS Canto II cytometer (BD Biosciences) and analyzed the data using FlowJo software (Tree Star). We sorted mouse and human T cells using a FACS Aria II cell sorter (BD Biosciences). We performed fluorescent cell imaging using a BZ-X8000 fluorescence microscope (Keyence).

A method for the generation of EBV-specific T cells was described previously (26). These EBV-specific CD8⁺ T cells were CFSE labeled (5 × 10⁴), then stimulated with 1.25 × 10⁴ autologous LCL or 5 × 10⁴ Dynabeads Human T-Activator CD3/CD28 microbeads for 60 hours in normal media with or without 10% AB serum.

To detect comparative telomere length in T cells, we performed flow cytometry using FISH and a fluorescein-conjugated PNA probe (Dako).

Total RNA was extracted using RNAiso Plus (#Z6012, Takara Bio) or ReliaPrep RNA Miniprep Systems (#4368814, Promega) and subjected to reverse transcription using a High Capacity Complementary DNA (cDNA) Synthesis Kit (Thermo Fisher Scientific). PCR analysis was performed using an iCycler iQ multicolor Real-Time PCR Detection System (Bio-Rad) and SsoFast EvaGreen Supermix (Bio-Rad). All primer sets yielded a single product of the correct size. Relative expression levels were normalized to 18SrRNA.

The expression levels of the Affymetrix Human Gene 1.0 ST Array data were calculated from CEL files using GeneSpring GX software 14.9 (Agilent). Low or no expressed genes were removed and genes were selected by one-way ANOVA with Tukey test (P < 0.05, iT_SCM vs. T_CM in Fig. 1 and S3, iT_SCM vs. activated in Fig. 6 and Supplementary Fig. S10), and picked over 2-fold expression in iT_SCM compared with that in T_CM or that in activated, respectively. Clustering analysis, principal component analysis and gene ontology analysis were performed using GeneSpring GX software 14.9. Red and blue colors indicate increased and decreased expression of genes, respectively.

All experiments using mice were approved by the Institutional Animal Care and Use Committee (IACUC; approval number 08004) of Keio University (Tokyo, Japan), and performed according to IACUC guidelines. The tumor transplantation and therapeutic model was performed according to the procedure reported by Sabatino and colleagues (14). NALM6 (1 × 10⁶) cells were intravascularly inoculated into NOD-SCID Il2rγ^null (NSG) mice 7 days before CAR-T transfer to establish therapeutic models. CAR-T cells (5 × 10⁵) were adoptively transferred into tumor-bearing mice. Animals were closely monitored for signs of graft-versus-host disease and other toxicity as evidenced by loss of fur, diarrhea, or conjunctivitis, as well as for leukemia-related hind-limb paralysis. When mice reached the humane endpoint in the tumor model, they were sacrificed, as approved by the Animal Ethics Committee of Keio University (Tokyo, Japan).

This study was approved by the Institutional Review Board of Keio University School of Medicine (approval number: 20120039), and conducted in compliance with the Declaration of Helsinki. Written informed consent was obtained from all individuals.

Statistical analysis was performed using Student t test, one-way ANOVA with the Tukey–Kramer method, two-way ANOVA with Tukey–Kramer method and the Kaplan–Meier method, using GraphPad Prism version 7 software (GraphPad Software). The variance among groups was estimated using the F test, and P values <0.05 were considered statistically significant. All data are presented as the mean ± SEM. Mice were randomly assigned to experimental groups. The investigators were not blinded to allocation during experiments and outcome assessment.

The data that support the findings of this study are available within the Supplementary Information files or from the corresponding author upon reasonable request. Microarray data that support the findings of this study have been deposited in GEO with the primary accession codes GSE23321, GSE93211, and GSE136765.

We previously reported that the coculture of EBV-specific T cells with OP9-hDLL1 cells induced EBV-specific iT_SCM cells and these iT_SCM cells exhibited a strong in vivo antitumor effect (26, 27). Gene set enrichment analysis (GSEA) indicated that highly expressed genes in iT_SCM cells were enriched in gene sets of “Myc targets,” “G₂–M checkpoint,” and “E2F targets” that were supposed to be related to the higher proliferation potential of iT_SCM cells, and “Oxidative Phosphorylation.” Recent studies suggested a strong correlation between OXPHOS and memory phenotypes (16, 17). Therefore, we investigated the mechanism of the enhanced antitumor activity of iT_SCM cells by focusing on T-cell metabolism.

To confirm the difference between EBV-specific T cells before and after coculturing with OP9-hDLL1 cells, genes that were differentially expressed between T_CM and iT_SCM were extracted and the genes related to OXPHOS were significantly upregulated in iT_SCM cells (Fig. 1A). Enhanced OXPHOS has been shown to be a hallmark for a strong antitumor effect of endogenous antitumor T cells or CAR-T cells (21, 22, 33). Thus, we compared the metabolic status by using a Seahorse Flux analyzer among naïve T cells and Epstein Barr virus (EBV)-specific T_EM, T_CM, and iT_SCM cells (26, 27). iT_SCM cells possessed higher basal and maximal respiration capacity compared with T_EM and T_CM cells (Fig. 1B). The ECAR and OCR of iT_SCM cells were significantly increased compared with those of naïve T cells, T_EM, and T_CM cells (Fig. 1C and D). The OCR/ECAR ratio of naïve T cells and iT_SCM cells was higher than that of T_EM cells and T_CM cells (Fig. 1E). CE-MS analysis also revealed that the total amount of glycolytic metabolites significantly increased in T_CM cells, whereas TCA cycle–related metabolites did not change among T_CM and iT_SCM cells (Supplementary Fig. S1A–S1C). The ratio of TCA cycle/glycolysis-related metabolite was increased by the coculture with OP9-hDLL1 cells (Supplementary Fig. S1C). In addition, mitochondrial amounts, PGC1A expression, ATP content, and the NADH/NAD ratio were all increased in iT_SCM cells compared with those in the other cells (Fig. 1F and G; Supplementary Fig. S1D). These data indicate that iT_SCM cells preferred OXPHOS to glycolysis compared with other T-cell subsets.

We have shown that iT_SCM cells exhibited stronger proliferative ability than other effector/memory T-cell subsets (26). To determine the fuel for this rapid proliferation of iT_SCM cells during the recall response, we examined the proliferative ability in the presence of the glycolysis inhibitor, 2-deoxy-d-glucose (2-DG) or the mitochondrial ETC inhibitor, rotenone. Glycolysis inhibition suppressed T_EM and T_CM proliferation more strongly than iT_SCM cells, whereas rotenone inhibited the iT_SCM recall response more efficiently than T_EM and T_CM cells (Fig. 1H). In addition, iT_SCM cells showed lower glucose uptake capacity compared with other T-cell subsets (Supplementary Fig. S1E). These results confirmed that the proliferation of iT_SCM depends on OXPHOS rather than glycolysis.

Activated T cells prefer glycolysis, OXPHOS, and glutaminolysis for the rapid production of ATP, whereas memory T cells preferentially engage de novo fatty acid synthesis and fatty acid consumption by OXPHOS (34). Glycolysis-derived pyruvate is also consumed as the energy source of mitochondria. To validate the energy dependency of iT_SCM induction, we first examined the effects of metformin, which disrupts OXPHOS and promotes glycolysis, on iT_SCM induction. Expectedly, metformin interfered with iT_SCM induction (Supplementary Fig. S2A). Exogenous glutamine addition did not promote iT_SCM induction and did not rescue the effect of metformin (Supplementary Fig. S2B). The pyruvate-enriched condition slightly enhanced iT_SCM induction, but this condition did not rescue the effect of metformin on iT_SCM induction (Supplementary Fig. S2C). These data are consistent with our observation that OXPHOS rather than glycolysis plays an important role in iT_SCM phenotypes and function. Furthermore, these data also suggest that iT_SCM cells do not prefer to consume both pyruvate and glutamine as fuel of OXPHOS.

To identify the contribution of fatty acid metabolism to iT_SCM phenotypes, we next examined fatty acid uptake, storage, and synthesis. iT_SCM cells showed low fatty acid uptake ability but contained larger amounts of intracellular neutral lipid droplets (Fig. 2A–C). The expression of fatty acid synthesis–related genes that regulate the entire fatty acid synthesis pathway from acetyl-CoA also increased and fatty acid transporters to mitochondria, l-carnitine, and carnitine palmitoyltransferase 1A (CPT1A) were highly expressed in iT_SCM cells (Fig. 2D–F; ref. 35). The OCR in the lipid-depleted state was drastically increased after loading the mitochondrial uncoupler, FCCP (Fig. 2G), indicating an extremely high mitochondrial reserve capacity in iT_SCM cells. The CPT1A inhibitor, etomoxir, rapidly reduced OCR in lipid-depleted medium (Fig. 2G). Lipid depletion significantly inhibited the recall response of T_EM and T_CM, but the recall response of iT_SCM cells did not change (Fig. 2H). Thus, these data indicate that iT_SCM cells use FAO from largely stored neutral lipids for OXPHOS in response to antigen stimulation.

To apply the method for iT_SCM generation to CAR-T therapy, we next produced anti-CD19 CAR–expressing iT_SCM cells by the coculturing of CAR-T cells with OP9-hDLL1 cells. As shown in Fig. 3A, to be engineered with CARs, T cells were first primed by anti-CD3/CD28 antibodies and then transduced by retrovirus carrying anti-CD19 CARs at 24 hours after activation. The CAR construct is shown in Supplementary Fig. S3A. Five days after retroviral transduction, CAR-expressing T cells were sorted out and expanded by further TCR stimulation for 6 to 8 days. These “CAR-T” cells were cocultured with OP9-hDLL1 cells for 11 days and converted into iT_SCM cells, designated as “CAR-iT_SCM” cells (Fig. 3A). Empty vector- and CAR-transduced T cells displayed CD45RA⁻ T_CM and T_EM phenotypes, whereas the coculture with OP9-hDLL1 cells efficiently converted these empty vector- and CAR-transduced T cells into CD45RA⁺CCR7⁺ T_SCM phenotypes (Fig. 3B).

Next, we examined the differences in the T-cell phenotypes between CAR-T and CAR-iT_SCM cells. Expression of early memory-associated molecules CD27, CD28, and CD62L was increased in CAR-iT_SCM cells but the effector markers, CD25, CD95, and PD-1 were reduced (Fig. 3C). Early memory-associated genes, EOMES, BCL6, TCF7, LEF1, KLF2, and BACH2 were also upregulated in CAR-iT_SCM cells compared with those in CAR-T cells (Fig. 3D). The CAR-iT_SCM phenotype potentiated recall responses to CD19-positive leukemia, and significantly suppressed activation-induced cell death (AICD; Fig. 3E and F).

To verify the antitumor effect of CAR-iT_SCM cells in vivo, CAR-T or CAR-iT_SCM cells were transferred into mice carrying leukemic (NALM6) cells. NALM6 cells were transferred into NSG mice, and CAR-T or CAR-iT_SCM infusion was performed 7 days later (Fig. 3G). CAR-T and CAR-iT_SCM infusion significantly reduced the number of NALM6 cells in the peripheral blood, whereas infused T cells were retained in both the peripheral blood and the spleen (Fig. 3H; Supplementary Fig. S3B). These data indicate an expansion of CAR-iT_SCM cells in the peripheral blood on day 4, and a decrease thereafter due to the eradication of leukemic cells. It is of note that 7 to 10 days after T-cell injection, leukemia cells almost completely disappeared in mice infused with CAR-iT_SCM cells, whereas remained in mice infused with CAR-T cells (Fig. 3H). Furthermore, a larger number of CD8⁺ T cells with the naïve/T_SCM phenotype (CD45RA⁺CCR7⁺) was detected in the spleen of CAR-iT_SCM-injected mice (Fig. 3I and J) and the survival of these mice was prolonged (Fig. 3K). These data indicate that CAR-iT_SCM cells have much higher antitumor potential than conventional CAR-T cells.

Next, we investigated how NOTCH contributes to iT_SCM formation. Both NOTCH1 and NOTCH2 but not NOTCH3 were highly expressed on human activated T cells and progressively downregulated during the OP9-hDLL1 coculture (Supplementary Fig. S4A). To define the effect of NOTCH signaling, the NICD region was expressed in activated T cells by a retrovirus vector (Supplementary Fig. S4B). Forced expression of NICD upregulated NOTCH target gene expression (Supplementary Fig. S4C). However, we failed to induce iT_SCM cells only by NICD transduction due to cell death during long-term culture. It has been reported that OP9-hDLL1 culture conditioned medium (CM) with NOTCH signaling can induce mature T cells from T-cell progenitors (36). Thus, we tried to culture NICD-transduced T cells in CM of OP9-hDLL1 cells (Fig. 4A). NICD-transduction and the culture in CM supported long-term cell proliferation for 11 days, and converted activated T cells into CD45RA⁺CCR7⁺ cells at a similar efficiency as iT_SCM cells induced by OP9-hDLL1 cells (Fig. 4B, CM). No additional effects of NICD transduction were not observed when NICD-transduced cells were cocultured with OP9-hDLL1 feeder cells (FC; Fig. 4B). NICD transduction also increased the mitochondrial mass and PGC1A expression (Fig. 4C) and neutral lipid storage (Fig. 4D) and upregulated the gene expression of fatty acid synthesis–related enzymes (Fig. 4E).

T_SCM cells have been shown to have a long-life due to high telomerase activity (37). Telomerase component genes were highly expressed by NICD overexpression (Supplementary Fig. S4D). Consistently, a significant increase in telomere length was observed in iT_SCM cells compared with other memory T cells (Supplementary Fig. S4E). These data suggest that, in addition to intracellular neutral lipid storage, telomere elongation is a mechanism responsible for the long-life of iT_SCM cells.

These data indicate that NOTCH signaling is necessary for mitochondrial biogenesis, fatty acid synthesis, and telomere elongation to acquire T_SCM phenotypes.

Next, to validate the transcriptional activity of NICD, we generated various deletion mutants of NICD and examined the ability of NICD mutants for iT_SCM induction. The ankyrin (ANK) domain of NOTCH is an integral part of the transcription complex, whereas the RBP-Jκ–associated molecule (RAM) domain may trigger allosteric changes in the structure needed for the derepression of transcription (38). The transactivation (TAD) domain may be important for LEF-1 activation (39). Induction of various NOTCH target genes was severely impaired by deletion of the ANK domain, but not the TAD domain, and moderately affected by the RAM domain deletion (Supplementary Fig. S5A). Consistently, T_SCM induction potential was destroyed by ANK domain deletion but not so severely by TAD or RAM domain deletion (Supplementary Fig. S5B). These data clearly demonstrate that transcriptional activation by NICD is necessary for iT_SCM induction.

Next, to identify a key transcription factor regulated by NOTCH signaling, we have searched iT_SCM-specific genes into pathways listed in a molecular signature database. Gene set enrichment analysis revealed that several cell-cycle–regulating pathways were enriched as iT_SCM-specific phenotypes (Fig. 5A). In particular, we noticed the forkhead box protein M1 (FOXM1), which is well known to regulate cell-cycle progression, G2–M checkpoint and other cell-cycle–related genes (Fig. 5A). FOXM1 has been reported to regulate not only the cell cycle but also stemness, mitochondrial function and redox network in various tumor cells (40–43). We confirmed a high expression of FOXM1 in iT_SCM cells (Fig. 5B). NICD overexpression increased FOXMl expression in T cells (Fig. 5C; Supplementary Fig. S5C), and we confirmed that the deletion of RAM, ANK and TAD domain of NICD decreased FOXM1 expression, which indicates that NICD-induced transcription plays an essential role in the induction of FOXM1 (Supplementary Fig. S5D).

To examine the roles of FOXM1 in iT_SCM phenotypes, we transduced a constitutive active form of N-terminal truncated FOXM1 (FOXM1ΔN; ref. 44) in T cells (Supplementary Fig. S6A) and examined the function by using the same culture method shown in Fig. 4A. FOXM1ΔN-transduction plus OP9-hDLL1-CM conferred T_SCM phenotypes to activated T cells-like coculture with OP9-hDLL1-FC (Fig. 5D and E). FOXM1-transduced cells expressed CD27 molecules at high levels (Fig. 5D and E), whereas FOXM1 inhibitor, thiostrepton treatment, and FOXM1 knockout suppressed the induction of T_SCM phenotype (CD45RA⁺CCR7⁺; Fig. 5F; Supplementary Fig. S6B) and NICD-induced iT_SCM formation was cancelled by thiostrepton treatment (Fig. 5G). In addition, the FOXM1 promoter contains two RBP-J–binding sites where NICD binds, and this region showed transcriptionally positive chromatin modification such as H3K27 acetylation (Supplementary Fig. S6C). Importance of the RBP-J–binding sites of the FOXM1 promoter is also supported by the requirement of the RAM (RBP-J–associated module) domain for induction of FOXM1 expression (see Supplementary Fig. S5D). These results suggest that FOXM1 is a direct downstream gene of Notch signaling and positively regulates the induction of T_SCM-like phenotypes.

We confirmed the enhanced antitumor potentials of T_SCM-like cells induced by FOXM1ΔN overexpression using the CAR-T system. Normal CAR or FOXM1ΔN tandemly jointed CAR (FOXM1ΔN-CAR) constructs were retrovirally transduced (Supplementary Fig. S7A) and these cells were cultured on OP9-hDLL1-FC or with OP9-hDLL1-CM (Fig. 6A). These CAR-transduced T cells were transferred to NALM6-bearing NSG mice (Fig. 6B). Twelve days after CAR-T transfer, few NALM6 cells were detected in CAR-iT_SCM- and CAR-T_SCM-like (FOXM1ΔN)-transferred mice, whereas leukemic cells remained in the other CAR-T–transferred mice (Fig. 6C). Both CAR-iT_SCM- and CAR-T_SCM-like (FOXM1ΔN) cells prolonged the survival of NALM6-bearing mice compared with the other groups (Fig. 6D). These data indicate that T_SCM-like cells induced by FOXM1ΔN and CM have similar antitumor potential to iT_SCM cells induced by coculture with OP9-hDLL1-FC, suggesting that FOXM1 is a key factor for generating T_SCM-like CAR-T cells.

Next, we investigated the role of FOXM1 in the mitochondrial changes of iT_SCM cells. Treatment with a FOXM1 inhibitor, thiostrepton, during coculture on OP9-hDLL1-FC decreased mitochondrial mass, PGC1A expression, and OCR, but increased ECAR (Fig. 6E and F). FOXM1 deficiency also decreased mitochondrial mass (Fig. 6G), indicating that FOXM1 is required for mitochondrial changes. Overexpression of FOXM1ΔN with OP9-hDLL1-CM increased mitochondrial mass and PGC1A expression to similar levels in iT_SCM cells induced by OP9-hDLL1-FC (Supplementary Fig. S7B). Furthermore, the increase in mitochondrial biogenesis induced by overexpression of NICD was cancelled by FOXM1 inhibition (Supplementary Fig. S7C). ETC inhibitor, rotenone, also partly inhibited iT_SCM formation either by OP9-hDLL1 coculture or by NICD or FOXM1ΔN overexpression (Fig. 6H). In contrast, CPT1A inhibition had little effect on FOXM1-driven iT_SCM induction (Supplementary Fig. S7D). Thus, although NOTCH and FOXMl tightly regulate both the mitochondrial biogenesis and iT_SCM phenotypes, CPT1A and FAO themselves may not be involved in iT_SCM surface phenotypes but are important for high proliferation ability of iT_SCM cells in glucose-low conditions.

To further confirm the relationship among activated T cells, gene expression profiles were analyzed among naïve T cells, iT_SCM cells induced by OP9-DLL1-FC, and iT_SCM-like cells induced by NICD or FOXM1. Principal component (PC) analysis and clustering analysis revealed that NICD- and FOXM1ΔN-induced T_SCM-like cells were most closely related to iT_SCM cells compared with naïve T cells and activated T cells (Fig. 6I; Supplementary Fig. S8A). OXPHOS and glycolysis profiles of both NICD- and FOXM1ΔN-induced T_SCM-like cells highly resembled iT_SCM cells (Supplementary S8B and S8C). These data confirmed that both NOTCH and FOXM1 promote the transcriptional program that induce iT_SCM cells.

We have reported that the NOTCH signaling induces T_SCM-like, “iT_SCM” cells from fully activated T cells. Unlike other T_SCM-generating methods that induce T_SCM-like cells from naïve T cells, our two-step culture is more suitable for generating CAR-iT_SCM cells, because T-cell activation is necessary for the transduction of the CAR gene. We also found that the gene sets of OXPHOS were highly enriched in iT_SCM cells. Therefore, we focused on OXPHOS because Myc targets, G₂–M checkpoint and E2F targets appear to be related to a higher proliferation capacity of iT_SCM cells. Here, we demonstrated that the coculture with NOTCH ligand-expressing feeder cells, or the forced expression of NICD or FOXM1ΔN conferred T_SCM-like properties, including OXPHOS preference, and potentiate the antitumor ability of CAR-T cells. Our data also indicated that mitochondrial biogenesis and fatty acid synthesis are important for iT_SCM induction and function.

The potent antitumor activity of iT_SCM cells appears to be associated with not only early memory phenotypes but also FAO-dependent OXPHOS of iT_SCM cells. Fraietta and colleagues reported that CAR-T cells from complete-responding patients with CLL were enriched in early memory–related genes with reduced glycolysis signatures (7). TCR stimulation with Wnt signaling from naïve T cells sustains the T_SCM property and leads to strong antitumor response in CAR-T therapy (13). Less-differentiated CAR-T induction was obtained by inhibiting the Akt–mTOR pathway, enhancing FOXO1 activity and potentiating mitochondrial function (23–25, 45–47). In addition, IL21 in combination with rapamycin has been reported to enhance generation of tumor antigen-specific T_SCM-like cells (48), and IL21 enhances T_CM formation in a FAO-dependent manner (49). Conversely, chronic CD4⁺ T cells are irreversibly dependent on OXPHOS and FAO to produce ATP for IFNγ synthesis, but PD-1 blockade unlocks the irreversible metabolism and reprograms to functional T cells (50). These reports suggest that the metabolic fitness regulates T-cell differentiation, memory formation, and exhaustion, and therefore maintains the antitumor ability of T cells.

Here, we identified FOXM1 as a key regulator for iT_SCM induction as a downstream target of NOTCH signaling. FOXM1 overexpression can mostly substitute for the effects of NOTCH signaling on CAR-iT_SCM properties, including OXPHOS preference, telomere elongation, and potent antitumor ability. FOXM1 has been characterized as an important regulating factor for cancer stemness, chemo-resistance, mitochondrial function, and redox pathways in tumor cells (40–43). We propose that FOXM1 is also an important factor for the generation of iT_SCM cells. Therefore, the induction of FOXM1 is a potent method for conferring T_SCM-like properties to CAR-T cells. In this sense, it is of interest that FOXM1 is also upregulated in T_SCM-like cells induced by the Wnt signaling. Our data suggest that NOTCH1 and FOXM1 are sufficient for the induction of T_SCM-cell surface phenotypes and the induction of OXPHOS-dominant ATP synthesis; however, FAO-dependent OXPHOS is not sufficient for induction of the entire iT_SCM program. The precise molecular mechanism of FAO and OXPHOS-dependent ATP synthesis in iT_SCM cells remains to be clarified.

W. Tomisato is an employee at Daiichi Sankyo Co., Ltd. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T. Kondo, W. Tomisato, A. Yoshimura

Development of methodology: T. Kondo

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T. Kondo, M. Ando, T. Srirat, S. Mise-Omata, T. Hishiki

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T. Kondo, M. Ando, N. Nagai, T. Srirat, M. Ohmura

Writing, review, and/or revision of the manuscript: T. Kondo, B. Liu, S. Chikuma

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. Kondo, M. Ando, M. Ikeda, H. Nishimasu, O. Nureki, N. Hayakawa, R. Uchibori, K. Ozawa

Study supervision: T. Kondo, A. Yoshimura

We thank Yoshiko Noguchi, Yasuko Hirata, Yukiko Tokifuji, Mika Inoue, Eiji Sugihara, Hideyuki Saya (Keio University) for their technical assistance and Taeka Hayakawa and Yutaka Kawakami (Keio University) for discussions. This work was supported by JSPS KAKENHI (S) JP17H06175, Challenging Research (P) JP18H05376, and AMED-CREST JP19gm1110009, TaNeDS grant program (Daiichi Sankyo Co., Ltd.), the Takeda Science Foundation, the Uehara Memorial Foundation, the Mochida Memorial Foundation for Medical and Pharmaceutical Research, Bristol–Myers Squibb Research grant, the Kanae Foundation, the SENSHIN Medical Research Foundation and the Keio Gijuku Academic Developmental Funds.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.