CREM is a regulatory checkpoint of CAR and IL-15 signalling in NK cells

Cell lines and culture conditions

Human cancer cell lines of Raji (CCL-86), K562 (CCL-243), SKOV3 (HTB-77) and 293T (CRL-3216) were obtained from the American Type Culture Collection (ATCC). Raji cells were cultured in RPMI-1640 medium with 10% FBS, 2 mM l-glutamine (Invitrogen) and 1% penicillin–streptomycin (Invitrogen). SKOV3 and 293T cells were cultured in DMEM medium (Invitrogen) supplemented with 10% FBS, 1% penicillin–streptomycin and 2 mM l-glutamine (Invitrogen). The UMRC3 cell line was obtained from Sigma-Aldrich (08090512) and cultured in DMEM medium with 10% FBS, 2 mM l-glutamine (Invitrogen) and 1% penicillin–streptomycin (Invitrogen). The PDX cell line BCX.010 of breast cancer was provided by F. Meric-Bernstam at MD Anderson Cancer Center (MDACC), and PATC148 cells were provided by A. Maitra (MDACC). Both BCX.010 and PATC148 cells were cultured in DMEM (Invitrogen) supplemented with 10% FBS, 1% penicillin–streptomycin and 2 mM l-glutamine (Invitrogen). Cells were passaged every 3–4 days. The CD70 gene in UMRC3 cells was deleted using CRISPR–Cas9 methods (detailed below) to generate CD70 KO UMRC3 cells. UMRC3 and BCX.010 cells were transduced to express GFP for fluorescence monitoring. Raji cells were transduced to express mCherry or GFP, and K562 cells were transduced to express mCherry for fluorescence monitoring by microscopy. Raji cells were modified to express firefly luciferase to enable bioluminescence in vivo imaging. All cells were maintained in 5% CO₂ at 37 °C and were authenticated by STR profiling at the MDACC Cell Line Characterization Core Facility. All cell lines were tested regularly for mycoplasma contamination using a MycoAlert Mycoplasma Detection kit (Lonza) and were only used when tested negative for contamination.

Cord blood NK cell isolation and expansion

Cord blood (CB) units were provided by the MDACC CB Bank under a protocol approved by the Institutional Review Board (Lab04-0249). Lymphocytes from CB were isolated through density gradient (Ficoll-Histopaque, Sigma). NK cells (CD56⁺CD3^–) were purified from lymphocytes by negative selection using a NK cell isolation kit (Miltenyi Biotec) following the manufacturer’s instructions. Isolated NK cells were preactivated for 16 ± 2 h using recombinant human (rh) IL-12 (BioLegend; 10 ng ml^–1), rhIL-18 (MBL International; 50 ng ml^–1) and rhIL-15 (BioLegend; 50 ng ml^–1) as previously described⁵⁶, washed 2 times with PBS to remove cytokines and expanded with irradiated (100 Gy) universal antigen presenting cells (uAPCs) at a feeder cell-to-NK cell ratio of 2:1 and cultured in 50% Click’s medium (EHAA, Fujifilm) and 50% RPMI-1640 (referred to hereafter as NK cell medium) supplemented with rhIL-2 (Proleukin, 200 U ml^–1, Chiron). Medium was changed every 2–3 days, and irradiated uAPCs were added every week to the NK cell culture to support expansion. Transduction of NK cells with retroviral constructs (detailed below) was performed on days 5–6 after expansion with uAPCs. An equal number of NK cells from each condition was electroporated for CRISPR–Cas9 KO and expanded on day 2 after transduction. Functional experiments were performed on days 14–15 (7–8 days after second expansion), unless mentioned otherwise. Cell counts were recorded using ViaStain AOPI staining solution (Nexcelom) and a cellometer Auto 2000 (Nexcelom) instrument on a regular basis to monitor NK cell proliferation or at the time of functional assays to ensure equal cell numbers between conditions.

Generation of retroviral constructs, virus production and CAR-NK cells

The construct for the CD70-targeting CAR (CD27(ECD).CD28.ζ.2 A.IL-15), referred to as CAR70–IL-15, incorporates the ECD of CD27 (the natural receptor for CD70 ligand) linked to the CD28 co-stimulatory domain and the CD3ζ signalling domain. It also includes the IL15 transgene. The CAR70 construct was generated through deletion of IL-15 by restriction digestion. The six tyrosine (Y) residues of the ITAMs in the CD3ζ region of CAR70 construct were changed to phenylalanine (F) and termed CAR70.3ζ.Y6F⁴⁴. The entire CD3ζ region of the CAR70 construct was deleted and termed CD27(ECD).

The construct for the TROP2-targeting CAR (TROP2scFv (clone hRS7).CD28.ζ.2 A.IL-15), referred to as CAR.TROP2–IL-15, consists of a scFv targeting TROP2 (derived from the sequence of the TROP2-targeting antibody–drug conjugate sacituzumab govitecan, human RS7 (hRS7)) linked to the CD28 co-stimulatory domain and the CD3ζ signalling domain. Similarly, the construct includes the IL15 transgene.

All CAR constructs and the retroviral construct encoding IL-15 were cloned into a SFG retroviral backbone by GeneArt Gene Synthesis (Thermo Fisher Scientific) to generate viral vectors. Transient retroviral supernatants were produced from transfected 293T cells with CAR plasmids along with packaging and envelope plasmids. CAR transduction efficiency was measured 48–72 h after transduction by flow cytometry.

CAR, CD16, NKp30, NKp46 and cytokine-stimulation assays

For CAR-stimulation assays, human CD27 ligand–CD70 protein (ACROBiosystems, CDL-H82D7) was plated in PBS in appropriate plates (96-well ELISA plates or 6-well plates) at a concentration of 1.2 μg ml^–1. Plates were kept at 4 °C on a plate shaker overnight. NK cells from the indicated conditions were placed in NK cell medium without any cytokine support for 24–72 h before stimulation. The next day, PBS was aspirated, and an equal number of NK cells from the indicated conditions was stimulated in the plates at 37 °C after a brief centrifugation step to promote uniform interaction of NK cells from the indicated conditions with the surface-coated CD70. NK cells were then collected at appropriate intervals according to the downstream assays (30 min for pCREB detection and 24 h for qPCR). For stimulation assays with CD16, NKp30 and NKp46, the following antibodies were used: anti-human CD16 (BD Biosciences, 3G8, 556617), anti-human NKp30 (R&D systems, 210845, MAB1849) and anti-human NKp46 (eBioscience, 9E2, 16-3359-85). Similar to the CAR-stimulation assays, antibodies were plated in PBS in appropriate plates at a concentration of 1 μg ml^–1. The assays were then performed as described for CAR stimulation.

For cytokine stimulation, NK cells were placed in NK cell medium without any cytokine support for 24–72 h before stimulation. The next day, NK cell medium containing various cytokine concentrations (50 pg ml^–1, 500 pg ml^–1 and 5,000 pg ml^–1) were prepared by serial dilution. NK cells were counted, pelleted and then resuspended in equal numbers in medium with the corresponding cytokine concentrations and incubated for an appropriate interval (30 min for pCREB detection and PKA calorimetric assay, and 24 h for qPCR). The following cytokines were used: rhIL-2 (Proleukin), rhIL-10 (StemCell Technologies, 78024.1), rhIL-12 (P70, BioLegend, 573004), rhIL-15 (BioLegend, 570304), rhIL-18 (MBL International, B003-5) and hIL-21 (Miltenyi Biotec, 130-095-767). For IL-15 antagonism experiments, a functional-grade IL-15 monoclonal antibody (eBioscience, 16-0157-82, clone ct2nu) was used at a concentration of 100 ng ml^–1 where indicated to neutralize the bioactivity of human IL-15.

Flow cytometry

NK cells or tumour cells were collected for flow cytometry, washed with PBS, pelleted and stained with Live/Dead Fixable Aqua Dead cell stain (ThermoFisher, 1:200) to determine their viability. After washing with FACS buffer (PBS with 1–2% FBS), cells were centrifuged and pellets were resuspended in antibody cocktail, mixed and incubated for 20 min at room temperature protected from light for surface staining. The transduction efficiency of IL-15 NK cells was measured using a conjugated goat F(ab′)2 anti-human lgG (H+L; Jackson ImmunoResearch) that recognizes the IgG hinge portion of the construct (as previously described)³. CAR expression on the CAR-NK cells transduced with the various CD70-targeting constructs was measured using anti-CD27 antibody. CAR expression on the TROP2-targeting CAR-NK cells was measured by incubating the cells in human TROP2–TACSTD2 protein, His tag (ACROBiosystems, TR2-H5223) at a concentration of 1.2 μg ml^–1 for 30 min at room temperature, after which they were washed with PBS and stained with anti-His antibody and the rest of the surface antibodies for 20 min at room temperature. The following antibodies were used for flow cytometry experiments: APC-Cy7 anti-human CD3 (BioLegend, HIT3a, 300318, 1:100); BV605 anti-human CD56 (BioLegend, HCD56, 318334, 1:50); BV650 anti-human CD16 (BD Biosciences, 3G8, 563173, 1:50); PerCP anti-human CD45 (BioLegend, HI30, 304026, 1:50); APC-Cy7 anti-human CD45 (BioLegend, HI30, 304014, 1:50); PE-CF594 anti-human CD27 (BD Biosciences, M-T271, 562297, 1:50); PE-Cy7 anti-human CD70 (BioLegend, 113-16, 355112, 1:50); AF-700 anti-mouse CD45 (BioLegend, QA17A26, 157616, 1:50); PE anti-human TROP2 (BioLegend, NY18, 363804, 1:50); anti-His-APC (BioLegend, J095G46, 362605, 1:50); PE anti-human CD326 (EpCAM; BioLegend, 9C4, 324206, 1:50); anti-human CD215 (IL-15Rα; BD Biosciences, JM7A4, 566589, 1:40); and anti-Ki67 (BD Biosciences, B56, 563462, 1:33). Staining for Ki67 required additional fixation and permeabilization steps as detailed below. Data were acquired using LSRFortessa (BD Biosciences), and analysis was performed using FlowJo software (v.10.8.2). All sorting was performed on a BD Biosciences Aria II Cell Sorter or a Beckman Coulter CytoFLEX SRT Cell Sorter at the MDACC North Campus Flow Cytometry and Cellular Imaging Core Facility. Representative gating strategies of NK cells are shown in Supplementary Fig. 1.

Intracellular staining

NK cells from the indicated conditions were stimulated by culturing them with cancer cells at an E/T ratio of 1:1 for 5–6 h in the presence of brefeldin A (10 μg ml^–1, BD Biosciences GolgiPlug, 555029) and monensin (1×, BioLegend, 420701) to inhibit protein transport. Unstimulated NK cells were used as negative controls, and NK cells stimulated with phorbol 12-myristate 13-acetate (Sigma-Aldrich, P8139) and inomycin (Sigma-Aldrich, I0634) were used as positive controls. Cells were collected after incubation, washed with PBS and stained using a Live/Dead Fixable Aqua Dead cell stain kit (ThermoFisher, 1:200) to identify viable cells. Cells were then stained with surface antibodies for 20 min at room temperature in the dark. Cells were then fixed and permeabilized using a BD Cytofix/Cytoperm kit (BD Biosciences, 554714) following the manufacturer’s instructions. Intracellular staining was performed with PE anti-human IFNγ (BioLegend, 506507, 1:40) and PE-Cy7 anti-human TNF (BioLegend, 502930, 1:40) antibodies for 30 min at 4 °C. Expression of IFNγ and TNF was assessed by flow cytometry and expressed as a percentage of CD56⁺CD3^– NK cells.

pCREB and CREM detection by flow cytometry

A FOXP3 transcription factor staining kit (eBioscience) was used for pCREB and CREM staining. In brief, cells (deprived of cytokine support for 24 h) were washed with PBS and stained with Live/Dead Fixable Aqua Dead cell stain (ThermoFisher, 1:200) as described above. After washing, cells were resuspended in NK cell medium and stimulated in 96-well plates with IL-15 or on antigen-coated plates as described above at 37 °C for 30 min. Following the stimulation step, cells were pelleted and permeabilized with FOXP3 Fix/Perm buffer at 4 °C for 45 min on a plate shaker. Cells were then washed twice with 1× Perm Wash buffer, after which they were stained with PE anti-human pCREB (pS133)/pATF-1 (pS63) antibody (BD Phosflow, J151-21, 558436, 1:20) or anti-human CREM (Santa Cruz Biotechnology, 22, sc-101530, 1:20) and appropriate surface antibodies, depending on the experiment, at room temperature for 30 min. Cells were then washed twice and data were acquired on a flow cytometer. Cells treated with FSK (Sigma-Aldrich, F6886) at a concentration of 20 μM were used as positive controls. In some experiments, cells were treated with the PKA inhibitor H89, dihydrochloride (Cell Signaling, 9844) at a concentration of 30 μM for 1–2 h before stimulation. For calcium chelation, a 500 mM stock solution of EGTA was prepared as follows: 6.645 g of 10 N NaOH with 19.0175 g of EGTA (Millipore, 324626) and MilliQ dH₂O to 100 ml. Cells in the corresponding conditions were treated with 10 mM EGTA for 1–2 h before stimulation.

Western blotting

NK cells as indicated were lysed using IP lysis buffer (Pierce IP Lysis Buffer, Thermo Scientific, 87788) supplemented with protease and phosphatase inhibitors (Halt protease and phosphatase inhibitor single-use cocktail, EDTA-free, Thermo Scientific, 78443) and incubated for 30 min on ice. Whole-cell lysates were obtained after centrifugation. A BCA Protein Assay kit (Pierce) was used to measure protein concentrations. Cell lysates from the indicated conditions were subjected to electrophoresis in equal amounts using SDS–PAGE and transferred to polyvinylidene difluoride membranes (Bio-Rad). Membranes were blocked with 5% milk (in PBST) or 5% BSA (in PBST) for 30 min, followed by primary antibody incubation overnight at 4 °C. The following primary antibodies were used: pCD3ζ Tyr142 (Abcam, EP265(2)Y, ab68235, 1:1,000); CD247/CD3ζ (Bethyl Laboratories, A305-212A, 1:2,000); CREM (Creative Diagnostics, CABT-B10032, 1:2,000); pCREB (Ser133; Cell Signaling, 87G3, 9198, 1:1,000); total CREB (Cell Signaling, 48H2, 9197, 1:1,000); STAT3 (Cell Signaling, 9139, 1:1,000); pSTAT3 (Cell Signaling, 9145, 1:1,000); STAT5A (Cell Signaling, 4807, 1:1,000); STAT5B (Cell Signaling, 3466, 1:1,000); pSTAT5A/B (EMD Millipore, 04-886, 1:1,000); STAT5-PY694 (Cell Signaling, 4322S, 1:1,000); STAT5 (Cell Signaling Technology, 94205S, 1:1,000); JAK1-PY1034/1035 (Cell Signaling, 74129S, 1:1,000); JAK1 (Cell Signaling, 3344S, 1:1,000); JAK2-PY1008 (Cell Signaling, 8082S, 1:500); JAK2 (Cell Signaling, 3230S, 1:1,000); S6-PS240/244 (Cell Signaling, 5364S, 1:1,000); S6 (Cell Signaling, 2217S, 1:1,000); SOCS1 (Cell Signaling, 55313S, 1:1,000); and β-actin (Sigma-Aldrich, AC-15, A5441, 1:5,000). The membranes were washed 3 times with PBST (5 min each) and then incubated with secondary antibody (donkey anti-rabbit IgG, HRP-linked whole Ab, Genesee Scientific, 84-854, and sheep anti-mouse IgG, HRP-linked whole antibody, Genesee Scientific, 84-848; both 1:3,000, 5% milk in PBST) for 60 min. Where applicable, the membranes were stripped to remove bound antibodies using Restore Western Blot Stripping buffer (Thermo Scientific) for further protein analyses. Protein signals were detected using ECL (Amersham) following the manufacturer’s instructions. Densitometry analyses were performed by evaluating the band intensity mean grey value of the indicated protein and normalizing it with the mean grey value of the total protein of the corresponding lane or the loading control (β-actin) using ImageJ 1.53t software.

PKA enzymatic assay

NK cells deprived of cytokines for 24 h were stimulated with the corresponding IL-15 concentrations as described above at 37 °C for 30 min. Cell lysates were then prepared as above. PKA activity was measured using a PKA Colorimetric Activity kit (Invitrogen, EIAPKA) following the manufacturer’s instructions.

CRISPR gene editing

For CRISPR–Cas9-mediated KO, pre-designed sgRNAs targeting CREM, ICER, CD70, STAT3, STAT5A and STAT5B were obtained from Integrated DNA Technologies (IDT). The gRNA sequences used are presented below.

For CREM-specific exon KO (Extended Data Fig. 3a,d,e): sgRNA-1: ACCACCTAGTATTGCTACCA; sgRNA-2: TCTTCAATCTTGGGAACACC.

For ICER-specific exon KO (encompassing both ICER-specific exon 1 and exon 2 (Extended Data Fig. 3a,d,e)): sgRNA-1: GCTGTAACTGGAGATGACAC; sgRNA-2: GCTCGATCTTACCACTAAGC.

For CD70: sgRNA-1: TCACCAAGCCCGCGACCAAT; sgRNA-2: GGCCATCTGCTCCTCCACGA.

For STAT3: sgRNA-1: GCAGAAAACTCTCACGGACG; sgRNA-2: TCTTCTGCCTGGTCACTGAC.

For STAT5A: sgRNA-1: CAAGTAGTGCCGGACCTCGA; sgRNA-2: CATTGACTTGGACAATCCCC.

For STAT5B: sgRNA-1: CATCAGATGCAAGCGTTATA; sgRNA-2: AAATAATGCCGCACCTCAAT.

To form the ribonucleoprotein (RNP) complex, sgRNAs (100 µM) were combined with Cas9 (Alt-R S.p. Cas9 Nuclease V3, 1081059) in the presence of P3 primary solution (P3 Primary Cell 4D-Nucleofector X Kit S, Lonza) at a sgRNA-to-P3-to-Cas9 ratio of 0.22:0.48:0.3 µl. The RNP complex was incubated for 20 min at room temperature. One million NK cells were collected, washed twice with PBS, pelleted and resuspended in 20 µl of the RNP complex in the presence of an electroporation enhancer (IDT, Alt-R Cas9 Electroporation Enhancer, 1075916) and P3 primary solution for electroporation. The final concentration for each electroporation was 2.2 µM sgRNA, 1.9 μM Cas9 nuclease and 5 µM Cas9 electroporation enhancer. The cells were electroporated in Nucleocuvette strips using the X unit of a 4D-Nucleofector device (Lonza) using the CM-137 electroporation program. After electroporation, the cells were left to rest for 10–15 min in a 37 °C incubator, after which they were transferred into prepared flasks with NK cell medium and feeder cells and IL-2, and cultured in a 37 °C incubator. The KO efficiency of CREM was evaluated by western blotting or by qPCR or PCR followed by gel electrophoresis using the following primers: forward: 5′-TGAATGAACTGTCCTCTGATGTG-3′; reverse: 5′-CCTGAGTTGCTTCAATATAACTAGAGA-3′. The KO efficiency for STAT3 and STAT5A/B was tested by western blotting.

CD70 KO in UMRC3 cells was performed using a Neon transfection system (Invitrogen). The sgRNAs were first diluted with nuclease-free water to a concentration of 44 µM. Cas9 was diluted with R buffer (Neon Electroporation kit, Invitrogen) at a Cas9 to R buffer ratio of 3:2. Diluted sgRNA and Cas9 were mixed to form the RNP complex at a 1:1 ratio and incubated for 20 min at room temperature. UMRC3 cells were collected and washed twice with PBS in aliquots of 100,000 cells each. The supernatant was removed, and the cells were resuspended in 12 µl of the RNP complexes along with an electroporation enhancer in R buffer. The final concentration for each electroporation was 1.8 µM sgRNAs, 1.6 µM Cas9 nuclease and 2.25 µM Cas9 electroporation enhancer. The cells were electroporated using a Neon Transfection system at 1,700 V, 20-ms pulse width and 1 pulse with 10 µl electroporation tips (Thermo Fisher Scientific, MPK5000). After electroporation, the cells were transferred into pre-warmed DMEM medium and cultured in a 37 °C incubator. The KO efficiency of CD70 was evaluated by flow cytometry.

Incucyte live cell spheroid and tumour rechallenge assays

GFP⁺ UMRC3 or BCX.010 spheroids were formed by plating 20,000 single cells in 100 μl medium in a 96-well clear round-bottom ultralow attachment spheroid microplate (Corning, 4520) followed by a 10-min centrifugation step at 120g. The plates were then placed in a 37 °C incubator for 24–48 h to allow for the spheroids to form. Once formed, the spheroids were treated with various NK cell conditions at the specified E/T ratios in technical duplicate or triplicate wells or left untreated as controls. Frames were captured with a ×10 objective at 4-h intervals over a number of days (as specified in each experiment). The green signal was quantified using an Incucyte S3 live-cell analysis system (Sartorius) in real-time and reported as the spheroid green image mean normalized to the green image mean immediately before the time of NK cell addition when the spheroid had already formed. Videos or images were exported using the same analysis system.

For the Raji tumour rechallenge assay, NK cells were cultured with Raji tumour cells labelled with mCherry and fresh tumour cells were added to this culture every 2–3 days in 96-well flat clear-bottom black microplates (Corning, 3904). Each well received 25,000 single Raji cells at each rechallenge, and duplicate or triplicate wells were used for each different condition. Images of each well were captured in real-time in five distinct regions per well. The counts of tumour cells for which the mCherry fluorochrome was detected were analysed using an Incucyte S3 live-cell analysis system that measures the number of target cells (fluorochrome labelled) in real-time.

xCELLigence killing and tumour rechallenge assays

Tumour cells were plated in 96-well RTCA E-plates (Agilent) for 24 h (to allow tumour cell adherence to the plate and growth) before NK cells were added at the designated E/T ratios. Only medium was added for tumour alone controls. Impedance was monitored over a number of days (specified individually for each experiment) at intervals of 15 min in a xCELLigence machine (Agilent) and reported as the normalized cell index, which was normalized to the cell index at the time NK cells were added using RTCA immunotherapy module software (Agilent).

For rechallenge assays using xCELLigence, the first killing assay was set up as described above. After that, new tumour cells were plated in new 96-well RTCA E-plates every 2–4 days and a new killing assay was started in the xCELLigence machine using RTCA immunotherapy module software. NK cells of the indicated conditions were transferred from the previous killing assay plate onto the newly plated tumours (after 24 h), and impedance was monitored over the next 2–4 days before the next rechallenge. Tumour growth is reported as the normalized cell index.

Mass cytometry and data analysis

Mass cytometry experiments and primary antibody conjugation were performed as previously described⁵⁷. In brief, cells were collected, washed with cell staining buffer (0.5% BSA in PBS) and incubated with 2.5 µM cisplatin (Sigma Aldrich) for viability assessment. Cells were washed, incubated with 5 µl human Fc receptor blocking solution (Trustain FcX, BioLegend) for 10 min at room temperature, and then stained for cell surface markers with a freshly prepared antibody mix for 30 min at room temperature on a shaker. After washing with cell staining buffer, samples were fixed and permeabilized using BD Cytofix/Cytoperm solution for 30 min in the dark at 4 °C, washed twice with Perm/Wash buffer and stained with antibodies directed against intracellular markers. Samples were then washed and stored overnight in 500 µl of 1.6% paraformaldehyde (EMD Biosciences) and PBS with 125 nM iridium nucleic acid intercalator (Fluidigm). The next day, samples were washed, filtered, counted and resuspended in MilliQ dH₂O supplemented with EQTM four element calibration beads at a concentration of 0.5 × 10⁵ per ml. Samples were acquired on a Helios instrument (Fluidigm) using Helios (v.6.5.358) acquisition software (Fluidigm). The following antibodies and corresponding metal tag isotopes were used: CD45 (Standard Biotools, 3089003B, HI30, 89Y, 1:167); CCR6 (Miltenyi Biotec, 130-108-023, REA190, 141Pr, 1:250); Eomes (Invitrogen, 14-4877-82, WD1928, 142Nd, 1:167); CD127 (Standard Biotools, 3143012B, A019D5, 143Nd, 1:200); GFP (BioLegend, 338002, FM264G, 144Nd, 1:250); CD70 (BioLegend, 355102, 113-16, 145Nd, 1:500); CD8a (Miltenyi Biotec, 130-122-281, REA734, 146Nd, 1:167); NKG2C (Miltenyi Biotec, 130-122-278, REA205, 147Sm, 1:125); TRAIL (Miltenyi Biotec, 130-126-490, REA1113, 148Nd, 1:125); CD25 (Standard Biotools, 3149010B, 2A3, 149Sm, 1:337); CD69 (Miltenyi Biotec, 130-124-326, REA824, 150Nd, 1:5,000); 2B4 (Miltenyi Biotec, 130-124-523, REA122, 151Eu, 1:5,000); CD95 (Miltenyi Biotec, 130-124-328, REA738, 152Sm, 1:500); panKIR (R&D systems, MAB1848, 180704, 153Eu, 1:337); CX3CR1 (Miltenyi Biotec, 130-122-286, REA385, 154Sm, 1:125); CD27 (Standard Biotools, 3155001B, L128, 155Gd, 1:250); CXCR3 (Standard Biotools, 3156004B, G025H7, 156Gd, 1:167); OX40 (Miltenyi Biotec, 130-095-212, REA621, 158Gd, 1:125); CD11c (Standard Biotools, 3159001B, Bu15, 159Tb, 1:250); T-bet (Standard Biotools, 3160010B, 4B10, 160Gd, 1:250); TIGIT (Miltenyi Biotec, 130-122-310, REA1004, 161Dy, 1:125); Ki67 (Standard Biotools, 3162012B, B56, 162Dy, 1:250); BTLA (Standard Biotools, 3163009B, MIH26, 163Dy, 1:67); CD73 (Miltenyi Biotec, 130-120-066, AD2, 164Dy, 1:125); TIM3 (Miltenyi Biotec, 130-122-333, REA635, 165Ho, 1:10); NKG2D (Standard Biotools, 3166016B, ON72, 166Er, 1:334); CREM (Creative Diagnostics, CABT-B10032, 4C6, 167Er, 1:500); KLRG1 (Miltenyi Biotec, 130-126-458, REA261, 168Er, 1:125); NKG2A (Standard Biotools, 3169013B, Z199, 169Tm, 1:500); CD161 (Miltenyi Biotec, 130-122-347, REA631, 170Er, 1:500); DNAM (Standard Biotools, 3171013B, DX11, 171Yb, 1:250); CD38 (Miltenyi Biotec, 130-122-307, REA572, 172Yb, 1:500); CXCR4 (Standard Biotools, 3173001B, 12G5, 173Yb, 1:167); PD1 (Miltenyi Biotec, 130-096-168, PD1.3.1.3, 174Yb, 1:125); LAG3 (Miltenyi Biotec, 130-124-529, REA351, 175Lu, 1:125); ICOS (Miltenyi Biotec, 130-122-304, REA192, 176Yb, 1:167); CD16 (Standard Biotools, 3209002B, 3G8, 209Bi, 1:167); CD57 (Miltenyi Biotec, 130-124-525, REA769, 115In, 1:500); CD39 (Miltenyi Biotec, 130-093-506, MZ18-23C8, Pt195, 1:100); perforin (Standard Biotools, 3196002B, B-D48, Pt196, 1:167); GZMB (Standard Biotools, 3198002B, GB11, Pt198, 1:167); CD56 (Miltenyi Biotec, 130-108-016, REA196, 106Cd, 1:167); CD2 (Miltenyi Biotec, 130-122-348, REA972, 111Cd, 1:334); HLA-DR (Miltenyi Biotec, 130-122-299, REA805, 112Cd, 1:250); NKp30 (Miltenyi Biotec, 130-092-554, AF29-4D12, 113Cd, 1:167); NKp46 (Miltenyi Biotec, 130-124-522, REA808, 114Cd, 1:125); and NKp44 (Miltenyi Biotec, 130-126-465, REA1163, 116Cd, 1:125).

Mass cytometry data were analysed using Cytobank. FCS files were first processed using FlowJo by removing beads and then gating singlets in Ir-191 and Ir-193 double-positive cells. NK cell populations were identified by Pt-195 (cisplatin)^low GFP^–CD45⁺CD56⁺. The gating strategy was applied to all files. Representative gating strategies of NK cells are shown in Supplementary Fig. 1. Data from 20,000 NK cell events for each condition were randomly subsampled in FlowJo using the DownSample plugin. Normalized data from donors representing the same condition were concatenated. Downstream analysis was performed on randomly sampled 10,000 events from each condition. NK cells from the various conditions were merged to create a single t-SNE CUDA map. The positive population for each marker was gated in Cytobank. The mean expression of each marker was normalized using the z score, then hierarchically clustered, and plotted as a heatmap, with the percentage positivity overlayed for each marker using Morpheus matrix visualization and analysis software (Broad Institute).

qPCR assay

NK cells were stimulated as described above. RNA was extracted using a RNeasy Plus Mini kit (Qiagen, 74134) according to the manufacturer’s protocol. cDNA was synthesized using an Iscript DNA Synthesis kit (Bio-Rad, 1708891). A qPCR reaction mix of 20 μl included 10 μl TaqMan Advanced Fast PCR master mix (Applied Biosystems, 4444557), 2 μl CREM or ICER PrimeTime Std qPCR assay (IDT), 2 μl cDNA and 6 μl nuclease-free water. The following primers were used.

CREM-specific exon (ENSE00003481920; Ensembl GRCh38.p14; Extended Data Fig. 3a) shared by the majority of the CREM non-ICER isoforms: primer 1 (FWD): 5′-ACTGAATGAACTGTCCTCTGATG-3′; primer 2 (REV): 5′-GTACTGCCATGGTAGCAATACT-3′.

ICER-specific exon 1 (ENSE00001890657, ENSE00001427126 or ENSE00001925783; Ensembl GRCh38.p14; Extended Data Fig. 3a): primer 1 (FWD): 5′-GATGTCAGTCCTCCTGCTTATC-3′; primer 2 (REV): 5′-CCTGTGTCATCTCCAGTTACAG-3′.

ICER-specific exon 2 (ENSE00001923116, ENSE00001835159 or ENSE00001384382; Ensembl GRCh38.p14; Extended Data Fig. 3a): primer 1 (FWD): 5′-TCAGTTCCTTTCCGCTTTGTA-3′; primer 2 (REV): 5′-TCAAGCAGACAACCACTTCA-3′.

qPCR was performed on an ABI 7500 Fast Real Time PCR system (Applied Biosystems). ΔC_T was calculated as the C_T of the target gene – the C_T of the corresponding internal control 18S. Relative expression was determined by normalizing the amount of each gene of interest to the experimental control using the ΔΔC_T method.

scRNA-seq

scRNA-seq data for CAR19–IL-15 NK cells from a non-curative in vivo model of lymphoma was obtained from a previous study¹⁴.

In data from the in vivo model, CAR19–IL-15 NK cells from time points with greater than 100 NK cells were retained (before and after infusion (day 7 and day 14)) and were processed using a standard Seurat workflow⁵⁸. In brief, data were normalized and scaled using the functions NormalizeData() with scale.factor = 10,000 and ScaleData(), respectively. A total of 2,000 variable genes were identified using FindVariableFeatures(). RunPCA() and RunUMAP() were run to generate a UMAP embedding using 20 principal components selected by inspecting the elbow plots. FindAllMarkers() was used to identify differentially expressed genes between cells before and after infusion, with log fold change threshold = 0 using Wilcox tests. Differentially expressed genes were identified at adjusted P values < 0.01 and absolute average log₂ fold change > 0.5. All analyses were performed in R (v.4.0.1) with Seurat (v.4.1.1). IL-15 activity was inferred from scRNA-seq using CytoSig⁵⁹.

Analysis of publicly available scRNA-seq datasets

scRNA-seq data were obtained from publicly available datasets and analysed for CREM expression. First, normalized CREM expression levels in various cell types from scRNA-seq cancer datasets were downloaded from the TISCH2 database (available at http://tisch.comp-genomics.org)^60,61. Datasets with available data for NK cell expression were included for this analysis.

To study CREM gene expression in TI-NK cells in the microenvironment of different types of cancer, we used the preprocessed scRNA-seq data and models of TI-NK cells prepared as previously described²⁶, which have been made available on Zenodo (https://doi.org/10.5281/zenodo.10139343) and as an online resource (http://nk-scrna.malmberglab.com/).

For the analysis of CREM gene expression in different types of NK cells, we studied the scRNA-seq datasets provided in previous study²⁸ that can be downloaded from the associated website (https://collections.cellatlas.io/meta-nk). The analysis of the above data and corresponding plotting were conducted using the Python based software scanpy. We followed the analytical steps detailed in the official website for scanpy software (https://scanpy.readthedocs.io/en/stable/tutorials/index.html).

Analysis of TCGA data

Gene expression and clinical data were obtained from TCGA (https://gdc.cancer.gov/about-data/publications/pancanatlas). Association of CREM with survival was performed using Cox regression implemented in the R package survival. Survival analysis was performed at the level of the entire cohort and individual cancers (with >400 samples) while controlling for tumour stage and age at diagnosis. For the survival analysis with the full cohort, we also controlled for the tumour type.

Regulon analysis

To identify key transcription factors and quantify the biological activity of their corresponding regulons in the pancreatic ductal adenocarcinoma scRNA‐seq dataset²⁷, we applied the single‐cell regulatory network inference and clustering (SCENIC) workflow, as previously described⁶². In brief, we used the Python implementation pySCENIC (with default parameters) on a high‐performance computing cluster. Regulatory interactions between a curated list of transcription factors and candidate target genes were inferred from scRNA‐seq co‐expression patterns using the GRNBoost2 algorithm implemented in Arboreto⁶³. This procedure generated an adjacency matrix linking each transcription factor to putative target genes along with an importance score. Next, we assembled candidate modules consisting of each transcription factor and its associated target genes. To distinguish direct from indirect regulatory targets, these modules were refined by selecting only those genes for which promoters contained the relevant transcription factor‐specific DNA motifs, as determined by RcisTarget motif enrichment analysis. We then quantified the relative biological activity of each refined module (regulon) at single‐cell resolution by calculating the AUC for every cell. Differential regulon activity across experimental groups was identified using the FindMarkers function in Seurat, and we visualized the top 50 differentially active regulons (adjusted P < 0.01) by generating a heatmap of the scaled AUC matrix with the DoHeatmap function.

ChIP assay

ChIP assays were performed at the MDACC Epigenomics Profiling Core as previously described^64,65 with some modifications. In brief, 20–30 million NK cells of the indicated conditions (NT, IL-15, CAR70–IL-15 and CAR70) were crosslinked with 1% formaldehyde for 10 min, quenched with 125 mM glycine for 5 min followed by chromatin preparation and sonication to obtain fragment sizes of 200–600 bp. In other experiments, NT NK cells were treated for 1 or 6 h with increasing concentrations of IL-15 (0, 500 and 5,000 pg ml^–1). Similarly, NK cells were then crosslinked with 1% formaldehyde for 10 min, quenched with 125 mM glycine for 5 min followed by chromatin preparation and sonication to obtain fragment sizes of 200–600 bp. ChIP was performed overnight with antibodies specific to pCREB (9198, Cell Signaling Technology, 1:50) and pSTAT3-Y705 (9145, Cell Signaling Technology, 1:100), STAT5B (13-5300, Invitrogen, 1:50), CREB (sc-240, 1:175), CREM (CABT-B10032, Creative Diagnostics, 1:100) and IgG (2729, Cell Signaling Technology). The immunocomplexes were collected the following day using DiaMag Protein A-coated magnetic beads (Diagenode, C03010020), washed and reverse-crosslinked overnight followed by DNA extraction. The DNA regions of interest were detected by SYBR real-time qPCR using oligonucleotides in putative promoter regions of CREM and STAT3 using primer pairs listed in Supplementary Table 1. For CREM ChIP–seq, input and CREM ChIP DNA libraries were prepared using a NEB Ultra II DNA library prep kit (New England Biolabs, E7645) following manufacturer’s instructions and subjected to next-generation sequencing to obtain about 30 million 50 bp paired-end reads per sample. Sequencing was performed at the MDACC Advanced Technology Genomics Core (ATGC) facility.

Bulk RNA-seq and analysis

NT NK cells, CREM WT CAR70–IL-15 NK cells and CREM KO CAR70–IL-15 NK cells were either cultured alone or incubated for 24 h with UMRC3 cells at an E/T ratio of 1:1. After that, NK cells were collected and sorted on a BD Biosciences Aria II cell sorter or a Beckman Coulter CytoFLEX SRT cell sorter at the MDACC North Campus Flow Cytometry and Cellular Imaging Core Facility. NK cells were sorted on single cells⁺live⁺GFP^–CD45⁺CD56⁺ cells. RNA was isolated from NK cells using a RNeasy Plus Mini kit (Qiagen, 74134) according to the manufacturer’s protocol. Barcoded, Illumina compatible, stranded mRNA libraries were prepared using a KAPA Stranded mRNA-Seq kit (Roche). In brief, 250 ng total RNA from each NK cell condition was captured using magnetic Oligo-dT beads. After bead elution and clean-up, the resultant PolyA RNA was fragmented using heat and magnesium. First-strand synthesis was performed using random priming followed by second-strand synthesis with the incorporation of dUTP into the second strand. The ends of the resulting double-stranded cDNA fragments were repaired, 5′-phosphorylated and 3′-A tailed, and Illumina-specific indexed adapters were ligated. The products were purified and enriched to generate the full-length library with nine cycles of PCR. The strand marked with dUTP was not amplified, which resulted in a strand-specific library. The libraries were quantified using a Qubit dsDNA HS Assay kit and assessed for size distribution using 4200 TapeStation High Sensitivity D1000 ScreenTape (Agilent Technologies) and then multiplexed, with 24 libraries per pool. The library pool was quantified by qPCR using a KAPA Library Quantification kit (Roche) and sequenced on a NovaSeq6000 SP flow cell (Illumina) using the 100 nucleotide paired-end format.

The raw FASTQ files were processed using the nf-core/rnaseq pipeline (v.3.14.0). All samples passed the built-in quality control. The resulting read mapping data in bigwig format and the expression quantification matrix in transcripts per million (TPM) were used for downstream analyses and visualization plots.

Bulk ATAC–seq and analysis

NK cells were prepared and sorted as for bulk RNA-seq preparation. ATAC–seq library preparation was performed at the MDACC Epigenomics Profiling Core following a previously described protocol^3,66 with minor modifications. Nuclei isolated from NK cells were tagmented using Tagment DNA TDE1 enzyme (Illumina) and the resulting libraries were purified using SPRISelect beads (Beckman Coulter). Libraries were sequenced on a NovaSeq6000 SP flow cell using the 100 nucleotide PE format with an 8 nucleotide single index.

For each bulk ATAC–seq sample, the pair-end reads from raw FASTQ files were aligned to the human genome (GRCh38) using bwa⁶⁷ mem mode with duplicated reads removed. The 5′ ends of ATAC–seq reads were shifted to the actual cut site of the transposase using the alignmentSieve module implemented in DeepTools. Peaks were called using MACS2 (ref. ⁶⁸) using the pair-end read information. The MACS2 outputs from multiple samples were loaded using DiffBind (v.3.8.4)⁶⁹. The peak sets from multiple samples were identified as the overlapping ones among samples using the UseSummarizeOverlaps function in DiffBind. Overall chromatin openness was visualized using Deeptools (v.3.5.2)⁷⁰. Motif activities for each sample were calculated using the function RunChromVAR in Signac (v.1.12.0)⁷¹ with the JASPAR motif database (2020 version)⁷². GeneActivities of each sample were calculated using the aggregation method implemented in Signac⁷³.

ChIP–seq analysis

The read processing of ChIP–seq data was the same as for ATAC–seq. We performed quality control through visualization of coverage enrichment along a gene body and filtered two samples (from two donors in the CAR70–IL-15 NK cell condition) that showed no enrichment of TSS regions. Thus, these samples were excluded from the analysis. Target genes were called for each sample when peaks were detected within 1 kb of the gene TSS using ChIPseeker (v.1.38.0)⁷⁴ and were removed if overlapping with the paired input (control) sample. Top targets (when aligned with RNA-seq) were called when the fold change was greater than 0.3 in log[TPM] between conditions. Functional enrichment of gene sets was performed using a hypergeometric test (HypeR package (v.2.0.0)⁷⁵). Differential enrichment of gene sets between conditions was performed using gsea (fgsea package (v.1.28.0)⁷⁶). Genome track visualization was performed using epiwraps (https://github.com/ETHZ-INS/epiwraps) using bigWig files as the original input.

Figure 5e was prepared using GSEA of upregulated and downregulated pathways in CREM KO versus WT CAR70–IL-15 NK cells as assessed by bulk RNA-seq. For this GSEA, only direct targets of CREM from CREM ChIP–seq were considered in each Hallmark pathway. Fig. 5h also represents GSEA. GSEA was applied as previously described⁷⁷ to test the enrichment of a given gene set by assessing whether the gene set members commonly rank at the top or bottom of a ranked gene list, with a model related to one-sided Kolmogorov–Smirnov tests.

Animal experiments

All procedures and experimental protocols involving mice were performed in accordance with the American Veterinary Medical Association and National Institutes of Health recommendations under protocols approved by the MD Anderson Cancer Center Institutional Animal Care and Use Committee (protocol number 00001263-RN01). We used 8–10-week-old NSG mice for the xenograft model for the mouse experiments to assess the antitumour activity of CREM KO CAR-NK cells in vivo. Where applicable, mice were randomized prior to therapy according to tumour size to ensure an equal tumour burden among all groups. Mice were injected and treated by an operator who was blinded to the treatment groups. Pathology analyses were performed by a pathologist who was blinded to the differences between the treatment groups and expected outcomes. 

For the Burkitt lymphoma Raji model, NSG mice were irradiated with 300 cGy. The following day, mice received 20,000 firefly luciferase-labelled Raji cells intravenously. Mice were then injected the same day with NK cells of the indicated conditions through the tail vein. To model exhaustion, we used older CAR70–IL-15 NK cells (23 days old) at a subtherapeutic dose (4 × 10⁶ CAR⁺ cells). Tumour growth was monitored using weekly bioluminescence imaging (Spectral Instrument Imaging (SII) system). Signal quantitation as the average radiance (p s^–1 cm^–2 sr^–1) was determined using Aura (v.4.0.7). Mice were either followed for survival or underwent timed euthanasia at days 10 and 20 after treatment. To study NK cell engraftment, blood was collected on days 10 and 20 after NK cell injection and analysed by flow cytometry. For the mice that underwent timed euthanasia, liver, lung, spleen and bone marrow were collected. The organs were processed and either stained to assess NK cell engraftment by flow cytometry or pooled for ex vivo functional assays. The following kits were used for tissue processing according to the manufacturer’s protocols: Lung Dissociation kit, mouse (Miltenyi Biotec, 130-095-927), Liver Dissociation kit, mouse (Miltenyi Biotec, 130-105-807) and Spleen Dissociation Kit, mouse (Miltenyi Biotec, 130-095-926).

In the metastatic PDX mouse model of breast cancer, female NSG mice received an injection of 300,000 BCX.010 cells through the tail vein on day −7, followed by irradiation on day −1. NK cells were injected on day 0 through the tail vein. We performed two independent experiments: one with timed euthanasia of the mice and one for survival using three CB donors infused at subcurative doses. One donor was used for the timed euthanasia experiment and two donors were used for the survival experiment. Flow cytometry analysis of NK cell kinetics was conducted by drawing blood on days 10 and 20 after NK cell injection. To investigate the effectiveness and migration of NK cells to tumour sites in the BCX.010 model, mice were euthanized on day 35 after NK cell injection in the timed euthanasia experiment. Lungs and livers were collected, fixed and paraffin-embedded. Tumour nodules were quantified by a pathologist through haematoxylin and eosin (H&E) staining of lung and liver sections using bright-field microscopy. Immunohistochemistry staining for hCD45 and GZMB was performed on lung and liver sections to identify NK cells and cytotoxic NK cells, respectively. Whole slide digital imaging was performed using an Aperio AT2 after immunohistochemical staining on a Leica Bond RX autostainer for hCD45 and GZMB on adjacent serial sections. Images were deconvoluted in HALO (v.3.6) Deconvolution module (v.1.1.8). Deconvoluted images were subsequently registered and fused using the same HALO software.

To investigate safety concerns with CREM KO CAR-NK cells, we performed an in vivo experiment based on the BCX.010 infusion model. As described above, mice engrafted with the BCX.010 cells were treated with CREM WT or CREM KO CAR70–IL-15 NK cells 1 week later. As a control, CREM KO CAR70–IL-15 NK cells were infused in mice without tumours. Two weeks later, at the peak of NK cell engraftment, mice were euthanized and the histology of sections of vital organs (liver, lungs and kidneys) was evaluated by a veterinary pathologist who is board certified by the American College of Veterinary Pathologists and the American Board of Toxicology. Blood was also collected from another group of mice 30 days after treatment with either CREM WT or CREM KO CAR70–IL-15 NK cells and analysed for haematological parameters and chemistry to evaluate for acute and subacute toxicity. For the BCX.010 model in vivo experiments, NK cells were infused at a dose of 1 or 3 × 10⁶ cells based on the pre-infusion in vitro activity for each donor.

For the orthotopic mouse model of human pancreatic cancer, NSG mice underwent surgical orthotopic implantation of 300,000 PATC148 cells directly into the pancreas. One week later, mice were intraperitoneally treated with the indicated NK cell products (5 × 10⁶ cells) after irradiation one day before. Blood was similarly drawn on days 10 and 20 after NK cell treatment. To assess tumour burden by histology, mice were euthanized on day 36 after NK cell treatment. Pancreases were collected, fixed and paraffin embedded. Tumour nodules were quantified after H&E staining by a pathologist.

Seahorse assays

Glycolysis measured on the basis of the extracellular acidification rate (ECAR) and mitochondrial respiration measured on the basis of the oxygen consumption rate (OCR) were assayed using an Agilent Seahorse XF Analyzer (Agilent) following the manufacturer’s protocol. Seahorse glycolysis stress tests were performed using 2 g l⁻¹ d-glucose, 2.5 μM oligomycin and 50 mM 2-deoxyglucose mixed with Hoechst 33342 (Invitrogen) dye. Seahorse Mito stress tests were performed using 2.5 μM oligomycin, 0.5 μM FCCP and 0.5 μM rotenone–antimycin A mixed with Hoechst 33342 (Invitrogen) dye. Technical triplicates for each biological donor replicate were used. Following the assays, viable cells were counted using live-cell imaging and counting in a Cytation 1 machine. Normalized OCR or ECAR data per 1,000,000 live NK cells are shown. The basal respiration was calculated using the following formula: last rate measurement before first injection − non-mitochondrial respiration rate. This value represents the minimum rate measurement after rotenone–antimycin A treatment. The maximal respiration was calculated using the following formula: maximum rate measurement after FCCP injection − non-mitochondrial respiration. The baseline glycolysis represents the non-glycolytic acidification (the last rate measurement before glucose injection). The glycolytic capacity was calculated using the following formula: maximum rate measurement after oligomycin injection − last rate measurement before glucose injection.

Statistics

Data were collected and organized using Microsoft Excel for Mac 2023. Statistical analyses were conducted with GraphPad Prism (Prism 9 and 10, GraphPad Software). Quantitative differences were assessed using either ANOVA for multiple groups (with multiple comparisons when applicable) or a t-test for two groups. Statistical significance was defined as P < 0.05. Mean values ± s.e.m. were used to represent the data, unless otherwise specified. The specific statistical tests used and corresponding sample sizes (n) are detailed in each figure legend. No statistical methods were used to predetermine sample sizes. 

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.