The nuclear factor ID3 endows macrophages with a potent anti-tumour activity – Nature

Materials

Mice

Animal procedures were performed in adherence with the Institutional Review Board (IACUC 15-04-006) at Memorial Sloan Kettering Cancer Center (MSKCC). Rosa26^LSL-YFP (ref. ⁵⁵), Rosa26^LSL-tdT(ref. ⁵⁶), Flt3^cre (ref. ⁵⁷), Tnfrsf11a^cre (ref. ⁵⁸), Id3^f/f (ref. ⁵⁹), Id3^−/− (ref. ⁶⁰), Cx3cr1^gfp/+ (ref. ⁶¹), CCR2^−/−(ref. ³⁰), Csf1r^f/f(ref. ⁶²), Spi1^f/f(ref. ⁶³), Cxcr4^gfp/+ (ref. ⁶⁴), Cxcr4^creERT2(ref. ⁶⁵), Clec4f^cre-tdT(ref. ²⁵), Rosa26^LSL-DTR(ref. ⁶⁶), p48^cre (ref. ⁶⁷), Trp53^LSL-R172H(ref. ⁶⁸), Kras^LSL-G12D(ref. ⁶⁹), CD45.1 mice and C57BL/6J mice were purchased from Jackson Lab. Mice were bred under SPF conditions, under a 12 h–12 h light–dark cycle, at around 21–22 °C with 30–70% humidity. A list of mouse strains and the genotyping protocol is provided in Supplementary Data 4.

Human tissue samples

All of the procedures performed in studies involving human participants were conducted according to the Declaration of Helsinki. Human tissues were obtained with patient-informed consent and used under approval by the Institutional Review Boards from Memorial Sloan Kettering Cancer Center (IRB protocols, 15-021).

A list of the reagents, plasmids, antibodies and qPCR primers purchased and used in this study is provided in Supplementary Data 5.

Mouse cell lines

KPC-1 and KPC-2 cell lines⁷⁰ were obtained from pancreas tissue from p48^creTrp53^LSL-R172HKras^LSL-G12D mice⁷¹. Pan02^72,73 (DCTD Tumour Repository) cells were provided by D. Lyden. The Colon adenocarcinoma cell line MC38^73,74 and melanoma cell line B16F10⁷⁵ (ATCC, CRL-6475 was a gift from J. D. Wolchok). Lewis lung carcinoma line LLC1 cells⁷⁶ (ATCC, CRL-1642) were purchased from ATCC. KPC-1, KPC-2 and PANC-1 cells were cultured in DMEM (Gibco) supplemented with 10% fetal bovine serum, 100 U ml⁻¹ penicillin and 100 μg ml⁻¹ streptomycin (Invitrogen). Panc02 cells, LLC1 cells, MC38 cells and B16F10 cells were cultured in RPMI 1640 (Gibco) supplemented with 10% fetal bovine serum, 100 U ml⁻¹ penicillin and 100 μg ml⁻¹ streptomycin (Invitrogen) and incubated at 37 °C in 5% CO₂.

Mouse primary cells

Mouse BMDMs were obtained as follows. Femur, tibia and iliac bones from C57BL/6J mice were flushed with PBS, red blood cells were lysed using red blood cell lysis buffer (eBioscience) and bone marrow cells were seeded per 15 cm non-tissue culture plate in DMEM (Thermo Fisher Scientific) with 10% FBS (Thermo Fisher Scientific), 20 ng ml⁻¹ M-CSF (315-02-50ug, PeproTech), 100 U ml⁻¹ penicillin–streptomycin (Thermo Fisher Scientific) for 7 days.

Mouse splenic NK cells

Mouse splenic NK cells were obtained as follows. Splenic cell suspensions from C57BL/6J mice were dissociated and passed through a 100 μm cell strainer (BD), red blood cells were lysed using red blood cell lysis buffer (eBioscience) and resuspended in 50 μl of blocking buffer containing anti-mouse CD16/32 antibodies (1:100) for 15 min at 4 °C, followed by staining with PE-anti-NKP46 antibodies for 30 min at 4 °C and anti-PE microbeads (Miltenyi Biotec) for 30 min at 4 °C, respectively. NKP46⁺ NK cells were isolated by passing stained samples through the Miltenyi Biotec Magnetic separation system using LS columns according to the manufacturer’s instructions.

Human cell lines

PANC-1 cells⁷⁷ (ATCC, CRL-1469) were purchased from ATCC. Human macrophages were obtained from hiPS cell lines derived from frozen peripheral blood mononuclear cells of two independent healthy donors. Written informed consent was obtained according to the Helsinki convention. The study was approved by the Institutional Review Board of St Thomas’ Hospital; Guy’s hospital; the King’s College London University and the Memorial Sloan Kettering Cancer Center. hiPS cells were derived according to published protocols⁷⁸ using Sendai viral vectors (Thermo Fisher Scientific; A16517). Newly derived iPS cell clones were maintained in culture for 10 passages (2–3 months) to remove any traces of Sendai viral particles and ensure that the cells remain stable during a prolonged culturing period. Over 90% of iPSCs in the derived lines expressed high levels of the pluripotency markers NANOG and OCT4, as determined using flow cytometry. Karyotyping analysis showed a normal karyotype (46, XX). iPS cell lines tested negative for mycoplasma contamination using MycoAlert Plus kit (Lonza). iPS cell clones that passed all of the quality-control checks were frozen down and used for downstream experiments. hiPS cells were maintained on irradiated CF1 mouse embryonic fibroblasts (MEFs; Thermo Fisher Scientific; A34181) in embryonic stem (ES) cell medium supplemented with 10 ng ml⁻¹ basic fibroblast growth factor (bFGF, Peprotech; 100-18B). The medium was changed every other day. Passaging was performed every 7 days at 1/4–1/6 dilution ratio depending on colony size. During passaging, iPS cells were detached as clusters by incubation for 13 min at 37 °C with collagenase type IV (250 UI ml⁻¹ final concentration) (Thermo Fisher Scientific; 17104019) and were pelleted at room temperature by centrifugation at 150g.

hiPSC-Macs

iPS cell clusters were resuspended in ES cell medium supplemented with 10 ng ml⁻¹ bFGF (Peprotech; 100-18B) and plated onto NUNC plates containing 12,500 to 16,000 MEFs per cm². hiPSC-Macs were obtained using a previously published protocol⁷⁹, modified as follows. At day 0 of differentiation, expanded hiPS cells were detached as described above and transferred (from a 150 mm plate, to four wells) for cultivation in six-well low-adhesion plates in ES cell medium supplemented with 10 μM ROCK inhibitor (Sigma-Aldrich; Y0503). The plates were kept on an orbital shaker at 100 rpm for 6 days to allow for the spontaneous formation of embryoid bodies with haematopoietic potential. At day 6 of differentiation, 200–500 μm cystic embryoid bodies were picked under a dissecting microscope and transferred onto adherent tissue culture plates (around 2.5 embryoid bodies per cm²) for cultivation in HD medium. At day 18 of differentiation, macrophages produced by embryoid bodies were collected from suspension and cultivated on tissue culture plates at a density of around 10,000 cells per cm² in MC medium for 6 days in before use for downstream experiments. All cells were cultured at 37 °C, 5% CO₂, in standard tissue culture incubators.

Human primary cells

Human NK cells

Human NK cells (IQ Biosciences, IQB-Hu1-NK5) were cultured in NK MACS medium (Miltenyi Biotec) supplemented with 10% heat-inactivated pooled human AB serum (Sigma-Aldrich), 100 U ml⁻¹ penicillin and 100 μg ml⁻¹ streptomycin (Invitrogen) and 20 ng ml⁻¹ hIL-2 incubated at 37 °C in 5% CO₂.

Human CD8 T cells

Human CD8 T cells (IQ Biosciences, IQB-Hu1-CD8T10) were cultured with RPMI supplemented with 10% heat-inactivated pooled human AB serum, 100 U ml⁻¹ penicillin and 100 μg ml⁻¹ streptomycin and 20 ng ml⁻¹ hIL-2 incubated at 37 °C in 5% CO₂.

Methodology

In vivo treatment with DT, CSF1R inhibitor, phosphatidylserine blockade, SIRPA blocking antibodies, and NK and CD8 T cell blocking antibodies

For DT-mediated depletion of KCs, Clec4f^creRosa26^LSL-DTR mice and Rosa26^LSL-DTR mice were intraperitoneally injected with 100 ng DT (D0564-1MG, Sigma-Aldrich) as a single dose or weekly injections²⁵, as indicated in the figure legends for the corresponding experiments. The efficiency of KC depletion was determined in Extended Data Fig. 1j.

For CSF1R inhibitor treatment with PLX5622³¹, C57BL/6J mice were fed ad libitum with PLX5622-impregnated chow (1,200 mg per kg, provided by Plexxicon) or control chow 2 weeks before injection of tumour cells.

For the SIRPA blockage assay, Clec4f^creId3^f/f mice and Id3^f/f littermates were intraperitoneally injected with control IgG (HRPN, BioXcell) or 250 μg anti-SIRPA (P84, BioXcell) 2 days before and every 2 days after tumour cell injection.

For phosphatidylserine blockade⁸⁰, C57BL/6J mice were injected i.v. with PBS or 1 μg MFG-E8(D89E)⁸¹ (gift from S. Nagata), 6 h before injection of tumour cells.

For NK cell and CD8 T cell depletion, 8–12-week-old Clec4f^creId3^f/f mice and Id3^f/f littermates were intraperitoneally injected with control IgG (HRPN, BioXcell), or 200 μg anti-NK1.1 (BE0036, BioXcell) and 200 μg anti-CD8 (BP0061, BioXcell) 1 day before tumour injection and every 4 days afterwards.

Metastasis-initiating cell assays

In vitro oncosphere formation

Sphere-formation assays⁸² were performed by sorting 1,000 CD9⁺CD133⁺ KPC-1 cells per well and 1,000 CD9⁻CD133⁻ KPC-1 cells per well and plating them in ultra-low-attachment 96-well plates (Corning) in DMEM/F-12 medium supplemented with B-27 serum (1:50, Invitrogen), 20 ng ml⁻¹ bFGF (R&D Systems) and 50 U ml⁻¹ penicillin–streptomycin for a total of 7 days. Images were acquired using the Leica DM IL inverted phase-contrast microscope with Leica application SuiteX software and quantified using ImageJ.

In vivo metastasis assay

In vivo metastasis assays were performed by sorting 5 × 10⁴ or 2 × 10⁵ CD47^brightCD9⁺CD133⁺ tumour cells and CD47^lowCD9^lowCD133^low tumour cells from KPC-1-GFP tumour cells, followed by intraportal injection into C57BL/6J mice for 1 week. Metastatic potential was determined by bioluminescence imaging analysis (see the ‘Short-term liver metastasis model (intraportal injection of tumour cell lines)’ section below).

Transduction of mouse and human tumour cell lines

To generate luci-tdT-, luci-GFP- and mtdT-expressing tumour cell lines, KPC-1, KPC-2, PAN02, MC38, B16F10 and LLC1 cells were seeded at a density of 5 × 10⁵ cells per well in six-well culture plates. Lentiviral supernatant carrying the luciferase-gfp gene (plasmid pFUGW-FerH-ffLuc2-eGFP, Addgene, 71393) or the luciferase-tdT gene (plasmid pUltra-Chili-Luc, Addgene, 48688) and 10 µg ml⁻¹ polybrene (Millipore-Sigma) were added to the tumour cell culture medium for 12 h. The medium was then replaced and, 72 h later, GFP⁺ or tdT⁺ tumour cells were FACS-sorted three times before in vivo injection in mice.

To generate mtdT-expressing tumour cells, KPC-1 and PANC-1 cells were seeded at a density of 5 × 10⁵ cells per well in a six-well culture plate. The lentiviral supernatant carrying retroviruses expressing the mtdT gene (plasmid pQC membrane tdTomato IX, Addgene, 37351) and 10 µg ml⁻¹ polybrene (Millipore-Sigma) were added to the tumour cell culture medium for 12 h. The medium was then replaced and, 72 h later, tdT⁺ tumour cells were FACS-sorted three times before use in experiments.

To generate Cd47-KO KPC tumour cells, sgRNAs targeting Cd47 sequence, sgCD47 5′-CCTTGCATCGTCCGTAATG-3′⁸³ were cloned into pSpCas9(BB)-2A-Puro (PX459) V2.0 (Addgene, 62988) according to the Addgene cloning protocol. To establish Cd47-KO KPC cell line, electroporation of pSpCas9-sgCD47 plasmid into 1 × 10⁶ KPC-1-luci-tdT cells was performed according to the Neon Transfection System protocol. Cells were FACS-sorted based on the loss of CD47 staining with anti-mouse CD47-AF647 antibodies (1:200, BioLegend) for three rounds to get pure populations of Cd47-KO cells.

Transduction of macrophages

For lentiviral transduction of BMDMs, BMDMs were seeded at a density of 1 × 10⁶ cells per well of six-well plate after 5 days culture of bone marrow cells. Lentiviral supernatant carrying control, mouse Id3 gene, mouse shMYC, shHES1, shE2A, shELK1, scramble (Santa Cruz) and 10 µg ml⁻¹ polybrene (Millipore-Sigma) were added to BMDM culture medium for 12 h. The medium was then replaced and, 48 h later, transduced cells were selected with 2 μg ml⁻¹ puromycin for 4 days. The transduced BMDMs were analysed by FACS, RT–qPCR, time-lapse imaging or in vivo rescue experiments.

For lentiviral transduction of KCs, KCs were seeded at a density of 8 × 10⁵ cells per well of six-well plates in the presence of 20 ng ml⁻¹ M-CSF and transduced with lentiviral supernatant carrying control, mouse shE2A, shELK1, VPX supernatant (pSIV3-VPX plasmids, a gift by M. Ménager) and 10 µg ml⁻¹ polybrene for 12 h. The medium was then replaced and, 48 h later, transduced cells were selected with 1 μg ml⁻¹ puromycin for 3 days. The transduced KCs were analysed by FACS.

For Lentiviral transduction of hiPSC-Macs, hiPSC-Macs were seeded at a density of 5 × 10⁵ cells per well of a six-well plate in the presence of 20 ng ml⁻¹ M-CSF and transduced with lentiviral supernatant carrying control, human ID3 gene, VPX supernatant and 10 µg ml⁻¹ polybrene for 12 h. The medium was then replaced and, 48 h later, transduced cells were selected with 1 μg ml⁻¹ puromycin for 3–4 days. The transduced hiPSC-Macs were analysed by RT–qPCR and time-lapse imaging.

Flow cytometry, cell sorting and cell counting

Blood and BM cell preparation

Mouse blood cells were obtained as follows. Mice were anaesthetized by intraperitoneal injection of ketamine/xylazine/acepromazine anaesthesia cocktail. Blood were collected using the cardiac puncture approach. In brief, mice were placed on their back and the needle of 1 ml syringes was inserted (pretreated with 1 ml 100 mM EDTA buffer) under the rib cage. The plunger was gently pulled to collect blood. Red blood cells were lysed using red blood cell lysis buffer (eBioscience). Mouse bone marrow cells were obtained as follows. The femur, tibia and iliac bones from mice were flushed with PBS, and red blood cells were lysed using red blood cell lysis buffer.

Tissue cell suspension preparation

Mice were perfused with 10 ml PBS under terminal anaesthesia, tissue samples were minced into small pieces and incubated in digestion buffer containing 1× PBS, collagenase D (1 mg ml⁻¹, Sigma-Aldrich), dispase (2.4 mg ml⁻¹, Thermo Fisher Scientific), DNase (0.2 mg ml⁻¹, Sigma-Aldrich) and 3% heat-inactivated fetal bovine serum (FBS, Invitrogen) for 30 min at 37 °C. Cells suspensions were dissociated and passed through a 100 μm cell strainer (BD) and resuspended in 50 μl of blocking buffer containing 1× PBS, 0.5% BSA, 2 mM EDTA, anti-mouse CD16/32 (1:100), 5% normal rat, 5% normal mouse and 5% normal rabbit serum (Jackson ImmunoResearch) for 15 min at 4 °C. The samples were stained with the indicated antibodies (a list of which is provided in Supplementary Data 5; 1:200) for 30 min at 4 °C. Flow cytometry was performed using the BD Biosciences LSR Fortessa flow cytometer with Diva software. All data were analysed using FlowJo v.10.6 (Tree Star).

Staining and gating strategies

Mouse liver macrophage/myeloid cell panels were as follows: Pop1 macrophages, Hoechst⁻CD45⁺CD3⁻CD19⁻Nkp46⁻Ly6G⁻F4/80⁺TIM4⁺; Pop2 macrophages, Hoechst⁻CD45⁺CD3⁻CD19⁻NKP46⁻Ly6G⁻F4/80⁺TIM4⁻MHCII⁺; Pop3 myeloid cells, Hoechst⁻CD45⁺CD3⁻CD19⁻NKP46⁻Ly6G⁻F4/80⁺TIM4⁻MHCII⁻. KC subsets: KC subset 1, Hoechst⁻CD45⁺CD3⁻CD19⁻NKP46⁻Ly6G⁻F4/80⁺TIM4⁺CD206⁺; KC subset 2, Hoechst⁻CD45⁺CD3⁻CD19⁻NKP46⁻Ly6G⁻F4/80⁺TIM4⁺CD206^high. Other mouse macrophage panels were as follows: kidney macrophages, Hoechst⁻CD45⁺CD3⁻CD19⁻NKP46⁻CD11b^lowF4/80^bright. Brain macrophages, Hoechst⁻CD45⁺CD3⁻CD19⁻NKP46⁻CD11b⁺F4/80⁺. Lung alveolar macrophages, Hoechst⁻CD45⁺CD11b⁻CD11c⁺CD64⁺SIGLECF⁺. Lung interstitial macrophages, Hoechst⁻CD45⁺CD11b⁺Ly6G⁻CD64⁺. Skin macrophages, Hoechst⁻CD45⁺CD3⁻CD19⁻NKP46⁻CD11b⁺F4/80⁺. Spleen RPM, Hoechst⁻CD45⁺CD3⁻CD19⁻NKP46⁻CD11b^lowF4/80⁺. Pancreas macrophages, Hoechst⁻CD45⁺CD3⁻CD19⁻NKP46⁻F4/80⁺. Mouse myeloid cells panels were as follows: cDC1, Hoechst⁻CD45⁺F4/80⁻Ly6G⁻CD11b⁻CD11c⁺MHCII⁺. cDC2, Hoechst⁻CD45⁺F4/80⁻Ly6G⁻CD11b⁺CD11c⁺MHCII⁺. Ly6C⁺ monocytes (blood), Hoechst⁻CD3⁻CD19⁻NKP46⁻CD11b⁺CD115⁺Ly6C⁺. Ly6C⁺ monocytes (spleen, liver), Hoechst⁻CD45⁺CD3⁻CD19⁻NKP46⁻F4/80⁻Ly6G⁻CD11b⁺Ly6C^high. Ly6G⁺ granulocytes (blood), Hoechst⁻CD3⁻CD19⁻NKP46⁻Ly6G⁺. Ly6G⁺ granulocytes (spleen, liver), Hoechst⁻CD45⁺CD3⁻CD19⁻F4/80⁻Nkp46⁻F4/80⁻Ly6G⁺. Mouse lymphocyte panels were as follows: NKT cells (liver), Hoechst⁻CD45⁺F4/80⁻TCRβ⁺CD1dTetramers⁺; γδT cells (liver), Hoechst⁻CD45⁺F4/80⁻CD3⁺TCRβ⁻TCRγδ⁺; CD3⁺ T cells (spleen, liver), Hoechst⁻CD45⁺F4/80⁻CD3⁺. CD8⁺ T cells (spleen, liver), Hoechst⁻CD45⁺F4/80⁻CD3⁺CD8⁺; CD4⁺ T cells (spleen, liver), Hoechst⁻CD45⁺F4/80⁻CD3⁺CD4⁺; CD19⁺ cells (spleen, liver), Hoechst⁻CD45⁺F4/80⁻CD3⁻CD19⁺; NKP46⁺ cells (spleen, liver), Hoechst⁻CD45⁺F4/80⁻CD3⁻NKP46⁺. CD3⁺ T cells (blood), Hoechst⁻Ly6G⁻NKP46⁻CD19⁻CD3⁺. CD8⁺ T cells (blood), Hoechst⁻Ly6G⁻Nkp46⁻CD19⁻CD3⁺CD8⁺; CD4⁺ T cells (blood), Hoechst⁻Ly6G⁻NKP46⁻CD19⁻CD3⁺CD4⁺; CD19⁺ cells (blood), Hoechst⁻Ly6G⁻NKP46⁻CD19⁺; NKP46⁺ cells (blood), Hoechst⁻Ly6G⁻CD19⁻CD3⁻Nkp46⁺. Mouse bone marrow panels were as follows: long-term HSCs (LT-HSCs), Hoechst⁻CD3⁻CD19⁻NKP46⁻Ly6G⁻Kit⁺SCA1⁺CD150⁺CD48⁻; short-term HSCs (ST-HSCs), Hoechst⁻CD3⁻CD19⁻NKP46⁻Ly6G⁻Kit⁺SCA1⁺CD150⁻CD48⁻; multipotent progenitors (MPPs), Hoechst⁻CD3⁻CD19⁻NKP46⁻Ly6G⁻Kit⁺SCA1⁺CD150⁻CD48⁺. Human lymphocyte intracellular staining panels were as follows: CD8⁺ T cells, Hoechst⁻CD8⁺IFNγ⁺TNF⁺. CD56⁺ NK cells, Hoechst⁻CD56⁺IFNγ⁺. See Supplementary Figs. 1–3 and Supplementary Tables 1 and 2.

Cell counting

Cells number was assessed using a cell counter (GUAVA easyCyte HT).

Cell sorting

Cell sorting was performed using the Aria III BD cell sorter. Single live cells were gated on DAPI⁻ and using forward scatter width (FSC-W) and FSC-A to exclude doublets.

Cytokine intracellular staining

For IFNγ and TNF intracellular staining in CD8⁺ and CD8⁻ T cells, mouse liver cell suspension, and human CD8⁺ T cells were treated with a cocktail of phorbol 12-myristate 13-acetate (PMA), ionomycin, brefeldin A and monensin (Thermo Fisher Scientific) for 4–12 h. Staining for IFNγ and TNF was performed using the eBioscience Transcription Factor Staining kit (Thermo Fisher Scientific) according to the manufacturer’s instructions. IFNγ and TNF production were analysed by flow cytometry.

IFNγ intracellular staining of mouse splenic NK cells and human CD56⁺ NK cells was performed using the eBioscience Transcription Factor Staining kit according to the manufacturer’s instructions. IFNγ production was analysed using flow cytometry.

Lineage tracing of bone-marrow-derived cells using genetic labelling

Bone-marrow-derived cells are labelled in Cxcr4^gfp/+and Cx3cr1^gfp/+ mice, and by a single injection of 4-OH TAM (37.5 mg per kg body weight) supplemented with progesterone (18.75 mg per kg body weight) in 6-week-old Cxcr4^creERT2Rosa26^LSL-tdTmice (ref. ⁶⁵) (Extended Data Fig. 2k,o). Cxcr4^gfp/+ and Cx3cr1^gfp/+ mice, and Cxcr4^creERT2Rosa26^LSL-tdT mice 2 weeks after 4-OH TAM injection, were injected with 1 × 10⁶ KPC-1 cells through the portal vein and euthanized 2 weeks later for the analysis of tdT⁺ cells, YFP⁺ cells or GFP⁺ cells among liver macrophage subsets.

Parabiosis

Generation of CD45.2 Id3
^+/+/CD45.1 and CD45.2 Id3
^−/−/CD45.1 parabionts

Female 6–8-week-old CD45.2 Id3^+/+ and CD45.2 Id3^−/− host parabionts were generated with age- and weight-matched female CD45.1 mice through parabiosis surgery described previously⁸⁴. Mice were maintained on a trimethoprim/sulfamethoxazole diet after surgery to minimize infection. After 8 weeks, parabiotic mice were separated and perfused with phosphate-buffered saline (PBS). Partner-derived CD45.1⁺ cells in TIM4⁺ KCs were determined by flow cytometry analysis.

Generation of CD45.2/CD45.1 parabionts

Female 6–8-week-old congenic CD45.1 and CD45.2 mice were connected through parabiosis surgery. After 8 weeks, CD45.2 parabionts were injected with 1 × 10⁶ KPC-1-luci-tdT cells through the portal vein to induce liver metastasis. Liver samples were collected 2 weeks after tumour injection. Flow cytometry and immunofluorescence imaging were performed to analyse liver macrophage exchange ratios.

RT–qPCR

Mouse KCs and BMDMs and human iPSC-macrophages were lysed directly on tissue culture plates and RNA was extracted using the quick-RNA Microprep kit (Zymo research; R1050) in accordance with manufacturer’s instructions. cDNA preparation was performed using Quantitect Reverse transcription kit (Qiagen; 205313) according to the manufacturer’s protocol. RT–qPCR was performed on the Quant Studio 6 Flex System with 10 ng cDNA per reaction using probes (Supplementary Data 5) and TaqMan Fast Advance Mastermix (Thermo Fisher Scientific; 4444557), or PowerUp SYBR Green Master Mix (Thermo Fisher Scientific; A25742), according to the manufacturer’s instructions. Expression values for each tested gene relative to a GAPDH endogenous control were calculated using the ΔΔC_t method according to the formula: \({2}^{-({{\rm{C}}}_{{t}_{{\rm{t}}{\rm{e}}{\rm{s}}{\rm{t}}{\rm{g}}{\rm{e}}{\rm{n}}{\rm{e}}}}-{{\rm{C}}}_{{t}_{GAPDH}})}\).

Cytology

Cytospin preparation was performed using Cytospin 3 (Thermo Fisher Scientific, Shandon) and Cytofunnels (Thermo Fisher Scientific, BMP-CYTO-DB25) by centrifuging sorted cells onto Super-frost slides (Thermo Fisher Scientific, 12-550-15) at 800 rpm for 10 min (medium acceleration). The slides were air-dried for at least 30 min and fixed for 10 min in 100% methanol (Thermo Fisher Scientific, A412SK-4). Methanol-fixed cells were stained in 50% May–Grünwald solution (Sigma-Aldrich, MG500-500mL) for 5 min, 5% Giemsa (Sigma-Aldrich, 48900-500mL-F) for 15 min and washed with Sorensons buffered distilled water (pH 6.8) three times for 2 min after each staining. Slides were mounted with Entellan New (Millipore, 1079610100) after air-drying, representative pictures were taken using an Axio Lab.A1 microscope (Zeiss) under a N-Achroplan ×100/01.25 objective.

Bioluminescence imaging

Depending on the experiment, bioluminescence imaging was conducted either on isolated organs (ex vivo) in long-term orthotopic pancreatic tumour experiments, or in anaesthetized mice (in vivo) in short-term models (below) on the In ViVo Imaging System spectrum (Perkin Elmer). Quantification of bioluminescence images was performed using LivingImage (v.2.60.1; Perkin Elmer). Photoradiance was measured as photons s⁻¹ cm⁻² sr⁻¹.

Immunofluorescence and whole-mount imaging

Immunofluorescence imaging of mouse liver

Mice were perfused with 10 ml PBS, liver samples were dissected and fixed overnight at 4 °C with PLP fixative in phosphate buffer⁸⁴. After the PBS wash, livers were dehydrated in 30% sucrose in PBS and embedded in OCT. Cryoblocks were cut at a thickness of 16 μm and blocked with PBS containing 5% normal goat serum (Jackson ImmunoResearch), 1% BSA and 0.3% Triton X-100 (Sigma-Aldrich) for 1 h at room temperature. The samples were incubated with anti-mouse-F4/80-eF450/AF647/AF488/eF570 (1:200, BM8, eBioscience), anti-mouse TIM4-AF647/PE (1:200, RMT4-54, BioLegend), anti-mouse CD45.1-AF488 (1:200, A20, BioLegend), chicken-anti-GFP (1:500, A10262, Invitrogen, recognize YFP), rabbit-anti-RFP (1:200, 600-401-379, Rockland), goat-anti mouse CLEC4F (1:200, AF2784, R&D Systems), anti-mouse CCL3 (1:200, 50-7532-82, Thermo Fisher Scientific), anti-mouse CCL4 (1:200, AF-451-NA, Thermo Fisher Scientific), anti-mouse CCL5 (1:200, 701030, Thermo Fisher Scientific), anti-mouse IL-12P70 (1:200, MM121B, Thermo Fisher Scientific), anti-mouse IL-15 (1:200, AF447-SP, Thermo Fisher Scientific), anti-mouse IL-18 (1:200, PA5-79481, Thermo Fisher Scientific) antibodies for 2 h at room temperature. Secondary antibody staining using anti-chicken-Alexa Fluor 488 (1:500; a11039, Thermo Fisher Scientific), anti-rabbit-Alexa Fluor 555 (1:500, A32794, Thermo Fisher Scientific), anti-rabbit-Alexa Fluor 647 (1:500, A32795, Thermo Fisher Scientific), anti-goat-Alexa Fluor 555 (1:500, a32816, Thermo Fisher Scientific), anti-goat-Alexa Fluor 647 (1:500, a21447, Thermo Fisher Scientific), anti-goat-Alexa Fluor 488 (1:500, a32814, Thermo Fisher Scientific), anti-sheep-Alexa Fluor 568 (1:500, A21099, Thermo Fisher Scientific), Streptavidin-Alexa Fluor 647 (1:500, 405237, BioLegend) were performed for 1 h at room temperature. Nuclei were counterstained with DAPI (Invitrogen). The sections were mounted with Fluoromount-G (eBiosciences). Images were acquired on a Zeiss LSM880 confocal microscope using an oil-immersion ×40/1.4 NA objective.

Immunofluorescence imaging of the metastatic liver of human patients with PDAC

Metastatic liver samples of human patients with PDAC were embedded in OCT. Cryoblocks were cut at a thickness of 16 μm, fixed with 4% paraformaldehyde (PFA) for 30 min, blocked with PBS containing 5% normal goat serum (Jackson ImmunoResearch), 1% BSA and 0.3% Triton X-100 (Sigma-Aldrich) for 1 h.

The samples were incubated with anti-human CD14-AF488 (1:200, BD), sheep-anti-human CK19 (1:200, AF3506, R&D Systems), rabbit-anti-human TIM4 (1:200, PA5-53346, Thermo Fisher Scientific), goat-anti-human IL-12 (1:200, AF-219-NA, R&D Systems), goat-anti-human IL-18 (1:200, AF2548, R&D Systems), mouse-anti-human IL-15 (1:200, MAB2471, R&D Systems), goat-anti-human CCL3 (1:200, AF-270-NA, R&D Systems), goat-anti-human CCL4 (1:200, AF-271-NA, R&D Systems) and goat-anti-human CCL5 (1:200, AF-278-NA, R&D Systems) antibodies for 2 h at room temperature. The samples were then incubated with anti-rabbit-Alexa Fluor 647 (1:500, A32795, Thermo Fisher Scientific), anti-rabbit-Alexa Fluor 555 (1:500, A32794, Thermo Fisher Scientific), anti-sheep-Alexa Fluor 568 (1:500, A21099, Thermo Fisher Scientific), anti-goat-Alexa Fluor 647 (1:500, a21447, Thermo Fisher Scientific) and anti-mouse-Alexa Fluor 555 (1:500, A-31570, Thermo Fisher Scientific) for 1 h. Nuclei were stained with DAPI for 10 min. Sections were mounted with Fluoromount-G (eBiosciences). Images were acquired on a Zeiss LSM880 confocal microscope using an oil-immersion ×40/1.4 NA objective.

Whole-mount immunofluorescence imaging of mouse liver

Liver pieces were fixed in 4% PFA diluted in PBS for 30 min at room temperature with agitation. The samples were permeabilized with 1× PBS containing 0.3% Triton X-100, 4% BSA for 1 h at room temperature and incubated with an anti-F4/80-eF450 (1:100, eBioscience), anti-TIM4-AF647 (1:100, BioLegend) antibody mix for 2 h at room temperature. Data were acquired using the LSM880 Zeiss microscope. Imaris v.9.3.1 (Bitplane) was used to analyse the acquired images.

Quantification of chemokines/cytokines and tumour materials relative MFI in KCs

Quantification of tdT relative MFI in tdT⁺ KC lysosomes

Immunofluorescence staining for F4/80, TIM4, LAMP1⁺ and tdT was performed on frozen liver sections from C57BL/6J mice 2 weeks after intraportal injection of 1 × 10⁶ KPC-1–tdT cells. Images were acquired on the Zeiss LSM880 confocal microscope. Imaris v.9.3.1 (Bitplane) was used to reconstruct the 3D surface of LAMP1⁺ lysosome or 97.5 μm² non-LAMP1⁻ region in tdT⁺ KCs. The tdT relative mean fluorescence intensity (MFI) in LAMP1⁺ lysosomes was determined by normalizing to a non-LAMP1⁻ region MFI of 1.

Quantification of chemokine/cytokine relative MFI in mouse KCs

Immunofluorescence staining for CCL3, CCL4, CCL5, IL-12P70, IL-15, IL-18, GFP, F4/80 and TIM4 was performed on frozen liver sections from C57BL/6J mice, Clec4f^creId3^f/f mice and Id3^f/f littermates 2 weeks after intraportal injection of 1 × 10⁶ KPC-1-GFP cells. Images were acquired on the Zeiss LSM880 confocal microscope. Imaris v.9.3.1 (Bitplane) was used to reconstruct the 3D surface of F4/80⁺TIM4⁺ KCs. Chemokine/cytokine relative MFI in F4/80⁺TIM4⁺ KCs was determined by normalizing to an MFI of 1 in the tumour-free region (distance from tumour > 50 μm) in C57BL/6J mice or Id3^f/f mice.

Quantification of chemokine/cytokine and CK19⁺ tumour material relative MFI in human KCs

Immunofluorescence staining for CCL3, CCL4, CCL5, IL-12, IL-15, IL-18, CK19, CD14 and TIM4 on frozen liver sections from human patients with PDAC metastatic liver. Images were acquired on the Zeiss LSM880 confocal microscope. Imaris v.9.3.1 (Bitplane) was used to reconstruct the 3D surface of CD14⁺TIM4⁺ KCs. The chemokine/cytokine and CK19⁺ tumour material relative MFI in CD14⁺TIM4⁺ KCs was determined by normalizing to a background MFI of 1.

Tumour growth, metastasis and rescue models

Endogenous KPC model

KPC mice heterozygous for p48^cre, Trp53^LSL-R172H and Kras^LSL-G12D alleles⁷¹ were generated by crossing p48^cre (ref. ⁶⁷), Trp53^LSL-R172H (ref. ⁶⁸) and Kras^LSL-G12D(ref. ⁶⁹) mice under SPF conditions. KPC mice were monitored on a regular basis to check for symptoms of abdominal distension; moribund animals were euthanized by CO₂ asphyxiation according to IACUC guidelines. Mice were euthanized at 6 months and livers were collected and fixed as described above. Quantification of CK19⁺ tumour material, chemokine and cytokine relative MFI was performed as follows. Frozen liver sections from KPC mice and control (Kras^LSL-G12DTrp53^LSL-R172H) mice were subjected to immunofluorescence staining for CCL3, CCL4, CCL5, IL-12P70, IL-15, IL-18, CK19, F4/80 and TIM4. Images were acquired on the Zeiss LSM880 confocal microscope. Imaris v.9.3.1 (Bitplane) was used to reconstruct the 3D surface of F4/80⁺TIM4⁺ KCs. CK19⁺ tumour material, chemokine and cytokine relative MFI in F4/80⁺TIM4⁺ KCs was determined by normalization to Kras^LSL-G12DTrp53^LSL-R172H mice MFI of 1.

Long-term orthotopic pancreatic tumour model

Orthotopic injection of pancreatic cell lines into the pancreas was performed according to a published protocol⁸⁵. Mice were anaesthetized under isoflurane gas, sterile sharp scissors were used to cut a single incision off the abdominal skin and muscle above the pancreas, the pancreas was gently positioned to allow slow injection into the pancreas of 2 × 10⁵ KPC-2-luci-tdT or 2 × 10⁵ KPC-2-luci-GFP cells per mouse, resuspended in 50 μl of PBS and Matrigel (354234, corning) at 2/1 ratio using 31 G insulin syringes (BD). The pancreas was gently placed back into the abdominal cavity. The muscle layer was closed using sterile absorbable vicryl suture (J463G, Ethicf on). Skin edges were closed with sterile 9 mm wound clips (Braintree Scientific). Then, 8 weeks after pancreatic orthotopic injection, the mice received retro-orbital injection of 1 mg d-luciferin (Goldbio Technology) in 100 μl sterile water, and the liver, spleen, lungs and pancreas were dissected for ex vivo analysis by bioluminescence imaging. Livers were collected and fixed as described above, and the percentage of tdT⁺TIM4⁺ cells in TIM4⁺ KCs was analysed by whole-mount imaging as described in the ‘Whole-mount immunofluorescence imaging of mouse liver’ section. (Extended Data Fig. 3f).

Conditional depletion of KCs by DT

For conditional depletion of KCs by DT, in the indicated experiments, Clec4f^creRosa26^LSL-DTR and Rosa26^LSL-DTR littermates received weekly intraperitoneal injection of 100 ng DT, starting 1 week after pancreatic orthotopic injection of tumour cells.

Short-term liver metastasis model (intraportal injection of tumour cell lines)

In total, 1 × 10⁶ KPC-1-luci-GFP cells, 5 × 10⁵ B16F10-luci-GFP cells, 1 × 10⁶ LLC1-luci-GFP cells, 1 × 10⁶ MC38-luci-GFP cells, or 1 × 10⁶ KPC-1-luci-tdT cells, 5 × 10⁵ B16F10-luci-tdT cells, 1 × 10⁶ LLC1-luci-tdT cells or 1 × 10⁶ Pan02-luci-tdT cells were resuspended in 50 μl PBS. Mice were anaesthetized with isoflurane gas and sterile sharp scissors were used to cut a single incision off the abdominal skin and muscle. While holding the median side of the incision aside with forceps, including the skin and peritoneal lining, a sterile cotton swab was used to carefully pull the large and small intestines out until the portal vein is visualized. After covering the intestines with the sterile gauze soaked in sterile PBS, a 31 G needle (BD) loaded with tumour cells was inserted into the portal vein below the liver and the full volume (50 µl) was slowly injected. The needle was then removed while simultaneously placing a sterile cotton tip applicator on the vein with pressure, for 5 min, to keep the injection site intact. The internal organs were gently placed back into the abdominal cavity. The muscle layer was closed using sterile absorbable vicryl suture (J463G, Ethicon). Skin edges were closed with sterile 9 mm wound clips (Braintree Scientific). Tumour-bearing mice were analysed as follows.

Conditional depletion of KCs by DT

In the indicated experiments, Clec4f^creRosa26^LSL-DTR and Rosa26^LSL-DTR littermates received intraperitoneal injection of 100 ng DT before and/or after the tumour cell grafts as described above.

Survival experiments

In the indicated experiments, the cohort of tumour-bearing mice was examined by a veterinarian twice a week for 5 weeks. Moribund animals were determined by the veterinarian and euthanized by CO₂ asphyxiation according to IACUC guidelines. Comparison of survival curves was performed using log-rank (Mantel–Cox) tests (Figs. 1g, 3j and 6g).

Liver tumour burden at 24 h

In the indicated experiments, CD45⁻GFP⁺ or CD45⁻tdT⁺ liver tumour cell numbers in tumour-bearing mice were analysed by flow cytometry. The percentage of GFP⁺TIM4⁺ cells or the percentage of tdT⁺TIM4⁺ cells in KCs was analysed by immunofluorescence staining.

Liver tumour burden at 2 weeks

In the indicated experiments, the liver tumour burden at 2 weeks was assessed by in vivo bioluminescence imaging. Chemokine and cytokine production by KCs was analysed using RT–qPCR and immunofluorescence staining, and the percentage of GFP⁺TIM4⁺ cells or the percentage of tdT⁺TIM4⁺ cells in KCs was analysed by immunofluorescence staining. The numbers of immune cells were analysed using flow cytometry and immunofluorescence staining. The production of IFNγ and TNF by NK cells or CD8 T cells was analysed using flow cytometry.

Rescue of KPC liver metastasis in Clec4f
^cre
Id3
^f/f mice by BMDMs

Clec4f^creId3^f/f mice (aged 6–12 weeks) received 1 × 10⁶ KPC-1-luci-GFP cells by intraportal injection, followed after 1 week (day 7 after tumour injection) by intraportal injection of either 1 × 10⁶ BMDMs expressing lenti-control or lenti-mId3 cells. The tumour burden was performed 14 days after tumour injection by bioluminescent images described above.

Rescue of LLC liver metastasis in wild-type mice by BMDMs

C57BL/6J mice (aged 6–8 weeks) received 1 × 10⁶ LLC1-luci cells by intraportal injection, followed after one week (day 7 after tumour injection) by intraportal injection of 1 × 10⁶ BMDMs expressing lenti-control, lenti-mId3 cells or not. The tumour burden was analysed 14 days after tumour injection by bioluminescence imaging as described above. Survival was analysed as described above. Comparison of survival curves was performed using log-rank (Mantel–Cox) tests.

B16F10 melanoma subcutaneous tumours and rescue by BMDMs

C57/BL6J mice (aged 6–12 weeks) received subcutaneous injection of 1 × 10⁶ B16F10-luci-GFP cells, into the left and right flank, followed by intratumour injection of 5 × 10⁵ BMDMs expressing lenti-control, lenti-mId3 cells at day 7 after tumour injection. Then, 7 days later, the tumour burden was assessed by in vivo bioluminescence imaging as described above. The numbers of immune cells and the production of IFNγ and TNF by NK cells or CD8⁺ T cells were analysed using flow cytometry.

Intravital imaging of liver KCs and KPC-1-mtdT tumour cells in vivo

C57BL/6J mice (aged 6–12 weeks) were injected with 1 × 10⁶ KPC-1-mtdT cells through intraportal injection as described above. Then, 2 weeks after tumour cell injection, the mice anaesthetized under isoflurane gas and anaesthesia was maintained through continuous inhalation of Isoflurane (0.5 l min⁻¹) in oxygen through a nose cone. The mice were then injected retro-orbitally with 5 μl CellEvent caspase-3/7-green reagent (Invitrogen), a four-amino-acid peptide (DEVD) caspase-3/7 cleavage reporter conjugated to a nucleic-acid-binding dye that becomes fluorescent when bound to DNA (Cas-Green) to monitor tumour cell apoptosis and death⁸⁶, and 10 μl of anti-TIM4-AF647 antibodies (BioLegend) in 50 μl PBS to label apoptotic/dead cells and KCs, respectively. Sterile eye lubricant was applied to both eyes to prevent corneal drying during the experiment. A 1.5 cm horizontal incision was cut off the skin and muscle above the liver, and the liver left lobe was extruded gently. The mouse was then inverted and positioned on a custom-made aluminium tray stage inserted through circular 2.5 cm diameter hole, covered with a glass coverslip that was attached with silicone grease. PBS-soaked sheets of paper were prepositioned on the cover slip to surround the area of the exposed liver, then the mouse was ready for intravital imaging. During the whole imaging period, PBS was gently added every 20 min on both sides of the mouse to keep the area moist. A thermostat-controlled heated chamber keeps the whole microscope, mice, tray and microscope objectives at 32 °C to prevent hypothermia during the experiment. Imaging was performed using the Zeiss LSM880 confocal laser scanning microscope. Acquisition of CellEvent caspase-3/7-green, tdT and AF647 fluorescence signals was performed in line in a single channel. The power used for each laser line was as follows; 1%, 488 nm; 1%, 568 nm; and 5%, 647 nm—the lowest required to obtain a sufficient signal for each fluorescent probe and chosen to minimize phototoxicity. Seven consecutive stacks with an interval of 2.5 µm were captured using a Zeiss Plan-Apochromat ×20/0.75 objective, with digital zoom set to 1, every 1 min per position for up to 8 h. Time-lapse videos and 3D surface reconstructions were generated using Imaris v.9.3.1 (Bitplane).

In vitro mouse coculture assays

Engulfment assay: ex vivo 3D co-culture and time-lapse imaging of KCs and KPC-1-mtdT tumour cells

F4/80⁺TIM4⁺ KCs were sorted from freshly isolated liver of C57BL/6J or Id3^−/− mice using digestion buffer and the antibody panel described in the ‘Flow cytometry, cell sorting and cell counting’ section. In total, 1 × 10⁴ KCs were mixed with 2 × 10³ KPC-1-mtdT cells (5:1 ratio), embedded in growth-factor-reduced Matrigel (356231, Corning) and cultured overnight in a 24-well µ-plate (Ibidi) with DMEM medium in the presence of 20 ng ml⁻¹ M-CSF and, when indicated, with 1 µg D89E, 60 mm latrunculin A, 50 µg ml⁻¹ anti-SIRPA (P84, BioXcell) or 20 μg ml⁻¹ anti-dectin-1 (R1-8g7, InvivoGen). Before imaging, the co-cultures were stained with 2 μM CellEvent caspase-3/7-green reagent (Invitrogen) and anti-F4/80-AF647 antibodies (1:200, BioLegend) for 30 min. Imaging was performed using the Zeiss LSM880 confocal laser-scanning microscope equipped with an imaging chamber maintained at 37 °C, 5% CO₂, 20% O₂ and 90% relative humidity. Five consecutive stacks at an interval of 2.5 µm were captured using the Zeiss LD C-Apochromat ×40/1.1 water-immersion objective (x = 212.55 µm, y = 212.55 µm, z = 15 µm) every 5 min per position for 20 h. The data were analysed using Imaris v.9.3.1 (Bitplane). For each sample, the percentage of engulfing KCs was determined by averaging the percentage of F4/80⁺ cells engulfing live tumour cells (CellEvent, cleavage caspase-3/7⁻ tumour cells) from at least three simultaneously imaged fields of view. Time to engulfment values were determined as the time from stable interaction between macrophages and tumour cells to tumour cell engulfment. Time to Cas cleavage was determined as the time from stable interaction between macrophages and tumour cells to the detection of caspase-3/7 cleavage. The total n numbers of macrophages tracked per each sample are indicated.

Engulfment assay: ex vivo 3D co-culture and time-lapse imaging of BMDMs and KPC-1-mtdT tumour cells

For BMDM/KPC-1–mtdT tumour cell time-lapse imaging, 1 × 10⁴ BMDMs expressing lenti-control or lenti-mouse-Id3 were mixed with 2 × 10³ KPC-1-mtdT cells (5:1 ratio), embedded in growth-factor-reduced Matrigel (356231, Corning) and cultured overnight in a 24-well µ-plate (Ibidi) with DMEM medium in the presence of 20 ng ml⁻¹ M-CSF and, when indicated, with 50 µg ml⁻¹ anti-SIRPA (P84, BioXcell). before imaging, the co-cultures were stained with 2 μM CellEvent caspase-3/7-green reagent (Invitrogen), anti-F4/80-AF647 antibodies (1:200, BioLegend) for 30 min. Imaging was performed using the Zeiss LSM880 confocal laser-scanning microscope and analysed as described in the ‘Engulfment assay: ex vivo 3D co-culture and time-lapse imaging of KCs and KPC-1-mtdT tumour cells’ section.

Production of chemokines and cytokines by KCs in a coculture assay with tumour cells

In total, 3 × 10⁵ KCs from Id3^f/f or Clec4f^creId3^f/f mice were seeded in a 12-well plate with 1.5 × 10⁵ KPC tumour cells or not. Then, 48 h later, the supernatants from the coculture were collected for the following study. The production of chemokines and cytokines by KCs was analysed using RT–qPCR described as above.

Role of supernatants in lymphoid cell activation

Mouse NK cell/supernatant coculture assay

Mouse splenic NK cells from C57BL/6J mice were seeded in a 96-well round-bottom plate at 3 × 10⁴ cells per well in 100 μl NK culture medium (RPMI supplemented with 10% FBS, 100 U ml⁻¹ penicillin, 100 μg ml⁻¹ streptomycin and 20 ng ml⁻¹ mIL-2) in the presence of 100 μl of the above-mentioned supernatant. Then, 3 days later, IFNγ production by NK cells was analysed using flow cytometry.

In vitro human coculture assays

Engulfment assay: ex vivo 3D co-culture and time-lapse imaging of hiPSC-Macs and PANC-1-mtdT tumour cells

For hiPSC-Mac/PANC-1–mtdT tumour cell time-lapse imaging, a total of 1 × 10⁴ hiPSC-Macs expressing lenti-control or lenti-human-ID3 were mixed with 2 × 10³ PANC-1-mtdT cells (5:1 ratio), embedded in growth-factor-reduced Matrigel (356231, Corning) and cultured overnight in a 24-well µ-plate (Ibidi) with RPMI1640 medium in the presence of 100 ng ml⁻¹ M-CSF. Before imaging, the co-cultures were stained with 2 μM CellEvent caspase-3/7-green reagent (Invitrogen), anti-CD14-AF647 antibodies (1:200, BioLegend) for 30 min. Imaging was performed using the Zeiss LSM880 confocal laser-scanning microscope and analysed as described in the ‘Engulfment assay: ex vivo 3D co-culture and time-lapse imaging of KCs and KPC-1-mtdT tumour cells’ section.

Production of chemokines, cytokines by human macrophage in coculture assay with tumour cells

In total, 10⁵ hiPSC-Macs expressing lenti-control or lenti-human-Id3 were seeded in a 12-well plate, and treated or not with 5 x 10⁴ PANC-1 tumour cells. Then, 48 h later, the coculture supernatants were collected for the following study. The production of chemokines and cytokines by hiPSC-Macs was analysed using RT–qPCR described as above.

Role of supernatants in lymphoid cell activation

For the human NK cell/supernatant coculture assay, human NK cells were seeded in a 96-well round-bottom plate at 10⁴ cells per well in 100 μl NK culture medium (see above) in the presence of 100 μl of the above-mentioned supernatant. Then, 3 days later, IFNγ production were analysed by flow cytometry. For the human CD8 T cell/supernatant coculture assay, human CD8⁺ T cells, stained with CFSE, were seeded in a 96-well round-bottom plate at 10⁴ cells per well in 100 μl CD8 culture medium (see above) in the presence of 100 μl of the above-mentioned supernatant, then PBS washed anti-hCD3/hCD28 activation beads were added to the medium. Then, 3 days later, the samples were treated with a cocktail of PMA, ionomycin, brefeldin A and monensin for 6 h. CFSE proliferation, TNF and IFNγ production by CD8 T cells were analysed using flow cytometry.

Identification of candidate functional E2A and ELK1 motifs

The position weight matrix (PWM) of E2A/TCF3 and ELK1 motifs was downloaded from the JASPAR database with motif IDs MA0522.1 and MA0028.2, respectively⁸⁷. To find motif matches, we first computed a motif score or PWM score for all of the putative regulatory elements of KCs at the Sirpa locus and filtered with a minimum PWM score cut-off that passed a false-positive rate of <0.2%. We then computed the DeepLIFT scores based on a deep learning model (see below) at every regulatory element and overlaid these scores with motif matches for E2A and ELK1 to predict functional motifs. Our final set of functional motifs all have a PWM score exceeding the score cut-off and at least three positions within top 20% based on DeepLIFT scores.

Training and interpretation of the deep learning model

The deep learning model was trained and interpreted as described previously⁸⁸. In brief, we adapted a strategy of AgentBind⁸⁹ and fine-tuned a pretrained DeepSEA model⁹⁰ using all active enhancers in KCs on the basis of previously published ATAC–seq and H3K27ac ChIP–seq data under Gene Expression Omnibus (GEO) GSE128338 (ref. ⁹¹). The AgentBind model consists of (1) pretraining convolutional neural networks, which infer important sequence context features and learn combinations and orientations of these features that are predictive of binding, using ChIP–seq and DNase-I-sequencing profiles collected from ENCODE18 and the Epigenomics Roadmap Project20 across dozens of cell types; and (2) fine-tuning an individual model for each transcription factor to identify bound versus unbound sequences, described previously³⁸. The DeepSEA model (deep learning–based sequence analyzer) is a fully sequence-based algorithmic framework for non-coding-variant effect prediction, described previously³⁹. The software used for this methodology was as follows: Python (v.3), Keras (v.2.3.1), tensorflow (v.2.1.0), scikit-learn (v.0.21.3), deeplift (v.0.6.10.0) and biopython (v.1.76). Training data were prepared as follows. Positive data labelled as 1 were 300 bp sequences of ATAC–seq peaks associated with strong levels of H3K27ac. We first obtained the processed data file from GEO GSE128338, which includes the reproducible ATAC–seq peaks merged from KCs of both healthy and NASH-diet mice and their tag counts of H3K27ac ChIP–seq in the expanded 2,000 bp regions⁹¹. We removed sex chromosomes and filtered the peaks with a minimum cut-off of 32 tags of H3K27ac ChIP–seq. The positive sequences were balanced with the same number of 300 bp negative sequences, which were GC-content-matched random genomic regions selected from the mm10 genome and were labelled as 0. During the training, we left out sequences on chromosome 8 for cross-validation and those on chromosome 9 for testing. The final model had an area under the receiver operating characteristic curve (auROC) equal to 0.828 on the testing data. We next used DeepLIFT⁹² to generate importance scores with single-nucleotide resolution using uniform nucleotide backgrounds. For each input sequence, we generated two sets of scores, one for the original sequence and the other for its reverse complement. The final scores were the absolute maximum at each aligned position. We defined predicted functional nucleotides by the top 20% (that is, top 60) positions within each input 300 bp sequence.

Experimental analysis of candidates E2A and ELK1 binding motifs using CUT&RUN

CUT&RUN was performed using the Epicypher (14-1048) kit according to the manufacturer’s protocol with modifications. A total of 2 × 10⁵ TIM4⁺ KCs was sorted from the liver and resuspended in 1 ml nucleus isolation buffer (0.5 mM Tris, pH 8.0, 0.5 mM EDTA, 5 mM magnesium chloride, 0.1 M sucrose, 0.05% Triton X-100, 1× EDTA-free protease inhibitor (11836170001, Sigma-Aldrich)) and incubated on ice for 10 min. Nuclei were resuspended in 100 μl wash buffer, DNA was purified using the QIAamp DNA Micro Kit, using 5 µl of the sample as the input. The rest of the sample was mixed with 10 µl concanavalin A beads and rotated at room temperature for 30 min. The supernatant was removed by placing beads and nuclei mixture on a magnetic stand. Nuclei were resuspended in 50 µl antibody buffer mixed with 3 µl rabbit-anti-E2A (gift from the K. Murre laboratory, made by D. Wiest), or 4 µl rabbit-anti-Elk1 (Cell Signaling Technology, 9182 S) and incubated overnight at 4 °C. The next morning, nuclei were washed in wash buffer twice, resuspended in 50 µl cell permeabilization buffer containing 2.5 µl pAG-MNase and incubated for 10 min at room temperature. After two washes, 1 µl of chromatin digest additive was added, and the samples were incubated at 4 °C for 2 h with rotation. After addition of 33 µl of stop buffer, the samples were incubated for 10 min at 37 °C. The tubes were then placed onto a magnetic stand and the supernatant containing enriched DNA was transferred to 1.5 ml tubes. DNA was purified using the Chip DNA Clean&Concentrator kit (Zymo research). CUT&RUN-enriched DNA and input DNA were analysed by qPCR on the QuantStudio (TM) 6 Flex System (Applied Biosystems) with the Power SYBR Green PCR Master Mix (Thermo Fisher Scientific, A25742) and calculated as the percentage of input.

Bulk RNA-seq analysis

In total, 80,000 KCs per sample were FACS-sorted into 1.5 ml Eppendorf tubes with 800 µl TRIzol LS Reagent (Thermo Fisher Scientific, 15596018) or in 1.5 ml Eppendorf tubes precoated with 10% BSA. RNA samples were submitted to the Integrated Genomics Operation (IGO) at MSKCC for quality and quantity analysis, library preparation and sequencing. In brief, phase separation in cells lysed in TRIzol Reagent was induced with chloroform. RNA was precipitated with isopropanol and linear acrylamide and washed with 75% ethanol. The samples were resuspended in RNase-free water. After RiboGreen quantification and quality control using the Agilent BioAnalyzer, 2 ng total RNA with RNA integrity numbers ranging from 9.4 to 10 underwent amplification using the SMART-Seq v4 Ultra Low Input RNA Kit (Clonetech, 63488), with 12 cycles of amplification. Subsequently, 10 ng of amplified cDNA was used to prepare libraries with the KAPA Hyper Prep Kit (Kapa Biosystems KK8504) using 8 cycles of PCR. The samples were barcoded and run on the HiSeq 4000 system in a paired-end 100 bp run, using the HiSeq 3000/4000 SBS Kit (Illumina). For RNA-seq data processing and analysis, sequenced reads from the RNA-seq were aligned to the mouse reference genome GRCm39 or mm10 using STAR (v.2.7.10a)⁹³. The aligned reads were quantified as gene counts using HTSeq⁹⁴ with GENCODE release M30⁹⁵. DESeq2⁹⁶ was applied to the gene counts table to identify differentially expressed genes (DEGs). Adjusted P values were determined with DEseq2 using the Benjamini–Hochberg method for multiple comparisons with two-sided tests. DEGs were ranked on the basis of their log₂-transformed fold change and associated P values (P_adj < 0.05). For gene set enrichment analysis, pathways enriched in the ranked DEGs were identified against the mouse Molecular Signatures Database (MSigDB)⁹⁷ pathway collection (P_adj < 0.25) using the fgsea package in R, and the most biologically informative lists are shown.

scRNA-seq analysis

The CRC dataset (GSE146409)⁹⁸ contained three patients with colorectal liver metastasis and a non-tumour individual. The PDAC dataset (GSE205013)⁹⁹ contained three patients with PDAC liver metastasis. The non-tumour dataset (GSE115469)⁵¹ control contained five non-tumour individuals. scRNA-seq analysis was conducted using the Seurat package (v.4.3.0) in R studio (4.2.0). For each dataset, quality control was performed by retaining cells with nFeature_RNA > 200 but <10,000, and mitochondrial content <10%. The PDAC and the non-tumour datasets were integrated using the SCTransform workflow. First, the two Seurat objects were merged, normalized and scaled with 2,000 features. Subsequently, the SCTransform() function was applied to the merged object with the dataset source regressed out. Linear dimensionality reduction was applied to the SCT assay and the first 50 principal components. Harmony (v.0.1.1) was used to correct the dataset and samples. The clustering analysis was based on the harmonized Seurat object. The first 40 principal components were used in the RunTSNE() and FindNeighbors() functions, whereas the resolution parameter was set to 1.8 in the FindCluster() function. For the other parameters unspecified above, the default values were used in the Seurat workflow. The CRC dataset was preintegrated by the authors, so we performed the analysis using the standard Seurat pipeline. Clusters were visualized in a two-dimensional t-distributed stochastic neighbour embedding (t-SNE) and were annotated using differential expressed marker genes based on the human protein atlas (https://www.proteinatlas.org/). The expression patterns of characteristic genes were presented in the t-SNE plot. Expression data of characteristic genes in KC and TAM clusters were extracted and presented in violin plots using the ggplot2 package (v.3.4.1). Average expression levels in each cluster were labelled on the violin plots. Adjusted P values were obtained with the Seurat FindMarkers() function using Wilcoxon tests and Bonferroni correction based on the total number of genes in the dataset.

Statistics and reproducibility

Analysis of bulk RNA-seq and scRNA-seq data is included in the corresponding sections. For other experiments, error bars in graphical data represent mean ± s.d. Statistical significance was determined using two-tailed Student’s t-tests and ANOVA for normally distributed data, or Mann–Whitney U-tests and Kruskal–Wallis tests when data were not normally distributed based on Shapiro–Wilk test or Anderson–Darling test (P < 0.05) (Supplementary Tables 1 and 2). Comparison of survival curves was performed using log-rank (Mantel–Cox) tests. P < 0.05 was considered to be statistically significant. Statistical analyses were performed using GraphPad Prism. The n values represent biological replicates unless otherwise specified in the legend. Experiments were repeated to ensure the reproducibility of the observations. No statistical methods were used to predetermine sample size. Results were obtained from at least three (Figs. 1a–e, 2a,f,g, 3h,e, 4g, 5e and 6a,h and Extended Data Figs. 1c, 2a–g, 4a,b, 5b,d,e, 6c–h, 7b–d,f, 8b and 10a,c) and two (Figs. 1h–j, 2b,d,e,h, 3a,c,d,j, 4f,h,i,l,k, 5c,g and 6g and Extended Data Figs. 1b,d,m, 2h, 3e, 5a,f, 7a,g, 8g and 10d) independent experiments.

Reporting summary

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