Mice, diets, and treatments
Table of Contents
Standard mouse diet feeding (ad libitum water and food access) and treatment regimens were as described previously17. Male mice were housed at the German Cancer Research Center (DKFZ) (constant temperature of 20–24 °C and 45–65% humidity with a 12-h light–dark cycle). Mice were maintained under specific pathogen-free conditions and experiments were performed in accordance with German law and the governmental bodies, and with approval from the Regierungspräsidium Karlsruhe (G11/16, G129/16, G7/17). Tissues from inducible knock-in mice expressing the human unconventional prefoldin RPB5 interactor were received from N. Djouder17,33. The plasmids for hydrodynamic tail-vein delivery have been described previously34,35,36,37. For interventional studies, male mice fed a CD-HFD were treated with bi-weekly for 8 weeks by intravenous injection of 25 μg CD8-depleting antibody (Bioxcell, 2.43), 50 μg NK1.1-depleting antibody (Bioxcell, PK136), 300 μg anti-PDL1 (Bioxcell, 10F.9G2), 200 μg anti-TNF (Bioxcell, XT3.11), 100 μg anti-CD4 (Bioxcell, GK1.5), or 150 μg anti-PD1 (Bioxcell, RMP1-14). PD1−/− mice were kindly provided by G. Tiegs and K. Neumann. Mice for Extended Data Fig. 3g were treated with anti-PD1 antibody (Bioxcell, RMP1-14) or isotype control (Bioxcell, 2A3) at an initial dose of 500 μg intraperitoneally (i.p.) followed by doses of 200 μg i.p. bi-weekly for 8 weeks. Mice for Extended Data Fig. 3h were treated i.p. with anti-PD1 (200 μg, Bioxcell, RMP1-14) or IgG (200 μg, Bioxcell, LTF-2). The treatment regimen for Extended Data Fig. 3i was as described elserwhere38.
Intraperitoneal glucose tolerance test and measurement of serum parameters were as described previously17.
Magnetic resonance Imaging
MRI was done in the small animal imaging core facility in DKFZ using a Bruker BioSpec 9.4 Tesla (Ettlingen). Mice were anaesthetized with 3.5% sevoflurane, and imaged with T2-weighted imaging using a T2_TurboRARE sequence: TE = 22 ms, TR = 2,200 ms, field of view (FOV) 35 × 35 mm, slice thickness 1 mm, averages = 6, scan time 3 min 18 s, echo spacing 11 ms, rare factor 8, slices 20, image size 192 × 192 pixels, resolution 0.182 × 0.182 mm.
Multiplex ELISA
Liver homogenates were prepared as for western blotting17 and cytokines or chemokines were analysed on a customized ELISA according to the manufacturer’s manual (Meso Scale Discovery, U-PLEX Biomarker group 1, K15069L-1).
Flow cytometry and FACS
Isolation and staining of lymphocytes
After perfusion and mechanical dissection, livers were incubated for up to 35 min at 37 °C with collagen IV (60 U final concentration (f.c.)) and DNase I (25 μg/ml f.c.), filtered at 100 μm, and washed with RPMI1640 (11875093, Thermo Fisher). Next, samples underwent a two-step Percoll gradient (25%/50% Percoll/HBSS) and centrifugation for 15 min at 1,800g and 4 °C. Enriched leukocytes were then collected, washed, and counted. For re-stimulation, cells were incubated for 2 h at 37 °C under 5% CO2 with 1:500 Biolegend´s Cell Activation Cocktail (with brefeldin A) (423304) and 1:1,000 Monensin Solution (420701). Live/dead discrimination was done using DAPI or ZombieDyeNIR according to the manufacturer’s instructions with subsequent staining of titrated antibodies (Supplementary Tables 12–14). Samples for flow cytometric-activated cell sorting (FACS) were sorted and samples for flow cytometry were fixed using eBioscience IC fixation (00-8222-49) or FOXP3 Fix/Perm kit (00-5523-00) according to the manufacturer’s instructions. Intracellular staining was performed in eBioscience Perm buffer (00-8333-56). Cells were analysed using BD FACSFortessa or BD FACSSymphony and data were analysed using FlowJo (v10.6.2). For sorting, FACS Aria II and FACSAria FUSION were used in collaboration with the DKFZ FACS core facility.
For UMAP and FlowSOM plots, BD FACSymphony data (mouse and human) were exported from FlowJo (v10). Analyses were performed as described elsewhere39.
Single-cell RNA-seq and metacell analysis (mouse)
Single-cell capturing for scRNA-seq and library preparation were done as described previously40. Libraries (pooled at equimolar concentration) were sequenced on an Illumina NextSeq 500 at a median sequencing depth of ~40,000 reads per cell. Sequences were mapped to the mouse genome (mm10), using HISAT (version 0.1.6); reads with multiple mapping positions were excluded. Reads were associated with genes if they were mapped to an exon, using the Ensembl gene annotation database (Ensembl release 90). Exons of different genes that shared a genomic position on the same strand were considered to represent a single gene with a concatenated gene symbol. The level of spurious unique molecular identifiers (UMIs) in the data was estimated by using statistics on empty MARS-seq wells and excluded rare cases with estimated noise >5% (median estimated noise overall for experiments was 2%). Specific mitochondrial genes, immunoglobulin genes, genes linked with poorly supported transcriptional models (annotated with the prefix “Rp-”), and cells with fewer than 400 UMIs were removed. Gene features were selected using Tvm = 0.3 and a minimum total UMI count >50. We carried out hierarchical clustering of the correlation matrix between those genes (filtering genes with low coverage and computing correlation using a down-sampled UMI matrix) and selected the gene clusters that contained anchor genes. We used K = 50, 750 bootstrap iterations, and otherwise standard parameters. Subsets of T cells were obtained by hierarchical clustering of the confusion matrix and supervised analysis of enriched genes in homogeneous groups of metacells41.
Velocity and correlation analyses of scRNA-seq data
Velocyto (0.6) was used to estimate the spliced and unspliced counts from the pre-aligned bam files42. RNA velocity, latent time, root, and terminal states were calculated using the dynamical velocity model from scvelo (0.2.2)43. Kendall’s rank correlation coefficient (τ) was used to correlate the expression patterns of biologically significant genes with latent time.
Preparation for mass spectrometry, data acquisition, and data analysis
After FACS purification, cells were resuspended in 50% (vol/vol) 2,2,2-trifluoroethanol in PBS pH 7.4 buffer and lysed by repeated sonication and freeze–thaw cycles. Proteins were denatured at 60 °C for 2 h, reduced using dithiothreitol at a final concentration of 5 mM (30 min at 60 °C), cooled to room temperature, alkylated using iodoacetamide at 25 mM (30 min at room temperature in the dark), and diluted 1:5 using 100 mM ammonium bicarbonate, pH 8.0. Proteins were digested overnight by trypsin (1:100 ratio, 37 °C), desalted using C18-based stage-tips, dried under vacuum, resuspended in 20 μl HPLC-grade water with 0.1% formic acid, and measured using A380.
We used 0.5 μg of peptides for proteomic analysis on a C18 column using a nano liquid chromatography system (EASY-nLC 1200, Thermo Fisher Scientific). Peptides were eluted using a gradient of 5–30% buffer B (80% acetonitrile and 0.1% formic acid) at a flow rate of 300 nl/min at a column temperature of 55 °C. Data were acquired by data-dependent Top15 acquisition using a high-resolution orbitrap tandem mass spectrometer (QExactive HFX, Thermo Scientific). All MS1 scans were acquired at 60,000 resolution with AGC target of 3 × 106, and MS2 scans were acquired at 15,000 resolution with AGC target of 1 × 105 and maximum injection time of 28 ms. Analyses were performed using MaxQuant (1.6.7.0), mouse UniProt Isoform fasta (Version: 2019-02-21, number of sequences 25,233) as a source for protein sequences. One per cent FDR was used for controlling at the peptide and protein levels, with a minimum of two peptides needed for consideration of analysis. GSEA was performed using ClusterProfiler (3.18)44 and gene sets obtained from WikiPathway (https://www.wikipathways.org/) and MSigDB (https://broadinstitute.org/msigdb)45,46,47.
Histology, immunohistochemistry, scanning, and automated analysis
Histology, immunohistochemistry, scanning, and automated analysis have been described previously17. Antibodies used in this manuscript are described in Supplementary Table 12. For immunofluorescence staining, established antibodies were used, coupled with the AKOYA Biosciences Opal fluorophore kit (Opal 520 FP1487001KT, Opal 540 FP1494001KT, Opal 620 FP1495001KT). For mRNA in situ hybridization, freshly non-baked 5 μm formalin-fixed paraffin-embedded sections were cut and stained according to the manufacturer’s (ACD biotech) protocol for manual assay RNAscope, using probes PDL1 (420501), TNF (311081) and CXCR6 (871991).
Isolation of RNA and library preparation for bulk RNA sequencing
RNA isolation17 and library preparation for bulk 3′-sequencing of poly(A)-RNA was as described previously48. Gencode gene annotations version M18 and the mouse reference genome major release GRCm38 were derived from https://www.gencodegenes.org/. Dropseq tools v1.1249 were used for mapping the raw sequencing data to the reference genome. The resulting UMI-filtered count matrix was imported into R v3.4.4. Before differential expression analysis with Limma v3.40.650 sample-specific weights were estimated and used as coefficients alongside the experimental groups as a covariate during model fitting with Voom. t-test was used for determining differentially (P < 0.05) regulated genes between all possible experimental groups. GSEA was conducted with the pre-ranked GSEA method46 within the MSigDB Reactome, KEGG, and Hallmark databases (https://broadinstitute.org/msigdb). Raw sequencing data are available at European Nucleotide Archive (https://www.ebi.ac.uk/ena/browser/home) under the accession number PRJEB36747.
Stimulation of CD8 T cells
Stimulation of CD8 T cells was as described elsewhere27.
Flow cytometry of human biopsies
Analysis of patient material (Supplementary Table 1) was performed on liver tissue (needle biopsies or resected tissue, BIOFACS Study KEK 2019-00114), which were obtained from the patient collection nAC-2019-3627 (CRB03) from the biological resource centre of CHU Grenoble-Alpes (nBRIF BB-0033-00069). Tissue samples were minced using scalpels, incubated (with 1 mg/ml collagenase IV (Sigma Aldrich), 0.25 μg/ml DNase (Sigma Aldrich), 10% FCS (Thermo Fisher Scientific), RPMI 1640 (Seraglob)) for 30 min at 37 °C, stopping enzymatic reactions with 2 mM EDTA (StemCell Technologies, Inc.) in PBS. After filtering through a 100-μm cell strainer, cells were resuspended in FACS buffer (PBS, EDTA 2 mM, FCS 0.5%) with Human TruStain FcX (Fc Receptor Blocking Solution) (Biolegend), incubated for 15 min at 4 °C and stained with antibodies (Supplementary Table 13).
Flow cytometry of human samples (Extended Data Fig. 9f) was approved by the local ethical committee (AC-2014-2094 n 03).
High-throughput RNA-seq of human samples
As previously reported, RNA-seq analysis was performed using the data from 206 snap-frozen biopsy samples from 206 patients diagnosed with NAFLD in France, Germany, Italy, and the UK and enrolled in the European NAFLD Registry (GEO accession GSE135251)51,52. Samples were scored for NAS by two pathologists53. Alternate diagnoses were excluded, including excessive alcohol intake (30 g per day for males, 20 g for females), viral hepatitis, autoimmune liver diseases, and steatogenic medication use. Patient samples were grouped: NAFL (n = 51) and NASH with fibrosis stages of F0/1 (n = 34), F2 (n = 53), F3 (n = 54) and F4 (n = 14). Collection and use of data of the European NAFLD Registry were approved by the relevant local and/or national Ethical Review Committee51. A correction for sex, batch, and centre effects was implemented. Pathway enrichment and visualization were as described elsewhere52,54,55.
Immunohistochemistry of NAFLD/NASH cohort
Sixty-five human FFPE biopsies from patients with NAFLD were included (Supplementary Table 3). Sequential slides were immunostained with antibodies against human CD8 (Roche, SP57, ready-to-use), PD1 (Roche; NAT105, ready-to-use), and CD4 (Abcam, ab133616, 1:500). All staining was performed on the VENTANA BenchMark autostainer at 37 °C. Immunopositive cells were quantified at 400× magnification in the portal tract and the adherent parenchyma.
Isolation of cells for scRNA-seq data analysis (human)
Analyses used liver samples from patients undergoing bariatric surgery at the Department of Surgery at Heidelberg University Hospital (S-629/2013). Samples were preserved by FFPE for pathological evaluation and single cells were generated by mincing, using the Miltenyi tumour dissociation kit (130-095-929) per the manufacturer’s instructions, filtering through a 70-μm cell strainer and washing. ACK lysis using the respective buffer (Thermo Fischer Scientific A1049201) was performed, and samples were stored in FBS with 20% DMSO until further processing (scRNA-seq analysis and mass cytometry).
Cells were thawed in a 37 °C water bath, washed with PBS + 0.05 mM EDTA (10 min, 300g at 4 °C), Fc receptor-block (10 min at 4 °C), stained with CD45-PE (3 μl, Hl30, 12-0459-42) and Live/Dead discrimination (1:1,000, Thermofischer, L34973), washed and sorted on a FACSAria FUSION in collaboration with the DKFZ FACS. Library generation was performed according to the manufacturer’s protocol (Chromium Next EM Single Cell 3′GEM, 10000128), and sequencing was performed on an Illumina NovaSeq 6000. De-multiplexing and barcode processing were performed using the Cell Ranger Software Suite (Version 4.0.0) and reads were aligned to human GRCh3856. A gene–barcode matrix containing cell barcodes and gene expression counts was generated by counting the single-cell 3′ UMIs, which were imported into R (v4.0.2), where quality control and normalization were executed using Seurat v357. Cells with more than 10% mitochondrial genes, fewer than 200 genes per cell, or more than 6,000 genes per cell were excluded. Matrices from 10 samples were integrated with Seurat v3 to remove batch effects across samples. PCA analysis of filtered gene–barcode matrices of all CD3+ cells, visualized by UMAP (top 50 principal components), and identification of major cell types using the highly variable features and indicative markers were performed. Pairwise comparisons of CD4+ T cells versus CD4+PD1+ T cells and CD8+ T cells versus CD8+PD1+ T cells were performed using the results of differential expression analysis by DESeq2 (v1.28.1)58, setting CD4+/CD8+ T cells as controls. Volcano plots were then generated using EnhancedVolcano (v1.6.0)59 to visualize the results of differential expression analysis.
Mass cytometry data analysis (human)
Antibody conjugates for mass cytometry were purchased from Fluidigm, generated in-house using antibody labelling kits (Fluidigm X8, MCP9), or as described before60,61. Antibody cocktails for mass cytometry were cryopreserved as described before62. Isolation of cells is described in ‘Isolation of cells for scRNA-seq data analysis (human)’. Cells were thawed, transferred into RPMI + benzonase (14 ml RPMI + 0.5 μl benzonase), and centrifuged for 5 min at 500g. The cell pellet was resuspended in 1 ml CSM-B (CSM (PBS 0.5% BSA 0.02% sodium azide) +1 μl benzonase), filtered through a 30-μm cell strainer, adjusted to 3 ml, counted, resuspended in 35 μl CSM-B and incubated for 45 min at 4 °C, and 100 μl CSM-B was added. Cells were pooled and stained with a surface antibody cocktail (Supplementary Table 15) for 30 min at 4 °C. Dead cell discrimination was performed with mDOTA-103Rh (5 min, room temperature). For intracellular staining, the FOXP3 intracellular staining kit from Miltenyi Biotec was used per the manufacturer’s instructions, followed by staining for intracellular targets for 30 min at room temperature. Cells were washed, resuspended in 1 ml of iridium intercalator solution, and incubated for 25 min at room temperature. Cells were washed with CSM, PBS, and MilliQ water, adjusted to a final concentration of 7.5 × 105 cells/ml and supplemented with 4-element EQ beads. The sample was acquired on a Helios mass cytometer and raw data were EQ-Bead-normalized using Helios mass cytometer and Helios instrument software (version 6.7). Compensation was performed in CATALYST (v1.86)63 and FlowCore (1.50.0). De-barcoding and gating of single, live CD45+ cells were performed using FlowJo (v10.6.2). Then, data from CD45+ cells were imported into Cytosplore 2.3.1 and transformed using the arcsinh(5) function. Major immune cell lineages were identified at the first level of a two-level hierarchical stochastic neighbour embedding (HSNE) analysis with default perplexity and iteration settings. HSNE with the same parameters was run on CD3+ cells to identify T cell phenotypes. Gaussian mean shift clustering was performed in Cytosplore and a heat map of arcsinh(5)-transformed expression values of all antibody targets was generated. Cell type identification was based on the transformed expression values and clusters showing high similarity were merged manually.
Histological and immunohistochemical analysis of NASH–HCC cohort
Four healthy samples, 16 samples from patients with NASH cases, and non-tumoral tissue adjacent to HCC tumours from patients of the following aetiologies were selected: NASH (n = 26), viral hepatitis (n = 19 HCV, n = 3 HBV), alcohol (n = 5), and other (n = 2). All samples were obtained from International Genomic HCC Consortium with IRB approval. After heat-induced antigen retrieval (10 mM sodium citrate buffer (pH 6.0) or Universal HIER antigen retrieval reagent (ab208572) for 15 min (3 × 5 min), the reaction was quenched using 3% hydrogen peroxide. Samples were washed with PBS and incubated with anti-CD8 (Cell Signaling, Danvers, MA) or anti-PD1 (NAT105, ab52587). DAB (3,3′-diaminobenzidine) was used as a detection system (EnVision+ System-HRP, Dako). PD1-positive cases were defined by considering median positivity by immunohistochemistry64 and using a cutoff of ≥1% of PD1-positive lymphocytes among all lymphocytes present on each slide. Analysis of human samples from the Department of Pathology and Molecular Pathology, University Hospital Zurich (Extended Data Fig. 10), was approved by the local ethics committee (Kantonale Ethikkommission Zürich, KEK-ZH-Nr. 2013-0382 and BASEC-Nr. PB_2018-00252).
Search strategy, selection criteria, and meta-analysis of phase III clinical trials
The literature search was done through MEDLINE on PubMed, Cochrane Library, Web of Science, and clinicaltrials.gov, using the following searches: ‘checkpoint inhibitors’, ‘HCC’, ‘phase III’, between January 2010 and January 2020, and complemented by manual searches of conference abstracts and presentations. Single-centre, non-controlled trials, studies with insufficient data to extract HRs or 95% confidence intervals, and trials including disease entities other than HCC were excluded. As conference abstracts were not excluded, quality assessment of the included studies was not performed. Three studies5,10,11 fulfilled the criteria and were included in the quantitative synthesis (Extended Data Fig. 10). The primary outcome of the meta-analysis was overall survival, defined as the time from randomization to death. HRs and CIs related to overall survival were extracted from the papers or conference presentations5,10,11. Pooled HRs were calculated using the random-effects model and we used the DerSimonian–Laird method to estimate τ2, and the generic inverse variance was used for calculating weights 65. To evaluate heterogeneity among studies, Cochran’s Q test and I2 index were used. P < 0.10 in the Q-test was considered to indicate substantial heterogeneity. I2 was interpreted as suggested in the literature: 0% to 40% might not represent significant heterogeneity; 30% to 60% may represent moderate heterogeneity; 50% to 90% may represent substantial heterogeneity; 75% to 100% represents considerable heterogeneity. All statistical pooled analyses were performed using RevMan 5.3 software.
A cohort of patients with HCC treated with PD(L)1-targeted immunotherapy
The retrospective analysis was approved by local Ethics Committees. Data from this cohort were published previously66. Patients with liver cirrhosis and advanced-stage HCC treated with PD(L)1-targeted immune checkpoint blockers from 12 centres in Austria, Germany, Italy, and Switzerland were included. The χ2 test or Fisher’s exact test were used to compare nominal data. Overall survival was defined as the time from the start of checkpoint inhibitor treatment until death. Patients who were still alive were censored at the date of the last contact. Survival curves were calculated by the Kaplan–Meier method and compared by using the log-rank test. Multivariable analysis was performed by a Cox regression model. Statistical analyses were performed using IBM SPSS Statistics version 25 (SPSS Inc., Chicago, IL).
A validation cohort of patients with HCC treated with PD1-targeted immune checkpoint blockers
A multi-institutional dataset that included 427 patients with HCC treated with immune checkpoint inhibitors between 2017 and 2019 in 11 tertiary-care referral centres specialized in the treatment of HCC was analysed. Clinical outcomes of this patient cohort have been reported elsewhere67,68. Inclusion criteria were: 1) diagnosis of HCC made by histopathology or imaging criteria according to American Association for the Study of Liver Disease and European Association for the Study of the Liver guidelines; 2) systemic therapy with immune checkpoint inhibitors for HCC that was not amenable to curative or loco-regional therapy following local multidisciplinary tumour board review; 3) measurable disease according to RECIST v1.1 criteria at commencement of treatment with immune checkpoint inhibitors. One hundred and eighteen patients with advanced-stage HCC were recruited with Child–Pugh A liver functional reserve, and documented radiologic or clinical diagnosis of cirrhosis. Ethical approval to conduct this study was granted by the Imperial College Tissue Bank (reference number R16008).
Statistical analyses
No statistical methods were used to predetermine sample size. The experiments were not randomized and the investigators were not blinded to allocation during experiments and outcome assessment. Data were collected in Microsoft Excel. Mouse data are presented as the mean ± s.e.m. Pilot experiments and previously published results were used to estimate the sample size, such that appropriate statistical tests could yield significant results. Statistical analysis was performed using GraphPad Prism software version 7.03 (GraphPad Software). Exact P values lower than P < 0.1 are reported and specific tests are indicated in the legends.
Sample sizes, biological replicates and statistical tests
Fig. 1a: PD1, n = 5 mice/group; CD8, ND n = 6 mice; CD-HFD n = 6 mice; WD-HTF n = 5 mice. Scale bar, 100 μm. Fig. 1b: n = 3 mice/group. Scale bar, 100 μm. Fig. 1c: ND n = 4 mice, CD-HFD n = 6 mice. Fig. 1d, e: n = 3 mice/group. Fig. 1f: tumour incidence: CD-HFD, n = 19 tumours/lesions in 25 mice; CD-HFD + anti-PD1, n = 10 tumours/lesions in 10 mice. Fig. 1h: ND, n = 3 mice; CD-HFD, n = 13 mice; CD-HFD + anti-PD1, n = 8 mice; intra-tumoral staining: CD-HFD, n = 11 mice; CD-HFD + anti-PD1, n = 8 mice. Data in Fig. 1a, h were analysed by two-tailed Student’s t-test. Data in Fig. 1f were analysed by two-sided Fisher’s exact test.
Fig. 2a, b: n = 3 mice/group. Fig. 2c: CD8+: ND, n = 6 mice; CD-HFD + IgG, n = 5 mice; CD-HFD + anti-PD1, n = 6 mice; CD8+PD1+: ND, n = 4 mice, CD-HFD + IgG, n = 6 mice; CD-HFD + anti-PD1, n = 6 mice. Fig. 2d, e: ND, n = 4 mice; CD-HFD + IgG, n = 8 mice; CD-HFD + anti-PD1, n = 6 mice. Fig. 2f: CD-HFD + IgG, n = 6 mice; CD-HFD + anti-PD1, n = 4 mice. Fig. 2g: ND, n = 30 mice; CD-HFD, n = 47 mice; CD-HFD + anti-PD1, n = 35 mice; CD-HFD + anti-PD1/anti-CD8, n = 9 mice; CD-HFD + anti-TNF, n = 10 mice; CD-HFD + anti-PD1/anti-TNF, n = 11 mice; CD-HFD + anti-CD4, n = 8 mice; CD-HFD + anti-PD1/anti-CD4, n = 8 mice. Fig. 2h: CD8+PD1+CXCR6+: ND, n = 30 mice; CD-HFD, n = 47 mice; CD-HFD + anti-PD1, n = 35 mice; CD-HFD + anti-PD1/anti-CD8, n = 9 mice; CD-HFD + anti-TNF, n = 10 mice; CD-HFD + anti-PD1/anti-TNF, n = 11 mice; CD-HFD + anti-CD4, n = 8 mice; CD-HFD + anti-PD1/anti-CD4, n = 8 mice. Fig. 2j: tumour incidence: CD-HFD, n = 32 tumours/lesions in 87 mice; CD-HFD + anti-CD8, n = 2 tumours/lesions in 31 mice; CD-HFD + anti-CD8/NK1.1, n = 0 tumours/lesions in 6 mice; CD-HFD + anti-PD1, n = 33 tumours/lesions in 44 mice; CD-HFD + anti-PD1/anti-CD8, n = 2 tumours/lesions in 9 mice; CD-HFD + anti-TNF, n = 3 tumours/lesions in 10 mice; CD-HFD + anti-PD1/anti-TNF, n = 3 tumours/lesions in 11 mice; CD-HFD + anti-CD4, n = 3 tumours/lesions in 9 mice; CD-HFD + anti-PD1/anti-CD4, n = 8 tumours/lesions in 9 mice. All data are shown as mean ± s.e.m. Data in Fig. 2e, g, h were analysed by one-way ANOVA and Fisher’s LSD test. Data in Fig. 2f were analysed by two-tailed Mann–Whitney test. Data in Fig. 2j were analysed by two-sided Fisher’s exact test.
Fig. 3a, b: control, n = 6 patients; NAFLD/NASH, n = 11 patients. Fig. 3c: control, n = 4 patients; NAFLD/NASH, n = 7 patients. Fig. 3d–h: mouse, n = 3; human, n = 3. All data are shown as mean ± s.e.m. Data in Fig. 3b, f were analysed by two-tailed Mann–Whitney test. Data in Fig. 3d were analysed by two-tailed Spearman’s correlation.
Fig. 4a: Hazard ratios are represented by squares, the size of the square represents the weight of the trial in the meta-analysis. Cochran’s Q-test and I2 were used to calculate heterogeneity. Fig. 4b: Kaplan–Meier curve displays overall survival of patients with NAFLD versus those with any other aetiology; all 130 patients were included in these survival analyses (NAFLD n = 13; any other aetiology n = 117). Fig. 4c: Kaplan–Meier curve displays overall survival of patients with NAFLD versus those with any other aetiology (NAFLD n = 11; any other aetiology n = 107). Data in Fig. 4b, c were analysed by Kaplan–Meier method and compared using log rank test.
Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this paper.