Kupffer cell programming by maternal obesity triggers fatty liver disease

Abstract

Kupffer cells (KCs) are tissue-resident macrophages that colonize the liver early during embryogenesis¹. Upon liver colonization, KCs rapidly acquire a tissue-specific transcriptional signature, mature alongside the developing liver and adapt to its functions^1-3. Throughout development and adulthood, KCs perform distinct core functions that are essential for liver and organismal homeostasis, including supporting fetal erythropoiesis, postnatal erythrocyte recycling and liver metabolism⁴. However, whether perturbations of macrophage core functions during development contribute to or cause disease at postnatal stages is poorly understood. Here, we utilize a mouse model of maternal obesity to perturb KC functions during gestation. We show that offspring exposed to maternal obesity develop fatty liver disease, driven by aberrant developmental programming of KCs that persists into adulthood. Programmed KCs promote lipid uptake by hepatocytes through apolipoprotein secretion. KC depletion in neonate mice born to obese mothers, followed by replenishment with naive monocytes, rescues fatty liver disease. Furthermore, genetic ablation of the gene encoding hypoxia-inducible factor-α (HIF1α) in macrophages during gestation prevents the metabolic programming of KCs from oxidative phosphorylation to glycolysis, thereby averting the development of fatty liver disease. These results establish developmental perturbation of KC functions as a causal factor in fatty liver disease in adulthood and position fetal-derived macrophages as critical intergenerational messengers within the concept of developmental origins of health and diseases⁵.

Fig. 1. Maternal obesity leads to FLD in the offspring.

a, Generation of the maternal obesity mouse model. b, ORO staining of offspring livers. Scale bars, 200 μm. Representative for n = 11, 7, 7, 9, 7 and 9 mice for CD_MCD_LCD, CD_MCD_LHFD, HFD_MCD_LCD, HFD_MHFD_LCD, HFD_MCD_LHFD and HFD_MHFD_LHFD groups, respectively, on 11 experimental days. c, ORO staining quantification of panel b by QuPath. Violin plots were created from 2–10 images per mouse, and the median and quartile of all images are shown. Each circle or triangle represents the mean of all the images per mouse. One-way analysis of variance (ANOVA) with Tukey’s multiple comparison test, comparing the mean of all biological samples and showing significances of only CD_MCD_LCD, CD_MCD_LHFD and HFD_MCD_LCD groups, was used. d, Liver lipidomics of all experimental groups visualized by PCA. n = 3, 6, 4, 6, 4 and 5 mice for CD_MCD_LCD, CD_MCD_LHFD, HFD_MCD_LCD, HFD_MHFD_LCD, HFD_MCD_LHFD and HFD_MHFD_LHFD groups, respectively. e, Bulk RNA-seq analysis of sorted KCs visualized by PCA. Each circle or triangle represents one mouse. n = 2, 6, 3, 5, 4 and 2 mice for CD_MCD_LCD, CD_MCD_LHFD, HFD_MCD_LCD, HFD_MHFD_LCD, HFD_MCD_LHFD and HFD_MHFD_LHFD groups, respectively. f, Horizontal co-expression network analysis of KC RNA-seq data with the 5,000 most variable genes across all samples (left). The network was clustered, and group fold change (GFC) across all conditions per identified cluster is depicted. The selected enriched pathways of each horizontal co-expression network analysis module by over-representation analysis are also shown (right). n = 2–6 mice per group. g, Heatmap of selected genes of the intersection of DEGs and the respective annotated gene set from panel f. OxPhos, oxidative phosphorylation. Schematic in a created in BioRender. Mass, E. (2025) https://BioRender.com/3ecv7rv. Source data

Fig. 2. KCs retain yolk sac origin after maternal obesity and cause lipid droplet accumulation via paracrine signalling.

a, Breeding scheme to generate Tnfrsf11a^Cre;Rosa26^LSL-YFP;Ms4a3^FlpO;Rosa26^FSF-tdTomato double fate mapper mice. b, Schematic of lineage-tracing strategy of macrophages using the double fate mapper model. EMP, erythromyeloid progenitor; GMP, granulocyte–monocyte progenitor; pMac, pre-macrophage. Green indicates cells expressing YFP, red indicates tdT, and yellow indicates both YFP and tdT. c, Labelling efficiency of the double fate mapper model in KCs isolated from 11–12-week-old CD_MCD_LCD (n = 4 on 2 experimental days) and HFD_MCD_LCD (n = 3 on 2 experimental days) mice. The bar graphs show the mean ± s.d. d, Schematic of hepatocyte culture isolated from chow diet-fed mice with KC addition isolated from CD_MCD_LCD or HFD_MCD_LCD mice, respectively, which were recorded for lipid accumulation for a 4-h time course. e, Lipid accumulation shown by normalized LD540 intensity through live imaging in ex vivo-cultured hepatocytes co-cultured for 4 h with KCs from CD_MCD_LCD or HFD_MCD_LCD mice. n = 3 for CD_MCD_LCD or HFD_MCD_LCD mice with technical duplicates. Two-way ANOVA with Sidak’s multiple comparison test on biological samples was used. f, Generation of KC depletion and cell transfer maternal obesity model. Diphtheria toxin (DT) indicated by the skull symbol. Monocyte and haematopoietic stem and progenitor cells (Mono/HSPCs) indicated by the red cell. g, Liver ORO staining in HFD_MCD_LCD offspring generated as shown in panel f. n = 4, 5 and 9 mice for Clec4f^+/+ with DT and Mono/HSPCs, Clec4f^DTR/+ with DT, Clec4f^DTR/+ with DT and Mono/HSPCs, respectively, on 7 experimental days. Scale bars, 200 µm. h, ORO staining quantification of panel g by QuPath. Violin plots were created from n = 10 images per mouse, and the median and quartile of all images are shown. Each triangle represents the mean of all images per mouse. A two-tailed Mann–Whitney test was used, comparing the mean of all biological samples. The illustrations of the mice in panel a, the schematics in panels b,d, the model in panel f, and the illustrations of the skull and red cell in panels g,h were created in BioRender. Mass, E. (2025) https://BioRender.com/5hrtyf4; https://BioRender.com/hh2dq2m; https://BioRender.com/hgceyko; https://BioRender.com/1don7ps. Source data

Fig. 3. Postnatal FLD is driven by HIF1α-dependent developmental programming of KCs.

a, Generation of the maternal obese model with myeloid Hif1a ablation. b, Representative liver ORO staining of n = 8, 5, 8, 6, 9, 7, 6 and 9 mice for CD_MCD WT, CD_MCD KO, CD_MHFD WT, CD_MHFD KO, HFD_MCD WT, HFD_MCD KO, HFD_MHFD WT and HFD_MHFD KO groups, respectively, on 15 experimental days. Scale bars, 200 μm. c, ORO staining quantification of panel b by QuPath. Violin plots were from n = 1–10 images per mouse, and the median and quartile of all images are shown. Each circle or triangle represents the mean of all images per mouse. One-way ANOVA with Tukey’s multiple comparison test was used. d, Liver lipidomics visualized by PCA. n = 7, 5, 5, 5, 4, 5, 3 and 7 mice for CD_MCD WT, CD_MCD KO, CD_MHFD WT, CD_MHFD KO, HFD_MCD WT, HFD_MCD KO, HFD_MHFD WT and HFD_MHFD KO groups, respectively. e, Bulk RNA-seq analysis of sorted KCs visualized by PCA. n = 5, 4, 4, 4, 3, 2, 3 and 4 mice for CD_MCD WT, CD_MCD KO, CD_MHFD WT, CD_MHFD KO, HFD_MCD WT, HFD_MCD KO, HFD_MHFD WT and HFD_MHFD KO groups, respectively. f, Chord diagram showing KC ligands (in orange; DEGs between HFD_MCD WT and HFD_MCD KO in panel e) and their respective receptors (in brown) expressed by hepatocytes. LFC, log fold change. g, Venn diagram showing differentially expressed proteins from KCs comparing HFD_MCD WT versus CD_MCD WT (light blue), HFD_MCD WT versus HFD_MCD KO (light orange) mice, and overlapping proteins of the two comparisons (‘Cross’). n = 3, 3, 4 and 3 mice for CD_MCD WT, CD_MCD KO, HFD_MCD WT and HFD_MCD KO groups, respectively, on 2 experimental days. h, Heatmap showing selected proteins from panel g. ^#HFD_MCD WT versus CD_MCD WT, ^$HFD_MCD WT versus HFD_MCD KO and *Cross. ANOVA multiple-sample testing (S0 = 0.5, false discovery rate < 0.1 with 250 randomizations) was used. Schematic in a was created in BioRender. Mass, E. (2025) https://BioRender.com/lcxdcmf. Source data

Fig. 4. Maternal obesity induces epigenetic and transcriptional changes in the KCs of the offspring.

a, snRNA/ATAC-seq of CD_MCD_LCD and HFD_MCD_LCD mice visualized by uniform manifold approximation and projection (UMAP) showing clusters depicting five different KC states. b, Dotplot showing genes specifically expressed by each KC state. c, Heatmap showing predicted transcription factor activity in each cluster of both CD_MCD_LCD and HFD_MCD_LCD groups. d, Violin plots comparing Apoe (left) and Apoa1 (right) expression in all KC states between CD_MCD_LCD and HFD_MCD_LCD groups. Two-tailed Wilcoxon rank-sum test was used. e, ATAC-seq peak coverage plot of the Apoe locus comparing CD_MCD_LCD and HFD_MCD_LCD groups for each KC state. A possible differentially exploited regulatory region controlling Apoe expression is highlighted in red. Transcription factor-binding sites of PPARγ–RXRα are highlighted with a blue vertical line and of PPARα with green vertical lines. f, Lipid accumulation shown by normalized LD540 intensity through live imaging in ex vivo-cultured hepatocytes supplemented without any factor (control), with TNF, APOE or APOA1. n = 4 mice per group with technical triplicates. Mean ± s.d. of biological samples is shown. One-tailed ratio paired Student’s t-test on biological samples comparing treated groups to control was used. Source data

Extended Data Fig. 1. Maternal obesity model and characterization of liver metabolic status in the offspring.

a, Body weight of mothers on control diet (CD, grey) or high-fat diet (HFD, green). n = 15 mice/group. Mean ± SD is shown. Two-tailed unpaired Student’s t-test. b, Homeostatic Model Assessment for Insulin Resistance (HOMA-IR) of mothers on CD or HFD performed after 8 weeks on respective diet. n = 15 mice/group. Mean is indicated. Two-tailed unpaired Student’s t-test. c, Heatmap showing log-transformed values of detected cytokines and chemokines in the sera of mothers on CD and HFD. Each column represents one mouse. n = 7 mice/group. Two-tailed unpaired Student’s t-test on the individual coefficients after linear fitting model and multiple testing correction using false discovery rate <0.05. d, Plots of results from c that show significance when using two-tailed unpaired Student’s t-test. Mean ± SD. e, Body, white adipose tissue (WAT) and liver weights of offspring at 11 weeks of age. n = 8, 8, 7, 6, 7, 7 mice for CD_MCD_LCD, CD_MCD_LHFD, HFD_MCD_LCD, HFD_MHFD_LCD, HFD_MCD_LHFD, HFD_MHFD_LHFD groups, respectively, on 11 experimental days. Bar graphs show the means. f, Hematoxilin-eosin (HE) staining of liver sections of the offspring. Scale bars, 100 μm. Representative pictures for the mice in e. g, Heatmap of liver lipidomics in Fig. 1d showing the condition-wise Log-Fold changes (LFC) against the CD_MCD_LCD of selected lipid groups. Two-tailed unpaired Student’s t-test on the individual coefficients after linear fitting model and multiple testing correction using false discovery rate <0.05 (* = p < 0.05, ** = p < 0.01, *** = p < 0.001). h, Metabolomics of serum from all experimental groups visualized by PCA. i, Enriched metabolic pathways in HFD_MCD_LCD compared to CD_MCD_LCD based on serum metabolomics data. h, i, n = 7 mice/group. One-way ANOVA – Tukey’s multiple comparison test. Source data

Extended Data Fig. 2. Flow cytometry analysis, FACS gating strategy and metabolic profiling of KCs and liver myeloid cells.

a, Flow cytometry gating strategy for KCs, Liver Capsular Macrophages (LCM), classical Dendritic Cells 1 and 2 (cDC1, cDC2), monocytes (Mono) and neutrophils (Neu). b, Cell number of different myeloid populations per gram of liver tissue quantified by flow cytometry. n = 4, 5, 4, 6, 4, 4 mice for CD_MCD_LCD, CD_MCD_LHFD, HFD_MCD_LCD, HFD_MHFD_LCD, HFD_MCD_LHFD, HFD_MHFD_LHFD groups, respectively, on 8 experimental days. Bar graphs show the means. One-way ANOVA – Tukey’s multiple comparison test, comparing and showing significances of only CD_MCD_LCD, CD_MCD_LHFD and HFD_MCD_LCD groups. c, Fluorescence activated cell sorting (FACS) gating strategy for sorting KCs for RNA-seq. d, Schematic illustration of metabolite transporters and enzymes assessed for the metabolic state of KCs. 1. GLUT1, glucose transporter 1; 2. PKM, pyruvate kinase M; 3. SDHA, succinate dehydrogenase A; 4. Cytc, cytochrome c; 5. ATP5A, ATP synthase F1 subunit alpha; 6. G6PD, glucose-6-phosphate dehydrogenase; 7. CD36, fatty acid translocase; 8. CPT1A, carnitine palmitoyl transferase 1A; 9. ACC1, acetyl-CoA Carboxylase 1. The schematic was created in BioRender. Mass, E. (2025) https://BioRender.com/axc6avr. e, Mean Fluorescence Intensity (MFI) of GLUT1, PKM, SDHA, CytC, ATP5A, G6PD, CD36, CPT1a and ACC1 in KCs isolated from 11–13 weeks CD_MCD_LCD and HFD_MCD_LCD mice. n = 5 mice/group on 1 experimental day. Bar graphs show the means. Two-tailed unpaired Student’s t-test. Source data

Extended Data Fig. 3. Characterization of serum and bone marrow immune status in the offspring.

a, Heatmap showing log-transformed values of detected cytokines and chemokines in the offspring sera. Each column represents one mouse. n = 5, 7, 6, 6, 5, 6 mice for CD_MCD_LCD, CD_MCD_LHFD, HFD_MCD_LCD, HFD_MHFD_LCD, HFD_MCD_LHFD, HFD_MHFD_LHFD groups, respectively, on 11 experimental days. No significance across conditions using Kruskal-Wallis test. b, Flow cytometry gating strategy for analysis of hematopoietic stem and progenitor cells in the bone marrow. LT-HSC, long-term hematopoietic stem cells; ST-HSC, short-term hematopoietic stem cells; GMP, granulocyte-macrophage progenitors; MEP, megakaryocyte-erythroid progenitors; CMP, common myeloid progenitors; CLP, common lymphoid progenitors. c, Cell number of different hematopoietic stem and progenitor cells quantified by flow cytometry. n = 6, 7, 3, 6, 4, 4 mice for CD_MCD_LCD, CD_MCD_LHFD, HFD_MCD_LCD, HFD_MHFD_LCD, HFD_MCD_LHFD, HFD_MHFD_LHFD groups, respectively, on 11 experimental days. Bar graphs show the means. No significance across conditions using One-way ANOVA – Tukey’s multiple comparison test. Source data

Extended Data Fig. 4. Analysis of Tnfrsf11a^Cre; Rosa26^LSL-YFP; Ms4a3^FlpO; Rosa26^FSF-tdTomatodouble fate mapper mouse model.

a, Ms4a3 locus indicating targeting strategy. b, Flow cytometry gating strategy for blood monocytes and microglia. c, Quantification of labelling efficiency of blood monocytes and microglia in 11–12 weeks old maternal lean (CD_MCD_LCD, n = 4 on 2 experimental days) and maternal obese (HFD_MCD_LCD, n = 3 on 2 experimental days) mice. Bar graphs show the mean ± SD. d, Flow cytometry gating strategy for KCs. Source data

Extended Data Fig. 5. KC depletion in Clec4f^DTR/+ mouse model.

a, Representative flow cytometry gating of F4/80⁺Tim4⁺ KCs from Clec4f^+/+ and Clec4f^DTR/+ mice 14 h, 24 h and 48 h after diphtheria toxin (DT) injection at P0. b, Quantification of F4/80⁺Tim4⁺ KCs (gating shown in a) in % of CD11b⁺ cells comparing Clec4f^+/+ and Clec4f^DTR/+ mice 14 h (n = 3 and 2 mice for Clec4f^+/+ and Clec4f^DTR/+, respectively), 24 h (n = 5 and 4 mice for Clec4f^+/+ and Clec4f^DTR/+, respectively) and 48 h (n = 3 and 5 mice for Clec4f^+/+ and Clec4f^DTR/+, respectively) after DT injection at P0. Bar graphs show the mean ± SD. Two-tailed unpaired Student’s t-test when sample size is no less than 3. c, Flow cytometry analysis of isolated bone marrow monocytes and hematopoietic stem and progenitor cells (HSPCs) from Rosa26^mTmG mice used for transplanting Clec4f^+/+ and Clec4f^DTR/+ mice 15 h after DT injection. d, Immunofluorescence staining of HFD_MCD_LCD Clec4f^DTR/+ liver at the age of 13 weeks showing the partial replacement of endogenous KCs by tdT⁺ bone marrow monocytes and HSPCs from Rosa26^mTmG mice. KCs are shown as Iba1⁺ and nuclei as DAPI⁺. Scale bar: 50 μm. Representative image of n = 9 mice on 4 experimental days. e, Representative H&E staining of livers in Fig. 2f–h. Scale bar: 200 μm. The illustrations were created in BioRender. Mass, E. (2025) https://BioRender.com/1don7ps. Source data

Extended Data Fig. 6. Characterization of Hif1α-WT and Hif1α-KO maternal lean and obese offspring.

a, Body, WAT and liver weight of 11-week-old offspring. Each circle/triangle represents one mouse. n = 12, 8, 7, 6, 8, 7, 7, 10 mice for CD_MCD WT, CD_MCD KO, CD_MHFD WT, CD_MHFD KO, HFD_MCD WT, HFD_MCD KO, HFD_MHFD WT, HFD_MHFD KO groups, respectively, on 15 experimental days. Bar graphs show the means. One-way ANOVA – Tukey’s multiple comparison test. b, Heatmap of liver lipidomics in Fig. 3d showing the condition-wise Log-Fold changes (LFC) against the CD_MCD of selected lipid groups. Two-tailed unpaired Student’s t-test on the individual coefficients after linear fitting model and multiple testing correction using false discovery rate <0.05 (* = p < 0.05, ** = p < 0.01, *** = p < 0.001). c, Pathological scoring of liver steatosis performed on HE stainings. n = 7, 7, 7, 8, 7, 8, 7, 11 mice for CD_MCD WT, CD_MCD KO, CD_MHFD WT, CD_MHFD KO, HFD_MCD WT, HFD_MCD KO, HFD_MHFD WT, HFD_MHFD KO groups, respectively, on 15 experimental days. Bar graphs show the means. One-way ANOVA – Tukey’s multiple comparison test. d, FACS-gating strategy to sort KCs and hepatocytes for RNA-seq. e, RNA-seq of FACS-sorted hepatocytes from all diet groups of Hif1α-WT and Hif1α-KO mice visualized by PCA. n = 5, 5, 5, 5, 2, 3, 6, 2 mice for CD_MCD WT, CD_MCD KO, CD_MHFD WT, CD_MHFD KO, HFD_MCD WT, HFD_MCD KO, HFD_MHFD WT, HFD_MHFD KO groups, respectively. Source data

Extended Data Fig. 7. Transcriptome analysis of KCs isolated from P0 pups.

a, Heatmap showing scaled expression of DEGs of KCs isolated from pups born to CD-fed (CD_M) or HFD-fed (HFD_M) mothers. n = 5 mice/group. b, Transcription factor (TF) activity analysis with genes from a. as input. The higher the score is, the higher the predicted activity is of the indicated TF in KCs born to obese mothers. c. Volcano plot of DEGs (blue: down, red: up) between P0 KCs born to HFD- and CD-fed mothers. HIF1α targets are highlighted in red. d. HALLMARK pathway overrepresentation-analysis of all upregulated DEGs shown in a. e, Variant supporting counts (vsc) of all Hif genes. n = 5 mice/group. Box plots show median with interquartile range and min/max. f, Representative images of immunofluorescence staining of HIF1α in F4/80⁺ KCs in P0 livers born to CD- and HFD-fed mothers. n = 4, 3 P0 pups born to CD-, HFD-fed mothers, respectively, on 3 experimental days. g, Quantification of HIF1α intensity in nuclei in % to cytoplasm in F4/80⁺ KCs comparing P0 pups born to CD- and HFD-fed mothers in f. n = 9, 12 cells from P0 pups born to CD-, HFD-fed mothers, respectively. Bar graphs show the mean ± SD. Two-tailed Mann-Whitney test. Source data

Extended Data Fig. 8. Characterization of myeloid cell immune infiltration and ex vivo KC proteomics in Hif1α-WT and Hif1α-KO livers.

a, Expression of selected chemo-attractants expressed by hepatocytes across diet groups of HIF1α mouse model. n = 5, 5, 5, 5, 2, 3, 6, 2 mice for CD_MCD WT, CD_MCD KO, CD_MHFD WT, CD_MHFD KO, HFD_MCD WT, HFD_MCD KO, HFD_MHFD WT, HFD_MHFD KO groups, respectively. Box plots show medians with interquartile range and min/max. b, Cell number of different myeloid populations per gram of liver tissue quantified by flow cytometry. n = 5, 3, 4, 5, 4, 5, 5, 4 mice for CD_MCD WT, CD_MCD KO, CD_MHFD WT, CD_MHFD KO, HFD_MCD WT, HFD_MCD KO, HFD_MHFD WT, HFD_MHFD KO groups, respectively, on 15 experimental days. Bar graphs show the means. One-way ANOVA – Tukey’s multiple comparison test. c, Dot plot showing ORA analysis of DEGs comparing HFD_MCD WT versus HFD_MCD KO conditions. d, Heatmap showing the relative mean expression and hierarchical clustering of differentially expressed proteins from isolated KCs comparing HFD_MCD WT vs. CD_MCD WT mice, HFD_MCD WT vs. HFD_MCD KO mice, and the overlapping proteins of the two comparisons. Same experiment as Fig. 3h, where selected proteins from the different clusters are displayed. e, Dot plot showing enriched pathways of the differentially expressed proteins from Fig. 3g. Source data

Extended Data Fig. 9. Single nucleus (sn)RNA/ATAC-seq analysis of CD_MCD_LCD and HFD_MCD_LCD KCs.

a, snRNA/ATAC-seq of CD_MCD_LCD and HFD_MCD_LCD livers visualized by Uniform Manifold Approximation (UMAP) showing all liver cell types. b, Dot plot showing the genes determining the cell type annotation of identified clusters shown in a. c, Representative immunofluorescence (IF) staining showing MHC-II expression in the liver of CD_MCD_LCD and HFD_MCD_LCD mice. All KCs and macrophages are labelled with Iba1. scale bar: 200 μm. Representative pictures of n = 11 CD_MCD_LCD and n = 7 HFD_MCD_LCD mice. d, Volcano plot showing assigned genes to the differentially detected snATAC-seq peaks between KCs from CD_MCD_LCD and HFD_MCD_LCD livers. e, GO term analysis of downregulated genes associated with snATAC-seq peaks. f, Violin plots comparing Gpnmb (upper) and Pck1 (lower) expression within all KC states comparing CD_MCD_LCD and HFD_MCD_LCD groups. Two-tailed Wilcoxon Rank Sum test. g, CellChat analysis using differentially expressed genes between CD_MCD_LCD and HFD_MCD_LCD KCs as sender molecules and all cell types as receivers. Coloured genes indicate significant (p < 0.05) differences of information flow between the conditions tested with Two-tailed Wilcoxon Rank Sum test.