Single-cell atlas of human liver and blood immune cells across fatty liver disease stages reveals distinct signatures linked to liver dysfunction and fibrogenesis

Abstract

Immune cells play a central yet poorly understood role in metabolic dysfunction-associated steatotic liver disease and metabolic dysfunction-associated steatohepatitis (MASLD/MASH), a global cause of liver disease with limited treatment. Limited access to human livers and lack of studies across MASLD/MASH stages thwart identification of stage-specific immunological targets. Here we provide a unique single-cell RNA sequencing atlas of paired peripheral blood and liver fine-needle aspirates from a full-spectrum MASLD/MASH human cohort. Our findings included heightened immunoregulatory programs with MASH progression, such as enriched hepatic regulatory T cells, monocytic myeloid-derived suppressor cells, TREM2⁺S100A9⁺ macrophages and S100^hiHLA^lo type 2 conventional dendritic cells. Hepatic cytotoxic T cell functions increased with inflammation, but decreased with fibrosis, while acquiring an exhausted signature, whereas natural killer cell-driven toxicity intensified. Our dataset proposes immunological mechanisms for increased fibrogenesis and vulnerability to liver cancer and infections in MASH and provides a basis for a deeper understanding of human immunological dysfunction in chronic liver disease and a roadmap to new targeted therapies.

Metabolic dysfunction-associated steatotic liver disease and metabolic dysfunction-associated steatohepatitis (MASLD/MASH) is the most common liver disease worldwide¹ and its incidence continues to increase. MASLD is associated with excessive fat in the liver (steatosis), with 20–30% of MASLD cases progressing to MASH, which is characterized by lobular inflammation and hepatocyte ballooning and degeneration. Patients with MASH are more likely to have progressive liver fibrosis, a major predictor of mortality, and 3–6% develop liver cancer². Although several drugs are currently in the pipeline, with resmetirom recently approved for MASH³, most patients do not have evidence of fibrosis improvement and the precise mechanisms driving its progression are incompletely understood.

Immune cells are central to inflammation and fibrogenesis in MASLD/MASH⁴. Increased levels of free fatty acids, oxidative damage, hepatocyte death and altered gut permeability contribute to the activation and hepatic recruitment of local and peripheral immune cells, eventually driving liver dysfunction and fibrosis. Despite the availability of high-throughput, single-cell technologies to assess the multifaceted functions of these immune cells, limited access to human liver samples and lack of studies across all stages of human MASLD/MASH complicate defining stage-specific immunological targets. Furthermore, most single-cell data in the MASLD/MASH space are currently derived from mouse models. Analogous human data are either compiled from available datasets or less well-stratified cohorts, most of which are derived from liver tissue or biopsies that require digestion and flow cytometry-based enrichment of immune cells⁵. Importantly, liver biopsies are becoming less common due to their replacement by noninvasive methods to assess fat accumulation and fibrosis.

Here we used liver fine-needle aspiration (FNA), a minimally invasive approach that provides samples greatly enriched in immune cells, to generate a resource of single-cell transcriptomics of freshly isolated immune cells from paired peripheral blood and liver FNAs from a comprehensive human cohort spanning MASLD/MASH stages. Our analysis tested the relevance of immune cell types for human MASH pathology and found distinct immune signatures linked to liver injury, namely intensified natural killer (NK) cell-driven cytotoxicity and enriched immunoregulatory hepatic regulatory T cells (T_reg cells), S100^hiHLA^lo type 2 conventional dendritic cells (cDC2s), monocytic myeloid-derived suppressor cells (M-MDSCs) and TREM2⁺S100A9⁺ macrophages.

Results

FNAs are amenable to the study of liver immune cells

We studied immune cells from liver FNAs and the corresponding peripheral blood mononuclear cells (PBMCs) of 25 patients spanning the MASLD/MASH spectrum. All patients showed the presence of lipid droplets at ultrasound and were histologically defined as simple steatosis (SS, n = 4), MASH F0 (n = 5), MASH F1 (n = 5), MASH F2 (n = 5), MASH F3 and F4 (n = 4) and no steatosis, inflammation or fibrosis (NS, n = 2) in the absence of meaningful alcohol consumption or viral hepatitis (Fig. 1a). Demographic and clinical parameters, histological staging and scoring, as well as noninvasive measurements of fibrosis, are included (Extended Data Table 1). For analytical comparisons, we grouped participants by MASH status as follows: NS–SS (n = 6), early MASH (MASH F0–F1, n = 10) and advanced MASH (MASH F2–F4, n = 9). Analysis of demographic data between these groups showed no significant differences in age, race and sex, and the groups had comparable body mass indices (Extended Data Table 1). Liver function blood tests for participants with MASH compared with those with NS–SS showed higher liver enzymes (alanine aminotransferase (ALT) and aspartate aminotransferase (AST)), reflecting liver injury, and greater calculated noninvasive fibrosis scores FIB-4 (fibrosis 4 index) and APRI (AST to platelet ratio index), alongside increased liver stiffness assessed by transient elastography (Fibroscan) and histological nonalcoholic fatty liver disease activity score (NAS) (Extended Data Table 1). We assessed patient serum levels of the macrophage activation marker sCD163 and found that they positively correlated with liver enzymes, liver stiffness, APRI and NAS, and were higher in patients with MASH compared with those with NS–SS (Extended Data Fig. 1), consistent with macrophage activation with MASH progression^6–8. These characteristics demonstrated typical biochemical features of MASLD/MASH, indicating that our cohort was representative of the MASLD-affected population.

Fig. 1 |. The landscape of liver and peripheral blood immune cells across the MASLD/MASH spectrum.

a, Graph showing the number of patients (n = 25) spanning MASLD/MASH disease stages, including NS (n = 2, male sex n = 1, median age = 50 (46–54) years), SS (n = 4, male sex n = 1, median age = 47.5 (25–67) years), early MASH (MASH F0, n = 5, male sex n = 2, median age = 60 (56–73) years; MASH F1, n = 5, male sex n = 1, median age = 34 (26–60) years) and advanced MASH (MASH F2, n = 5, male sex n = 2, median age = 53 (31–58) years; MASH F3 and F4, n = 4, male sex n = 1, median age = 61.5 (50–65) years) who provide liver FNA, clinical liver biopsy and peripheral blood. b, Merged Uniform Manifold Approximation and Projection (UMAP) of single cells color coded by cell type (clustering settings: 20 principal components (PCs); Louvain resolution, 0.05) based on analysis of scRNA-seq of 168,634 cells from the liver FNAs and PBMCs as in a. c, Dot-plot of expression of marker genes in each cell type as in b. Shade represents the average scaled expression and size the percentage of cells that express the gene. d, Stacked bar chart showing the proportion of cell types as in b in patients with NS–SS, early MASH and advanced MASH in liver and PBMCs. e, Box plot showing the proportion of CD14⁺ monocytes, CD14⁺CD16⁺ monocytes, CD16⁺ monocytes, cDCs, B cells and macrophages as a percentage of a patient’s liver FNA or PBMC sample (nonparenchymal cells for FNAs). Data are presented as median values (horizontal line), 25th or 75th quartile values (bounds of boxes) and nonoutlier minimum or maximum (whiskers) (NS–SS (n = 6), early MASH (n = 10) and advanced MASH (n = 9); permutation test). *Significance determined by false discovery rate (FDR) <0.05 and log₂(false discovery (FD)) > 0.58.

We surveyed single-cell transcriptomic (single-cell RNA sequencing (scRNA-seq)) data of 168,634 high-quality cells sequenced immediately after isolation from paired liver FNAs (52,961 cells) and PBMCs (115,673 cells). Liver FNAs consisted of cell suspensions, did not require digestion, captured few embedded cells (for example, hepatocytes) and were, therefore, naturally enriched for immune cells. The scRNA-seq data from both FNAs and PBMCs were merged and processed together for comparison (Methods and Extended Data Fig. 2). Major cell types, 17 represented in every patient, were identified (Fig. 1b–d): CD4⁺ T cells (CD3⁺CD4⁺), CD8⁺ T cells (CD3⁺CD8⁺), mucosal-associated invariant T (MAIT) cells (CD3⁺SLC4A10⁺KLRB1^hi), γδ T cells (CD3⁺TRGC1⁺), B cells (MS4A1⁺CD79⁺), plasma cells (JCHAIN⁺IGH^hi), CD16⁺ NK cells (NCR1⁺FCGR3A⁺), CD16⁻ NK cells (NCR1⁺FCGR3A⁻), cDCs (CLEC10A⁺), plasmacytoid DCs (pDCs, CLEC4C⁺), proliferating cells (MKI67⁺HMGB2^hi), CD14⁺ monocytes (CD14⁺FCGR3A⁻), CD14⁺CD16⁺ monocytes (CD14⁺FCGR3A⁺), CD16⁺ monocytes (FCGR3A⁺CD14⁻) and a few liver macrophages (MARCO⁺CD5L⁺C1QA⁺), hepatic stellate cells (HSCs; SPARC⁺ACTA2⁺) and hepatocytes (ALB⁺APOA1⁺).

We compared major cell type proportions, as a percentage of each sample, across compartments and disease stages. Compared with PBMCs, liver FNAs contained higher proportions of CD16⁻ NK cells, MAIT cells and CD8⁺ T cells and lower proportions of B cells⁹ (Extended Data Fig. 3a). Liver FNA proportions of CD14⁺ monocytes, CD14⁺CD16⁺ monocytes, CD16⁺ monocytes, macrophages, B cells and cDCs increased in MASH compared with NS–SS (Fig. 1e), indicating recruitment to the liver as the dominant site of injury. We did not see changes in T lymphocyte or NK cell frequencies with disease before analyzing with respect to their subtypes (Extended Data Fig. 3b). We also used the public database Reactome to perform transcriptomic pathway enrichment analysis on major cell types in relation to patient disease parameters. In macrophages, we found a significant positive correlation between reactive oxygen or nitrogen species production (Reactome ID R-HSA-1222556), which are essential for phagocytic functions, and steatosis assessed by the controlled attenuation parameter (CAP, measured by Fibroscan; Extended Data Fig. 3c), consistent with macrophage activation in response to increasing liver lipotoxicity¹⁰. In HSCs, we found a positive correlation between collagen formation (Reactome ID R-HSA-1474290) and NAS, consistent with increased fibrogenesis with MASLD/MASH progression, and, in hepatocytes, a negative correlation between a pathway for cell growth regulatory proteins (insulin-like growth factor and insulin-like growth factor-binding proteins; Reactome ID R-HSA-381426) and ALT, consistent with decreased hepatocyte function as liver injury progresses (Extended Data Fig. 3c).

We also found upregulation of interferon-stimulated genes (ISGs; Reactome ID R-HSA-1169410) in MASH compared with NS–SS consistently throughout many cell types, in both compartments (B cells, CD14⁺ monocytes, CD16⁺ monocytes, CD16⁺ NK cells, MAIT cells, γδ T cells, CD8⁺ T cells and CD4⁺ T cells) and only in liver FNAs (CD16⁻ NK cells, pDCs and macrophages) or PBMCs (CD14⁺CD16⁺ monocytes) (Extended Data Fig. 4a). To investigate sources of interferons (IFNs) in the liver, we analyzed liver FNAs for IFN transcript expression across all cell types per sample and found IFNG expression predominantly in NK cells (CD16⁺ NK cells and CD16⁻ NK cells combined) and CD8⁺ T cells (Extended Data Fig. 4b). Expression of genes upstream of IFNα and IFNβ (Extended Data Table 2) were highest in pDCs (Extended Data Fig. 4c), indicating that pDCs were a predominant source of IFNα and IFNβ, and the proportion of pDCs as a percentage of patient liver FNAs increased in MASH compared with NS–SS (Extended Data Fig. 4d). To assess how strongly ISGs were upregulated in MASH compared with viral liver infection, we compared our liver FNA scRNA-seq data with published datasets of liver FNA myeloid cells from patients with untreated hepatitis C virus (HCV)⁸. The ISG expression of all monocytes combined and all macrophages combined was higher in HCV infection compared with NS–SS and MASH (Extended Data Fig. 4e), consistent with the role of ISGs as antiviral effectors¹¹. To validate the ISG expression patterns that we observed in MASH, we stained patient liver biopsies for ISG proteins (IFIT3 and MX1) in macrophages (CD68⁺) via immunofluorescence. We found that the proportion of IFIT3⁺ and MX1⁺ macrophages, as a fraction of CD68⁺ cells, increased from SS to MASH F1 and MASH F2, respectively (Extended Data Fig. 4f), aligning with reports that suggested IFNs, possibly triggered by sustained low-grade inflammation and gut leakage, were key to MASLD pathogenesis¹². Together these analyses demonstrated that transcriptomic data generated from liver FNAs accurately reflected known liver disease phenomena.

ISG^hi and S100^hi CD14⁺ monocytes are enriched in MASH

To investigate monocyte heterogeneity, we analyzed our scRNA-seq data for all monocytes (24,017 cells, liver FNAs and PBMCs combined) together with monocyte-derived cells—cDCs (881 cells) and macrophages (1,434 cells)—to account for monocyte differentiation paths. Unsupervised clustering and pathway analyses of these cells resulted in five distinct subgroups of CD14⁺ monocytes (Fig. 2a–d): immature monocytes (CD14⁺S100A10^hi), which showed low expression of major histocompatibility complex class II (MHC-II) and S100 genes, enrichment in translational pathways and overexpression of S100A10, a negative regulator of toll-like receptor (TLR) function¹³; ISG^hi monocytes (CD14⁺MX1⁺ISG15⁺), characterized by upregulated ISGs; inflammatory monocytes (CD14⁺FOSB⁺), which overexpressed immediate-early genes (EGR1 and IER2) and the proinflammatory IL1B; differentiating monocytes (CD14⁺CSF3R^high), which upregulated monocyte-to-macrophage differentiation genes (CSF3R and RGS2); and S100^hi monocytes (CD14 ⁺S100^hiHLA^lo), which showed high expression of alarmins S100A8/A9/A12, TLR4 signaling pathways and low expression of MHC-II genes, characteristic of an M-MDSC signature¹⁴, and showed higher expression of the M-MDSC gene signature¹⁵ compared with other monocyte subtypes (Fig. 2e and Extended Data Table 2).

Fig. 2 |. Monocytes, monocyte-derived cells and their connectivity across the MASLD/MASH spectrum.

a,b, UMAPs of monocytes, macrophages and cDCs color coded by cell type (a) and FNAs versus PBMCs (b) in cells as in Fig. 1b (clustering settings: 15 PCs; Louvain resolution, 0.6). c, Heatmap showing the top differentially expressed genes of each monocyte subcluster as in a. d, Pathway enrichment analysis for each monocyte subcluster as in a. The one-sided hypergeometric test with Bonferroni’s correction was used. DDX58, retinoic acid-inducible gene I (RIG-I); IFIH1, melanoma differentiation-associated protein 5 (MDA5). e, Box plot showing M-MDSC scores in each cell per monocyte subcluster for liver FNA cells as in a. Data are presented as median values (horizontal line), 25th or 75th quartile values (bounds of boxes) and nonoutlier minimum or maximum (whiskers): CD16⁺ mono (n = 3,617), CD14⁺CD16⁺ mono (n = 3,608), immature mono (n = 5,862), S100^hi mono (n = 3,299), ISG^hi mono (n = 2,575), inflammatory mono (n = 299) and differentiating mono (n = 4,116). Two-sided Wilcoxon’s rank-sum test relative to the reference group (S100^hi mono) was used: ^****P < 0.0001. f, Enrichment of monocyte subsets as in a in liver FNA cells versus PBMCs. Data are presented as observed log₂(fold-changes) ± 95% bootstrap confidence intervals (CIs), n values as in f, two-sided permutation test, significance determined as FDR < 0.05 and log₂(FD) > 0.58. g, Box plot showing the proportion of CD14⁺ monocyte subsets (immature mono, S100^hi mono, ISG^hi mono and differentiation mono) in NS–SS (n = 6), early MASH (n = 10) and advanced MASH (n = 9), shown as a percentage of liver FNA (nonparenchymal cells) or PBMC sample per patient. Data are presented as median values (horizontal line), 25th or 75th quartile values (bounds of boxes) and nonoutlier minimum or maximum (whiskers); permutation test: *Significance determined by FDR < 0.05 and log₂(FD) > 0.58. h, Cluster connectivity by PAGA with immature monocytes as the root node in a Reingold–Tilford tree plot layout for monocyte clusters as in a, with CD14⁺ monocyte subsets highlighted in gray. i, Slingshot analysis showing pseudotime progression of the monocyte or monocyte-derived cell lineage, shown by UMAP (left) and strip plot (right). NS, not significant; ZAP-70, ζ-chain-associated protein kinase 70.

We next analyzed the proportion of these monocyte clusters, as a percentage of each patient sample, across compartments and disease stages. Liver FNAs (all stages combined) had a higher proportion of inflammatory monocytes compared with PBMCs (Fig. 2f). Liver FNA proportions of immature monocytes, S100^hi monocytes (M-MDSCs), ISG^hi monocytes and differentiating monocytes increased in MASH compared with NS–SS (Fig. 2g), possibly reflecting increased monocyte turnover with inflammation-driven myelopoiesis and recruitment to the liver with induction of M-MDSCs (S100^hi monocytes) to dampen inflammation and differentiation to macrophages. To identify gene signatures that might drive monocyte recruitment to the liver, we investigated chemokine receptor genes in liver monocytes per sample (all monocyte clusters combined) and found CCRL2 and CMKLR1 significantly increased in MASH compared with NS–SS (Extended Data Fig. 5a). We then examined the expression of the gene encoding the chemoattractant protein chemerin that binds to both CCRL2 and CMKLR1 (RARRES2) across all liver FNA cell types and found a significant positive correlation between fat accumulation (CAP) and RARRES2 expression in hepatocytes (Extended Data Fig. 5b), consistent with reported associations between the level of chemerin and liver injury in MASLD^16,17.

Finally, to infer differentiation paths between monocyte clusters and monocyte-derived macrophages, we investigated the connectivity between these populations using the immature monocytes (CD14⁺S100A10^hi) as the origin. Cluster similarity (partition-based graph abstraction (PAGA)) and pseudotime (slingshot.R)¹⁸ analyses indicate that immature monocytes differentiate into S100^hi, ISG^hi or inflammatory or intermediate monocytes, with S100^hi and inflammatory clusters most strongly connecting to the differentiating monocyte cluster (Fig. 2h,i), consistent with reports that, although inflammatory monocytes develop into inflammatory macrophages, M-MDSCs (S100^hi) can differentiate into immunosuppressive macrophages that could be profibrotic¹⁹. We also investigated cell paths of monocyte differentiation without specifying a starting point of progression. Cell velocity (velocyto.R)²⁰ and unbiased pseudotime (slingshot.R)¹⁸ analyses implied a de-differentiation pathway away from macrophages to the differentiating monocyte cluster that intensified in advanced MASH compared with early MASH (Extended Data Fig. 6), consistent with preclinical work suggesting that, on MASH development, although monocyte-derived macrophages progressively replaced Kupffer cells (KCs), surviving KCs progressively lost their identity²¹. Together these results confirmed increased recruitment of monocytes to the liver with MASLD/MASH progression and pinpointed S100^hi monocytes (M-MDSCs) as contributors to liver fibrogenesis.

S100A9⁺ macrophages enrich with fibrosis in MASH

Liver FNAs captured a small number of macrophages (1,434 cells) in which unsupervised clustering and differential gene analyses resulted in seven clusters separated into two distinct groups: a group of transitioning monocytes (FCN1⁺NEAT1⁺, clusters PreMP_1–3) that shared monocyte features and expressed PLCG2 and CSF3R, involved in monocyte-to-macrophage differentiation and phenotype modulation^22,23, and a group of macrophages (FTL⁺TUBA1B⁺, clusters MP_0–3) (Fig. 3a,b). Within the transitioning monocytes (PreMP_1–3), PreMP_2 monocytes (ATP2B1-AS1⁺S100A12^hi) were enriched in liver FNAs and highly expressed proinflammatory genes (FOS and FOSB), whereas PreMP_1 and PreMP_3 were enriched in the periphery (Fig. 3a,b), possibly reflecting transitional monocyte states induced by systemic signals. Within the macrophages (MP_0–3), MP_0 macrophages (TUBA1B^hiHLA^hi) upregulated MHC-II but shared features with transitioning monocytes, likely representing differentiating macrophages (Fig. 3b); MP_3 macrophages (FOLR2⁺CD5L⁺) exhibited a strong KC signature enriched in MHC-II genes and anti-inflammatory or immunosuppressive functions (CD5L, MARCO, FOLR2, SELENOP and C1Qs) (Fig. 3b); and MP_1 and MP_2 macrophages exhibited an overlapping signature, possibly representing transitional states, where MP_2 macrophages (LGALS3⁺FABP5⁺) resembled lipid- or scar-associated macrophages (LAMs or SAMs) (TREM2, SPP1, CD9, LGALS1/3, FABP3/4 and ANXA1), suggested as controlling fibrosis^24,25, but also maintained expression of S100A9 (normally lost on macrophage differentiation¹⁹) and the transcription factor CEBPB (together with S100A9 reported to induce the immunosuppressive macrophage phenotype¹⁹) in 35% of cells (Fig. 3c). Cluster modularity analysis between each subcluster indicated that MP_0 and MP_3 had discrete gene signatures and suggested connectivity between the transitioning monocytes (PreMP_1–3) and macrophages (MP_0–3) through MP_0 (Fig. 3d). Unsupervised clustering of MP_0 segregated into MHC-II⁺ and S100A9⁺CEBPB⁺ subclusters (Fig. 3e), suggesting that MP_0 consists of a pool of macrophages becoming proinflammatory or immunosuppressive.

Fig. 3 |. Macrophage and pre-macrophage subsets and association with MASLD/MASH progression.

a, UMAPs of macrophages (MPs) and pre-macrophages (PreMPs) color coded by subcluster (top), compartment (FNAs or PBMCs, middle) and disease stage (NS–SS, early MASH and advanced MASH, bottom) in cells as in Fig. 1b (clustering settings: 15 PCs; Louvain resolution, 1.4). b,c, Marker gene expression dot-plot for all PreMP or MP subclusters (PreMP_1, PreMP_2, PreMP_3, MP_0, MP_1, MP_2 and MP_3) (b) and only MP_1, MP_2 and MP_3 (c) as in a. Color represents the average scaled expression of a gene in a cluster and size the percentage of cells that express a gene in a cluster. d, Matrix plot of pairwise cluster modularity index for each PreMP or MP subcluster as in a. e, Subclustering of MP_0 macrophages (clustering settings: 7 PCs; Louvain resolution, 0.4) as UMAP (left) and heatmap of differentially expressed genes (right). f, Differential abundance graph for MPs and PreMPs as in a depicting closely related cells (nodes), the layout of which is determined by the UMAP position of the neighborhood index cell. Node size represents the number of cells in the neighborhood. Lines represent the number of cells shared between adjacent neighborhoods. Color represents the log(fold-change) (log(FC)) differential abundance of neighborhoods in advanced versus early MASH. g, Volcano plot of differentially expressed genes in pooled MP_0, MP_1, MP_2 and MP_3 macrophage subsets in FNAs between early and advanced MASH (left) and the corresponding enriched pathways in early (middle) and advanced (right) MASH. One-sided hypergeometric test with Bonferroni’s correction was used. TriC/CCT, T-complex protein Ring Complex also known as Chaperonin Containing TCP-1 (CCT). h, IF showing staining for S100A9 and CD68 in livers from patients with SS (n = 1), MASH F1 (early MASH, n = 1) and MASH F2 (advanced MASH, n = 1) (top and bottom left) and percentage of S100A9⁺ cells among CD68⁺ cells in patients with NS (n = 1), SS (n = 1), MASH F1 (n = 1) and MASH F2 (n = 1) (bottom right). i, Representative immunohistochemistry images of S100A9 staining in male wild-type C57BL/6J mice fed with HF-CDAA every day for 4 (n = 4), 8 (n = 4) and 12 (n = 4) weeks, starting at 8 weeks of age. ER, endoplasmic reticulum.

Neighborhood-based differential abundance analysis revealed that MP_1 and MP_2 macrophages were enriched in advanced MASH compared with early MASH (Fig. 3f). Differential gene analysis comparing advanced MASH with early MASH macrophages (MP_0–3) revealed the upregulation of S100A9 and LAM- or SAM-associated genes, as well as downregulation of genes involved in antigen presentation (MHC-I or MHC-II), pattern recognition (C1Qs), phagocytosis regulation (selenoprotein or SELENOP) and damaged protein removal (UBC) (Fig. 3g), suggesting dysfunctional phagocytic, antioxidant and T cell-stimulating functions in advanced disease, consistent with a diminishing KC signature and replacement by monocyte-derived macrophages²¹. We used immunofluorescence (IF) staining to validate the observed expansion of S100A9⁺ macrophages in advanced MASH at the protein level in corresponding patient liver biopsies and found that the fraction of S100A9⁺CD68⁺ macrophages (as a percentage of CD68⁺ macrophages) was undetectable in NS and SS, emerged in MASH F1 and further increased in MASH F2 (Fig. 3h). Immunohistochemistry staining for S100A9 in liver tissue from mice fed a choline-deficient, amino acid-defined diet containing 60% fat by calories (HF-CDAA) to induce progressive fibrotic MASH²⁶ found that S100A9 sequentially increased from week 4 to week 12 of the HF-CDAA diet (Fig. 3i). Together, our results indicated the enrichment of immunosuppressive S100A9⁺ macrophages in response to continuous inflammation, which could further promote fibrosis in advanced MASH.

Liver pDC and S100^hiHLA^lo cDC2 frequencies mirror NAS

We next analyzed the composition and distribution of DCs, the main primers of T cell responses, in the liver FNAs (284 cells) and PBMCs (1,072 cells). Unsupervised clustering delineated three major subsets: pDCs (CLEC4C⁺), which are potent producers of type 1 IFNs, cDC1s (CLEC9A⁺), which are specialized in presenting antigens to CD8⁺ T cells, and cDC2s (CLEC10A⁺), which present antigens to CD4⁺ T cells and were further divided into HLA^hiS100^lo cDC2 and S100^hiHLA^lo cDC2 populations (Fig. 4a,b). Differential gene analyses showed that pDCs were enriched in proinflammatory genes (MZB1, SPIB and UGCG), enabling them to secrete high amounts of IFNα and genes such as TLR7 and GZMB²⁷, reported to mediate inflammation resolution and suppression of T cell expansion, respectively (Fig. 4c,d). Furthermore, cDC1s expressed XCR1 (Fig. 4c,d), encoding the receptor for lymphotactins XCL1 or XCL2, the interaction of which plays an important role in DC-mediated, CD8⁺ T cell, cytotoxic immune response²⁸. Although the HLA^hiS100^lo cDC2 cluster was enriched in antigen-presenting genes, S100^hiHLA^lo cDC2s enriched for regulatory genes, including TGFBR1 and S100A9, were reported to suppress DC cytokine production²⁹ (Fig. 4c,d). Marker gene analysis showed that S100^hiHLA^lo cDC2s also enriched for transcription factors of lipid metabolism (SREBF1/2), cholesterol uptake (LDLR), mevalonate or cholesterol pathway genes (HMGCSR, HMGCR, MVD and FDFT1), tricarboxylic acid (TCA) rate-limiting catalyst (IDH1) and maturity marker (FAS), which are markers of mature regulatory DCs (mregDCs) shown to promote T_reg cell function³⁰ (Fig. 4e).

Fig. 4 |. DC subsets and association with MASLD/MASH progression.

a,b, UMAPs of DCs color coded by subcluster (a) and compartment (liver FNAs or PBMCs) (b) in cells as in Fig. 1b (clustering settings: 15 PCs; Louvain resolution, 0.13). c,d, Marker gene expression in DCs as in a (c) and heatmaps of differentially expressed genes in cDC1, HLA^hiS100^hi cDC2, S100^hiHLA^lo cDC2 and pDC subclusters (d) as in a. e, Expression dot-plot showing marker genes, including SREBF1/2, LDLR, HMGCSR, HMGCR, MVD, FDFT1, IDH1 and FAS in cDC1, HLA^hiS100^hi cDC2, S100^hiHLA^lo cDC2 and pDC subclusters. The color represents the average scaled expression of a gene in a cluster and size the percentage of cells that express a gene in a cluster. f, Enrichment of cDC1s (n = 86 cells), HLA^hiS100^lo cDC2s (n = 389 cells), S100^hiHLA^lo cDC2s (n = 406 cells) and pDCs (n = 363 cells) as in a in liver FNAs versus PBMCs. Data are presented as observed log₂(FC) ± 95% bootstrap CI; two-sided permutation test: significance determined as FDR < 0.05 and log₂(FD) > 0.58. g, Frequency changes of cDC1, HLA^hiS100^hi cDC2, S100^hiHLA^lo cDC2 and pDC subclusters as in a among total PBMCs or liver FNA cells in relation to MASLD/MASH progression measurements. The two-sided Spearman’s test was used.

We next analyzed the proportion of these DC clusters, as a percentage of each patient sample, across compartments and disease stages. Differential abundance analysis between compartments showed a higher proportion of XCR1⁺ cDC1s in liver FNAs compared with PBMCs (Fig. 4f). Correlation analysis between DC cluster frequencies and patient MASLD/MASH progression measurements (histologically defined fibrosis stages, inflammation and steatosis grading and NAS; continuous disease parameters ALT, AST, APRI and FIB-4; and liver stiffness and CAP) revealed a decrease in the frequency of XCR1⁺ cDC1s with liver fibrosis measurements (histological fibrosis stages, APRI, FIB-4 and liver stiffness) in the liver and peripheral blood (Fig. 4g), consistent with previous reports³¹. The HLA^hiS100^lo cDC2 and S100^hiHLA^lo cDC2 subsets, in contrast, tended to increase with liver fibrosis measurements in the periphery (Fig. 4g), supporting reports that cDC1s correlate inversely with NAS activity and cDC2s in peripheral blood³¹. The S100^hiHLA^lo cDC2 (mregDC) subset was also enriched in the liver with MASLD/MASH progression measurements (Fig. 4g), consistent with findings in mice that cDC2s replace cDC1s in the liver with MASH³¹. Altogether, our data suggested the emergence of a pathogenic population of liver DCs that suppressed T cells with MASH progression.

T_H17 cells and GZMK⁺PDCD1⁺ CD8⁺ T cells enrich with fibrosis

Next, we analyzed T cells, the predominant immune population in our transcriptomic atlas (36,515 liver FNA cells and 72,744 PBMCs). Unsupervised clustering and differential gene analysis delineated 14 T cell subsets: CD4⁺ naive T (T_N) cells (CCR7^hiCD45RA⁺), CD4⁺ central memory T (T_CM) cells (CD45RO⁺SELL^hi), helper type 1 T (T_H1) cells (CCL5⁺HOPX⁺) and type 17 T (T_H17) cells (KLRB1⁺RORC⁺), CD4⁺ cytotoxic T lymphocyte (CTL) cells (GZMK⁺GZMA⁺), naive (RTKN2⁺CCR7^hi) and memory (RTKN2⁺CCR7^lo) T_reg cells, CD8⁺ T_N cells (CCR7^hiCD45RA⁺), CD8⁺ T_CM cells (CD45RO⁺SELL^hi), CD8⁺ effector memory T (T_EM) cells (CD45RO⁺SELL^lo), two subsets of CD8⁺ T_EM re-expressing CD45RA (T_EMRA) cells, T_EMRA 1 (CD45RA⁺SELL^loGNLY^hi) and T_EMRA 2 cells (CD45RA⁺SELL^loGNLY^lo) (both expressing GZMH and GZMB), MAIT cells (KLRB1⁺SLC4A10⁺) and γδ T cells (TRGC1⁺TRDC⁺) (Fig. 5a–d). Marker gene analysis showed that CD45RO⁺SELL^lo CD8⁺ T_EM cells expressed GZMK, the tissue-residency marker CXCR6, the exhaustion or activation marker PDCD1 (encoding PD-1) and the immunomodulatory chemokine XCL1 (Fig. 5e).

Fig. 5 |. T cell subsets and association with MASLD/MASH progression.

a,b, UMAPs of T cells color coded by subcluster (a) and compartment (FNA or PBMC) (b) in cells as in Fig. 1b (clustering settings: 13 PCs; Louvain resolution, 0.6). c,d, Marker gene expression in T cells as in a (c) and heatmaps of differentially expressed genes for CD4⁺ T cell subclusters (left) and CD8⁺ T cell subclusters (right) (d) as in a. e, Expression dot-plot for marker genes in CD8⁺ T_N cell, CD8⁺ T_CM cell, CD8⁺ T_EM cell, CD8⁺ T_EMRA 1 cell and CD8⁺ T_EMRA 2 cell subsets as in a. The color represents the average scaled expression of a gene in a cluster and size the percentage of cells that express a gene in a cluster. f, Enrichment of T cell subsets (MAIT cells (n = 5,891 cells), CD8⁺ T_EM cells (n = 8,186 cells), γδ T cells (n = 3,744 cells), CD4⁺ CTLs (n = 4,227 cells), CD8⁺ T_EMRA 2 cells (n = 6,270 cells), CD8⁺ T_EMRA 1 cells (n = 6,403 cells), CD8⁺ T_CM cells (n = 2,070 cells), T_H17 cells (n = 7,037 cells), naive T_reg cells (n = 1,427 cells), CD8⁺ T_N cells (n = 7,468 cells), T_H1 cells (n = 2,679 cells), CD4⁺ T_CM cells (n = 21,366 cells), CD4⁺ T_N cells (n = 27,813 cells) and memory T_reg cells (n = 4,653 cells)) as in a in liver FNAs versus PBMCs. Data are presented as observed log₂(FC) ± 95% bootstrap CI; two-sided permutation test: significance determined as FDR < 0.05 and log₂(FD) > 0.58. g, Frequency changes of T cell subsets as in a among total PBMC T cells or FNA T cells in relation to MASLD/MASH progression measurements. The two-sided Spearman’s test was used. h, NMF-derived functional programs expression by T cell UMAP plots (top), top genes for each NMF program (middle) with literature-based delineated functions (bottom). GZMH, granzyme H; GZMB, granzyme B. i, Matrix plot of pairwise correlation coefficients among NMF programs. The two-sided Spearman’s test was used. j, Matrix plot showing the median NMF expression of each T cell subset as in a. k, NMF program expression changes of T cells in PBMCs or FNAs in relation to MASLD/MASH parameters. The two-sided Spearman’s test was used. l, Enrichment of each NMF program in FNAs compared with PBMCs. The two-sided Wilcoxon’s test was used.

We next analyzed the proportion of these T cell clusters, as a percentage of each patient’s T cells, across compartments and disease stages. Differential abundance analysis between compartments showed that CD45RO⁺SELL^lo CD8⁺ T_EM cells and KLRB1⁺SLC4A10⁺ MAIT cells were enriched in liver FNAs compared with PBMCs (Fig. 5f). Within the CD4⁺ T cell subsets, we observed a positive correlation of CD45RO⁺SELL^hi CD4⁺ T_CM cells and KLRB1⁺RORC⁺ T_H17 cells in both liver and blood with fibrosis, with similar trends in CCL5⁺HOPX⁺ T_H1 cells in both compartments and RTKN2⁺CCR7^lo memory T_reg cells in the liver (Fig. 5g), consistent with reports describing higher T_H17 cell frequency in the liver and higher T_H17 cell to naive T_reg cell ratio in the periphery with MASH^32,33, possibly reflecting opposing proinflammatory, anti-inflammatory and/or regulatory agents, which together could drive fibrosis. Within the CD8⁺ T cell subsets, the effector memory populations showed opposite patterns between compartments with MASLD/MASH progression: CD45RO⁺SELL^lo CD8⁺ T_EM cells, which harbored an exhausted phenotype, significantly increased in the liver with liver stiffness but decreased in the blood with steatosis grading, whereas the CD8⁺ T_EMRA populations (CD45RA⁺SELL^loGNLY^hi CD8⁺ T_EMRA 1 cells and CD45RA⁺SELL^loGNLY^lo CD8⁺ T_EMRA 2 cells) decreased in the liver with steatosis grading and increased in the blood with CAP (Fig. 5g). This increase of CD45RO⁺SELL^lo CD8⁺ T_EM cells expressing an exhausted and immunomodulatory signature (PDCD1 and XCL1) with liver stiffness aligns with reports of accumulation of CD8⁺ T cells harboring a similar signature in the mouse and human liver with MASH^31,34,35. Naive T cells (CD4⁺ T_N cells, CD8⁺ T_N cells and naive T_reg cells) significantly decreased in the liver with MASLD/MASH progression measurements (Fig. 5g), possibly reflecting increased differentiation toward the mature phenotypes that increased with liver injury.

Next, we identified functional transcriptional programs in the T cell subsets and investigated their distribution across compartments and changes with MASLD/MASH progression measurements. Using non-negative matrix factorization (NMF)³⁶, we defined 12 distinct transcriptional programs with different distributions in the T cell subsets (Fig. 5h,i). Each T cell subset’s median expression of each transcriptional program revealed that the T cell subsets expressed different combinations of these functional programs (Fig. 5j). Correlation analysis showed NMF11, which represented a T cell activation program expressed across all T cells, positively correlated with steatosis and NAS in the liver (Fig. 5k), consistent with increasing T cell activation with MASLD/MASH progression. We used IF staining to validate the observed increase in T cell activation with MASH progression at the protein level in corresponding patient liver biopsies and found that the T cell activation marker, inducible T cell costimulator (ICOS), was enriched in advanced MASH (F2) compared with SS and early MASH (F1) (Extended Data Fig. 7a–c), while, at the transcriptional level, its ligand (ICOSLG) was mainly expressed in B cells and monocytes (Extended Data Fig. 7d). NMF3–NMF5 were mainly expressed in CD4⁺ T cell subsets involved in inflammation and its regulation (KLRB1⁺RORC⁺ T_H17 cells, CD45RO⁺SELL^hi CD4⁺ T_CM cells, CCL5⁺HOPX⁺ T_H1 cells, RTKN2⁺CCR7^hi CD4⁺ naive T_reg cells and RTKN2⁺CCR7^lo CD4⁺ memory T_reg cells) and increased with liver fibrosis and inflammation measurements in both the liver and peripheral blood (Fig. 5k), possibly reflecting the increased frequencies of CD4⁺ T cell-priming cDC2s. NMF6–NMF8 represented cytotoxic programs that were differentially expressed in CD8⁺ effector memory subsets (CD45RO⁺SELL^lo CD8⁺ T_EM cells, CD45RA⁺SELL^loGNLY^hi CD8⁺ T_EMRA 1 cells and CD45RA⁺SELL^loGNLY^lo CD8⁺ T_EMRA 2 cells) and NMF6–NMF8 correlated positively with inflammation measurements, but negatively with fibrosis measurements in both the liver and PBMCs (Fig. 5k), with NMF6 and NMF8 enriched in the liver FNAs compared with PBMCs (Fig. 5l). These results indicated enrichment of proinflammatory CD4⁺ T cell subsets (T_H17 cells and T_H1 cells) with parallel enrichment of immunoregulatory memory T_reg cells and exhausted CD8⁺ T_EM cells with MASLD/MASH progression and overall loss of cytotoxic T cell functions as liver fibrosis progressed, possibly due to reduced frequencies of CD8⁺ T cell-priming cDC1s.

B cell states shift with MASH progression

Unsupervised clustering of B cells (2,467 liver FNA cells and 13,054 PBMCs) delineated five B cell subsets (Fig. 6a,b). Differential gene expression analyses revealed two TCL1A⁺CD27⁻ naive B cell populations³⁷, including a transitioning subset (PLCG2⁺ naive B cells) expressing PLCG2, which facilitates the development of mature B cells upon B cell receptor (BCR) activation³⁸, where both naive B cell subsets expressed ICOSLG (Fig. 6c–e). It also revealed three CD27⁺TCL1A⁻ memory B cell populations that expressed the immune sensor AIM2 (ref. 39), including a subset (FCRL5⁺ memory B cells) that upregulated FCRL5, which encodes a BCR-signaling regulatory protein described to enrich in atypical memory B cells⁴⁰, and a subset (ITGB1⁺ memory B cells) that highly expressed proinflammatory genes (S100A4/6/10) and ITGB1, which are involved in B cell differentiation and phosphoinositide 3-kinase signaling⁴¹ (Fig. 6c–e).

Fig. 6 |. B cell subsets and association with MASLD/MASH progression.

a,b, UMAPs of B cells and plasma cells color coded by subcluster (a) and compartment (FNAs or PBMCs) (b) in cells as in Fig. 1b (clustering settings: 11 PCs; Louvain resolution, 0.5). c,d, Marker gene expression in B cells and plasma cells as in a (c) and heatmaps of differentially expressed genes per B cell subcluster as in a (d). e, Expression of ICOS ligand per B cell subcluster as in a. f, Frequency changes of B cell subclusters as in a among total PBMCs or liver FNA cells in relation to MASLD/MASH progression measurements. The two-sided Spearman’s test was used. g, Enrichment of plasma cells (n = 296 cells), memory B cells (n = 3,226 cells), naive B cells (n = 9,947 cells), FCRL5⁺ memory B cells (n = 540 cells), ITGB1⁺ memory B cells (n = 1,127 cells) and PLCG2⁺ naive B cells (n = 680 cells) as in a in FNAs versus PBMCs. Data are presented as observed log₂(FC) ± 95% bootstrap CIs; two-sided permutation test: significance determined as FDR < 0.05 and log₂(FD) > 0.58.

We investigated the frequency of B cell subsets as a percentage of each patient’s liver FNA cells or PBMCs, for changes regarding histological and noninvasive measurements of liver disease. Frequencies of TCL1A⁺CD27⁻ naive B cells, particularly of PLCG2⁺ naive B cells, positively correlated in the liver with MASLD/MASH progression, whereas frequencies of CD27⁺TCL1A⁻ memory B cells, including ITGB1⁺ memory B cells, correlated negatively in the liver but positively in the periphery with MASLD/MASH progression (Fig. 6f), possibly reflecting MASH-driven B cell turnover and peripheral recirculation. Although we could not profile the BCR repertoire due to the use of 3′-scRNA-seq, analysis of each patient’s B cell κ chain to λ chain ratio suggested lack of clonal expansion (Extended Data Fig. 7e). JCHAIN⁺IGH^hi plasma cells expressed genes essential for their maturation and proper assembly or secretion of antibodies (TNFRSF17 and FKBP11), were positively correlated with MASLD/MASH progression measurements in peripheral blood (Fig. 6f), and enriched in the liver compared to the blood (Fig. 6g). These results suggested the emergence of potentially pathogenic B cell populations with MASLD/MASH progression, including enrichment of naive B cell populations that expressed ICOSLG in the liver and possibly drove increased T cell activation via ICOS–ICOSLG interaction, and enrichment of memory B cells and plasma cells in the peripheral blood, contextualizing elevated serum levels of immunoglobulins with MASH progression^42,43.

Liver tissue-resident NK cells enrich with MASH

Unsupervised clustering of NK cells (7,404 liver FNA cells and 8,674 PBMCs) delineated 4 CD16⁺ (FCGR3A⁺) and 3 CD16⁻ subsets (Fig. 7a–c). Differential gene expression analyses revealed that CD16⁺ NK cells had low expression of NCAM1 (CD56^dim) and high expression of chemotaxis (CX3CR1) and cytotoxic (PRF1, GZMA, GZMB, GZMH and FGFBP2) factors (Fig. 7d,e). Their subsets included a cluster (CD16⁺CCL4^hi NK cells) with high expression of proinflammatory chemokines or ligands (CCL3, CCL4 and CCL4L2), a cluster (CD16⁺S100B⁺ NK cells) with high expression of GZMH and a cluster (CD16⁺PTGDS⁺ NK cells) with distinct expression of a gene involved in prostaglandin metabolism (PTGDS) (Fig. 7d,e). CD16⁻ (CD56^bright) NK cells consisted of a cluster (CD16⁻IL7R⁺ NK cells) with high expression of cytokine receptors (IL7R and IL18R1), immunocheckpoint membrane receptor NKG2A (KLRC1), transcription factor (TCF7) and cell migration or tissue-homing genes (CD44, GPR183, SELL and COTL1) and a cluster (CD16⁻IL7R⁻ NK cells) that contained CD160^hiCXCR6^hi tissue-resident NK cells that upregulated expression of an immunometabolic checkpoint (CMC1)⁴⁴, GZMK and XCL1, which recruits XCR1⁺ crosspresenting DCs²⁸ (Fig. 7d–e) and is enriched in liver FNA (Fig. 7f).

Fig. 7 |. NK cell subsets and association with MASLD/MASH progression.

a,b, UMAPs of NK cells color coded by subcluster (a) and compartment (FNAs or PBMCs) (b) in cells as in Fig. 1b (clustering settings: 16 PCs; Louvain resolution, 0.25). c,d, Marker gene expression in NK cells as in a (c) and heatmap of differentially expressed genes in NK cell subclusters as in a (d). e, Expression dot-plot for marker genes in NK cell subclusters as in a. The color represents the average scaled expression of a gene in a cluster and size the percentage of cells that express a gene in a cluster. f, Enrichment of tissue-resident NK cells (n = 2,237 cells), CD16⁻IL7R⁻ NK cells (n = 483), CD16⁺CCL4^hi NK cells (n = 596), CD16⁻IL7R⁺ NK cells (n = 1,549), CD16⁺ NK cells (n = 9,671), CD16⁺PTGDS⁺ NK cells (n = 1,251) and CD16⁺S100B⁺ NK cells (n = 290) as in a in liver FNA cells or PBMCs. Data are presented as observed log₂(FC) ± 95% bootstrap CIs; two-sided permutation test: significance determined as FDR < 0.05 and log₂(FD) > 0.58. g, Frequency changes of NK cell subsets as in a among total PBMCs or liver FNA cells in relation to MASLD/MASH parameters. The two-sided Spearman’s test was used. h, NMF-derived functional program expression by NK cell UMAP plots (top), top genes for each NMF program (middle) and with literature-based delineated functions (bottom). i, Matrix plot of pairwise correlation coefficients among NMF programs. The two-sided Spearman’s test was used. j, Matrix plot showing the median NMF expression of each NK cell subset as in a. k, NMF program expression changes of NK cells in PBMCs and liver FNA cells in relation to MASLD/MASH parameters. The two-sided Spearman’s test was used. l, Enrichment of each NMF program in liver FNA cells compared with PBMCs. The two-sided Wilcoxon’s test was used.

We investigated the frequency of NK cell subsets as a percentage of each patient’s liver FNA cells or PBMCs for changes regarding histological and noninvasive markers of liver disease. We observed the strongest frequency changes in CD16⁻ NK cell subsets: CD160^hiCXCR6^hi tissue-resident NK cells positively correlated with fibrosis in the liver, whereas other CD16⁻ NK cells correlated positively in the periphery and negatively in the liver (Fig. 7g), possibly implying a recirculation of these cells to the periphery, consistent with a report proposing differentiated NK cell recirculation as an alternative to a stable pool of tissue-resident NKs⁴⁵. Identification of functional transcriptional programs in the NK cell subsets using NMF³⁶ defined nine distinct transcriptional programs with different distributions in the NK cell subsets (Fig. 7h,i). Each NK cell subset’s median expression of each transcriptional program revealed that the NK cell subsets expressed different combinations of these functional programs (Fig. 7j). Investigation of functional program correlations with MASLD/MASH progression measurements indicated that inflammatory NMF5 (most highly expressed by CD16⁺CCL4^hi NK cells) positively correlated with MASLD/MASH progression measurements in the periphery, whereas immunomodulatory NMF6 (most highly expressed by CD16⁻IL7R⁺ NK cells) positively correlated with MASLD/MASH progression measurements in the liver (Fig. 7k). Cytotoxic programs (NMF1, NMF2 and NMF4) were enriched in CD16⁺ NK cell subsets and positively correlated with MASLD/MASH progression measurements in the liver (NMF4) and the periphery (NMF2) (Fig. 7k). Together, our data implicated damage-driving NK cell cytotoxicity as an alternative to declining CD8⁺ T cell cytotoxicity with MASLD/MASH progression.

HSC communication changes with MASLD/MASH progression

Last, we investigated the interactions between captured liver immune and nonimmune cells across MASLD/MASH stages using CellChat⁴⁶. The predicted total number of significant ligand–receptor interactions for each pair of cell types indicated that monocytes were the main drivers of cell-to-cell communication, increasing from NS as early as in SS and through advanced MASH, whereas lymphocyte interactions peaked in early MASH but decreased in advanced MASH (Fig. 8a). Notably, S100^hiHLA^lo cDC2s (mregDCs) showed strong and increasing communication with RTKN2⁺CCR7^lo CD4⁺ memory T_reg cells and CD45RO⁺SELL^lo CD8⁺ T_EM cells (Fig. 8a), suggesting that this immunoregulatory subset contributed to decreased T cell cytotoxicity. Total predicted ligand–receptor interactions for hepatocytes peaked in SS (Fig. 8a), possibly reflecting the initiation of immune responses upon stress, then diminished in MASH (Fig. 8a), reflecting a shift toward prevailing immune-driven pathogenesis.

Fig. 8 |. Hepatic immune cell crosstalk and changes with MASLD/MASH progression.

a, Predicted ligand–receptor interactions for each pair of cell types that were captured by liver FNA cells, for NS, SS, early MASH (F0–F1) and advanced MASH (F2–F4). Values of zero (black) reflect insufficient cell numbers for interaction analysis. b, Radar plot of the total number of cell–cell interactions (both incoming and outgoing) between HSCs and other cell types in SS, early MASH (F0–F1) and advanced MASH (F2–F4). c, Ligand–receptor interaction probability per stage between HSCs and CD8⁺ T_EM cells (left) and macrophages (right).

Within this ligand–receptor interaction analysis, we focused on immune cell crosstalk with HSCs, the key mediators of liver fibrogenesis. Comparison across disease stages showed that HSC crosstalk with monocytes, macrophages and cDC2s increased with disease progression, whereas interactions with other immune cells peaked in early MASH compared with SS and advanced MASH (Fig. 8b). Individual ligand–receptor interactions between HSCs and immune cells changed across disease stages (Fig. 8c and Extended Data Fig. 8a,b), confirming intensified HSC crosstalk with immune cells in early MASH versus SS. Most identified ligand–receptor interactions were reciprocal (from HSCs to immune cells and from immune cells to HSCs) and included signaling by chemotaxis, proinflammatory, growth, differentiation, activation and/or antigen presentation factors: in early MASH versus SS (interleukin (IL)-13–IL-13 receptor subunit α2 (IL-13RA2)), bone morphogenetic protein (BMP)_2/4–(BMPR_1/2 or activin A receptor type II (ACVR2_A/B)), ciliary neurotrophic factor (CNTF)–(CNTFR or leukemia inhibitory receptor factor (LIFR), growth arrest-specific 6 (GAS6)–AXL, growth hormone (GH)1–GHR and human leukocyte antigen (HLA) signaling) and in advanced MASH versus early (growth or differentiation factor (GDF)_5/2–(BMPR_1/2 or ACVR2_A/B), serine protease 3 (PRSS3)–(partitioning defective 3 homolog (PARD3) or coagulation factor 2 receptor (F2R))⁴⁷, vascular endothelial growth factor receptor 1 (VEGFR1) and IL-1 signaling) (Extended Data Fig. 8a,b). Unilateral ligand–receptor interactions (from HSCs to immune cells or from immune cells to HSCs) in early MASH versus SS included increased CCL14–CCR1 recruitment signaling from HSCs to immune cells, whereas communication with HSCs involved profibrotic transforming growth factor (TGF)β signaling (from pDCs, MAIT and CD4⁺ T cell subsets) and proinflammatory resistin (RETN)–TLR4 (from monocytes) (Extended Data Fig. 8a,b). Apoptotic TRAIL or TNF superfamily member 10 (TNFSF10)–TNF receptor superfamily member 10 (TNFRSF10) signaling from CD16⁻IL7R⁺ NK cells also increased in early MASH versus SS (Extended Data Fig. 8a,b), possibly reflecting the antifibrotic activity of this subset⁴⁸, whereas HSC-to-macrophage communication involved fibroblast growth factor (FGF)9 (ref. 49) signaling (characteristic of activated HSCs) to promote the profibrotic macrophage phenotype (Fig. 8c). In advanced MASH versus early MASH, HSC-to-macrophage signaling involved proinflammatory macrophage migration inhibitory factor (MIF)–(CXCR2 + CD74) to promote macrophage recruitment, whereas TGFβ signaling to HSCs increased from CD45RO⁺SELL^lo CD8 T_EM cells (which harbored an exhausted phenotype) (Fig. 8c). Although we could not assess HSC subsets and zonation in tissue, our analysis suggested that reciprocal interactions, in which HSCs activated immune cells and influenced their recruitment and phenotype and, in turn, immune cells modulated HSC activation and fibroblast proliferation, changed across MASLD/MASH stages.

Discussion

Here we described a single-cell transcriptomic resource of human immune cells from paired liver FNAs and peripheral blood across MASLD/MASH stages. Our findings included enrichment of immunoregulatory cell types with MASH progression, namely hepatic T_reg cells, M-MDSCs, TREM2⁺S100A9⁺ macrophages and S100^hiHLA^lo cDC2s. Liver cytotoxic T cells acquired an exhausted phenotype, whereas NK cell-driven toxicity intensified with fibrosis progression.

We observed a marked enrichment of monocyte subsets, monocyte-dominated cell crosstalk (with lymphocytes, DCs and HSCs) that increased as early as SS, and intensified overall T cell activation with MASH progression in the liver. These phenomena indicate recruitment and activation of the mononuclear phagocytic system in the liver, possibly initiated by chemotactic hepatocyte stress responses, where enriched monocytes prime or suppress T cell responses. Notably, enriched monocyte subsets included M-MDSCs (‘S100^hi’ monocytes), which are known to increase during inflammation and suppress T cell functions¹⁴. M-MDSCs have been shown to differentiate into immunosuppressive S100A9⁺ macrophages (which could be profibrotic) through persistent expression of S100A9 (ref. 15) and we found upregulation of S100A9 in macrophages in advanced MASH (MASH F2–F4) compared with early MASH (MASH F0–F1), suggesting a mechanism to promote fibrosis. Persistent expression of S100A9 in M-MDSC-derived macrophages was reported to regulate their suppressive activity via the transcription factor CEBPB¹⁵ and, in our data, 35% of MP_2 macrophages (which harbored a LAM or SAM signature) expressed S100A9 and CEBPB. These observations align with a report that TREM2⁺ LAMs represent an immunosuppressive state and mainly originated from S100A8⁺ monocytes⁵⁰, possibly M-MDSCs. It remains to be investigated, in a larger human macrophage dataset, whether LAM or SAM and S100A9⁺ macrophages represent different populations.

Our observation that a possible de-differentiation from macrophages into monocytes increased in advanced MASH (MASH F2–F4) compared with early MASH (MASH F0–F1) is similar to reports in preclinical work, where, on MASH development, surviving KCs progressively lost their identity²¹. The de-differentiation of human macrophages into monocyte-like cells has been shown to be induced by bacterial toxins that elevate cAMP, consistent with reported cAMP elevation in the MASH liver and its known effects on lipid metabolism^51,52. Increased cAMP signaling may be a molecular pathway that drives KC de-differentiation in MASH, although this possibility requires further experimental validation.

We found a marked increase in proinflammatory T_H1 cell and T_H17 cell frequency and function with MASH in the liver and peripheral blood, consistent with previous reports^32,33, alongside an increase in RTKN2⁺CCR7^lo memory T_reg cells, possibly as a feedback mechanism to reduce inflammation and suppress T cell function. We delineated a fibrosis-related enrichment of CD8⁺ T_EM cells harboring a previously reported exhausted signature^31,34,35 in the liver, whereas the frequency of CD8⁺ T_EMRA cells increased in the peripheral blood with liver steatosis as assessed by CAP, consistent with several studies documenting an increase of CD8⁺ T cells in either the liver or blood in patients with MASH compared with healthy donors^31,34,35. The CD8⁺ T cell subsets in our liver FNA dataset showed cytotoxic activity that positively correlated with patient liver inflammation scores, but negatively correlated with histological and noninvasive markers of fibrosis, indicating that, although cytotoxic T cell functions contributed to liver disease pathogenesis associated with inflammation, they waned as fibrosis advances. These T cell functional changes were mirrored by decreased frequencies of XCR1⁺ cDC1s (which prime CD8⁺ T cells) and increased frequencies of cDC2s (which prime CD4⁺ T cells), consistent with observations in mice that cDC1s are replaced by cDC2s with MASH progression³¹. Of note, the peripheral blood frequency of XCR1⁺ cDC1s has been reported to be greater in patients with MASLD/MASH compared with healthy controls⁵³, yet, when assessed within the MASLD/MASH spectrum, they were found to decrease with disease progression³¹, as in our data. We found that liver S100^hiHLA^lo cDC2s (mregDCs) were enriched and increasingly interacted with RTKN2⁺CCR7^lo memory T_reg cells with disease progression, possibly driving an immunosuppressive environment to further hamper GZMK⁺GZMA⁺ CD4⁺ CTL function during fibrosis. These observations indicate a shift from the antigen-specific, cytotoxic functions of T cells during MASH that could contribute to increased vulnerability during infections or toward developing liver cancer.

Our dataset captured pDCs and NK cells, which are less well understood in the context of MASLD/MASH. In particular, liver tissue-resident NK cells were enriched with disease progression and expressed, alongside proinflammatory chemokines, XCL1/2, which recruits XCR1⁺ cDC1s and could represent an attempt to rescue CD8⁺ T cell priming. This was accompanied by an enrichment in cytotoxic and inflammatory functions in CD16⁺ NK cells in the liver and blood with histological and noninvasive markers of fibrosis, suggesting a shifting role for NK cell subsets in damage-driving inflammation and non-antigen-specific cytotoxicity in MASH. A beneficial, antifibrotic NK cell interaction (CD16⁻IL7R⁺ NK cell signaling to HSCs through TRAIL) was increased only in early MASH (MASH F0–F1) and CD16⁻IL7R⁺ NK cells decreased in frequency with disease progression. Despite the known technical difficulties of capturing IFN transcripts by scRNA-seq⁵⁴, analysis of upstream genes delineated pDCs as strong producers of IFNs. Furthermore, pDCs were enriched in the liver with disease progression, possibly driving the increased ISG expression that we observed with MASH progression across cell subsets (for example, CD14⁺, CD16⁺ and CD14⁺CD16⁺ monocytes, CD16⁺ and CD16⁻ NK cells, MAIT cells and CD8⁺ and CD4⁺ T cells), which can drive immune exhaustion. In addition, pDCs increased signaling to HSCs through TGFβ in early MASH (MASH F0–F1) compared with SS, potentially contributing to the profibrotic milieu.

In conclusion, we generated a human cross-compartment scRNA-seq atlas of immune cells spanning the MASLD/MASH spectrum and uncovered contributions of immune cells to the hepatic inflammatory and profibrotic milieu and multiple types of immunoregulation leading to T cell exhaustion, which can contribute to increased vulnerability to liver cancer and infections in patients with MASH. Our dataset represents a resource that provides a basis for a deeper understanding of immunological plasticity and dysfunction in chronic liver disease in humans and provides a roadmap for new targeted antifibrotic therapies.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41590-025-02255-y.

Methods

Characteristics of patients with MASLD

We obtained samples from 25 patients (8 men and 17 women, aged 25–73 years) undergoing clinical liver biopsy for MASLD/MASH diagnosis at the Massachusetts General Hospital. No statistical methods were used to predetermine sample sizes but our sample sizes were similar to those reported in previous publications⁸. This was a cross-sectional study (involving all 25 patients in each transcriptomic analysis), which did not involve treatment and was designed by the investigators and approved by the Mass General Brigham Institutional Review Board and the sponsor. All patients gave written consent before participation. Compensation consisted of US$250 for a liver FNA, US$25 for a blood draw and a parking voucher. Inclusion criteria comprised age ≥18 years, fatty liver infiltration on ultrasound, computed tomography scan or magnetic resonance imaging and the ability to provide informed consent. Exclusion criteria included infection with hepatitis B virus, HCV or HIV, or other causes of chronic liver disease, including steatogenic medications and significant alcohol consumption (defined by >2 drinks per d in men or >1 drink per d in women). Each liver biopsy (with >15 mm length or >10 portal tracts) was reviewed by 2 single-blinded liver pathologists at Massachusetts General Hospital. When staging differed, the more advanced scoring was considered. Liver tissues were scored for the presence of fibrosis (modified Brunt’s stage, F0–F4) and assigned scores for the grade of steatosis (grade: 0 = <5% steatosis; 1 = 5–33%; 2 = 34–66%; 3 = ≥66%), hepatocyte ballooning (0 = no ballooning; 1 = few; 2 = many) and lobular inflammation per 200× (0 = no foci; 1 = ≤2 foci; 2 = 2–4 foci; 3 = ≥ 4 foci). The NAS was calculated as a sum of the scores for steatosis grade, lobular inflammation and hepatocyte ballooning. MASH was defined based on the Clinical Research Network recommendations as a score of ≥1 in each of the three components. Baseline demographics, including age, gender and race, as well as baseline clinical and laboratory values, were collected for all patients (Extended Data Table 1).

Sample collections and FNA procedures

We collected liver (FNAs) and corresponding peripheral blood and serum samples. The liver FNAs and PBMCs were processed fresh whereas the serum was frozen at −80 °C until use. The liver FNA procedure was guided by ultrasound using a 25G Chiba needle and consisted of four passes. The aspirate was air flushed into the bottom of the 5-ml collection tube and the needle was immediately flushed with a fresh 0.5 ml of cold medium (Roswell Park Memorial Institute-1640, Sigma-Aldrich) into the same tube. A new 25G needle and syringe were used for each subsequent pass. FNA collections were immediately placed on ice for transfer to the laboratory for processing

Cell preparation for 10x single-cell sequencing

Liver FNAs consisted of cell suspensions that did not require digestion. The least hematic FNA passes (usually one to two passes) were directly subjected to red cell depletion (EasySep RBC Depletion Reagent, STEMCELL Technologies, cat. no. 18170). PBMCs were isolated from acid citrate dextrose-treated blood by density centrifugation using Ficoll-Paque (GE Healthcare Life Sciences). The cell viability of the freshly purified PBMCs and liver FNA cell suspensions was assessed using Trypan Blue (0.4%, Thermo Fisher Scientific, cat. no. 15250061) and was always >95%. We obtained between 15,000 and 30,000 cells per liver FNA pass. The cell concentration for each sample type was adjusted to 1,000 cells μl⁻¹ for loading. In one case, the liver aspirate contained a small liver tissue piece that we dispersed mechanically, which corresponded to the only FNA sample where hepatocytes were the major population (Fig. 1d).

Library preparation and sequencing

Samples were loaded independently on to the 10x Genomics single-cell-G chip with a target capture of 10,000 cells: cell samples were processed on the Chromium instrument using the Next GEM single-cell 3′-kit (v.3.1) reagents, including primers. The quality of the resulting complementary DNA and sequencing libraries was assessed using Tapestation 4200 (Agilent Technologies). Libraries were sequenced on Illumina NextSeq 2000 instrument in the paired-end 28-bp or 90-bp mode.

Quality control and treatment of datasets

Raw sequencing data files (.bcl files) were converted to FASTQ format using the mkfastq pipeline from CellRanger v.2.1.1 (10x Genomics). Reads were aligned to the human reference genome GRCh38 release 97 from Ensembl (www.ensembl.org) using the CellRanger count pipeline. The default filtering parameters of CellRanger were applied to obtain a gene expression matrix (unique molecular identifier counts per gene per cell) for each library. In addition, cells with a library size <25%, genes detected in <3 cells, cells with mitochondrial content >35% and doublets were excluded. Cells across all samples highly expressed the housekeeping gene, B2M, confirming the quality of our data (Extended Data Fig. 2a).

ScRNA-seq analyses

ScRNA-seq datasets from liver FNAs and PBMCs were combined per donor. The dataset of all 25 donors (50 samples) was then pooled and analyzed using the Seurat R package. We performed quality control, filtering for high-quality cells with <35% mitochondrial content, >300 genes and at least 1,000 unique molecular identifiers. Cells presenting aberrant co-expression of genes from distinct clusters were detected as doublets and removed using the R package scDblFinder. After filtering, an average of 1,170 genes per cell were detected. The Harmony algorithm was applied to correct for batch effects, projecting cells on to a shared embedding where cells are clustered according to cell type rather than patient-specific or sample-specific effects. The Harmony embedding was used as an input to generate a low-dimensional representation of cells and perform clustering. Cell clusters were generated using Louvain modularity optimization algorithm on a K-nearest neighbor graph and low-dimensional representations were created using the Seurat implementation of Uniform Manifold Approximation and Projection (UMAP). The bluster R package was used to calculate root mean square deviation per subcluster as an assessment of internal cluster heterogeneity and pairwise cluster modularity based on a null model of random connections between nodes as a quantification of how distinct gene expression between clusters is, with values close to 1 indicating a strong difference from other clusters. These clusters were annotated according to differentially expressed genes (DEGs) and verified by annotation transfer of established expert annotations of human single-cell data in the Azimuth PBMC reference dataset. DEGs for each cell cluster were obtained using the Seurat implementation of Wilcoxon’s rank-sum test (FindAllMarkers function). The P values were adjusted using Bonferroni’s method. An adjusted P value (P_adj) of 0.05 and a log₂(fold-change) (log₂(FC)) of 0.25 were used to identify significant DEGs.

Of note, we excluded neutrophils from analysis, which certainly also play an important role in MASLD/MASH as the first responders of the innate immune system. Neutrophils and granulocytes have high levels of RNases and other inhibitory compounds, leading to their high sensitivity to degradation after sample collection and 10× processing steps. In our data, they were inconsistently captured by the 10× fluidic approach and, when captured (in liver FNAs), they had low levels of RNA content (in blood, they were filtered out during PBMC extraction via Ficoll).

Frequencies of cell populations or subpopulations and changes with disease parameters

Frequencies of immune cell populations and subpopulations were calculated as a percentage of sample (percentage of nonparenchymal in the case of FNAs), therefore representing relative proportions instead of concentration per volume or tissue features. Changes in the frequency of these cell subsets per compartment (PBMCs versus FNAs), and for each compartment per histological disease stage (NS–SS, early MASH and advanced MASH), were assessed by permutation testing from the scProportionTest R package or, for neighborhood-based analyses, by differential abundance analysis from the miloR R package. Spearman’s correlations of frequencies to clinical and histological parameters were performed using the R package stats (v.3.6.2).

Decomposition of cellular programs and changes with disease parameters

Functional transcriptional programs were inferred from single-cell data using NMF. This was performed using the nmf function from the R package Rcpp Machine Learning Library (RcppML) for cell types that had suitable numbers of cells for NMF, that is, T lymphocytes and NK cells, independently. The distribution of the resulting programs per cell subset was assessed by cell subset median expression of the NMFs. Changes in these programs between compartments and within each compartment with clinical and histological parameters were assessed using Wilcoxon’s and Spearman’s tests, respectively, from the R package stats (v.3.6.2).

Pathway enrichment analysis

To assess pathway enrichment, we first identified DEGs in the condition or population of interest (at least 10% of cells expressing gene, P < 0.05, log₂(FC) > 0.05). We used the R package clusterProfiler to visualize pathway enrichment assessed by the enrichGO function and the Gene Ontology Biological Processes database. All pathways shown are significant, according to the Benjamini–Hochberg-corrected P < 0.05.

Calculations of gene set enrichment scores

We utilized the AUCell R package to calculate gene set enrichment scores⁸. Gene sets used for these analyses were derived from the Reactome database (reactome.org) to create pathway-based signatures.

Trajectory inference

We inferred trajectories to liver monocyte and monocyte-derived cells (DCs and macrophages) using clusters that we generated via Seurat, as described above (Fig. 2). Partition-based graph abstraction (PAGA) was used to estimate the connectivity of clusters (partition) as implemented in Scanpy, a Python-based toolkit for analyzing single-cell gene expression. Pseudotime analyses¹⁸ were performed to assess progression through biological processes where cell subsets were ordered, based on their gene expression using slingshot R package. Two pseudotime approaches were used: (1) UMAP dimensionality reduction with pre-existing annotated clusters (Fig. 2i) and (2) principal component analysis dimensionality reduction without clustering (Extended Data Fig. 6b). Cell velocity analyses²⁰ were conducted to infer directions and speed of gene expression changes along an inferred trajectory, using the velocyto.R R package for each group of MASH stages—early MASH (F0–F1) and advanced MASH (F2–F4)—independently.

Inference of ligand–receptor interactions

Ligand–receptor interactions between hepatic cell populations were inferred using the CellChat R package based on the expression of known ligand–receptor pairs in different populations⁴⁶. Populations were annotated as described previously. Detected ligand–receptor interactions were first exported and displayed as heatmaps, presenting the total number of interactions between all pairs of captured hepatic cell populations and subpopulations, pooling data per histological stage: NS, SS, early MASH (F0–F1) and advanced MASH (F2–F4). We then investigated total interactions between HSCs and immune cells, which we displayed by radar plot (using the radarchart function from the fmsb R package), for which we identified individual interactions that were differentially expressed between stages (Extended Data Fig. 8a,b) and showed interaction probability per stage for two cell types (Fig. 8c). We also highlighted macrophage and pDC interactions in Extended Data Fig. 8c, but did not discuss in ‘Results’ due to limitations of space.

Liver tissue multiplex immunofluorescence

Formalin-fixed, paraffin-embedded (FFPE) liver sections were obtained from three MASH patients (SS, F1, F2) and one control (no steatosis, no inflammation) from the Massachusetts General Hospital pathology core facility. The multiparameter immunostaining was performed using the Opal 3-plex manual detection kit (Akoya, cat. no. NEL810001KT), which involves repeating sequential staining cycles using one primary antibody per cycle. In brief, FFPE liver tissue sections were first deparaffinized, rehydrated, fixed and photobleached. Next, slides were subjected to heat-based antigen retrieval (AR6 buffer), then blocked. Antibody dilutions and incubation time were adapted from parameters proven to work in tissue by standard immunohistochemistry. Each staining cycle consisted of blocking, primary antibody incubation, introduction of horseradish peroxidase (mouse and rabbit), deposition of an Opal fluorophore (different for each antibody or cycle) and antibody stripping via microwave treatment. Staining cycles were repeated until the three targets of interest were detected. The best antibody or fluorophore combination and sequence were decided based on the antibody tissue distribution, signal brightness and target stability to heat treatment.

Two sets of primary antibodies or fluorophores were used: set 1 (Fig. 3h): (1) CD68 rabbit polyclonal antibody (pAb) (Abcam, cat. no. 125157), 1:100, overnight at 4 °C; Opal-520 fluorophore, 7 min; (2) S100A9 (D5O6O) rabbit monoclonal antibody (mAb) (Cell Signaling Technology, cat. no. 34425), 1:400, overnight at 4 °C; Opal-570 fluorophore, 10 min. Set 2 (Extended Data Fig. 5f): (1) CD68 rabbit pAb, 1:100, overnight at 4 °C; Opal-520 fluorophore, 7 min; (2) MX1 (D3W7I) rabbit mAb (Cell Signaling Technology, cat. no. 37849), 1:160, overnight at 4 °C; Opal-570 fluorophore, 10 min; (3) IFIT3 rabbit pAb (Proteintech, cat. no. 15201–1-AP), 1:100, overnight at 4 °C; Opal-650 fluorophore, 10 min.

Before imaging, slides were mounted using ProLong Gold antifade reagent with DAPI (Invitrogen, cat. no. P36935). Multispectral images for each slide were acquired using the EVOS M7000 Imaging System (Invitrogen, cat. no. AMF7000). Three-dimensional deconvolution was performed using Celleste 6 Image Analysis Software (Invitrogen, cat. no. AMEP4942) and deconvolved images were analyzed with ImageJ software (v.1.53t)². Macrophages were identified by Otsu thresholding of CD68 fluorescence. Regions of interest for each cell were defined using ImageJ’s built-in particle analysis function. The number of DAPI⁺ nuclei was counted for each slide. Mean fluorescence intensities for markers of interest were calculated for each cell type (CD163, S100A9, MX1 or IFIT3 for macrophages). Cells were subsequently classified as positive or negative for these markers based on background intensity and population distribution. Liver tissue staining for ICOS was performed by Akoya Biosciences using Opal multiplex immunohistochemistry reagents.

Murine liver tissue immunohistochemistry

Progressively fibrotic murine steatohepatitis was induced in male wild-type C57BL/6J mice starting from 8 weeks of age as previously described in detail²⁶. Briefly, mice were fed HF-CDAA for 4 (n = 4), 8 (n = 4) and 12 (n = 4) weeks, as time points encompassing disease progression to advanced fibrosis. Mice fed regular chow (n = 4) served as normal controls. To validate the enriched expression of S100A9 in vivo, we performed S100A9 immunohistochemistry in FFPE liver sections staining using primary rabbit mAb (D3U8M, Cell Signaling Technology, cat. no. 73425) 1:800, overnight at 4 °C.

Measurement of soluble CD163 in patient serum

To assess the macrophage activation status, we measured levels of soluble CD163 in all 25 patients using ELISA (Quantikine ELISA, R&D Systems).

Statistical analyses

Unless otherwise noted, nonparametric statistical tests were applied, indicated in each figure caption. All tests were carried out as two-tailed tests and significance levels were defined as P < 0.05. When tests were performed per granular liver disease stage (SS or NS, F0, F1, F2, F3 or F4), at least four participants were included to enable comparison using standard statistical tests.

Extended Data

Extended Data Fig. 1 |. Serum levels of monocyte and macrophage activation marker, soluble (s)CD163.

Levels of sCD163 significantly correlated with (a) non-invasive markers of fibrosis, liver stiffness (kPa) and APRI score; and (b) liver injury enzymes, ALT and AST, two-sided Spearman correlation. (c) Levels of sCD163 were significantly higher in patients with histologically defined MASH, compared to healthy controls (n = 10), and in advanced MASH (n = 9), compared to NS-SS (n = 6); and (d) in patients with high NAS score (>4, n = 11) compared to low NAS score (<5, n = 14) (the box plots represent the median and interquartile ranges, and the whiskers depict the minimum and maximum of the data set), two-sided Mann–Whitney unpaired U test. *P < .05, **P < .01, ***P < .001, ****P < .0001. Abbreviations: kPa, kilopascal; APRI, aspartate aminotransferase–to-platelet ratio index; ALT, Alanine aminotransferase; AST, Aspartate aminotransferase; NAS; MASLD (former NAFLD) activity score. NS, no steatosis/inflammation/fibrosis.

Extended Data Fig. 2 |. Additional information on the quality of scRNA-seq dataset.

(a) Ridge plot of the housekeeping gene B2M expression per sample, per compartment. (b) Stacked bar chart of sample proportion in each cell type, per compartment. (c) Separate clustering UMAPs for liver FNAs (n = 55,234) and PBMCs (n = 123,753), color-coded by cell type (clustering settings: 20 PCs, Louvain Resolution=0.05).

Extended Data Fig. 3 |. Broad characterization of cell type enrichment per compartment and per liver disease stage.

(a) Box plots showing per-sample (n = 25) frequencies of CD16⁻ NK cells, MAIT cells, CD8⁺ T cells and B cells, and the ratio of CD4⁺/CD8⁺ T cells, per compartment: data are presented as median values (horizontal line), 25th/75th quartile values (bounds of boxes), and non-outlier minimum/maximum (whiskers); two-sided Wilcoxon. (b) Box plot of the proportion of each cell type as a percentage of each sample in FNA (left) and PBMC (right): data are presented as median values (horizontal line), 25th/75th quartile values (bounds of boxes), and non-outlier minimum/maximum (whiskers); NS-SS (n = 6), early MASH (n = 10), advanced MASH (n = 9); two-sided permutation test; significance (*) determined by false-discovery rate (FDR) < 0.05 and log2 false discovery (log2FD)>0.58. (c) Scatter plot of average pathway enrichment score per patient (n = 25) of (left) ROS/RNS production in macrophages by controlled attenuation parameter (CAP reflects steatosis), (middle) collagen formation in hepatic stellate cells (HSCs) by MASLD/NAFLD activity score (NAS), and (right) insulin-growth factor binding protein (IGF) and IGF-binding protein (IGFBP) in hepatocytes by liver enzyme ALT: error bands represent 95% confidence interval; two-sided Pearson.

Extended Data Fig. 4 |. Interferon-induced protein validation and liver sources of interferons.

Box plot by cell type per patient of average (a) ISG score per disease stage (NS-SS (n = 6), early MASH (n = 10), advanced MASH (n = 9)) in liver FNA (left) and PBMCs (right), (b) IFNG expression in liver FNAs (n = 25), and (c) upstream gene expression for interferon-alpha (left) and interferon-beta (right) in liver FNAs (n = 25): data are presented as median values (horizontal line), 25th/75th quartile values (bounds of boxes), and non-outlier minimum/maximum (whiskers); two-sided Wilcoxon; upstream genes are shown in Extended Data Table 2. (d) Magnification of Extended Data Fig. 3a: box plot of the percentage of pDCs out of each sample in FNA (left) and PBMC (right): data presented as in 5a; two-sided permutation test; significance (*) determined by false-discovery rate (FDR) < 0.05 and log2 false discovery (log2FD)>0.58. (e) Comparative ISG expression in liver monocytes (left) and macrophages (right) between patients without steatosis or with steatosis (NS-SS), those with MASH and those with untreated HCV: data are presented as median values (horizontal line); two-sided Wilcoxon; ***P < .001. (f) Immuno-fluorescent (IF) validation of MX1 and IFIT3 in macrophages in human liver tissue biopsies. Representative image merged and per channel (left) and expression fractions of MX1 + , IFIT3 + , and MX1 + IFIT3+ out of macrophages (CD68 +) for three disease stages (right): simple steatosis (SS, n = 1), early MASH (F1, n = 1), and advanced MASH (F2, n = 1).

Extended Data Fig. 5 |. Additional data on monocyte association with MASLD/MASH progression.

(a) Monocyte chemotaxis receptors: box plots of average CCRL2 (left) and CMKLR1 (right) expression in FNA by patient: data are presented as median values (horizontal line), 25th/75th quartile values (bounds of boxes), and non-outlier minimum/maximum (whiskers); NS-SS (n = 6), early MASH (n = 10), advanced MASH (n = 9); two-sided Wilcoxon (b) Monocyte chemotaxis ligands: scatter plot of average RARRES2 expression in hepatocytes per patient (n = 25) by CAP score: error bands represent 95% confidence interval; two-sided Pearson.

Extended Data Fig. 6 |. Investigation of the monocyte and monocyte-derived cell paths without specifying a starting point of cell progression.

(a) Velocity analysis of monocyte and monocyte-derived cells (macrophages and cDCs) shown by UMAP in early MASH (left) and advanced MASH (right) and (b) unbiased pseudotime via slingshot analysis, color-coded per cell type.

Extended Data Fig. 7 |. Investigation of T-cell activation marker ICOS, its ligand, and B-cell light chain kappa versus lambda expression.

(a) Immunofluorescent staining of ICOS in human liver biopsies for three disease stages: simple steatosis (SS, steatosis grade 1, n = 1), early MASH (F1, steatosis grade 1, n = 1), and advanced MASH (F2, steatosis grade 2, n = 1). Box plots of (b) average ICOS expression by T-cell subpopulation per patient FNA (n = 25), (c) average T_reg ICOS expression by histological steatosis grade per patient FNA (0–1: n = 10; 2: n = 11, 3 n = 3), two-sided Wilcoxon, and (d) average ICOSLG expression by cell type per patient FNA (n = 25): data are presented as median values (horizontal line), 25th/75th quartile values (bounds of boxes), and non-outlier minimum/maximum (whiskers). (e) Bar chart showing the fraction of B cells that express kappa (IGKC) or lambda (IGLC2–7) light chain genes for liver FNA (left) and PBMC (right).

Extended Data Fig. 8 |. Additional data on immune cell crosstalk.

Changes in predicted individual ligand-receptor interactions between (a) SS versus early MASH and (b) early versus advanced MASH, from HSCs to immune cells (left) and from immune cells to HSCs (right). (c) Radar plot of the total number of cell-cell interactions (both incoming and outgoing) between macrophages and other cell types (left) and between pDCs and other cell types (right).

Extended Data Table 1.

Patient characteristics.

Baseline characteristics	NS-SS (n=6)	Early MASH (F0-F1) (n=10)	Advanced MASH (F2-F4) (n=9)
Age, years - median (range)	50 (25 – 67)	57 (26–73)	55 (31 –65)
Male sex - no, (%)	2 (33.3)	3 (30.0)	3 (33.3)
Race - no. (%) White	6 (100)	9 (90)	9 (100)
Body mass index, kg/m2 - median (range)	30.8 (24.4–44.1)	35.6 (29.1 –52.8)	34.0 (27.5–37.4)
Platelet count, K/ul - median (range)	267 (169–315)	215 (181 –521)	233 (163–311)
Total Bilirubin, mg/dL- median (range)	0.55 (0.4–0.7)	0.7 (0.2– 1.3)	0.6 (0.3– 1.1)
Total Protein, g/dL - median (range)	7.65 (7.0– 8.3)	7.5 (6.8–8.2)	7.5 (6.3–8.6)
Albumin, g/dL - median (range)	4.35 (4.1 –4.9)	4.5 (3.9–5,2)	4.4 (4.2 – 5.0)
Globulin, g/dL - median (range)	3.3 (3.0–4.0)	3.0 (2.4–35)	3.2 (2.0 –3.6)
ALP, Ul/L- median (range)	125 (83– 169)	86 (44– 127)	76 (45– 90)
ALT*, Ul/L - median (range)	21 (8–42)	37 (25– 150)	52 (24– 189)
AST**, Ul/L - median (range)	20 (12–44)	35 (17–65)	55 (22–153)
Liver stiffness&, kPa - median (range)	4.4 (2.8–9.4)	8.65 (3.7–25.1)	13.5 (10.2–24.5)
CAP score - median (range)	281 (209 – 395)	346 (261 –381)	359 (241 –400)
FIB-4 score#-median (range)	1.14 (0.39– 1.84)	1.05 (0.35–2.00)	1.84 (0.69–3.25)
APRI score# - median (range)	0.29 (0.13–0.47)	0.44 (0.19–0.80)	0.76 (0.18–2.05)
Histology staging	2 NS, 4 Steatosis	5 F0, 5 F1	5 F2, 1 F3, 3 F4
NAS score - median (range)	0.5 (0–3)	4 (2–6)	5 (3–6)
sCD163$, ng/mL - median (range)	588.9 (332.8–652.0)	716.5 (539.9– 1909.8)	1387.5 (491.3– 1862.1)

^*ALT levels were significantly higher in NASH, early (p=0.015) and advanced (p=0.0002), compared to simple steatosis/NS

^**AST levels were significantly higher in NASH, early (p=0.049) and advanced (p=0.01), compared to simple steatosis/NS and in advanced NASH compared to early NASH (p=0.03)

^$Liver stiffness was significantly higher in NASH, early (p=0.02) and advanced (p=0.0002), compared to simple steatosis/NS and in advanced NASH compared to early NASH (p=0.016)

^#FIB-4 and APRI scores were significantly higher in advanced NASH compared to early NASH (p=0.025 and 0.031, respectively) and to steatosis/NS (p=0,039 and 0.02, respectively)

^$SCD163 levels were higher in NASH, advanced (p=0.01) and early (p=0.08), compared to simple steatosis/NS and levels were significantly higher in all 3 study groups compared to healthy controls (n=9, median: 356, range: 221 – 474, p<0.05)

Abbreviations:

ALP, alkaline phosphatase

ALT, Alanine aminotransferase

AST, Aspartate aminotransferase

kPa, kilopascal

CAP, controlled attenuation parameter

FIB-4, liver fibrosis 4, fibrosis assessment based on patient age, platelet count, AST and ALT values

APRI, aspartate aminotransferase-to-platelet ratio index

NAS, MASLD (former NAFLD) activity score

NS, no stealosis/no inflammation/no fibrosis as assessed by histology staging

sCD163, soluble CD163

Extended Data Table 2.

Gene sets used for custom scoring

Genes in M-MDSC score	Genes upstream of IFN-α	Genes upstream of IFN-β
S100A8	TLR7	DDX58
PADI4	TLR8	IFIH1
MTND5P32	TLR9	MAVS
S100P	MYD88	TLR3
STEAP4	IRAK1	TICAM1
PROK2	IRAK4	NAP1L1
QPCT	TRAF6	TBK1
F5	IRF7	IKBKE
RFLNB	IRF5	IRF3
DYSF		IRF7
CLEC4D
BPI
RBP7
VNN3
LIN7A
TSPAN2
NFE2
MTARC1
MEGF9
CYP1B1
RNF24
S100A12
EMB
ALOX5AP
FAM13A
MCEMP1
VCAN
FOLR3
AATK
MMP25
SYNE1
WLS
METTL9
DGAT2
GCA
PLBD1
GLT1D1
CDA
PCNX1
CXCL1
CCPG1

Supplementary Material

Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41590-025-02255-y.

Data availability

The human RNA-seq data are deposited in the NIH dbGaP portal (accession no. phs004044) and can be used only for studying health, medical or biomedical conditions. Source data are provided with this paper.