Acute liver failure is regulated by MYC- and microbiome-dependent programs

Acute liver failure (ALF) is a fulminant complication of multiple etiologies, characterized by rapid hepatic destruction, multi-organ failure and mortality. ALF treatment is mainly limited to supportive care and liver transplantation. Here we utilize the acetaminophen (APAP) and thioacetamide (TAA) ALF models in characterizing 56,527 single-cell transcriptomes to define the mouse ALF cellular atlas. We demonstrate that unique, previously uncharacterized stellate cell, endothelial cell, Kupffer cell, monocyte and neutrophil subsets, and their intricate intercellular crosstalk, drive ALF. We unravel a common MYC-dependent transcriptional program orchestrating stellate, endothelial and Kupffer cell activation during ALF, which is regulated by the gut microbiome through Toll-like receptor (TLR) signaling. Pharmacological inhibition of MYC, upstream TLR signaling checkpoints or microbiome depletion suppress this cell-specific, MYC-dependent program, thereby attenuating ALF. In humans, we demonstrate upregulated hepatic MYC expression in ALF transplant recipients compared to healthy donors. Collectively we demonstrate that detailed cellular/genetic decoding may enable pathway-specific ALF therapeutic intervention. A single-cell map of transcriptomic changes during acute liver failure unveils new insights into pathogenesis and potential therapeutic targets.


Results
A single-cell transcriptomic atlas of liver in healthy and ALF settings. We began our cell-specific exploration of the liver in homeostasis and ALF by characterizing healthy and liver-damage adult (8-week-old) specific pathogen-free (SPF)-housed C57BL6 mice, using the acute APAP and TAA ALF models (Methods, Fig. 1a and Extended Data Fig. 1a). Of note, both TAA and APAP elicit oxidative stress through similar mechanisms 2,9 , and an ensuing intense liver inflammation, further contributing to liver damage 10 .
To profile the hepatic nonparenchymal cellular populations in naïve and ALF settings, we depleted hepatocytes from liver cellular samples by centrifugation. Half of the resultant cellular fraction was further enriched for hepatic stellate cells (HSCs) by flow cytometry gating on intrinsic retinoid fluorescence of this cell population (Extended Data Fig. 1b) 11 . The remaining half remained unaffected, to enable unbiased quantification using single-cell genomics. Using droplet-based, single-cell RNA sequencing (RNA-seq), we analyzed 6,592 cells from three 8-week-old healthy male C57BL6 mice housed under SPF conditions at our facility, as well as 10,609 cells from APAP-treated mice and 8,500 from TAA-treated mice (Fig. 1b). Using hierarchical clustering, we identified 40 different cell populations that could be divided into six major types: immune cells, endothelial cells, HSCs, hepatocytes, cholangiocytes and mesothelial cells, collectively resulting in a high-resolution liver cell atlas (Methods and Supplementary Figs. [1][2][3][4][5][6][7][8][9][10][11][12]. We annotated cell clusters using conventional markers, major histocompatibility complex II (MHCII) expression and by comparison of their specific gene expression patterns to the Immgen database 12 (Fig. 1c and Extended Data Fig. 1c).
Within stellate cells we found four distinct populations and, based on their markers, we classified these as Lrat high quiescent stellate cells, Col1a1-positive fibrotic stellate cells, Acta2-positive ALF-activated stellate cells (referred to as AAs) and cycling stellate cells. In the endothelial cell population, we identified three clusters bearing different transcriptional signatures depending on their localization-the most abundant being a liver sinusoidal endothelial cell (LSEC) population and two additional, smaller   our studies, these transcriptional programs did not coexist within a single cell.
To further dissect the function of AAs cells, we performed differential expression analysis between quiescent and activated stellate cell populations and found that 421 genes induced by the toxic ALF insult are related to gene expression and translation (Fig. 2f), suggestive of a potentially increased protein production in these cells. This observation coincided with the higher number of detected transcripts in the activated population (Extended Data Fig. 1d-e), including secreted factors such as Ccl2, Ccl7, Csf1, Tagln, Tagln2 and Thbs1 (Fig. 2g). Within the cytokines induced in AAs during ALF were members of the interleukin-6 family, including Il6, Il11 and Lif 22,23 . Interestingly, receptors for these interleukins were expressed by different cell types, suggesting that responsiveness to interleukin-6 family cytokines may represent a possible 'division of labor' in cellular signaling (Fig. 2h).
Gene Ontology enrichment analysis of AAs also revealed terms associated with cell death: among upregulated genes within these terms we found Trp53 and Cdkn1a. In the presence of stress, p53 induces expression of the gene Cdkn1a that triggers cell cycle arrest, leading to senescence or apoptosis 24 . Increased cellular transcriptional activity, coupled with markedly induced senescence-associated secretory phenotype (chemokines, Timp1 and Ereg) expression collectively suggested that ALF-associated AAs cells may feature senescence rather than apoptosis 25 . Interestingly, a trend towards a lower number of proliferating cells was noted in ALF in all resident populations, further supporting the notion that liver cells may undergo cell cycle arrest (Extended Data Fig. 2a).
Importantly, AAs cells in ALF induced by either APAP or TAA clustered together, suggesting that differences in stellate cell activation states between these conditions are somewhat minor. To examine potential molecular differences between the two models, we performed differential expression with DESeq2 using pseudobulk from single-cell populations in the APAP and TAA samples (Extended Data Fig. 2b), highlighting 45 genes overlapping with the stellate cell activation signature-for example, Il11, Itga5 and Timp1 were higher in APAP while genes related to stress response, such as Mt1 or Hif1a, were higher in TAA. Together, this suggests that key transcriptional changes involving cytokines and extracellular matrix proteins are similarly upregulated in ALF regardless of the liver insult.
AAe cells. Endothelial cells regulate blood flow in the liver through vasoconstriction, form a barrier for molecules and immune cell liver trafficking through regulation of fenestration and partake in blood clearance through endocytosis 26 . Liver sinusoidal endothelial cells (Fig. 2i), but not venous and arterial endothelial cells (Extended Data Fig. 2a), assumed an activated phenotype upon and TAA-treated (n = 4) mice; significance was determined using a two-sided Wilcoxon test. Boxplot shows 25th to 75th percentiles with the 50th denoted by a line; whiskers show 1.5× interquartile range, or maximum or minimum if smaller. c, Balloon plots showing mean normalized and scaled expression of collagens in stellate cell subpopulations. d, Heatmap showing z-score for expression of genes from extracellular matrix GO category GO:0031012 that are significantly upregulated in AAs. e, Heatmap showing z-score for expression of genes from GO category stress fiber GO:0001725 that are significantly upregulated in AAs. f, GO term enrichment analysis of genes upregulated in AAs in comparison to quiescent cells. GO analysis was performed with GProfiler using standard settings; P values shown are corrected for multiple hypothesis testing using the g:SCS algorithm. g, Violin plots showing expression of chemokines, cytokine and extracellular matrix regulators in stellate cell populations. h, Balloon plot showing normalized and scaled expression of IL6 family cytokines and their receptors in all cell types. i, Percentage of endothelial cell populations in all endothelial cells in control (n = 3), APAP-treated (n = 4) and TAA-treated (n = 4) mice; significance was determined using a two-sided Wilcoxon test. Boxplot defined as in b. j, GO term enrichment analysis of genes upregulated in AAe in comparison to sinusoidal endothelial cells. GO analysis was performed as in f. k, Percentage of Kupffer cell populations in immune cells in control (n = 3), APAP-treated (n = 4) and TAA-treated (n = 4) mice; significance was determined using a two-sided Wilcoxon test. Data points as in b; boxplot defined as in b. l, GO term enrichment analysis of genes upregulated in AAk in comparison to Kupffer cells. GO analysis was performed as in f. m, Balloon plots showing significantly upregulated ligands in populations of stellate, endothelial and Kupffer cells and corresponding receptors, and their normalized and scaled expression in all cell types. ALF induction. Gene Ontology term enrichment analysis of upregulated genes in AAe revealed terms related to gene expression and terms associated with vascular remodeling (Fig. 2j) Fig. 2c). Comparison of AAe between the APAP and TAA models revealed 101 differentially expressed genes, including Fos and Junb, that function in stress response 27 (Extended data Fig. 2d). This suggests that TAA may elicit more oxidative stress than APAP.
AAk cells. Acute liver failure was associated with activation of, on average, 51.5% of Kupffer cells (Fig. 2k). Gene Ontology analysis of upregulated genes revealed terms related to chemotaxis, cell migration, immune response and apoptosis ( Fig. 2l and Extended Data Fig. 2e). Similar to stellate cells, apoptosis-related terms are probably associated with cell cycle arrest because we did not observe hallmarks of apoptosis such as increase in the percentage of mitochondrially encoded transcripts (Extended Data Fig. 3a). Kupffer cell activation was similar in APAP-and TAA-induced ALF, with 26 genes differentially expressed between the two disease models, including several interferon-responsive genes.
ALF-associated cell-to-cell signaling. Hepatic communication networks involving co-residing cells are frequently altered in liver disease 28,29 . To explore certain liver cell-to-cell communication motifs in steady state and ALF, we filtered differentially expressed genes between AAs, AAe and AAk and their quiescent equivalents, for ligands from a dataset curated by Ramilowski et al. 30 . We identified 18 ligands that can be grouped into the categories TGFβ ligands, chemokines, cytokines and growth factors, then identified matching receptors for these ligands. Due to redundancy in ligand-receptor interactions (for example, Ccl2 chemokine was shown to bind to Ccr1, Ccr2, Ccr3 and Ccr4), we grouped receptors into the same functional categories as ligands and found major signaling modules: chemokines target mostly the immune compartment, TGFβ targets mainly stellate and endothelial cells while growth factors and cytokines seem to potentially affect all cell types (Fig. 2m). The roles and contributions of these cell-and ligand-specific 'division of labor' networks in steady state and liver disease merit further studies.
Liver failure-associated cellular infiltration. We next investigated the characteristics of infiltrating cells during ALF. Indeed, in parallel to reduced cell proliferation of resident cells during ALF (Extended Data Fig. 2a), we identified populations of expanded hepatic subsets that did not have corresponding quiescent counterparts and thus probably represented infiltrating cells. Ly6C-positive monocytes expressed Ccr2; its ligand, Ccl2, was previously reported to be responsible for monocyte recruitment 31 (Extended Data Fig. 3b). In contrast, the neutrophil infiltrating fraction did not express Ccr2 and was probably recruited via a different mechanism 32,33 , possibly the highly expressed Ccr1 or Cxcr2 ( Fig. 2m and Extended Data Fig. 3c-d). Because both the infiltrating ALF-associated neutrophil and monocyte subsets were heterogeneous, we further dissected them in decoding potential distinct functional roles of their subsets.
Heterogeneity of neutrophils. We identified two neutrophil subpopulations (Fig. 3a), the larger subset representing classical, tissue-resident neutrophils 34 and the smaller expressing Ccl3, Ccl4, Cxcl2 and Csf1, suggesting that these cells were probably the proinflammatory subtype 35 . These neutrophils also expressed Nfe2l2 encoding the NRF2 transcription factor, known for regulation of the antioxidant transcriptional program 36 . Interestingly, Cxcr2 was downregulated upon neutrophil activation, possibly suggesting that it may be involved in mediation of infiltration (Fig. 3b). Neutrophil infiltration and activation were more pronounced in TAA-treated as compared to APAP-treated mice (Fig. 3c).
Heterogeneity of Ly6C-positive monocytes. Ly6C-positive monocytes have been suggested to infiltrate the liver in a number of pathologies 37 . We identified two populations of Ly6C-positive monocytes: the main population of 3,507 cells massively infiltrated the liver in ALF, while a small subpopulation of only 71 cells was not affected by the disease. Differential expression, GO and transcription factor binding site analyses revealed that the small population is most similar to Ly6C-positive monocytes and that it features an upregulated response to interferon ( Fig. 3d-i). The main population of Ly6C-positive monocytes exhibited further underlying heterogeneity. Diffusion maps 38 revealed that gene expression heterogeneity in this monocyte cluster stems from two processes, one consisting of monocyte homing to the liver and the other induction of MHCII complex gene expression (Fig. 3j). Of note, monocyte homing led to a gradual loss of expression of Ly6c2 and Sell, coupled with increased expression of Cxcl16, C1qa, Hmox1 and cathepsins (Fig. 3k).

Molecular activation patterns in ALF-induced resident cellular subsets.
Importantly, some of the upregulated genes in AAs, AAe and AAk, such as Ccl2, Nfe2l2 or Mt1, were common to these three cell types, suggesting a possible common activation signature in ALF. Indeed, we identified as many as 77 commonly expressed genes in AAs, AAe and AAk (Fig. 4a). A Monte Carlo simulation estimating the odds of such a commonality being random, using 10 9 iterations, did not observe a single instance of overlap with as many genes, strongly hinting at a common transcriptional response program underlying this expression pattern. An enrichment analysis of transcription factor binding site motifs within the promoters of this gene set yielded multiple different MYC binding motifs, suggesting that it may be a regulator of this response (Fig. 4b). To further corroborate this enrichment, we performed permutation analysis of a number of MYC binding sites within 77 randomly chosen genes: 10 10 iterations resulted in the distribution of a mean 197 binding sites and a maximum value of 343 binding sites, while within the common activation signature we discovered 402 MYC binding sites (Extended Data Fig. 4a). Furthermore, expression of the Myc transcript also trended towards upregulation in ALF but did not reach statistical significance (Extended Data Fig. 4b). At the protein level, Boxplots defined as in Fig. 2b. g, Percentage of monocyte IFN in immune cells in control mice (n = 3), APAP-treated (n = 4) and TAA-treated (n = 4) mice; significance was determined using a two-sided Wilcoxon test. Boxplots defined as in Fig. 2b. h, GO term enrichment analysis of genes upregulated in monocyte IFN in comparison to Ly6C-positive monocytes. GO analysis was performed with GProfiler using standard settings; P values shown are corrected for multiple hypothesis testing using the g:SCS algorithm. i, Transcription factor binding sites enriched in the promoters of genes upregulated in monocyte IFN in comparison to Ly6C-positive monocytes. Transcription factor binding site analysis was performed with GProfiler using standard settings; P values shown are corrected for multiple hypothesis testing using the g:SCS algorithm. j, Diffusion maps explaining heterogeneity within Ly6C-positive monocytes. k, Diffusion maps depicting expression of genes that change during the homing process.
we noted a significant elevation of MYC in mice treated with APAP or TAA in comparison to controls (Fig. 4c, Extended Data Fig. 4c and source data for Extended Data Fig. 4c), while phosphorylated MYC remained unchanged (Extended Data Fig. 4d and source data for Extended Data Fig. 4c). Collectively, these data suggested that a common Myc-regulated program might commonly control the activation state of AAs, AAe and AAk during ALF.
MYC inhibition leads to amelioration of ALF. Given these results, we reasoned that MYC induction might contribute to the altered stellate, endothelial and Kupffer cell states described above, while inhibition of MYC transcriptional activity might potentially attenuate resident cellular response to ALF-induced signals. Such inhibition, including that of the upregulation of Ccl2, the key chemokine promoting monocyte recruitment, may also lead to an impairment in Ly6C-positive monocyte infiltration, thereby further contributing to attenuation of ALF-induced hepatic damage. To test our hypothesis, we induced ALF with APAP or TAA and cotreated mice with the MYC inhibitor KJ-Pyr-9 (ref. 39 Fig. 4e). Indeed, we observed a significant reduction in monocyte infiltration in mice induced with ALF and cotreated with the MYCi inhibitor, suggesting that MYC may play a role in induction of the inflammatory response to liver damage in this setting (Fig. 4d,e). Serum aspartate transaminase (AST) and alanine transaminase (ALT) activity in both ALF models (Fig. 4f,g), and mortality in APAP-administered mice (Fig. 4h), were likewise attenuated upon MYC inhibition. Histologically, hematoxylin and eosin (H&E) staining of liver sections demonstrated, in both ALF models, that MYC inhibition led to reduced hepatic damage ( Fig. 4i-k).
To corroborate these results, we performed single-cell RNA-seq in mice receiving MYCi in the APAP and TAA ALF models, or in the absence of acute hepatic insult, to examine the effect of MYC inhibition on liver-cell-specific gene expression patterns. In the absence of ALF, MYCi did not exert a notable effect on the gene expression landscape ( Fig. 4l and Extended Data Figs. 4f and 5). Closer examination, by differential expression analysis of pseudobulk counts between samples, revealed that, during steady state, MYC inhibition led to differential expression of 152 genes in stellate cells, 99 in Kupffer cells and 9 in endothelial cells. Gene Ontology analysis of differentially expressed genes in the presence of MYCi in stellate cells revealed downregulation of genes coding for ribosome proteins and other components of the translation machinery, coupled with upregulation of terms related to developmental processes which included, among others, genes from the AP1 family (Fos, Jun and Junb), as well as Col3a1 and Cxcl12 (Extended Data Fig. 6a). Similarly, in endothelial cells, MYC inhibition during steady state induced downregulation of AP1 family genes, while in Kupffer cells MYC inhibition drove downregulation of genes related to antigen processing and presentation (mainly MHCII genes) (Extended Data Fig. 6b-c).
Importantly, during ALF induction in the presence of MYCi, activated populations (AAs, AAe and AAk) did not arise, and instead new cellular states of stellate, endothelial, Kupffer, dendritic and T cells were observed, these being markedly different from the populations found in the absence of MYCi and did not cluster together with conventional activation states ( Fig. 4l and Extended Data Fig. 4f). In addition, single-cell transcriptomic data reaffirmed that, in the presence of MYCi, a near-total abrogation of hepatic Ly6C-positive monocyte infiltration was noted in both APAPand TAA-induced ALF (Extended Data Fig. 6d-e). Interestingly, neutrophil infiltration was not affected by MYCi (Extended Data Fig. 6d-e), probably explained by lack of downregulation of the main neutrophil chemoattractant Cxcl2, in contrast to marked suppression of the monocyte chemoattractant Ccl2 (Fig. 4m). Importantly, expression of the vast majority of the 77 genes constituting the common activation signature in activated stellate, Kupffer and endothelial cell subsets was markedly attenuated in MYCi-treated mice. Only two genes, metallothionein 1 and 2 (Mt1 and Mt2), remained unaffected by MYC inhibition, suggesting that expression of these oxidative stress-response genes is regulated by a different mechanism ( Fig. 4l and Extended Data Fig. 6f-h).
Interestingly, MYC inhibition in ALF led to lower total gene expression. The strongest effect was observed in stellate cells, where the median number of detected transcripts dropped almost twofold, from 1,956 to 917. This effect was not a result of technical differences between samples, because the median number of detected transcripts in other cell types did not mirror such a difference (Extended Data Fig. 1d). One potential explanation for this reduction is apoptosis, which is associated with rapid messenger RNA decay and decrease in mitochondrial content 40 . Indeed, a strong downregulation of transcript number, with no increase in the percentage of mitochondrial reads, suggested that stellate cells may undergo cell death in the absence of MYC activity (Extended Data Fig. 3a). The process seemed to be specific to stellate cells, because endothelial and Kupffer cells did not exhibit these cell death hallmarks. Moreover, upregulation of the senescence and cell-cycle-arrest marker Cdkn1a (coding p21) in activated cells was attenuated compared to AAe and AAk in stellate cells upon MYC inhibition (Extended Data Fig. 6I). Gene Ontology analysis of genes upregulated in activated endothelial cells in the presence of MYC inhibition revealed terms related to apoptosis and its negative regulation and metabolism, while such analysis of Kupffer cells demonstrated mainly changes in immune-response-related terms (Extended Data Fig. 6j).
Microbiome modulation of the MYC program during ALF. We next sought to examine potential microbiome contributions to this ALF program. To this end, we induced disease in the APAP and TAA models following depletion of the microbiome of naïve or ALF-induced mice by a 2-week, wide-spectrum antibiotic treatment (ABX, 1 g l -1 ampicillin, neomycin, metronidazole and 0.5 g l -1 vancomycin in drinking water) 41 . To control for possible direct antibiotic impacts on liver physiology and ALF, we also induced ALF in germ-free mice (GF), which are devoid of a microbiome (Fig. 1a). Microbiome characterization by 16S rRNA gene V4 region amplicon sequencing of colon and jejunum content during disease induction demonstrated no major differences in relative abundance, other than an increase in alpha diversity during ALF (Extended Data Fig. 7a-c).
Comparison of cell numbers in naïve and ALF-induced GF and SPF mice demonstrated no new distinct cell populations, but Transcription factor binding site analysis was performed with GProfiler using standard settings; P values shown are corrected for multiple hypothesis testing using the g:SCS algorithm. c, Quantification of MYC expression levels in healthy and ALF mice from immunoblots; control (CTRL), n = 20; APAP, n = 15; TAA, n = 15; significance was determined using a one-sided Wilcoxon test. Boxplot defined as in Fig. Fig. 2. d-    in both ALF models fewer AAs, AAe and AAk cells were noted upon microbiome depletion (Extended Data Fig. 7d). Importantly, APAP-induced ALF in GF mice was associated with a significantly reduced infiltration of Ly6C-positive monocytes compared to SPF mice, in agreement with the attenuated APAP-induced liver toxicity noted in GF mice 42 (Fig. 5a). Moreover, liver damage in both the APAP and TAA models, as assessed by serum ALT and AST activity (Fig. 5b,c and Extended Data Figs. 4f and 8a) and histology (Fig. 5d,e), was milder in GF and ABX-treated mice as compared with microbiome-intact mice.
To determine whether functional differences in hepatic resident cells had contributed to this enhanced microbiome-induced monocyte infiltration, we next performed differential expression between GF and SPF mice using pseudobulk counts for each cell type with DESeq2. In naïve mice, the analysis revealed only two differentially expressed genes between SPF and GF mice-Cxcl14 in stellate cells and Trf in cholangiocytes (DESeq2 false discovery rate (FDR) adjusted P = 3.65 × 10 −6 and 1.23 × 10 −6 , respectively; Fig. 5f) 43 . In contrast, comparison of differential gene expression between APAP-administered SPF and GF mice revealed 127 differentially expressed genes between AAs cells, 87 in AAe cells and 17 in AAk cells. We did not find differentially abundant genes between AAs, AAe and AAk cells in GF and SPF mice in the TAA model. Among the more abundant genes in APAP ALF-induced SPF mice were Ereg and Thbs1 in stellate cells, Inhbb in endothelial cells and Cxcl2 in Kupffer cells (Fig. 5g) which, collectively, may have contributed to the difference in infiltration of Ly6C-positive monocytes between SPF and GF mice during ALF. Moreover, interleukin-6 family members (Il6, Il11 and Lif) and activation markers such as Thbs1, Timp1, Cd44, Itga5, Il17ra and Ereg featured higher expression levels in activated stellate cells in SPF mice, in agreement with the previously shown dependence of Ereg expression on the microbiome via TLR4 signaling in a hepatocellular carcinoma mouse model 5 . Under SPF conditions in AAe, we observed higher levels of activated endothelial cell markers, including Lrg1, Inhbb, Thbd and Bhlhe40, as compared to GF mice. Many stellate and endothelial cell genes upregulated in SPF mice were related to translation machinery. In Kupffer cells we found six genes to be upregulated in SPF mice, including Marco and Cxcl2. Only a few genes in GF mice were expressed at higher levels than in SPF mice-for example, Cox17 and Comt in Kupffer cells (Extended Data Fig. 8b-e). Depletion of the microbiome with antibiotics led to similar gene expression changes that were intermediate in extent between those for GF and SPF (Fig. 5b-g). Importantly, these observed microbiome gene-expression effects raised the possibility that the entire MYC-regulated ALF signature may be affected by the microbiome. Indeed, mean expression of the MYC-regulated gene signature in stellate, endothelial and Kupffer cells was significantly higher in SPF than in GF and ABX-treated mice (Fig. 5h). Together, these results suggest that microbiome-mediated upstream signals may regulate MYC during ALF.

TLR signaling and downstream adapters are necessary for activation of the MYC program in ALF.
We hypothesized that the microbiome regulates the MYC program in AAs, AAe and AAk cells during ALF through triggering of TLR signaling. In this scenario, signaling by damage-associated molecular patterns (DAMPs) originating from damaged liver cells, coupled with portal venous microbial-associated molecular patterns (MAMPs) originating from the gut microbiome, jointly drive TLR-induced MYC activation in these cells, leading to downstream immune cell infiltration and exacerbated disease. Indeed, a reporter cell assay (Methods) identified portal vein TLR2, TLR4, TLR5, TLR9, NOD1 and NOD2 agonists upon induction of TAA ALF, and TLR4, TLR9 and NOD2 agonists upon induction of APAP ALF (Extended Data Fig. 9).
To test whether potential TLR involvement exists downstream of these MAMPs, we utilized MyD88-Trif double-knockout (MyD88-Trif dKO) mice, which lack both adapter proteins necessary for TLR signaling, and performed single-cell RNA-seq under both naïve and APAP-treated conditions in these mice and compared to wild-type (WT) controls. In steady state, all cellular states in MyD88-Trif dKO mice were similar to WT mice, except for MyD88-Trif dKO Kupffer cells, which clustered separately from the respective cells in WT mice (Extended Data Figs. 5 and 10a). Interestingly, MyD88-Trif dKO Kupffer cells featured higher expression of interferon-responsive factors as compared to WT Kupffer cells (Extended Data Fig. 10b). monocytes within all immune cells in GF and SPF mice; significance was determined using a one-sided Wilcoxon test; n = 10 for each group. b,c, Activity of AST and ALT in mouse serum from APAP (b) and TAA (c) liver failure models in GF, ABX and SPF mice; significance was determined using a one-sided Wilcoxon test. APAP, n = 10 + 10 for ABX and SPF and n = 10 + 9 for GF, from two independent experiments; TAA, n = 10 for each group. Boxplots defined as in Fig. 2b. d,e, Histology scores of H&e-stained liver sections from APAP (d) and TAA (e) liver failure models in GF, ABX and SPF mice; significance was determined using a one-sided Wilcoxon test. APAP, n = 10 + 10 for ABX and SPF and n = 10 + 9 for GF, from two independent experiments; TAA, n = 10 for each group.  Kupffer Kupffer *** *** P = 0.243 *** *** *** *** *** P = 0.117 *** *** *** *** *** P = 0.236 *** *** *** Importantly, upon induction of APAP-driven ALF, MyD88-Trif dKO stellate and endothelial cells become aberrantly activated, assuming a transcriptional state distinct from that of ALF-induced WT mice and nearly identical to that of MYCi-treated mice (Extended Data Fig. 5). MyD88-Trif dKO Kupffer cells also became aberrantly activated, but their activation state was distinct from that of both APAP-induced and MYCi-treated, APAP-induced WT mice (Extended Data Fig. 10b). Similarly, neutrophils assumed an activated pattern markedly different from that observed in ALF-induced WT controls (Extended Data Figs. 4f and 10a). In corroboration of these findings, stellate, endothelial and Kupffer cells in MyD88-Trif dKO mice expressed metallothioneins Mt1 and Mt2 in response to APAP but failed to express Ccl2, Ccl7, Acta2, Csf1 and Inhbb, similarly to MYCi-treated mice in which MYC transcriptional activity was inhibited (Fig. 5i-k and Extended Data Fig. 10c-e).
Furthermore, expression of the 77-gene MYC-induced 'signature' in stellate, endothelial and Kupffer cells was significantly attenuated in MyD88-Trif dKO mice as compared to WT littermate controls, similarly to that observed upon MYCi treatment (Fig. 5l). Moreover, ALF-induced monocyte infiltration was blocked in the absence of TLR signaling in MyD88-Trif dKO mice while that of neutrophils remained unaffected. (Fig. 5m,n).
We next sought to study the downstream events by which TLR activation during ALF leads to activation of the MYC program during ALF. One apparent candidate pathway is MAPK, relaying signals from TLRs sensing MAMPs and DAMPs to regulate downstream MYC-dependent gene expression. In support of such pathway involvement in ALF are observations suggesting that TLR4 signaling regulates microbiota-dependent Ereg expression in hepatocellular carcinoma in stellate cells 5 , the senescence-associated secretory   Fig. 6 | The MAPK pathway relays signaling from TLR to MYC. a, FACS analysis of infiltration of Ly6C-positive monocytes, shown as a percentage of all immune cells from mice receiving APAP and inhibitors of MAPK pathway proteins; significance was determined using a one-sided Wilcoxon test; n = 5 for each group. b,c, Serum activity of AST (b) and ALT (c) for mice receiving APAP and inhibitors of MAPK pathway proteins; significance was determined using a one-sided Wilcoxon test; n = 5 for each group. d,e, Histology scores (d) of H&e-stained liver sections from APAP liver failure model (e); significance was determined using a one-sided Wilcoxon test; n = 5 for each group. Boxplot in a-d defined as in Fig. 2b; numbers above plots indicate P values. f, Immunohistochemistry of MYC in human controls and ALF samples; n = 5 for controls and n = 7 for ALF. Boxplot defined as in Fig. 2b; significance was calculated using a two-sided Wilcoxon test. g, examples of MYC immunohistochemistry. Arrows indicate positive staining for MYC. All scale bars, 100 μm. phenotype being downregulated in the absence of TLR2 (ref. 44 ), and the strong induction of Map3k8 expression (coding for TPL2) noted in the presence of MYCi during ALF (Extended Data Fig.  10f). To test for MAPK pathway involvement in ALF, we selected six proteins from the pathway for which small molecule inhibitors are available: IRAK4, RIP1, TAK1, TPL2, ERK1/2 and p38, and tested them in APAP-induced ALF [45][46][47][48][49][50] . Indeed, we observed significant reduction of monocyte infiltration in mice receiving inhibitors of IRAK4, TAK1 and p38 (Fig. 6a). AST activity in serum was significantly lower in the presence of IRAK4, RIP1, TAK1 and p38, while liver-specific ALT activity was lower in the presence of IRAK4, RIP1 and p38 (Fig. 6b,c). Histopathological analysis reaffirmed these results, demonstrating significantly reduced liver damage in mice subjected to IRAK4, TAK1 or p38 inhibition (Fig. 6d,e). Importantly, ERK1/2 inhibition did not induce any trend towards lower ALF severity, suggesting that it may not be involved in the observed regulation. Together, these results suggest that the ALF MYC program in stellate, endothelial and Kupffer cells is regulated via upstream TLR signaling, probably activated by tissue damageand microbiome-associated DAMPs and MAMPs. Hepatic TLR signaling, in turn, regulates MYC via activation of the MAPK pathway. TLR and MAPK pathway inhibition, or microbiome depletion, induces marked suppression of this cell-specific MYC program, thereby driving a significant attenuation of ALF.
MYC is upregulated in human ALF. Finally, we aimed to determine whether the noted MYC involvement in animal models of ALF could be observed in human patients. To this end, we quantified by immunohistochemistry the levels of MYC in hepatic liver sections obtained from seven patients with ALF (Extended Data Fig. 10g). As 'healthy' controls we used liver samples obtained from five cadaveric liver donors (Extended Data Fig. 10g). Indeed, a significant increase in nuclear MYC protein levels was noted in patients with ALF as compared to controls (Fig. 6f,g). The functional implication of this MYC upregulation, and whether inhibition of MYC signaling by the above-mentioned checkpoints may impact the course and outcome of human ALF, merit further studies.

Discussion
In this work, using two ALF animal models we uncovered new HSC, LSEC and myeloid cellular states characterized by distinct transcriptional signatures. We suggest that, during ALF, both microbiota-derived MAMPs and necrosis-derived DAMPs 51 signal to TLRs in resident stellate, sinusoidal endothelial and Kupffer cells, which activate MYC through IRAK4-and p38-dependent singaling 52,53 . Activated MYC in these cells, in turn, impacts downstream gene expression, leading to liver infiltration of Ly6C-positive monocytes. Importantly, MYC inhibition prevented activation of these ALF-associated subsets, thereby leading to significant amelioration of liver damage.
Our findings may constitute a first step towards the identification of new therapeutic targets in human ALF. Currently, beside liver transplantation, intravenous N-acetylcysteine constitutes the sole APAP-induced ALF treatment, by replenishing glutathione reserves depleted in APAP detoxification. This intervention is only partially effective and is accompanied by adverse effects, including anaphylactoid reactions in as many as 15% of cases. Even fewer therapeutic options are available for other ALF etiologies. Identification of MYC signaling as a potential regulatory axis of cellular response to ALF may enable disruption of ALF-induced liver pathology and damage, and merits further studies in human patients.
Future studies utilizing other ALF models in murine and human ALF should assess the commonalities and distinctions between cellular subsets and gene expression profiles in different ALF entities, and determine these cellular and genomic dynamics during hepatic regeneration from ALF. For example, the inflammasome-IL1 signaling axis may constitute another potential MyD88-MAPK-MYC-dependent avenue of regulation of inflammation and merits further studies. Future research may evaluate the context-specific contribution of distinct commensals and their products to the MYC-dependent gene signature. With these limitations notwithstanding, we suggest that cell-and pathway-specific molecular elucidation of ALF may allow utilization of host and microbiome inhibitors of signaling (such as MYCi and P38 inhibitors, researched in the cancer context) as future interventions in ALF.

Online content
Any methods, additional references, Nature Research 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/ s41591-020-1102-2.

Methods
Mouse models. Mice were kept in a standard conventional 12/12 h light/dark cycle, 21-24 °C, 55% relative humidity with 12-16 air changes h -1 , and fed commercially available standard chow and water ad libitum. Eight-week-old C57BL6 male mice were injected intraperitoneally with either 500 mg kg -1 body mass APAP in PBS or 300 mg kg -1 of TAA in PBS, 20 h before sample collection. To avoid known circadian effects on this model, we performed all injections between 13:00 and 14:00. Control mice were injected with vehicle (PBS). For antibiotic treatment, mice were given a cocktail of ampicillin (1 g l -1 ), neomycin (1 g l -1 ), vancomycin (0.5 g l -1 ) and metronidazole (1 g l -1 ) in drinking water for 2 weeks. MyD88 and Trif double-KO mice were 8-week-old males in a C57BL6 background 54 . All experimental procedures involving mice were approved by the local institutional animal care and use committee. Liver cell isolation. Liver cells were isolated using a modified protocol of Mederacke et al. 11 . In brief, using a peristaltic pump we performed retrograde liver perfusion into the inferior vena cava with three solutions: (1) EGTA (8 g l -1 NaCl, 0.4 g l -1 KCl, 88 mg l -1 NaH 2 PO 4 •H 2 O, 120 mg l -1 Na 2 HPO 4 •H 2 O, 2.38 g l -1 HEPES, 0.35 g l -1 NaHCO 3 , 0.19 g l -1 EGTA, 0.9 g l -1 glucose) for 2 min; (2) pronase (0.4 mg ml -1 protease in EBS buffer: 8 g l -1 NaCl, 0.4 g l -1 KCl, 88 mg l -1 NaH 2 PO 4 •H 2 O, 120 mg l -1 Na 2 HPO 4 •H 2 O, 2.38 g l -1 HEPES, 0.35 g l -1 NaHCO 3 , 0.42 g l -1 CaCl 2 ) for 5 min; and (3) collagenase D (0.1 U ml -1 collagenase D in EBS buffer) for 7 min. The liver was then dissected, placed in cold EBS solution and shaken vigorously with forceps to allow separation of single cells. The solution was filtered through 100-μm mesh and hepatocytes were depleted by centrifugation at 30g for 5 min. Cells were then collected by centrifugation at 580g and resuspended in cold PBS. To enrich for stellate cells, we sorted cells with retinoid fluorescence in the channel excitation 405, emission 450/40 using a BD FACSAria III. We then mixed stellate and unsorted cells, spun them down, resuspended them in PBS with 0.04% BSA and counted them using a Neubauer chamber, before proceeding to single-cell RNA-seq.
10X library preparation and sequencing. Single cells were captured and processed using the 10X Genomics Chromium 3′ Single Cell RNA-seq protocol according to the manufacturer's manual. Subsequently, the libraries were sequenced using NextSeq 500/550 High Output Kit v.2.

Measurement of monocyte infiltration.
Mouse livers were finely chopped with sterile scissors and then digested with 4 ml of prewarmed 0.4 mg ml -1 protease and 0.1 U ml -1 collagenase D (EBS buffer; Liver cell isolation) for 30 min at 37 °C, with shaking. Next, 10 ml of cold PBS was added and the suspension was filtered through 100-μm mesh. To deplete hepatocytes, samples were centrifuged at 30g for 5 min and the supernatant was transferred to new tubes. Cells were collected by centrifugation at 580g. To lyse red blood cells, 1 ml of Gibco ACK Lysing Buffer was added and cells were incubated at room temperature for 1 min. Subsequently, cold PBS was added and cells were collected by centrifugation at 580g.
Histology. Samples from the left lobe of the liver were fixed in 4% formaldehyde, embedded in paraffin, sectioned and stained with H&E. Slides were scored by a blinded veterinary pathologist for necrosis and hemorrhage, on a scale from 0 (healthy) to 5 (most severe).
Inhibitors. The MYC inhibitor KJ-Pyr-9 was injected intraperitoneally 2 h after injection with 500 mg kg -1 APAP or corresponding PBS vehicle. Next, 10 mg of KJ-Pyr-9 (Tocris, no. 5306) was dissolved in 1 ml of DMSO and combined with Tween80 and 5% dextrose (1:1:8 by volume). Mice were injected with this mixture or corresponding vehicle (0.5 ml per 20 g) to give a final dose of 25 mg kg -1 body mass.
Inhibitors of MAPK pathway proteins (see table below) were injected intraperitoneally, 1 h after injection with 500 mg kg -1 APAP or corresponding PBS vehicle. They were then dissolved in 5% DMSO in PBS to give a final injection volume of 400 μl per 20-g mouse. Control mice were injected with 5% DMSO in PBS vehicle.

Inhibitor
Target Dose (mg kg -1 ) Measurement of liver enzyme activity. Measurement of serum ALT and AST level activity was performed initially using a Liver-1 test on an Arkray SPOTCHEM EZ SP-4430, to validate the model (Fig. 1a). All following measurements were done using a Roche Cobas 111 Serum analyzer. Immunohistochemistry. Sections 4 μm in thickness were deparaffinized, rehydrated, treated for 30 min with 6 ml of H 2 0 2 + 200 ml of 70% methanol + 2 ml of HCl to block endogenous peroxidase activity, and washed in PBS. Antigen retrieval was done using citric acid. The sections were incubated with blocking solution, processed using an AB blocking kit and incubated overnight with anti-cMYC monoclonal antibody (13-2,500, 1:25; Invitrogen). The sections were then incubated with mouse biotin, processed with the ABC kit and stained with DAB and hematoxylin. They were then dehydrated, cleared in xylene and coverslipped. Sections were viewed using a microscope under ×20 magnification to monitor the color of the nucleus: a cell was considered positive if the nucleus was stained red/brown. The total numbers of positive and negative nuclei were determined automatically using the 'Image pro' computer program, followed by training the software on manually selected positive and negative cells.
Single-cell RNA-seq data analysis. Mapping. Single-cell RNA-seq data were demultiplexed, mapped to the GRCm38 mouse genome and unique molecular identifiers were counted using the Cell Ranger Single-Cell Software Suite 2.1.1 and bcl2fastq 2.17.1.14.
Filtering and doublet removal. First, cells with <100 detected transcripts and >15% mitochondrial reads were removed. We performed clustering and identified populations of thrombocytes, erythrocytes, neutrophils and mast cells. Next we performed a second filtering using 600 detected transcripts, but did not include the above-mentioned cell populations in this step because these cells have small transcriptomes and they would have been lost. A second step was necessary because there were many low-quality cells with a low threshold. Doublets were then identified by finding clusters of cells expressing gene expression patterns of two cell types simultaneously. The marker sets used were as follows: for stellate cells, Dcn, Colec11, Ecm1, Cxcl12, Sod3, Angptl6, Rgs5, Reln, Tmem56, Rbp1, G0s2 55 . Genes present in fewer than three cells were first removed, then highly variable genes were identified as those having a mean of nonzero values 0.0125-3 and s.d. >0.5. Dimensionality reduction was done with principal component analysis, with the first 50 principal components used for clustering.
Because the liver population is very complex, we decided to perform stepwise clustering. We first checked for the expression of Ecm1, Ptprc, Ptprb, Epcam, Msln and Alb in the clusters and divided the cells accordingly into seven groups: stellate, immune, endothelial, mesothelial and cycling cells, and cholangiocytes and hepatocytes. Immune cells were classified in the same way and, based on the expression of Agdre1, Cd5l, Ncr1, Cd3e, Cd79b, Retnlg, Cx3cr1 and Stmn1, were split into seven groups: B, T and natural killer cells, neutrophils, Kupffer cells, monocytes and remaining immune cell types. Within these groups, cells were clustered using Seurat FindClusters (Supplementary Figs. 1-12).
Cell type marker identification and annotation. For each cluster, marker genes were identified with the Seurat FindMarkers function and, based on crosschecking of identified markers with known marker genes and comparison to the ImmGen database (available at http://www.immgen.org/), we annotated clusters with cell types. Because many clusters were assigned to macrophages or dendritic cells, we added key marker genes to their cell type description to make comparison to other data easier for the wider audience. Clusters that represented the same cell type were merged-that is, Ly6C-positive monocytes and Cxcr6 T cells.
Functional analysis of cell populations. Gene Ontology analysis and transcription factor binding motif analysis was performed using g:Profiler 56 with default settings, and multiple hypothesis testing adjustment using all mouse genes as background control. The log 10 FDR-adjusted P values were plotted as barplots. For heatmaps, GO lists were obtained from Ensembl BioMart.
Mapping of ligand-receptor interactions. Differentially expressed genes between quiescent stellate, sinusoidal endothelial and Kupffer cells and their corresponding activated counterparts were filtered for ligands from a previously published database 30 . Corresponding receptors were then identified in the database, and both normalized and scaled expression of ligands and receptors were plotted as balloon plots.

Comparison of gene expression between clusters.
To compare gene expression between clusters, such as quiescent versus acute stellate cells, we used the Seurat FindMarkers function to define the identity of both clusters 55 .
Comparison of gene expression within a cluster between conditions. To compare gene expression between samples, such as activated stellate cells in GF APAP-induced versus activated stellate cells in SPF APAP-induced mice, we calculated pseudobulk by adding reads from all cells within each cluster in a sample. We then used DESeq2 with default parameters to determine differential expression 57 .
Diffusion maps. Diffusion maps were calculated using the destiny R package 38 . First, Ly6C-positive monocytes were clustered revealing four different subclusters; then, using the Seurat FindMarkers function, we identified the top 50 specific genes for each cluster. Normalized data were filtered for genes specific to these subsets, and these data were used to calculate diffusion maps using Euclidean distances and local-scale parameter sigma, without rotated eigenvalues, and taking the ten nearest neighbors.
16S V4 amplicon sequence analysis. 16S amplicon sequences were analyzed using Qiime2 (ref. 58 ), and sequencing reads were demultiplexed with demux plug-in. Thirty-one poor-quality bases were trimmed from the reverse read, and one base from forward read, combined, denoised and amplicon sequence variants (ASVs) was called with dada2. Sequences were aligned using Mafft, masked and a phylogenetic tree constructed using phylogeny fasttree; reads were then rarefied to 20,000 reads per sample. Taxonomic assignment to ASVs was done using feature-classifier classify-sklearn and Greengenes 13_8, 99% operational taxonomic units. Differential abundance analysis was done with a two-sided Wilcoxon test and Benjamini-Hochberg FDR correction.

Data integrity check. Figures, supplementary figures and supplementary
information panels were checked for data integrity using the Proofig pipeline. Fig. 2 | Cell abundance changes in acute liver failure and differences between APAP and TAA models. a, Percentage of cell populations in control mice (n = 6), APAP (n = 8) and TAA (n = 6) treated mice, significance was determined using two-sided Wilcoxon test. Boxplot defined as in extended Data Fig. 1a. Data points from SPF samples denoted as •, GF -■ and ABX -▲ b, Heatmap showing differentially expressed genes in AAs between APAP and TAA treated mice. c, Heatmap showing differentially expressed genes in AAe between APAP and TAA treated mice. e, Violin plots showing normalised and scaled expression of example chemokines upregulated in activated Kupffer cells. d, Violin plots showing normalised and scaled expression of example chemokines, cytokines and extracellular matrix modifiers upregulated in activated endothelial cells. e, Violin plots showing normalised and scaled expression of example chemokines upregulated in activated Kupffer cells. Fig. 6 | Effect of MYC inhibition on gene expression. a, b, Gene ontology term enrichment analysis of genes differentially expressed in healthy mice and healthy mice treated with MYCi in stellate cells and in Kupffer cells. GO analysis was performed with GProfiler using standard settings, p-values shown are corrected for multiple hypothesis testing using g:SCS algorithm. c, Volcanoplots showing differentially abundant genes healthy mice and healthy mice treated with MYCi. Y-axis value depicts multiple hypothesis testing corrected p-value calculated using DeSeq2 package. d, e, Barplot showing infiltration of (d) Ly6C-positive monocytes and (e) neutrophils in the presence and absence of MYCi. Different colors of bars denote subpopulations of neutrophils; legend as in extended Data Fig. 5. f-h, Boxplots showing expression of 77-gene signature in healthy mice, mice treated with APAP or TAA and mice treated with APAP and MYCi SPF (n = 3, cS=1999, ce=1463, cK=659), SPF + APAP (n = 4, cS=4339, ce=1517, cK=265), SPF + TAA (n = 4, cS=910, ce=1456, cK=285), in presence of MYCi: SPF + APAP + MYCi (n = 2, cS=251, ce=303, cK=125) and SPF + TAA + MYCi (n = 2, cS=198, ce=512, cK=233), *** denotes p-value < 0.001, n = number of mice, cS = number of stellate cells, ce = number of endothelial cells, cK = number of Kupffer cells. Boxplot defined as in extended Data Fig. 1a. Significance was determined using one-sided Wilcoxon test. p-values in stellate cells: SPF + APAP vs SPF + APAP + MYCi 1.135⋅10 −15 , SPF + TAA vs SPF + TAA + MYCi 2.662⋅10 −14 ; in endothelial cells: SPF + APAP vs SPF + APAP + MYCi 5.196⋅10 −6 , SPF + TAA vs SPF + TAA + MYCi 4.935⋅10 −6 ; in Kupffer cells: SPF + APAP vs SPF + APAP + MYCi 1.287⋅10 −8 , SPF + TAA vs SPF + TAA + MYCi 1.517⋅10 −7 (i) Violin plot showing normalised and scaled expression of Cdkn1a in three activated cells types. j, Gene ontology term enrichment analysis of genes differentially expressed in APAP and TAA treated mice with and without MYC inhibitor in stellate, endothelial and Kupffer cells. GO analysis was done as in extended Data Fig. 6a-b.