All methods and protocols were carried out according to the approved guidelines of the Vrije Universiteit Brussel (VUB, Belgium) and according to European Guidelines for the Care and Use of Laboratory Animals. Animal experiment protocols were approved by the Ethical Committee of Animal Experimentation of the Vrije Universiteit Brussel (VUB, Belgium, 14-212-4). BalbC mice aged 11–14 weeks were housed in a controlled environment in conventional cages and were allowed food and water ad libitum. Acute liver injury in BalbC mice was induced by a single intraperitoneal injection with 15 μl carbon tetrachloride (CCl₄, 87031, Sigma-Aldrich, St. Louis, MO, USA) and 85 μl mineral oil (Sigma-Aldrich, St. Louis, MO, USA) per 30 g bodyweight. Blood, total liver and cells were collected from healthy mice and after 1, 3 and 7 days of CCl₄ administration. Mice were anesthetized using 100 μL Dolethal^® (Vetoquinol, France). Analysis of alanine aminotransferase (ALT) was performed using a SPOTCHEM EZ SP-4430 (A.Menarini Diagnostics, The Netherlands). At the start and end of the experiment mice were weighted. Daily observation of the mice showed only a mild effect on animal welfare.

Non-parenchymal cells (NPCs) were retrieved as previously described (27). Red blood cell lysis (Miltenyi Biotec, Germany) was performed, and NPCs were washed with PBS + 0.1% Bovine Serum Albumin (BSA). NPCs were resuspended in BPE buffer (PBS with 5% BSA and 2 mM EDTA) with 1 ul anti-mouse Fc block™ (Becton-Dickinson, Belgium) reagent added per 10⁷ cells for 10 min at 4°C. Cells were washed and incubated in 600 μL PBS+0.1% BSA with 5 μl CD32-PE (ab30357, Abcam, UK), 2 μL CD45-FITC (11-0451-85, eBioscience, USA) and 10 μl F4/80 Alexa-647 per 10⁷ cells (MF48021, Life Technologies) for 15 min at 4°C. After incubation with the antibodies, cells were washed and resuspended in a buffer solution without calcium and supplemented with DNase I (3:1, 10104159001, Roche, Switzerland) before cell isolation using FACS (FACS Aria IIu, BD Biosciences, Belgium). FACS was used to sort viable cells (negative selection based on propidium iodide) and LSECs were selected and sorted based on a positive signal for CD32 (27–29) and a negative signal for UV, F4/80, CD45. CD32b is expressed in all LSECs across the liver sinusoid (Supplementary Figure 1A) (30). Potential doublets with HSCs, KCs, and immune cells were excluded (cfr. Supplementary Figure 1B. Utmost right FACS plot with circled LSEC population). Stainings were performed on cytospins after isolation and showed a high purity (95%) of LSECs using this sorting strategy (Supplementary Figure 1C).

Total RNA was extracted from FACS-isolated LSECs using ReliaPrep RNA Cell Miniprep System (Z6012, Promega, USA), RNA concentrations and quality measurements were performed using a Bioanalyzer 6000. Preparation of samples and sequencing, using Clontech SMARTseq v4 kit (R400752, Takara, Japan) and NovaSeq S2 (2 × 100 bp), was performed by the BRIGHTcore of the Vrije Universiteit Brussel. Single-end sequencing was run on Illumina NextSeq 500 High.

Mouse liver tissues were fixed with formalin for 48 h at 4°C. Liver tissues were stored in 70% EtOH and were used for sectioning (Leica, The Netherlands) of 100 μm liver sections in 4% UltraPure™ Low Melting Point Agarose (Invitrogen, USA) using a vibratome (Leica, The Netherlands). Sections were kept in 70% EtOH until usage. Upon usage sections were rehydrated in 50% EtOH for 10 min and rinsed for 10 min with PBS. For permeabilization, sections were incubated with PBS + 0.2% Triton for 20 min at room temperature. After permeabilization, sections were washed two times with PBS and blocked with 3% BSA-PBS for 2 h at room temperature. Sections were incubated overnight at room temperature with the following primary antibodies; Lyve1 (2 μg/mL, AF2125, R&D systems, Canada), Ki67 (0.5 μg/mL, 14-5698-82, Thermofisher, USA) and CD32b (10 mg/mL, AF2125, R&D systems, Canada). PHEM buffer (10 mM PIPES, 25 mM HEPES, 10 mM EGTA, 2 mM MgCl2^*6H2O) was used for CD32b staining instead of PBS in all steps. Vibratome sections were washed three times with PBS for 10 min and were incubated for 1 h with the following secondary antibodies (1:200); Donkey-anti-goat Alexa488 (A11055, Thermofisher, USA) and Donkey-anti-rat Alexa 647 (ab150155, Abcam, UK). Sections were washed three times with PBS, incubated for 10 min with 70% EtOH and then incubated with 1% Sudan Black (199664, Sigma-Aldrich, Belgium) in 70% EtOH. Sections were rinsed with PBS and mounted with Mowiol (9002-89-5, Sigma-Aldrich, Belgium) with DAPI (D9564, 10 μg/mL, Sigma-Aldrich, Belgium) and visualized by EVOS M7000 (Thermofischer, USA) and Zeiss Axioscan (Zeiss, Germany). Quantification was performed with HALO 3.1 image analysis platform (Indica labs Inc., USA).

Liver tissues were embedded in paraffin, sliced in 5 μm sections and deparaffinized with Xylene. For H&E stainings sections were rehydrated, washed with PBS and counterstained with Harris Hematoxylin (1:10 Roth, Newport Beach, CA, USA) before being rinsed with acid water followed by 10 min wash with tap water. Sections were incubated with eosin for 5 min, shortly rinsed, dehydrated and mounted with DPX mounting medium (Sigma-Aldrich, Belgium). For Collagen 4 staining, sections were rehydrated, washed with PBS-0.05%Tween (PBST) and endogenous peroxidase was quenched with 3% H₂O₂ in methanol. Samples were washed three times with PBST for 5 min and incubated with 2% BSA-PBS for 1 h at room temperature. Col4 antibody (2 μg/ml, ab6586, Abcam, UK) was dissolved in 1% BSA-PBS and incubated overnight at 4°C. Sections were washed and incubated with Dako EnVision+ System- HRP Labeled Poly (K4003, Dako, Denmark) for 30 min at room temperature. Sections were washed with PBST, incubated with DAB substrate for 3 min at room temperature. Finally, samples were rinsed, counterstained with Harris Hematoxylin (1:10) and mounted with DPX mounting medium (Sigma-Aldrich, Belgium) and imaged visualized with Leica Aperio CS2 (Leica, The Netherlands). Quantification was performed with Orbit image analysis (31).

One-tailed Kruskal Wallis with Dunnett's multiple comparisons test was applied for the statics of ALT measurements, CD32b, Lyve1 and Lyve1/Ki67 stainings. Calculations were made using GraphPad Prism 9. Ns > 0.05, ^*P ≤ 0.05, ^**P ≤ 0.01.

To identify the presence of healthy or dysfunctional LSECs in RNA profiling data sets from human or mouse livers we set out to identify genes that are enriched in LSECs from healthy or cirrhotic livers. To this end, we used scRNAseq data of healthy and cirrhotic human livers reported by Ramachandran et al. (25). First, we identified LSECs and ECs expressing known LSEC and EC markers in healthy livers (Figure 1A; Supplementary Figure 3). Next, we identified an additional cell population which was not present in the endothelial cell population of healthy livers (Figure 1B). These cells were CD34⁺PLVAP⁺VWA1⁺ positive which strongly resembled the scar-associated endothelial cell population identified by Ramachandran et al. (25). These cells were restricted to cirrhotic livers, expressed pro-fibrogenic genes and displayed an immunomodulatory phenotype (25). We refer to this population as damaged LSEC/ECs (Figure 1B) as some of the markers expressed in this population show a sinusoidal expression pattern in cirrhotic livers (25) but damaged ECs cannot be excluded. Subsequently, we defined genes that were higher expressed in healthy LSECs or damaged LSEC/EC population compared to all other liver cells (endothelial cells, macrophages, stellate cells, cholangiocytes, innate lymphoid cells (ILC), dendritic cells, T and B cells, hepatocytes and plasma cells) from healthy and cirrhotic livers with a fold change of at least two, and every gene should be expressed in at least 50% of cells within the healthy LSEC or damaged LSEC/EC population. This resulted in, respectively, 60 and 48 genes that were higher expressed in LSECs from healthy- or cirrhotic human livers in comparison to other liver cell types (Figures 1A,B). To identify genes that can further distinguish LSECs from healthy or diseased livers, we performed differential expression analysis between both populations resulting in a list of genes expressed higher in healthy LSECs compared to damaged LSEC/ECs from cirrhotic livers (Figure 1C). By combining genes that are enriched in LSECs from healthy or cirrhotic livers vs. other cells with genes that are higher expressed in one of the conditions vs. the other, we could create two genesets: a geneset for LSECs from healthy livers (n = 48) and a geneset for damaged LSEC/EC from cirrhotic livers (n = 15) (Figure 2A; Supplementary Table 1). Next, we performed gene ontology analyses to summarize the overlap in biological functions of genes included in each geneset. Genes that were enriched in healthy LSECs were part of GOs that are related to scavenging function and viral entry, both important characteristics of LSECs (46, 47). Genes that are enriched in damaged LSEC/EC belong to GOs that are related to dysfunctional LSECs, such as vascular development, migration and matrix organization, which are typical features of liver fibrosis (48, 49) (Figure 2B).

Next, we wondered whether we could use these gene sets to visualize an enrichment of damaged LSECs/ECs in microarray gene expression data from human livers. We therefore performed gene set enrichment analysis (39, 50) (GSEA) with the two LSEC gene sets on microarray gene expression data from human livers with different etiologies to identify the presence of healthy or damaged LSECs/ECs in advanced cirrhotic liver diseases. Figure 2C shows that gene sets that were highly expressed in healthy LSECs were substantially enriched in transcriptomes of healthy livers and diseases livers with early stage liver fibrosis (F0–F1). Gene sets that were highly expressed in damaged LSEC/ECs were enriched in cirrhotic livers (vs healthy livers) (45); ASH (vs alcoholic steatosis) (44), HCC (vs normal tissue) (45), NASH (F4–F3 vs. F0–F1) (43) and HBV (F3–4 vs. F0–1) (42) (Figure 2D). Taken together, our analysis suggests that LSECs transform into a more damaged endothelial cell phenotype in all advanced liver diseases that we investigated.

LSECs play a crucial role in the regenerative response after an acute injury that can either lead to liver regenerative or a maladaptive fibrotic response (17). Yet, all studies and datasets we have used so far only reflected chronic liver injury. Therefore, we wanted to know whether the gene sets could also demonstrate LSEC phenotype changes after an acute injury. To this end, acute liver injury in mice was induced with a single dose of CCl₄ and livers were collected at 1, 3, and 7 days after injection (Figure 3A). Blood analysis shows acute liver injury (high ALT levels) at 24 h after a single dose of CCl₄, which decreases to baseline levels at day 7 (Figure 3B). Hematoxylin eosin staining shows necrotic areas that appear at 1 day and are more pronounced after 3 days demonstrating that liver injury is still present at that time point (Figure 3C). However, after 1 week the liver appears to have recovered from the injury. When we further examine LSECs through staining, we see an increased trend of Lyve1 protein levels indicating that LSECs are still present and sinusoids are intact. However, we observed a temporary loss of CD32b expression after 1 and 3 days of CCl₄ treatment, indicating at least partial LSEC dysfunction which is restored after 1 week.

To further analyse LSECs after acute liver injury, livers were collected and LSECs were isolated via FACS at 1, 3, and 7 days after CCl₄ administration (Figure 3A; Supplementary Figure 1B) and transcriptome analysis was performed on the freshly isolated LSECs. Four samples were included for each condition apart from LSECs after 1 day of CCl₄-treatment, because 2 samples did not meet the quality standards for RNA sequencing (low RIN values). Although this reduces the statistical power, still more than 2,000 genes were differentially expressed when compared to healthy LSECs (Supplementary Figure 4). Principal component analysis (PCA) demonstrates separated clusters for each timepoint indicating a change in LSEC transcriptome after exposure to CCl₄ (Figure 4A). Interestingly, LSECs appear not to restore to the healthy LSEC cluster after CCl₄ induced injury, indicating that LSECs after 1 week CCl₄ have a different phenotype compared to healthy LSECs. Next, pathway analysis was performed on LSECs from CCl₄-treated livers compared to healthy LSECs (40). All enriched pathways were clustered in Cytoscape (Supplementary Figure 5) and graphically represented in Figure 4B. Shortly after the induction of acute liver injury, several pathways related to ROBO signaling and inflammation become significantly enriched (NES > 1, FDR <0.25). After 3 days, pathways involved in angiogenesis, ECM (extra cellular matrix) production and cell cycle are induced. Interestingly, a considerable amount of cell cycle pathways are strongly active after 3 days of CCl₄ but seem to become inactive again after 7 days. This was confirmed by the presence of Ki67⁺Lyve1⁺ positive LSECs in livers 3 days after CCl₄, indicating that indeed LSECs are proliferating at day 3, but not anymore after 7 days (Figure 4C). Pathways regarding ECM production were elevated after 3 days of CCl₄ and remained elevated after 7 days. NCAM signaling, important for the inhibition of fibroblast growth factor signaling (51), shows a similar trend. One of the dysregulated ECM genes is Collagen 4 which has been described to be produced by LSECs (52–54). Upon acute injury we indeed see an induction of Collagen 4 expression on day 3, which shows a sinusoidal pattern (Figure 4C).

To compare human with mouse LSEC dysfunction, we analyzed the expression of human LSEC gene sets (Figures 1, 2) in mouse LSECs after an acute CCl₄-induced liver injury. Genes that are enriched in healthy human LSECs show diverse expression patterns in mouse LSECs after acute liver injury (Figure 5A). Typical LSEC genes such as STAB2 and CLEC4G are downregulated upon liver injury, in contrast to genes such as LYVE1, CLEC1B, and CD36 which are upregulated at early timepoints, indicating that these genes cannot always discriminate healthy LSECs from damaged LSECs. Genes that were expressed higher in healthy LSECs were selected for the generation of a restricted healthy LSEC signature that should identify healthy LSECs in mice and human samples. This signature contains both novel (PLPP3, NTN4 and OIT3) and well-established (CLEC4G and STAB2) genes for LSECs which show a high expression in healthy human LSECs (Figure 5B). To generate also a more restricted gene signature that can specifically identify LSECs in damaged livers instead of both damaged LSECs and ECs, we first identified genes that were differentially expressed in the damaged LSEC population in comparison to healthy endothelial cells (Figure 5C). These differentially expressed genes were compared to previously identified enriched gene sets from damaged LSECs/ECs (Figure 2A) which resulted in a damaged LSEC signature that contained only four genes: Fabp4/5 and Vwf/a1 (Figure 5D). These four genes were all upregulated in CCl₄-induced liver injury after 1 day or 3 days. Moreover, these four genes are highly expressed in the damaged human LSEC population of the Ramachandran et al. (25) data set (Figure 5E). The expression of two healthy and damaged LSEC signature genes were validated using qPCR and confirmed the RNAseq data (Supplementary Figure 2A).

Next, we wanted to examine if scRNAseq data sets of LSECs from healthy and diseased livers can be identified as such with these two LSEC signatures. As samples can differ quite a lot between studies due to a different definition of healthy subjects, different isolation methods, different scRNAseq approaches, different etiologies and species we validated our newly generated LSECs signatures in 4 independent scRNAseq data sets of healthy human livers (21, 24) and healthy or diseased (NASH and fibrotic) mouse livers (26, 32) (Supplementary Figure 6). Using the LSEC signature we could show a higher gene signature score in LSEC-related populations in healthy human livers compared to all other cell populations (Figures 6A,B). Moreover, the damaged LSEC signature shows a low gene signature score in all liver cell types in healthy human livers except for a slightly higher gene signature score for periportal LSECs and (portal) ECs. These results confirm that quantification of LSEC signatures (scores) can be used to identify LSECs in scRNAseq data of human healthy livers. Unfortunately, we could not validate our signatures in a different scRNAseq data set of cirrhotic patients due to the lack of publicly available human data. Next, we validated the LSEC signatures using scRNAseq data of healthy and diseased (NASH and fibrotic) mouse livers (26, 32). In both data sets, the LSEC/EC populations from control livers have a high LSEC signature score, but is also still present (but lower) in NASH and fibrotic livers. More importantly, the damaged LSEC signature score is higher in LSEC/EC population from NASH livers, and to a lesser extend in CCl₄ livers, indicating that LSECs are damaged and can be identified in NASH and fibrotic livers using these 4 genes (Figures 6C,D). These findings demonstrate that the LSEC signatures can be used to identify and distinguish damaged LSECs from healthy LSECs in scRNAseq data of human and mouse livers.