Results

Blood flow-derived haemodynamic stimuli induce vascular Wnt expression

To identify zonation-enforcing biophysical parameters in the liver, we surveyed the literature for conditions that affect hepatic metabolic zonation. Intriguingly, pathological conditions that alter hepatic blood flow, such as heart failure, cirrhosis, or portosystemic shunt, were found to result in loss of glutamine synthetase (GS), a strictly pericentrally zonated metabolic enzyme that is tightly regulated by angiocrine Wnt25,26. Together with the finding that pericentral EC are generally more stretched and exposed to blood cells of higher velocity27,28, we consequently hypothesised that blood flow may be a major determinant of the LSEC transcriptional programme and, hence, a defining biophysical parameter for establishing hepatic zonation. To test this hypothesis, liver samples of mice suffering from congenital hepatic portosystemic shunts29, which led to chronically reduced blood flow to the liver, were analysed (Fig. 1a and Supplementary Fig. 1a). Compared to healthy controls, the hepatic zonation pattern of mice with shunts was significantly perturbed, as evidenced by downregulation of GS (Fig. 1b, c)26. A similar loss of hepatocytic GS expression was also observed for Wnt-deficient mutant mice4,9–11,21, which hinted towards the mechanosensitive expression of vascular Wnt.

Fig. 1: Blood flow-derived biomechanical stimuli modulate vascular Wnt signalling.a Illustration of a congenital portosystemic shunt. Aberrant vessel formation between the portal vein and vena cava hijacks blood flow routed to the liver. Created in BioRender. Lee, KH. (2026) https://BioRender.com/5olnoti. b Representative immunofluorescence image of glutamine synthetase (GS) in the liver of the shunt (n = 6) and control mouse (n = 8). Male mice were included in the study. Scale bars, 100 μm. c Percentage of GS positive area in the whole liver section (control, n = 8 mice; shunt, n = 6 mice). d–g Relative RT-qPCR gene expression analysis of Klf2, Angpt2, Wnt2 and Wnt9b in wild-type primary LSEC freshly isolated from each mouse were cultured under different shear conditions (static control, n = 6; low shear condition, n = 12; high shear condition, n = 10). LSEC were isolated from both female and male mice. h Experimental scheme of 2/3rd partial hepatectomy (PHx) model. Created in BioRender. Lee, KH. (2026) https://BioRender.com/f7leoqi. Early kinetics of haemodynamic signatures were captured during the liver regeneration process (upper panel). Liver to body weight ratio of wild-type male mouse after PHx (sham: n = 5, 3 h: n = 6, 6 h: n = 4, 12 h: n = 6, D-1: n = 6, D-2: n = 6, D-4: n = 6 mice; lower panel). i–l Temporal kinetics of Klf2, Angpt2, Wnt2 and Wnt9b expression during liver regeneration. Relative RT-qPCR gene expression analysis was performed on MACS-bead sorted LSEC post-PHx. () indicates fold difference. c–g Data are presented as mean ± s.e.m. d–g The gene expression was normalised to Pecam1 and relative fold changes were calculated in comparison to the static control. h–l Data are presented as mean ± s.d (background area). The gene expression was normalised to Pecam1 and relative fold changes were calculated in comparison to sham-operated controls (dotted line). c Two-tailed unpaired t-test was performed. d–g, h–l One-way ANOVA followed by Bonferroni test was performed.

To determine whether flow directly affected the expression of vascular Wnt, primary LSEC were isolated from wild-type adult murine livers and exposed to different shear conditions in vitro (Supplementary Fig. 1b). As proof of principle, the expression of shear-responsive endothelial genes Klf2 and Angpt2 was assessed30. As expected, Klf2 expression was significantly induced under shear conditions, whereas Angpt2 expression was repressed under high shear conditions (Fig. 1d, e). In line with our hypothesis, the expression and secretion of bioactive vascular Wnt2 and Wnt9b were induced in a flow-dependent manner under fluidic shear stress (Fig. 1f, g and Supplementary Fig. 1c). To further dissect the mechanism of flow-regulated Wnt regulation, a specialised in vitro fertilisation dish was used to discriminate the laminar shear and oscillatory shear response (Supplementary Fig. 1d). Predictably, oscillatory shear distinctly induced oscillatory shear-sensitive markers, Icam1 and Vcam1 (Supplementary Fig. 1e–h), whereas Klf2 and Klf4 expressions were more susceptible to laminar shear stress (Supplementary Fig. 1i–l). Both oscillatory shear stress as well as laminar shear stress induced the expression of Klf2 and Wnt2 under high shear conditions (Supplementary Fig. 1i–p). The expression of Wnt9b, however, remained unchanged under either condition. These results indicated that vascular Wnt2 expression was directly regulated by oscillatory or laminar shear stress (Supplementary Fig. 1e–p). In parallel, the findings were further corroborated by parallel flow assays (Supplementary Fig. 1q). Upon activation of high laminar flow and high oscillatory flow, a distinct shear response was detected. As expected, expression of Klf2 and Klf4 was upregulated at high laminar flow, whereas expression of Icam1 and Vcam1 was induced at high oscillatory flow (Supplementary Fig. 1r, s). In line with our previous result, Wnt2 expression was induced upon high laminar flow and more profoundly at high oscillatory flow conditions (Supplementary Fig. 1t). Hence, Wnt2 expression was more sensitive towards oscillating shear, confirming that vascular Wnt expression is sensitive towards oscillating shear stress31,32. Overall, these results corroborate that the expression of vascular Wnt factors was induced by the haemodynamic response in vitro.

To confirm these findings in vivo, we surgically pursued a model of increased hepatic blood perfusion by employing 2/3rd partial hepatectomy (PHx). This procedure results in the overloading of the regenerating lobe with excessive blood, thereby inducing high shear conditions (Fig. 1h). Consistent with previous reports33–35, a transient upregulation of the shear-response genes Klf2, Klf4, Nos3, Gja1 and Gja4 (Fig. 1i and Supplementary Fig. 2a–d) and a downregulation of growth inhibitory angiocrine factor Angpt2 (Fig. 1j) was observed during the early inductive phase of liver regeneration7. Notably, Wnt2 and Wnt9b expression was transiently increased upon the induction of shear-responsive genes (Fig. 1k, l). In line with these findings, shear-sensitive Klf2 was upregulated immediately after the shear induction, whereas Wnt2 expression was induced only 12 h after the high laminar shear in vitro (Supplementary Fig. 2e, f). Concurrently, the transient activation of mechanosensitive genes coincides with a temporal increase in blood perfusion, vessel area, and vessel volume during early stages of PHx36. To exclude surgery-related effects, hepatic blood flow was also induced in vivo by treating mice pharmacologically with a vasodilating agent, Riociguat (Supplementary Fig. 2g)37. In line with the surgical approach, vasodilation stimulated the haemodynamic response in liver EC and consequently induced vascular Wnt expression (Supplementary Fig. 2h–k) along with a dilated vasculature and upregulated mechanosensitive ITGB1 and VEGFR3 phosphorylation (Supplementary Fig. 2l–q). To further validate these findings, we employed an ex vivo perfusion model that allowed for the direct modulation of hepatic flow (Supplementary Fig. 2r)23,38. In line with the previous results, Klf2, Klf4, and Wnt2 showed marked upregulation under accelerated flow (Supplementary Fig. 2s–v). Taken together, these data establish that biomechanical activation of hepatic EC triggers vascular Wnt expression.

Hypoxia and oxidative stress do not induce vascular Wnt expression

Besides blood flow, other biophysical parameters show similar zonation gradients in the liver. For instance, oxygen availability is strictly zonated alongside the porto-central liver axis, with the pericentral region being a highly hypoxic microenvironment39. To investigate the effect of hypoxia on LSEC zonation, freshly isolated murine primary LSEC were incubated under hypoxic and atmospheric oxygen conditions in vitro. As expected, Slc2a1 expression was upregulated in response to the hypoxic challenge (Supplementary Fig. 3a). Hypoxia did not induce pericentrally zonated Wnt factors, but rather seemed to repress Wnt2 expression, whereas no significant effect was observed for Wnt9b (Supplementary Fig. 3b, c). Chemical stabilisation of hypoxia-inducible factor (HIF) using cobalt chloride (CoCl2) further validated this finding as Slc2a1 expression was induced, validating the hypoxic response, whereas Wnt2 and Wnt9b showed markedly reduced expression upon CoCl2 treatment (Supplementary Fig. 3d–f). Similarly, exposing mice to seven days of normobaric hypoxic challenge40, as shown by marked induction of Hif1a, Epas1, and Slc2a1 expression, had no effect on hepatic vascular Wnt expression or hepatic metabolic zonation (Supplementary Fig. 3g–m). Concordantly, induction of acute hypoxic response by short-term inhibition of HIF prolyl-hydroxylase in vivo did not induce vascular Wnt expression nor alterations in pericentral hepatocyte identity (Supplementary Fig. 3n–q). Extending the treatment period to one week did not alter vascular Wnt expression (Supplementary Fig. 3r–u). Taken together, these data excluded a prominent role of hypoxia as a modulator of vascular Wnt and hepatic zonation.

As the cytochrome enzymatic activity of hepatocytes is restricted to the pericentral region16,39, we tested whether the resulting reactive oxygen species (ROS) may contribute to the zonated vascular Wnt expression profile. For this purpose, mice were treated with the antioxidant agent, N-acetylcysteine (NAC), to reduce hepatic oxidative stress. Similar to hypoxia, relieving oxidative stress did not affect vascular Wnt expression or metabolic hepatocyte zonation (Supplementary Fig. 3v–y). Collectively, these data indicate that blood flow-mediated haemodynamic factors and not hypoxia or oxidative stress modulate vascular Wnt expression.

Vascular Wnt factors sustain liver homeostasis and regeneration

To unravel the physiological consequences of vascular Wnt signalling in sustaining the spatial division of hepatocytes and LSEC, we established an EC-specific inducible knockout mouse model of Evi/Wls, a cargo receptor for Wnt ligands, and the Wnt potentiator Rspo3, as well as the Evi/Wls and Rspo3 double knockout model (Double-iECKO). Comparing the different mouse strains revealed synergistic effects of Wnt and Rspo3 signalling in maintaining metabolic liver zonation, as Double-iECKO mice exhibited an aggravated loss of GS compared to the single knockouts (Fig. 2a, b). In contrast, the loss of vascular Wnt and Rspo3 showed divergent effects on CYP2E1 expression. Whereas Evi/Wls and Double-iECKO mice showed a marked reduction of CYP2E1, Rspo3-iECKO mice did not display loss of CYP2E1 (Fig. 2c, d). This may be explained by the overall abundance of vascular Wnt factors, with Wnt2 in particular being more widely expressed throughout the liver compared to angiocrine Rspo3. As Wnt ligands and Rspo3 are short-range acting molecules, the difference in abundance could compensate for the loss of Rspo3 in Rspo3-iECKO mice41 and by other cellular sources such as hepatic stellate cells via pericrine Rspo3 signalling42.

Fig. 2: Endothelial-derived Wnt ligands and Rspo3 sustain the spatial division of labour in the liver.a Representative immunofluorescence image of GS (red) in the liver. The genotypes are indicated below. Scale bars, 100 μm. b Percentage of GS positive area in the whole liver section (control, n = 10; Evi/Wls iECKO, n = 9; Rspo3 iECKO, n = 6; Double-iECKO, n = 11 mice). c Representative immunofluorescence image of CYP2E1 (red) in the liver. The genotypes are indicated below. Vessels are highlighted by COLIV (grey) staining, cell boundaries are marked by Phalloidin staining (cyan), and cell nuclei are visualised with DAPI (blue). Scale bars, 100 μm. d Percentage of CYP2E1 positive area in the whole liver section (control, n = 7; Evi/Wls iECKO, n = 5; Rspo3 iECKO, n = 4; Double-iECKO, n = 6 mice). e Uniform Manifold Approximation Projection (UMAP) of the total hepatocyte population from control (left panel, n = 3) and Double-iECKO mice (middle panel, n = 4). Cells are coloured according to central hepatocyte signature gene scores. Log-normalised density plot of the average central hepatocyte gene score calculated for control hepatocytes (blue line) and Double-iECKO hepatocytes (red line; right panel). f UMAP of the same dataset, with cells coloured according to their portal hepatocyte gene score (left and middle panel). Log-normalised density plot of the average portal hepatocyte gene score calculated for control (blue line) and Double-iECKO hepatocytes (red line; right panel). g Gene set enrichment analysis (GSEA) of the total hepatocyte population comparing Double-iECKO and littermate controls. Differential genes were identified using DESeq2 comparing pseudobulks reflecting the biological replicates across genotypes. A two-tailed Wald test and Benjamini–Hochberg correction were used. Threshold for significance was set to p < 0.05 and q < 0.1. Normalised enrichment score is shown for each GO term. The red bar shows upregulated, and the blue bar represents downregulated GO terms. h Log-normalised density plot of average gene score calculated for the detoxification pathway. i Limma-normalised GSEA of proteomics data followed by two-tailed Student’s t-test and Benjamini–Hochberg correction comparing central hepatocytes of Double-iECKO (n = 3) and control (n = 3). Colour refers to the adjusted p-value, and dot size represents the number of proteins. +/− sign represents downregulated or upregulated GO, respectively. () indicates fold difference. a–i Both female and male mice were considered in the study. b, d Data are presented as mean ± s.e.m. b, d–f, h Average gene scores from individual mice were acquired and a two-tailed unpaired t-test comparing control and Double-iECKO was performed.

To assess the functional consequences of the hepatic zonal defect, semi-targeted metabolic screening of whole liver lysates and plasma samples from Double-iECKO and littermate controls was performed. Notably, accumulation of ammonium, potassium, and sugar molecules, including fructose and hexose caused by loss of GS activity and perturbed pericentral glycolysis, respectively, was observed in the liver (Supplementary Fig. 4a–d). Systemically, proline, alanine, oxalic acid, and β-alanine were found to be accumulated, hinting towards periportal dominance with an increased portal hepatocyte-modulated amino acid catabolism (Supplementary Fig. 4e–h). In parallel, Wnt gain of function experiments showed a reversed phenotype with reduced amino acid abundance confirming Wnt-dependent metabolic functions43.

Long-term tracing of Double-iECKO mice revealed that loss of Wnt signalling was not compensated for, as evidenced by the persistent loss of GS that was additionally accompanied by an overall reduction in liver mass (Supplementary Fig. 4i–k). It was recently reported that hepatocyte progenitor cells exist across the porto-central liver axis and are marked by distinct gene expression profiles depending on their spatial location44–50. To elucidate which progenitor pool was affected and consequently led to the reduced liver mass in Wnt-deficient mice, the gene expression of hepatic progenitor markers was assessed. Interestingly, gene expression of the periportal progenitor marker genes Sox9 and Mfsd2a (Supplementary Fig. 4l, m)44,45, and the midzonal progenitor marker genes Igfbp2 and _Hamp2_46 were not altered between Double-iECKO and control mice (Supplementary Fig. 4n, o), whereas pericentral progenitor marker genes Axin2 and _Lrg5_21,51 showed a downregulation upon vascular Wnt loss (Supplementary Fig. 4p, q). This suggested that vascular Wnt is crucial in maintaining the pericentral progenitor niche and that pericentral progenitor hepatocytes are essential for the maintenance of liver mass.

As angiocrine Wnt was established to exert crucial maintenance functions under homeostatic conditions, we performed 2/3rd PHx on Double-iECKO and littermate control mice to test the importance of angiocrine Wnt during liver regeneration. In line with the previous findings9,11, Double-iECKO mice showed impeded liver regeneration characterised by a reduced liver-to-body weight ratio and fewer proliferating hepatocytes during the inductive phase of liver regeneration (Supplementary Fig. 5a–d). Moreover, a similar reduction of liver mass was observed during the angiogenic phase of liver regeneration (Supplementary Fig. 5e), indicating that hepatocyte proliferation was not simply attenuated but reduced in general (Supplementary Fig. 5f–h). This reduced liver regeneration capacity was mainly caused by the loss of pericentral hepatic progenitors (Supplementary Fig. 5i–n). To further corroborate this result, the number of proliferating hepatocytes was quantified in a zone-specific manner using GS expression and bile ducts as anchors for each zone (Supplementary Fig. 5o). In line with a previous report42, midzonal hepatocytes dictated hepatic regeneration. Consequently, fewer proliferating hepatocytes were observed in midzonal and central zones upon the loss of angiocrine Wnt factors, whereas the portal zone remained unaffected, suggesting Wnt-dependency of midzonal and central hepatocytes during liver regeneration (Supplementary Fig. 5p–r). Overall, these data denote that vascular Wnt factors serve as the guardians of liver zonation, size, and function.

Vascular Wnt factors orchestrate the spatial division of labour in the liver

To molecularly define the mechanisms by which vascular Wnt signalling is maintaining liver homeostasis, we performed scRNA-seq using the CEL-seq2 protocol52 on FACS-sorted hepatocytes and LSEC of Double-iECKO and littermate control mice (Supplementary Fig. 6a, b). The samples with successful gene deletion, as evidenced by immunofluorescence staining of GS, were further processed (Supplementary Fig. 6c). To validate the genetic recombination, pseudobulk differential gene expression (DEG) analyses were performed on hepatocyte and LSEC datasets. In line with the loss of GS, Glul expression was downregulated in hepatocytes (Supplementary Data 1 and Supplementary Fig. 6d). Additionally, targeted genes, including Rspo3 and Wls expression, were downregulated in LSEC dataset, confirming the robust gene deletion (Supplementary Data 1 and Supplementary Fig. 6e). Consequently, the scRNA-seq data were further processed and the data were visualised in a Uniform Manifold Approximation and Projection (UMAP). The hepatocyte and LSEC data clustered into three subgroups, which based on previously established marker gene sets, could be assigned to the portal, midzonal, or central area of the liver lobule (Supplementary Fig. 6f–i).

Next, the scRNA-seq data of LSEC and hepatocytes from Double-iECKO and control mice were spatially resolved by aligning the transcriptomes alongside a pseudo porto-central axis8,17,22,53 and the expression of previously published zonation marker genes was assessed16,17,22. Loss of vascular Wnt did not affect the zonal expression pattern of the pericentral hepatocyte marker Nt5e or the periportal hepatocyte marker _Cdh1_53. Moreover, the previously reported gradual expression of _Kit_8,17,22 was maintained in LSEC of Double-iECKO mice (Supplementary Fig. 6j–l). To orthogonally validate the sequencing data, hepatocytes and LSEC from Double-iECKO and control mice were spatially sorted as bulk samples and proteomic analysis was performed. To this end, hepatocytes were discriminated on the basis of CD73 and CD324 expression in FACS and four equally sized fractions were sorted, reflecting the portal, periportal, pericentral, and central hepatocytes (Supplementary Fig. 6m). Similarly, four equally sized fractions of the LSEC samples were sorted according to c-Kit expression, with c-Kit high cells being located close to the central vein and c-Kit low cells near the portal vein (Supplementary Fig. 6n).

In line with previous studies9,12,4,10,21, the data corroborated that the metabolic fate of hepatocytes was strongly dependent on vascular Wnt factors. In general, vascular Wnt factors were dispensable for midzonal hepatocyte marker genes (Supplementary Fig. 6o). In parallel, hepatocytes isolated from Double-iECKO mice showed a loss of the pericentral-like hepatocyte gene signature (Fig. 2e, Supplementary Fig. 7a, b and Supplementary Data 1 and 2), while gaining periportal-like hepatocyte identity (Fig. 2f and Supplementary Fig. 7c, d), which was also observed at the proteome level (Supplementary Fig. 7b, d and Supplementary Data 3). Gene set enrichment analysis (GSEA) on DEG of Double-iECKO and littermate control hepatocyte pseudobulks reflecting the biological replicates highlighted perturbed metabolic functions, such as detoxification and lipid metabolism, in Wnt-deficient mice (Fig. 2g, h and Supplementary Fig. 7e, f). To further examine the functional role of vascular Wnt in a spatial context, the total hepatocyte population was split into three different subsets (Supplementary Fig. 6f, h). Subsequent GSEA on DEG comparing the Double-iECKO and control spatial pseudobulks further substantiated the loss of pericentral and a shift towards periportal hepatocyte identity in Wnt-deficient mice.

In the periportal zone, GO terms related to pericentral hepatocyte functions including lipogenesis and lipid phosphorylation were further downregulated in Double-iECKO pseudobulks, whereas portal hepatocyte functions including gluconeogenesis were found to be enriched (Supplementary Fig. 7g). This shift was also observed for midzonal hepatocytes, which displayed increased periportal-specific amino acid catabolism and reduced pericentral xenobiotic functions (Supplementary Fig. 7h), as well as for pericentral hepatocytes, which showed strong periportal hepatocyte-like features, while losing their pericentral identity (Supplementary Fig. 7i). Validating the transcriptomic approach, GSEA on differentially expressed proteins comparing portal and central hepatocytes from control mice similarly revealed the central bias of xenobiotic catabolic processes and the portal bias of urea cycle (Supplementary Fig. 7j).

Next, central hepatocytes of Double-iECKO and control mice were compared. In line with the transcriptomic data, loss of vascular Wnt was found to induce a metabolic shift as indicated by the reduction in xenobiotic functions and the acquisition of amino acid catabolic functions in mutant central hepatocytes (Fig. 2i). This is in line with previous work that highlighted a vascular Wnt-regulated hepatic xenobiotic core in toxication studies using carbon tetrachloride and acetaminophen10,11. Collectively, these data indicate that angiocrine Wnt signalling does not only shape the pericentral hepatocyte niche but is crucial in maintaining overall hepatocyte function and, hence, hepatic performance.

Vascular Wnt factors shape LSEC function in an autocrine manner

Analyses of the LSEC dataset revealed that the overall LSEC zonation pattern4,17 was maintained in Double-iECKO mice (Fig. 3a, b and Supplementary Fig. 8a–d); yet, loss of vascular Wnt led to functional perturbations. Double-iECKO LSEC showed a reduction in canonical Wnt signalling, metabolic programmes, such as nitric oxide (NO) and ROS metabolism, as well as vasodilation (Supplementary Fig. 8e). Indeed, liver tissues from Double-iECKO mice showed less eNOS expression compared to their control (Supplementary Fig. 8f, g). Similar to the hepatocyte dataset, LSEC were split into three subgroups based on the expression of zonal identifier genes reflecting portal, midzonal and central LSEC (Supplementary Fig. 6g, i). Zonal pseudobulks reflecting the biological replicates were then compared between Double-iECKO and control samples and GSEA was performed. Interestingly, loss of Wnt led to the disturbed regulation of blood pressure, as well as NO metabolism in portal venous and midzonal LSEC (Supplementary Fig. 8h, i), whereas pericentral LSEC displayed loss of canonical Wnt signalling and a mitigated response to various milieu factors including shear stress, hypoxia, antioxidation, mechanotransduction and calcium ions (Fig. 3c).

Fig. 3: Molecular atlas of LSEC reveals spatial division of Wnt responsiveness and consequent autocrine Wnt function.a UMAP of total LSEC population from control (left panel, n = 3) and Double-iECKO mice (middle panel, n = 3). Cells are coloured by central LSEC signature gene scores. Log-normalised density plot of average gene scores calculated for central LSEC signatures (right panel). b UMAP of the same data with cells coloured according to portal LSEC gene score. Log-normalised density plot of average gene score calculated for portal LSEC (right panel). c GSEA of central vein EC population comparing Double-iECKO and littermate control. Differential genes were identified using DESeq2, comparing pseudobulks reflecting the biological replicates of the different genotypes. Wald test and Benjamini–Hochberg were used. Threshold for significance was set to p < 0.05 and q < 0.4. Normalised enrichment score is shown for each GO term. The red bar shows upregulated, and the blue bar represents downregulated GO. d Violin plot of average gene score calculated for shear stress associated genes, and e for gap junction molecules comparing control (n = 3) from Double-iECKO mice (n = 3). For violin plots, the horizontal line represents the median value, the lower and upper quartiles mark the 25th and 75th percentiles, and the whiskers indicate the minimum and maximum values. f–h Spatial expression of Wnt receptors, Fzd4 and Lrp6, and Wnt suppressor Apc. Zonation score was calculated based on the previously identified reference genes. Each line represents one biological replicate. Colour indicates the genotype. Shadowed area indicates s.e.m. i Representative immunofluorescent image of LRP6 (red) co-stained with the hepatic zonal marker GS (grey) and VE-Cadherin (cyan) for vasculature (left panel). Scale bar shows 20 µm. The asterisk indicates the VE-Cadherin and LRP6 double-positive area. LRP6 expression was further highlighted in grey (right panel). j Percentage of COLIV and LRP6 double-positive area in the liver lobule (n = 6). k Representative immunofluorescent image of APC (red) co-stained with hepatic zonal marker GS (grey) and VE-Cadherin (cyan) for vasculature (left panel). Scale bar shows 10 µm. The asterisk indicates the VE-Cadherin and APC double-positive area. APC expression was further highlighted in grey (right panel). l Percentage of COLIV and APC double positive area in the liver lobule (n = 6). m Model of vascular Wnt zonation. CV central vein, PV portal vein. () indicates fold difference. a–l Both female and male mice were incorporated in the study. j, l Data are presented as mean ± s.e.m. a, b, d, e Average gene scores from individual mice were computed and subjected to statistical analysis. a, b, d, e, j, l Two-tailed unpaired t-test was performed.

We further validated the data by calculating the average gene scores for genes which are associated with shear stress and gap junctions, and found all genesets to be downregulated in LSEC upon loss of Wnt (Fig. 3d, e). To orthogonally validate the data, GSEA was performed on differentially expressed proteins from the central vein LSEC subpopulation comparing Double-iECKO and littermate control mice. In line with the transcriptomic approach, the analysis revealed a perturbed calcium response in mutant LSEC (Supplementary Fig. 8j). Collectively, the data revealed the existence of an autocrine vascular Wnt function in LSEC, supporting the view of pericentral hepatocytes not being the main vascular Wnt-addicted hepatic cell type.

To further investigate the autocrine vascular Wnt function, the zonal expression pattern of Wnt receptors and the Wnt-processing machinery in the hepatic vasculature were assessed. The expression of Wnt signalling receptors Lrp6 and Fzd4 was biased towards the central vein, whereas _Apc_54, a Wnt suppressor, was enriched at the portal vein (Fig. 3f–h). Immunostaining of LRP6, FZD4 and APC confirmed that LRP6 protein expression was zonated towards the central vein, whereas FZD4 was evenly expressed across the liver axis (Fig. 3i, j and Supplementary Fig. 8k, l). APC was preferably localised around the portal vein (Fig. 3k, l), suggesting that pericentral LSEC were poised for canonical Wnt signalling, whereas periportal LSEC were sealed through the lack of receptor expression and the default suppression of the pathway.

To further validate our findings, we performed comprehensive analyses on publicly available data and confirmed the vascular Wnt zonation55 (Supplementary Fig. 8m). As Lrp6 and Fzd4 showed a similar zonation pattern to vascular Wnt factors, we tested whether their expression was also regulated by flow-induced shear stress. Applying shear stress to freshly isolated primary LSEC markedly induced the expression of Lrp6 and Fzd4, pointing towards a similar gene expression regulation of Wnt ligands and Wnt receptors in LSEC in vitro (Supplementary Fig. 8n, o). Taken together, these data suggest an autocrine Wnt signature within the hepatic endothelium whose activity is biased towards the central vein (Fig. 3m).

Vascular Wnt factors fine-tune angiocrine gene signatures and LSEC morphology

To examine the functional consequences of the autocrine Wnt factors, we applied shear stress to primary LSEC isolated from Double-iECKO and littermate control mice (Fig. 4a). For this purpose, freshly isolated LSEC were cultured and genetic recombination was induced in vitro (Supplementary Fig. 9a, b). Upon shear activation, Double-iECKO LSEC showed less induction of the shear response genes Klf2 and Klf4 (Fig. 4b and Supplementary Fig. 9c). Additionally, Wnt2 induction was similarly impeded (Fig. 4c). To validate this finding in vivo, 1/3rd PHx was performed on Double-iECKO and control mice and the early haemodynamic response of LSEC was traced (Supplementary Fig. 9d). In line with the in vitro data, Double-iECKO LSEC showed less induction of shear response genes compared to sham-operated mice (Fig. 4d, e). Similarly, pseudobulk DGE analysis of total LSEC showed reduced gene expression of Klf2 and Klf4 (Supplementary Data 1 and Fig. 4f–h).

Fig. 4: Vascular Wnt factors coordinate LSEC orientation and angiocrine profiles.a Experimental scheme of in vitro shear stress model. Primary LSEC were isolated via liver perfusion and MACS- bead sorting. After seeding in 6-well plates, genetic recombination was induced by administering 2 μM of hydroxy-tamoxifen. LSEC were then applied to either static (0) or high shear conditions (14 dynes/cm2). Created in BioRender. Lee, KH. (2026) https://BioRender.com/dtxlnfw. b Relative gene expression of shear sensing Klf2 and c Wnt2 in control and Double-iECKO LSEC cultured under different shear conditions (static control, n = 9 mice; shear control, n = 8 mice; static Double-iECKO, n = 8 mice, shear Double-iECKO, n = 7 mice). d Relative gene expression of shear sensing Klf2 and e Klf4 6 h post 1/3rd partial hepatectomy (PHx) (sham, n = 6 mice; control, n = 4 mice; Double-iECKO, n = 4 mice) from MACS-bead sorted LSEC. f Dot plot of the total LSEC dataset comparing the control to the mutant samples. Gene expression of Wls, Rspo3, Klf2, Klf4, Gja1 and Gja4 were visualized. Average gene expression, in z-score, is shown in the colour code and the dot size illustrates the percentage of EC that express the target genes. Violin plot from the total LSEC scRNA-seq dataset comparing the control to the mutant samples; g Klf2, h Klf4 mRNA expression. i Relative gene expression of angiocrine factors Wnt2 and j Angpt2 24 h post 2/3rd PHx (sham, n = 3 mice; control, n = 4 mice; Double-iECKO, n = 6 mice) from MACS-bead sorted LSEC. k Phalloidin staining with subsequent in silico segmentation was performed to visualise shear stress fibres comparing static and high shear conditions of primary Double-iECKO and control LSEC. Colour code represents fibre orientation relative to the shear axis (0°). Scale bars, 5 µm (static control, n = 9 mice; shear control, n = 8 mice; static Double-iECKO, n = 8 mice, shear Double-iECKO, n = 7 mice). l Rose plot represents cell alignment angle normalised to the direction of shear (0°). m Quantification of cell alignment angle relative to the direction of the shear (0°). Each dot represents a single image. () indicates fold difference. b–m Both female and male mice were incorporated in the study. b–e, i, j Gene expression was normalised to Pecam1. d, e, i, j, the relative fold change was calculated with respect to sham-operated mice. b, c Data are presented as mean ± s.e.m. d, e, i, j Data are presented as mean ± s.d. d, e, i, j One-way ANOVA followed by Bonferroni test was performed. b, c, j, m Two-tailed unpaired t-test was performed.

Next, Double-iECKO and control mice were subjected to 2/3rd PHx to assess the instructive angiocrine expression profile, which reaches its peak 24 h after surgery (Supplementary Fig. 9e). Wnt2 induction was impaired in mutant LSEC (Fig. 4i). Furthermore, Angpt2 expression was dysregulated during the induction phase of liver regeneration (Fig. 4j). In line with this, Angpt2 expression was retained in isolated Double-iECKO primary LSEC upon high shear activation in vitro (Supplementary Fig. 9f).

We next sought to decipher the macroscopic impact of the disturbed molecular shear sensors in Double-iECKO LSEC and analysed the orientation of F-actin filaments and traced the morphological changes upon shear stimulation (Fig. 4k). Typically, under laminar flow, EC, including LSEC align in direction to the flow by rearranging their cytoskeleton. While control LSEC were able to align in response to flow, Double-iECKO LSEC displayed poor alignment (Fig. 4l, m), which was further confirmed by a pump-based flow experiment (Supplementary Fig. 9g, h). To assess whether Wnt-deficient LSEC in vivo showed similar morphological abnormalities, high-resolution electron microscopy was used to examine the morphological features of hepatic EC in liver sections of Double-iECKO and littermate control mice. Strikingly, mutant central venous EC showed structural defects as evidenced by thickening and detachment of the basement membrane, while overall mimicking pseudo-capillarisation through the loss of fenestration (Supplementary Fig. 9i). Intriguingly, central vein EC from Double-iECKO were found to be significantly smaller compared to their control and the lumen size of the hepatic vasculature was reduced in Double-iECKO mice, which implicates autocrine Wnt functions in shaping LSEC size (Supplementary Fig. 9j–r). In summary, Wnt signalling was found to be crucial for LSEC function by regulating cell morphology and fine-tuning the expression dynamics of angiocrine gene signatures.

Endothelial autocrine Wnt controls gap junction expression

During the process of mechanotransduction, flow stimulates mechanoreceptors and induces mechanosensing ion channel activity, thereby mediating the calcium-directed downstream pathway56. The single-cell data revealed connexin 37 (Cx37, Gja4) and connexin 43 (Cx43, Gja1) as highly Wnt-dependent effector genes that modulate ion current and calcium influx57. Additionally, several genes, including Arhgap29, Il1a, Nfkbiz, Cacng2, Psmd2 and Slco3a1, were identified as Wnt-regulated genes (Supplementary Data 2). Loss of vascular Wnt led to a significant reduction of Cx37 and Cx43 expression in vivo (Supplementary Data 1, Fig. 4f and Supplementary Fig. 10a–f) and in vitro (Supplementary Fig. 10g, h). To further dissect the zone-dependent function of vascular Wnt signalling, spatial coordination of gap junctional molecules was examined. In situ hybridisation of Gja1 revealed a reduction of mRNA expression at the central vein upon vascular Wnt loss, whereas Gja1 expression did not differ in the portal vein (Fig. 5a, b and Supplementary Fig. 10i, j). Similarly, GJA4 protein expression was markedly downregulated in the central vein of the vascular Wnt-deficient liver, whereas GJA4 abundance was unaffected in the portal vein (Fig. 5c, d and Supplementary Fig. 10k, l). Remarkably, similar to vascular Wnt factors, the expression of both gap junction molecules showed regulation by flow-induced shear stress and a spatial expression pattern with central bias (Supplementary Fig. 2c, d and Supplementary Fig. 10c, d). Overall, these results show that vascular Wnt is short-ranged and promiscuous, modulating spatially distinct gap junctional molecule expression in an autocrine manner.

Fig. 5: Vascular Wnt factors fine-tune gap junction expression.a Representative in situ hybridisation image of Gja1 (red) on DAPI (blue) co-stained with hepatic zonal marker GS (cyan) and VE-Cadherin (white) for vasculature at the central vein. Gja1 expression was further highlighted in red (right panel). Gja1 expression between the mutant (n = 5) and littermate control (n = 5) was compared. Asterisks indicate the Gja1, VE-Cadherin and DAPI triple-positive area. b Corresponding quantification of Gja1, VE-Cadherin and DAPI triple-positive area on liver sections of control mice and mutant mice. c Representative immunofluorescence image of GJA4 (red) co-stained with hepatic zonal marker GS (cyan) and VE-Cadherin (white) for vasculature at the central vein. GJA4 expression was further highlighted in red (right panel). GJA4 expression between the mutant (n = 6) and littermate control (n = 6) was compared. Asterisks indicate the GJA4 and VE-Cadherin double-positive area. d Corresponding quantification of GJA4 and VE-Cadherin double-positive area on liver sections of control mice and mutant mice. e Experimental scheme for pharmacological manipulation of Wnt signalling pathway. Created in BioRender. Lee, KH. (2026) https://BioRender.com/pahnmbj. f Relative gene expression of gap junctional molecule Gja4 in wild-type primary LSEC after LGK974 (n = 9) or g CHIR99021 treatment (n = 8). h Relative gene expression of gap junctional molecule Gja1 in wild-type primary LSEC after stimulation with different WNT5A concentrations (n = 6). i Immunoblotting of GJA1 and Tubulin (left) and quantification (right) and j immunoblotting of GJA4 and Tubulin (left) and quantification (right) comparing WNT9B treated primary LSEC (n = 3) and its control (n = 3). k Immunoblotting of GJA1 and Tubulin (left) and quantification (right) and l immunoblotting of GJA4 and Tubulin (left) and quantification (right) of comparing WNT2 treated (n = 3) and its control (n = 3). m Live cell calcium imaging of wild-type primary LSEC treated with either water (n = 5) or gap junction channel blocker Carbenoxolone (CBX, n = 6), positive for calcium indicator, Cal-520, after calcium administration (grey background). The fluorescence intensity was normalised to the resting state. n Maximum normalised fluorescence intensity ratio of (m). o Live cell calcium imaging of Double-iECKO (n = 6) and control LSEC (n = 6) positive for calcium indicator, Cal-520, after calcium administration (grey background). The fluorescence intensity was normalised to the resting state. p Maximum normalised fluorescence intensity ratio of (o). () indicates fold difference. a, c Scale bar shows 10 µm. CV indicates the central vein area. a–d, f–p Both female and male mice were incorporated in the study. h–l Data are presented as mean ± s.d. b, d, f, g, m–p Data are presented as mean ± s.e.m. b, d, f, g, i–l, n, p Two-tailed unpaired t-test was performed. h One-way ANOVA followed by the Bonferroni test was performed.

To demonstrate the vascular Wnt control of gap junction expression, freshly isolated primary LSEC were either treated with Wnt antagonist (LGK974) or Wnt agonist (CHIR99021) in vitro. A reduction of Axin2 expression under antagonistic and induction of Axin2 under agonistic conditions validated the experimental setup (Fig. 5e and Supplementary Fig. 11a, b). As hypothesised, Gja4 expression was significantly decreased upon LGK974 treatment (Fig. 5f), whereas agonistic treatment with CHIR99021 induced Gja4 expression (Fig. 5g). In contrast, Gja1 was not regulated by canonical Wnt signalling, which led us to explore the role of non-canonical Wnt ligands in regulating gap junction expression (Supplementary Fig. 11c, d). Strikingly, WNT5A, a prototypic non-canonical Wnt ligand, triggered Gja1 and Gja4 expression (Fig. 5h and Supplementary Fig. 11e). Likewise, recombinant WNT2 and WNT9B, mimicking the pericentral Wnt-milieu of LSECs, induced protein levels of GJA1, whereas GJA4 levels did not change significantly (Fig. 5i–l). Furthermore, consistent with the transcriptomic data, LGK974-directed Wnt inhibition led to downregulation of GJA4 (Supplementary Fig. 11f, g). Concurrently, CHIR99021-derived Wnt activation induced GJA4 expression, whereas GJA1 expression was only regulated upon Wnt activation (Supplementary Fig. 11h–k). This indicated the fine tuning of gap junctional expression via both, the canonical and non-canonical Wnt signalling pathways in the liver and further provided evidence for an autocrine vascular Wnt function in LSEC that is regulating the gap junction expression.

We next examined the functional relevance of junctional molecules by performing live cell calcium imaging on primary LSEC from Double-iECKO mice and littermate controls in vitro. Treating primary LSEC with the gap junction blocker carbenoxolone inhibited channel activity, as evidenced by reduced calcium flux, thereby confirming the regulatory properties of gap junctional molecules in LSEC (Fig. 5m, n). Calcium tracing in mutant LSEC revealed perturbed channel function and calcium signalling in both steady and shear conditions (Fig. 5o, p and Supplementary Fig. 11l, m), which could be mimicked by Wnt antagonistic treatment of wild-type LSEC with LGK974 (Supplementary Fig. 11n, o). Overall, these data demonstrated perturbed gap junction-mediated calcium propagation upon vascular Wnt loss in vitro, highlighting vascular Wnt control of gap junctional molecules.