YAP Orchestrates Heterotypic Endothelial Cell Communication via HGF/c-MET Signaling in Liver Tumorigenesis

The Hippo signaling pathway negatively controls the subcellular localization and activity of the transcriptional coactivator yes-associated protein (YAP). Accordingly, inactivation of Hippo pathway constituents or nuclear enrichment of YAP has been associated with tumor formation as exemplified for many solid cancer entities such as hepatocellular carcinoma (HCC; ref. 1).

So far, different tumor cell intrinsic mechanisms have been described for YAP that contribute to liver overgrowth and malignant transformation. These include dedifferentiation of hepatocytes associated with elevated proliferation and accumulation of oncogenic mutations (2, 3). In addition, the Hippo/YAP axis transcriptionally controls the expression of genes critically involved in replication, chromosome segregation, as well as DNA repair and dysregulation of these factors increases the risk of chromosomal instability (4–6). However, recent data suggested that next to tumor cell autonomous effector mechanisms, aberrant YAP activation in malignantly transformed cells influences immune cells to create a tumor-supportive environment. For example, YAP affects recruitment and polarization of myeloid-derived suppressor cells and immune evasion in an HCC cancer model (7). This data point to a connection between YAP activation in tumor cells and immune cells or nonmalignant cells (e.g., nonparenchymal liver-resident cells) via paracrine-acting growth factors, cytokines, and chemokines. In particular, how YAP controls tumor cell extrinsic conditions supporting tumor growth and tumor progression through the induction of a proangiogenic response in endothelial cells (EC) is not well understood.

The liver vascular network consists of phenotypically different ECs (8–10). Organotypic liver sinusoidal endothelial cells (LSEC) connect the vessels of the portal field with the central vein (porto-central axis) and are characterized by morphological features such as sieve plate fenestrations and the absence of a basement membrane. In contrast, portal and pericentral vessels are lined by a continuous endothelium (continuous endothelial cells, CEC), which is covered by an additional layer of smooth muscle cells in the portal field to withstand higher pressure and pulsatile flow. Furthermore, LSECs and CECs differ in their endocytotic capacity, antigen presentation (11), and marker gene expression (e.g., LSEC-specific Lyve-1, CD32b, Clec4g, Stabilin-2; ref. 12). It was therefore hypothesized that EC subtypes may exert specialized functions with regard to liver development, metabolic zonation, and the regulation of the blood-flow rate (13, 14). Clinically, dynamic alterations of LSECs and CECs are frequently observed in chronic liver disease and hepatocarcinogenesis. It is well known that the LSEC phenotype is progressively lost and replaced by non-fenestrated ECs—a process called capillarization (15–17). If and to which extent this process is controlled by YAP-mediated extrinsic communication patterns derived from tumor cells is not well understood.

In this study, we aimed to dissect the dynamic alterations in EC population abundance and communication networks in normal livers as well as YAP-induced hepatocarcinogenesis. Our data unravel central paracrine hubs connecting liver EC subtypes [e.g., hepatocyte growth factor (Hgf)/c-Met] as well as tumor cell–derived regulators (e.g., osteopontin, Opn), which contribute to the formation of a tumor-supporting vascular microenvironment.

Antibodies used for Western immunoblotting or immunohistochemical staining as well as corresponding dilutions, primers, and siRNA sequences are listed in Supplementary Tables. A detailed description of the in vivo and in vitro analyses as well as protocols for mouse work, isolation of primary cells (CECs, LSECs, macrophages, hepatocytes), sample preparation, real-time PCR, Western immunoblotting, expression profiling, tissue-microarray analyses, bioinformatics as well as statistical procedures can be found in Supplementary Materials and Methods.

The mouse EC line SVEC4-10, the human liver EC line Sk-Hep1, as well as the human liver cancer cell line HepG2 were obtained from ATCC (LGC Standards). The murine liver cancer cell line Hep55.1C was obtained from Cell Lines Services GmbH (CLS). Cells were cultured in DMEM (SVEC4-10, Sk-Hep1, Hep55.1C) and RPMI-1640 (HepG2) supplemented with FCS (10%) and 1% penicillin/streptomycin. All cell lines were grown in at 37°C and 5% CO₂ in a 95% humidity atmosphere for up to 30 passages. Cell lines were authenticated by STR-analysis (DSMZ) and routinely checked for Mycoplasma contamination.

C57/Bl6N were bought from Janvier Laboratories. LAP-tTA/Col1A1-YAP^S127A mice were used in this study (2). For transgene repression of constitutively active YAP^S127A, mice received 2 mg/mL doxycycline in their drinking water supplemented with 10 mg/mL sucrose. For transgene induction, doxycycline was withdrawn at the age of 10 weeks. Control mice (with doxycycline) and animals with YAP^S127A expression were sacrificed 8–15 weeks after YAP^S127A induction.

In general, material from three to four mice was pooled to obtain sufficient EC numbers for one biological sample. If the liver weight of YAP^S127A mice in pooled samples exceeded the weight of 5 g, the pooled sample was declared as “late.” If individual YAP^S127A livers had less than 5 g, the sample was declared as “early.” Average liver weights in the “early” and “late” subgroups were 3.29 ± 0.82 g and 5.98 g ± 2.01 g, respectively. Normal liver weight was 1.3 ± 0.2 g.

Animal work was authorized by the German Regional Council of Baden-Württemberg (Regierungspräsidium Karlsruhe, ref. numbers G-30/13, G-57/14, G-65/14, G-201/14, G-176/16, G-30/16, G-187/19, G-201/13). All experiments were performed in accordance with the institutional regulations of the IBF (Interfakultäre Biomedizinische Forschungseinrichtung, University of Heidelberg) under specific pathogen-free (SPF) conditions. The mouse colony was housed under a 12 hours light/dark cycle with free access to water and food. Exclusion and termination criteria were defined in the ATBW criteria.

Human HCCs were surgically resected at the University Hospital and histologically classified according to established criteria by two experienced pathologists. The study was approved by the institutional ethics committee of the Medical Faculty of Heidelberg University (application no. 206/05). Transcriptome and clinical data of 242 human patients with HCC and single-cell data have been published previously (18–20).

Expression data were deposited in the Gene Expression Omnibus database (available: http://www.ncbi.nlm.nih.gov/geo/; accession numbers GSE128042 (CECs vs. LSECs), GSE128044 (wt, early, late of CECs and LSECs), and GSE128046 (HGF stimulation).

For a first characterization of EC diversity in mouse livers, we analyzed different surface marker combinations to discriminate between CECs and LSECs. By using the pan-EC markers CD146 and CD31 in combination with the sinusoidal markers Lyve-1, CD32b, and Stabilin-2 (Stab2) both cell-types were clearly separated using tissue immunofluorescence (IF). Although CECs lining the vessels near the portal tract were CD146⁺⁺/Lyve-1⁻ (CD146⁺⁺/CD32b⁻, CD31⁺⁺/Lyve-1⁻, CD31⁺⁺/CD32b⁻, CD146⁺⁺/Stab2⁻), sinusoidal LSECs were CD146⁺/Lyve-1⁺ (CD146⁺/CD32b⁺, CD31⁺/Lyve-1⁺, CD31⁺/CD32b⁺, CD146⁺/Stab2⁺; Fig. 1A, Supplementary Fig. S1A). To exclude the presence of contaminating Lyve-1–positive lymphatic ECs, colocalization of Lyve-1 and the LSEC marker Stab2 was performed. Almost complete colocalization of these two markers illustrated that lymphatic ECs were not detectable in the chosen experimental setup (Supplementary Fig. S1B; ref. 21).

Validity of this histomorphological approach was confirmed by fluorescence intensity-based plotting of portosinusoidal transition zones, showing CD146 positivity in all EC subtypes, whereas Lyve-1 expression was restricted to LSECs (Fig. 1B). Quantitative measurements illustrated a consistent expression of CD146 in portosinusoidal transition zones, whereas Lyve-1 was exclusively detectable in sinusoids (Supplementary Fig. S1C).

Primary isolated murine ECs exhibited the typical cobblestone morphology in bright field microscopy and actively took up acetylated low-density lipoprotein (Supplementary Fig. S1D). IF and FACS analyses illustrated a mixed EC cell-population, including CD146⁺/Lyve-1⁻ (CECs; 4±2%) and CD146⁺/Lyve-1⁺ cells (LSECs; 95±2%; Supplementary Fig. S1E). Scanning electron microscopy confirmed the presence of LSEC-specific sieve plate fenestrations in >95% of all isolated cells. In the group of sieve plate-positive cells, an expected cellular porosity of 8.67% ± 4.58% was observed (22), whereas ECs without fenestrations represented CECs (Supplementary Fig. S1F and S1G). A capillarization-inducing cultivation protocol was used to analyze network formation of both EC subtypes (23). Indeed, IF stains for CD146/Lyve-1 followed by comparative analysis of EC networks illustrated that CECs formed significantly more junction points than LSECs (Fig. 1C).

One requirement for a molecular characterization of CECs and LSECs was the isolation of pure cell fractions. For this, MACS-based pre-purified CD146⁺ ECs were sorted for CD146, CD31, and Lyve-1 using a FACS-based protocol (Supplementary Fig. S2A and S2B). The approach was confirmed by real-time PCR using CEC markers [von Willebrand factor (Vwf) and endothelin-1 (Edn1)] and LSEC markers [Lyve-1 and c-type lectin domain family member b (Clec1b); Supplementary Fig. S2C; ref. 24].

Comprehensive transcriptome analysis revealed that 247 genes were differentially expressed between CECs and LSECs (Supplementary Fig. S2D). As expected, genes that characterized EC identity, including Stabilin1/2, CD34, MAF BZIP transcription factor (Maf), and transcription factor EC (Tfec), were differentially expressed between both EC subtypes (Fig. 1D). Gene set enrichment analysis (GSEA) revealed an accumulation of differentially expressed genes involved in Hippo and TGFβ signaling in CECs (Supplementary Fig. S2E). In addition, CECs showed positive enrichments of gene sets involved in cell mobility (e.g., Itgb4 and Itga2), cell adhesion (e.g., Ncam1 and L1cam), and cellular guidance (e.g., Sema3f and Sema3d), suggesting that CECs exhibit the molecular requirements for active migration.

Several genes differentially expressed between CECs and LSECs have not been described as potential subtype markers before. We selected basal cell adhesion molecule (Bcam) and thrombomodulin (Thbd/CD141), which were highly expressed in CECs but not in LSECs (Fig. 1E). IF stains and real-time PCR analysis illustrated highest signal intensity for Bcam and Thbd in CECs of the portal tract but not in LSECs (Fig. 1E, Supplementary Fig. S2F). In detail, MFI measurements revealed that the relative intensities of Bcam and Thbd were even higher in vessels of the portal tract than CD146 or CD31, suggesting that these proteins represent specific markers for continuous liver ECs.

Finally, validity of the bulk cell sorting approach was confirmed using single-cell RNA sequencing (scRNA-seq) data derived from healthy human liver tissues (18). These results illustrated the presence of two distinct subgroups of ECs, which are characterized by the presence CEC markers (e.g., CD34, EPHA4, VWF, and BCAM) or LSEC markers (e.g., LYVE-1, STAB2, VEGFR3, and CLEC4G; Fig. 1F, Supplementary Fig. S3A and S3B).

These results demonstrate the existence of two EC subtypes in mouse and human livers with distinct histomorphological, functional, and molecular characteristics.

We were then interested if dysregulation of YAP in hepatocytes could affect the vascular network at different stages of tumorigenesis. To approach this question experimentally, we used the LAP-tTA/YAP^S127A transgenic mouse model for the inducible and hepatocyte-specific expression of constitutively active YAP^S127A, which allows the investigation of EC subtypes under normal, premalignant, and malignant conditions without interfering fibrosis and inflammation (2).

YAP^S127A-induced hepatomegaly and tumor formation within 8–15 weeks after transgene induction was associated with the expansion of YAP-positive and mitotically active cells (Fig. 2A; refs. 2, 5). Tissue-based real-time PCR revealed an induction of YAP target genes (Ctgf and Cyr61) together with expression changes of EC markers (e.g., reduced LSEC markers Lyve-1/Vegfr3) in the process of liver cancer development, implicating alterations of the EC composition (Fig. 2B). Indeed, CD146/Lyve-1 IF stains revealed that Lyve-1 positivity decreased in peri- and intratumoral areas of YAP^S127A-transgenic livers in comparison with portosinusoidal and sinusoidal-central transition zones of wild-type (wt) mice (Fig. 2C, Supplementary Fig. S4A).

Finally, dynamic changes within the EC population were quantitatively examined in normal, hyperplastic, and tumor-bearing livers by FACS analysis. The results demonstrated a significant reduction of LSECs from 95% ± 2% in wt livers to 39% ± 30% in late tumor stages, whereas the number of CEC increased from 4% ± 2% in wt livers to 61% ± 30% in tumors (Fig. 2D). The model-independent shift from an LSEC to a CEC-phenotype especially near tumor borders was confirmed in autochthonous (AlbTag) and implantation (Hep55.1C) liver cancer models (Supplementary Fig. S4B and S4C). Because of similar results in all models, we concluded that the EC population switch represented a model-independent process in areas with active angiogenesis.

Functionally, proliferative angiogenesis contributed to the observed expansion of CECs. This was illustrated by elevated proliferation of ECs isolated from YAP^S127A transgenic mice compared with EC from wt animals (Supplementary Fig. S4D). In addition, Ki67/CD146 co-stains (Supplementary Fig. S4E) and bioinformatic analysis of CEC expression data derived from wt mice and YAP^S127A transgenics further supported that active cell division contributes to CEC expansion (Supplementary Fig. S4F).

These results unveil a progressive replacement of LSECs by CECs in YAP-induced hepatocarcinogenesis, which recapitulates a phenotype observed in human chronic liver disease and cancer.

Our previous results demonstrated an EC subpopulation shift in YAP-driven hepatocarcinogenesis; however, it is unknown to what extent CECs and LSECs dynamically modify gene expression in this process in vivo. To define this dynamic molecular response in tumorigenesis, CECs and LSECs were isolated from wt, YAP^S127A-induced hyperplastic (early) as well as tumor-bearing (late) livers and were subjected to transcriptome analysis. CEC- and LSEC-specific marker gene expression validated the purity of both EC populations used for this analysis at all stages of the experiment (Fig. 3A).

In total, 4,877 genes with Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway annotation were significantly regulated when multiple comparisons between CECs and LSECs as well as between EC subtypes at different stages of tumorigenesis were performed (Supplementary Fig. S5A). Principal component analysis (PCA) illustrated highly concordant gene expression within all six biological groups (CEC- and LSEC-specific clusters) as well as the molecular evolution of both cell types within carcinogenesis (Fig. 3B). To identify CECs- and LSEC-specific properties in tumor angiogenesis, a panel of 12 published genes that characterize polarized cells within angiogenic sprouts was tested (tip-cell signature; refs. 25–27). Most signature genes were highly expressed in CECs (e.g. Jag1) or gradually increased from wt to late stage CECs (e.g. Lamb1), which was indicative of a higher degree of cell migration in this EC subtype (Fig. 3C). In accordance with these results, a differential expression of genes in the migration-associated KEGG pathways ‘focal adhesion’, and ‘cell adhesion and ECM interaction’ was detected (Supplementary Fig. S5B).

Since we hypothesized that YAP dysregulation in hepatocytes affects the vascular compartment in a growth factor/cytokine/chemokine-dependent manner, YAP^S127A-transgenic mice were treated with the FDA-approved multi-kinase inhibitor cabozantinib to block paracrine communication. Indeed, cabozantinib treatment abolished CEC expansion and reduced the expression of tip-cell signature genes in YAP^S127A-transgenic mice, confirming the presence of functionally relevant communication networks between different cells types such as EC subpopulations (Supplementary Fig. S6A–S6C).

To further characterize whether close spatial distance between CECs and LSECs would allow the direct communication of both cell types via paracrine factors, co-stains of CD146/Lyve-1 as well as Cxcr4 (which is a tip-cell marker upregulated in migratory ECs) were performed. The results illustrated that tumor-associated CECs (CD146⁺⁺/Lyve-1⁻, Cxcr4⁺/Lyve-1⁻) were predominantly located in close proximity with LSECs (CD146⁺/Lyve-1⁺, Cxcr4⁻/Lyve-1⁺; Supplementary Fig. S7A and S7B). This spatial organization of CECs and LSECs especially at the tumor border of YAP^S127A-induced tumors further argues for the existence of communication hubs in areas with active angiogenesis.

To identify potential cross-talks connecting CECs and LSECs subtypes in YAP^S127A-induced tumor formation, the expression of genes involved in cellular signaling, including ligands and receptors, was compared (Fig. 3D). Next to static, cell-type–specific differences [e.g., Tgfβ2 in CECs and C–X–C motif chemokine ligand 9 (Cxcl9) in LSECs], real-time PCR analyses confirmed that several known ligand/receptor pairs were dynamically regulated in both EC subtypes. This was indicative of the existence of heterologous communication networks connecting CECs and LSECs in tumorigenesis (Fig. 3E). These cytokine/cytokine receptor pairs included C–X–C motif chemokine ligand 12 (Cxcl12)/C–X–C motif chemokine receptor 4 (Cxcr4) and Vegfc/fms-related tyrosine kinase 4 (Flt4/Vegfr3). Equally, Hgf was highly expressed only by LSECs in wt, early, and late phases of liver tumorigenesis, whereas its receptor c-Met was progressively and exclusively induced in CECs.

Because of the functional importance of the Hgf/c-Met signaling axis in different biological processes such as migration, we decided to confirm that YAP activation in hepatocytes/tumor cells positively affects c-Met expression in CECs. For this, we first showed a stepwise CEC-specific induction of c-Met in tissue specimens from YAP^S127A transgenic mice and in the orthotopic Hep55.1C implantation model (Fig. 3F). Indeed, a clear colocalization between ECs with high CD146 expression and c-Met was detectable in tumors with active angiogenesis (Supplementary Fig. S7C).

To further substantiate that tumor cell–derived YAP is a critical regulator of the vascular niche and c-Met expression in ECs, additional in vivo experiments were performed. First, YAP^S127A expression was turned off in mice after tumor initiation, which not only led to elevated Lyve-1 positivity but also decreased tip-cell genes and Opn expression (Supplementary Fig. S7D–S7F, Opn see below). Second, hydrodynamic tail vein injection experiments were performed (28). Combined overexpression of YAP^S127A and a stabilized β-catenin isoform (β-catenin^T41) was compared with the oncogene combination β-catenin^T41/myr-AKT (myristoylated AKT). As expected, co-injection of both oncogene combinations led to the formation of tumor nodules in liver tissues (28). However, YAP^S127A co-injection with β-catenin^T41 led to a more pronounced vascular positivity of c-Met expression compared with tumors caused by injection of β-catenin^T41/myr-AKT (Supplementary Fig. S7G).

In sum, these data demonstrate that CECs and LSECs actively adjust their expression at the molecular level to form potential communication hubs, including the ligand/receptor combination Hgf/c-Met in the process of YAP-induced tumorigenesis.

To confirm the cell-type–specific expression of Hgf and c-Met in EC subtypes in an alternative approach, a time-dependent differentiation culture model for primary mouse ECs was established (24). Here, short-term cultivation of ECs was characterized by an LSEC phenotype (≤2 days), whereas long-term cultured ECs lost the LSEC and acquired the CEC phenotype (5 days). Indeed, a shift from an LSEC to a CEC phenotype after EC cultivation for 5-day was confirmed by elevated expression of different CEC markers as well as an increase of CD146-specific MFI (Supplementary Fig. S8A and S8B). More importantly, Hgf transcript levels were clearly reduced, whereas c-Met mRNA increased in cells with an CEC phenotype (Fig. 4A). Indeed, Hgf levels in cell culture supernatants illustrated that freshly isolated ECs with an LSEC phenotype secreted significantly higher amounts of Hgf than long-term cultured ECs with a CEC phenotype or primary hepatocytes (Fig. 4B).

To further substantiate that Hgf was produced by LSECs and that c-Met is increasingly expressed by tumor-associated ECs with CEC phenotype, in situ hybridization was performed. In healthy murine liver tissue, Hgf transcripts were exclusively detectable in sinusoidal cells with protracted morphology, whereas c-Met positivity was not observed (Supplementary Fig. S8C). In livers from YAP^S127Atransgenic mice, Hgf mRNA was more pronounced in ECs outside tumor nodules (LSECs phenotype) compared with intratumoral CECs. In contrast, c-Met positivity was only detected in ECs alongside of tumor cells, which confirms the data derived from immunohistochemical stains (Supplementary Fig. S8C; Fig. 3F). Predominant LSEC-dependent Hgf expression was confirmed at the transcript level for primary isolated LSECs, CECs, liver macrophages, and hepatocytes derived from wild-type and YAP^S127A transgenic mice (Supplementary Fig. S8D). In sum, these experiments illustrate predominant Hgf expression by LSECs and c-Met expression by CECs after tumor initiation.

To further define the biological relevance of Hgf on CECs in vitro and because even FACS-sorted primary CECs are not suitable for functional experiments, SV40-transformed SVEC4-10 cells were used. Although, these cells were of lymphatic origin, and therefore can only serve as surrogate model for liver-derived CECs, SVEC4-10 cells displayed several CEC characteristics such as vascular cell adhesion molecule-1 (VCAM1) expression as well the ability to form tube-like networks (29). Moreover, SVEC4-10 cells expressed c-Met, lacked the expression of typical LSEC markers (e.g., Lyve-1), and therefore served as an Hgf-sensitive CEC-related model system for functional and mechanistic analyses (Supplementary Fig. S8E and S8F). Expression profiling revealed that HGF administration for 3 or 6 hours led to the significant regulation of 1,717 genes in SVEC4-10 cells (Fig. 4C). Gene ontology analysis showed a regulation of genes involved in processes associated with blood vessel development (e.g., “angiogenesis”) and cell mobility (e.g. “invasion”; Fig. 4D). Together with our previous finding that CECs and LSECs differentially express cell mobility-associated genes, we hypothesized that the paracrine crosstalk between LSECs and CECs via Hgf/c-Met predominantly affects CEC migration.

Indeed, SVEC4-10 cell-population–based, three-dimensional spheroid sprouting assays illustrated that Hgf significantly increased total spheroid size through the enhanced formation of sprouts in a surrounding matrix (Fig. 4E). To substantiate the impact of Hgf on a single-cell level, motion trajectories of SVEC4-10 cells were quantified after HGF administration using time-lapse microscopy (Fig. 4F). In addition, c-Met inhibition in SVEC4-10 cells using gene-specific siRNAs diminished cell invasiveness (Supplementary Fig. S9A and S9B).

In sum, these data demonstrate that Hgf secretion by LSECs in YAP transgenic livers activates migration/invasion in ECs with a CEC phenotype.

As exemplified for the Hgf/c-Met signaling pathway, our data pointed to a dynamic communication between EC subtypes in the process of YAP-dependent premalignant liver capillarization and hepatocarcinogenesis. To describe potential mechanisms how hepatocytes and tumor cells initiate these changes in the vascular niche, we aimed to identify paracrine-acting factors actively secreted by YAP^S127A-positive parenchymal cells. Proteome array analysis revealed that several cytokines in conditioned medium from YAP^S127A transgenic hepatocytes were elevated, such as osteopontin (Opn), placental growth factor (Plgf), and serpin family E member 1 (SerpinE1; syn. plasminogen activator inhibitor, Pai1; Fig. 5A and B; Supplementary Fig. S10A). For further analysis, we focused on Opn, which showed the highest dynamic range and lowest variability in the supernatants of cultured hepatocytes.

Indeed, confirmatory experiments illustrated high-level expression of Opn in primary hepatocytes at the mRNA level and in plasma of YAP^S127A-transgenic mice (Fig. 5C and D). On the basis of previous studies, which revealed a physical interaction between YAP and the transcription factor TEA domain transcription factor 4 (Tead4) in hepatocytes (5), we hypothesized that Tead4 might transcriptionally regulate OPN expression through binding to gene-promoter regions. Indeed, siRNA-mediated inhibition of TEAD4 reduced OPN expression in a human liver cancer cell line (Fig. 5E). Furthermore, data derived from the Encyclopedia of DNA Elements (ENCODE) database corroborated the presence of TEAD4-binding sites within the human OPN (syn. SPP1) and CTGF promoters (pos. control) in two cell lines (Supplementary Fig. S10B). This was confirmed by chromatin immunoprecipitation (ChIP) analyses, illustrating binding of TEAD4 at the OPN/SPP1 promoter as predicted by a JASPAR database (Fig. 5F; ref. 30).

Finally, the biological impact of Yap-dependent Opn on ECs and cells with a CECs phenotype with regard to c-Met expression was tested in vitro using primary isolated cells. Indeed, Opn administration led to a moderate but consistent induction of c-Met expression at the transcript level in primary ECs and at the protein level in SVEC4-10 cells (Fig. 5G and H). This increase of c-Met expression in SVEC4-10 cells was mediated by the PI3K/AKT pathway but not via MAKP, STAT3, or NF-kB signaling (Supplementary Fig. S10C). To confirm the induction of c-Met transcription by the PI3K/AKT pathway, a chemical AKT phosphorylation activator was used to mimic Opn-induced PI3K/AKT pathway activation in vitro (31). As expected, SC-79 administration increased c-Met expression levels in SVEC4-10 cells (Supplementary Fig. S10D).

In summary, hepatocyte-derived secreted factors such as YAP-induced Opn serve as initial effectors that induce heterologous communication networks between CECs and LSECs, for example, by sensitizing CECs toward LSEC-derived Hgf.

To confirm the findings of our in vivo studies and screening approaches in human tissue samples, transcriptome data derived from 242 patients with human HCC (HCCs and corresponding adjacent liver tissues) were analyzed (20).

The results confirmed the loss of cells with an LSECs phenotype because a significant reduction of LSEC marker signature genes (incl. LYVE-1 and STAB2) was detectable in HCCs compared with liver tissues, whereas CEC markers (including BCAM and VWF) were elevated (Fig. 6A). Interestingly, the presence of a tip-cell gene signature (consisting of JAG1, LAMB1, NRP1, CXCR4, PGF, KDR, ESM1), which was characteristic of the CEC phenotype, not only significantly correlated with OPN transcript levels (e.g., CXCR4, r = 0.34, P ≤ 0.001; LAMB1, r = 0.39, P ≤ 0.001) but was also statistically associated with poor clinical outcome of patients with HCC (Fig. 6B). In addition, OPN overexpression in HCCs significantly associated with poor overall survival and early cancer recurrence (Supplementary Fig. S11A).

Immunohistochemical stains of liver tissues and HCCs with good (G₁/G₂) and poor (G₃/G₄) differentiation were performed (Fig. 6C and D). Similar to our previous results, semiquantitative analysis revealed a progressive and statistically significant induction of YAP in cytoplasm and nuclei in tumor cells (r = 0.41 and 0.45; P ≤ 0.001; refs. 5, 32). Although CD146 expression increased (r = 0.45, P ≤ 0.001), Lyve-1 was strongly reduced in the group of HCCs (r = −0.34, P ≤ 0.001). Similarly, elevated OPN abundance significantly correlated with tumor progression (r = 0.34; P ≤ 0.001). In situ hybridization confirmed a significant reduction of HGF-producing cells in the course of tumor dedifferentiation (r = −0.22; P ≤ 0.05), which positively correlated with the loss of LYVE-1 expression (r = 0.39; P ≤ 0.01; Supplementary Fig. S11B).

Finally, we aimed to confirm our findings using scRNA-seq data derived from human patients with liver cancer (19). In the group of ECs, a high degree of cellular heterogeneity was detectable, with one population being positive for LYVE-1, CD32b, VEGFR3, and HGF (Fig. 6E). In contrast with this EC population with typical LSEC marker gene expression, another distinct subpopulation could be identified with CEC marker expression, including EDN1, CXCR4, and c-MET. In addition, in the tumor cell population a significant correlation between OPN and typical YAP target genes was detectable supporting the YAP-dependent regulation of OPN in these cells (e.g., with CYR61, r = 0.49; UHMK1, r = 0.42; Jag-1, r = 0.52, P ≤ 0.001).

Together, these results confirmed the association between YAP activation, OPN overexpression, and the replacement of LSECs by CECs in human hepatocarcinogenesis.

One central goal of our study was the molecular characterization of CECs and LSECs in different stages of YAP-induced hepatocarcinogenesis. To address this, we used a model for the inducible and hepatocyte-specific expression of constitutively active YAP^S127A (2). This model leads to the development of hepatomegaly and rapid tumor formation within a few weeks without extensive fibrosis. Importantly, our model of YAP-induced changes in the microenvironment through heterologous communication between hepatocytes/HCC cells and ECs is supported by recent findings illustrating tumor cell-extrinsic functions of this oncogene (7). In addition, the central role of the Hippo/YAP pathway in the context of angiogenesis is already discussed as the major YAP target genes CTGF and CYR61 are important regulators of angiogenesis (33). Moreover, YAP contributes to angiogenesis in cholangiocellular carcinoma via the transcriptional regulation of proangiogenic microfibrillar-associated protein 5 (MFAP5; ref. 34). Therefore, the here proposed mechanism of YAP-induced OPN expression that induces heterologous communication between EC subpopulations via the HGF/c-MET pathways represents one possibility of how this oncogene creates a tumor-supportive hepatic vascular compartment. Even though multiple paracrine communication cues might be induced by YAP, this mechanism is of particular interest, because its activation depends on the spatial proximity of liver tissue (predominantly LSECs) with tumor border (predominantly CECs). This supports the idea of spatially active proangiogenic compartments in tumor development and progression.

Heterotypic communication regulates cell-type–specific functions and is required for organ development and tissue homeostasis. The liver vascular niche is a key regulator of organ functionality under physiological and pathological conditions. For example, a dynamic interlineage cross-talk in three-dimensional liver bud organoids consisting of hepatic endodermal, endothelial, and mesenchymal cells is crucial for liver development (35). EC-derived angiocrine signaling regulates organ patterning in liver development, fibrosis, and organ reconstitution after liver damage (36–38). Recent findings illustrate the relevance of ECs as integrators of mechanical forces (blood perfusion) into angiocrine signals and biological responses, such as hepatocellular proliferation and organ growth (39). Under pathophysiological conditions, the importance of ECs in tumor formation and progression has been demonstrated for various tumor entities (40); however, the impact of dynamically expressed and liver EC-derived angiokines in different stages of hepatocarcinogenesis is poorly understood.

The concept of EC heterogeneity across organs as well as their distinct characteristics in homeostasis, regenerative processes, and disease is well established (41). However, little is known about the molecular characteristics of EC subtypes within tissue-specific vascular beds. For the liver, ultrastructural and histo-morphological data illustrate the existence of heterogeneous EC compartments with specific morphological features: The porto-central axis lining liver sinusoid (LSECs) and the portal/pericentral micro-vessels composed of CECs (8, 9). First attempts to characterize hepatic ECs on a molecular level were conducted by comparative expression profiling of liver and lung microvascular ECs (24). This analysis revealed a gene set representative for ECs forming discontinuous micro-vessels, including transcriptional regulators (e.g., TFEC and MAF), scavenger receptors (e.g., Stabilin-1/-2) and cytokine/growth factor signaling (e.g., the WNT pathway). Recently, unpaired and paired single cell analyses further describe liver EC heterogeneity (10, 18); however, a combined and detailed spatial, molecular, biochemical, and functional description of hepatic EC subtypes is not yet available. Our results demonstrate exclusive features of distinct EC populations in the liver, including the expression of signaling pathway constituents (e.g., Hippo and TGFβ pathway) and functionality (e.g., ECM interaction and cell adhesion). However, it is worth mentioning that other hepatic EC subtypes exist (e.g., lymphatic ECs) and that CECs as well as LSECs can be subtyped into more distinct subpopulations (42).

Indeed, data of this study illustrate a replacement of LSECs by CECs in hepatocarcinogenesis and describe new CEC subtype biomarkers, including BCAM and THBD. In combination with arterio-venous discriminating molecules such as EFNB2–EPHB4, these factors may represent valuable tools for a subsequent dissection of hepatic vascular compartments (43).

Hepatic vascular remodeling or “capillarization” is defined by the loss of LSEC characteristics (including marker expression and fenestration), an increase in CEC marker expression and altered ECM deposition in the perisinusoidal space (44). Our data illustrate that this progressive shift in EC composition is detectable in the earliest phases of liver tumorigenesis. This is confirmed by previous findings showing that the loss of LSEC characteristics might act as a tumor driver already in early liver disease (45). Our results also suggest that CEC expansion is not only based on transdifferentitation but also on CEC proliferation and active migration; however, additional cellular mechanisms may contribute to the observed EC class switch. For example, EC plasticity could lead to cellular transdifferentiation (24, 44) and endothelial-to-mesenchymal transition (EndMT) could give rise to ECs to form a continuous endothelium (46). In addition, bone marrow–derived CSF1R⁺ cells have been identified as endothelial progenitors that integrate into existing blood vessels (47, 48). Finally, recent data also suggest that bone-marrow–derived progenitor cells can replace damaged LSECs by incomplete differentiation and may therefore be a source of ECs with continuous phenotype (49).

The dynamic and tumor stage–dependent expression patterns derived from CECs and LSECs led us draw several conclusions. First, YAP induces a rapid heterologous communication network, which is already detectable in early and premalignant stages of hepatocarcinogenesis. This network is central for remodeling the vascular niche, which may be a mandatory step for the “angiogenic switch” needed for tumor progression (50). Second, angiocrine signaling not only connects ECs with parenchymal hepatocytes (38) or nonparenchymal HSCs (51), but may also form a communication network between EC subtypes. Interestingly, the dynamic regulation of several ligand/receptor pairs in CECs and LSECs is suggestive of a multifactorial connection between both cell types (e.g., HGF/c-MET, VEGFC/VEGFR3, CXCL12/CXCR4), which may affect their spatiotemporal responsiveness and their biological properties at the tumor–liver interface.

Our data and other studies already illustrated that LSECs represent the major cellular source for Hgf (36, 39). However, the progressive increase of c-Met in CECs, in combination with constitutively high HGF production by LSECs, raised our particular interest, as it indicated that CECs gain their responsiveness toward HGF in liver cancer development. Indeed, the relevance of the HGF/c-MET axis in aberrant vascularization, for example, in neural tumor development and during liver regeneration has been confirmed in previous work (36, 52). It is worth mentioning that next to LSECs, ECM constituents with low-affinity–binding sites for HGF (e.g., collagens) may serve as additional source for this growth factor (53). Thus, HGF derived from LSECs may be stored as ECM-bound HGF, which is released in the course of expansive capillarization in tumorigenesis. However, we want to emphasize that for some mechanistic and functional experiments in this study a surrogate model system was utilized (SVEC4-10 cells of axillary lymph node origin), because primary isolated liver CEC fractions could not be used due to technical limitations.

Induction of the YAP-dependent heterologous communication network in YAP^S127A mouse model must initially originate from genetically modified hepatocytes. Among the hepatocyte-derived soluble factors, OPN is transcriptionally regulated by YAP/TEAD4 and induces c-MET expression in liver ECs. On the basis of our results, we conclude that OPN represents one communication hub connecting tumor cells with CECs, which sensitizes this EC subtype for LSEC-derived HGF. Indeed, the relevance of OPN as a central regulator of angiogenesis has already been confirmed by several studies, such as in pancreatic cancer and breast cancer (54, 55). If YAP-dependent extrinsic functions control heterologous communication and capillarization in other HCC models that equally show YAP overexpression (e.g., AlbTag, Hep55.1C, hydrodynamic tail vein injection of oncogenes), or whether YAP-independent mechanisms contribute to this phenotype, needs to be clarified in future studies.

In summary, the integration of time-resolved data allowed us to reconstruct a spatial communication network consisting of YAP-dependent OPN secretion in tumor cells, which controls the cross-talk between EC populations via the HGF/c-MET signaling axis. This heterologous communication contributes to the formation of a tumor-supporting microenvironment.

D. Kazdal reports personal fees from AstraZeneca, Bristol-Myers Squibb GmbH, and Pfizer Pharma GmbH outside the submitted work. K. Breuhahn reports grants from German Research Foundation (DFG) and German Cancer Aid during the conduct of the study. No disclosures were reported by the other authors.

S. Thomann: Conceptualization, data curation, formal analysis, investigation, visualization, methodology, writing-original draft. S.M.E. Weiler: Investigation, methodology. S. Marquard: Investigation, methodology. F. Rose: Investigation, methodology. C.R. Ball: Resources, formal analysis, visualization. M. Tóth: Investigation, methodology. T. Wei: Investigation, methodology. C. Sticht: Resources, data curation, methodology. S. Fritzsche: Investigation, methodology. S. Roessler: Resources, writing-review and editing. C. De La Torre: Resources, software, investigation, methodology, resources, software, investigation, methodology. E. Ryschich: Resources, investigation. O. Ermakova: Software, investigation, visualization, methodology. C. Mogler: Conceptualization, methodology. D. Kazdal: Software, formal analysis, visualization. N. Gretz: Resources, software, methodology. H. Glimm: Resources, visualization. E. Rempel: Software, formal analysis, visualization. P. Schirmacher: Conceptualization, resources, supervision. K. Breuhahn: Conceptualization, data curation, formal analysis, supervision, funding acquisition, writing-original draft, project administration, writing-review and editing.

The authors thank Jennifer Schmitt, Michaela Bissinger, Sandra Förmer, Jutta Scheuerer, and Nina Hofmann for excellent technical assistance. We also want to thank F.D. Camargo for providing the Col1A1-YAP^S127A transgenic mice and S. Goerdt and K. Schledzewski for providing the Stabilin-2 antibody. Tissue samples were provided by the Tissue Bank of the National Center of Tumor Diseases (NCT, Heidelberg, Germany) in accordance with the regulations of the Tissue Bank and the approval of the Ethics Committee of Heidelberg University. We thank the Center of Model System and Comparative Pathology (CMCP; Tanja Poth, Heike Conrad, Diana Lutz, Karin Rebholz, and Veronica Geissler) for their support. We also want to thank Ulrike Ganserer and Dr. Ingrid Hausser for conducting scanning electron microscopy as well as Martin Dittmer for providing us with support in programming. We want to thank Katherine Zach for article proofreading. This study was supported by a grant from the German Cancer Aid (Deutsche Krebshilfe/Dr. Mildred Scheel Stiftung, DKH 111524 to K. Breuhahn) and the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation)—SFB/TRR 209 (314905040 to K. Breuhahn, P. Schirmacher, and S. Roessler). P. Schirmacher was supported by the European Union's Horizon 2020 Research and Innovation Program under grant agreement no. 667273 (HEP–CAR). S. Thomann was a member of the MD/PhD program of the Hartmut Hoffman Berling International Graduate School (HBIGS) at Heidelberg University and was financially supported by the Heidelberg Medical Faculty. S. Thomann received start-up funding from the German Society of Pathology and the LBBW Foundation.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.