Hepatic Endothelial Notch Activation Protects against Liver Metastasis by Regulating Endothelial-Tumor Cell Adhesion Independent of Angiocrine Signaling

Paget's “seed and soil” hypothesis claimed organotropism of cancer metastasis. Among other organs, the brain, the bones and the liver are prime examples for organotropic metastasis. Although prostate cancer preferentially metastasizes to the bones, colorectal cancer metastasizes to the liver. Originally, it has been assumed that preferential metastasis of colorectal cancer to the liver reflects the fact that the liver is the first sieve for colorectal cancer cells that have immigrated into the circulation. Evidence is now accumulating that tumor cell-intrinsic factors as well as microenvironment-mediated, organ-specific factors are of paramount importance for active control of organotropic metastasis. This notion is also supported by the fact that tumors not derived from the gastrointestinal tract also metastasize to the liver such as lung and breast cancer as well as malignant melanoma. Regarding uveal melanoma, organotropic metastasis to the liver is clinical proof of principle for Paget's hypothesis.

In colorectal cancer, several tumor cell-intrinsic mechanisms for organotropic liver metastasis have been described including factors regulating epithelial-mesenchymal transition such as miR-200c (1), which is itself dependent on the PKCζ/ADAR2 axis (2). Molecular subtyping further defines a curable oligometastatic state in colorectal liver metastasis characterized by immune activation. On the contrary, adverse outcome is characterized by VEGFA amplification (3). After they have metastasized, CRC cells also undergo metabolic reprogramming thus taking advantage of the new microenvironment in the liver as a metabolically active organ (4). Metabolic exploitation clearly requires organ-specificity. In other cancers, similar tumor cell–intrinsic mechanisms determine liver metastasis. In melanoma, for example, MSX1-induced neural crest-like reprogramming promotes hepatic metastasis (5).

Besides tumor cell–intrinsic factors, the hepatic niche provides organ-specific cues for organotropic metastasis. Tissue-resident macrophages, that is, Kupffer cells, but also monocyte-derived macrophages support liver metastasis, especially when M2 polarization prevails (6). Among other mechanisms, colorectal cancer-derived VEGFA stimulates macrophages to secrete the chemokine Cxcl1 causing recruitment of Cxcr2-positive myeloid-derived suppressor cells (7). In melanoma, NK-cell activity is determined by genetic immune predisposition protecting from liver metastasis (8). In addition, hepatic stellate cells (9), but also innervation determine metastatic susceptibility of the hepatic niche (10). Last, but not least, hepatocytes (HC) should not be neglected as active members of the hepatic niche. Notably, HC-derived VEGF promotes “off-drug” metastasis after discontinuation of anti-VEGF therapy of colorectal cancer (11).

Within the hepatic niche, the major player to control metastasis is the liver sinusoidal endothelium representing the first line of contact for circulating tumor cells in the liver. Liver sinusoidal endothelial cells (LSEC) promote liver metastasis by either adhesive or angiocrine mechanisms. E-selectin was one of the first molecules to be shown to mediate hepatic metastasis in colorectal cancer (12), followed by an overwhelming body of evidence for ICAM1 to exert a major role in liver metastasis of different cancers (13). More recently, endothelial lectins such as Clec4g were shown to mediate colorectal cancer metastasis to the liver (14). In addition, hepatic endothelial fibronectin deposits have been shown to bind circulating tumor cells via talin1, a component of focal adhesions (15). The importance of LSEC in the control of liver metastasis is also underlined by the long-known fact that cirrhotic livers harbor less metastases than non-cirrhotic livers (16). Liver cirrhosis/fibrosis is accompanied by transdifferentiation of LSEC into continuous capillary-like ECs, a process termed sinusoidal capillarization. Sinusoidal capillarization is characterized by deposition of an endothelial basement membrane including matrix molecules such as collagens, which are also known to promote liver metastasis (17).

Moreover, secretion of angiocrine factors is increasingly recognized as a highly relevant function of LSEC. Among others, LSEC-derived inflammatory cytokines such as IL1 (18), MIF, and Cxcl12 (19) promote liver metastasis. Notably, gut microbiome-mediated primary-to-secondary bile acid conversion induces secretion of Cxcl16 by LSEC regulating NK-cell accumulation (20). Surprisingly, angiocrine signaling pathways such as Wnt signaling (21–25), Hgf signaling (26), and Bmp2 signaling (27, 28), described by us and others to govern liver development, liver function, and liver regeneration, have not been checked for their role in liver metastasis. In addition, known superordinate molecular regulators of angiocrine signaling in the liver such as Gata4 (29) have also not been scrutinized for their involvement in hepatic metastasis.

In this respect and regarding the molecular mechanisms of metastatic organotropism, there is indication that Notch signaling may exert opposing effects in metastasis to the lung as compared with the liver. Endothelial overexpression of Notch was shown to promote lung metastasis by upregulation of VCAM1 (30), whereas liver metastasis was promoted by Notch inhibition (31). In the latter study, however, tools were missing to elucidate whether the effects of Notch inhibition on liver metastasis were mediated by HEC or other cells in the hepatic niche and effects of Notch activation in HEC were not investigated.

Therefore, we here generated a novel EC subtype-specific Cre driver mouse line using the Clec4g promoter. When crossed with Rosa26^{N1ICD-IRES-GFP} mice to enhance Notch signaling in HEC, liver metastasis in various models of malignant melanoma as well as colorectal carcinoma was significantly reduced. Wnt signaling in the liver was markedly impaired with effacement of metabolic liver zonation. However, reduced metastatic susceptibility of the liver vasculature was not due to these angiocrine alterations as hepatic metastasis in endothelial Wnt secretion-deficient Stab2-iCre^tg/wt;Wls^fl/fl mice (25) was not reduced. On the contrary, cancer cell adhesion to hepatic endothelium was significantly reduced in Clec4g-Cre^tg/wt;Rosa26^{NICD-IRES-GFP} mice accompanied by a special form of sinusoidal capillarization and a strong reduction in ICAM1, but not VCAM1 expression. Altogether, we show here that endothelial Notch signaling controls metastatic susceptibility in an organ-specific and opposite manner exploiting regulation of adhesive mechanisms rather than by influencing hepatic angiocrine signaling.

A BAC encoding the genomic Clec4g locus (RPCI-23-239A7) was obtained from SourceBioscience. A cDNA containing a codon improved version of cre recombinase (32) followed by polyA signal sequence was fused into exon 1 of Clec4g via homologous recombination. In frame recombination into Clec4g was realized via PCR added homologous arms of 45 bp in length flanking the 5′ and 3′ region of the start codon. Correct Clec4g-cre recombinant BAC was confirmed by sequencing, linearized by Not1 endonuclease, purified with Sepharose CL4B (GE Life Sciences) and injected into the pronuclei of fertilized C57Bl/6NCRL oocytes. Three transgenic founder mice (B6.Tg(Clec4g-cre)1.1-1.3) were identified by PCR using primer pairs PC1 and PC2 (see below) and bred with C57BL/6N mice. For characterization transgenic progeny of all three founder lines were crossed to Rosa26:eYFP^fl/fl [B6.129 × 1-Gt(ROSA)26Sor^tm1(EYFP)Cos/J; (JAX 006148); ref. 33] reporter animals and analyzed as described (25, 27, 29). Only the strain B6.Tg(Clec4g-cre)1.1 (Clec4g-Cre^tg/wt) was further used in this study.

Stab2-Cre^tg/wt or Clec4g-Cre^tg/wt mice were crossed to a mouse line bearing a fragment of the Notch1 gene and a GFP inserted into the Rosa26 locus [Gt(ROSA)26Sor^{tm1(Notch1)Dam}/J; JAX 008159; ref. 34] to generate Stab2-Cre^tg/wt or Clec4g-Cre^tg/wt; Rosa26^{N1ICD-IRES-GFP} (NICD^OE-HEC) mice. Genotyping was performed at P28 using primer pairs PC1 and PC2 for Cre, PY1, PY2, and PY3 for Rosa26^{N1ICD-IRES-GFP} and for Rosa26:eYFP (Supplementary Table S1). All animals used were on a C57BL/6 background and siblings were chosen as Ctrl.

See Supplemental Experimental Procedures.

The mouse melanoma cell line B16F10 luc2 was purchased from Perkin Elmer. The transformed mouse melanoma cell line Wt31 (35) was a gift from O. Sansom (Beatson Institute for Cancer Research, Glasgow, United Kingdom). MC38 colorectal carcinoma cells were kindly provided by S. Herzig (Helmholtz Zentrum, Munich, Germany). Cell authentication was conducted by pigmentation status, morphology or bioluminescence. All cell lines were regularly tested mycoplasma-free by PCR. All cells were maintained in RPMI with 10% (v/v) FCS and 100 U/mL penicillin/streptomycin at 37°C, 5% CO₂ (Thermo Fisher Scientific). For in vivo experiments always the same passage of corresponding cell lines was used. After thawing they were not passaged more than three times. Maximum culture time prior to in vivo experiments was one week.

For metastasis experiments female mice were used between 11 to 12 weeks of age. After anesthesia with Isoflurane 2.5 × 10⁶ Wt31 melanoma cells were slowly injected into the tail vein. To perform intrasplenic injection of tumor cells mice were anaesthetized by Isoflurane and a laparotomy on the lateral left side was performed. The spleen was put out of the peritoneum using a cotton stick. 1.5 × 10⁵ B16F10 luc2 melanoma cells or MC38 colorectal carcinoma cells or 3.0 × 10⁵ Wt31 melanoma cells were carefully injected into the spleen. 15 min after injection a splenectomy was performed to avoid the growth of an intrasplenic tumor and an increase in tumor burden. The peritoneum and skin were closed by suture (Vicryl 6.0; Ethicon). For rehydration 0.9% NaCl was applied subcutaneously. Carprofen was used as analgetics and animals were monitored carefully after procedure. At indicated days tumor load was quantified by bioluminescence imaging (BLI; see Supplementary Experimental Procedures). The animals were sacrificed either at day 19 (Wt31 melanoma, intravenous injection), day 14 (B16F10 luc2 melanoma, Wt31 melanoma, splenic injection) or day 21 (MC38 colorectal carcinoma, splenic injection), organs were removed and analyzed for melanoma colonization. To study initial tumor cell adhesion and retention the mice were analyzed 90 min after intrasplenic injection of B16F10 luc2 melanoma cells.

After anaesthesia with Isoflurane 2.0 × 10⁵ B16F10 luc2 melanoma cells were injected into the tail vein to study melanoma lung colonization. The animals were sacrificed and analyzed at day 14.

C57Bl/6 Wt mice were injected with either 500 μg of InVivoMab anti-ICAM1 (Clone: YNI.7.4; BE0020-1; Hölzel Diagnostika Handels GmbH) or corresponding InVivoMab rat IgG2b isotype control (Clone: LTF-2; BE0090; Hölzel Diagnostika Handels GmbH) antibodies 24 hours prior to injection of tumor cells into the spleen.

All animals received humane care in compliance with the Guide for the Care and Use of Laboratory Animals published by the National Academy of Sciences and all animal experiments were approved by the animal ethics committee of Baden-Wuerttemberg (Regierungspraesidium Karlsruhe).

All statistical analyses and graphical displays were performed with GraphPad Prism7 (Graph Pad) and mean ± SEM is presented. For statistical analysis, an unpaired, two-tailed t test was applied if not indicated differently in figure legends. The quantification of immunofluorescence was tested by a one-sample t test. For statistical analysis of sampling distributions Fisher exact test was used. Differences between data sets with P < 0.05 were considered statistically significant.

Endothelial subtype-specific Stab2-Cre driver mice (29) were crossed with Rosa26^{N1ICD-IRES-GFP} mice to express constitutively active Notch1 intracellular domain (NICD) in HEC. The resultant Stab2-Cre^tg/wt; Rosa26^{N1ICD-IRES-GFP} mice died in utero before E12.5, likely due to yolk sac defects (Supplementary Fig. S1). Therefore, an alternative approach was required to enhance Notch signaling preferentially in HEC. Apart from Stab2, Clec4g (LSECtin) is another protein expressed by LSEC, but not by vascular EC of most other organs (Supplementary Fig. S1). Therefore, transgenic mice with Clec4g promoter-driven Cre-expression were generated. Clec4g-Cre^tg/wt;Rosa26:eYFP^fl/wt reporter (Clec4g-eYFP reporter) mice displayed reporter activity in HEC from E13.5 onwards (Supplementary Fig. S2). At E13.5 only a small fraction of HEC showed endothelial positivity of YFP, whereas at E17.5 almost all microvascular HEC including LSEC and central vein EC were YFP-positive. In adult liver, activity was seen in all microvascular HEC, but not in Kupffer cells or stellate cells or any other cell types. In other adult tissues, activity was observed in some EC of the heart, whereas EC of the lung, kidney, bone marrow (BM), and spleen were negative (Fig. 1A; Supplementary Fig. S2). In addition, a small number of nonendothelial cells also showed reporter activity in different tissues except for the liver. Costainings with CD45, CD68, CD11b, c-Kit, TER-119, and CD61 showed that these cells do not belong to a homogeneous cell population and may represent reporter activation in a small fraction of precursor cells (Supplementary Fig. S2). Clec4g-Cre^tg/wt mice were then crossed to Rosa26^{N1ICD-IRES-GFP} mice to generate Clec4g-Cre^tg/wt;Rosa26^{N1ICD-IRES-GFP} mice (NICD^OE-HEC). NICD^OE-HEC mice were viable (Fig. 1B) and displayed transgenic nuclear enhanced GFP (nEGFP) expression in HEC, indicating NICD expression, whereas all other cell types of the liver remained negative (Fig. 1C; Supplementary Fig. S3).

Because endothelial Notch signaling is known to affect metastasis to the lung (30), models of hepatic metastasis of melanoma and colorectal carcinoma were studied in NICD^OE-HEC mice. In two different melanoma models, liver colonization was strongly reduced in NICD^OE-HEC mice. Wt31 melanoma cells that were injected either intravenously (Fig. 1D) or intrasplenically (Supplementary Fig. S4) showed fewer hepatic metastasis (P = 0.0022). In the B16F10 melanoma model, no visible metastases were detected in NICD^OE-HEC mice after intrasplenic injection as compared with controls (Ctrl; P = 0.0014; Fig. 1E). BLI confirmed the absence of hepatic metastases after injection of B16F10 melanoma into NICD^OE-HEC mice (P = 0.0047; Fig. 1F). Furthermore, reduced liver metastasis was also observed after intrasplenic injection of the colorectal carcinoma cell line MC38 (P = 0.0008; Fig. 1G). Therefore, activation of Notch signaling in HEC strongly reduced metastasis to the liver. These findings demonstrate that Notch activation in lung EC and HEC has opposing effects on metastasis to the respective organs.

To understand how endothelial Notch activation affects metastatic susceptibility, NICD^OE-HEC mice were analyzed in detail. Mutant mice appeared grossly normal and had a normal total body weight. A macroscopic screening of the internal organs did not reveal obvious alterations of the brain, kidney, and lungs (Supplementary Fig. S5). However, liver/body weight ratio was significantly reduced in NICD^OE-HEC mice (P < 0.0001, P = 0.014; Fig. 2A) and most of the livers showed slightly dilated vessels on the surface (Fig. 2B). A moderate increase in spleen/body weight ratio and lung/body weight ratio was observed (Supplementary Fig. S5). The heart/body weight ratio was increased, as well, but both heart function and wall structure were not altered (Supplementary Fig. S5). Routine histochemical stainings (H&E, PAS, Sirius Red) revealed no increased inflammation or major fibrosis in NICD^OE-HEC livers (Supplementary Fig. S5). Clinical chemistry did not reveal abnormal levels of hepatic enzymes (AST, ALT, GLDH), total protein, glucose, or triglycerides in mutant mice, whereas cholesterol was significantly reduced and CHE was significantly increased in NICD^OE-HEC mice (Fig. 2C; Supplementary Fig. S5). Expression of Hmgcr and Cyp7a1, which represent key enzymes of cholesterol synthesis, however, were not altered in NICD^OE-HEC mice (Supplementary Fig. S6). NICD^OE-HEC mice displayed significantly elevated bilirubin levels (Supplementary Fig. S6). This correlated with higher expression levels of the hepatic enzyme Hmox1, an essential enzyme-mediating heme catabolism, in NICD^OE-HEC mice (Supplementary Fig. S6). Increased Hepcidin expression (qRT-PCR) and normal Prussian blue staining were observed in NICD^OE-HEC livers (Supplementary Fig. S6), indicating that the Bmp-2/Bmp-6/hepcidin axis, which controls iron homeostasis, was functional in NICD^OE-HEC.

Recently, we have demonstrated that liver growth and liver/body weight ratio are controlled by angiocrine Wnt signaling (25). As NICD^OE-HEC mice also displayed a reduced liver/body weight ratio, we hypothesized that angiocrine signaling might be altered. Therefore, HEC-derived angiokines Hgf, Bmp2, Wnt2, and Wnt9b were analyzed by in situ hybridization in NICD^OE-HEC and Ctrl livers (Fig. 3A–D). NICD^OE-HEC mice showed a significant decrease in hepatic Wnt2 and Wnt9b expression, whereas Hgf and Bmp2 were not altered. To assess whether angiocrine Wnt signaling was indeed significantly reduced, β-catenin-target genes were analyzed in HCs. Axin2 expression was diminished in pericentral HC of NICD^OE-HEC mice as assessed by qRT-PCR and in situ hybridization (Fig. 4A and B). Likewise, marker proteins of metabolic liver zonation such as Glul, RhBg, and CYP2E1, which are characteristic for pericentral HC were decreased in NICD^OE-HEC mice (Fig. 4C). Vice versa, Arg1 and E-Cadherin expression was extended in NICD^OE-HEC mice (Fig. 4C). As expected, membranous β-catenin expression was found in HC of NICD^OE-HEC and Ctrl mice (Fig. 4D). In summary, enhanced Notch activity in NICD^OE-HEC mice led to inhibition of angiocrine canonical Wnt signaling followed by impairment of metabolic zonation and growth of the liver.

Because of the regulation of angiocrine Wnt secretion by hepatic endothelial Notch activation, we hypothesized that the reduced metastatic susceptibility observed in NICD^OE-HEC mice could also be mediated by alterations of the angiocrine Wnt/β-catenin axis. To test this hypothesis, hepatic metastasis was studied in Wls^HEC-KO mice with endothelial deficiency of the Wnt cargo receptor Wls. These mice show abolished angiocrine Wnt ligand secretion and a similar phenotype in regard to metabolic zonation and liver growth as NICD^OE-HEC (25). Wt31 melanoma cells were injected into the tail vein and hepatic metastasis was quantified. Here, no difference in Wt31 melanoma metastasis to the liver was observed (P = 0.6580; Fig. 4E). This indicates that the reduced metastatic susceptibility in NICD^OE-HEC mice was neither directly mediated by decreased angiocrine Wnt ligand secretion from HEC nor by its sequelae, that is, impaired metabolic zonation and decreased liver size.

To identify Notch-dependent endothelial alterations beyond angiocrine Wnt signaling, primary LSEC isolated from Ctrl and NICD^OE-HEC mice were compared by microarray gene expression profiling. Overall, 311 genes were significantly upregulated with a fold change (FC) > 2 and P < 0.05 whereas 197 were downregulated with a FC < 0.5 and P < 0.05. Activation of Notch was evident by strong upregulation of Hey1 and Hes1, which was confirmed by qRT-PCR and in situ hybridization (Fig. 5A–D; Supplementary Fig. S6). To comprehensively assess endothelial gene regulation, a heat map containing LSEC-associated and continuous EC-associated genes was analyzed. Clusters of genes characteristic of liver sinusoidal versus continuous EC were mostly changed in opposite directions with sinusoidal genes being mostly downregulated whereas continuous EC genes were mostly upregulated in NICD^OE-HEC (Fig. 5E). A subset of these EC genes regulated by Notch signaling was further studied in more detail. Upon immunofluorescent analysis, NICD^OE-HEC mice showed increased endothelial expression of CD31, Emcn, and Lyve1 throughout the liver lobule in contrast to the zonated expression pattern of these markers seen in Ctrl mice (Fig. 6A). The increase of Cd31, Emcn, and Lyve1 was confirmed by qRT-PCR (Supplementary Fig. S7). In addition, a moderate increase in expression of PODXL was observed (Fig. 6A). In contrast, CD32b and Stab2 expression strongly decreased (Fig. 6B). Expression of the adherens junction protein VE-Cadherin did not differ between NICD^OE-HEC and Ctrl mice (Supplementary Fig. S7). As alterations of the extracellular matrix could promote metastasis (15), its composition was also analyzed. Although periendothelial deposition of collagen 1, collagen 3, collagen 4, and fibronectin was not altered, Lama4 was strongly increased in the subendothelial space of Disse (Fig. 6C and D; Supplementary Fig. S8; Supplementary Table S2). Thus, balanced Notch activity controls the zonated expression of LSEC membrane proteins and production of matrix molecules such as Lama4. This process is independent of the Wnt/β-catenin axis as Wls^HEC-KO do not show effaced LSEC zonation. As Lama4 can activate stellate cells (29), the numbers of Desmin-positive stellate cells were assessed. Here, no difference in the number of stellate cells was found (Fig. 6C and D).

As Notch activation significantly altered the expression of surface proteins in HEC, we extended our analysis to additional adhesion molecules potentially involved in metastasis. Indeed, ICAM1 expression was significantly downregulated in HEC of NICD^OE-HEC mice (Fig. 7A), whereas VCAM1 expression increased (Fig. 7B). Downregulation of ICAM1 was also independent of the Wnt/β-catenin axis as ICAM1 did not differ between Wls^HEC-KO and Ctrl mice (Fig. 7C and D).

Altered expression of surface molecules involved in tumor cell adhesion in NICD^OE-HEC mice, but not Wls^HEC-KO indicated that the lower metastatic susceptibility in NICD^OE-HEC livers may be explained by reduced tumor cell adhesion. To test this hypothesis, initial adhesion and retention analyses were performed in vivo. Livers and lungs were assessed 90 minutes after intrasplenic injection of B16F10 melanoma cells. Here, NICD^OE-HEC mice showed decreased melanoma cell retention in the liver in comparison to Ctrl (P = 0.0022; Fig. 7E). However, increased lung colonization (P = 0.0006; Supplementary Fig. S4) at day 14 and higher numbers of tumor cells adhering to the lungs after 90 minutes (P = 0.0022; Supplementary Fig. S4) were found in NICD^OE-HEC mice after intrasplenic injection of B16F10 melanoma cells. Similar results were obtained after intrasplenic injection of Wt31 melanoma cells (Supplementary Fig. S4; Supplementary Table S3). Importantly, no differences in lung colonization were observed after intravenous injection of B16F10 melanoma cells into the tail vein of NICD^OE-HEC and Ctrl mice excluding primary differences in the metastatic susceptibility of the lungs of Ctrl and NICD^OE-HEC mice (Supplementary Fig. S4). These findings indicate that adhesion and retention of tumor cells in the hepatic vascular bed in NICD^OE-HEC mice are decreased whereas passage to the lungs is increased.

To test whether downregulation of ICAM1 could mediate the decreased tumor cell adhesion and retention observed in NICD^OE-HEC mice, wild-type (Wt) mice were treated with either anti-ICAM1 antibody or the corresponding isotype control antibody. Indeed, anti-ICAM1 antibody treatment decreased ICAM1 expression in LSEC (Fig. 7F) and reduced tumor cell retention in the liver (P = 0.0262; Fig. 7G).

Altogether, Notch1 activation in HEC reduced endothelial tumor cell adhesion and retention and finally metastatic burden in the liver presumably mediated by reduced ICAM1 expression on HEC. These findings are in line with prior results demonstrating that inhibition of Notch signaling increases hepatic metastasis (31). In contrast to our work, however, Banerjee and colleagues did neither prove that this effect was EC-specific nor demonstrate that Notch activation provides protection from metastasis in vivo. As increased endothelial Notch signaling in loco can facilitate metastasis to the lung (30), our data highlight the organ-specific, opposing role of endothelial Notch signaling in controlling the organotropism of metastatic spread to the liver.

Hepatic metastasis still represents a harbinger of potentially fatal tumor progression in many cancers. Despite considerable progress, metastatic spread to the liver is associated with a poor prognosis, and often treatment options are limited. As endothelial Notch activation increases pulmonary metastasis in malignant melanoma and lung cancer (30), targeting endothelial Notch activity has recently been suggested as a therapeutic modality to inhibit lung metastasis. In contrast, Notch inhibition in the liver was demonstrated to increase metastasis (31). Because of the use of pharmacologic Notch inhibitors as well as of global, but not cell type-specific, heterozygous deficiency of Notch in this study, Notch effects on liver metastasis in these models could not be attributed to Notch inhibition in EC as compared with other cellular constituents of the hepatic niche. Therefore, we here aimed to elucidate the role of endothelial Notch signaling for hepatic metastasis.

Previously, conditional, but not cell type–specific deletion of the Notch target protein RBP-J as well as of Notch1 itself was shown to cause vascular defects in the liver (36, 37). In an inducible, global EC-specific model vascular defects were indeed tied to diminished endothelial Notch activity (38). Conversely, enhanced endothelial Notch signaling did not cause gross hepatic alterations (39). However, deficiency of Notch ligands Dll4 or Jagged1 led to embryonic lethality due to major vascular defects (40). In line with these findings, Stab2-iCre^tg/wt;Rosa26^{NICD-IRES-GFP} mice in our study died in utero, indicating that onset of promoter activity and dosage effects determine the extent of the effects that Notch overexpression or deficiency may entail.

By generating the Clec4g-Cre model here, we provide a novel endothelial subtype-specific driver mouse line. In comparison to the Stab2-Cre model published by us recently (29), promoter activity of the Clec4g model as investigated in reporter mice starts later at E13.5 (Supplementary Fig. S2). The endothelial distribution pattern is more restricted mainly sparing lung, bone marrow, and kidney (Fig. 1A; Supplementary Fig. S2). With this mouse model, hepatic endothelial Notch signaling was successfully enhanced by overexpression of NICD (NICD^OE-HEC; Fig. 1B) without affecting Notch activity in EC of many other organs. The Clec4g-Cre model therefore offers a promising tool for future studies of LSEC biology in vivo. Focusing on cancer biology and metastatic spread to the liver, EC-subtype specific mouse models will be valuable tools to investigate effects of hepatic endothelial alterations on metastasis.

Studying metastatic organotropism to the liver, NICD^OE-HEC were significantly protected from hepatic metastasis in models of metastatic melanoma (Wt31, tail vein, and splenic injection; B16F10, splenic injection) and colorectal carcinoma (MC38, splenic injection). As putative mechanisms for reduced liver metastatis in NICD^OE-HEC mice, angiocrine functions were analyzed and downregulation of endothelial Wnt ligand expression was observed. This resulted in decreased liver/body weight ratio and effacement of metabolic liver zonation. Similar to our findings, Duan and colleagues demonstrated that Tamoxifen-inducible, short-term, pan-endothelial Notch activation reduced hepatic angiocrine Wnt functions. In contrast to our findings, however, hepatic angiocrine Wnt functions became relevant only upon challenging the liver while organ size and metabolic liver zonation remained unaffected (41). Compared with our Clec4g-Cre model, these reduced angiocrine Wnt effects can likely be explained by partial and mosaic-like transgene activation in HEC in Tamoxifen-inducible Cdh5-promoter-driven systems (23). Notably, metabolic alterations in NICD^OE-HEC mice were similar to those observed in Wls^HEC-KO (25). Contrastingly, hepatic metastatic susceptibility was only decreased in NICD^OE-HEC mice, but not Wls^HEC-KO mice. Therefore, it can be deduced that endothelial Notch signaling controls organotropic metastasis to the liver independently from angiocrine Wnt secretion in HEC.

In addition to the sequelae of angiocrine Wnt deficiency, NICD^OE-HEC mice showed effacement of LSEC zonation and features that were reminiscent of sinusoidal capillarization, that is, enhanced expression of CD31, Emcn, and Lama4. Sinusoidal capillarization is regularly observed in fibrotic/cirrhotic liver diseases (42) associated with fewer liver metastases. In addition, Notch activation has been described to occur in several hepatic cell types including ECs (37, 43, 44). In contrast to classical sinusoidal capillarization (45), however, NICD^OE-HEC showed increased endothelial expression of Lyve1 whereas expression of VE-Cadherin was not altered (Supplementary Fig. S7). Remarkably, effacement of LSEC zonation and overexpression of continuous EC-associated surface proteins were only present in NICD^OE-HEC mice, but not Wls^HEC-KO mice, indicating that angiocrine Wnt signaling is only one of several downstream targets of Notch activation in hepatic endothelium.

Previous reports have shown that Notch signaling regulates expression of various adhesion molecules (30, 31). In the liver, we identified downregulation of ICAM1 as an organ-specific hallmark of endothelial Notch activation. Functionally, downregulation of ICAM1 may preclude adhesion of B16F10 and Wt31 melanoma cells as well as MC38 colorectal carcinoma cells in the liver sinusoids of NICD^OE-HEC mice. Supporting this notion, increased expression of ICAM1 in HEC and higher serum levels of ICAM1 have been found to be associated with liver metastasis in malignant melanoma and various carcinomas (13, 46). In addition, decreased expression of the ICAM1 receptor LFA-1 on colorectal cancer cells causes a reduction in liver metastasis. As anti-ICAM1 treatment similarly reduced melanoma cell adhesion in Wt mice Notch-mediated downregulation is the likely mechanism by which endothelial Notch activation prevents tumor cell adhesion in the liver. Interaction of tumor cells with ICAM1 on HEC could either involve binding of inducible ligands of ICAM1 on tumor cells (47) or occur via linker molecules like Fibrin (48). Despite unaltered Notch signaling in lung EC, increased metastatic lung colonization was observed in NICD^OE-HEC after intrasplenic injection of B16F10 melanoma cells. This finding is likely caused by reduced adhesiveness of HEC and enhanced liver passage of melanoma cells, resulting in subsequent arrest in the vascular bed of the lungs. As lung EC are generally not affected by Notch activation in NICD^OE-HEC mice, the effect cannot be explained by intrinsic lung endothelial alterations. Moreover, this was excluded as no difference in lung colonization was observed after intravenous injection of B16F10 into the tail vein of NICD^OE-HEC and Ctrl mice. These findings indicate that Notch controls metastatic susceptibility in the liver via the control of tumor cell adhesion and retention.

In summary, enhanced endothelial Notch activation in the liver prevented metastasis, whereas enhanced endothelial Notch activation in the lung promoted metastasis. This demonstrates that endothelial Notch signaling affects metastasis to different organs based on pre-established, organotypic, endothelial differentiation programs. As deregulation of Notch is associated with cancer development, for example, in hepatocellular carcinoma (49), modulating Notch activity has been proposed as a valuable therapeutic target. In malignant melanoma, inhibition of Notch suppresses tumor cell growth and enhances treatment efficacy of targeted therapies (50). Although global Notch inhibition would likely protect from melanoma metastasis to the lungs, enhanced risk of liver metastasis would be a disadvantageous side effect. In contrast, selective enhancement of Notch signaling in HEC could prevent hepatic melanoma metastasis. Altogether, therapeutic translation will require approaches that target Notch in an organ-, cell-, and context-specific fashion to avoid unwanted side effects.

No potential conflicts of interest were disclosed.

Conception and design: S.A. Wohlfeil, C. Géraud, S. Goerdt

Development of methodology: S.A. Wohlfeil, M. Winkler, P.-S. Koch, S. Goerdt

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S.A Wohlfeil, V. Häfele, B. Dietsch, K. Schledzewski, M. Winkler, J. Zierow, T. Leibing, V. Olsavszky, P.-S. Koch, C. Géraud, S. Goerdt

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S.A Wohlfeil, V. Häfele, M. Winkler, J. Zierow, M.M. Mohammadi, J. Heineke, C. Sticht, V. Olsavszky, P.-S. Koch, C. Géraud, S. Goerdt

Writing, review, and/or revision of the manuscript: S.A Wohlfeil, K. Schledzewski, M. Winkler, J. Heineke, P.-S. Koch, C. Géraud, S. Goerdt

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S.A Wohlfeil, P.-S. Koch, S. Goerdt

Study supervision: C. Géraud, S. Goerdt

The authors thank Hiltrud Schönhaber, Stephanie Riester, Jochen Weber, Alexandra Demory, Günter Küblbeck, Cathleen Fichtner, and Carolina De La Torre for excellent technical support. We acknowledge both O. Sansom and S. Herzig for generously providing cells. For support with statistical analysis, we thank C. Weiß. We acknowledge the support of the Core Facility Live Cell Imaging Mannheim at the Centre for Biomedicine and Medical Technology Mannheim (German Research Foundation grant DFG INST 91027/9-1 FUGG and DFG INST 91027/10-1 FUGG). This work was supported in part by grants from the German Research Foundation (Deutsche Forschungsgemeinschaft/DFG) GRK2099/RTG2099, project 7 (to C. Géraud and S. Goerdt), and SFB TR23, project B1 (to S. Goerdt and C. Géraud). S.A. Wohlfeil was supported by the Translational Physician Scientist Program of the Medical Faculty Mannheim, Heidelberg University (funded by the Ministry of Science, Research and the Arts, Baden-Wuerttemberg).

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.