Endothelial Zeb2 preserves the hepatic angioarchitecture and protects against liver fibrosis

Abstract Aims Hepatic capillaries are lined with specialized liver sinusoidal endothelial cells (LSECs) which support macromolecule passage to hepatocytes and prevent fibrosis by keeping hepatic stellate cells (HSCs) quiescent. LSEC specialization is co-determined by transcription factors. The zinc-finger E-box-binding homeobox (Zeb)2 transcription factor is enriched in LSECs. Here, we aimed to elucidate the endothelium-specific role of Zeb2 during maintenance of the liver and in liver fibrosis. Methods and results To study the role of Zeb2 in liver endothelium we generated EC-specific Zeb2 knock-out (ECKO) mice. Sequencing of liver EC RNA revealed that deficiency of Zeb2 results in prominent expression changes in angiogenesis-related genes. Accordingly, the vascular area was expanded and the presence of pillars inside ECKO liver vessels indicated that this was likely due to increased intussusceptive angiogenesis. LSEC marker expression was not profoundly affected and fenestrations were preserved upon Zeb2 deficiency. However, an increase in continuous EC markers suggested that Zeb2-deficient LSECs are more prone to dedifferentiation, a process called ‘capillarization’. Changes in the endothelial expression of ligands that may be involved in HSC quiescence together with significant changes in the expression profile of HSCs showed that Zeb2 regulates LSEC–HSC communication and HSC activation. Accordingly, upon exposure to the hepatotoxin carbon tetrachloride (CCl4), livers of ECKO mice showed increased capillarization, HSC activation, and fibrosis compared to livers from wild-type littermates. The vascular maintenance and anti-fibrotic role of endothelial Zeb2 was confirmed in mice with EC-specific overexpression of Zeb2, as the latter resulted in reduced vascularity and attenuated CCl4-induced liver fibrosis. Conclusion Endothelial Zeb2 preserves liver angioarchitecture and protects against liver fibrosis. Zeb2 and Zeb2-dependent genes in liver ECs may be exploited to design novel therapeutic strategies to attenuate hepatic fibrosis.


Mouse experiments were approved by the KU Leuven Animal Ethics Committee (Ethics Committee
Human liver biopsies used for immunostaining were collected in Gey's buffer after obtaining informed consent from the donors (patients undergoing elective cholecystectomy at the University Hospital UZ Leuven). Biopsies were fixed overnight in 4% paraformaldehyde (PFA), embedded in paraffin and sectioned for immunofluorescence (IF) staining (primary and secondary antibodies used for IF staining are listed in Table S1). Biopsies were stained with Sirius Red and none of the biopsies showed fibrosis. Human umbilical vein ECs (HUVECs) were isolated as previously described 7 from umbilical cords of babies delivered by Caesarean section after full-term pregnancy with informed consent from the mother. The use of human biopsies, umbilical cords and human cells was approved by the Ethics Committee of University Hospitals Leuven (No B32220152525871) and experiments were performed in accordance with the Committee's guidelines and the principles of the Declaration of Helsinki.

Maintenance and hepatotoxic models
To study the effect of endothelial Zeb2 on (vascular) maintenance, tamoxifen-treated mice were euthanised under anaesthesia as described above at 1, 2, or 4 weeks after the last tamoxifen injection (Fig. S1D). Unless indicated otherwise, mRNA expression changes are shown at 2 weeks posttamoxifen and protein expression or structural changes at 4 weeks post-tamoxifen. To study the response to acute liver injury, mice received one i.p. injection with high-dose CCl4 (0.6 µL/g in mineral oil) or with mineral oil alone as control (vehicle) and were sacrificed 24 hours later (Fig.  S1E). To study the effect of mild fibrosis, mice were injected with low-dose CCl4 (0.2 µL/g in mineral oil) or vehicle 3 times with one day in between and were sacrificed 24 hours (progression cohort) or 1 week (regression or 'R' cohort) after the last injection (Fig. S1F). For chronic (septal) fibrosis, mice received high-dose CCl4 (0.6 µL/g in mineral oil) or vehicle 3 times per week (Monday-Wednesday-Friday) for 4 weeks and were sacrificed 24 hours (progression cohort) or 1 week (regression or 'R' cohort) after the last injection (Fig. S1G). For (immuno)histological analyses, mice were euthanised under anaesthesia as described above, livers were subsequently perfusion-fixed with zinc-formalin (Sigma) and processed for paraffin sectioning. For other analyses, tissues were isolated, snap-frozen and stored at -80°C until further use.

Cell isolation and gene profiling
To study the organ-specific Zeb2 expression, GFP + ECs were isolated from Tie2-GFP hearts, brains and livers (yielding > 95% pure populations consisting for > 99% of microvascular ECs) 8,9 and comparative gene expression was performed by quantitative (q) real-time (RT)-PCR. To establish primary cultures for fenestrae analysis, livers were digested and plated in EGM2 medium supplemented with EGM2-MV (containing 0.5 ng/mL VEGF-A; Lonza) onto gelatin-coated cell culture vessels, cells were allowed to settle for 36 hours, washed and fixed with 2.5% glutaraldehyde in 0.1 M sodium cacodylate buffer. To simultaneously isolate the four main hepatic cell types, EC KO and their WT (Cre-negative) littermates were perfused via the portal vein for 5 minutes with SC-1 solution (8 g/L NaCl, 400 mg/L KCl, 75.5 mg/L NaH2PO4.H2O, 120.45 mg/L Na2HPO4, 2.38 g/L HEPES, 350 mg/L NaHCO3, 190 mg/L EGTA, 900 mg/L glucose, pH 7.3), followed by 5 minutes 0.04% pronase E (Merck) and 5 minutes 0.02% collagenase P (Boeringer-Manheim) both in 8 g/L NaCl, 400 mg/L KCl, 75.5 mg/L NaH2PO4.H2O, 120.45 mg/L Na2HPO4, 2.38 g/L HEPES, 350 mg/L NaHCO3, 560 mg/L CaCl2.2H2O, pH 7.3). The liver was excised and incubated at 37°C for 15 minutes in a solution containing 0.03% collagenase P, 0.03% pronase E and 0.001% DNase (Grade II, Boeringer-Manheim). The resulting suspension was filtered through a 100 µm strainer and centrifuged 2 times for 2 minutes at 50 g to separate the hepatocytes (HEPs) from the nonparenchymal cell (NPC) fraction. Hepatocytes were immediately snap-frozen after centrifugation. Red blood cells were removed from the NPC fraction by incubation with red blood cell lysis buffer  Table S1.
For RNA sequencing, RNA was isolated from ECs, HSCs, Kupffer cells and hepatocytes from WT and EC KO mice (n=2 per cell type and genotype) using Reliaprep (Promega). RNA samples were quality controlled and prepared according to the Smart-seq2 method. 10 In brief, polyA + RNA was reverse transcribed using an oligo(dT) primer. Template switching by reverse transcriptase was achieved using an LNA-containing TSO oligo. The reverse transcribed cDNA was pre-amplified with primers for 18 cycles followed by clean-up. Tagmentation was performed on 500 pg of the preamplified cDNA with Tn5 followed by gap repair. The tagmented library was extended with Illumina adaptor sequences by PCR for 14 cycles and purified. The resulting sequencing library was measured on a Bioanalyser and equimolar loaded onto a flow cell and sequenced according to the Illumina TruSeq v3 protocol on the HiSeq2500 with a single read 50 bp and dual 9 bp indices. Illumina adapter sequences and poly-A stretches were trimmed from the reads. 11 The remaining sequences were aligned to the mouse GRCm38 reference sequence using HISAT2 (version 2.1.0). 12 Transcript abundance level (transcript count) was generated using HTSeq (version 0.9.1) based on the ENSEMBL 84 gene annotation. 13 The transcript counts were further processed using R software environment for statistical computing and graphics (version 3.4.0). Data normalisation, removal of batch effect and other variance was performed using EDASeq R and RUVSeq packages. 14,15 Differential expression analysis was performed using the edgeR R package, 16 applying the negative binomial general linear model (GLM) approach. Differentially expressed genes with false discovery rate (FDR) < 0.05, Benjamini-Hochberg multiple testing correction and expression level in WT samples of > 1 counts per million (CPM) were retained and used for further processing (i.e., principal component analysis, cluster analysis, Volcano plots, heat maps, Gene Enrichment analysis by Enrichr 17 and ligand-target prediction analysis by NicheNet (using an open source R implementation; https://github.com/saeyslab/nichenetr). 18 The RNA sequencing datasets generated for this study are available in the NCBI GEO repository (https://www.ncbi.nlm.nih.gov/geo/), under series number GSE150699. To obtain RNA from whole liver tissue, the liver was homogenised in TRIZol and RNA was isolated according to manufacturer's instructions. For quantitative qRT-PCR, cDNA was made using the GoScriptÔ reverse transcription system (Promega) according to the manufacturer's protocol. QRT-PCR was performed with the ABI system using Sybr green (ABI). Gapdh was used as reference gene after validation (Note S1; for primer sequences, see Table S2).

In vitro angiogenesis assays and lentivirus-mediated knock-down
HUVECs were cultured in EBM2 medium supplemented with EGM2-MV (Lonza) in gelatin-coated culture flasks. Plasmids containing ZEB2-shRNA or GFP (control) in a lentiviral backbone were purchased from Sigma and lentiviruses were produced in human embryonic kidney (HEK)293 cells.
Obtained viruses were titrated on HUVECs and the lowest concentration leading to 100% transduction efficiency was used in subsequent experiments. Cells were transduced with viruses, medium was changed, 1 and 3 days after transduction and on day 6 cells were either lysed in TRIzol for RNA isolation or passaged for functional analyses.
For proliferation assays, 5,000 cells were plated per well on gelatin-coated 24-well plates, cells were counted 6 days later and doubling time was calculated. For tube formation assays, 50,000 cells were plated on Matrigel-coated 24-well plate and tube formation was assessed 24 hours later. Branch length was quantified using the angiogenesis analyser plug-in for Image J (NIH). For scratch wound assays, 50,000 cells were plated in a 24-well plate, and 24 hours later when confluence was reached, a standardised scratch was made and the gap distance between the cells was measured immediately after scratching and again 24 hours later. To assess chemotactic migration, 25,000 cells were seeded in starvation medium onto collagen type I-coated Boyden chambers with 8 micron pores (Costar) placed into 24-wells filled with EBM2-EGM2-MV medium and cells were allowed to migrate for 24 hours, as described. 19 Cells were stained with Wright-Giemsa solution (Sigma) and migrated cells were counted. To evaluate sprouting, spheroids were made by incubating 600 HUVECs per spheroid for 24 hours in hanging droplets containing 0.24 g/L methylcellulose in EBM2-EGM2-MV medium. Spheroids were embedded in 2 mg/mL collagen (Corning) and collagen gels were covered with EBM2-EGM2-MV medium. Sprouting was analysed 24 hours later.

Scanning electron microscopy
To perform morphometric analysis on vascular corrosion casts, mice were euthanised under anaesthesia as described above and livers were consecutively perfused with 50 U/mL heparin in PBS, 2.5% glutaraldehyde in PBS and VasQtec resin (VasQtec Zurich, Switzerland; prepared according to manufacturer's instructions). Livers were isolated and kept in 2.5% glutaraldehyde in PBS at 40°C to allow the resin to harden and livers were subsequently saponified in 5% KOH at 40°C for 2-5 days, casts were dried and sputter-coated with gold and images were recorded on a JEOL scanning electron microscope at 0.

Ultrafast ultrasound imaging
Ultrafast ultrasound acquisition was performed on the left (lateral) liver lobe at 2 or 4 weeks after the last tamoxifen injection on a research ultrafast platform (INSERM "Biomedical Ultrasound" ART) comprising an ultrasound acquisition board (128 channels, 62.5 MHz, Verasonics, Vantage 256) and a 15 MHz ultrasonic probe (pitch 0.11 mm, 128 elements, Vermon, France) mounted on 4 motors (3 translation and 1 rotation, Pi, Germany) (Fig. S8). The scanner was driven by a real-time Doppler acquisition software platform (INSERM "Biomedical Ultrasound" ART and Iconeus, Paris, France). Mice were anaesthetised with ketamine (75 µg/g i.p.) and medethomide (1 µg/g i.p.). Anaesthesia depth was checked by toe pinch and, if necessary, mice received another i.p. injection with 7.5 µg/g ketamine and 0.1 ug/g medothomide. A 3-dimensional printed water tank consisting of TPX polymethylpentene film was placed on the mouse abdomen with transmission gel between the mouse skin and the film and the tank was filled with water. The probe was placed in the water tank in contact with the liver and several planes of the lateral left liver lobe were acquired every 0.5 mm. The animals received subcutaneously atipamezole (1 µg/g) to reverse the anaesthesia, the average imaging time was 45 minutes per mouse. The ultrasensitive ultrasound sequence consisted of 11 compounded plane waves (angles from -10° to 10°) designed to acquire a full 30 seconds at a continuous framerate set to 500 Hz. A singular value decomposition clutter filter was applied on blocks of 200 frames (400 ms) to separate blood signal from tissue signal 21 by removing the 60 largest singular values. In order to filter different flow velocities, additional bandpass frequency filters were applied on the 400 ms subsets of data acquisition, giving access to blood volume movies corresponding to three different velocity ranges (2 to 4 mm/s, 4 to 6 mm/s and 6 mm/s and upper). 22 Those high framerate power-Doppler movies were computed by integrating the energy of windows of 20 consecutive filtered ultrafast frames (40 ms). Breathing motion (which can induce significant out of plane movements and artefacts) was eliminated by performing frame classifications on the high frame-rate power-Doppler movies. For this purpose, a k-means clustering algorithm (2 clusters, correlation distance metric, 100 replicates) was applied to the power-Doppler movies in order to remove the outlier images. The remaining frames were then median-averaged to construct one power Doppler image per band-speed. Regions of interest (ROI) were manually drawn based on the anatomical and vascular landmarks of the liver and the median value of each ROI was estimated for the different bands.

Morphometric analysis and assessment of liver fibrosis and function
For histology, mice were euthanised under anaesthesia as described above and perfusion-fixed via the heart with zinc-formalin (Sigma), the left liver lobe, heart and brain were dissected out, fixed overnight in zinc-formalin, dehydrated, embedded in paraffin and 7 µm sections were prepared. To assess general morphology and liver damage, sections were stained with haematoxylin and eosin (H&E) and collagen was stained using Sirius red, Masson's trichrome (Sigma HT15) or antibodies against fibrillar Collagen types I and III. To analyse the zonated architecture of the liver vasculature, we used antibodies against cytokeratin (Ck)19 and Endomucin to identify portal triads and central veins, 23 respectively. To further characterise the reaction to liver repair/damage, sections were stained with antibodies against a-smooth muscle actin (aSMA), Desmin and the inflammatory marker Cd45.
To characterise endothelial changes, IF stainings were performed for Pan-endo, Cd31, Cd32, Cd34, vWF, Laminin, Lyve1, Endomucin, Wheat Germ Agglutinin (WGA) lectin and Collagen type IV. To assess Cre-mediated recombination in EC KO mice or demonstrate endothelial expression of Zeb2 in Zeb2-eGFP reporter mice, livers were co-stained for eGFP and ETS-related gene (Erg). Alternatively, recombination was assessed by FACS on monocellular liver suspensions stained for Pan-endo. To analyse recombination in different zones, recombined ECs were counted on serial cross-sections stained for Erg/eGFP and pericentral hepatocyte zonation marker Cytochrome P450 Family 2 Subfamily E member 1 (Cyp2e1). Hepatocyte zonation was analysed by IF staining using antibodies against pericentral markers Cyp2e1 or Glutamate ammonia ligase (Glul) and periportal/midzonal marker Arginase1 (Arg1). Antibodies were tested using a negative control (no primary antibody) and WT healthy mouse liver sections as positive control where we compare the expression pattern with stainings found on the manufacturers' websites and in literature; for antibodies and concentrations see Table S1. Where necessary, amplification was performed using a Cy3-or fluorescein-tyramide kit according to the manufacturer's instructions (Perkin Elmer). To demonstrate endothelial Zeb2 expression on human liver paraffin sections, these were co-stained for ZEB2 and Erg (Table S1). Microscopy images were analysed in Image J and FACS data were analysed with FACS DIVA software. To assess fibrosis by hydroxyproline content, livers were hydrolysed in 6 M HCl at 110°C overnight. Liver homogenates were dried and dissolved in 50% isopropanol. Hydroxyproline was detected after oxidation by chloramine T using Ehrlich reagent. To estimate liver function, plasma alanine transferase (ALT) was measured using a Spotchem EZ system analyser (Arkray) and GPT/ALT strips (Menarini diagnostics).

Statistics
Data are expressed as mean ± standard error of the mean (sem), unless stated otherwise. Student's ttest was used to compare 2 groups. One-way ANOVA with Bonferroni post-hoc test was used to compare > 2 groups. P<0.05 was considered statistically significant. Graphpad Prism 8 was used for statistical analysis.