Well-differentiated liver cancers reveal the potential link between ACE2 dysfunction and metabolic breakdown

Angiotensin-converting enzyme 2 (ACE2) is the receptor of the Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) causing Coronavirus disease 2019 (COVID-19). Transmembrane serine protease 2 (TMPRSS2) is a coreceptor. Abnormal hepatic function in COVID-19 suggests specific or bystander liver disease. Because liver cancer cells express the ACE2 viral receptor, they are widely used as models of SARS-CoV-2 infection in vitro. Therefore, the purpose of this study was to analyze ACE2 and TMPRSS2 expression and localization in human liver cancers and in non-tumor livers. We studied ACE2 and TMPRSS2 in transcriptomic datasets totaling 1503 liver cancers, followed by high-resolution confocal multiplex immunohistochemistry and quantitative image analysis of a 41-HCC tissue microarray. In cancers, we detected ACE2 and TMPRSS2 at the biliary pole of tumor hepatocytes. In whole mount sections of five normal liver samples, we identified ACE2 in hepatocyte’s bile canaliculi, biliary epithelium, sinusoidal and capillary endothelial cells. Tumors carrying mutated β-catenin showed ACE2 DNA hypomethylation and higher mRNA and protein expression, consistently with predicted β-catenin response sites in the ACE2 promoter. Finally, ACE2 and TMPRSS2 co-expression networks highlighted hepatocyte-specific functions, oxidative stress and inflammation, suggesting a link between inflammation, ACE2 dysfunction and metabolic breakdown.


Results
Higher expression of ACE2 and TMPRSS2 mRNAs in HCCs with mutated β-catenin. We first searched for relationships of ACE2 and TMPRSS2 mRNA expression levels with HCC aggressiveness. To this end, overall and disease-free survival analyses were carried out in the TCGA dataset (n = 370). After applying the exclusion criteria described in Supplementary Fig. 1a, the survival dataset consisted of 256 patients. In consistency with previous reports [21][22][23] , higher ACE2 and TMPRSS2 mRNA expression were associated with better overall and disease-free survival (Supplementary Fig. 1b and 2a). ACE2 and TMPRSS2 mRNAs were also detected in 47 non-tumor livers in the TCGA dataset. Whereas ACE2 mRNA expression was 1.7-fold higher in non-tumors livers than in HCCs, TMPRSS2 expression did not differ between both groups ( Supplementary Fig. 2b).
In HCCs, the expression of genes associated with relatively good prognosis is frequently preserved in welldifferentiated tumors because they maintain the phenotype of adult hepatocytes 24,25 . To verify if this applied to ACE2 and TMPRSS2, we measured their expression levels in HCCs classified according to their likeness to normal hepatocytes. Thus classified, HCCs are divided into two major classes, namely "low-proliferation" and "high-proliferation" 25 . Four HCC subclasses result from the interaction of, on one hand, the metabolic phenotype and, on the other hand, the proliferation/differentiation ratios of tumor cells 24,26 . Hence, in the "low-proliferation" class of well-to-moderately differentiated HCCs, the periportal-type (PP) and perivenous-type (PV) subclasses refer to the preservation of the metabolic zonation functions of normal hepatocytes 24,26 . At the opposite end of the spectrum, in the "high-proliferation" class of moderately-to-poorly differentiated HCCs, the extracellularmatrix-type (ECM) and STEM-type subclasses, refer to tumors developing an important stromal and vascular support and expressing cancer stem cell markers 24,26 . In this context, we found higher ACE2 mRNA expression in perivenous-type HCCs than in the other subclasses ( Fig. 1a; Supplementary Fig. 2c) in both the Désert's cDNA microarray meta-dataset (1133 HCCs) 26 (Fig. 1a) and in the TCGA dataset (370 HCCs) (Supplementary Fig. 2c). Of note, ACE2 mRNA expression was higher in HCCs expressing previously validated β-catenin pathway activation signatures 26,27 (Fig. 1b) and in HCCs carrying sequenced β-catenin (CTNNB1) activating mutations ( Supplementary Fig. 2c). In turn, TMPRSS2 was expressed at higher levels in periportal-type HCCs ( Supplementary Fig. 2c, d), which we previously demonstrated to carry wild-type CTNNB1 and to belong to the "low-proliferation" class of well-to-moderately differentiated HCCs that preserve the periportal metabolic functions of normal hepatocytes 26 . The MERS-CoV receptor, DPP4, was also related to β-catenin pathway activation (Supplementary Fig. 2e-g). ACE2 DNA was hypomethylated in tumors carrying CTNNB1 mutations (Fig. 1c), which is consistent with transcriptional activation the ACE2 gene. By contrast, TMPRSS2 was hyper-methylated in HCCs carrying CTNNB1 mutations (Supplementary Fig. 3a; Supplementary Table 1), which is consistent with transcriptional repression. Neither ACE2 nor DPP4 mRNA expression were related TERT and TP53 gene mutations, which are the two most frequent mutations in HCCs ( Supplementary Fig. 3c, d). In turn, TMPRSS2 mRNA levels were higher in HCCs carrying wild-type TP53 ( Supplementary Fig. 3d), which is consistent with the above-described higher expression of TMPRSS2 in the well-differentiated, non-proliferative class of HCCs, where TP53 gene mutations are rare 26,28 . Indeed, TP53 mutations are most frequent in poorly-differentiated HCCs (S2 29 ; G3 30 ), showing an aggressive, cancer stem cell phenotype 26,31 .
In line with the higher ACE2 and DPP4 mRNA expression in HCCs showing increased β-catenin pathway activation and CTNNB1 activating mutations, in silico analysis of 5,000 base pairs upstream of the transcription start sites of ACE2 and DPP4 DNAs revealed putative T-cell factor 4/LEF-1 consensus transcription factor binding sites, respectively ( Fig. 1d; Supplementary Fig. 4a, b).
ACE2 is predominantly immunodetected in perivenous-type HCCs carrying CTNNB1 mutations and preserving hepatocyte polarization. In consistency with the above findings, scoring of immunohistochemical signal for ACE2 and for the β-catenin target gene GLUL (glutamine synthetase) in a tissue microarray of 41 HCCs (Fig. 1e,  www.nature.com/scientificreports/ mRNA expression data of five genes as follows: β-catenin activation score = [(GLUL × LGR5 ×ODAM) ÷ (VNN1 × HAL)] 26 . Thus, we measured mRNA expression for these five genes by real-time PCR on mirror frozen samples from the same HCCs as those included in the TMA, applying our previously described method 26,32,33 . GLUL and ACE2 protein expression were correlated with β-catenin activation scores (GLUL, R = 0.69, p < 0.001, n = 41; ACE2, R = 0.51, p = 0.001, n = 38). β-catenin activation scores ranged from − 2 to 20, median = 4 (Supplementary Table 5). As previously described 26 , high scores (above the median), define perivenous-type HCCs. On average, GLUL and ACE2 proteins were respectively 4.0-and 2.6-fold higher in HCCs with high β-catenin activation scores (Fig. 1g). These results indicate that HCCs preserving a perivenous-type metabolic phenotype express higher level of ACE2 protein 26 . HCCs with mutated CTNNB1 belong to the class of well-to-moderately differentiated tumors that preserve adult hepatocyte features 25 . Thus, they are frequently cholestatic and contain pseudo-glandular structures with bile plugs, where hepatocytes are polarized and preserve the basal and apical poles 31 . Co-immunolabeling for ACE2 and ABCC2 (a.k.a. MRP2, a marker of the biliary pole in hepatocytes) in HCCs, detected ACE2 at the biliary pole and within the lumen of pseudo-glandular tumor structures and in slit-like bile canaliculi (Fig. 2a,b). By contrast, ACE2 was not detected at the basal hepatocyte pole, which was highlighted by the capillary endothelial cell marker CD34 (Fig. 2c). To quantify ACE2 protein expression at the biliary pole of HCC cells, we measured the fluorescence area for separate ACE2 and ABCC2 signals and for dual (ACE2 + ABCC2) staining in 96 TMA spots corresponding to 32 HCCs. To correct for differences in cell densities among HCCs, we plotted specific signal areas as the fraction of the DAPI area (i.e., nuclear staining) (Fig. 2d,e). Mean ± SD ABCC2/DAPI or ACE2/DAPI ratios were: ABCC2/DAPI, 33 ± 27%; ACE2/DAPI, 16 ± 34%; (ACE2 + ABCC2)/DAPI, 3 ± 6%. Of note, 19 ± 25% of all ACE2 expression (i.e., (ACE2 + ABCC2)/ACE2) was located at the biliary poles of HCC cells (Fig. 2d,e).
TMPRSS2 was also detected at the apical pole of some of the pseudo-glandular formations or within slitlike trabecular structures and in cell-cell borders ( Supplementary Fig. 5). In contrast with ACE2, the signal for TMPRSS2 was mild in HCCs, suggesting low protein expression levels, which were unrelated to CTNNB1 mutations. Neither ACE2 nor TMPRSS2 (not shown) were detected in myofibroblasts as suggested by dual ACE2 and ACTA2 staining ( Supplementary Fig. 5).
In five non-tumor samples from patients undergoing resection of colon cancer metastases and with minimal inflammatory changes in the liver, ACE2 was co-detected with the sinusoidal endothelial cell marker CLEC4M (a.k.a. DC-SIGNR) (Fig. 3a,c,e) and the bile canaliculi marker ABCC2 (a.k.a. MRP2) (Fig. 3b,d,f). The presence of ACE2 within bile canaliculi is consistent with the facts that the extracellular part of ACE2 can be cleaved off the cell surface by ADAM17, thus shedding functional ACE2 fragments 34 ; and that ACE2 is an abundant component of the normal human bile proteome 35 . Also, ACE2 was co-detected with CD34 in capillary vessels of the periportal vascular plexus (Fig. 3g). By contrast, ACE2 was not detected in CD68-positive sinusoidal cells (Kupffer cells, Fig. 3h) or ACTA2-positive vascular smooth muscle cells or myofibroblasts (Fig. 3i,j). As expected, the antibodies used detected high levels of ACE2 in the apical intestinal epithelium, Bowman's capsule and convoluted tubes in the kidney; in turn, TMPRSS2 was detected in the apical compartment of the epithelial lining of prostatic glands ( Supplementary Fig. 5).
In summary, dual labeling of HCC and of non-neoplastic liver with cell-type specific antibodies showed ACE2 in the epithelial lining of bile ducts and in bile canaliculi, i.e., the biliary pole of normal and differentiated tumor hepatocytes. In non-neoplastic liver, CLEC4M-positive sinusoidal endothelial cells and CD34-positive capillary endothelial cells of the periportal vascular plexus also contain ACE2.
ACE2 and TMPRSS2 co-expression networks highlight metabolic functions typical of normal hepatocytes. By Weighted Gene Correlation Network Analysis 36 comparing periportal-type (n = 342) and perivenous-type (n = 225) HCCs within the Désert's dataset (n = 1133), we found that ACE2 and TMPRSS2 were related through interacting metabolic functions (Fig. 4a), including detoxification, aminoacid catabolism, lipid metabolism, fatty acid oxidation and catabolism of organic compounds, which is probably related to the carboxypeptidase functions of ACE2, involved in the degradation of inflammatory polypeptides 37,38 (Fig. 4b). These findings are in line with the above-described expression of ACE2 and TMPRSS2 in well-to-moderately differentiated HCCs preserving the metabolic programs of adult hepatocytes. Next, we specifically investigated the ACE2 co-expression network in HCCs and analyzed associated gene functions (Supplementary Tables 2, 3, 4). Not surprisingly, ACE2 was associated with hepatocyte-specific functions such as aminoacid metabolism, xenobiotic detoxification, fatty acid uptake and oxidation. High ACE2 expression was associated with low expression levels of oxidative stress markers and cytokine-mediated inflammatory signals ( Supplementary Fig. 7a, b). These findings raise the hypothesis of a link between inflammation, ACE2 dysfunction and metabolic breakdown that may be relevant to the pathogenesis of COVID-19.

Discussion
In this work, we show that ACE2 and TMPRSS2 mRNAs and proteins are detected at higher levels in well-tomoderately differentiated HCCs than in poorly differentiated tumors. In particular, ACE2 is detected at the biliary pole in tumors preserving hepatocyte polarization. By contrast, in non-tumor livers, ACE2 is detected not only in the epithelial lining of bile ducts and within hepatocyte's bile canaliculi, but also in the CLEC4M-positive sinusoidal endothelium and the CD34-positive capillary vessels of the periportal plexus. TMPRSS2, in turn, is not detected in normal liver, but is detected at the biliary pole in well-to-moderately differentiated HCCs, which is consistent with the functional association of ACE2 and TMPRSS2 through periportal and perivenous metabolic  26 , in non-proliferative-moderately-to-well-differentiated, periportal-type and perivenous-type HCCs 24 . The co-expression network was identified by Weighted Gene Correlation Network Analysis and the WGCNA 36 R package, followed by visualization with Cytoscape 49 . Genes co-expressed with ACE2, light-blue nodes with black borders and black links; genes co-expressed with TMPRSS2, white nodes with red borders and black links; genes co-expressed with both ACE2 and TMPRSS2, light-blue nodes with red borders and blue links. Link thickness is proportional to the correlation coefficient between a given node and ACE2 or TMPRSS2. www.nature.com/scientificreports/ networks. Of note, high-resolution 500-nm optical sections revealed ACE2 within hepatocyte's bile canaliculi in histologically normal liver. This finding agrees with the high content of N-glycosylated, lectin-binding ACE2 in the normal human bile proteome 35 and with the role of ACE2 in aminoacid uptake in the gut and in the regulation of the splanchnic blood flow 37 .
In HCCs carrying activating β-catenin (CTNNB1) mutations, ACE2 DNA is hypomethylated, and ACE2 mRNA expression upregulated. Conversely, TMPRSS2 mRNA is lower in HCCs carrying mutated CTNNB1 and its DNA hypermethylated. This finding agrees with the fact that only ACE2 (and DPP4, but not TMPRSS2) contain TCF/LEF-1 binding sites that are predicted to respond to β-catenin transcriptional activation. Neither ACE2, nor DPP4 or TMPRSS2 are associated to TERT or TP53 gene mutations. On the contrary, HCCs containing mutated TP53 show relatively lower levels of TMPRSS2 (Supplementary Fig. 3D). This association does not necessarily imply a direct regulation of TMPRSS2 by TP53, but may be coincidental. In fact, well-differentiated, periportaltype HCCs express the highest levels of TMPRSS2 and, as we previously showed, have the lowest prevalence of TP53 mutations 26 . Nonetheless, the relationships of ACE2 and TMPRSS2 with histological, transcriptomic and genomic HCC subclasses converge to demonstrate that tumors preserving hepatocyte-like features, such as hepatocyte trabeculae with bile canaliculi and/or pseudo-glandular structures with an apical biliary pole, express higher levels of the SARS-CoV-2 receptor and co-receptor couple.
The above findings agree with further observations in this work: the ACE2 co-expression network included liver-enriched genes involved in hepatocyte-specific functions, such as aminoacid metabolism (GNMT), xenobiotic detoxification (CYP3A4), fatty acid uptake, bile acid reconjugation (SLC27A5), enterohepatic circulation of bile acids and cholesterol homeostasis (SLC10A1). ACE2 was also co-expressed with SIRT1 and CD36 and the fatty acid oxidation enzymes ACADL, EHHADH, ACADM and ECI2, and alternative glucose metabolism pathways (CRYL1), as well as enzymes involved in gluconeogenesis (PCK2). This network is in line with the role of ACE2 as a regulator of fatty acid uptake in hepatocytes through the angiotensin 1-7 -MAS1-SIRTUIN1-CD36 axis 39 . Indeed, ACE2 knock-out mice show low levels of peripheral lipid uptake, but high hepatocyte uptake of triglycerides, resulting in steatosis, lipid peroxidation, oxidative stress and inflammation 39 . In turn, the induction of oxidative stress by ACE2 deficiency may be related to its effects on glucose and lipid metabolism because the ACE2/Ang 1-7 /MAS axis increases glucose uptake, decreasing insulin resistance. Thus, ACE2 knock-out mice exhibit high oxidative stress, whereas exposure to Ang 1-7 reduces oxidative stress in liver cells through insulin signaling 40 . Taken together, this body of evidence matches the metabolic landscape of HCCs carrying activating CTNNB1 mutations. These tumors oxidize fatty acids as a source of energy to feed the oxidative phosphorylation pathway 41 , but resist to oxidative stress 42 and suppress inflammation 43 .
ACE2 expression was related to gene functions that smother inflammation. Indeed, ACE2 was related to low levels of the inflammation-related transcripts BIRC3, MSN, MMP9, LTB, TNFRSF4, IL4R, CCL20 and IL32, among others (Supplementary Table 2). ACE2 exerts anti-inflammatory effects through carboxyterminal cleavage of vasoactive polypeptides, such as Apelin-13, which induces vasodilation, and des-arg9-bradykinin, a neutrophil chemoattractant 44 . In turn, deletion of ACE2 leads to tissue injury and oxidative stress, through upregulation of the AT1 receptor and inflammatory cytokines like IL1β, IL6, IL17, RANTES, ICAM1, TNFα, TNFRSF1A and downregulation of the anti-inflammatory cytokine IL10 45 . This picture is surprisingly similar to that seen in the spectrum going from immune dysregulation to macrophage activation syndrome in moderate-to-severe COVID-19 4,46 .
The anti-inflammatory role of ACE2 goes well with its up-regulation in response to liver injury, such as bileduct ligation 47 and with the evidence that ACE2 produced by hepatocytes limits fibrogenesis through degradation of AngII to Ang 1-7 37 . In human liver cirrhosis, ACE2 protein is increased by > 90 folds in hepatocytes. In vitro, hypoxia increases ACE2 activity in hepatocytes, enhancing acetylcholine-induced vasodilatation 38 . As hypoxia activates Wnt/β-catenin signaling, it would be interesting to investigate whether ACE2 expression in perivenous hepatocytes depends on β-catenin 34,48 .
This body of matching data suggests that ACE2 may regulate energy metabolism, protecting against oxidative stress and inflammation. As both TMPRSS2 and ACE2 are functionally connected through metabolic pathways in the liver, it may be reasonable to hypothesize that ACE2 depletion resulting from SARS-CoV-2 infection 18 may contribute to metabolic dysfunction and inflammation. Our results point to the relevance of further studies on the signaling pathways and pathological contexts leading to cell surface ACE2 availability and to the possible impact of viral infection on the disruption of the ACE2 metabolic network.

Materials and methods
Patients, samples and datasets. Human hepatocellular carcinoma (HCC): immunohistochemistry was performed on a tissue microarray (TMA) built from formalin-fixed-paraffin-embedded (FFPE) routine tissue blocks of 41 HCC cases and two histologically normal liver controls, for which the exon 3 of β-catenin (CTNNB1) had been Sanger-sequenced, as we previously described 26 . In addition, three formalin-fixed, paraffin-embedded non-tumor liver tissue blocks from patients undergoing resection of liver metastases from extra-hepatic cancers were added to the immunohistochemistry analysis. To calculate β-catenin activation scores based on the mRNA expression of GLUL, LGR5, ODAM, VNN1 and HAL, we retrieved real-time PCR mRNA expression data from our previously published work 26 corresponding to frozen fragments of the 41 tumors arrayed in the TMA. Quality control matching of frozen and FFPE fragments has been previously described in an 80 HCC dataset including the 41-tumor subset 32,33 . Tissue samples and patient data were collected and anonymized after obtaining written informed consent from the patients. De-identification was performed in accordance with the Health Insurance Portability and Accountability Act (HIPAA) Privacy Rule. The study was approved by INSERM's Institutional Advisory Board (approval number  www.nature.com/scientificreports/ relevant guidelines and regulations. Two different immunohistochemical methods were used to detect primary antibodies: peroxidase-labeled secondary antibodies, revealed by a chromogenic substrate or multiplex immunofluorescence followed by analysis with a confocal slide scanner. The Désert's microarray meta-dataset composed of 1133 human HCCs was previously described 26 . In this dataset, HCCs expressing a β-catenin-activated transcriptomic program were identified with a 5-gene signature composed of GLUL, LGR5, ODAM, VNN1 and HAL, which predicts activating CTNNB1 mutations with high sensitivity (0.86-0.91) and specificity (0.83-1.0) 26 . β-catenin pathway activation was confirmed with an independent 23-gene signature 27 . The publicly available TCGA dataset was composed of 370 HCCs and 47 matching non-tumor samples. RNA sequencing, mutation and methylation data were extracted as described 26 , using the TCGAbiolinks R package. Raw mRNA expression was median-normalized (DESeq R package). DNA methylation was quantile-normalized (preprocessCore R package); probes with B-value over 0.8 or under 0.2 were respectively considered as hypermethylated or hypomethylated. The R-package Limma was used to identify differential methylation. See Supplementary Information for a detailed description of immunohistochemical methods and statistical analyses and Supplementary Table 6 for primary, secondary antibodies, fluorochromes used, epitope unmasking and incubation conditions.

Data availability
Datasets analyzed in this study are available at The Cancer Genome Atlas or at the Gene Expression Omnibus databases. Accession numbers and standard, publicly available R packages used are provided in the main text and supplementary files. The procedures applied for merging nine publicly available transcriptomic datasets into a meta-dataset totaling 1133 hepatocellular carcinomas have been previously published 26 . A filtered, normalized and batch-effect-corrected user-friendly version of this meta-dataset is available from the corresponding author upon reasonable request.