Autophagy is a gatekeeper of hepatic differentiation and carcinogenesis by controlling the degradation of Yap

Activation of the Hippo pathway effector Yap underlies many liver cancers, however no germline or somatic mutations have been identified. Autophagy maintains essential metabolic functions of the liver, and autophagy-deficient murine models develop benign adenomas and hepatomegaly, which have been attributed to activation of the p62/Sqstm1-Nrf2 axis. Here, we show that Yap is an autophagy substrate and mediator of tissue remodeling and hepatocarcinogenesis independent of the p62/Sqstm1-Nrf2 axis. Hepatocyte-specific deletion of Atg7 promotes liver size, fibrosis, progenitor cell expansion, and hepatocarcinogenesis, which is rescued by concurrent deletion of Yap. Our results shed new light on mechanisms of Yap degradation and the sequence of events that follow disruption of autophagy, which is impaired in chronic liver disease.

H epatocellular carcinoma (HCC) is the third leading cause of cancer-related mortality worldwide 1 with a rising incidence attributed to advanced liver disease from multiple etiologies 2 . HCC develops as a multistep process on a background of chronic liver injury leading to inflammation, stromal activation, fibrosis, and regeneration 3 . Pathways involved in the malignant transformation of hepatocytes from dysplastic nodules to early stage HCC include oxidative stress responses, deregulation of the protein folding machinery and dedifferentiation with reexpression of fetal genes 4,5 .
Autophagy has a dual role in cancer. During tumor initiation autophagy removes damaged organelles and reactive oxygen species to maintain genomic stability and promote oncogeneinduced senescence, thereby inhibiting malignant transformation. However, in advanced tumors and metastases autophagy may promote tumor cell survival in low-nutrient conditions and chemotherapy-induced stress 14,15 .
Autophagy's role in liver is primarily tumor-suppressive, yet its full activity in hepatocarcinogenesis is unclear. Rodent models with liver-specific deletion of the autophagy-related proteins 5 (Atg5) 16 or Atg7 17-19 display hepatic metabolic dysfunction with steatosis, ER stress and marked hepatomegaly and develop benign tumors (adenomas), but not HCCs. Central to this phenotype is signaling by p62/Sqstm1-Keap1-Nrf2, because loss of either p62/ Sqstm1 20 or Nrf2 in autophagy-deficient double knock out mice normalizes liver size 16,21 and decreases tumorigenesis in p62/ Sqstm1/Atg7 DKO mice 18 . When autophagy is impaired, the autophagy adaptor protein p62/Sqstm1 accumulates and binds Kelch-like protein 1 (KEAP1), a negative regulator of the oxidative master regulator nuclear factor erythroid 2-related factor 2 (NRF2, NFE2L2). NRF2 stabilization leads to a deleteriously high antioxidative response with induction of downstream targets such as NAD(P)H quinone oxireductase 1 (Nqo1) or gluthathion S-transferase 1 (Gst1) exacerbating liver injury and metabolic reprogramming 22 . The p62/Sqstm1-Keap1-Nrf2 axis has been thus identified as the pathway linking autophagy inhibition in the liver to hepatomegaly and tumorigenesis. However, the current paradigm does not account for the dramatic tissue remodeling and progenitor cell activation within autophagy deficient livers prior to the onset of cancer.
Yap is the major nuclear effector of the Hippo signaling pathway and mediates organ size control, proliferation, differentiation, and stemness. Yap can be activated by multiple stimuli including DNA damage 23 , ER stress 24 , and mechanical/shear stress 25 , but its inhibition is tightly regulated and includes signaling by the Hippo core kinase cassette 26 . Yap may also be inhibited by cytoplasmic sequestration by dystrophin-glycoprotein complex 27 or Yap/Amot, which can result in the translocation of Yap to the cytoplasm or to cell junctions 28,29 . Upon its activation, Yap translocates into the nucleus, where it interacts with Tead1-4 as transcriptional co-activators to promote a range of biologic activities, including cell survival, proliferation, polarity, and, most importantly, organ size control 30 . Liver-specific deletions of Hippo pathway components (e.g., Mst1/2, Sav1, Nf2) or overexpression of Yap, lead to prominent hepatomegaly, progenitor cell expansion and hepatocarcinogenesis [31][32][33][34] . Ectopic overexpression of Yap in hepatocytes promotes their plasticity, with dedifferentiation to a cholangiocytic phenotype 35 .
Yap activation is an early event in liver cancer development 36 . Up to 65% of HCCs harbor dysregulation of the Hippo/Yap pathway, which is associated with a significantly poorer prognosis 37,38 . Strikingly, despite comprehensive efforts, no germline or somatic mutations have been uncovered 39 . Thus, mechanisms underlying Yap dysregulation in hepatocarcinogenesis remain obscure.
Here, we identify Yap as an autophagy substrate and as an essential downstream mediator of tissue remodeling, progenitor cell activation and hepatocarcinogenesis in autophagy-deficient livers. Our results shed new light on mechanisms of Yap dysregulation in HCC by implicating impaired autophagy as a potential driver of Yap stabilization and activity. Because autophagy is impaired chronic liver disease, the findings provide a potential basis for enhanced Yap activity associated with hepatocarcinogenesis 7,40 .

Results
Autophagy maintains hepatic organ size and differentiation. To explore the mechanisms underlying autophagy's regulation of hepatic growth and tumorigenesis, we generated mice with conditional deletion of Atg7 in hepatocytes (Albumin-CRE: Atg7 F/F mice) (Fig. 1a). Consistent with earlier studies 18,19,41 mice with liver-specific deletion of the autophagy-related protein 7 (Atg7) displayed massive hepatomegaly with up to 8.5 fold increased relative liver weight compared to CRE negative littermates at 3 months of age ( Supplementary Fig. 1A, B, C). By histological analysis, hypertrophic and hyperplastic hepatocytes were prominent with a marked thickening of hepatocyte plates that were composed of at least 2-3 cells instead of single cells based on HE staining and on the pattern of cell membranous staining for β-catenin ( Supplementary Fig. 1D, F). Hepatocytes were markedly dedifferentiated based on HNF4α staining (Fig. 1b) and whole liver RNA array analysis in which hepatocyte-specific genes including Albumin, Transthyretin and complement factors (C2, C3) were significantly downregulated (Fig. 1c). Immunostaining showed Epcam + progenitor cell expansion (Fig. 1d). Blinded histologic scoring demonstrated increased lobular and portal inflammation, steatosis and ballooning ( Supplementary Fig. 1E) and significant fibrosis (Supplementary Fig. 1G, H). Hepatocyte proliferative activity was markedly increased as assessed by Ki67 staining and quantification ( Supplementary Fig. 1I, J). All Atg7 KO mice developed dysplastic nodules at 8 months, which progressed to HCC at 12 months (Supplementary Fig. 2A, B, C). In contrast to previous reports 17 , tumors had typical features of HCCs including pseudoglandular transformation, loss of reticulin staining and diffuse Gst1 staining 42 (Supplementary Fig. 2D, E).
Autophagy maintains hepatic Hippo tumor suppressor pathway. Yap is a key regulator of hepatocellular fate 35 , stem cell renewal, and liver size with overexpression or dysregulation of the Hippo/Yap signaling pathway leading to liver outgrowth, progenitor cell expansion and tumorigenesis 32,34 . We thus analyzed livers of Atg7 KO mice at 3 months and tumors at 12 months of age and found increased cytoplasmic and nuclear Yap (Fig. 1e). consistent with activation of the Hippo pathway effector. Primary hepatocytes isolated from Atg7 KO mice also displayed increased nuclear Yap staining ( Supplementary Fig. 3B). Gene set enrichment analysis (GSEA) of the livers' transcriptome revealed significant enrichment of Yap activation signatures previously identified in Mst1/Mst2 KO mice 43 and in human cancer cells 44,45 (Fig. 1f). Significant induction of genes associated with Yap activation and Yap target genes was confirmed by gene array (Fig. 1g) and qRT-PCR (Fig. 1h).
To analyze the immediate response to Atg7 deletion, we generated an inducible conditional Atg7 KO mice by crossing Atg7 floxed mice into the tamoxifen-inducible CRE line driven by the Albumin promoter 46,47 (Fig. 1i). This model allows for time dependent, hepatocyte-specific gene deletion in adult liver. Immunoblotting of whole liver lysates from tamoxifen-inducible, conditional Atg7 KO mice revealed increased total Yap protein within 7 days post tamoxifen injection, concomitant with effective Atg7 deletion and abrogation of autophagy characterized by reduced conversion of LC3-I to LC3-II (Fig. 1j). In contrast, protein levels of the Yap paralog Taz/Wwtr1 were consistently increased from 14 d ( Supplementary Fig. 3D). Immunostaining for Taz/Wwtr1 showed significant staining in non-parenchymal cells rather than hepatocytes ( Supplementary Fig. 3E). P62/ Sqstm1, a selective autophagy adaptor protein, which accumulates upon inhibition of autophagy 7 and which mediates Nrf2 stabilization via Keap1 21 was increased at 28 days post tamoxifen injection, indicating that the early increase in Yap at d7 is independent of p62/Sqstm1 (Fig. 1j). Quantitative band densitometry showed significantly increased Yap/β-Tubulin and decreased P-Yap/Yap ratios, consistent with Yap activation (Fig. 1k, l).    Fig. 1 Autophagy maintains Hippo/Yap pathway in liver. a Control (Atg7 F/F ) and Atg7 KO mice (Alb-CRE:Atg7 F/F ) were analyzed at 3 months (n = 4) and 12 months of age (n = 6 and 7 respectively). b Immunostaining for HNF4α in control and Atg7 KO mice. c Gene expression array analysis of whole liver RNA from controls and Atg7 KO animals. Data indicate row z-scores. d Immunostaining for Epcam in control and Atg7 KO mice. e Immunostaining for Yap in control and Atg7 KO liver sections at 3 months and 12 months. Tu, tumor. f Gene set enrichment analysis for enrichment of Hippo/Yap gene signatures and plotting of NES scores. NOM P-value *P < 0.05, **P = 0.001, ****P < 0.0001. NES, normalized enrichment score. g Gene expression array analysis of whole liver RNA from controls and Atg7 KO for Yap downstream targets. Data indicate row-z scores. h qRT-PCR analysis of whole liver RNA from control and Atg7 KO mice at 3 months for Yap downstream targets. Data normalized to Gapdh mRNA expression and expression in control mice. Data represent two independent experiments. Mean ± SD. **P < 0.01, ***P < 0.001 by two tailed t-test. i Tamoxifen (TAM)-inducible, hepatocyte-specific Atg7 KO (ERT2-Alb-CRE:Atg7 F/F ) and respective controls were analyzed 7, 14, and 28 days after TAM injection. j Immunoblotting of whole liver lysates from control and Atg7 KO mice, 7, 14, and 28 days after tamoxifen injection. k Quantitative band densitometry of Yap and β-tubulin immunoblots normalized to Yap/βtubulin ratios in control mice. **P < 0.005 by two tailed t-test compared to controls. l Band densitometry of P-Yap and Yap immunoblots normalized to P-Yap/Yap ratios in control mice. **P < 0.005 and *P < 0.05 by two tailed t-test compared to controls. Scale bar 100 µm. Insets in b, d, e in the lower right corners correspond to small boxes within the associated lower magnification images Yap is degraded by autophagy. The Hippo/Yap signaling pathway contains a highly conserved core kinase cassette which, when engaged, phosphorylates and inactivates Yap by βTrCP-mediated proteasomal degradation 38,48 . Yap may also be inactivated by sequestration in the cytoplasm 27,28,32 . Since previous data indicated normal proteasomal activity in autophagy-deficient neurons 49 we hypothesized that Yap is degraded by autophagy. To test our hypothesis, we stably infected the murine hepatocyte line AML12 with lentiviral shRNAs (scrambled or shAtg7 50 ) and verified Atg7 knockdown by qRT-PCR and immunoblotting ( Supplementary Fig. 4A, B, C). shAtg7-AML12 cells displayed increased p62/Sqstm1 typical of autophagy inhibition and increased nuclear Yap localization by immunofluorescence (Fig. 2a), which was also verified by immunoblotting Yap in cytoplasmic and nuclear fractions (Fig. 2b). Gapdh and Nucleoprotein p62 (Nup62) (nuclear Nup62 62 kD, cytoplasmic precursor 61 kD protein) 51 served as markers for the respective fractions. We observed increased nuclear Nup62 in shAtg7 cells despite loading equal amounts of protein per lane, however, increased expression of Nup62 and other nucleoproteins have been described to be cell cycle-regulated with increases of Nup62 from G1 to G2/M phase 52  Similarly, shAtg7-AML12 cells transfected with a Tead4-Luciferase reporter 35 displayed significantly more luciferase activity compared to controls, consistent with increased Yap activation (Fig. 2d). By cycloheximide chase assay Yap half-life was increased in shAtg7-AML12 cells (Fig. 2e, f) while proteasomal activity was maintained ( Supplementary Fig. 4D). We thus tested if Yap is degraded by autophagy. Leupeptin/ NH 4 Cl which inhibits lysosomal degradation including autophagy, further increased Yap protein levels in scram-AML12 cells but not in shAtg7-cells (Fig. 2g), confirming that increased Yap levels in shAtg7-cells are a consequence of reduced lysosomal   degradation. Furthermore, Tead4-Luciferase activity was significantly increased in AML12 cells incubated with Leupeptin/NH 4 Cl compared to cells incubated with vehicle alone (Fig. 2h). In the human hepatocyte line THLE5B 53 , YAP protein was also increased upon inhibition with the lysosomal inhibitor Leupeptin/NH 4 Cl or by the macroautophagy inhibitor 3-methyladenine ( Supplementary Fig. 4E). Increased YAP protein levels and Tead4-Luciferase activity indicative of YAP activation was also observed in shATG5-THLE5B cells ( Supplementary Fig. 4J, K). Combined, these result indicate that YAP activation results from a general deficiency in the autophagy pathway rather than from loss of ATG7 alone.
To confirm that YAP colocalized with lysosomes, THLE5B cells were transfected with YAP-DsRed and incubated with Lyso®TrackerGreen, which enabled tracking of Yap and lysosomes. Incubation with Leupeptin increased the appearance of lysosomes and the colocalization of YAP and lysosomes, indicating lysosomal degradation of YAP (Supplementary Fig. 4F, G). Similarly, THLE5B cells transfected with YAP-DsRed and GFP-LC3 also showed colocalization of YAP and autophagosomes upon Leupeptin/NH 4 Cl incubation ( Supplementary  Fig. 4H,I). Finally, endogenous YAP colocalized with autophagosomes in cells transfected with GFP-LC3 (Fig. 2i, j) upon Leupeptin/NH 4 Cl incubation, further confirming that YAP is degraded by autophagy.
Moreover, tumor size and number were significantly decreased in Atg7/Yap het DKO mice and Atg7/Yap DKO mice compared to Atg7 KO at 12 months after tamoxifen administration (Fig. 3b, e, f, Supplementary Fig. 6D, E) thereby identifying Yap as a driver of liver growth and hepatocarcinogenesis when autophagy is impaired.
Deletion of Yap in Atg7/Yap DKO mice reduced hepatocyte size (Fig. 3g-i) and significantly improved portal and lobular inflammation, ductular reaction, steatosis, and fibrosis in Atg7/ Yap DKO mice ( Supplementary Fig. 6F-H). Restored hepatic differentiation was demonstrated by increased immunostaining for HNF4α and qRT-PCR of whole liver RNA for Albumin (Fig. 3g, j) in agreement with previous findings linking Yap overexpression to hepatic dedifferentiation 35,55 . In addition, progenitor cell expansion, which is observed with Yap overexpression 35 was significantly decreased in Atg7/Yap DKO mice as assessed by immunostaining and qRT-PCR for Cd133, Cd44, and Epcam (Fig. 3g, j).
Yap is a potentially druggable driver of HCC in Atg7 KO. To assess if Yap-dependent proliferation in autophagy-impaired livers ( Supplementary Fig. 5G, H) is targetable, we tested the effect of verteporfin, a small molecule Yap-Tead4-inhibitor 59 in culture and in vivo. Verteporfin has also been shown to be safe in phase I/phase II clinical trials 60 . In vivo, treatment of Atg7 KO mice with verteporfin for 21 days significantly decreased the number of Ki67 + nuclei in Atg7KO mice (Fig. 5b, c), which was also associated with reduced expression of the Yap target gene Cyr61 (Fig. 5d). In cultured scram-and shAtg7-AML12 cells, verteporfin significantly decreased proliferative activity (Fig. 5e). Similarly, verteporfin significantly diminished Tead4-Luciferase activity in scram-and shAtg7-AML12 cells (Fig. 5f).
Atg7-KO gene signature aligns with human NASH and HCC profiles. To analyze the relevance of our findings to human liver disease and hepatocarcinogenesis, we defined a Atg7-KO gene signature (differentially expressed genes in liver tissues from Atg7-KO mice compared to control mice) and compared it to the transcriptome profile of 72 human non-alcoholic fatty liver disease (NAFLD) liver tissues with known disease severity. GSEA analysis showed significant enrichment of the Atg7-KO gene signature in NAFLD samples that were comprised primarily of samples from advanced NAFLD (F3 or F4), along with enrichment for YAP target genes (Fig. 6a). Similarly, analysis of 374 human HCC transcriptome profiles indicated simultaneous enrichment of the Atg7-KO gene signature and YAP activation in 42.2% (158/374) of HCCs as well (Fig. 6b). Integrative transcriptomic analysis of HCCs has indicated that these tumors may be classified according to three molecular subclasses, S1, S2, and S3 61 . Interestingly, HCC subclass S1, which is characterized by steatohepatitic HCCs, a histological subtype which arises in NAFLD/NASH 62 and immune cell infiltration 63 , was significantly enriched for the Atg7-KO gene signature, whereas subclass S3 had no such alignment with the autophagy-deficient signature. Immunofluorescence analysis of human HCCs with known molecular subclass showed nuclear YAP staining and enhanced cytoplasmic staining for p62/SQSTM1 within the same cells (Fig. 6c), indicating that impaired autophagy flux and concomitant YAP activation occur in human disease.

Discussion
By utilizing a regulable hepatocyte-specific promoter to analyze Atg7 −/− mice, we have uncovered autophagy as the gatekeeper of hepatic differentiation, growth regulation and carcinogenesis by controlling degradation of Yap. Hepatocyte-specific Atg7 −/− mice display enhanced Yap target gene expression and increased Yap protein that drives hepatocyte proliferation, leading to gross hepatomegaly. These abnormalities are significantly attenuated by heterozygous and homozygous loss of Yap allele(s) in Atg7/Yap Het and Atg7/Yap DKO double knockout mice. Thus, Yap is a driver of tissue remodeling and carcinogenesis when autophagy is impaired.
Dysregulation of liver growth regulation in Atg7 KO mice is due to increased cell size (hypertrophy) and Yap-dependent increase in cell number (hyperplasia). Hepatomegaly in Atg7 KO mice was previously attributed to accumulation of non-degraded cell components including proteins, lipids and damaged organelles 17,18,41 . However, our data demonstrate that inhibiting Yap-mediated hepatocyte proliferation and hyperplasia in Atg7/ Yap DKO mice can significantly attenuate the increased liver size and significantly decrease tumorigenesis.   Unlike previous studies 20,21 the dramatic phenotype consequences of autophagy loss in liver occurs independent of p62/ Sqstm1-KEAP1-NRF2 signaling in our study. p62/Sqstm1 stabilizes Nrf2 by negatively regulating Keap1, thereby abrogating polyubiquitination and proteasomal degradation of Nrf2 21 ; this stabilizes Nrf2 and allows its nuclear translocation. In previous studies, Atg7/p62 (Alb-CRE:Atg7 F/F , p62 −/− ) and Atg7/Nrf2 (Mx1-CRE: Atg7 Atg7 F/F , Nrf2 −/− ) double knock-out mice reportedly rescue the dramatic liver phenotype, associated with a significant decrease in ubiquitinated proteins consistent with improved proteostasis despite inhibition of autophagy 18,21 . In Atg7/p62 double knock-out mice there is decreased tumorigenesis in aged Atg7/p62 DKO animals. Thus, the contribution and relationship of p62/Sqstm1, Nrf2, and Yap in the context of autophagy impairment to liver injury, liver growth regulation and hepatocarcinogenesis remains to be clarified. p62/Sqstm1 is an independent oncogenic driver in AAV-p62/Sqstm1-infected hepatocytes through oncogenic effects of mTOR and Nrf2 signaling. In contrast, Atg7/Yap double knockout significantly attenuates hepatocarcinogenesis despite increased expression of p62/Sqstm1 and Nrf2, and Nrf2 downstream targets. Thus, Yap is a significant driver of hepatocarcinogenesis when autophagy is impaired. However, tumorigenesis was not completely abrogated in Yap/Atg7 double knock-out mice and the remaining tumorigenesis could be due to p62/Sqstm1-Nrf2 signaling. A contribution of the Yap paralog Taz/Wwtr1 cannot be excluded, however immunostaining demonstrates an increase in Taz/Wwtr1 in non-parenchymal cells, and qPCR analysis showed significant decrease in Hippo downstream targets Cyr61 and Areg, indicating efficient suppression of Yap/Taz targets in Atg7/Yap DKO mice.
Autophagy is clearly a tumor-suppressive pathway in our models. In previous studies autophagy has been ascribed a bifunctional role in tumorigenesis-tumor suppressive in the initial phase and tumor promoting in advanced stages 8 . Previously, tumors in hepatic Atg7 KO mice have been described as benign adenomas 17 ; however, tumors in ERT2-Alb-CRE:Atg7 F/F and Alb-CRE:Atg7 F/F mice fulfill characteristics of HCCs. Differences in the genetic background and microbial environment might contribute to the observed phenotypic differences between ours and previous studies, as the microbiome can affect hepatocarcinogenesis, but future studies will need to clarify the underlying mechanisms.
Our results provide an important link between metabolic alterations and mechanisms for Yap activation in hepatocarcinogenesis. With the epidemic of obesity and metabolic syndrome, hyperlipidemia and insulin resistance are frequently observed concomitant to chronic viral hepatitis and NASH 2 and have been shown to decrease hepatic autophagic and lysosomal degradation 40 . Comprehensive integrative molecular analyses (transcriptomic, genomic and epigenomic data) as well as immunological analyses have provided a molecular and immunological classification of HCCs which can be differentiated into a proliferative (S1 and S2) and non-proliferative class 39,[64][65][66] . Of these, S1 subclass is associated with poor prognosis, chromosomal instability and activation of pathways associated with cell proliferation and survival and immune exhaustion 66   and alignment of the Atg7 KO gene signature in HCCs of the S1 subclass indicates that metabolic pathways of Yap activation might contribute to this subclass. Our results have uncovered a mechanism of autophagydependent degradation of Yap which, when impaired, promotes dedifferentiation, inflammation, fibrosis, and hepatocarcinogenesis. Because Yap activity is strongly implicated in human hepatocarcinogenesis, our studies provide a rationale for chemopreventive or anti-tumorigenic strategies through inhibition of Yap or enhancement of autophagy by drugs such as verteporfin 60 or carbamazepine, which is currently in phase II trial for α1-antitrypsin deficiency (NCT01379469) and has anti-fibrotic effects in murine disease models 67 .

Methods
Animals. Alb-CRE mice (Jax 003574) were purchased from Jackson Laboratory and crossbred with conditional Atg7 19 . Nrf2 −/−21 were crossed with Alb-CRE:Atg7 F/F to generate Atg7/Nrf2 DKO mice. ERT2-Alb-CRE mice 46 crossed with conditional Atg7 19 and/or, conditional Yap mice 54,21 mice were utilized in this study. Agematched, male and female mice were used and did not show sex-bias differences. ERT2-Alb-CRE mice (6-8 weeks) and respective controls were given 0.1 mg tamoxifen (Sigma) (diluted in corn oil) for 5 consequential days. Verteporfin (Sigma) diluted in DMSO and injected i.p. (50 mg kg −1 ) every other day for 21 days. All animal studies and procedures were approved by the Animal Care and Use Committee of Icahn School of Medicine at Mount Sinai.
Immunohistochemistry. For immunohistochemistry, liver samples were fixed in 10% formalin overnight, transferred to 70% ethanol and processed for paraffin embedding. For immunostaining, samples were deparaffinized and heat induced antigen retrieval performed with either 10 mM sodium-citrate, pH 6 or Tris-EDTA, pH 9 or DAKO target retrieval solution (Dako, S2375) depending on the antigen. Chromogen development after incubation with primary and secondary antibodies was performed using the Dako Envision Kit (Dako). For immunofluorescence studies, AML12 cells, THLE5b cells or primary murine hepatocytes were grown on coverslips, fixed with acetone or 50% acetone/50% methanol for 10 min and rehydrated in PBS. Liver sections from either 4% PFA-perfused livers and sequential sucrose-cryopreservation were analyzed, fixed with acetone, or 50%   Morphometric quantitative analysis of cell size and nucleus size from liver sections was performed by ImageJ analysis. Two hundred ten to 270 cells per genotype (n = 3) were analyzed and plotted by frequency distribution analysis (Prism 7). Colocalization analysis was performed by ImageJ analysis software (plugin JACoP v2).
Immunoblotting. Fifty micrograms of snap frozen liver were homogenized by Qiagen TissueLyser in RIPA buffer containing proteinase inhibitors (Roche complete Mini proteinase inhibitor, 2 tab/10 ml buffer) and 1x HALT phosphatase inhibitors (Thermoscientific, #78428). Lysates were sonicated and then centrifuged for 10 min at 18,800 × g. THLE5B and AML12 cells were lysed directly in RIPA buffer containing inhibitors, sonicated and centrifuged. Protein concentration of supernatants were determined by BioRad protein assay (BioRad). For immunoblot analysis, 30-50 µg of protein were analyzed per SDS-PAGE and tank blotting. Antibodies and dilutions used are indicated in Supplementary Table 1. Semiquantitative band densitometry was performed using ImageJ software. Uncropped scans of blots are provided in the Supplementary Information.
Luciferase assays. AML12 and THLE5B cells were transfected at 70% confluency using Mirus TransIT-LT1 transfection reagent and analyzed at 100% confluency to control for cell-contact mediated inhibition of Hippo pathway by Dual-Glo Luciferase Assay (Promega). pRL-Tead4-Luciferase and Renilla control plasmid were gifts of Yimlamai and Camargo 35 . Incubation with verteporfin (10 µg ml −1 ) or vehicle (PEG300) were carried out for 24 h.
qRT-PCR and primer sequences. For qRT-PCR analysis, RNA was extracted from snap frozen liver or cells by Trizol (Ambion) followed by purification using the RNeasy Kit (Qiagen). One to 5 µg RNA were transcribed to cDNA using the RNA to cDNA EcoDry Premix (Clontech). IQ SYBR Green Supermix (Biorad) was used for quantitative PCR on the LightCycler480 System (Roche Diagnostics). Samples were analyzed in triplicates. Data are represented as the relative expression after normalizing to housekeeping genes (Gapdh, β-actin, 18S, Tbp) expression, respectively. Primer sequences are listed in Supplementary Table 2.
Gene expression analysis. Liver tissue was incubated in RNAlater (Ambion) and RNA extracted by Trizol (Ambion) according to the manufacturer's instructions, and further purified by the RNeasy Kit (Qiagen). RNA from 2 month old, female and male, age-matched Alb-CRE:Atg7 F/F and CRE-negative littermates (n = 4 per group) were analyzed on a MouseWG-6 v2.0 Expression BeadChip according to the manufacturer's protocol. Raw scanned data were normalized by using cubic spine algorithm implemented in the GenePattern genomic analysis toolkit (www. broadinstitute.org/genepattern). Probe-level expression data were collapsed into gene-level by calculating the median of multiple probes, and converted to human genes based on an orthologous mapping table provided by the Jackson laboratory (www.informatics.jax.org). Gene expression analysis was determined by R software and expressed as row z-score.
Bioinformatics and statistical data. Molecular pathway deregulations were determined in the genome-wide transcriptome dataset by surveying a comprehensive collection of pathway gene sets in Molecular Signature Database (MSigDB, www.broadinstitute.org/mdigdb) and a collection of liver cancer-related gene signatures in literature using Gene Set Enrichment Analysis (GSEA, www. broadinstitute.org/gsea).
Transcriptome profiles of 72 NAFLD-affected liver tissues were obtained from NCBI Gene Expression Omnibus database (www.ncbi.nlm.nih.gov/geo, accession number GSE49541). Genome-wide transcriptome profiles of 374 human HCC tissues were obtained from The Cancer Genome Atlas data portal (https://gdc. cancer.gov). Transcriptomic molecular HCC subtypes we were determined using Nearest Template Prediction (NTP) algorithm as previously described 61,69 . Atg7-KO gene signature was defined as differentially expressed genes in liver tissues from the Atg7-KO mice compared to control mice by t-test-based LOOCV validation (four biological replicates for Atg7-ko and control group, respectively). Transcriptional target genes of Hippo pathway (REACTOME_SIGNALING_BY_HIPPO), transcriptional target genes of Yap1 (CORDENONSI_YAP_CONSERVED_SIGNATURE) were obtained from Molecular Signature Database (MSigDB) (www.broadinstitute.org/msigdb). Modulation of the gene sets in each individual sample was determined by modified gene set enrichment analysis 70 .
Human HCC samples. Formalin-fixed, paraffin-embedded (FFPE) HCC tissues (n = 8) were analyzed by immunohistochemistry. The de-identified human HCC samples were obtained from Mount Sinai Biorepository with IRB-approved written informed consents from patients. The study had been approved by the Institutional Review Board of Icahn School of Medicine at Mount Sinai.

Data availability
The Gene Expression Omnibus accession number for the transcriptome profiles reported in this paper is GSE67676. The Gene Expression Omnibus accession number for 72 NAFLD-affected livers is GSE49541. Genome-wide transcriptome profiles of 374 human HCC tissues were obtained from The Cancer Genome Atlas data portal (https://gdc.cancer.gov).