Amplification of Glyceronephosphate O-Acyltransferase and Recruitment of USP30 Stabilize DRP1 to Promote Hepatocarcinogenesis Free

World-wide, hepatocellular carcinoma (HCC) is among the most lethal of human cancers. Although numerous inherited and acquired conditions can predispose to the development of HCC, infection by hepatitis viruses B and C accounts for more than 50% of all cases (1–3). Other risk factors include food contamination with aflatoxin, alcoholic and nonalcoholic cirrhosis, and nonalcoholic fatty liver disease. Excessive alcohol consumption and smoking further increase the overall risk (1–3). Gene mutations involving TP53 and CTNNB1, numerous copy-number variations, epigenetic alterations, and dysregulation of several miRNAs also likely play key roles in HCC initiation and/or progression (4).

Glyceronephosphate O-acyltransferase (GNPAT), which is mainly localized in peroxisome membranes, is essential for ether lipid (EL) synthesis. Defects in the biosynthetic pathways of EL are responsible for rhizomelic chondrodysplasia punctata, a serious peroxisomal disorder characterized by proximal limb shortening, recurrent respiratory tract infections, congenital cataracts, and seizures (5). In mice, GNPAT deficiency is associated with male infertility, defective eye development, and impaired neurotransmission (6). GNPAT deficiency in mice also seriously affects thymic maturation of semi-invariant natural killer T (iNKT) cells and iNKT cell antigen production (7). It has also been reported that the GNPAT variant p.D519G is associated with higher serum iron and serum ferritin levels following an iron challenge in healthy women and that it significantly worsens the iron overload in individuals with the p.C282Y^+/+ variant of the HFE gene (1–3). In addition, ectopic expression of alkylglycerone phosphate synthase, which is also essential for EL synthesis and which associates with GNPAT during the synthesis of plasmalogens (5), alters the balance of structural and signaling lipids to fuel cancer growth. However, there is little information as to how GNPAT functions in liver tumorigenesis.

Dynamin-related protein 1 (DRP1) plays an important role in mitochondrial and peroxisome membrane fission (8). DRP1 contains an N-terminal GTPase domain, a bundle signaling element, a middle domain (MD), and a GTPase effector domain (9). Extensive investigations have revealed that defects in DRP1 can cause neurodegenerative disorders such as Alzheimer disease, Huntington disease, and neonatal lethality (10, 11). Other studies indicate that the C452F mutant of DRP1 may lead to energy deficiency and cardiomyopathy. Inhibition of DRP1-mediated mitochondrial fission by metformin efficiently attenuates diabetes-induced atherosclerosis (12, 13). Finally enforced expression of DRP1 in HCC cells not only leads to mitochondrial fission but also enhances HCC growth by promoting G₁–S phase transition via pathways that involve p53 and NF-kB (14). Enforced overexpression of Drp1 in rat fibroblasts leads to a state of chronic increased mitochondrial fission, decreased mitochondrial mass, and a concurrent attenuation of oxidative phosphorylation and ATP production. This in turn is accompanied by the activation of AMP-dependent protein kinase that presumably represents an abortive attempt to restore a normal energy balance (15). In Saccharomyces, cerevisiae Drp1 is also involved in peroxisome fission, especially under peroxisome-inducing growth conditions (12, 13).

DRP1 undergoes multiple posttranslation modifications including phosphorylation, SUMOylation, S-nitrosylation, and ubiquitination. Phosphorylation of DRP1 at Ser₆₁₆ by PDK1 in mitosis and PKCδ under stress conditions increases mitochondrial fragmentation, and phosphorylation of DRP1 at the same site by ERK2 promotes MAPK-driven tumor growth and cell programming (16–19). Similarly, phosphorylation of DRP1 at Ser₆₃₇ by c-AMP and PKA inhibits its GTPase activity, whereas its dephosphorylation by calcineurin and PP2A modulates the translocation to mitochondria and dynamic balance (20–22). DRP1 SUMOylation by MAPL stabilizes the mitochondrial/ER platform required for apoptosis, whereas de-SUMOylation by SENP5 and SENP2 blocks cell death (23–25). In contrast, DRP1 de-SUMOylation by SENP3 facilitates DRP1 localization at mitochondria, thereby promoting fragmentation and cytochrome c release (26). Previous studies have shown that MARCH5 and PARK2 degrade DRP1 protein and maintain normal mitochondrial morphology (9, 27, 28).

Here, we report that GNPAT is upregulated in patients with HCC and is recruited by USP30 to deubiquitinate and stabilize DRP1, thus regulating mitochondrial morphology and hepatocarcinogenesis. USP30 is a deubiquitinylase that normally localizes to the outer mitochondrial membrane and acts as a PARK2 inhibitor, thus preventing autophagy (27, 28). These findings identify a relationship between GNPAT and USP30-mediated DRP1 stabilization that contributes to HCC development. Further, they reveal a previously unappreciated cross-talk between mitochondrial and peroxisomal biogenesis and function that may be important in both normal and neoplastic cell metabolism.

HCC samples were used as previously described (29). Written-informed consent was obtained from all patients at the Union Hospital in Wuhan, China. The studies were conducted in accordance with Declaration of Helsinki and approved by the review board of Wuhan University. The diagnoses of all samples were confirmed by histologic review.

Human embryonic kidney HEK293 cells and human retinal pigmented epithelium cells immortalized with hTERT were obtained from the American Type Culture Collection in 2009, whereas normal human hepatocytes HL-7702, human liver tumor-derived cell lines BEL7402, FHCC98, Hep3B, Huh7, HepG2, HCCLM9, and MHCC97L were obtained from China Center for Type Culture Collection and Cell Bank of Type Culture Collection of Chinese Academy Sciences between 2014 and 2015, respectively. All cells, regularly authenticated by short tandem repeat analysis and tested for absence of Mycoplasma contamination, were used within 5 passages after thawing and cultured as previously described (29). Cycloheximide (R750107), proteasome inhibitor MG132 (M8699), and autophagy inhibitor 3-methyladenine (3-MA; M9281) were purchased from Sigma-Aldrich. Antibodies used in this study are listed in Supplementary Table S1.

The human GNPAT, DRP1, KDM1A, Myc, and Ubiquitin-coding sequences were amplified and cloned into pHAGE-CMV-MCS-PGK-3 × Flag and pCMV-HA vectors. Mutations in the Ubiquitin and DRP1 cDNA sequences were generated by overlap extension PCR. Human GNPAT, DRP1, USP30, Myc, and KDM1A shRNA vectors were obtained from GENECHEM. siRNAs and short hairpin DNAs to inhibit murine Gnpat were designed and synthesized by GENEWIZ. The latter were annealed and then inserted into pLKO.1-puro vector. The DRP1-luciferase vector was generated by cloning the human DRP1 coding sequences to the luciferase gene tail of pcDNA3.1(+)-5′flag Luc (a kind gift from Dr. Ping Wang, Life Sciences of East China Normal University, Shanghai, China; ref. 30). To detect GNPAT and DRP1 localization in peroxisomes, the peroxisomal targeting sequence serine–lysine–leucine was inserted into the pmCherry-N1 vector (Clontech; ref. 29). GNPAT and DRP1 sgRNAs were designed by an online tool (http://www.genome-engineering.org) and cloned into the Lenti-CRISPER v2 vector (Addgene; ref. 31). Primers used in this study are listed in Supplementary Table S2. Transfection and the establishment of stable cell lines were performed as previously described (29). Cell viability was measured by MTT assay according to the manufacturer's instructions (Promega).

Subcellular fractionation was performed as described previously (32, 33). In brief, cells were trypsinized, collected in PBS, resuspended in HEPES–mannitol–sucrose buffer, and homogenized through a 25G needle. Nuclei and unlysed cells were pelleted from the homogenate, and the mitochondria were recovered from the supernatant by centrifugation. The crude mitochondrial pellet was resuspended and washed for further experiments.

Transfected cells were lysed in 1 mL lysis buffer [20 mmol/L Tris (pH 7.4), 300 mmol/L NaCl, 1% Triton, 1 mmol/L EDTA, 10 mg/mL aprotinin, 10 mg/mL leupeptin, and 1 mmol/L PMSF]. Sepharose beads were washed 3 times with 1 mL lysis buffer containing 150 mmol/L NaCl. Coimmunoprecipitation (Co-IP) and immunoblot analyses were performed as described elsewhere (34).

Chromatin immunoprecipitation (ChIP) was performed as described previously (34, 35). In brief, intracellular protein–DNA complexes were cross-linked with 1% formaldehyde, sonicated, and subjected to chromatin-conjugated immunoprecipitation using specific antibodies. After reversal of cross-links, precipitated DNA was purified and analyzed by qPCR with the primers indicated in Supplementary Table S2.

For qRT-PCR, total RNA was isolated using TRIzol (Invitrogen) followed by DNase (Thermo Scientific) treatment. Reverse transcription was performed with a cDNA Synthesis kit (Promega), and qPCR was performed using SYBR Green Master Mix (Bio-Rad) using standard protocols. All primer sequences used are shown in Supplementary Table S1. β-Actin was used as an internal control.

For lipid uptake, 10⁶ cells were trypsinized, collected in PBS, and stained at 37°C for 30 minutes in the dark using BODIPY-493/503 (Sigma; D3922). Flow cytometry was performed using a FACStar flow cytometer (BD Biosciences; refs. 36–38).

For lipid synthesis, cells were cultured in a 6-well dish in growth medium. For labeling, ³H-acetic acid (1.59 Ci/mmol, Perkin Elmer, 37 KBq/mL) was added to the wells and incubated at 37°C overnight. The next day, cells were disrupted with lysis buffer (100 mmol/L Tris-HCl, pH 7.5, 150 mmol/L NaCl, 2% SDS, 10% glycerol, 12.5 mmol/L EDTA, 1% Triton X-100, and 0.5% NP-40) per well and incubated at 37°C for 30 minutes. Organic extraction was performed in chloroform following the protocol of Bligh and Dyer, and lipid-soluble products were measured by scintillation counting (Tri-Carb 2910 TR; Perkin Elmer; refs. 36–38).

Cellular nonesterified fatty acid (NEFA) levels were determined with a commercially available ELISA kit (AO42-2; Jiancheng Bio.) according to the manufacturer's instruction. Serum TG levels were measured using the Infinity Triglycerides Reagent (Thermo Scientific).

Human GNPAT promoter sequences were inserted into pGL3-basic luciferase vector (Promega). Luciferase assays were performed as described previously (34).

All animal studies were approved by the Animal Care Committee of Wuhan University. For xenograft experiments, 4-week-old male BALB/c nude mice were purchased from SLAC Laboratory Animal Corporation and maintained in microisolator cages. Detailed procedures were described previously (29).

For the DEN/carbon tetrachloride (CCl₄)–induced HCC model, C57/B6 mice were intraperitoneally administrated 25 mg/kg of diethylnitrosamine (DEN; Sigma) on day 15 of life. Following this, CCl₄ (Sigma) was then injected weekly for 12 weeks (n = 10 per group). Concentrated shRNA-lentiviruses were injected by tail vein. After 28 weeks, mice were sacrificed for analysis.

Liver sections were fixed in 10% formalin and embedded in paraffin according to standard protocols. Livers were sectioned and stained with hematoxylin and eosin (H&E). Ki-67 staining was performed according to the supplier's protocol (mouse monoantibody, 1:200, Sc-25280; Santa Cruz Biotechnology).

Confocal microscopy was performed as previously described (35). Briefly, cells were fixed with 4% paraformaldehyde at room temperature, permeabilized, and then stained with the primary antibody against DRP1 (1:100), GNPAT (1:100), or TFAM (1:100) for 1 hour. After exhaustive removal of the primary antibody by washing, a second incubation was performed with horseradish peroxidase–conjugated goat anti-mouse (A-11005 or A-21050; Invitrogen) or goat anti–rabbit-488 antibodies (A-11008; Invitrogen; 1:500). Fluorescence imaging was performed on a Leica SP8 confocal microscope and images analyzed with Laika AF Lite software. To determine peroxisome morphology, cells were randomly selected for quantitative analysis.

This was performed as described previously (32, 33). In brief, cells were fixed in 2.5% glutaraldehyde (Sigma) in PB buffer. After three washes in PB buffer, cells were scratched, collected, and centrifuged. Then cells were postfixed in 1% osmium tetroxide (OSO4). After further washes with PB, cells were dehydrated in 50%, 75%, 95%, and 100% ethanol, and then further dehydrated in 100% acetone twice. The samples were then infiltrated in 1:1 acetone/epoxy resin, 1:2 acetone/epoxy resin, 100% epoxy resin, and finally 100% epoxy resin at 65°C for polymerization. The blocks were cut into micrometer sections with a diamond knife, picked up on 200-mesh grids, stained, and observed according to the standard procedures with Transmission Electron Microscope (JEM-1400plus).

Datasets for HCC were downloaded from the Cancer Genome Atlas (TCGA) data portal (http://www.tcga-data.nci.nih.gov). GNPAT expression data were assessed, and Kaplan–Meier curves were calculated in human HCC tissues from the TCGA mRNA dataset. GSE datasets were analyzed from NCBI GEO databases (https://www.ncbi.nlm.nih.gov), which were reported from indicated gene expression profile array. Cancer Cell Line Encyclopedia (CCLE) datasets (https://portals.broadinstitute.org/ccle/home) were used to analyze gene expression and DNA copy number in HCC cell lines.

Survival curves were plotted by the Kaplan–Meier method with GraphPad Prism 5 software using the log-rank test and correlation with a Spearman rank correlation test. Other statistical significance (P < 0.05) was assessed by the Student t test. Data were presented as the mean ±SD.

To identify deregulated metabolic genes involved in HCC progression, an unbiased bioinformatic screening of 2,783 metabolic enzyme genes was conducted in the HCC database from TCGA (39). Interestingly, 896 of these genes were commonly upregulated in HCCs (1.5-fold threshold) with the mRNA abundance from 442 of these genes being >100 RPKM (Reads Per Kilobase Million) by RNA-sequence analysis (Fig. 1A and B). As tumor cells almost always harbor unstable genomic DNA (40), we sought candidates from the above group whose overexpression could be explained by gene amplification. Interestingly, 36 such examples were noted as being highly correlated (R > 0.5). Nine of these 36 genes reside on chromosome 1q, which is one of the most frequently amplified chromosomal regions in HCCs (Fig. 1C, R > 0.62; refs. 1, 4). To evaluate the role of each of these 9 genes on human HCC cell growth, 3 siRNAs against each were synthesized and used to inhibit their expression in human HCC cells HepG2, Hep3B, and MHCC97L. MTT assays showed that GNPAT silencing significantly inhibited growth of each of the three cell lines (Fig. 1D; Supplementary Fig. S1A and S1B), and hence, we focused on GNPAT in the subsequent work described below.

To study the clinical significance of GNPAT expression in HCC, the publicly available TCGA, GSE14520, GSE40367, and CCLE datasets of HCC cases and cell lines were analyzed. On average, GNPAT mRNA levels were expressed at higher levels in HCCs relative to normal hepatic tissues in all tested databases (Fig. 1E; Supplementary Fig. S1C–S1E). Human HCC cell lines also contained similarly elevated levels of GNPAT transcripts than did nontransformed hepatocyte lines (Supplementary Fig. S1F). In addition, Western blot assays showed GNPAT protein levels to be significantly upregulated in HCC tissues and cell lines (Supplementary Fig. S1G and S1H). In patients with HCC, those with stage II/III/IV had significantly higher GNPAT expression than those with stage I (Fig. 1F). Also in patients with HCC, those with high GNPAT expression had significantly worse survival than those with low GNPAT expression (Supplementary Fig. S1I).

The GNPAT gene, located on Chr.1q42.2, was found to be significantly amplified in both HCCs and most HCC cell lines. As shown in Fig. 1G–I and Supplementary Fig. S1J–S1L, GNPAT mRNA levels in HCCs also generally correlated with the degree of GNPAT gene amplification (Log₂ SNP > 0). Interestingly, patients with HCC with amplified GNPAT genes had significantly worse survival than those without gene amplification (Fig. 1J). As a control, FAM98A, localized next to GNPAT, expression was not upregulated and not associated with patient survival in HCCs (Supplementary Fig. S1M). These findings suggest that GNPAT amplification and overexpression are involved HCC pathogenesis and/or contribute to other clinically relevant features of the tumor.

To better study how GNAPT contributes to HCC development, we performed SDS-PAGE and mass spectrometry following ectopic expression of GNPAT to identify its interacting partner proteins. Interestingly, DRP1, which tends to localize to mitochondria and peroxisomes, was identified as one of the most prominent interactors (Fig. 2A; Supplementary Fig. S2A). The interaction between endogenous GNPAT and DRP1 was also confirmed by Co-IP experiments in the HepG2 and Hep3B cell lines (Fig. 2B). To identify the region of each protein responsible for the interaction, various DRP1 and GNPAT truncation mutants were generated and coexpressed in HEK293 cells. We found the GTPase domain of DRP1 and the middle domain of GNPAT to be the critical interacting region of each protein (Fig. 2C and D). Immunofluorescence staining further indicated that GNPAT and DRP1 were colocalized in mitochondria and peroxisomes (Fig. 2E; Supplementary Fig. S2B; refs. 41, 42). These results were confirmed by subcellular fractionation experiments (Fig. 2F).

Mitochondrial functions are related to mitochondrial dynamics, whereas DRP1 has a critical role in mitochondrial morphology (8, 43). To investigate whether GNPAT affects mitochondrial dynamics, we examined mitochondrial morphology in cells with GNPAT inhibition. As shown in Fig. 2G and Supplementary Fig. S2C and S2D, GNPAT inhibition impaired mitochondria fission as indicated by increased numbers of elongated mitochondria and decreased numbers of fragmented ones. Also these defects could be rescued with overexpression of shRNA-resistant GNPAT and DRP1 but not GNPAT ΔCD or DRP1 ΔGTP. These findings suggest that the regulation of mitochondrial morphology mediated by the GNPAT–DRP1 interaction may be mediated at the level of mitochondria fission.

Next, the effect of GNPAT on DRP1 expression was evaluated. GNPAT inhibition in Hep3B and HepG2 cells resulted in a marked decrease of DRP1 at the protein level without affecting mRNA abundance (Supplementary Figs. S2C and S3A). This indicated that GNPAT likely affects DRP1 protein turnover. GNPAT inhibition was associated with a significantly shortened DRP1 half-life after treatment with cycloheximide, which blocked de novo protein synthesis (Fig. 3A; Supplementary Fig. S3B). To identify the pathway(s) responsible for GNPAT-mediated DRP1 degradation, NH₄Cl, MG132, and 3-MA were used to inhibit lysosomal, proteasomal, and autophagocytic pathways of protein degradation, respectively. GNPAT inhibition–mediated degradation of DRP1 was completely inhibited by the proteasome inhibitor MG132 but not by NH₄Cl or 3-MA (Fig. 3B; Supplementary Fig. S3C). These results suggest that GNPAT inhibition mediated degradation of DRP1 through the proteasome pathway.

Given that deubiquitinases (DUB) modulate protein stability, we hypothesized that GNPAT recruits one or more DUBs to deubiquitinate and stabilize DRP1. As GNPAT and DRP1 are mainly expressed in mitochondria and peroxisomes, 14 DUBs known to localize to these organelles were selected for further examination. To identify DUBs that could stabilize DRP1, firefly luciferase was fused to the N terminus of DRP1, and the fusion protein (DRP1-Luc) was used as a reporter of DRP1 stability in the presence of each of the above 14 DUBs (30). Our results showed that ectopic expression of USP30 and USP48 increased the luciferase activity of DRP1-Luc (Fig. 3C; Supplementary Fig. S3D). Consistent with this, both GNPAT and DRP1 copurified with immunoprecipitated USP30 and USP48 (Supplementary Fig. S3E). Further assays showed that ectopic GNPAT expression increased USP30-mediated luciferase activity of DRP1-Luc, whereas GNPAT inhibition produced the opposite results. In contrast, USP48-mediated luciferase activity of DRP1-Luc was not affected by GNPAT expression or inhibition (Fig. 3D and E). Co-IP experiments confirmed that USP30 but not USP48 was recruited by GNPAT to promote DRP1 stability (Supplementary Fig. S3F). Also USP30 inhibition in Hep3B and HepG2 cells resulted in a marked decrease of DRP1 at the protein level (Supplementary Fig. S3G).

To investigate whether USP30 could deubiquitinate DRP1, HA-tagged wild-type (WT) USP30 and its catalytically inactive (CI, C77S) mutant were generated. Co-IP assays showed that WT but not mutant USP30 deubiquitinated DRP1 (Fig. 3F). Further studies showed that ectopic GNPAT expression increased USP30-mediated DRP1 deubiquitination (Fig. 3G; Supplementary Fig. S3H), whereas USP30 depletion with ectopic GNPAT expression impaired DRP1 deubiquitination (Supplementary Fig. S3I). To identify the type of GNPAT recruiting USP30-mediated deubiquitination of DRP1, several vectors expressing HA-tagged mutant Ubiquitin (K6O, K11O, K27O, K29O, K33O, K48O, and K63O), which contained substitutions of arginine for all lysine resides except the lysine at this position, respectively, were used. Deubiquitination of DRP1 was detected only in the presence of WT and K48O, consistent with the idea that GNPAT enhanced USP30-mediated K48-linked deubiquitination of DRP1, but not other position-linked deubiquitination of DRP1 (Fig. 3H). Further experiments were performed with the mutant ubiquitin K48R, which included a single lysine to arginine substitution at position 48. Mutant on K48 could not remove the ubiquitination in contrast to WT, which classified that GNPAT downregulated the K48 ubiquitin chain on DRP1 (Supplementary Fig. S3J). These results indicate that GNPAT promotes USP30-mediated K48-linked deubiquitination of DRP1.

Additional DRP1 ubiquitination sites were further confirmed with USP30 inhibition. Public proteomic database assays showed three potential ubiquitin sites of DRP1 protein (K160/238/532), all of which were mutated to arginine to generate single, double, or triple “KR” mutants. We found that DRP1 ubiquitination was gradually decreased in parallel with the number of KR mutations. Notably, the triple KR mutant almost completely abolished DRP1 ubiquitination (Fig. 3I). In contrast, USP30 inhibition increased the WT DRP1 ubiquitination, but did not affect ubiquitination of the triple DRP1 mutant (Supplementary Fig. S3K). This suggests that these three arginine residues are all critical sites of ubiquitination.

We next investigated the effects of GNPAT and DRP1 modulation on HCC development. Because GNPAT and DRP1 are highly expressed in HepG2 and Hep3B cells, these two cell lines were exploited for further study (Supplementary Fig. S1H). shRNA-mediated inhibition of either GNPAT or DRP1 significantly retarded cell proliferation in vitro (Fig. 4A and B; Supplementary Figs. S2C and S4A) and in vivo xenograft growth (Fig. 4C and D; Supplementary Fig. S4B).

GNPAT is critical for fatty acid synthesis, and DRP1 promotes glucose metabolism, presumably by promoting mitochondrial fission, reducing reliance on Oxidative phosphorylation, and increasing glycolysis (44, 45). Depletion of GNPAT or DRP1 also decreased lipid uptake (Fig. 4E), lipogenesis (Fig. 4F), and NEFA levels (Fig. 4G), indicating that both proteins regulate lipid metabolism. Knockout of GNPAT and DRP1 using a CRISPR/Cas9-based approach confirmed the above results (Supplementary Fig. S4C–S4F). Collectively, these results indicate that both GNPAT and DRP1 are critical for HCC tumor growth in vitro and in vivo and that this growth might be influenced by lipid levels.

Because GNPAT and DRP1 inhibition diminishes the malignant phenotypes of HCC cells, we explored whether GNPAT's oncogenic effects were DRP1-dependent. HepG2 and Hep3B cell lines stably expressing ectopic GNPAT or DRP1 grew more rapidly than their control counterparts both in vitro and in vivo. These effects were reversed by inhibiting DRP1, indicating that it was the direct mediator of GNPAT's growth-promoting effects (Fig. 5A–F; Supplementary Fig. S5A). Further ectopic overexpression of DRP1 but not DRP1 ΔGTP or GNPAT ΔCD was able to rescue the phenotypic changes in HCC cells mediated by GNPAT inhibition (Fig. 5A–F; Supplementary Figs. S2C, S4A, S5A–S5C). These results indicate that DRP1 is necessary for and mediates GNPAT-malignant phenotypes.

We next examined the expression of GNAPT and DRP1 in an autologous HCC model induced by a combination of DEN and CCl₄. Consistent with our prior observation in human samples, GNAPT and DRP expressions were all upregulated at various times during disease development (Fig. 6A; Supplementary Figs. S1G and S6A).

To determine whether GNPAT and DRP1 inhibition could attenuate autologous HCC tumor growth in the above model, concentrated lentivirus-containing shGnpat and shDrp1 expression vectors were administered to DEN/CCl₄-treated mice for 4 weeks (weeks 8–11), and the tumor burden was assessed at week 28. Tumors treated with lentivirus-expressing shGnpat and shDrp1 were smaller and showed reduced expression of GNAPT and DRP1 as well as genes involved in lipogenesis and fatty acid uptake (Fig. 6B–F; Supplementary Fig. S6B–S6D). These mice also showed lower serum levels of NEFA and TG (Fig. 6G; Supplementary Fig. S6E).

GNPAT mRNA levels tended to be expressed at much higher levels in HCC tissues and cell lines than could be accounted for even by gene amplification. This suggested an additional level of gene copy-number–independent control that we hypothesized was transcriptionally mediated. Examination of the GNPAT promoter identified conserved c-Myc (Myc) binding E-boxes, located within CpG islands 5 kb upstream and 2 kb downstream of the transcription start site (TSS; Fig. 7A). Myc is a master transcriptional factor whose targets contribute to growth and metabolism under both normal and neoplastic conditions (46). As shown in Fig. 7B and C, Myc inhibition resulted in downregulation of both GNPAT mRNA and protein levels in both HepG2 and MHCC97L HCC cells but not GNPAT-amplified Hep3B cells (Supplementary Fig. S7A and S7B), thereby indicating that at least some portion of GNPAT upregulation is likely mediated by the excessive levels of Myc expression associated with HCC. Indeed, a ChIP assay performed in HCC cells revealed that Myc binds the GNPAT promoter region containing the most conserved and proximal upstream E-boxes (Fig. 7D; Supplementary Fig. S7C), thereby supporting the notion that GNPAT expression levels are directly regulated by Myc.

Because Myc associates with several histone demethylases to facilitate cancer progression (47–50), we hypothesized that its interaction with one or more of these regulates GNPAT transcription. Therefore, ChIP experiment was performed to investigate which of the seven known histone demethylases bind to the GNPAT promoter in association with Myc. Our results showed that immunoprecipitation of KDM1A efficiently enriched the GNPAT promoter region containing the proximal E-box in control HepG2 and MHCC97L cells but not in cells in which Myc levels were depleted by the expression of an shRNA (Fig. 7D; Supplementary Fig. S7C and S7D). The direct interaction between endogenous Myc and KDM1A was also confirmed in HepG2 and MHCC97L cells by Co-IP experiments (Supplementary Fig. S7E). Furthermore, the use of deletion mutants showed that the region encompassing residues of 320–439 of Myc and containing the entirety of its bHLH-ZIP dimerization domain is required to interact with KDM1A (Supplementary Fig. S7F, left), whereas residues 474–852 of KDM1A were required to interact with Myc (Supplementary Fig. S7F, right). KDM1A is the main histone demethyltransferase that demethylates histone H3 at lysine 4 (H3K4me2), thereby mediating epigenetic activation (46). We therefore determined whether the H3K4me2 could also be detected at the GNPAT promoter. Indeed, immunoprecipitation of H3K4me2 efficiently enriched the E-box–containing GNPAT promoter region in control HepG2 and MHCC97L but not in Myc-depleted cells (Fig. 7D; Supplementary Fig. S7C). These results suggest that Myc recruits KDM1A to bind the GNPAT promoter at a functional E-box site and that KDM1A-mediated histone demethylation of H3K4me2 contributes to Myc-induced GNPAT activation.

Next, the potential role of KDM1A in regulating GNPAT expression was evaluated. ShRNA-mediated KDM1A inhibition significantly decreased GNPAT mRNA and protein levels in control HepG2 and MHCC97L cells but not if they were first depleted of Myc, but not in Hep3B cells (Fig. 7B and C; Supplementary Fig. S7A–S7C). We then tested the effects of Myc and KDM1A in a luciferase reporter assay of the GNPAT promoter. Coexpression of both plasmids in HepG2 and MHCC97L cells increased luciferase activity in reporters containing either of the WT E-box sites but not in those with mutated sites. Either of the two E-boxes appeared to be sufficient for upregulating the activity of the reporter, although their effects did not appear to be additive (Fig. 7E; Supplementary Fig. S7G).

Previous studies have shown that Myc is activated in and essential for HCC progression (46, 51). In HepG2 and MHCC97L cells, the ectopic expression of GNPAT but not a dominant-negative form of the protein could partially rescue the oncogenic phenotypes in Myc-depleted HCC cells (Fig. 7F, left, G, and H; Supplementary Fig. S7H, top). This suggested that GNPAT is a relatively important Myc target with regard to transformation and can recapitulate this complex function in the presence of reduced or absent Myc levels as has been reported for only a limited number of previously described Myc targets (52, 53).

Finally, we asked whether GNPAT is required for the full expression of Myc oncogenic phenotypes. We first showed that the stable expression of Myc in HepG2 and MHCC97L cells enhanced both their in vitro growth and in vivo tumorigenicity. However, these phenotypes were impaired following GNPAT inhibition (Fig. 7F, right, G, and H; Supplementary Fig. S7H, bottom). These results suggest that GNPAT partially mediates and is required for the optimal expression of Myc-mediated oncogenic phenotypes.

Because reprogrammed metabolism is now appreciated to be a hallmark of cancer and a promising therapeutic target (54), the identification of key enzymes that mediate this reprogramming and contribute to cancer growth and/or survival is of paramount importance. In the current work, an unbiased bioinformatic screening of 2,783 metabolic enzymes identified GNPAT, a member of the EL biosynthetic pathway, as an important facilitator of HCC development. Although some of GNPAT's overexpression is due to gene amplification of a critical region on Chr.1q42.2, the remainder appears to involve transcriptional activation mediated by a complex comprised of Myc and the histone demethylase KDM1A.

Surprisingly, GNPAT was found to interact with and stabilize DRP1 located in mitochondria and peroxisomes. Recent reports have revealed that DRP1 interacts with the head group of phosphatidic acid and saturated acyl chains of phospholipid (55), which further affirms our results. It is well known that DRP1 plays a key role in mitochondrial fission (9), indicating that the interaction of GNPAT and DRP1 is essential for mitochondrial function. Significant cross-talk between mitochondrial and peroxisomal compartments has also been documented (8) as evidenced by the fact that DRP1 is important for mediating fission events in both organelles. It is tempting to speculate that GNPAT is one of the factors that mediate this activity and/or their physical interaction.

DRP1 is also involved in many posttranslational modifications that regulate mitochondrial and peroxisome function (9). In support of this, our study has shown that GNPAT recruits USP30 to deubiquitinate and stabilize DRP1. USP30 is known to impair PARK2-mediated mitophagy and pexophagy and, along with GNPAT and DRP1, localize to the peroxisome (27, 28).

The genetic suppression of either GNPAT or DRP1 profoundly impaired xenograft growth and further attenuated DEN/CCl₄-induced hepatocarcinogenesis in mice. Because previous results had shown that GNPAT regulates free cholesterol and plasmalogen homeostasis (44), our results confirmed that GNPAT inhibition also decreased lipogenesis and fatty acid uptake in HCCs. Thus, it is tempting to speculate that GNPAT's regulation of tumor growth may involve both its regulation of mitochondrial morphology and metabolic functions. Such functions might include regulating the supply of critical metabolites necessary for tumor growth as well as protection against the high levels of reactive oxygen species that are typically generated by rapidly dividing tumors, particularly those with elevated Myc levels. Independent of this, DRP1 has recently been reported to increase leptin sensitivity and glucose responsiveness (45), and our results indicate that DRP1 inhibition also attenuated lipogenesis and fatty acid uptake. Although the precise mechanism for this remains unclear, it is conceivable that this occurs via both mitochondrial and peroxisomal pathways involved in fatty acid oxidation. Recent work in animal models has demonstrated that both Myc-driven HCCs and β-catenin–driven hepatoblastomas redirect fatty acid metabolism away from oxphos-linked energy-generating pathway in favor of anabolic pathways in which fatty acids are used directly for the purpose of new membrane synthesis to sustain rapid growth (36, 37).

Myc plays a pivotal role in cell proliferation, tumorigenesis, and metabolism and is highly expressed in most tumors where it regulates target genes involved in these and other processes that support tumor growth (46, 51). We found that the complex comprised of Myc and KDM1A directly regulates GNPAT transcription in HCC cells and is required for maximal expression of some of the Myc-driven phenotypes. The resultant stabilization of DRP1 and the presumed increase in mitochondrial fission, arising as a consequence of the interaction of GNPAT and the USP30 deubiquitinase, could potentially provide an explanation for the significant reduction in overall mitochondrial DNA content (and probably mitochondrial mass) that has been observed in HCCs, hepatoblastomas, and multiple other human cancers (46, 51). Collectively, these findings suggest that targeting of GNPAT might be an effective therapeutic strategy in HCC, particularly for the 40% to 70% of tumors that are believed to be largely Myc-driven.

In summary, we propose a working model in which HCC progression is at least partially driven by gene amplification or a complex comprised of Myc/KDM1A, which transcriptionally activates GNPAT. The interaction of GNPAT with the USP30 deubiquitinase stabilizes DRP1 by inhibiting its Ub-dependent degradation, thereby regulating mitochondrial morphology, lipid metabolism, and liver tumorigenesis. Several critical new points of potential therapeutic intervention have been identified that will need to be investigated in future studies.

No potential conflicts of interest were disclosed.

Conception and design: L. Gu, Y. Zhu, E.V. Prochownik, Y. Li

Development of methodology: L. Gu, Y. Zhu

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Gu, Y. Zhu, Xi Lin, Y. Li

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Gu, Y. Zhu, Xi Lin, Y. Li, K. Cui, Y. Li

Writing, review, and/or revision of the manuscript: L. Gu, E.V. Prochownik, Y. Li

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E.V. Prochownik

Study supervision: Y. Li

We thank Professors Bo Zhong, Xiaodong Zhang, Zan Huang, and Min Wu (Wuhan University, Wuhan, China) for the USP30 shRNAs, the DUBs’ vectors, ubiquitin vectors, and demethylase antibodies, respectively, and Ping Wang (East China Normal University, Shanghai, China) for pcDNA3.1(+)-5′flag Luc vector.

This study was supported by grants from the National Natural Science Foundation of China (81772609 and 81472549) to Y. Li.

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.