Mitophagy promotes sorafenib resistance through hypoxia-inducible ATAD3A dependent Axis

The identification of novel targets for recovering sorafenib resistance is pivotal for Hepatocellular carcinoma (HCC) patients. Mitophagy is the programmed degradation of mitochondria, and is likely involved in drug resistance of cancer cells. Here, we identified hyperactivated mitophagy is essential for sorafenib resistance, and the mitophagy core regulator gene ATAD3A (ATPase family AAA domain containing 3A) was down regulated in hypoxia induced resistant HCC cells. Blocking mitophagy may restore the sorafenib sensitivity of these cells and provide a new treatment strategy for HCC patients. Hypoxia induced sorafenib resistant cancer cells were established by culturing under 1% O2 with increasing drug treatment. RNA sequencing was conducted in transfecting LM3 cells with sh-ATAD3A lentivirus. Subsequent mechanistic studies were performed in HCC cell lines by manipulating ATAD3A expression isogenically where we evaluated drug sensitivity, molecular signaling events. In vivo study, we investigated the combined treatment effect of sorafenib and miR-210-5P antagomir. We found a hyperactivated mitophagy regulating by ATAD3A-PINK1/PARKIN axis in hypoxia induced sorafenib resistant HCC cells. Gain- and loss- of ATAD3A were related to hypoxia-induced mitophagy and sorafenib resistance. In addition, ATAD3A is a functional target of miR-210-5p and its oncogenic functions are likely mediated by increased miR-210-5P expression. miR-210-5P was upregulated under hypoxia and participated in regulating sorafenib resistance. In vivo xenograft assay showed that miR-210-5P antagomir combined with sorafenib abrogated the tumorigenic effect of ATAD3A down-regulation in mice. Loss of ATAD3A hyperactivates mitophagy which is a core event in hypoxia induced sorafenib resistance in HCC cells. Targeting miR-210-5P-ATAD3A axis is a novel therapeutic target for sorafenib-resistant HCC.


Background
Mitophagy is a mitochondrial form of autophagy that is critical for mitochondrial quality control and homeostasis [1,2]. In response to various stimuli, mitochondria undergo self-depolarization and the damaged organelles are identified by autophagosomes which then fuse with lysosomes to complete the degradation process [3,4]. Defective or excessive mitophagy is a pathological factor in chronic diseases including parkinson's disease (PD), diabetes, myocardial ischemia reperfusion (MIR) injury, ankylosing spondylitis (AS) and cancers [5]. Several mitophagic pathways have been identified that counteract therapy-induced mitochondrial damage [5,6] . Guo et al. reported that reversing mitophagy inhibition by the PIK3CA/AKT1/MTOR /RPS6KB1 pathway reduced the risk of colorectal cancer (CRC) development [7]. Katreddy et al. found that EGFR down-regulation cleared ovarian cancer cells by activating the mitophagy/ mTORC2/Akt axis [8]. Furthermore, mitophagy suppression via knocking down key mitophagy receptors such as PINK1, FUNDC1 or AMBRA1 could chemosensitize cancer cells [9,10]. The E3 ubiquitin-ligase ARIH1 and not Parkin mediates PINK1-induced mitophagic responses to cisplatin/etoposide in breast and lung adenocarcinomas [11]. In fact, the predominance of certain mitophagy receptors or mediators in specific cancer subtypes is a decisive factor in therapeutic resistance via mitochondrial clearance [6,12], and the underlying mechanisms are still unclear.
The ubiquitin-dependent PINK1/Parkin pathway is the most widely studied mammalian mitophagy cascade. PINK1 is an outer mitochondrial membrane (OMM) Ser/Thr kinase that is stabilized upon mitochondrial depolarization, and induces Parkin E3 activity and recruits Parkin by phosphorylating it at Ser-65 [13]. Recent studies has shown that ectopic expression of PINK1 plays a dual role in cancer development and drug treatment [14]. Wang et al. reported that the induction of PINK1/Parkin-mediated mitophagy sensitized tongue cancer cells to ZnO NPs [15]. In contrast, Villa et al. showed that ARIH1/HHARI triggered PINK1dependent mitophagy in the breast and lung adenocarcinomas and protected them from chemotherapyinduced death [11]. In the recent study, our group showed that the AAA domain containing 3A (ATAD3A) protein could prevent excessive accumulation of PINK1 and therefore inhibited unnecessary mitophagy in stem and progenitor cells [16]. As an up-regulator of PINK1, ATAD3A spans the inner mitochondrial membrane with its two terminal domains in the outer membrane and the matrix [17]. However, little is known regarding its role in cancer development and drug sensitivity under hypoxia microenvironment of solid tumors. microRNAs are members of a large class of endogenous small noncoding RNAs that control gene expression and regulate a wide array of biological processes by binding to the 3′-untranslated regions (3′-UTR) directly to regulate target mRNA expression, and eventually promote target mRNA degradation or translational inhibition [18]. Several studies have reported miR-210, as one of the highly up-regulated miRNAs in hypoxic cells that is involved in numerous biological processes of the human body involving regulating mitochondrial metabolism, promoting the angiogenesis, proliferation and apoptosis. miR-210 has two versions, miR-210-3P and miR-210-5P. miR-210-3p is the guide-strand that integrates into the RISC (RNA induced silencing complex), whereas miR-210-5p is the passenger-strand that is inactivated through degradation [19,20]. miR-210-5P is up-regulated in several malignant tumors that are associated with a variety of functionally important targets involved in cancers [21]. However, whether miR-210-5p could regulate mitophagy to participate in drugsensitivity remains to be examined in depth.
Hepatocellular carcinoma (HCC) is the sixth most common fatal malignancy and the major cause of cancer-related deaths worldwide [22]. Most patients are diagnosed at the advanced stage which is highly recalcitrant to the current therapies [23]. The multi-target tyrosine kinase inhibitor sorafenib is the FDA-approved first-line systematic therapy for advanced HCC patients, which increased median survival from 7.9 to 10.7 months [24]. It blocks tumor cell proliferation and angiogenesis by inhibiting the Rad/Mitogen-activated protein (MAP)/extracellular signal-regulated kinase (ERK)/ MEK signaling cascade, as well as the kinase activity of vascular endothelial growth factor receptor (VEGFR) and platelet-derived growth factor receptor (PDGFR)-β. However, almost all HCC patients have developed sorafenib resistance within a few months [25]. Since the drug targets several signaling pathways, the tumors acquire resistance through different mechanisms, such as activation of compensatory signaling cascades and a hypoxic microenvironment [26]. It is crucial to understand the precise molecular mechanisms underlying sorafenib resistance in order to identify novel therapeutic targets and improve the clinical outcome of HCC patients. Although, sorafenib-induced autophagy contributes to drug resistance, it is unknown whether hyperactivated mitophagy is also involved in sorafenib resistance in HCC [27][28][29]. Regarding hypoxia is a major stimulus of mitophagy and also an important cause of sorafenib resistance [30][31][32], our objective is to determine the potential relationship between mitophagy and hypoxia-induced sorafenib resistance in HCC. We found that hypoxia induced sorafenib resistance in hepatoma cells was accompanied by hyperactivated mitophagy and a downregulation of ATAD3A expression. Thus, ATAD3A is a crucial mediator of hypoxia induced mitophagy signaling in HCCs and a novel therapeutic target for reversing sorafenib resistance.

Patients and tissue microarray
Liver cancer tissues and paried normal tissues were collected form the First Affiliated Hospital of Third Military Medical University and Shanghai Biochip Company Ltd (Shanghai, China). The informed consent was obtained from all patients.

Cell culture
Human hepatoma cell lines Huh7 and LM3 were obtained from the American Type Culture Collection (ATCC) and authenticated by the Cell Bank of Type Culture Collection of Chinese Academy of Science. The cells were maintained under recommended conditions. For hypoxia treatment, the cells were either cultured in a sealed hypoxia chamber (Thermo fisher, Inc.) containing 1% O 2 , 5% CO 2 and 94% N 2 , or treated with 100 mM CoCl 2 for 24 or 48 h. Sorafenib-resistant Huh7 (Huh7-SR) and LM3 (LM3-SR) cells were enriched by steadily increasing the drug dose, while hypoxia-induced sorafenib resistant cell lines (Huh7-H-SR and LM3-H-SR) were established by culturing under 1% O 2 with increasing drug doses.

Virus production
The pLVX-CMV-EGFP-3FLAG-PGK-Puro lentiviral vector expressing full-length human ATAD3A was purchased from SunBio (Shanghai. China). The pMAGic2.1-CMV-HygroR-U6 shRNA lentivirus vector purchased from SunBio (Shanghai. China), and the shRNA sequences are shown in Additional file 1. Human HEK293T cells (American Type Culture Collection) were cultured in 6-well plates till~70% confluent, and co-transfected with 2 μg overexpression or knockdown virus vector, 1 μg pMD2.G and 1 μg psPAX2 lentivirus packaging vectors using Lipofectamine 2000 (Invitrogen) according to the manufacturer's protocol. The cells were maintained in high-glucose DMEM containing10% FBS, 2 mM glutamine, and100 units/ml penicillin and streptomycin. The supernatants with virus were harvested twice at 48 h and 72 h, filtered through a 0.45 μm syringe filter and frozen in liquid nitrogen.

Target cell transduction
HCC cells were cultured in complete DMEM, and 1.5 × 10 5 cells were mixed with 450 μl virus-containing supernatant in the presence of 4 μg/ml polybrene (Sigma). The cells were seeded into a 6-well plate, and the medium was changed 12 h post-infection. After 48 h, the infected cells were trypsinized and seeded into 10 cm culture dish with 4 μg/ml puromycin (Thermo Fisher Scientific). The stably transduced cells were selected over 48 h, and harvested 6 days post-infection to determine knockdown efficiency.

Western blotting
Total protein was extracted from cancer cells using Mammalian Protein Extraction Buffer (P0013, Beyotime, Beijing, China) supplemented with protease inhibitor cocktail (87786, ThermoFisher, USA). The mitochondrial fractions were separated using Mitochondria Isolation Kit (number: SM0020, Solarbio, Beijing, China) according to the manufacturer's protocol, and the protein was extracted as above. Equal amounts of protein lysates were separated by SDS-PAGE gel and electrotransferred to PVDF membrane (Millipore, USA). After blocking in 5% milk-PBST for 2 h at 37°C, the membranes were incubated overnight with primary antibodies (Additional file 2) at 4°C and with secondary antibodies at 37°C for 2 h. The positive bands were visualized with Immobilon Western Chemiluminescent HRP Substrate detection reagent (Millipore, USA), and acquired using a ChemiDoc™ imaging System (Bio-Rad, USA).

Quantitative real-time PCR
Total RNA was extracted from cancer cells using Trizol Reagent (Invitrogen, USA), and qRT-PCR was performed using SYBR Prime Script RT-PCR kit (TaKaRa, Japan) on a Rotor-Gene 6000 real-time genetic analyzer (Corbett Life Science, USA). The primer sequences and the product sizes are listed in Additional file 3. Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was used as the internal control. The PCR conditions were as follows: denaturation at 95°C for 2 mins, followed by 40 cycles of amplification and quantification (95°C for 5 s, 55°C-57°C for 30 s), and melting curve (55°C-95°C, with 0.5°C increment each cycle). Each sample was tested in triplicates.

Apoptosis assay
For apoptosis assays, the suitably treated cells were washed twice with cold PBS, and resuspended in binding buffer at the density of 1 × 10 6 cells/ml, and distributed into 500 μl aliquots (1 × 10 5 ) in 2 ml tubes. 5 μL of Annexin V-FITC were added to each tube. Cells were incubated at room temperature (25°C) for 15 min in a dark environment, and then analyzed on a FACS Calibur (BD Biosciences, USA), and results were calculated using Cell Quest software (BD Biosciences, USA).

Luciferase reporter assay
The 3′-UTR region of human ATAD3A was amplified by PCR from genomic DNA and cloned downstream to the firefly luciferase coding region in the pMIR-REPORTTM plasmid. The 293 T cells were seeded in 96-well plates, and co-transfected with 100 ng/ml reporter plasmid and 50 nM miR-210-5P or NC mimics using Lipofectamine 2000 24h late. Luciferase activity were measured after 72 h using the Dual-Glo Luciferase Assay System (Promega, Madison, WI, USA). All experiments were performed in triplicates.

Colony formation assay
For colony formation assay, 5 × 10 2 cells were seeded in 6-well plates with sorafenib 5.13 μM (Huh7) or 7.92 μM (LM3) as appropriate. The conditioned medium was removed 3 days later and replaced by medium containing 10% serum for 2 weeks. The ensuing colonies were stained by crystal violet and counted. The experiment was performed thrice.

Mitophagy assay
Western blotting was used to test LC3 mobility shift by analyzing the expression levels of LC3 I/II, and mitochondrial marker protein TOMM20 and TOMM70. Subcellular localization of mitophagy bodies was tracked by transmission-electron-microscopy (TEM, JEM-1230, Japan). The cells were fixed in cooling 2% glutaraldehyde and 0.1 M cacodylate buffer at 4°C overnight, then exposed to phosphate buffer containing 1% osmium tetroxide for 1 h. Following dehydrated in different concentrations of acetone, these cells were infiltrated and embedded into Epon. The embedded cells were sectioned and stained with 3% uranyl acetate and lead citrate. Mitochondrial content was analyzed by Mito-Tracker staining (number: C1049, Beyotime Biotechnology, Shanghai, China) according to the manufacturer's protocol and then analysed by FACS (BD Biosciences, USA), and results were calculated using Flow Jowo software (BD Biosciences, USA).

Lactate measurement
The suitably treated cells were seeded in 6-well plates in triplicate at the density of 5*10 4 cells/well. After culturing for 2 days, the number of viable cells was counted, and the culture medium was collected. Lactate levels were detected in the latter using a specific analytical kit (Nanjing Jian cheng Bioengineering Institute, China), and normalized to cell number and calculated as relative units per cell.

RNA sequencing and bioinformatic data analysis
LM3-shControl and LM3-shATAD3A cells were generated as before and total RNA was extracted. RNA libraries were constructed using a TruSeq Stranded mRNA LT Sample Prep Kit (Illumina, San Diego, CA, USA) according to the manufacturer's instructions, and sequenced on an Illumina platform (HiSeqTM 2500 or Illumina HiSeq X Ten) into 125 bp/150 bp paired-end reads by OE Biotech (Shanghai, China). The differentially expressed genes (DEGs) were identified using Cuffdiff with p value < 0.05 and fold change > 2 as the criteria. Gene set enrichment analysis was performed using software.broadinstitute.org/gsea. Heatmaps were generated with the heatmap package of R program.

Tumorigenicity analysis in vivo
Female BALB/c (nu/nu) athymic nude mice, 5 weeks of age, were purchased from hfkbio Inc. Mice were maintained in specific pathogen-free conditions: 20-24°C, 12/12 h of dark/light cycle, 60 ± 5% of humidity, and plastic cage (3-4 mice/cage). Bedding materials were changed every week, and environmental enrichment was done with sterile materials. All animal experiments were approved by the Institutional Animal Care and Use Committee of the Medicine University of Electronic Science and Technology of China's. Control and miR-210-5P antagomir-transfected Huh7-shControl and Huh7-shATAD3A cells were inoculated subcutaneously (n = 5 each, 2 × 10 6 cells per mouse) in the right flank of 4-6 week old BALB/c athymic female nude mice. Once the tumors grew to approximately 100mm 3 (around 5 days), the mice were intraperitoneally injected with sorafenib tosylate (10 mg/kg) daily for a week, and monitored every 4 days for the appearance of subcutaneous tumors. We conducted the sacrifice of mice at 28 days, placed the mice in the chamber and introduced 100% carbon dioxide. After we removed each tumor, we maintained the carbon dioxide flow for a minimum of 1 min after respiration ceases. The tumor weight and tumor volume (TV; mm3) were calculated. Tumor volume was calculated as d2 × D/2 (d and D represent the shortest and the longest diameters, respectively). All animal experiments were performed in accordance with the guidelines of the ARRIVE reporting guidelines.

Statistical analysis
Statistical analyses were performed using GraphPad Prism 6.0 (GraphPad, La Jolla, CA, USA). Two tailed unpaired t-test or one-way ANOVA were used to compare two or multiple groups. Survival curves were plotted using the Kaplan-Meier method and compared by the log-rank test. All data was presented as mean ± SD or mean ± SEM. P < 0.05 was considered statistically significant.

Hyperactivated mitophagy plays a vital role in hypoxiainduced sorafenib resistance
As shown in Supplementary Fig. S1a-1d, HCC cells cultured under hypoxia showed significantly lower apoptosis rates and better cell viability response to sorafenib treatment in a time-dependent manner. To further analyze the role of hypoxia in sorafenib resistance, we established sorafenib resistant Huh7 (Huh7-SR) and LM3 (LM3-SR) cell lines, as well as the corresponding hypoxia-induced sorafenib resistant cell lines (Huh7-H-SR and LM3-H-SR) in the presence of increasing doses of sorafenib under 1% O 2 culture condition (Fig. 1a-c). Compared to the respective control, Huh7-H-SR and LM3-H-SR cells showed increased levels of HIF-1α and drug resistance-related genes, including ABC20, ABCB1, ABCC1 and LRP ( Supplementary Fig. S1e). Interestingly, extensive mitophagy was observed in Huh7-H-SR cells compared to the untreated controls, hypoxia treated Huh7 and Huh7-SR cells (Fig. 1d). Consistent with this, the Huh7-H-SR cells showed highest level of PINK1 expression, along with accumulation of PINK1, Parkin and LC3-II in the mitochondrial fraction (Fig. 1e). We also characterized a higher ubiquitination of multiple mitochondrial proteins via Parkin in Huh7-H-SR cells (Fig. 1e). Hypoxia-induced sorafenib resistant Huh7 and LM3 cells lines significantly upregulated PINK1, LC3-II, HIF-1α, FUNDC1 expression, and downregulated TOMM20 and ATAD3A expression (Fig. 1f). Furthermore, TCGA analysis results showed that PINK1 levels were positively correlated with that of ABCB1 and ABCG2 in HCC patients (Fig. 1g-h), and HIF-1α was positively correlated with ABCC1 and negatively with TOMM20 and TOMM70 ( Supplementary Fig. S1f-1 h). These results indicated that activated mitophagy is the likely cause of hypoxia-induced sorafenib resistance. In agreement with this, the mitophagy inhibitor 3MA sensitized Huh7-H-SR cells to sorafenib, resulting in a significant decrease in colony yield and proliferation rates (Fig. 1i-j). Furthermore, PINK1 knockdown in Huh7-H-SR cells also sensitized the cells to sorafenib (Fig. 1k-l). Taken together, hyperactivated mitophagy plays a pro-survival role in the hypoxia-induced sorafenib resistant HCC cells.

Hypoxia-induced sorafenib resistance is dependent on ATAD3A-mediated mitophagy
The mitochondria transmembrane protein ATAD3A was significantly downregulated in the Huh7-H-SR and LM3-H-SR cells compared to the respective controls ( Supplementary Fig. S2a). To determine the biological relevance of ATAD3A in HCC, we established stable ATAD3A-overexpressing and knockdown Huh7 and LM3 cell lines ( Fig. 2c and Supplementary Fig.S2b-2c). Mitophagy was hyperactivated in the shATAD3A LM3 cells compared to the control (Fig. 2a), which paralleled a marked increase in the polyubiquitination and accumulation of global mitochondrial proteins as seen in the hypoxia-induced sorafenib resistant cells (Fig. 2b). In addition, PINK1 and LC3-II were significantly upregulated, and TOMM20, TOMM70 were downregulated following ATAD3A knockdown (Fig. 2c). Transcriptome sequencing analysis of the control and shATAD3A LM3 cells identified 319 upregulated and 302 downregulated genes in the latter (Supplementary Fig. S2i-2j). Gene set enrichment analysis (GSEA) (Fig. 2f) showed that the shATAD3A cells were commonly enriched with drug resistance genes (NES = -1.939, P = 0.00828 for GEFITINIB  resistance; NES = 1.8435; P = 0.02 for TAMOXIFEN resistance) (Fig. 2d, e). The expression levels of these genes were validated by qRT-PCR (Fig. 2g, h). We also confirmed that the expression level of these genes in LM3-H-SR were similar to that in sh-ATAD3A cells compared to shControl (Supplementary Fig. S2f). In addition, shATAD3A and hypoxia-induced sorafenib resistantLM3 cells desensitized the HCC cells to sorafenib, resulting in a significant decrease in apoptotic index ( Fig. 2i and Supplementary Fig. S2g-2 h), whereas ATAD3A overexpression augmented the sorafenibinduced cell death (Fig. 2j, k and Supplementary Fig.  S2d, 2e). Loss of ATAD3A downregulated the proapoptotic PARP, Bax and Cleaved-Caspase 3 expression (Fig. 2l), decreased mitochondrial mass (Fig. 2m), and increased lactate production ( Supplementary Fig. S3a-3b) in Huh7 and LM3 cells. Taken together, ATAD3A plays a vital role in mitophagy and sorafenib insensitivity in HCC cells.

ATAD3A participates in hypoxia-induced mitophagy and sorafenib sensitivity
To further determine the biological significance of hypoxia and ATAD3A expression in HCC cells, we introduced ectopic ATAD3A in the hypoxia cultured cells. As shown in Fig. 3a, hypoxia-induced mitophagy was abrogated by ATAD3A overexpression, which also upregulated TOMM20 and TOMM70 in Huh7 and LM3 cells (Fig. 3b). Ectopic ATAD3A also restored the mitochondrial mass (Fig. 3c, d) and decreased lactate concentration ( Supplementary Fig. S3c, 3d) upon hypoxia-treated LM3 and Huh7 cells. Under hypoxia condition, cancer cells were insensitive to sorafenib treatment, while, gain of ATAD3A expression mitigated the insensitivity as indicated by the fewer colonies (Fig. 3e, f) and increased apoptotic index in LM3 cells (Fig. 3g, h). Western blot confirmed that PARP, Bax and Cleaved-Caspase3 were downregulated in hypoxia cells which were restored in ATAD3A overexpression cell under sorafenib treatment. Furthermore, we confirmed that ATAD3A expression restored sorafenib sensitivity in hypoxia treated Huh7 and LM3 cells via the apoptotic related proteins (Fig. 3i).

ATAD3A is a direct functional target of hypoxia-induced mir-210-5p
To determine the mechanism underlying hypoxiamediated regulation of ATAD3A expression, we performed a systematic bioinformatics analysis of gene-gene interaction networks based on mutations, copy number alterations, mRNA expression profiles, and protein expression profiles using the STRING database (http:// string-db.org/cgi/network). ATAD3A and the hypoxiaresponsive transcriptional factor HIF-1α are located in a network containing 40 nodes (Fig. 4a), suggesting a potential link between HIF-1α signaling and ATAD3A. Furthermore, in situ HIF-1α protein expression in 85 HCC tissues was negatively correlated with that of ATAD3A (Fig. 4b). As shown in Fig. 4c and Supplementary Fig. S4a, hypoxia treatment significantly down regulated ATAD3A expression in hepatoma cells. In addition, both 1% O 2 and CoCl 2 induced HIF-1α expression and downregulated ATAD3A by western blot in Huh7 and LM3 cells (Fig. 4d), and decreased ATAD3A expression in Huh7 cells as a time-dependent manner ( Supplementary Fig. S4b). In addition, we found a loss of ATAD3A gene expression in 1% O 2 and CoCl 2 treated Huh7 and LM3 cells (Fig. 4e-f and Supplementary Fig.  S4c). Bioinformatics analysis identified hsa-miR-210 and hsa-miR-193b as the most hypoxia-responsive miRNAs ( Supplementary Fig. S4d), and ATAD3A as the potential target gene of miR-210-5P (Fig. 4g). Consistent with this, 1% O 2 and CoCl 2 increased miR-210-5p expression after 24 h (Fig. 4h and Supplementary Fig. S4e). Furthermore, miR-210-5P suppressed the expression of reporter gene carrying the wild-type but not mutant 3′-UTR of human ATAD3A gene (Fig. 4j, k). Hepatoma cells transfected with miR-210-5P mimic significantly inhibited ATAD3A expression, whereas miR-210-5p inhibitor had the opposite effect (Fig. 4i, l and Supplementary Fig. S4f). These findings indicated that ATAD3A is a direct target of the hypoxia induced miR-210-5P. This was confirmed by treating hepatoma cells with CoCl 2 and/or the miR-210-5P inhibitor, which showed that hypoxia-induced downregulation of ATAD3A was abrogated by blocking miR-210-5P (Fig. 4m-o). Taken together, inhibition of ATAD3A under hypoxia is dependent on miR-210-5P.
Clinical significance of ATAD3A expression in HCC and the miR-210-5P-ATAD3A axis in exnografts formation in vivo To investigate the expression level of ATAD3A in HCC tissues and their paired normal tissues. We generated a tumor sample tissue array containing 135 hepatocellular carcinoma samples with survival information and 85 cases with paired normal tissues to determine ATAD3A expression by IHC. Results showed that ATAD3A expression in HCC patients was heterogeneous, ranging from completely absent to strongly expressed and the representative images of hepatocellular carcinoma samples with strong, moderate or low ATAD3A expression were shown in Fig. 6a. ATAD3A expression level was lower in cancerous liver tissues compared to the adjacent normal tissues (Fig. 6b-c), and lower expression of ATAD3A was associated with higher-grade HCC   Each bar represents the mean ± SEM of three independent experiments. *P < 0.05 and ** P < 0.01 patients (Fig. 6d). 14% of hepatocellular carcinoma samples had strong ATAD3A expression, whereas percentages of moderate and no ATAD3A expression were 21% and 65%, respectively (Fig. 6e). Kaplan-Meier analysis showed that lower ATAD3A expression in HCC was correlated closely with poor overall survival (OS) of the patients (P = 0.0184; Fig. 6f). Western blot analysis also showed that ATAD3A had a significantly lower expression in 6 HCC tissues compared to their paired normal tissues (Fig. 6g). Taken together, these data indicated that ATAD3A is down regulated in HCCs and indicative of poor clinical outcome of HCC patients. To further determine the tumorigenicity of sh-ATAD3A and whether miR-210-5P antagomir could attenuate the tumor progression role of sh-ATAD3A under sorafenib treatment in vivo. Huh7-shcontrol and Huh7-shATAD3A cells that have been transfected with or without antagomir of miR-210-5p construct were subcutaneously implanted into the right flank of nude mice (Fig. 6h).
Results showed that sh-ATAD3A in Huh7 cells significantly increased the tumor weight and tumor volume compared to the control group in the presence of sorafenib treatment, and miR-210-5P antagomir reduced the tumor size and tumor volume both in control and sh-ATAD3A groups (Fig. 6i-l). Collectively, these data provide evidence for an important inverse regulating relationship between ATAD3A and miR-210-5P in vitro and in vivo. Taken together, our findings point to an inverse regulatory relationship between ATAD3A and miR-210-5P in mediating mitophagy in the sorafenib-resistant HCC cells under hypoxia (Fig. 6m).

Discussion
Hypoxia is a hallmark of solid tumors and is the consequence of absent or abnormal vasculature in the tumor microenvironment. It is essential for tumor progression, angiogenesis, metastasis, invasion, immune escape and chemoresistance [33]. Various studies showed that sorafenib could stabilize HIF-1α expression under hypoxia and increase the expression of HIF-1α targets like MDR1, GLUT-1, VEGF and miRNAs (e.g. miR-210), and the whole progress played vital roles in cancer biology [34][35][36]. In addition, evidence indicated that hypoxia-induced mitophagy mainly depends on BNIP3/NIX and the FUNDC1 pathway [37,38]. In the present study, we found that hypoxia induced mitophagy also dependent on the loss of ATAD3A, which could increase PINK1 accumulation in the outer mitochondrial membrane. ATAD3A located on the mitochondrial membrane which regulated fragmentation, fission, protein transport and other aspects of mitochondrial physiology [39][40][41][42]. ATAD3A deletion triggered mitophagy through the PINK1/Parkin pathway in the hematopoietic system and induced cisplatin sensitivity by increasing mitochondrial fragmentation. In our present study, hypoxia-induced sorafenib resistance in HCC cells was associated with lower ATAD3A expression and upregulation of BNIP3/NIX, and loss of ATAD3A increased sorafenib insensitivity. Since PINK1 deletion nearly completely abrogated sorafenib resistance of Huh7 cells under hypoxia, we hypothesized that ATAD3A played a vital role in hypoxia-induced sorafenib resistance through the mitophagic pathway. Under both basal conditions and times of abnormal stress, the process of mitophagy was to prevent the accumulation of dysfunctional mitochondria and preserved mitochondrial homeostasis [43]. While, how mitophagy affected drug sensitivity was still confusing. The removal of damaged mitochondria through mitophagy could dampen the toxicity of chemotherapy [44]. Mitophagy is also described as a possible mechanism for preventing ROS production [45]. And cancer cells were likely to be more sensitive to the additional oxidative damage promoted by ROS [46][47][48]. Therefore, we presumed that hypoxia induced hyperactivated mitophagy may decrease ROS production and then result in sorafenib insensitivity.
HIF-1α is a key transcriptional factor that allows cells and tissues to adapt to hypoxia. Bioinformatics analysis predicted ATAD3A as a direct target of miR-210-5P, the predominant miRNA induced under hypoxia in multiple solid tumors and a surrogate marker for tumor hypoxia [49]. It promoted cancer cell proliferation, metabolism, angiogenesis, metastasis, invasion, chemo-resistance, and correlated with patient prognosis [50][51][52][53]. Mitochondrial homeostasis was crucial for cellular survival during hypoxia [52], and in line with this, miR-210-5P blockade reversed hypoxia-induced down-regulation of ATAD3A and sensitized the HCC cells to sorafenib by promoting mitophagy.

Conclusions
In summary, the lack of reliable and robust predictive biomarkers of sorafenib sensitivity and treatment response has hindered the development of personalized therapy in HCC. Loss of ATAD3A increased HCC tumor growth in the presence of sorafenib treatment. However, miR-210-5P antagomir abolished the tumorigenic effect of ATAD3A knockdown. Thus, ATAD3A is an effective target for overcoming resistance to sorafenib and also a predictive marker for sorafenib response and clinical outcomes. Our findings also point to the role of ATAD3A-PINK1-Parkin signaling pathway in hypoxiainduced sorafenib resistance in HCC. Further investigation is warranted to examine the feasibility of miR-210-5P inhibition or ectopic ATAD3A expression combined with sorafenib as a novel therapeutic strategy against HCC. In addition, additional response predictors need to be assessed in order to identify patient subgroups most likely to benefit from sorafenib, achieve optimal risk-tobenefit ratio and maximize the treatment efficacy for each patient.
Additional file 1. Target sequences and ORF expression clone for PINK1 and ATAD3A in the study.

Additional file 2. Antibodies used for immunostaining analysis in our dataset
Additional file 3. Sequences of the primers used for qRT-PCR.
Additional file 4: Supplementary Fig. 1. Hypoxia induced sorafenib resistance in hepatoma cells. a,b Sorafenib sensitivity under normoxic or hypoxia condition of LM3 and Huh7 cells measured by apoptosis rates (%). c, d Cell viability of LM3(c) or Huh7 (d) cells under hypoxic or normoxic conditions with sorafenib treatment. e qRT-PCR analysis of indicated genes in control, hypoxia treated, sorafenib resistant and hypoxiainduced sorafenib resistant Huh7 cells. f. Immunoblot showing expression levels of indicated proteins in control, hypoxia, SR and Huh7-H-SR cells. g-i Relation between HIF-1α expression and that of ABCC1 (g), TOMM20 (h) and TOMM70 (i) in HCC patients from TCGA database. Each bar represents the mean ± SEM of three independent experiments. *P < 0.05 and ** P < 0.01. Additional file 6: Supplementary Fig. 3. ATAD3A plays a vital role in lactate production of hepatoma cells. a,b Lactate levels in the supernatants of Huh7-shControl and Huh7-shATAD3A (a) or LM3-shControl and LM3-shATAD3A (b) cells. c,d Lactate levels in Huh7-Vector and Huh7-ATAD3A (c) or LM3-Vector and LM3-ATAD3A (d) cells with/without 1% O 2 exposure for 24 h. e,f Lactate levels in Huh7 (e) and LM3 (f) cells with/ without miR-210-5P inhibitor transfection and with/ without CoCl 2 treatment. g, h Lactate levels in control and miR-210-5P mimictransfected Huh7 (g) and LM3 (h) cells. Each bar represents the mean ± SEM of three independent experiments. *P < 0.05 and ** P < 0.01. Additional file 7: Supplementary Fig. 4. ATAD3A is down-regulated in hypoxia and a potential target of miR-210-5P. a Immunofluorescent staining of HIF-1α (green) and ATAD3A (red) expression in normoxic and hypoxic LM3 cells (1% O 2 treatment for 24 h). Scale bar = 200 μm. b Immunoblot showing expression levels of indicated proteins in Huh7 cells with/without CoCl 2 treatment for different durations. c qRT-PCR analysis of ATAD3A expression in LM3 cells with/without CoCl 2 treatment. d Bioinformatics analysis of the hypoxia-responsive miRNAs from digestive cancers. e qRT-PCR analysis showing miR-210-5P levels in Huh7 and LM3 cells treated with DMSO or CoCl 2 for 24 h. f qRT-PCR analysis showing the expression of ATAD3A in LM3 cells treated with miR-210-5P mimic and inhibitor. Each bar represents the mean ± SEM of three independent experiments. *P < 0.05 and ** P < 0.01.