The up‐regulation of NDRG1 by HIF counteracts the cancer‐promoting effect of HIF in VHL‐deficient clear cell renal cell carcinoma

Abstract Background Hypoxia‐inducible factors (HIFs) are thought to play important roles in the carcinogenesis and progression of VHL‐deficient clear cell renal cell carcinoma (ccRCC). Methods The roles of HIF‐1/2α in VHL‐deficient clear cell renal cell carcinoma were evaluated by bioinformatics analysis, immunohistochemistry staining and Kaplan‐Meier survival analysis. The downstream genes that counteract the cancer‐promoting effect of HIF were analysed by unbiased proteomics and verified by in vitro and in vivo assays. Results There was no correlation between the high protein level of HIF‐1/2α and the poor prognosis of ccRCC patients in our large set of clinical data. Furthermore, NDRG1 was found to be up‐regulated by both HIF‐1α and −2α at the cellular level and in ccRCC tissues. Intriguingly, the high NDRG1 expression was correlated with lower Furman grade, TNM stage and longer survival for ccRCC patients compared with the low NDRG1 expression. In addition, NDRG1 suppressed the expression of series oncogenes as well as the proliferation, metastasis and invasion of VHL‐deficient ccRCC cells in vitro and vivo. Conclusions Our study demonstrated that HIF downstream gene of NDRG1 may counteract the cancer‐promoting effect of HIF. These results provided evidence that NDRG1 may be a potential prognostic biomarker as well as a therapeutic target in ccRCC.

PHD) through oxygen. 1 Once hydroxylated, HIF-α is subjected to conjugate with the von Hippel-Lindau tumour suppressor protein (pVHL), which is the substrate-recognition component of E3 ubiquitin ligase complex. pVHL recruits an E3 ubiquitin ligase that catalyses polyubiquitination of HIF-α, thereby targets it for proteasomal degradation. [2][3][4] There are hypoxia and/or anoxia regions in 50% to 60% of solid tumours. 5 Under hypoxic conditions, PHD activity is inhibited and the pVHL fails to recognize HIF-α and make it ubiquitylation, resulting in the accumulation of HIF-1/2α subunits in the cytoplasm. 3 Then, HIF-1/2α subunits translocate to the nucleus, form dimerization with HIF-1β and subsequently activate directly or indirectly numerous genes that are involved in cell proliferation, migration and invasion, angiogenesis, metabolic shift towards glycolysis and survival. For example, the production of vascular endothelial growth factor (VEGF), PDGF-β and transforming growth factor-alpha (TGF-α) and several glycolytic enzymes is up-regulated by HIF, which endow cancer cells surviving advantages under hypoxia conditions. 6 HIF-1/2α have been associated with poor prognosis in a broad range of human cancers including astrocytoma, breast, melanoma, ovarian and prostate cancers. 7 The critical role of HIF in cancer has led to its recent recognition as an ideal target for small molecule interventions. Many of the novel anti-cancer drugs have been developed to inhibit HIF activity directly or indirectly. HDAC inhibitors and compounds have been shown to increase HIF-α degradation. And DNA intercalating drugs, including echinomycin and anthracyclines, such as doxorubicin and daunorubicin inhibit HIF transcriptional activity by blocking its binding to DNA. Receptor tyrosine kinases inhibitors (such as gefitinib and erlotinib) and mTOR inhibitors (such as rapamycin) are thought to reduce tumour angiogenesis by indirectly reducing the synthesis of HIF-α subunits. [8][9][10][11] Kidney cancer is one of the most common cancers. The incidence and mortality of kidney cancer have been increasing in many countries. Renal cell carcinoma (RCC) is the most prevalent subtype of kidney cancer (90%). About 75%-80% of adult RCC are clear cell renal cell carcinomas (ccRCCs). Importantly, more than 90% of ccRCC tumours harbour biallelic inactivation of VHL via point mutation, deletion or methylation, which occur at the earliest stage of tumour formation. 12 In ccRCC, the alterations of VHL mostly disrupt the function of pVHL so that HIF-1/2ɑ cannot be degraded and are accumulated in the cancer cell even under normoxic conditions. Inactivation of VHL is thought to be an early event in the pathogenesis of ccRCC. 13 But many molecular studies have supported the crucial role of HIF-1/2α in the carcinogenesis and progression of ccRCC. [14][15][16] Surprisingly, in this study we found that there was no correlation between the high protein level of HIF-1/2α and the poor prognosis of ccRCC patients in our large set of clinical data. Furthermore, N-myc downstream-regulated gene 1 (NDRG1) was found to be up-regulated by HIF-1/2α in ccRCC tissues and its high expression was correlated with longer survival for ccRCC patients compared with the low NDRG1 expression. Our findings indicated that the up-regulation of NDRG1 by HIF-1/2α may counteract the cancer-promoting effect of HIF-1/2α in VHL-deficient ccRCC.

| Clinical samples
A total of 645 paraffin-embedded tumour samples and 260 paired adjacent normal tissues were obtained from ccRCC patients who underwent partial or radical nephrectomy in Department of Urology, Shanghai Renji Hospital, Shanghai Jiaotong University (Shanghai, China).

| Western Blot (WB) analysis and antibodies
Proteins were extracted from cells by using RIPA buffer with protease inhibitors. A 20 µg of proteins was loaded and separated by electrophoresis. Then, proteins were transferred to nitrocellulose membrane.

| Quantitative real-time PCR (qRT-PCR)
Total RNA was extracted from cells using TRIzol® reagent (Life Technologies) following the manufacturer's instruction. Then, RNA was converted to complementary DNA (cDNA). Quantitative real-time polymerase chain reaction (qRT-PCR) was performed on cDNA using genespecific primers in the presence of SYBR Green (Applied BioSystems).
Data S9 lists the gene-specific primers used in this study. Each measurement was performed in triplicate with at least three independent experiments. Transcript levels were normalized with GAPDH, and relative mRNA levels in experimental samples were normalized to controls.

| Lentiviral shRNA, siRNAs and transfection
Pairs of complementary oligonucleotides against VHL, HIF-2α and NDRG1 were synthesized, annealed and ligated into pSIREN-RetroQ (Clontech, Mountain View, CA, USA). The target sequence for VHL, HIF-2α and NDRG1 was described on our previous research. 17,18 Expression lentivirus for NDRG1 and corresponding control empty vector was transfected in Caki-1 cell lines according to the manufacturer's instructions. After transfection, the virus supernatant was collected, filter-sterilized and added to cells in six-well plate containing polybrane with a final concentration of 4 mg/mL. The stably transfected cells were selected by adding puromycin (2 mg/mL).
The siRNA was either designed against the HIF-1α gene (siHIF-1α or siHIF-1α NC). RCC4 cells were treated with siHIF-1α or siRNA-NC complexed with in vivo liposome at a final concentration of 100 nmol/L. After 24 hours, media was replaced with DMEM media with 10% serum.

| Cell proliferation assays
Cells were plated in triplicate in 96-well plates in 100 μL appropriate growth medium. At indicated time points, each well was pulsed by addition of 10 μL of CCK-8 assay (Dojindo, Kumamoto, Japan), followed by incubation at 37°C for 3 hours. Absorbance readings at a wavelength of 450 nm were taken.

| Colony formation assay
Cells (100 cells per well) were seeded in a six-well plate and cultured for 10 days. Colonies were then fixed with 4% paraformaldehyde for 15 minutes and stained for 10 minutes with 0.5% crystal violet.
Wells were washed with water to remove excess dye; dried plates were taken images, and the number of colonies was counted.   Tumours were imaged to observe luciferase expression on day 28 after tumour cell injection. Briefly, the mice were anesthetized and then injected (i.p.) with luciferin in a volume of 100 μL at a dose of 150 mg/kg. Images were captured using an IVIS-Lumina II imaging system (Caliper, USA) at a peak time of 10-15 minutes after injection.

| Data analysis
Differentially expressed proteins and mRNA were determined based on fold change log2 (log2FC)> |0.5| and unpaired t test P-value < .05 by using Scaffold 4 software (version4.7.2, Proteome Software Inc, Portland, OR, USA). 18 Differentially methylated regions (DMRs) identification, the mapped reads were used for detection of DMRs with statistically significant. And differentially methylated genes were determined based on log2FC > 1 and P-value < .05. The networks functional analyses and functional annotation and clustering were performed by using QIAGEN's Ingenuity Pathway Analysis (IPA, 2019-Summer, QIAGEN).

| Statistical analysis
The pre-treated level RNA-seq data and corresponding clinical information of ccRCC patients were collected from The Cancer Genome Atlas (TCGA) database (http://cance rgeno me.nih.gov/). All statistical ANALYSIS was performed by using IBM SPSS Statistics (version 24.0) and GraphPad Prism (version 6.0).

| HIF-1/2ɑ cannot indicate the ccRCC patient's prognosis
Given the critical role of HIF pathway in the carcinogenesis, we sought to evaluate the effect of the HIFs on the ccRCC patients systematically. To this aim, VHL was reintroduced into the VHL- and S5). Next, we used QIAGEN's Ingenuity Pathway bioinformatic analysis to analyse the interaction network of HIF-1/2α with these potential HIF-regulated genes. As shown in Figure 1B, we found that both HIF-1α and HIF-2α (EPAS1) have the ability to activate oncogenes deeply involved in cancer biology, such as VEGF, CXCL8 and PGR. And the bar chart of the functional annotation and clustering revealed that these potential HIF-regulated genes were involved in series of cancers, such as gastrointestinal cancer, endocrine gland cancer, head and neck cancer ( Figure 1C). The analysis of expression of putative HIF-1/2α-regulated gene in the 786-O cells suggests that HIF-1/2α may be associated with poorer overall survival in ccRCC.
Then, 331 specimens collected from ccRCC patients were stained with HIF-1α/2α antibody, respectively, and scored according to the IHC as described in experimental procedures. Next, we used Kaplan-Meier analysis to evaluate the relationship between HIF-1/2α protein level and patient survival time. We found that high-HIF-1α protein level could not indicate good or poor survival compared with low-HIF-1ɑ protein level (progression-free survival (PFS), P = .972; overall survival (OS), P = .651) ( Figure 1D and F).
Similarly, there was no statistical significance between HIF-2α protein level and patient survival (PFS, P = .613; OS, P = .972) ( Figure 1E and G). Further analysis revealed that there were no major statistical differences in baseline clinical pathologic parameters between the low and the high protein levels of HIF-1/2α, except that HIF-2α protein level is associated with Fuhrman grade (P = .012) (Tables 1   and 2). In conclusion, neither high protein level of HIF-1α nor HIF-2α can indicate the ccRCC patient's poor prognosis. However, from the bioinformatics results of the interaction network and bar chart in Figure 1B and C, the role of HIF in the development of ccRCC should be very important. So, we hypothesized that HIF activates downstream tumour suppressor genes in addition to the activating of oncogenic genes.

| The gene-regulatory network revealed that NDRG1 played a role in ccRCC
We then identified the potential tumour suppressor genes regulated by HIF-1/2α in VHL-deficient ccRCC cancer cells by subtractive proteomics strategy. In order to narrow down the range of genes regu- To determine whether NDRG1 suppresses tumour progression in ccRCC, we analysed the interaction of NDRG1 with the genes downstream of VHL. Three different levels of VHL-regulated gene, including protein, mRNA and methylation of genes, were combined (Data S1). The interaction network between these VHL-regulated genes and NDRG1 was shown in Figure 2C. NDRG1 was demonstrated to interact with series oncogenes, including JUN, STAT3, CXCL5, XRCC6, CD38, PGR and IL4. In addition, NDRG1 was shown to be down-regulated by VHL and involved in carcinoma, apoptosis, cell death and migration of tumour cell lines. This suggests that NDRG1 played an important role in ccRCC.

| NDRG1 was regulated by HIF-1/2α
To validate the proteomic results that HIF-1α/2α regulated NDRG1, reintroduction of VHL into VHL-deficient 786-O and RCC4 cells decreased the protein level of HIF-1/2α. And we also found the protein levels of NDRG1 were also decreased ( Figure 3A and B). Vice versa, both the protein of HIF-1α and NDRG1 were significantly up-regulated by silencing VHL expression in Caki-1 cells ( Figure 3C). Next, we disturbed the HIF-1α expression in RCC4 cells and knocked down HIF-2α in 786-O cells. As shown in Figure 3D, the deletion of HIF-1α in RCC4 decreased the

| High NDRG1 predicted a better prognosis in ccRCC
To determine the clinical implication of NDRG1 in ccRCC, we analysed the expression of NDRG1 in 260 ccRCC tissue samples and matched adjacent renal tissues by IHC according to Wilcoxon matched-pairs signed rank test. We found that NDRG1 was predominantly localized in the cytoplasm ( Figure 4A), and NDRG1 expression was increased in cancer issues compared with the paired adjacent tissue samples (P < .0001) ( Figure 4B). Then, we analysed an additional cohort of ccRCC specimens. Higher expression of NDRG1 protein in cancer issues was further confirmed through Mann-Whitney test (normal: n = 260, tumour: n = 645) (P < .0001) ( Figure 4C). Consistently, mRNA levels of NDRG1 in 72 ccRCC patients were significantly higher compared to their paired normal counterparts in TCGA data set (P < .0001) ( Figure 4D). And the mRNA level of NDRG1 from tumour tissues was higher than that from normal renal tissues (normal: n = 72, tumour: n = 530) (P < .0001) ( Figure 4E). These data indicated that the expression of NDRG1 was increased in ccRCC.
Next, we examined whether NDRG1 expression was associated with clinicopathologic parameters. The protein expression of NDRG1 was significantly decreased in the higher ccRCC Fuhrman grade (P = .006) and TNM stage (P = .026) ( Figure 4F and G, Table 3). In detail, the NDRG1 protein expression was negatively correlated with cancer pT stage (P = .001), pM stage (P = .026) and tumour size (P = .026) ( Table 3). We also observed that the mRNA level of NDRG1 expression was significantly lower in higher Fuhrman grade (P = .038, Figure 4H and Table 4) except TNM stage ( Figure 4I, Table 4). Furthermore, we evaluated the relationship between NDRG1 expression and patient survival by Kaplan-Meier analysis. We found that the overall survival time of the patients with high NDRG1 expression (n = 275) was longer than that of the patients with low NDRG1 expression (n = 370) (P = .001) ( Figure 4J). Consistently, compared with patients with low mRNA level of NDRG1 (n = 260), patients with high mRNA level of NDRG1 (n = 270) also had longer overall survival time (P = .007) according to the analysis of TCGA data ( Figure 4K).
We next analysed the correlation between NDRG1 expression and metastasis based on the follow-up information and TMA data set by linear-by-linear association. The incidence of metastasis in initially diagnosed patients was significantly lower in high NDRG1 expression patient than that in low NDRG1 expression ones (P = .021) (n = 645; Figure 4L). When excluding initially diagnosed ccRCC patients with metastasis, the patients with low NDRG1 expression were apt to have metastasis after radical nephrectomy (P = .015) (n = 589; Figure 4M).

| NDRG1 suppressed the proliferation and metastasis of ccRCC tumour cells both in vitro and vivo
To investigate the biological role of NDRG1 in tumour formation, we introduced two independent NDRG1 shRNAs into 786-O F I G U R E 3 NDRG1 was regulated by HIF-1/2α. (A-E) The protein expression of the shown genes was detected by Western blot, respectively; β-actin was used as the loading control.

(A) RCC4 and 786-O cells (B) were infected with VHL vector (VHL) or empty vector (EV). (C) Caki-1 cells were transfected with VHL shRNAs (shVHL) or non-specific control (NC). (D) RCC4 cells were infected with HIF-1α siRNAs (siHIF-1α) or non-specific control (NC). (E) 786-O cells were transfected with HIF-2α shRNAs (shHIF-2α) or non-specific control (NC). (F) Representative images of HIF-1α
, HIF-2α and NDRG1 protein expression from the IHC staining of human clear cell renal cell cancer specimens. Patient I: The protein levels of HIF-1α, HIF-2α and NDRG1 were low. Patient II: The protein levels of HIF-1α, HIF-2α and NDRG1 were medium. Patient III: The protein levels of HIF-1α, HIF-2α and NDRG1 were high (scale bar, 50 μm). (G and H) The protein level of NDRG1 is significantly correlated with HIF-1α (G) and HIF-2α (H). The protein levels were evaluated according to IHC scores, representing very low (score 0-2), low (score 3-4), high (score 5-8) and strong (score 9-12). The subjects were divided into four groups according to the IHC scores of HIFs in the tumours. (I) The protein level of NDRG1 is significantly correlated with the average of HIF-1α and HIF-2α in ccRCC tissues. The protein levels are evaluated according to IHC scores, representing very low (score < 3), low (score < 5), high (score < 8) and strong (score [8][9][10][11][12]. The subjects were divided into four groups according to the IHC scores of the average of HIF-1α and HIF-2α. . Compared with control group, luciferase signals in the mice from 786-OshNDRG1-2/3 groups were significantly higher at 28 days after injection ( Figure 5J), which was consistent with more foci in the lungs of the mice in 786-OshNDRG1-2/3 groups ( Figure 5K). These results suggested that NDRG1 suppressed metastasis of ccRCC cells.
It may be that the growth of 786-O/shNDRG1-2/3 cells was faster, which resulted in stronger metastasis in the lungs of mice.

| High HIF-1/2α predicted a poor prognosis in ccRCC after adjustment of the expression of NDRG1
As shown in Figure 1, there was no significant difference in PFS and OS between the high-and low-expression groups of HIF-1/2α.  However, the expression level of NDRG1 in ccRCC patients with high levels of HIF-1/2α protein was significantly higher than that in ccRCC patients with low levels of HIF-1/2α protein ( Figure S2a and b). In addition, the higher NDRG1 was found to suppress the development of ccRCC. Next, we adjusted ccRCC patients in high-HIF-1/2α level group and low-HIF-1/2α level group with the NDRG1 expression level so that there was no significant difference in the expression of NDRG1 between the two ccRCC groups ( Figure S2c and d). And then, we found that the PFS and OS time of the patients with high-HIF-1α protein level (n = 91) were significantly shorter than that of the patients with low-HIF-1α protein level (n = 91) (PFS, P = .026; OS, P = .031) ( Figure 6A  This suggests that NDRG1 inhibits tumour proliferation and metastasis by suppressing the expression of genes involved in carcinogenesis.

| D ISCUSS I ON
As documented in the past decades, HIF-1/2α is thought to be deeply involved in the carcinogenesis and progression of tumours and the expression of HIF-1/2α has been supposed to be tight with the prognosis in various cancers. 6,7 In our study, the bioinformatic analysis of putative genes regulated by HIF-1/2α in 786-O cells demonstrated that the HIF-1/2α should be associated with poor prognosis of ccRCC.
However, the implication of HIF-1/2α in the prognosis of ccRCC is controversial and largely unclear. Piotr M et al have shown that HIF-1/2α expression is associated with an inferior survival. 19,20 In contrast, it have been reported that HIF-1α expression in ccRCC was associated with the good prognosis, and HIF-1α acts as a tumour suppressor in ccRCC. 21 We collected large set of tumour tissue specimens and follow-up information of 331 ccRCC patients. Surprisingly, we found no correlation between HIF-1/2α protein expression and outcome of patients. The inconsistence between these other studies and ours may be caused by the relative smaller sample size in these other studies. Our work may provide relatively credible evidence for the role of HIF-1/2α in ccRCC prognosis because of the large set of tumour tissue specimens.
Our findings did not demonstrate the cancer-promoting effect of HIF. It suggested that some genes downstream of HIF-1/2α may act as a tumour suppressor and counteract the cancer-promoting effect of HIF-1/2α. We identified the common genes regulated both by VHL and hypoxia in RCC4 and 786-O cell lines. The common genes that were regulated by both VHL and hypoxia may be potential HIFregulated genes. Finally, we identified NDRG1 as a candidate downstream of HIF to suppress tumour progression.
NDRG1 is down stream of n-myc/c-myc, and the later could suppress its expression. NDRG1 can be up-regulated at protein and mRNA levels by hypoxia, cellular iron depletion and DNA damage through HIF-1α-dependent and -independent mechanisms. 22,23 However, its role is controversial in various tumours. In colon, prostate, breast and pancreatic cancer, NDRG1 suppresses tumour proliferation and metastasis. [24][25][26][27][28] In contrast, NDRG1 stimulates carcinogenesis in tumours of the liver, bladder, oesophagus and cervix. [29][30][31][32] However, its role in ccRCC is largely unknown. We constructed the interaction comprehensive networks be- In addition, it has been reported that the downstream genes of HIF play an anti-tumour role in ccRCC. 38 Cav-1 and VCAM-1 was suppressed by NDRG1. CCND1 is a key regulator of G1/S transition and cell proliferation. SIRT1 regulates ageing and resistance to oxidative and DNA damage stress by inhibiting cellular apoptosis or senescence. Consistent with their role, in cancer cells, they act as oncogenes by promoting tumorigenesis. 40,41 It is best known that about 90% of cancer-related deaths are caused by metastatic disease rather than primary tumours. 42 The characteristic that malignant tumour cells have the ability of metastasis is mainly achieved through epithelial-mesenchymal transition (EMT). EMT is a series of cell-biological programmes coordinated by master EMT-inducing transcription factors (EMT-TFs), especially Snail, Slug and Zeb2. 43 In addition, VEGFα, MMP2, ADAM12 and Vim, which are produced by cancer cells, have been reported to disrupt vascular integrity. They have previously been beneficial during primary tumour invasion and have proven useful at the invasion-metastasis cascade. 44 VCAM-1expressing carcinoma cells are able to obtain the ability to metastatic colonization by activating AKT signalling. 45 The modulated profiling of these proliferation and metastasis-related genes in NDRG1-silencing cells indicated the potential molecular events downstream of NDRG1.
It has been reported that NDRG1 and the mitogen-inducible gene 6 (MIG6) form a complex through direct interaction in the cytoplasm, which facilitates lysosomal process of epidermal growth factor receptor (EGFR) and down-regulates the EGFR expression. 46 In this work, we used proteomics to screen for genes regulated by hypoxia and VHL, which did not fully represent the downstream genes of HIF-1/2α. So, there may be other tumour suppressor genes downstream of HIF-1/2α that we missed. In addition, both HIF-1α and HIF-2α can up-regulate NDRG1 expression; thus, further research is necessary to determine the different ability of HIF-1α and HIF-2α in ccRCC on regulating NDRG1.
In conclusion, we demonstrated HIF downstream gene of NDRG1 counteracts the cancer-promoting effect of HIF. And we revealed the expression pattern, biological function and potential regulatory mechanism of NDRG1 in ccRCC. These results provide evidence that NDRG1 may be a potential prognostic biomarker as well as a therapeutic target in ccRCC.

CO N FLI C T O F I NTE R E S T
The authors declare that they have no competing interests. University.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data generated or analysed during this study are included in this article or if absent are available from the corresponding author upon reasonable request.