HIF-1α promoted vasculogenic mimicry formation in hepatocellular carcinoma through LOXL2 up-regulation in hypoxic tumor microenvironment

A total of 201 primary tumor specimens were obtained from the Tumor Tissue Bank of the Tianjin Cancer Hospital (Tianjin, China). The HCC specimens were collected from patients who underwent hepatectomy between 2001 and 2009. A diagnosis of HCC in these samples was verified by pathologists. Detailed pathological and clinical data were collected for all the samples, including the Edmondson tumor grade, metastasis and survival duration. The use of these tissue samples was approved by the Institutional Research Committee.

Immunohistochemical staining was done to validate the expression of HIF-1α and LOXL2 in HCC tumor tissues. The tissue sections were deparaffinized, hydrated and rehydrated based on standard protocols. Antigen retrieval was performed, and non-specific binding sites were blocked. The sections were then incubated with rabbit monoclonal anti-HIF-1α (ZA-0562, USA) and rabbit polyclonal anti-LOXL2 (1:800, Cat. #GTX105085; GeneTex, California, USA) primary antibodies overnight at 4 °C, and the secondary antibody was incubated with the samples for 30 min at 37 °C. The color was developed using a 3,3’-diaminobenzidine chromogen (DAB) solution.

Immunohistochemical staining with CD31 was performed on the sections as described above prior to PAS staining. Then, the slides were treated with periodic acid solution for 10 min and rinsed with distilled water for 5 min. In a dark chamber, the slides were submerged in Schiff solution for 15 min at 37 °C. After washing the slides under running water for 20 min, all of the sections were counterstained with hematoxylin, dehydrated, and mounted.

The protein expression levels were quantified according the intensity and percentage of positive tumor cells. At least 10 randomly selected microscope fields per slide were counted with approximately 100 tumor cells per field. The extent of positivity (“extent of distribution” of positive cells) was graded on the following scale: 0 for <10% positive cells, 1 for <25% positive cells, 2 for <50% positive cells, and 3 for more than 50% positive cells. The intensity of the staining was scored on a scale of 0–3 as follows: 0, no appreciable staining in the tumor cells; 1, barely detectable staining in the cytoplasm and/or nucleus compared to the stromal elements; 2, readily visible brown staining; and 3, dark brown staining in tumor cells obscuring the cytoplasm and/or nucleus. The product (staining index) of intensity and percentage scores were utilized to determine the result. For statistical analysis, a total score <4 was defined as negative/low expression, while scores ≥4 were defined as positive/high expression.

The Bel7402 and HepG2 cell lines were obtained from the American Type Culture Collection. Both these cell lines were cultured in RPMI 1640 and MEM supplemented with 10% fetal bovine serum (FBS; Invitrogen). We used cobalt chloride (CoCl₂) to mimic hypoxic conditions. Cells were seeded in dishes or plates and grown for 24 h in complete medium. The medium was removed, and cells were washed with PBS. Afterwards, the cells were treated with 150 μM CoCl₂ and incubated for 48 h.

HIF-1α and LOXL2 suppression was mediated by lentiviral infection using OmicsLink short hairpin RNA (shRNA) Expression Clones (Catalog no. HSH008831-LVRH1MH and HSH010830-LVRU6GP, GeneCopoeia; indicated in the figures as shHIF-1α and shLOXL2, respectively). CSHCTR001-LVRU6MH and CSHCTR001-LVRU6GP encoding non-specific shRNA were also used as negative controls.

LOXL2 overexpression was achieved by cloning the ORF into a lentiviral vector that induced elevated expression of the LOXL2 gene (Catalog no. EX-Y2020-LV201; indicated in the figures as LOXL2). EX-NEG-Lv201 was used as a negative control. Cells were transduced using a Lenti-Pac HIV Expression Packaging Kit (Catalog no. HPK-LvTR-40) according to the manufacturer’s protocol. Puromycin was used as the selection marker to obtain a stable cell line.

Total RNA was extracted using TRIzol reagent (Tiangen Biotech, Beijing, China) according to the manufacturer’s instructions; cDNA was prepared using the PrimeScript™ RT reagent Kit With gDNA Eraser (TaKaRa). Quantitative PCR (qPCR) was performed using the 7500/7500 Fast Real-Time PCR System (Applied Biosystems) with Tli RNaseH Plus (TaKaRa, RR820A). The following primers were used for qRT-PCR. HIF-1α, forward, 5’-GTCGGACAGCCTCACCAAACAGAGC-3’; reverse, 5’-GTTAACTTGATCCAAAGCTCTGAG-3’; LOXL2,forward,5’-CATCTGGATGTACAACTGCCACATA-3’;reverse,5’-AGCCCGCTGAAGTGCTCAA-3’; CDH1,forward, 5’-GAGTGCCAACTGGACCATTCAGT-3’;reverse, 5’-AGTCACCCACCTCTAAGGCCATC-3’; CDH5,forward, 5’-AGCCAGCCCAGCCCTCAC-3’; reverse: 5’CCTGTCAGCCGACCGTCTTTG-3’. Vimentin,forward,5’-TGACATTGAGATTGCCACCTACA-3’;reverse,5’- TCAACCGTCTTATACAGAAGTGTCC-3’. Glyceraldehyde3-phosphate dehydrogenase (GAPDH) was used as an endogenous control (forward primer, 5’-CCTGGCCAAGGTCATCCATGAC-3’; reverse primer, 5’-TGTCATACCAGGA- AATGAGCTTG-3’), and relative fold changes were calculated using the 2^-∆∆Ct method.

Whole cell lysates were separated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis and transferred onto polyvinylidene difluoride membranes. Blots were blocked and incubated with antibodies targeting LOXL2 (1:1000, Cat. #GTX105085; GeneTex, California, USA), HIF-1α (1:200, Cat. #Ab1; Abcam, Cambridge, UK), E-cadherin (1:200, Cat. #ZS-7870; Zhongshan Chemical Co, Beijing, China), vimentin (1:1000, Cat. #ab92547; Abcam, Cambridge, UK) and VE-cadherin (1:500, Cat. #ab33168; Abcam, Cambridge, UK). The membranes were then incubated with a secondary antibody (1:2000, Cat. #sc-2055, Cat. #sc-2004, Santa Cruz, CA, USA). The blots were developed using an enhanced chemiluminescence detection kit (Amersham Pharmacia Biotech, Piscataway, NJ, USA). To analyze protein loading, a monoclonal β-actin antibody (1:2000, Cat. #sc-47778; Santa Cruz, CA, USA) was used.

Cells were plated onto coverslips and fixed in ice cold methanol for 10 min. The cells were blocked with 5% FBS and incubated with the primary antibodies and the FITC-conjugated secondary antibodies. The sections were counterstained with 4,6diamino-2-phenylindole and observed under a fluorescence microscope (80i, Nikon) at × 200 magnification.

HepG2 and Bel7402 HCC cells (1x10⁵ cells) in 100 μl of MEM and RPMI 1640 without FBS were seeded into Matrigel matrix-coated upper 24 wells (1 mg/mL; BD Biosciences) containing polyethylene terephthalate filters with 8 μm porosity (Invitrogen). The lower chamber was filled with 10% FBS-containing medium. The cells were incubated for 48 h, and non- invading cells were removed from the upper surface of the membrane. The cells that invaded the Matrigel matrix and adhered to the bottom surface of the membrane were fixed with methanol and stained with 0.5% crystal violet. The number of invading cells was counted using an inverted light microscope (100× magnification) (Nikon). Each experiment was performed in triplicate.

For wound-healing assays, HepG2 and Bel7402 HCC cells were plated in 12-well culture plates. When the cells formed a monolayer, a straight scratch was made in the center of each well using a micropipette tip, and the cells were washed with PBS and incubated in serum-free medium. Cell motility was assessed by measuring the movement of the cells into the scratch in each well. We opened these pictures through drawing software and used scaleplate to measure the length of the wound. The migration rate (MR) was monitored after 24, 48 and 72 h. The following formula was used to calculate MR at different time points: MR = (d − d′)/d, where d is the length of the wound at time 0, and d′ is the length at other different time points. Each experiment was performed in triplicate.

In vitro VM vitro was evaluated using a 3D culture system. To create the 3D culture, Matrigel (BD, USA) was thawed at 4°Cand added to each well of a 96-well plate (30 μl/well). The plates were placed on ice and then moved into a 37 °C incubator containing 5% CO₂ for 12 h to solidify the Matrigel. Tumor cells in complete medium were then seeded onto the gel and incubated at 37 °C for 24 h. The formation of capillary-like structures was observed under phase contrast microscopy (100× magnification). Each experiment was performed in triplicate.

Total RNA from three groups of HCC cells was extracted using TRIzol reagent (Tiangen Biotech, Beijing, China). Briefly, the extracted RNA was labeled and hybridized on an Affymetrix Gene Expression Microarray (design ID: OE2015Q1548; Agilent Technologies, Santa Clara, CA, USA) by Oebiotech Co. (Shanghai, China). Statistical analyses and data normalization were performed using GeneSpring (version 13.1) software (Agilent Technologies). Differentially expressed genes were then identified by observing fold changes as well as by calculating the P-values using t-tests. The thresholds set for up-regulated and down-regulated genes were a fold change ≥2.0 and a P-value <0.05.

Analysis was performed using SPSS 21.0. The pathological and clinical characteristics of the two groups in hepatocellular carcinoma cases were assessed by the χ²test. Mean values were assessed using a two-tailed Student’s t test for paired data. Survival curves were estimated using the Kaplan-Meier method and compared by a log-rank test. Statistical significance was defined as p < 0.05.

We retrospectively evaluated the expression of HIF-1α and LOXL2 as well as the prevalence of VM among a cohort of 201 hepatocellular carcinoma specimens. The results showed that HIF-1α protein was highly expressed in 94 of 201 HCC sample tissues (46.76%; Fig. 1a and Table 1). LOXL2 protein was highly expressed in 102 of 201 HCC sample tissues (50.75%; Fig. 1a and Table 1). Elevated HIF-1α and LOXL2 expression was associated with an increase in tumor grade and VM (p < 0.05, Table 1). Finally, Kaplan–Meier survival analysis indicated that the HIF-1α and LOXL2 high expression groups have poor overall survival compared with the low expression groups (p < 0.05, Fig. 1b and c).

Tube cavities lined with PAS-positive, CD31-negative tumor cells and red blood cells observed within the cavities were considered VM channels (red arrows, Fig. 1d). Channels positive for both PAS and CD31 were defined as EDVs (green arrows, Fig. 1d).

In 48 specimens (23.88%), VM was observed. Notably, our results demonstrated that the presence of VM correlates with the tumor grade, metastasis, and poor prognosis (P <0.05, Table 2 and Fig. 1e).

To understand the clinical relevance of HIF-1α and LOXL2, we also explored the relationship between HIF-1α and LOXL2 by conducting a correlation analysis of HIF-1α and LOXL2 in 201 HCC patients. The data showed that HIF-1α expression was positively correlated with LOXL2 in the HCC samples (Table 3).

To confirm that HIF-1α induces LOXL2 expression in an HCC cell model, we treated HepG2 and Bel7402 HCC cells that stably expressed short hairpin RNA (shRNA) against HIF-1α with the hypoxia-mimetic agent CoCl₂ (150 μM) for 48 h. The protein and mRNA expression levels of HIF-1α and LOXL2 of these two cell lines were assessed by Western blot and qRT-PCR, respectively, and the results showed that LOXL2 expression was up-regulated by HIF-1α, while HIF-1α knockdown ameliorates LOXL2 up-regulation in HepG2 and Bel7402 cells (Fig. 2a and b). Furthermore, immunofluorescence revealed that high LOXL2 expression occurs following HIF-1α overexpression. Conversely, cells decreased-expressing HIF-1α demonstrated loss of LOXL2 expression (Fig. 2c). The results demonstrated that LOXL2 expression was induced by HIF-1α.

Previous observations demonstrated that HIF-1α and LOXL2 exert similar effects on the aggressive phenotypes in HCC cells and that HIF-1α regulates LOXL2 expression. Based on these results, we performed a rescue assay to assess whether the effects of HIF-1α on HCC cells are mediated by LOXL2 expression. Cells were co-transfected with LOXL2 and the HIF-1α short hairpin plasmid, and overexpression of LOXL2 was confirmed to rescue the decrease in the protein and mRNA levels of LOXL2 caused by shHIF-1α expression and treatment with CoCl₂ (Fig. 3c and d). As expected, restoring LOXL2 expression mostly blocked the inhibitory influence of HIF-1α knockdown on migration and invasion (Fig. 3a and b). Cells transduced with shLOXL2 plasmid were treated with CoCl₂, and the low LOXL2 expression was confirmed to prevent the increase in the protein and mRNA levels of LOXL2 caused by HIF-1α (Fig. 3c and d). Ectopic LOXL2 expression counteracted the increased migration and invasion of HCC cells induced by HIF-1α (Fig. 3a and b).

EMT is an important cellular process during tumor invasion and migration, and the tumor cells that exhibit the ability to undergo EMT also have a greater ability to invade and metastasize; thus, EMT-related indexes were evaluated using Western blotting and quantitative RT-PCR. The results confirmed that elevated E-cadherin expression and down-regulation of vimentin, a mesenchymal marker, in HepG2 and Bel7402 cells treated with CoCl₂ following HIF-1α knockdown. Conversely, cells overexpressing LOXL2 demonstrated a loss of E-cadherin and up-regulation of vimentin expression (Fig. 3c and d). At the same time, cells with LOXL2 knockdown that were treated with CoCl₂ exhibited increased E-cadherin expression and down-regulation of vimentin. The gray analysis showed that these differences were statistically significant (Fig. 3c and d).

To investigate the function of LOXL2 in HIF-1α-induced VM, we used a Matrigel-based tube formation assay. HIF-1α silencing in HCC cells inhibited VM in Matrigel, whereas LOXL2 overexpression facilitated the development of VM tubes. Treating cells with CoCl₂ resulted in the formation of tube-like structures on the surface of the Matrigel, whereas loss of LOXL2 triggered the disappearance of these structures (Fig. 4a). The expression of vascular endothelial-cadherin (VE-cadherin), a marker of vascular mimicry (VM), was also explored by Western blot and qRT-PCR. The results showed that HIF-1α down-regulation can suppress VE-cadherin protein expression, while VE-cadherin protein expression was increased when LOXL2 was overexpressed. In addition, cells transfected with the LOXL2 short hairpin plasmid had lower VE-cadherin expression than control cells (Fig. 4b and c).

Next, we assessed the mRNA expression profiles in hepatocellular carcinoma HepG2 cells (named Control group), HepG2 cells transfected with plasmid LOXL2 to upregulate LOXL2 expression (named LOXL2 group), HepG2 cells treated with CoCl₂ (named CoCl₂ group). Comparison results showed that 70 genes were differently expressed between LOXL2 group and Control group, including 48 upregulated genes and 22 downregulated genes (Fig. 5a). 1059 genes showed differential expression between CoCl₂ group and Control group, including 771 upregulated genes and 288 downregulated genes (Fig. 5b). Next, we analyzed these differently expressed genes whose fold change ≥ 2 in three groups. Moreover, we conducted venn analysis between LOXL2vsControl and CoCl₂vsControl to make further study on LOXL2 induced by hypoxia. The results of venn analysis indicated 65differentially expressed genes (DEGs) that may associate with LOXL2 promoted by hypoxia (Fig. 5c). Among the 65 genes, 12 genes, including HLTF, CENPF, ASPM, NIPBL, SLK, LRPPRC, PIK3C2A, SMC2, TAF9B, AGL, ATAD2, ATP5E, have been considered as cancer-related genes by reviewing other researchers’ studies in PubMed. Then quantitative real-time PCR was carried out to validate the changes in expression levels of the 12 genes (Fig. 5d-e). Among the 12 genes, the expression of HLTF, CENPF, ASPM, NIPBL, SLK, PIK3C2A, SMC2, TAF9B and ATAD2 were consistent with the results of microarray analysis and all these genes have an upregulated tendency in HepG2 cells transfected with plasmid LOXL2 or treated with CoCl₂. And the expression of AGL, LRPPRC and ATP5E have no change in LOXL2 group and CoCl₂ group compared to Control group.

In order to make functional interpretation for the gene expression changes, GO analysis was performed. Among the upregulated genes between LOXL2 group and Control group, we found several GO terms from the top 20 GO terms of biological process which may relate to our research, such as mitotic cell cycle, developmental growth and positive regulation of cell growth (Fig. 6a). Meantime, the biological process of GO analysis was applied for the downregulated DEGs in LOXL2 group, including chromatin organization, small GTPase mediated signal transduction and blood coagulation et al (Fig. 6b).

Following GO analysis for up-and downregulated DEGs in CoCl₂ group, significant GO terms of biological process were collected. For upregulated DEGs, cell division, mitotic nuclear division, regulation of DNA replication and DNA damage response, signal transduction by p53 class mediator from the top 20 GO terms of biological process may relate to our research (Fig. 6c). For downregulated DEGs, GO analysis of biological process such as diacylglycerol metabolic process, response to amino acid and low-density lipoprotein particle clearance may not relate to cancer (Fig. 6d). In addition, the biological process of GO analysis of upregulated DEGs between LOXL2vsControl and CoCl2vsControl, such as mitotic cell cycle, developmental growth, stem cell maintenance and mitotic nuclear division may associate with our research (Fig. 6e and Additional file 1: Table S4). However, the GO analysis of the downregulated DEGs identified biological process which were not cancer related and might not play roles in our study.