Dickkopf‐1‐promoted vasculogenic mimicry in non‐small cell lung cancer is associated with EMT and development of a cancer stem‐like cell phenotype

Abstract To characterize the contributions of Dickkopf‐1 (DKK1) towards the induction of vasculogenic mimicry (VM) in non‐small cell lung cancer (NSCLC), we evaluated cohorts of primary tumours, performed in vitro functional studies and generated xenograft mouse models. Vasculogenic mimicry was observed in 28 of 205 NSCLC tumours, while DKK1 was detected in 133 cases. Notably, DKK1 was positively associated with VM. Statistical analysis showed that VM and DKK1 were both related to aggressive clinical course and thus were indicators of a poor prognosis. Moreover, expression of epithelial‐mesenchymal transition (EMT)‐related proteins (vimentin, Slug, and Twist), cancer stem‐like cell (CSC)‐related proteins (nestin and CD44), VM‐related proteins (MMP2, MMP9, and vascular endothelial‐cadherin), and β‐catenin‐nu were all elevated in VM‐positive and DKK1‐positive tumours, whereas the epithelial marker (E‐cadherin) was reduced in the VM‐positive and DKK1‐positive groups. Non‐small cell lung cancer cell lines with overexpressed or silenced DKK1 highlighted its role in the restoration of mesenchymal phenotypes and development of CSC characteristics. Moreover, DKK1 significantly promotes NSCLC tumour cells to migrate, invade and proliferate. In vivo animal studies demonstrated that DKK1 enhances the growth of transplanted human tumours cells, as well as increased VM formation, mesenthymal phenotypes and CSC properties. Our results suggest that DKK1 can promote VM formation via induction of the expression of EMT and CSC‐related proteins. As such, we feel that DKK1 may represent a novel target of NSCLC therapy.


Background
Primary lung cancer is the leading cause of cancer mortality in the world [1]. Some advances in diagnosis and treatment have made significant strides towards improving these trends; however, the survival rate of patients with lung cancer remains poor. Thus, further research is critically necessary to expand our understanding of this disease and identify novel, actionable targets. Non-small cell lung cancer (NSCLC) comprises approximately 80% of all lung cancer cases [1,2].
In 1999, Maniotis et al., detected a novel mode by which highly aggressive uveal melanoma maintain blood supply. Specifically, the authors identified blood vessels that are completely surrounded by tumour cells, a condition that was designated vasculogenic mimicry (VM) [3]. Since the discovery of VM, research has focused on the molecular mechanism by which VM formation occurs in an effort to identify actionable targets. Epithelial-mesenchymal transition (EMT) and cancer stem-like cell (CSC) have been associated with the formation of VM in several tumours [4][5][6]. As the Wnt signalling pathwaywhich provides important clues to embryonic development and tumorigenesis -is implicated in both EMT and CSC, we evaluated negatively correlated with the existence of VM, and may reduce proliferation, migration and invasion of colon cancer cells [8]. By contrast, DKK1 exhibits a potential oncogenic function as a result of elevated expression detected in Wilms' tumour, hepatoblastoma and hepatocellular carcinoma (HCC) [11,12]. A previous report has shown that DKK1 is also highly expressed in NSCLC, and may be useful as a novel diagnostic and prognostic marker for lung cancer [13].
However, the relationship between DKK1 and VM in NSCLC remains unknown. In this study, we aimed to identify the potential contribution of DKK1 in the formation of VM. We hypothesized that DKK1 promotes VM formation via induction of EMT and development of CSC characteristics. To evaluate or premise, we obtained large cohorts of human NSCLC tissues to identify the clinical and biological overlap between VM and DKK1 expression. Subsequently, cell culture and xenograft mouse models were used for in vitro and in vivo studies, respectively.

Materials and methods Patients
Tissue specimens were obtained from 205 patients who had undergone surgical resection for lung cancer in Tianjin Medical University Cancer Institute and Hospital from October 1990 to November 2010. These 205 NSCLC samples included 79 cases of squamous cell carcinoma, 75 cases of adenocarcinoma and 51 cases of large cell cancer. The diagnoses of these samples were verified by two pathologists according to the standards of classification [2,14]. Clinicopathological parameters were obtained from patients' clinical records and pathological reports. Total survival time, final follow-up examination and diagnosis of metastasis were recorded from the date of surgery. This study was approved by the Ethical Committee of Tianjin Medical University.
Immunofluorescence, immunohistochemistry and CD31/periodic acid Schiff double-staining Immunohistochemistry was performed as described by Sun et al. [4,15]. The details of the primary antibodies are listed in Table S2. Secondary antibodies were purchased from Zhongshan Golden Bridge (Beijing, China). Positive and negative controls were run for each batch. PBS was used as substitute for primary antibodies in the negative controls. The results were evaluated according to the method described by Bittner et al. [16].

Western blot analysis
Western blot was performed according to previously published protocols [13]. A monoclonal beta-actin antibody (Santa Cruz, Dallas, TX, USA) was used for protein loading analysis, and the other primary antibodies are described above. Secondary antibodies were also purchased from Santa Cruz Biotechnology.

Cell proliferation assay
Cells were cultured in 96-well plates at a concentration of 1 9 10 3 cells per well and incubated for various periods (1, 2, 3, 4, 5 and 6 days). At a specified time, MTT (Sigma-Aldrich, St. Louis, MO, USA) solution was added. After 4 hrs incubation at 37°C, dimethyl sulfoxide was used to dissolve the purple crystals by shaking for 10 min. on a table concentrator. The optical density was determined at 490 nm on Spectra Max M2 (Molecular Devices, Sunnyvale, CA, USA).

Invasion assay
Cell invasion was detected using transwell cell culture inserts with 8 lm membrane pore size (BD Biosciences, San Jose, CA, USA). Matrigel (2 mg/ml; BD Biosciences) was placed on the upper surface of the chamber. Subsequently, 200 ll cell suspension (2 9 10 5 cells/ml) without bovine serum was added to the upper chamber, and 500 ll RPMI-1640 containing 20% FBS was added to the lower chamber. Following incubation at 37°C with 5% CO 2 for 48 hrs, the passed cells were fixed, stained and counted. Each experiment was performed in triplicate.

Migration and wound-healing assay
Migration assay was performed in the same manner as an invasion assay, except that the inner surface was not supplied with Matrigel and the incubation time was 24 hrs. For the wound-healing assay, cells were placed in a 12-well plate to form a monolayer 1 day before the assay. After making a uniform straight scratch with a pipette tip, cells were incubated at 37°C with 5% CO 2 , and cell motility was assessed by measuring the speed of wound closure at intervals. Each experiment was performed in triplicate.

3D culture assay
This assay was conducted as previously described [4,15]. The cells on Matrigel were pictured with a microscope (TS100-F; Nikon, Shinagawa- Ku, Tokyo, Japan). Six pictures in each group were collected, and the number of VM channels was counted.

Animal studies
A total of 20 male BALB/c nude mice, 3-to 4-weeks of age (purchased from Beijing H\FK Bioscience Co., Ltd., Beijing, China), were divided randomly into four groups: H460-control group, H460-transfection group, A549-control group and A549-transfection group. Each mouse was subcutaneously injected in the right armpit with 5 9 10 6 cells. The tumour size was measured every 3 days for 21 days with sliding calliper. The tumour volume was calculated using the following formula: volume = (length [mm] 9 width 2 [mm])/2. Tumour samples were fixed in formalin, embedded in paraffin, slide cut (4 lm thick) and immunochemically stained. All animal experiments were conducted according to the guidelines of Tianjin Medical University, China.

Statistical analysis
All data in the study were evaluated using SPSS17.0 software (SPSS, Chicago, IL, USA). Survival data were analysed by Kaplan-Meier analysis. C, Pearson chi-squared test, Spearman correlation analysis and t-test were used as needed. All P-values were two-sided, and P < 0.05 was considered a statistically significant test.

Association of VM and DKK1 with clinicopathological features in human NSCLC samples
Based on our previous studies [4,15], classical endothelial vessels were lined with shuttle-like endothelium, whereas the channels surrounded by tumour cells with the presence of red blood cells were deemed VM. These channels do not display necrosis nor were infiltrating inflammatory cells observed (Fig. 1A). As shown by CD31/periodic acid Schiff (PAS) double staining, endothelium-dependent vessels were positive for both CD31 and PAS, whereas the tumour cells lining VM channels were negative for CD31. The base membrane-like structure between red blood cells and the tumour cells were positive for PAS (Fig. 1B). In our study, VM was detected in 28 (13.66%) of 205 specimens. The frequency of VM was significantly associated with histological classification, differentiation, T stages, clinical stages, and distant metastasis (P < 0.001, <0.001, 0.021, 0.001 and <0.001 respectively). Poor differentiation, high distant metastasis, and advanced stage were observed in samples in the VM-tumours. Among the three histological types, the frequency of VM was greatest in large cell lung cancer (17/51, 33.33%), followed by adenocarcinoma (7/ 75, 9.33%) and squamous cell carcinoma (4/79, 5.06%). No statistically significant correlations were observed between VM and other clinicopathological features including: age, gender, tumour size, tumour location, pleural invasion, lymph node metastasis and therapy (P > 0.05; Table S1).
To verify the clinical significance of VM, we investigated the relationship between VM and survival outcome. The average survival period for VM-positive patients was 25.95 months, whereas for VM-negative patients, it is 68.07 months (Fig. 1E). The total survival period of patients with VM was significantly shorter than that of patients without VM (P = 0.001). In addition, Kaplan-Meier survival analysis demonstrated that DKK1 had an effect similar trend, as VMpositive versus VM-negative groups, in regard to survival. The total survival time for patients of the DKK1-negative group was longer than that of the DKK1-positive group (P = 0.012). The average survival period for patients of the DKK1-positive group was 55.89 months, whereas that of the DKK1-negative patients was 76.12 months (Fig. 1F).

Relationship between VM, DKK1 and EMT/CSC/ VM-related proteins in human NSCLC tissue
To characterize the role of DKK1 in VM formation, the relationship between VM and DKK1 was examined in NSCLC tissues. Dickkopf-1 was positively associated with VM (P = 0.003, r = 0.203). The positive expression of DKK1 in the VM group (25/28, 89.29%) was significantly higher than that in the non-VM group (108/177, 61.02%). Dickkopf-1-positive and VM-positive samples included 25 cases, and DKK1/VM-negative samples included 69 cases. All of 205 NSCLC samples had been used in IHC staining for EMT and CSC-related proteins.
Relationships between VM and EMT/CSC-related proteins were also detected. As shown in Table 1, no significant relationship was found between EpCAM and VM (P > 0.05), whereas the differential expression levels of EMT-related proteins (E-cadherin, vimentin, Twist, and Slug) and CSC-related proteins (CD44, nestin and CD34) between VM-positive and VM-negative groups was significant (P < 0.05, Fig. 2A and B). b-catenin nuclear expression and VMrelated proteins (VE-cadherin, matrix metalloproteinase2 (MMP2) and MMP9) were also overexpressed in the VM group (P < 0.05, Table 1, Fig. 2C).
The relationship between DKK1 and these related proteins was also examined. The results were similar to those of VM, with the CD34 as the only exception. Dickkopf-1 was significantly associated with EMT-related proteins (E-cadherin, vimentin, Twist, and Slug), CSC-related proteins (CD44 and nestin), VM-related proteins (VEcadherin, MMP2 and MMP9) and b-catenin nuclear expression (P < 0.05). No statistically significant relationship was found between EpCAM, CD34, and DKK1 (P > 0.05, Table 1, Fig. 2).

Effects of DKK1-transfection on EMT-related proteins of NSCLC cells in vitro
The expression of DKK1 was detected among three lung cancer lines representing distinct histological entities, namely, H460 (large cell lung cancer), H1299 (lung adenocarcinoma) and A549 (lung adenocarcinoma). Based on Western blot results, H460 cells were selected as the lowest expression of DKK1, while A549 cells demonstrated the lowest highest levels (Fig. S1A). Following transfection, DKK1 expression was evaluated by western blotting to confirm the transfection efficiency (Fig. S1A). The effect of DKK1 on VM formation was also investigated.
During stable transfection of the DKK1 plasmid into H460 cells, remarkable morphological changes were observed. H460 cells are typically polygonal and cobble-stone-like, which are features of epithelial phenotypes. Following transfection with DKK1, the cells gradually transformed into a spindle shape, which is a mesenchymal phenotype (Fig. 3A). Moreover, increased expressions of vimentin, Slug and Twist were detected, while expression of E-cadherin, which is an epithelial marker, decreased ( Fig. 3B and D).
By contrast, A549 cells exhibited more molecular characteristics of epithelial cells following transfection of DKK1-shRNAinto these cells. E-cadherin expression increased in the transfected cells, whereas protein levels of vimentin, Slug, and Twist were reduced ( Fig. 3B and D).

Effects of DKK1 overexpression/deletion on CSC-related proteins of NSCLC cells in vitro
Based on our characterization of human primary tumours, the expression of CSC-related proteins that were significantly associated with VM and DKK1 were examined. CD44 and nestin were both overexpressed in H460-DKK1 cells and down-regulated in A549-siDKK1 cells (Fig. 3C). The expression of b-catenin was also investigated in these transfected cells. We found that b-catenin expression  decreased in A549-siDKK1 cells and increased in H460-DKK1 cells ( Fig. 5B and D). The immunofluorescent staining for b-catenin confirmed the result of the western blot, and it also indicated that more b-catenin located in the nuclei of H460-DKK1 cells (Fig. 3D).

Effects of DKK1 overexpression/silencing on migration and invasion capability of NSCLC cells in vitro
To investigate the impact of DKK1 on migration, a wound-healing assay and a transwell assay without Matrigel were both performed. In the wound-healing assay, H460-DKK1 cells migrated more rapidly than H460 cells, and the migration distances were 0.39 and 0.28 mm respectively (P < 0.05, Fig. 4A). Moreover, the cell numbers were higher for H460-DKK1 cells that were passed, as compared to controls (191 versus 100, P < 0.05) (Fig. 4A). In the Matrigel transwell assay, the passed cells were also fixed, stained and counted. There was approximately a fivefold increase in cell invasion observed in the H460-DKK1group compared with controls (P < 0.05, Fig. 4B). In keeping with these results, A549-siDKK1 cells showed lower migration and invasion abilities than the control (P < 0.05, Fig. 4D and E).

Abnormal expression of DKK1 regulated the proliferation ability of NSCLC cells in vitro
To evaluate the impact of DKK1 on proliferation, an MTT assay was performed. As expected, we observed increase in the growth rate of H460 cells overexpressing DKK1 (P < 0.05, Fig. 4A-C). Moreover, suppression of growth was also found as early as the fourth day in A549-siDKK1 cells (P < 0.05, Fig. 4B and C).

Effects of DKK1 overexpression/deletion on VM formation of NSCLC cells in vitro
We evaluated the formation of VM in NSCLC cells using a 3D culture assay model. Compared with the control group, there were more pipe-like structures that formed in the H460-DKK1 group, and the inner walls of these pipes were smoother ( Fig. 5A and B). Moreover, as shown in Figure 5C and D, A549 cells were unable to form channels following transfection with DKK1-shRNA. By contrast, the control group had numerous pipeline structures. Vasculogenic mimicry-related proteins were also evaluated following DKK1 overexpression or suppression. In agreement with the results of the 3D culture, MMP2, MMP9 and VE-cadherin expressions increased in the H460-DKK1 group and decreased in the A549-siDKK1 group compared with their corresponding controls ( Fig. 5E and F).

Effects of DKK1-transfection on xenograft in vivo
To further validate the effect of DKK1 on NSCLC cells, xenograft mouse models were performed. Following the subcutaneous transplantation of H460-DKK1, A549-siDKK1 and their corresponding control cells in nude mice, xenografts were established for all 40 mice. Compared to the corresponding control group, we found that   transplanted tumour cells in H460-DKK1 group grew more rapidly, and the tumours in A549-siDKK1 group developed at a slower rate (P < 0.05, Fig. S2A and Fig. 6A). The tumour volume of the H460-DKK1 group was larger than that in parental H460 cells (1736.09 and 841.60 mm 3 ) respectively (P < 0.05, Fig. S2A). By contrast, the tumour size in the A549-siDKK1 group (268.95 mm 3 ) was smaller than that in the control (533.85 mm 3 ) (P < 0.05, Fig. 6A).
Vasculogenic mimicry was also examined in seven of the mice in the H460-transfection group, and was found to be higher than that in the control group (4/10, P < 0.05, Fig. S2G). Vasculogenic mimicry was found in two mice in the A549-transfection group, and was lower than that in the A549-control group (5/10, P < 0.05, Fig. 6G).

Discussion
Vasculogenic mimicry has been identified in several malignant tumours, including HCC, glioblastoma and breast cancer [17][18][19][20]. Based on these studies, VM results in more aggressive cancer and is associated with a poor prognosis. We found similar findings, in NSCLC specimens through the identification of VM which was associated with poor differentiation, metastasis, advanced stage, and a shorter survival period. With the unique structure of VM, tumour cells are exposed directly to the blood flow, which enable them to enter the microcirculation and metastasize to other organs with greater ease. In addition, VM most often occurs in large cell lung cancer, followed by lung adenocarcinoma and squamous cell carcinoma. Large cell lung cancer is the most pernicious subtype, which may have contributed to this result. We also evaluated the expression of DKK1 in human NSCLC tissues, and found that DKK1 was related with histological classification and differentiation similar to that in VM. Moreover, overexpression of DKK1 positively correlated with the existence of VM and high expression of some VM-related proteins (MMP2, MMP9 and VE-cadherin).
Several studies have shown that VM is usually regarded as an example of aggressive tumour cells with remarkable differentiation plasticity [20,21]. In this process, more aggressive tumour cells transdifferentiated, altered their cell markers, acquired more embryonic stem characteristics, and acted as endothelium cells to form the VM structure. Therefore, we can assume that the actions of EMT and CSC may be involved in VM formation.
Epithelial-mesenchymal transition describes a series of events during which epithelial cells lose many of their epithelial characteristics and take on properties typical of mesenchymal cells, implying complex changes in cell architecture and behaviour [22,23]. Wnt signalling pathway is essential to the EMT induction. Cell division cycle 6 (Cdc6), a component of Wnt signalling pathway, has been shown to inhibit the translational activity of E-cadherin [24]. Overexpression of Cdc6 can suppress E-cadherin and lead to EMT in a lung cancer cell line [24]. Dickkopf-1 is also a regulator of Wnt signalling pathway. Our data highlight a correlation between the existence of VM, overexpression of DKK1 and decreased E-cadherin expression, and increased vimentin expression in NSCLC tissues. Moreover, overexpression of DKK1 in lung tumour cells decreased E-cadherin expression and increased vimentin expression. b-catenin, the central molecule of the Wnt pathway, and its nuclear expression, which has been shown to induce EMT and is used as a mesenchymal marker, was also positively related to VM and DKK1. Furthermore, VM and DKK1 demonstrated a correlation with Slug and Twist, which are two important positive regulators of EMT. Results of in vitro and in vivo experiments further indicate that DKK1 can sufficiently induce EMT and promote VM formation.
The CSC theory may explain the biological heterogeneity of solid tumours [25]. Similar to normal adult stem cells, cancer stem cells possess high tumorigenic ability, as well as self-renewal and pluripotency that could differentiate into different cell types [26]. Our data suggest that CSC phenotype was associated with VM and concomitantly with DKK1 overexpression. Our in vivo studies also showed that DKK1-overexpression in lung cancer cells exhibited CSC phenotype and were more aggressive. These xenograft tumours were more vas- cular and displayed more VM than cells from normal lung cancer cells. Cancer stem-like cell may also directly contribute to tumour angiogenesis by converting into endothelial cell [18,27]. Previous studies have reported that the cells engaged in VM may come from cancer cells endowed with transdifferentiation and stemcell plasticity, and may represent an incomplete differentiation of CSC towards endothelial lineage [18,20,28,29]. Smadja et al. also found that DKK1 enhanced angiogenic properties of endothelial colony-forming cells in vitro and was required for endothelial colony-forming cell and mesenchymal stem cell angiogenic phenotypes in vivo in breast cancer [30]. Based on previous studies and our data, we postulate that overexpression of DKK1 can lead to transdifferentiation of lung cancer cells, resulting in the loss of their epithelial cell phenotypes and the occurrence of more mesenchymal cell phenotypes. Meanwhile, in the EMT process, these tumour cells also develop cancer stem cell features, which enhance their abilities of proliferation, invasion, migration and tumorigenesis. These cells can act as endothelium cells, which are known as mesenchymal cells, to form the VM channel. Thus, the tumour with VM shows more aggressive behaviour, can metastasize more easily, and leads to poorer prognosis. These CSCs are assumed to generate erythroid cells, as Zhang reported [31]. Dickkopf-1 seems to have opposing roles in different tissues and types of tumours. Similar to DKK1, the functions of the E2F transcription factors can vary significantly in malignancies of the digestive system [32]. Previously, Qi et al. in our laboratory found that DKK1 inhibited EMT and VM in colon cancer. Considering the conflicting results in our laboratory, we   1683 speculate that the differences in results were because of the use of colon cancer cells in the previous study [8], whereas we used NSCLC [8]. Ultimately, the role of DKK1 in cancer is not completely understood [33]. Dickkopf-1 can bind to the Wnt co-receptor LRP5/6 and transmembrane proteins Kremen1/2 to form a tripolymer that block the binding of Wnt, LRP5/6 and Fz receptor to cause the degradation of b-catenin [34]. Hence, DKK1is able to inhibit the b-catenin-mediated transcription of the EMT-related molecule including snail, slug and twist1 in breast cancer, melanoma and colon cancer [35][36][37]. Interestingly, DKK1 is also a downstream target of Wnt signalling [33], and thus there is a negative feedback loop that activation of the canonical Wnt signalling (but not non-canonical Wnt signalling) which causes up-regulation of DKK1 expression [10]. The DKK1 promoter can be transactivated by the b-catenin/ T-cell factor (TCF) complex [38]. Dickkopf-1 is overexpressed in human hepatoblastomas, multiple myeloma, Wilms' tumours and oesophageal carcinomas, which display unregulated activation of Wnt signalling [11,33].
In conclusion, this study describes the previously unrecognized role of DKK1 and authenticates the hypothesis that DKK1 promotes VM formation by inducing EMT-related proteins and by developing CSC characteristics in NSCLC. Our findings may benefit future studies on the mechanism of NSCLC progression and provide new hope for NSCLC diagnosis and therapeutic strategies.