δ-Catenin Promotes Bevacizumab-Induced Glioma Invasion

Glioblastoma is the most destructive brain tumor. Despite the advances in basic and clinical research, patients with glioblastoma have a poor prognosis, with a median survival time of around 14 months (1). Characteristics of malignant glioma include it being fast-growing, invasive, aggressive, and capable of angiogenesis. Recently, clinical trials to investigate drugs directed at molecular targets have been performed (2, 3).

Vascular endothelial growth factor (VEGF), which promotes angiogenesis, is abundantly expressed by glioma cells. The results of trials with bevacizumab, an anti-VEGF antibody, showed significant effects for treatment of recurrent glioblastoma patients (4–6). The phase III AVAglio and RTOG 0825 studies indicated that the combination of bevacizumab, radiation, and temozolomide in newly diagnosed patients with glioblastoma did not show a statistically significant increase in overall survival (OS). However, these trials did indicate a statistically significant improvement in progression-free survival (PFS), (refs. 7, 8). Anti-VEGF treatments demonstrated that the enhancement on contrast-enhanced MRI was reduced, however, the outcome in patients was not improved with these treatments (9).

Piao and colleagues found that anti-VEGF therapy made tumors more resistant, invasive and aggressive, as the epithelial-to-mesenchymal transition took place (10). Lu and colleagues also showed that VEGF was able to regulate tumor invasiveness by recruiting protein tyrosine phosphatase 1B to MET/VEGFR2 heterocomplexes, which inhibited MET phosphorylation and tumor migration (11). In our previous research, increased glioma cell invasion due to anti-VEGF therapy was associated with extracellular matrix (ECM) changes, and an integrin inhibitor (cilengitide) reduced anti-VEGF therapy-induced glioma cell invasion (12).

The process of tumor cell invasion leads to the production of adhesion molecules and ECM. Studying the mechanism of bevacizumab-induced glioma invasion will help to develop novel glioma therapeutic strategies (13). In this study, we investigated the changes in expression of adhesion molecules in gliomas cells exposed to bevacizumab.

The human glioma cell lines U87ΔEGFR, U251MG, A172, and Gli36 were prepared and maintained as described previously (14). Bevacizumab was provided by Genentech/Roche/Chugai Pharmaceutical Co. The human glioblastoma-derived cancer stem cell line, MGG23 cells were provided by Dr Hiroaki Wakimoto at Massachusetts General Hospital. The human glioblastoma-derived cancer stem cell line, MGG23 cells were cultured as previously described (15, 16). U87ΔEGFR, U251, and A172 were authenticated by Promega via short tandem repeat profiling in December 2016. Mycoplasma is negative in all cells.

This study was conducted in strict accordance to the recommendations in the Guide for the Care and Use of Laboratory Animals in Japan. All procedures and animal protocols were approved by the Committee on the Ethics of Animal Experimentation at Okayama University as described previously (12).

We prepared 1.0 × 10⁵ U87ΔEGFR cells/μL as described previously (12). These cells were injected into athymic mice (balb/c-nu/nu; CLEA Japan, Inc.) or athymic rats (F344/N-nu/nu; CLEA Japan, Inc.). Tumor cells were implanted into the right frontal lobe of athymic mice (2 μL: 3 mm lateral and 1 mm anterior to the bregma at a depth of 3 mm) and athymic rats, as described previously. We administered PBS or bevacizumab (athymic mice or rats: 6 mg/kg) intraperitoneally, three times per week, starting on day 5 after tumor cell implantation.

We assessed the survival time of the U87ΔEGFR mouse glioma model using a Kaplan–Meier survival analysis.

Athymic animals were sacrificed at 18 days after tumor implantation, following six administrations of PBS or bevacizumab. The maximum transverse diameter of tumors was measured. Hematoxylin and eosin (HE) staining was performed as described previously (17).

Eighteen days after tumor implantation, U87ΔEGFR mouse models treated with bevacizumab or PBS were sacrificed (n = 4 per group). RNA was extracted as previously described (10). RNA analysis was conducted with an Extracellular Matrix & Adhesion Molecules RT Profiler PCR Array (PAHS-013Z; Qiagen), according to the manufacturer's instructions. Expression of genes encoding 84 human cell adhesion molecules and ECM components was evaluated in brain tumor tissue after bevacizumab treatment relative to control (PBS). We defined a gene as being upregulated when the bevacizumab treatment/control average intensity ratio was >2.0, and downregulated when the bevacizumab treatment/control ratio was <0.5. Data were extracted using the following criteria: P-value < 0.05. The array included controls to assess cDNA quality and DNA contamination. All data are deposited in the Gene Expression Omnibus (accession number GSE126168).

We isolated total RNA from tumor cells incubated with bevacizumab (0, 1, 5, 10, and 100 μg/mL for 24 hours) or PBS with an RNeasy Mini Kit (Qiagen). In vivo, total RNA was extracted from the brain tumor tissue of mice that had been treated with PBS or bevacizumab using the TRIzol reagent (Invitrogen) as the manufacturer's instructions. qRT-PCR was performed as previously described (18). β-Actin was used as an internal control.

We assessed gene expression of COL1A1 (Collagen, type I, alpha 1), ITGB2 [Integrin, beta 2 (complement component 3 receptor 3 and 4 subunit)], VCAM1 (vascular cell adhesion molecule 1), and δ-catenin (CTNND2; cadherin-associated protein, delta 2 [neural plakophilin-related arm-repeat protein]) in athymic mice harboring U87ΔEGFR brain tumors

Primers were as follows: COL1A1: F (forward), 5′-cctggatgccatcaaagtct-3′, R (reverse) 5′-gaatccatcggtcatgctct -3′; ITGB2 F (forward), 5′-gaccagagcatttccaacacc -3′, R (reverse), 5′-tttaacggtttcagggaacg-3′; VCAM1 F (forward), 5′-atttcactccgcggtatctg-3′, R (reverse), 5′-ccaaggatcacgaccatctt -3′; CTNND2 F (forward), 5′-gccggaagtgattcagatgt -3′, R (reverse), 5′-ctacggtggacttcggtcat -3′; and β-actin F (forward), 5′-agagctacgagctgcctgac-3′, R (reverse), 5′-agcactgtgttggcgtacag-3′.

The avidin–biotin–peroxidase complex method (Ultrasensitive; MaiXin) was performed for IHC. δ-catenin mouse monoclonal antibody (1:300 dilution; Abcam, Inc.) was used for the staining. Mouse immunoglobulin was used as a negative control. Both antibodies were stained as previously described (15). Hematoxylin was used for counterstaining. We evaluated the positivity of cytoplasmic immunostaining in tumor cells.

For scratch wound assays, data were collected at 6, 12, 18, and 24 hours after scratching. We incubated cells to confluence using serum-starved DMEM. Scratches were made using a 200-μL pipette tip and monolayer wounds were made. The medium was changed to DMEM containing either bevacizumab or vehicle (PBS). Glioma cells were assessed for movement after 18 or 24 hours of exposure to bevacizumab. Serum was starved during the assay.

The invasion assay in vitro was performed using a 24-well plate and ThinCert (8 μm-pore, 24-well format; Greiner Bio-one) according to the manufacturer's instructions. The lower chamber was filled with DMEM with 10% FBS as a chemoattractant with or without 5 μg/mL bevacizumab. After a 24-hour incubation, noninvading cells were scraped from the top compartment. The insert filters were stained with 5% Giemsa solution. The number of invading cells was counted on the lower surface of the filter.

The in vitro invasion assay was performed using 96-well ultra-low attachment wells (Corning, Inc.) according to the manufacturer's instructions. Briefly, 1 × 10⁵ cells were seeded in low-serum DMEM in the bottom of wells. After a 24-hour incubation, siRNAs were transfected into cells. After 24 hours, bevacizumab or vehicle (PBS) was added to the wells, and subsequently Matrigel was added. Digital photomicrographs of the spheroid midplane were taken daily using with the BZ-8100 microscope (Keyence). The radius of invasion was calculated using ImageJ (http://rsb.info.nih.gov/ij/), as previously described (19).

The water-soluble tetrazolium-1 (WST-1) assay was performed according to the manufacturer's protocol (Roche). Briefly, 2 × 10³ cells in 100 μL of medium were placed in a 96-well microculture plate and incubated at 37°C for 24 to 72 hours. Ten microliters of WST-1 solution was then added and cells were incubated for 4 hours. Optical absorbance at a test wavelength of 450 nm and reference wavelength of 600 nm was measured with a microplate reader (Multiskan FC; ThermoFisher Scientific).

Total cell protein was extracted in lysis buffer, and quantified using the Bradford method (20). Samples of 50 μg of protein were separated by SDS-PAGE. Samples were transferred to polyvinylidene fluoride membranes and incubated overnight at 4°C with anti-δ-catenin antibodies (1:1,000; Abcam), anti-VEGFR2 antibodies (1:1,000; Cell Signaling Technology), and a mouse monoclonal antibody against ß-actin (1:5,000; Sigma). The membrane was incubated with HRP-conjugated appropriate secondary antibodies for 1 hour. Immunoreactive bands were visualized using the ECL Prime Western Blotting Detection Reagent (GE Healthcare) and the VersaDoc 5000 MP (Bio-Rad Laboratories). We quantified relative protein levels using ß-actin as a loading control.

Lipofectamine 3000 transfection reagent (ThermoFisher Scientific) was used for transfection of small-interfering RNAs (siRNA) into U87ΔEGFR cell lines, according to the manufacturer's instructions. The δ-catenin siRNA sequence was #1:5′- UCG CCU CAG UCA AAG AAC AGG AAU U -3′; and #2: 5′-CCC ACA GGA UUA UUC UAC AGG UGA A-3′. A nonsilencing siRNA sequence was used as a negative control [ThermoFisher Scientific Inc. (Dharmacon, Stealth RNAi siRNA)]. Protein levels were assessed after 48 hours using Western blotting. Targeted downregulation of δ-catenin in U87ΔEGFR cells was achieved by transfection of an siRNA targeted to δ-catenin [5′-(UCU CGA GCU GGU GGA CUC CUG UAU U)-3′; ThermoFisher Scientific), according to the manufacturer's instructions.

Human δ-catenin plasmids were purchased from GeneCopoeia (Catalog No. LPP-Prodld-LV202-050). The lentivirus encoding human δ-catenin plasmids, scramble short hairpin RNA (shRNA), and shRNA against human δ-catenin and VEGFR2 were prepared using the pLKO.1, psPAX2, and pMD2.G plasmids (Addgene); 293FT cells; and the FuGENE6 Transfection Reagent according to manufacturer's recommendation. The shRNA sequences against human VEGFR2 were as follows:

shVEGFR2#1: 5′-CCGGCCACAGATCATGTGGTTTAAACTCGAGTTTAAACCACATGATCTGTGGTTTTT-3′; and

shVEGFR2#2: 5′-CCGGCTGGAATGAATACCCTCATATCTCGAGATATGAGGGTATTCATTCCAGTTTTT-3′.

The antitumor effect of bevacizumab was tested in animals harboring intracerebral U87ΔEGFR glioma cells. The survival time of mice treated with bevacizumab was longer than that of control mice (median survival = 17.5 and 26 days, respectively; P = 0.0033; Fig 1A).

As assessed using the maximum transverse diameter, tumor volume was lower in the brains of rats treated with bevacizumab compared with untreated controls at 18 days after tumor inoculation (P = 0.0209; Fig. 1B).

The mouse glioma model with U87ΔEGFR cells showed angiogenic growth and well-demarcated borders in the brain (Fig. 1C, a). However, anti-VEGF therapy with bevacizumab increased cell invasion (Fig 1C, b). In the mouse model, we observed similar histologic results to previous rat data (12).

To illustrate the molecular mechanisms related to the anti-VEGF treatment invasiveness, gene expression in tumor tissues was compared between U87ΔEGFR orthotopic mice in the bevacizumab and the control. The resultant plot arranges genes according to their biologic and statistical significance (Fig. 1D). Several genes exhibited differential expression in bevacizumab-treated U87ΔEGFR glioma tissue compared with control U87ΔEGFR glioma tissue, comprising 11 upregulated genes (Table 1) and four downregulated genes (Table 2). For the upregulated genes, the following 11 genes were identified: δ-catenin, VCAM1, ITGB2, MMP7, MMP1, ITGA8, ANOS1, HGDC, TGFBI, COL1A1, and COL4A2. For the downregulated genes, the following four were identified: LAMA2, HAS1, VTN, and ITGA7.

To confirm the results from the qRT-PCR array analysis, COL1A1, ITGB2, VCAM1, and δ-catenin, which are associated with the ECM, adhesion molecules, and tumor cell invasion, were verified using qRT-PCR analysis. Relative expression levels of COL1A1, VCAM1, ITGB2, and δ-catenin in the U87ΔEGFR mouse model with bevacizumab were upregulated compared with the control group by 2.12-fold, 1.59-fold, 2.38-fold, and 16.5-fold, respectively (P < 0.05; Fig. 1E). δ-catenin expression was markedly elevated.

Using IHC, bevacizumab treatment led to high δ-catenin expression in U87ΔEGFR cells compared with the control group (Fig. 1F). δ-catenin was expressed at the invasive tumor border in the mouse glioma model of U87ΔEGFR cells. These data indicate that δ-catenin is related to bevacizumab-induced glioma invasion in an in vivo glioma model.

We investigated the cytotoxic effect of bevacizumab on glioma cells in vitro. Incubation with bevacizumab for the indicated time did not alter the glioma cell proliferation rate (Supplementary Fig. S1).

Wound closure was significantly enhanced by bevacizumab (Fig. 2A; Supplementary Fig. S2). Low-dose bevacizumab also elevated the motility of U251MG, A172, Gli36 cell lines (Fig. 2B). This assay indicated that bevacizumab treatment influences glioma cell migration.

δ-catenin expression was also confirmed by qRT-PCR and Western blot analysis in vitro. Relative expression of δ-catenin in U87ΔEGFR cells treated with bevacizumab was increased by 23-fold compared with controls (P < 0.05; Fig. 2C). In Western blotting, δ-catenin protein expression was augmented in a dose-dependent manner (Fig. 2D). The expression of δ-catenin was also increased in U251MG, A172, and Gli36 glioma cell lines compared with untreated cells (Fig. 2E).

We established a δ-catenin overexpressed tumor cell line after lentivirus transfection of U87ΔEGFR cells. δ-catenin was upregulated by δ-catenin overexpressed transfection compared with control. Expression levels of COL1A1, VCAM1 were also upregulated, but ITGB2 was not significantly changed by δ-catenin overexpressed transfection (Fig. 3A).

δ-catenin expression levels were confirmed using Western blotting analysis (Fig. 3B). In the two-chamber assay, δ-catenin-overexpressing U87ΔEGFR cells were more invasive compared with the control (Fig. 3C, *P < 0.05). Mice in the δ-catenin-overexpressed cells group showed unclear tumor borders. We assessed the depth of invasion from the tumor border to the invading cells. The depth of tumor invasion in the U87ΔEGFR cells expressing δ-catenin was significantly increased compared with empty vector-transfected U87ΔEGFR cells (Fig. 3D). δ-catenin overexpression was confirmed at the invasive tumor border in the mouse orthotopic glioma model with overexpressing δ-catenin U87ΔEGFR cells (Supplementary Fig. S3).

δ-catenin expression levels were confirmed by qRT-PCR analysis after siRNA transfection of U87ΔEGFR cells. Relative expression of δ-catenin in U87ΔEGFR cells transfected with the δ-catenin siRNA was decreased compared with bevacizumab-treated cells (P < 0.05; Fig. 4A). Western blotting data also showed that U87ΔEGFR (Fig. 4B and C) and glioma stem cell (GSC, MGG23; Fig. 4D) δ-catenin levels were lower after siRNA or shRNA transfection than bevacizumab incubation. COL1A1, ITGB2, and VCAM1 expression patterns were examined when δ-catenin was knocked down. COL1A1, ITGB2, and VCAM1 expression levels were detected (Fig. 4E). In the scratch assay, U87ΔEGFR cells harboring the δ-catenin siRNA were less invasive than bevacizumab-treated cells (Fig. 4F and G, *P < 0.05, **P < 0.01). Moreover, in the two-chamber invasion assay, U87ΔEGFR were less invasive after transfection of the δ-catenin shRNA compared with bevacizumab incubation (Fig. 4H and I).

Matrigel invasion assays showed that δ-catenin siRNA transfection inhibited bevacizumab-induced glioma cell invasion (Fig. 5Aa, B; U87ΔEGFR, Ab; MGG23).

Moreover, we made U87ΔEGFR cells expressing δ-catenin shRNA or scramble shRNA and implanted each cell type into the brain of athymic mice. Tumors in the PBS group showed well-defined borders (Fig. 5C, a), whereas bevacizumab-treated cells expressing scramble shRNA consistently invaded normal brains and the tumor border became irregular (Fig. 5C, b). The invasion depth in the U87ΔEGFR-expressing δ-catenin shRNA samples treated with bevacizumab was significantly less compared with U87ΔEGFR-expressing scramble shRNA treated with bevacizumab group (Fig. 5D, P < 0.05). Decreased δ-catenin expression was confirmed at the tumor in the mouse orthotopic glioma model with δ-catenin-knockdown U87ΔEGFR cells treated with bevacizumab (Supplementary Fig. S3). We made VEGFR-knockdown GBM cells using shRNA against human VEGFR2. The δ-catenin expression level was not changed when VEGFR2 was knocked down. Cell viability and migrating cells were decreased (Supplementary Fig. S4). Invasion cells under bevacizumab treatment were also decreased when VEGFR2 was knocked down (Fig. 5E).

This study demonstrated that anti-VEGF therapy prolonged the overall survival of mice. In the glioma orthotopic models, bevacizumab treatment inhibited angiogenesis, but promoted tumor invasion. Analysis on qRT-PCR arrays of xenograft tumor cells demonstrated that bevacizumab alters gene expression of some ECM and adhesion molecules. qPCR array analysis in the U87ΔEGFR orthotopic mouse model showed that bevacizumab led to upregulation of COL1A1, VCAM1, ITGB, and δ-catenin (CTNND2). This observation affirmed the previously reported findings that bevacizumab increases perivascular ECM in tumors containing collagen fibers (12). δ-catenin gene expression was potently upregulated. In this study, δ-catenin was associated with bevacizumab-induced glioma invasion in vivo and in vitro.

Previously, we showed a decreasing the number of vessels in the tumor xenograft model (10), and we demonstrated prolonged survival when treated with bevacizumab. The VEGF autocrine signaling loop is suppressed, the Akt and Erk pathways are activated, and tumor growth and invasion are stimulated by anti-VEGF therapy (21). Molecules within the ECM microenvironment such as proteoglycans and collagens may influence the process of tumor invasion during anti-VEGF therapy (22).

Cell–cell adhesion molecules include the cadherin family and the immunoglobulin superfamily (IgSF). Cell–matrix adhesion molecules include the integrin family and components of the ECM fibronectin, laminin, collagen, tenascin, BAI1 (brain specific angiogenesis inhibitor), CCN1 (cysteine-rich 61/connective tissue growth factor/nephroblastoma overexpressed), and proteoglycans (3, 18, 23–30). Comprehensive analysis of adhesion molecules by qRT-PCR array revealed upregulated expression of the δ-catenin gene. By qRT-PCR array analysis, anti-VEGF therapy increased gene expression of COL1A1, VCAM, ITGB2, and δ-catenin compared with the control group. These data reflect the underlying mechanisms of bevacizumab-induced glioma invasion. The fold-change in expression was markedly higher for δ-catenin than the other genes.

δ-catenin is a member of the p120 subfamily, and these proteins are almost solely expressed in nervous system structures. δ-catenin is crucially associated with nervous system functions such as neurite elongation and dendritic morphogenesis (31, 32). δ-catenin belongs to the p120 catenin subgroup. p120 catenin binds to E-cadherin at the proximal membrane side, regulating cell growth through stimulation of c-Src (33), and affects cell motility via changes to the cytoskeleton through actin filament movements (34). δ-catenin was discovered because it is able to bind to presenilin-1 (35) that was related to mutated genes found in familial Alzheimer's disease (36). δ-catenin plays a significant role in dendritic morphogenesis, which is related to changes in small GTPase activity.

For the relationship between δ-catenin and other ECM molecules, we showed an upregulation of COL1A1 with overexpression of δ-catenin, but even with δ-catenin knockdown, COL1A1 was still upregulated. Although little is known about the relationship between COL1A1 and δ-catenin, δ-catenin binds to E-cadherin (37), and cadherin is associated with COL1A1 (38). Moreover, the expression of another type of catenin, β-catenin, was reported to result in spontaneous, progressive skin fibrosis with thickened collagen fibers (39). In stabilized β-catenin cells, there was an increase in the relative expression of COL1A1 mRNA. The relationship between δ-catenin and COL1A1 may occur through a similar mechanism as that of β-catenin. However, δ-catenin binds to E-cadherin in a competitive manner with p120 catenin (37). For δ-catenin knockdown, p120 catenin might bind to cadherin and induce COL1A1 upregulation. Further studies on the detailed mechanisms of the relationship between δ-catenin and other ECM molecules are required.

δ-catenin is strongly expressed in the normal brain, especially in neurons. Mutations in the δ-catenin gene, which loses its function in mesenchymal glioblastoma, might be the key event that causes aggressiveness in this glioblastoma subtype (40). The δ-catenin mutations are associated with transformation of the glioma into a very aggressive mesenchymal phenotype. We used the IVY glioblastoma atlas project database and showed that δ catenin is highly expressed at the tumor edge and in the infiltrating tumor (Fig. 5F). δ-catenin might be associated with astrocytoma progression and could be a potential biomarker for the behavior of astrocytoma cells (41). In this study, we showed that δ-catenin expression was upregulated by low concentrations of bevacizumab in vitro and in a bevacizumab-induced invasive model in vivo.

siRNA and shRNA-mediated knockdown of δ-catenin in U87ΔEGFR cells markedly decreased bevacizumab-induced cell invasion. δ-catenin was reported to promote the invasion of colorectal cancer cells by binding to E-cadherin in a competitive manner with p120 catenin (37). δ-catenin was found to stimulate astrocytoma cell invasion through upregulation of Rac1 activity (41). Therefore, δ-catenin is strongly associated with tumor cell invasion.

It has been reported that VEGF directly and negatively regulates tumor cell invasion (42, 43). The main VEGF signaling circuit is VEGFR2, which is related to all critical endothelial functions such as proliferation, migration, and vessel formation (44–46). VEGF binding to VEGFR2 has also been reported to inhibit invasiveness (43). In our VEGFR-knockdown experiment, δ-catenin expression level was unchanged. However, invasion cells under bevacizumab treatment were decreased. There is a possibility of different routes of bevacizumab-induced invasion. Further information about the detailed mechanisms of the relationship between the VEGF pathway and δ-catenin is required.

This study showed that some ECM factors were altered and that anti-VEGF therapy induced glioma cell invasion. The findings suggest that δ-catenin is implicated in bevacizumab-induced glioma cell invasion.

The Student t test and ANOVA were used to test for statistical significance. Data are presented as the mean ± SD and SE. Differences were considered to denote statistical significance when P < 0.05. All statistical analyses were performed using SPSS statistical software, version 20 (SPSS, Inc.).

No potential conflicts of interest were disclosed.

Conception and design: J. Ishida, Y. Otani

Development of methodology: J. Ishida, T. Ichikawa, Y. Otani, Y. Tomita, Y. Hattori,

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T. Shimizu, J. Ishida, T. Ichikawa, Y. Otani,

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T. Shimizu, J. Ishida, K. Kurozumi, T. Ichikawa, Y. Tomita,

Writing, review, and/or revision of the manuscript: T. Shimizu, K. Kurozumi, T. Ichikawa, Y. Otani, T. Oka, Y. Tomita,

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. Shimizu, T. Ichikawa, A. Uneda, Y. Matsumoto, I. Date

Study supervision: K. Kurozumi, I. Date

We would like to thank M. Arao, N. Uemori, and Y. Ukai for their technical assistance. The following medical students also contributed to the animal experiments: Y. Inoue. Bevacizumab was generously provided by Genentech/Roche/Chugai Pharmaceutical Co. We thank Jodi Smith, PhD, from Edanz Group (www.edanzediting.com/ac) for editing a draft of this manuscript. This study was supported by grants-in-aid for Scientific Research from the Japanese Ministry of Education, Culture, Sports, Science, and Technology to K. Kurozumi (No. 23592125; No. 26462182) & I. Date (No. 26670644).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.