PBX3/MEK/ERK1/2/LIN28/let-7b positive feedback loop enhances mesenchymal phenotype to promote glioblastoma migration and invasion

Glioma gene expression data were downloaded from three datasets: Repository for Molecular Brain Neoplasia Data (Rembrandt; http://​caintegratorinfo​.​nci.​nih.​gov/​rembrandt); and the National Center for Biotechnology Information Gene Expression Omnibus (NCBI-GEO) datasets GSE4290 and GSE59612 (http://​www.​ncbi.​nlm.​gov.​geo/​; accession nos. GSE4290 and GSE59612). Forty-five GBM samples were collected from the Department of Neurosurgery, The First Affiliated Hospital of Nanjing Medical University between 2011 and 2014. Informed consent for the use of samples was obtained from all patients. Our study was approved by the institutional review board and ethics committee of Nanjing Medical University.

U87 and U251 cells were purchased from the American Type Culture Collection (ATCC, Manassas, VA). Cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM; Hyclone, USA) supplemented with 10% fetal bovine serum (Gibco, Invitrogen, Carlsbad, CA, USA) at 37 °C in a humidified atmosphere with 5% CO₂.

Lentiviruses carrying PBX3 or vectors or siRNA-PBX3 or siRNA-NC were purchased from Genepharma (Shanghai, China). Stable U87 and U251 cells were established by lentiviral infection and puromycin selection as manufacturer’s protocol. Although, our previous study showed that PBX3 protein levels in H4 and U118 cell lines were significantly lower than that in U87 and U251 cells [27], we did not select H4 and U118 cells for overexpression studies. The main reason is the low ovexpression efficiency of PBX3 in both H4 and U118 cells. Let-7b mimic and inhibitor were purchased from RiboBio (Guangzhou, China) and transfected into cells using Lipofectamine RNAiMAX Reagent (Life Technologies, Grand Island, NY) according to manufacturer’s protocol. siRNAs or plasmids were synthesized by Genepharma and transfected into cells using Lipofectamine 2000 Transfection Reagent (Invitrogen).

Recombinant human transforming growth factor-β (TGFβ), U0126 and phorbol 12-myristate 13-acetate (PMA) were purchased from Sigma, and diluted according to each manufacturer’s protocol. The dose of each drug was used as previously described [17, 31].

Total RNA extraction and qRT-PCR for mRNA were performed as previously described [26]. Primers used in this study were listed as follows: PBX3 forward 5′-CAAGTCGGAGCCAATGTG-3′ and reverse 5′-ATGTAGCTCAGGGAAAAGTG-3′; N-cadherin forward 5′-GAAGGAGGTGGGGAGGAAGATA-3′ and reverse 5′-GGTGGTCTCTGACGAGGTAAACA-3′; ZEB1 forward 5′-AGTTTACCTTCCAGCAGCCCTAC-3′; and reverse 5′-AGCTCTTCTGCACTTGGTTGTG-3′; Slug forward 5′-AGACCCCCATGCCATTGAAG-3′ and reverse 5′-GGCCAGCCCAGAAAAAGTTG-3′; CD44 forward 5′-CACAACAACACAAATGGCTG-3′ and reverse 5′-CAATGCCTGATCCAGAAAAA-3′; IL-6 forward 5′-TCCAGAACAGATTTGAGAGTAGTG-3′ and reverse 5′-GCATTTGTGGTTGGGTCAGG-3′; HMGA2 forward 5′-CACTTCAGCCCAGGGACAAC-3′ and reverse 5′-GCCTCTTGGGCGTTTTTCTC-3′; β-actin forward 5′-GTGATCTCCTTCTGCATCCTGT-3′ and reverse 5′-CCACGAAACTACCTTCAACTCC-3′; qRT-PCR for let-7b was performed using commercially available TaqMan® MicroRNA Assays (#4373168, Applied Biosystems, Darmstadt, Germany) according to manufacturer’s protocol. The specificity of PCR was confirmed by melting curves and PCR product sequencing.

Protein extraction, quantification and immunoblotting were performed as our previously described [27]. The antibodies used in this study were: PBX3 (1:500, ab109173), N-cadherin (1:750, ab18203), ZEB1 (1:500, ab180905), Slug (1:750, ab180714), CD44 (1:750, ab51037), c-myc (1:500, ab32072), LIN28 (1:500, ab46020) and β-actin (1:2000, ab8226) all from Abcam (Cambridge, UK). HMGA2 (1:1000, no. 8179), STAT3 (1:1000, no. 9139), phospho-STAT3 (1:500, no. 9145), MEK1/2 (1:1500, no. 4694), phospho-MEK1/2 (1:1000, no. 4370), ERK1/2 (1:2000, no. 9194,) and phospho-ERK1/2 (1:1000, no. 4370) all from Cell Signaling Technologies (Danvers, MA, USA).

The invasive capabilities of glioma cells were determined as our previously reported [26].

Tissues were fixed in 4% paraformaldehyde for two 48 h and paraffin embedded in a regular way. Tissue sections (4 μm) were treated and stained with the following antibodies: PBX3 (1:75, ab109173), N-cadherin (1:100, ab18203), ZEB1 (1:150, ab180905), Slug (1:150, ab180714), CD44 (1:100, ab51037).

Chromatin-immunoprecipitation (ChIP) assays were carried out using a ChIP Assay Kit (#17408, Millipore) according to the manufacturer’s protocol. The chromatin fragments were immunoprecipitated with 2 μg of antibodies against either c-myc (ab32) or Jun (ab31419). The primer sequences used in this study were as follows: LIN28 (c-myc) forward 5′-GGGAGGGCCCATTCATTTC-3′ and reverse 5′-GGGTCCCCAAAGCAGATACA-3′; Wnt5a (c-myc) forward 5′-GTCGGGAAGTGGTCAAGGTT-3′ and reverse 5′-AAGTGCCAGAGACAGATGCT-3′; CyclinD1 (Jun) forward 5′-GTCCCAGGCAGAGGGGAC-3′ and reverse 5′-CGGCAATTTAACCGGGAGA-3′; β-globin (negative control) forward 5′-AGTGCCAGAGCCAAGGA-3′ and reverse 5′-CAGGGTGAGGTCTAAGTGATGACA-3′; and rRNA (internal control) forward 5′-ATTAGTCAGCGGAGGAAAAGAAAC-3′ and reverse 5′-TCGCCGTTACTGAGGGAATC-3′.

To determine whether let-7b directly binds to the PBX3 3′-UTR, dual luciferase reporter assays were conducted. Wild-type (WT) and mutated putative let-7b-binding sites were amplified and cloned into the XbaI site of a pGL3 control vector (Invitrogen). The following reporter assays were performed in a regular way as our previously described [26].

The collected supernatant was serially diluted, and levels of IL-6 of different groups were measured by enzyme-linked immunosorbent assay (ELISA; R&D Systems Inc., Minneapolis, MN).

Female Bagg albino (BALB)/c nude mice at 5 weeks were purchased from the Shanghai Experimental Animal Center of the Chinese Academy of Sciences and maintained in specific pathogen-free conditions for 1 week. To established intracranial GBMs, LV-siRNA-NC- or LV-siRNA-PBX3-transfected U87 cells (2.5 × 10⁵) were stereotactically injected (12 mice for each group). Three weeks later, all mice were killed by rapid decapitation and brains were extracted and stored in liquid nitrogen. All experimental procedures were in accordance with the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health and approved by the Nanjing Medical University Animal Experimental Ethics Committee.

Statistical analyses were performed using GraphPad Prism 5.0 (GraphPad Software, Inc., LaJolla, CA, USA). Data were presented as mean ± SD. The significance of differences between groups was tested by two-tailed Student’s t-test. Correlations between PBX3 and N-cadherin, ZEB1, Slug, CD44, let-7b, HMGA2, and IL-6 expressions were analyzed with Pearson’s correlation method. Values of p < 0.05 were considered statistically significant.

To evaluate the expression profiles of PBX3 between mesenchymal and proneural gliomas, mRNA data from Rembrandt, GSE4290 and GSE59612 were analyzed. Results showed that PBX3 expression was increased in mesenchymal gliomas compared with proneural gliomas (Fig. 1a-c). Using the expression data from the Rembrandt dataset, we performed GSEA to verify whether we could detect some hallmarks of cancer, including mesenchymal transition,in the patients with high-level of PBX3. Our results showed that hallmarks, including “Epithelial_Mesenchymal_Transition”, were significantly enriched in the patients with high-level of PBX3 (Fig. 1d). We chose four mesenchymal transition markers (N-cadherin, ZEB1, Slug, and CD44) from the gene list identified by GSEA, which are reported to be involved in glioma mesenchymal transition [6, 32‐35], and analyzed the relationship between these markers and PBX3 expression. Pearson correlation analyses using mRNA data from Rembrandt dataset were performed. Results showed significant and positive correlations between PBX3 and N-cadherin, ZEB1, Slug and CD44 (Additional file 1: Figure S1A-D). To further confirm our observations, we measured PBX3 and mesenchymal markers mRNA levels in 45 GBM samples by qRT-PCR and observed high expression levels of N-cadherin, ZEB1, Slug and CD44 in GBMs with high PBX3 expression (n = 23) than those with low PBX3 expression (Additional file 1: Figure S1E-H). Having demonstrated that PBX3 is upregulated in mesenchymal gliomas and correlated with mesenchymal markers, we wondered whether PBX3 is required for maintenance of mesenchymal phenotype of GBM cells. To achieve this goal, we ectopically expressed PBX3 using lentiviral vectors and knocked down PBX3 expression using lentiviral siRNAs in U87 and U251 cells (Fig. 1e and h). Results showed that overexpression of PBX3 significantly promoted a fibroblast-like morphology and enhanced the expression of mesenchymal markers (Fig. 1 f and g). By contrast, knockdown of PBX3 triggered a cobblestone-like morphology and decreased the expression of mesenchymal markers (Fig. 1 i and j). On the basis of the mesenchymal promoting role of PBX3 in glioma, we focused on migration and invasion assays. As expected, PBX3 overexpression promoted glioma migration and invasion, while PBX3 downregulation inhibited glioma migration and invasion (Additional file 2: Figure S2). Collectively, these results indicated that PBX3 is associated with mesenchymal transition in gliomas.

TGF-β is a master regulator of mesenchymal transition in GBM and induces the expression of mesenchymal genes, such as N-cadherin, ZEB1, Slug and CD44 [17, 36, 37]. Moreover, GSEA results showed that GBMs with high-level of PBX3 were enriched for the TGF-β signaling (Additional file 3: Figure S3). Therefore, we decided to explore the role of PBX3 in response to TGF-β. First, we examined the expression of PBX3 mRNA and protein upon exposure to TGF-β. The levels of PBX3 mRNA and protein in U87 and U251 cells were significantly upregulated during the 48-h post-incubation period (Fig. 2a and b). Next, we tested the effect of PBX3 downregulation on the TGF-β induced upregulation of N-cadherin, ZEB1, Slug and CD44. TGF-β induced upregulation of N-cadherin, ZEB1and CD44 were completely antagonized, whereas the accumulation of Slug was partially reversed by the knockdown of PBX3 (Fig. 2c). Finally, would-healing and transwell assays demonstrated that TGF-β-induced GBM cells migration and invasion were significantly decreased by PBX3 knockdown (Fig. 2d and e). In summary, these results demonstrated that PBX3 is required for the TGF-β-induced mesenchymal transition in GBM cells.

To confirm our hypothesis that PBX3 promotes glioma mesenchymal transition and then triggers migration and invasion via activating LIN28/let-7 axis, we first measured the expression of LIN28 and let-7 upon PBX3 overexpression. Results showed that overexpression of PBX3 increased LIN28 protein levels while decreased let-7 levels, with let-7b being the most downregulated one (Fig. 3a, Additional file 4: Figure S4 and Additional file 5: Figure S5A), suggesting that PBX3 positively regulates LIN28/let-7 axis. As let-7b represents the most significantly downregulated let-7 family microRNAs upon PBX3 overexpression, we selected let-7b for the following experiments. Next, we tested whether PBX3-driven GBM mesenchymal phenotype was dependent on LIN28/let-7b axis. LIN28 knockdown prevented PBX3-mediated increases in cell migration and invasion and reserved PBX3-induced mesenchymal phenotype (Fig. 3b and c and Additional file 5: Figure S5B and C). Consistently, overexpression of let-7b reversed PBX3-driven GBM cell migration, invasion and mesenchymal transition (Fig. 3d-f and Additional file 5: Figure S5D-F). To further confirm our hypothesis, we also measured LIN28 and let-7b expression under PBX3 knockdown conditions. Results showed that downregulation of PBX3 decreased LIN28 protein levels and increased let-7b expressions(Fig. 3g and Additional file 5: Figure S5G). As expected, overexpression of LIN28 or inhibition of let-7b expression reversed PBX3 downregulation-induced decreased cell migration, invasion and mesenchymal phenotype (Fig. 3h-l and Additional file 5: Figure S5H-L). Collectively, these results demonstrated that LIN28/let-7b axis is required for PBX3-driven mesenchymal transition and for increased migration and invasion.

Having demonstrated that PBX3 activates LIN28/let-7b pathway in GBM, we next investigated whether PBX3 regulates LIN28/let-7b axis via ERK1/2-dependent mechanism as our previously expected. To achieve this goal, we need to investigate whether MEK/ERK1/2 pathway mediated PBX3-driven GBM mesenchymal transition. First, we determined the changes in expression of phospho-MEK, MEK, phospho-ERK1/2 and ERK1/2 in GBM cells with altered expression of PBX3. As expected, phospho-MEK and phospho-ERK1/2 expression were increased after overexpression of PBX3 and decreased upon PBX3 downregulation (Fig. 4a and d). These data clearly demonstrated that PBX3 could activate MEK/ERK1/2 pathway in GBM cells. Next, we inhibited MEK/ERK1/2 activation in PBX3-overexpressed cells by U0126 and activated MEK/ERK1/2 pathway in PBX3-downregulated cells by PMA to determine whether PBX3-mediated GBM cell migration, invasion and mesenchymal transition were dependent on MEK/ERK1/2 pathway. As shown in Fig. 4a, the upregulation of phospho-MEK and phospho-ERK1/2 expression by PBX3 overexpression were significantly reduced by U0126 incubation. In the meantime, the promotive effects of PBX3 on cell migration, invasion and mesenchymal transition were markedly reversed by U0126 treatment (Fig. 4a-c). Consistently, the decreased expression of phospho-MEK and phospho-ERK1/2 expression by PBX3 downregulation were restored by PMA treatment. Meanwhile, the inhibitory effects after PBX3 downregulation on cell migration, invasion and mesenchymal transition were reversed by PMA incubation (Fig. 4d-f). These results confirmed the hypothesis that PBX3 promotes GBM migration, invasion and mesenchymal transition via activation of MEK/ERK1/2 pathway.

To demonstrate that PBX3 regulates LIN28/let-7b axis via ERK1/2-dependent mechanism. We inhibited ERK1/2 signaling in PBX3-overexpressing cells and activated ERK1/2 signaling in PBX3-knockdown cells. As shown in Fig. 5a and b, ERK1/2 inhibition reversed PBX3 overexpression induced LIN28 upregulation and let-7b downregulation, while ERK1/2 activation restored PBX3 depletion induced LIN28 downregulation and let-7b upregulation. Thus, these data suggest that PBX3 promotes LIN28/let-7b axis via ERK1/2-dependent mechanism. Previous studies have demonstrated that ERK1/2 regulates LIN28/let-7b axis via c-myc-dependent way and c-myc regulates LIN28 expression by directly binding to its promoter [30, 38]. As c-myc plays central roles in GBM progression, including mesenchymal transition [39], we proposed that ERK1/2 signaling promotes LIN28/let-7b axis in GBM cells via c-myc-dependent pathway. To test whether the hypothesis is true and whether the transcriptional factor c-myc might mediated PBX3/ERK1/2-driven mesenchymal transition, we overexpressed c-myc in ERK1/2-inhibited cells and knocked down c-myc in ERK1/2-activated cells. Results showed that LIN28 downregulation and let-7b upregulation induced by ERK1/2 inhibition were prevented by c-myc overexpression (Fig. 5c). Consistently, LIN28 upregulation and let-7b downregulation induced by ERK1/2 activation were reversed by c-myc knockdown (Fig. 5d). Furthermore, ChIP assays confirmed previous studies that c-myc can directly bind to the promoter of LIN28 and this biding was depended on ERK1/2 signaling (Fig. 5e-g). Finally, we found that silencing c-myc can reduced PBX3 overexpression induced increased migration, invasion and mesenchymal transition, while overexpression of c-myc can restore PBX3 inhibition induced decreased migration, invasion and mesenchymal transition (Additional file 6: Figure S6). Collectively, our finding demonstrated that PBX3 activates LIN28/let-7b axis in GBM cells via ERK1/2-c-myc-dependent mechanism.

Previous studies have demonstrated that PBX3 can be negatively regulated by various miRNAs, including let-7 family members (miR-98, let-7c, and let-7d) [19, 20, 26]. However, whether PBX3 can be regulated by let-7b in GBM remains undetermined. To test whether PBX3 is a direct target of let-7b, luciferase report assays were performed. Fig. 6a shows the predicted interaction between let-7b and the target sites. Relative luciferase activity was significantly inhibited by let-7b when the PBX3 plasmid containing the wild 3′-UTR and mut1 3′-UTR were present. However, mutations in binding site 2 (mut2) or both the binding sites were mutated abrogated the suppressive effect of let-7b (Fig. 6b). Moreover, western blot analysis showed that let-7d suppressed PBX3 and the downstream targets MEK/ERK1/2, c-myc and LIN28. In contrast, suppression of let-7b enhanced PBX3 and its downstream signaling (Fig. 6c). Taken together, these results suggest that PBX3 is a bona fide target of let-7b and there exists a positive feedback loop between PBX3 and let-7b. Moreover, other targets of let-7 family, which are mesenchymal transition relevant factors, such as HMGA2 and IL-6/STAT3, were also evaluated in our present study. As shown in Fig. 6d, let-7b can also inhibit HMGA2 and IL-6/STAT3 pathways. As PBX3 regulates LIN28/let-7b axis, we wondered whether PBX3 regulates let-7b targets in glioma cells. Therefore, we tested the effect of PBX3 silencing on the expression of let-7b targets. As shown in Fig. 6e, PBX3 silencing decreased the mRNA levels of HMGA2 and IL-6 and protein ratios of p-STAT3/STAT3 in both U87 and U251 cells. Indeed, our GSEA results showed that hallmark of “IL-6_JAK_STAT3 signaling” is significantly enriched in the patients with high-level of PBX3 (Additional file 7: Figure S7).

We further measured let-7b, HMGA2 and IL-6 mRNA levels in 45 GBM samples and Pearson’s correlation analyses were performed. Results showed that let-7b inversely correlated with PBX3, HMGA2 and IL-6 expression (Fig. 6f), supporting that PBX3, HMGA2 and IL-6 are targets of let-7b. Until now, we established a positive feedback loop between PBX3 and let-7b, which involves MEK/ERK1/2, c-myc and LIN28, and ultimately affects HMGA2 and IL-6/STAT3 pathways to promotes GBM mesenchymal transition, migration and invasion (Fig. 6g).

To further confirm the role of PBX3 in GBM invasion and assess the therapeutic potential of PBX3 depletion in vivo, LV-siRNA-NC- and LV-siRNA-PBX3-transfected U87 cells were intracranially injected into nude mice. As shown in Fig. 7a, downregulation of PBX3 consistently led to decreased tumor invasion compared with the control group. In addition, PBX3 depletion reduced tumor growth and resulted in the formation of significantly smaller tumors when compared with the control group (Fig. 7b and c). These results demonstrated that targeting PBX3 inhibits GBM invasion and growth in vivo. Next, we performed PBX3, N-cadherin, ZEB1, Slug and CD44 staining to assess the effects of PBX3 on GBM mesenchymal transition. Results showed that inhibition of PBX3 suppressed the expression of N-cadherin, ZEB1, Slug and CD44 (Fig. 7d), indicating that downregulation of PBX3 inhibits GBM mesenchymal transition in vivo. Finally, we investigated whether MEK/ERK1/2/c-myc/LIN28/let-7b pathway was suppressed upon PBX3 knockdown. Our western blot and qRT-PCR results demonstrated that PBX3 knockdown inhibited p-MEK, p-ERK1/2, c-myc, LIN28, HMGA2, IL-6 and p-STAT3 protein levels and increased let-7b expression (Fig. 7e and f). Taken together, these data support our in vitro findings and reinforce the hypothesis that PBX3 has a critical role in the maintenance of a mesenchymal and invasive phenotype in human GBM.