Identification of the FOXM1D and PKM2 Interaction
In late stage colorectal cancer, FOXM1D, a new isoform of FOXM1, promotes tumor metastasis by binding to and further activating ROCKs [28]. In our study, we found that anti-Flag antibody immunoprecipitated more proteins, including PKM2 rather than ROCK2, in the Co-IP assay and subsequent mass spectrometric analysis of HeLa cells overexpressing Flag-FOXM1D (Fig. 1a, and Additional file 4: Supplementary Data). Next, we confirmed the interaction between PKM2 and FOXM1D by mutual Co-IP analysis using anti-Flag and anti-HA antibodies in 293T cells transfected with HA-PKM2 and Flag-FOXM1D (Fig. 1b). The result of GST-pulldown assay also demonstrated this interaction. GST-FOXM1D pulled down HA-PKM2 expressed in 293T cells, which was determined by both anti-HA and anti-PKM2 antibodies (Fig. 1c).
To explore the precise binding region between FOXM1D and PKM2, we designed truncated mutants based on the structural domain of PKM2 (Fig. 1d) [33], and further expressed them with HA tag in 293T cells (Fig. 1e). The result of a Co-IP assay using anti-HA antibody demonstrated that PKM2 mutants A, B, and C, as well as intact PKM2 could capture the co-expressed Flag-FOXM1D (Fig. 1e), indicating that the domains A1 and/or B of PKM2 are most likely the binding site for FOXM1D. We further identified the precise binding region in PKM2 using the GST-pulldown assay. As shown in Fig. 1f, GST-FOXM1D bound to the B but not the A1 domain of PKM2, as expressed in 293T cells. In addition, we also designed truncated mutants of FOXM1D that were based on the coding exons due to absence of the FOXM1 crystal structure (Fig. 1g) [28]. We expressed these with Flag tag in 293T cells, and using the Co-IP assay with the anti-Flag antibody we mapped the binding mutants of FOXM1D to the co-expressed PKM2 as mutant A, B, and D (Fig. 1h). Our findings suggested that exon II and III of FOXM1D encoded the binding region for PKM2, therefore clearly demonstrated the interaction between FOXM1D and PKM2.
FOXM1D Promotes Tumor Aerobic Glycolysis
Given that exons II and III of FOXM1D are shared by FOXM1A/B/C/D [28], we thus detected whether or not FOXM1A/B/C could physically bind to PKM2. We co-transfected 293T cells with plasmid expressing HA-PKM2, Flag-FOXM1A/B/C/D, or empty vector, then performed a Co-IP assay using anti-Flag antibody. As we expected, all of the four Flag-tagged FOXM1 isoforms bound to HA-PKM2 (Additional file 1: Fig. S1a) and exhibited little effect on the transcription of PKM2 (Additional file 1: Fig. S1b). Because PKM2 catalyzes the rate-limiting step of aerobic glycolysis, it plays a crucial role in the regulation of the Warburg effect [4, 34]. As FOXM1A/B/C/D all interact with PKM2, we sought to detect the effect of ectopic expression of FOXM1A/B/C/D on glycolysis by measuring the ECAR and OCR in HeLa cells. FOXM1D induced an appreciable increase in ECAR, which reflects overall glycolytic flux (Fig. 2a), along with a decrease in OCR, which is an indicator of mitochondrial respiration (Fig. 2b). Meanwhile FOXM1A/B/C failed to induce an obvious alteration in either ECAR or OCR when compared to the vector control (Fig. 2a-b). Considering that LoVo and SW-480 cells display the highest and lowest expression levels of FOXM1D, respectively, in the six examined colorectal cell lines [28], we further compare the effect of FOXM1D on glycolysis between them. We measured the ECAR and OCR in SW-480 cells that over-expressed FOXM1D and in LoVo cells where FOXM1D was knocked down by specific siRNA. We found that ectopic FOXM1D resulted in a notable elevation in the ECAR, whereas insufficient FOXM1D resulted in an ECAR reduction (Fig. 2c). In contrast, ectopic FOXM1D yielded a significant reduction in the OCR, while insufficient FOXM1D resulted in its elevation (Fig. 2d). Together, these results demonstrated that FOXM1D could effectively promote the Warburg effect.
FOXM1D Inhibits PK Activity of PKM2 Tetramer by Assembling a Heterooctamer
It has been reported that PK activity of PKM2 may be regulated by its post-translation modification (PTM). Phosphorylation at Y105 may inhibit the formation of the active PKM2 tetrameter [35], acetylation at K305 may increase PKM2 degradation [13], and acetylation at K433 may interfere with FBP binding [7]. Considering that FOXM1A/D, but not FOXM1B/C, significantly regulate glycolysis without changing the PKM2 expression level as described previously, we first detected the PTM status of PKM2 by immunoblotting in HeLa cells with ectopic FOXM1A/D expression. The results showed that the levels of phospho-Y105 ,acetyl-K305 and -K433 unchanged after ectopic FOXM1A/D expression (Fig. 3a), indicating that another mechanism was involved in the regulation of glycolysis. In addition, we found that ectopic FOXM1A/D expression failed to change the level of phospho-S37 of PKM2, which is associated with nuclear translocation [15] (Fig. 3a).
The PK activity of PKM2 is also allosterically activated by FBP, whose binding to PKM2 causes the switch from a less active monomeric/dimeric form to an active tetrameric from [18, 36, 37]. Next, we determined the monomer: dimer: tetramer equilibrium of PKM2 using cross-linking experiments with glutaraldehyde in HeLa cells with ectopic FOXM1A/D expression. Following glutaraldehyde cross-linking, the proportion of the tetramer in FOXM1D-overexpressing cells was notably reduced, while the monomer increased when compared with FOXM1A-overexpressing and control cells (Fig. 3b). Interestingly, we observed that an obvious band ran close to the location of stacking gel (arrow in Fig. 3b). To understand whether or not this macromolecular band was associated with the complex of PKM2 and FOXM1D, we induced PKM2 insufficiency (Fig. 5j) by using two specific siRNAs in HeLa cells with ectopic FOXM1A/D (up panel in Fig. 3c). Immunoblotting assays using anti-FOXM1 antibody revealed the presence of another notable band, other than a normal FOXM1 band, that was also located close to the stacking gel in the cell extracts treated with glutaraldehyde. This band was present in both the scramble siRNA treated HeLa cells (left bottom panel in Fig. 3c). However, PKM2 insufficiency provoked this band much weaker only in FOXM1D- but not FOXM1A-overexpressing cells compared to scramble control, thus resulting in a dissociated new band close to the normal FOXM1 (middle and right bottom panels in Fig. 3c). These results suggested that FOXM1D, but not FOXM1A, could form a large heterogeneous polymer with PKM2 to regulate glycolysis.
In order to determine the size and composition of the heterogeneous polymer, we used gel chromatography to separate FOXM1D and PKM2 in the extract of FOXM1A/D-overexpressing HeLa cells. The result of the immunoblotting assay for each fraction demonstrated that FOXM1D, PKM2, and FBP appeared simultaneously in fractions 28-35, whose size was a bit greater than the 660 kD marker in the fraction between 35 and 36 (Fig. 3d). The four B domains of the PKM2 tetramer were located at the outside corner [38], and the binding site of FOXM1D on PKM2 was located in the B domain, both of which may favor the binding of four FOXM1D molecules to the PKM2 tetramer without steric hindrance. Therefore, we hypothesized that this heterogeneous polymer was most likely a heterogeneous octamer, i.e., a complex of four PKM2 (~60 kD), four FBP (~40 kD), and four FOXM1D (apparent ~130 kD). In addition, we observed that ectopic FOXM1D decreased the proportion of PKM2 tetramer accompanied by the reduced FBP (fractions 43-47), and increased those of the PKM2 dimer (fractions 48-52) and monomer (fractions 53-54) when compared with the vector control (Fig. 3d). It is notable that FOXM1A failed to form such a heterogeneous octamer with PKM2 in fractions 28-35 (Additional file 1: Fig. S2), indicating that this physical interaction may be unstable. In addition, ectopic FOXM1A seemed to result in a slight increase in the proportion of PKM2 tetramer, accompanied by an increase in FBP (fractions 43-47) and PKM2 dimer (fractions 48-52) and a reduction of PKM2 monomer (fractions 53-54) (Additional file 1: Fig. S2).
Given that the PKM2 tetramer displayed high PK activity, we next measured the enzyme activity of each fraction. The results showed that the heterogeneous octamer, which existed in fractions 28-33 in the extracts of FOXM1D-overexpressing HeLa cell, showed undetectable PK activity (Fig. 3e). The PK activity of fractions 43-52, which contained PKM2 tetramer and dimer, decreased in the extracts of FOXM1D-overexpressing compared with FOXM1A-overexpressing and control HeLa cells (Fig. 3e). Interestingly, we found that fractions 38-42 from the extracts of the FOXM1A-overexpressing and control, but not FOXM1D-overexpressing HeLa cell displayed notable PK activity (Fig. 3e). This PK activity may result from the complex of PKM2 tetramer and other proteins, as FBP existed in these fractions and FOXM1D may have disrupted this complex (Fig. 3d).
To further confirm the physical interaction between FOXM1A/D and PKM2, we expressed and purified recombinant Flag-FOXM1A/D and Flag-FBP, and purchased PKM2. The results of SDA-PAGE (Additional file 1: Fig. S3a) and gel chromatography (Additional file 1: Fig. S3b-e) demonstrated the high purity and the principal existence form of the monomer. In the presence of excess FBP, the tetramer is the predominant form of PKM2, with trivial amount of monomer and without the dimer forms (Additional file 1: Fig. S3f). However, in the absence of FBP but the presence of FOXM1A or FOXM1D, PKM2 formed a heterodimer with FOXM1A (Additional file 1: Fig. S3g) or FOXM1D (Fig. 3f). Furthermore, in the presence of both FBP and FOXM1A, PKM2 preferred to form a homologous tetramer exclusive of binding to FOXM1A, although we did observe a tiny heterooctamer curve (Additional file 1: Fig. S3h). In contrast, in the presence of FBP and FOXM1D, all of the PKM2 tetramer further bound to the four molecules of FOXM1D, leading to the assembly of a heterooctamer (Fig. 3g). This finding indicated that FOXM1D and FOXM1A displayed different binding activity to PKM2 tetramer. In the functional test, supplement of FBP alone, but not FOXM1A or FOXM1D alone, elevated the PK activity of PKM2 remarkably (Fig. 3h). Addition of FOXM1A weakly elevated the above FBP-induced PK activity of PKM2 but without statistical significance, however, addition of FOXM1D significantly reduced this PK activity in a dose-dependent manner (Fig. 3h). Moreover, when providing excessive FOXM1D (1.33-fold of PKM2 in molar ratio), about half (49%) of the PK activity of PKM2 was abolished. Therefore, these findings reveal that FOXM1D binds to the PKM2 tetramer to assemble a heterogeneous octamer, thereby reducing PK activity by about a half.
FOXM1D Promotes Tumor Angiogenesis
It has been recognized that the Warburg effect is closely associated with tumor angiogenesis [39-41], however, the underlying mechanisms remains elusive. Therefore, we detected the effect of ectopic FOXM1D expression on tumor angiogenesis. We collected the supernatant of FOXM1A/B/C/D-overexpressed or control HeLa cells, then used those to treat HUVECs before detecting the ability for tube formation. The results showed that ectopic expression of only FOXM1D, but not of FOXM1A/B/C and control significantly enhanced angiogenic sprouting and tube formation in HUVECs (Fig. 4a-c). This finding indicating that the supernatant of FOXM1D-overexpressing HeLa cells may contain pro-angiogenesis factor(s).
To verify this angiogenic effect in vivo, we qualified the tumor blood vessels by microangiography [42]. Considering that FOXM1B accelerates tumor growth more potently than FOXM1D [28], we selected tumor tissues of similar sizes to compare their microvessels. These tumor tissues were collected from previously generated tumor-bearing mice that were implanted with vector control, FOXM1B-, or FOXM1D-overexpressing HeLa cells [28]. The images of tumor blood vessels show that ectopic FOXM1D, but not FOXM1B visibly enhanced angiogenesis compared to vector control (Fig. 4d). Furthermore, the quantitative results revealed that, although the tumor sizes were comparable in each group, the density, branches, nodes, maximal diameter, and average diameter of the microvessels in FOXM1D-, but not FOXM1B-overespressing tumors were significantly greater than those of vector control tumors (Fig. 4e). Together, these results suggested that FOXM1D may strongly promote tumor neovascularization, most likely through releasing the pro-angiogenesis factor, and more importantly, that the ample number of blood vessels induced by ectopic FOXM1D expression failed to accelerate tumor growth synchronously.
FOXM1D Upregulates VEGFA Expression Mediated by PKM2 and NF-κB
VEGF/VEGF-receptor signaling has been well-established as a key mediator of tumor angiogenesis [43]. Therefore, to understand the molecular mechanism of the FOXM1D angiogenic effect and its relationship with PKM2-mediated glycolysis, we first detected the expression level of VEGFA in the cell lysate of FOXM1D-overexpressing HeLa cells. Unexpectedly, ectopic FOXM1D even reduced the VEGFA level compared to vector control (Fig. 5a). However, we observed that VEGFA levels were significantly increased in the supernatants of FOXM1A/D- but not of FOXM1B/C-overexpressing HeLa cells (Fig. 5b). This result indicated that, in FOXM1D-overexpressed HeLa cells, VEGFA was most likely secreted extracellularly. Although FOXM1B has been reported to increase VEGF transcription in glioma cells [44], we observed that only FOXM1D, but not FOXM1A/B/C caused a marked upregulation in VEGFA transcription as detected by quantitative real-time PCR (qRT-PCR) (Fig. 5c).
Considering the facts that: 1) nuclear PKM2-controlled HIF-1α and NF-κB regulate VEGFA transcription in hypoxic conditions [22]; 2) nuclear PKM2 promotes angiogenesis by directly interacting with the NF-κB p65 subunit [23]; 3) FOXM1 directly interacts with the NF-κB subunit p65 [45]; and 4) FOXM1D, PKM2, and p65/p50 co-existed in fractions 34-36 in the gel chromatography experiment (Fig. 3d), we therefore hypothesized that FOXM1D may promote VEGFA transcription by interacting with PKM2 and NF-κB. Although we previously failed to identify any NF-κB subunits as FOXM1D binding proteins by Co-IP and LC/MS analysis (Additional file 4: Supplementary Data), we still conducted the Co-IP assay in order to detect the interaction between FOXM1A/B/C/D and NF-κB. In the vector control and Flag-FOXM1A/B/C/D-overexpressing HeLa cells, we further transfected HA-p65 or -p50 expressing plasmid. The results of the Co-IP assay revealed that anti-Flag antibody captured HA-tagged p65 and endogenous PKM2 in all of the FOXM1A/B/C/D-overexpressing cells (Fig. 5d-e), while it could capture HA-tagged p50 only in Flag-FOXM1A-, and especially Flag-FOXM1D-overexpressing cells (Fig. 5e). In addition, anti-HA antibody captured Flag-tagged FOXM1A/B/C/D in HA-p65-overexpressing cells (Fig. 5d) and endogenous PKM2 in either HA-p65- or HA-p50-overexpressing cells (Fig. 5d-e), however, anti-HA antibody bound only slightly to Flag-FOXM1A, but bound strongly to Flag-FOXM1D (Fig. 5e). Furthermore, both nuclear subunits p65 and p50 increased simultaneously in HeLa cells with ectopic expression of FOXM1A/B/C/D (Fig. 5f), resulting in the reduced levels of p65 and p50 in cytoplasm but not in total cell lysate (Additional file 1: Fig. S4). These results suggest that FOXM1D may up-regulate VEGFA transcription by promoting NF-κB nuclear translocation.
Next, we detected the effect of PKM2, the critical regulator in glycolysis, on FOXM1D-induced VEGFA transcription. Two specific siRNAs against PKM2 reduced PKM2 mRNA levels. However, si-PKM2 #1 induced the restoration of PKM1 mRNA, while si-PKM2 #2 downregulated PKM1 transcription (Fig. 5g). We further observed that the PKM2 insufficiency induced by two siRNAs almost completely abrogated FOXM1D-induced VEGFA transcription (Fig. 5h), secretion (Fig. 5i), and translation (Fig. 5j) in FOXM1D-overexpressed cells. Therefore, PKM2 may effectively regulate VEGFA expression independent of PKM1.
To confirm that PKM2 and NF-κB may regulate VEGFA transcription as transcriptional factors, we conducted a ChIP assay in FOXM1D-overexpressing HeLa cells in order to identify whether or not PKM2 and NF-κB bind directly to the same VEGFA promoter region. The specific antibody against PKM2, p50, or p65, but not isotype IgG enriched this promoter fragment remarkably (Fig. 5k-m), which was amplified by qRT-PCR with the same primer pair (Fig. 5k). We further knocked-down the expression of PKM2 and p65 to verify the binding specificity. Insufficiency of PKM2 significantly impaired the capability of enriching the VEGFA promoter fragment not only by PKM2 antibody but also by p65 or p50 antibody, and insufficiency of NF-κB subunit p65 also displayed the similar results (Additional file 1: Fig. S5). Together, these results revealed that FOXM1D binds to PKM2, the NF-κB subunits p65 and p50, further promotes their nuclear translocation, thus leading to increased VEGFA expression as regulated by both PKM2 and NF-κB.
The Binding of FOXM1D to Importin 4 Promotes the Nuclear Translocation of PKM2 and NF-κB
FOXM1D locates predominantly in the cytoplasm, while PKM2 and NF-κB translocate into nucleus to regulate VEGFA transcription upon ectopic FOXM1D. Therefore, we further explored how FOXM1D promotes nuclear translocation of PKM2 and NF-κB. The result of Co-IP analysis demonstrated the interaction between FOXM1A/B/CD and PKM2 (Additional file 1: Fig. S1a). Next, we employed ICC analysis to verify their localization (Fig. 6a). As reported previously [28], FOXM1B/C locate predominantly in the nucleus while FOXM1A/D locate predominantly in the cytoplasm (Fig. 6a). Ectopic expression of all four FOXM1 isoforms did not alter the expression level of PKM2, while ectopic expression of FOXM1B/C/D, but not FOXM1A, promoted the nuclear translocation of PKM2 (Fig. 6a-b). Due to the higher nuclear PKM2 translocation, we thus observed a slight reduction in the level of PKM2 in the cytoplasm of FOXM1B/C/D- but not FOXM1A- overexpressing cells (Fig. 6b). In addition, we also found that ectopic expression of the four FOXM1 isoforms could effectively promote the nuclear translocation of endogenous NF-κB by ICC assay (Additional file 1: Fig. S4) and Western blotting (Fig. 5f).
Next, we interrogated the mechanism by which FOXM1D promotes PKM2 and NF-κB translocation into nucleus. Although importin-α/β are responsible for the import of many proteins from the cytoplasm to the nucleus, a host of importins related to β-importins have been described to interact with the nuclear localization signal of specific proteins and import them into the nucleus independently of importin-α. For example, importin-4 transports vitamin D receptor [46] as well as transition protein 2 [47], and importin-7 transports HIV-1 intracellular reverse transcription complexes [48] into the nucleus. Interestingly, we observed that both importin 4 and 7 were captured by FOXM1D with anti-Flag antibody (Additional file 4: Supplementary Data). Using Co-IP and ICC assays, we demonstrated that importin 4 bound to FOXM1A/D but not to FOXM1B/C (Fig. 6c-d), indicating that FOXM1B/C employed a different approach to inducing nuclear translocation of PKM2 and NF-κB from FOXM1A/D. In addition, results of the gel chromatography experiment indicated that importin 4 universally located in fractions of 34-40 (Fig. 3d), suggesting that it binds to a variety of proteins including PKM2 and NF-κB. Furthermore, insufficiency of importin 4 induced by specific siRNAs abrogated the effect of ectopic FOXM1D to increase VEGFA transcription, which most likely resulted from the subsequently reduced nuclear levels of PKM2 and NF-κB (Fig. 6e-f). We also found that the levels of VEGFA in total lysate (Fig. 6e) decreased accordingly. These results suggest that FOXM1D promotes the nuclear translocation of PKM2 and that NF-κB may be mediated by importin 4, thus leading to the upregulation of VEGFA.
FOXM1D Promotes VEGFA Release by Interacting with VPS11
Our study found that ectopic expression of FOXM1D remarkably increased VEGFA transcription (Fig. 5c), however, the level of VEGFA protein decreased conversely in the cell lysate (Fig. 5a) and only slightly increased in the supernatant (Fig. 5b). The above evidence did not seem enough to support the potent pro-angiogenic effect of FOXM1D (Fig. 4). We therefore interrogated the underlying mechanism for FOXM1D-dependent VEGFA extracellular release. FOXM1 has been demonstrated to interact with HSP70, a critical exosome biomarker [49]. Using Co-IP and LC/MS analysis, we previously identified that FOXM1D also interacted with VPS11 (Additional file 4: Supplementary Data), an exosome component [50]. It has been shown that VPS11 plays an important role in vesicle trafficking and fusion of lysosomes and endosomes [51]. Considering the importance of the exosome in angiogenesis [52, 53], we therefore hypothesized that FOXM1D might facilitate VEGFA release via the exosome by interacting with VPS11. Using a Co-IP assay, we first verified that anti-Flag antibody could capture the endogenous VPS11 in all of the Flag-FOXM1A/B/C/D-overexpressing HeLa cell extracts (Fig. 7a), indicating that FOXM1A/B/C/D interacted physically with VPS11 in the cell extracts. We also employed an ICC assay to detect the subcellular co-localization of FOXM1A/B/C/D and VPS11. Due to the distinct subcellular location of VPS11 and FOXM1B/C, we failed to observe their virtual interaction (Fig. 7b). In contrast, we did observe sporadic formation of FOXM1A, while formation of FOXM1D exhibited the massive dot-like overlap with VPS11 (Fig. 7b). This finding suggested the strong involvement of FOXM1D in the formation of the exosome.
To understand the function of ectopic FOXM1D on regulating exosome and VEGFA release, we further collected the exosomes from the supernatant of vector control or FOXM1A/B/C/D-overexpressing HeLa cells, in which the cell number and supernatant volumes were same. Ectopic FOXM1A/B/C slightly elevated, while FOXM1D strongly elevated the number of exosomes in the supernatant, as determined by HSP70 and VEGFA levels (Fig. 7c). Further, VPS11 insufficiency induced by specific siRNAs (upper panel in Fig. 7d-e) failed to alter the effect of ectopic FOXM1D on increasing VEGFA transcription (bottom panel in Fig. 7d). However, it is possible that VPS11 insufficiency may result in exosome dysfunction, thus reducing the level of VEGFA in the exosome (Fig. 7f) and supernatant (Fig. 7g), while elevating VEGFA levels in total cell lysate (Fig. 7e). Therefore, these results suggested that FOXM1D interaction with VPS11 promoted exosome and VEGFA release.