A cytoskeleton regulator AVIL drives tumorigenesis in glioblastoma

Glioblastoma is a deadly cancer, with no effective therapies. Better understanding and identification of selective targets are urgently needed. We found that advillin (AVIL) is overexpressed in all the glioblastomas we tested including glioblastoma stem/initiating cells, but hardly detectable in non-neoplastic astrocytes, neural stem cells or normal brain. Glioma patients with increased AVIL expression have a worse prognosis. Silencing AVIL nearly eradicated glioblastoma cells in culture, and dramatically inhibited in vivo xenografts in mice, but had no effect on normal control cells. Conversely, overexpressing AVIL promoted cell proliferation and migration, enabled fibroblasts to escape contact inhibition, and transformed immortalized astrocytes, supporting AVIL being a bona fide oncogene. We provide evidence that the tumorigenic effect of AVIL is partly mediated by FOXM1, which regulates LIN28B, whose expression also correlates with clinical prognosis. AVIL regulates the cytoskeleton through modulating F-actin, while mutants disrupting F-actin binding are defective in its tumorigenic capabilities. Genes that modulate the cytoskeleton have been associated with increased cell proliferation and migration. Here, the authors show that AVIL, an actin regulatory protein, is overexpressed in glioblastomas and mediates oncogenic effects through regulation of FOXM1 stability and LIN28B expression.

G lioblastoma (GBM) is the most common primary brain tumor and among the deadliest of human cancers. Despite advances in surgery, radiation and chemotherapy, survival of patients affected by GBM remains dismal (~15 months after diagnosis) [1][2][3] . Clearly, better treatment options, and identification of selective therapeutic targets are urgently needed.
Oncogene addiction describes a phenomenon according to which tumor cells become reliant on the activity of a particular oncogene and die once this activity is inhibited 4,5 . Many of the targeted cancer therapies exploit this concept 6 . It is perhaps best exemplified by the successful use of imatinib in the therapy of chronic myelogenous leukemia (CML) 7 . In CML, the major driver of tumorigenesis is the BCR-ABL fusion oncogene; imatinib inhibits the constitutively active BCR-ABL protein kinase, to which leukemic cells become addicted. Other successful examples include trastuzumab targeting ERBB2 addiction 8 , and vemurafenib targeting BRAF addiction 9 . The challenge is to find such key oncogenes. Even though large sets of genome and transcriptome data are available to facilitate the identification of driver mutations in cancer, true signals are often buried in a large number of passenger events. In contrast to adult cancers, pediatric tumors tend to have fewer point mutations and structural changes. While studying a pediatric cancer, rhabdomyosarcoma, we discovered a gene fusion, which results in the juxtaposition of a house-keeping gene next to the AVIL gene. Suspecting that other tumors may also dysregulate AVIL expression, we examined AVIL in adult cancers and found its critical role in the tumorigenesis of GBM. We believe that the same approach can be applied to the discovery of other oncogenes.
The cytoskeleton of the cells plays important roles in addition to maintain the cell size and shape. Many critical processes including cell proliferation, migration, and even transcriptional regulations have been connected to the cytoskeleton 10 . Various genes that modulate cytoskeleton have been associated with enhanced infiltrative and proliferative capacity 11 . For instance, in GBM, CTTN, an actin nucleating factor is overexpressed, and this overexpression is associated with an enhanced infiltrative capacity, and poor prognosis 12,13 . Here, we report an oncogene, AVIL, which encodes a protein that regulates F-actin dynamics and cytoskeleton. We found that AVIL is overexpressed in GBM cells including GBM stem cells, and that AVIL overexpression is crucial for GBM proliferation and migration. Mechanistically, AVIL functions upstream of FOXM1. FOXM1 is a member of FOX family. While it is silenced in differentiated cells, it is overexpressed in a number of solid tumors including GBMs 14 . It has been also reported to mediated critical processes of tumorigenesis, such as tumor invasion, angiogenesis, and metastasis [14][15][16][17][18] . On the other hand, let-7 family of microRNAs functions as tumor suppressors and inhibits glioma malignancy 19 . We showed multiple lines of evidence supporting that AVIL regulates FOXM1 stability, which in turn regulates LIN28B/let-7. These findings support the critical role of cytoskeleton dynamics in GBMs, and connect cytoskeleton regulation to the stability of FOXM1 and let-7 expression.

Results
AVIL is frequently upregulated in glioblastomas. Previously, we identified a gene fusion in alveolar rhabdomyosarcoma, a pediatric cancer 20 . We noticed that even though PAX3-FOXO1 is the most well-known fusion in this type of rhabdomyosarcoma, MARS-AVIL has the highest number of reads in the RNA-Seq data ( Supplementary Fig. 1a). MARS encodes methionyl-tRNAsynthetase. It is a house-keeping gene, expressed in all examined tissues (Supplementary Fig. 1b). AVIL is known as a member of the villin/gelsolin family, that regulates actin filament reorganization 21 . The expression of AVIL is more restricted, being low or undetectable in most tissues ( Supplementary  Fig. 1c). As with many gene fusions, including IGH-MYC, and TMPRSS2-ERG, the expression of a proto-oncogene may be misregulated by subjecting itself to the control elements of its fusion partner. We suspected that fusion to MARS in rhabdomyosarcoma is one mechanism to misregulate AVIL gene expression, and that AVIL may be misregulated by other mechanisms in other cancers. We found that the AVIL locus is amplified in 15-18% of glioblastoma cases in The Cancer Genome Atlas (TCGA) studies 22,23 via cBioPortal analysis (Fig. 1a). We confirmed such a copy number gain by FISH analyses, using a probe covering the AVIL locus (Fig. 1b). The AVIL locus is amplified in two glioblastoma cell lines, SF767 and A172, but not in three other glioblastoma lines, U87, U251, T98G, or in an immortalized astrocyte culture.
12q13-15 is frequently amplified in multiple tumor types including glioma 24 . AVIL is located on 12q14, 45 kb away from CDK4 toward centromere, and around 10 MB from MDM2 toward telomere. Given the importance of CDK4 and MDM2 in tumorigenesis, it is possible that AVIL amplification is only a byproduct of a larger fragment amplification. However, AVIL expression seems to be upregulated through additional mechanisms: at the protein level AVIL was not detected in the normal astrocytes; in contrast, all of the tumor cell lines tested had higher AVIL protein expression, including the ones that do not have AVIL locus amplification (Fig. 1c). This difference was not obvious at the RNA level ( Supplementary Fig. 2), suggesting that some post-transcriptional mechanism may be at play.
The overexpression of AVIL protein in GBMs was further confirmed by immunohistochemistry assays. Staining in normal human cerebral cortical gray matter was low, and largely restricted to a subpopulation of pyramidal neurons. Normal white matter demonstrated minimal immunoreactivity. In contrast, all the glioblastomas we tested had much stronger signals (Fig. 1d, and Supplementary Fig. 3).
We then examined the REMBRANDT database, which has microarray data for 28 non-tumor brain tissues, 148 astrocytomas (WHO grade II or III), and 228 GBM (WHO grade IV) cases 25 . Two different microarray probes showed that AVIL expression correlates with tumor grade, with the highest levels in GBMs (Fig. 1e). In our own collection of eight non-tumor brain tissues, and 36 glioblastoma cases, we also confirmed the significant difference in AVIL RNA expression levels between the two groups ( Fig. 1f). To examine the expression levels of AVIL across the major subclasses of glioma recently established and implemented in diagnostic neuropathology, we interrogate the RNAsequencing dataset from the TCGA diffuse glioma study 26 . Here we subdivided the TCGA samples into GBMs with IDH-wild type, GBMs with IDH-mutant, Lower-grade glioma (LGG) with IDH-wild type, LGG with mutant IDH and 1p/19q codeletion, and LGG with IDH mutant without 1p/19q codeletion. Consistent with REMBRANT data, GBMs with or without IDH mutation have high level of AVIL expression. Within LGG, the group with IDH WT is the highest and not statically different from the GBMs. LGG IDH MT without 1p/19q codeletion is slightly higher than the group with codeletion (Fig. 1g).
We then examined the AVIL protein in 12 non-tumor brain tissues, and 14 GBMs. The AVIL protein was absent, or barely detected in the non-tumor cases, but higher levels of AVIL protein expression were seen in almost all of the GBM cases (Fig. 1h). These results further suggest that in addition to the 15-18% of cases that have copy number gain, a higher percentage of glioblastomas may use post-transcriptional/translational mechanisms to up-regulate AVIL expression. They provide indirect evidence that AVIL overexpression may be involved in GBM tumorigenesis rather than being solely a passenger event, a consequence of CDK4 and MDM2 locus amplification.
In order to determine whether AVIL protein expression levels correlate with tumor grade and/or IDH mutation status, we performed a semiquantitative analysis of AVIL IHC data on a set of diffuse gliomas, consisting of grade II astrocytomas and oligodendrogliomas (n = 11, grade III astrocytomas and oligodendrogliomas (n = 14) and grade IV glioblastomas (n = 15) ( Supplementary  Fig. 3i). Using a 0-3 scoring system (1 => 0% <10% tumor cells positive; 2 => 10% <50% tumor cells positive; 3 => 50% tumor cells positive), we found a positive correlation with grade. WHO grade II tumors mean = 1.14 ± 0.40 (SEM); WHO grade III tumors mean = 1.93 ± 0.23 and WHO Grade IV tumors (glioblastoma) mean = 2.72 ± 0.18. Using unpaired T-tests, a significant difference is detected between grade IV and each other group (comparison of grade IV and grade III, p = 0.0126; comparison of grade IV and grade II, p = 0.0072). Because IDH1 mutation status has proven to be a more important prognostic factor than histologic grade, we then compared AVIL IHC scores between IDH1 (R132) mutant and wild-type tumors, with the observer blinded to IDH1 status. We found that IDH1 (R132H) mutant gliomas had a mean score of 1.31 ± 0.31 (SEM) while IDH1 wild-type gliomas had a mean score of 2.36 ± 0.21 (SEM), significantly differing with a two-tailed P value of 0.0063 ( Supplementary Fig. 3j). O v a ri a n B re a s t (S a n g e r) O v a ri a n  T1  T2  T3 T4  T5  T6 T7  N8  N9  N10  N11 N12  T8  T9  T10  T11  T12  T13  T14   4

ARTICLE
To further validate the clinical significance of AVIL in human gliomas, we examined the relationship between AVIL expression, and patient survival in 343 glioma cases in the REMBRANDT project 25 . A three-class model, in which patients were stratified according to AVIL expression showed a clear positive correlation between elevated AVIL expression, and shorter survival (upregulated vs. intermediate, p = 1 × 10 −5 ; upregulated vs. all other, p = 4 × 10 −7 , log-rank test) ( Fig. 2a. The same trend was also observed with a different microarray probe using a two-class model ( Supplementary Fig. 4a). Suspecting that this correlation is due to the overall higher expression of AVIL RNA in GBM as a group (Fig. 1e), we then focused on GBM cases. Within GBMs in TCGA database, we found no association between AVIL RNA level and patient survival ( Supplementary Fig. 4b). However, when we examined AVIL protein expression using our collections of clinical GBM collections, we saw a strong inverse correlation between AVIL protein expression and patient survival (R = −0.82, p = 0.0012) ( Supplementary Fig. 4c). Kaplan-Meier survival analyses also demonstrated a correlation between poor prognosis and the high AVIL protein expression (log-rank test p = 0.0005) (Fig. 2b).
The term Lower-grade glioma (LGG) includes the grade II and III gliomas. However, the management of LGG is one of the most controversial areas in clinical neuro-oncology, with survival ranges from 1 to 15 years. We queried a recent TCGA lowergrade glioma dataset 27 for AVIL expression. Out of 286 samples that had RNA-sequencing data, the high AVIL group (two fold or higher than average) had a much shorter overall survival than the low AVIL group (p = 1 × 10 −5 , log-rank test). The median survival for the low AVIL group was 75.1 months. In contrast, the high AVIL group had a median survival of only 23.1 months, comparable to those of GBM patients ( Fig. 2c and Supplementary  Fig. 5a). Consistently, the two groups also had a significant difference in disease-free survival (p < 0.01) ( Supplementary  Fig. 5b). The difference in patient survival based on AVIL expression is more significant than that based on traditional histopathologic classification and grading.
AVIL overexpression is crucial for GBM tumorigenesis. We then directly tested whether overexpression of AVIL is necessary for the tumorigenesis of GBM. Two independent siRNAs effectively silenced AVIL expression in A172 GBM cells (Supplementary Fig. 6a). By a week, the cells transfected with either AVIL-targeting siRNA almost completely died out (Fig. 3a). Consistently, we observed a significant induction of cleaved Caspase3 around 96-120 h after transfection in these cells (Fig. 3b). The same effect was seen in U251 ( Fig. 3c and Supplementary Fig. 6b). In contrast, no growth inhibition was seen in the astrocyte cultures ( Fig. 3c and Supplementary Fig. 6c). We also measured the effect of silencing AVIL on the migratory ability and invasiveness of A172 and U251 cells. When AVIL was silenced at 48-72 h, a dramatic reduction in the cell movement was observed by a wound-healing assay in both cell lines (Fig. 3d). Similarly, the number of cells that invaded through trans-well was dramatically reduced in both cell lines (Fig. 3e). The wound healing and trans-well assays were performed at early time points after siRNA transfection, and the duration of the experiments were shorter than the doubling time. Nonetheless, the effect of AVIL silencing on cell survival and proliferation complicates the evaluation of cellular motility. To confirm the effect of AVIL on cell migration, we performed live-cell imaging on A172 cells, following the track of movement for individual cells 24 h after Fig. 1 AVIL is overexpressed in glioblastomas. a Cross-cancer analysis of mutations and copy number variation from cBioPortal. The AVIL locus is amplified in about 15% (TCGA GBM provisional), or 18% (TCGA GBM 2008 study) of GBMs. b FISH analysis using a probe covering the AVIL locus in GBM cell lines SF767, A172, U87, and an immortalized astrocyte culture. c Western blot measuring AVIL protein expression in GBM cells, astrocytes, siAVIL treated GBM cells. GAPDH was used as an internal loading control. d Immunohistochemical staining of AVIL in non-neoplastic brain and glioblastomas. Scale bars in A-D represent 50 microns; scale bar in the insert represents 10 microns. e AVIL expression in REMBRANDT database with microarray data of 28 non-tumor brain tissues, 148 astrocytomas (WHO grade II or III), and 228 GBM (WHO grade IV) cases. Results from two microarray probes are shown. f qRT-PCR summary of our own collection consisted of 8 non-tumor brain tissues, and 36 GBM cases. AVIL level was normalized against that of GAPDH. Data are presented as mean values ± SD. g RNA-Seq analysis on diffuse glioma study indicates that GBMs have higher level of AVIL than LGG in general, but LGGs with wild-type IDH1 have a comparable high level of AVIL to GBMs. LGG with IDH MT without 1p/19q codeletion is slightly higher than the group with codeletion. h Western blot measuring AVIL protein in 12 non-tumor brain biopsies from epilepsy patients (N1-N12), and 14 GBM (T1-T14) samples. For box plots in e and g, box, 25-75 percentile; lower whisker, least data value excluding outliers; upper whisker, highest data value excluding outliers; bar in middle, median. Lower outliers are calculated as less than Q1 − 1.5*IQR; Upper outliers are calculated as greater than Q3 + 1.5*IQR, where Q1 is value corresponding to 25th percentile and Q3 is value corresponding to 75th percentile and IQR is interquartile range. For all qPCR, P value was calculated by standard two-tailed t-test. *p < 0.05, **p < 0.01, ***p < 0.001.  transfection. Consistently, obvious reduction was observed when AVIL was silenced (Fig. 3f, g and Supplementary Movie 1). To confirm whether AVIL plays an important role in tumorigenesis in vivo, we tested the effect of silencing AVIL in tumor initiation with a widely used U251 intracranial xenograft model. In this model, tumor-bearing animals usually die in about one month 28 . We implanted U251 cells that were freshly infected with lentivirus expressing shAVIL1, or shCT in the brains of immune-deficient mice. After 4 weeks, all control mice had reached significant tumor volumes, detected by MRI (Fig. 3h). In contrast, we could hardly observe any tumor formation in the shAVIL1 group by MRI. Consistently, dramatic differences in tumor volumes were observed between the two groups (Supplementary Fig. 7a). All of the shCT group animals died within 35 days, or had to be euthanized due to abnormal behavior caused by tumor burden. All of the shAVIL1 group animals displayed no sign of disease, until the day we terminated the experiment, with the exception of two (we euthanized one shAVIL1 mouse at the same time as a shCT mouse as a control. Another shAVIL1 mouse developed an infection after the MRI imaging, requiring euthanasia) ( Supplementary Fig. 7b). These results support the crucial role that the overexpression of AVIL plays in tumorigenesis in vivo.
Like other cancer types, GBMs harbor a subpopulation of glioblastoma initiating/stem cells (GICs/GSCs) that govern tumor initiation, maintenance, and recurrence after therapy 29 . GSCs partly mediate resistance to chemo and radiotherapy, and have become vital targets for the reversal of chemo-resistance 30   GIC/GSC cells in our collection express a high level of AVIL, while a neural stem cell culture 31 had almost no detectable AVIL expression (Fig. 4a, b). Consistently, we found that silencing AVIL in two GSCs, GSC11 and GSC627, resulted in dramatic reduction of cell proliferation, while no obvious difference was observed in the neural stem cell culture (Fig. 4c). Neurosphere formation was also dramatically reduced in the two GSCs (Fig. 4d). Not surprisingly, the expression of stemness markers was reduced, whereas both astrocyte and oligodendrocyte differentiation markers were induced when AVIL was silenced ( Supplementary Fig. 8). Using live-cell imaging, we traced the movement of GSC individual cells in 3D culture. Similar to GBM   (Fig. 4g). In addition, more apoptotic cells were found in the shAVIL group ( Supplementary Fig. 9).
AVIL is a bona fide oncogene. To further confirm that high levels of AVIL are tumorigenic, we examined the effects of AVIL overexpression in astrocyte, U251, and U87 cells. AVIL overexpression led to increased proliferation rates, and increased migration in all three cell types ( Fig. 5a-c).
To test whether AVIL can function as a bona fide oncogene, we performed the classic focus assay on NIH3T3 cells expressing, or not, AVIL. We observed significantly larger and higher numbers of foci in cells transfected with AVIL than with an empty vector control (Fig. 5d). Most GBMs have genetic alterations in three core signaling ways [32][33][34] , including RTK/RAS pathway with alteration in EGFR/PDGFRA/PI3K/PTEN/NF1/RAS, the p53 pathway with genetic alteration in TP53/MDM2/MDM4/ p14ARF, and the RB1/CDK4/p16INK4A/CDKN2B pathway. To examine the potential connection and collaborative effect of AVIL with these three pathways, we performed a modified oncogene cooperativity assay, where we introduced alone the EGFR vIII mutant 35 , shRNA targeting TP53, or shRNA targeting RB, or in combination with AVIL overexpression. As shown in Fig. 5e, all three resulted in an increased number of foci. Impressively, overexpressing AVIL alone has triggered a larger number of foci compared with any of these three known oncogenic factors. In addition, combing AVIL with these known factors did not result in more foci, but rather less in the case of shRb and AVIL combination. These results suggest that AVIL is sufficient to trigger focus formation and is at least as powerful, if not stronger, than the known factors.
To evaluate the transforming ability of AVIL in vitro, we performed soft agar assay to test anchorage independency for astrocytes overexpressing AVIL. Compared with empty vectortransfected controls, a large number of colonies were observed in astrocytes transfected with an AVIL expressing vector (Fig. 5f).
Lastly, we tested whether the overexpression of AVIL would transform astrocytes in vivo. We injected 2 million astrocyte cells stably expressing either AVIL or a control plasmid subcutaneously into the flanks of NIH-III nude mice. No tumor was seen in any of the ten injections for the control group. In contrast, five out of 14 injections of the AVIL overexpression group had visible tumors within ten days of injection (5G). To confirm that the masses formed in the group of AVIL overexpression are truly tumor, we examined them histologically. Indeed, the cells from AVIL overexpressed astrocytes revealed a neoplasm with histologic features of malignancy including necrosis (N), frequent mitotic figures (arrows), and apoptotic figures (asterisks) (Fig. 5h).
We then conducted RNA-Seq on the whole transcriptome changes, and identified a large number of gene expression affected. Among them, all these major oncogenic pathways involved in GBM tumorigenesis were enriched by a Gene Set Enrichment Analysis (GESA) (Supplementary Fig. 10). On the other hand, we found no significant changes of AVIL expression when we introduced EGFR mutant, shTP53 or shRB into these astrocytes ( Supplementary Fig. 11). These results together with our observation that AVIL is overexpressed in all GBMs we tested, suggest that AVIL overexpression is a common theme, and it may even function upstream of the three known oncogenic signaling pathways.
AVIL binds to F-actin. AVIL is a member of the villin/gelsolin family of actin-regulatory proteins 21 . The gene encodes a protein also known as advillin, which has been reported to affect cell movement, and is involved in the formation of filopodia-like structures in fibroblasts, as well as a role in ciliogenesis 36 . Consistently, our results showed that silencing AVIL reduced migration in GBM cells, whereas overexpression enhanced their migration (Figs. 3 and 5). However, the exact activities of AVIL on actin cytoskeleton have not been extensively examined. We first visualized colocalization of GFP-tagged AVIL, with phalloidin staining of F-actin (Fig. 6a, b and Supplementary Fig. 12). We confirmed direct binding of recombinant AVIL to actin filaments by high-speed ultracentrifugation ( Fig. 6c and Supplementary Fig. 13). More AVIL was found to be in the pellet fraction when actin was added.
AVIL's activity depends on F-actin binding. The dependence of GBM cells on AVIL to attach and spread was further evidenced when we monitored the cell shape and area when newly plated onto fibronectin substrate. Control cells spread much more than the siAVIL1 transfected cells, reflected by overall silhouettes of the cells and the areas cells occupied (Fig. 6d, e).
We further analyzed the effect of AVIL knock-down on the dynamics of cellular protrusions and retractions on the cell edge as those processes were previously established to be dependent on actin cytoskeleton dynamics 37 . Although the knock-down of AVIL did not affect the rate of the membrane retraction, the rate of forward protrusion at the cell edge was increased compared to controls ( Supplementary Fig. 14). Even though the reduced migration cannot be directly explained by this defect, the result is consistent with an imbalance in the actin cytoskeleton regulation caused by the loss of AVIL. This finding supports a role for AVIL in the regulation of actin dynamics in cells.
AVIL's headpiece domain is situated at the C-terminus and shares sequence and structure similarity with villin's headpiece. The villin headpiece domain was previously shown to be crucial for F-actin binding and bundling 38 . Based on the structural models of the villin and advillin headpiece domains 21 , and sequence conservation, we selected six residues in the AVIL headpiece for mutagenesis and biochemical studies (Fig. 6f). We Fig. 4 AVIL overexpression is crucial for GSCs/GICs. a qRT-PCR measuring AVIL mRNA level in GSCs, two GBM cell lines, and control astrocytes. AVIL RNA expression was normalized against that of GAPDH. b Western blot measuring AVIL protein in astrocyte, A172, NSC, and four GSC cell lines. c Cell growth measured by MTT. Shown are the results of GSC11 (upper), GSC627 (middle), and NSC (lower) cells transfected with shAVIL1, shAVIL2, or control shCT (n = 6). d Neurosphere formation was inhibited with AVIL silencing. GSC11 (upper) and GSC627 (lower) cells were transfected with shAVIL1, shAVIL2, or control shCT. e Movement of GSC11 cells suspended in Matrigel matrix was observed over 24 h period of time by live-cell imaging, starting 6 h after transection with either siCT (left) or siAVIL (right). Shown are representative images depicting the last timepoint of the experiment with overlaid lines tracking the movement of individual cells. f Mean velocities of all cells tracked in the experiment depicted in e. n > 300 cells quantified per condition (box, 25-75 percentile; whisker, 5-95 percentile; bar in middle, median) (two-sided Student's t-test). g Tumor volume comparison between GSC11 cells infected with shCT expressing virus and shAVIL virus (n = 6). Data are presented as mean values ± SD in a, c, and g. P value was calculated by standard two-tailed t-test. *p < 0.05, **p < 0.01, ***p < 0.001. NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-17279-1 ARTICLE NATURE COMMUNICATIONS | (2020) 11:3457 | https://doi.org/10.1038/s41467-020-17279-1 | www.nature.com/naturecommunications expressed Myc-tagged AVIL, either wild-type or mutants, in glioblastoma cells in which endogenous AVIL was silenced and performed co-immunoprecipitation with anti-Myc antibody to assess the interaction between actin and AVIL mutants. The majority of the AVIL headpiece mutants exhibited reduced binding to actin (Fig. 6g).
We then chose two mutants (AVIL K808C and AVIL F819C) that yielded the lowest amount of actin in the co-immunoprecipitation   experiment and are also highly conserved for further studies. Consistently, both recombinant AVIL mutant proteins exhibited reduced binding to filamentous actin ( Fig. 6h and Supplementary  Fig. 15). Compared to the wild-type AVIL, these mutants do not promote cell proliferation to the same rate (Fig. 6i). More importantly, the mutants failed to rescue cell migration caused by AVIL silencing (Fig. 6j).
AVIL regulates the stability of FOXM1. We then performed whole transcriptome analyses on U87 cells silenced for AVIL, and with AVIL overexpression, together with controls (Fig. 7a). We focused on those candidates whose expression was inversely changed when AVIL was silenced or overexpressed. We noticed that many of such candidates including AURKA, AURKB, CDC20, KIF11, SPC25, and TTK etc., are known targets of transcription factor, FOXM1. Gene Set Enrichment Analysis (GESA) revealed highly similar profiles between FOXM1 targets 39 , and differentially expressed genes caused by AVIL silencing/overexpression (Fig. 7b). In addition, using BART, a bioinformatics tool for predicting functional factors 40 , FOXM1 was predicted to be the top factor, far more significant than other factors (Fig. 7c).  Fig. 6 AVIL binds to F-actin, and the interaction is crucial for its tumorigenic activity. a AVIL colocalize with F-actin. U87 cells stably expressing GFP-AVIL (green) were stained for F-actin (Phalloidin, red) showing co-localization of AVIL and F-actin. b Fluorescence colocalization between F-actin and AVIL. Scatterplot representing the intensity range of red and green pixels in the image as shown in a. c AVIL binds directly to F-actin. Silver-stained SDS-PAGE analysis of actin high-speed pelleting of F-actin in the presence or absence of AVIL. d Silencing AVIL resulted in reduced ability of A172 cells to spread. e Cell area was plotted for A172 cells transfected with siCT, or siAVIL1. Significant difference (p = 0.0014) was observed by the two-sided Rank-Sum test (non-parametric test for non-normally distributed data). n > 50. f Amino acid sequence alignment of the C-terminal region of advillin and villin from eight species. Arrows indicate positions selected for mutagenesis; red arrows indicate positions of mutations used in experiments represented in g-j.
g Mutations in the headpiece region of AVIL significantly reduce binding to β-actin in cells. Co-immunoprecipitation of β-actin by antibodies specific to Myc-tag in cell lines expressing Myc-tagged AVIL WT or indicated mutants. Mutants indicated in red were subsequently utilized in experiments represented in h-j. IBimmunoblotting, IPco-immunoprecipitation. h AVIL mutants are deficient in binding to F-actin. Silver-stained SDS-PAGE analysis of actin high-speed pelleting of F-actin in the presence or absence of AVIL in a buffer containing EGTA. i Mutations in AVIL headpiece region are not as effective as WT AVIL in stimulating cell proliferation (two-sided Student's t-test). U87 cells were transfected with siAVIL and grown for 24 h followed by transfection with a plasmid overexpressing AVIL WT or indicated AVIL mutants. After 48 h, cells were counted. j Headpiece mutants failed to rescue deficiency in cell movement (two-sided Student's t-test). Total movement of U87 cells expressing AVIL WT or indicated mutants in the wound-healing experiment. Data are presented as mean values ± SD in e, i, and j. FOXM1 is overexpressed in many cancers including glioma 14 , where it has been demonstrated to play a role in the pathogenesis, progression, and metastasis 15,41 . We found the reduced cell growth and migration abilities caused by AVIL silencing can be at least partially rescued by overexpressing FOXM1 (Fig. 7d), suggesting that FOXM1 functions downstream of AVIL. Perturbing AVIL had no effect on FOXM1 mRNA (Fig. 7e), but affected FOXM1 at the protein level (Fig. 7f), suggesting a posttranscriptional mechanism of AVIL on FOXM1. Indeed, FOXM1 half-life was reduced by silencing AVIL, and increased by overexpressing AVIL (Fig. 7g). Given the activity of AVIL on F-actin, we suspect that disturbing F-actin dynamics may also perturb FOXM1 stability. Indeed, when we used F-actin polymerization inhibitor Cytochalasin D, or the depolymerization inhibitor Jasplakinolide to treat U87 cells, we also observed significant reduction of FOXM1, supporting that AVIL loss or wholesale disruption of F-actin dynamic result in reduced FOXM1 protein (Supplementary Fig. 16) Consistently, using proteasome inhibitor MG132, the effect of AVIL silencing on FOXM1 was fully abolished (Fig. 7h).
Among the genes that were inversely regulated by AVIL silencing vs. overexpressing, LIN28B caught our attention. LIN28B belongs to the group of RNA-binding proteins that play critical roles in embryonic development, and in tumorigenesis 42 . Several studies have demonstrated that LIN28B can promote proliferation, and invasion of cancer cells 43,44 . It has been postulated that LIN28B is regulated by FOXM1 26 . Indeed, we found that silencing AVIL or FOXM1 resulted in reduced LIN28B expression, whereas overexpressing AVIL or FOXM1 enhanced LIN28 expression (Fig. 8a and Supplementary Fig. 17). Supporting that LIN28B is downstream of AVIL, we found that LIN28B expression partially rescued both reduced cell proliferation and migration caused by AVIL silencing in U87 and U251 cells (Fig. 8b, and Supplementary Fig. 18). In GSC cells, similar observation was made supporting that LIN28B mediates at least part of tumorigenic effect of AVIL ( Supplementary Fig. 19). Consistently, AVIL mutants defective in F-actin binding also failed to induce LIN28B to the same extent as the wild-type AVIL (Fig. 8c). Clinically, high LIN28B expression correlates with poor prognosis in patients with GBMs (Fig. 8d), and lower-grade gliomas (Fig. 8e).
LIN28B is best known to negatively regulate the biogenesis of tumor-suppressive microRNAs, let-7. Consistently, let-7 members have been shown to inhibit glioma cell malignancy 19 , and are expressed at lower levels in high-grade gliomas 19 . Indeed, we observed that silencing AVIL induced all of the members of let-7 that are expressed (let7-f, let7-g, and mir-98 are not expressed in U87 cells) (Fig. 8f). Conversely, AVIL overexpression resulted in the down-regulation of most let-7 members. Recently, the overexpression of several let-7 downstream targets including HMGA2, IGFBP1, and IGFBP3, were found to define a distinct patient group with poor prognosis in pancreatic cancer 45 . HMGA2 and IGFBP1 have been shown to antagonize the tumor-suppressive activity of let-7 46 . Since let-7 expression level is not available in the TCGA datasets, we examined HMGA2, IGFBP1, and IGFBP3 level. Impressively, higher levels of all three genes are correlated with worse overall survival, as well as diseasefree survival (Supplementary Fig. 20).

Discussion
Glioblastoma (GBM), WHO classification Grade IV Astrocytoma, is the most common, and most aggressive malignant primary brain tumor in humans 47 . Survival of patients affected by GBM has remained low, despite advances in surgery, radiation, and chemotherapy 1-3 . About 50% of patients diagnosed with GBM die within one year, and 90% die within three years 48 . More therapeutic targets and treatment options are clearly needed. Our study shows that AVIL is overexpressed in the vast majority, if not all of human glioblastomas. We found that GBM cells, including GSC/GIC cells are dependent on the overexpression of AVIL for increased survival and migration. Silencing AVIL induced GBM cell death in vitro, and prevented/reduced GBM xenograft formation and growth in animal models. In contrast, normal astrocytes or neural stem cells express very low levels of AVIL, and silencing AVIL had no obvious effect. Taken together, this demonstrates that AVIL is crucial for GBM tumorigenesis, which are addicted to the AVIL overexpression. Therefore, AVIL may be a promising selective therapeutic target, inhibition of which may effectively suppress GBM growth and invasion, yet spare normal brain cells.
Tumor cells use multiple "tricks" to dysregulate some oncogenes, which at the same time give credence to the genes as key players in tumorigenesis and malignancy. Our study showed that AVIL is dysregulated via multiple mechanisms. It forms a fusion with MARS in rhabdomyosarcomas. It is amplified by copy number gain in some glioblastomas, and upregulated at the transcriptional/translational level in other GBMs. In this regard, it is similar to anaplastic lymphoma kinase (ALK). ALK has been proven to be a remarkably promiscuous oncogene 49 . It contributes to the development of a notable assortment of tumor types from different lineages through a variety of genetic mechanisms. It can form NPM-ALK fusions in~60% of anaplastic large-cell lymphomas (ALCLs) 50 , or EML-ALK in a subset of non-small-cell lung cancers 51 ; additionally, it can have gain-offunction point mutations in neuroblastomas 52 , and copy number gain in several tumor types 53 . Accordingly, the ALK inhibitor, Crizotinib, was proven to be an effective, and safe treatment for lung cancer patients harboring EML-ALK 54 , leading to the quick approval of the drug by the FDA. Considering gene fusions often result in aberrant expression of one of the proto-oncogenes, our strategy was to start from a gene fusion in pediatric cancer, then extend to adult tumors including GBM. We believe that the approach may lead to the discovery of other promiscuous oncogenes.
AVIL is known as a member of the villin/gelsolin family, which regulates actin filament reorganization 21 . We demonstrated that F-actin interacting capability of advillin is crucial for the migration of GBM cells, and for its downstream signaling pathways. Advillin is 75% homologous to villin, and 65% homologous to gelsolin and adseverin. Recently, AVIL mutants were found to be involved in the alteration PLCE1 action, and contribute to the Steroid-resistant nephrotic syndrome 55 . Even though there are studies suggesting connections between these actin-regulatory proteins and cancer 56,57 , no direct evidence has shown any of the family members to be oncogenic, or tumor suppressive. However, given the importance of actin dynamic in cell growth and motility, it is not surprising to have one or more villin/gelsolin family members affect tumorigenesis. Indeed, both gelsolin and villin are known to regulate apoptosis, cell migration, invasion, and EMT 58,59 .
Consistent with our observation with AVIL, FOXM1 has been reported to play a critical role in tumorigenesis of GBM. It can promote cell proliferation, invasion, and transform immortalized astrocytes in vitro and in vivo 14,15,17,18 . However, how is FOXM1 overexpressed in cancer is not all clear. We have accumulated the following evidence that AVIL regulates FOXM1 stability through regulating F-actin dynamics: (1) FOXM1 level was reduced with AVIL silencing, and increased with AVIL overexpression; (2) FOXM1 half-life is reduced when AVIL is silenced; (3) FOXM1 half-life is extended with AVIL overexpression; (4) FOXM1 level is enhanced with MG132, a proteasome inhibitor; (5) even with AVIL silencing, in the presence of MG132, FOXM1 remains the same as siCT; and (6) F-actin polymerization inhibitor Cytochalasin D, and depolymerization inhibitor Jasplakinolide treatment also resulted in significant reduction of FOXM1. In addition, we showed that AVIL functions upstream of FOXM1, which in turn regulates LIN28B transcription, and that AVIL mutants, which are defective in F-actin binding failed to rescue reduced LIN28B caused by AVIL silencing. Thus, our study revealed a connection between cytoskeleton and FOXM1, and established a signaling axis of AVIL/FOXM1/LIN28B/let-7.

Methods
Cell culture. Glioblastoma cell line U87 cells were cultured in Minimum Essential Medium Eagle (MEM), supplemented with 1 mM sodium pyruvate, 1% nonessential amino acids, 0.15% sodium bicarbonate, and 10% fetal bovine serum (FBS); T98G cells were cultured in MEM with 10% FBS; A172 cells were cultured in Dulbecco's modified Eagle's medium (DMEM) with 4.5 g/L glucose, and supplemented with 10% FBS; U251 cells were cultured in Roswell Park Memorial Institute medium with 5% FBS. Immortalized human astrocytes (a kind gift from Dr. Russ Pieper, University of California San Francisco) were grown in DMEM/F12 with 4.5 g/L of glucose, supplemented with 10% FBS. NSC cells were purchased from Millipore. GSC/GIC cells were gifts from Dr. Krishna Bhat, MD Anderson. They were cultured Neurobasal Media supplemented with 0.5mM L-Glutamine, B27 supplement, N2 supplement, 50 ng/ml bFGF and 50 ng/ml EGF. All cells were grown at 37°C in 5% CO 2 -95% O 2 . All cell lines unless specified were originally obtained from ATCC. They were not further authenticated.
Clinical samples. Fresh-frozen samples of normal brain and tumors were collected under an approved University of Virginia Institutional Review Board protocol. Tumors were macrodissected, and RNA and protein were extracted according to well-established protocols.
PCR and real-time PCR. RNA was extracted using TRIzol reagent (Invitrogen), and quantified with Nanodrop (Thermo). cDNA was generated by AMV-RT kit (NEB), and a random hexamer primer. Real-time qPCR was carried out on the StepOne Plus system from Applied Biosystems using SYBR mix (Thermo). Primer sequences are listed in Supplementary Table 1.
Western blotting. To measure protein levels, cell lysates were resolved by denaturing gel electrophoresis, before electrotransfer to a Protran nitrocellulose membrane. The membrane was subjected to western blot analysis with antibodies against the proteins of interest. The following antibodies and dilutions were used: . a LIN28B was reduced upon AVIL silencing, and induced upon AVIL overexpression. U87 cells were transfected with siAVIL or siCT, and empty vector control (CT) or AVIL expressing vector. LIN28B level was measured by qRT-PCR, and normalized to that of GAPDH, and then to siCT. b LIN28B mediates at least some of the AVIL effect. U87 cells transfected with siAVIL or siCT were further transfected with LIN28B expression vector, or control plasmid (CT). Cell proliferation was measured by cell counting. Cell motility was measured by wound-healing assay. c Headpiece mutants are not as effective as WT in inducing LIN28B expression. LIN28B expression was measured by qRT-PCR and is represented relative to GAPDH levels. d Clinical correlation between the expression of LIN28B and patient survival. A two-class model, stratified by LIN28B expression using TCGA GBM data. The LIN28B high group has a significantly worse survival (p < 0.01, two-sided log-rank test). e With TCGA lower-grade glioma data, LIN28B high group is found to have worse overall survival (p = 0.00013) (two-sided log-rank test). f Let-7 family members were induced upon AVIL silencing, and suppressed upon AVIL overexpression. Let-7 family members that are expressed in U87 were measured by qRT-PCR. Their expression levels were plotted against that of siCT, or CT. *p < 0.05, **p < 0.01, ***p < 0.001. Data are presented as mean values ± SD in a-c, and f. FISH. DNA probes for fluorescence in situ hybridization (FISH) were labeled by nick translation with SpectrumGreen or SpectrumRed 2′-deoxyuridine-5′-triphospahte (Abbott). Cells were grown in 8-chamber slides and fixed with methanol:acetone = 1:1 fixation solution for 10 min at 4°C. BAC Fish clone, RP11-143123 was purchased from BACPAC.
Immunohistochemistry. Immunohistochemistry for AVIL was performed on a robotic platform (Ventana discover Ultra Staining Module, Ventana Co., Tucson, AZ, USA). Tissue sections were deparaffinized using EZ Prep solution (Ventana). A heat-induced antigen retrieval protocol set for 64 min was carried out using a TRIS-ethylenediamine tetracetic acid (EDTA)-boric acid pH 8.4 buffer (Cell Conditioner 1). Endogenous peroxidases were blocked with peroxidase inhibitor (CM1) for 8 min before incubating the cells with AVIL antibody (Abcam, Ab72210) at 1:100 dilution for 60 min at room temperature. Antigen-antibody complex was then detected using DISCOVERY anti-rabbit HQ HRP detection system and DISCOVERY ChromoMap DAB Kit (Ventana Co.). All the slides were counterstained with hematoxylin subsequently; they were dehydrated, cleared, and mounted for the assessment.
For cleaved caspase3, and Ki67, A heat-induced antigen retrieval protocol set for 64 min was carried out using a TRIS-EDTA-boric acid pH 8.4 buffer (CC1). Endogenous peroxidases were blocked with peroxidase inhibitor before incubating the section with cleaved Caspase 3 antibody (Cell Signaling, Cat #9661) at 1:400 dilution for 1 h at room temperature, antigen-antibody complex was then detected using DISCOVERY anti-rabbit HQ HRP detection system and DISCOVERY ChromoMap DAB Kit (Ventana Co.) The slide was next incubated with Ki67 antibody (Abcam, Ab16667) at 1:300 for 1 h at room temperature and then incubated with DISCOVERY anti-rabbit HQ HRP detection system and the DISCOVERY Purple kit. All slides were counterstained with hematoxylin subsequently; they were dehydrated, cleared, and mounted for the assessment.
Immunofluorescence and confocal microscopy. For GFP-AVIL, U87 cells were transfected with GFP-AVIL expressing plasmid, and seeded on chamber slide and subsequently co-fixed with 4% PFA and 0.5% Triton-X 100 for 10 min at 37°C and blocked with 5% BSA in PBS. Cells were then incubated with Texas Red™-X Phalloidin (ThermoFisher Scientific) for 45 min at room temperature, washed with PBS, and stained with DAPI. Cells were subsequently examined under a Zeiss LSM 510 laser scanning fluorescence confocal microscope at 400x nominal magnification. For other immunofluorescence experiments, fixed cells were typically blocked and incubated overnight with primary antibodies. Subsequently, the slides were washed with PBS, incubated with fluorescence-conjugated secondary antibodies for 1 h at room temperature, before mounted and examined by Zeiss LSM 510 laser scanning fluorescence confocal microscope.
Cell migration and invasion assay. The effect of AVIL on cell migration was assayed by a wound-healing assay. Briefly, cells were cultured to confluency. A wound was created by scraping the cells using a 10 ul plastic pipette tip, and the medium was replaced with fresh medium. Images were captured immediately after the scratch, and again 6 h later. Cell migration was qualitatively assessed by the size of the gap within the confluent monolayer culture at the end of the experiment. Eight gaps were measured. For siRNA experiment, the measurement took place around 48-72 h after transfection.
The effect of AVIL on cell invasion was assessed by a transwell invasion assay. siAVIL or siGL2 transfected glioblastoma cells (1 × 10 5 ), after 72 h transfection were suspended in 300 μL 0.1% FBS medium, and added to the upper chamber of the wells. The lower chamber contained 600 μL of 10% FBS medium. The plate was kept in air with 5% CO 2 for 8 h at 37°C. The cells on the upper membrane surface were then mechanically removed. The cells that had migrated to the lower side of the collagen IV-coated membrane were fixed and stained with 0.1% crystal violet. Migrated cells were counted in five randomly chosen fields under a microscope, and the average number of these cells per field was calculated.
Live-cell imaging. The cellular movement was analyzed by live-cell imaging. Briefly, A172 were plated to 40-60% confluency in DMEM + 10% FBS without Phenol Red, followed by transfection with siGL2 or siAVIL. 2 h prior to start of imaging media was supplemented with 1 µM SiR-DNA (Cytoskeleton) dye. 24 h after siRNA transfection images were collected on a Zeiss Axio-observer-Z1 epifluorescent microscope in humidified chamber in 5% CO 2 environment, at 37°C every 20 min over the period of 24 h. Resulting movies were processed ImageJ (https://imagej.nih.gov/ij/), and cell movement was tracked semi-automatically based on DNA staining by TrackMate plugin for ImageJ.
Analysis of the movement of GSC11 glioma stem cells was done similarly as described above with following modifications. 2 h after the transfection with either siGL2 or siAVIL GSC11 cells were transferred to the 24-well clear bottom plate in media containing 50% MatriGel matrix (Corning) supplemented with 1 µM SiR-DNA. Images were collected similarly to described above, starting 6 h after transfection. Subsequent analysis was performed utilizing TrackMate plugin in ImageJ.
For the experiment depicted in Fig. S14, A172 cells expressing mCherry-Lifeact were plated onto Nunc™ Lab-Tek™ II Chambered Coverglass with 1.5 Borosilicate glass (ThermoFisher Scientific) previously coated with 1 µg/ml fibronectin and subsequently transfected with siGL2 or siAVIL at~40% confluency. 24 h after transfection media was replaced with DMEM + 10% FBS without Phenol Red (Gibco), supplemented with 20 mM HEPES buffer. Movies were collected for 10 min by inverted confocal microscope (Zeiss, Jena, Germany) using a Nakagawa spinning disc and with frames taken every 2 s. Following movies were processed and analyzed in ImageJ.
Cell morphology assessment. All materials were from Life Technologies unless otherwise indicated. In-house-made dishes with glass coverslip bottoms were preadsorbed with 2 µg/ml fibronectin overnight. Cells were stained with 5 µM DiI in OptiMEM for 15 min, then rinsed with PBS and returned to normal growth medium until the experiment. Cells were seeded on glass-bottomed dishes in serum-free, CO2-independent medium (CCM1, Hyclone), and cultured for 1.5 h. Each sample was imaged live on an Olympus Fluoview 1000 laser scanning confocal microscope with a 10× (0.3 NA) objective. The samples were maintained at 37°C with a stage heater. The DiI-stained cells were excited using the 543 nm line of a HeNe laser. Images were acquired at a resolution of 0.795 µm/pixel. Settings were adjusted to minimize photodamage.
Cell area measurements were automated using custom MATLAB scripts. Briefly, cell images were subjected to an interactive threshold, resulting in silhouettes, which were then automatically quantified using MATLAB's built-in region properties function. Cell areas (µm 2 ) of the control and AVIL knock-down groups were compared using the Rank-Sum test for non-normally distributed data. The sample sizes for the control and knock-down groups were 55 and 51 cells, respectively.
Tumor formation in vivo. The mouse work was performed under the study protocol 4234 as approved by the University of Virginia Institutional Animal Care and Use Committee. Immunocompromised SCID/NCr BALB/c adult male mice (6-8 weeks old) were used. All animals were housed in sterilized plastic cages under specific pathogen-free conditions, at 22°C, 12/12 light/dark cycle, 55% humidity. U251 or GSC11 cells were transfected with control shRNA, or shAVIL1 for 12 h. The transfected cells were then counted, and 3 × 10 5 were stereotactically (Stoelting) implanted into the right corpus striatum of the mice. Cerebral magnetic resonance imaging was performed on anesthetized mice at 4 weeks post implantation. Ten to fifteen minutes before scanning, 30 ul of Magnevist brand gadopentetate dimeglumine was injected intraperitoneally. T1-weighted serial coronal images of each brain were acquired at 1 mm intervals with a 5 × 5 mm field, and a 256 × 256 pixel resolution. For image analysis and tumor volume quantification, a luminosity histogram was first generated for a selected area of the left cerebrum that was grossly tumor-free. This served as an internal control. Pixel luminosity mean and SD were noted. Histogram generation was repeated on a similar selection from the right cerebrum that contained all enhanced tumor. Pixels in the right cerebrum, greater than two SDs above the left cerebrum control luminosity mean, were recorded as representing enhanced tumor for a given image. This procedure was then repeated for all the images showing an enhanced tumor for a given brain, thus generating a sum of enhanced tumor pixels for each brain. Tumor volume relates to tumor pixels in a linear manner, and was calculated based on the image acquisition, interval distance, and resolution. Astrocytes stably expressing AVIL or an empty vector were injected subcutaneously into the flanks of NIH-III Nude mice. Around 2 million cells were used per injection. The animals were monitored twice a week. For drug treatment, 1.5 million U87 cells were injected subcutaneously into the flanks of NIH-III Nude mice.