A first-in-class small molecule inhibitor targeting AVIL exhibits safety and antitumor efficacy in preclinical models of glioblastoma

Abstract

Glioblastoma (GBM) is the most common and deadliest malignancy of the brain. Despite decades of intense research, there has been little change to the overall survival of patients with glioblastoma. Our lab recently identified the actin-binding protein advillin (AVIL) as being overexpressed, oncogenic, and necessary for tumorigenesis in glioblastoma. Here, we further examined AVIL expression in GBMs and found it was enriched across molecular subtypes and states, including GBM stem cells and temozolomide-resistant samples. In contrast, we found that AVIL was scarcely expressed in normal human brain tissue. In addition, Avil knockout in mice had no adverse effects, suggesting there may be a wide therapeutic window for therapies targeting AVIL. Using high-throughput small molecule screening, we identified a direct inhibitor of AVIL that bound to the protein and also blocked AVIL binding to its substrate, actin. It induced a similar transcriptome profile to that of AVIL silencing by siRNA, and caused down-regulation of FOXM1 and LIN28B, two known downstream targets of AVIL. Moreover, it exhibited selectivity towards tumor cells, sparing astrocytes and neural stem cells in vitro. In vivo, we found that the compound readily crosses the blood-brain barrier and could be delivered orally. We then demonstrated efficacy in five GBM mouse models without evidence of side effects. In summary, we have identified an efficacious first-in-class compound targeting a oncogene in GBM. Further optimization of the molecule may offer an effective therapeutic intervention for GBM.

One sentence summary:

A first-in-class compound targeting the oncogene AVIL demonstrated efficacy and safety in mouse models of glioblastoma.

Editor’s Summary:

Abolishing AVIL. Glioblastoma (GBM) is the most common primary brain tumor in adults and has a dismal prognosis. New therapeutic targets are thus urgently needed. Here, Xie et al. build upon their prior work where they identified advillin (AVIL) as an oncogene in GBM by developing a first-in-class compound to inhibit AVIL. The authors first verified that AVIL is indeed a driver of GBM growth, then show that knocking down or pharmacologically inhibiting AVIL can impair GBM and GBM stem cell growth and migration, extending survival in multiple mouse models of GBM including TMZ-resistant models. Notably, the AVIL inhibitor did not appear to affect healthy cells including astrocytes, supporting the potential for AVIL inhibition in the clinic. Together, these data suggest that a new target for GBM therapy has been found and support further development of this and other AVIL inhibitors. –Courtney Malo

INTRODUCTION

Glioblastoma (GBM) is the deadliest neurologic malignancy with approximately 12,000 new cases annually in the US alone and more than 300,000 cases reported worldwide(1). Despite advances in surgery, radiation, and chemotherapy, the survival of patients with GBM remains dismal (about 15 months after diagnosis)(2–5). Currently, the only commonly used treatment, temozolomide with radiation, provides just a 2.5 month increased survival over radiation alone(6). Consequently, the development of additional therapeutics and exploration of new areas of drug development are critical, unmet needs in the field.

In a previous study, we discovered AVIL forming a gene fusion with a housekeeping gene, MARS, in a pediatric tumor, rhabdomyosarcoma(7, 8). Suspecting that other cancers may dysregulate AVIL using different mechanisms, we found that AVIL is overexpressed in all of the GBMs we tested. In some cases, this was due to locus amplification; in others, overexpression occurred at the transcriptional or translational level.

AVIL is a member of the gelsolin/villin family of actin-binding proteins and regulates F-actin dynamics(9, 10). AVIL can bind, nucleate, sever, and bundle actin filaments. Advillin is 75% homologous to villin, and 65% homologous to gelsolin and adseverin. Even though there are studies suggesting connections between these actin-regulatory proteins and cancer (11, 12), no direct evidence has shown any of the family members to be oncogenic or tumor suppressive. However, given the importance of actin dynamics in cell growth and motility, it is not surprising to have one or more such proteins affect tumorigenesis. Indeed, both gelsolin and villin are known to regulate apoptosis, cell migration, invasion, and EMT(13, 14).

We previously showed that silencing AVIL killed GBM cells including GBM stem/initiating (GSC/GIC) cell cultures and inhibited growth of in vivo xenografts in mice, but had no effect on control astrocytes or neural stem cells. Conversely, overexpressing AVIL promoted cell proliferation and migration in GBMs and astrocytes in vitro, enabled fibroblasts to escape contact inhibition, and transformed immortalized astrocytes. In addition, we found that AVIL expression correlated with patient prognosis, with high AVIL expression being associated with worse outcomes. These evidences support that AVIL is a bona fide oncogene(15, 16). On the other hand, AVIL is very sparsely expressed throughout the body and, as such, has the potential to be a therapeutic target with limited toxicity.

This led us to investigate pharmacologic targeting of AVIL as a potential therapeutic intervention for GBM. Actin-binding proteins are often considered undruggable due to their lack of commonly targeted structures, binding pockets, and active sites, such as those found in kinases and other enzymes(17). However, our work supports the feasibility of targeting AVIL with a small-molecule and demonstrates its efficacy both in vitro and in multiple in vivo models of GBM, paving the way for further development of the AVIL inhibitor and its derivatives as a targeted therapy for GBM.

RESULTS

AVIL is highly expressed in glioma and expression is correlated with worse prognosis

Previously, we have examined AVIL expression by microarrays from the REMBRANDT database and glioma RNA sequencing (RNA-seq) in The Cancer Genome Atlas (TCGA)(15, 18). Here, we confirmed high AVIL expression in gliomas compared with normal brain tissues in an independent glioma dataset, the Chinese Glioma Genome Atlas (CGGA)(19). Consistent with TCGA (fig. S1A), AVIL was elevated in the CGGA glioma samples compared with normal marginal tissue (Fig. 1A), It was expressed at high levels in GBM; it was also more highly expressed in lower grade gliomas (LGGs) without chr. 1p/19q codeletion versus those with the codeletion as well as more highly expressed in LGGs with wild-type IDH compared with LGGs with mutant IDH (Fig. 1B). AVIL expression was observed in all four molecular subtypes (classic, neuronal, proneuronal, and mesenchymal) among different GBM tumors, as well as in single cells of neural-progenitor-like (NPC-like), oligodendrocyte-progenitor-like (OPC-like), and astrocyte-like (AC-like), and mesenchymal-like (MES-like) states within the same tumors (fig. S1B to D)(20). In both CGGA and TCGA, increased AVIL expression was correlated with worse clinical outcomes (Fig. 1C and fig. S1E). The higher expression of AVIL in GBM compared with non-neoplastic tissues, grade II gliomas, and III gliomas was also observed at the protein level by immunohistochemistry (Fig. 1D and E, fig. S2).

Figure 1. AVIL is a candidate therapeutic target for GBM.

(A) Analysis of AVIL expression using RNA-seq data from CGGA comparing tumor samples (n=325) to adjacent non-neoplastic margins (n=20). (B) Analysis of AVIL RNA expression in CGGA samples broken into LGG with mutant IDH with (LGG-mt-codel) (n=45) or without 1p/19q codeletion (LGG-mt-no_codel) (n=61), LGG with wild-type IDH (LGG-wt) (n=40), and GBM (n=127). (C) Survival probability based on AVIL expression in primary (upper) and recurrent (lower) gliomas using data from CGGA (two-sided log-rank test). (D) AVIL (brown) IHC staining on non-neoplastic brain tissues, LGG tissues, and GBM tissues. Scale bar, 50 μm. (E) Quantification of IHC stain scoring and frequency performed by a pathologist. (F) Experimental scheme of a tet-on shRNA in vivo system. 2X10⁶ U87 cells expressing tet-on shAVIL were subcutaneously injected into athymic nude mice. When tumors became visible, the animals were switched to the drinking water containing doxycycline drinking on day 7 and ended on day 30 (n=6). (G) Tumors were harvested and imaged at day 30. (H and I) Tumor weight (H) and volume (I) were measured and plotted at day 30. For box plots in (A) and (B), the box represents the 25th to 75th percentile, whiskers represent the 5th to 95th percentile, and the bar in middle represents median. (J) Construction of Avil knockout mouse. (K) Genotyping PCR for Avil knockout and wild-type (WT) mice. (L) H & E staining for key organs from Avil knockout and wild-type litter mates, For (H) and (I), data are presented as mean values ± SD. Statistical significance was calculated using t-test with Bonferroni correction for multiple hypothesis testing in (A) and (B), two-sided log-rank test in (C), Cochran–Armitage test in (E), or unpaired two-tailed t-test in (H) and (I). **p < 0.01, ***p < 0.001. ****p<0.0001

AVIL silencing suppresses growth of tumor xenografts

Previously, we demonstrated that AVIL silencing via RNAi suppressed tumor formation in vivo(15). To evaluate whether AVIL may be a viable therapeutic target, we established an inducible system, where GBM cell lines U87 and A172 were stably infected with Tet-On shRNA viral constructs targeting AVIL (shAVIL3–5). Upon addition of doxycycline, GFP expression was detected (fig. S3A and B) and both AVIL RNA and protein expression was suppressed in shAVIL virally infected cells (fig. S3C to H). Consistent with our previous observations, AVIL silencing led to a reduction in cell number in the presence of doxycycline (fig. S3E and H). We then selected one shAVIL3 for in vivo evaluation. In this GBM cell line xenograft model, doxycycline was added to the water and fed to the animals at 7d post-xenograft when tumors became visible (Fig. 1F). We observed a significant reduction in tumor weight (p<0.01) and volume (p<0.01) in the shAVIL3 group versus the control shCT group (Fig. 1G to I).

To better understand the role of AVIL in a living system and to evaluate potential side effects of targeting AVIL, we generated an Avil constitutive knockout (KO) mouse using CRISPR/Cas9 to delete a key portion of the gene and induce a frame shift of the remaining coding region (Fig. 1J and K). Avil KO showed no overt effect on the animals, which appeared healthy and had no gross tissue abnormalities detected by a pathologist (Fig. 1L). We also assessed the effect of Avil on animal fertility by breeding Avil KO mice with wild-type mice, and found no difference in the litter size compared to wild-type breeding (fig. S4). These results highlight AVIL as a target with a potentially wide therapeutic window and provide the rationale for our efforts to develop inhibitors for AVIL.

AVIL displays actin binding, nucleating, severing, and bundling activities

AVIL is a member of the villin/gelsolin family, which regulates actin filament reorganization(9). Previous characterization of AVIL demonstrated that it affects cell movement and is involved in the formation of filopodia-like structures in fibroblasts(21). Additionally, AVIL plays a role in ciliogenesis(21). These findings are consistent with our observations that silencing AVIL reduced cell migration, whereas overexpressing AVIL enhanced migration(15). AVIL has been suggested to have similar actin regulatory activities as Villin(9); however, these activities remain to be been directly evaluated. Therefore, we analyzed the actin binding, severing, nucleating, and bundling activities of AVIL. We first produced recombinant AVIL protein (fig. S5A to C) and confirmed its direct binding to filamentous (F-) actin by high-speed ultracentrifugation (Fig. 2A). Next, we analyzed the severing activity of AVIL on F-actin by negative stain electron microscopy (Fig. 2B). We observed that the number of filaments visible within the field of view was more than doubled in the presence of AVIL (F-actin alone, 28 ± 5.8 per micrograph; F-actin + AVIL, 66.5 ± 14.4). Additionally, actin filaments were significantly shorter (p<0.001) (Fig. 2C) and filament count was greater in the presence of AVIL (p<0.01) (Fig. 2D and fig. S5D). Upon addition of AVIL, real-time analysis of pyrene-labeled F-actin demonstrated actin depolymerization, supporting its actin-severing capability (Fig. 2E). Similar to other gelsolin family members, AVIL also showed a concentration-dependent effect on the elevated rate of actin polymerization, confirming actin-nucleating activity (Fig. 2F). AVIL contains a headpiece domain that was previously shown to bind directly to F-actin, independently of calcium(22), and is essential for the actin-bundling activity of Villin(23). Low-speed pelleting of F-actin (Fig. 2G) and electron microscopy imaging (Fig. 2H, fig. S5E) confirmed that AVIL possesses actin bundling activity. In summary, AVIL regulates actin dynamics by binding G- and F-actin and exhibiting actin severing, nucleating, and bundling activities, thereby having the capacity to promote cell migration and cytoskeletal reorganization.

Figure 2. AVIL possesses actin binding, severing, nucleating, and bundling activity.

(A) Silver-stained SDS-PAGE analysis of high-speed pelleting of F-actin in the presence or absence of AVIL. (B) Representative negative stain electron microscopy images of F-actin incubated for 24 hours in the presence or absence of AVIL. Scale bar, 100 nm. (C) Quantification of negative stain electron microscopy images. Histogram representing F-actin filament length in the presence or absence of AVIL. Small insert shows box and whisker plot of the filament length distribution, where whiskers represent 5^th to 95th percentile; p = 6.71527×10⁻⁹⁷ by Student’s t-test. (D) The number of filaments per micrograph was quantified. (E). AVIL protein was added to pyrene-labeled F-actin and pyrene fluorescence was monitored and plotted versus time as a measure of actin depolymerization activity. (F) AVIL protein was added to pyrene-labeled G-actin before polymerization. Pyrene fluorescence was monitored and plotted versus time to measure actin nucleation activity. (G) Silver-stained SDS-PAGE analysis of low-speed pelleting of F-actin in the presence or absence of AVIL. (H) Representative images from low-speed pelleting of F-actin in the presence of AVIL are shown to demonstrate actin bundling. Actin bundles are indicated by white arrows. Scale bar = 100 nm. (I) Wild-type AVIL (AVIL WT), the K808C mutant, or the F819C mutant was added to pyrene-labeled G-actin before polymerization. Pyrene fluorescence was monitored and plotted versus time. (J) AVIL WT, K808C, or F819C was added to pyrene-labeled F-actin, and pyrene fluorescence was monitored and plotted versus time. (K) AVIL WT, K808C, or F819C was incubated with varying concentrations of pyrene-labeled F-actin, and pyrene fluorescence was measured. Actin critical concentration (CC) was calculated as an intersection between linear interpolations of G-actin and F-actin fluorescence measurements. (L) Silver-stained SDS-PAGE analysis of low-speed pelleting of F-actin in the presence or absence of AVIL WT, K808C, or F819C with EGTA. (M) Representative negatively stained electron micrographs from (L). White bar = 50 nm. (N) Counts of stable astrocyte cell lines overexpressing AVIL WT, K808C, or F819C mutants three days post-plating. (O) Total movement of astrocyte cell lines expressing K808C or F819C mutants 6 hours after wound-healing experiment. CT, empty vector control. (P) 3×10⁶ stable astrocytes overexpressing AVIL WT or the two headpiece mutants were injected into athymic nude mice to form tumors (n=7). Tumors were harvested at the endpoint and imaged. (Q and R) Tumor weight (Q) and volume (R) were measured and plotted. (S) Stable astrocyte cell lines overexpressing AVIL WT or the RRI, RKR, E71L, or Y458D AVIL mutants were quantified at three days post-plating. (T) Total movement of astrocyte cell lines expressing AVIL WT or the indicated mutants in the wound-healing experiment. (U) 3×10⁶ stable astrocytes overexpressing AVIL WT or the indicated mutants were injected into athymic nude mice to form tumors (n=8). Tumors were harvested and imaged at the endpoint. (V and W) Tumor weight (V) and volume (W) were measured and plotted. (X) Stable astrocyte cell lines overexpressing AVIL WT or the double and triple mutants were injected into athymic nude mice to form tumors (n=5). Tumors were harvested and imaged at the endpoint. (Y and Z) Tumor weight (Y) and volume (Z) were measured and plotted. Mean ± SD are shown. For (E and F) and (I to K), a.u. indicates arbitrary units. Data were analyzed by unpaired two-tailed t-test in (C) and (D), or one-way ANOVA with Dunnett’s multiple-comparisons test in (N), (O), (Q), (R), (S), (T), (V), (W), (Y) and (Z). Comparisons are shown relative to AVIL-WT. ns, p > 0.05, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Loss of actin bundling activity reduces AVIL’s oncogenic effect

The headpiece domain of Villin was previously shown to be crucial for actin binding and bundling(23–26). AVIL also contains a C-terminal headpiece domain. Previously, we have identified two headpiece domain mutants (AVIL K808C and AVIL F819C) that disrupt the binding of AVIL to actin(15). Accordingly, both mutants failed to induce the expression of LIN28B, a downstream target of AVIL(15). We examined the effects of the two mutants on various actin dynamics. Firstly, actin polymerization was not affected by these mutations (Fig. 2I), indicating that nucleation through the calcium-dependent core domains was not disrupted. Secondly, the mutants were not defective but rather more potent in inducing actin depolymerization (Fig. 2J). We then analyzed the effect of AVIL on actin critical concentration. Almost no differences were observed between the headpiece mutants and wild-type AVIL (Fig. 2K), indicating that actin levels reach similar equilibrium regardless of mutations. This suggests that the ability to cap the actin barbed ends is comparable between wild-type AVIL and the headpiece domain mutants. Finally, we analyzed the ability of AVIL to bundle actin filaments. AVIL headpiece mutants displayed low actin-bundling activity (Fig. 2L and M). Assuming that actin bundling and severing activities of AVIL are exclusive, the lack of bundling activity could explain the elevated actin depolymerization rate for the two mutants, as AVIL is presumed to remain bound to actin barbed ends after severing.

To test the oncogenic ability of these AVIL headpiece domain mutants, we established a stable AVIL-overexpressing astrocyte cell line. We first confirmed similar expression of mutated AVIL to the wild-type AVIL control by immunoblotting (fig. S5F). The mutant-transfected cells showed reduced growth rates compared with the cells expressing wild-type AVIL (p<0.05); migrative ability was also impaired in the F819C mutant cell line (Fig. 2N and O). We then implanted the AVIL-overexpressing astrocyte cell line into mice to measure tumorigenesis in vivo. We observed a consistent decrease in tumor take rate in both AVIL bundling mutants (Fig. 2P). Additionally, decreases in both tumor weight and volume were demonstrated (p<0.01) (Fig. 2Q and R), suggesting that defects in actin bunding can reduce the tumorigenic activity of AVIL.

To evaluate the effect of other actin-modulating activities of AVIL, we designed and constructed a series of additional AVIL point mutations in core gelsolin-like (G-) domains. These mutations were based on previously established mutations in other gelsolin family members, including: an AVIL “RRI” mutation (AVIL ⁸³DDYLGG to NTLLGE, ¹²²GMKHV to AMHKTS) in the G1 domain and a RKR to AAA mutation (AVIL^{R135A,K142A,R143A}) in the G2 domain, predicted to specifically disrupt actin severing activity(27),(28); an AVIL^Y458D mutation in the G4 domain, predicted to disrupt actin nucleating activity(29); and an AVIL^E71L mutation in the G1 domain, predicted to disrupt both actin severing and capping(30). We then established astrocyte cell lines stably overexpressing these mutants (fig. S5G). Similar to the two headpiece mutants, all mutants that affected actin dynamics resulted in reduced oncogenicity, evidenced by reduced cell proliferation (p<0.05) (Fig. 2S), motility (p<0.05) (Fig. 2T), and tumor formation (p<0.01) (Fig. 2U to W). We then established astrocyte lines overexpressing double mutants (RRI+Y458D, RRI+E71L) or a triple mutant (RRI+E71L+Y458D). Compared with cells expressing wild-type AVIL, they all showed significant reduction in cell proliferation (p<0.05), movement (p<0.05), and tumor formation (p<0.0001) (Fig. 2X to Z, fig. S6). These results provide rationale to suppress the oncogenic effect of AVIL by disrupting its regulation of actin dynamics.

A small-molecule screen revealed AVIL-interacting compounds

We used a small-molecule microarray platform(31, 32) to identify putative binders of the AVIL protein. Approximately 50,000 small molecules, including commercially available compounds, products of diversity-oriented synthesis, and known bioactive compounds, were screened with recombinant AVIL protein (fig. S7A). From these small molecules, we identified 90 AVIL-interacting compounds (Fig. 3A). We then conducted step-wise cellular assays to screen candidate compounds based on their half-maximal inhibitory concentration (IC₅₀ values, beginning with the GBM cell line U87, then three additional GBM cell lines, (U251, A172, and TG98) and immortalized astrocytes (fig. S7B). From three prime candidates, we selected one compound, Compound A (Fig. 3B) for further investigation due to its wide difference in IC₅₀ between tumor cell lines and astrocytes. Compound A effectively killed four GBM cell lines, U87, A172, U251, and T98G, at lower IC₅₀ values (10–30 μM) than the astrocyte cell line (>120 μM) (Fig. 3C). For comparison, we used the current standard therapeutic compound, temozolomide (TMZ), on the same cell lines. Consistent with other reports(33, 34), TMZ was not very effective in vitro, with IC₅₀ values ranging from approximately 500 μM to >1 mM on the four GBM cell lines, and a similar IC₅₀ on the astrocyte line to TMZ-resistant T98G cells (Fig. 3D). Combination of Compound A and TMZ showed similar efficacy on GBM cell lines to Compound A alone (Fig. 3E, fig. S7C to E). Representative brightfield images of Compound A treated cells showed that cells shrunk and died (Fig. 3F). Consistently, crystal violet staining of cell viability showed that Compound A killed U87 cell cultures at 10 μM, but TMZ was not effective even at 400 μM (Fig. 3G). In colony formation assays, Compound A did not affect astrocyte viability until the highest dose (80 μM), whereas TMZ had a deleterious effect even at the lowest dose (200 μM) (Fig. 3H). Similar effects of Compound A and TMZ were observed on A172 and U251 cells (fig. S7F and G).

Figure 3. Compound A is an AVIL-interacting compound.

(A) Small-Molecule Microarray screen. Assay positives with a Robust Z score greater than 3-σ from the median are highlighted. Orange dots indicate the top three candidates. (B) Chemical structure of Compound A. (C) IC₅₀ values of Compound A on four GBM cell lines and astrocytes. (D) IC50 values of TMZ on four GBM cell lines and the astrocyte cell line. Data in (C) and (D) graphed as % of solvent (DMSO) control treated cells. (E) Cell viability rate measured by MTT assay at 3 days post-treatment of Compound A, TMZ, or combination. (F) Representative images of astrocytes, U251, and U87 treated with Compound A at 20 μM. (G and H) Crystal violet staining of U87 cells (G) and the astrocyte cell line (H) treated with DMSO solvent and various concentration of Compound A, TMZ, or combination. (I) Thermal shift assay. Graphs represent negative of the first derivatives of melting curves between 30 and 50°C of AVIL incubated with increasing concentrations of Compound A. Small inserts represent full melting curves (20 to 90°C). Melting temperature (T_m) was determined as the lowest point of the curve. (J) Microscale thermophoresis assay was conducted to evaluate the interaction of recombinant AVIL with Compound A and the K_d value was calculated. Y axis is the difference in normalized fluorescence [‰]. (K) Immunoblot assaying protein expression of AVIL, FOXM1 and LIN28B in A172 and U87 cells treated with DMSO or Compound A. GAPDH is the loading control. (L) LIN28B gene expression was measured in A172 and U87 cells treated with control DMSO or with Compound A by qRT-PCR. For qRT-PCR, AVIL expression was normalized to GAPDH. Plotted is the fold change in Compound A group relative to the DMSO control group. (M and N) Mice bearing U87 xenografts (n=10) were treated with Compound A or DMSO control and tumor volume (M) and tumor weight (N) were measured. Data are presented as mean values ± SD in (C to E) and (L to N). Statistical significance was calculated using one-way ANOVA with Tukey’s post-hoc multiple-comparisons test in (E) or unpaired two-tailed t-test in (L), (M), (N). ns, p > 0.05, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Thermal shift assays were conducted to confirm binding of Compound A with recombinant AVIL protein. We observed dose-dependent destabilization after Compound A binding to AVIL (Fig. 3I). Microscale thermophoresis was used as a secondary assay for evaluating binding potential and revealed the K_d of Compound A was around 6 μM (Fig. 3J). Previously, we demonstrated that AVIL regulates FOXM1 which, in turn, regulates the expression of LIN28B(15). Consistently, loss of AVIL with siRNA caused a substantial decrease in the expression of FOXM1 and LIN28B after silencing, whereas overexpression of AVIL resulted in increased expression of FOXM1 and LIN28B (fig. S7H). To assess whether the small-molecule inhibitor of AVIL had the same effect, we repeated the immunoblotting with both Compound A treated A172 and U87 cells and saw a reduction of FOXM1 and LIN28B expression after treatment, whereas abundance of AVIL itself was not affected (Fig. 3K). Consistently, LIN28B was also down-regulated at RNA level upon treatment with Compound A (Fig. 3L). We then conducted a pilot test on U87 xenografts subcutaneously injected into athymic nude mice. With the limited volume of compound, we performed three intratumoral injections (fig. S7I) and found a significant decrease in overall tumor volume (p<0.05) and weight (p<0.05) after Compound A treatment compared with vehicle control (Fig. 3M and N, fig. S7J).

Compound A exhibits on-target effects

A critical aspect of characterizing a small-molecule inhibitor is the evaluation of on- and off-target effects and their contribution to the overall changes observed following treatment. Therefore, we characterized transcriptome changes induced by Compound A treatment in U87 cells and compared these results to changes induced by loss of AVIL using siRNA knockdown. Both approaches showed similar overall changes in gene ontology (GO) terms, with common perturbations in cell cycle, mitosis, centromere, and cytoskeleton-related genes being down-regulated; in contrast, genes related to apoptosis and stress response were up-regulated (Fig. 4A and B). Consistently, Gene Set Enrichment Analysis (GSEA) revealed that differentially expressed genes in Compound A treated cells were enriched among the gene sets of siAVIL targets (Fig. 4C). We have previously identified, through transcriptomic profiling, a gene set associated with AVIL perturbation and FOXM1 appeared to be strongly associated with AVIL oncogenic function(15). GSEA analysis of ranked genes showed the similar enrichment of FOXM1 targets in Compound A treated cells, similar to siAVIL transfected cells (Fig. 4D).

Figure 4. Compound A exhibits on-target activity.

(A) Shown are gene ontology terms enriched in the genes down-regulated by Compound A treatment or siRNA targeting AVIL in U87 cells. Plotted is statistical significance (-Log₁₀(p-value)) of each term. (B) Shown are gene ontology terms enriched in the genes up-regulated by Compound A treatment or siRNA targeting AVIL in U87 cells. (C and D) Gene set enrichment analyses of Compound A down-regulated genes for the gene set down-regulated by siAVIL (C) or FOXM1 target gene set (D). FDR, False Discovery Rate; NES, Normalized Enrichment Score. (E) HEK293T cells were transfected with a Myc-tagged AVIL expressing plasmid. Cells were then treated with Compound A or DMSO solvent control. Immunoprecipitation using Myc antibody was performed. Immunoblotting using β-actin was performed. (F) Astrocyte cell lines overexpressing empty vector control (CT) or AVIL were treated with 20 μM Compound A. Representative images were taken at 72 hours following treatment. (G) IC₅₀ values of Compound A on cell lines of astrocytes overexpressing empty vector control (CT) or AVIL. (H) Immunoblotting was used to determine AVIL deletion in clone 4 and 11 by CRISPR/Cas9 in U87 cells. (I) Crystal violet staining of these clones and control sgRNA transfected cells, as well as parental U87 cells, treated with 20 μM Compound _A. (J) IC₅₀ values of Compound A on wild-type (WT) and AVIL KO U87 cells (clone 11). Data in (G) and (J) are presented as mean ± SD.

To measure AVIL’s interaction with its substrate, β-actin, we then transfected HEK293T cells with Myc-tagged AVIL and immunoprecipitated AVIL by Myc tag. Following the compound treatment, we observed only low amounts of β-actin protein interacting with AVIL (Fig. 4E). Moreover, we evaluated Compound A on the wild-type astrocyte cell line and the astrocyte cell line overexpressing AVIL to demonstrate specificity. In contrast to astrocytes expressing an empty vector (IC₅₀ value of 97 μM), astrocytes with AVIL overexpression were sensitive to the Compound A at an IC₅₀ value of 28 μM, with most cells eradicated at 20 μM (Fig. 4F and G).

One of the most robust ways to demonstrate targeting specificity is to simply remove the target and test the sensitivity of the cells. If the compound is on-target, it should no longer function at therapeutic doses(35). We used CRISPR/Cas9 to create AVIL knockout clones of U87 cells (Fig. 4H). As previously demonstrated, most cells with AVIL deletion died immediately. As such, we were unable to establish long-term stable cells with AVIL deletion. However, in a transient system, we saw complete loss of efficacy of the Compound A at 20 μM and a right shift of IC₅₀ values in clone 11 that had AVIL homozygous deletion (Fig. 4I and J). Together, these experiments suggest that Compound A elicits on-target effects on cell viability.

Compound A exhibits anti-tumor efficacy in vivo

We scaled up the synthesis of Compound A, which allowed us to assess additional animal models. To evaluate the performance of Compound A in vivo, we first performed subcutaneous implantation of U87 GBM cell line in athymic nude mice. After 14 days of tumor growth, we began treatment with the Compound A at 50 mg/kg every other day for 20 days by intraperitoneal (i.p.) injection (Fig. 5A). We observed a significant reduction in overall tumor size (p<0.001) and volume (p<0.001) after Compound A treatment (Fig. 5B to D).

Figure 5. Compound A exhibits efficacy against GBM models in vivo.

(A) 2X10⁶ U87 cells were injected subcutaneously into athymic nude mice. On day 14, 50 mg/kg Compound A or solvent control was delivered by i.p. injection every other day for 20 days (n=8). (B) Tumors were harvested and imaged at the end point. (C) Tumor volumes were monitored by caliper and plotted along the treatment time window. (D) Tumor volumes and weights were plotted at the endpoint (Day 34). (E) Syngeneic CT2A cells expressing luciferase (2X10⁵) were injected intracranially. On day 14, 2 mg/kg Compound A or control solvent in 5 μl was delivered into the injection site (n=7). The intracranial injection treatment was repeated again on day 18, and the experiment was stopped on day 60. (F) Bioluminescence imaging was performed before the treatment on day 14 and seven days after treatment began. (G) Bioluminescence signal was measured with the intensity of the light emitted, normalized by the area and solid angle. (H) Animal survival was plotted on a Kaplan-Meier curve. (I) 80 mg/kg Compound A was administered to three mice orally. At 30 minutes following treatment, blood, brain, and CSF were collected. Compound A was measured by mass spectrometry. Measurement of three mice were averaged and plotted. (J) Similar to (E), 2X10⁵ CT2A cells were injected intracranially. On day 14, 50 mg/kg Compound A or solvent control was administered orally every day for ten days. The experiment was stopped on day 60 (n=8). (K) Bioluminescence images from before the treatment on day 14 and seven days after treatment began are shown. (L) Bioluminescence from (K) was quantified and plotted. (M) Animal survival from (J) was plotted on a Kaplan-Meier curve. (N) IHC staining for Ki67 and cleaved PARP were performed on tumors harvested from (J). Shown are representative images. (O) Quantification of the frequency of Ki67⁺ or cleaved PARP⁺ cells out of total cells in the field from (N). Three mice with 8 to 10 images from each mouse were used. For (C), (D), (I), and (O), mean ± SD are shown. Statistical significance was calculated using two-way repeated-measures ANOVA with Geisser–Greenhouse correction and Šídák’s multiple-comparisons test in (C), two-way repeated-measures ANOVA with Šídák’s multiple-comparisons test in (G) and (L), unpaired two-tailed t-test in (D) and (O), two-sided log-rank (Mantel–Cox) test in (H) and (M), or one-way ANOVA with Tukey’s post-hoc multiple-comparisons test in (I). ns p > 0.05, *p < 0.05, **p < 0.01, ***p < 0.001. Artwork created using BioRender.com

We next tested the therapeutic efficacy of Compound A in a CT2A syngeneic mouse model for GBM. We first measured the effect of AVIL silencing in CT2A cell line in vitro, where we observed a reduction in cell proliferation (fig. S8A to C). Consistently, Compound A treatment eradicated CT2A cell cultures (fig. S8D). We then intracranially inoculated CT2A cells into immunocompetent C57BL/6 mice and allowed the tumor to grow for 14 days before treatment began. The compound was intracranially administered twice through the same boreholes at 2 mg/kg in 5 μl volume (Fig. 5E). CT2A cells were transduced with luciferase, allowing us to monitor the tumor growth by bioluminescence imaging (Fig. 5F). Quantitation of bioluminescence demonstrated a gain in signal from tumor cells in all control tumors, and loss or stabilization of signal in mice treated with Compound A (Fig. 5G). Overall survival was also improved in the group treated with Compound A compared with control-treated mice (Fig. 5H).

Compound A can be delivered orally and crosses the BBB

As previously mentioned, the primary chemotherapeutic administered as part of the standard of care for GBM is TMZ. Despite poor efficacy in vitro, it is used because of its capability of crossing the blood-brain barrier (BBB). By prediction, Compound A has an AdaBoost-MACCSFP BBB score of 4.9, and an SVM_MACCSFP BBB Score of 0.026, both predicting capability of crossing the BBB. SwissADME prediction also predicted Compound A to be BBB penetrant, meeting Lipinski’s rule of five and having good drug-like properties (fig. S9A). We performed an MDCK-MDR1 permeability assay and found that Compound A exhibited good permeability in both directions (fig. S9B). Additionally, we conducted pharmacokinetics analyses on Compound A and found that it has a T_1/2 of approximately 2 hours by both oral gavage (PO) and intravenous injection (IV) (fig. S9C to E). Given that the T_max by PO was around 30 minutes, we collected blood, brain tissue, and cerebrospinal fluid (CSF) 30 minutes following oral gavage. We detected enriched signal of Compound A in the brain tissues with very little in the CSF (Fig. 5I), confirming BBB penetration of Compound A.

We then tested the effect of Compound A by repeating the syngeneic, intracranial grafting of CT2A cells and initiating treatment at 14 days after tumor inoculation, but through oral gavage of 50 mg/kg dosed daily for 10 days (Fig. 5J). Although not as potent as intracranial delivery, oral administration of the compound significantly reduced the change in tumor growth compared with the control group (p<0.001) (Fig. 5K and L). Additionally, overall survival was significantly extended in the group treated with Compound A (p<0.001) (Fig. 5M). We then harvested tumor tissues and performed immunohistochemistry (IHC). We observed a significant decrease in positivity for the Ki67 proliferation marker as well as an increase in cleaved PARP positivity, a marker of apoptotic cells, in the group treated with Compound A (p<0.01) (Fig. 5N and O).

Furthermore, using a U87 intracranial xenograft model, we tested the effect of Compound A, TMZ, and their combination. Fourteen days after tumor cell injection, we began the treatments of Compound A by oral gavage (50 mg/kg), TMZ by i.p. injection (15 mg/kg), or combination of both (fig. S10). The treatments were stopped on day 40. A survival benefit was observed in all treatment groups compared with control, although no further benefit was seen when Compound A was combined with TMZ versus Compound A or TMZ alone.

Compound A demonstrated efficacy in a GSC tumor model

Similar to other cancer types, GBMs harbor a subpopulation of glioblastoma initiating/stem cells (GIC/GSCs) that govern tumor initiation, maintenance, and recurrence after therapy(36–38). GSCs partly mediate resistance to chemo and radiotherapy and have become vital targets for the reversal of chemo-resistance(39). Although difficult to culture, GSCs are more representative of disease progression in patients. We first examined AVIL expression in multiple GSC cell lines(40). Quantitative polymerase chain reaction (qPCR) demonstrated significantly higher expression of AVIL in GSCs compared with the normal astrocyte cell line and with the GBM cell line A172 (p<0.01) (Fig. 6A). AVIL expression was also confirmed at the protein level by immunoblotting (Fig. 6B, fig. S11A). siRNA-mediated silencing and Compound A inhibition of AVIL in GSC11 and GSC627 cell lines significantly reduced the expression of common stem cell markers, SOX2, OLIG2, CD133, and LIN28B, and increased the expression of differentiation markers GFAP and CLDN11 (p<0.001) (Fig. 6C to F). We grew GSC11 and GSC627 neurospheres in 3D cell cultures and compared Compound A performance on GSCs versus neural stem cells (NSCs). We observed an IC₅₀ in GSCs (6 μM) that showed around nine-fold lower than the NSC control cells (55 μM) (Fig. 6G) and an overall reduction in the number of neurospheres and cells in culture, whereas no change was observed with the NSC control cells (Fig. 6H to M). The increased sensitivity of GSCs to Compound A compared with GBM cells is consistent with their elevated AVIL expression. The effects of Compound A were further confirmed in two additional GSC lines, GSC262 and GSC267 (fig. S11B and C).

Figure 6. Compound A is efficacious against GSC tumor models in mice.

(A) qRT-PCR measuring AVIL RNA expression in eight GSC lines compared with astrocytes and GBM line A172. (B) Immunoblot detected AVIL protein expression in GSCs, astrocytes, and A172. (C to F) qRT-PCR measuring AVIL, stem cell markers (SOX2, OLIG2, CD133, and LIN28B), and differentiation markers (GFAP, Claudin 11, and Tuj1). (C and D) GSC11 (C) and GSC627 (D) were transfected with shRNAs targeting AVIL or control shRNA. (E and F) GSC11 (E) and GSC627 (F) were treated with Compound A at 10 μM or DMSO control for 48 hours. (G) IC₅₀ values of Compound A on NSC, GSC11, and GSC627. (H) Neurosphere imaging showing GSC11 and GSC627 treated with Compound A or DMSO control. (I to M) Shown is quantification of neurospheres in GSC11 (I), viable cell numbers in GSC11 (J), quantification of neurospheres in GSC627 (K), viable cell numbers in GSC627 (L), and viable cell numbers in NSC (M) (n=5). (N) 1X10⁵ GSC11 cells were injected intracranially. On day 38, the first MRI was performed to detect tumors, and treatment started with 50 mg/kg Compound A or solvent control given orally daily until day 51. The second MRI was performed on day 45 (n=5). (O) Tumor volume was quantified and plotted. (P) Representative MRI images were shown. Tumor areas were highlighted by a red circle. (Q) Animal survival was plotted on a Kaplan-Meier curve. For qRT-PCR, AVIL expression was normalized to GAPDH. For (A), (C to F), and (I to M) mean ± SD are shown. Statistical significance was calculated using one-way ANOVA with Dunnett’s multiple-comparisons test in (A), (C) and (D), unpaired two-tailed t-test in (E), (F), (I), (J), (K), (L) and (M), two-way repeated-measures ANOVA with Šídák’s multiple-comparisons test in (O), or two-sided log-rank (Mantel–Cox) test in (Q). ns, p > 0.05, *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. Artwork created using BioRender.com

We then intracranially implanted GSC11 cells and treated with Compound A at 50 mg/kg daily by oral gavage after the first positive signal was detected by magnetic resonance imaging (MRI) (Fig. 6N). We saw tumor growth stabilization and a significant lower tumor volume in the mice treated with Compound A compared with vehicle control (p<0.05) (Fig. 6O). MRIs conducted at early and late points showed a gain in tumor volume of GSC xenografts when treated with the control but limited progression after treatment with the compound except for one (Fig. 6P). Consistently, we observed an extension in survival in the Compound A-treated mice (Fig. 6Q).

Compound A is effective on PDXs and TMZ-resistant tumors

Patient-derived xenografts (PDX) are a valuable tool for characterizing human disease in mice. We measured AVIL mRNA expression in 19 PDX models obtained from Mayo Clinic(41), discovering a broad spectrum of elevated expression in most models compared with the control astrocyte cell line (p<0.05) (Fig. 7A). High AVIL expression was confirmed by immunoblotting (Fig. 7B). Compound A effectively killed PDX cultures with IC₅₀ values around 10 μM (fig. S12). TMZ resistance is one of the most challenging aspects of GBM treatment and is commonly seen in the clinic(42, 43). We sought to address whether pharmacologic inhibition of AVIL would still show efficacy after TMZ resistance was acquired. Clinical samples from patients with or without TMZ treatment exhibited no obvious difference in the level of AVIL expression (fig. S13A). We then obtained GBM39 and GBM12 TMZ-sensitive and TMZ-resistant PDX samples from Mayo Clinic (44) (there are two different strains of TMZ resistant GBM12 cells). In two PDX systems, the TMZ-resistant cells exhibited increased AVIL mRNA expression (p<0.05) (Fig. 7C) as well as AVIL protein levels (Fig. 7D, fig. S13B and C), compared with their TMZ-sensitive counterparts. Consistently, TMZ-resistant cells were more sensitive to AVIL inhibition by Compound A treatment than their TMZ-sensitive parental PDX (Fig. 7E). Similar observations were observed in two independent derivatives of GBM12 TMZ-resistant PDXs, GBM12_TMZ 2096 and GBM12_TMZ 3080 (fig. S13D).

Figure 7. Compound A demonstrates antitumor activity against GBM PDX models.

(A) AVIL RNA expression in 19 PDX samples compared to astrocytes and GBM cell line A172. (B) Immunoblot-detected AVIL protein expression in PDXs, an astrocyte cell line, and A172 cells. (C and D) AVIL RNA measured by qRT-PCR (C) and AVIL protein measured by immunoblot (D) in GBM12, GBM39, and corresponding TMZ-resistant cell lines, GBM12_TMZ and GBM39_TMZ. (E) IC₅₀ values of Compound A on GBM39 and GBM39_TMZ. (F) GBM39 and GBM39_TMZ PDXs were implanted subcutaneously into athymic nude mice. On day 3, treatment began with 50 mg/kg Compound A or solvent control delivery orally on a five-days on / two-days off schedule. The experiment was terminated on day 48 (n=10). (G to J) GBM39 (G and H) and GBM39_TMZ (I and J) tumors were harvested at the endpoint, imaged (G and I), and measured (H and J). (K) Blood was collected after treatment with DMSO solvent control or Compound A at the last day (day 48) of the experiment. Representative blood count features are plotted. WBCB indicates total white blood cell count (WBC) determined by basophil/lobularity method; WBCP indicates WBC count determined by peroxidase staining; PDW indicates platelet distribution width; RBC indicates red blood cell count; PLT indicates platelet concentration. (L) Organs were harvested after treatment with DMSO solvent control and Compound A for on the last day of the experiment. Histological examination was performed by a pathologist. Representative H&E slides are shown. Scale bar= 50 μm. (M) Maximum tolerance dose assay with body weights (g) from mice treated with DMSO solvent control and Compound A at 50 to 200 mg/kg treated by daily oral gavage for seven days. Animal weights were monitored and plotted (n=3). For (A), (C), (E), (H to K) and (M) mean ± SD are shown. Statistical significance was calculated using one-way ANOVA with Dunnett’s post-hoc test in (A), unpaired two-tailed t-test in (C), (H), (J) and (K), or two-way repeated measures ANOVA with Dunnett’s post-hoc test comparing each dose group with the vehicle control at each time point in (M). ns, p > 0.05, *p < 0.05, **p < 0.01, ****p < 0.0001. Artwork created using BioRender.com

TMZ-sensitive and TMZ-resistant GBM39 cells(44) were subcutaneously injected into athymic nude mice, grown for 3 days, and then treated with either vehicle or 50 mg/kg Compound A by oral gavage for 5 days on/2 days off (Fig. 7F). In the TMZ-sensitive parental PDX, we saw a significant decrease in tumor weight and volume (p<0.01) (Fig. 7G and H). A similar reduction in tumor weight and volume was also observed in the TMZ-resistant PDX model (p<0.05) (Fig. 7I and J). Consistently, when we performed immunohistochemistry (IHC) on the tumors, we found significant decreases in signal for Ki67 staining in both GBM39 and GBM39_TMZ groups treated with Compound A (p<0.0001) (fig. S14A and B). Positive signal for the AVIL downstream marker, FOXM1, was also significantly reduced in both GBM39 and GBM39_TMZ groups treated with Compound A (p<0.05) (fig. S14C and D). We then repeated the experiment with intracranial xenografts of GBM39 and GBM39_TMZ cells (fig. S15). Similar results were seen in this orthotopic model, suggesting that GBM39_TMZ is at least as sensitive to Compound A as GBM39.

Throughout these animal studies, we did not observe any toxicities. We analyzed complete blood count (CBC) and found no difference in Compound A treated versus solvent treated mice (Fig. 7K). We also harvested vital organs, including the brain. Pathologist analysis of necropsy samples found no indication of organ toxicity after treatment (Fig. 7L). Body weights throughout the treatment and all blood counts at the end point were also normal (fig. S16). In a maximum tolerated dose experiment, we increased the daily dose to 200 mg/kg for one week and observed no changes to mouse body weight, even at the highest dose of the compound (Fig. 7M). We repeated the toxicity study in a C57BL/6 immunocompetent mouse strain, and found no abnormalities in CBC or serum chemistry (fig. S17).

AVIL has been reported to be expressed in peripheral sensory neurons(45). To assess the potential off-tumor on-target toxicity of the compound on sensory neurons, we isolated primary dorsal root ganglion sensory neurons from mice, and exposed them, along with U87 glioblastoma cells, to 20 μM Compound A. After 48 hours of treatment, live-cell imaging documented preserved neuronal morphology and viability in cultured sensory neurons, whereas U87 cells displayed progressive morphological changes, culminating in cell death (fig. S18). These observations suggest a therapeutic window for AVIL inhibition, with robust activity against tumor cells and minimal acute toxicity to sensory neurons.

DISCUSSION

GBM presents particular challenges for therapeutic development. Due to the unique nature of the brain, including its BBB protection and the fact that it is poorly infiltrated by the immune system, new directions and targets need to be explored to better our outcomes in GBM management. We propose AVIL represents a favorable target for innovative therapeutic strategies. As a target, AVIL has several attractive features. First, It is highly expressed in GBM compared with healthy brain tissue. We have previously reported that AVIL expression is enriched in a large portion of GBMs(15). Here, we further confirmed high AVIL expression in GBM, including all four molecular subtypes and four single cell states(20). Moreover, AVIL is even more highly expressed in GSCs and high expression was retained in TMZ-resistant samples. Additionally, GBMs are oncogene-addicted to AVIL. Our previous study has shown that AVIL alone is sufficient to transform immortalized astrocytes, and it seems to be a merging point for three main signaling pathways in GBMs(15). Silencing AVIL is sufficient to kill GBM cells and prevent in vivo xenograft formation. Finally, targeting AVIL appears safe. We previously observed no effect of AVIL silencing on normal astrocytes or on neural stem cells(15). Here, we further demonstrated that AVIL is a safe target in that Avil knockout mice are healthy and fertile.

Not all oncogenes are druggable with small-molecule inhibitors. In this work, we detailed a first-in-class inhibitor of AVIL in GBM cells and accumulated evidence supporting that the compound is on target. Compound A selectively killed GBM cells at concentrations ranging from 10 to 20 μM, with little effect on astrocytes at this dose. It also preferentially killed GSCs at concentrations of approximately 5 μM, sparing neural stem cells, which minimally express AVIL. Mechanistically, we found that Compound A blocked AVIL’s interaction with F-actin and that AVIL targets, LIN28B and FOXM1, were suppressed after treatment similar to siRNA-induced AVIL silencing. Likewise, treatment with Compound A resulted in similar whole transcriptome changes to those observed upon AVIL silencing. Moreover, cells with higher AVIL expression were more sensitive to Compound A. Finally, CRISPR/CAS9 deleted AVIL clones of GBM cells were resistant to Compound A.

Previously, we identified AVIL activates FOXM1-LIN28B pathway. FOXM1 is a well-known transcription factor critical for cell proliferation. It is aberrantly expressed in the majority of human cancers and has also been implicated in poor prognosis(46–50), leading to its designation as the 2010 Molecule of the Year(51). LIN28B is an RNA-binding protein that negatively regulates the biogenesis of the let-7 family of tumor-suppressor microRNAs. Even though small molecule inhibitors have been developed for FOXM1 and LIN28B, targeting AVIL offers more specificity to cancer cells, offering an advantage over inhibitors for FOXM1 and LIN28B inhibitors. FOXM1 and LIN28B play important roles in normal physiology. For example, homozygous deletion of Foxm1 in mice is lethal(52–54) and conditional deletion of Foxm1 in various murine cell types has been shown to inhibit cell proliferation(54–56); Similarly, Lin28a/b-deficiency leads to dwarfism in mice(57). To that end, existing small molecule inhibitors for FOXM1 or LIN28B have been shown to be toxic, are often non-specific, and do not have “clinically” relevant potency(51). Finally, existing FOXM1 or LIN28B inhibitors are not ideal for GBM treatment due to of their poor penetration across the BBB(58, 59).

Our study has some limitations. The AVIL protein structure has not been solved and an AlphaFold predicted model did not agree with our mutagenesis results, preventing further molecular docking efforts. The IC₅₀ of Compound A is still at micromolar range making it unlikely to be directly used in the clinic. More potent derivatives are being developed. Even though we have tested the effect of the compound on peripheral sensory neurons, no in vivo model was used to test the compound’s effect on mechano-sensing and pain-sensing Future studies should examine these effects to assess on-target side effects.

Nonetheless, Compound A demonstrated efficacy in multiple in vivo models of GBM, including subcutaneous U87 xenografts, CT2A syngeneic intracranial tumors, GSC intracranial xenografts, and subcutaneous and intracranial TMZ-sensitive and TMZ-resistant PDX xenograft models. Moreover, Compound A exhibited IC₅₀ values 10- to 100-fold lower than TMZ and offers a large therapeutic window. Importantly, Compound A can cross the BBB and can be delivered orally. Besides its efficacy in the multiple animal models studied here, we did not observe any toxicity or abnormality at the tissue level or in complete blood counts. Even at the maximal dose of 200 mg/kg daily for seven days, we did not observe any effect on animal body weights. Though work needs to be done to develop more potent compounds for clinical use, the identification and evaluation of Compound A as a small molecule inhibitor for AVIL have provided us confidence that targeting AVIL is a promising therapeutic approach for the devastating disease of GBM.

Materials and Methods

Study design.

The objectives of this work were twofold: to determine whether the oncogene AVIL represents a suitable therapeutic target, and to generate the first small-molecule inhibitor directed against it. To begin, we assessed AVIL upregulation in glioblastoma by combining bioinformatic analyses with immunohistochemistry. To explore potential toxicities associated with AVIL loss, we generated Avil knockout mice and monitored physiological consequences. Next, we confirmed AVIL’s role in regulating F-actin organization using biochemical assays and electron microscopy. By examining a panel of AVIL mutants, we found that alterations disrupting any aspect of actin dynamics consistently diminished AVIL’s oncogenic activity. These observations provided the rationale for conducting a small-molecule microarray screen aimed at identifying compounds capable of binding recombinant AVIL. The highest-ranking candidate from this screen (referred to as Compound A) was subsequently evaluated for both functional efficacy and target engagement in multiple cell lines. We further assessed its therapeutic potential and safety profile across several mouse models. Animals were allocated to groups at random; blinding was not employed. Sample sizes were informed by previous studies and preliminary data. All cell-based experiments were independently replicated at least three times, and animal cohorts consisted of a minimum of five animals per group. Data points from individual experiments or animals are displayed as distinct dots in the figures to illustrate biological variability, with replicate numbers detailed in figure legends.

TCGA and CGGA analyses.

For TCGA and CGGA survival modeling, raw STAR counts were downloaded from the CGGA website(18, 19, 60) and TCGAbiolinks was used for TCGA data(61). Counts were normalized with limma and converted to z-score to ensure different data sets were comparable. Overall survival was plotted for individual genes using the R packages survival, survminer, with figures generated using ggplot.

Tumor formation in vivo.

Animal studies have been approved by IACUC at the University of Virginia. For the subcutaneous U87 models, 2×10⁶ cells were injected into the flanks of athymic nude mice. For the CT2A syngeneic model, 2×10⁵ cells were stereotactically (Stoelting) implanted into the right corpus striatum of immunocompetent C57BL/6 adult male mice (6–8 weeks old). For the U87 intracranial model, 2X10⁵ cells were stereotactically injected into right corpus striatum of athymic nude mice. IVIS imaging was performed on anesthetized mice at two weeks and three weeks post implantation. For the GSC model, 1×10⁵ GSC11 cells were stereotactically (Stoelting) implanted into the right corpus striatum of immunocompromised athymic nude adult male mice (6–8 weeks old). MRI imaging was performed on anesthetized mice at 1 to 4 weeks post implantation. For PDX models, blocks of PDX tissue were implanted under the skin in the flanks of athymic nude mice or 2×10⁵ cell culture suspension was stereotactically injected into right corpus striatum of athymic nude mice. For stable astrocyte cell lines expressing wild-type AVIL or AVIL mutants, 3×10⁶ cells were injected subcutaneously into the flanks of athymic nude mice. All animals were monitored daily. Tumor sizes were measured by a caliber or IVIS imaging. Tumor volume values were calculated through the formula Volume= 0.5 × L × W². Treatment starting points were determined by the time most mice have tumors of 100–200 mm³ in size. Compound A was dissolved in 50% DMSO, 15% ethanol, and 35% water and delivered as described in figure legends.

Statistical analysis.

Individual-level data are presented in data file S1. Statistical analyses were performed using GraphPad Prism version 10.5.0. Data are presented as biological replicates and displayed as mean ± SD, or as ranges, as indicated in the corresponding figure legends. Sample sizes (n), exact statistical tests used, and adjusted P values are reported within each figure and legend. Comparisons between two independent groups were conducted using unpaired, two-tailed t-tests. For experiments involving more than two groups, one-way ANOVA was applied followed by either Dunnett’s multiple-comparisons test (for comparisons against a single control) or Tukey’s post-hoc test (for all pairwise comparisons). Time-course and other repeated-measures experiments were analyzed using two-way repeated-measures ANOVA with either Šídák’s post-hoc multiple-comparisons test or Dunnett’s post-hoc test; Geisser–Greenhouse correction was applied when sphericity assumptions were not met. Survival analyses were performed using the two-sided log-rank (Mantel–Cox) test. When survival curves involved multiple treatment arms, pairwise comparisons versus control and between treatment groups were adjusted using the Holm–Šídák method. Categorical trends were evaluated using the Cochran–Armitage test. Sequencing counts were normalized with limma and converted to z-scores.

Supplementary Material

Supplemental Materials

Material and Methods

Fig. S1 to S18

Tables S1 and S2

MDAR Reproducibility Checklist

Data files S1 and S2

Data, code, and material availability.

All data associated with this study are in the paper or supplementary materials. Microarray data for this study have been deposited onto the GEO database under accession numbers GSE95164 and GSE268857. Single cell RNA-seq data of 28 adult and pediatric IDH wild-type GBM tumors were acquired from GSE131928. Non patented material is available to academic researchers via a materials transfer agreement with the University of Virginia by contacting H.L. University of Virginia utilizes the_UBMTA implementing letter for material transfers. If the recipient is not a UBMTA signatory, the AUTM template is used.