Loss of ATRX promotes aggressive features of osteosarcoma with increased NF-κB signaling and integrin binding

Osteosarcoma (OS) is the most common primary bone cancer diagnosed in humans and is most common among children and adolescents. It is a highly lethal cancer with a propensity for lung metastasis. At present, at least one-third of people diagnosed with OS die from this disease, even if a diagnosis is made and aggressive treatment with surgical excision and combination chemotherapy (i.e., doxorubicin, methotrexate, and cisplatin) is started early in the course of the disease (1, 2). Even for patients who do respond to therapy and survive OS, a normal life expectancy is unlikely, due to the toxicity of current treatment regimens. For patients who develop metastatic disease, the prognosis is particularly bleak: almost 70% of patients do not survive beyond 5 years (3). Despite advances in understanding the molecular and genetic features underlying OS, patient outcomes have not improved significantly in over 30 years. Progress in OS research has been hindered, in part, by the rarity of the disease, with fewer than 1000 human cases diagnosed annually in the United States (2), as well as by the genomic complexity of OS, with its considerable inter- and intratumoral heterogeneity (4–7). For all of these reasons, there remains an urgent need for a better understanding of the molecular underpinnings of OS biology and the resulting altered cellular pathways that may be targetable to provide more specific, refined, and effective therapies for OS.

Recent genomics studies have revealed a high rate of structural variation among OS tumors, including somatic mutations and copy number alterations, as well as a variety of single nucleotide variations or recurrent point mutations (5, 6, 8). Although the TP53 and RB1 genes show the most common recurrent alterations in OS, there are also commonly recurrent somatic alterations in other candidate driver genes, such as ATRX (5, 9). In fact, across a variety of cancers, several recent studies have identified frequent loss-of-function mutations in ATRX, including in gliomas, pancreatic neuroendocrine tumors, melanomas, and soft tissue sarcomas (10–16) (Figure 1A). The Cancer Genome Atlas identified ATRX as the 14th most frequently altered gene across all cancers surveyed in the study (17–19). In 288 OS tumors surveyed by the American Association for Cancer Research Project Genomics Evidence Neoplasia Information Exchange (GENIE) Consortium, ATRX was one of the most frequently mutated genes, second only to TP53 (18–20) (Figure 1B). Additionally, at the protein level, several recent studies examining both human and canine OS tumors have found that between 20% and 30% lack nuclear expression of ATRX (11, 21, 22). Importantly, the actual incidence of ATRX mutation may be underestimated by modern next-generation sequencing technologies because these methods do not detect many complex indels with short sequence reads. Indeed, Ye et al. (23) identified ATRX as one of several oncogenic driver genes with frequent somatic complex indels in tumors across several cancer types that were overlooked by studies using next-generation sequencing modalities. Despite the known frequency and consistency across both human and canine OS, the impact of this ATRX loss on OS biology is not fully understood.

Figure 1

Frequency of ATRX mutations. (A) Frequency of ATRX mutations across cancers in TCGA Pan-Cancer Atlas, as accessed from cBioPortal (17–19). (B) ATRX is the second most frequently mutated gene in 288 osteosarcomas surveyed by the AACR Project GENIE Consortium as accessed from cBioPortal (18–20).

α Thalassemia and intellectual disability syndrome X-linked (ATRX) is a member of the SWI/SNF family of chromatin remodeling factors. In humans, germline loss of function of the ATRX gene causes the α thalassemia and intellectual disability, X-linked syndrome, for which the gene is named (24). The ATRX protein contains 2 highly conserved domains: the SWI/SNF helicase domain, which regulates chromatin remodeling, and the ATRX-DNMT3-DNMT3L (ADD) domain, which controls DNA methylation patterns and transcriptional repression (25, 26). With this ADD domain, ATRX forms dimers with death domain–associated protein (DAXX), and this complex acts as a histone chaperone to deposit histone variant H3.3 to GC-rich regions of the genome, including the pericentric, ribosomal, and telomeric repeat sequences (27, 28). In the context of cancers, nearly all published studies have focused on the correlation between loss of ATRX expression and activation of the alternative lengthening of telomeres (ALT) pathway for telomere maintenance (21, 22, 29, 30). However, while replicative immortality is one of the key hallmarks of cancer (31), it is not likely to be solely sufficient to promote oncogenesis. Very recently, a few studies have explored other tumor-promoting changes that occur in cancers with ATRX deficiency, including increased cellular motility in glioma cells and TGF-β activation with CDH1 (E-cadherin) downregulation in liver cancer cells (16, 32, 33).

Based on its frequent loss in OS and given the important role of ATRX in chromatin remodeling and methylation patterns, we sought to define the phenotypic and mechanistic impacts of ATRX loss in OS. We hypothesized that loss of ATRX would increase aggressive cellular features of OS, including tumor initiation, migration, invasion, and metastasis, and we describe here our investigations into the impact of this ATRX loss on OS biology, using a range of models to examine each specific cellular phenotype of aggression. Using RNA-Seq and ATAC-Seq, we examined changes in cellular pathways that correspond with ATRX loss. These analyses pinpointed alterations in the NF-κB and several extracellular matrix–related (ECM-related) pathways. Analysis of chromatin binding motif enrichments identified common overlap of ATRX binding sites with ETS family transcription factor motifs, which is notable because ETS family proteins play an important role in osteogenic differentiation (34, 35). Using high-throughput collateral sensitivity screens, we found that OS cells with ATRX KO were sensitized to an integrin inhibitor. Further examination of these cells demonstrated increased integrin β3 expression with ATRX loss. Treatment of ATRX-KO cells with the integrin inhibitor was sufficient to reverse the phenotypes of aggression, particularly migration and invasion, and this drug partially reversed the nuclear upregulation of the NF-κB transcription factors. Examining publicly available pancancer sequencing data, we similarly found enriched integrin signaling with ATRX alteration. The relationship between ATRX expression and integrin binding, NF-κB activation, and ETS family transcription factor binding may impact other known diseases with ATRX loss, including other cancers. Our data show that ATRX mutations sensitize OS cells to integrin inhibition. Future studies are needed to explore integrin inhibition as a potentially new targeted therapy for ATRX-deficient OS.

ATRX loss promotes tumor initiation. We hypothesized that ATRX loss of expression in OS would correlate with the acquisition of aggressive tumor phenotypes, including alterations in tumor initiation, growth, migration and invasion, and metastasis (Supplemental Figure 2A; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.151583DS1). We used a range of in vivo and in vitro models to examine each of these specific phenotypes. To examine how ATRX loss alters tumor initiation in OS, we chose to work with a previously established transgenic Osterix-Cre mouse model with conditional (floxed) alleles of both p53 (p53^fl/fl) and Rb (Rb^fl/fl) and with a Tet-off cassette providing an additional level of temporal control (36, 37). The transgene expression of Cre recombinase is driven by promoter sequences of Osterix, a principal regulator of bone differentiation, and is therefore mostly restricted to committed osteoblast progenitors (36, 38) (although recent studies show Osterix expression in additional subsets of cells; refs. 39, 40). These mice with homozygous deletion of Rb and p53 show completely penetrant OS development, typically between 4 and 8 months of age (37), occurring most frequently in the jaw and head, rear limb, hip, ribs, and vertebra. To determine if Atrx loss would decrease the time to tumor initiation, we added a floxed Atrx allele (Atrx^fl/fl/y) to the Osx-Cre⁺p53^fl/flRb^fl/fl to create Osx-Cre⁺p53^fl/flRb^fl/flAtrx^fl/fl/y. We removed a doxycycline diet at time of weaning and monitored mice of each genotype for tumor development. To more comprehensively identify tumors that developed in any bone, we performed monthly fluoroscopy on a subcohort of 10 females and 10 males of each genotype to further increase our sensitivity in tumor detection (Supplemental Figure 1A). Tumors were collected, and gene recombination of tumors was confirmed by PCR to distinguish the KO allele from the floxed, nonrecombined allele (Supplemental Figure 1B). Consistent with our hypothesis, loss of Atrx significantly increased the rate of tumor initiation compared with the p53/Rb KO alone (Figure 2A; log-rank, P = 0.0021).

Figure 2

ATRX loss promotes tumor initiation and growth. (A) Loss of ATRX expression increased the rate of tumor formation in an Osterix-Cre–driven mouse model of OS. Kaplan-Meier, log-rank P = 0.0021, experimental cohorts Osx-Cre⁺p53^fl/flRb^fl/fl (n = 26 males and 25 females) or Osx-Cre⁺p53^fl/flRb^fl/flATRX^fl/fl/y (n = 22 males and 25 females). (B and C) Western blot and qPCR results showing knockdown (KD) of ATRX with 2 shRNA constructs in the 143B human OS cell line. Dunnett’s multiple-comparison test, P < 0.0001 for both NS versus shATRX-1 and NS versus shATRX-2 for qPCR results. (D) Established xenografts of the 143B human OS cell line with ATRX shRNA KD showed greater final tumor volume compared with NS controls (1-way ANOVA with multiple comparisons, P = 0.04 shATRX-1 versus NS, P = 0.006 shATRX-2 versus NS; n = 3 males and 2 females per treatment group). (E) Western blot results show effective CRISPR-Cas9 KO of ATRX expression in the 143B human OS cell line. (F and G) Established xenografts of the 143B cell line with ATRX KO had a faster rate of tumor growth (1-way repeated-measures ANOVA, log transform, P = 0.02) and larger final tumor volumes compared with WT cells (Student’s t test, P = 0.03; n = 5 males and 5 females per treatment group). (H) Histology of xenograft tumors. The expected ATRX expression status of all WT and KO tumors was validated by IHC for ATRX. All xenograft tumors stained strongly for Ki-67 (greater than 95% positive), suggesting high cellular proliferation in all tumors. No significant differences were observed in the xenograft tumors with IHC staining for cleaved caspase 3 with less than 5% of cells staining positively in all tumors. Positive control is shown.

ATRX loss promotes tumor growth. Given the role of ATRX loss in speeding the time to tumor initiation, we next sought to determine if ATRX loss would lead to increased tumor growth. To do this, we chose to compare tumor growth using a xenograft mouse model in which tumors would be easily detected and would develop in the same location (in the s.c. flank) for all mice. We stably transduced human 143B OS cells with a nonsilencing (GFP) shRNA or 1 of 2 independent shRNA constructs targeting ATRX (shATRX-1 and shATRX-2). ATRX knockdowns were confirmed via Western blotting and quantitative PCR (qPCR) (Figure 2, B and C). We then injected the control and ATRX shRNA knockdown 143B cells s.c. in SCID-beige mice and monitored tumor growth rates over time. ATRX knockdown significantly enhanced tumor growth in these xenografts for both shRNAs (Figure 2D). To further validate these findings, we also developed a CRISPR-Cas9 KO of ATRX in the 143B human OS cell line (Figure 2E and Supplemental Figure 2B), and we repeated the previous xenograft experiment with ATRX-KO or WT cells injected s.c. in SCID-beige mice. Consistent with the results of the shRNA-mediated knockdown study, ATRX KO increased both the rate of tumor growth and final tumor volume as compared with WT 143B cells (Figure 2, F and G).

Histological analysis of xenograft tumors and in vitro cells show no significant differences in cell proliferation or apoptosis with ATRX loss. In order to investigate whether the increased tumor size found with ATRX KO was due to differences in tumor cell proliferation or apoptosis, IHC was performed on the formalin-fixed, paraffin-embedded xenograft tumors harvested from these mice. First, we did verify by IHC that the ATRX KO was retained in the final tumors harvested (Figure 2H). Both WT and KO tumors stained very strongly for Ki-67, with greater than 95% of cells staining positive, supporting high cellular proliferation in all tumors (Figure 2H). The 143B cell line is known to be an especially proliferative cell line, and perhaps this high baseline of proliferation limits the ability to detect differences in proliferation, if present, between tumors derived from WT and ATRX-knockdown cells. Additional in vitro studies of the 143B nonsilenced or shRNA knockdown cells examining change in percent confluence over time in the Incucyte live cell imager showed similarly high proliferation rates of all cells, with no significant differences across cell types (Supplemental Figure 2C). Thus, we were unable to conclude that the tumor size differences were due to changes in rate of proliferation. We then performed IHC for cleaved caspase 3 to compare apoptosis, but again, no significant differences were found, with less than 5% of cells for either KO or WT tumors exhibiting positive staining (Figure 2H). Thus, we were also unable to conclude that the tumor size differences were due to resistance to apoptosis.

ATRX loss promotes tumor migration and invasion. We next decided to investigate the impact of ATRX loss on changes in cell motility, including migration and invasion, using standard in vitro scratch wound and transwell migration/invasion chamber assays. Assays were repeated with 2 well-established OS cells lines, 143B and MG-63, to test for consistency despite the different genetic mutation background found in each cell line. To examine migration, we first performed scratch wound assays with human 143B cells with shRNA-mediated knockdown of ATRX and with human MG-63 cells with CRISPR/Cas9-mediated KO of ATRX (Figure 3, A–D, and Supplemental Figure 2B). For both cell lines, wound closure rate was significantly increased with ATRX knockdown/KO, supporting increased migration in ATRX-null OS cells. We also tested if ATRX loss would increase transwell migration/invasion using Boyden chamber migration/invasion assays. As with the scratch wound assays, CRISPR/Cas9-mediated KO of ATRX in both the 143B and MG-63 cell lines increased both migration and Matrigel invasion (Figure 3, E–H).

Figure 3

Tumor migration, invasion, and rate of metastasis increase with loss of ATRX expression. (A) In a scratch wound assay, 143B cells with shRNA KD of ATRX showed faster wound closure than the nonsilenced (NS) control cells (1-way repeated-measures ANOVA, P < 0.0001 for both shATRX-1 and shATRX-2 compared with NS; n = 8 replicates per cell type; 2 experiments). (B) Representative images of scratch wounds at 3 time points. Scale bars: 200 μm. (C) Western blot showing effective CRISPR-Cas9 KO of ATRX expression in the MG-63 human OS cell line. (D) Similarly, in the MG-63 cell line, KO cells showed significantly faster wound closure than WT cells (1-way repeated-measures ANOVA, P < 0.0001; WT, n = 13 replicates; KO, n = 10 replicates; 2 experiments). (E and F) 143B-KO cells showed increased migration and invasion, respectively, in uncoated and Matrigel-coated transwell plates (1-tailed Student’s t test [E] and 2-tailed Student’s t test [F]; uncoated: P = 0.04, n = 8 replicates; Matrigel: P = 0.002, n = 8 replicates; 2 experiments). Scale bars: 100 μm. (G and H) MG-63–KO cells also showed increased migration and invasion, respectively, in uncoated and Matrigel-coated transwell assays (2-tailed Student’s t test; uncoated: P < 0.0001, n = 12 replicates; Matrigel: P < 0.0001, n = 12 replicates; 2 experiments). (I) After 24 hours of growth on a bed of Matrigel, WT cells remain in tight clusters, whereas KO cells form a network of connecting “tubes” or branching networks through the matrix. Scale bars: 100 μm. (J) After 96 hours of growth, the distinct differences in morphology between the 143B WT and KO cells are even more apparent. Scale bars: 200 μm. (K) Luciferase-labeled 143B WT or KO cells were injected into the subperiosteal space of the tibia of SCID-beige mice. ATRX KO correlates with an increased rate of lung metastasis at 1 week after amputation (1-tailed Student’s t test, P = 0.026; WT: n = 5 males and 10 females; KO: n = 6 males and 11 females). (L) Experimental design for orthotopic injections with LUC-labeled cells to study lung metastasis.

Interestingly, when we plated the 143B WT and KO cells on a bed of Matrigel, there was a distinct difference in appearance after 24 hours, and it was even more apparent after 96 hours. KO cells formed a network of connecting “tubes” in the Matrigel, permitting more cell-to-cell contact, whereas the WT cells grew in more confined clusters within the Matrigel (Figure 3, I and J). These findings support an enhanced ability of KO cells to invade through the Matrigel, perhaps by secreting enzymes and ECM proteins to form this branching network. When repeating this experiment with the less-proliferative MG-63 cell line, the differences in morphology were less distinct but still notable after 72 hours (Supplemental Figures 3, A and B).

ATRX loss promotes tumor metastasis to lungs. The in vitro migration and invasion assays suggest that ATRX loss may promote metastatic dissemination of OS cells. To further examine the role of ATRX deficiency in driving metastasis, we used an orthotopic metastasis model of OS. To do this, luciferase-labeled WT or ATRX-KO cells were injected into the subperiosteal space of the tibia in SCID-beige mice (Supplemental Figure 2D). These mice developed tibial OS in approximately 2 weeks, after which the affected legs were amputated, and metastatic progression to the lungs was quantified using luciferase imaging with the In Vivo Imaging System (Caliper Life Sciences Inc., PerkinElmer; see Supplemental Methods), a method previously validated in our lab as an accurate and accessible assessment of in vivo metastatic tumor burden (41). Most of the mice developed lung metastases if given enough time; however, consistent with our in vitro migration and invasion assays, at 1 week after amputation, ATRX KO led to significantly more lung metastases as compared with WT cells, supporting the conclusion that ATRX loss promotes lung metastasis (Figure 3, K and L).

ATRX loss promotes NF-κB pathway activation and downregulates ECM proteins. Given the important roles of ATRX as both a chromatin remodeler and regulator of histone and DNA methylation, we hypothesized that loss of ATRX would have broad impacts on gene expression across the genome and that the resulting alterations to multiple cellular pathways would collectively promote more aggressive cancer phenotypes. To examine this, we performed an integrated genomics analysis of RNA-Seq and ATAC-Seq profiles using these nonsilenced or shRNA-knockdown 143B cells. Analysis of the RNA-Seq data by Gene Set Enrichment Analysis pinpointed enrichment of several pathways relevant to OS upon ATRX knockdown, including upregulation of the NF-κB pathway and downregulation of ECM proteins (Figure 4, A and B, and Supplemental Figure 4, A and B). To further understand the mechanisms underlying these gene expression alterations, we analyzed genomic changes in chromatin openness using ATAC-Seq. Analysis of the ATAC-Seq findings revealed altered chromatin openness across the genome upon ATRX shRNA knockdown compared with the nonsilenced control, supporting the global genomic importance of the role of ATRX as a chromatin remodeler (Supplemental Figure 5A). We then cross-referenced our ATAC-Seq results with our RNA-Seq data and found that transcriptionally upregulated genes were significantly enriched in regions of more open chromatin upon ATRX loss (Figure 4C). Similarly, genes that were downregulated with ATRX knockdown significantly corresponded with regions of more closed chromatin (Figure 4C). These data suggest that the significant alterations in NF-κB and ECM pathways may derive from ATRX-mediated effects on chromatin state. The upregulation of the NF-κB pathway was validated using an ELISA for nuclear extracts of our cell types. Consistent with the RNA-Seq data, NF-κB transcription factor family activity was upregulated in nuclear extracts from both 143B and MG-63 ATRX-KO cell lines compared with WT cells (Figure 5A).

Figure 4

RNA-Seq identifies upregulation of several NF-κB pathways and downregulation of several extracellular matrix (ECM) pathways. (A and B) RNA-Seq of 143B human OS cells with either the NS control or 1 of 2 shRNA knockdowns of ATRX show upregulation of NF-κB pathways and downregulation of various ECM-related pathways with KD of ATRX expression. (C) ATAC-Seq changes in chromatin openness with ATRX shRNA KD correlate with RNA-Seq findings of gene expression changes, particularly when the chromatin peaks are found in regions of introns or transcriptional start sites (Benjamini and Hochberg correction for P value calculations as shown). These data suggest that the significant alterations in NF-κB and ECM pathways may derive from ATRX-mediated effects on the chromatin state. Gold crosses indicate types of DNA regions that showed the most significant correlations with gene expression changes.

Figure 5

ATRX KO increases nuclear expression of NF-κB transcription factors, and an analysis of ATRX chromatin binding motifs highlights common ETS transcription factor family binding motifs. (A) NF-κB ELISA shows increased relative nuclear expression of this family of transcription factors (TFs) with ATRX KO when comparing nuclear extracts from both the 143B and MG-63 WT or KO cell lines (Multiple t tests). Statistically significant comparisons for 143B: p65, P = 0.0004; p50, P = 0.000009; cRel, P = 0.04; p52, P = 0.009; n = 3 replicates for each cell type and TF. Statistically significant comparisons for MG-63: p65, P = 0.01; RelB, P = 0.024; n = 3 replicates for each cell type and TF. (B) Top differentially deviated binding motifs found with ATRX shRNA knockdown in the 143B cell line most closely resemble the ETS transcription factor family binding motifs.

Analysis of common ATRX binding motifs pinpoints ETS family transcription factor binding. To examine potential common binding motifs for ATRX across the genome, we analyzed our sequencing data using chromVar to look at known transcription factor binding motifs in the Jasper motif database as well as all 6 k-mers. The top differentially enriched motifs correspond most closely to ETS family transcription factors, suggesting an important interaction between ATRX and these transcription factors (Figure 5B).

OS cells with ATRX loss display collateral sensitivity to pharmacological inhibition of integrin signaling. Our collective data suggest that ATRX loss provides a selective advantage to OS cells by reducing barriers to tumor initiation, increasing tumor growth rate, increasing migratory/invasive capacity, and enhancing survival in the metastatic niche. Given the host of adaptations and survival benefits conferred on cells with ATRX loss, we hypothesized that these cells would be sensitized to loss of a secondary pathway. To identify potential actionable collateral sensitivities, we performed a high-throughput drug screen of 2100 bioactive compounds on MG-63 WT or KO OS cells. Among the compounds for which this differential viability was outside of the 99th percentile CI, most (74%) displayed increased resistance upon ATRX loss (Figure 6, A and B). These compounds included heat shock protein inhibitors KW-2478, XL888, and VER-49009. Interestingly, however, we also observed increased sensitivity of ATRX-null cells to the integrin inhibitor SB273005 (Figure 6A). SB273005 is a nonpeptide antagonist of the αvβ3 and αvβ5 integrins (42, 43). We further validated this integrin inhibitor sensitization upon ATRX loss using IC₅₀ assays, confirming a significant drug sensitization with ATRX loss (Figure 6C). Integrins are known key interactors with ECM components and directly activate the NF-κB pathway, all of which is consistent with our observations that both NF-κB and ECM pathways are altered upon ATRX loss.

Figure 6

ATRX KO sensitizes cells to pharmacological inhibition of integrin signaling in vitro. (A) Comparison of normalized cell viability for all 2100 drugs in a bioactive compound screen. ATRX-KO cells were most substantially sensitized to the integrin inhibitor SB273005. (B) Among the compounds for which this differential viability was outside of the 99th percentile confidence interval, most (74%) displayed increased resistance upon ATRX loss. (C) SB273005 IC⁵⁰ curves show significant sensitization of KO cells to the integrin inhibitor SB273005. Nonlinear regression, extra sum-of-squares F test, P = 0.006.

The integrin inhibitor SB273005 has a high affinity for αvβ3 integrins; therefore, we hypothesized that there would be an increase in expression of integrin αvβ3 at the cell surface with loss of ATRX expression. To test this, we performed immunofluorescent imaging of our cell lines. As predicted, we saw significantly increased expression of integrin β3 in our KO cells in both the 143B and MG-63 cell lines (Figure 7, A–C, and Supplemental Figure 3C). One of the key matrix components to which integrins αvβ3 and αvβ5 bind is osteopontin; therefore, we examined expression of the gene SPP1. SPP1 mRNA expression was upregulated in our ATRX-knockdown cells (Supplemental Figure 3D), and we hypothesize that there may be increased secretion of this phosphoprotein correlated with ATRX loss. Future studies will further investigate these ECM-integrin relationships and how they correlate with ATRX deficiency.

Figure 7

ATRX-KO cells have greater expression of integrin β3. (A and B) 143B ATRX-KO cells show significantly increased expression of integrin β3 compared with WT cells (2-tailed Student’s t test, P < 0.0001; KO, n = 49; WT, n = 52). Scale bars: 25 μm. (C) Example of integrin β3 expression at cell-to-cell adhesions. Scale bar: 10 μm.

We next tested the integrin inhibitor SB273005 in vivo with xenograft mouse tumors to examine its efficacy as a therapeutic for ATRX-deficient OS. For this experiment, we chose to use an established ATRX-null OS cell line, U-2 OS, to see if this cell line would respond in vivo as the MG-63–KO cells did in vitro. As observed in our high-throughput in vitro screen that treatment with SB273005 significantly reduced in vivo tumor growth in xenografts formed from s.c. flank injections of these U-2 OS cells (Figure 8, A and B).

Figure 8

ATRX KO sensitizes cells to pharmacological inhibition of integrin signaling in vivo, and integrin inhibition is sufficient to reverse aggressive phenotypes seen with ATRX loss. (A and B) Treatment with the integrin inhibitor SB273005 significantly reduced tumor growth in ATRX-null U-2OS cells (1-way repeated-measures ANOVA, P < 0.0001 for volume change over time; 1-tailed Student’s t test, P = 0.02 for final tumor volume change; vehicle-treated: n = 4 females, 5 males; integrin inhibitor-treated: n = 5 females, 5 males). (C) The integrin inhibitor reversed the increased rate of migration conferred by loss of ATRX in the MG-63 cell line (repeated-measures ANOVA, P < 0.0001; WT vehicle, n = 19 replicates; WT drug, n = 26 replicates; KO vehicle, n = 18 replicates; KO drug, n = 25 replicates). (D and E) The integrin inhibitor SB273005 reversed the increased migration and invasion observed with the MG-63 ATRX-KO cells in both uncoated transwell assays (D) and Matrigel-coated transwell assays (E). Scale bars: 100 μm. (Tukey’s multiple-comparison test for uncoated wells: ATRX-KO vehicle versus WT Vehicle P = 0.028, KO Vehicle versus WT Drug P = 0.035, KO Vehicle versus KO Drug P = 0.0117, n = 12 wells, 2 experiments. Tukey’s multiple-comparison test for Matrigel-coated wells: ATRX-KO vehicle versus WT vehicle, P < 0.0001; KO vehicle versus WT drug, P < 0.0001; KO vehicle versus KO drug, P < 0.0001; n = 12 wells; 2 experiments). (F) Integrin inhibitor treatment partially reversed the upregulation of NF-κB transcription factors p65 and RelB in the ATRX-KO cells (Tukey’s multiple-comparison tests for p65: ATRX-KO vehicle versus WT vehicle, P < 0.0001; KO vehicle versus WT drug, P < 0.0001; KO vehicle versus KO drug, P = 0.028; n = 3. Tukey’s multiple-comparison tests for RelB: ATRX-KO vehicle versus WT vehicle, P < 0.0001; KO vehicle versus WT drug, P < 0.0001; KO vehicle versus KO drug, P = 0.0002; n = 3).

Integrin inhibition is sufficient to reverse aggressive phenotypes seen with ATRX loss. Given the functional connection between αvβ3 and αvβ5 signaling, NF-κB signaling, and invasive phenotypes in cancer, we tested if integrin signaling inhibition could reverse the increase in phenotypes of aggression that we found with ATRX knockdown/KO. We first repeated our scratch wound assays with WT and KO cells treated with vehicle control or with 8 nM SB273005. The KO cells treated with vehicle retained a significant increase in rate of wound closure relative to both vehicle- and drug-treated WT cells; however, the KO cells treated with SB273005 had a wound closure rate that was very similar to the WT cells (Figure 8C and Supplemental Figure 3E). Thus, this experiment supports that the integrin inhibitor is sufficient for reversal of the increased migratory phenotype conferred by loss of ATRX. These results suggest that the enhanced migratory capability of ATRX-deficient cells is likely due to increased integrin expression and binding. Similarly, treatment of KO cells with the integrin inhibitor reversed the increased migration and invasion observed in the transwell migration and Boyden invasion chamber assays (Figure 8, D and E).

Integrin inhibition partially reverses upregulation of NF-κB signaling. Given the reported relationship between integrin binding and NF-κB signaling, we also investigated whether the integrin inhibitor would be sufficient to reverse the upregulation of the NF-κB transcription factor family in the ATRX-KO cells. Both KO and WT MG-63 cells were incubated for 24 hours with either SB273005 or the vehicle control, nuclear extractions were performed, and the NF-κB ELISA was repeated. The 2 significantly upregulated transcription factors seen in our prior experiment with the MG-63 KO cells, p65 and RelB, were both significantly rescued by integrin inhibitor treatment compared with vehicle control (Figure 8F). These results support a close correlation between the increased migratory and invasive phenotypes found with ATRX deficiency, integrin binding, and NF-κB pathway activation.

Integrin signaling is enriched in ATRX-altered tumors in ICGC/TCGA Pan-Cancer Analysis of Whole Genomes data set. In order to explore whether similar cellular signaling alterations are found in human cancers with altered ATRX expression, we used cBioportal to examine the gene expression data of the ICGC/TCGA Pan-Cancer Analysis of Whole Genomes data set (18, 19, 44). Importantly, in line with our own experimental findings, we did find integrin signaling to be enriched in the ATRX-altered subset of tumors (Figure 9, A and B). Additionally, survival in the subset of patients with ATRX-altered tumors was significantly decreased compared with those with ATRX WT expression (Figure 9C). These results further support a close correlation between ATRX deficiency, survival, and integrin signaling.

Figure 9

Integrin signaling is enriched in ATRX-altered tumors in the ICGC/TCGA Pan-Cancer Analysis of Whole Genomes dataset. (A) Oncoprint of ATRX status. ATRX was found to be altered in 9% of 2583 samples after “hiding mutations and copy number alterations of unknown significance” in the ICGC/TCGA Pan-Cancer Analysis of Whole Genomes dataset on cBioPortal (18, 19, 44). (B) Integrin signaling was enriched in the ATRX-altered tumors. Overrepresentation Analysis (ORA) was performed on expression data between ATRX-altered and unaltered groups (92). Genes submitted to ORA had higher expression in ATRX-altered groups than unaltered groups, and all identified genes had q < 0.05. The query was submitted to the Panther, KEGG, and Wikipathway cancer databases. Enrichment ratios for identified pathways with FDR < 0.05 are displayed. (C) Kaplan-Meier survival curve based on ATRX status. Hazard ratio = 3.917; 95% CI, 1.335–11.5; P < 0.036. Note, not all ATRX-altered tumors have deletions; aside from homdel, there are 2 missense mutations and 1 amplification represented in the altered group.

Our experimental results support the hypothesis that ATRX loss, common in human OS, plays a key role in enhancing the aggressiveness of OS. Loss of expression of this gene increases multiple oncogenic phenotypes, including increased tumor initiation, growth, migration, invasion, and metastasis. Our investigations into the underlying cellular mechanisms driving these aggressive OS phenotypes with ATRX deficiency point to changes in the NF-κB pathway, ECM protein expression, ETS transcription factor binding, and integrin expression — specifically, integrins αvβ3 and αvβ5. ATRX-deficient cells display substantially increased sensitivity to integrin signaling inhibition. Additionally, examination of pancancer sequencing data supports these correlations between ATRX deficiency, survival, and integrin signaling across a range of cancer types.

NF-κB activity is known to increase tumor cell proliferation, suppress apoptosis, and promote angiogenesis (45, 46). NF-κB signaling also enhances tumor invasiveness by inducing and maintaining epithelial-mesenchymal transitions required for tumor metastasis (47). Specifically in OS, Felx et al. (48) found that NF-κB pathway activation played a central role in proliferation in the MG-63 human OS cell line, and Zhao et al. (49) reported that the NF-κB pathway was a key regulator of OS tumor growth, metastasis, and resistance to chemotherapeutics. We demonstrated increased expression of integrin β3 with ATRX KO in our cells. Scatena et al. (50) linked NF-κB activation with integrin αvβ3 binding to its ECM ligands. ECM components include various proteins and growth factors, such as osteopontin, fibronectin, collagens, proteoglycans, and laminins. OS cells adhere to the matrix via cell-surface receptors, primarily integrins, which bind to these ECM proteins (51). The ECM plays a critical role in tumor migration and metastasis, as tumor cells use integrin binding as well as various ECM-degrading proteases — including MMPs — to invade and metastasize (52). Several studies have examined the general interplay between these cellular pathways and OS biology. Li et al. (53) were able to inhibit OS metastasis with the combined blockade of both NF-κB signaling and integrin β1 expression in MG-63 cells. Very recently, Shi et al. (54) showed that expression of avβ3 integrins and fibronectin were both correlated with poor clinical prognosis and decreased survival in OS patients. These findings point to a precise mechanism through which OS with ATRX loss behave more aggressively in the clinical setting.

Our successful attenuation of xenograft tumor growth, as well as tumor cell migratory and invasive capabilities with the integrin inhibitor SB273005, supports the importance of integrin binding for increased OS aggression correlated with ATRX loss. In ecological contexts, it is often the case that an advantage in one environment leads to a collateral sensitivity to another environment. Given the host of adaptations and survival benefits conferred on cells with ATRX loss, we hypothesized that these cells would harbor some collateral sensitivity to loss of a secondary pathway. SB273005 is a potent, orally active nonpeptide integrin inhibitor with a high affinity for integrin αvβ3 (binding affinity constant, K_i = 1.2 nmol/L) and a somewhat lower affinity for integrin αvβ5 (K_i = 0.3 nmol/L) (43). Research on this inhibitor is somewhat limited to date but has included studies of its effect on bone resorption and osteoporosis, arthritis, and the production of Th2 cells and cytokine IL-10 in pregnant mice (42, 43, 55). Gomes et al. (56) studied breast adenocarcinoma cells in whole blood under flow conditions and found that a combination of SB273005 and lamifiban (a nonpeptide antagonist specific for platelet αIIbβ3) successfully inhibited adhesion to the vascular ECM. Several other αvβ3-targeting drugs have advanced to clinical trials for treatment of various solid tumors, including cilengitide, etaracizumab, and the small molecule GLPG0187 (57–62). Despite success in early clinical trials, many of these therapies did not produce clinical outcomes that were significantly improved compared with standard treatment regimens. Based on our own findings, one might ask whether a more defined target population, based on specific cellular signaling pathway alterations and gene mutations, such as ATRX loss, would improve these outcomes. Indeed, such a similar dependence on specific mutations for drug efficacy has been demonstrated with the use of PARP inhibitors most successfully in breast cancer tumors harboring mutations in the BRCA1 and BRCA2 genes (63).

In addition to altered integrin expression, we discovered increased mRNA expression of osteopontin following ATRX loss. Osteopontin (SPP1) is an important component of the ECM in bone, and high expression and secretion of osteopontin in numerous cancer types — including breast, prostate, lung, gastrointestinal, hepatocellular, cervical, and bladder cancers — have been clinically correlated with poor prognosis and shortened survival times (64–73). Gaumann et al. (74) also showed that strong expression of osteopontin correlated with progression of malignancy and metastasis in poorly differentiated sarcomas. Song et al. (75) showed that targeting osteopontin expression with miR-4262 could reduce cell invasion and migration in OS cells. In line with our study findings, binding of osteopontin to integrin αvβ3 directly activates the NF-κB pathway (50). Future assays will examine changes in the secreted proteins, including osteopontin, to further characterize how ATRX expression levels affect these ECM proteins.

The examination of ATRX binding motifs from our sequencing data shows a close similarity to ETS family transcription factor binding motifs. ETS family proteins are also involved in osteogenic differentiation and have been reported to play an important role in osteoblast development and bone formation (34, 35). Studies show that the ETS family of transcription factors plays a crucial role in driving malignancy of tumor cells by prevention of apoptosis, support of angiogenesis, and promotion of invasion and metastasis (34, 35, 76–82). ETS factors are well-known critical mediators of ECM remodeling and invasive properties in cancers, regulating a wide spectrum of ECM-related target genes (34, 83). Also fitting with our own experimental findings, ETS transcription factors are known to share crosstalk with NF-κB signaling, and ETS1 and ETS2 have been shown to bind directly to the promoter regions of integrins αv and β3, as well as their ligand, osteopontin (77, 81, 84–88).

Our research has not completely elucidated how ATRX protein expression alters ETS transcription factor activity, but the enrichment of the ETS binding motifs with changes in chromatin accessibility suggest that functional ATRX may repress these transcription factors by maintaining closed chromatin, effectively preventing the tumor-promoting sequelae that would otherwise occur with active ETS protein binding. Intriguingly, Li et al. (89) demonstrated that ETS1 and DAXX protein colocalize to the promyelocytic leukemia (PML) nuclear bodies. DAXX/Ets1-associated protein 1 (DAXX/EAP1) is able to bind to the N-terminal part of ETS1 and cause repression of transcriptional activation of ETS1 target genes, including MMP1 and BCL2 (89). In more recent studies, researchers have found that ATRX also localizes to the PML nuclear bodies and forms a binding complex with DAXX at this location (90, 91). Future studies should explore whether ATRX also binds with ETS1 in PML bodies, similar to DAXX/EAP1, and in this way, directly represses transcriptional activation of ETS1 target genes.

As we have described, the interactions between ETS binding, NF-κB activation, integrin αvβ3 expression, and osteopontin expression have been closely linked with more aggressive tumor biology in prior studies across cancer types, but the association with ATRX expression has not been noted previously. This relationship may pertain to the underlying cellular signaling pathways and pathogenesis of other ATRX-deficient diseases, including other cancers and ATRX syndrome. In a large pancancer sequencing data set, integrin pathway signaling was enriched in this ATRX-mutated population, supporting the hypothesis that our findings will apply to a wide range of ATRX-deficient cancers. Our graphical model of these signaling pathways based upon our experimental findings is shown (Figure 10). Understanding how ATRX alters these cellular pathways in OS may aid in identifying new targeted therapeutics for ATRX-deficient OS tumors, including the potential use of current and novel integrin inhibitors. Future research should investigate further the role of ECM alterations in OS metastasis and the efficacy of integrin inhibitors as a targeted therapy for specific subsets of OS. How ATRX loss alters these pathways in other cancers should also be explored.

Figure 10

Graphical working model. Numerous studies have shown a close interaction between ETS family transcription factor binding, NF-κB pathway activation, and integrin expression and binding — all of which collectively promote increased cellular motility and metastasis in cancers. Our study has revealed that ATRX expression suppresses these pathways to prevent the oncogenic phenotypes of migration, invasion, and metastasis and that integrin inhibition may be a potential targeted therapy for ATRX-deficient OS.

Supplemental Methods are available online with this article.

Data availability. The high-throughput sequencing data from this study have been submitted to the NCBI GEO repository under accession no. GSE167546.

Statistics. For bar graphs, all data are presented as means ± SEM. All data were analyzed for statistically significant differences using the 1- or 2-tailed Student’s t test (for 2 comparisons) or ANOVA with Tukey’s multiple-comparison test. Effects over time were analyzed with repeated-measures ANOVA. Tumor-free survival curves were estimated using the Kaplan-Meier method and compared statistically using the log-rank test. JMP Pro 15.0 and/or Prism 9.0 (GraphPad Software, Inc.) were used for the statistical analyses. Any P value less than 0.05 was considered statistically significant.

Study approval. All animal studies were approved by the IACUC at Duke University.