Relapse-associated AURKB blunts the glucocorticoid sensitivity of B cell acute lymphoblastic leukemia

Genome-Wide Identification of Genes That Influence Sensitivity to GC-Induced Cell Death.

To determine the contribution of each gene in the genome to cell growth and GC-induced cell death in B-ALL, we used a next generation shRNA screen (9, 17). We performed this screen in NALM6 cells, which we demonstrated previously to be a useful cell line model for the response of patient specimens and patient-derived xenograft samples to GCs (9). We targeted each known protein coding gene (∼20,000) with an average of 25 shRNAs delivered by lentivirus. Starting with 6 billion cells, we performed the screen with three biological replicates as described previously, except in spinner flasks rather than still tissue culture flasks to accommodate the vastly greater number of genes screened (9, 18, 19). Infected cells were then treated three times with vehicle or 35 nM dex (∼EC₅₀) for 3 d each time, washing the drug out in between. By comparing the enrichment of integrated shRNA expression cassettes in the vehicle vs. initially infected cells, we calculated the effect of each gene on growth (γ score). By comparing the enrichment in cells treated with dex vs. vehicle, we calculated the effect on dex sensitivity (ρ score). The dex sensitivity scores were highly consistent between biological repeats (SI Appendix, Fig. S1A and SI Methods has details). This design not only identified high-confidence hits but also, identified genes that both contribute to and restrain the response of cells to GCs (17, 18, 20).

Hundreds of genes significantly affected growth (γ scores). Significance was calculated by Mann–Whitney U (MW) and Kolmogorov–Smirnov (KS) tests, which generally agreed well, except for a cohort of genes that exhibited greater significance by MW (SI Appendix, Fig. S1B). This is likely because shRNAs that affect growth severely dropped out of the screen, and the MW test is less sensitive to a smaller number of shRNAs. Nonetheless, ∼5% of genes (1,030, q value ≤ 0.1) (21) affected growth, with the majority (910 genes) impeding growth when depleted (Fig. 1A, Dataset S1, and SI Appendix, Fig. S1C). Depletion of components involved in translation, mammalian target of rapamycin (mTOR) signaling, and energy production had a negative impact on growth (SI Appendix, Fig. S1D). The B cell receptor pathway was also important, likely because signaling through this pathway enhances growth and survival of NALM6 cells (9). AURKB, an important mitosis facilitator that proved to be highly relevant to this study, was also a significant contributor to cell growth.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Genes that affect growth and sensitivity to dex in the B-ALL cell line NALM6. (A) Growth or γ values are calculated by averaging enrichment of each shRNA in the cell population at the end of the growth period (TF) vs. at initial infection (T0) (SI Appendix, SI Methods). Confidence values (P values) are calculated by MW test comparing enrichment of shRNAs vs. thousands of control shRNAs. Points are colored if the genes significantly (q value ≤ 0.1) decrease (green) or increase (purple) growth on depletion. (B) ρ or dex sensitivity values are calculated for each gene as effect of depletion on average enrichment of shRNAs on dex treatment (R1–R3) vs. growth control (TF) (SI Appendix, SI Methods). Confidence values (P values) are calculated by MW test comparing enrichment of shRNAs vs. thousands of control shRNAs. Green points, genes that sensitize cells to dex when depleted; purple points, those that render cells more resistant when depleted. (C) Effector plot for genes that effect dex-induced cell death. From genes with a significant (P value ≤ 0.05) effect on dex-induced cell death (B and Dataset S1), a subset was identified that overlaps with the set of genes consistently and significantly regulated (KS test; q value ≤ 1e-3) by GCs in B-ALL primary samples, patient-derived xenografts, and cell lines (9). These genes are plotted by effect of depletion on cell death (1, perfectly protective; −1, twofold increase in sensitivity) vs. average regulation in response to dex. Genes with up-regulation (purple) or repression (green) by dex that contributes to dex-induced cell death are effectors. Genes with up-regulation (orange) or repression (yellow) that opposes cell death are buffering genes (refer to SI Appendix, Fig. S2). (D) Volcano plot depicts the effect of depletion on dex sensitivity for all known nuclear receptor coregulators (points are a subset of B).

Hundreds of genes were also found to affect dex sensitivity (ρ scores). The 276 most significant genes (q value ≤ 0.1) were split between those that contribute to GC-induced cell death (130; protective on knockdown) (Fig. 1B, purple) and those that attenuate the toxicity of dex (146; sensitizing on knockdown) (Fig. 1B, green). Importantly, depletion of GR (NR3C1) was the most protective, highest-confidence hit in the screen. At the other end of the spectrum, MBNL1 was the most significant sensitizing hit. Sensitizing hits are of immediate therapeutic interest, as inhibitors of these proteins are good candidates for sensitizing cells to dex. Similar to γ, the MW and KS tests for ρ scores were consistent but did not agree for a cohort of genes, again likely because of shRNA dropout during dex treatment (SI Appendix, Fig. S1E). This dropout likely caused an underestimation of the significance of some genes (false negatives) but would not diminish the confidence of significant hits. For example, five AURKB shRNAs dropped out after dex treatment, presumably because they rendered cells extremely sensitive to dex. This resulted in apparently marginal significance (P value = 0.15), despite a substantial phenotype value (ρ = −0.30) (SI Appendix, Fig. S1F). In subsequent analyses, we therefore use a q value ≤ 0.1 cutoff to identify significant genes and a more relaxed cutoff (P value ≤ 0.05) for pathway and exploratory data analyses (top genes).

Identification of GC-Regulated Cell Death Effector Genes.

Previously, we identified genes that are consistently regulated by GCs in B-ALL by combining measured changes in gene expression with existing data for primary samples, patient-derived xenografts, and cell lines (9). Using a stringent cutoff (KS test; q value ≤ 1e-3), we found that 1,040 genes were regulated across dex-sensitive B-ALL samples, with about one-half being activated and one-half being repressed. By comparing the genes regulated by dex in B-ALL cells with the top hits from our shRNA screen, we identified GC-regulated genes that affect cell death (Dataset S2).

Effectors are genes with regulation by dex that contributes to cell death. If activation of a gene contributes to cell death, then depletion would make cells more resistant (Fig. 1C and SI Appendix, Fig. S2A, purple) or would increase cell growth (SI Appendix, Fig. S2 B, purple and C, purple). If repression contributes to cell death, then depletion would render the cells more sensitive to dex (Fig. 1C and SI Appendix, Fig. S2A, green) or would impede cell growth (SI Appendix, Fig. S2 B, green and C, green). In total, 60 genes were classified as effectors by their effect on dex sensitivity (ρ), and 108 genes were classified as effectors by their effect on growth (γ). These effectors include BH3 apoptosis family members (e.g., BCL2, BCL1L11), B cell receptor/PI3K pathway components (e.g., BTK, PIK3CD), and B cell development genes (e.g., LEF1, MEF2C). Two splicing factors, MBNL1, which is mutated in some leukemias (22⇓–24), and SRRM1 (25), are strong, negative effectors, suggesting that GC regulation of splicing contributes to B-ALL death.

The full screen also revealed a number of genes that buffer dex efficacy. Buffering genes are genes with regulation that would counteract the cytotoxicity of GCs. These include genes that are up-regulated by dex and have depletion that renders cells more sensitive and genes that are repressed and have depletion that renders cells more resistant (Fig. 1C and SI Appendix, Fig. S2, orange and yellow points). The most prominent among the buffering genes is FKBP5, which is among the top three most dex-activated genes and strongly sensitizes cells when depleted. Buffering genes with a direct link to leukemia include PREX1 and MAP2K1 (MEK1), which activate RAS/MAPK pathways promoting survival (26), and LMO2, which is an oncogene in T cell acute lymphoblastic leukemia (27). These results indicate that GCs are not cleanly cytotoxic; instead, they regulate genes that oppose their own toxicity in B-ALL. These genes reflect the diverse biology of GCs, misregulation of which may be a source of resistance in some B-ALLs. This unexplored set of buffer genes is an excellent potential target for therapies, as inhibition would counteract the regulation of buffering genes, increasing the cytotoxicity of GCs.

Specific Nuclear Receptor Coregulators Affect Dex Sensitivity.

Pathway analysis of the screen indicated that the estrogen and GC signaling pathways have a significant impact on GC-induced cell death (SI Appendix, Fig. S1G). Because knockdown of estrogen receptor (ER) itself did not have an effect (q value = 0.78), identification of the ER pathway is likely because ER and GR interact with many of the same coregulators and transcription factors.

Our results clearly indicate that nuclear receptor coregulators significantly affect GC sensitivity. Of the 337 proteins in this class identified online and compiled from the literature for this study (https://www.nursa.org/nursa/index.jsf) (28⇓–30), about one-quarter (80) were among the top genes (P value ≤ 0.05) implicated in GC-induced cell death (Dataset S3), with 37 being highly significant (q value ≤ 0.1) (Fig. 1D). Depletion of 33 coregulators reduced sensitivity to dex (e.g., PTGES3, EHMT1), indicating that these coregulators contribute to GC-induced cell death. The most prominent hit was PTGES3 (also known as p23), a chaperone that serves both as a coregulator for GR and as an enzyme producing the prostaglandin PGE2 (31). Depletion of the other 47 coregulators sensitized NALM6 cells to dex, including CREBBP, which is frequently mutated in cancers, KMT2D, and components of the NuRD complex (including HDAC2, MAT1, SPEN, MBD2/3, and GATAD2B) (32). This indicates that these cofactors attenuate GC-induced cell death and would be good targets for inhibitors to sensitize B-ALL. We, therefore, hypothesized that specific coregulators would cooperate with GR to regulate genes involved in GC-induced cell death.

Depletion of Coregulators Generally Attenuates Dex Regulation and Causes Misregulation of Specific GC-Regulated Genes.

To understand mechanistically how coregulator depletion affects dex sensitivity, we measured the effect of depleting selected coregulators with shRNA on dex regulation of genes. We chose EHMT1 and EHMT2, two known GR coregulators, because (i) they have strong significant positive ρ scores (Fig. 1D) (EHMT1 had one of the top scores in the screen), (ii) EHMT2 is homologous to EHMT1 and heterodimerizes with it, and (iii) we have extensively characterized their mechanism of action with GR (16, 33, 34). We then treated three biological replicates of NALM6 cells and two controls (uninfected, scrambled shRNA) with dex or vehicle, measured gene expression by microarray (Illumina HT12 v4), and calculated differential gene expression (lumi/limma, R/Bioconductor) (Dataset S4).

Depletion of each coregulator resulted in fewer genes significantly regulated by dex (q value < 0.01) (SI Appendix, Fig. S3). Much of this effect is due to a general attenuation of dex regulation of genes, with depletion of each coregulator attenuating both activation and repression of genes compared with controls (Fig. 2 A and B). Depletion of EHMT1 had a more pronounced effect (average of 38% attenuation compared with ∼20% for EHMT2), likely because of a concomitant reduction in EHMT2 protein level (35). The majority of genes seems to fit this trend (r² = 0.86–0.88), but other genes exhibited a more significant change in regulation (Fig. 2 A, red dots and B, red dots). Some cell death effector genes did not follow this trend (MEF2C, LEF1, ZMIZ1) and seemed to be unaffected by cofactor depletion. Other effector genes are reliant on a specific coregulator (EHMT1), including MYC and RAG1 (9). Thus, these coregulators have both general and specific effects on dex regulation of genes that contribute to B-ALL cell death.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Each coregulator supports GC regulation of a subset of GR target genes. EHMT1 (A and C) and EHMT2 (B and D) were knocked down and treated with dex for 4 h (three biological replicates), and total RNA was analyzed on Illumina microarrays. (A and B) Genes significantly regulated in control (scrambled shRNA) or knockdown in response to dex were plotted. Each point represents log2 change in expression after dex exposure for control (x axis) or knockdown (y axis) for each gene. The dotted lines represent the linear least squared regression fits to the points, and the flanking curved lines are 99% confidence intervals about that line. Red dots, genes that do not fit (P value ≤ 0.01) a slope of one (solid line), representing no change. (C and D) Expression levels in dex-treated cells for genes significantly regulated by dex under any condition are plotted for control (x axis) vs. knockdown (y axis). Lines of slopes of one (solid lines) representing no change in regulation are plotted for comparison. Genes significantly different (P value ≤ 0.01) are red. Genes referred to in the text are labeled.

Separate from an effect on regulation by dex, coregulator depletion also significantly changed mRNA levels of some genes in the presence of dex. This is important, because the ultimate expression level of a gene on addition of dex governs the biological response rather than the fold change. Comparing the mRNA levels of dex-treated control cells with coregulator-depleted cells (Fig. 2 C and D), 468 (EHMT1-KD) and 305 (EMHT2-KD) genes exhibited significantly different mRNA levels (Fig. 2 C, red dots and D, red dots). Most of these genes overlapped with the genes that failed to be properly regulated by dex on coregulator depletion (Fig. 2 A and B). This analysis revealed that some repressed genes, including MYC and SOX4, remained aberrantly high and that some potential cell death effector genes were underexpressed on coregulator depletion, including TSC22D3 (also known as GILZ), TXNIP, and NFKBIA (36⇓⇓–39) (Dataset S4). Overall, two-thirds of cell death effector genes are no longer significantly regulated by dex on EHMT1/2 depletion in NALM6 cells (Dataset S2: the “NALM6 Effectors” sheet has 18 genes, and the “Misreg EHMT1 or 2 KD Effectors” sheet, which shows the Effector genes that are misregulated by knockdown of EHMT1 or EHMT2, contains 12 of those 18 genes). This abrogation of regulation explains how inhibiting EHMT1/2 could severely blunt sensitivity to dex and shows that the complex is a key driver of dex-induced cell death. In addition, although depletion of EHMT1/2 also attenuates regulation of a similar fraction of buffering genes, the regulation of key buffering genes remains intact. For example, FKBP5, expression of which protects cells against dex-induced cell death, is still strongly up-regulated by dex in EHMT1/2 knockdown. Thus, coregulators have both general and specific effects on the basal and induced expression levels of potential effector genes.

To test whether TSC22D3, TXNIP, and NFKBIA are bona fide effector genes, we measured the effect of shRNA-mediated depletion on GC-induced cell death in NALM6 cells (SI Appendix, Fig. S4). For each gene, reduction in expression was accompanied by a significant reduction in sensitivity to dex.

An Intact GR-EHMT1/2-CBX3 Complex Is Required for Full Dex Potency.

We previously established the interplay of EHMT1/2, CBX3, and AURKB in A549 cells (16). EHMT1 and EMHT2 directly associate with GR and are automethylated or methylated by each other. Methylation forms a binding interface for CBX3, which is required for full coactivator complex activity when associated with GR. AURKB opposes EHMT1/2 interaction with CBX3 by phosphorylating EHMT1/2, reducing the activity of GR at a subset of genes that require EHMT1/2 (Fig. 3A). According to our screen, the depletion of each member of the coregulator complex produced a phenotype that was consistent with its role in dex regulation of cell death effector genes in NALM6 cells [i.e., depletion of GR, EHMT1, EHMT2, and CBX3 was protective against dex, while depletion of AURKB sensitized NALM6 cells to dex (although the AURKB effect was not significant in the screen)] (Fig. 1 B and D).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

EHMT2-EHMT1-CBX3 facilitates GC-induced cell death. (A) Model for regulation of dex-induced genes involved in B-ALL cell death by GR, EHMT2, EHMT1, CBX3, and AURKB. (Upper) GR recruits EHMT2/EHMT1. Methylated EHMT2/EHMT1 recruits CBX3, which recruits RNA polymerase II to activate transcription of cell death genes. (Lower) Phosphorylation of EHMT2/EHMT1 by AURKB prevents CBX3 recruitment, reduces death gene activation by GC, and reduces leukemia cell death. (B) Methylation and phosphorylation of EHMT2 (Upper) and EHMT1 (Lower) in NALM6 cells was analyzed by immunoprecipitation (IP) with control IgG or antibody against EHMT2 or EHMT1 followed by immunoblot with indicated antibodies. (C) NALM6 cells expressing shEHMT2, shEHMT1, shCBX3, or an shNS were treated with twofold dilutions of dex for 72 h. Cell survival was measured by a fluorescence metabolic assay. EC₅₀, the concentration at which one-half of the cells remain alive, is compared with vehicle controls. Error bars indicate SEM of four independent experiments; P values indicate comparison of each coregulator depletion with shNS using the paired t test. **P ≤ 0.01; ***P ≤ 0.001. (D) Pre-B 697 cells expressing shRNA against EHMT2, EHMT1, or shNS were treated and analyzed as in C. *P ≤ 0.05. (E–G) NALM6 cells depleted or not for EHMT2 (E), EHMT1 (F), or CBX3 (G) were treated with 100 nM dex (+) or ethanol (−) for 24 h, and indicated proteins were examined by Western blot.

To test this model experimentally, we first confirmed that both methylation and phosphorylation on EHMT1/2 were intact in NALM6 cells (Fig. 3B). We also confirmed that NALM6 cells depleted of EHMT2, EHMT1, or CBX3 by shRNA were more resistant to dex than controls [nonspecific sequence (shNS)] (Fig. 3C and SI Appendix, Fig. S5 A–C). Another dex-sensitive B-ALL cell line (pre-B 697) also became less sensitive to dex when EHMT2 or EHMT1 was depleted (Fig. 3D and SI Appendix, Fig. S5D), indicating that this mechanism is not limited to the NALM6 cell line. Decreased cleavage of Caspases 3 and 7 and Poly(ADP-Ribose) Polymerase 1 (PARP1) in NALM6 cells depleted of EHMT1/2 or CBX3 after 24 h of dex treatment confirmed attenuation of apoptosis rather than simply growth (Fig. 3 E–G).

We next tested whether the coregulator complex was important for regulation of cell death effector genes. Coregulator depletion (Fig. 4A) significantly decreased both expression and dex regulation of TSC22D3, TXNIP, and NFKBIA but had no significant effect on dex regulation of FKBP5, a gene that did not require EHMT2 or EHMT1 (Fig. 4B). To test whether these genes are direct targets of these cofactors, we first validated the GC-induced binding of GR (9) to neighboring response elements using ChIP-qPCR (Fig. 4C). Using antibodies to CBX3, we found that it was also recruited to effector gene GR response elements on treatment with dex but not to the GR binding site associated with FKBP5 (Fig. 4D). This indicates that dex induces recruitment of the GR-EHMT1/2-CBX3 complex for full regulation of cell death effector genes, and this mechanism applies only to dex-regulated genes that require EHMT1/2.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

EHMT2, EHMT1, and CBX3 are coactivators for a subset of GR target genes. (A) Immunoblot showing EHMT2, EHMT1, GR, CBX3, and GAPDH protein levels in extracts from NALM6 cells analyzed in B. Depletion of EHMT1 resulted in depletion of EHMT2 due to stabilization of EHMT2 protein by EHMT1 protein (16). (B) NALM6 cells expressing shRNA against EHMT2, EHMT1, CBX3, or shNS were treated 8 h with 100 nM dex or an equivalent volume of ethanol. mRNA for indicated GR target genes was measured by qRT-PCR and normalized to β-actin mRNA. Results shown are mean ± SEM for three independent experiments. P value, paired t test comparing each shRNA with shNS. ns, not significant. *P ≤ 0.05; **P ≤ 0.01. (C and D) CBX3 is selectively recruited to EHMT2/EHMT1-dependent GR target genes in response to dex. NALM6 cells were treated with 100 nM dex or ethanol for 4 h. ChIP was performed with antibody against GR (C) or CBX3 phosphorylated at S93 (CBX3-S93p) (D). Immunoprecipitated DNA was analyzed by qPCR using primers that amplify GR binding regions associated with indicated GR target genes. Results are normalized to input chromatin and shown as mean ± SEM for three independent experiments. P value, paired t test. ns, not significant. *P ≤ 0.05; **P ≤ 0.01.

Transcriptome Analysis of Paired Diagnostic/Relapsed B-ALL Samples Identifies Genes Associated with Relapse.

To test whether alterations in GR cofactor levels contribute to relapse or resistance, we examined patient data. Previously, three studies measured mRNA levels of B-ALL cells from patients at diagnosis and after relapse (GSE3912, GSE18497, GSE28460). Analyzed individually, these studies identified dozens of genes misexpressed on relapse (13⇓–15). A recent metaanalysis of these three studies (40) identified hundreds more, but some misexpressed genes (e.g., AURKB) were not identified, and others had surprisingly low-confidence values (e.g., false discovery rate = 0). We, therefore, performed a combined analysis of these datasets to identify genes that are misexpressed on relapse.

Because expression levels were measured on different microarray platforms, we processed each set separately and then combined the results. Arrays were transformed and normalized [affy (41), R/Bioconductor], with differentially expressed probes identified by paired analysis [limma (42)]. We then combined the P values for each gene using the Fisher method. This analysis identified 350 “relapse genes” with expression that was significantly different from diagnostic samples (q value ≤ 0.05). The majority of these (292) were overexpressed (Fig. 5A). Using a more relaxed significance cutoff (P value ≤ 0.01) for pathway and comparative analyses, we identified 539 top hits, most of which (427) were again overexpressed (Dataset S5).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

Genes differentially expressed in B-ALL at relapse vs. diagnosis. (A) Three studies collecting paired RNA samples from B-ALL patients at diagnosis and relapse (GSE3912, GSE18497, GSE28460) were combined. A fold change and a P value for each gene were first calculated for each dataset. The fold changes were then averaged, and the P values were combined using Fisher’s method to generate the volcano plot. Genes with a q value ≤ 0.05 are colored red. Outlying genes are labeled. (B) Ingenuity pathway analysis of relapse genes indicates that cell cycle genes are highly enriched. (C) Upstream analysis of relapse genes shows an enrichment for prostaglandin signaling, specifically through PTGER2. (D) AURKB is overexpressed on relapse. Boxplots depict the relative expression of AURKB in B-ALL patient blood samples taken sequentially at diagnosis and relapse for the three different studies. Notches in boxplots represent 95% confidence intervals, with the combined q values shown.

The results of this combined analysis are consistent with and extend previous analyses. Consistent with previous analyses (43, 44), the top relapse genes were related to cell cycle (e.g., CCNB2, CDK1) and replication (Fig. 5B). However, genes relating to metabolism and nucleotide synthesis were also prevalent. Other relapse genes, such as BIRC5 (Fig. 5A), increase cell survival by opposing apoptosis (45). Underexpressed genes were not related to cell cycle but instead, immune cell function (e.g., Toll-like receptor signaling and IL-1 signaling). These data suggest that relapse results from an increase in proliferation and survival and a decrease in immune cell characteristics. Another pathway significantly enriched among relapse genes indicates activation of prostaglandin signaling through PTGER2 (Fig. 5C). The significance of this finding suggests a strong link between prostaglandin signaling and relapse that has not been reported previously.

Combining Data from shRNA Screen and Genes Misexpressed at Relapse Identifies Resistance-Relapse Genes.

To link misexpression of genes to resistance, we overlapped the top relapse gene hits (Fig. 5A) (P value ≤ 0.01) with our top hits (P value ≤ 0.05) from the shRNA screen for growth (γ) (Fig. 1A) or dex sensitivity (ρ) (Fig. 1B). About 22% of relapse genes (117) affected growth or survival (SI Appendix, Fig. S6A). Of these, the genes that contribute to relapse are either overexpressed and decreased growth when depleted or underexpressed and increased growth when depleted (SI Appendix, Fig. S6B, orange and yellow, respectively). We call the 103 genes that fit these criteria “resistance relapse” genes. These include previously identified cell cycle genes (e.g., CCNB2, CDK1) (13⇓–15, 40) (Dataset S6 and SI Appendix, Fig. S6 C and D) and newly identified genes, such as AURKB (Figs. 1A and 5A). Both epigenetic (e.g., DNMT1, H2AZ, CBX5, PRDM2) and transcription factors (e.g., PRDM2, KLF7, and CITED2) were also identified as resistance relapse genes.

Fewer relapse genes (69 genes) affect GC sensitivity (SI Appendix, Fig. S7A). Of these, 48 fit the criteria for dex relapse resistance (Dataset S6 and SI Appendix, Fig. S7 B and C). These also include specific cell cycle (BIRC5, CDK1, and CCNB2) and DNA repair (PARPBP) genes, the overexpression of which increases resistance to dex (SI Appendix, Fig. S7D). Each of these represents a potential target for combination therapy with dex. This reveals a previously unrecognized interaction between TOP2A, targeted by anthracyclines (e.g., doxorubicin) in some standard and relapsed therapy, and dex that helps explain their effectiveness in combination.

Surprisingly, few GR coregulators were misexpressed on relapse. Of the 23 coregulators misexpressed (P value ≤ 0.01), only 3 (CCNA2, PA2G4, and PSME3) affect dex sensitivity (Dataset S3). Accordingly, few misexpressed coregulators are predicted to interact directly with GR (e.g., SMARCE1). Instead, cell cycle genes, such as CCNA2, CCNE1, and AURKB, are more prevalent and may regulate the activity of GR during the cell cycle. We focused on AURKB, because we had previously shown that it regulates the activity of the EHMT1/2-CBX3 complex (Fig. 3A) and was significantly overexpressed at relapse vs. diagnosis (Fig. 5 A and D), and its overexpression is predicted to enhance growth (SI Appendix, Fig. S6C), suggesting that it may contribute to relapse.

AURKB Inhibitor Enhances GC Sensitivity of B-ALL Cell Lines and Primary Cells from Patients with Relapsed B-ALL.

According to our model, AURKB blunts dex toxicity in B-ALL by phosphorylating EHMT1/2, which interferes with recruitment of CBX3 (Fig. 3A). Thus, we hypothesized that inhibition of AURKB would sensitize NALM6 cells to dex by enhancing dex regulation of cell death effector genes. To test this, we used two AURKB-specific inhibitors: ZM447439 (46, 47) and AZD2811 (also known as AZD1152-HQPA) (48⇓⇓–51). Inhibiting AURKB is toxic to NALM6 cells, consistent with our screen data and presumably due to cell cycle inhibition (52). Inhibition of AURKB with either ZM447439 or AZD2811 reduced phosphorylation of EHMT2 (SI Appendix, Fig. S8 A and B) and significantly sensitized NALM6 cells to GC-induced cell death (Fig. 6A), while inhibition of aurora kinase A (Alisertib) had no effect (SI Appendix, Fig. S8C). AURKB inhibition by AZD2811 potentiated the apoptotic activity of dex as evidenced by increased cleavage of apoptotic markers in dex-treated NALM6 (Fig. 6B). In addition to NALM6 cells, AURKB inhibitors also enhanced the sensitivity of RCH-ACV, a dex-resistant B-ALL cell line (Fig. 6C). Importantly, in NALM6 cells, AZD2811 enhanced dex-induced expression of dex effector genes that utilize EHMT2, EHMT1 and CBX3 but not the EHMT2/EHMT1-independent FKBP5 (Fig. 6D). Thus, the effect of the AURKB inhibitor on cell survival involves its selective regulation of EHMT2/EHMT1-dependent GR target genes.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 6.

AURKB inhibitors sensitize NALM6 cells to GC-induced cell death. (A) NALM6 cells were treated with indicated dex concentration in addition to 0.75 μM ZM447439 (Upper), 16 nM AZD2811 (Lower), or equivalent volume of DMSO for 72 h; cell survival was measured by fluorescence metabolic assay. Values measured with dex were normalized to those with ethanol. Percentage of survival, mean ± SEM of four independent experiments; P values for individual dex concentrations, paired t test. F test comparing the two curves: P ≤ 0.001. *P ≤ 0.05; **P ≤ 0.01. (Insets) EC₅₀ values, where one-half of the cells remain alive, compared with vehicle controls. Error bars indicate SEM of four independent experiments; P values compare AURKB inhibitor treatment with DMSO using paired t test. In these graphs, survival of cells treated with aurora kinase inhibitor alone was set to 100%. Samples treated with dex plus kinase inhibitor were normalized to survival observed with kinase inhibitor alone; this more clearly depicts effect of kinase inhibitor on dex sensitivity. **P ≤ 0.01. (B) To NALM6 cells pretreated for 24 h with DMSO or 16 nM AZD2811, ethanol (−) or 100 nM dex (+) was added for an additional 24 h. Indicated proteins were examined by immunoblot. (C) RCH-ACV cells were treated with indicated dex concentration in addition to 0.75 μM ZM447439, 16 nM AZD2811, or DMSO for 72 h; cell survival was measured by fluorescence metabolic assay. Normalization was performed as in A. Percentage of survival, mean ± SEM of five individual experiments; P value for each dex concentration, paired t test. F test comparing the two curves: P ≤ 0.001. *P ≤ 0.05; **P ≤ 0.01; ***P ≤ 0.001. (D) NALM6 cells pretreated with AZD2811 (16 nM) or DMSO for 24 h were treated for 8 h with 100 nM dex or ethanol. mRNA for indicated GR target genes was measured by qRT-PCR and normalized to β-actin mRNA. Results shown are mean ± SEM for three independent experiments. P value, paired t test. ns, not significant. *P ≤ 0.05; **P ≤ 0.01. (E) Model for GC-induced cell death. GC-induced cell death is a balance between effector gene regulation driving cell death and buffer genes that oppose it. Enhanced regulation of effector genes in response to GCs and inhibition of AURKB tip the balance toward cell death. We infer (gray arrows) that overexpression of AURKB attenuates effector gene regulation but retains regulation of some key buffering genes, favoring survival.

We then tested the combination of AZD2811 and dex in two patient-derived xenograft lines derived from relapsed B-ALL, LAX7R (KRAS G12A), and LAX56 [t(Y;7)(p1.3;p13)] (53, 54). LAX7R and LAX56 were previously shown to be resistant to both dex and vincristine, another component of standard B-ALL combination chemotherapy. Treatment of these cells with 16 nM AZD2811 alone for 5 d (Fig. 7) reduced cell survival to 70–80% compared with the vehicle-treated control. Treatment with 100–200 nM dex alone reduced survival to 40–60%, whereas lower concentrations of dex had little if any effect on cell survival. In contrast, in combination with 16 nM AZD2811, even 0.1 nM dex reduced cell survival to 10–30%, and lower survival was achieved for LAX56 with higher dex concentrations. Thus, AZD2811 enhanced the dex sensitivity of these two dex-resistant primary B-ALL lines.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 7.

AURKB inhibitors sensitize primary B-ALL cells from relapsed patients to GC-induced cell death. LAX7R (A) or LAX56 (B) primary human B-ALL cells were cocultured with OP-9 feeder cells for 5 d in the presence of indicated drugs; cell survival was determined by staining with Annexin/7-AAD. P value, a t test; different symbols indicate P values between different groups. Values are mean ± SD for three biological replicates and are representative of three independent experiments. *P ≤ 0.05 for significance between combination group and respective dex-treated groups; **P ≤ 0.01 for significance between combination group and respective dex-treated groups; ***P ≤ 0.001 for significance between combination group and respective dex-treated groups; ⁺P ≤ 0.05 for significance between dex- or AZD2811 (16 nM)-treated group and DMSO control group;⁺⁺P ≤ 0.01 for significance between dex- or AZD2811 (16 nM)-treated group and DMSO control group; ⁺⁺⁺P ≤ 0.001 for significance between dex- or AZD2811 (16 nM)-treated group and DMSO control group; ^#P ≤ 0.05 for significance between AZD2811 + dex combination group and AZD2811 group; ^##P ≤ 0.01 for significance between AZD2811 + dex combination group and AZD2811 group; ^###P ≤ 0.001 for significance between AZD2811 + dex combination group and AZD2811 group.