KDM6B Counteracts EZH2-Mediated Suppression of IGFBP5 to Confer Resistance to PI3K/AKT Inhibitor Treatment in Breast Cancer

This article is featured in Highlights of This Issue, p. 1793

PIK3CA is found to be the most frequently mutated gene (up to 45%) in luminal breast cancer, resulting in aberrant activation of the PI3K/AKT signaling pathway (1). Accordingly, small-molecular inhibitors of PI3K/AKT have been actively investigated in Phase II/III clinical trials in luminal breast cancers with PIK3CA mutations. Despite promising results obtained from preclinical studies, the clinical efficacy of single-agent PI3K inhibitor is limited or sub-optimal (2), probably due to intrinsic and acquired resistance. As such, multiple drugs combination strategies with an aim to enhance the efficacy of PI3K inhibitors are being tested in various late-phase clinical trials (3).

Epigenetic alterations such as aberrant histone modifications are implicated in cancer initiation, progression, and metastasis (4, 5). Among them, EZH2-mediated histone three lysine 27 tri-methylation (H3K27me3) has been shown to be oncogenic to promote cancer development by repressing tumor-suppressor genes (6). EZH2 can also work as a tumor suppressor (7, 8), and loss of H3K27me3 due to deletion of EZH2 causes T-cell acute lymphoblastic leukemia (T-ALL; ref. 9). Global reduction of H3K27me3 caused by a point mutation at H3 also drives pediatric brain cancer progressions (10–12). Moreover, high-grade breast, ovarian and pancreas cancers have been found to harbor low global H3K27me3 which is correlated with increased recurrence and poor survival (13–15). These findings suggest that both the gain and loss of H3K27me3 can contribute to tumorigenesis in a context-dependent manner.

The level of H3K27 methylation is tightly regulated. Removal of H3K27 methylation mark is catalyzed by two H3K27 demethylases, KDM6A and KDM6B and both of which have been implicated in cancer progression (16). The two enzymes may have redundant functions in the regulation of stem cell identity and animal development but also sometimes exhibits contrasting functions in oncogenesis such as T-ALL (17). In breast cancer, KDM6B and estrogen receptor (ER) coordinate to promote the expression of anti-apoptotic BCL2 by inactivating EZH2-mediated H3K27me3 at its promoter to enhance cell survival (18), whereas KDM6A and MLL4 co-regulate some proliferative and survival genes (19). KDM6B also promotes epithelial to mesenchymal transition that mediates breast tumor invasion and metastasis (20). These findings suggest that KDM6B or KDM6A-mediated demethylation of H3K27 may have a role in promoting cancer development and therapeutic targeting this molecular event may present new therapeutic opportunities.

The level of H3K27 methylation can be regulated by oncogenic signaling pathway, which is exemplified by PI3K/AKT-mediated inhibitory phosphorylation of EZH2, resulting in inhibition of H3K27me3 (21). Alternatively, the inhibition of H3K27me3 may also rise from the upregulation of KDM6B/KDM6A demethylase in cancer. Here, we investigated a role of KDM6B in regulating PI3K/AKT inhibitor response in breast cancer and identified a novel KDM6B target gene, IGFBP5, being crucial in affecting the apoptotic response of PI3K inhibitor. We demonstrated that therapeutic targeting of KDM6B is effective in overcoming PI3K inhibitor resistance.

GDC-0941 (ref. 22; catalog number: Axon-1377) and MK2206 (ref. 23; catalog number: Axon-1684) was purchased from Axon Medchem. GSKJ4 (ref. 24; catalog number: 12073-5) was purchased from Cayman Chemicals.

The following primary antibodies were used for Western analysis: total PARP (catalogue number: 9542), cleaved PARP (catalogue number: 9541), EZH2 (catalogue number: 3147), phospho-ERK 1/2 (catalogue number: 9101), total ERK 1/2 (catalogue number: 9102), phospho-AKT (S473; catalogue number: 9271), phospho-AKT (T308; catalogue number: 2965), H3 (catalogue number: 9175 and 3638), H3K27me3 (catalogue number: 9733), survivin (catalogue number: 2802), and tubulin (catalogue number: 2146) were purchased from Cell Signaling Technology. KDM6B (catalog number: 38113) and phospho-EZH2 S21 (catalog number: 84989) were purchased from Abcam. KDM6A (catalog number: ABE409) was purchased from Merck Millipore. Actin (catalog number: A5441) was purchased from Sigma-Aldrich. Bim (catalog number: 559685) was purchased from BD Biosciences. ChemiDoc MP Imaging Systems and Image Lab software (Bio-Rad) were used to detect chemiluminescence intensity.

Samples were prepared by using Blue Loading Buffer Pack (Cell Signaling Technology, catalog number: 7722), and were denatured at 95°C for 5 minutes. Equal amounts of protein extract were loaded and run on SDS-polyacrylamide gels then transferred to polyvinylidene fluoride (PVDF) membranes by semi-dry transfer. Membranes were blocked for 1 hours at room temperature in TBST supplemented with 5% non-fat milk, and incubated overnight at 4°C with primary antibody diluted in the same blocking buffer. After three washing in TBST, membranes were incubated for 1 hour at room temperature with horseradish peroxidase (HRP)–conjugated secondary antibodies. After a further three times washing with TBST, membranes were incubated with Pierce ECL2 Western Blotting Substrate (Thermo) and exposure with ChemiDoc MP Imaging Systems and Image Lab software (Bio-Rad).

Cancer subtype–specific gene expression analyses were performed using data on UCSC cancer browser, https://genome-cancer.ucsc.edu/proj/site/hgHeatmap/ (TCGA RNAseq RPKM level 3 value, N = 1,215). Correlation between gene expressions and overall survival of 1,789 patients was performed using the GOBO algorithm (http://co.bmc.lu.se/gobo/).

All breast cancer cell lines were obtained by the ATCC. MCF-7, T-47D, MDA-MB-361, MDA-MB-415, MDA-MB-231, BT-549, Hs 578T, and MDA-MB-157 breast cancer cell lines were grown in DMEM medium supplemented with 10% FBS, and their identities have been authenticated. SKBR3 cells were grown in McCoy's 5A medium. HCC1806, HCC70, and HCC1937 were grown in RPMI medium supplemented with 10% FBS. HMEC and MCF10A normal breast epithelial cell line were grown in DMEM/F12 supplemented with 5% horse serum, 20 ng/mL epidermal growth factor, 0.5 mg/mL hydrocortisone, 100 ng/mL cholera toxin and 10 μg/mL insulin. All media were also supplemented with 5,000 U/mL penicillin/streptomycin. All cells were grown at 37 °C in a 5% CO₂ atmosphere.

KDM6B siRNAs, IGFBP5 siRNAs, and EZH2 siRNAs listed below were purchased from IDT (Integrated DNA Technologies, Singapore). For transient knockdown experiments, siRNAs were transfected into cells using Lipofectamine RNAimax (Thermo Fisher Scientific) following manufacturer's instructions. Forty-eight hours after transfection, cells were used for further experiments.

• siKDM6B #1 (5′-GCTACACCTTGAGCACAAACGGAAC-3′),

• siKDM6B #2 (5′-GCCGAATTCAAGATCCTACCTGATG-3′),

• siIGFBP5 #1 (5′-GGAGAACAGTAAGATGGATGGTTCC-3′),

• siIGFBP5 #2 (5′-GGATAGCACAGTTGTCAGACAAGAT-3′),

• siKDM6A #1 (5′-GGCATTACCTTAACCAAAGAGAGCA-3′),

• siKDM6A #2 (5′-GGACTCTCACAAAGCTGATGAGCTG-3′),

• siEZH2 #1 (5′-TGGTCTCCCCTACAGCAGAA-3′),

• siEZH2 #2 (5′-AGAGTATGTTTAGTTCCAATCGTCA-3′)

To generate the IGFBP5-overexpressing plasmid, complementary DNAs of human IGFBP5 were amplified from pcDNA3-IGFBP5-V5 (Addgene plasmid: 11608) by PCR and inserted into the GFP-based PMN retroviral expression vector (a gift from Dr. Linda Penn, University of Toronto, Canada).

Cells were plated at a density of 1,000 cells/well in a 96-well optical bottom plate (Corning) and left to incubate overnight before small interfering RNA or inhibitor treatment. Cell proliferation or viability was determined using CellTiter-Glo (Promega) following the manufacturer's instruction. Quantitation of DNA content using flow cytometry was used to analyze cell cycle and quantify the sub-G₁ population, which is reflective of the number of hypo-diploid cells undergoing a late stage of apoptosis following endonucleases activity. Briefly, treated cells were fixed with 70% ethanol for one hour at 4°C and stained with propidium iodide (50 μg/mL) for 30 minutes in the dark. To verify cell apoptosis independently, treated cells were also stained with JC-1 dye to analyze mitochondrial permeability using the BD MitoScreen (catalog number: 551302) following manufacturer's instruction. Briefly, cells were harvested and collected by Centrifuge at 400 × g for 5 minutes at room temperature. After removing the supernatant, cells were gently resuspended in 0.5 mL of freshly prepared JC-1 Working Solution. Cells were incubated the in JC-1 Working Solution for 10 to 15 minutes at 37°C in a CO₂ incubator. Then cells were washed twice with 1 × Assay Buffer and at 400 × g for 5 minutes. The stained cells were analyzed using FACScalibur and CellQuest software (BD Biosciences) after resuspending in 0.5 mL of 1 × Assay Buffer.

Total RNA was isolated and purified using the RNeasy Mini Kit (Qiagen). Human HT-12 V4.0 expression Beadchip (Illumina) was used for microarray hybridization and data were analyzed using GeneSpring software (Agilent Technologies). The differential gene expression was determined by 1.5-fold cutoff and statistical analysis (P < 0.05) using student t test. After analysis differentially expressed gene sets were uploaded to Ingenuity Pathway Analysis (IPA; www.ingenuity.com) for gene ontology analysis. All microarray data have been deposited at the deposited at the National Centre for Biotechnology Information's Gene Expression Omnibus (accession no. GSE74655). Reverse-transcription and quantitative PCR assays were performed using High Capacity cDNA Archive kit and KAPA SyBr Fast qPCR kit (KAPA Biosystems). PCR reactions were analyzed in 96-well format using the Applied Biosystems PRISM 7500 Fast Real-Time PCR system. For quantification of mRNA levels, GAPDH was used as internal controls. Real-time primer sequences are as follows: KDM6B (Forward primer CGAGTGGAATGAGGTGAAGA, and reverse primer GCAGCAGGCAGAACTTGAT); EZH2 (Forward Primer TGGTCTCCCCTACAGCAGA and reverse primer TCCTGATCTAAAACTTCATCTCCCA); IGFBP5 (Forward primer GGTTTGCCTCAACGAAAAGA and reverse primer CGGTCCTTCTTCACTGCTTC); KDM6A (Forward primer TTTGTCAATTAGGTCACTTCAACCTC and reverse primer AAAAAGGCAGCATTCTTCCAGTAGTC).

Chromatin immunoprecipitation (ChIP) assay was performed as described previously (25). KDM6B (catalog number: ab38113) and H3K27ac (catalog number: ab4729) from Abcam, EZH2 (catalog number: 39876) from Active Motif, H3K27me3 (catalog number: 9733) from Cell Signaling Technology and IgG (catalog number: sc2027) from Santa Cruz Biotechnology were used. The immunoprecipitated DNA was analyzed by real-time quantitative PCR and enrichments of target proteins on examined DNA regions were quantitated relative to the input DNA. The following primers were used to amplify the indicated regions surrounding the transcription start sites of IGFBP5: P1(forward AGATCAGGATCTGGGGGTGT and reverse TCTTGCTCGTTCAGTTCAGG); P2(forward TCATTGTGTTCACCCTGCTC and reverse GGAATGTAAGAAAGGGGCAAG); P3(forward AGTGTGGGCTTTTTCCCTTT and reverse AAACCCCAAACCCTAACACC); P4(forward ACCTGCTCTACCTGCCAGAA and reverse AGCGAGAGTGCAGGGATAAA); P5(forward AACACCCCACATCCTTGGTA and reverse CCACAGGCAATCATCTTCAA).

All experiments were repeated at least three times unless stated otherwise, and data are expressed as mean ± SEM (standard error of the mean). Statistical analyses were performed using GraphPad PRISM6. The Student t test and ANOVA followed by Tukey's, or Dunnett's tests were used to calculate the P values (*, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001).

Because the global level of H3K27me3 has been observed to be downregulated in high-grade solid tumors, including breast cancer (13, 14, 26), we hypothesized that a skewed balance between the expression/activity of EZH2 and KDM6B/KDM6A could result in the downregulation of H3K27me3. Our analysis using the breast cancer online database of GOBO (http://co.bmc.lu.se/gobo) shows that luminal A and luminal B tumors have higher expression of KDM6B and KDM6A but lower EZH2 levels compared with basal-like tumors (Fig. 1A; Supplementary Fig. S1A). Similarly, in breast cancer cell lines, KDM6B and KDM6A also showed higher expression in luminal lines compared with basal lines (Supplementary Fig. S1B). Moreover, western blot analysis also showed that KDM6B is highly expressed in luminal breast cancer cell lines, especially in T-47D, MCF-7 and MDA-MB-361 cells (Supplementary Fig. S1C). Analysis of the breast cancer survival rates using GOBO (http://co.bmc.lu.se/gobo/gobo.pl) shows that KDM6B, but not KDM6A, is positively correlated with the poor overall survival in luminal A and luminal B but not in basal tumors (Fig. 1B; Supplementary Fig. S2). Given the previous reports showing that a low level of H3K27me3 is correlated with poor survival in breast cancer (13–15), we hypothesized that KDM6B is a key regulator of H3K27me3 in breast cancer.

Given a potential oncogenic role of KDM6B in luminal breast cancer, we sought to investigate whether pharmacologic inhibition of KDM6B could provide therapeutic benefits. We treated a panel of breast cancer cells consist of varying PIK3CA mutations with small-molecule GSKJ4, which has been shown to be an effective inhibitor of H3K27me3 demethylase KDM6B/KDM6A (24). Strikingly, GSKJ4 preferentially induced cell death in luminal breast cancer cell lines (MCF-7, T-47D, and MDA-MB-361) that carry activating PIK3CA mutations such as E545K and H1047R (Fig. 1C), with significantly lower IC₅₀s, compared with PIK3CA wild-type cells (Fig. 1D). A time-course analysis shows the effect of GSKJ4 on MCF-7, T-47D, and MDA-MB-361 cells but not on PIK3CA wildtype luminal MDA-MB-415 or non-cancerous MCF10A cells (Fig. 1E). A dose–response analysis shows that GSKJ4 induced upregulation of H3K27me3 and increased PARP cleavage in MCF-7, T-47D and MDA-MB-361 cells but not in MDA-MB-415 and MCF10A cells (Fig. 1F), indicating a preferential effect of GSKJ4 on PIK3CA mutant cancer cells. Similar to GSKJ4 treatment, knockdown of KDM6B induced H3K27me3 in all MCF-7, T-47D, MDA-MB-361 cells, SKBR3, MDA-MB-415, BT20 and SUM159 cells but only resulted in obvious PARP cleavage in PIK3CA-mutant luminal breast cancer cells: MCF-7, T-47D and MDA-MB-361 cells (Fig. 1G; Supplementary Fig. S3), which were accompanied by a deficiency in cell proliferation and colony formation (Fig. 1H). These findings suggest that KDM6B has a role in conferring a survival advantage in PIK3CA-mutant luminal breast cancer.

We next tested whether a combination of GSKJ4 with a PI3K inhibitor GDC-0941 or an AKT inhibitor MK2206 would produce more cell death. Indeed, the combination treatment resulted in more cell death compared with single-agent treatment in MCF-7 and T-47D cells but not in SKBR3, BT20 and SUM159 cells as determined by propidium iodide DNA staining of cells in Sub-G₁ phase (Fig. 2A). An independent apoptosis assay, JC-1 that correlates mitochondrial health with apoptosis further confirmed the result (Fig. 2B). Western blot analysis shows that, compared with single-agent treatment, the combination treatments induced more PARP cleavage which correlated with increased H3K27me3 (Fig. 2C). Of note, the inhibitory phosphorylation of EZH2 at serine 21 (p-EZH2 S21) was reduced by GDC-0941 or MK2206 treatment which was more evident in the combined treatment conditions (Fig. 2C). Consistent with pharmacologic inhibition, genetic depletion of KDM6B also sensitized MCF-7 and T-47D cells but not SKBR3, BT20 and SUM159 cells to GDC-0941 (Fig. 2D). These data suggest that co-targeting KDM6B and PI3K/AKT effectively induced more up-regulation of H3K27me3, which was accompanied by more robust apoptosis.

Co-inhibition of PI3K and KDM6B may increase the H3K27me3 level by simultaneously relieving inhibitory EZH2 phosphorylation at serine 21 while suppressing the KDM6B demethylase activity. This led us to hypothesize that an increased H3K27me3 may lead to inhibition of an active gene transcription program which might be important to coordinate or mediate the apoptotic response to the combination treatment. As these three inhibitors all induced H3K27me3, we sought to determine the common gene set which can be repressed by all the three inhibitors. Microarray gene expression analysis of MCF-7 cells treated with GSKJ4, GDC-0941, and MK2206, revealed a common set of 35 genes whose expression were downregulated (using a 1.5-fold cutoff, P < 0.05) by all the three inhibitors (Fig. 3A). We hypothesized that among these 35 genes there are potential candidates that are co-regulated by both KDM6B and PI3K/AKT, which might be responsible for the combinatorial killing effect. We first analyzed 1,215 breast tumors from TCGA and found that 14 out of 35 genes were expressed significantly higher in luminal tumors compared with basal tumors (Supplementary Fig. S4A). Further analysis of gene expression and patient survival data (GOBO) identified IGFBP5 as the only gene (out of the above 14 genes) being both upregulated in luminal tumors and significantly associated with poor survival in luminal A breast cancer patients (Fig. 3B and C; Supplementary Fig. S4B). Consistent with the mRNA level, secreted IGFBP5 proteins are also much higher in the luminal breast cancers (Supplementary Fig. S4C). These in silico analyses of cancer genomic databases and clinical information support IGFBP5 being a key candidate regulated by KDM6B and PI3K/AKT/EZH2 in luminal breast cancer.

RT-PCR and Western blotting further validated the microarray data and showed that GSKJ4, MK2206 or GDC-0941 treatment downregulated the IGFBP5 expression, which was more evident in combination treatments (Fig. 3D). KDM6B knockdown was also able to reduce IGFBP5 expression in both MCF-7 and T-47D cells (Fig. 3E), but KDM6A knockdown did not affect the IGFBP5 expression (Supplementary Fig. S4D), excluding a role of KDM6A in regulating IGFBP5 expression. Furthermore, depletion of IGFBP5 in MCF-7 and T-47D cells resulted in PARP cleavage, decreased survivin and p-ERK 1/2, as well as increased Bim expression (Fig. 3F). Interestingly, although reduced IGFBP5 was also caused by KDM6B knockdown in MDA-MB-415 and BT20 cells (Supplementary Fig. S5A), neither KDM6B knockdown nor IGFBP5 knockdown results in substantial cell death in SKBR3, MDA-MB-415, BT20 and SUM159 cells (Supplementary Figs. S3 and S5B). Ectopic expression of IGFBP5 partially rescued cell death induced by treatment with GSKJ4 and GDC-0941(Fig. 3G). Together, these data indicate that downregulation of IGFBP5 is necessary for the effective apoptosis induction by co-targeting KDM6B and PI3K/AKT.

To determine whether IGFBP5 is a direct target of KDM6B and EZH2 (whose activity is regulated by PI3K/AKT through inhibitory phosphorylation; ref. 21), we performed ChIP analysis in MCF-7 cells. Using known EZH2 targets CDKN1C and CDH1 as positive controls, we show that both KDM6B and EZH2 were enriched in the IGFBP5 gene promoter flanked by H3K27me3 and H3K27Ac marks (Fig. 4A). Depletion of EZH2 by two independent siRNAs resulted in decreased H3K27me3 and increased IGFBP5 expression (Fig. 4B). Similarly, pharmacologic inhibition of EZH2 by GSK126 also reduced H3K27me3 and augmented the IGFBP5 expression (Fig. 4C). However, this effect of GSK126 on both H3K27me3 and IGFBP5 was reversed by co-treatment with GDC-0941(Fig. 4C), as it is known that inhibition of PI3K/AKT by GDC-0941 can induce H3K27me3 by removal of AKT-induced inhibitory phosphorylation of EZH2. Moreover, overexpression of a constitutively active mutant of PIK3CA- PIK3CA^E545K in 293T cells upregulated IGFBP5 through inhibition of EZH2 activity whereas re-introduction of EZH2 in these PIK3CA^E545K-expressing cells partially reversed the IGFBP5 level (Fig. 4D). These findings support a direct and counteractive regulation of IGFBP5 by KDM6B and PI3K/AKT-regulated EZH2. Moreover, consistent with the combined effect on IGFBP5 expression, co-treatment of GSKJ4 with MK2206 or GDC-0941 resulted in increased H3K27me3 but reduced H3K27Ac at the IGFBP5 promoter (Fig. 4E). These data indicate that co-targeting of KDM6B and PI3K/AKT was able to downregulate IGFBP5 effectively through modulating H3K27me3 and H3K27Ac markers, resulting in apoptosis.

To determine whether therapeutic targeting of KDM6B is relevant in breast cancer cells resistance to a PI3K inhibitor, we generated three PIK3CA mutant breast cancer cell lines (MCF-7, T-47D, and MDA-MB-361) that had acquired resistance to PI3K inhibitor treatment following prolonged exposure to GDC-0941 (Fig. 5A). Interestingly, although the three GDC-0941-resistant lines were refractory to GDC-0941 treatment compared with the parental lines, they remained sensitive to GSKJ4 treatment at a level comparable to their corresponding parental lines (Fig. 5B). Similarly, depletion of KDM6B reduced expression of IGFBP5 and induced PARP cleavage and cell death equally effectively in both parental and resistant cells (Fig. 5C). These data suggest that although these cells become resistant to the PI3K inhibitor, they remain sensitive to KDM6B inhibition.

Strikingly, IGFBP5 expression was upregulated in GDC-0941-resistant cells compared with parental cells (Fig. 6A and B), which was however accompanied by decreased p-AKT and increased H3K27me3 (Fig. 6B). This finding suggests that PIK3/AKT-mediated regulation on EZH2/H3K27me was not predominant in regulating IGFBP5 expression in GDC-0941–resistant cells. Examination of the IGFBP5 gene promoter by ChIP showed an increase in H3K27Ac and a decrease in H3K27me3 in GDC-0941–resistant cells as compared with the parental cells (Fig. 6C), suggesting an epigenetic switch at the chromatin level leading to the upregulation of IGFBP5. Similar to KDM6B knockdown, IGFBP5 knockdown equally or even more prominently induced PARP cleavage and cell death in GDC-0941–resistant lines compared with parental lines (Fig. 6D). Collectively, these findings suggest that IGFBP5 expression is promoted by both KDM6B and PI3K/AKT/EZH2 through suppressing H3K27me3 at the IGFBP5 promoter and KDM6B may play a major role in regulating IGFBP5 expression to confer survival advantage upon resistance to PI3K inhibitor (Fig. 6E).

PIK3CA mutations occur in up to 40% of luminal breast tumors, which provides a rationale for therapeutic targeting of PI3K/AKT signaling in PIK3CA-mutant breast cancers (27, 28). However, in multiple clinical trials, PI3K or AKT inhibitors used as a single-agent have repeatedly demonstrated lower than expected efficacy with partial tumor regression and long-term disease stabilization only in a small number of patients (29). Here, we provide evidence that inhibition of an epigenetic regulator, the H3K27 demethylase KDM6B, can enhance the apoptotic response to PI3K/AKT inhibitors in breast cancer cells.

We show that IGFBP5 is a direct target of KDM6B and that IGFBP5 expression level is a crucial modulator of PI3K/AKT inhibitor response. IGFBP5 is also a target of EZH2 whose activity towards H3K27me3 is negatively regulated by PI3K/AKT-mediated phosphorylation. Thus, we propose that epigenetic (KDM6B overexpression) and oncogenic (PI3K activation) deregulations may converge on the chromatin level, resulting in a low enrichment of H3K27me3 at the IGFBP5 promoter, resulting in IGFBP5 overexpression in luminal breast cancer. As such, combined inhibition of both KDM6B and PI3K/AKT is necessary to inhibit IGFBP5 expression to trigger effective apoptosis. This strategy demonstrates that epigenetic approaches to increase H3K27me3 and thus suppress some key pro-survival oncogenes might be exploited to enhance therapeutic response. Our study supports a potential use of GSKJ4 in combination with PI3K inhibitor in luminal breast cancers with PIK3CA mutation. Of note, GSKJ4 has been shown to be effective in cervical cancer, acute lymphoblastic leukemia, and prostate cancer (17, 30, 31), and recent research also showed that GSKJ4 is an effective inhibitor targeting breast cancer stem cells (32). Our findings advocate the use of GSKJ4 or more specific KDM6B inhibitors to treat luminal breast cancer with PIK3CA mutation in future clinical trials.

Although both H3K27me3 demethylases KDM6B and KDM6A showed higher expression in luminal than basal tumors, only KDM6B is significantly associated with patient survival, suggesting an oncogenic role of KDM6B but not KDM6A in breast cancer. Consistent with this notion, we found that depletion of KDM6B, but not KDM6A, downregulated IGFBP5. The oncogenic role of KDM6B in prognosis is consistent with previous findings that low H3K27me3 predicts poor prognosis, though a low level of EZH2 shows the opposite (13–15). These data suggest that KDM6B could be a more important regulator of H3K27me3 in luminal breast tumors compared with EZH2 whose epigenetic activity is often restricted by inhibitory phosphorylation mediated by oncogenic signaling such as AKT.

Our study suggests a role of IGFBP5 in luminal breast cancer to promote survival. Although a tumor-suppressor role of IGFBP5 has also been shown (33–35), our data are consistent with the previous reports showing that IGFBP5 facilitates survival in breast cancer MCF-7 cells (33) and promotes resistance to IGF-IR inhibitor (36). Furthermore, high level of IGFBP5 has been found in tumors and has the pro-metastatic capacity (37), and overexpression of IGFBP5 was associated with advanced tumor stage, poor clinical outcomes in breast cancer patients and urothelial carcinoma patients (38, 39). As a critical mediator of PI3K/AKT and MAPK/ERK survival pathways, IGFBP5 not only acts to inhibit IGF binding to the insulin receptor to inhibits the downstream PI3K signaling output but also functions in the cytoplasm to promote growth in an IGF-independent manner (40–42). Therefore, one possibility is that IGFBP5 has both tumor suppressor role and the oncogenic role and its fine regulation determine its main activity in a context-dependent manner. Interestingly, upon acquiring resistance of breast cancer cells to GDC-0941, IGFBP5 was upregulated, and its promoter underwent an epigenetic switch from H3K27me3 to H3K27Ac. Importantly, breast cancer cells acquiring resistance to GDC-0941 showed remained dependency on KDM6B and IGFBP5 for survival, suggesting when PI3K signaling is inhibited, the cells may still rely on IGFBP5 expression to confer a survival advantage. These findings support the notion that upfront KDM6B inhibition in luminal breast cancer might be able to sensitize PI3K inhibitor treatment as well as to forestall resistance to PI3K inhibitors.

Our study provides an example of how H3K27me3-mediated repression of an oncogene such as IGFBP5 can be targeted to achieve a therapeutic effect. In this scenario, increasing H3K27me3 by targeting H3K27me3 demethylase might be a useful approach to inhibit the expression of a crucial oncogene, though there is a concern that the global effect on H3K27me3 might also restore the expression some tumor-suppressor genes to counteract the benefit.

In summary, our findings demonstrate that targeting KDM6B in combination with PI3K inhibition represents a novel strategy to improve initial response and overcome resistance to PI3K inhibitors. This combination treatment effectively reduces IGFBP5 in sensitive and PI3K inhibitor-resistant cells to induce apoptosis. Given that IGFBP5 is overexpressed in luminal breast cancer and further upregulated upon acquiring resistance to PI3K inhibition, it warrants further study to determine whether IGFBP5 level can be used as a predictive biomarker to monitor response to KDM6B/PI3K inhibition and development of drug resistance. Our strategy of combining KDM6B and PI3K inhibition represents a new paradigm in harnessing epigenetic therapeutics to overcome drug resistance. Further studies are thus warranted to evaluate this combination in a clinical setting.

No potential conflicts of interest were disclosed.

Conception and design: W. Wang, K.G. Lim, Q. Yu

Development of methodology: W. Wang, K.G. Lim, M. Feng

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): W. Wang, K.G. Lim, Y. Bao, D.S.B. Hoon

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): W. Wang, K.G. Lim, M. Feng, Y. Bao, H. Zhang, D. Marzese, D.S.B. Hoon, Q. Yu

Writing, review, and/or revision of the manuscript: W. Wang, K.G. Lim, Y. Cai, D. Marzese, D.S.B. Hoon, Q. Yu

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K.G. Lim, M. Feng, P.L. Lee, Y. Chen, Q. Yu

Study supervision: Q. Yu

This work was supported by the Agency for Science and Technology of Singapore (A*STAR to Q.Yu); Margie Petersen Breast Cancer Program (JWCI; to D.S.B. Hoon) and National Medical Research Council (NMRC) of Singapore (OFIRG16may081; to Q. Yu).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.