A high-throughput drug combination screen identifies an anti-glioma synergism between MTH1 and PI3K inhibitors


 Background

Glioblastoma multiforme (GBM) is the most common and lethal type of primary brain tumor. More than half of GBMs contain mutation(s) of PTEN/PI3K/AKT, making inhibitors targeting the PI3K pathway very attractive for clinical investigation. However, so far, PI3K/AKT/mTOR inhibitors have not achieved satisfactory therapeutic effects in clinical trials of GBM. In this study, we aimed to develop a high-throughput screening method for high-throughput identification of potential targeted agents that synergize with PI3K inhibitors in GBM.
Methods

A Sensitivity Index (SI)-based drug combination screening method was established to evaluate the interactions between BKM120, a pan-PI3K inhibitor, and compounds from a library of 606 target-selective inhibitors. Proliferation, colony and 3D spheroid formation assays, western blotting, comet assay, γ-H2AX staining were used to evaluate the anti-glioma effects of the top-ranked candidates. The drug combination effects were analyzed by the Chou-Talalay method.
Results

Six compounds were successfully identified from the drug screen, including three previously reported compounds that cause synergistic antitumor effects with PI3K/mTOR inhibitors. TH588, an MTH1 inhibitor exhibited significant synergy with BKM120 in suppressing the proliferation, colony formation and 3D spheroid formation of GBM cells. Further investigation revealed that both DNA damage and apoptosis were markedly enhanced upon combination treatment with TH588 and BKM120. Finally, activation of PI3K or overexpression of AKT compromised the anti-glioma efficacy of MTH1 inhibition.
Conclusions

The screening method developed in this study demonstrated its usefulness in the rapid identification of synergistic drug combinations of PI3K inhibitors and targeted agents.


Abstract Background
Glioblastoma multiforme (GBM) is the most common and lethal type of primary brain tumor. More than half of GBMs contain mutation(s) of PTEN/PI3K/AKT, making inhibitors targeting the PI3K pathway very attractive for clinical investigation. However, so far, PI3K/AKT/mTOR inhibitors have not achieved satisfactory therapeutic effects in clinical trials of GBM. In this study, we aimed to develop a highthroughput screening method for high-throughput identification of potential targeted agents that synergize with PI3K inhibitors in GBM.

Methods
A Sensitivity Index (SI)-based drug combination screening method was established to evaluate the interactions between BKM120, a pan-PI3K inhibitor, and compounds from a library of 606 targetselective inhibitors. Proliferation, colony and 3D spheroid formation assays, western blotting, comet assay, γ-H2AX staining were used to evaluate the anti-glioma effects of the top-ranked candidates.
The drug combination effects were analyzed by the Chou-Talalay method.

Results
Six compounds were successfully identified from the drug screen, including three previously reported compounds that cause synergistic antitumor effects with PI3K/mTOR inhibitors. TH588, an MTH1 inhibitor exhibited significant synergy with BKM120 in suppressing the proliferation, colony formation and 3D spheroid formation of GBM cells. Further investigation revealed that both DNA damage and apoptosis were markedly enhanced upon combination treatment with TH588 and BKM120. Finally, activation of PI3K or overexpression of AKT compromised the anti-glioma efficacy of MTH1 inhibition.

Conclusions
The screening method developed in this study demonstrated its usefulness in the rapid identification of synergistic drug combinations of PI3K inhibitors and targeted agents.

Background
Glioblastoma multiforme (GBM) is the most common and lethal type of brain cancer in adults. Current first-line therapy for GBM is a combination of surgery, chemotherapy and radiotherapy [1]. Despite such aggressive treatment, the overall survival of treated GBM patients is still shorter than 15 months [2], so the development of novel and effective therapy for GBM is a dire need. However, in the last 4 decade, most clinical trials evaluating new therapeutics in GBM have encountered failures. Major issues need to be solved in the development of future anti-GBM therapy including but not limited to: the insufficient delivery of drug to the brain, genetically heterogeneous nature of GBM, and aberrant activation of multiple signal pathways [3].
The phosphatidylinositol 3-kinase (PI3K) pathway affects several cellular processes that are essential for cancer progression, including cell growth, survival, motility, DNA repair and metabolism [4]. In GBM, several components of the PI3K pathway are frequently mutated. According to the cancer genome atlas (TCGA) GBM project, the frequency of genetic alterations in the PI3K pathway, such as mutation or amplification of EGFR, activated mutations of PI3CA (p110) or PIK3R1 (P85), and loss of PTEN, were estimated to occur in up to 88% of GBMs [5]. Therefore, targeting PI3K and its downstream effectors, such as AKT and mTOR, seems to be a promising strategy for treating GBM.
Indeed, a variety of targeted agents in PI3K pathways are currently under evaluation in different stages of clinical trials [6].
Unfortunately, most agents targeting the PI3K-AKT-mTOR pathway have failed to achieve satisfactory effects in the treatment of GBM so far [7,8]. Modest monotherapy activity, unexpected statistically significant toxicity, and limited benefits in survival have been observed for PI3K inhibitors [9][10][11].
Thus, deep insights into the bottleneck of the anti-glioma effect of PI3K inhibitor treatment are needed. For example, recent research showed that drug resistance is likely driven by compensatory activation of alternative pathways and reorganization of protein networks [12]. Moreover, given the importance of the PI3K pathway in GBM genesis and development as well as the high incidence of mutations in this pathway, novel strategies for more effective combination regimens should be explored to enhance the anti-GBM effect of PI3K inhibitors. Importantly, synergistic combination therapies can improve efficacy, reduce the dose of individual drugs, and avoid potential drug toxicity [13].
In this study, we aimed to develop a method for efficiently identifying targeted agents that work with PI3K inhibitors in synergy to suppress the growth of GBM cells in a high-throughput format. We selected BKM120, a pan-PI3K inhibitor of all catalytic subunit isoforms of class I PI3Ks (p110α, p110β, and p110γ). Interestingly, BKM120 can cross the blood-brain barrier relatively easy because its brain penetration is not restricted by ABCB1 and ABCG2, the two drug-efflux transporters responsible for elimination of most exogenous compounds in the brain. Therefore, it represents an attractive candidate drug for GBM treatment [14][15][16].
To find the agents may sensitize the anti-glioma effect of BKM120 with known targets, we chose a drug library composing of target selective small molecular inhibitors. These small molecules targeting signaling proteins involved in several well-described signaling pathways, including neuronal signaling, metabolism, epigenetics, DNA damage, etc.
Lastly, it would be a laborious work if we want to identify drug combinations with synergistic effects from hundreds of compounds using the conventional dose-response matrix method. Instead, we developed a novel strategy that only the effect of one a fixed concentration of BKM120 need to be tested. This strategy allows high-throughput screening and quick discovery of potential useful combination therapies for GBM treatment.

Antiproliferation assays
To assess the antiproliferative activity of the targeted agents, cells were seeded in 96-well plates at a density of 1,200 (U251) or 2,000 (all other cell lines) cells per well and were allowed for growth for 24 h. Subsequently, cells were treated with different concentrations of BKM120, either alone or in combination with different concentrations of another targeted agent. Cellular metabolic activity was 6 assessed after 96 h treatment by incubation with Alamar Blue solution (Yeasen, Shanghai, China) for another 4 h, following measurement of fluorescence at 534/584 nm using an Fluoroskan Ascent FL analyzer (Thermo Fisher, CA, USA).

Colony formation and 3D spheroids for mation assay
To assess the impact of drug treatment on clonogenicity of GBM cells, 400 cells per well were seed ed in 12-well plates for 24 h prior to treatment of BKM120, targeted agent, or a combination of both.
Twelve days after incubation, the colonies were washed with phosphatebuffered saline (PBS) and fixed with 4% paraformaldehyde for 15 min and stained with a Giemsa Stain solution (Yeasen) for 30 min. For the 3D spheroids formation assay, 400 cells per well were seeded in an ultra-low binding 96well plates with a round bottom (Costar, MA, USA). After 9 days of treatment, digital images of the spheroids were photographed using a phase contrast microscope. The volumes of 3D spheroids were calculated by formula (V = 0.5 * Length * Width 2 ) based on the major and minor axial length (Length: major axial length; Width: minor axial length).

Drug combination screening
A target-selective inhibitors library consisted of 606 target selective small molecule inhibitors (#L3500) were purchased from Selleck Chemicals. U251 cells were seeded in 96-well plates at a density of at 2000 cells per well 24 h prior to screening. The drug screen was performed in following steps: First, cells were treated with dimethyl sulfoxide (DMSO) (0.2‰), BKM120 (1μM), agent from the drug library (10μM), and combination of BKM120 and targeted agent. SI values were calculated as the formula given in Figure 2A. As a result, 110 targeted agents exhibited excellent synergies with BKM120 (SI≥0.12) and were selected for the further validation. Next, the drug sensitivity for U251 of 110 candidate drugs was reassessed as same as the first step and SI was measured again. Finally, six top ranked compounds in the second step were chosen for further assessment of their antiproliferative effects at concentrations of 10 μM or 2 μM in the presence of 1 μM BKM120 across 8 GBM cell lines, and those continuously achieved high SI values were chosen for further investigation.

Western blotting
Cells were lysed and homogenized in RIPA buffer, and the protein concentrations were determined 7 using a BCA Protein Assay Kit (Thermo Fisher, CA, USA). Protein samples were separated using 10% sodium dodecyl sulfate-polyacrylamide gels, transferred to PVDF membranes (Merck Millipore ,MA, USA) and the membrane was blocked with 3% bovine serum albumin for 1 h. The blots were incubated with specific primary antibodies for 2h at room temperature or 4 ºC overnight, and then incubated with corresponding horseradish peroxidase-conjugated secondary antibodies for 1 h.
Finally, it was visualized using an ECL kit (Beyotime, Shanghai, China) and the density of the immunoreactive bands was analyzed using Imager software (Tanon, Shanghai, China).

Comet assay
Cells were treated with different drugs for 24 h or 100 mM H 2 O 2 for 8 min as positive control of DNA damage. Then all cells were harvested, washed and resuspended in PBS at a concentration of 2x10 6 cells/ml. Next, normal melting agarose was dissolved in the above cell suspension to generate a 0.75% agarose mixture. 100 μl and 450 μl cell-agarose mixture were separately added to a fully frosted slide with a 10 min interval to form two layers in the slide and were kept at 4°C for 30 min.
After removing the coverslip, the slide was incubated in lysis buffer (10 mM Tris pH 10.0, 2.5 M NaCl, 0.1 M EDTA2Na, 10% DMSO and 1% Triton X-100) at 4°C overnight. Then the slide was subjected to electrophoresis in Mini Horizontal Cells (BioRad, CA, USA) filled with ice-cold alkaline electrophoresis buffer (0.3 N NaOH, 1 mM EDTA) at 300 mA and 25 V for 30 min. Soaking in neutralization buffer (0.4 M Tris-HCl, pH 7.5) for 5 minutes, the slide was counter-stained with 5 mg/ml DAPI (Yeasen). Images were acquired by a Zeiss LSM 510 confocal microscope and were analyzed using a Comet Score software (CASP, CASPLab).

γ-H2AX Staining
Cells received treatment for three days were harvested and fixed in 75% ice cold ethanol and kept at -20 ºC overnight. Cells were spun down and washed with ice-cold PBS containing 2% FBS for three times. The cell suspension was incubated with an Anti-phospho-histone H2AX (Ser139) (BioLegend Cat. No. 613404, CA, USA) antibody for 1 h at room temperature. After a brief wash, 10,000 stained cells were collected and analyzed by a BD FACS Verse flow cytometer.
shRNA construction and MTH1 knockdown MTH1 shRNA sequences were designed by an online tool from Sigma-Aldrich website or obtain from literature. Two MTH1 targeting sequences: shRNA#1 5'-GAAATTCCACGGGTACTTC-3' and shRNA#2 5'-CGACGACAGCTACTGGTTT-3' were synthesized and subcloned into a pLKO.1-TRC vector via Age1/EcoRI restriction sites. For transfection, the above plasmids, empty vector together with the packaging plasmids pVSVg (#8454, Addgene) and psPAX2 (#12260, Addgene) were transfected into HEK293T cells at approximately 60% confluence using polyethylenimine (PEI) for 48 h. The culture medium of U251 cells was replaced with the above lentivirus-containing supernatants with the addition of 1:1000 of polybrene and was centrifuged for 2 hrs at 37 ºC for viral infection. After puromycin selection for 3-7 days, the infected cells were collected for validation of MTH1 knockdown.

Cell apoptosis assay by flow cytometry
Cells received drug treatment for 24 h were collected, washed with ice-cold PBS, and evaluated for apoptosis by double-staining with FITC and PI with an Apoptosis Detection Kit (Yeasen Biotech).
Stained cells were analyzed with a FACS Calibur flow cytometer and the obtained data were analyzed by FlowJo 7.6.

Analysis of PTEN expression-drug sensitivity interactions from public database
We took advantage of the Wooster cell line dataset which consists of over 300 cell lines from 19 cancer types (also see https://cabig.nci.nih.gov/caArray_GSKdata/) for analysis of the correlation between PTEN expression and their sensitivities to PI3K inhibitors. All cell lines were divided into three groups: resistant, intermediate sensitive and sensitive according to their IC50 upon treatment of BKM120 or GSK1059615. The PTEN expressions of all cell lines were obtained from Oncomine (https://www.oncomine.org/).

Statistics
All experiments except the drug screen were performed at least in triplicate with more than two independent experiments. The data are expressed as the mean ± SD. Statistical analyses were performed with GraphPad Prism 5.0 software (San Diego, CA, USA). The isobologram analysis of the two drug combinations was performed using CompuSyn 2.0 software (ComboSyn, Inc., Paramus, NJ, 9 USA) [17]. Statistical significance was determined using the two-tailed Student's t-test unless otherwise mentioned, with the following P values considered significant: *: P < 0.05; **: P < 0.01; ***:

BKM120 blocked PI3K-AKT signaling and exhibited cell line-dependent antiglioma effects
We first investigated the antiproliferative effect of BKM120 using cell viability and colony formation assays across eight GBM cell lines. BKM120 exhibited general growth inhibitory effects in a dosedependent manner, but limited responsiveness was observed for several cell lines, such as U251, compared with sensitive cell lines like U87 or T98G ( Fig. 1A and 1B). Next, we selected BKM120 sensitive and insensitive cell lines for further investigation of signaling pathway perturbation.
Exposure of U251, U87 and T98G cells to BKM120 resulted in suppression of AKT and S6 phosphorylation in a dose-dependent manner, suggesting that the PI3K-AKT signaling was sufficiently blocked even in the BKM120 insensitive cell line (Fig. 1C).
PTEN deletion or mutation is a critical event frequently occurred during GBM development. More than half of the GBM cell lines harbored mutations in the PTEN gene (Additional file 1: Figure S1A), and this could affect the activities of PI3K inhibitor as an anti-cancer agent [18]. We took advantage of the PTEN WT and KO isogenic LN18 cell lines that our lab established previously to investigate the relationship between the PTEN status and the anti-glioma effect of BKM120. As shown in Figure S1B, PTEN loss did not confer sensitivity to BKM120 in LN18 cells. Furthermore, we analyzed the Wooster cell line dataset which comprise over 300 cell lines treated with 19 drugs. Again, whether these cell lines were sensitive or resistant to BEZ235 or GSK1059615, another pan PI3K inhibitor, were independent of the levels of PTEN expression ( Figure S1C). We also calculated Spearman's correlation between PTEN protein expression using the data obtained from the cell line reverse-phase protein array in CCLE (Cancer Cell Line Encyclopedia project) and GDSC (Genomics of Drug Sensitivity in Cancer project). We found that low PTEN expression even confers marked increased resistance instead of sensitivity to BKM120 treatment, for a negative (p < 0.05) correlation between the expression of PTEN and the IC50 of another PI3K inhibitor AS605240 was observed ( Figure S1D). Drug combination screening identified compounds synergizing anti-glioma effect with BKM120 Evaluation of the drug combination effect can be carried out via various methods and models.
However, most of them are based on a dose response matrix, e.g., the Highest Single Agent (HSA) and Bi-level intergrated system synthesis (Bliss) independence models [19][20][21], but such approaches are laborious when screening hundreds of drugs. To simplify this process, we established a method for high-throughput drug combination screening in which a Sensitivity Index (SI) score was introduced to quantify the influence of the addition of another drug. SI was calculated for each library drug according to the formula in Fig. 2A. The SI scores ranged from − 1 to + 1, with positive values indicating BKM120-sensitizing effects.
From the above results, we determined that 1 µM BKM120 was an appropriate concentration for the combination drug screen, and the U251 cell line was chosen for its insensitivity to BKM120. We took advantage of a drug library containing 606 highly selective small molecule inhibitors targeting multiple cancer-related signaling pathways (Additional file 2: Figure S2). To calculate the SI for individual drugs in the drug library, U251 cells received the following treatments: vehicle (DMSO), 1 µM BKM120, 10 µM library drug and a combination of 1 µM BKM120 and 10 µM library drug (Fig. 2B).
The first step resulted in the identification of 110/606 (1.8%) inhibitors with SI ≥ 0.12, a threshold at which potential BKM120-sensitizing effects were expected. In the second step, the antiproliferative activity of these inhibitors was then re-assessed at 10 µM on U251 ± 1 µM BKM120 to confirm the candidates identified in the first step. Finally, the 6 top ranked compounds displaying relatively strong synergistic effects with BKM120 in the second step were chosen for further assessment of their antiproliferative effects at 10 µM and 2 µM in the presence of 1 µM BKM120 across 8 GBM cell lines (Fig. 2B).
Validation of the synergism between BKM120 and the candidate agents For the six top candidates selected above, SI were re-assessed to ensure their abilities to sensitize cells to the antiproliferative effect of BKM120. All candidate compounds had a BKM120-sensitizing effect (SI > 0), and five of them exhibited a strong sensitizing effect (SI > 1.0) (Fig. 3A). Next, we assessed the antiproliferative effect of the combination treatment of eight GBM cell lines. Although differentiated responses were observed across different cell lines and across different treated concentrations, the mean SI of all compounds were all above 0, suggesting an overall BKM120sensitizing effect of these candidates. A brief summary of the top six candidates were given (Fig. 3B).
Finally, the top four candidate compounds, TH588 (an MTH1 inhibitor), CGP57380 (an MNK1 inhibitor), GW9662 (a PPAR inhibitor) and 4-hydroxytamoxifen (an estrogen receptor inhibitor), were chosen for further evaluation. As shown in colony formation assays, these compounds almost eliminated all colonies of U251 in the presence of BKM120, while a single drug or BKM120 failed to show such a strong effect (Fig. 3C). Among them, ABT263, CGP57380 and 4-hydroxytamoxifen have already been reported for their synergistic effects in combination with BKM120 [22][23][24]. Altogether, this evidence supports that our proof-of-principle screening method is useful to identify potential drug combinations that have synergistic interactions with BKM120.
Characterization of the anti-glioma effect of the combination treatment of BKM120 and TH588 Because TH588 was the top candidate and has not yet been reported elsewhere for its combined effect with BKM120, it was chosen for further investigation. U251 and SNB19 cells were treated with a 5-by-5 dose titration matrix of TH588 and BKM120, and the viability of the cells was determined. As shown in Fig. 4A and 4D, the expected versus observed dose-response surface plot of the 5-by-5 matrix viability data demonstrated synergistic effect of two agents against U251 and SNB19 GBM cells. In addition, the combination treatment led to marked reduced viability of GBM cells. Moreover, most dots (each dot represents a combination of a specific concentration of two agents) had a CI value below 1.0 in the Fa-CI plots, suggesting the interaction between TH588 and BKM120 were synergistic ( Fig. 4B and 4E). We also applied classical synergy models including HSA and Bliss [19][20][21] to determine the HSA and Bliss values, which are a readout for synergistic inhibition and depict the difference between expected inhibition and observed inhibition. Most of the HSA and Bliss values were above 0, indicating synergistic interactions between the two drugs ( Fig. 4C and 4F).
Furthermore, the combination of TH588 and BKM120 also severely suppressed colony formation in eight GBM cell lines (Fig. 4G) and 3D spheroid formation in SNB19 and LN229 cells (Fig. 4H-K), confirming that the effect of the combination of the two drugs exceeds these of each single agent in suppressing the growth and proliferation of GBM cells in different models.
The combination of BKM120 and TH588 synergistically induces oxidative DNA damage and apoptosis Previous works suggest that PI3K inhibitors interfere with DNA synthesis or repair through nucleoside depletion [25]. MTH1 inhibitor TH588 has been reported to cause DNA damage by increasing 8-oxodG or 2-OH-dA incorporation into DNA, leading to DNA base mispairing, mutation and cell death [26]. To reveal the synergistic mechanism, we used a comet assay, a sensitive method for detecting 8-oxodG or 2-OH-dA at the single-cell level to evaluate the DNA damage caused by the combination treatment [27]. The significant elongated tailing ratio of combination treatment relative to control or single agent treatment suggests that more abundant DNA-damaged fragments and accumulation of 8-oxodG in DNA were presented after treatment of both MTH1 and PI3K inhibitor ( Fig. 5A and 5B).
The nucleosomal histone protein H2AX is rapidly phosphorylated at serine 139 (γ-H2AX) at the double strand break (DSB) site [28], so γ-H2AX foci formation is usually used to quantify DSBs. Therefore, we reasoned that the PI3K inhibitor also enhanced the DSB damage of DNA caused by TH588. To confirm that, cells treated with BKM120 and/or TH588 was stained with γ-H2AX and analyzed by flow cytometry. As shown in Fig. 5C ~ 5D and Figure S3(Additional file 3), the γ-H2AX-positive fraction of U251 cells increased significantly upon treatment with BKM120 and TH588 compared with that in cells that received single-agent treatment, suggesting that the combination of PI3K and MTH1 inhibitors dramatically elevates DSB levels in glioma cells. Moreover, the combination of BKM120 and TH588 caused a significant increase in the apoptotic fraction of cells that received treatment (Fig. 5E ~ 5F and Additional file 4: Figure S4). In comparison, those that received BKM120 or TH588 at the same dose did not show differences relative to cells in the control group that received only vehicle. These results confirm those of earlier experiments showing that the combination treatment markedly reduced cell viability and colony formation.
The PI3K/AKT signaling is a determinant of the responsiveness to MTH1 inhibition To assure that the antiproliferative activity of TH588 is due to its selective targeting of MTH1 but not an off-target effect, stable MTH1 knockdown U251 together with control cells transduced with only 13 empty vector control were established. The expression of MTH1 in two MTH1 knockdown cell lines was validated by Western blotting (Fig. 6A). As shown in Fig. 6B, MTH1-silenced U251 cells were generally sensitive to BKM120, and the half maximal inhibitory concentration (IC50) of two MTH1-silenced cells (2.59 and 1.63) were lower than that of control cells (2.99).
To further investigate whether the TH588 synergizing effect with PI3K inhibitor only limit to BKM120, we also assessed the combined anti-glioma effect of TH588 with other PI3K and AKT inhibitors. For GDC-0941, another pan-PI3K inhibitor, it effectively suppressed the proliferation of GBM cells with TH588, and the Combination Index (CI) values were lower than 1 at most concentrations, suggesting the interaction between two agents were synergistic (Fig. 6C). For GDC-0068 and MK-2206, two highly selective AKT inhibitors, most CI values were also below 1 when they were combined with TH588 ( Fig. 6D and 6E). The above evidence suggested that both PI3K and AKT inhibitors enhanced the antiproliferation effect of MTH1 blockade or knockdown, indicating that the PI3K-AKT pathway activation may be an important determinant for the anti-glioma efficacy of MTH1 inhibition.
To find more evidence supporting this finding, we also checked the levels of PI3K-AKT signaling activation in eight GBM cell lines and categorized them into low, medium or high groups according to the expression levels of p-AKT (Fig. 6F). Next, we performed a proliferation assay by treating these GBM cell lines with TH588 and analyzing the corresponding IC50. Although the number of cell lines used was limited, GBM cells with highly activated PI3K-AKT signaling tended to be more resistant to TH588 (IC50 values for the p-AKT low, medium and high groups were > 14.56, 9.19 and 8.98 µM, respectively, Fig. 6G). Based on this observation, we established LN229 isogenic cell lines with PTEN wild type, PTEN knockout, and PIK3CA carrying constitutively activated mutations (p.N345K and p.E542K) [29] and investigated their responses to MTH1 inhibition. It is known that loss-of-function mutations of PTEN and gain-of-function mutations of PIK3CA can both activate the PI3K-AKT pathway.
Indeed, the levels of p-AKT and its downstream component p-S6 were elevated in all three mutant cell lines compared with LN229 parental cells (Fig. 6H). In line with the above results, both PTEN knockout and PIK3CA mutant cells were less responsive to TH588 than their parental cell lines (Fig. 6I). Finally, we transfected a plasmid carrying the intact AKT gene or an empty vector into HeLa cells, and it led 14 to elevation of p-AKT, p-S6 and p-4EBP1 expression in AKT-overexpressing cells relative to empty vector-transfected cells (Fig. 6J). Similarly, AKT-overexpressing cells showed significantly elevated TH588 tolerance compared with that in empty vector-transfected control cells (Fig. 6K). Altogether, we found that the anti-glioma effect of TH588 was determined by the activation of the PI3K-AKT signaling pathway, which may provide an explanation for the synergistic interaction between PI3K and MTH1 inhibitors.

Discussion
In this study, we developed a simple and efficient screening method for high-throughput identification of agents that synergize with a PI3K inhibitor. The primary screening was based on the Sensitivity Index, which was calculated as the difference between the observed combined effects and expected combined effects. As a result, at least six promising BKM120 synergistic combinations were identified and validated. Among them, three top-ranked candidates, namely, MNK1, estrogen receptor and BCL-2 inhibitors, have been reported previously for their antitumor effects in combination with PI3K/mTOR inhibitors, confirming the usefulness of this screening method. Importantly, in this study, we successfully identified a novel combination of BKM120 and the MTH1 inhibitor TH588, which had a potent anti-glioma effect. Further investigation suggested that the synergy between PI3K and MTH1 inhibitors was not only limited to BKM120 and TH588, but also can be extended to AKT or other PI3K inhibitors together with TH588.
Single-agent PI3K inhibitors failed to exhibit a significant therapeutic response in clinical trials of malignant gliomas for multiple reasons. The data presented here showed that BKM120 alone suppressed the PI3K-AKT pathway as well as arrested GBM cell proliferation, but this effect was doseand cell line-dependent. In addition, GBMs carrying PTEN mutations or deletions were previously thought to be sensitive to PI3K inhibitors. However, we failed to find that these tumors responded better to PI3K inhibitors than PTEN wild-type GBM. Given the importance of the PI3K-AKT signaling pathway and the fact that PTEN was mutated or deleted in 40%-60% of GBMs, it is necessary to explore effective combined treatment to boost the anti-glioma effect of PI3K inhibitors. As a result of the drug screening, six compounds that significantly enhanced the anti-glioma effect of BKM120 in the PTEN-deficient GBM cell line were successfully identified, and the MTH1 inhibitor TH588, which had the highest SI score in the screen, was further studied. The human mutT homologue MTH1 is a human 8-oxo-dGTPase that removes oxidized bases in the nucleotide pool and DNA, thus avoiding ROS-induced DNA misincorporation, mutations and cell death [26]. In contrast, normal cells repair oxidative damage to DNA via alternative pathways. The severe dependence on MTH1-mediated DNA repair in cancer cells makes MTH1 an attractive target in cancer treatment [30,31]. Recent reports indicated that GBM cells were also heavily dependent on functional MTH1 to maintain stemness and tumorigenesis [32,33]. Interestingly, previous work reported the effectiveness of the combination of TH588 and everolimus, an mTORC1 inhibitor, in the treatment of neuroendocrine cancer cells [34].
However, brain delivery across the blood-brain barrier of everolimus is limited [7]. In this study, we reported the synergy between TH588 and BKM120, a pan-PI3K inhibitor penetrating the blood-brain barrier more efficiently than rapalogs [35]. Some evidence from previous works supports our finding that there are functional interactions between PI3K pathway inhibition and ROS-induced DNA damage repair [25,26] and PARP-mediated base excision repair [36]. Indeed, we found a significant increase in the proportion of segmented tail DNA and phosphorylated H2AX under combined treatment with TH588 and BKM120, suggesting that the DNA damage repair pathway was indeed affected by the PI3K pathway and that blockade of both causes severe consequences for cancer cells.

Conclusions
In conclusion, with a simple and efficient screening method, we successfully identified several promising and effective combinations of targeted drugs with BKM120 that potently inhibit GBM proliferation and growth. Among them, the MTH1 selective inhibitor TH588 excelled other top ranked candidate targeted agents and exhibited a strong synergistic anti-glioma effect with BKM120, thus offering a potential combination therapy for further evaluation in GBM models. Overall, the screening method used in this study provides a useful and reliable tool for finding novel and effective combination therapies for refractory tumors such as GBM.

Ethics approval and consent to participate
Not applicable.

Consent for publication
The content of this manuscript has not been previously published and is not under consideration for publication elsewhere and it has been read and approved by all the co-authors.

Availability of data and materials
The analysed data sets generated during the study are available from the corresponding author on reasonable request.

Competing interests
The authors declare no conflicts to disclose.     (F) Quantification of apoptotic cells from F; shown is the average ± SD, n =3 (*p < 0.05, **p < 0.01).

Supplementary Files
This is a list of supplementary files associated with this preprint. Click to download.