Drug sensitivity prediction model
Our approach utilizes predicted IC50 values for 272 drugs from patient samples in TCGA obtained through CancerRxTissue [7] with in silico techniques, followed by evaluation in relevant GBM cellular models. Our workflow involves three stages (Fig. 1a): 1) identifying potential drugs; 2) in silico analyses; and 3) in vitro validation of predicted effects. In the first stage, data processing involves stratifying glioma patients based on WHO CNS classification and other characteristics to evaluate IC50 values, a critical parameter for understanding drug efficacy. We then select potentially effective drugs using the median of the medians of predicted IC50 values as a cut-off. The second stage consists of: 1) predicting blood-brain barrier (BBB) permeability; 2) evaluating target expression to ensure drug specificity; and 3) selecting relevant GBM cell lines based on target expression levels. Third stage involves experimental validation of predicted outcomes.
To independently validate the prediction model developed by Li et al. [7], we first explored drugs commonly used in GBM. As expected, the algorithm predicts that treatment with TMZ will exert a stronger antitumoral effect (lower IC50) in glioma patients with methylation of the MGMT promoter, which leads to reduced expression of the DNA repairing enzyme (Fig. 1b). In addition, the algorithm predicts that specific inhibitors of the mutated enzyme isocitrate dehydrogenase (mIDH) will exhibit a stronger effect (lower IC50) in patients with mIDH gliomas than in wtIDH glioma patients (Fig. 1c). These findings suggest that the algorithm holds predictive potential for determining drug sensitivity in these patients.
Identification and experimental validation of chemotherapeutic drugs with therapeutic potential for GBM
After validating the potential utility of the predictive model, our goal was to identify alternative therapeutic strategies for adult gliomas among a pre-defined list of chemotherapeutic drugs (Fig. 2a). Initially, we assessed drug permeability across blood-brain barrier (BBB) using various in silico tools [11, 12]. TMZ, carmustine, cyclophosphamide, fluorouracil, and cisplatin were predicted to cross the BBB, while contradictory results were obtained for Etoposide BBB permeability (Online Source 2, Fig. 6a), our analyses predicted that GBM will exhibit high sensitivity to this drug. Moreover, Etoposide has been utilized for intracranial neoplasia in clinical settings [13]. While carmustine and TMZ exhibited the highest predicted IC50, i.e. lowest predicted antitumoral effect, Etoposide and Cisplatin emerged as potential drugs for which GBM would exhibit high sensitivity (Fig. 2a). Thus, we selected these drugs, as well as TMZ, the gold-standard for treatment in patients, for further in silico and in vitro validation. Comparing the predicted IC50 of these drugs between mIDH and wtIDH gliomas (Fig. 2b), we found that while wtIDH gliomas are predicted to exhibit a lower predicted response to TMZ and Cisplatin than mIDH gliomas, Etoposide would elicit a stronger effect (lower IC50) in wtIDH gliomas. Using neurospheres derived from mIDH and wIDH gliomas [14], we validated these predictions in vitro (Fig. 2B). Concentration-response curves in U-251 MG, LN-229, and U-87MG ATCC cells showed that these cells were more sensitive to Etoposide followed by Cisplatin than to TMZ (Fig. 2c), aligning with the in silico prediction. To assess potential toxicity, we compared the cytotoxic effects of these drugs between murine normal astrocytes and GBM neurospheres (Fig. 2d). Normal astrocytes exhibited some sensitivity to TMZ, with GBM neurospheres showing very limited response. In contrast, Cisplatin and Etoposide demonstrated potent antitumor effects in this GBM model with no apparent toxicity in normal astrocytes (Fig. 2d). We selected Etoposide to further evaluate its therapeutic potential due to its superior antitumoral effect and favorable therapeutic window.
We found expression of DNA topoisomerase IIα (TOP2A), a target for Etoposide and other chemotherapy agents [15], in GBM sections from The Human Protein Atlas (THPA) [9] (Fig. 3a). Since our model predicted increased sensitivity to Etoposide in wtIDH than in mIDH gliomas, we compared the expression TOP2A across these glioma types and normal brain tissue. Our analysis revealed overexpression of TOP2A in wtIDH glioma biopsies compared to mIDH gliomas and normal brain (Fig. 3b).
Using information available at THPA [9] we found that the expression of TOP2A was higher in U-251 cells, intermediate in LN-229 cells and lower in U-87 cells (Fig. 3c). We then assessed the sensitivity of these cell lines to Etoposide at high and low concentrations. As depicted in Fig. 3d, 2.5 µM Etoposide demonstrated robust cytotoxicity across the three commercial cell lines, with the maximum response (20 µM) positively correlating with TOP2A expression levels. Treatment of mIDH (G01) and wtIDH (G08) glioma patient-derived cells with a fixed concentration of Etoposide confirmed higher sensitivity in wtIDH cells (Fig. 3e). Analyzing mRNA expression levels of GBM biopsies, we observed a positive correlation between TOP2A and several epithelial-mesenchymal transition (EMT) markers (Fig. 3f, Online Source 2, Fig. 6). To validate these findings, we used a wound closure assay, demonstrating that Etoposide inhibited U-251 cell migration, a hallmark of EMT (Fig. 3g).
Identification of alternative drugs with therapeutic potential for GBM
Among all 272 drugs (Fig. 4a I; Online Source 3) we identified several BBB-permeable drugs (BBB+) (Online Source 2, Fig. 7a) with high predicted sensitivity in GBM (low IC50) (Fig. 4a II; Online Source 4). We proceeded selecting potential candidates for drug repurposing based on rigorously defined criteria: target upregulation in the tumor compared to normal tissue (Fig. 4b); predicted higher efficacy compared to TMZ (Fig. 4c) and low reported toxicity in previous clinical trials for other diseases. Following these criteria, we compiled a shortlist of five drugs: Sepantronium bromide (specific BIRC5 inhibitor), Daporinad (specific NAMPT inhibitor), CUDC-101 (a potent inhibitor against HDAC, EGFR, and HER2 targets), HG6-64-1 (specific BRAF inhibitor), and QL-XII-47 (BMX and BTK inhibitor) (Fig. 4a II). Although all these drugs met the established parameters, only the target for Daporinad, nicotinamide phosphoribosyltransferase (NAMPT), was effectively present in GBM biopsies from IHC samples sourced from THPA (Fig. 4d). Moreover, overexpression of NAMPT was associated with worse prognosis in wtIDH glioma patients (Fig. 4e; Online Source 2, Fig. 7b). Consequently, we selected Daporinad for preclinical in vitro assessment of therapeutic potential.
Using the data from THPA we found expression of NAMPT in commercial GBM cell lines, which was as follows: U-87 cells: NAMPThigh; U-251 cells: NAMPTmedium; and LN-229 cells: NAMPTlow (Fig. 5a). Daporinad showed antitumoral effects in all these commercial cell lines, revealing a potent effect at remarkably low concentrations and showing that in conditions of higher NAMPT levels, such as in U-87 cells, the effect of Daporinad was lower (Fig. 5b). This antitumoral effect was extended to cell cultures derived from GBM biopsies [16] (Fig. 5c). Additionally, Daporinad exhibited no evident effect on normal murine astrocytes while in murine GBM neurospheres its IC50 was comparable to that in human GBM cells (Fig. 5d).
Meta-analysis of GBM biopsies from TCGA revealed a positive correlation between NAMPT, and markers associated with EMT (Fig. 5e; Online Source 2, Fig. 8). Thus, we evaluated the effect of Daporinad on the migration of GBM cells, finding that Daporinad effectively inhibited the migration of U-251 and LN-229 cells (Fig. 5f).
Identification of drug combinations with potential therapeutic efficacy in GBM
To identify potential drug combinations for GBM, we explored correlations between drug response and NAMPT mRNA expression levels in biopsies from GBM patients. We observed a positive correlation between TMZ ln(IC50) values and NAMPT expression levels (Online Source 2, Fig. 9a), suggesting that patients with elevated tumor NAMPT levels will be less sensitive to TMZ. Lower predicted efficacy of a specific drug in patients with higher NAMPT levels suggests that its combination with Daporinad could enhance chemosensitivity. Consequently, we combined Daporinad with TMZ in patient derived GBM cell cultures with different NAMPT expression levels, i.e. NAMPThigh G02 GBM cells and NAMPTlow G09 GBM cells (Online Source 2, Fig. 9b). We treated these cells with Daporinad (60 nM), TMZ (150 µM) or both. This assay validated the predicted higher sensitivity of G09 NAMPTlow GBM cells to TMZ while, interestingly, TMZ increased the viability of NAMPThigh G02 GBM cells (Online Source 2, Fig. 9c). As shown above, Daporinad exerted a cytotoxic effect that was higher in NAMPTlow G09 cells than in NAMPThigh G02 GBM cells (Online Source 2, Fig. 9c). Tallying with the results from the predictive model, Daporinad sensitized both GBM cells to TMZ (Online Source 2, Fig. 9c).
Following the same rationale, we evaluated the predicted effect of Etoposide based on NAMPT expression. We identified a negative correlation between ln(IC50) values of Etoposide and NAMPT mRNA expression levels in GBM biopsies, suggesting that Etoposide would exert stronger antitumoral effects in patients with high NAMPT levels (Online Source 2, Fig. 9d). Thus, it is possible that Daporinad might not be an ideal choice for combination with Etoposide. To validate this prediction, we assessed cell death by propidium iodide (PI) exclusion in NAMPThigh U-87 cells and NAMPTlow U-251 treated with Daporinad (10 nM), Etoposide (1 µM), or both. As we observed when assessing cell viability, Etoposide showed similar effects in both cell lines, while Daporinad exerted a stronger cytotoxic effect in U-251 cells in comparison with U-87 cells (Online Source 2, Fig. 9e). In accordance with the prediction model, Daporinad did not improve the response of these cells to Etoposide (Online Source 2, Fig. 9e).
Combining drugs with TMZ could be a promising approach to address GBM challenges, such as invasiveness and resistance to conventional therapies. We conducted an integrative analysis, assessing predicted sensitivity to TMZ and gene expression in canonical pathways [17] (Online Source 2, Fig. 10, Fig. 11). TCGA GBM patients were classified into "low" and "high" expression groups for each pathway and predicted IC50 values for TMZ were evaluated in both groups. Pathways with higher TMZ IC50 values in the 'high' expression group suggest that combining TMZ with drugs that inhibit those pathways could enhance its efficacy. This analysis revealed potential therapeutic combinations of TMZ with inhibitors of these pathways that were previously tested in clinical trials for other diseases (Online Source 5)