Validation of the prognostic value of 9-gene ATE score for IDH wild-type glioblastoma

We read with great interest the article by Johnson et al.¹ They successfully developed a gene expression-based prognostic signature for isocitrate dehydrogenase (IDH) wild-type glioblastoma (GBM) using NanoString gene expression data from six clinical trial datasets. Developed by elastic net penalized Cox regression analysis, the ATE score, short for ARTE, TAMIGA, and EORTC 26101 trials, was significantly prognostic for overall survival (OS) in both training and validation cohorts. However, the prognostic signature was developed based on the NanoString gene expression platform, which was not commonly used compared with high-throughput sequencing (HTS) platforms. Hence, we aimed to validate the prognostic value of ATE score in the Cancer Genome Atlas (TCGA) GBM cohort, which was measured by HTS assays.

The level-3 RNA sequencing (AffyU133a platform) and somatic mutation data (whole-exome sequencing) of GBM patients were downloaded from the TCGA database (https://portal.gdc.cancer.gov/). After excluding samples without intact clinical information, we finally enrolled 302 IDH wild-type GBM patients. Multivariate Cox regression analysis was performed on the 9 genes to generate “9-gene TCGA ATE score.” All patients were divided into a high- and low-score group using the median value of ATE score. Kaplan-Meier survival analysis demonstrated high-score patients showed moderately significantly poorer OS (log-rank P = 4.675 × 10⁻²) compared with low-score patients (Figure 1A). When including the full set of clinical variables, multivariate analysis indicated ATE score remained significantly prognostic for OS in the TCGA cohort, consistent with that in ATE, AVAglio, and UCLA cohorts (Figure 1B).

Fig. 1.

(A) Kaplan-Meier survival analysis for OS using the ATE score in the TCGA GBM cohort. (B) Forest plots showing hazard ratios with 95% confidence intervals and P values for clinical variables and ATE score. (C) Comparisons of the infiltration level of stromal (stromal score) and immune (immune score) cells, the ESTIMATE score, and tumor purity in high- and low-ATE score group by boxplots. (D) Comparisons of the abundances of 22 immune cells in high- and low-ATE score group by boxplots. (E) Comparisons of the TMB in high- and low-ATE score group by boxplots. (F) Comparisons of the abundances of blood-derived and microglial TAMs in high- and low-ATE score group by boxplots. (G) Comparisons of the expressions of six immune checkpoint molecules in high- and low-ATE score group by boxplots. (H) Boxplots showing the TIDE scores in high- and low-ATE score group. (I) Comparisons of the proportions of nonresponders and responders to immunotherapy in high- and low-ATE score group. (J) Boxplot showing the difference in the AUC values of TMZ estimated by pRRophetic algorithm based on CTRP database between high- and low-ATE score group. Lower AUC value indicated increased sensitivity to TMZ therapy. “ns” means P > .05, * means P < .05, ** means P < .01, *** means P < .001, and **** means P < .0001. Abbreviations: AUC, area under the dose-response curve; KPS, Karnofsky performance status; MGMT, O⁶-methylguanine DNA methyltransferase; OS, overall survival; TAM, tumor-associated macrophage; TIDE, Tumor Immune Dysfunction and Exclusion; TMB, tumor mutation burden; TMZ, temozolomide.

As the authors said that they can hardly assess tumor microenvironment (TME) based on NanoString data. Fortunately, the evaluation of TME, especially immune infiltration patterns, can be realized in TCGA cohort. ESTIMATE was employed to evaluate the TME, and CIBERSORT was utilized to quantify the compositions of 22 types of tumor-infiltrating immune cells (TIICs) based on transcriptomic profiles of GBM samples.^2,3 As shown in Figure 1C, the stromal and immune scores were significantly higher in high-score group, indicating higher abundances of intratumoral immune and stromal cells. In contrast, tumor purity significantly decreased in the high-score group (P = 3.4 × 10⁻⁶). Only 5 TIICs, mainly adaptive immune cells, showed significant difference between two groups (Figure 1D).

Additionally, tumor mutation burden did not differ significantly between groups (Figure 1E), whereas two immunosuppressive cells, blood-derived and microglial tumor-associated macrophages, were more abundant in high-score group (Figure 1F). The expressions of PD-L2 were higher in high-ATE score group, whereas other checkpoint molecules did not differ significantly between groups (Figure 1G). Tumor Immune Dysfunction and Exclusion (TIDE) algorithm was used to predict the likelihood of immunotherapy responses in GBM patients.⁴ We found no significant difference in immunotherapy sensitivity between two groups (Figure 1H), and also no significant difference in proportion of responders to immunotherapy (34.4% vs 38.4%, high vs low) (Figure 1I). Furthermore, the pRRophetic package in R was used to predict the chemotherapeutic response to temozolomide (TMZ), which was determined by the area under the dose-response curve based on the Cancer Therapeutics Response Portal (CTRP v.2.0, https://portals.broadinstitute.org/ctrp) by integrating the expression profiles of GBM cell lines and TCGA samples.⁵ Higher levels of AUC values indicated more resistance to TMZ therapy in the high-ATE score group (z test, P < .001; Figure 1J).

Studies seeking to identify clinically applicable predictors for prognosis and therapeutic responses for IDH wild-type GBM have been rarely reported in the literature.^1,6 Based on the prognostic signature for IDH wild-type GBM developed by Johnson et al, we successfully validated the ATE score based on HTS data. Furthermore, the associations between ATE score and TME and therapeutic response to immunotherapy and TMZ were also evaluated as a vigorous supplement to the original study.

Funding

This study was supported by the Chinese Academy of Medical Sciences Innovation Fund for Medical Sciences (grant number: 2016-I2M-2-001), the Beijing Municipal Natural Science Foundation (grant number: 7202150 and 19JCZDJC64200(Z)), Tsinghua University-Peking Union Medical College Hospital Initiative Scientific Research Program (grant number: 2019ZLH101), and the Graduate Innovation Fund of the Chinese Academy of Medical Sciences and Peking Union Medical College (grant number: 2019-1002-73).

Conflict of interest statement. The authors have no relevant competing interests to disclose.

Authorship statement. Z.H.W. and L.G. performed the data curation and analysis. Z.H.W., Y.N.W., and M.Q.C. analyzed and interpreted the results. Z.H.W., H.X., B.X., Y.W., and W.B.M. drafted and reviewed the manuscript. All authors read and approved the final manuscript.