Mutant-allele tumor heterogeneity in malignant glioma effectively predicts neoplastic recurrence

Abstract

Intra-tumor heterogeneity (ITH) is one of the most important causes of therapy resistance, which eventually leads to the poor outcomes observed in patients with glioma. Mutant-allele tumor heterogeneity (MATH) values are based on whole-exon sequencing and precisely reflect genetic ITH. However, the significance of MATH values in predicting glioma recurrence remains unclear. Information of patients with glioma was obtained from The Cancer Genome Atlas database. The present study calculated the MATH value for each patient, analyzed the distributions of MATH values in different subtypes and investigated the rates of clinical recurrence in patients with different MATH values. Gene enrichment and Cox regression analyses were performed to determine which factors influenced recurrence. A nomogram table was established to predict 1-, 2- and 5-year recurrence probabilities. MATH values were increased in patients with glioma with the wild-type isocitrate dehydrogenase (NADP⁽⁺⁾) (IDH)1/2 (IDH-wt) gene (P=0.001) and glioblastoma (GBM; P=0.001). MATH values were negatively associated with the 2- and 5-year recurrence-free survival (RFS) rates in patients with glioma, particularly in the IDH1/2-wt and GBM cohorts (P=0.001 and P=0.017, respectively). Furthermore, glioma cases with different MATH levels had distinct patterns of gene mutation frequencies and gene expression enrichment. Finally, a nomogram table that contained MATH values could be used to accurately predict the probabilities of the 1-, 2- and 5-year RFS of patients with glioma. In conclusion, the MATH value of a patient may be an independent predictor that influences glioma recurrence. The nomogram model presented in the current study was an appropriate method to predict 1-, 2- and 5-year RFS probabilities in patients with glioma.

Keywords: glioma; intra-tumor heterogeneity; mutant-allele tumor heterogeneity; prediction; recurrence.

Figure 1.

MATH values in different glioma subtypes. (A) Distribution of MATH values among all patients with glioma (LGG and GBM). The MATH values ranged between 5.81 and 90.60, with a mean of 44.61 and a median of 43.28. (B) Boxplot showing the MATH values of patients with different IDH1/2 glioma types. The MATH values found in IDH1/2-wt ranged between 10.12 and 89.15, with a mean of 47.37±16.03. The MATH values found in patients with IDH1/2-mut ranged between 5.81 and 90.60, with a mean of 41.68±14.70. (C) Boxplot showing the MATH values found in patients with glioma with different histological classifications. The MATH values in O, OA, A and GBM were 41.39±15.17, 42.18±15.26, 44.52±14.56 and 46.86±16.13, respectively, and the ranges were 5.81–90.60, 13.07–77.60, 16.87–83.22 and 10.12–89.15, respectively. (D) Boxplot showing the MATH values obtained for patients with different World Health Organization grades of glioma. The MATH values obtained in grade II, III and IV were 41.77±15.15, 43.62±14.84 and 46.86±16.13, respectively, and the ranges were 5.81–90.60, 11.40–83.22 and 10.12–89.15, respectively. Data are presented as the mean ± SD. *P<0.05, ***P<0.001. A, astrocytoma; GBM, glioblastoma; IDH, isocitrate dehydrogenase (NADP⁽⁺⁾); LGG, low grade glioma; MATH, mutant-allele tumor heterogeneity; mt, mutant; O, oligodendroglioma; OA, olio-astrocytoma; wt, wild-type.

Figure 2.

Kaplan-Meier survival curves for patients with glioma with high and low MATH levels. (A) The 5-year RFS rate in 757 patients with glioma (LGG and GBM) with high and low MATH levels. Among patients in the low-MATH group, the RFS rates were 56% at 2 years and 28% at 5 years, whereas the rates were 43 and 17%, respectively, in the high-MATH group (P=0.001). (B) The 5-year RFS rate in 389 patients with glioma with the IDH-wt subtype and high or low MATH levels. The Kaplan-Meier estimates revealed RFS rates of 18% at 2 years and 4% at 5 years in the low-MATH group compared with respective rates of 8 and <0.001% in the high-MATH group (P=0.001). (C) The 5-year RFS rates in 368 patients with the IDH-mut subtype and high or low MATH levels. Patients in the low-MATH group had an RFS rate of 83% at 2 years and 51% at 5 years, whereas the corresponding rates were 80 and 40% in the high-MATH group (P=0.594). (D) The 5-year RFS rates in 345 patients with GBM with high or low MATH levels. Patients in the low-MATH group had RFS rates of 16% at 2 years and 3% at 5 years, whereas those in the high-MATH group had rates of 7 and <0.1%, respectively (P=0.017). The log-rank test was used to determine significance. GBM, glioblastoma; IDH, isocitrate dehydrogenase (NADP⁽⁺⁾); LGG, low grade glioma; MATH, mutant-allele tumor heterogeneity; mut, mutant; RFS, recurrence-free survival; wt, wild-type.

Figure 3.

Somatic mutation sample frequency pattern analysis and gene set enrichment analysis. (A) Mutated genes with higher sample frequency in the low-MATH group of patients. (B) Mutated genes with higher sample frequency in the high-MATH group of patients. (C) Comparison of six common high sample frequency mutation genes (IDH1, TP53, TTN, ATRX, MUC16 and EGFR) at different levels in the MATH group. Comparisons were made using the χ² test. (D) Mutated genes enriched in biological pathways with fold-changes >100. Compared with low-MATH glioma, high-MATH glioma exhibited more gene mutations in the BH3 anti-apoptotic signaling pathway and fewer in the dolichyl phosphate biosynthesis signaling pathway. The gene expression set observed in the high-MATH group was enriched in (E) ‘cell adhesion molecules cams’ (P=0.030; FDR q<0.25) and (F) ‘cytokine cytokine receptor interaction’ signaling pathways (P=0.018; FDR q<0.25). ATRX, ATRX chromatin remodeler; CIC, capicua transcriptional repressor; DNAH3, dynein axonemal heavy chain 3; EGFR, epidermal growth factor receptor; ES, enrichment score; FDR, false discovery rate; FLG, filaggrin; IDH1, isocitrate dehydrogenase (NADP⁽⁺⁾) 1; KEGG, Kyoto Encyclopedia of Genes and Genomes; LRP2, LDL receptor related protein 2; MATH, mutant-allele tumor heterogeneity; MUC16, mucin 16; MUC17, mucin 17; NF1, neurofibromin 1; OBSCN, obscurin, cytoskeletal calmodulin and titin-interacting RhoGEF; TP53, tumor protein p53; TTN, titin.

Figure 4.

Nomogram prediction table for glioma RFS probability. Each patient can be assigned a point value from the point scale for each factor. The cumulative number of points are identified on the total points bar and used to determine the 1-, 2- and 5-year RFS probabilities. A, astrocytoma; IDH, isocitrate dehydrogenase (NADP⁽⁺⁾); MATH, mutant-allele tumor heterogeneity; O, oligodendroglioma; OA, oligo-astrocytoma; RFS, recurrence-free survival; TTN, titin; WHO, World Health Organization.

Figure 5.

Internal validation of the nomogram designed to predict the probability of glioma RFS at 1, 2 and 5 years. ROC curves of the nomogram to predict the probability of glioma RFS at (A) 1, (B) 2 and (C) 5 years. Calibration curve for predicting the probability of RFS at (D) 1, (E) 2 and (F) 5 years. The grey line indicates the ideal nomogram RFS line. The vertical bars represent the 95% CI. AUC, area under the curve; RFS, recurrence-free survival; ROC, receiver operating characteristic.