INTRODUCTION
Glioblastoma multiforme (GBM) is the most aggressive and common form of primary central nervous system tumors, with an incidence rate of 3-4 cases per 100000 individuals [20,23]. Despite numerous research efforts, substantial advancements in improving overall survival (OS) rates have been intangiblev [23]. Consequently, the median survival time for patients with GBM remains less than 15 months, and the 5-year survival rate is below 5% [7,12,20,23,26,31]. With the advent of molecular diagnostics, as defined in the 2021 World Health Organization (WHO) Classification of Tumors of the Central Nervous System (5th edition), potential therapeutic targets and biomarkers are being actively investigated, offering significant prognostic value [22,31]. Since GBM is inherently heterogeneous both intertumorally and intratumorally, understanding gene expression facilitates decoding GBM molecular complexity and developing targeted therapies through detailed examination of genetic, transcriptomic, and proteomic variations [2]. GBM’s heterogeneity, with diverse cell types and genetic profiles, complicates treatment but molecular approaches can facilitate the categorization of this complex disease, advancing GBM analysis. The selection of TWIST2, GATA3, and HES5 as the focus of this study was driven by their established roles in oncogenesis and their involvement in key tumorigenic pathways, and their potential correlation with GBM survival rates, making them ideal candidates for exploring tumor diversity. TWIST2 is implicated in epithelial-mesenchymal transition (EMT) [25,27], a process known to contribute to tumor invasion and metastasis, which is critical in GBM’s aggressive nature. GATA3, though better known for its role in breast cancer, is involved in neural development and may play a role in mesenchymal-to-epithelial transition (MET) [17,34,38], a key process in tumor differentiation. HES5, a downstream effector of Notch signaling, is vital in maintaining neural progenitor cells (NPCs) and has been linked to cancer stem cell maintenance, which could be critical in GBM recurrence and resistance to therapy [6,10,18]. These transcription factors were selected specifically because they represent potential points of vulnerability in GBM biology that have yet to be extensively studied in this context. Analysis of their expression in GBM may reveal robust biomarkers and new therapeutic targets, thereby enhancing our understanding and treatment of the disease.
Since the late 1960s, when the concept of EMT was first introduced by Hay et al., this process has received significant attention in cancer research due to its role in tumor cell invasion, progression, and metastasis [5,12,15,24,29,30,36]. The EMT process involves epithelial cells losing their tightly packed, polarized characteristics to acquire a migratory mesenchymal phenotype [37]. Recently, GBM tumors with a mesenchymal signature have garnered significant attention because of their highly aggressive nature [14]. TWIST2, a highly conserved basic helix-loop-helix transcription factor, is known to play a critical role in embryogenesis and tumorigenesis, particularly in the context of EMT [25,27]. However, its specific contribution to GBM is still poorly understood and underexplored in the literature. While TWIST2 is a potentially important gene in the context of GBM, more research is required to fully elucidate its role.
GATA3 is a transcription factor crucial for embryonic development, immune responses, and endothelial-to-hematopoietic transition, and its expression is strongly associated with the estrogen receptor-positive phenotype and T cell lineage commitment [4,11,21,32,35]. However, GATA3’s function in GBM remains largely unexplored. Although it is known to influence neuronal maturation and MET [17,34,38], there has been limited research investigating its significance in GBM. Given its involvement in developmental processes and cellular differentiation, GATA3 represents a candidate gene for further study in the context of GBM, where its role is still not fully understood.
Notch signaling is crucial for the maintenance of both normal neural and brain tumor stem cells. HES5, a key Notch pathway effector, regulates early brain development by sustaining NPCs [6,10]. As a Notch target bHLH transcription factor, HES5 is highly expressed in NPCs and decreases as differentiation proceeds [1,19]. While HES5 has been linked to EMT-related gene expression and cancer stem cell maintenance [18], its specific role in GBM remains poorly investigated. Further research is required to determine how HES5 contributes to GBM pathogenesis and resistance to therapy. This study aims to explore TWIST2, GATA3, and HES5’s prognostic significance in GBM, helping to clarify its function as a potential biomarker and therapeutic target.
MATERIALS AND METHODS
This study was approved by the Institutional Ethics Committee of Kangwon National University Hospital, with a waiver of informed consent granted (IRB 2024-07-017).
Data collection and processing
The Cancer Genome Atlas (TCGA) provides extensive molecular, clinical, and histopathological data, significantly advancing cancer research and enabling comprehensive cataloging of the human transcriptome [3,8,13]. The GTEx project, a large-scale research initiative, aims to study the relationship between gene expression and genotype across various human tissues, providing crucial data on the impact of genetic variation [28].
We collected mRNA expression profiles and clinical data of patients with GBM from TCGA (https://portal.gdc.cancer.gov/) and normal tissue data from the Genotype-Tissue Expression (GTEx) Project databases (https://commonfund.nih.gov/gtex).
In this study, we integrated GTEx data with the TCGA dataset, focusing on TWIST2, GATA3, and HES5. Level 3 HTSeq-fragments per kilobase of transcript per million mapped reads (FPKM) data were normalized to transcripts per million reads (TPM), and RNA sequencing data in TPM format from the UCSC Xena database were used. We compared gene expression levels in GBM tissues to those in normal brain tissues to ensure tissue-specific relevance and minimize confounding variables, with the aim of identifying changes in gene expression specifically associated with the pathological state of GBM. In a database of 172 patients with GBM, certain basic characteristics exhibited missing data (n=7). Consequently, data from 165 patients with GBM were utilized for the survival analysis.
Statistical analysis
Baseline characteristics were presented using descriptive statistics, and continuous variables were compared using the Mann-Whitney U test based on their normality. Survival analysis was conducted using Cox proportional hazards models, with prognostic accuracy assessed through Harrell’s C-index and the area under the receiver operating characteristic (ROC) curve (AUC) [9]. The proportional hazards assumption was evaluated using Schoenfeld residuals. OS was analyzed using the Kaplan-Meier method, with the log-rank test used to determine the significance of survival differences predicted by each gene’s cutoff point. The Maxstat package was used to calculate the optimal cutoff value, categorizing patients into high-risk and low-risk groups [16,33]. Statistical significance was set at p-values less than 0.05. All statistical analyses were conducted using R software (version 4.3.2; R Foundation for Statistical Computing, Vienna, Austria).
RESULTS
TWIST2
The comparison of TWIST2 gene expression levels between GBM and normal brain tissues revealed a significantly higher expression in GBM tissues (Fig. 1A). In GBM patients, the median expression level of TWIST2 in GBM tissues (n=172) was 0.4427 (interquartile range [IQR], 0.2470-0.8301), which was substantially higher than the median expression level in normal brain tissues (n=2356), which was 0.1497 (IQR, 0.0713-0.2935) (Table 1). The p-value calculated using the Wilcoxon test was <0.001, indicating a highly significant difference between the two groups. These results suggest that TWIST2 could serve as a potential biomarker for GBM.
The association between TWIST2 gene expression levels and OS was analyzed using Kaplan-Meier curves and hazard ratios (HRs) (Fig. 1B and Table 2). The cutoff points for the high and low expression groups were determined to be 0.3927 using the Maxstat method, which identifies the optimal threshold by maximizing the test statistic.
The Kaplan-Meier curves in Fig. 1B visually depict the survival probabilities over time for different risk groups based on gene expression levels. The survival curve shows that individuals in the high-expression group (red line) have significantly lower OS than those in the low-expression group (blue line). The steep decline in the survival curve for the high-expression group indicated a higher rate of adverse events (deaths) over time. The low-expression group showed a slower decline in survival probability, suggesting better OS outcomes. The separation of the curves between the high and low expression groups underscores the impact of gene expression on OS. The log-rank test p-value was <0.001, indicating a highly significant difference in survival between the low-expression and high-expression groups for TWIST2.
In the cohort study, the high expression group (n=96) had an event rate of 87.98 per 100 person-years (PYR), whereas the low expression group (n=69) had an event rate of 49.87 per 100 PYR. The HR for the high-expression group was 2.04 (95% confidence interval [CI], 1.38-3.01; p<0.001), indicating that higher TWIST2 expression levels are associated with a significantly increased risk of adverse outcomes. These findings suggested that TWIST2 expression is a prognostic indicator in patients with GBM.
The ROC curves and the predictive performance metrics for TWIST2 are presented in Fig. 1C and Table 3. The ROC curve illustrates the trade-off between sensitivity (true positive rate) and 1-specificity (false positive rate) across different thresholds. The AUC measures the model’s ability to distinguish between positive and negative cases, with an AUC of 0.567 indicating moderate discriminative power for TWIST2.
Sensitivity, the true positive rate, is 0.781, indicating that the model correctly identified 78.1% of the true positive cases. Specificity, the true negative rate, was 0.404, indicating that the model correctly identifies 40.4% of the true negative cases. The positive predictive value (PV+), or the proportion of positive results that were true positives, was 0.543, whereas the negative predictive value (PV-), the proportion of negative results that were true negatives, was 0.258.
These metrics suggest that while TWIST2 has high sensitivity, its specificity and predictive values are low, indicating potential limitations in its predictive accuracy. Nonetheless, these findings highlight the potential of TWIST2 as a biomarker for identifying high-risk patients.
GATA3
GATA3 expression was significantly higher in GBM tissues than that in normal brain tissues (Fig. 2A). In patients with GBM, the median expression level of GATA3 in GBM tissues (n=172) was 0.2805 (IQR, 0.1404-0.5534), which was substantially higher than the median expression level of 0.1580 (IQR, 0.0812-0.3088) in normal tissues (n=2356) (Table 4). The p-value calculated using the Wilcoxon test was <0.001, indicating a highly significant difference between the two groups. These findings suggest that GATA3 may serve as a potential biomarker of GBM.
The relationship between GATA3 expression levels and OS was assessed using Kaplan-Meier survival analysis and HR (Fig. 2B and Table 5). A cutoff point of 0.3018 distinguished the high- and low-expression groups. The Kaplan-Meier curve shows that patients in the high-expression group (red line) had significantly lower OS than those in the low-expression group (blue line), with a p-value of 0.0019.
In the study cohort, the high expression group (n=83) had an event rate of 87.59 per 100 PYR, whereas the low-expression group (n=82) had an event rate of 55.73 per 100 PYR. The HR for the high-expression group was 1.83 (95% CI, 1.25-2.69; p=0.002), indicating that higher GATA3 expression levels are associated with a significantly increased risk of adverse outcomes. These findings indicated that GATA3 expression is a prognostic indicator in patients with GBM.
The predictive performance of GATA3 gene expression was assessed using a ROC curve and the associated metrics, as shown in Fig. 2C and Table 6. GATA3 had an AUC of 0.536, indicating modest discriminative ability.
The sensitivity was 0.360, indicating that the model correctly identified 36% of the true-positive cases, whereas the specificity was 0.769, correctly identifying 76.9% of true-negative cases. The PV+ was 0.646, and the PV- was 0.226.
These results suggest that while GATA3 has reasonable specificity and PV+, its sensitivity and PV- are low, limiting its effectiveness as a predictive biomarker.
HES5
A comparison of HES5 expression levels between GBM and normal brain tissues demonstrated a significantly higher expression in GBM tissues (Fig. 3A). In patients with GBM, the median expression level of HES5 in cancer tissues (n=172) was 1.3263 (IQR, 0.7126-1.8134), which was substantially higher than that in normal tissues (n=2356) of 0.0855 (IQR, 0.0434-0.1760) (Table 7). The p-value calculated using the Wilcoxon test was <0.001, indicating a highly significant difference between the two groups. These findings suggested that HES5 may serve as a potential biomarker in GBM.
The relationship between HES5 expression levels and OS was assessed using Kaplan-Meier survival analysis and HR (Fig. 3B and Table 8). A cutoff point of 1.3470 distinguished the high-and low-expression groups. Interestingly, the Kaplan-Meier curve demonstrated that patients in the high-expression group (red line) had significantly higher OS than those in the low-expression group (blue line), with a p-value of 0.0039. This unexpected result suggests that HES5 may play a dual role in GBM progression, potentially acting as both a marker of aggressiveness and an indicator of improved survival. This phenomenon warrants further investigation to elucidate the underlying mechanisms.
In the study cohort, the high expression group (n=79) had an event rate of 55.07 per 100 PYR, whereas the low expression group (n=86) had an event rate of 85.79 per 100 PYR. The HR for the high-expression group was 0.57 (95% CI, 0.39-0.83; p=0.004), indicating that higher HES5 expression levels are associated with a significantly decreased risk of adverse outcomes. These findings indicate that HES5 expression is a prognostic indicator in patients with GBM.
The predictive performance of HES5 gene expression was assessed using a ROC curve and associated metrics, as shown in Fig. 3C and Table 9. HES5 has an AUC of 0.538, indicating modest discriminative ability.
The sensitivity was 0.939, indicating that the model correctly identified 93.9% of the true-positive cases, whereas the specificity was 0.173, correctly identifying 17.3% of true negative cases. The PV+ was 0.438, and the PV- was 0.287.
Despite its high sensitivity, the low specificity indicates difficulty in correctly identifying true- negative cases. This suggests that although HES5 expression is useful for identifying positive cases, it may not be as reliable for distinguishing negative cases. Therefore, HES5 could be considered a potential biomarker for positive case identification.
Multigene model
Lastly, we looked at the illustrative comparison of predictive accuracy for TWIST2, GATA3 and HES5 using Harrell’s C-Index (Fig. 4). Individually, TWIST2-based model had a C-index of 0.5700 (95% CI, 0.5160-0.6240), GATA3-based model had a C-index of 0.5720 (95% CI, 0.5180-0.6260), HES5-based model had a C-index of 0.5680 (95% CI, 0.5150-0.6220), all suggesting moderate prognostic performance with statistically significant p-values (TWIST2 : <0.001; GATA3 : 0.002; HES5 : 0.004). The confidence interval indicates statistical uncertainty around the C-index estimate, which should be considered when interpreting the prognostic accuracy of a model. A C-index value closer to 1 indicates perfect prediction accuracy, whereas a value of 0.5 indicates no predictive ability. The combined model of TWIST2, GATA3, and HES5 showed a C-index of 0.6260 (95% CI, 0.5620-0.6890) with a highly significant p-value of <0.001, indicating superior predictive performance. These findings suggest that a combination of these genes provides more accurate predictions than each individual gene, emphasizing the enhanced predictive power of the multigene model.
DISCUSSION
Gene expression analysis and prognostic modeling
The present analysis reveals significant overexpression of TWIST2, GATA3, and HES5 in GBM, with robust statistical support (p<0.001). TWIST2 demonstrated a striking differential expression in GBM, with a median of 0.4427 (IQR, 0.2470-0.8301) compared to 0.1497 (IQR, 0.0713-0.2935) in normal tissues. Similarly, GATA3 exhibited elevated expression in GBM tissues, with a median of 0.2805 (IQR, 0.1404-0.5534), surpassing the levels observed in normal tissues (0.1580; IQR, 0.0812-0.3088). HES5 emerged as the most differentially expressed, with a median expression of 1.3263 (IQR, 0.7126-1.8134) in GBM, vs. 0.0855 (IQR, 0.0434-0.1760) in normal tissue. These findings underscore the potential pathogenic role of these transcription factors in GBM. The multigene model incorporating TWIST2, GATA3, and HES5 achieved a Harrel’s C-index of 0.6260 (95% CI, 0.5620-0.6890; p<0.001), significantly outperforming individual gene models (TWIST2 : 0.5700; GATA3 : 0.5720; HES5 : 0.5680).
This enhancement in predictive accuracy demonstrates the added value of a multigene approach in capturing the intricate molecular landscape of GBM. While the individual gene models revealed moderate prognostic capabilities, the combined model more effectively reflects the interplay of factors that drive tumor progression, offering a clearer understanding of patient outcomes.
Mechanistic insights and clinical implications
The mechanistic relevance of TWIST2, GATA3, and HES5 in GBM biology is strongly supported by existing literature, reflecting their key roles in tumor dynamics. While the application of advanced techniques like the Lasso penalty could have further refined the model by selecting the most predictive variables from high-dimensional data, the selection of TWIST2, GATA3, and HES5 remains strongly substantiated by their established contributions to GBM progression. TWIST2 is intimately involved in facilitating EMT, which is essential for tumor invasion and metastasis. HES5, within the Notch signaling pathway, is critical for the maintenance of cancer stem cell populations, a major factor in GBM’s resistance to conventional therapies. Additionally, GATA3, though more traditionally associated with other cancers, plays an emerging role in modulating tumor differentiation in GBM.
The clinical significance of these findings is considerable. Inhibiting TWIST2 could provide a means of impairing EMT, thereby reducing metastatic potential. Similarly, targeting HES5 within the Notch signaling pathway presents an opportunity to disrupt cancer stem cell maintenance, offering a promising strategy to counter GBM’s notorious resistance to existing therapies. These therapeutic insights underscore the potential for developing targeted treatments, including TWIST2 inhibitors and Notch pathway modulators, to enhance the efficacy of current treatment regimens.
Molecular diagnostics and relevance of TCGA
The 2021 revision of the WHO Classification of Tumors of the Central Nervous System reflects a broader shift toward integrating molecular diagnostics to refine GBM classification, incorporating markers such as isocitrate dehydrogenase mutations, TERT promoter mutations, and EGFR amplification. Although TCGA dataset predates the integration of these contemporary markers, it remains a robust and indispensable resource for examining gene expression in GBM. The use of TCGA in this study enables a comprehensive analysis of TWIST2, GATA3, and HES5, transcription factors that are not traditionally included in GBM’s molecular landscape but whose emerging relevance is evident from their mechanistic roles.
Incorporating TWIST2, GATA3, and HES5 into established prognostic models would substantially enhance current classification frameworks by addressing key biological processes, including EMT and stem cell maintenance. These factors, in complementing conventional markers, provide a more nuanced understanding of GBM pathogenesis and prognosis. Furthermore, expanding the molecular scope to include novel, as-yet-undiscovered genes will enhance the precision of patient stratification, facilitating more personalized treatment approaches.
Statistical considerations and future directions
The statistical analysis employed in this study yielded significant p-values for each of the transcription factors examined, providing strong support for their prognostic relevance in GBM. While future investigations may benefit from incorporating additional methods to control for false discovery rates, such as Benjamini-Hochberg corrections, the current focus on these three transcription factors allows for a precise and targeted exploration of their biological significance. This focused approach is particularly valuable in highlighting the specific prognostic utility of TWIST2, GATA3, and HES5 in GBM.
In the broader context of ongoing research, applying advanced statistical refinements will further enhance the robustness of comparisons across larger gene sets. That said, the insights derived from this study already provide a solid foundation for future research. The results underscore the necessity of validating these findings in larger, independent cohorts, where the use of more comprehensive statistical techniques will undoubtedly serve to further solidify the clinical implications of TWIST2, GATA3, and HES5. Moreover, continued exploration of the interactions between these transcription factors and existing therapeutic modalities offers a promising avenue for novel treatment strategies. This could lead to more personalized therapeutic interventions, addressing the inherent complexity of GBM biology and potentially improving patient outcomes.
This study provides compelling evidence for the prognostic value of key biomarkers in GBM, setting the stage for more refined predictive models and therapeutic approaches. Further validation in larger cohorts is crucial to understanding the biological mechanisms behind their synergistic effects. Integrating additional clinical and molecular data will sharpen predictive accuracy and improve their practical utility, moving us closer to more personalized and effective treatments for GBM.
CONCLUSION
Predictive accuracy is a crucial asset for patients, especially those with a limited life expectancy. In this groundbreaking study, we examined the expression of three pivotal genes and assessed their significance in predicting survival and risk in GBM using advanced bioinformatic tools.
Our findings reveal a compelling clinical correlation between gene expression patterns, patient survival rates, and high-risk factors. At the heart of this research lies the innovative concept of a multigene model, which ingeniously groups seemingly unrelated genes to unlock significant insights that were previously unattainable.
We stand at the threshold of a new era in which rigorous research and relentless validation will further illuminate the intricate roles played by these genes in GBM progression. The potential therapeutic implications are vast and offer hope for transforming GBM treatment paradigms. This study not only paves the way for future breakthroughs, but also epitomizes the essence of scientific innovation, heralding a new dawn in the fight against one of the most formidable forms of cancer.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation 
Print 
