Attributing Meaning to Molecular Interaction Networks by Leveraging Clinical and Omic Data: The Missing Link between Tumor Biology and Treatment Strategies in Glioma

Abstract

The pace of data growth in the molecular space has led to the evolution of sophisticated approaches to data aggregation and linkages, such as IPA, STRING, KEGG, and others. These tools aim to generate molecular interaction networks harnessing growing molecular data at all levels to link tumor biology knowledge to signaling pathways and matched analyses. Potentially actionable biomarkers, however, are evaluated based on clinically associated prognosis, and necessary computational approaches should be vetted for interpretability through a clinical lens. Intersectional clinical and computational expertise is needed to link omics, molecular interactions, and clinical data to address the missing link between tumor biology and treatment strategies.

1. Introduction

As increasingly large-scale data streams in genomic, transcriptomic, proteomic, and metabolic contexts are becoming available to researchers, molecular signals are emerging at an unprecedented pace. This rise in data and signal has rightfully fostered the emergence of molecular interaction networks toolsets, including Ingenuity Pathway Analysis (IPA) [1], Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) [2], Kyoto Encyclopedia of Genes and Genomes (KEGG) [3] and others to group, annotate and classify emerging signals. While highly sophisticated and evolving, these interactions directly lead to emerging evidence. Still, they are increasingly more challenging to implement with specified training and expertise, and the pressure for rapid interpretation can render the ongoing generation of information biased and may limit the identification of connections in novel signals. Despite the rapid rise in signal, the clinical space and clinical implementation functions in a realm driven by studies that generate information wherein one or very few interventions are analyzed with limited molecular data acquired. Often clinical studies are aimed at testing a specific hypothesis but are difficult to link to large-scale data since they carry clinical but relatively less omic data. Clinical links to omic data and molecular interaction networks are lacking, limiting emerging data classification when employing molecular interaction networks. Molecules widely studied reemerge in linkage analyses but still lack a clear mechanistic context that can be effectively exploited for therapeutic gain. Sex differences have increasingly been studied in glioma, but specific classification via molecular interactions is evolving. Equally so, there is no linkage between tumor burden, response to particular interventions over time, and survival endpoints are generally employed with limited data on progression. Some clinical characteristics such as age and sex are linked to prognosis and biological triggers and thus present important avenues for classifying molecular networks to allow for a superior understanding of biology and to identify predictive biomarkers. Signaling routes that are gaining momentum and experience in glioma, including Notch, NF-κB signaling, ferroptosis, the lipidome, fatty acid metabolism, and the propagation, evolution, and sustainment of stem cells, all overlap in interaction and require more research to define in omic terms. Treatment strategies have centered on signals surpassed by rapidly evolving knowledge in the field and do not connect to molecular subtypes emerging in proteogenomic analyses. The emerging rapidly changing landscape of databases and interaction networks is surfacing new connections in large-scale omic data, which will be discussed, highlighting pathways shared among analyses leading into biological mechanisms and therapeutic strategies existing and ongoing to eventually leverage clinical and omic data in glioma.

2. The growing landscape of molecular interaction network databases and connections to glioma

Molecular interaction networks are growing to meet researcher demands, with over 50,000 publications in Web of Science as of mid-2023. Notably, in this context, 21% of publications are related to biochemistry and molecular biology, and 4.3% to oncology. The growth of interest and research in this space is evidenced by a rise in publications from 41 in 2012 to 280 in 2022 [4]. Paralleling this interest in aggregating data into networks is the growth in tools making this possible. There are several reasons for the existence of multiple avenues of aggregating, analyzing, and visualizing molecular data since each meets different needs, either with different data sources or approaches to linkage (Figure 1). As toolsets gather information from evidence-based medicine from various sources and aggregate this in classification networks, there is source and data overlap but also distinctions (Table 1). STRING (Search Tool for the Retrieval of Interacting Genes/Proteins) gathers protein interaction information from the annotated proteomes maintained by SWISS-PROT with functional information on genes derived from a manually curated orthology database, Clusters of Orthologous Genes (COG) when available and when not available then automatically derived by an automatic method resembling the COG methodology [2] to generate protein association networks. The linkage of data sources that are widely available to other data sources, e.g., TCGA (The Cancer Genome Atlas), CGGA (Chinese Glioma Genome Atlas), and Gene Expression Omnibus (GEO), facilitates the identification of networks and the assignment of meaning to novel protein signals [6]. As exemplified by Li et al., STRING can be employed to connect prognosis-related genes to the glioma microenvironment, which in their study was accomplished by employing TCGA and GSE4290 data sets to arrive at several previously lesser known or unknown genes that share both a link to the glioma microenvironment and prognosis [7]. Mischkulnig et al. employed TCGA data and STRING to link heme biosynthesis regulation to glioma aggressiveness by connecting heme biosynthesis RNA expression to molecular markers associated with glioma grade [8]. The KEGG (Kyoto Encyclopedia of Genes and Genomes) database [3], initially created in 1995, is a conglomerate of 16 databases and functions by gathering published literature to map omic signals to pathways (Figure 2). KEGG has been employed to compare normoxic and hypoxic glioblastoma stem-like cell lines using co-expression analysis and linkage to the gene expression dataset GSE117 to identify hub genes, which were then subsequently screened with STRING to examine protein-protein interactions [9]. As an example of linkage of a novel signal, KEGG has been employed to analyze the possible function of FUBP3, identified by Li et al. as the only gene associated with survival out of 5 intersection genes in 2 GEO datasets [10]. FUBP3 was found to have an immunohistochemical expression in GBM and adjacent tissue with KEGG, linking it to association with immune surveillance in GBM. IPA (Ingenuity pathway analysis), developed by Qiagen, is a combination of functional analysis and databases that employs a variety of data sources, including public data and proprietary data, and allows linkage with experimental data uploaded by users of the platform. It also employs manual curation of existing published data to allow for annotation. IPA can thus evolve to include a wide array of canonical pathways, as exemplified by Ghosh et al., with core canonical pathways identified in GBM [11]. As novel omic signals are identified in response to management, these can be linked to canonical pathways [12]. WGCNA (Weighted Gene Co-expression Network Analysis) analyzes proteogenomic connections by generating mapping co-expression of molecules to modules [13]. Candidate biomarkers can thus be linked to molecular mechanisms as exemplified by Yang et al., wherein WGCNA was utilized to link the gene expression profile of GSE50161 to 47 tissue samples, of which 34 were surgical brain tissue samples, 47 from patients with GBM and 13 from normal brain, in this case, pediatric epilepsy patients [14]. In this analysis, the modules identified using WGCNA were examined using an independent TCGA data set to arrive at several molecules that may be utilized as signals in liquid biopsy. Co-expression modules in WGCNA have been employed to link expression profile data to survival. Zhou et al. used TCGA expression profile data and clinical information to screen out genes related to the prognosis of GBM, dividing patients into high- and low-risk groups [15]. The biological processes identified using GO analysis were cell cycle but less intuitively progesterone-mediated oocyte maturation and oocyte meiosis related with CDCA5 and CDCA8 identified as genes of interest with expression levels related to overall survival. GSEA (Gene Set Enrichment Analysis) [16] is a computational technique that is also often employed, producing sets of genes based on a differential expression that allows ranking and calculation of enrichment scores that can then be connected to biological processes. Hypoxia risk signatures have been obtained using TCGA and CGGA for low-grade glioma (LGG) and GBM using GSEA [17].

Figure 1.

The complex interplay results in molecular interaction networks based on one-to-one and one-to-many relationships between available clinical data and published omic data. Unexplored clinical data and unidentified or unpublished markers do not connect to molecular networks despite one-to-many relationships between data components [5].

Database	Source of data and output	Usage in literature for Glioma
STRING (Search Tool for the Retrieval of Interacting Genes/Proteins) [2, 6]	https://string-db.org/; SWISS-PROT, COG; Protein-protein interactions	Li et al., 2020 Bioinformatic Profiling of Prognosis-Related Genes in Malignant Glioma Microenvironment [7]; Mischkulnig et al. 2022. The impact of heme biosynthesis regulation on glioma aggressiveness: Correlations with diagnostic molecular markers [8]
KEGG (Kyoto Encyclopedia of Genes and Genomes) [3]	https://www.kegg.jp/; Public databases, published literature curation, protein structure databases, genome and chemical databases; Pathways, Interactions	Guvem et al., 2022 Screening the Significant Hub Genes by Comparing Tumor Cells, Normoxic, and Hypoxic Glioblastoma Stem-like Cell Lines Using Co-Expression Analysis in Glioblastoma. [9]; Li et al., 2022. Identification of a key glioblastoma candidate gene, FUBP3, based on weighted gene co-expression network analysis [10]
IPA (Ingenuity Pathway Analysis) [1]	http://www.ingenuity.com/; Public databases published literature curation, proprietary databases, and researcher-uploaded experimental data; Pathways, Interactions	Ghosh et al., 2017. Core Canonical Pathways Involved in Developing Human Glioblastoma Multiforme (GBM) [11]; Tasci et al. 2023. RadWise: A Rank-Based Hybrid Feature Weighting and Selection Method for Proteomic Categorization of Chemoirradiation in Patients with Glioblastoma. [12]
WGCNA (Weighted Gene Co-expression Network Analysis) [13]	Data analysis technique; Correlated genes and proteins along modules of co-expressed genes	Yang et al. 2018. Candidate Biomarkers and Molecular Mechanism Investigation for Glioblastoma Multiforme Utilizing WGCNA [14]; Zhou et al., 2021. Construction of co-expression modules related to survival by WGCNA and identification of potential prognostic biomarkers in glioblastoma. [15]
GSEA (Gene Set Enrichment Analysis) [16]	Gene sets defined based on differential expression; Ranked lists of gene sets	Lin et al., 2020. Characterization of Hypoxia Signature to Evaluate the Tumor Immune Microenvironment and Predict Prognosis in Glioma Groups [17]

Table 1.

Major omic interaction databases and bioinformatics applications and connections to glioma literature.

Figure 2.

KOBAS bubble plot of KEGG pathways enriched in the set of significantly differentially expressed proteins between pre-and post-treatment based on a paired t-test in GBM. (A). Upregulated genes. (B). Downregulated. The bubble size reflects the KOBAS hypergeometric test p-value broken into ranges. Colors reflect clusters of related pathways [18].

There is currently no single database or bioinformatics analysis that is considered the gold standard, and thus, most studies employ several approaches in part to validate that the signals map to similar networks or pathways when using a dataset. Most studies employ proteogenomic data from TCGA and Clinical Proteomic Tumor Analysis Consortium (CPTAC) or examine original data in conjunction with repository data from public databases (Table 2).

Author	Title	Sample /Technique	Clinical factors	Survival analysis	Time points
Lam KHB, et al. Nat Commun. 2022 Jan 10;13 [1]:116 [19].	Topographic mapping of the glioblastoma proteome reveals a triple-axis model of intra-tumoral heterogeneity.	20 patients; Tumor tissue/Mass spectrometry	Age, sex, resection	No	One (tissue)
Duhamel M, et al. Nat Commun. 2022 Nov 4;13 [1]:6665 [20].	Spatial analysis of the glioblastoma proteome reveals specific molecular signatures and markers of survival.	96 patients; 46 tumors analyzed mass spectrometry-based spatially-resolved proteomics guided by mass spectrometry imaging; Integration of protein expression and clinical information, a 5-protein signature is associated with survival. The expression of these 5 proteins was validated by immunofluorescence on an additional cohort of 50 patients.	Age, sex, KPS, resection, tumor location, molecular features	Yes	One (tissue)
Rose M, et al. Front Immunol. 2021 Sep 27;12:746168 [21]	Surfaceome Proteomic of Glioblastoma Revealed Potential Targets for Immunotherapy.	Cell lines; Proteins verified in patient GBM using spatial proteomic guided by MALDI-mass spectrometry	none	No	One (cell lines, tissue)
Syafruddin SE, et al. BMC Cancer. 2021 Jul 23;21 [1]:850 [22]	Integration of RNA-Seq and proteomics data identifies glioblastoma multiforme surfaceome signature.	RNA-Seq data from TCGA GBM (n = 166) and GTEx normal brain cortex (n = 408)	Age, sex, treatment	No	One (tissue)
Yanovich-Arad G, et al. Cell Rep. 2021 Mar 2;34 [10]:108787 [23].	Proteogenomics of glioblastoma associates molecular patterns with survival.	87 GBM patients high-resolution mass spectrometry proteomics and RNA sequencing (RNA-seq). Integrative analysis of protein expression, RNA expression, and patient clinical information	Age, sex, KPS, resection, tumor location, IDH status, treatment	Yes	One (tissue)
Wang LB et al. Cancer Cell. 2021 Apr 12;39 [4]:509-528.e20 [24].	Clinical Proteomic Tumor Analysis Consortium. Proteogenomic and metabolomic characterization of human glioblastoma.	Integrated analysis of genomic, proteomic, post-translational modification, and metabolomic data on 99 treatment-naive GBMs	Age, sex, BMI, race, ethnicity, smoking history, tumor location	Yes	One (tissue)
Krauze AV et al. Frontiers in Oncology. Pending print [18]	Glioblastoma survival is associated with distinct proteomic alteration signatures post chemoirradiation in a large-scale proteomic panel.	Proteogenomic analysis pre vs. post chemoirradiation in 83 GBM patients	Age, sex, KPS. RPA, tumor location, resection, MGMT status, management, radiation volumes	Yes	Two Serum proteome pre and post-treatment

Table 2.

Major publications linking the GBM proteome to clinical factors.

Several barriers emerge in attributing meaning to molecular interaction networks, however. It is worthwhile noting that most published studies that aim to link clinical and omic data do so by leveraging several databases to filter and assign meaning to the measured signal. There is also significant concern that incorrect meaning may be attributed to molecule modification in different scenarios based on the malignancy studied, the treatment, and the tissue wherein the signal was obtained (Figure 1). Measurements augment this concern in specific settings with only one time point. The signals arrived at using the abovementioned studies are all largely distinct, given different data acquisition means, various data linkages, and analysis timing as databases evolve. This causes difficulty in biological understanding since few signals are common between studies, and novel markers are often not validated in independent cohorts and might not present an opportunity for validation due to difficulty in the sample acquisition (e.g., tumor tissue), cost, or limited sample available for multiple analyses. Lack of validation poses significant challenges. Thus results need to be interpreted cautiously since several alternative versions of the same linkage are conceivable (Figure 1), given additional published data. It is anticipated that as connections grow, the annotation will improve. However, the annotation will continue to be subject to bias as new clinical data and unidentified or unpublished markers will not link potentially missing meaningful biological connections that may not be reflected or surfaced in interaction databases and bioinformatics applications. This is augmented by the ambiguity of captured biological signals and attribution of differentially expressed signals to upregulated and downregulated signaling pathways (Figure 2), making interpretation difficult. The attribution of meaning to omic data will be discussed in the next section.

3. Attribution of meaning to omic data using proteogenomic analysis

The meaning assigned to novel signals along molecular networks can align with specific cells cell types, specific pathways, clinical aspects of the disease, or specific outcomes (Table 2). Increasingly studies link tissue culture, animal, and human data with outcomes data along molecular interaction networks. The attribution of meaning to omic data can be enhanced by connecting it to mechanistic roles via pathways (Figure 2), to other omic data (Figure 3), and to clinical features to elicit linkage to patient outcomes. Existing and established connections may be the result of data linked to a specific disease process or intervention or a novel scenario, such as the study of a novel intervention or agent. Additional connections can be based on data sources such as imaging and pathology slides that involve data analysis based on structural or textural features. The previous section discussed tool sets linking largely undefined signals to molecular interaction pathways. The results of these analyses have added insight to our understanding of glioma heterogeneity. The work of Lam et al., using a liquid chromatography-tandem mass spectrometry (LC-MS/MS)-based proteomic analysis on tumor tissue to generate an atlas that connects the levels of 4794 proteins to hallmark histomorphologic niches to define intra-tumoral molecular patterns [19]. The proteomic patterns identified were based on anatomical niches aligning less strongly with the Verhaak signature [26] and were not included due to low concordance between protein and RNA. This study ultimately resulted in three groups defined by KRAS_targets, MYC_targets, and Hypoxia to define heterogeneity in GBM [19]. When comparing the groups to drug response employing 543 drugs in the CTRPv2 dataset to GBM cells grown in culture, heterogeneity was again observed but also diminished response based on hypoxia signatures with some differential response in MYC-enriched vs. KRAS-enriched acknowledging limitations of the study. Without the histological annotation, but rather using spectrometry-based spatially-resolved proteomics guided by mass spectrometry imaging, Duhamel et al. examined GBM heterogeneity in 46 tumors in relation to survival, resulting from arriving at three molecular groups associated with immune, neurogenesis, and tumorigenesis signatures [20]. The groups identified were, by contrast, not associated with unique histological areas. Group A (cluster 2) was associated with neuro-developmental genes, linked to neurogenesis and axon guidance; group B (cluster 1) with microglial activation, including iron transporters and proteins involved in coagulation, and group C (cluster 3) was associated with tumor growth. The authors concluded that heterogeneity is microenvironment-specific and presented a 21 gene signature and 5 proteins associated with survival. Two notable proteogenomic approaches have recently connected multi-omic data [23, 24]. These comprehensive data linkage studies reveal the complexity of the interplay between proteomic, transcriptomic, and genomic data. A recent MS-based proteomics and RNA-seq analysis identified molecular differences associated with survival. The analysis showcases the complexity of data linkage and difficulty in sample acquisition, which utilized 84 samples, 54 with high-resolution proteomic data, 65 with high-quality RNA-seq data, and 32 with both [23]. Given the importance of this data and the burden of disease in glioma, such a sample size illustrates the difficulties faced by the field in integrating omic analysis to arrive at meaningful and actionable conclusions. The identified proteomic modules were associated with survival, whereas the RNA modules did not, unless the IDH mutated samples were included in the analysis. A good correlation between proteomics and transcriptomics was noted in sex -correlating modules [23]. This data also linked increased fatty acid oxidation with shorter survival and oxidative phosphorylation with longer survival time. A subsequent study involving integrated proteogenomic and metabolomic data from 10 platforms, including RNA sequencing, DNA methylation arrays with whole genome sequencing investigating 99 GBMs from CPTAC and 10 unmatched GTEx normal brain samples revealed four immune subtypes with histone acetylation associations, lipid and metabolome profiles. The key signaling pathways that emerged were RTK/RAS, PI3K/AKT, and p53/cell cycle [24]. Dysregulation in signaling involving RTK, PI3K, WNT, and NOTCH pathways was present in all tumors with distinct drug connectivity analysis along EGFR and NF1 altered signatures. Another promising facet of analysis, given the relevance to druggable targets, is the surfaceome. Data from GBM cell lines examining the surfaceome with spatial proteomic-guided MALDI-mass spectrometry identified 87 overexpressed proteins, with 7 already in clinical trials and 3 of unknown interaction [21]. An integrated analysis of RNA-Seq data from TCGA and GTEx normal brain cortex databases was carried out to examine the surfaceome in GBM and identified 2381 dysregulated genes, of which 395 were surfaceome related, and a 6 gene signature was arrived at, including HLA-DRA, CD44, and SLC1A5, EGFR, ITGB2, PTPRJ with several clinically approved drugs potentially effective [22]. These studies take the large proteogenomic signals and arrive at pathways of significance that allow the classification of glioma. However, the connection to prognosis is inconsistent, and although there are mechanistic commonalities among studies, specific protein signals vary, with very few in common among studies. Existing or evolving pathways that may not yet map in molecular interaction pathways will require a referencing process to evolving data. This is embedded in several of the approaches in Table 2, using manual curation and proprietary linkages. In glioma, the most common linkages are made to hypoxia, angiogenesis, epithelial-mesenchymal transition, glioma stem cells, and the interplay of metabolism and immune response with these hallmarks of cancer. Because certain molecules are more widely studied and important data is present to draw from, such as p53 and EGFR, irrespective of the data input, p53 or EGFR will emerge as linkage centric. Sex differences have increasingly been studied in glioma, but specific classification via molecular interactions is evolving. Equally, there is no linkage along the lines of tumor burden with no omic signature of tumor burden currently defined. Some clinical characteristics of glioma are linked to glioma type and prognosis and thus present important avenues for classification, such as age at diagnosis, which correlates with the type of glioma and prognosis [27]. Novel proteins in any one malignancy will, however, map to networks wherein the protein has been described in another malignancy, limiting the ability to investigate its meaning in specific scenarios.

Figure 3.

The proteome results from gene expression and subsequent modification of resulting proteins. Filtration of the signal along the path from genome to transcriptome to proteome with transcription and then translation of information as mechanistic steps challenging linkages along molecular interaction networks [25].

4. Emerging pathway and omic signatures in glioma

Pathways that have emerged in previous studies have been described in other studies and merit more in-depth analysis. Several, including Notch, NF-κB signaling, Ferroptosis, lipidome, fatty acid metabolism, stem cell propagation, evolution, and sustainment, appear more prominently and will be highlighted here. Notch has emerged in several recent proteogenomic analyses [18, 23, 24] as a potential mediator of glioma proliferation, invasion, stemness, and resistance to treatment [28, 29]. It has been connected to sex differences and response to treatment [30] and is a promising immunotherapy target [31]. Notch, possibly regulated through posttranslational modification, has been implicated in other cancers and linked to regulating multiple genes, including EGFR and NF-κB [32]. NF-κB has relationships to subunits linked to poorer outcomes at the RNA level and longer survival at the protein level, while Notch signaling was tumor-promoting overall [18, 23] with evolving directionality with respect to the protein level. Ferroptosis, a novel mechanism of cell death, has been increasingly described in glioma with connections to the lipidome and known mediators, including p53 and MAPK, and iron metabolism [33, 34]. It is increasingly associated with adaptive resistance mechanisms and may be employed in future therapeutic avenues [35]. Given the connection to iron metabolism, ferroptosis-based interactions in proteogenomic analysis via heme signaling and peroxisome classifications [18, 36, 37]. Several agents, including temozolomide, target ferroptosis [38]. Its links to the lipidome will likely benefit from increasing analysis as connections between the metabolome and the lipidome and tumor resistance are growing in glioma. Fatty acid metabolism is of increasing interest for multiple reasons. Gliomas have an increased level of lipid content as compared to normal tissues, and the brain has a higher lipid content than other tissues, making the lipidome an exciting avenue of research since it can allow for the classification of omic signal [39]. The lipidome is liked to radiation resistance [40] and metabolism, including relationships to IDH mutation, EGFR, PTEN, and MGMT promoter mutation via association with alterations in lipid metabolism [41]. Stem cells as mediators of treatment resistance have been an essential facet of glioma research and have now been connected to virtually every signaling pathway. In their histomorphologic niche analysis, Lam et al. found that the infiltrating tumor region was associated with both neuronal systems and stem cell-related pathways [19], while another proteogenomic study identified a cluster related to the regulation of stem cell maintenance [23]. Lower survival in glioma has been linked to stem cell burden [18, 23]. Still, there is currently no robust stem cell signature in part because the formation and sustainment of stem cells are deeply linked to the microenvironment, metabolism, and adaptive resistance, fostering heterogeneity and is also a dynamic process that is only partially captured in single time point, single tissue sample analyses [42].

5. Tumor biology and treatment strategies in glioma

Despite the data proliferation in the glioma, including its analysis along genomic, transcriptomic, and proteomic data sets now extending over 10 years [43], management has remained the same. Typical management is still comprised of maximal safe resection followed by concurrent chemoirradiation and adjuvant temozolomide. Tumor failure ensues almost universally, and upon recurrence, rarely management, including chemotherapy and re-irradiation, results in relatively limited benefit. The use of additional agents to the standard of care has yet to significantly improve outcomes and is implemented on trial with no standard of care agent yet to be implemented. Meanwhile, only 10% of patients benefit from genomic analysis.

Molecular subtypes based on RNA sequencing have not been connected to prognosis and have become challenging to connect to other omic data based on clinical samples [26, 44]. Biomarkers for response and resistance are lacking, and prognostic markers have struggled to reach predictive roles [45]. Criteria such as REMARK (Reporting Recommendations for Tumor Marker Prognostic Studies) have needed to be more consistently implemented, and a plethora of emerging markers are difficult to conceptualize for inclusion in clinical trials [46]. As a result, emerging novel agents had theoretical backing for biological mechanisms of action that should have translated into clinical benefit, but limited benefit emerged for multiple reasons.

Several targets and approaches predominate. EGFR has been extensively studied since it is the receptor tyrosine kinase most amplified in glioma. However, its mechanistic implications remain unclear, and despite several attempts at targeting EGFR in clinical trials, no significant outcome benefits have been obtained. This is likely multifactorial, including tumor heterogeneity and a variety of mechanisms for EGFR activation, which is complex with both overexpression of the receptor and ligand production [47]. Therapeutic attempts have included tyrosine kinase inhibitors (Erlotinib, Gefitinib, Lapatinib, Osimertinib), monoclonal antibodies (Cetuximab), immunotherapy (CAR-T cells targeting EGFRvIII) and others. There are several possible reasons for the lack of benefit noted, including tumor heterogeneity and the wide-ranging signaling that occurs downstream of genetic EGFR alterations that can lead to resistance and phenotypic diversity. There is also no specific biomarker that defines possible responses to EGFR-mediated interventions. IDH1 and 2 mutations have also been widely studied and are an integral component of glioma classification, given the association with prognosis [48]. The 2021 classification of glioma includes IDH alteration to separate low-grade and high-grade gliomas that are IDH mutated from GBM, which is IDH-wildtype [49]. It should be noted that existing data does not yet reflect this classification, and IDH status in most glioma cohorts does not yet include IDH status, although this is likely to change as more data emerges [50]. The mechanism by which IDH mutation, which is common in lower-grade gliomas and uncommon in GBM, results in superior outcomes is unclear. IDH mutation impacts tumor growth by inhibiting glutaminase and altering metabolism [51]; however, IDH mutation also fosters the hypermethylation phenotype with potentially conflicting results for clinical therapeutic interventions [52]. Targeting the IDH mutation is evolving with recent data in this space [53]. In a double-blind phase 3 trial, Vorasidenib, an inhibitor of mutant IDH1 and IDH2 enzymes, improved progression-free survival and delayed the time to intervention in patients with grade II IDH mutated glioma. There are currently 475 interventional recruiting trials in glioma. Thus, it is anticipated that this data growth, matched by biospecimen acquisition and accompanied by biomarkers, will increasingly define the mechanisms of resistance and response in glioma to link biology and treatment strategies. Potentially actionable biomarkers, however, are evaluated based on clinically associated prognosis, which can be based on outdated endpoints. Progression is based on Response Assessment in Neuro-Oncology (RANO) criteria that include no omic signatures or relationships [54]. Imaging features that have evolved to integrate computational analysis are equally not currently included [55, 56, 57, 58]. The rapid evolution of computational approaches and their application should be vetted for interpretability through a clinical lens. Intersectional clinical and computational expertise is needed to link omics, molecular interactions, and clinical data to address the missing link between tumor biology and treatment strategies. Expertise that functions at the intersection of computational approaches and medicine is on the rise and should be included when new trials are designed and implemented to harness maximal data output for future analyses.

6. Conclusions

Molecular interaction databases and bioinformatics applications have been growing in number and usage to identify novel signals in glioma and annotate emerging data. However, challenges remain, including basing conclusions on any type of biospecimen, time point, or intervention and lack of transparency as to how linkages in bioinformatics platforms were generated. Emerging signals are primarily distinct from one another given different data acquisition means, various data linkages, and analysis timing as databases evolve. This causes difficulty in biological understanding since few signals are shared between studies and novel markers are often not validated in independent cohorts. Results are growing and perpetuate inclusion into databases but can generate circular connections, which need to be interpreted cautiously since several alternative versions of the same linkage are conceivable given additional published data and validation. As associations grow, the annotation will improve but can be subject to bias as new clinical data and unidentified or unpublished markers will remain unlinked. Biologic signal ambiguity in the attribution of differentially expressed signals to upregulated and downregulated signaling pathways is the subject of ongoing studies, and thus directionality of signal change is evolving. Endpoints employed to study large-scale omic data need to be updated to reflect potentially actionable biomarkers, and intersectional expertise needs to be implemented in clinical trials to allow for future data integration and analysis.

Conflict of interest

The authors declare no conflict of interest.

Abbreviations

CGGA	Chinese Glioma Genome Atlas
COG	Clusters of Orthologous Genes
CPTAC	Clinical Proteomic Tumor Analysis Consortium
GBM	Glioblastoma
GEO	Gene Expression Omnibus
GO	Gene Ontology
IPA	Ingenuity Pathway Analysis
KEGG	Kyoto Encyclopedia of Genes and Genomes
KOBAS	KEGG Orthology-Based Annotation System
LGG	Low-Grade Glioma
MGMT	O6-Methylguanine-DNA Methyltransferase
MSigDB	The Molecular Signatures Database
RANO	Response Assessment in Neuro-Oncology
REMARK	Reporting Recommendations for Tumor Marker Prognostic Studies
ssGSEA	Single-sample Gene Set Enrichment Analysis
STRING	Search Tool for the Retrieval of Interacting Genes/Proteins
TCGA	The Cancer Genome Atlas
WGCNA	Weighted Gene Co-expression Network Analysis