A systematic computational analysis of the endosomal recycling pathway in glioblastoma

Glioblastoma (GBM) is the most common and aggressive brain cancer in adults. The standard treatment is brutal and has changed little in 20 years, and more than 85% of patients will die within two years of their diagnosis. There is thus an urgent need to identify new drug targets and develop novel therapeutic strategies to increase survival and improve quality of life. Using publicly available genomics, transcriptomics and proteomics datasets, we compared the expression of endosomal recycling pathway regulators in non-tumour brain tissue with their expression in GBM. We found that key regulators of this pathway are dysregulated in GBM and their expression levels can be linked to survival outcomes. Further analysis of the differentially expressed endosomal recycling regulators allowed us to generate an 8-gene prognostic signature that can distinguish low-risk from high-risk GBM and potentially identify tumours that may benefit from treatment with endosomal recycling inhibitors. This study presents the first systematic analysis of the endosomal recycling pathway in glioblastoma and suggests it could be a promising target for the development of novel therapies and therapeutic strategies to improve outcomes for patients.


Introduction
Glioblastoma (GBM) is a devastating adult brain cancer with high rates of recurrence and resistance to treatment.More than a quarter of a million people worldwide develop brain cancer each year, and the incidence of GBM is increasing due to aging populations [1].It is a low survival cancer; without treatment average survival is 3 months and this increases to 10-13 months with treatment.The standard of care, maximal surgical resection followed by radiation and chemotherapy with temozolomide, is brutal and has not evolved since it was established almost 20 years ago [2].There is a clear unmet need to develop new and more effective therapies to treat people with GBM, in order to increase survival rates and quality of life.
One of the characteristics of GBM that makes it so difficult to treat is the very high degree of inter-and intra-tumoral heterogeneity, with the broad range of genetic mutations between patients affecting prognosis and response to therapy [3].GBM was the first cancer type that was comprehensively analysed by genomics methods, and careful analysis revealed that the epidermal growth factor receptor (EGFR) gene is amplified or mutated in 40-60% of GBM cases [4,5].EGFR is a member of the ERBB family of receptor tyrosine kinases (RTKs), which also includes HER2, HER3 and HER4.It is a cell surface receptor that regulates signalling pathways which control cell division and survival, and its aberrant activation has been associated with a number of cancers, including lung adenocarcinoma, breast and head and neck cancers.There are several monoclonal antibody and small molecule therapies that specifically target EGFR and are in use in the clinic.A few EGFR-targeting tyrosine kinase inhibitors (TKIs) have been tested in clinical trials to treat GBM, but the results have been disappointing with no improvement in clinical outcome observed [6].A third generation EGFR TKI called osimertinib holds greater promise as it can efficiently cross the blood-brain barrier and is currently undergoing clinical trials [7,8].
RTKs are continuously internalised from the plasma membrane into organelles called early endosomes by a process called endocytosis, which affects the specificity and duration of their signalling.From the early endosome they are either sent to lysosomes for degradation or are returned to the plasma membrane to be re-used.This return pathway is called the endosomal recycling pathway and is the main cellular mechanism for controlling the composition of the plasma membrane.A typical cell turns over the entire contents of its plasma membrane every 2 hours [9].We and others have reported that genes encoding regulators of endosomal recycling are frequently mutated, amplified, overexpressed or deleted in many cancers, including glioblastoma [10,11].This can lead to hyperactivation of endosomal recycling and cause an imbalance in the level of RTKs and other clinically relevant proteins at the cell surface, resulting in a consequent upregulation of cell proliferation and motility signals.
Furthermore, we have recently shown that inhibition of endosomal recycling in breast cancer cells with small molecule inhibitors leads to a reduction in total cellular HER2 and HER3 protein levels.We reported that blocking the recycling of these receptors results in their diversion to lysosomes where they are degraded.We also showed that endosomal recycling inhibitors synergise with HER2-targeting therapies, in both drug sensitive and drug resistant HER2-positive breast cancer [12].Based on this and other findings from our laboratory [13], we proposed that the endosomal recycling pathway represents a novel and underexploited target for developing anticancer drugs.
To investigate whether disrupting endosomal recycling might be a novel strategy for treating glioblastoma, we set out to determine if the endosomal recycling pathway is dysregulated in this cancer type.Using publicly available genomic, transcriptomic and proteomic datasets we demonstrate that the expression of key regulators of endosomal recycling are frequently altered in GBM, and are linked to poorer survival.These findings suggest that targeting the endosomal recycling pathway in GBM is worthy of further investigation.

Samples and datasets
The mRNA expression data and corresponding clinical information for 527 glioblastoma (GBM) tumours and 10 non-tumour brain tissues from the TCGA cohort, and 219 GBM and 28 non-tumour samples from the REMBRANDT cohort were acquired from The Cancer Genome Atlas (TCGA).The gene expression data of 54 tissues from 948 donors was downloaded from the Genotype-Tissue Expression (GTEx) project (dbGaP Accession phs000424.v8.p2).
All statistical analyses were carried out using R 4.2.2 software.Endosomal recycling genes were categorized into 3 groups based on their median gene expression level.Group 1: brain median gene expression difference with the non-brain tissues log [fold change (FC)] < 1. Group 2: brain median gene expression in brain was 1 log [fold change (FC)] or greater higher than the median expression in non-brain tissues.Group 3: median gene expression in brain was 1 log [fold change (FC)] or more lower than the median expression in non-brain tissues.

Identification of differentially expressed genes
Gene expression in non-tumour brain and GBM samples were analysed using the edgeR package in R, and differentially expressed genes were identified with the following criteria: [log2 fold-change (FC)] > 0.5 and P < 0.05 [14].Gene expression and survival data for individual genes were imported from http://gliovis.bioinfo.cnio.es/intoGraphPad Prism v.10, which was used to generate the Kaplan-Meier curves.
The mass spectrometry proteomics dataset PXD014606 [15] was downloaded from the PRIDE database repository.Quantification of upand downregulated endosomal recycling components was performed in Microsoft Excel, by calculating the log2 ratios of the individual endosomal recycling regulators in glioma biopsies compared to control non-tumour brain tissue.Box plots displaying the median, the first and third quartile, and whiskers indicating the minima and maxima were generated in GraphPad Prism v.10.

Prognosis analysis
Overall survival (OS) analysis was performed based on Kaplan-Meier curves to explore the prognostic value of potential GBM-associated endosomal recycling genes.Log-rank P-value <0.05 was considered significant.In addition, univariate Cox regression analysis of gene expression level was performed using the "survival" R package and used to screen for prognosis-related genes in GBM (p < 0.05 was considered significant).
An optimal risk scoring model based on the prognostic potential of the endosomal recycling genes was calculated using a Least Absolute Shrinkage and Selection Operator (LASSO) regression analysis.The "glmnet" R package was used and a risk score was applied to all glioblastoma samples in the dataset and the samples were separated into low-and high-risk groups using the median as a cut-off.A Cox regression model incorporating age and the log-rank test were used to assess overall survival (OS) of the two groups in the whole dataset.
Receiver operating characteristic (ROC) curves to evaluate the effectiveness and accuracy of the risk score in predicting the prognosis were generated by the "timeROC" R package.

Characterisation of the gene expression profile of endosomal recycling regulators
We have taken a computational approach to determine if the endosomal recycling pathway is altered in glioblastoma.To this end, we compiled a list of 71 genes that have been reported in the literature to regulate the endosomal recycling pathway (Table 1).The bulk expression for each gene was analysed using the Genotype-Tissue Expression (GTEx) Portal to determine their tissue-specific gene expression profiles [16].This allowed us to divide the genes into three groups based on their transcript levels; Group 1: ubiquitously expressed (less than 1 log-fold difference between the median expression in brain tissue and the median expression in all other tissues, n = 41); Group 2: preferentially expressed in the brain (the median expression in brain tissue is 1 log-fold or more greater than the median expression in all other tissues, n = 12); Group 3: not expressed in the brain (1 log-fold or more lower median expression in brain tissue compared to the median expression in all other tissues, n = 21).We chose EPS15, KIF5A and EHD2 as representatives of groups 1, 2 and 3, respectively (Fig. 1).

Endosomal recycling regulator expression is dysregulated in GBM
Next, to determine if endosomal recycling genes are differentially expressed between glioblastoma and non-tumour brain tissue, we analysed their expression in the TCGA_GBM microarray expression dataset, which contains 528 GBM samples and 10 non-tumour brain samples.We also analysed the same genes in an independent brain cancer gene expression dataset, the Repository for Molecular Brain Neoplasia (REMBRANDT) collection, which includes 219 GBM and 28 non-tumour samples and uses the same microarray gene expression chip as the TCGA dataset [17].The expression of 16 endosomal recycling regulator genes were upregulated and 14 were downregulated in the TCGA dataset (Fig. 2A).Similarly, 12 genes were significantly upregulated and downregulated in the REMBRANDT dataset (Fig. 2B).Comparison of the results from both datasets revealed that 7 of the upregulated genes and of the downregulated genes were shared between both datasets, thus ~ 19% of the endosomal recycling genes were consistently differentially expressed in GBM (Fig. 2C).These were selected for further analysis.We performed univariate cox regression analysis to reveal the relationship between their expression and prognosis (Fig. 2D).Six genes, CMTM6, KDELR1, RAB8A, EHD2, CAV1 and COMMD4, were identified as risk factors (hazard ratio >1).Five genes, KIF5A, TBC1D9, KIF5C, VAMP2 and RAB11FIP2, had a protective effect (hazard ratio <1).
Next, we looked at a representative example from each group in more detail (EPS15 from group 1, KIF5A from group 2, and EHD2 from group 3).The mRNA expression of EPS15 and KIF5A was significantly downregulated in both the TCGA and REMBRANDT glioblastoma datasets, and EHD2 was upregulated in both datasets (Fig. 3A and B).GBM is a form of glioma, which accounts for 78% of all malignant brain tumours.Gliomas can be divided into four grades.Low-grade glioma (grades 1 and 2) includes oligodendrogliomas and diffuse astrocytomas.Anaplastic astrocytoma and GBM are high-grade gliomas (grades 3 and 4).By definition, GBM is always a grade 4 glioma.There is a statistically significant downregulation in the expression of EHD1 between astrocytoma and GBM, but no difference between GBM and the two lowgrade gliomas (Fig. 3C).In contrast, KIF5A and EHD2 displayed differential expression between GBM and all other gliomas (Fig. 3C).Receiver operating characteristic (ROC) curve analysis of each of these differentially expressed genes revealed that all had high accuracy in differentiating non-tumour brain tissue from GBM, and KIF5A and EHD2 could also accurately distinguish oligodendroglioma and astrocytoma from GBM (Fig. 3D-Table 2).
There was a strong negative correlation between KIF5A mRNA expression and tumour grade, and a strong positive correlation between EHD2 expression and grade (Fig. 4A).EPS15 did not display the same level of correlation.To determine if there is a link between gene expression and survival, the TCGA_GBMLGG dataset was divided into high-and low-gene expression cohorts (median cutoff) for each gene, and Kaplan-Meier survival curves plotted.There was no difference in survival between the high-and low-EPS15 expression cohorts.However, the high KIF5A expression cohort had considerably better prognosis, while the high EHD2 expression cohort had much poorer overall survival (Fig. 4B).Almost identical patterns were observed in the REMBRANDT collection (Fig. S1).
Recently, Buser et al. performed an unbiased quantitative proteomics analysis of human glioma [15].They used mass spectrometry to compare protein expression levels from 4 human glioma biopsies with 4 control human brain biopsies.We analysed the protein expression of 12 of the 14 differentially expressed endosomal recycling regulators in this glioma proteomics inventory (CMTM6 and KDELR1 were missing from this dataset).The protein expression levels of these endosomal recycling regulators closely matched their transcriptomic expression (Fig. 4C).

Construction of a prognostic model with endosomal recycling genes
Finally, we used a LASSO Cox regression model to overcome overfitting and to identify the most robust genes for the construction of a gene signature that could be beneficial for GBM prognosis (Fig. 5A and  B).We used the data from the TCGA_GBM database as a training set, and the REMBRANDT GBM data as a validation set.We consequently identified an 8-gene signature that was significantly correlated with OS in GBM (Fig. 5C).Of these, CMTM6, AP2A2, and DNM1 showed a positive correlation with the risk score while KIF5B, ATP9A, COMMD3, RAB11B and VPS26A had negative correlations (Table 3).Based on these, we determined the corresponding risk scores of GBM patients, and found that patients with high-risk scores had significantly shorter survival times than those with low-risk scores, in both datasets (Fig. 5C and D).The predictive power of the gene-set was confirmed by ROC curve analysis (Fig. 5E and F).

Discussion
Glioblastoma is among the most difficult cancers to treat, with one of the lowest 5-year survival rates.It has high rates of recurrence and resistance to treatment, largely due to its cellular heterogeneity and extensive tumour invasion into surrounding brain tissue, making complete surgical resection impossible [3].Furthermore, the recurrent tumour often has a more malignant phenotype [6].To combat these characteristics of GBM it is important to gain a thorough understanding of the mechanisms that lead to resistance, with a view to identifying common features that may be targeted for therapeutic intervention.To that end, we set out to investigate whether the endosomal recycling pathway, an intracellular membrane trafficking pathway that is dysregulated in other cancers, is also disrupted in glioblastoma [18].
The repertoire of proteins that a cell maintains at the plasma membrane determines how it interacts with neighbouring cells, which nutrients it can internalise, and how it responds to extracellular signals.These cell surface proteins are continuously internalised into early endosomes by a process called endocytosis.From here they are either sorted into late endosomes and lysosomes, where they are degraded, or returned to the plasma membrane.Disruption of this process has been associated with numerous neurological and metabolic diseases, as well as many cancers.Bacteria and viruses also often hijack parts of the membrane trafficking machinery during their life cycle [9].The endosomal recycling pathway mediates the transport of internalised cell surface proteins back to the plasma membrane, either directly from early endosomes, or indirectly via an endosomal recycling compartment [19].Dozens of proteins are known to regulate this trafficking pathway, and the expression of many of these is altered in breast, ovarian, head and neck squamous cell carcinoma, and lung cancers [20][21][22][23][24].
Recently, Buser et al. used an unbiased quantitative proteomic analysis to find that multiple components of the endocytosis pathway were massively downregulated in human gliomas compared to nontumour brain tissue.They showed that this resulted in increased cell surface receptor levels and persistent receptor tyrosine kinase signalling from the cell surface.They proposed that defective endocytosis creates a selective advantage for glioma tumour progression [15].The involvement of a defective endocytic pathway in the development of glioma has also been reported by Wang et al. who used a bioinformatics approach to generate a signature of endocytosis genes that could be used to stratify low-grade gliomas into high-and low-risk groups [25].
Since dysregulated endosomal recycling can increase the aggressiveness of other cancers by upregulating receptor signalling and promoting migration and invasion, we used cancer genomics to determine if this pathway might also be aberrantly activated in GBM.
40-60% of glioblastomas carry mutations or increased copy numbers of the EGFR gene [4].Recycling is associated with prolonged signalling, while degradation attenuates signalling [26].Mechanistically it can be envisaged that upregulation of the endosomal recycling pathway would further increase the density of EGFR (and other RTKs) on the cell surface, by shifting the balance between recycling and degradation towards recycling.This would result in an amplification of downstream signal transduction, such as the ERK and Akt pathways, and promote tumour development and invasion into the surrounding tissue.Indeed, a recent study using cultured GBM cells has reported that upregulated endosomal recycling does indeed lead to increased EGFR at the cell surface and sustained proliferative signalling [27].
However, the link between dysregulation of endosomal recycling and cancer progression is complicated by the fact that, depending on the cancer type (or even subtype), the same gene can play a tumour promoting or a tumour suppressive role.For example, Rab25 has been reported to act as an oncogene in breast, ovarian, lung, GBM and gastric cancer and as a tumour suppressor in colon, oesophageal, head and neck and triple-negative breast cancer [28].Similarly, both overexpression and downregulation of RCP (Rab11FIP1) have been reported to promote breast tumour progression [20,29,30].Gene expression data of normal brain tissue and GBM biopsies, acquired from publicly available cancer genomics databases, has been used to study the underlying mechanisms of GBM progression, to construct prognostic signatures and to identify novel drug targets [25,[31][32][33].We used a similar approach to gain a better understanding of the pathophysiological role that endosomal recycling may play in GBM.We first examined the expression pattern of 71 genes that encode regulators of this trafficking pathway.We identified 14 genes that were differentially expressed in two independent glioblastoma transcriptomics datasets.High expression of 6 of these were associated with decreased OS, while high expression of 5 had a protective effect.Further, we showed that the expression levels of a selection of these genes could accurately distinguish low-grade gliomas from high-grade gliomas, and their differential expression correlated with tumour grade.Using the Buser et al. proteomics dataset [15], we confirmed that the differential gene expression we observed from the transcriptomics datasets correlated closely with the protein expression levels.
To further analyze the informativeness of endosomal recycling regulators for prognosis, a risk-scoring model of GBM was used to construct an 8-gene endosomal recycling prognostic signature based on their expression.Kaplan-Meier survival curves and ROC analysis in training and validation datasets confirmed the effectiveness of this model at stratifying GBM into higher and lower risk groups.
It is well established that dysregulated endosomal recycling promotes tumour progression in other cancers.To our knowledge, this is the first time that the endosomal recycling pathway has been analysed systematically in GBM.Our findings show that key components of the endocytic recycling machinery are aberrantly expressed in glioblastoma and suggest that they contribute to the aggressiveness of this cancer.Thus, inhibition of this pathway with small molecule inhibitors may have therapeutic benefit.We have previously shown that chemical and genetic inhibition of endosomal recycling reduces the aggressiveness of lung cancer cells [34].There are a number of recycling inhibitors available, and we have recently reported that small molecule endosomal recycling inhibitors (ERIs) synergise with HER2-targeting therapies in drug-sensitive and drug-resistant breast cancer cells [12].Unpublished work from our laboratory using GBM cell lines has found that these ERIs also synergise with the EGFR-targeting TKIs gefitinib and osimertinib.
The main limitation of this study is that the data comes entirely from publicly available databases and the involvement of the dysregulated  endosomal recycling regulators in GBM aggressiveness has not been verified experimentally.Further work will involve modulating the expression of the individual endosomal recycling regulators identified in this study in GBM cell lines and in vivo models, to determine their specific roles in RTK trafficking, cell proliferation, and invasion.

Fig. 1 .
Fig. 1.RNA expression of endosomal recycling genes in normal tissues.Violin plots of GTEx bulk tissue gene expression for representative endosomal recycling genes from groups 1, 2 and 3. Horizontal lines indicate median gene expression.

Fig. 2 .
Fig. 2. Differentially expressed endosomal recycling genes.Volcano plots of differentially expressed endosomal recycling genes in the TCGA (A) and REM-BRANDT (B) GBM cohorts.Red dots are significantly upregulated genes, blue dots are significantly downregulated (p value < 0.05 and log 2 FC > 0.5).C List of the endosomal recycling genes that are differentially regulated in both the TCGA and REMBRANDT datasets.LogFC values indicated.D Forest plots of the differentially expressed endosomal recycling genes indicating their relationship with overall survival.Indicated are hazard ratio and 95% confidence interval.

Fig. 3 .
Fig. 3. Endosomal recycling gene expression in glioma.Comparison of the expression of the representative differentially expressed endosomal recycling genes in the TCGA (A) and REMBRANDT (B) cohorts.C Expression of endosomal recycling genes in the TCGA_GBMLGG dataset.D Receiver operating characteristic (ROC) curves indicating the accuracy of gene expression at distinguishing glioma subtype.

Table 1
The list of endosomal recycling regulator genes analysed in this study.

Table 1 (
continued ) (continued on next page) L.J. Joyce and A.J. Lindsay