Increased LGALS3 expression independently predicts shorter overall survival in patients with the proneural subtype of glioblastoma

Abstract In the current study, we tried to study the expression of LGALS3 and LGALS3BP, their potential as prognostic markers and the possible genetic/epigenetic mechanisms underlying their dysregulation in different subtypes of glioblastoma (GBM). An in silico retrospective study was performed using large online databases. Results showed that LGALS3 and LGALS3BP were upregulated at both RNA and protein levels in GBM tissue and were generally associated with shorter overall survival (OS) in GBM patients. However, in subgroup analysis, we only found the association in proneural subtype. The copy number alterations did not necessarily lead to LGALS3/LGALS3BP dysregulation. In the proneural subtype of GBM patients, hypermethylation of the two CpG sites (cg19099850 and cg17403875) was associated with significantly lower expression of LGALS3. In univariate and multivariate analysis, LGALS3 expression independently predicted shorter OS in the proneural subtype of GBM (HR: 1.487, 95% CI: 1.229‐1.798, P < 0.001), after adjustment of age, gender, IDH1 mutations, temozolomide chemotherapy, radiotherapy and LGALS3BP expression. In comparison, LGALS3BP lost the prognostic value in multivariate analysis. Based on these findings, we infer that LGALS3 expression serves as an independent biomarker of shorter OS in the proneural subtype of GBM, the expression of which might be regulated in an epigenetic manner.

with IDH1 mutations or those belonging to the G-CIMP phenotype have significantly improved OS. 5 Besides, MGMT promoter methylation is also a powerful and independent indicator of therapeutic responses among GBM patients who received chemotherapy with alkylating agents. 6 However, these markers have their own limitations in clinical use. IDH1 mutations are rare in primary GBM cases. 7 In comparison, the prognosis value of MGMT promoter might depend on the chemotherapeutic substances used 6 and might lose the value under some conditions. 8,9 Therefore, it is quite meaningful to further explore other potential prognostic markers in GBM. However, GBM is a not a homogenous disease, but has heterogeneous histological features. 10,11 The Cancer Genome Atlas (TCGA) researchers divided GBM into four subtypes (Proneural, Neural, Classical and Mesenchymal) according to their gene expression pattern. 12 Besides the genomic differences, these subtypes varied significantly in survival length, patient age and treatment response. Therefore, exploring specific prognostic marker of the GBM subtypes might support better therapeutic management.
The galectins are a group of beta-galactoside-binding proteins that involve in regulating cell-cell and cell-matrix interactions. 13 Till now, galectin-1, -2, -3, -4, -7, -8, -9, -10 and -12 have been identified in humans. Galectin-3 encoded by LGALS3 gene plays an important role in cell adhesion, growth, differentiation, apoptosis, as well as angiogenesis in both normal and cancerous tissues. 14 Its upregulation might serve as a valuable prognostic marker in multiple cancers, such as breast cancer, 15 gastric cancer, 16,17 colorectal cancer 18 and liver cancer. 19 Upregulated LGALS3 was also observed in GBM tissues. 20,21 Galectin-3-binding protein (Gal-3BP) is a secreted glycoprotein encoded by LGALS3BP gene. In the microenvironment of human neuroblastoma, Gal-3BP interacts with Galectin-3 (Gal-3) in bone marrow mesenchymal stem cells (BMMSC) and induces transcriptional upregulation of IL-6, via the Gal-3BP/Gal-3/Ras/MEK/ERK signaling pathway. 22,23 These findings suggest that these two genes may collaboratively participate in the pathological process of cancer.
In the current study, we tried to study the expression of LGALS3 and LGALS3BP, their potential as prognostic markers and the possible genetic/epigenetic mechanism underlying their dysregulation in different subtypes of glioblastoma (GBM), by using large online databases.

TCGA-GBM
The level-3 data in TCGA-GBM were obtained using the UCSC Xena Browser (https://xenabrowser.net/heatmap/). 24 The recurrent tumors and cases that had a history of neoadjuvant treatment were excluded. Affymetrix Human Genome U133 Array Strip (AffyU133a) data was used to quantify gene expression. Based on these criteria, a total of 508 primary tumor cases and 10 adjacent normal tissue cases were included in this study. The genomic, clinicopathological and survival data of the included patients were downloaded. Briefly, the data included IDH1 mutations, gene expression subtypes, age at initial pathologic diagnosis, longest tumor dimension, gender, karnofsky performance score (KPS), overall survival (OS) time, temozolomide chemotherapy status, radiation therapy status, LGALS3/ LGALS3BP expression, LGALS3/LGALS3BP DNA copy number alterations (CNAs, calculated by an algorithm called Genomic Identification of Significant Targets in Cancer 2.0 [GISTIC2]) 25 and their DNA methylation (quantified by Infinium HumanMethylation27 BeadChip). In the GISTIC2, CNAs were defined as −2: homozygous deletion; −1: heterozygous deletion, 0: copy-neutral, +1: low-level copy gain, +2: high-level amplification.
MGMT-STP2 model, 26 which includes two CpG sites (cg12434587 and cg12981137) was used determine the MGMT promoter methylation status.

| Data mining in the R2 web-based application
The association between LGALS3/LGALS3BP expression and OS in GBM patients was also analyzed using genomic and survival data in GSE16011 from GEO datasets, 27 using the R2 web-based application (https://hgserver1.amc.nl/cgibin/r2/main.cgi). Only the GBM cases in GSE16011 were included in Kaplan-Meier survival analysis. The best cutoff was identified using scan model.

| Data mining in the HPA
LGALS3 and LGALS3BP protein expression in normal brain tissues (typically cerebral cortex and hippocampus) and GBM tissues was examined using data from the Human Protein Atlas (HPA) (available fromwww.proteinatlas. org). 28,29 Immunochemical images and protein scoring data of LGALS3 and LGALS3BP were retrieved.

| Statistical analysis
Statistical analyses were conducted using SPSS 25.0 software (SPSS, Chicago, IL) and GraphPad Prism 7.04 (GraphPad Inc, La Jolla, CA). One-way ANOVA with Bonferroni post hoc tests and Welch's t test (unequal variances t test) were performed to assess the statistical differences. Kaplan-Meier OS curves were generated using GraphPad Prism 7.04. Receiver operating characteristic (ROC) analysis for death detection was performed to identify the best cutoff (Youden index) for gene expression in Kaplan-Meier curves. The difference between the curves was compared using the Logrank test. The independent prognostic value of LGALS3/ LGALS3BP expression in proneural subtype was assessed using the univariate and multivariate Cox regression models. P < 0.05 was considered as statistically significant.

| Both LGALS3 and LGALS3BP RNA expression were upregulated in GBM tissues compared with adjacent normal tissues
In TCGA-GBM, 508 cases of primary GBM and 10 adjacent normal tissues were subjected to AffyU133a microarray analysis of gene expression. Using the array data, we compared the expression of LGALS3 and LGALS3BP in GBM and the adjacent normal tissues, as well as the different subtypes of GBM ( Figure 1A). Results showed that both LGALS3 and LGALS3BP were significantly upregulated in GBM tissues compared with adjacent normal tissues ( Figure 1A-C). Since the distinct molecular subtypes of GBM show varying prognosis and responses to aggressive chemotherapy and radiotherapy, we also examined the expression profiles of these two genes in the subtypes. Group comparison showed that LGALS3 expression varied significantly among the four subtypes, in which the mesenchymal and proneural subtypes had the highest and lowest expression, respectively ( Figure 1D). In comparison, LGALS3 expression in the neural and proneural subtypes was significantly lower than that in the mesenchymal and classical subtypes ( Figure 1E).

| LGALS3 and LGALS3BP protein expression was not detectable in glial cells in normal brain tissues, but was detectable in GBM tissues
Using IHC staining images and protein expression scoring in the HPA, we examined LGALS3 and LGALS3BP protein expression in normal brain and GBM tissues. According to the data in the HPA, LGALS3 and LGALS3BP protein expression was not detectable in glial cells in normal brain tissues (Figure 2, left). In comparison, among 9 cases of GBM LGALS3 and LGALS3BP were expressed at the protein level in GBM tissues.

| LGALS3 and
LGALS3BP upregulation was associated with unfavorable OS in GBM patients By comparing LGALS3 and LGALS3BP expression between the deceased and living GBM cases, we found that the deceased group had significantly elevated LGALS3 and LGALS3BP expression (P = 0.005 and P = 0.006, respectively) ( Figure 3A,B). Then, we tried to explore the association between the expression of these two genes and the survival outcomes by generating Kaplan-Meier curves, using the survival data in TCGA-GBM. Via setting the best cutoff identified in the ROC analysis for death detection, we found that the high LGALS3 expression group and the high LGALS3BP expression group had significantly shorter OS compared to the respective low expression group (P = 0.007 and P = 0.013, respectively) ( Figure 3C,D). To verify these trends, we also used genomic and survival data in another study (Tumor Glioma French database, GSE16011 from GEO datasets 27 ), which included 156 cases of GBM out of 284 glioma cases. Results confirmed the association between F I G U R E 2 LGALS3 and LGALS3BP protein expression in normal brain tissues and in glioblastoma (GBM) tissues. IHC staining of LGALS3 (up) and LGALS3BP (down) protein expression in normal brain tissues (left, including cerebral cortex and hippocampus) and in GBM tissues. Images credit: The Human Protein Atlas. Images and protein scoring were obtained from: v18.proteinatlas.org, via: https:// www.proteinatlas.org/ENSG00000131981-LGALS3/tissue/cerebral+cortex; https://www.proteinatlas.org/ENSG00000131981-LGALS3/ tissue/hippocampus; https://www.proteinatlas.org/ENSG00000131981-LGALS3/pathology/tissue/glioma#ihc; https://www.proteinatlas.org/ ENSG00000108679-LGALS3BP/tissue/cerebral+cortex; https://www.proteinatlas.org/ENSG00000108679-LGALS3BP/tissue/hippocampus and https://www.proteinatlas.org/ENSG00000108679-LGALS3BP/pathology/tissue/glioma#ihc | 2035 HE Et al.
the high gene expression and unfavorable OS in GBM patients ( Figure 3E,F).

| Kaplan-Meier OS analysis in the four molecular subtypes of GBM
Since we found that both LGALS3 and LGALS3BP expression varied significantly in the four subtypes of GBM, we then tried to explore whether the association between the high gene expression and unfavorable survival was consistent in the four subtypes. Subgroup analysis suggested that LGALS3 and LGALS3BP expression had no prognostic value in terms of OS in mesenchymal, classical and neural subtypes ( Figure  4A,C,E,G). In comparison, the associations were confirmed in proneural subtype ( Figure 4D,H).

| LGALS3 expression was not correlated with its DNA CNAs, but was associated with IDH1 mutations in proneural GBM
Using genomic data in TCGA-GBM, we further investigated the potential genomic and epigenetic alterations associated with dysregulation of these two genes. The proneural subtype was prone to haveLGALS3 heterozygous deletion (50/125, 40%) ( Figure 5A). In comparison, LGALS3BP amplification (28/125, 22.4%) was more common than heterozygous deletion (6/125, 4.8%) ( Figure 5B). However, these genetic alterations did not necessarily lead to LGALS3/LGALS3BP dysregulation ( Figure 5C-D). Previous studies found that IDH1 mutation is an important prognostic marker in GBM patients and is associated with hypermethylation status of

F I G U R E 3 LGALS3/LGALS3BP expression and overall survival (OS) of glioblastoma (GBM). A and B, Plot charts comparing LGALS3 (A)/
LGALS3BP (B) expression between the living and deceased GBM cases in TCGA-GBM. C and D, Kaplan-Meier curves of OS in GBM patients in TCGA-GBM. Patients were separated into two groups according to the best cutoff of LGALS3 (C)/LGALS3BP (D) expression identified in the ROC analysis for death detection. E and F, Kaplan-Meier curves of OS in GBM patients in GSE16011 from GEO datasets. 27 Patients were separated into two groups according to the best cutoff of LGALS3 (E)/LGALS3BP (F) expression a series of genes. 12,27 By checking the correlation between these two genes and IDH1 mutations, we found that the IDH1 mutation group had significantly lower LGALS3 expression (P < 0.001, Figure 5E). But this association was not observed in LGALS3BP expression (P = 0.70, Figure 5F).

DNA methylation status in proneural GBM
Since IDH1 mutations are generally associated with hypermethylation status, we then checked whether LGALS3 expression was related to its DNA methylation level. By examining the correlation between LGALS3 expression and its DNA methylation (methylation 27k data), we found that the group of patients with hypermethylation of the two CpG sites (cg19099850 and cg17403875) had significantly lower expression of LGALS3 ( Figure 6A, green box). In comparison, this association was not observed in terms of LGALS3BP expression ( Figure 6B). By separating the patients according to median LGALS3/LGALS3BP methylation, we found that the LGALS3 hypermethylation group had significantly better OS ( Figure 6C). No significant association was observed between LGALS3BP methylation and OS ( Figure 6D).

| LGALS3 expression was an independent prognostic indicator of OS in proneural subtype, after the adjustment of IDH1 mutations
To explore the independent prognostic value of LGALS3 expression in terms of OS in proneural subtype, we performed univariate and multivariate analysis based on COX regression model. The clinicopathological, genomic and survival data used were given in supplementary Table S1. Univariate analysis showed that older age, male patients, no IDH1 mutations, no temozolomide chemotherapy, no radiotherapy, increased LGALS3 expression and increased LGALS3BP expression were risk factors of shorter OS (Table 1). In multivariate analysis, increased LGALS3 expression was independently associated with shorter OS (HR: 1.487, 95% CI: 1.229-1.798, P < 0.001) ( Table 1), after adjustment of other risk factors. However, LGALS3BP expression had no independent prognostic value (Table 1).

| DISCUSSION
A series of studies confirmed the oncogenic properties of LGALS3 expression in glioma. Galectin-3 expression is correlated with the malignant potential of tumors in the central nervous system. 30 Its expression could be induced under hypoxic and nutrient deprived microenvironments, which protects cells from cell death. 20 In addition, it also enhances GBM cell motility. 31 However, previous studies found that LGALS3 expression varied significantly in different types of glioma, such as supratentorial pilocytic astrocytoma, astrocytoma, anaplastic astrocytoma and glioblastoma. 21,32 These findings suggest that Galectin-3 might have varying regulatory effects on different types of glioma. In terms of LGALS3BP, it is characterized as an onco-protein that regulates the malignant behaviors of multiple cancers. [33][34][35] In this study, we confirmed the upregulation of LGALS3 and LGALS3BP at both RNA and protein level in GBM tissues. Besides, we found that LGALS3 expression was generally associated with shorter OS in GBM patients. Since we observed that these two genes had different expression in the 4 subtypes of GBM, we further performed subtype group analysis to check the robustness of the finding. In subgroup analysis, we only found the association between LGALS3/LGALS3BP expression and unfavorable OS in proneural subtype, but not in other subtypes.
Among the subtypes, the proneural subtype has the most frequent mutations in theIDH1 gene. In addition, the proneural subtype had the longest median OS, around 17 months. 36 In this study, we found that among 60 cases of proneural subtype patients with IDH1 mutations examined, only 13 cases had mutations. Therefore, although IDH1 mutations might be a powerful prognostic marker, it is still necessary to explore other robust prognostic biomarkers for the large proportion of proneural patients without IDH1 mutations. In univariate analysis, we found that the promoter methylation status of MGMT might not be a prognostic marker in proneural subtype, when not considering the therapeutic strategy. This finding is consistent with previous studies that suggest the prognostic value of MGMT promoter methylation dependents on the chemotherapeutic substances used. 6 By conducting univariate and multivariate analysis, we confirmed that LGALS3 expression was an independent prognostic indicator of shorter OS in the proneural subtype of GBM (HR: 1.487, 95% CI: 1.229-1.798, P < 0.001), after the adjustment of age, gender, IDH1 mutations, temozolomide chemotherapy, radiotherapy and LGALS3BP expression. In comparison, although LGALS3BP expression is a risk factor in univariate analysis, it lost the prognostic value in multivariate analysis. These findings suggest that LGALS3 expression might be a valuable prognostic biomarker in the proneural subtype of GBM.
LGALS3 methylation expression is associated with the loss of the galectin-3 expression in the mucinous colorectal carcinomas, 37 prostate cancer 38 and breast cancer, 39 suggesting that its expression is controlled in an epigenetic manner. Since IDH1 mutations lead to hypermethylation status and G-CIMP phenotypes 4 and is associated with suppressed LGALS3 expression, we examined the correlation between the methylation status of LGALS3 and their expression. Results showed LGALS3 DNA hypermethylation was associated with decreased LGALS3 expression. This phenomenon suggests that DNA methylation is a potential epigenetic alteration leading to LGALS3 dysregulation in the proneural subtype of GBM.
This study also had some limitations. Firstly, some clinical information, such as surgical resection and the extent of resection were not recorded in TCGA-GBM. This might impair the credibility of the findings. Secondly, although the Tumor Glioma French database was used to validate the prognostic value of LGALS3, this database had no information of molecular subtypes. Therefore, more detailed validation analysis based on a large cohort of the proneural subtype is required in the future.

| CONCLUSION
LGALS3 expression serves as an independent biomarker of shorter OS in the proneural subtype of GBM, the expression of which might be regulated in an epigenetic manner.