Establishment of a nomogram with EMP3 for predicting clinical outcomes in patients with glioma: A bi‐center study

Abstract Aim To demonstrate the clinical value of epithelial membrane protein 3 (EMP3) with bioinformatic analysis and clinical data, and then to establish a practical nomogram predictive model with bicenter validation. Methods The data from CGGA and TCGA database were used to analyze the expression of EMP3 and its correlation with clinical prognosis. Then, we analyzed EMP3 expression in samples from 179 glioma patients from 2013 to 2017. Univariate and multivariate cox regression were used to predict the prognosis with multiple factors. Finally, a nomogram to predict poor outcomes was formulated. The accuracy and discrimination of nomograms were determined with ROC curve and calibration curve in training and validation cohorts. Results EMP3 was significantly higher in higher‐grade glioma and predicted poor prognosis. In multivariate analysis, high expression of EMP3 (HR = 2.842, 95% CI 1.984–4.071), WHO grade (HR = 1.991, 95% CI 1.235–3.212), and IDH1 mutant (HR = 0.503, 95% CI 0.344–0.737) were included. The nomogram was constructed based on the above features, which represented great predictive value in clinical outcomes. Conclusion This study demonstrated EMP3 as a novel predictor for clinical progression and clinical outcomes in glioma. Moreover, the nomogram with EMP3 expression represented a practical approach to provide individualized risk assessment for glioma patients.


| INTRODUC TI ON
Glioma is the most common malignant tumor of the central nervous system, with a 5-year overall survival (OS) rate of approximately 36%, and accounts for more than 70% of intracranial malignant tumors. [1][2][3] Regardless of tumor malignancy and aggressive treatment, the average median OS time is still only 12-18 months. 4,5 Although various therapeutic modalities are available, such as surgical resection, chemotherapy, radiotherapy, and immunotherapy, patients with glioma have low survival rates. With the development of high-throughput microarray technology, gene expression profiling has been widely used to determine signatures or biomarkers associated with tumor progression and clinical prognosis. [6][7][8] Although many studies of the prognostic biomarkers have been reported, few have been validated in the clinical setting. Genetic signatures identified from four different published microarray datasets have been validated in glioma cohorts. [9][10][11] However, the clinical value of these genetic signatures is undetermined and has not been applied in clinical practice.
The tumor immune microenvironment participates in oncogenesis and tumor progression, and affects clinical prognosis. 12 Several studies have demonstrated the correlation between the tumor immune microenvironment and N6-methyladenosine (m6A) modification; however, the potential role of m6A modification in immune infiltration is still unclear, especially in glioma. m6A is the most common mRNA modification in diverse cells and has various regulatory functions in tumorigenesis, progression, and immune modulation. 13 In our current study, we built a model with integrated m6A and immune infiltration data to improve the overall prediction of outcome for glioma patients without using a large number of samples to verify its clinical practicality. 14  It has been reported that EMP3 promotes tumor growth and metastasis via the PI3K/AKT pathway, which is highly expressed in upper urinary tract urothelial cancer (UTUC) and hepatocellular carcinoma (HCC). 17,18 However, several studies considered EMP3 to be a tumor suppressor gene in several solid cancers, including low-grade glioma (LGG), esophageal carcinoma, and lung cancer. [19][20][21] One recent study demonstrated that EMP3 has oncogenic properties in high-grade glioma (HGG), and its overexpression might also predict poor clinical prognosis in primary glioblastoma multiforme (GBM). 22,23 Given the limited evidence that EMP3 may be a biomarker to predict the prognosis in gliomas, our study was performed to determine the relationship between EMP3 and immune infiltration and clinical outcomes and to establish a practical nomogram predictive model to guide the clinical therapy and assess prognosis.

| Immune cells and bioinformatic analysis
The single sample gene set enrichment analysis (ssGSEA) was used to define an enrichment fraction that represents the absolute enrichment of genomes in each sample with R package "GSVA" within a given dataset. Normalized enrichment fractions can be calculated for each type of immune cell. Genome set signature of 28 immune cells were obtained from a previous study. 24 Gene Set Variation Analysis from R package GSVA was performed to obtain the immune profile of the glioma samples.

| Immunohistochemical analysis
Human-derived glioma samples were fixed in 4% paraformaldehyde at room temperature for 24 h and then embedded in paraffin. The blocks were cut into 8 μm slices for the following analysis. Sections were blocked with 5% goat serum at room temperature for 1 h and then were stained with EMP3 (1:100, Santa Cruz, sc-81797, USA).
After washing with PBS, the sections were incubated with secondary antibody (1:5000, Beijing Zhongshan-Golden Bridge Technology Co., Ltd) at 37°C for 30 min. The ABC method (Vector Laboratories) was used at room temperature. The images were observed using an AX-80 microscope (Olympus, Tokyo, Japan) and were analyzed with ImageJ software.

| Real-time PCR
Total RNA was extracted from human glioma core region and adjacent tissues using TRIzol reagent (Invitrogen, Carlsbad, CA, USA

| Nomogram construction and validation
The patients from the Second Hospital affiliated to Zhejiang University were considered as training cohorts, and the patients from Shanghai General Hospital were included in validation cohorts.
In the survival analysis, the association between traits and overall survival was assessed using Cox regression model. The Kaplan-Meier survival curves were plotted and compared by log-rank tests with R packages "survival" and "survminer". In the training cohort, independent predictive risk factors were screened by multivariate Cox regression analysis. Based on independent factors, the nomogram model was established with the R software version 4.0.1 (https:// www.r-proje ct.org) to predict outcomes in 1-, 2-, and 3-year followups. The performance evaluation of the nomogram includes c-index, calibration curves, and ROC curve, which were verified by the calibration curve generated by the validation cohort. ROC curves, sensitivity, and specificity values were generated using R package "pROC".

| Statistical analysis
All statistical analysese were performed using R software 4.0.1, SPSS analysis tools (IBM Corp.), or the Prism8 software program (GraphPad Software). Continuous variables were reported as Mean ± SD. For the clinical data, Student's t-test was used for continuous variables and chi-square or Fisher's exact test for categorical variables. All tests were two-sided. Spearman correlation analysis was used for correlation analysis. Both univariate and multivariable Cox regression were performed to estimate clinical outcomes. We used a multivariable model with a forward stepwise regression procedure to screen out the potential predictors that have been reported or assumed to be predictive of poor outcome.
Based on the multivariable analysis, we used modeling nomograms to predict prognosis at 1-, 2-, and 3-year follow-ups. R package "meta" was used for Meta-analysis. p < 0.05 was considered significant for all statistical analyses.

| Increased EMP3 expression predicts poor prognosis in gliomas patients
In the initial study, we focused on determining the mechanistic role and clinical value of EMP3 in glioma. To investigate the expression of EMP3 in gliomas at different stages, we analyzed EMP3 mRNA expression in 3 datasets. We observed that the expression of EMP3 was elevated in glioblastoma. In the CGGA dataset, WHO grade III (n = 334) and grade IV (n = 388) tumors had a significantly higher expression of EMP3 than grade II (n = 291) ( Figure 1A) tumors. In the TCGA-GBMLGG dataset, a significant increase in EMP3 expression was confirmed in grade III (n = 244) and IV (n = 150) tumors ( Figure 1B). Furthermore, an upward trend was also observed in the Rembrandt dataset ( Figure 1C).
After elucidating the correlation between EMP3 expression and tumor malignancy, we aimed to demonstrate the prognostic value of EMP3. According to the median value of EMP3 expression, patients were divided into high and low EMP3 expression groups. The Kaplan-Meier curve and survival comparison analysis showed that patients with high EMP3 levels from the CGGA (HR = 1.77, 95% CI = 1.59-1.98), TCGA (HR = 3.69, 95% CI = 2.82-4.83), and Rembrandt (HR = 1.68, 95% CI = 1.46-1.92) datasets had worse overall survival (OS) rates than those with low EMP3 levels ( Figure 1D-F). To improve the stability of this result, we used a fixed-effects model to summarize the HRs of these three cohorts. The results confirmed that patients with high EMP3 expression had significantly shorter OS times than patients with low EMP3 levels (RR = 2.00, 95% CI = 1.84-2.18, Figure 1G).
To further validate these results, IHC for EMP3 was performed to evaluate EMP3 expression in patient-derived glioma tissues from two institutions. As expected, there was a significant increase in EMP3 in high-grade glioma (HGG) compared with low-grade glioma (LGG) ( Figure 1H-J). In addition, as expected, in comparison with adjacent tissues, a significant increase in EMP3 was revealed in the tumor core region ( Figure S1). According to the above data, the expression of EMP3 increased with the progression of glioma, suggesting that EMP3 may be involved in the development of tumor malignancy.

| EMP3 regulates immune infiltration and immune activation in gliomas
Infiltration of immune cells plays a critical role in a variety of cancers, which may lead to different clinical outcomes. The correlation between the immune profile and prognosis has been reported in several cancers, especially gliomas. The specific presence of EMP3 in lymphoid tissues, including the spleen and thymus, is thought to be directly or indirectly involved in immune system regulation. 25 However, the role of EMP3 in the tumor immune microenvironment remains unclear. Therefore, we aimed to explore the correlation between EMP3 levels and the status of immune infiltration to reveal the underlying mechanism in affecting the prognosis of gliomas.
Twenty-eight types of immune cells were systematically estimated from the CGGA dataset using the ssGSEA algorithm ( Figure 2A). The Spearman method was used to evaluate the relationship between EMP3 expression and immune cell infiltration, which revealed a close correlation between EMP3 expression and the infiltration of cells showed significant differences between the high and low EMP3 subsets ( Figure 2C).
In our present study, we observed that the majority of immune

cells, both of protumor immune cells, including MDSCs and regulatory T cells (Tregs), and antitumor immune cells, such as natural killer (NK)
T cells, were significantly increased in high-grade glioma ( Figure 3A).
Notably, we observed that patients with high expression of EMP3 also had high expression of the therapeutic targets PD1/PDL1 and CTLA4, which are considered to be immune checkpoint proteins ( Figure 3B).
Although the increase in antitumor immune cells in high-grade glioma seems to be paradoxical, the significant infiltration protumor immune cells might counteract the infiltration of antitumor immune cells and disrupt the balance between the two cell types. To further explore the existence of malignant gliomas with a protumor immune phenotype, manually curated gene sets associated with both adaptive and innate immune responses were used to quantify the immune phenotype ( Figure 3C). As is shown in the heatmap, with an increase in EMP3 expression, the tumor microenvironment is more likely to present respond to immune checkpoint blockades (ICBs). This is consistent with the conclusion drawn above that FCER1G plays a critical role in the activation of the immune response. In addition, the results of GSVA revealed a high correlation between the FCER1G and the activated PDL1 pathway (r = 0.57, p < 0.05), activated CTLA4 pathway (r = 0.42, p < 0.05), and T cell-mediated immunity (r = 0.53, p < 0.05) ( Figure 3D).
The above findings suggest that EMP3 is involved in immune infiltration remodeling in glioma and is closely associated with T-cell infiltration, which plays a significant role in immunosurveillance evasion in malignant glioma. 26 Table 1. Detailed information on baseline characteristics, tumor malignancy, comprehensive histopathological biomarkers, and EMP3 expression levels is summarized.

| The clinical value of EMP3 for predicting tumor characteristics and survival outcomes
Although there was a slight difference in Ki67 expression between the two cohorts, there were no differences in the patients' demographics and the other tumor characteristics.
To validate the clinical role of EMP3, according to the median expression levels of EMP3, we divided the patients from the two centers into a high EMP3 group (n = 82) and a low EMP3 group (n = 97).
By univariate analysis of clinical features, we observed that EMP3 was more likely to be associated with high malignancy (p = 0.032), high Ki67 expression (p = 0.005), and wild-type IDH (p < 0.001).
However, there were no significant differences in age, sex, PHH3 levels, ATRX mutation status, or MGMT methylation levels. In addition, of the patients with tumors with high EMP3 expression, 51.5% and 96.9% had a poor prognosis at 1 year and 3 years after surgery, respectively (Table 2).   (Table 3).

| Univariate and multivariate analysis of risk factors for overall survival of glioma patients
To determine the clinical predictive value of EMP3 expression, we

| Establishment of nomogram and validation of predictive accuracy for poor outcome
To better apply the results of this study to clinical practice, we   Then, we analyzed these three independent predictors using decline curve analysis (DCA) to confirm the predictive ability of the nomogram for the 1-, 2-, and 3-year overall survival rates of the patients. The curves demonstrated that the nomogram with combined factors was significantly better than individual factors alone in prognostic prediction. In addition, high expressed EMP3 is of great predictive value and accuracy in prognostic prediction of glioma patients ( Figure 4D). To demonstrate the generality of the nomogram model, a validation cohort set was used to further confirm the prediction value of the model ( Figure 4E). According to the above analysis, the establishment of this model is undoubtedly of substantial significance to both clinician workers and patients.

TA B L E 3
Comparison of Clinical characteristics and comprehensive histopathological biomarkers between favorable and unfavorable outcomes at 1-year and 3-year follow-ups  20,21 The above findings suggested that the function of EMP3 in solid tumors might be multifaceted and dependent on the specific type of cancer.

1-year follow-up 3-year follow-up
Although EMP3 was initially identified as a tumor suppressor in low-grade gliomas, its inhibitory role is still controversial. EMP3 expression was significantly higher in GBM than in non-neoplastic white matter and led to worse OS rates in WHO grade II-III glioma. 23,35 Another recent study reported that EMP3 directly interacts with TGFBR2 in glioma cells, which subsequently activates the TGFβ/

Smad2/3 pathway and enhances tumor progression in vitro and in
vivo. 22 In our present study, we determined that EMP3 enhanced glioma progression and showed clinical value for prognostic prediction.
Tumor initiation and progression is a complex process that requires interactions among cancer cells, the tumor microenvironment, and the immune system. 36   There are several limitations in our study. First, the patient outcome data of patients were recorded from outpatient and telephone interviews at 1, 2, and 3 years, for which there was referral bias due to unacceptable and incoordinate patients with unfavorable neurological status or clinical outcomes. Thus, a potential underestimation of poor prognosis might have influenced the overall estimate.
Additionally, because of the relatively limited number of patients, there is potential for minor bias to skew the interpretation.

| CON CLUS ION
This study demonstrated that EMP3 is a novel independent predictor for clinical diagnosis, prognosis, and immune infiltration in glioma patients. These results are of great clinical significance and will contribute to prognostic prediction and the development of individualized therapies. Although more clinical data from other institutions are required for further validation of our nomograms, individualized quantitative risk assessment using the present nomograms would be a practical approach for predicting prognosis and counseling patients.

CO N FLI C T O F I NTE R E S T
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be constructed as a potential conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
Publicly available datasets were analyzed in this study. These data can be found here: http://gliov is.bioin fo.cnio.es/. The supplementary material for this article can be found online. All processed data and R codes used in this study can be obtained from the corresponding author on reasonable request.