Introduction

The oncogenic role of sex steroid receptors in classical hormone-responsive tissues such as the breast and prostate are well-established. However, the same relationship is not as well understood in gliomas. First discovered in 1983, androgen receptor (AR) expression in malignant gliomas was described in several studies1. Epidemiological evidence supports a sex difference in glioblastomawith increased prevalence in males than females at a ratio of 1.5:12. Caroll et al. found that the expression of the AR mRNA and protein was present in all astrocytic neoplasms examined, regardless of histological grades3. Similarly, Chung et al. detected 40% of grade 4 gliomas and 75% grade 3 gliomas expressed AR using immunohistochemistry techniques. Gender differences in this population were not significantly different: the rate of AR expression in female gliomas was 39%, which is comparable to male gliomas of 47%4,5. Our institutional results showed no significant difference of AR expression between genders. We also detected AR overexpression in a greater proportion of patients, specifically, 74% out of a total of 58 glioblastoma) display positive AR expression in > 10% tumor cell nuclei, with the other eleven patients showing 1–10% positivity (93% with > 1% positivity) and only four patients’ tumor were found to be completely devoid of AR staining5.

While overexpression of AR in brain neoplasms has been described, their roles in the pathogenesis and prognostic significance remain largely obscure. AR is a member of a superfamily of ligand-activated transcription factors that are potentially oncogenic in gliomas6. In vitro studies have demonstrated that AR activation influences glioblastoma cell proliferation, migration, and invasion7. Additionally, several studies suggested a correlation between grade of anaplasia and AR expression. Yu et al. found that AR expression was upregulated in glioblastoma tissue as compared to normal brain tissue by Western blotting assays8. Bao et al. further demonstrated that AR expression increases according to pathological grade, with WHO grade 4 presenting highest expression9. Overall, these studies suggest that AR regulation plays a key role in the pathological process in malignant gliomas, which in turn could dictate the clinical outcome of the disease.

In our study, we aim to study AR expression and its association with the prognosis of malignant gliomas. Understanding the prognostic significance of AR expression may help us gain insights into the biology and natural history of this diverse disease, and design appropriate treatment strategies that may be optimized based on the personalized prognostic factors. We utilized The Cancer Genome Atlas (TCGA) database, which provides convenient access to genetic and clinical data of a large cohort of Caucasian patients with malignant gliomas and allows for database mining for potential correlation between gene expression and overall survival. For additional validation, we used another mRNA microarray dataset from the Chinese Glioma Genome Atlas (CGGA).

Materials and methods

RNAseq data of 160 glioblastoma tissues and 510 lower grade glioma (LGG) tissues were obtained from The Cancer Genome Atlas (TCGA) database that has clinical information including overall survival data. LGG contains World Health Organization (WHO) grade 2 and 3 glioma as described in published TCGA study in 201510. We used RNAseq data of 207 normal brain tissues from the GTEx database (https://gtexportal.org) as the control for the glioblastoma. RNAseq data of 249 glioblastoma tissues, 443 LGG tissues, and 19 normal brain tissues were obtained from the Chinese Glioma Genome Atlas (CGGA) database (http://www.cgga.org.cn/)11,12. We first updated all grading per WHO CNS 4th edition for all cases13. The new grading criteria per WHO CNS 5th edition were adopted as much as the data allows including separating the previously GBM with IDH mutation (IDH1 and/or IDH2) to grade 4 astrocytoma with IDH mutation and recategorize all LGG with IDH mutation (IDH1 and/or IDH2) and 1p19q co-deletion to be oligodendroglioma but did not regrade LGG per molecular markers due to lack of complete information. The clinical data of glioblastoma patients and LGG patients were obtained from both the TCGA (cases from 2007 to 2020) and CGGA databases (cases from 2004 to 2020). The protein expression profiles of 408 glioblastoma tissues and 854 LGG tissues were obtained from the Cancer Proteome Atlas (TCPA) database (https://tcpaportal.org/tcpa/).

For LGG in CGGA database, the numbers of the tissues of primary oligodendroglioma, WHO grade 2 (O), primary astrocytoma, WHO grade 2 (A), primary astrocytoma, WHO grade 3 (AA), and primary glioblastoma from CGGA database are 45, 85, 82, and 140, respectively. For the recurrent tissues, the numbers of the rO, rA, rAA, and rGBM from CGGA database are 15, 34, 70, and 109, respectively. Isocitrate dehydrogenase (IDH) mutations were obtained through pyrosequencing from the samples in the CGGA and TCGA databases10,14.

For TCGA database, we included 227 out of 510 primary LGG tissue with IDH mutation. We included 162 out of 606 patients with grade 4 tumor (glioblastoma and grade 4 astrocytoma with IDH mutation) with complete clinical information. For CGGA database, we analyzed 129 out of 136 primary grade 4 tumor (glioblastoma and grade 4 astrocytoma with IDH mutation). We also included 224 out of 282 primary LGG tissue with both IDH mutation and 1p/19q codeletion status with downloadable molecular RNA-seq data.

The correlation between AR gene expression level and AR protein expression level were obtained from cBioportal for Cancer Genomics database (https://www.cbioportal.org/). We used log2 (TPM) of AR in this study to compare AR mRNA expression between different groups. Z scores were used for protein expression comparisons. TPM: transcript per million. The cut-off values for high AR expression (high AR) and low AR expression (low AR) levels were based on the median values of both protein and RNA expression levels. Student t-test (two groups) or one-way ANOVA (more than two groups) was performed using GraphPad Software (Version 8.3.1, San Diego, CA). Overall survivals were compared between different groups with Kaplan–Meier analysis. Results were considered statistically significant if p < 0.05.

Results

AR expression across gliomas of different pathologic grades and molecular subtypes

Analyzing 681 tumor samples and 207 normal samples from TCGA and GTEx databases, AR mRNA expression was found to be significantly higher in glioblastoma compared to LGG, which in turn was higher than normal brain (p < 0.01) (Fig. 1A). Similar analysis with CGGA database also demonstrated higher AR gene expression in glioblastoma versus LGG (p = 0.016) (Fig. 1B). AR protein expression level is higher in glioblastoma than in LGG (p < 0.0001) (Fig. 1C). Protein and gene expression levels had a linear correlation with the spearman r equals 0.65 (p < 0.0001) (Fig. 1D). Figure 1E shows the mutation and copy number alteration (CNA) data for different types of gliomas in TCGA database. The descriptive data for each subtype of glioma are shown in Supplementary Table 1 and 2.

Fig. 1
figure 1

AR gene and protein expression levels of glioblastoma, lower grade glioma (LGG), and normal brain. (A) AR gene expression levels of glioblastoma, LGG from the Cancer Genome Atlas (TCGA) database, and AR gene expression levels of normal brain from Genotype-Tissue Expression (GTEx) database. (*p < 0.01) (B). AR gene expression levels of different grades of glioma from Chinese Glioma Genome Atlas (CGGA) database (p = 0.016). (C). AR protein level comparison between glioblastoma and LGG patients from the Cancer Proteome Atlas (TCPA) (****p < 0.0001). (D) Linear relation between AR protein expression level (Y axis) and AR expression at mRNA level (X axis) (spearman r = 0.65) (E) AR mutations and copy number alteration (CNA) analyses in different gliomas showing very low percentages of gene mutations or amplifications. (AO: Anaplastic Oligodendroglioma (74 cases); A: Astrocytoma (279 cases); AA: Anaplastic Astrocytoma (313 cases); glioblastoma: Glioblastoma (2630 cases); AO: Anaplastic Oligoastrocytoma (408 cases); O: Oligodendroglioma (408 cases); DA: Diffuse Astrocytoma (116 cases); OA: Oligoastrocytoma (276 cases); DG: Diffuse Glioma (1118 cases); GS: Gliosarcoma (21 cases)).

Full size image

When taking a closer look at a different subtype of gliomas including both primary and recurrent patients, anaplastic astrocytoma glioma has the highest AR gene expression level in all types of gliomas in CGGA database. glioblastoma and lower grade astrocytoma have similar AR gene expression levels, which are higher than oligodendroglioma (p < 0.01) (Fig. 2A). For different molecular subtype glioma, glioblastoma (with wildtype IDH) has higher AR expression level compared with grade 4 astrocytoma with mutant IDH although the difference was not statistically significant. LGG with wildtype IDH and LGG with mutant IDH but 1p/19q non-codeletion have higher AR gene expression levels compared with LGG with mutant IDH and 1p/19q codeletion (p < 0.001) (Fig. 2B). As expected, among different subtypes of the primary gliomas, glioblastoma had the worst overall survival (OS). Glioblastoma (with wildtype IDH) that has higher AR expression level numerically has significantly worse OS compared with grade 4 astrocytoma with mutant IDH (data not shown). In our study, due to existence of some uncertainty of defining grade 2 vs. 3 gliomas in grading criteria, we combined grade 2 and 3 into lower grade glioma for the above analysis as has been done previously10. We did divide grade 2 and 3 and performed the analysis accordingly in Fig. 1S. Grade 2 and 3 patients when divided based on molecular markers including IDH mutation status and 1p/19q codeletion status have no significant difference in terms of prognosis, justifying the rational of combing grade 2 and 3. Per WHO CNS 5th edition, all LGGs with IDH mutation and 1p19q co-deletion are now oligodendroglioma and all IDH wildtype LGGs are categorized into astrocytomas. We acknowledge that due to lack of complete studies on some of the molecular markers, our data in grade 2 and 3 gliomas most likely contain a small number of molecular GBM or grade 4 astrocytoma 15.

Fig. 2
figure 2

AR gene expression levels in different subtypes of LGG and glioblastoma patients and their overall survival comparisons in the CGGA database. (A) AR gene expression levels in different subtypes of LGG patients in CGGA database (O: newly diagnosed oligodendroglioma WHO grade 2; rO: recurrent oligodendroglioma WHO grade 2; A: newly diagnosed astrocytoma WHO grade 2; rA: recurrent astrocytoma WHO grade 2; AA: anaplastic astrocytoma WHO grade 3; rAA: recurrent anaplastic astrocytoma WHO grade 3; GBM: glioblastoma; rGBM: recurrent glioblastoma). (B) AR gene expression levels among LGG patients with wildtype IDH, LGG patients with mutant IDH with or without 1p/19q codeletion, glioblastoma patients with wildtype IDH, and patients with IDH-mutant grade 4 astrocytoma (p < 0.0001). **p < 0.01; ***p < 0.001; ****p < 0.0001.

Full size image

AR gene expression is a prognostic marker for lower grade glioma patients but not for glioblastoma patients

The LGG patients from both TCGA and CGGA were analyzed for their prognosis/survival based on different AR expression levels. Patients with lower AR expression at mRNA level have significantly better OS compared with those with higher AR expression (p < 0.0001) in TCGA database (Fig. 3A). LGG in CGGA database confirmed the same finding (p = 0.0025) (Fig. 3B).

Fig. 3
figure 3

Overall survival (OS) comparison between LGG and glioblastoma patients with higher and lower AR expression levels at both mRNA and protein expression levels. (A) Comparison of OS between LGG patients with higher versus lower AR gene expression at mRNA level in TCGA database (p < 0.0001). (B) Comparison of OS between LGG patients with higher versus lower AR gene expression at mRNA level in CGGA database (p = 0.0025). (C) Comparison of OS between LGG patients with higher versus lower AR gene expression at protein level in TCGA database (p = 0.03). (D) Comparison of OS between glioblastoma patients with higher versus lower AR gene expression at mRNA level in TCGA database (p = 0.51). (E) Comparison of OS between glioblastoma patients with higher versus lower AR gene expression at mRNA level in CGGA database (p = 0.44). (F). Comparison of OS between glioblastoma patients with higher versus lower AR expression at protein level in TCGA database (p = 0.058).

Full size image

The same correlation was also found between the protein levels of AR and OS. The LGG patients from TCGA with lower AR protein expression level have better OS compared with those with higher AR protein level (p = 0.03) (Fig. 3C).

In contrast, for glioblastoma patients, there is no significant difference of OS between patients with higher AR gene expression at mRNA level and those with lower AR gene expression for both TCGA and CGGA databases (Fig. 3D,E). We also did not observe any significant difference of OS between patients with higher AR protein expression level and those with lower expression in TCGA database as well (Fig. 3F).

We found that LGG patients with mutant IDH have lower AR expression level compared with those with higher AR expression level in TCGA database. For LGG patients in CGGA database, there is no significant difference of AR expression level between the patients with mutant IDH and those with wildtype IDH (Fig. 2S). Furthermore, we separated the LGG patients based on IDH mutation status in both TCGA and CGGA databases. We found that for LGG patients with mutant IDH status in CGGA database, patients with lower AR gene expression level have better OS (p < 0.05) compared with those with higher AR expression level (Fig. 3S B). For LGG patients with wildtype IDH in TCGA database, the patients with lower AR expression levels also have significantly better OS rate than the patients with higher AR expression level (p < 0.01) (Fig. 3S C). For LGG patients with wildtype IDH in TCGA database or LGG patients with mutant IDH in CGGA databases, there are no significant differences between the patients with lower AR expression level and those with higher AR expression level (Fig. 3S A and D). Again caution of data interpretation is advised due to high likelihood of containing molecular GBM or grade 4 astrocytoma due to incomplete molecular/chromosomal information required by WHO CNS 5th criteria in current LGG databases.

Gender is not associated with AR gene expression levels in gliomas

Both male and female LGG patients have similar AR gene expression levels in both TCGA and CGGA databases (Fig. 4A,B). There is no significant difference in OS between male and female LGG patients although female patients appear to have more long-term survivors (Fig. 4C,D).

Fig. 4
figure 4

AR gene expression levels and OS comparison in male and female LGG patients. (A) AR mRNA expression levels between male and female LGG patients in TCGA database (p = 0.94) (B) AR mRNA expression levels between male and female LGG patients in CGGA database (p = 0.72). (C) Overall survival between male and female LGG patients in TCGA database (p = 0.10). (D) Overall survival between male and female LGG patients in CGGA database (p = 0.43).

Full size image

For glioblastoma patients, male and female patients have again similar AR gene expression levels in TCGA (p = 0.32) and in CGGA databases (p = 0.72) (Fig. 5A,B). There is no significant OS difference between male and female patients in either database (Fig. 5C,D). It also showed no difference when combining glioblastoma and grade 4 astrocytoma cases or when separated (data not shown). However, there are more female patients in TCGA database who survived for long term, i.e., longer than 1000 days although the numbers are low after long-term follow-up (Fig. 5C), which is not shown in CGGA database (Fig. 5D).

Fig. 5
figure 5

AR gene expression levels and OS comparison in male and female glioblastoma patients. (A) AR gene expression levels between male and female glioblastoma patients in TCGA database (p = 0.32) (B) AR gene expression levels between male and female glioblastoma patients in CGGA database (p = 0.72). (C) Overall survival between male and female glioblastoma patients in TCGA database (p = 0.88) (D) Overall survival between male and female glioblastoma patients in CGGA database (p = 0.83).

Full size image

Discussions

Despite decades of research, there is a limited number of prognostic markers identified for gliomas, with less that can potentially serve as a therapeutic target. Among these markers, EGFRvIII, the most common EGFR mutation that occurs in up to 30% of high-grade gliomas especially glioblastoma, is correlated with a worse prognosis16. However, the antibody targeting EGFRvIII has failed to improve OS in phase III clinical trial casting doubt on the therapeutic value of targeting this mutation in glioblastoma17. Another potential marker gene, isocitrate dehydrogenase (IDH) is mutated only in about 10% glioblastoma, and the therapeutic roles targeting IDH mutation in gliomas are still under investigation18. There is thus an urgent need to identify novel and highly specific markers for gliomas to improve treatment efficacy.

Our previous study has demonstrated that the majority of glioblastoma overexpresses AR5. Current results have further shown that AR expression levels correlate to the grade and aggressiveness of the gliomas. For instance, glioblastoma expresses the highest level of AR compared with lower grade gliomas which, in turn, show overexpressed AR than normal brain controls. However, we did not observe the prognostic value of AR based on its mRNA expression levels within glioblastoma population but a significant role only in LGG. Interestingly, we found LGG patients with lower AR expression level have better OS compared with those with higher AR expression level regardless of IDH mutation status in both TCGA and CGGA databases (Fig. 3S).

One possibility to explain the difference between the AR prognostic values between LGG and glioblastoma is that the pathogenesis between LGG and glioblastoma differs from each other with AR contributing to LGG tumorigenesis but not in the process of glioblastoma development. However, our previous studies as well as from results from other laboratories demonstrated that AR antagonists can suppress glioblastoma growth both in vitro and in vivo5,8,19. These results strongly suggest that AR can be a novel target in treating gliomas, both in LGG and glioblastoma, possibly better serving as a predictive value in high grade glioma (HGG)/glioblastoma based on the higher expression levels detected among gliomas. Another possibility to explain the difference between the AR prognostic values between LGG and glioblastoma is that AR expression could be more saturated in glioblastoma in terms of promoting tumor development than LGG thus less prognostic within glioblastoma population. The third possibility is that in HGG/glioblastoma, there are common expressions of anomaly of AR whose functions are not only dependent on expression levels but AR response to androgen signaling. For example, studies from Zalcman et al. demonstrated a high proportion of patients’ glioblastoma expresses alternatively spliced AR mRNA thus with prematurely terminated C-terminal of AR being possibly ligand- independent forms of AR in HGG/glioblastoma which may not be present in LGG20. Our study, although did not involve AR splicing pattern analyses, did show a very low rate of mutations in AR in gliomas arguing against the idea that mutant AR is the dominant “anomaly of AR” present in HGG/glioblastoma. The common presence of alternatively spliced AR isoforms in glioblastoma but not in LGG could explain why the AR expression level is not prognostic within glioblastoma but is in LGG. The fourth possibility is that glioblastoma patients commonly receive concurrent chemo- and radiation therapy but not so in LGG tumors21. The more aggressive treatment as standard therapy in glioblastoma compared to LGG could have masked the prognostic value of AR in glioblastoma, in which case we may propose to stratify glioblastoma patients based on AR expression levels to decide whether to offer concurrent chemotherapy with temozolomide. Other possibilities to explain the observed prognostic value of AR expression levels in LGG but not in glioblastoma include the higher complexities/heterogeneities within glioblastoma than lower grade gliomas. We did observe that glioblastoma patients have the largest variation of AR expression levels compared to the other subtypes of gliomas (Fig. 2A,B). AR gene expression, among many other gene expression variations within the high heterogeneity of glioblastoma, may have lost its independent significance as a prognostic factor. Interestingly, we did not observe any gender difference in AR expression levels in either LGG or glioblastoma, or within Chinese (CGGA) or Caucasian populations (TCGA). These results indicate that AR expression could be an intrinsic marker for gliomatogenesis, irrespective to genetic background, which make this gene an ideal candidate as a therapeutic target, particularly considering the high frequency of overexpression of AR in glioblastoma as previously reported by our group5. However, it is important to note here that recent studies revealed sex-specific disparities in glioblastoma regarding incidence, molecular characteristics, and survival rates with male patients generally having poorer prognosis, suggesting hormonal influences and genetic differences play key roles in disease progression and outcomes22,23,24. It is interesting to point out that our results did not reveal any difference between genders in terms of AR expression levels, which seems to be contradicting the findings from others. However, we hypothesize that AR itself is a marker for carcinogenesis but its expression level is likely maximized in high grade gliomas/glioblastomas. The gender difference of survival likely could be due to the difference in the expression levels of the hormon/ligand of AR, i.e., testosterone or DHT, between genders, which dictated the survival. This is supported by the finding from Werner et al. that supplying exogenous testosterone promote glioblastoma growth in female mice19. However, further studies to confirm the hypothesis are needed and to test other possibilities including epigenetic regulation of AR between gliomas at different grades. Despite no significant differences in AR gene expression levels between genders in TCGA and CGGA databases, the broader evidence points to the importance of understanding how sex-specific hormonal and molecular pathways affect glioma pathogenesis and patient survival.

One of the intrinsic limitations of the current study is lack of a direct correlations of AR mRNA expression levels and AR protein levels, which were obtained from different patient population/database in current studies. However, our data suggests there is a strong correlation of AR expression at mRNA and protein levels. While our study showed extremely low possibilities of AR gene mutations in sequence or AR gene amplification, we have not determined whether and to what extent the AR protein is translated from alternatively spliced AR mRNA(s). Thus, the question of whether AR alternative splicing pattern can be prognostic in the glioma patients in addition to its mRNA/protein expression levels remain unknown. In terms of the grading system of glioma, we have spent extensive effort to regrade the gliomas in both databases per WHO CNS 5th edition15. We first deleted the subtype of lower grade glioma such as oligoastrocytoma that had been taken out of the WHO CNS 4th edition to ensure all grading of tumor is at least updated to the 4th edition13. When considering the added molecular information into the WHO CNS 5th Edition 2021, we the adopted with more accurately subtyping of gliomas taking into account of the IDH mutation and 1p/19q co-deletion information to redefine astrocytoma (1p19q not co-deleted) and oligodendroglioma (IDH mutant and 1p19q co-deleted) in grade 2 and 3 categories. In addition, the WHO CNS 5th Edition only defines IDH wildtype grade 4 glioma to be glioblastoma, WHO grade 4. But for grade 4 glioma with IDH mutation they are now classified as grade 4 astrocytoma, IDH mutant but no longer as glioblastoma reflecting the better prognosis of IDH mutant glioma even as grade 4 which is confirmed by our data by taking IDH mutant GBM (per WHO CNS 4th edition used in the databases) out of the GBM category in current study. Our study also shows that IDH mutant glioblastoma (WHO 4th Edition) or grade 4 astrocytoma, IDH mutant (WHO 5th Edition) have significantly lower AR expression level than IDH wildtype glioblastoma and oligodendrogliomas have lower AR expression than astrocytomas at the same grade. These data suggest that AR might be a better prognostic marker than histologic subtyping. However, we acknowledge the limitation of our study not being able to further define the grade of IDH wildtype astrocytoma per WHO 5th edition criteria with no results reported for TERT promotor mutation, or concurrent gain of whole chromosome 7 and loss of whole chromosome 10 (+ 7/−10) status in TCGA or CGGA databases. As for the third criterion that upgrades LGG to glioblastoma per WHO CNS 5th edition, i.e., IDH wildtype glioma with EGFR gene amplification with histological LGG, they only comprise 2.4% of all LGG in TCGA thus very unlikely to change the results from our study. It is also worth noting that it also remains arguable whether these molecular features truly reflect the grading and/or survival of the patients despite of the updates in WHO CNS 5th edition25.

Our research on AR expression in glioblastoma patients, influenced by recent findings and a detailed study, reveals AR's complex role in glioblastoma pathogenesis, particularly in contributing to radiation resistance, suggesting that anti-androgen therapy could enhance radiation treatment efficacy. Integrating AR expression with key prognostic indicators like MGMT promoter methylation and IDH1 mutation enriches our prediction of patient outcomes, underscoring the importance of personalized therapy strategies that incorporate AR expression as a crucial element of a comprehensive prognostic framework19.

Our laboratory has investigated in the past years the potential mechanistic link between AR expression and glioma pathogenesis, focusing on AR-regulated genes related to tumor aggressiveness and stemness. We have demonstrated the important linkage between AR expression and cancer stemness5. Our most recent research has demonstrated the potential of combining androgen receptor antagonists (ARAs) with radiotherapy (RT) to significantly improve glioblastoma treatment outcomes. Through both in vitro and in vivo studies, we've shown that ARAs not only exhibit significant radiosensitizing effects but also play a crucial role in immunomodulatory mechanisms by interacting with TGF-β/LIF/STAT3 signaling pathways26. These preclinical studies, together with our current clinical database research, strongly support to conduct future prospective research such as clinical trials targeting AR in glioblastoma patients.

Conclusions

Our current studies suggest the negative prognostic value of AR in gliomas particularly in lower grade gliomas (LGG) (grade 2 and 3) and its potential role as a therapeutic target for gliomas as it is commonly overexpressed across the grades. AR expression levels do not differ significantly between genders with is consistent between TCGA and CGGA databases. Further prospective studies to better exam AR expression/prognostic value in molecular GBM and/or grade 4 astrocytoma as well as in better defined LGGs are warranted due to lack of complete information to recategorize these tumors out of the LGG database as a major limitation of our current study.