CD80 and CD86: expression and prognostic value in newly diagnosed glioblastoma

Purpose Strategies to modulate the tumor microenvironment (TME), including the vascular and immune components, have opened new therapeutic avenues with dramatic yet heterogeneous intertumor ecacy in multiple cancers, including brain malignancies. Therefore, investigating molecular actors of TME may help understand the interactions between tumor cells and TME. Immune checkpoint proteins such as a Cluster of Differentiation 80 (CD80) and CD86 are expressed on the surface of tumor cells and inltrative tumor lymphocytes. However, their expression and prognostic value in glioblastoma (GBM) is still unclear. Methods In this study, we have investigated, in a retrospective local discovery cohort and a validation TCGA dataset, expression of CD80 and CD86 at mRNA level and their prognostic signicance in response to standard of care. CD80 and CD86 at the protein level were also investigated in the discovery cohort. Results Both CD80 and CD86 are expressed heterogeneously in GBM at mRNA and protein levels. In a univariate analysis, the mRNA expression of CD80 and CD86 was not signicantly correlated with OS in both ONT and TCGA datasets. On the other hand, CD80 and CD86 mRNA high expression was signicantly associated with shorter PFS (p<0.05). These ndings were validated using the TCGA cohort; higher CD80 and CD86 expressions were correlated with shorter PFS (p<0.05). In multivariate analysis, CD86 mRNA expression was an independent prognostic factor for PFS in the TCGA dataset only (p<0.05). Conclusion Additional studies are warranted to validate our ndings and to explore the expression of CD80 and CD86 in GBM patients treated with immunotherapy and, more specically, with CTLA-4 inhibitors. patients. Furthermore, it could be used as a biomarker for patients’ stratication for future clinical trials. Our study suffers from the limitation of retrospective studies with a limited number of patients. Nonetheless, our results were validated in an independent dataset and support investigations of immune checkpoint molecules as potential prognostic biomarkers and potential predictive biomarkers of response to immunotherapies in GBM.


Introduction
Glioblastoma (GBM) is the most common and aggressive glioma in adults. The latest World Health Organization (WHO) guideline classi es GBM as grade IV glioma [1]. Over the last years, massive efforts have led to a better understanding of the pathology and the genetic of GBM [2].
To date, the most effective and approved standard therapeutic regimen is maximum surgical resection of the tumor followed by concurrent chemoradiation and adjuvant chemotherapy with temozolomide [1]. Despite this very intensive therapeutic regimen, newly diagnosed GBM patients have a dismal outcome with a median overall survival (OS) below 18 months [3]. The main known prognostic factors are (i) age, (ii) Karnofsky performance status -KPS-, (iii) MGMT promoter methylation status, and (iv) IDH mutational status [4].
Immunotherapies have dramatically improved melanoma prognosis [5] and other non-neurological solid tumors [5]. In the setting of primary brain cancer, results from clinical trials are still disappointing [6]. Nonetheless, speci c GBM patients responded, supporting the identi cation of biomarkers to stratify patients in the prescription of immunotherapies. Immune checkpoint proteins such as Cluster of Differentiation 80 (CD80; known as B7-1) and CD86 (known as B7-2) are expressed on the surface of tumor cells [7]. Furthermore, CD80 protein expression was observed in in ltrative tumor lymphocytes in melanoma [8]. Cytotoxic T-lymphocyte-associated antigen-4 (CTLA-4) and Cluster of Differentiation 28 (CD28) are located on T cells. Both CD28 and CTLA-4 proteins bind to their ligands on the antigen-presenting cells and major histocompatibility complex (MHC) [9]. CTLA-4 has a higher a nity to CD80 and CD86, and when bound to its ligands, T cells remain inactive and exhausted [10].
Antibodies targeting CTLA-4 were used in preclinical studies in multiple solid tumors, resulting in many ongoing clinical trials [11]. Ipilimumabanti-CTLA4-has also shown responses in patients with brain metastases, highlighting e cacy within the central nervous system [12]. Expression of the most studied immune checkpoint proteins, programmed death-ligand (PD-L1), was inversely correlated with overall survival in GBM patients [13]. However, the expression of CD80 and CD86 in GBM tissues and their prognostic signi cance has not been reported yet. This study investigated the mRNA and protein expression of CD80 and CD86 in GBM patients, aged below 70 and with KPS above 70 treated with the standard of care. In addition, we have investigated possible correlation with prognosis.

Patient samples
OncoNeuroTek (ONT) is a local brain tumor tissue bank collecting samples from patients operated at the University Hospital La Pitié-Salpêtrière.
All samples were collected with informed consent from patients. The inclusion criteria of the discovery local cohort (47 patients) were as follow: (i) newly diagnosed and histologically veri ed GBM, (ii) age at diagnosis is below 70 years, (iii) KPS above 70%, (v) known MGMT promoter methylation status, (vi) known IDH status, (vii) treated with the standard rst-line therapeutic regimen including chemoradiation and adjuvant temozolomide and, (viii) a documented clinical follow-up. The validation cohort (121 patients, TCGA cohort) clinical information and RNAsequencing data (read counts) were downloaded from the National Cancer Institute's Genomic Data commons (GDC) Data portal and from the NCBI GEO GSE62944, respectively.

Immunohistochemistry (IHC) staining
Para n-embedded tissue blocks (5-7 µm) from biopsies of newly diagnosed GBM patients were received from the ONT biobank. Tissue sections were depara nized using xylene and rehydrated. For antigen retrieval, each slide was embedded in citrate buffer at pH 4.0 and heated for 15 min in the microwave at 800w. 10% goat serum with 5% fetal bovine serum in 0.2% triton phosphate buffer saline was used as a blocking buffer. 3% hydrogen peroxide was used to block tissue peroxidation. Anti-human CD80 antibody (A16039; Abclonal) and anti-human CD86 antibody (A2353; Abclonal) were used at 1:500 dilution in blocking solution and incubated on the tissue slides overnight at room temperature.
Avidin-Biotin Complex (ABC) kit was used as a signal enhancer before the incubation in 3,3′-Diaminobenzidine (DAB). Slides were embedded in hematoxylin dye and rinsed with tap water for nuclear staining; gradual alcohol and xylene baths were used for dehydration and mounted with a hydrophobic mounting medium (Sigma, 24845633). All stained tissues were scanned via ZEISS Axio Scan 40x for bright eld imaging.

Quanti cation of IHC staining
Following all slides' imaging, three regions of interest with known dimensions (528*528 µm) were randomly selected for each tissue section and quanti ed using an in-house quanti cation Fiji code. Shortly, each image was imported to the Fiji program [14]. Using the color deconvolution tool, the area positive for DAB staining was isolated and quanti ed using a semi-automated in-house generated code. The percentage of DAB positive areas were calculated, and the mean value from the three images was calculated and used in the survival analysis.

Statistical analysis
A Violin plot was used to visualize our data's full distribution (GraphPad Prism) [14]. Spearman correlation between the expression values (RNA or protein) and age was evaluated to discard age bias. Survival analysis was performed by an open-source validated approach [15,16] by nding a supervised cut-off value for the CD80 or CD86 expression independently using the function, which determines the cut point based on the highest/lowest value of the log-rank statistics (low or high expression values), and then using these categories for Kaplan-Meier analysis or Cox proportional hazard regression modeling testing at each variable independently or to adjust for multiple variables including CD80/CD80 expressions and MGMT promoter methylation status P-values lower than 0.05 were considered signi cant [17,18].

Patients and tumors characteristics
Forty-seven patients with a con rmed GBM diagnosis ful lled the inclusion criteria: 14  CD80 and CD86 expression at mRNA and protein level At the mRNA level, CD86 expression was quantitatively higher than CD80 expression (Supplementary gure 1-A). In agreement with mRNA expression, IHC analysis showed that the expression of CD86 is higher than CD80 in our discovery cohort (Supplementary gure 1-B). Based on the IHC staining, CD80 and CD86 are observed in the cell membrane and/or cytoplasm Our patient's cohort was used as a discovery cohort, while the TCGA dataset was used as a validation cohort. In a univariate analysis, mRNA expression of CD80 and CD86 was not signi cantly correlated with OS in both the ONT cohort and TCGA dataset (Table 1). On the other hand, CD80 and CD86 mRNA high expression was signi cantly associated with shorter PFS (p = 0.04 and p=0.005, respectively; Figure 2, A-B). Moreover, these ndings were validated using the TCGA cohort; higher CD80 and CD86 expressions were correlated with shorter PFS (p-value; 0.0428, 0.00283; Figure 2, C and D). Interestingly, higher CD86 protein expression was associated with shorter PFS in the ONT cohort (P<0.005; Table 2). CD80 and CD86 protein expression were not available in the TCGA dataset for validation purposes.
As expected, MGMT promoter methylation was associated with longer PFS and longer OS in the ONT cohort (p<0.05 and p<0.05 respectively) and TCGA dataset (p<0.05 and p<0.05 respectively) ( Table 1 and 2). Furthermore, IDH mutations were also associated with better OS and PFS in the TCGA database (p<0.05 and p<0.05 respectively); however, in the ONT cohort, the limited number of IDH-mutant GBM did not allow a robust analysis (n=2). In multivariate analysis, CD80 mRNA expression did not provide additional prognostic information to MGMT promoter methylation in the ONT cohort. On the other hand, multivariate analysis of CD86 mRNA expression was an independent prognostic factor for PFS in the TCGA dataset only (p<0.05; Figure 3). We have observed a similar trend (p=0.27; Figure 3) in the ONT cohort, yet the trend was not signi cant, which could be related to the lower patient numbers (n=47) in the ONT cohort compared to (n=121) in the TCGA database.

Discussion
CD80 and CD86 molecules play an essential role in in uencing the immune recognition of GBM cells. They bind to the CD28 molecule with a costimulatory signal for T-lymphocytes activation. On the other hand, they bind to CTLA-4, resulting in an immunosuppressive effect. CTLA-4 has a higher a nity to CD80 and CD86, making these molecules' role in immunosuppressive effect higher than their costimulatory effect [18].
The current study has linked CD80 and CD86 expression on GBM tumor microenvironment to PFS. We observed a low correlation between mRNA and protein expression of CD80. However, a better correlation was observed between CD86 protein and mRNA expression. Low correlation between the mRNA and protein expression might be due to post-transcriptional mechanisms involved in turning mRNA into protein.
Another reason could be related to the stability of both mRNA and protein in our patient's samples. Finally, there is a possible error and noise in protein quanti cation and mRNA extraction that could in uence mRNA stability and protein expression [17].
The number of patients (n = 47) in the ONT cohort is lower than the number of patients in the TCGA dataset (n = 121). The higher number of TCGA GBM samples could be one reason that affected the statistical analysis and provided a better prognostic value than the ONT cohort. Indeed, GBM samples' availability with comprehensive clinical and biological annotations and ful lling the inclusion criteria is a limitation for a larger cohort. Larger patient cohort is needed to evaluate the prognostic value of CD80 and CD86 expression in GBM samples. In our protein analysis, co-staining of CD80 and CD86 is needed to determine these proteins' expression in different immune cell populations. Furthermore, other immune checkpoint proteins could be evaluated in future studies.
The expression of 50 immune checkpoint molecules was investigated in breast cancer. The study showed that high expression of costimulatory immune checkpoint molecules was associated with better PFS. However, no signi cant effect on prognosis was associated with CD80 and CD86 expression in the selected cohort [19]. Feng et al. reported that low expression of CD80 is a predictive biomarker for poor prognosis in adenocarcinoma [20]. Furthermore, CD80 and CD86 were found to be potential biomarkers for better prognosis survival in nasopharyngeal carcinoma [21]. Additionally, the molecular characterization of PDL1 expression was correlated with other checkpoint proteins, i.e., CD80, highlighting that higher levels of immunosuppression are associated with GBM than lower-grade gliomas (LGG) [22]. In myeloma cell lines, silencing the CD28-CD86 pathway resulted in myeloma cells' signi cant cell death [23]. A recent study constructed a more robust model, using GBM and LGG data from the TCGA and CGGA (Chinese Glioma Genomic Atlas), and identi ed that low expression of CD86 molecules is a good prognostic indicator for OS. PFS analysis was not applied in this study [24].
In 2017, Berghoff et al. described a speci c signature to predict the success of TMZ in MGMT-methylated patients. They showed that the TME signature could be used to indicate an individual's TMZ sensitivity. The TME was identi ed to be different between IDH mutant and wildtype. A richer tumor in ltrative lymphocyte (TIL) and a higher expression of PDL1 were observed in IDH-wildtype tumors [25].  [27]. In line with these ndings, our study supports that the expression of immune checkpoint inhibitors may inhibit T-lymphocyte and anti-tumor reaction.
CD86 expression could be used as potential biomarkers predicting the e cacy of ipilimumab in GBM patients. Furthermore, it could be used as a biomarker for patients' strati cation for future clinical trials. Our study suffers from the limitation of retrospective studies with a limited number of patients. Nonetheless, our results were validated in an independent dataset and support investigations of immune checkpoint molecules as potential prognostic biomarkers and potential predictive biomarkers of response to immunotherapies in GBM. Cox-P (proportional hazards) multivariate analysis of CD86 protein expression and mRNA expression. CD86 was found to be an independent prognostic factor in TCGA database (P=0.0019); mRNA expression of CD86 is a more predictive prognostic factor than MGMT methylation. A non-signi cant trend was observed in our ONT cohort

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. SuppFigures.pptx