The samples were collected from the Department of Neurosurgery of West China Hospital between January 2011 and September 2019, which had adequate tumor samples for TMAs. Written informed consent was obtained from all patients before enrolment. The clinical follow-up was continuously performed until the death of patients. Hospital records and pathology reports of all patients were retrieved and reviewed. Brain tissues were collected from traumatic brain injury patients who underwent injured brain resection. The TMAs were prepared by Shanghai Outdo Biotech (Shanghai, China).

Immunohistochemical (IHC) staining was performed on TMAs using the following primary antibodies: PD-L1 (1:200, 4.37 µg/mL, 13684, Cell Signaling Technology), B7-H3 (1:400, 1.49 µg/mL, 14 058S, Cell Signaling Technology), B7-H4 (1:400, 0.66 µg/mL, 14572, Cell Signaling Technology) and VISTA (1:200, 0.05 µg/mL, 54979, Cell Signaling Technology).15–18 TMA slides were stained using standard immunohistochemistry techniques as previously described.15 The IHC staining was scanned using Pannoramic MIDI (3D HISTECH). The histoscore of the membrane and nuclear staining quantification was assessed according to the formula (3+ per cent cells)×3+(2+per cent cells)×2+(1+per cent cells)×1, and the formula total intensity/total cell number was used to assess the histoscore of pixel quantification. In this case, the normalized score is between 0 and 300. This method has previously been described.19

Immunofluorescence staining (IF) staining was performed on TMAs using the following primary and secondary antibodies: VISTA (the same as IHC) and CD3 (1:400, 1.63 µg/mL, ab699, Abcam), Ionized calcium binding adaptor molecule 1 (IBA-1, 1:100, 25 µg/mL, ab15690, Abcam), PD-L1 (1: 200, 2.5 µg/mL, ab210931, Abcam), goat anti-rabbit Alexa-594 antibody (1:500, 4 µg/mL, ab150088, Abcam), goat anti-mouse (Alexa Fluor 488, 1:500, 4 µg/mL, ab150117, Abcam). Sections were then cover slipped using hard-set mounting media containing DAPI (ab104139, Abcam).

The fluorescent IHC staining was scanned using Pannoramic MIDI (3D HISTECH). Quantitative measurement of the fluorescence signal was performed using the AQUA method of quantitative IF (QIF) as previously reported.20 Briefly, The QIF score of VISTA and PD-L1 signal for each antibody in the tumor were calculated by dividing the target VISTA and PD-L1 pixel intensities in the area of the tumor by the DAPI positivity. Tumor-infiltrating lymphocytes and tumor-associated macrophages (TAMs) were defined as cells that have CD3 and IBA-1 stain positivity, respectively. Scores were normalized to the exposure time and bit depth at which the images were captured, allowing scores collected at different exposure times to be comparable. Cases were considered as target expressers when the QIF score was above the signal detection threshold determined using the negative control preparation and visual inspection.

Before total DNA from fresh tumor tissue was extracted, the content of the tumor cells was confirmed by instantaneous sections. DNA was extracted by using Animal Genomic DNA Quick Extraction Kit (D0065 M, Beyotime, China). Primer pairs were used for gene amplification of CTNNB1 exon 3 (Primer sequence Forward: 5’-TCACTGAGCTAACCCTGGCT-3’, Reverse: 5’- GCATTCTGACTTTCAGTAAGGCAAT-3’) and BRAF V600E (Primer sequence Forward: 5’-AGGAAAGCATCTCACCTCATCCT-3’, Reverse: 5’-AGGAAAGCATCTCACCTCATCCT-3’) . PCR was performed by applying the default conditions: 5 min 95℃ for initial denaturation, which was followed by 30 cycles of amplification that consist of denaturation at 95℃ for 60 s, annealing at 60℃ for 45 s, and extension at 72℃ for 90 s with final extension 72℃ for 12 min. PCR products were sent to TsingKe for sequencing analysis.

All values are expressed as means±SEM. Statistical data analysis was performed using GraphPad Prism V.5.03 (GraphPad Software La Jolla, California, USA) statistical package and R language (V.3.6.1). Interdependence between staining and clinical data was calculated using the X² displayed by cross-tables. Student’s t-test was used for two-group analysis. The Cox proportional hazards regression model and Kaplan-Meier plots were generated by the survival package, and the log-rank test was used to compare survival curves between high and low expression groups. The cut-off points were determined by the survivalROC package. Statistical significance was defined as a p<0.05.

A total of 132 patients were enrolled, including 86 males and 46 females (male/female ratio 1.86:1, table 1, online supplemental table 1). Patients’ ages had a bimodal distribution and ranged from 2 to 75 years with the median age of 40 . We chose the lowest point (35 years) between the two peaks to stratify age. Pathologically, there were 89 ACPs and 43 PCPs. Within this cohort, 117 patients had primary tumors and 15 patients had recurrent tumors. The clinical presentations of patients in the study included the following: vision loss, headache, fatigue, drowsiness and unconsciousness. The average tumor diameter in this cohort was 3.6 cm (ranges from 1.2 to 9.2 cm). GTR and subtotal resection (STR) were achieved in 104 (78.8%) and 28 (21.2%) cases, respectively.

A total of 110 out of 132 patients (83.3%) were followed up for an average of 18.2 months, which ranges from 0.2 to 84 months (online supplemental table 1). Within the perioperative period, 26 deaths have occurred. For the 87 GTR patients being followed up, 6 (6.9%) had shown tumor recurrence within an average of 30 months, which ranges from 2 to 82 months. Among the 28 STR patients being followed up, 10 (35.7%) had experienced tumor recurrence within an average of 16.2months, which ranges from 6 to 53 months.

In this cohort, the cut-off values of PD-L1, B7-H3, B7-H4 and PD-L1 expressions were determined by survivalROC package, which was 70, 182, 0 and 20, respectively. Elevated expressions of PD-L1, B7-H3, B7-H4 and VISTA were significantly associated with some clinicopathological characteristics (table 1). For example, PD-L1 expression was associated with age, histopathological, tumor morphology and hydrocephalus, whereas VISTA expression was connected with age, histopathological and hydrocephalus.

Using IHC, we explored the expression characteristics of PD-L1, B7-H3, B7-H4 and VISTA in CP. Consistent with the previous study,12 the patterns of PD-L1 staining were predominantly membranous, and 101 (76.5%) positive cases were detected in this set of data. In ACP, PD-L1 expression was present in the regions of well-keratinized squamous tumor epithelium (figure 1A). Notably, we also found strong staining around the cholesterol crystals, which was not seen in previous reports (figure 1A). Compared with ACP, PCP had a much higher expression of PD-L1 (figure 1B), which was observed in multiple layers of tumor cells with the strongest expression surrounding the fibrovascular stroma and decreased with increasing distance from the stroma (figure 1A).

B7-H3 was detectable to varying degrees in all samples (100%). It was mainly expressed on the cell membrane of tumor cells and was also observed in vascular endothelial cells and tumor-infiltrating immune cells in the two subtypes of CPs (figure 1A). The average expression scores of B7–H3 in both ACP and PCP were higher than in normal brain tissues, but no significant difference was detected between ACP and PCP (figure 1B). Intriguingly, we also found that less than one half of CP tissues (53/132, 40.2%) have weak to moderate expression of B7–H4 (figure 1A), which is currently reported to be expressed only in a few type of tumor cells.21 B7–H4 mainly expressed on the surface of CP cell membrane far from the basal cells. There was no difference in expression level between the two subtypes of CPs (figure 1B).

IHC staining demonstrated that 106 (80.3%) CP specimens, including 67 ACPs (histoscore range from 10 to 90) and 39 PCPs (histoscore range from 10 to 230), were positive for VISTA staining. The expression of VISTA was detected both on the cell membrane and in the cytoplasm (figure 1A). It was predominantly presented in the regions of epithelial whorls in ACP and on multiple layers of tumor cells which are far from the basal cell in PCP (figure 1A). The VISTA histoscore was significantly higher in PCPs than in brain tissues and ACPs. However, we found no statistical significance between normal brain tissues and ACPs (figure 1B).

We further compared the expression of PD-L1, B7-H3, B7-H4 and VISTA in primary tumors versus recurrent tumors as well as between preoperative radiotherapy and non-radiotherapy groups. The results suggested that there was no significant difference between the groups (online supplemental figure 1).

All CP cases contain numerous immune cells in the stroma and epithelium. Among these cases, we also found that tumor-associated immune cells expressed B7 family ligands/receptors, of which VISTA expression was the highest (figure 2A). To explore the possible role of VISTA in the immune microenvironment of CPs, we measured its level in CD3⁺ tumor-infiltrating T cells and IBA-1⁺ TAMs by colocalization IF staining. As shown in figure 2B–D, VISTA was detected in all immune cell subsets but was significantly higher in IBA-1⁺ TAMs than in CD3 positive lymphocyte in both ACP (figure 2C) and PCP (figure 2D). Additionally, we found that there was a negative association between CD3 and VISTA, and a positive association between VISTA and IBA-1 (figure 2E–F).

We further analyzed CTNNB1 and BRAF mutation status in CPs by PCR. 60% (21 out of 35) cases exhibited activating mutations within exon 3 of the β-catenin gene CTNNB1 in ACPs (figure 3A) and 56% (14 out of 25) cases exhibited BRAF V600E mutation in PCPs (figure 3B). In ACPs, CTNNB1 mutation could be linked to the high expression of PD-L1, B7-H3 and VISTA. Moreover, there was a significant difference (p=0.0363) in B7-H3 expression between the two subtypes of CP (figure 3A). Notably, in PCPs, BRAF V600E mutation group showed a significantly higher VISTA expression comparing to the wild type group (p=0.0117) (figure 3B).

We further analyzed the relationship between CP tumor size and B7 family ligands/receptors expression. We found that VISTA expression and tumor size had an inverse association (R²=0.04871, p=0.0110); but the expression of the other three molecules had no association with tumor size (figure 4A). Kaplan-Meier survival curves by B7 family ligands/receptors expression levels are shown in figure 4B. Interestingly, strong VISTA expression was correlated with favorable outcomes in CPs. Calculated median OS in CP patients with high VISTA expression was 7 years compared with a median OS of 2.25 years in VISTA negative or low patients (p=0.0027). However, elevated expression of PD-L1, B7-H3 or B7-H4 was not associated with OS nor PFS in this cohort.

The univariate Cox regression model revealed that histopathological, tumor location, the extent of resection and VISTA expression were all correlated with PFS (p<0.05), whereas tumor location, hydrocephalus and VISTA expression were associated with OS (table 2). We selected the characteristics with p<0.1 in univariate for multivariate analysis. After adjustment, we found that only VISTA was an independent prognostic factor for OS (p=0.018), while only the extent of resection was an independent prognostic factor for PFS of CP patients (p=0.025).

Recently, some reports showed that VISTA overexpression is correlated with other B7 family ligands/receptors, such as PD-L1, PD-1 and CTLA-4. When investigating the molecules potentially associated with VISTA in the tumor microenvironment, we observed that VISTA expression was associated with PD-L1 in CPs. However, there was no significant association between the expression of VISTA and the expression of B7-H3 and B7-H4 (figure 5A). Furthermore, we explored the spatial distribution relationship between the expression of PD-L1 and VISTA in tumor tissues by colocalization IF staining. We found that PD-L1 and VISTA colocalized with each other only in immune cells within the tumor stroma but not in tumor cells (figure 5B,C).