Correlation of alteration of HLA-F expression and clinical characterization in 593 brain glioma samples

HLA-F expression is downloaded from CGGA which includes 109 grade II, 72 grade III, and 144 grade IV and from the GSE 16011 array set which includes 24 grade II, 85 grade III, and 159 grade IV. To further validate the protein expression of HLA-F in different grades, we randomly selected 15 different grade glioma patients (5 patients for each grade) from Beijing Ditan Hospital for immunohistochemistry (IHC). All the patients are histologically diagnosed by two neuropathologists according to the 2007 WHO classification guidelines, and clinical information is downloaded from each website. This research is approved by the Ethics Committee of Beijing Ditan Hospital.

Isocitrate dehydrogenase 1 (IDH1) mutations are the most common mutations in gliomas, especially in low-grade gliomas and secondary GBM [22]. In the CGGA RNA-seq set, the genomic DNA was isolated from frozen tissues using a QIAamp DNA Mini Kit (Qiagen) according to the manufacturer’s protocol [22]. IDH1 status in the CGGA RNA-seq set was tested by pyrosequencing. TCGA subtype was based on previous research [23].

IHC was performed to detect HLA-F protein expression. The tumor tissues excised during the operation were immediately placed in 10% formalin for fixation, followed by dehydration, paraffin embedding, and sectioning. Anti-HLA-F antibody (Abcam) was used at a dilution of 1:100. Each slide was individually reviewed and scored by two independent neuropathologists, HLA-F protein expression (semi-quantitative scoring) = expression intensity × expression area. Expression intensity was scored using a 4-point scale from 0 to 3. Expression area was scored using a 5-point scale from 0 to 4.

Significantly related genes with HLA-F expression were retrieved by using the Pearson correlation analysis. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyzed the related gene sets and relevant biological function on the DAVID website (http://​david.​ncifcrf.​gov/​). In addition, the heatmap package of R language was used to list the genes positively related with HLA-F expression.

We performed Kaplan-Meier analysis to explore the prognostic value of HLA-F. Cox proportional hazards model analysis was performed to verify HLA-F as an independent prognostic factor. OS was estimated from the date of diagnosis to the date of either death or last follow-up. Student’s t test was performed to explore the expression distribution in different groups. Pearson correlation analysis was used to find genes related to HLA-F expression. R language packages (ggplot2, pheatmap, pROC, and corrgram) are used for other statistical computations and drawing figures. All differences were considered statistically significant at the level of p < 0.05.

Firstly, 325 samples with HLA-F expression profile and whole follow-up information are obtained from the CGGA RNA-seq set. The GSE 16011 array set was used for validation. In Fig. 1a, the results showed that HLA-F mRNA expression was positively associated with grade (p < 0.05). Moreover, HLA-F expression was higher in the isocitrate dehydrogenase 1 wild-type (IDH1 wt) group than in the IDH1 mutation (IDH1 mut) group (p < 0.05, Fig. 1b). Fortunately, all the above results were validated in 268 samples from the GSE 16011 array set (Fig. 1c, d). To further explore the relation between tumor grade and HLA-F protein expression, immunohistochemistry (IHC) was performed (Fig. 2a, b), and the results showed that the protein expression of HLA-F (semi-quantitative scoring) was higher in high-grade gliomas (p < 0.05).

To explore the expression distribution of HLA-F, we evaluated the expression pattern of HLA-F in different molecular subtypes as defined by the Cancer Genome Atlas (TCGA) network [23]. The results (Fig. 3a, b) showed that HLA-F was significantly upregulated in the mesenchymal molecular subtype than in other subtypes in both the CGGA RNA-seq set and GSE 16011 array set. To further validate this finding, receiver operating characteristic curves (ROC) for HLA-F expression and mesenchymal subtype of all grade gliomas are performed. Interestingly, the areas under the curves (AUC) are 80.5% and 76.5% in the CGGA RNA-seq set and GSE 16011 array set, respectively (Fig. 3c, d).

All 593 patients from the CGGA RNA-seq set and GSE 16011 array set have HLA-F mRNA expression information, and the median value was used as the cutoff point (0.2 for CGGA, 10.8 for GSE 16011). Then, we evaluated the survival time of each group and found that the patients with higher HLA-F expression have remarkably shorter OS in all grade gliomas and GBM in the CGGA RNA-seq set (Fig. 4a, b; p < 0.0001 for all grade gliomas, p = 0.0587 for GBM). Similar results were obtained from the GSE 16011 array set (Fig. 4c, d; p < 0.0001 for all grade gliomas, p = 0.0238 for GBM). Then, to verify HLA-F expression as an independent prognostic factor (Table 1), univariate and multivariate Cox regression analyses were performed in CGGA RNA-seq. The results of univariate regression showed that HLA-F (p < 0.001) along with grade (p < 0.0001), age (p < 0.0001), and IDH1 status (p < 0.0001) predicted the OS in all grade gliomas. In multivariate regression, HLA-F showed significant results (p = 0.027) after adjusting for grade (p < 0.0001), age (p = 0.687), and IDH1 status (p = 0.023). The above results suggested that HLA-F plays an important role in the progression of gliomas. Next, biologic function analysis should be performed to further validate our findings.

To investigate the relationship of HLA-F and another HLA-I molecular family, we performed Pearson correlation analysis in the CGGA RNA-seq set and GSE 16011 array set. The results (Fig. 5a, b) showed that the HLA-F expression was significantly associated with HLA-A, HLA-B, HLA-C, and HLA-E. To explore the biological processes associated with HLA-F expression in gliomas, Pearson correlation analysis between HLA-F expression and other genes in whole genome gene profiling of 325 patients in the CGGA RNA-seq set was performed. The results revealed that 183 genes (R > 0.5) are positively related to HLA-F expression (Fig. 6a). Biological process (BP) and KEGG analysis are also performed through the DAVID website (https://​david.​ncifcrf.​gov). The results are listed in Fig. 6b, wherein 5 BPs for 183 genes were observed including antigen processing and presentation, defense response to virus, innate immune response, negative regulation of viral genome replication, and positive regulation of catalytic activity. As for KEGG analysis, the proteasome, Epstein-Barr virus infection, antigen processing and presentation, insulin signaling pathway, and phagosome were included. In summary, all the above results indicated that HLA-F could affect glioma-related immune activities.