CYPA and CD147 expression analysis in human glioblastoma: Evidence for a correlation with reduced survival

Glioblastomas (GBMs) are fatal WHO grade IV astrocytomas with a median survival of 14.6 months. Immunohistochemical (IHC) studies indicate that cyclophilin A (CYPA), and its receptor CD147 are elevated in GBM, but the impacts on disease progression are poorly understood. We determined their occurrence in diffuse astrocytomas (WHO grades II-IV) using IHC and evaluated how gene expression affects patient outcomes in The Cancer Genome Atlas (TCGA) and REMBRANDT datasets (GBM and Low-Grade-Glioma: LGG). While IHC staining of CYPA was evident across all tumour grades, CD147 staining increased with tumour grade. However, while analysis of the TCGA grade II/III tumour dataset showed a trend towards high CD147 gene expression and reduced patient survival (HR=2.34, p =0.064, n=132), the results were not statistically significant. By contrast, analysis of the TCGA GBM dataset showed that high CYPA expression was significantly associated with reduced survival (HR=1.30, p =0.009; n=593). Furthermore, Kaplan-Meier analysis of both the TCGA and REMBRANDT ( p =0.052; n=176) GBM datasets showed a comparable reduction in survival of about 3-months for patients with high CYPA expression. We found that CYPA gene amplification (≥2.2 copies) occurred in 72% of patients (REMBRANDT GBM dataset) and was correlated with reduced survival (log-rank p = 0.022). Finally, we observed a correlation between CYPA and CD147 gene expression in grade II/III (R=0.481, p =1.81x10 -6 ) and grade IV (R=0.132; p =0.0012) tumours. Together, our data identify CYPA and CD147 as potential pathogenic determinants in glioblastoma.

To date, the limited numbers of studies reporting CYPA and CD147 expression in GBM have predominantly relied on immunohistochemical (IHC) data. While useful, IHC is only semi-quantitative, with staining levels confined to simple categories (e.g. 0-3). Thus, adopting IHC These datasets were previously validated [24,25] and certified for use without limitation (http://cancergenome.nih.gov/publications/ publicationguidelines). The GBM cohort of 593 patients comprised 360 men and 233 women (mean age 57.7 years), with 457-recorded deaths from all causes, at the time of the study. The cohort of 132 patients with astrocytoma consisted of 70 men and 62 women (mean age 44.5 years) comprising grade II (n=43) and grade III (n=89) tumours, with 26-recorded deaths from all causes, at the time of the study. It should be noted that the TCGA datasets denoted as low-grade or lower grade glioma (LGG) included WHO grade II/III astrocytomas. Data from the Repository for Molecular BRAin Neoplasia DaTa (REMBRANDT) National Cancer Institute 2005, was analysed using the REMBRANDT web-based interface for simple calculations, accessed 2014 April 30. The original REMBRANDT database is now accessible via the link: https://gdoc.georgetown.edu/gdoc/workflows/index.

Statistical and bioinformatics analyses
The GBM and astrocytoma cohorts from The Cancer Genome Atlas (TCGA) [26] were statistically analysed using the R programming environment [27]. Kaplan-Meier survival curves showing the prognostic significance of CYPA and CD147 overexpression, were produced using the R "survival" package. Cox regressions were calculated using the R "coxph" function from the "survival" package with ties addressed by Efron's method. For both Cox regressions and Kaplan-Meier survival curves, all deaths were used as an endpoint. As indicated, analyses were adjusted for age and gender. The validity of proportional hazards assumptions were assessed by scaled Schoenfeld residuals [21]. Correlation coefficients were calculated using the R "cor. test" function and a simple linear model fitted to CYPA and CD147 expression data using the "lm" function.
Kaplan-Meier and gene copy number data from REMBRANDT were analysed using the REMBRANDT web-based interface and downloaded without manipulation (Figure 3b; Figure 4). Copy number was measured using the SNP 100K chip data. The REMBRANDT GBM data set provided expression intensities for all 8 CYPA probe tested. The median expression intensity for all CYPA Affymetrix probes was 8.11 (range: 6.7 -9.7; standard deviation of 1.3), meaning all probes reported a similar CYPA expression. The CYPA gene copy number versus patient survival data reported in Figure 4, was derived from probe 217602_at, which had an expression intensity of 8.97, was close to the mean intensity value for all probes.

Microscopy and imaging
Microscopy and imaging: Image acquisition was performed using an Olympus IX70 microscope fitted with a cooled CCD digital camera (DP70; Olympus, Japan) under software control (DP controller; Olympus).

Staining of CYPA and CD147 in astrocytoma (WHO grades II-IV)
Astrocytoma tumour sections (N=30) were stained for CYPA and CD147 (Table 1, Figure 1). CD147 staining was moderate-to-strong in GBM tissue, moderate in WHO grade III analplastic astrocytoma and mostly undetectable in WHO grade II diffuse astrocytomas (N=13). Based on these data, CD147 staining increased with tumour grading. Cyclophilin A staining was moderate-to-strong in GBM (N=14), moderate in grade III (N=3) and weak-to-moderate in grade II (N=13) astrocytoma. Compared to CD147, while CYPA staining increased with tumour grade, the delineation between grades was less marked.
to correlate candidate tumour gene expression and patient survival comes with caveats. Also, while increased CD147 staining is reportedly prognostic [14] in GBM, staining intensity may not necessarily equate with actual changes in protein levels. This is because the contribution of CD147 to tumorigenesis appears to correlate with the degree of receptor glycosylation [17,20,21], which is independent of protein expression. Similarly, while the cytokine behaviour of CYPA may be a contributing factor in GBM pathogenesis, as is reported for other malignancies [12,19,22], its extracellular location means it is unlikely to be detected by IHC.
The availability of high quality tumour specific genetic data in large depositories like The Cancer Genome Atlas (TCGA) and REMBRANDT can provide powerful quantitative evidence to corroborate and/or inform the findings of IHC studies. Thus, this study evaluated CYPA and CD147 expression in GBM and in low grade astrocytoma using both strategies.

Patient tissue
The Western Australian, state government pathology service, Path West, provided archived astrocytoma (confirmed pathology) and nonneoplastic brain tissue. All subjects gave their informed consent for inclusion before they participated in the study. The study was conducted in accordance with the declaration of Helsinki and the permission to use human tissue was granted by the Sir Charles Gairdner Group, Human Research Ethics Committee (approval #2009-016).

Immunohistochemistry
Gliomas were classified and graded according to current World Health Organisation (WHO) Guidelines [23]. Specimens were received fixed in 10% buffered neutral formalin, assessed macroscopically and routinely processed into paraffin embedded blocks. Following processing, 4μm sections of tissue were cut, transferred onto glass slides, deparaffinized, re-hydrated and subjected to antigen retrieval. Sections were probed with rabbit polyclonal anti-CYPA antibody (1:3000, Biomol, USA) or mouse monoclonal anti-CD147 (1:10000, Santa Cruz, USA). Anti-mouse and anti-rabbit HRP-tagged secondary antibodies were used according to the manufacturers instructions (EnVision TM DAKO, USA). Sections were stained with 3,3'-diamino-benzidine and counterstained with Haematoxylin.

Analysis of tissue immunoreactivity
Tissue sections were scored according to the method of Gu et al [8]. In brief, the numbers of cells showing membrane and cytoplasmic staining from ten representative microscope fields were totalled. The percentage immunoreactivity was determined for each tissue section and a semi-quantitative value was derived according to the following scheme: 0 = absent/<5%; 1+ = weak/6-25%; 2+ = moderate/26-50% and; 3+ = strong/>51%.

Publicly available datasets
To investigate the clinical significance of CYPA and CD147 mRNA overexpression in GBM, we downloaded the clinicopathologic and intratumoral mRNA expression data from 593 Glioblastoma (GBM) and 132 Low Grade Glioma (LGG) astrocytoma patient datasets at The Cancer Genome Atlas (TCGA) (https://tcga-data.nci.nih.gov/ tcga/tcgaDownload.jsp; accessed October 2013). The intratumoral mRNA levels provided were measured using Illumina HiSeq 2000 RNA Sequencing Version 2 and were subsequently normalised.
By contrast, CD147 gene expression in grade III astrocytoma correlated with reduced patient survival (HR 2.34, p=0.064; n=132) when compared to grade II astrocytoma. Within the astrocytoma grade III cohort, Kaplan-Meier analysis of patient survival returned a hazard ratio of 1.92 (p=0.14, n=89) for high expression of CD147 compared to low CD147 gene expression (Figure 2). In this instance, the number of patients was relatively small and the results were not significant.
A Pearson's correlation coefficient determination revealed a strong correlation between CD147 and CYPA expression in astrocytoma grade III (R = 0.481 and a p-value of 1.81 x 10-6; data not shown). In other words, high CD147 gene expression was associated with high CYPA gene expression.  High CD147 expression returned a HR value of 1.92 (p=0.14, n=89) relative to low CD147 expression. As only 2 deaths were recorded for grade II astrocytoma patients, a Kaplanmeier analysis was not performed.

Analysis of CYPA and CD147 gene expression in grade II/III astrocytomas
To test our IHC findings, we analysed grade II/III astrocytoma (N=132) expression data (low-grade glioma: LGG cohort, accessed April 2014) from the The Cancer Genome Atlas (TCGA) network [28]. For both genes, there was a slight (but non-significant), 1.04 fold higher expression level in grade III tumours compared to grade II tumours (CYPA, p=0.25; CD147, p=0.498; n=132). Cox-regression analysis of CYPA expression (adjusted for age and gender) showed that the slight difference between grade II and grade III tumours did not affect overall survival (HR=0.9, p=0.81; n=132, Table 2).

Analysis of CYPA and CD147 gene expression in GBM
Cox-regression analysis of CYPA and CD147 gene expression data from the TCGA GBM cohort (N=593; Table 3) using multivariate analysis (adjusted for age and gender) showed that high CYPA expression was associated with reduced patient survival compared to low CYPA expression (Hazard Ratio: HR, of 1.30, p=0.009, N=593).
Kaplan-Meier survival plots for CYPA showed that high expression decreased median patient survival by approximately 3 months ( Figure   3a). To test these findings, we analysed CYPA and CD147 gene expression (N=176) in a second GBM dataset, the REMBRANDT database.
We found strong agreement with the TCGA dataset, in that higher CYPA expression was associated with a reduction in median patient survival (p=0.052) of 85 days or 2.75 months (Figure 3b). CD147 gene expression was not associated with reduced patient survival (HR 1.06 p=0.860, 0.76-1.26).

CYPA gene copy number correlates with patient survival
The consistent distribution pattern of CYPA gene expression, viz. 33% and 67% (high versus low, respectively) and patient survival observed in both GBM datasets suggested a potential genetic origin. To explore this possibility, we derived a Kaplan-Meier survival plot as a function of CYPA gene copy number using data from the REMBRANDT dataset. We found that the CYPA gene was amplified by ≥ 2.2 copies in 72.2% of patients (determined using the AFFymetrix reporter: 217602_at) and that this increase significantly correlated with reduced patient survival (log rank p-value of 0.022, n=90; Figure 4). It should be noted that the AFFymetrix reporter: 217602_at maps to the first intronic sequence of the human Ppia gene.

Discussion
A key finding of this study is that CYPA gene expression is predictive of patient outcome. Specifically, multivariate cox-regression and Kaplan-Meier analysis of CYPA gene expression in the TCGA and REMBRANDT GBM datasets revealed a statistically significant relationship between high CYPA expression and reduced overall patient survival. We found substantial consistency across both GBM datasets,     where a third of all patients had tumours with high CYPA expression and that median survival among this group was reduced by about 3 months. By contrast, a comparison of CYPA staining in GBM and low-grade astrocytoma was less clear, although a weak correlation was suggested. For example, while staining intensity was generally robust in GBM and grade III astrocytomas, CYPA staining was also evident in grade II astocytomas. Similarly, CYPA gene expression showed no association between tumour grade and overall patient survival in the TCGA LGG dataset. Thus, while differential CYPA gene expression can predict GBM patient outcome, it cannot distinguish between lowgrade glioma (grade II/III astrocytoma) and GBM. While elevated CYPA expression has been previously noted in GBM tissue using IHC [10], this is the first study to link especially high gene expression with reduced survival.
As CYPA is encoded on chromosome 7, a chromosome frequently amplified in GBM isolates (29), we asked if CYPA amplification could account for the difference in gene expression observed. However, we found that increased CYPA copy number, which occurred in >72% of patients (according to REMBRANDT), did not equate with the 33% of patients that showed particularly high CYPA gene expression. Not surprisingly, the genetic origins of CYPA gene expression in GBM are more complex and are likely to involve regulatory, transcriptional and gene dosage effects.
Consistent with previous reports [13,14,30] we observed robust CD147 staining in GBM tissue. However, we found no difference in patient survival (TCGA and REMBRANDT) between high and low CD147 gene expression. Our findings are in line with a colorectal cancer study that showed an association between CD147 staining and tumour severity, but not between mRNA levels and tumour severity [31]. The extent to which CD147 is glycosylated, which purportedly increases with tumour grade and virulence [21], may account for the disparity between our gene expression data and the findings described in IHC studies [13,14,30]. In this instance, the correlation between differential staining intensity and malignancy may arise if the antibodies used exhibited a greater affinity for glycosylated CD147. This notion is supported by the work of Reithdorf et al (2006) who showed that removing specific carbohydrate moieties from glycosylated CD147 could alter and even abolish antibody reactivity [17]. Nonetheless, as CD147 staining was evident in grade III astrocytomas and GBMs, but virtually absent in grade II astrocytomas, we suggest that CD147 staining could be used to distinguish grade II astrocytomas from grade III astrocytomas and GBMs. This view is also supported by our multivariate analysis of CD147 gene expression of the LGG (TCGA) dataset which showed that CD147 expression was associated with a HR of 2.34 for grade III tumours (p=0.064; n=132) compared to grade II tumours. Within the astrocytoma grade III cohort, Kaplan-Meier analysis of patient survival returned a hazard ratio of 1.92 (p=0.14; n=89) for high expression compared to low CD147 expression. However, the number of patients was relatively small and the value was not significant.
This study also revealed a correlation between CD147 and CYPA expression in both the GBM (R=0.132; p=0.002) and LGG datasets (R = 0.481; p=1.81 x 10 -6 ). To our knowledge this study is the first study to report such a finding in GBM, or for any human disease, including other malignancies. This is despite considerable experimental evidence that CYPA and CD147 participate and/or interact in many disease related pathways [19]. Although interesting, the association between CYPA and CD147 expression remains speculative as the molecular basis for this interaction was not investigated. Nonetheless, previous studies suggest that transcriptional activation of both CYPA and CD147, could arise via the hypoxia inducible factor (HIF) [32,33] and/ or the transcription factor Sp1 [33]. Evidence also suggests that CYPA directly increases CD147 expression, at least in cholangiocarcinoma cells [19]. Finally, that CYPA and CD147 share a regulatory pathway(s) warrants further investigation.
As HIF mediates transcriptional activation of CYPA expression, a potentially novel pathway linking CYPA and CD147 could involve hypoxic and/or oxidative stress induced secretion of CYPA [34,35]. As activation of the HIF pathway is a driver of tumour growth and angiogenesis in GBM [36,37], it would be interesting to determine if GBM cells, like other cancer cells [19], can secrete CYPA. In this scenario, extracellular CYPA could contribute to GBM tumourigenesis by activating pro-survival signaling [2, 4,38], cell proliferation [19], cell invasion/migration [39][40][41] and MMP release [42], presumably via the CD147 receptor. Interestingly, extracellular CYPA has been detected in the CSF of patients with diffuse pontine glioma [43] and in an experimental model of prion disease [44].

Conclusion
Our findings highlight the need to consider both IHC and gene expression data when assessing the relevance of a given disease determinant/marker. Our findings also reveal that CYPA gene expression is an independent prognostic factor of patient outcome in GBM and a potential therapeutic target that warrants further investigation.