MicroRNA signature constituted of miR-30d, miR-93, and miR-181b is a promising prognostic marker in primary central nervous system lymphoma

MicroRNAs (miRNAs) are small RNA molecules that inhibit gene function by suppressing translation of target genes. However, in primary central nervous system lymphoma (PCNSL), the biological significance of miRNAs is largely unknown, although some miRNAs are known to be prognosis markers. Here, we analyzed 847 miRNAs expressed in 27 PCNSL specimens using microarray profiling and surveyed miRNA signature for prognostic prediction. Of these, 16 miRNAs were expressed in 27 PCNSL specimens at a frequency of 48%. Their variable importance measured by Random forest model revealed miR-192, miR-486, miR-28, miR-52, miR-181b, miR-194, miR-197, miR-93, miR-708, and let-7g as having positive effects; miR-29b-2*, miR-126, and miR-182 as having negative effects; and miR-18a*, miR-425, and miR-30d as neutral. After principal component analysis, the prediction formula for prognosis, consisting of the expression values of the above-mentioned miRNAs, clearly divided Kaplan-Meier survival curves by the calculated Z-score (HR = 6.4566, P = 0.0067). The 16 miRNAs were enriched by gene ontology terms including angiogenesis, cell migration and proliferation, and apoptosis, in addition to signaling pathways including TGF-β/SMAD, Notch, TNF, and MAPKinase. Their target genes included BCL2-related genes, HMGA2 oncogene, and LIN28B cancer stem cell marker. Furthermore, three miRNAs including miR-181b, miR-30d, and miR-93, selected from the 16 miRNAs, also showed comparable results for survival (HR = 8.9342, P = 0.0007), suggestive of a miRNA signature for prognostic prediction in PCNSL. These results indicate that this miRNA signature is useful for prognostic prediction in PCNSL and would help us understand target pathways for therapies in PCNSL.

Introduction Primary central nervous system lymphoma (PCNSL), a rare subgroup of diffuse large B-cell lymphoma (DLBCL) arising in the central nervous system (CNS), is an aggressive malignant variant of nodal non-Hodgkin lymphoma (NHL) [1,2]. PCNSLs account for 3% of all primary CNS tumors and 1% of NHLs in adults [3]. Most PCNSLs are immune privilege site-associated DLBCLs, according to the WHO diagnostic criteria [1]. Despite intensive treatments including high-dose methotrexate (HD-MTX) based polychemotherapy with whole brain radiotherapy, the median overall survival (OS) time is approximately 4 years for PCNSLs and shows a poorer prognosis than that of extracerebral DLBCLs [4][5][6].
MicroRNAs (miRNAs) are small noncoding regulatory RNAs consisting of approximately 20-mer nucleotides, that inhibit gene function through suppression of translation of target genes [7]. Previous studies have reported that various miRNAs are involved in cell proliferation, differentiation, cancer, and cell death in all living organisms [8,9]. The mechanism of gene silencing by miRNAs is well known and is referred to as RNA interference (RNAi) [10]. Dysregulations of miRNA expression and RNAi mechanism are related to tumor malignancy in chronic lymphocytic leukemia [11,12] and acute lymphoblastic leukemia [13]. Although molecular biology of miRNA in PCNSL is largely unknown, the expression pattern of miRNAs has been reported in PCNSL and non-CNS DLBCL [14][15][16]. In PCNSL, miR-199a, miR-214, miR-193b, and miR-145 were down-regulated [17]. Inversely, miRNAs that were up-regulated include miR-17-5p and miR-20a (associated with MYC pathways), miR-9 and miR-30b/c (associated with blocking of terminal B-cell differentiation), and miR-155 (associated with cytokine-dependent expression) [17,18].
Many miRNAs are known to be potential biomarkers for diagnosis and prognosis in PCNSL [19][20][21][22][23][24][25]. In addition, there are potential biomarker miRNAs for PCNSL in the cerebrospinal fluid [26,27] and in the serum [28]. However, to date, the comprehensive function, significance, and effectiveness of miRNAs as biomarkers in the clinical studies of PCNSL have not been elucidated.
Here, we deciphered the miRNA signature through analysis between the expression patterns of miRNAs and their correlation to the prognosis in 27 PCNSL specimens. First, we selected 16 miRNA candidates from the 847 miRNAs detected using microarray technique. Then based on principal component analysis (PCA) after Random forest analysis and clustering analysis, we determined that miR-181b, miR-30d, and miR-93 constituted a miRNA signature in PCNSL. The results here indicate that this miRNA signature is useful for prognostic prediction in PCNSL and would help us understand target pathways for therapies in PCNSL.

Clinical specimens
Twenty-seven patients with PCNSL were diagnosed and treated at Chiba University, Toyama Prefectural Central Hospital, Wakayama Medical University School of Medicine, and Yamaguchi University. The study was approved by The Ethics Committee of Kyoto Prefectural University of Medicine (RBMR-C-1082-1), and experiments were performed in accordance with institutional guidelines. Written informed consent was obtained from all the patients.

RNA extraction and microarray hybridization
Total RNA was extracted from approximately 100 mg of each tumor tissue using Isogen (Nippongene, Toyama, Japan). The quality of the extracted RNA was verified with a Bioanalyzer System using RNA Pico Chips (Agilent Technologies, Tokyo, Japan). Approximately 1 μg of RNA, amplified twice, was used for hybridization with Affymetrix GeneChip miRNA Array, comprising of 30,424 probes (Affymetrix, Inc., Tokyo, Japan). After hybridization, the array chips for target detection were processed, washed, and stained using the Fluidics Station 450. The High-Resolution Microarray Scanner 3000 was used for scanning the signal, and GCOS Workstation Version 1.3 was employed for image-quality analysis (Affymetrix, Inc.). The values of miRNA expression were determined using Affymetrix Expression Console Software according to manufacturer's instructions (Affymetrix, Inc.). Arrays were normalized using a quantile normalization to impose the same empirical distribution of intensities and a Z-score was calculated as a standard deviation from their means, as described [29]. The microarray data was uploaded to the Gene Expression Omnibus (GEO) (GSE122011).

Random survival forests analysis
Random survival forests analysis was used to determine the variable importance factors distinguishing expression of miRNAs with microarray raw data, as described [29,32,33]. The values of variable importance reflecting the relative contribution of each variable to the prediction for the survival time, and they were estimated by randomly permuting its values and recalculating the predictive accuracy of the model, which were expressed as the log rank test statistics. The method was implemented by using the randomForestSRC package of the statistical software R.

Clustering analysis
Expression of miRNAs in the 27 PCNSLs was clustered with the hierarchical method using the JMP built-in modules (SAS Institute, Inc., Tokyo, Japan), as described [34].

Kaplan-Meier survival analysis
The Kaplan-Meier method was used to estimate survival distributions for each subgroup with the log-rank test among subgroups using the JMP built-in modules (SAS Institute Inc.), as described [34].

Statistics
Statistical analyses were performed using the JMP built-in modules (SAS Institute Inc.) as described [34]. The P-values < 0.05 were considered statistically significant.
The two subgroups divided by a Z 2 score (= 0.1067) also extremely divided the Kaplan-Meier curves (HR = 23.45498, 95% CI: 4.448828-245.78973, P < 0.0001, in multivariate analysis) (Fig 2B and Table 5). Expression of miR-30d, miR-93, and miR-181b was validated by the qPCR method, although the correlation between the results of microarrays and qPCRs were hardly detected (S3 Fig). These results suggested that the 16 universally expressed miRNAs on the microarray function as useful prognostic markers in PCNSL, and especially, the three selected miRNAs including miR-30d, miR-93, and miR-181b are valid for the miRNA signature as a promising prognostic marker in PCNSL, whereas the exact expression of miRNAs should further be addressed in a large data set. Representative GO terms (GO numbers) of the three miRNAs were cell migration (0030335), macrophage activation (0042116) and    Table 3). Besides, the targets of the three miRNAs included cell growth-related genes RBAK, GPR137C, HTR1F, and MAP3K2 (also known as MEKK2); apoptosis-related gene PDCD2L; and immune-related gene PDCD1LG2 (also known as PD-L2) (S1 Table). These results suggested that it would be a hint for development of target therapies in PCNSL.
Thus, we should also address comprehensive expression profiling of miRNAs in a larger population of PCNSL, as well as a non-CNS DLBCL study [43].
In this study, we selected 16 universally expressed miRNAs from 847 miRNAs detected using microarray, and then determined miR-30d, miR-93, and miR-181b as the CNS miRNA signature for prognosis prediction. This study was carried out on specimens from 27 patients with PCNSL and their miRNA expressions data were analyzed using a Random forest model, principal component analysis, and Kaplan-Meier method. Interestingly, miR-30d targets programmed cell death 2-like (PDCD2L), autophagy related 12 (ATG12), serotonin receptor 1F, G-protein coupled (HTR1F) (S1 Table). The miR-93 targets programmed cell death 1 ligand 2 (PDCD1LG2, also known as PD-L2), which is related to cancer immunotherapy, and G-protein coupled receptor 137C (GPR137C), and mitogen-activated protein kinase kinase kinase 2 (MAP3K2, also known as MEKK2) in the RAS-MAPK signaling (S1 Table). The miR-181b targets retinoblastoma (RB)-associated KRAB zinc finger (RBAK) in the RB signaling (S1 Table). Altogether, these data propose that the signature constituted of miR-30d, miR-93, and miR-181b is involved in cell death and proliferation, and immune tolerance. Our data demonstrated that a subgroup associated with lower expression of miR-93 showed poor prognosis (Fig 1D  and 1E), which could be explained by the genetic interaction between miR-93 and PDCD1LG2. Thereby, it might have acquired immune tolerance with PD-L2/PD-1 axis within the PCNSL microenvironments. In Hodgkin lymphoma, miR-30d and miR-93 are known to be up-regulated and down-regulated, respectively [44]. Furthermore, the plasma level of miR-93 is associated with higher mortality in 25 patients with lymphoma [44], which suggests that they are reliable biomarker candidates. Differentially expressed miR-30d is also predicted for CNS-DLBCL [45]. Although the data is limited by the small sample size and there is less evidence for the above-mentioned function and signaling pathways at present, it would help us to better understand biological significances of the CNS signature of miRNAs in PCNSL.