High-Throughput MicroRNA Profiling of Vitreoretinal Lymphoma: Vitreous and Serum MicroRNA Profiles Distinct from Uveitis

Purpose: Vitreoretinal lymphoma (VRL) is a non-Hodgkin lymphoma of the diffuse large B cell type (DLBCL), which is an aggressive cancer causing central nervous system related mortality. The pathogenesis of VRL is largely unknown. The role of microRNAs (miRNAs) has recently acquired remarkable importance in the pathogenesis of many diseases including cancers. Furthermore, miRNAs have shown promise as diagnostic and prognostic markers of cancers. In this study, we aimed to identify differentially expressed miRNAs and pathways in the vitreous and serum of patients with VRL and to investigate the pathogenesis of the disease. Materials and Methods: Vitreous and serum samples were obtained from 14 patients with VRL and from controls comprising 40 patients with uveitis, 12 with macular hole, 14 with epiretinal membrane, 12 healthy individuals. The expression levels of 2565 miRNAs in serum and vitreous samples were analyzed. Results: Expression of the miRNAs correlated significantly with the extracellular matrix (ECM) ‒receptor interaction pathway in VRL. Analyses showed that miR-326 was a key driver of B-cell proliferation, and miR-6513-3p could discriminate VRL from uveitis. MiR-1236-3p correlated with vitreous interleukin (IL)-10 concentrations. Machine learning analysis identified miR-361-3p expression as a discriminator between VRL and uveitis. Conclusions: Our findings demonstrate that aberrant microRNA expression in VRL may affect the expression of genes in a variety of cancer-related pathways. The altered serum miRNAs may discriminate VRL from uveitis, and serum miR-6513-3p has the potential to serve as an auxiliary tool for the diagnosis of VRL.


Introduction
Vitreoretinal lymphoma (VRL) is a rare malignancy arising in the vitreous and subretinal space. This disease is characterized by high rate of relapse and may manifest before, after, or simultaneously with central nervous spread, which is generally incurable due to a lack of active systemic therapies. In Japan, multicenter epidemiologic surveys of uveitis were conducted in 2002 [1] and 2009 [2]. According to these surveys, the prevalence of VRL is increasing because of the advances in diagnostic tests and the increase in recognition of the disease. However, VRL remains a masquerade syndrome and The patient control group consisted of 40 patients having uveitis with vitreous opacity (eight males and 32 females; aged 68.1 ± 14.1 years; 20 patients with ocular sarcoidosis and 20 with uveitis of unknown etiology) as well as 12 patients with macular hole and 14 with epiretinal membrane (12 males and 14 females; aged 68.7 ± 9.5 years) and 12 healthy individuals (six males and six females; aged 53.9 ± 20.5 years). None of the control patients had associated vitreoretinopathy. VRL patients were divided into two groups according to clinical features: diffuse vitreous opacities with or without multiple subretinal white lesions. Demographic, clinical and laboratory data (including white blood cell count and soluble IL-2 receptor level) of VRL patients and controls are shown in Table 1.
VRL patients were age-matched with patients with uveitis (age comparison, p = 0.38), patients with macular hole/epiretinal membrane (p = 0.56), and healthy individuals (p = 0.40). All patients were Asian adults. All patients underwent brain magnetic resonance imaging (MRI). Vitreous samples were harvested from the mid-vitreous region at the start of a standard 3-port 25-gauge vitrectomy. The vitreous was removed by a vitreous cutter before intraocular infusion. Then, a complete vitrectomy with infusion of balanced salt solution was performed. Undiluted and diluted vitreous specimens were delivered immediately to the Cytology and Molecular Laboratory (performed at SRL, Tokyo, Japan). The samples were stored immediately at −80 • C until assay.
The diagnosis of VRL was based on clinical features and results of cytology, cytokine analysis, and PCR test using vitreous sample after diagnostic vitrectomy. Cytopathologic and immunocytochemical evaluations were performed on undiluted vitreous samples to evaluate features of the lymphoma, as described previously [21]. The samples were diagnosed without knowledge of the findings of IgH chain gene rearrangement or concentrations of IL-10 and IL-6. Undiluted vitreous specimens were also used in analysis of miRNA. Diluted vitreous samples were used in PCR for analysis of IgH chain gene rearrangement to detect B-cell clonal expansion (performed at SRL, Tokyo, Japan).
Written informed consent was obtained from all the participants in the study. The study was approved by the Ethical Committee of the Tokyo Medical University Hospital, Tokyo, Japan (2016-162). All investigations were conducted according to the principles of the Helsinki declaration.

Blood Sample Collection
From each participant enrolled in this study, a sample of venous blood (approximately 5.0 mL) was collected in BD vacutainer tube using 21-gauge needle, then centrifuged (1000× g, room temperature, 15 min) to collect serum, which was stored at −80 • C until assay.

Microarray Analysis
Total RNA was extracted from serum and vitreous samples using miRNeasy Mini Kit (Qiagen GmbH, Hilden, Germany) following the manufacturer's instructions. Gene tip miRNA was extracted from the serum and vitreous samples using 3D-Gene ® RNA extraction reagent from a liquid sample kit (Toray Industries, Inc., Kamakura, Kanagawa, Japan) and concentrated. Fluorescent labeling of RNA was performed using 3D-Gene ® miRNA Labeling kit. Labeled RNA was hybridized to a 3D-Gene ® Human miRNA Oligo Chip (Toray Industries, Inc.) designed to detect 2565 mature human miRNA sequences registered in miRBase release 21 (http://www.mirbase.org/). The chip was scanned using a 3D-Gene ® Scanner, miRNAs with signals higher than the background signal were selected (positive call), and only miRNAs with positive call were used in subsequent analyses. The miRNA signal values were standardized by global normalization (log conversion of data and median alignment) [22].

Bioinformatic Analysis and Statistical Analysis
Relative expression level of each miRNA was validated using one-way analysis of variance or t-test (p < 0.05). MicroRNAs with expression levels of at least 2-fold difference (|log 2 fold change| > 1 and p < 0.05) in the test sample versus control sample were analyzed.
Principal component analysis was used to discriminate between biological samples based on the distances of a reduced set of new variables (principal components), using the top three principal components to depict the results in three dimensions. Unsupervised hierarchical clustering analysis was performed using an algorithm based on Pearson correlation and the average linkage method. Differentially expressed genes and miRNAs in any two groups were identified using criteria comprising p value and fold change. Genes targeted by significantly modulated miRNAs were identified using miRDB database, an integrative database for prediction of human functional microRNA targets [23]. Pathway enrichment analyses of miRNA gene targets and differentially expressed genes in VRL were performed using DIANA-mirPath v.3.0 [24]. The web-based computational tools DIANA-mirPath v.3.0 and DAVID 6.8 (https://david.ncifcrf.gov/) were used to predict the target genes and altered pathways of the differentially expressed miRNAs. Cytoscape 3.7.1 (http://manual.cytoscape.org/en/stable/) was used to create plots of the relations between Extracellular matrix (ECM)-receptor interaction and microRNA, and between IL-10 and microRNA.
Statistical analyses were performed using JMP version 13 (Business Unit of SAS, Cary, NC, USA). Continuous variables were compared between two groups using Student's t-test. Differences were considered significant at p values less than 0.05. Machine learning algorithms were implemented by random forest algorithm using R environment (https://cran.r-project.org/).

High-Throughput Vitreous and Serum miRNA Profiling in VRL Patients
Since a global miRNA expression analysis with updated coverage of miRNA sequences for VRL has not been reported, we provide detailed descriptions of miRNAs associated with VRL by investigating the transcription of a large number of miRNA sequences in vitreous and serum of VRL patients by microarray strategy. Vitreous samples obtained from six patients (eight eyes) with VRL and 26 age-and sex-matched macular hole/epiretinal membrane and 16 uveitis patients were analyzed using dedicated and high-density array with a coverage of more than 2500 human miRNA transcripts and all mature miRNA sequences registered in miRBase release 21.

Vitreous miRNAs
Microarray analysis revealed a large number of modulated miRNAs in vitreous samples, which satisfied the t-test p value criterion (p < 0.05) and the fold change criterion (>2), showing a robust and statistically significant differentiation between VRL and macular hole/epiretinal membrane. Such a large number of modulated transcripts clearly reflect the high performance of the array in the detection of a wide range of miRNA sequences. We thereafter refined our analysis by selecting only modulated miRNAs annotated as having high confidence in miRDB database (http://www.mirdb.org/), which would render our results more informative. Using these criteria, we selected 1771 modulated miRNAs as shown in Supplementary Materials Table S1.
Interestingly, 1714 (control 1436 + uveitis 278) miRNAs were upregulated and only 57 (control 42 + uveitis 15) were downregulated in the vitreous ( Table 2).  As shown in Figure 1a, there was a significant separation between the vitreous samples of VRL patients and those of macular hole/epiretinal membrane patients. A miRNA pathway analysis of these vitreous miRNAs suggested that the Hippo signaling pathway was the most relevant among the pathways enriched in VRL-modulated miRNA target genes obtained by gene union analysis ( Figure 2).  As shown in Figure 1a, there was a significant separation between the vitreous samples of VRL patients and those of macular hole/epiretinal membrane patients. A miRNA pathway analysis of these vitreous miRNAs suggested that the Hippo signaling pathway was the most relevant among the pathways enriched in VRL-modulated miRNA target genes obtained by gene union analysis ( Figure 2). As shown in Figure 1a, there was a significant separation between the vitreous samples of VRL patients and those of macular hole/epiretinal membrane patients. A miRNA pathway analysis of these vitreous miRNAs suggested that the Hippo signaling pathway was the most relevant among the pathways enriched in VRL-modulated miRNA target genes obtained by gene union analysis ( Figure 2).  Histogram presenting pathways enriched in genes targeted by VRL-modulated miRNAs and differentially expressed miRNAs in vitreous of VRL patients compared with macular hole/epiretinal membrane patients. y axis: −log 10 (p-value). TGF: transforming growth factor, ECM: extracellular matrix, FoxO: forkhead box O, Rap: rhoptry-associated protein, AMPK: AMP-activated protein kinase, PI3K-Akt: phosphatidylinositol 3-kinase-protein kinase B, MAPK: mitogen-activated protein kinases, mTOR: mechanistic target of rapamycin, cGMP-PKG: cyclic guanosine monophosphate -dependent protein kinase, HIF-1: hypoxia-inducible factor-1, HTLV-1: human T-lymphotropic virus type-1.
Further pathway union analysis showed that the ECM-receptor interaction pathway was the most significantly influenced pathway (p < 0.05), followed by prion diseases, glycosphingolipid biosynthesis -lacto and neolacto series, TGF-beta signaling pathway, and Hippo signaling pathway (Supplementary Materials Table S2).

Serum miRNAs
Likewise, we used the same methodology to perform serum miRNA profiling in patients with VRL and healthy individuals. The serum samples of 13 patients with VRL, 12 healthy controls and 34 uveitis patients were tested. A statistical analysis showed that 36 serum miRNAs were significantly upregulated, while 49 serum miRNAs were significantly downregulated. A hierarchical cluster analysis of miRNAs showed that there was a separation to some extent between serum samples of patients with VRL and those of controls ( Figure 3). protein kinases, mTOR: mechanistic target of rapamycin, cGMP-PKG: cyclic guanosine monophosphate -dependent protein kinase, HIF-1: hypoxia-inducible factor-1, HTLV-1: human Tlymphotropic virus type-1.
Further pathway union analysis showed that the ECM-receptor interaction pathway was the most significantly influenced pathway (p < 0.05), followed by prion diseases, glycosphingolipid biosynthesis -lacto and neolacto series, TGF-beta signaling pathway, and Hippo signaling pathway (Supplementary Materials Table S2).

Serum miRNAs
Likewise, we used the same methodology to perform serum miRNA profiling in patients with VRL and healthy individuals. The serum samples of 13 patients with VRL, 12 healthy controls and 34 uveitis patients were tested. A statistical analysis showed that 36 serum miRNAs were significantly upregulated, while 49 serum miRNAs were significantly downregulated. A hierarchical cluster analysis of miRNAs showed that there was a separation to some extent between serum samples of patients with VRL and those of controls ( Figure 3). A miRNA pathway analysis of these miRNAs suggested that fatty acid biosynthesis was the most relevant pathway among the pathways enriched in VRL-modulated miRNA targets genes obtained by gene union analysis ( Figure 4).  Further pathway union analysis showed that the pathway of morphine addiction was the most significantly influenced pathway (p < 0.05), followed by fatty acid biosynthesis, thyroid hormone synthesis, mucin type O-glycan biosynthesis, and Hippo signaling pathway (Supplementary Histogram presenting pathways enriched in genes targeted by VRL-modulated miRNAs and differentially expressed miRNAs in serum of VRL patients compared with macular hole/epiretinal membrane patients. y axis: −log 10 (p-value). TGF: transforming growth factor, GABA: gamma-aminobutyric acid, FoxO: forkhead box O, MAPK: mitogen-activated protein kinases, cAMP: cyclic adenosine monophosphate, AMPK: AMP-activated protein kinase, Rap: rhoptry-associated protein, mTOR: mechanistic target of rapamycin.
Further pathway union analysis showed that the pathway of morphine addiction was the most significantly influenced pathway (p < 0.05), followed by fatty acid biosynthesis, thyroid hormone synthesis, mucin type O-glycan biosynthesis, and Hippo signaling pathway (Supplementary Materials Table S4).
These results indicate that, even in serum, patients with VRL had a significantly distinct miRNA profile.

Comparison Between Vitreous and Serum miRNA Profiles of VRL Versus Uveitis
To verify whether serum or vitreous miRNA profile can discriminate VRL from uveitis with vitreous opacity, we used the same methodology for vitreous and serum miRNA profiling in VRL and compared with the results in uveitis with vitreous opacity. A statistical analysis showed that 278 vitreous miRNAs and 15 serum miRNAs were upregulated, while 15 vitreous miRNAs and one serum miRNA were downregulated in VRL compared to uveitis with vitreous opacity (Table 3). A hierarchical cluster analysis of miRNAs showed that there was a separation to some extent between vitreous and serum samples of patients with VRL and those of patients with uveitis ( Figure 5).
A hierarchical cluster analysis of miRNAs showed that there was a separation to some extent between vitreous and serum samples of patients with VRL and those of patients with uveitis ( Figure  5). MiR-6513-3p, 138-2f-3p and 4445-3p were upregulated commonly in vitreous and serum, and miR-6513-3p had the largest area under the ROC curve of the three, with a value of 0.84. These results indicate that miR-6513-3p may be a potential diagnostic biomarker of VRL, which is important for differentiation of VRL from uveitis with vitreous opacity (Figure 6).

Pathway Enrichment Analysis of miRNAs Deregulated in VRL
In the second part of our analysis, we aimed to identify all the molecular pathways that were targeted by the selected miRNA by performing pathway enrichment analyses based on the annotated gene targets registered in DIANA-mirPath. Detailed analysis of the most relevant pathways of VRL was done by performing a search using the KEGG database for potential compound identities and relevant pathways. The software allows evaluation of the miRNA regulatory effects and identification of regulatory pathways based on predicted and validated miRNA-target interactions.

Pathway Enrichment Analysis of miRNAs Deregulated in VRL
In the second part of our analysis, we aimed to identify all the molecular pathways that were targeted by the selected miRNA by performing pathway enrichment analyses based on the annotated gene targets registered in DIANA-mirPath. Detailed analysis of the most relevant pathways of VRL was done by performing a search using the KEGG database for potential compound identities and relevant pathways. The software allows evaluation of the miRNA regulatory effects and identification of regulatory pathways based on predicted and validated miRNA-target interactions. The predominant pathways were involved in ECM-receptor interaction (Figure 7).

Pathway Enrichment Analysis of miRNAs Deregulated in VRL
In the second part of our analysis, we aimed to identify all the molecular pathways that were targeted by the selected miRNA by performing pathway enrichment analyses based on the annotated gene targets registered in DIANA-mirPath. Detailed analysis of the most relevant pathways of VRL was done by performing a search using the KEGG database for potential compound identities and relevant pathways. The software allows evaluation of the miRNA regulatory effects and identification of regulatory pathways based on predicted and validated miRNA-target interactions. The predominant pathways were involved in ECM-receptor interaction (Figure 7).  Interestingly, according to pathway analysis using DIANA-mirPath based on combinations of differentially expressed ECM-receptor interaction pathways, a well described pathway for breast cancer [25] was suggested to play a central role in molecular pathology of VRL when compared to uveitis with vitreous opacity. Figure 8 shows the predicted genes targeted by miRNAs in the ECM-receptor interaction pathway, in the vitreous of VRL patients. Interestingly, according to pathway analysis using DIANA-mirPath based on combinations of differentially expressed ECM-receptor interaction pathways, a well described pathway for breast cancer [25] was suggested to play a central role in molecular pathology of VRL when compared to uveitis with vitreous opacity. Figure 8 shows the predicted genes targeted by miRNAs in the ECMreceptor interaction pathway, in the vitreous of VRL patients.

Comparative Analysis of Selected miRNA Gene Targets and Differentially Expressed Genes in VRL
To better define the role played by miRNAs in the pathogenesis of VRL, we sought to select miRNAs that are able to target genes modulated in VRL. Therefore, we used a more sophisticated

Comparative Analysis of Selected miRNA Gene Targets and Differentially Expressed Genes in VRL
To better define the role played by miRNAs in the pathogenesis of VRL, we sought to select miRNAs that are able to target genes modulated in VRL. Therefore, we used a more sophisticated integrative database for human microRNA target prediction (mirDB: http://www.mirdb.org/) to obtain a list of genes that are targeted by each of the selected miRNAs, with very high scores of over 95 (fold change > 7). DAVID analysis suggested the relationship of the miRNAs with pathways as shown in Supplementary Materials Table S5.
This analysis found no significant relationship with ECM-receptor interaction, but detected laminin gamma (LAMC) 3 related to ligands that bind to CD44. Thus, we were able to identify miRNAs that may regulate gene modulation involved in the pathogenesis of VRL. Supplementary Materials Table S6 shows the targeted genes and their corresponding targeting miRNAs.

Vitreous IL-10 Levels and miRNAs
Because elevated vitreous IL-10 level is essential to establish a diagnosis of VRL, we examined the correlation between vitreous IL-10 levels and the expression of 30 miRNAs (Figure 9). Using the miRDB database, we identified the miRNAs that target IL10. Among them, we used Cytoscape platform to show miRNAs that were upregulated or downregulated in this analysis. This analysis revealed that miR-1236-3p correlated significantly and positively with vitreous IL-10 levels in VRL patients ( Figure 10). Using the miRDB database, we identified the miRNAs that target IL10. Among them, we used Cytoscape platform to show miRNAs that were upregulated or downregulated in this analysis. This analysis revealed that miR-1236-3p correlated significantly and positively with vitreous IL-10 levels in VRL patients ( Figure 10).
Cytoscape platform to show miRNAs that were upregulated or downregulated in this analysis. This analysis revealed that miR-1236-3p correlated significantly and positively with vitreous IL-10 levels in VRL patients ( Figure 10).

Association between Clinical Features and miRNAs
VRL patients were divided into two groups according to clinical features: diffuse vitreous opacities with or without multiple subretinal white lesions. We evaluated the association between vitreous miRNA expression and clinical features ( Figure 11). J. Clin. Med. 2020, 9,1844 13 of 21 VRL patients were divided into two groups according to clinical features: diffuse vitreous opacities with or without multiple subretinal white lesions. We evaluated the association between vitreous miRNA expression and clinical features ( Figure 11). The vitreous expression levels of 4 miRNAs (miR3677-3p, miR-423-3p, miR-548ay-3p, miR-520h) were significantly (p < 0.05) elevated in VRL manifesting diffuse vitreous opacity with multiple subretinal white lesions than in VRL with diffuse vitreous opacity alone. On the other hand vitreous expression levels of 8 miRNAs (miR-107, miR-1269b, miR-4461, miR-7706, miR-6503-5p, miR-526b-5p, miR-7848-3p, miR-4422) were significantly (p < 0.05) elevated in VRL manifesting diffuse vitreous opacities alone than in VRL with vitreous opacities and multiple subretinal white lesions.

Machine Learning and miRNA Expression Validation in VRL
Machine learning analysis using random forest algorithm revealed that among the 17 Figure 11. Results of analyses of differentially expressed miRNAs for VRL manifesting vitreous opacity alone vs. vitreous opacity with subretinal infiltration. Heatmaps depict expression profiles of mRNAs in vitreous (a) and serum (b) The spectrum of dark red to dark blue color corresponds to high to low values.

Discussion
In the present study, we assessed the expression of 2565 miRNAs in both vitreous and serum samples obtained from VRL patients, disease controls and healthy controls, and identified miRNAs that are biomarkers of VRL, defined as the area under the ROC curve > 0.8 for VRL versus uveitis with vitreous opacity. An area under the ROC curve of 0.8 is considered moderately accurate for prediction. High throughput miRNA analysis provides a huge number of miRNAs to facilitate comprehensive analysis of disease status, and to elucidate miRNA profiles and pathways that distinguish VRL from chronic uveitis with vitreous opacity and from healthy status. Identification of vitreous and serum miRNA alterations may improve current diagnostic methods for VRL by substantiating the evidence of disease pathogenesis. Interestingly, among all the differentially expressed miRNAs observed in this study, the most prominent putative biomarker of VRL was miR-361-3p, as calculated by random forest. The ROC curve analysis of miR-361-3p yielded area under the curve of 0.921 (95% CI 0.806-1) for discriminating VRL from vitreous opacity with uveitis, indicating diagnostic value. miR-361-3p has been reported to be modulated in solid tumors such as colon, lung, pancreas, cervical, thyroid, cervical cancer and retinoblastoma [26][27][28][29][30][31]. These findings allow us to speculate that increased vitreous miR-361-3p may contribute to carcinogenesis in VRL patients, and the performance of vitreous miR-361-3p needs further validation before clinical use.
In this analysis, we found 1478 dysregulated miRNAs in vitreous samples of VRL patients compared with patients with macular hole/epiretinal membrane. As the number of differentially expressed vitreous miRNAs is relatively high, we focused on miRNAs with more than seven-fold changes in expression (miR-30b-3p, miR-1290 and miR-21-5p; Supplementary Materials Table S1). The majority of data about these miRNAs come from studies focused on various types of cancers, regarding their roles in cell proliferation, apoptosis, division, migration and invasion. Some of the studies analyzed hepatocellular carcinoma, non-small lung carcinoma, and esophageal squamous cell carcinoma [32][33][34][35]. The cellular infiltrate present in the vitreous of VRL patients suggests that lymphoma-derived and infiltrated leukocyte-derived miRNAs may contribute to the vitreous miRNA profile obtained in this study. However, resident cells may produce other miRNAs. Therefore, the differentially expressed miRNAs we identified in this study may better reflect the total miRNAs, rather than miRNA production from only the lymphoma and infiltrating leukocytes.
Early diagnosis of VRL is often challenged by the absence of accurate early diagnostic and prognostic biomarkers, because VRL may masquerade as chronic uveitis. Although vitreous IL-10 analysis had the highest diagnostic sensitivity [5,36], serum IL-10 is not detected in all patients with VRL. Effective primary screening using peripheral blood has not been reported. To the best of our knowledge, this is the first report of the potential of using circulating miRNAs as candidate biomarkers to identify VRL patients, especially those with vitreous opacity. Furthermore, the strong point of our study is that we studied a larger number of miRNAs (2565 miRNAs) in VRL patients compared to previous reports [16,17], which allows reliable statistical analysis. A lack of predictive serum biomarker is the main reason for delayed diagnosis of VRL. In the absence of unique cytological or laboratory features in peripheral blood which may facilitate diagnosis, diagnostic vitrectomy is currently used to define and diagnose VRL. However, diagnostic vitrectomy potentially carries the risk of complications. Therefore, less invasive tools to differentiate VRL from vitritis are needed. Recent studies have found that miRNAs are present in plasma in detectable levels. Due to their small size and stem-loop structure, they are more stable than messenger RNAs in body fluids, are resistant to degradation, and are easily and rapidly measurable [37][38][39][40]. Compared to diagnostic vitrectomy, blood-based biomarker assays are relatively economical and non-invasive methods to detect VRL, with additional advantages of easy accessibility and low risk associated with sample collection. Thus, we assessed the serum miRNA expression levels and found that the levels of ten miRNAs (mir-1273d, 133b, 146a-5p, 181a-5p, 193b-3p, 221-3p, 326, 345-5p, 422a, 4655-3p) were different between VRL and macular hole/epiretinal membrane patients, whereas the levels of three miRNAs (miR-6513-3p, 4445-3p, 138-2-3p) were higher in VRL than in uveitis with vitreous opacity. Among the differentially expressed miRNAs in vitreous obtained by comparing VRL versus uveitis with vitreous opacity, three modulated (upregulated) vitreous miRNAs (miR-6513-3p, 138-2-3p and 4445-3p) may be used for distinguishing VRL from uveitis with vitreous opacity, since principal component analysis of these miRNAs clearly separated patients in two distinct groups, as shown in Figure 6. In our study, ROC curve analysis showed that serum miR-326 expression level could discriminate VRL from macular hole/epiretinal membrane, miR-6513-3p could discriminate VRL from uveitis with vitreous opacities, and a combination of these two miRNAs may have even higher performance. Further evaluation is needed to confirm these very preliminary results. Clinically, discriminating between VRL and uveitis with vitreous opacity using serum miRNAs may support clinical decision making and allow timely initiation of treatment, thereby avoiding vitrectomy.
Previous studies have investigated miRNA profiles in diffuse large B-cell lymphoma (DLBCL) [41][42][43][44][45][46]. miR-181a-5p is reported to be upregulated in peripheral blood of acute lymphoblastic leukemia compared with normal peripheral blood mononuclear cells [47]. miR-422a has been associated with better overall survival in patients with DLBCL, and is related to patient's response to chemotherapy [48]. MiRNA-133b is one of the most widely researched RNAs and has been implicated in many cancers such as ovary, lung, liver and esophageal cancers [49][50][51][52]. miR-146a-5p is a tumor suppressor in malignancies, regulating proliferation and apoptosis. Its role in inflammation is less well studied but has been suggested to be a potential biomarker for rheumatoid arthritis [53][54][55]. Furthermore, similar to rheumatoid arthritis patients, miR-146a-5p expression levels are reduced in hepatocellular cell carcinoma [56]. miR-146a is involved in B cell hyperplasia [57], and miR-146a-5p has been reported to be a useful prognostic biomarker for DLBCL [58]. The possible roles of miR-326 as tumor suppressor miRNA have been established in various carcinomas. Given the above background, we investigated the possible pathways by which miRNA may be involved in VRL. According to KEGG analyses, the functions of most pathways and genes related to VRL are linked to Huntington's disease, circadian entrainment, ECM-receptor interaction, antigen processing and presentation, and morphine addiction pathways. These findings suggest that the pathogenesis of VRL may be related to disruption of these biological pathways. The above signaling pathways, particularly ECM-receptor interaction pathway, are closely related to the pathogenesis of VRL. The ECM-receptor interaction pathway plays a crucial role in breast cancer and kidney cancer [25,59], and miRNAs have been shown to regulate this pathway [60]. The KEGG pathway analysis showed substantial overlap among the signal transduction pathways of target gene sets involved in cancer transcriptional dysregulation pathway, proteoglycan, and cancer pathways, suggesting that these pathways may be closely related to cancers including VRL.
CD44 expression is upregulated in DLBCL [61][62][63]. In addition, there is a report that CD44 expression is upregulated via ECM-receptor interaction in kidney cancer [59]. Thus, ECM receptor interaction may be an important pathway also in VRL.
There are some discrepancies regarding miR-92, miR-19b, and miR-21. Kakkassery et al. [16]. reported that these three miRNAs were significantly upregulated in VRL patients. On the contrary, another study that assayed 168 miRNAs reported that these three miRNAs were not differentially expressed [17]. Our results of increased miR-21 expression in VRL patients are in concordance with the report by Kakkassery et al. [16]. Especially, miR-21 plays a pivotal role in the regulation of master regulator transcription factor essential in B-cell biology [64].
Pathways related to upregulated or downregulated miRNAs were mostly related to proteoglycans in cancers, possibly due to its post-transcriptional regulation by miRNAs. Proteoglycans are well known to be involved in DLBLC, contributing to tumor growth and metastasis [61][62][63]. Our findings of the relevance of proteoglycans in DLBLC are consistent with early studies by Nagel et al. [65], who reported a prognostic gene expression signature in DLBLC tumor tissue, which included the proteoglycan component. Hippo signaling is also an important network involved in the regulation of cancer cells and has been reported to be dysregulated in hematological malignancies including myelogenous leukemia [66][67][68]. A wide variety of inflammatory conditions strongly suggest that these miRNAs function in immune-mediated pathways. Pathway enrichment analysis for miRNA target genes supports the hypothesis that these miRNAs regulate inflammatory and interleukin signaling, consistent with inflammatory conditions in VRL [69].
The genes targeted by seven upregulated miRNAs (fold change > 7) in VRL are primarily related to pathways in cancers. Several lines of evidence support the role of modulated miRNAs in the pathogenesis of DLBCL. According to the miRDB database (http://www.mirdb.org/), downregulation of vitreous miR-4795-3p and miR-29b-2-5p may be related to the reported increase of their target gene MYD88, which is strongly associated with VRL pathogenesis [70,71]. The induction of miRNAs related to signaling in VRL is an interesting finding. According to the canonical miRNA-mRNA interaction, upregulation of these posttranscriptional elements in VRL may promote downregulation of important genes responsible for B-cell lymphoma. In fact, 21-5p has been associated with survival and proliferation of malignant cells in DLBCL [43,72].
We could not find any literature on the roles of miR-4445-3p in carcinogenesis and the immune system. Expression levels of this miRNA in both vitreous and serum were higher in VRL patients than in controls, suggesting that this miRNA is oncogenic. High expression of this miRNA in vitreous and serum could provide insight into its role, although further study is needed to analyze the its function. Since only very few miRNAs and their target genes have been identified, and the functions of most miRNAs remain unclear [73], such insight into the interplay of miRNAs in VRL would help further steer rational diagnostic and therapeutic strategies and ultimately improve patient outcomes.
MiR-1236-3p is significantly downregulated in gastric cancer and lung cancer relative to healthy persons [74,75], although to our knowledge its function remains unknown. Vitreous IL-10 strongly correlates with miR-1236-3p expression. IL-10 is an inhibitory cytokine that regulates tumor immune responses. In our series, all patients with VRL showed serum IL-10 levels below detection limit. MiR-1236-3p has been shown to regulate the PI3K/Akt signaling pathway [74], which is important for B-cell proliferation and differentiation.
Currently, it remains unclear why VRL mainly manifests diffuse vitreous opacity and multiple retinal or subretinal white lesions. An interesting finding of our study is that significantly different microRNA expression patterns exist between vitreous opacity with subretinal infiltration and vitreous opacity alone in patients with VRL, suggesting that miRNAs as regulators of molecular pathways impact the phenotypes of VRL. Lymphoma cells, infiltrating leukocytes, and ocular resident cells such as retinal pigmented epithelium and glial cells can produce miRNAs [76][77][78][79]. Therefore, the results presented here most likely represent the combined miRNA expression from lymphoma cells, leukocytes, and ocular resident cells, possibly interacting with each other. This study may provide valuable insight into the pathogenesis of each phenotype of VRL.
The present study had several limitations. First, the study was carried out using retrospectively collected samples. Consequently, sample handling conditions before microarray analysis, such as the interval between centrifugation and storage and the storage temperature, were not strictly controlled. Although miRNAs are more stable than messenger RNA, various processes can affect their levels in serum [38,39]. Further prospective studies are required to confirm our findings. Second, due to the relatively small sample size because VRL is a rare disease, the statistical significance is limited. Third, this investigation was a single-center case-control study, which does not allow calculation of positive and negative predictive values. However, because of the rarity of VRL and vitritis, conducting prospective cohort studies is difficult in terms of cost and time. Fourth, the large number of female patients and multiple forms of uveitis included in the control group may introduce some variability and confounders to the control group, which may prevent identification of other lymphoma-specific miRNA expression. Further study is needed to validate the present findings by including more patients with VRL compared with a carefully matched control group.

Conclusions
This work represents the first analysis performed on a large number of miRNAs integrated with study of the profiles of gene expression in VRL. Using this approach, we are able to identify the specific molecular pathways in which regulation by these miRNAs may occur. Further studies are needed to confirm the present observations and to clarify whether expression of the miRNAs is related to extraocular involvement and poor prognosis. Nonetheless, this study sheds light on some epigenetic aspects of VRL by identifying specific miRNAs, which may represent promising candidates for the identification of disease biomarkers and the design of novel therapeutic strategies in VRL.

Supplementary Materials:
The following are available online at http://www.mdpi.com/2077-0383/9/6/1844/s1. Table S1: Vitreous miRNAs modulated in vitreoretinal lymphoma versus macular hole/epiretinal membrane subjects in descending order of fold change. Table S2: Pathway union analysis of vitreous miRNAs in patients with vitreoretinal lymphoma versus macular hole/epiretinal membrane patients in descending order of significance. Table S3: Vitreous and serum miRNAs modulated in vitreoretinal lymphoma versus control in descending order of p-value. Filled in gray are miRNAs that were elevated in both the vitreous and serum. Table S4: Pathway union analysis of serum miRNAs in patients with VRL versus healthy control. Table S5: Pathway analysis for miRNA with fold change greater than 7. Table S6: ECM target genes with fold changes greater than 7 and scores of 95 or higher.