This article is mentioned in:

Introduction

Age-related macular degeneration (AMD) is a leading cause of irreversible blindness worldwide, and affects 6.5% of the US population aged ≥40 years (1). AMD is classified into wet and dry AMD. The latter is the most common phenotype, which is characterized by an increase in the number and diameter of drusen, progressive atrophy of the retinal pigment epithelium (RPE), pigmentary irregularities and a graded loss in visual acuity (2–4). Previous studies have indicated that environmental and genetic factors influence the disease, including smoking, obesity and dietary fat intake (5–8).

It has been observed that oxidative damage and inflammation are involved in the pathogenesis of AMD (9,10). Furthermore, the complement system, a vital component of innate immunity and a complex network that contributes to inflammation, has been identified to be associated with the pathogenesis and progression of AMD (11). Clusterin, also known as apolipoprotein J because it is present in high-density lipoprotein complexes, has been indicated to be a negative regulator of the complement cascade and exhibits increased mRNA expression levels in the RPE cells of AMD patients (12). Clusterin is involved in cholesterol transportation and protein aggregation in human plasma, and a deficiency of this complement regulatory protein may contribute to inflammation and neovascularization (13–15).

Epigenetic mechanisms, including DNA methylation, chromatin remodeling, histone modification and non-coding RNA-mediated regulation, have been demonstrated to influence gene expression and interactions between the genetics and the environment during development (16). Non-coding RNA consists of long non-coding RNAs (lncRNAs) and microRNAs (miRNAs). Recent epigenetic and genetic evidence has indicated that miRNAs, such as miR-146A, miRNA-9, miRNA-155 and miRNA-125b are upregulated to modulate a family of potentially pathogenic genes involved in Alzheimer's disease and cancer (17–20). In 2011, Salmena et al (21) proposed a competing endogenous RNA (ceRNA) hypothesis, which is now supported by further evidence (22–24). The hypothesis suggests that a regulatory network exists whereby protein-coding and non-coding RNAs compete for binding to miRNAs by sharing miRNA-binding sites. Nevertheless, there is a lack of data on expression patterns of specific lncRNAs in AMD patients. In addition, it is unknown whether lncRNAs participate in the regulation of clusterin-related inflammation, or whether some lncRNAs are induced to aberrantly express or to participate in an altered ceRNA network after RPE cells are exposed to clusterin.

The aim of the present study was to elucidate the effects of clusterin on the expression of coding and non-coding RNA in RPE cells. The expressed mRNAs, miRNAs and lncRNAs were profiled in RPE cells treated with or without clusterin, and the ceRNA network was characterized. To the best of our knowledge, this was the first study to investigate the disease-specific lncRNA expression patterns and the associated ceRNA network in RPE cells.

Materials and methods

Cell culture

The adult human RPE cell line D407 (Procell Life Science and Technology Co., Ltd., Wuhan, China) was cultured in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum (Thermo Fisher Scientific, Inc., Waltham, MA, USA), 100 µg/ml streptomycin, 100 U/ml penicillin and 2.5 µg/ml amphotericin B at 37°C and 95% humidity with a 5% CO2 atmosphere.

MTT viability assay

D407 cells were seeded in 96-well plates at a density of 1×104 cells/well and allowed to reach 90% confluence. The growth medium was subsequently replaced with medium supplemented with clusterin at 0 (control), 0.1, 1.0 or 10 µg/µl. Cells were incubated for 48 h and analyzed using an MTT assay. Briefly, the medium was removed and replaced with DMEM supplemented with 5 mg/ml MTT reagent. Following incubation for 4 h, cells were solubilized with 95% dimethyl sulfoxide and the absorbance of the solution was measured at 590 nm using an ELISA plate reader. Each sample well was assessed in triplicate, and all assays were run in triplicate. Statistical analysis was performed using one-way analysis of variance and Tukey's post hoc test. P<0.05 was considered to indicate a statistically significant difference. Data were presented as the mean ± standard error of the mean.

RNA extraction

Total RNA was isolated from D407 cells treated without clusterin (0 µg/µl; control) or 1.0 µg/µl clusterin using TRIzol reagent (Invitrogen; Thermo Fisher Scientific, Inc.) according to the manufacturer's instructions. The extracted RNA sample was treated with DNase I (Invitrogen; Thermo Fisher Scientific, Inc.) to remove the DNA. The RNA quality was analyzed using a NanoDrop spectrophotometer (Thermo Fisher Scientific, Inc.) and 10% gel electrophoresis. RNA samples were stored at −80°C for sequencing library construction.

Construction of sequencing library and RNA-sequencing (Seq)

An mRNA library was constructed according to the manufacturer's instructions using TruSeq Stranded mRNA Library Prep kit (Illumina, Inc., San Diego, CA, USA). Briefly, mRNA was enriched using oligo (dT) magnetic beads and then broken to ~200 bp fragments. The first-strand cDNA was synthesized using random hexamer primers, and then the second strand was synthesized. The purified DNA fragments were then sequenced using an Illumina Hiseq 2000 (Illumina, Inc.) by Guangzhou RiboBio Co., Ltd. (Guangzhou, China). The small RNA library was constructed according to Illumina's recommended protocols and sequenced on a Hiseq 2000 platform at RiboBio Co., Ltd. All sequences were deposited onto the GEO database (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE87167).

Data analysis

Sequence reads were aligned to the hg19 human genome using Bowtie (25) with Tophat (26). Only reads that had ≤2 mismatches and 2 gaps to the hg19 genome were accepted. To assess the transcript level in the sample (slug-ml-3) and control (NC-3), the reads per kilobase of transcript per million mapped reads value was calculated. The number of mapped reads was calculated for each miRNA and the read count of each unique sequence was normalized using the reads per million (RPM) normalization, according to the formula: RPM = (number of reads mapping to miRNA reference / total number of mapped reads) × 106. Differential expression analysis was performed using Cuffdiff (26). Protein-coding transcripts and lncRNAs with fold change (FC)>1.5 or FC<0.67 and P<0.05 were considered to be differentially expressed in D407 cells treated with clusterin, whereas miRNAs with FC>2 or FC<0.5 (P<0.05) were considered aberrantly expressed.

Each sequence was annotated by aligning to known RNA sequences downloaded from Rfam11.0 (http://rfam.xfam.org/; rRNA, snRNA and snoRNA), miRBase (http://www.mirbase.org/release version 21.0; miRNA), UCSC (http://genome.ucsc.edu; tRNA) and pirnabank (http://pirnabank.ibab.ac.in/; piRNA). Novel miRNAs were predicted using the miREAP algorithm (27) according to the manual. RNA secondary structures were predicted using RNAfold (28).

Functional annotation of differentially expressed mRNAs

Gene Ontology (GO) (29) and Kyoto Encyclopedia of Genes and Genomes (KEGG) (30) pathway analysis of differentially expressed mRNAs was performed using the database for annotation, visualization and integrated discovery (DAVID, version 6.7) (31). The obtained GO terms and KEGG categories were filtered at a threshold significance value of P<0.05.

Construction of ceRNA network

The ceRNA network was constructed and visualized using Cytoscape software as preciously described (32). The process included prediction of lncRNA-miRNA interactions by PITA (http://genie.weizmann.ac.il/pubs/mir07/mir07_prediction.html) (33), RNAhybrid (http://bibiserv.techfak.uni-bielefeld.de/rnahybrid/) (34) and miRanda (http://www.microrna.org) (35), targeting of mRNAs by miRNAs based on experimental data using miRTarBase (http://www.microrna.gr/tarbase) (36), miRwalk (http://mirwalk.uni-hd.de/) (37) and miRanda, and filtration of specific miRNA-lncRNA and miRNA-mRNA pairs, where the interaction pairs containing differentially expressed miRNA and lncRNA or mRNA were retained. A network map was constructed and visualized using Cytoscape software v3.5.1 (32).

Results

Effect of clusterin on cell viability

In consideration of the abnormal secretion of clusterin in RPE cells from AMD donors (38), clusterin activity on the viability of D407 cell viability was investigated using MTT assay. The viability of D407 cells was not affected by clusterin at tested concentrations (Fig. 1).

Figure 1. Viability of clusterin-treated D407 cells determined by MTT assay. Data are presented as the mean of three independent experiments. The bars represent standard deviation. OD, optical density.

RNA profiling in RPE cells

Total RNA isolated from D407 cells treated or untreated with clusterin was sequenced. After filtering out reads without a 3′, 5′ adaptor sequence and poly-A reads, the slug-ml-s and NC-3 mRNA libraries had 12,378,042 and 12,385,191 mapped reads, respectively (Table I). Notably, the distribution of the mapped reads in the experimental and control groups were similar between control and clusterin-treated RPE cells. The small RNA sequences were annotated according to their overlap with the sequences of known RNAs. The annotated results suggested that small RNA libraries from slug-ml-3 and NC-3 contained the same classes of RNAs (Table II). Furthermore, the abundance of each class was similar between the two libraries.

Table I. The distribution of mapped reads in experimental and control groups.

Table I.

The distribution of mapped reads in experimental and control groups.

Region	Slug-ml-3 mapped reads	NC-3 mapped reads
All	12,378,042	12,385,191
Upstream/downstream	171,120	160,934
Exonic	66,229	80,603
Intergenic	3,191,258	2,709,256
Intronic	1,116,005	1,116,007
NcRNA_exonic	7,642,921	8,092,812
NcRNA_intronic	110,292	134,888
Splicing	2,720	2,856
UTR	77,497	87,835

[i] Nc RNA, non-coding RNA; UTR, untranslated region.

Table II. Annotation of noncoding RNAs.

Table II.

Annotation of noncoding RNAs.

	Slug-ml-3		NC-3
Type	Total sRNA	%	Total sRNA	%
All	13,393,799	100.0	13,233,217	100.0
miRNA	5,547,804	41.4	6,436,571	48.7
tRNA	472,065	3.5	452,319	3.4
rRNA	1,648,269	12.3	1,417,235	10.7
snRNA	621,981	4.6	435,469	3.3
snoRNA	423,883	3.2	440,867	3.3
piRNA	504,767	3.8	451,597	3.4
Y_RNA	126,656	1.0	126,592	1.0
Others	4,048,374	30.2	3,472,567	26.2

[i] sRNA, small non-coding RNA; miRNA, micro RNA; tRNA, transfer RNA; rRNA, ribosomal RNA; snRNA, small nuclear RNA; snoRNA, small nucleolar RNA; piRNA, Piwi-interacting RNA; slug-ml-3, experimental group; NC-3, control group.

Identification of differentially expressed coding and non-coding genes

The expression levels of differentially expressed mRNA, miRNA and lncRNA were further analyzed between slug-ml-3 and NC-3 cells. A total of 321 genes were significantly differentially expressed, including 71 up- and 250 downregulated genes. Among them, 67 important genes are presented in Fig. 2A.

Figure 2. Hierarchical clustering analysis using Genesis clustering software of differentially expressed genes and miRNAs. Differentially expressed (A) genes and (B) miRNAs identified from human retinal pigment epithelial cell line D407 treated with or without clusterin. Slug-ml-3, experimental group; NC-3, control group; miRNA, microRNA.

Aberrantly expressed miRNAs were detected based on the threshold of FC>2 or FC<0.5, and P<0.05. When miRNA expression was compared between D407 cells treated with and without clusterin, 36 miRNAs were observed as aberrantly expressed (Fig. 2B). To identify novel miRNAs from the reads mapped to unannotated genomic regions, the described miRNA discovery algorithm, miREAP, which is a tool for discovering novel miRNAs from small RNA deep sequencing data, was employed. A total of 71 and 76 candidate miRNAs were identified and considered to be novel in slug-ml-3 and NC-3, respectively.

Differentially expressed lncRNAs were also detected between slug-ml-3 and NC-3 cells. In total, 296 previously annotated lncRNAs and 90 novel lncRNAs were aberrantly regulated in cells treated with clusterin.

ceRNA network generation

miRNAs interact with lncRNAs via miRNA response elements within the ceRNA network. Therefore, potential miRNA response elements in lncRNAs were investigated. The results indicated that 22 of 36 specific miRNAs may interact with 75 of 386 lncRNAs (data not shown). To establish a lncRNA-miRNA-mRNA network (ceRNA network), mRNAs that were targeted by miRNAs were investigated. Using the list of miRNAs that may interact with lncRNAs, miRNA targeting mRNAs were searched for using miRTarBase, miRwalk and miRanda. From this analysis, 23 miRNAs were identified (data not shown). Most of the targets of these miRNAs were AMD-associated genes such as engrailed homeobox (EN)1, EN2, endothelin 2, myosin regulatory light chain interacting protein and complement component 5a receptor 2 (C5AR2). Based on these data, a ceRNA network (Fig. 3) was established; 75 lncRNAs and 32 miRNAs were involved in the proposed ceRNA network.

Figure 3. ceRNA network in retinal pigment epithelial D407 cells. Blue circular nodes represent long-noncoding RNAs, orange squares represent mRNA and red inverted triangles represent miRNA. ceRNA, competing endogenous RNA.

Functional analysis of differentially expressed mRNAs

In order to elucidate the molecular model of action of clusterin in AMD; biological processes and pathways regulated by clusterin were analyzed using DAVID. Enriched GO terms including Biological Process, Cellular Component and Molecular Function annotations were assessed. The results indicated that differentially expressed mRNAs were predominantly associated with the regulation of lipid transport across the blood brain barrier, receptor catabolic process, postsynaptic membrane organization and phospholipid transport (Fig. 4A and B). To understand the signaling pathways involved in the ceRNA network, the mRNAs were analyzed using the DAVID algorithm. The principally enriched pathways were environmental information processing, genetic information processing, cellular processing, organismal systems, drug development, human diseases and metabolism (Fig. 4C and D). The results indicated that differentially expressed mRNAs were enriched in the adherens junction and cell adhesion molecules in addition to extracellular matrix-receptor interactions. Furthermore, these mRNAs were largely involved in diseases including small cell lung cancer, arrhythmogenic right ventricular cardiomyopathy and hypertrophic cardiomyopathy.

Figure 4. GO enrichment analysis of differentially expressed genes. (A and B) GO analysis and (C and D) KEGG pathway enrichment analysis of differentially expressed genes identified from human retinal pigment epithelial cell line D407 treated with or without clusterin. GO, gene ontology; BP, biological process; CC, cellular component; MF, molecular function; KEGG, Kyoto Encyclopedia of Genes and Genomes; EIP, environmental information processing; GIP, genetic information processing; CP, cellular processes; HD, human diseases; M, metabolism; OS, organismal systems. ***P<0.001; **P<0.01; *P<0.05.

Discussion

lncRNAs are considered as important regulators of biological processes in human diseases (39,40). However, few studies have reported the lncRNAs expression profile in retinal disorder-associated diseases including AMD, using a high-throughput sequencing strategy (41). Increasing evidence has demonstrated that there are interactions between mRNA and lncRNA (42) or lncRNA and miRNA (43,44) in cancer development, suggesting that lncRNAs may function as a part of ceRNA network. However, ceRNA networks in AMD are still poorly explored. The present study served to identify miRNA and lncRNA expression and distribution in RPE cells following treatment with clusterin, based on genome-wide expression profiling. Furthermore, a ceRNA network was constructed among mRNAs, miRNAs and lncRNAs, to provide a systematic view of lncRNA-miRNA-mRNA interactions.

Based on next generation RNA sequence data, 386 lncRNAs were observed to be differentially expressed between RPE cells treated with or without clusterin. Abnormally expressed miRNAs and mRNAs were also identified, and 75 lncRNAs were identified to be involved in the ceRNA network, suggesting they may serve role in AMD development. Among the lncRNAs involved in the ceRNA network, NR_036461 (HIST2H2BC) was reported to promote cell proliferation and invasion in triple-negative breast cancer (45). Furthermore, NR_102376 (small integral membrane protein 27) was identified to be a potential biomarker for human breast cancer (46). From these results, it was concluded that there was a differential expression pattern of certain lncRNAs between RPE cells with and without clusterin treatment. Further studies are required to elucidate the role of these differentially expressed lncRNAs.

To improve the predicted network, three databases including PITA, RNAhybrid and miRanda were applied to filter the pair-wised relationships based on lncRNA-miRNA expression correlations. This strategy served to identify additional lncRNAs, compared with using a single database. To enhance the reliability of the data, validated miRNA target genes were acquired from the miRTarBase database that had associated experimental supporting data.

Several genes involved in inflammation and cell adhesion were identified to participate in the ceRNA network after treating RPE cells with clusterin. For instance, engrailed (EN)1, EN2, endothelin 2, myosin regulatory light chain interacting protein, complement component 5a receptor 2 (C5AR2), lncRNA NR_036461 (HIST2H2BC) and NR_102376 (TOPORS-AS1) miRNA hsa-miR-1273g-3p, hsa-miR-5582-3p and hsa-miR-1293 were identified to be associated with AMD. Moreover, we found that hsa-miR-1273g-3p was closely associated with inflammation. C5AR2, which is mediated by hsa-miR-1273g-3p, has been indicated to be involved in C5a-mediated human mast cell adhesion and proinflammatory mediator production (47). Inflammation is well recognized as a molecular mechanism of AMD (48). Therefore, differentially expressed lncRNAs may serve important roles in AMD development.

In conclusion, the present study identified lncRNAs specifically induced by clusterin in RPE cells and identified an abnormal expression pattern of these lncRNAs. A ceRNA network was constructed that may provide a novel method to identify AMD-related lncRNA alterations in AMD. These results suggest that lncRNAs specifically identified in RPE cells may participate in a complex ceRNA network and in unknown regulatory pathways in the network, and may contribute to our understanding of the biological mechanism underlying AMD.

Acknowledgements

The current study was supported by the China Social Assistance Fund (grant no. BJ-LM2015001 J).

References