Screening of key genes and miRNAs involved in age-related hearing loss


 Background

Age-related hearing loss (ARHL) is the most common sensory deficit and refers to the gradual loss of hearing function with age. We aimed to research the differential expression genes during the occur of ARHL and explore microRNAs that maybe regulate these genes.
Methods

Search the GEO data GSE6045, GSE35234, GSE62173, GSE45026 from NCBI's Gene Expression Omnibus database, and analyze in R. Normalize GEO data by RMA method in R, and use linear microarray data model (LIMMA) method in R to compare young and old mice to find differential genes and microRNAs.
Results

109 up-regulated and 121 down-regulated important genes were identified respectively. Functional enrichment shows that they are significantly enriched in protein digestion and absorption, neuroactive ligand-receptor interaction, and PI3K-Akt signaling pathway. Among the top 20 Hub genes, 7 down-regulated genes (Col3a1, Col1a2, Sparc, Col4a2, Col2a1, Lox, Sparc, Ctgf) have verified targeted miRNAs, which have interaction with differential miRNAs of GSE45026. Finally, we designated miR-29a-3p, miR-29b-3p, miR-29c-3p, and miR-124-3p as key miRNAs involved in the development of age-related hearing loss.
Conclusions

These low expression genes of Col3a1, Col1a2, Sparc, Col4a2, Col2a1,Lox, Sparc, and Ctgf maybe key genes of ARHL, and probably regulated by miR-29a-3p, miR-29b-3p, miR-29c-3p, and miR-124-3p.

Compared with congenital and early onset hearing loss, our understanding of the biochemical process and molecular biology in this case is more limited. In order to explore its molecular mechanism, some scholars have used age-related hearing loss model mice to conduct high-throughput microarray platform to detect gene expression changes during the occurrence of age-related hearing loss [12,13].
High-throughput microarray platform is a new tool for detecting genetic changes and discovering biomarkers for diseases [14], in this way, we can test the change of gene and other little biomarkers. However, due to the insu cient number of samples, individual microarray studies often show a bias towards the identi cation of high-abundance molecules, so the results are often biased [15]. But by integrating multiple microarray data sets, su cient samples can be provided to reduce this bias and produce more convincing results.
In this study, we obtained some abnormally expressed genes by mining three GEO datasets. Subsequently, we performed functional enrichment analysis on these abnormally expressed genes. Then the miRNA upstream of the hub gene was predicted and veri ed in the GSE45026 dataset. Finally, we found some differentially expressed genes and miRNAs that were never thought to be related to hearing loss. These markers may be used as potential targets for age-related hearing loss detection in the future.

Methods
Microarray research, data sets and sample characteristics in the GEO database We query the GEO (http://www.ncbi.nlm.nih.gov/geo/) gene expression database for gene chips about agerelated hearing loss, which collects submitted high-throughput gene expression data. According to the following standards, the chip is considered to meet our research requirements: (1) Research about agerelated hearing loss. (2) The mouse model that used has been proved will occur age-related hearing loss.
According to the above standards, 4 microarray data including GSE6045, GSE35234, GSE62173, and GSE45026 were collected from the GEO database. Among them, GSE6045, GSE35234, GSE62173 analysis results are differential genes, and GSE45026 analysis results are differential miRNAs. The datasets were subjected to principal component analysis (PCA) for dimensionality reduction and quality control. If the quality of a particular sample is not good enough, it is excluded for subsequent analysis. Details of each microarray chip are provided in Table 1.

Differential expression analysis
Since the datasets we collected from different platforms with different handling, so we used the RMA[16] method in R to standardize the data rstly, then the unnormalized original data was summarized in the form of a matrix. Next, the linear microarray data model (LIMMA) [17] method was used in R to compare the young and old mouse to nd different genes and different miRNAs, and it is considered that there is a signi cant difference in genes when P.Value < 0.05 and | logFC | > 0.5.

GO and KEGG pathway analysis
For the DEGs, we used clusterPro ler[18] package to perform go and KEGG pathway enrichment analysis in R to nd all the potential functions of DEGs in the network (P <0.05).

Hub gene screening
Search for DEGs from the recognized STRING 10.0 (search tool for searching for interacting genes; https://string-db.org/) database and download PPI pairs, then import the PPI data into Cytoscap [19] to obtain the entire PPI network, and nally, use the CytohHubba [20] plug-in of Cytoscap software to screen out 20 hub genes.
Prediction and veri cation of miRNAs targeting hub genes First, we used the default important parameters of the miRTarBase database (http://mirtarbase.mbc.nctu.edu.tw/php/search.php) to predict the miRNA targeting the hub genes and then veri ed the obtained miRNAs in the differential miRNAs of GSE45026.

Principal component analysis verifying independence of each group
To distinguish the signi cant differences between young and old GEO data samples, we performed PCA to reduce the dimensions and evaluate the independence of each group (Fig 1). The results show that there are signi cant differences between young and old samples in the data sets (GSE6045, GSE62173, GSE35234, GSE45026).

Differential expression analysis
Because there may be errors in the results of a single experiment, it is necessary to nd several pieces of evidence to support, to increase the accuracy of the results. After standardizing the data using the RMA [16] method, the limma [17] package was used to identify differentially expressed genes according to thresholds of P.Value < 0.05 and | logFC | > 0. 5

Predict miRNAs targeting the hub genes and verify in GSE45026
We used miRTarbase (http://mirtarbase.mbc.nctu.edu.tw/php/search.php) to predict miRNAs targeting hub genes, and found that only 7 down-regulation hub genes (Col3a1, Col1a2, Sparc, Col4a2, Col2a1,Lox, Sparc, Ctgf) had results( gure 6A). Then we veri ed these miRNAs in the miRNAs differentially expressed by GSE45026 and found that miR-29a-3p, miR-29b-3p, miR-29c-3p and miR-124-3p were veri ed in GSE45026( gure 5B). So we highly suspect that miR-29a-3p, miR-29b-3p, miR-29c-3p and miR-124-3p are important molecules involved in the regulation of age-related hearing loss. Discuss ARHL commonly occurs in old age and brings a bad experience to life. In addition to affecting hearing, some studies have shown a signi cant correlation between hearing sensitivity and the incidence of Alzheimer's disease [21,22],as well as the difference between hearing sensitivity and cognitive function among non-demented individuals [23]. Due to the lack of good phenotyping methods and su cient sample size, several published GWAS data on ARHL have not found reproducible results [24]. We are still unclear about the molecular mechanism of ARHL.
Because of the di culties in human cohort studies, researchers began to use animal models to help determine the pathogenesis and genetics associated with ARHL. Mice are the main model organisms for studying the auditory function and aging of mammals [25]. For a long time, it has been reported that certain strains have good hearing for the elderly (e.g. CAST, CBA/CaJ, CBA/J, C3H/HeH), while others show a gradual decline in auditory function (e.g. BALB, C57BL/6, DBA/ 2J). The GSE datasets selected in this study are all mouse strains whose hearing loss occurs with age.
When conducting GO analysis on up-regulated genes, we found that they mainly focused on defense response to bacterium, ligand − gated ion channel activity, and s receptor complex. Previous studies have shown that hearing loss is mainly due to the loss of internal or external hair cells, which damages IHC ribbon synapses or spiral ganglion neurons [27]. It now appears that the biological processes of gene enrichment, cell composition, and molecular function are consistent with the occurrence of ARHL.
To systematically analyze the relationship and function of the important DEG in ARHL, we mapped DEG to STRING database and obtained PPI network. As we all know, genes with higher node degree in PPI network usually play more roles. To further identify the key genes in ARHL, we selected the top ten up-regulated and down-regulated hub genes for further analysis. The analysis results showed that seven down-regulated genes (Col3a1, Col1a2, Sparc, Col4a2, Col2a1,Lox, Sparc, Ctgf) may be the key genes of ARHL.
All in all, our study analyzed the GEO database data and found differential genes for age-related hearing loss, and found some key miRNA molecules, which may be potential biological targets for predicting agerelated hearing loss.