RWR-algorithm-based dissection of microRNA-506-3p and microRNA-140-5p as radiosensitive biomarkers in colorectal cancer

Radiotherapy resistance is one of the main causes for treatment failure in colorectal cancer (CRC), and it is associated with the deregulation of certain microRNAs. In this study, we constructed the microRNA-mRNA network consisting of 2275 microRNAs and 7045 target genes, collected the known microRNAs related to CRC-radiosensitivity (CRCR) (n=18) as the seed nodes, and applied the algorithm of random walk with restart (RWR) to the network to identify novel CRCR-related microRNAs (n=263). In functional analysis, 263 novel microRNAs shared a high proportion of the same biological processes and pathways with the known microRNAs. In topological analysis of the sub-network of the 263 microRNAs and their targets, hsa-mir-506-3p and hsa-mir-140-5p were identified as network hub nodes. In plasma, radiosensitive patients had a higher expression level of hsa-mir-506-3p and hsa-mir-140-5p than radioresistant patients. In experimental validation, both hsa-mir-506-3p and hsa-mir-140-5p over-expression could obviously decrease the cell proliferation, survival rate and colonality in CRC cells after radiation. In conclusion, this study combined the novel network-based method with experimental validation, and identified two novel radiosensitive biomarkers of hsa-mir-506-3p and hsa-mir-140-5p in CRC.


INTRODUCTION
Colorectal cancer (CRC) is one of the most common cancers worldwide, with an estimated 1.4 million new cases and 693,900 deaths per year [1]. Currently, surgical resection remains the only curative treatment for CRC. However, about 20% to 40% CRC patients are initially diagnosed at a locally advanced, unresectable, nonmetastatic stage termed "locally advanced CRC" [2]. Neoadjuvant chemoradiation is the standard therapeutic strategy for these patients. Unfortunately, many CRC patients are radioresistant and have a poor prognosis [3]. Therefore, it is necessary to elucidate the underlying radioresistant mechanisms in CRC, which may ultimately improve the therapeutic outcomes.
MicroRNAs are a class of small non-coding RNAs containing about 18~24 nucleotides. MicoRNAs function as post-transcriptional regulators, and usually cause translational suppression or RNA degradation via specific binding to 3′-untranslated region (3′-UTR) of their target genes [4]. Accumulating evidence has suggested a strong association between microRNAs deregulation and tumor radioresistance. For example, microRNA-21 expression could promote the radioresistance in a variety of cancer cell lines, including breast, lung, glioblastoma and nasopharyngeal cancers [3,[5][6][7][8]. Several microRNAs were also found dysregulated in radioresistant CRC cell lines, and upregulation or downregulation of these microRNAs could protect the cells from radiation effects. However, AGING more novel microRNAs biomarkers are needed to illuminate the underlying mechanism of CRC radioresistance.
In recent years, studies have investigated the mechanism of several diseases with the help of advanced computational methods, such as the networkbased method. Some studies adopted the gene-gene interaction network, which allowed us to identify novel disease genes based on known disease genes [9]. In this study, we adopted a classic network-based algorithm of random walk with restart (RWR), and developed a three-stage filtration strategy to identify novel CRCradiosensitivity (CRCR)-related microRNAs based on a microRNA-mRNA network and known CRCR-related microRNAs. For the network hub microRNAs, we also conducted an experimental validation.

CRCR-related microRNAs identified by the RWR algorithm
Eighteen microRNAs were collected by literature review, which has been experimentally validated in association with CRCR (Table 1). By integrating three microRNA databases, a microRNA-mRNA network was constructed consisting of 2275 microRNAs and 7045 target genes. Then, we selecting those 18 microRNAs as seed nodes, and conducted the RWR algorithm analysis on the network (Figure 1). After permutation test, 263 novel CRCR-related microRNAs were identified (Supplementary Table 1). There were 1386 common targets between the 18 microRNAs and 263 microRNAs ( Figure 2).

Functional enrichment analyses
Functional enrichment analyses were conducted on the targets of the 18 microRNAs and 263 novel microRNAs respectively. In biological processes, there was a high overlapped rate of 70% between the targets of the two groups ( Figure 3). Most of these biological processes focused on the biosynthesis and of RNA and protein, and cell proliferation. In pathways, there was a relatively high overlapped rate of 44% between the targets of the two groups. Most of these pathways focused on the tumor related signaling pathways.

Topological analysis of the sub-network of 263 microRNAs and their targets
The topological features of the sub-network of 263 microRNAs and their targets were evaluated by the degree, betweenness and closeness centrality. We listed the overlap of top 10 microRNAs with topological features in each dimension ( Figure 4). Two microRNAs (hsa-mir-506-3p and hsa-mir-140-5p) were identified as hub nodes in the network (Table 2).

DISCUSSION
In this study, we built a microRNA-mRNA network based on the experimental validation computational prediction databases. Eighteen known CRCR-related microRNAs were obtained from the literature and chose as seed nodes, and then we applied the network-based RWR algorithm to identify novel CRCR-related microRNAs. Finally, 263 novel microRNAs were identified. Several microRNAs have been validated in association with the radiosensitivity in multiple cancers. For example, has-mir-339-5p could increase the radiosensitivity of lung cancer cells by targeting phosphatases of regenerating liver-1 (PRL-1) [10]. Hypoxia-responsive has-mir-301a-5p and has-mir-301b-5p promoted radioresistance of prostate cancer cells via down-regulating NDRG2 [11]. Hsa-mir-99a-5p enhanced the radiation sensitivity of non-small cell lung cancer by targeting mTOR, and inhibition of the glucocorticoid receptor resulted in an enhanced hsa-mir-99a-5p-mediated radiation response in stem-like cells from human prostate cancers [12,13].

AGING
To validate the robustness of the prediction, we adopted two different methods. First, functional analysis was conducted respectively on the targets of the 18 microRNAs and 263 microRNAs. We found a significant overlap between the two groups, especially like the biological processes of the regulation of transcription, biosynthesis, cell proliferation and programmed cell death, as well as the pathways of MAPK signaling, PI3K-AKT signaling and TGF-β receptor signaling. These results indicated a similar role of the 263 microRNAs in the radioresistance of CRC, just like the 18 microRNAs.
To further validate our results, we also extracted the sub-network of the 263 microRNAs and their targets. After topological analysis of the sub-network, we found two hub nodes of hsa-mir-506-3p and hsa-mir-140-5p. Plasma samples of 18 CRC patients were collected to investigate the association between expression levels of the two microRNAs and subsequent radiosensitivity.   Finally, we found radiosensitive patients had a significantly higher expression of both hsa-mir-506-3p and hsa-mir-140-5p than radioresistant patients. Both hsa-mir-506-3p and hsa-mir-140-5p over-expression could decrease the cell proliferation, survival rate and clonality in CRC cells after radiation. Hsa-mir-506-3p has been found in association with the progression and prognosis of non-small lung cancer and esophageal squamous cell cancer [14,15]. Hsa-mir-140-5p could attenuate chemotherapeutic drug-induced cell death by regulating autophagy through IP3K2 in human osteosarcoma cells, and hsa-mir-140-5p could also play a therapeutic role for the treatment of non-small cell lung cancer [16,17]. These indicated the accuracy and efficacy of our prediction. However, well-designed experiments were also needed to validate our results.
As a classical method, RWR algorithm was not as complicated as the methods in some studies mainly focusing on bioinformatics analyses. Nevertheless, we made further innovations in this study. First, we constructed the microRNA network through the bridge of microRNA-mRNA interaction data. Most of previous studies focused on the application of RWR algorithm to the mRNA network, and identified mRNA or protein biomarkers. Second, our study focused on not only the identification of novel CRC-radiosensitive microRNAs, but also the experimental validation. This combined method could enhance the reliability of bioinformatics analysis, especially in the subsequent clinical practice.
In conclusion, this study combined the novel networkbased method with experimental validation, and identified two novel radiosensitive biomarkers of hsamir-506-3p and hsa-mir-140-5p in CRC.

Random walk with restart (RWR) algorithm analysis
As a classic ranking algorithm, the RWR algorithm always simulates a random walker starting from a seed node or several seed nodes and walking on a certain network [18]. Finally, possible novel nodes are identified and ranked from high to low probabilities. The algorithm has been adopted to search novel disease genes or other related problems [19][20][21].
After the seed nodes are selected, the initial probability P0 for each seed node is set as 1/N (where N was the number of seed nodes), while zero for non-seed nodes. Then, the RWR algorithm simulates a random walker that moved on certain network starting from seed nodes. For formulation, let Pi denote a vector representing the probability of each node after the ith moving procedure is complete. After each moving procedure, Pi is updated as follows: where M is the column-wise normalized adjacency matrix of the network; r is the restart probability of returning to the seed nodes at every step (r was set to 0.8 in this study to indicate the importance of the known microRNAs). When the L1-norm of the difference between two successive vectors is less than 1×10 6 , the vector becomes stable, the RWR algorithm stops and output Pi+1 as the final vector. Each component in this vector indicates the probability of a node being traitrelated. To remove the false positives, we also conducted a permutation test (1000 iterations). A false discovery rate (FDR) of less than 0.05 is considered significant.

Identification of candidate microRNAs
In this study, we adopted the RWR algorithm in the microRNA-mRNA network to identify candidate CRCR-related microRNAs. First, we reviewed the databases of PubMed and Embase, and chose certain microRNAs as the seed nodes. The microRNAs were included if meeting the following criteria: (i) reported in association with CRCR; (ii) conducted an experimental validation; (iii) availability of the experiment details. We excluded the microRNAs without experimental validation, or only detected by bioinformatics analysis.
The microRNA-mRNA network was constructed according to the interaction data from two computational prediction databases of miRDB (http://mirdb.org/) and TargetScan (http://www.targetscan.org/), and the experimental validation database of miRTarBase (http://mirtarbase.cuhk.edu.cn/php/index.php). To obtain more reliable results, we only included the interactions overlapped across the databases.
Then, the sub-network of candidate microRNAs and their targets were extracted from the microRNA-mRNA network. The network degree, betweenness and AGING closeness centrality were calculated respectively to analyze topological features of the sub-network [22]. The overlapped microRNAs with top 10 topological features in each dimension were regarded as novel CRCR biomarkers for experimental validation.

Functional enrichment analyses
To investigate the potential function of novel microRNAs, gene ontology (GO) analyses of biological processes and pathways are performed using the online tool ToppGene (https:// toppgene.cchmc.org/). A FDR of less than 0.05 is chosen as the cut-off criteria.

Plasma samples and cell culture
The plasma samples were obtained from 18 CRC patients who subsequently underwent radiotherapy at Renmin Hospital of Wuhan University, from September to October in 2018. All patients received no other therapies before radiotherapy, and were provided written informed consent before sample collection. The present study protocol was approved by the ethics committee of Renmin Hospital of Wuhan University. The radiotherapy efficacy was classified into sensitive (n=10) and resistant (n=8) groups according to response evaluation criteria in solid tumors (RECIST).

Colony forming assay
After transfection, cells were irradiated at the doses of 2, 4, and 6Gy using 250kV X-rays (Philips, the Netherlands). Then, 1000 cells were incubated in 6-well plate for 2 weeks to form visualized colonies of survived cells. Finally, cells were fixed with formaldehyde, stained with 0.1% crystal violet and air-dried. The number of clones was quantified. Plating efficiency was calculated as the ratio of the visible colonies from all seeded cells. The survival fraction was determined by the percentage of irradiated cells compared with the control.

Cell proliferation assay
After transfection and irradiation, cells were seeded in 96-well plate with 1×10 5 /well and cultured for 24, 48 and 72h. Then, 10μl CCK8 solution (Beyotime, China) was added into each well, and the absorbance at 450nm was measured after incubating for 2h.

Statistical analysis
All statistical analyses are performed using SPSS 20.0 (IBM, USA) and GraphPad Prism 7.0 (GraphPad, UAS). Data are presented as mean ± SD, and t test and one-way ANOVA are used to compare two or multiple groups. Receiver operating characteristic (ROC) curve analysis is conducted to evaluate the diagnostic ability of microRNAs. P<0.05 is considered statistically significant.

AUTHOR CONTRUBUTIONS
FL and WD conceived and designed the bioinformatics analysis. FL and XC conducted the experimental validation. All authors contributed to the discussion. FL, XC, PP and WD contributed to the drafting of the article. All authors have revised the manuscript critically in terms of its important intellectual content and have approved the final version to be published. FL, XC and WD are responsible for the integrity of the work as a whole.