miRCurator: A rapid curation and analysis tool for plant microRNA studies

Abstract

Plant microRNAs (miRNAs) are tiny, non-coding RNAs that regulate the biological pathways either by inducing translational repression or messenger RNA decay. The advent of sequencing technologies increases the number of biological tools and computational methods to identify miRNAs with their targets inside the genome and transcriptome. In this study, we present a rapid curation software tool, the so-called miRCurator, which filters out unwanted microRNA candidates based on three significant criteria: (1) the pre-miRNA sequence should be deprived of multi-branched loops; (2) less than four mismatches on between mature miRNA and miRNA* strand are allowed; (3) no mismatches at DICER cut points are permitted on a mature miRNA strand. miRCurator is a stand-alone user-friendly package that does not rely on cutting-edge computing platforms. We have examined our tool on different organisms, which are shown as case studies, and it was able to successfully identify all the unwanted miRNA candidates. The developed tool helps to identify genuine miRNAs and provides reliable outcomes in an automated fashion by eliminating false-positive predictions.

1. Introduction

RNA silencing is an evolutionarily conserved mechanism in eukaryotes, which assists plants to carry out different functional processes in the organism. This mechanism is induced by double-stranded RNA (dsRNA) or hairpin structured RNA (hpRNA), consisting of common elements including DICER or DICER-like (DCL), and Argonaute (AGO) family proteins. RNA silencing in plants can be classified into four distinct pathways: microRNA (miRNA) pathway, trans-acting small interfering RNA (tasiRNA) pathway, RNA-directed DNA methylation pathway, and exogenic RNA silencing pathway. Plants have evolved multiple RNA silencing factors for diverse RNA silencing pathways. For instance, the model plant Arabidopsis encodes four DCLs, six RNA-dependent RNA polymerases (RDRs), and ten AGOs, plus several other factors [1]. RNA silencing mechanism of the plants is also used in biotechnology to increase disease resistance, improve the commercial traits of the plants, change their flowering time, and develop high-value industrial products [1].

In this study, we mainly focus on the microRNA elements and their identification methodologies in the genome and transcriptome. Plant microRNAs are tiny (i.e., ~ 21 nucleotide) endogenous and non-coding RNAs, which regulate the gene expression either by inducing translational repression or messenger RNA decay. Plant miRNAs are generated from specific stem regions of single-stranded hairpin precursors, which is one of the distinct features from other small RNAs. The abundance of intracellular miRNA is regulated under multiple levels of control mechanism including transcription, processing, RNA modification, RNA-induced silencing complex (RISC) assembly, miRNA-target interaction, and turnover [2]. With the emergence of next-generation sequencing technologies, new computational methods and approaches are being developed to analyze the big data resulted from sequence information. These new methodologies provide better economy and more information accuracy to the users in a shorter computation time. Similar computational tools are also produced for plant miRNA identification from genomic and/or transcriptomic data, such as miRkwood [3], miRPlant [4], C-mii [5]. Some of these tools are available on the web to offer remote access to a broader range of users.

Other computational approaches benefit from the homology-based prediction for miRNA identification. The homology-based conserved approach is implemented in SUmiRFind and SUmiRFold Perl scripts [6]. For the homology-conservation method, the reference miRNA sequences should be selected from the current release of miRBase [7], which is the most detailed and updated database. SUmiRFind exploits BLASTN algorithm [8] with parameters, and records the important hits in a file. SUmiRFold script creates the secondary structure of these hit sequences by folding them using the UNAFold, an implementation of Zuker algorithm [9]. The output file from the SUmiRFold script is in a tabular format, which includes the following information for each putative miRNA: The new miRNA sequence, conserved miRNA sequence, pre-miRNA sequence, sequence ID, mature miRNA length, pre-miRNA length, number of mismatches to the query, pre-miRNA stem-loop start and end sites, hairpin location, MFE (ΔGkcal/mol), %GC content and MFEI.

This output file should be re-checked and some of the candidates may be removed by the users manually since there might be some miRNA candidates in the resulted file that do not satisfy these criteria: (1) the pre-miRNA sequence should not consist of multibranched loops; (2) less than four mismatches on between mature miRNA and miRNA* strand are allowed [10]; (3) no mismatches at DICER cut points are permitted on the mature miRNA strand. Nonetheless, the output file may have thousands of miRNA candidates based on the size of the source file; therefore, a manual quality control procedure is time consuming for the users. To ease the process, we developed a script, the so-called miRCurator, that eliminates these improper miRNAs from the SUmiRFold output file. Our script is a stand-alone user-friendly package that rapidly omits false-positive miRNA candidates, and generates report in both tabular and visual formats. Our previous studies have been subjected to miRCurator as case studies for the curation of miRNAs from the kiwifruit genome [11], hazelnut genome [12,13], spinach transcriptome [14], fabaceae family members [15], and medicinal plant transcriptome [16,17].

2. Materials and Methods

In this section, we provide the detail of our script in two subsections: In Section 2.1, the requested data structure is explained, by which the information is sent to be processed. In Section 2.2, we explain the developed tool in detail.

2.1. Preparation

Before all the steps, SUmiRFind, SUmiRFold and miRCurator, contigs (assembled raw reads) should be prepared as inputs for the miRNA identification in trancriptomic data, whereas, in genomic data, raw reads may be used if they are longer than 200 nt. The methodology for miRNA identification using homology conservation approach is summarized in Figure 1.

Figure 1.

An overview of miRNA identification method.

As shown in Figure 1, the input data (e.g., whole genome assembly, long raw genomic reads, and transcriptome assembly) are used for SUmiRFind script in FASTA file format. In this step, the inputs align to known miRNA sequences with four or fewer mismatches. This alignment file is used for SUmiRFold part to search for the secondary structures. The SUmiRFold output files are provided in tabular format and they can be transformed to Microsoft Excel.

In order to prepare the inputs for miRCurator, users need to arrange the SUmiRFold output file on MS Excel with the following structure:

• The first column includes links to postscript files (file_name.fsa_1.ps). These addresses can also be used as a hyperlink to double-check results,

• the second and third columns are Mature Start and Mature End, respectively, and

• the fourth column is the Unique Hit ID.

2.2. miRCurator Script

After acquiring the data in MS Excel, the developed code (miRCurator) can be added as a Visual Basic for Applications (VBA) script. Since the VBA is embedded in Microsoft Excel, researchers can utilize the tool directly without further complication. The code can be divided into two main algorithms: (1) Multi-branched loop detector, and (2) Gene structure analyzer.

In order to conduct analysis on each gene, miRCurator reads each row of the Excel file and loads the corresponding ct file, which is created by the SUmirFold Perl script [6]. The associated gene in the Excel file is painted in yellow when the script cannot locate the ct file. In the generated ct file, the sequence of bases and their connections are mentioned. miRCurator reads the entire ct file and generates an n_base × 8 matrix (C_nbase×8) where n_base is the number of bases in the corresponding miRNA sequence. The columns in C matrix represents: (1) the index of current base in the miRNA sequence; (2) the corresponding base (i.e., A, C, G, and U); (3) the index of the previous base in the sequence; (4) the index of the next base in the sequence; (5) the index of the paired base; (6) the index of current base (repeated); (7 and 8) these columns specify whether the current base is paired (otherwise, zero values are associated to both for that row).

Algorithm 1 in miRCurator checks the sequence and counts the number of loops by looking at the 5th column of the C matrix. The time complexity of the Algorithm 1 is polynomial, in order of where m is the number of genes in the Excel file and is the maximum sequence length in each gene. As displayed in Figure 2, the tip of each branch can be detected using the suggested algorithm. An advantage of this method is faster scan speed, which is due to ignoring long stems in the first loop (i.e., the condition over C_i,5 in the 5th line).

Algorithm 1:

Multi-branched loop detector

Figure 2.

Demonstrating Algorithm 1 over dG = −61.4CM014314.1aau – miR172.

If the number of loops is more than one (i.e., multi-branched loop), miRCurator will mark the corresponding FASTA file with the red color for visual cues.

The second part of the miRCurator checks the gene for unacceptable structures. If the DICER cut point is located at exactly the mature strand start, it will be reported as unacceptable. Moreover, if the number of DICER cuts between the mature strand start and mature strand end is more than four then miRCurator will mark the corresponding gene as unacceptable. The reported genes in the second algorithm are painted in cyan color to provide visual cues for users who wants to double check the results. If a gene is filtered by both algorithms, it will be painted in magenta color (the combination of red and cyan). The time complexity of the second part is in order of . Figure 3 depicts the results of the second algorithm on a gene, in which the DICER cut is located at the mature strand start.

Figure 3.

Illustrating a sample, in which the DICER cut coincides with the mature strand start.

Using a 64-bit computer that benefits from an Intel Core i5-4200U CPU and 4 GB memory, the developed script processed 4970 genes in 77.16 seconds. According to the MTM-1 method [18] and experts’ opinions, a manual review of such a dataset can take over a day.

After processing the whole dataset, miRCurator asks the user whether s/he is interested in a visual summary. The script generates a visual representation of results, which can be loaded with Academic Presenter [19] if the user approves. Due to its advantages and flexibility, Academic Presenter is being widely used for data visualization purposes [12,20]. As illustrated in Figure 4, the generated graph is a vector-based image that classifies the members into three categories: Group A represents those genes that are painted in cyan in the MS Excel file, Group B includes those that are eliminated because of multi-branched loops (marked with red color in the spreadsheet), and the magenta ones, at the center, are rejected by both algorithms (A ∩ B).

Figure 4.

Generated visual for a sample dataset. As the number of branches increases in a gene, the radius of the corresponding blob grows and the color shifts toward red.

3. Results and Discussion

In all the studies mentioned in Section 2, the miRNA genes are identified by SUmiRFind and then, SUmiRFold script is used to decide the secondary structures by folding them. Using miRCurator, we successfully identified the pre-miRNAs, which have hairpins with multi-branched loops, inappropriate DICER cut sites, and more than four mismatches on their mature strand. Table 1 reports the number of pre-miRNAs, and the remaining pre-miRNAs after the filtration with miRCurator in detail for ten datasets. On average, miRCurator eliminates around 40.4% of ill-structured pre-miRNAs. To ameliorate the error in the calculations, the CPU time is computed as the average of three runs in each case study.

Table 1:

The summary of miRCurator’s output in various datasets from previously published studies.

We also conduct regression together with analysis of variance (ANOVA) on results in Table 1 to quantify the correlation between the CPU time (y) and regressors, which are the number of ill-structured entries (x_A = |A|) and entries with multi-branched loops (x_B = |B|). To prevent over-fitting, we set the constant to zero (y = β₁x_A + β₂x_B)³. The results show that there is a strong correlation between regressors and CPU time (R² = 0.97). Additionally, the ANOVA test rejects H₀-hypothesis, in which the coefficients of regressors are presumed zero (β₁ = β₂ = 0). Our analysis shows that the CPU time is more affected by x_B (β₂ = 0.028) than x_A (β₁ = 0.024). Figure 5 depicts the difference between the predicted and observed CPU time with respect to x_A and x_B.

Figure 5.

Predicted CPU time ŷ and observed CPU time y versus x_A and x_B.

Compared to the other Linux-based tools in the field [28], miRCurator is quick, accurate, and user-friendly since it provides outputs in MS Excel file format. In addition to these, our tool creates a detailed report at the end of the process that depicts all types of unwanted structures with their identification numbers. In the generated report, the users can perceive the information about the number of the multi-branched loops in each pre-miRNAs (the size of the red blobs), the number of high mismatch containing mature miRNAs and the number of the mature miRNAs having inappropriate DICER-cut points (blue-colored blobs). Although the generated MS Excel file by miRCurator with restricted color options provide visual cues, Academic Presenter’s image provide additional information using more color options, by which the users can identify the number of loops in pre-miRNAs. As a result of this, the users can conveniently track the various number of multi-branched formations.

The users may benefit from these outputs as the different number of multi-branched loops in a miRNA can be important to calculate the free energies of the secondary structures [29]. We also show the potential genes in the report, which are the combination of the multibranched loops and mismatches. By having all these outputs, the users can investigate the miRNA genes in detail for further studies.

Author Contributions

Conceptualization, methodology, validation, data curation, formal analysis, B.A.; software, visualization, D.E.A.; writing—original draft preparation, B.A. and D.E.A.;

Funding

This research received no external funding.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The interested readers can acquire the VBA script and a small case study from the supplementary material.

Conflicts of Interest

The authors declare no conflict of interest.

Footnotes

• 1 Downloaded from Medicinal Plant Genomics Resource http://mpgr.uga.edu/index.shtml

• 2 Downloaded from Medicinal Plant Genomics Resource http://mpgr.uga.edu/index.shtml

• 3 With this regression surface, we assume that datasets with no multi-branched loop or unwanted structure are empty.

Abbreviations

The following abbreviations are used in this manuscript:

miRNA

microRNA

dsRNA

double-stranded RNA

hpRNA

hairpin structured RNA

DCL

DICER-like

AGO

Argonaute

tasiRNA

trans-acting small interfering RNA

RDRs

RNA-dependent RNA polymerases

RISC

RNA-induced silencing complex

VBA

Visual Basic for Applications

ANOVA

analysis of variance