Prediction and identification of natural antisense transcripts and their small RNAs in soybean (Glycine max)

Prediction of NATs in soybean

We analyzed 66,213 soybean transcripts downloaded from the Phytozome database (http://www.phytozome.net/index.php) [26]. Over 13% (8,634) of the transcripts had at least one antisense transcript in soybean. Among these transcripts, over 50% (4,788) had only one antisense transcript, while the others had from 2 to 177 antisense transcripts (Figure 1). A total of 26,216 NATs were identified in soybean. The NATs were categorized into cis-NATs and trans-NATs according to the transcript origin from the genomic loci. Mapping of the NAT transcripts to the soybean genome identified 994 cis-NATs and 25,222 trans-NATs (Additional files 1 and 2).

cis-NATs and trans-NATs in soybean

The cis-NATs can be classified into three types: convergent (with 3’-ends overlapping); divergent (with 5’-ends overlapping); and enclosed (with one transcript completely overlapping the other) [15]. Among the 994 soybean cis-NATs, 468 were arranged in the enclosed orientation; 291 were convergent; and 235 were divergent (Additional file 1). In contrast, most of the cis-NATs from Arabidopsis and rice are convergent [11, 15].

cis-NAT overlaps length are usually longer than trans-NAT overlaps length [14], and this was also true for soybean NATs. The cis-NAT overlaps length ranged from 31–2,808 bp (308 bp average), whereas the trans-NAT overlaps length ranged from 31–1,716 bp (87 bp average). The overlapping length of the majority of trans-NATs (74.87%) was shorter than 100 bp, and only 7.31% were longer than 200 bp (Figure 2).

For the trans-NATs, one transcript commonly had many antisense transcripts [15, 24]. The number of antisense transcripts ranged from 1 to 177 in soybean, possibly a consequence of the homologous genes in the gene families frequently having the same antisense transcript [24]. The soybean genome has gone through at least two rounds of polyploidy and subsequent diploidization events. Segmental duplications and chromosome-level homology are common in the soybean genome [33–36], and approximately 75% of genes have multiple copies [37]. Some transcripts can form both cis-NATs and trans-NATs [15]. Of the 8,634 transcripts in soybean, 1,200 transcripts were involved in both cis- and trans-NATs (Figure 3). These genes may be regulated by cis- and/or trans-NATs.

Identification of NATs in soybean

We identified 17 transcripts using RT-PCR amplification. These 17 transcripts can form 21 trans-NATs. One transcript may form NATs with multiple antisense transcripts [15]. We identified Glyma01g09920.1, Glyma04g05850.1 and Glyma08g42710.1 as having the same five antisense transcripts. The overlapping region in the sense transcripts had similar sequences (Additional file 3). Glyma14g13230.1 can form NATs with Glyma02g34100.1, Glyma10g23170.1, Glyma14g22790.1, Glyma18g16420.1, Glyma14g22790.1 and Glyma20g06230.1. The overlapping region of Glyma02g34100.1, Glyma10g23170.1, Glyma18g16420.1 and Glyma14g22790.1 had the same sequences, while Glyma20g06230.1 can form NAT at another location on the Glyma14g13230.1 transcript.

Small RNAs originating from NATs

As NATs can generate small RNAs [14, 15] we searched for the presence of small RNAs in our library for the 8,634 transcripts. We identified 2,286 transcripts able to give rise to small RNAs. 189,348 small RNA sequences, representing 27,465 unique small RNAs, were generated from these transcripts. The most abundant unique small RNAs were 21, 22, and 24 nt in length (Figure 4). In soybean, these 2,286 transcripts could form 179 cis-NATs and 6,629 trans-NATs (6,808 total; Additional file 4).

Most of the small RNAs were derived from one of the NAT transcripts in Arabidopsis[15]. Both cis- and trans-NATs mostly generated small RNAs from one strand of the NAT in soybean (Figure 5). Among the cis-NATs, 75.4% (135) generated small RNAs from only one strand of the NAT, and 9.5% (17) generated small RNAs equally from both transcripts. For the trans-NATs, 30.4% (2,019) generated small RNAs from only one strand, and 19.9% (1,321) generated small RNA equally from both strands.

Small RNAs originated from both the overlapping and non-overlapping regions of NATs [15]. The distribution of small RNAs in these two regions varies in different plants [38]. In soybean, the average densities (the number of small RNA loci per kilobase) of the unique and total small RNAs in the overlapping regions were 103.84 and 517.80, respectively, and 48.72 and 344.24 for the entire NATs. T-tests for the unique (P < 0.0001) and total (P < 0.0001) small RNAs suggested that both were enriched in the overlapping region.

The NATs degradome in soybean

NATs can produce small RNAs, which suggests that these transcripts are excised by Dicer-like proteins. We searched for the degradome tags of the 6,808 NATs that could produce small RNAs. A total of 122 cis-NAT and 4,425 trans-NAT transcripts were identified as having degradomes (Additional file 4). Most degradome tags were derived from one NAT transcript (Figure 5): 53.3% (65) cis-NATs, and 50.2% (2,222) trans-NATs, generated tags from only one transcript. This was consistent with the small RNA bias towards one strand of NATs.

Identification of NAT-derived small RNA targets in soybean

nat-siRNA can regulate gene expression by guiding target mRNA degradation at the posttranscriptional level [9, 19]. The targets of siRNAs can be globally identified by analyzing the degradome [27–32]. We searched the nat-siRNA targets by analyzing the soybean degradome and identified 446 target genes for the 165 nat-siRNAs (Additional file 5). Of these 165 nat-siRNAs, 83 were derived from trans-NATs, 81 from cis- or trans-NATs, and only one was generated from a cis-NAT. Regarding the 446 target genes, 203 were targeted by a nat-siRNA derived from the corresponding NAT sense strand, and 75 were targeted by a nat-siRNA produced from the corresponding antisense strand. The nat-siRNAs targets not only the transcript of their own NATs but also that of other transcripts. A total of 168 genes were identified as targets of nat-siRNAs, these nat-siRNAs were not produced from target sense or antisense transcripts.

miRNAs may be involved in the formation of NATs in soybean

Some NATs can form stem-loop structures and generate mature miRNAs. In rice, some miRNAs are derived from the overlapping transcript antisense of MADS box transcripts, and act to guide MADS transcript cleavage [39]. We used the UNAfold program to simulate folding of 2,286 transcripts identified as being able to produce small RNAs [40]. Five transcripts were predicted to contain a stem-loop structure characteristic of pre-miRNA (Additional file 6). These transcripts were Glyma02g02440.1, Glyma04g38430.1, Glyma05g03670.1, Glyma05g32980.1, and Glyma05g37200.1. Further analysis revealed that Glyma04g38430.1 and Glyma05g32980.1 were miR166 genes; Glyma05g37200.1 produced miR319; and Glyma02g02440.1 and Glyma05g03670.1 generated small RNAs randomly from both sense and antisense strands (Additional file 7). These five genes may be involved in the biogenesis of both miRNAs and NATs. There are two possible pathways by which miRNAs could be generated from these transcripts. One pathway occurs when the sense and antisense transcripts are co-expressed in the same cell, form a double RNA duplex, and produce nat-siRNAs. This then guides the generation of small RNAs from their sense or antisense transcripts [9]. Another pathway occurs when the sense and antisense transcripts are not co-expressed in the same cell; these transcripts can fold into a hairpin and produce miRNAs.

Targets of miRNAs may be involved in the formation of NATs. We collected 596 candidate targets of miRNAs and searched for targets that could form NATs. 86 miRNA targets were identified as having antisense transcripts (Additional file 8). These targets could form cis- and trans-NATs. Analysis of the soybean degradome of these 86 targets validated 28 as being miRNA targets [31, 32].

NATs may form complex regulatory networks in soybean

It has been suggested that NATs form complex regulatory networks in plants [15]. One transcript often has many antisense transcripts, and these can form NATs with other transcripts. In soybean, 1,200 transcripts were predicted to form both cis- and trans-NATs (Figure 3). 11 transcripts had multiple cis-NATs. Soybean commonly has one transcript that has many antisense transcripts forming the trans-NATs. Of the 8,634 transcripts that form NATs, 3,846 contain multiple (2–177) antisense transcripts (Figure 1).

The nat-siRNAs play important roles in plant development. NATs produce nat-siRNAs via a process mediated by Dicer-like RNA-dependent RNA polymerase and Suppressor of Gene Silencing 3. The nat-siRNA is then incorporated into the RISC and directs the cleavage of a complementary mRNA [9, 17]. With high-throughput sequences of small RNAs from different soybean tissues, we detected 6,808 NATs that produced at least one small RNA (Additional file 4). These small RNAs potentially regulate gene expression at the posttranscriptional level. In recent years, deep sequencing of the degradome has been used extensively to globally identify small RNA targets. Analysis of the soybean degradome database enabled identification of 446 genes as targets of 165 nat-siRNAs. These nat-siRNAs targets included NAT sense or antisense transcripts, and other transcripts (Additional file 5).

miRNAs and their targets may be involved in NAT regulatory networks. Five transcripts with pre-miRNA stem-loop structures had antisense transcripts. These transcripts may generate nat-siRNAs or miRNAs; this is dependent on whether the transcripts are co-expressed with antisense transcripts in the same cell. Furthermore, we detected 86 miRNA targets that had antisense transcripts in soybean. These miRNA targets might be regulated by their antisense transcripts.

NATs may form complex regulatory networks in soybean (Figure 6). In these networks, gene expression is regulated by other genes forming cis- or trans-NATs. NATs can produce nat-siRNAs that self-target their NAT transcripts and other gene transcripts. Some NATs produce miRNAs to regulate expression of other genes, and some miRNAs guide the cleavage of NATs.

Plant material and RNA isolation

Soybean (Glycine max) seeds of the cultivar Williams82 were planted in the experimental station of the Institute of Crop Sciences at the Chinese Academy of Agricultural Sciences, in Beijing in May. Flowers were collected and quickly frozen in liquid nitrogen and then stored at −70°C for use in future RNA isolation. Leaves and roots were collected from 12 days old soybean seedlings. Total RNA from different tissues was isolated separately using TRIzol reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s instructions. RNA samples were evaluated by electrophoresis on a 1% agarose gel.

Sequence datasets

Soybean gene sequences and annotations were downloaded from the Phytozome database (version 1.0; http://www.phytozome.net/index.php) [26]. The small RNAs and the degradome were previously identified with deep sequencing in our laboratory. Information for the soybean small RNAs and degradome is from the NCBI-GEO database (accession no. GSE33380). The soybean miRNAs were downloaded from miRBase (version Release 18; http://microrna.org/) [41].

Prediction of NATs in soybean

NATs were detected by aligning predicted Glycine max cDNA sequences to each other. If a pair of overlapping genes were matched at opposite strands with an E-value ≤ 1e-9¹⁹, then they were defined as a NAT pair. The NAT pair was located on the soybean genome to identify cis- and trans-NATs. If a pair of NATs was located at the same genome locus, they were considered a cis-NAT pair. If they were located at different genomic loci, they were considered a trans-NAT pair. Based on the overlap between the two transcripts, the cis-NATs were categorized into three types: convergent (3’-ends overlap); divergent (5’-ends overlap); and enclosed (full overlap).

Identification of NATs by RT-PCR

Several NATs were identified by use of RT-PCR. We designed gene-specific primers to amplify cDNAs based on their NAT transcript sequence (Additional file 9). 50 μg leaf RNA, 25 μg root RNA, and 25 μg flower RNA were added to a tube and mixed gently, these RNAs were treated with DNase I (Fermentas, Harrington, Ontario, Canada) for 30 minutes at 37°C, and then purified with phenol-chloroform. A total of 4 μg purified RNA was used in a 20 μl RT reaction containing 2 μl gene-specific RT primer (10 μM) (Additional file 9), 4 μl 5× reaction buffer, 1 μl RiboLock RNase inhibitor (20 u/μl), 2 μl dNTP (10 mM), and 1 μl (200 u/μl) RevertAid M-MuLV reverse transcriptase; this was carried out using the RevertAid First Strand cDNA Synthesis kit (Fermentas, Harrington, Ontario, Canada) according to the manufacturer’s instructions. 1 μl of the first strand cDNA sample was used as template for subsequent PCR reactions in 25 μl reactions using gene-specific primers with the following cycle conditions: 95°C, 30 s; 55°C, 30 s; 72°C, 1 min; the run was for 35 cycles. The RT-PCR products were evaluated by electrophoresis on 2% agarose gel. 1 μl of RT-PCR product was ligated into the pGEM-T vector using the pGEM-T easy vector system (Progema, Madison, WI, USA) according to the manufacturer’s instructions; next, 2 μl ligation reaction was transformed into TOP10 competent cells. Five clones of each gene were sequenced (Additional file 10).

Analysis of small RNAs

The small RNAs were screened against the Sanger Non-coding RNA Database (http://www.sanger.ac.uk/resources/databases/rfam.html) to eliminate rRNAs, tRNAs, and snoRNAs [42]. Small RNAs that were identical to the transposable elements identified in the G. max genome, downloaded from SoyTE (http://www.soybase.org/soytedb/), were also removed. Small RNAs were aligned to NAT transcripts using SOAP [43]. Sequences that were identical to NAT transcripts were considered as NAT-derived siRNAs. The significance of the enrichment of small RNAs in the overlapping regions of NATs was calculated according to the method previously described by Chen et al. [25]. Briefly, the number of unique small RNAs generated from the overlapping region (No) and NAT transcripts (Nt), and their corresponding lengths (Lo and Lt) were determined. The ratios No/Nt and Lo/Lt were used to calculate the density of small RNAs in the overlapping (Do) and entire (Dt) regions of the NAT. The ratio of Do/Dt was considered to be the enrichment score, and a standard χ² test was performed to test the significance of the enrichment of small RNAs in the overlapping regions of NATs.

Identification of nat-siRNA targets

Degradomes were mapped onto NAT transcripts using SOAP. The sequences that identically matched NAT transcripts were considered to be the NAT degradome. The locus sequence, containing the 20 bp upstream and downstream regions of the NAT degradome, was extracted as the long degradome from the transcript. Next, a search for small RNA targets was performed as described by Schwab et al. [44]. Total NAT-derived small RNAs were used to query the long degradome sequences, and small RNAs and complementary cDNA pairs for potential target sites were obtained using Patscan set at the default parameters: three mismatches, zero insertions, and zero deletions were permitted [45]. Only hits with fewer than two mismatches in positions 1–9, no mismatches in positions 10 and 11, and fewer than three mismatches after position 11 in the small RNAs were considered good target sequences.

Identification of pre-miRNAs and miRNA targets involved in the NAT network

Flanking sequences of the small RNAs that matched identically with the NAT transcripts were obtained as described by Sunkar and Zhu [46]. Fragment sequences 200 bp upstream and downstream of NAT-derived small RNAs were extracted from the NAT transcripts. Simulation of folding was then performed using UNAfold [40]. Identified secondary structures were checked for miRNA features using MirCheck [47]. The miRNA targets in soybean were predicted using the psRNATarget server (http://plantgrn.noble.org/psRNATarget/) [48]. Validated targets of miRNAs in soybean were obtained from Song et al. and Hu et al. [31, 32]. The targets containing the antisense transcripts were considered to be NAT-related miRNA targets.