Exploration of miRNA-mediated fertility regulation network of cytoplasmic male sterility during flower bud development in soybean

Abstract

Cytoplasmic male sterility (CMS) plays an important role in the production of soybean hybrid seeds. MicroRNAs (miRNAs) are a class of non-coding endogenous ~ 21 nt small RNAs that play crucial roles in flower and pollen development by targeting genes in plants. To dissect the function of miRNAs in soybean CMS, a total of 558 known miRNAs, 10 novel miRNAs, and 466 target genes were identified in flower buds of the soybean CMS line NJCMS1A and its restorer line NJCMS1C through small RNA sequencing and degradome analysis. In addition, miRNA-mediated editing events were also observed, and the two most frequently observed editing types (A → G and C → U) were validated by cloning and sequencing. And as the base editing occurred, some targets were filtered, such as gma-miR2118b-P6GT with Glyma.08G122000.2. Further integrated analysis of transcriptome and small RNA found some miRNAs and their targets’ expression patterns showing a negative correlation, such as gma-miR156b/GmSPL9a and gma-miR4413b/GmPPR. Furthermore, opposite expression pattern was observed between gma-miR156b and GmSPL9 during early stage of flower bud development. Taken together, the regulatory network of gma-miR156b/GmSPL9 and gma-miR4413b/GmPPR with flower bud development in soybean CMS was developed. Most importantly, previous reports showed that these targets might be related to pollen development and male sterility, suggesting that both conserved and species-specific miRNAs might act as regulators of flower bud development in soybean CMS. These findings may provide a better understanding of the miRNA-mediated regulatory networks of CMS mechanisms in soybean.

Electronic supplementary material

The online version of this article (10.1007/s13205-018-1543-1) contains supplementary material, which is available to authorized users.

Introduction

Soybean is one of the most important oil crops in the world. However, soybean yields are relatively low and the utilization of heterosis is one of the effective ways to improve them. Cytoplasmic male sterility (CMS) is a simple and efficient pollination control system that plays an important role in the production of hybrid seed. CMS is caused by mitochondrial genes with coupled nuclear genes, and CMS-based hybrid seed technology uses a three-line system: the CMS line, the maintainer line, and the restorer line (Chen and Liu 2014). The CMS line is propagated by crossing with the maintainer line and by crossing the CMS line with the restorer line, the male fertility of the F₁ plant can be restored by the Rf (restorer of fertility) gene(s) that come from the nuclear genome of the restorer line. Whether it is the production of CMS or F₁ fertility restoration, a series of changes will occur at both the molecular and physiological levels. In this process, the interaction between genes is very obvious. Therefore, the molecular mechanism of CMS is a hot research topic and there may be a high degree of epigenetic regulation involvement.

MicroRNAs (miRNAs) are a class of non-coding endogenous ~ 21 nt small RNAs (Bartel 2004) that play vital regulatory roles in plant growth and development, such as flowering-time regulation (Nie et al. 2015), leaf development (Samad et al. 2018), accumulation of anthocyanins (Liu et al. 2017; Zhao et al. 2017), response to the phytohormone abscisic acid (Duan et al. 2016), root development (Liu et al. 2018), cotton fiber elongation (Wang et al. 2016a), pollen or flower bud development (Shen et al. 2011; Yang et al. 2013; Jiang et al. 2014; Ding et al. 2016; Zhang et al. 2016), and so on. Among all of the miRNA families, MIR156 is one of the most conserved families in plants (Wang et al. 2016b) and regulates the vegetative phase change floral transition, integrated into flowering pathways by target squamosa promoter binding protein-like (SPL) transcription factors (Spanudakis and Jackson 2014). Moreover, the miR156 targeted or non-targeted SPL genes function together to redundantly secure male fertility in Arabidopsis (Xing et al. 2010). Compared with conserved miRNA, the expression of species-specific miRNA is relatively low, and they can be induced to function under certain special conditions (Tang 2010).

Our previous results have shown that epigenetic modification, such as DNA methylation and circular RNAs, are involved in flower bud development of soybean CMS (Li et al. 2017b; Chen et al. 2018). miRNAs, as a form of epigenetic modification, were purported to take part in the process of male sterility in plant (Xing et al. 2010; Yu et al. 2013). In this study, a number of miRNAs and their targets in soybean were identified by comparing a CMS line (NJCMS1A) and its restorer line (NJCMS1C), based on small RNA sequencing and degradome analysis. Most importantly, the regulatory network of gma-miR156b/GmSPL9 and gma-miR4413b/GmPPR with flower bud development in soybean CMS was found by small RNA sequencing, transcriptome and qRT-PCR comparative analysis. These results might shed light on the potential regulatory role of miRNAs in controlling flower bud development of soybean CMS.

Materials and methods

Plant materials, sample collection, and total RNA extraction

The soybean cytoplasmic male-sterile line NJCMS1A was developed through consecutive backcross procedures with the cultivar N8855 as donor parent and N2899 (hereafter designated as NJCMS1B) as recurrent parent (Gai et al. 1995; Ding et al. 1999, 2002). “Zhongdou 5” (hereafter designated as NJCMS1C for further evaluation of the genetic mechanism of male fertility restoration) was identified as the restorer of NJCMS1A in a previous study (Yang et al. 2007). NJCMS1A and NJCMS1C were planted in the summer of 2013 at Jiangpu Experimental Station, National Center for Soybean Improvement, Nanjing Agricultural University, Nanjing, Jiangsu, China, for sample collection for small RNA sequencing and degradome sequencing. The male-sterile plants and fertile plants were identified through three methods, namely, the dehiscence of anthers, germination rate of pollen, and performance of plants at maturity. Cytological observation showed that the male abortion of NJCMS1A occurred mainly at the early binucleate pollen stage (Fan 2003). Because it is very difficult to judge the precise developmental stage of pollen from the appearance of the flower buds in soybean, during the flowering period in the summer of 2013, flower buds of different sizes up to and including the abortion stage were collected and mixed from NJCMS1A and NJCMS1C plants, and then immediately frozen in liquid nitrogen and stored at − 80 °C for further use. Flower buds were collected from NJCMS1A and NJCMS1C in the summer of 2015 for quantitative real-time PCR (qRT-PCR). Total RNAs from the flower buds of NJCMS1A and NJCMS1C were extracted using TRIzol reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer’s protocol.

Small RNA sequencing library construction and bioinformatics analysis

Two small RNA libraries were constructed and sequenced by the Beijing Genomics Institute (BGI, Shenzhen, Guangdong Province, China) using the Illumina HiSeq 2000 System (TruSeq SBS KIT-HS V3, Illumina, Santa Clara, CA, USA). After sequencing, raw sequencing reads were processed into clean reads by filtering out adaptor contaminants, oversized insertions, low-quality reads, poly(A) tags, and small tags (< 18 nt). The small RNA tags were mapped to the soybean genome (Wm82.a2.v1) using SOAP to analyze their expression and distribution on the genome. All small RNA sequences were used to query the RNA sequences for deposited repeats (RepeatMasker output) in GenBank (NCBI GenBank). Rfam (V 11.0) was used to separate the small RNAs that matched rRNA, scRNA, snoRNA, snRNA, and tRNA and remove the degraded mRNA fragments (Wm82.a2.v1) in the small RNA tags that aligned to exons and introns. Next, to identify conserved miRNAs, clean reads were compared with known miRNAs of soybean deposited at miRBase 21.0 (http://www.mirbase.org/). Finally, the other sequences that did not map to known miRNAs and other kinds of small RNA were referred to as unannotated sequences for identification of new members of known miRNA families and for novel miRNA prediction.

Potentially novel miRNAs were identified using MIREAP (https://source-forge.net/projects/mireap/), and their secondary structures were predicted by the mfold web server (http://mfold.rna.albany.edu/?q=mfold/RNA-Folding-Form) (Zuker 2003). The criteria used for selecting novel miRNAs must meet the following characteristics based on Meyers et al. (2008), Zhang et al. (2006), Kozomara and Griffiths-Jones (2014), and miRBase 21.0: (1) the candidate miRNA-5p and miRNA-3p are derived from opposite stem arms with minimal matched nucleotide pairs exceeding 16 nt and with maximal size differences of up to 4 nt; (2) the most abundant reads from each arm of the precursor must pair in the mature miRNA duplex with a 2-nt 3ʹ overhang; (3) the number of asymmetric bulges within the miRNA-5p/miRNA-3p duplex must be one or fewer, and the size of the asymmetric bulges must be two bases or smaller; (4) the miRNA-5p and miRNA-3p are simultaneously on the two arms of pre-miRNA secondary structures, and one of the mature miRNAs must be no fewer than five; (5) the candidate miRNA precursor must have high negative minimum free energy (MFE) and minimum folding energy index (MFEI), with MFE < − 0.2 kcal/mol/nt and MFEI > 0.85.

Degradome library construction, data analysis, and target identification

Two degradome libraries were constructed according to the methods of Addo-Quaye et al. (2008) and German et al. (2008) and sequenced on the Illumina HiSeq 2000 (TruSeq SBS KIT-HS V3, Illumina, BGI, Shenzhen, Guangdong Province, China). Alignments with scores up to 4.5, where G:U pairs scored 0.5 and no mismatches were found at the site between the 10th and 11th nucleotides of the corresponding miRNAs, were considered potential targets.

Quantitative real-time PCR validation

Stem-loop qRT-PCR and qRT-PCR were carried out to validate differential expression levels of miRNAs and miRNA targets, respectively (Chen et al. 2005). All primers (Table S1) were designed based on the mature miRNA and mRNA sequences and synthesized commercially (Invitrogen, Shanghai, China). According to the procedures provided in the iScript Select cDNA Synthesis Kit (containing GSP enhancer solution, BIO-RAD, USA), 1 µg of total RNA was reverse transcribed by iScript reverse transcriptase using stem-loop primers. The qRT-PCR analysis was carried out using iTaq Universal SYBR Green Supermix (BIO-RAD, USA) on a Bio-Rad CFX96 instrument (CFX96 Touch, BIO-RAD, USA). All reactions were run with three biological replicates, and gma-miR1520d (Kulcheski et al. 2010) was used as the internal control gene. The relative expression levels of miRNAs and genes were quantified using the $2^{-\mathrm{\Delta }\mathrm{\Delta }{C}_t}$ method. Student’s t test was performed to compare miRNA or mRNA differences in expression between NJCMS1A and NJCMS1C. Expression levels that significantly differed (p < 0.05) according to Student’s t test were labeled as: “*”, “**”, and “***”, which represent p < 0.05, p < 0.01, and p < 0.001, respectively, which indicated significant difference levels between NJCMS1A and NJCMS1C.

Results

Overview of small RNA library data sets

Two small RNA libraries were constructed using total RNA obtained from flower buds of the CMS line NJCMS1A and its restorer line NJCMS1C and sequenced by Illumina Solexa high-throughput sequencing technology. After removing adaptor contaminants and low-quality reads etc., a total of 11,974,839 and 12,072,909 clean reads were obtained from the two libraries, respectively, ranging from 18- to 30-nt in length (Fig. 1a). Using the SOAP software, more than 80% small RNA tags were perfectly mapped to the soybean genome (Table 1).

Fig. 1.

Length distribution of small RNA sequences and degradome-seq clean tags in flower buds of NJCMS1A and NJCMS1C. a Length distribution of small RNA sequences derived from the small RNA libraries. b Length distribution of degradome-seq clean tags derived from the degradome libraries

Table 1.

Summary of small RNA sequences and degradome sequences

Type	Small RNA-seq		Degradome-seq
	NJCMS1A	NJCMS1C	NJCMS1A	NJCMS1C
Statistical information
Total reads	12,190,078 (100.00%)	12,267,565 (100.00%)	12,351,703 (100.00%)	12,669,765 (100.00%)
Clean reads	11,974,839 (98.93%)	12,072,909 (99.15%)	12,288,476 (99.49%)	12,613,736 (99.55%)
Genome mapping statistics
Total count of clean tags
Total tags	11,974,839 (100.00%)	12,072,909 (100.00%)	12,288,476 (100.00%)	12,617,159 (100.00%)
Mapping to genome	9,808,568 (81.91%)	9,985,396 (82.71%)	10,926,341 (88.92%)	11,244,308 (89.12%)
Unique count of clean tags
Total tags	3,668,753 (100.00%)	4,339,928 (100.00%)	6,332,697 (100.00%)	6,482,757 (100.00%)
Mapping to genome	2,682,237 (73.11%)	3,204,022 (73.83%)	5,494,191 (86.76%)	5,638,108 (86.97%)

Upon analyzing the small RNA tags between NJCMS1A and NJCMS1C, we found that NJCMS1A and NJCMS1C comprised 14.86% and 18.31% of the total sRNAs (Fig. 2a), respectively. Only 12.64% of unique sRNAs were shared between the CMS line and its restorer line (Fig. 2b). Moreover, the NJCMS1C-specific unique sRNAs were greater than those of NJCMS1A, which means that most of the unique sRNAs found in NJCMS1C flower buds were different from those in NJCMS1A (Fig. 2b).The small RNAs annotation for the two samples is shown in Fig. 2c, d. The majority of small RNAs were 21–24 nt, and 24-nt small RNAs were the most abundant (Fig. 1a), which is in accordance with the previously reported miRNAs of soybean (Song et al. 2011).

Fig. 2.

Distribution of total sRNAs (tags) and unique sRNAs (tags) among small RNA libraries and degradome libraries. a, b Distribution of total sRNAs and unique sRNAs in small RNA libraries of NJCMS1A and NJCMS1C. c, d Annotation of sRNAs in small RNA libraries of NJCMS1A and NJCMS1C. e, f Distribution of total tags and unique tags in degradome libraries of NJCMS1A and NJCMS1C

Identification of known miRNAs

The mappable small RNA sequences were aligned with the known soybean miRNAs using miRBase 21.0. A total of 499 known pre-miRNAs corresponding to 558 mature miRNAs that belong to 222 families were detected (Table S2, S3). The sequences of some miRNAs were tagged with “D” (Table S2), representing that these were either shorter or longer than the mature miRNAs in miRBase 21.0; these can be classified as isomiRNAs and may share the same target genes with the mature miRNAs.

Identification of novel miRNAs

To identify novel miRNAs, all unannotated and mappable small RNAs were aligned to known plant miRNAs in miRBase 21.0. The small RNAs that could not be mapped to known plant miRNAs were classified as novel miRNAs. The five criteria described in “Materials and methods” were utilized to increase predictive accuracy. As a result, ten novel miRNAs (five pairs) that belong to four families were identified in this study (Table 2 and Table S3, S4). Precursors of these novel miRNAs were identified by MIREAP and varied from 85 to 93 nt in length, with MFE values ranging from − 0.47 to − 0.57 kcal/mol/nt. The MFEI values ranged from 1.15 to 1.65 (Table S4), all of them being greater than 0.97 and the average being 1.36. The length of all of the mature miRNAs was 21 nt. Some of the novel miRNAs, such as novel_mir_003a to novel_mir_003b, shared the same mature sequence but had different precursors that came from different loci. The secondary structure for the precursor of novel_mir_003a was selected as an example to be shown in Fig. S1.

Table 2.

Novel miRNAs identified in NJCMS1A and NJCMS1C

miRNA Name	Novel miRNA Sequence	Length (nt)	NJCMS1A Count	NJCMS1C Count	3p/5pa
novel_mir_001-5p	TGGAGAAGCAGGGCACATGCT	21	78	136	5p
novel_mir_001-3p	CTTGTGTCCTACTTCTCCAGC	21	4	11	3p
novel_mir_002-5p	GAGCAATTCTCCTTTGGCAGA	21	2	0	5p
novel_mir_002-3p	TGCCAAAGGAGAATTGTCCTG	21	6	0	3p
novel_mir_003a-5p	TTTATCGGTAACATCATCATC	21	261	0	5p
novel_mir_003a-3p	TGATGATGCTACTGATAAATA	21	4	0	3p
novel_mir_003b-5p	TTTATCGGTAACATCATCATC	21	261	0	5p
novel_mir_003b-3p	TGATGATGCTACTGATAAATA	21	4	0	3p
novel_mir_004-5p	TAAACGTTTGATCCCTTGTAT	21	8	0	5p
novel_mir_004-3p	ACAAGGGATCAAACATTTAAG	21	1	0	3p

^aArm of this mature sequence on the predicted secondary structure

Overview of miRNA base edits

In the present study, miRNAs which might have base edit were detected by aligning unannotated sRNA tags with mature miRNAs from miRBase21.0. A total of 174 miRNA members exhibited base edit and the miRNA editing frequency was 0.13–100.00% (Table S5). Some miRNAs, such as gma-miR1516a-3p, were not found by sequencing, but its editing event was observed in this study (Table S5-1).The miRNA editing events had truly occurred in soybean, as confirmed by our sequencing results (Fig. S2). Most of the miRNA editing events occurred at nucleotide positions of 5–17 and the miRNA editing patterns were similar between the two lines, but the editing frequency at position of 11 is slightly lower (Fig. 3a). The most dominant nucleotide substitution types were A → G, C → U and G → U (Fig. 3b).

Fig. 3.

Summary and validation of miRNA editing events in flower buds of NJCMS1A and NJCMS1C. a Summary of miRNA editing frequency at each miRNA nucleotide position in both NJCMS1A and NJCMS1C. b Summary of miRNA editing types and their frequency in both NJCMS1A and NJCMS1C. c, d Validation of miRNA editing types inferred from miRNA precursors and mature miRNAs by high-throughput sequencing and cloning. Sequencing chromatograms from the two observed editing types (C → T, A → G) of gma-miR159e and gma-miR4403 are shown. The edited positions are highlighted in pink. The upper panel indicates the part of precursor miRNAs cloned from genomic DNA, and the lower panel indicates the mature miRNAs cloned from cDNA reverse transcribed with stem-loop RT primers. Primers are listed in Table S1

To validate the occurrence of the miRNA editing, the two most frequently observed edited types (A → G and C → U) were examined. The precursor miRNA sequence was cloned from the soybean genome, and the mature miRNA sequence was cloned from the cDNA reverse transcribed by the stem-loop reverse primers. Both sequencing and validate results confirmed that nucleotide substitutions had truly occurred during the mature miRNA processing (Table S5 and Fig. 3c, d). However, the reason for the variant editing type and their function in mature miRNA processing or gene expression regulation remain elusive and warrant further study.

Comparative analysis of miRNAs

Throughout all the identified miRNAs in flower buds of NJCMS1A and NJCMS1C, 76 differentially expressed miRNAs (including 74 known miRNAs and two novel miRNAs) with more than two-fold relative change were identified (Table S6). Among the differentially expressed miRNAs, 28 miRNAs were upregulated in NJCMS1A, and the remaining miRNAs showed higher expression levels or were only expressed in flower bud of NJCMS1C (Table S6). For example, gma-miR4393a and gma-miR5667 were only expressed in NJCMS1A, and gma-miR4346, gma-miR4372a, gma-miR4400, gma-miR4994-5p, and gma-miR5785 were only detected in NJCMS1C. In particular, gma-miR4346 showed a high relative expression level and may have a unique function in the process of pollen development in NJCMS1C. Of the differentially expressed novel miRNAs, most of them were only expressed in NJCMS1A or NJCMS1C (Table S6).

To examine miRNA expression, ten selected differentially expressed miRNAs were assayed by stem–loop qRT-PCR. As showed in Fig. 4, nine of these were consistent with the sequencing reads. However, gma-miR6300 was found to be upregulated in NJCMS1C, but was enriched in NJCMS1A based on the high-throughput sequencing analysis and this result warrants further study.

Fig. 4.

Verification results of differentially expressed miRNAs between NJCMS1A and NJCMS1C. The y axis indicates the miRNA relative expression level (log2) generated from qRT-PCR analysis and high-throughput sequencing. The results were obtained from three biological replicates, and the error bars indicate the standard error of the mean of $2^{-\mathrm{\Delta }\mathrm{\Delta }{C}_t}$, with NJCMS1C as a control

Degradome library construction and comparative analysis of miRNA targets between NJCMS1A and NJCMS1C

Here, we performed degradome sequencing and analysis to search for the miRNA targets in NJCMS1A and NJCMS1C. A total of 12,288,476 and 12,617,159 clean reads were generated from the NJCMS1A and NJCMS1C libraries, respectively (Table 1). More than 98.00% of the tags were 20 or 21 nt in length (Fig. 1b), as expected from the normal length distribution peak of degradome fragments. Of the unique tags, 41.89% and 43.24% were from NJCMS1A and NJCMS1C, respectively, and only 14.87% of them were shared between NJCMS1A and NJCMS1C (Fig. 2e, f).

To better understand the function of miRNA in flower bud development, we performed a conjoint analysis of the miRNA and degradome together. A total of 466 distinct transcripts targeted by 200 miRNAs were detected (Table S7 and Fig. S3). These transcripts included SPL family protein, GAMYB protein, auxin response factor, homeobox-leucine zipper protein, nuclear transcription factor Y, GRAS family transcription factor, and transcription factor APETALA2, which have essential roles in gene regulation (Table S7). Most importantly, a set of pentatricopeptide-repeat-containing (PPR) proteins were targeted by gma-miR1508, gma-miR160, gma-miR171o, gma-miR4413b and gma-miR5674, which may be a group of Rf proteins that suppress CMS gene expression through posttranscriptional mechanisms such as editing, cleavage, and degradation of the target mRNAs (Table S8; Chen and Liu 2014).

To determine if the base edit in positions 2–8 and 10–11 of a mature miRNA can change its target recognition, a combined analysis of the degradome library and base-edited miRNAs was conducted. As a result, 122 distinct transcripts targeted by 307 base-edited miRNAs were detected (Table S9). Most of the targets of the base-edited miRNAs were the same as for the corresponding miRNAs for which no base editing had occurred (Table S10). However, some targets were filtered out by the base editing. For example, there were 75 targets of gma-miR156b and gma-miR156f, but as the base edited, only 61 targets were detected (Table S10). Interestingly, we also detected some new targets that were not targeted by miRNA when no base editing occurred, such as Glyma.05G230700.2 to Glyma.05G230700.4, which were targeted by gma-miR166h-P8CT, gma-miR166k-P8CT, and gma-miR166u-P8CT (Table S10), but not by gma-miR166h, gma-miR166k, and gma-miR166u. Furthermore, with base editing in position 10–11 of the mature miRNA, some miRNA targets were changed. For example, no targets were detected for gma-miR4403 in this study, but as a result of a base edit at position 11, one target (Glyma.17G050600.1) was found (Fig. S4 and Table S10).

Integrated analysis of miRNA and mRNA expression between NJCMS1A and NJCMS1C

To better understand the function of differentially expressed miRNAs in flower bud development of soybean, we performed an integrated analysis to identify inverse miRNA–mRNA expression based on small RNA and transcriptome comparative analysis (Li et al. 2017a). In consideration of the expression of some miRNAs and their targets were changed more significantly in sporogenic tissues but were submerged by the expression in other tissues of the flower buds, so a cutoff of > 1.5-fold change was used in this section as that of small RNA and transcriptome sequencing in Chinese cabbage CMS (Wei et al. 2015). As a result, 13 miRNAs (6 miRNA families) with their 7 inverse expressed target genes that had a fold change > 1.5 or < − 1.5 were identified (Table S11 and Fig. S5). Among them, the conserved miRNA, gma-miR156b, was more than 1.5-fold greater in NJCMS1A. Twenty-three target genes were identified to be its targets in this study (Table S7). However, only one target showed a negative correlation with gma-miR156b (Table S11). Despite the conserved miRNAs, gma-miR4392 and gma-miR4413b were two soybean-specific miRNAs, which were also altered in NJCMS1A (Table S11). This result indicated species-specific miRNA may also participate in the flower bud development in plant. In addition, 3 conserved miRNA families, included gma-miR169, gma-miR1514 and gma-miR397, were more than 1.6-fold greater in the NJCMS1C. Of the targeted genes, 4 were found to be showing more than 2.0-fold higher in NJCMS1A (Table S11).

miRNA–target network underlying flower bud development of soybean CMS

Based on the integrated analysis of the expression profiles of miRNAs and their targets, six miRNA–targets pairs were found, in which miRNAs and target genes are negatively regulated at the expression level (Table S11 and Fig. S5). However, only half of the miRNA families were upregulated and suppressed the expression of three targets (Table S11 and Fig. S5). MIR156 is one of upregulated miRNA families and our present study found that 20 GmSPL genes were targeted by gma-miR156b (Table S7). Unfortunately, only GmSPL9a was downregulated in flower bud of soybean CMS (Table S11 and Fig. 5). It is possible that the expression of other GmSPL genes was very sensitive to the stage of flower bud development. To prove this point, qRT-PCR analysis was conducted to detect the expression level of gma-miR156b and some of its targets included GmSPL9b and GmSPL9b (Table S12). Consistent with expectations, there was a negative relationship at expression level between gma-miR156b and GmSPL9a or GmSPL9b during early flower bud development (Fig. 6). In addition, our present study found a set of GmPPR genes was targeted by gma-miR4413b, which is a nonconserved miRNA and only identified in soybean now (miRBase 21.0 and Table S8). Most importantly, both transcriptome data and qRT-PCR analysis found gma-miR4413b was upregulated and one of the GmPPR genes was downregulated in soybean CMS, simultaneously (Fig. 5c, d). This finding indicated the possible existence of a crosstalk network between miRNA–target module and flower bud development of soybean CMS (Fig. 7).

Fig. 5.

Relative expression level of gma-miR156b/GmSPL9a and gma-miR4413b/GmPPR between NJCMS1A and NJCMS1C. a, c, e Relative expression level of gma-miR156b/GmSPL9a and gma-miR4413b/GmPPR in small RNA sequencing and transcriptome analysis, respectively. b, d, f Relative expression level of gma-miR156b/GmSPL9a and gma-miR4413b/GmPPR in qRT-PCR analysis

Fig. 6.

Relative expression level of gma-miR156b, GmSPL9a and GmSPL9b in flower bud of three periods. a The flower buds of three periods that were sampled for qRT-PCR analysis. b–d Relative expression level of gma-mi156b, GmSPL9a and GmSPL9b in flower bud of soybean CMS

Fig. 7.

The speculated model of miRNA–targets regulatory network of soybean CMS during flower bud development. The up-regulated miRNAs and down-regulated mRNAs are in red and green, respectively. Solid lines indicate a pathway supported by experimental evidence in this study. Dotted lines show a speculated pathway supported by reports from literature. Pictures in left were morphology of anthers during flowering day in NJCMS1A and NJCMS1C, which exhibited anther indehiscence and dehiscence, respectively. Pictures in right were I₂–KI staining of pollen during flowering day in NJCMS1A and NJCMS1C, whose pollen were sterile and fertile, respectively

Discussion

Exploration of miRNAs and their targets participating in flower bud development of soybean CMS

In plants, many miRNAs seem to be universally expressed across diverse tissues, including the flower bud or pollen. Previous reports have found 100 known miRNAs in the Rf3 and rf3 pollens of maize by solexa sequencing (Yu et al. 2013). In this study, 499 known pre-miRNAs corresponding to 558 mature miRNAs and 10 novel miRNAs were identified in flower buds of the CMS line and its restorer line in soybean. Among all these miRNAs, 76 differentially expressed miRNAs with more than two-fold change between the flower buds of NJCMS1A and NJCMS1C were identified (Table S6). We speculate that these miRNAs might be involved in a specific process of flower bud development. By degradome analysis, 466 targets were chosen and predicted to be cleaved by 200 miRNAs (Table S7). Among the identified targets of miRNAs, some have previously been shown to be involved in floral organ or pollen development, such as miR156 with SPL (Xing et al. 2010), miR172 with APETALA2 (Aukerman and Sakai 2003), miR159 with GAMYB or GAMYB-like genes (Alonso-Peral et al. 2010), and so on.

Moreover, we detected 174 miRNA members in which base editing had occurred (Table S5). Wei et al. (2009) divided the miRNA sequence into three major areas; positions 2–8 and 9–16 of a mature miRNA were called the seed region and region A, respectively, and the remainder was denoted region B. miRNA editing events were first found in animals (Wei et al. 2009) and subsequently revealed to also occur during plant development (Yan et al. 2015). Previous report showed that the seed region is used for target recognition; if the seed region and region A or the seed region and region B work together, the miRNA can regulate different species-specific targets (Wei et al. 2009). As miRNAs function in plants by either target cleavage or translational repression, editing of the 10th/11th positions of miRNA will also lead to altered targets. To verify the above results, we combined the analysis of the degradome library and base-edited miRNAs (seed region and the 10th/11th positions of miRNA). The results showed that most of the targets of base-edited miRNAs were the same as those of the unedited miRNAs (Table S10). However, as the base editing occurred, some targets were filtered. For example, Glyma.08G122000.2 was targeted by gma-miR2118a-3p and gma-miR2118b-3p, but when the base G at position 6 was changed to T, it no longer targeted Glyma.08G122000.2 (Table S10). On the other hand, when the base C at position 11 of gma-miR2118a-3p and gma-miR2118b-3p was changed to T, we identified a new target (Glyma.08G005700.1) by degradome analysis (Table S10).

Several miRNA regulatory networks might be involved in regulation of flower bud development of soybean CMS

It is widely accepted that CMS is caused by mitochondrial genes with coupled nuclear genes, and the male fertility of the F₁ plants can be restored by the Rf gene(s) that comes from the nuclear genome of the restorer line (Chen and Liu 2014). In this process, the interaction between genes is very obvious, the nuclear genes might change their expression as target genes in response to the mitochondrial signaling, which subsequently affects flower bud development. In this study, altered expression of miRNAs and their targets genes was exhibited between NJCMS1A and NJCMS1C, demonstrating important roles of miRNAs regulation network in flower bud development of soybean CMS. For example, GmNF-YA3/GmNF-YA5 and GmLAC4 were target genes of gma-miR169 and gma-miR397 (Table S11), respectively. Transcriptome sequencing showed these target genes were all more than twofold change expressed in NJCMS1A (Table S11), and previous studies reported their overexpression affects male gametogenesis and induces nondehiscent anthers in plant (Mu et al. 2013; Zhang et al. 2014; Nasrin et al. 2010). Based on the expression analysis of miRNAs and their corresponding targets, gma-miR156b/GmSPL9 and gma-miR4413b/GmPPR regulatory network was selected as two candidate miRNAs–targets combinations that might be involved in flower bud development of soybean CMS.

MIR156 and SPL are one of the highly conserved miRNA and gene families in plant, respectively, and different members of MIR156 and SPL showed highly dynamic expression patterns at different stages of flower development in Arabidopsis (Xing et al. 2010). Former studies have shown that miR156 plays essential regulatory roles in floral development and male fertility by targeting SPL genes, and the loss of function of miR156/7-targeted SPL genes led to a semi-sterile phenotype in Arabidopsis (Xing et al. 2010). Yu et al. (2013) reported that miR156 had a higher expression level in the pollen of S-type CMS in maize. Moreover, it had higher expression levels in the anthers from the three anther developmental stages of the GMS mutant in cotton (Wei et al. 2013). In Arabidopsis, the pollen fertility of triple mutant spl8 spl9 spl15, spl8 spl2 spl9, and quadruple mutant spl8 spl2 spl9 spl15 was lower than that of the spl8 single mutant (Xing et al. 2010), which provided evidence that the AtSPL9 do affect pollen fertility. Most importantly, overexpression of AtSPL9 in the spl8 mutant could restore its fertility partially (Xing et al. 2010), indicating that AtSPL9 is crucial to pollen development. It has been reported that SPL9 is essential for anther development, and overexpression of miR157 in cotton leads to microspore abortion and anther indehiscence under high-temperature stress by regulated GhSPL9, etc. (Ding et al. 2017). In this study, gma-miR156b negatively regulated GmSPL9a that displayed differential expression between NJCMS1A and NJCMS1C, indicating that miR156/SPL9 regulatory network may also participate in flower bud development of soybean (Fig. 7).

In present study, a total of 4 GmSPL9 genes were predicted to be targeted by gma-miR156b (Table S12). However, only GmSPL9a reached more than 1.5-fold expression difference between NJCMS1A and NJCMS1C (Table S11). Previous studies indicated that SPL gene plays a key role in early anther development by promoting other early anther genes required for cell division, specification, and differentiation, resulting in pollen mother cell and tapetum formation (Schiefthaler et al. 1999; Yang et al. 1999; Xing et al. 2010). So we also speculated that GmSPL9 genes were also involved in flower bud development of soybean CMS in early stage of flower bud development. To verify this view, we used qRT-PCR to measure the expression of gma-miR156b, GmSPL9a and GmSPL9b at three periods during flower bud development of soybean. As a result, gma-miR156b was upregulated in the period I and II, and GmSPL9a and GmSPL9b were downregulated before period III and II (Fig. 6), respectively, which means that the targeted GmSPL9 genes exhibit an opposite expression pattern with that of gma-miR156b during certain stage of flower bud development (Fig. 6). Thus, miR156 and SPL9 may also be absolutely required for proper early flower bud development and cell proliferation and differentiation in soybean CMS, which warrants further study.

PPR proteins are classified as Rf proteins that suppress CMS gene expression through posttranscriptional mechanisms such as editing, cleavage, and degradation of the target mRNAs in many plants such as rice, Brassica, radish, and sorghum (Chen and Liu 2014; Brown et al. 2003; Hu et al. 2012; Iwabuchi et al. 1999; Jordan et al. 2010; Kazama and Toriyama 2003). Recently, Wang et al. (2016c) identified the Rf gene for soybean M-type CMS, Rf-m, which is located in a PPR gene-rich region on chromosome 16. This region contains 19 putative open reading frames (ORFs), and seven of them encode putative PPR proteins. Two of these (Glyma.16G161900 and Glyma.16G163100) were targeted by gma-miR4413b. Gma-miR4413b is a soybean-specific miRNA and was upregulated (1.5-fold) in NJCMS1A, and its target genes were all PPR proteins (Table S8). Moreover, we found Glyma.16G161900 was significantly downregulated in NJCMS1A according to the transcriptome data in our lab (Li et al. 2017a, b). Quantitative real-time PCR also showed that gma-miR4413b was upregulated in NJCMS1A, and Glyma.16G161900 was only expressed in NJCMS1C (Fig. 5d). In addition, the Glyma.16G161900 was placed in the mitochondria (Table S8) and was homologous with RPF1, which is a PPR gene and supposed to play a role in fertility recovery of Arabidopsis CMS by editing nad4 mRNA in mitochondria (Hölzle et al. 2011). Therefore, we speculated that the upregulation of gma-miR4413b reduced the expression level of GmPPR in soybean CMS, so that CMS gene can be expressed normally and lead to pollen sterility, eventually. All these results indicated that gma-miR4413b might play a specific function during the formation of soybean CMS (Fig. 7), which warrants further study.

Conclusion

In this study, a large number of miRNAs were identified during flower bud development in the soybean CMS line NJCMS1A and its restorer line NJCMS1C. By deep sequencing, 558 known miRNAs, 10 novel miRNAs, and 174 base-edited miRNA members were identified. Among the identified miRNAs, 76 differentially expressed miRNAs were discovered with greater than twofold changes between NJCMS1A and NJCMS1C. By degradome analysis, a total of 466 distinct transcripts targeted by 200 miRNAs and 122 distinct transcripts targeted by 307 base-edited miRNAs were detected. Small RNA sequencing, transcriptome and qRT-PCR comparative analysis found that gma-miR156b/GmSPL9 and gma-miR4413b/GmPPR regulatory networks might be involved in flower bud development of soybean CMS. Further functional studies on these differentially expressed miRNAs will need to provide a better understanding of the miRNA-mediated regulation mechanisms during the CMS occurrence in the soybean N8855-CMS line.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

CMS

Cytoplasmic male sterility

MFE

Minimum free energies

MFEI

Minimal folding energy indices

miRNA

MicroRNA

PPR

Pentatricopeptide repeat-containing

qRT-PCR

Quantitative real time PCR

Rf

Restorer of fertility

SPL

Squamosa promoter-binding protein-like

Author Contributions

SPY, JYG and XLD conceived and designed the experiments. XLD, HZ, HR, YWL, LFC, TLW, LJ and XQL performed the experiments. XLD analyzed the data. XLD, HZ and HR contributed reagents/materials/analysis tools. XLD, YWL, LFC, TLW, LJ and XQL conceived the qRT-PCR experiments and analyzed the data. XLD, SPY conceived the experiments and wrote the manuscript. All authors read and approved the final manuscript.

Data Availability

All of the small RNA-seq and degradome-seq data were submitted to the National Center for Biotechnology Information (NCBI) under the accession number PRJNA304685 and PRJNA504545.

Conflict of interest

All the authors declare that they do not have conflict of interest.

Ethical approval

This article does not contain any study with human subjects or animals performed by any of the authors.