A total of 15 E. sibiricus accessions were used in this study. Most of them were from the species range in China, comprising cultivars and wild accessions (Table 1). Six plant introduction (PI) accessions were originally obtained from the U.S. Department of Agriculture Germplasm Resources Information Network (GRIN). Other accessions were obtained from the State Key laboratory of Grassland Agro-ecosystems, Sichuan Agricultural University, and Sichuan Academy of Grassland Science. The seeds of each accession were germinated in plastic boxes with moistened blotter paper at room temperature. After germination, seedlings were maintained in a greenhouse under a 25/15°C day/night temperature regimes until they were 8 weeks old. Ten individual plants of each accession were transplanted to the experimental field in Yuzhong campus of Lanzhou University, Gansu, China (latitude 35°34′ N, longitude 103°34′ E, elevation 1720 m). After transplanting plants were well watered immediately and no fertilizer was applied to the plants afterward. To avoid potential seed contaminants, all accessions used in this study were previously identified based on some important phenotypical characteristics such as inflorescence, stem, leaf, and seed. Particularly, this species is small-anthered and long-awned bunchgrass.

The SS of 15 accessions were determined by measuring BTS that required to detach seeds from pedicels (Li et al., 2006), which is inversely proportional to SS degree. Thirty randomly chosen florets from five inflorescences of each plant were examined at each of the five developmental stages, 0, 7, 14, 21, and 28 days after heading (DAH), and their average BTS values were calculated. Other four seed related traits, including SL, SW, AL, KW were measured according to the methods described by Zhao et al. (2015). The mean value, standard deviation (SD), coefficient of variation (CV) of each trait, and correlation analysis among different traits were carried out using Statistical Package for Social Sciences program (SPSS 19.0, SPSS Inc., USA) software. According to the mean value of each trait for each accession, a heatmap was constructed using the Heatmap Illustrator (HemI 1.0) program (Deng et al., 2014). All phenotypic data were normalized under the logarithmic relations at first, due to the different conditions of measurement, then used to map the visualize color matrix. Each line and column represented different accessions and different traits, respectively. Hierarchical clustering was conducted based on average linkage method and the similarity metric with Pearson distance (Deng et al., 2014).

Histological analysis of flower-pedicel structure was carried out at the same five development stages concurrent with SS measurements. In order to reduce variation due to the spikelet position at each developmental stage, the three central spikelets of each florescence were used. The pedicels of each accession were fixed in FAA solution [ethylalcohol (70%): glacial acetic acid: formaldehyde (37∼40%) = 90 : 5 : 5)] (Glime, 2013; Kantar et al., 2016) after vacuum dried for 10 min in a vacuum drier and stored at 4°C in 15 M ethanol. They were then dehydrated in a gradient of ethanol solutions (50, 70, 90, and 100%) for 60 min (Htun et al., 2014). After transparency in the dimethylbenzene and soaking in the paraffin, paraffin-embedded pedicels were sectioned longitudinally to a thickness of 8 μm by the paraffin slicing machine (Jinhua YIDI Medical Appliance CO., LTD, Zhejiang, China). The sections were stained for 20 min with safranin, then 2 min with fast Green (Zhongtai, Shanghai, China). After staining, the pedicel structures were examined using a fluorescence microscope (Olympus Corporation, Japan). In order to detect the relationship between abscission layer development and SS degree at each of five developmental stages, the pedicel junctions after detachments of seed were examined by scanning electron microscopy (Konishi et al., 2006).

A total of 15 accessions were sampled and used for genetic diversity analysis using 20 EST-SSR primers. Genomic DNA was extracted from the young leaves according to an SDS method (Shan et al., 2011). The quantity and quality of DNA samples were detected using NanoDrop ND1000 spectrophotometer (Thermo Scientific, USA) and agarose gel electrophoresis, then diluted to 25 ng/μL and stored at -20°. The PCR amplification, SSR genotyping as well as data analysis were carried out using the methods described by Xie et al. (2015). Structure analysis of the 15 accessions was performed using the methods described by Zhang et al. (2016).

The AZ tissues were harvested according to methods described by Li et al. (2006). Each collected flower-pedicel structure consisted of an approximately 1 mm region of the pedicel and 1.5 mm of the flower. The CE and PG were extracted according to the methods described by Awad and Lewis (1980) and Amid et al. (2016). The enzyme activity was assayed in AZs at the same five developmental stages used for BTS and histological analyses, following the manuscript’s protocol of plant CE ELISA kit and plant PG ELISA kit (Shanghai Enzyme-linked Biotechnology Co., Ltd., China), respectively.

Abscission zone tissues of the two accessions (PI655199 and LQ04) with contrasting SS degree were collected at two of the five developmental stages: 21 days and 28 DAH. The two stages were selected based on results of SS, histological as well as enzyme activity analysis. Each collected flower-pedicel structure consisted of an approximately 1- mm region of the pedicel and 1.5 mm of the flower, which included the AZ (Li et al., 2006). Approximately 30 mg of this AZ tissue was collected. Total RNA was extracted using Plant total RNA Kit (TIANGEN, Beijing, China) according to the manufacturer’s instructions. RNA concentration and quality was measured using an Agilent 2100 Bioanalyzer (Agilent Technologies, Inc., Waldbronn, Germany). Based on the our previous transcriptome data of AZ in E. sibiricus (data not published), five candidate genes: PAL, GLU, CesA, PG and XIP1 involved in SS were selected for the gene expression analysis. Gene-specific primers were designed using Primer Express software (Applied Biosystems) and were shown in Supplementary Table S1. The SYBR Premix Ex Taq^TM II quantitative PCR system (Takara, Dalian) was applied for qRT-PCR analysis, following the manufacturer’s instructions, and reactions occurred on a Bio-Rad iQ5 real-time PCR instrument (Bio-Rad, Hercules, CA, USA). Expression levels of these candidates were calculated relative to reference gene GAPDH using the 2^-ΔΔCt method (Livak and Schmittgen, 2001).

The morphological variation and genetic diversity of 15 E. sibiricus accessions were studied using five seed related traits: SS, SL, SW, AL and KW, and 20 EST-SSR markers, respectively. Three of five traits: SS, SW and KW had the coefficient variation (CV) value more than 30%. The greatest morphological variation was found for SS (CV = 64.55%), followed by SW (CV = 45.89%), KW (CV = 40.85%), AL (CV = 21.76%) and SL (CV = 14.19%) (Table 2). SS showed significant correlation with SL, AL and KW (Table 3). The time course changes in SS degree of 15 E. sibiricus accessions indicated that BTS value differed at each developmental stage (Figures 1A–C). In general, although there was a decline of BTS at 7 DAH for wild accessions, BTS value of cultivars maintained generally a stable level from 0 DAH to 14 DAH. Notably, the average BTS value of wild accessions decreased quickly at 21 DAH, and dropped below 40 gf, while the average BTS value of cultivars remained above 75 gf. At 28 DAH, the BTS value of cultivars varied from 17.51 gf to 53.85 gf, with a mean of 36.78 gf. The BTS value of wild accessions ranged from 9.48 gf to 32.33 gf, with an average of 17.37 gf. Generally, cultivars had higher BTS value than wild accessions, of which PI655199 had the highest average BTS value (144.51 gf) and LQ04 had the lowest average BTS value (47.17 gf) across the whole developmental stages. Clustering analysis based on seed traits can divided most cultivars and wild accessions into two different groups (Figure 2). The dendrogram was concordant with morphological variability.

The high level of morphological variation was similar to genetic diversity detected by EST-SSR markers. A total of 20 EST-SSR primers amplified 249 bands, with an average of 12.45 per primer. The percentage of polymorphic bands (PPB) was 94.78%, suggesting a high degree of genetic diversity among these accessions (Table 4). The value of genetic similarity among accessions ranged from 0.51 (between PI655199 and Y1005) to 0.90 (between “Chuancao No.2” and “Hongyuan”), with an average of 0.72 (Table 5). Structure analysis showed all accessions were assigned into three major groups (Figure 3). In general, those data suggested rich morphological and genetic variation, and provided valuable germplasm resources from which the mechanisms of SS may be investigated.

In order to further detect the relationship between SS and degradation degree of abscission layer, we conducted the histological analysis of pedicel structure of 15 E. sibiricus accessions with SS variation. Anatomical investigation with longitudinal sections indicated abscission layers were already present at heading in all 15 E. sibiricus accessions. They occurred on both sides of vascular bundle, which could be stained dark red by safranin (Figure 4A). The shape and arrangement orientation of abscission layers were different when compared with surrounding cells. Generally, abscission layers had two or three layers of cells that had an elliptic shape and a well-organized shape. In addition, the cell size of abscission layer was smaller than that of the parenchyma cells in the rachilla. For most accessions, degradation of abscission layers occurred at 14 DAH. LQ04 had more serious degradation of abscission layers than low SS genotype PI655199 at 28 DAH. A well-defined boundary between pedicel and flower was present in high SS genotype at 21 and 28 DAH (Figure 4B, Ia, IIb). Additionally, scanning electron microscopy showed a smooth fracture surface on the rachilla in high SS genotype at 21 and 28 DAH while rough and irregular surface was observed in low SS genotype (Figure 4B, Ic, IId). Based on these staining results, lignin content increased along with five developmental stages, there was more lignin in the AZ and surrounding pedicel tissue of low SS genotype PI655199 than in high SS genotype LQ04 at 21 and 28 DAH (Figure 4B, Ie, IIf).

Specific activity of CE varied among 15 E. sibiricus accessions at different stages during seed development (Figures 5A–C). Generally, CE exhibited a different trend of activity between cultivars and wild accessions. The biggest differences of CE activity between cultivars and wild accessions were observed at 14 and 21 DAH. For cultivars, the mean CE activity was higher (181 U/L) than wild accessions (122 U/L) at 14 DAH while that activity was lower (144 U/L) than wild accessions (210 U/L) at 21 DAH (Figure 5A).

During the first 7 days after heading, the CE activity value was similar between cultivars and wild accessions and were maintained at more than 165 U/L (Figure 5A). The CE value of cultivars and wild accessions began to decrease after 7 DAH. The CE value of wild accessions began to increase rapidly at 14 DAH, and reached the highest value (210 U/L) at 21 DAH and remained above 150 U/L at 28 DAH. In comparison, the CE value of cultivars decreased quickly at 7 DAH, and dropped below 150 U/L at 21 DAH. At the same stage, wild accessions had the highest CE value.

Similarly, specific activity of PG varied among 15 E. sibiricus accessions at different stages during seed development (Figures 6A–C). The biggest differences of PG activity between cultivars and wild accessions were found at 0 and 14 DAH. For the cultivars, the mean PG activity was higher (132 pg/mL and 142 pg/mL) than wild accessions (76 pg/mL and 101 pg/mL) at 0 and 14 DAH. But, at 21 and 28 days after heading, the average PG activity of wild accessions was higher than cultivars, and their value were between 97 pg/mL and 112 pg/mL (Figure 6B).

Generally, PG and CE activity varied at each of five developmental stages for both cultivars and wild accessions, but they exhibited similar trend of activity during seed development (Figures 7A,B), indicating their roles and interaction in differentiation of SS. High SS accessions showed higher CE and PG activity at seed maturity stage (21 and 28 DAH). For example, high SS genotype LQ04 (CE = 176 U/L, PG = 159 pg/mL) had higher CE and PG activity than low SS genotype PI655199 (CE = 47 U/L, PG = 49 pg/mL) at 21 DAH.

Five SS candidate genes including PAL, GLU, XIP1 and CesA and PG were selected for expression analysis. The results showed that all these genes were differentially expressed in the AZ at two development stages, 21 day and 28 DAH (Figure 8). We used High-21 as a benchmark for relative expression analysis. CesA were down-regulated in low SS genotype PI655199 at 28 DAH (Low-28). The expression of other four genes (PAL, GLU, XIP1 and PG) was up-regulated in Low-28. The relative expression of PG for high SS genotype LQ04 was higher than that of low SS genotype PI655199 at 28 DAH. In particular, the relative expression of XIP1 for PI655199 at 28DAH was almost 116 times higher than that of LQ04 at 21 DAH (Low-21). These results indicated the involvement and role of these candidate genes in SS.

Seed shattering is a commonly observed trait in wild grass species and many forage cultivars (Zhao et al., 2015). This study showed wide variation in SS among 15 accessions. The BTS value decreased quickly after 14 DAH, and reached the lowest BTS value at 28 DAH. Cultivars had higher average BTS value than wild accessions during seed development. Various direct and indirect factors influencing SS were reported, i.e., the size, strength, shape, and the flexibility of the glumes and compactness of the inflorescence and spikelets (McWilliam, 1980). In this study, we found a significant correlation between SS and three seed related traits: AL and KW and SL, similar to previous reports. The awn is a morphological characteristic found in most cereal crops, such as rice, wheat, barley, and sorghum. In wild plants, long awns have been reported to aid in seed dispersal, however, long awn is disadvantageous for grain harvest (Magwa et al., 2016). In practice, high yielding plants are likely more susceptible to shattering, if harvest is delayed (Elgersma et al., 1988). Many studies also indicated the expression of SS may be modified by environmental factors such as a deficit or surplus of water, high/low temperature (Taylor and Whitelaw, 2001; Taghizadeh et al., 2009). Consequently, in practice wide germplasm collections, recurrent selection with multiple location and year testing, large seed field trials should be important to selection for seed retention and improvement of SS in this species.

Previous studies have showed that abscission was generally caused by the development of abscission layer, and revealed a correlation between the SS and the degree of degradation for abscission layers (Jin et al., 1982; Ebata and Tashiro, 1990; Jin et al., 1992; Fukuta et al., 1994a,b; Oba et al., 1995; Watanabe et al., 2003; Konishi et al., 2006; Li et al., 2006; Akasaka et al., 2011). In rice, abscission layers were described as a band of small cells that begin to form before heading and finish cell expansion to maximum size at the stage of heading (Jin, 1986; Oba et al., 1995; Roberts et al., 2002; Watanabe et al., 2003; Lin et al., 2007). Based on our results abscission layers were already present at heading in E. sibiricus, and included two small cell layers, which were considered to be more advantageous to cell separation and breakage formation (Gubert et al., 2014). For all accessions tested, seed didn’t shed at heading stage but began to shed at 21 DAH. Histological analysis of AZ at 21 and 28 DAH showed different degradation degree among different accessions. High SS accessions had a smooth fracture surface of rachilla when compared with low SS genotypes, suggesting the high level of degradation. This may resulted from the increased hydrolytic enzymes activity found in AZ. Abscission is related to cleavage and degradation of cell wall components by two major cell wall hydrolytic enzymes PG and CE (Fuller et al., 2010; Liljegren, 2012; Estornell et al., 2013; Tada et al., 2015). PG and CE have a crucial effect on the degradation of abscission layer, which can contribute to the shattering of seed and other organs (Sexton et al., 1980; Kalaitzis et al., 1997; Agrawal et al., 2002). PG have been implicated as positive regulators of cell separation, fruit ripening, abscission, cell growth and dehiscence (Jenkins et al., 1999; Sander et al., 2001; Atkinson et al., 2002, 2012; González-Carranza et al., 2007; Xiao et al., 2014). Typically, a research reported that decreasing PG activity could reduce seed loss prior to harvest (Swain et al., 2011).

In our study, PG and CE activity varied across five developmental stages. Similar trend of activity was found in pod shattering process of soybean (Agrawal et al., 2002). Wild accessions had higher PG and CE activity when compared with cultivars at seed physiological maturity, indicating the involvement and roles of PG and CE in SS difference among cultivars and wild accessions. A balance and interaction between PG, CE and other enzymes may be the key factor that regulates and determines the abscission process. Previous studies showed PG can hydrolyse specially pectin which is a dominant component and a fiber-enforced composite material in plant cell wall and has important consequences for cell adhesion (Chanliaud and Gidley, 1999; Wen et al., 1999; Willats et al., 2001; Daher and Braybrook, 2015). Then the decreased PH resulted from cell wall pectin hydrolyzation, in turn, might pose several direct potentially and indirect consequences for wall properties and other enzyme activities (Willats et al., 2001; Pelloux et al., 2007; Wolf and Greiner, 2012). Thus, the consequences might elevate CE activity and give rise to a stronger hydrolyzation and degradation of abscission layer cell wall. Our results indicated that CE and PG exhibited similar trend of activity during seed development, which could confirm partially an interaction between PG and CE in hydrolyzing cell wall.

Lignin is a highly branched polymer of phenylpropanoid compounds, and a component of the plant cell wall. The secondary cell wall provides the strength necessary to make cells suitable for transport and support which is imparted by the presence of lignins in the secondary cell walls (Barceló, 1997; Rogers and Campbell, 2004). In grass species, lignin comprises approximately 20% of the secondary cell wall, filling pores between the polysaccharides (Vogel, 2008). Lignin plays an important roles in plant growth, development and defense responses. Lignin deposition generally occurs when cell growth is completed and the cell wall undergoes secondary thickening (Rogers and Campbell, 2004; Wang et al., 2013). Plants are exposed to different stress, which may change lignin content and composition (Moura et al., 2010). It has been believed that lignification is a mechanism for stress defense in plants, particularly for plant–microbe interactions, this has been suggested as defense responses of host plants to the stress (Bhuiyan et al., 2009). Lignin is also identified as a major factor in recalcitrance of cell walls to digestion, particularly during enzymatic hydrolysis (Chen and Dixon, 2007). A previous study in rice also showed that SS can be induced by inhibiting lignin biosynthesis, and decreasing lignin levels in the AZ and surrounding pedicel tissues could lead to an increase in SS (Yoon et al., 2014). At seed physiological maturity, the total seed weight pressing the underlying abscission layers increases, which may result in easier breaking of abscission layers with lower lignin content. In this study, staining of pedicels of two contrasting genotypes at 21 and 28 days after heading showed that lignin deposition was much lower in high SS accession than in low SS accession. Meanwhile, high SS accession had lower BTS value when compared with low SS accession. These results implied higher lignin deposition found in low SS genotype may play a role in resistance to SS.

The shattering habit is a complex polygenic trait that is controlled by many genes (Konishi et al., 2006). Previous studies showed many glycosyl hydrolase family genes play an important role in modifying plant cell wall structure and component during tissue development (Zhou et al., 2006). CE (1,4,-β-glucanase), the first hydrolytic enzyme reported, play a critical role in wall loosening during abscission (Roberts et al., 2002). In rice, the gene OsCel9D (synonym OsGLU1) encodes an endo-1,4,-β-glucanase gene with cellulose function, and mutations of this gene reduce cell elongation and cellulose content, and increase the pectin content, therefore hampering the abscission process in SS (Zhou et al., 2006). Xylan is the major component of hemicelluloses. Xylanase inhibitors (XIs) can inhibited the activity of xylanase, which catalyze the hydrolysis of the β-1,4-xylosidic bonds in xylan (Xin et al., 2014). The low SS genotype PI655199 showed much higher expression of XIP1 when compared with high SS genotype. This indicated that the expression of this gene might be associated with a reduction of SS. PAL is the key gene required for monolignol biosynthesis, suppression of this gene took place early in the monolignol biosynthetic pathway, which causes significant reduction in lignin content (Chen and Dixon, 2007). A similar result was found in this study, where PAL was down-regulated and lignin content was relatively lower in high SS genotype LQ04, corresponding to increased SS. The different expression of these candidate genes indicated the involvement and role in SS, but which genes played a key role in difference of SS in E. sibiricus still remains unknown.