Proteomic Analysis of Silk Viability in Maize Inbred Lines and Their Corresponding Hybrids

A long period of silk viability is critical for a good seed setting rate in maize (Zea mays L.), especially for inbred lines and hybrids with a long interval between anthesis and silking. To explore the molecular mechanism of silk viability and its heterosis, three inbred lines with different silk viability characteristics (Xun928, Lx9801, and Zong3) and their two hybrids (Xun928×Zong3 and Lx9801×Zong3) were analyzed at different developmental stages by a proteomic method. The differentially accumulated proteins were identified by mass spectrometry and classified into metabolism, protein biosynthesis and folding, signal transduction and hormone homeostasis, stress and defense responses, and cellular processes. Proteins involved in nutrient (methionine) and energy (ATP) supply, which support the pollen tube growth in the silk, were important for silk viability and its heterosis. The additive and dominant effects at a single locus, as well as complex epistatic interactions at two or more loci in metabolic pathways, were the primary contributors for mid-parent heterosis of silk viability. Additionally, the proteins involved in the metabolism of anthocyanins, which indirectly negatively regulate local hormone accumulation, were also important for the mid-parent heterosis of silk viability. These results also might imply the developmental dependence of heterosis, because many of the differentially accumulated proteins made distinct contributions to the heterosis of silk viability at specific developmental stages.


Introduction
The maize silk is functionally equivalent to the stigma and style of a typical pistil. It is a specialized elongated tissue that begins to senesce about 8-10 days after it emerges from the husks [1]. In normal conditions, pollination is completed within 1-2 days after silk emergence. Thus, a period of 8-10 days of silk viability is normally sufficient for seed setting. However, in hybrid seed production where plants are emasculated, a female parent with a long period of silk viability is critical for a good seed set. Thus, elucidating the molecular mechanisms of silk viability is necessary for the production of elite inbred lines and hybrid selection, which in turn contribute to maize hybrid seed production and field production. five sampling stages, D 4 , D 6 , D 8 , D 10 , and D 12 , the average seed setting rate of three biological replications for the inbred lines Xun928 and Lx9801 was 99.3%, 97.4%, 95.0%, 26.0%, 13.5%, and 96.3%, 89.3%, 78.8%, 16.3%, 1.9%, respectively. The inbred line Zong3 sustained a high seed setting rate of 100% for all sampling stages. The ANOVA results showed that the difference of seed setting rate was significant between the parental inbred lines and their corresponding hybrids (P < 0.05), except for D 4 between the inbred line Xun928 (P = 0.079), Zong3, and their hybrid combination Xun928×Zong3. The different sampling stages of Xun928, Lx9801, Xun928×Zong3, and Lx9801×Zong3 also showed significant differences (P < 0.01). Thus, least-significant difference (LSD) multiple comparisons were performed and showed that the seed setting rate between each sampling stage was significantly different both for Xun928 and Lx9801. However, non-significant difference was detected between D 4 and D 6 both for the two hybrids Xun928×Zong3 and Lx9801×Zong3.
The seed setting rate significantly decreased from D 8 to D 10 in both Xun928 and Lx9801 and the two hybrids. However, the decrease in the seed setting rate was slower in the hybrids than in the two inbred lines because of heterosis in the hybrids. Compared with those of the parental inbred lines, the seed setting rate of the hybrids fell between the mid-parent and highparent values; i. e, seed setting rate showed partial dominant heterosis. Based on the phenotype of seed setting rate and heterostic degree, only the sampling stages with a significant difference at the 0.01 level were used for the proteomic analysis. Thus, the hybrids at D 8 , D 10 , and D 12 were used in the proteomic analysis of heterosis, and the inbred lines Xun928, Lx9801, and Zong3 at stages D 6 , D 8 , D 10 , and D 12 were used in the proteomic analysis of silk viability (Table 1).

Differentially accumulated proteins
For the 2-DE analysis, only protein spots that showed the same trend in the three biological replicates were retrieved (S1 and S2 Figs). After normalization and ANOVA, only 3, 7, and 16 differentially accumulated protein spots were obtained for the inbred lines Xun928, Lx9801, and Zong3, respectively. These protein spots, which showed maximum changes more than 1.5-fold (P < 0.05) during the four sampling stages, were manually excised and analyzed by MS (Table 2 and S2 Table). Among the 26 differentially accumulated protein spots, 17 and 7 protein spots showed the lowest and the highest levels at D 6 , respectively (S3 Table).  6 , D 8 , D 10 , and D 12 represents days after silk emerged above ligule of the husk outer leaf. Setting percentage was calculated by averaging three biological replications and each replication. Each replication consisted of 10 intact ears. g : P-value is significance level among different sampling stages. h : P-value is significance level between hybrids Xun928×Zong3 and its two parental lines at each sampling stage. i : P-value is significance level between hybrids Lx9801×Zong3 and its two parental lines at each sampling stage. *: Significant at 0.05 level. **: Significant at 0.01 level.
Meanwhile, 8 and 4 out of the 14 protein spots corresponding to D 6 showed the highest and the lowest levels at D 12 , respectively. Among them, protein spot 46 gradually accumulated during silk development, while protein spot 61 gradually diminished. For the heterosis analysis, 46, 47, and 37 protein spots with maximum changes of more than two-fold (P < 0.05) between the hybrid Xun928×Zong3 and its corresponding parents were retrieved at D 8 , D 10 , and D 12 , respectively (Table 3 and S2 Table). The corresponding numbers of protein spots with more than two-fold changes between the hybrid Lx9801×Zong3 and its two parents were 24, 37, and 24, respectively. Out of the 215 differentially accumulated proteins, about 57% (122 protein spots) were additively accumulated and 43% (93 protein spots) were non-additively accumulated in the two hybrids. Among the non-additively accumulated proteins, five interaction patterns were observed (Table 3); "−", "− −", "+", "+ −", and "+ +", accounting for about 34% (32 protein spots), 3% (3 protein spots), 42% (39 protein spots), 12% (11 protein spots), and 9% (8 protein spots) of the non-additively accumulated proteins, respectively. The "− −" pattern was only detected in the hybrid Xun928×Zong3 at D 12 , and the "+ +" pattern was only detected at D 10 and D 12 in the two hybrids. The "+ −" pattern was found in the hybrid Xun928×Zong3 at D 8 and D 12 and the hybrid Lx9801×Zong3 at D 10 and D 12 . The other two major non-additive accumulation patterns "+" and "−" were well distributed across the three sampling stages in each hybrid.

Differentially accumulated proteins identified as important for silk viability and its heterosis
Three proteins differentially accumulated during silk development were identified, including gi|413944345 (protein spot 63 in Zong3), gi|414869037 (protein spot 8 in Xun928), and gi| 195635735 (protein spot 16 in Xun928). These three proteins also differentially regulated the heterosis of silk viability in the two hybrids (Tables 2 and 3). gi|413944345, which differentially regulated silk development in the common paternal line Zong3, showed differential accumulation in the two hybrids Xun928×Zong3 and Lx9801×Zong3 at almost all sampling stages. gi| 414869037 and gi|195635735, which were specific for silk development in the inbred line Xun928, contributed to the heterosis of silk viability only in the hybrid Xun928×Zong3 (protein spots 155 and 239 at D 10 and D 12 ; protein spot 96 at D 8 ), and not in Lx9801×Zong3. The functional category analysis showed that these three proteins were involved in anthocyanin biosynthesis, methionine metabolism, and suberin biosynthesis.

Functional category and KEGG pathway enrichment classifications of differentially accumulated proteins
The differentially accumulated proteins associated with silk viability and its heterosis was in similar functional categories. Unknown proteins comprised a large proportion of the differentially accumulated proteins, accounting for 38% and 40% of the proteins related to silk viability and its heterosis, respectively. The proteins involved in metabolism group accounted for the largest proportion of the differentially accumulated proteins, accounting for 43% of proteins related to silk viability and 42% of proteins related to the heterosis (Fig 1). Proteins involved in protein biosynthesis and folding, including transcription, translation, folding, sorting and degradation, were the second most abundant group and were specific to the heterosis of silk viability. Other important categories, based on protein abundance, were stress and defense response, plant hormone biosynthesis and signal transduction, and cellular processes. In the largest category, metabolism, there were six and eight subcategories of proteins involved in silk viability and its heterosis, respectively. Among them, methionine metabolism and flavonoid metabolism were important for both silk viability and its heterosis (Tables 2 and 3, Fig 1), and lipid metabolism and energy metabolism were specific to the heterosis of silk viability. The protein-protein interaction networks involved in silk viability and its heterosis were analyzed by searching the String database (Fig 2). Three proteins were implicated in silk viability and its heterosis: gi|413944345 (KOG1192), gi|414869037 (KOG2263) and gi|195635735 (NOG293481). These proteins had only one or two interacting proteins and were distributed at a remote node in the network. Some reductases or dehydrogenases (S4 Table) were located at the interaction nodes and played an important role in the protein-protein interaction networks both for silk viability and its heterosis; for example, KOG1502-KOG2450-KOG0022 in the interaction network for silk viability and KOG1502-KOG1577-KOG2450-KOG0022-KOG 0725 in the interaction network for the heterosis of silk viability. The two important branches for the heterosis of silk viability were ATP energy production (KOG1758-KOG1353-KOG 1350-KOG1626-) and protein metabolism (KOG0177-KOG0179-KOG0863-KOG1688-) ( Fig 2B). Additionally, two glutathione S-transferases (gi|195619648 KOG0406, gi|162460516 KOG0867) might play a crucial role in providing energy and proteins for the entire proteinprotein interaction network for the heterosis of silk viability. Meanwhile, Kinases (KOG1367,          cut-off level were defined as non-additive proteins. Based on this premise, "+" and "-" represented protein spot intensities identified in F 1 hybrids that were similar to the high parent and low parent values, respectively. "+ +" and "--" represented the protein spot intensities identified in F 1 hybrids that were significantly different from the high parent and the low parent values, respectively. "+ -" represented the protein spot intensities identified in F 1 hybrids that fell in between the mid-parent and the high parent or the mid-parent and the low parent values. d Hybrids Xun928×Zong3 and Lx9801×Zong3 are abbreviated as XZ and LZ, respectively. D 8 , D 10 , and D 12 represent sampling stages. e GenBank accession number of protein spot.
f Protein name in NCBI database.
g Proteins scores were derived from ions scores as a non-probabilistic basis for ranking protein hits. h Confidence interval of the identified protein.
i Gene name retrieved from maize sequence (http://ensembl.gramene.org/Zea_mays/Info/Index) by cDNA blast.   Proteins Involved in Silk Vigor and Its Heterosis KOG2440), enolase (KOG2670), and isomerase (KOG1643) increased the complexity of the protein-protein interaction network for the heterosis of silk viability compared with the network for silk viability, which was complicated by cell cytoskeleton proteins (Fig 2A).

Discussion
Comparison of protein categories related to phenotypes of the inbred lines and their corresponding hybrids The results in this study revealed that several functional categories of proteins corresponded to the seed setting rate phenotype of the inbred lines and its corresponding hybrids. For the inbred lines, proteins involved in flavonoid metabolism, methionine metabolism and cytokinin signaling, made the highest contributions to contributed the highest silk viability in the inbred line Zong3 (Table 2). Compared to the inbred line Lx9801, the stronger silk viability of the inbred line Xun928 was attributed to proteins involved in methionine metabolism, and these proteins also contributed to the heterosis of silk viability in the hybrids, but the proportions of their contributions differed. Proteins involved in flavonoid metabolism were important for heterosis of silk viability at all sampling stages in the two hybrids (Table 3). However, proteins involved in methionine metabolism contributed differently to the heterosis of silk viability at different developmental stages of silks in the two hybrids: at D 10 and D 12 for Xun928×Zong3, and at D 8 and D 10 for Lx9801×Zong3. These results implied that proteins contributing to silk viability were not always as important for the heterosis of silk viability in the hybrids. Compared with the hybrid Lx9801×Zong3 at the three sampling stages, the hybrid Xun928×Zong3 accumulated more proteins involved in protein biosynthesis and folding, stress and defense responses, signal transduction and cell detoxification in response to genetic and environmental changes. Thus, the hybrid Xun928×Zong3 showed stronger heterosis than the hybrid Lx9801×Zong3.

Potential regulation networks revealed by differentially accumulated proteins related to silk viability and its heterosis
Methionine metabolism and salvage cycle. Nutrient supply is a basic requirement for successful fertilization. The nutrients in pollen, however, can support only about 2 cm of tube growth in the maize silk [21]. Thus, the maize silk must provide enough nutrients to support pollen tube growth over a longer distance. Consistent with this, many differentially accumulated proteins were related to cysteine and methionine metabolism (Tables 2 and 3).
Proteins involved in methionine supply were important for silk viability. Methionine functions not only as a building block for protein synthesis, but also as a signaling molecule in communicating intracellular metabolic events to receptors on the cell surface. Therefore, methionine could supply appropriate signals to support pollen tube growth and guidance in the maize silk. In plants, methionine synthase (MeSe EC 2.1.1.12; protein spots 8, 121, 155, 239) catalyzes the terminal step of the methionine synthesis pathway by transferring a methyl group to homocysteine (Hcy), producing methionine. However, this de novo synthesis is energetically expensive and highly tissue-specific. To save energy consumption, about 80% of the methionine is recycled [22]. S-Adenosylmethionine (AdoMet)-dependent transferase (protein spots 148, 194, 238) plays a critical role in methionine recycling by transferring the methyl group from AdoMet to S-adenosylhomocysteine (AdoHcy). This is not the only methionine recycling pathway in plants (S3 Fig). Methylthioribose-1-phosphate isomerase (protein spot 64), which catalyzes the phosphorylated methylthioribose (MTR) to methylthioribulose-1-P, is the first and ubiquitous enzyme for methionine recycling. The accumulation of adenine (Ade), a by-product of MTR formation, inhibits methionine recycling. On the other hand, Ade is also a substrate for phosphoribosyl Transferase 1 (APT1) (protein spot 46), which regulates cytokinin levels by converting active cytokinin forms to inactive ones. Loss of APT1 activity leads to excess accumulation of cytokinins, inducing a myriad of cytokinin-regulated responses, such as delayed leaf senescence, anthocyanin accumulation, and downstream gene expression [23].
AdoMet, as the major product of methionine metabolism, is an important cofactor that modulates various biological activities [24,25]. As the major methyl-group donor, AdoMet can regulate transmethylation reactions at the levels of DNA metabolism, RNA metabolism, and protein post-translational modifications [26]. AdoMet is also involved in metabolic and developmental regulation, since it is a substrate for thesynthesis of nicotianamine, ethylene (1-aminocyclopropane-1-carboxylate synthase), and polyamines [22]. AdoMet metabolism is complicated by its interaction with plant growth hormones such as cytokinins and auxins [27]. Thus, AdoMet is involved in regulating plant developmental by fine-tuning gene transcription, cell proliferation, and the production of secondary metabolites [28,29].
In this study, positive regulators of nutrients production were identified in the inbred lines Zong3 (protein spots 46, 64) and Xun928 (protein spot 8), but not in the inbred line Lx9801. In the hybrid combinations, relatively more positive regulators (protein spots 148, 155, 238, 239) were identified at the late developmental stages (D 10 and D 12 ) in Xun928×Zong3. However, only two positive regulators (protein spots 121, 194) were identified in hybrid Lx9801×Zong3 and differentially accumulated at D 8 and D 10 . These results were consistent with the stronger silk viability of the inbred lines Zong3 and Xun928 than that of Lx9801, as well as the high seed setting rates (84.6% for D 10 and 80.2% for D 12 ) and mid-parent heterotic degrees (34.4% for D 10 and 41.4% for D 12 ; data not shown) during the late sampling stages in the hybrid Xun928×Zong3. For comparison, Lx9801×Zong3 showed seed setting rates of 70.4% and 66.9% at D 10 and D 12 ; and mid-parent heterotic degrees of 21.1% and 31.3% at D 10 and D 12 (Tables 1-3).
Photosystem and energy metabolism. Photosynthesis provides fuel for plant growth by converting light energy into chemical energy. Ribulose-1,5-bisphosphate carboxylase/oxygenase (RuBisCO), catalyzes the first major step of the Calvin cycle (carbon fixation) to produce energy-rich carbohydrates. This reaction uses ATP as an energy source and NADPH as reducing power, and is often the rate-limiting step in photosynthesis [30]. In all eukaryotes, Rubisco is an oligomer consisting of eight large subunits bound to eight small subunits. The large subunits (protein spots 67, 106, 178) contain the enzymatically active substrate binding sites and are synthesized in the chloroplast. The small subunits are synthesized in precursor form by cytoplasmic ribosomes. Assisted by the RuBisCO large subunit-binding protein (protein spot 40), mature small subunits assemble with large subunits to form the oligomeric holoenzyme in the stroma. Several studies have shown that increased expression levels of RuBisCO subunits could increase photosynthetic efficiency by increasing catalytic activity and/or by decreasing the oxygenation rate [31].
ATP synthase (EC 3.6.3.14), a key enzyme in energy metabolism, is widely involved in oxidative and photosynthetic phosphorylation and plays an important role in many processes in plants. It consists of two rotary motors: the membrane-integrated CF o and the hydrophilic CF 1 . CF o mainly participates in proton transport through thylakoids, whereas CF 1 contains the nucleotide binding, catalytic, and regulatory sites of the ATP complex [32]. CF 1 contains five subunits: α (protein spot 199), β, γ, δ (protein spot 182), and ε [32,33]. The gene encoding the CF 1 α subunit, atpA, was shown to be related to cold resistance, and the transcript level of atpA was positively correlated with ATP synthase activity [34]. Mutation of the atpA gene in a cytoplasmic male sterile line caused an energy supply shortage during flower development, resulting in abnormal microspore development compared with its maintainer [35].
In this study, proteins involved in energy metabolism differentially accumulated in the inbred line Zong3 (protein spot 40) and the hybrid Xun928×Zong3 at D 8 and D 10 (protein spots 67, 106, 178, 182). A sufficient energy supply may be important to support stronger silk viability of Zong3, compared with those of Xun928 and Lx9801, and the stronger heterosis of silk viability in Xun928×Zong3 than in Lx9801×Zong3.
Protein metabolism and cell senescence. Proteins have a vast array of functions within living organisms, including catalyzing metabolic reactions, replicating DNA, responding to stimuli, and transporting molecules from one location to another. Many elaborate regulation mechanisms are involved in converting DNA sequences into functional proteins. Translation, the assembly of proteins by ribosomes, is an essential part of the protein biosynthetic pathway and requires initiation and elongation complexes [21]. Eukaryotic translation initiation factor 5A (eIF-5A) (protein spots 72, 146) not only regulates protein synthesis but also acts as an important determinant of cell proliferation and senescence. In dividing and dying cells, different isoforms of eIF-5A execute its biological switching function in response to physiological and environmental cues [36][37][38]. The elongation factor-1 (EF1) complex (protein spots 112, 161, 251) is responsible for the enzymatic delivery of aminoacyl tRNAs to the ribosome. EF1A is responsible for the selection and binding of the cognate aminoacyl-tRNA to the acceptor site of the ribosome. EF1 delta (protein spots 112, 251), functions as a guanine nucleotide exchange factor in regenerating active EF1A-GTP from inactive EF1A-GDP. During and after protein synthesis, polypeptide chains often fold into their native secondary and tertiary structures, whether they are used in the cell or secreted. To achieve their final correct states, cellular and secreted proteins require the help of several other folding proteins or chaperones. Protein disulfide isomerase (protein spot 252) catalyzes protein-folding, allowing proteins to reach their final correctly folded state without enzymatic disulfide shuffling [39]. Unneeded or damaged proteins are transferred to proteasomes, an active complex composed of α subunits and β subunits (protein spots 229, 288), to be degraded into amino acids that are used to synthesize new proteins. At all sampling stages, many differentially accumulated proteins involved in protein biosynthesis and correct folding were identified in the hybrid Xun928×Zong3. However, more proteins involved in proteasomes differentially accumulated during the late sampling stages (D 10 and D 12 ) in the hybrid Lx9801×Zong3. These results implied that the hybrid Lx9801×Zong3 might consume more resources during normal metabolism, which weakened its silk viability, especially at the late silk developmental stages.
During normal plant development, the insoluble polyesters suberin and cutin form extracellular lipophilic barriers to prevent membrane leakiness [40]. However, membranes become leaky when the cell begins to senescence. This process is usually accompanied by the accumulation of proteins involved in lipid metabolism. In this study, O-methyltransferase (protein spots 89, 150, 163, 171, 236, 243, 272), the first rate-limiting enzyme in suberin synthesis, and GDSLmotif lipase/hydrolase (protein spot 96), an enzyme involved in the hydrolysis and transfer of activated monomers in cutin synthesis [41], differentially accumulated in the hybrid Xun928× Zong3. Proteins involved in lipid metabolism only differentially accumulated in the hybrid Lx9801×Zong3. These results implied that membrane leakiness might occur earlier in Lx9801×Zong3 than in Xun928×Zong3 at the late silk developmental stages. Thus, silk viability was lost earlier in Lx9801×Zong3 than in Xun928×Zong3. This pattern of protein accumulation might also explain the faster decrease in the seed setting rate and the weaker heterotic degree in the hybrid Lx9801×Zong3 at the late silk developmental stages.
Phenylpropanoid metabolism and plant hormones regulation. The plant hormone auxin regulates cell elongation, division, differentiation, and morphogenesis. Many proteins involved in elaborate temporal and spatial regulation of auxin metabolism, transport, and signaling have been identified. Auxin-binding protein 1 (ABP1: protein spot 193) mediates cell elongation and, directly or indirectly, cell division. In previous studies, ectopic and inducible expression of ABP1 conferred auxin-dependent cell expansion in tobacco cells that normally lack auxin responsiveness [42] and antisense suppression of ABP1 eliminated auxin-induced cell elongation and reduces cell division. A homozygous null mutation of ABP1 was embryolethal in Arabidopsis [43]. Auxin-induced swelling of proteoplasts and intact guard cells can also be attributed to ABP1 [44,45]. Pyrophosphate-energized vacuolar membrane proton pump 1 (protein spot 69) facilitates auxin transport and regulates auxin-mediated developmental processes by modulating apoplastic pH [46]. Flavonoids in the phenylpropanoid pathway are another regulator of active auxin and have species-specific roles in nodulation, fertility, defense, and ultraviolet protection. Flavonols have been shown to negatively regulate the polar auxin transport (PAT) by competing for free auxin with auxin efflux carriers such as PIN and ABCB (PGP proteins) in vivo [47]. Dihydroflavonol 4-reductase (DFR4, EC1.1.1.219: protein spot 53) and UDP-glucoside: flavonoid glucosyltransferase (EC 2.4.1.115: protein spots 63, 76,79,99,117,119,122,123,138,152,159,188,197,200,201,202,212,224,231,241,285) are the first and last enzymes in the anthocyanin biosynthetic pathway, respectively. Their differential accumulation may be related to competition for the dihydroflavonol substrate with the flavonol branch, and thus, could indirectly affect PAT. The relatively higher contents of anthocyanin biosynthetic enzymes corresponded to higher levels of glutathione S-transferase-like proteins (protein spots 127, 216, 217, 298), which transport anthocyanins from the ER to the vacuole [48] in the hybrid Lx9801×Zong3 at the three sampling stages.
Proteins involved in the regulation of hormone levels (ABP1, pyrophosphate-energized vacuolar membrane proton pump 1; anthocyanin biosynthesis pathway; and cytokinin-O-glucosyltransferase) were identified as being important for both silk viability and its heterosis. These proteins differentially accumulated in the hybrid Xun928×Zong3, whereas only those involved in anthocyanin biosynthesis differentially accumulated in the hybrid Lx9801×Zong3. The flexibility of the systems regulating hormone levels may explain the increase in silk viability in the hybrid Xun928×Zong3, resulting in the high seed setting rate and strong heterosis.
In summary, proteins gi|413944345, gi|414869037, and gi|195635735 were attractive and might be related with silk viability as well as its heterosis. Significant correlation (r = 0.827 Ã for gi|413944345; r = -0.365 Ã for gi|414869037; r = 0.556 Ã for gi|195635735) was detected between these protein spots accumulation level and seed setting rate. Thus, we could propose the following hypotheses regarding proteins related to silk viability and its heterosis: methionine salvage, protein synthesis, and ATP supply function as positive regulators of silk viability, and therefore, contribute to strong silk viability and its heterosis in Zong3 and Xun928×Zong3, respectively. Active fatty acid metabolism, a signal for cell wall degradation, and anthocyanins, which negatively regulate local hormone accumulation, were related to weaker silk viability in Lx9801 and Lx9801×Zong3. The metabolism of cutin and suberin, which were derived from phenylpropanoid precursors, might confer stronger silk viability (Zong3 and Xun928) and stronger heterosis of silk viability in hybrids by slowing the silk aging process, especially during the late stages of silk development.

Conclusions
In this study, the heterosis of silk viability could be mainly attributed to additive accumulation of differentially regulated proteins, although proteins that accumulated in a non-additive manner made a similar contribution. Simple additive and dominant effects at a single locus, as well as complex epistatic interactions of metabolic pathway genes at two or more loci, resulted in partially dominant silk viability heterosis in the hybrids. For silk viability, most important differentially accumulated proteins were those involved in methionine metabolism for nutrient supply, phenylpropanoid metabolism for hormone homeostasis, protein biosynthesis and metabolism for genetic information processing, and carbon fixation for energy generation.

Plant materials
Three typical inbred lines, Zong3, Xun928, and Lx9801, with different silk viability were used in this study. Among more than one hundred inbred lines, the silk viability of the inbred line Zong3 was extremely high. The inbred lines Xun928 and Lx9801 had relatively weak silk viability. To assay the heterosis of silk viability in different genetic backgrounds, two hybrids, Xun928×Zong3 and Lx9801×Zong3 were created in this study. The two hybrids and the three inbred lines were planted on the farm of Henan Agricultural University (Zhengzhou, China; E 113°42 0 , N 34°48 0 ) in summer of 2013, when the daily average temperature was 14.3°C. The annual average rainfall is 640.9 mm in this region. Each plot consisted of ten 5-m-long rows, with 20 cm of in-row spacing and 67 cm of inter-row spacing. Only the middle rows were sampled to avoid edge effects. Before the silks emerged from the husk, ear shoots were totally covered with bags to avoid pollen contamination. To evaluate the silking time accurately, the silking time of each ear of the materials was recorded in the field. Day 1 (D 1 ) was marked as the day that the silks emerged above the ligule of the outer leaf of the husk. Silks were removed from the mid-base region of each ear at D 4 , D 6 , D 8 , D 10 , and D 12 and immediately frozen in liquid nitrogen in the field. Each sample was collected with three biological replications and 10 ears were mixed for each replication. At the same time, the ear for each sample was saturationpollinated by hand on the sampling day to measure the seed setting rate. Pollination was completed between 9 and 10 a.m. (below 37°C) to ensure consistent pollination efficiency. The ears were harvested at physiological maturity, and only the seeds at the mid-base (5-15 rounds from the base) were used to calculate the seed setting rate of the cob according to silk development characteristics [54]. The seed setting rate was calculated by dividing the total number of spikelets by the number of fully grown seeds.

Protein extraction and MS identification
Each genotype was assayed with three biological replications corresponding to each sampling stage. Frozen silks (1 cm, approx. 1.0 g) of each biological replication (mixture of silks from ten different plants) were fully ground in liquid nitrogen and then extracted in 10 mL pre-cooled trichloroacetate (TCA) buffer (10% w/v TCA in acetone with 0.07% β-mercaptoethanol) with vortexing for 2 h at 20°C. After centrifugation at 15,000 × g for 30 min, the supernatant was discarded and the precipitate was rinsed with 10 mL chilled buffer (80% acetone with 0.07% β-mercaptoethanol) four times by centrifuging for 10 min at 15,000 × g. The final cleaned precipitate was freeze-dried under a vacuum. The dried protein pellet per 1 mg was resuspended in 20 μL buffer (8 M urea, 2 M thiourea, 4% (w/v) CHAPS and 40 mM dithiothreitol (all from Solarbio)). The protein was quantified using a Bio-Rad protein assay with bovine serum albumin as a standard and used for two-dimensional gel electrophoresis . Three technical replications were assayed for each biological replication. For each technical replication, equal amounts of total protein extract (800 μg) were used for isoelectric focusing (IEF). Immobilized dry strips (24 cm, Imobiline drystrips, Bio Rad, Hercules, CA, USA) with a linear gradient of pH 4-7 were rehydrated for 16 h at 50 V. The IEF conditions for separating proteins were as followes: slow 250 V for 30 min, rapid 250 V for 2 h, rapid 500 V for 2 h, rapid 1,000 V for 2 h, linear 9,000 V for 5 h, rapid 10,000 V for 10 h, and a constant 500 V for the final 12 h at 20°C. Strips were immediately equilibrated in 10 mL of two types of SDS equilibration buffer for 15 min each. Buffer 1 contained 0.375 M Tris-HCl pH 8.8, 6 M urea, 20% glycerol, 4% SDS, and 2% DTT and buffer 2 contained 0.375 M Tris-HCl pH 8.8, 6 M urea, 20% glycerol, 4% SDS, and 2.5% iodoacetamide. IPG gel strips with the proteins were embedded into the top of a polyacrylamide gel (12%) after equilibration and separated at a constant voltage of 50 V for 30 min. Then, a constant voltage of 200 V was maintained until the electrophoresis was finished.
Digital images of the gels stained with Coomassie brilliant blue G250 were obtained with a scanner (UMAX Power Look 2100 XL). Spot detection and matching was performed with the default parameters using the "spot detection wizard" function in PDQuest 8.0 software. The "find spot centers" function was used with default auto-noise smoothing and background subtraction. A Gaussian model was selected to generate a master gel for each image file. All the gels were matched to the reference master gels selected and normalized in automated mode followed by manual group correction. The normalization parameters were "total quantity in valid spots", "total density in gel image", "mean of log ratios", and "local regression model". After normalization, ANOVA was used to calculate the significance of differences in the relative abundance of protein in individual spot features among the developmental stages of a certain inbred line, as well as among hybrids and their corresponding inbred lines at each developmental stage. For protein spots further assayed by MS, the maximum intensity variation criterion was set to ! 1.5-fold and ! 2-fold (P < 0.05) among different sampling stages for each inbred line, and between hybrid and parental inbred lines at each sampling stage, respectively.
The selected proteins were excised manually from gels, subjected to in-gel digestion with trypsin, and then destained using 25 mM ammonium bicarbonate in 50% (v/v) acetonitrile for 15 min at room temperature. The discolored spots were vacuum-dried and incubated with modified porcine trypsin at 37°C overnight. After centrifugation, the supernatant was collected and vacuum-dried, and then the precipitate was re-dissolved in 60% acrylonitrile/0.1% trifluoroacetic acid (TFA) (100 μL) for 15 min to obtain the peptides. Then, a 0.3 mL peptide sample and 0.3 mL matrix consisting of 10 mg/mL α-cyano-4-hydroxycinnamic acid in 50% acetonitrile and 0.1% TFA was analyzed on a matrix-assisted laser desorption ionization-time of flight (MALDI-TOF) mass spectrometry (MS). The parameters of the MS were set with 4000 Series Explorer software (Applied Biosystems). The lists of theoretical peptide MS from each peptidemap-fingerprinting (PMF) combined with MS/MS were used to search the NCBI (National Center for Biotechnology Information) database without repetition for homologous sequences using MASCOT 2.2 software (www.matrixscience.com). The search criteria were as follows: 1) peptide mass tolerance of 100 ppm; 2) maximum of a single missed tryptic cleavage; 3) fragment mass tolerance of 0.4 Da; and 4) carbamidomethylation by cysteine residues as fixed modifications and oxidation by methionine residues as dynamic modifications. Only proteins with a MASCOT score > 60 with 95% confidence and at least two matched peptides were accepted. Gene Ontology (GO) annotations and the theoretical Mr/pI for the identified proteins were retrieved from http://www.geneontology.org/ and http://www.expasy.ch/tools/pi_ tools.html, respectively. Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis was carried out using the blast function in BLAST2GO. The protein-protein interaction network was analyzed by the publicly available program STRING (http:// string-db.org/). Clusters of Orthologous Groups (COG) of proteins functions were used to construct the networks. Only an interaction networks with a high confidence (0.700 for silk viability or 0.900 for heterosis of silk viability) and no more than five interactors were retained. The eukaryotic orthologous groups (KOGs) were considered prime selections for a single protein spot.

Data analysis
Protein spots that had no significant difference in average spot intensity from the mid-parent value at the 0.05 level were considered additively accumulated (A). The accumulation pattern of each non-additive protein was classified as described by Hoecker et al. [55]. Average spot intensities of proteins that deviated significantly from the mid-parent value of the parental lines at P < 0.05 level were defined as non-additive proteins. "+" and "−" were used to indicate that the protein spot intensity identified in the F 1 hybrid was similar to the high parent and low parent values, respectively. "+ +" and "− −" indicated that the protein spot intensity identified in the F 1 hybrid was significantly different from the high parent and low parent values, respectively. "+ −" indicated the protein spot intensity identified in the F 1 hybrid fell in between the mid-parent and high parent or mid-parent and low parent values. ANOVA, LSD, and correlation analysis were performed with the corresponding function in Excel 2007.  [28]. Enzymes: 1, cystathionine gsynthase; 2, cystathionine b-lyase; 3, methionine synthase; 4, AdoMet synthetase; 5, AdoMetdependent methylase; 6, AdoHcy hydrolase; 7, 1-aminocyclopropane-1-carboxylicacid synthase; 8, AdoMet decarboxylase; 9, threonine synthase. Note that AVG inhibits both cystathionine g-synthase and 1-aminocyclopropane-1-carboxylic acid synthase. (TIF) S1