In-Depth Understanding of Camellia oleifera Self-Incompatibility by Comparative Transcriptome, Proteome and Metabolome

Oil-tea tree (Camellia oleifera) is the most important edible oil tree species in China with late-acting self-incompatibility (LSI) properties. The mechanism of LSI is uncertain, which seriously hinders the research on its genetic characteristics, construction of genetic map, selection of cross breeding parents and cultivar arrangement. To gain insights into the LSI mechanism, we performed cytological, transcriptomic, proteomic and metabolomic studies on self- and cross-pollinated pistils. The studies identified 166,591 transcripts, 6851 proteins and 6455 metabolites. Transcriptomic analysis revealed 1197 differentially expressed transcripts between self- and cross-pollinated pistils and 47 programmed cell death (PCD)-control transcripts. Trend analysis by Pearson correlation categorized nine trend graphs linked to 226 differentially expressed proteins and 38 differentially expressed metabolites. Functional enrichment analysis revealed that the LSI was closely associated with PCD-related genes, mitogen-activated protein kinase (MAPK) signaling pathway, plant hormone signal transduction, ATP-binding cassette (ABC) transporters and ubiquitin-mediated proteolysis. These particular trends in transcripts, proteins and metabolites suggested the involvement of PCD in LSI. The results provide a solid genetic foundation for elucidating the regulatory network of PCD-mediated self-incompatibility in C. oleifera.


Introduction
Oil-tea tree (Camellia oleifera) is the most economically important forest species native to China [1]. The tree is cultivated for its seeds, which are the source of edible oil rich in unsaturated fatty acids and various active ingredients, such as squalene, tocopherol, vitamins, carotene, glycerol, triterpene alcohol and saponin, etc. [2,3]. Oil-tea camellia can be used for production of high value-added skin care products, cosmetics, Vaseline, soaps, shampoos and medical products such as oleic acid and its esters [4]. Although the flowering rate of C. oleifera is high, the fruit-setting rate is extremely low under natural conditions (less than 5%), mainly due to the genetic characteristics of self-incompatibility (SI) and improper pollinating cultivar arrangement. Understanding the mechanism of its SI system can potentially increase seed yield of C. oleifera.
SI is a precise genetic mechanism that prevents self-fertilization and promotes cross-pollination during the evolution of flowering plants [5][6][7]. It is estimated that about half of the flowering plants exhibit SI [8]. The research on the SI mechanism has become a frontier area in plant biology [9][10][11].

Cytological Observation of SI Type in C. Oleifera
During the first 48 h after pollination, the pollen tube growth rate was basically the same between the self-and cross-pollinated plants ( Figure 1A-F). However, after 48 h, the growth rate of the self-pollinated pollen tubes was significantly lower than that of the cross-pollinated. The self-pollinated pollen tubes stopped growing between 48 and 72 h after pollination, and the LSI reactions appeared, such as tissue distortion and folding ( Figure 1I-L). It had been found that LSI reactions in C. Oleifera were caused by abnormal thickening of the pollen tube wall with unidentifiable organelle [33], which are the characteristics of PCD [34]. In the meantime, the pollen tubes entered the ovule by micropyle at 72 h after cross-pollination ( Figure 1G). These observations support the classification of C. Oleifera SI as LSI.

Illumina-and SMRT-Based RNA Sequencing and Error Correction
Illumina-and single molecule real-time (SMRT)-seq platforms were used to perform in-depth analysis of gene expression during oil-tea camellia LSI ( Figure 2). To identify as much longer transcripts as possible, equal amounts of total RNA from each sample were pooled together and reverse-transcribed for SMRT-seq. Total RNA with OD 260/280 ≥ 1.81, RIN ≥ 9.4 and 28S/18S ≥ 1.5 were used to build libraries (Table S2). To minimize bias that favors sequencing of shorter transcripts, two size-fractionated cDNAs (>2 kb and <2 kb) were constructed and subsequently sequenced in two SMRT cells (Table  S3). A total of 481,878 full-length non-chimeric transcripts were identified, which accounted for 88.3% of reads of insert (ROIs) ( Figure S1A). Finally, correction of the polished isoform sequences from the second-generation sequencing data was done with proofreading error correction software [39]. Using 166,591 redundant transcripts as the reference sequence for Illumina-seq, the total mapping rate was 81.82% ( Figure 2). The PCA diagram of the transcriptome showed that the samples were tightly clustered together, indicating the reliability of the experiment and the reasonable selection of the samples ( Figure S1F). In order to better understand the advantages of the combination of Illumina-and SMRT-based RNA sequencing, we compared the Illumina de novo assembled transcripts (281,140) to the PacBio Iso-Seq consensus transcripts (166,591) ( Figure S2). The mean length (1896 bp) and N50 (2198 bp) were longer among the Iso-Seq consensus transcripts than those among the Illumina RNA-seq data ( Figure S2A). Moreover, Iso-Seq technology yielded more contiguous transcripts and contained a higher proportion of intact annotation rate (94.16%) and open reading frames (ORFs) (90.89%) than those only using the Illumina method ( Figure 2 and Figure S2B-F).

Functional Annotation and Categorization of Transcripts
To predict the functions of the 166,591 transcripts, we used software Diamond to perform functional annotation using NR, GO, KEGG, KOG and Swiss-Prot. A total of 132,738 transcripts were successfully matched to known proteins in at least one of the five databases, and 33,612 transcripts received high scores with known proteins in all five databases ( Figure S1B).
To classify the functions of the C. oleifera transcripts, assignment of GO terms was performed using software Diamond from the NR database. In total, 111,856 transcripts were matched to GO terms, which were classified into three major functional categories, including biological process, cellular component and molecular function ( Figure S1C). A total of 77,810 transcripts were predicted using the KEGG database, which were divided into five branches (cellular processes, environmental information processing, genetic information processing, metabolism and organismal systems) ( Figure S1D). According to the KOG database, 68,629 transcripts participated into 25 functional groups ( Figure S1E).

Transcript Expression Profiles and Enrichment in Self-and Cross-Pollinated Plants
There were 656, 1260, 609, 699, 867, 858, 2776 and 2759 differentially expressed transcripts (DETs) in the eight pairs of sample comparisons SP48_vs_SP72, CP48_vs_CP72, SP48_vs_CP48, SP72_vs_CP72, SP48_vs_NP48, CP48_vs_NP48, SP72_vs_NP72 and CP72_vs_NP72, respectively ( Figure 3A). There were no common DETs in all of the eight comparison groups ( Figure 3B). We identified 1197 DETs between self-and cross-pollinated pistils (Table S4), including 342 significantly up-regulated transcripts in SP48_vs_CP48 and 399 in SP72_vs_CP72 and 267 significantly downregulated transcripts in SP48_vs_CP48 and 300 in SP72_vs_CP72 ( Figure 3A). According to the Venn diagram of the DETs, there were 368 specific DETs in self-and cross-pollinated pistils ( Figure 3B). KEGG annotation showed that 1197 DETs were enriched in 19 classes of KEGG pathways and classified in five groups ( Figure 3C). The significantly enriched pathways included pentose and glucuronate interconversions and ascorbate and aldarate metabolism in the carbohydrate metabolism class, biosynthesis of amino acids in the amino acids metabolism class, carbon fixation in photosynthetic organisms in the energy metabolism class, ATP-binding cassette (ABC) transporter in the membrane transport class, mitogen-activated protein kinase (MAPK) signaling pathway plant and plant hormone signal transduction in the signal transduction class, aminoacyl-tRNA biosynthesis in the translation class and ubiquitin-mediated proteolysis in the folding, sorting and degradation class.
Hormone signal transduction and MAPKs are associated with responses to SI [22,29,30]. A previous study provided evidence that MAPKs could induce PCD [41][42][43]. In our results, DETs in the selfand cross-pollinated pistils were significantly enriched in plant hormone signal transduction, MAPK signal transduction pathways and the ubiquitin-mediated proteolysis pathway ( Figure 4A). As shown in Figure 4A, there were 27 transcripts (COI1s, JAZs, ETRs, CTR1, EIN2, EIN3s and EBF1/2s) encoding JA and ETH-signaling-related components and 18 transcripts (ETRs, CTR1, EIN2, EIN3s and EBF1/2s) encoding MAPK-signaling-related components. The expression of XRN4 was observed predominantly in SP48, and the encoded protein was involved in the degradation of de-capped mRNAs, nonsense mediated decay, microRNA decay and essential for proper development [44]. RCHY1.2, TRIP12.2, TRIP12.3 and TRIP12.5 had relatively high expression levels in SP72; the expression levels of STAH1 was higher in CP48 and RCHY1.1, RCHY1.3 and RCHY1.4 were more highly expressed in CP72 than other combinations, which were three types of E3 ubiquitin ligase related to ubiquitin-mediated proteolysis [45]. The qRT-PCR analysis of six transcripts (JAZ2, EIN2, XRN4, SIAH1, RCHY1.4 and TRIP12.2) revealed consistent expression patterns with those generated by RNA-seq data ( Figure 4B). Moreover, calmodulin CaM (cb21070_c0/f2p0/995)-encoding proteins involved in MAPK signal transduction and the expression level of CaM in CP72 was 23.7 times than that in SP72. In addition, there were many differentially expressed transcription factors in self-and cross-pollinated pistils, for example, WRKY33 (cb8503_c4/f1p0/2186) and B-ARR (cb1922_c15/f1p0/2614) had significantly higher expression in CP72 than in SP72. The two transcripts code for proteins related to MAPK signaling pathway and plant hormone signal transduction, respectively. These gene products could be related to LSI in C. oleifera.

Identification of PCD Candidate Transcriptsin Self-and Cross-Pollination
Cytological observation showed that C. oleifera pollen tube growth inhibition ( Figure 1) after self-pollination is PCD-related [33]. Twenty putative PCD-related genes (10 positive and 10 negative) from literatures [46][47][48][49][50][51] were identified. In Table S5, we can find that 73 PCD-related transcripts were expressed in mature pollen at lower levels than those in self-, cross-and non-pollinated pistils. Forty-seven transcripts showed significant differences in SP and CP by Duncan's test ( Figure 5A). Most of the positive genes exhibited relatively high levels of expression in self-pollinated pistils, such as catalase isozyme 2 (CAT2), cryptochrome-1 (CRY1), E3 ubiquitin-protein ligase (RING1), CO(2)-response secreted protease (RSP) and UBP1-associated protein 2C (UBA2C). Relatively, the expression levels of negative genes were lower after self-pollination and higher in cross-or non-pollinated pistils, such as alpha carbonic anhydrase 4 (ACA4), BONZAI 3 (BON3), CBS domain-containing protein (CBSX5) and respiratory burst oxidase homolog protein A (RBOHA). The qRT-PCR analysis of nine transcripts revealed consistent expression patterns with those generated by RNA-seq data ( Figure 5B). These expression patterns suggested that the products of these PCD-related genes could have important roles in the inhibition of pollen tube growth of C. oleifera.

Proteomic Changes of C. oleifera in Response to SI
The PCA diagram of proteome showed that the samples were tightly clustered together, indicating that the high repeatability of the proteome experiment ( Figure S3). A total of 6851 proteins were identified in the pistils after self-and cross-pollination of C. oleifera, which accounted for 4.52% of total predicted proteins by transcriptome analysis. There were 294 differentially expressed proteins (DEPs) in the four combinations, including SP48_vs_SP72, CP48_vs_CP72, SP48_vs_CP48 and SP72_vs_CP72 (Table S6). The results of the Venn chart showed 229 DEPs in SP48_vs_CP48 and SP72_vs_CP72 ( Figure 6A), including 6 and 121 up-regulated proteins in SP48_vs_CP48 and SP72_vs_CP72, respectively, and 8 and 96 downregulated proteins in SP48_vs_CP48 and SP72_vs_CP72, respectively ( Figure 6B). The 294 DEPs were functionally categorized into 15 classes ( Figure 6D). Other than proteins of unknown function (54.1%), the most abundant classes were translation (11.1%), followed by signal transduction (6.8%); carbohydrate metabolism (5.1%); energy metabolism (4.8%); lipid metabolism (4.5%); folding, sorting and degradation (4.5%) and cell growth and death (3.8%).

Overall Metabolite in Different Groups of C. Oleifera
The PCA diagram of metabolome showed that each group had six replicates; one of the SP48 replicates and one of the SP72 replicates were separated from the rest, which were tightly clustered together in ( Figure S4). We deleted these two replicates and performed a metabolome differential analysis. A total of 6455 metabolites were identified in the four C. oleifera groups (SP48, SP72, CP48 and CP72). We detected a total of 60 differentially expressed metabolites (DEMs) in SP48_vs_SP72, CP48_vs_CP72, SP48_vs_CP48 and SP72_vs_CP72 (Table S7). The numbers of DEMs for each comparison were 33, 31, 28 and 18, including four, seven, three and three up-regulated metabolites and 29, 24, 25 and 15 downregulated metabolites, respectively ( Figure 7B). Venn diagrams showed that there were 43 DEMs in self-and cross-pollinated pistils at 48 h and 72 h ( Figure 7A).

Integrated Transcriptomic, Proteomic and Metabolic Datasets
There were 51 metabolic pathways which were shared by at least two DETs, DEPs and DEMs in self-and cross-pollinated pistils from the transcriptome, proteome and metabolome datasets ( Figure S5 and Table S8). Twenty-three metabolic pathways appeared associated to the three sets, such as fatty acid biosynthesis, aminoacyl-tRNA biosynthesis, alanine, aspartate and glutamate metabolism, arginine and proline metabolism, arginine biosynthesis, β-alanine metabolism, phenylalanine metabolism, pyrimidine metabolism and valine, and leucine and isoleucine degradation. Likewise, 19 additional pathways were found in common between DETs and DEPs ( Figure S5 and Table S8), such as plant MAPK signaling pathways, plant hormone signal transduction, RNA degradation, ubiquitin-mediated proteolysis and oxidative phosphorylation, etc. Moreover, another seven were additionally shared by DETs and DEMs, such as starch and sucrose metabolism, and ABC transporters, while only two metabolic pathways (cAMP signaling pathway and valine, leucine and isoleucine biosynthesis) were indicated by DEPs and DEMs.
To further understand the relationship between DETs expression and the amounts of DEPs and DEMs, we analyzed the correlation between them using Pearson's correlation statistics (Table S9). First, DETs with the same expression pattern could be found by trend analysis, as shown in Figure 8 Nine trends (G1-G9) were found, with p < 0.05. In the nine trend graphs, 190 transcripts that responded strongly to self-and cross-pollination reactions associated with 226 proteins and 38 metabolites (Table  S9). Among them, the expression level of G1 in SP48 was relatively higher than other groups, and four DEPs and 16 DEMs had the same trend of synthesis. Conversely, 32 DETs in G2 had lower expression in SP48 than other groups and related to one DEP and three DEMs. G3 and G7 were expressed at the lowest in SP72 and CP48, respectively. Thirteen DETs in G4 were abundantly expressed in SP48, SP72 and mature pollen, while G6 were abundantly expressed in CP. The expression trends of G5 and G8 were opposite. The former was significantly higher in CP72 than other groups, and the latter was significantly lower in CP72 than other groups. Interestingly, in G9, 35 DETs, 32 DEPs and 5 DEMs had the highest expression in SP72, but the expression levels in CP72 were the lowest. These screened transcripts, proteins and metabolites could be used as the basis for studying the molecular mechanism of LSI in C. oleifera.

Discussion
C. oleifera seeds are the source of healthy edible oil. However, the seed yield is low due to low natural seeding rates caused by the LSI of oil-tea tree [28,53,54]. The current study observed the difference of pollen tubes in the pistils between self-and cross-pollinated pistils by cytological observation. We found that the pollen tube would grow to the upper part of the ovary at the base of the style and then showed abnormalities, such as swelling, bifurcation and wavy, and stopped growing between 48 h and 72 h after self-pollination. The cross-pollinated pollen tubes entered the ovary at 48 h and entered the embryo sac through the micropyle at 72 h. It has been reported that the inhibition of pollen tubes in the self-pollinated oil-tea tree belonged to a type of PCD [33]. All these findings confirmed that the SI of C. oleifera was the LSI type by PCD. Furthermore, the current study used transcriptomics, proteomic and metabolomic technologies to study the molecular level differences between self-and cross-pollinated pistils in C. oleifera. Data analysis revealed that a large regulatory network existed in the LSI process with PCD in self-pollinated pollen tubes. The network included the transcripts, protein products and metabolites associated with plant MAPK signaling pathways, plant hormone signal transduction, ABC transporters and ubiquitin-mediated proteolysis. The relationship among PCD, metabolic pathways and SI are described in details as follows.

PCD-Regulated Genes Regulate the Inhibition of Pollen Tube in Self-Pollination
It has been reported that PCD specifically occurred in incompatible pollen tubes of Pyrus pyrifolia [21,34]. Chen [55] speculated that after the self-pollination signal was recognized, it was amplified by a cascade system, which changes the function of cell metabolism and eventually leads to PCD in the pollen tube. Transmission and scanning electron microscopy showed that the death of the C. oleifera pollen tube in the self-pollinated pistils belonged to PCD [33].
PCD could be controlled by both positive and negative regulators. In this study, 20 putative key PCD-related genes were detected based on their expression patterns, which included 10 positive genes (ADHIII, AMC4, CAT2, CRY1, RING1, RSP, POB1, SRC2, UBA2C and XCP1) and 10 negative genes (ACA4, BON3, CBSX5, CNX1, CRLK2, IMPA4, P2C63, RBOHA and VPS45) [46][47][48][49][50][51] (Table S5). Most of the positive genes exhibited relatively high levels of expression in SP by Duncan's test, such as CAT2, CRY1, RING1, RSP and UBA2C ( Figure 5A). Relatively, the negative genes showed low expression levels in SP and high expression in cross-or non-pollinated pistils, such as ACA4, BON3, CBSX5 and RBOHA ( Figure 5A). It has been reported that S-RNase could specifically cause degradation of RNA in the microfilament skeleton of pollen tubes, which causes the occurrence of pollen tube PCD in Pyrus pyrifolia [21,56]. We also found that six DETs and one DEP were associated with RNA degradation (Tables S4 and S6), which could trigger a signaling cascade resulting in PCD culminated by factors that determine style incompatibility [57,58]. The results suggested the important role of PCD-related genes and protein produced in the inhibition of the pollen tube growth of LSI in C. Oleifera.

MAPK Signaling Pathway Involvement in SI of C. Oleifera
MAPKs were known to be functionally involved in the activation of defense and stress responses [43,59,60]. It has been reported that SI and the plant innate immunity systems have common pathways [23,61]. SI inhibition of pollen tube growth could be regarded as a stress response, so it was speculated that MAPKs might be involved in the SI response [62]. A MAPK kinase p56 was implicated in the SI-induced signaling cascade in Papaver pollens, which was specifically expressed in incompatible pollen [42].
In this study, we found that 16 DETs and two DEPs were associated with the MAPK signaling pathway. Among them, ETR, CTR1, EIN2, EIN3 and EBF1/2-encoding proteins that participate in plant hormone signal transduction (Figure 4). CaM4 and WRKY33 were associated with MAPK signal transduction, and their expression levels in CP72 were 23.7 and 8.7 times higher than that in SP72, respectively (Table S3). Studies have shown that pollen tube growth requires a Ca 2+ gradient at the tip. SI makes the extracellular calcium influx eliminating pollen tube tip Ca 2+ gradient and inhibits the pollen tube growth [63]. As a transcription factor, WRKYs can act as positive or negative regulators in a defense response [64]. We speculate that CaM and WRKY33 could respond to LSI and might play a negative regulatory role in the PCD process of self-pollinated pollen tubes in C. oleifera.

Plant Hormone Signal Transduction Involvement in SI of C. Oleifera
Plant hormones play an important regulatory role in plant PCD. Studies have shown that the PCD process in the development of maize endosperm required a balance between ABA and ethylene [65][66][67]. This suggested that plant hormones might work together to regulate plant cell PCD. DETs after selfand cross-pollination were significantly enriched in plant hormone signal transduction pathway and plant-pathogen interaction pathway using transcriptomics in pears [22]. Likewise, it also demonstrated that jasmonate (JA) and abscisic acid (ABA) might enhance the expression level of S-RNase, which caused the PCD of the pollen tube in SI. It has been reported that the entries of both self and non-self S-RNase into pollen tubes of apples (Malus domestica) stimulated JA production, in turn inducing the accumulation of MdMYC2 transcripts, a transcription factor in the JA signaling pathway widely considered to be involved in plant defense processes [23].
In this study, we found that there were 16 DETs encoding JA and ETH signaling-related components and one DEP involved in plant hormone signal transduction. Among them, COI1.2 and COI1.3 were significantly higher expressed at SP72 than in the other groups (Figure 4), and the expression level in pollen was the lowest. COI1 is an F-BOX domain protein that forms a complex with SCF, which is involved in JA-regulated plant defense responses to biotic stress [68]. The SCF complex selectively degrades non-self S-RNase in the SI reaction by a ubiquitin-mediated protease degradation system, thereby causing the pollen tube PCD [69,70].

ABC Transporters Involvement in SI of C. Oleifera
ABC transporters are a large family of transmembrane transporters, also known as ATP-binding cassette transporters, and are currently the largest and most versatile family of proteins [71]. It has reported that MdABCF assisted in transportation of either self or non-self S-RNase into the pollen tube. Moreover, MdABCF coordinated with the cytoskeleton to transport S-RNase [52]. In our study, it was found that six DETs-encoded proteins and 16 DEMs in SP and CP are related to ABC transporters (Tables S4 and S7). The DETs included ABCB (cb1496_c2/f1p0/2420, cb1707_c15/f1p0/2047, cb1707_c16/f1p0/3057), ABCC (cb485_c32/f1p0/4653, cb485_c8/f1p0/2223) and ABCG (cb8540_c10/f1p2/2041), with higher expression at SP48 than CP48. DEMs included metabolites such as l-aspartate, choline, sucrose and maltotriose, which had higher production in SP. Other studies have shown that ABC transporters were closely related to transportation of plant hormones [72], which were closely related to plant SI. ABC transports might be involved in the regulatory network of LSI in C. oleifera.

Ubiquitin-Mediated Proteolysis Involvement in SI of C. Oleifera
It has been reported that incompatible pollen SCR could activate SRK in a S-haploid-specific manner in the SSI plants [73]. This process leads to an intracellular signal transduction cascade, ensuing downstream reactions with an E3 ubiquitin ligase [74], which is associated with ubiquitin-mediated proteolysis. In GSI plants, S-RNase could be specifically prevented from being degraded by ubiquitin 26S proteasome after self-pollination, retaining its activity as a cytotoxin to degrade its own pollen tube tRNA, thereby inhibiting the pollen tube growth and leading to the occurrence of the SI reaction [69,70,75,76]. We found the DETs JAZ and EIN3 directly affected the third metabolic pathway ubiquitin-mediated proteolysis, including three genes SIAH1, RCHY1 and TRIP12, which encode for three types of the E3 ubiquitin ligase (Figure 4) [77]. SIAH1, RCHY1 and TRIP12 were reported to play important roles in plant salt tolerance, drought, high temperature, low temperature stress and ABA signaling pathways. Further studies showed that ubiquitin ligase reduced reactive oxygen species (ROS) production and cell death, thereby positively regulating the salt stress response [78]. In recent years, more and more studies have found that SI could cause plant immune responses. For example, increased PA levels in pollen tubes initially played a protective role in incompatible pollen, until sustained PbrS-RNase activity reaches the point of no return and pollen tube growth ceases in pears [21]. A gamma-thionin protein from apples, MdD1, was required for defense against the S-RNase-induced inhibition of the pollen tube prior to self/non-self recognition [23]. Our results showed that the ubiquitin-mediated proteolysis pathway might directly be involved in the LSI or could assume a self-defense role in the LSI process of C. oleifera.

Plant Materials
Two oil-tea tree varieties, "Hua Xin" and "Hua Jin", developed by the Central South University of Forestry and Science and widely cultivated in the Hunan Province of China were selected for the study. These two varieties bloomed from October to November each year, and fruits ripened in late October of the following year. The experimental materials were collected from an orchard in Wangcheng District, Changsha City, Hunan Province (latitude 28 • 05 N, longitude 113 • 21 E).

Pollination Treatment, Sample Collection and Cytological Observation
Three pollination treatments were designed, including (1) self-pollination (SP): "Hua Xin" × "Hua Xin", (2) cross-pollination (CP): "Hua Xin" × "Hua Jin" and (3) non-pollination (NP): pistil of "Hua Xin" after emasculation. Mature pollens of "Hua Xin" and "Hua Jin" were collected from anthers in advance during the bud stage and placed in sulfate paper bags and kept at 25 • C for 8 h. Mature pollen (Pn) samples of "Hua Xin" (6 g) were stored in liquid nitrogen for RNA-seq. Pollination treatments were conducted in the field between 9:00-11:00 AM or 1:00-3:00 PM on sunny days in October 2017. For SP and CP, the flowers were emasculated and pollinated. The stigmas were not stained by any pollen until they became receptive. After pollination treatments, all the pistils were protected by sulfate paper bags. At 48 h and 72 h after pollination, pistils were collected and stored at −80 • C. There were 7 groups of samples (SP48, SP72, CP48, CP72, NP48, NP72 and Pn) for transcriptome, with 3 biological replicates of each group. Four groups of samples (SP48, SP72, CP48 and CP72) were used for proteome and metabolome, with 3 and 6 biological replicates, respectively (Table S1). Each replicate contained 60 pistils. In addition, 120 pistils of each treatment were used for cytological observation with the paraffin sectioning method [53] and quantitative real-time PCR analysis.

PacBio cDNA Library Construction and Third-Generation Sequencing
Total RNA was extracted one by one from 21 samples (Table S1). Ten microliters of total RNA from each sample were pooled. The pooled RNA was reverse-transcribed into cDNA using Clontech SMARTer PCR cDNA Synthesis Kit (Takara, Dalian, China) according to the protocol provided by the manufacturer for third-generation sequencing. The SMRTbell template was annealed to the sequencing primer and bound to polymerase and sequenced on the PacBio RS II platform to obtain a representative full-length transcriptome for C. oleifera [79]. Proofreading error correction software was used for large-scale high-accuracy PacBio correction through iterative short-read consensus by second-generation sequencing [39]. The final transcriptome isoform sequences were filtered by removing the redundant sequences with software CD-HIT-v4.6.7 [80] [85]. The open reading frames (ORFs) were detected using the ANGLE software [86] for transcript sequences to obtain the coding sequences (CDS), protein sequences and UTR sequences.

Proteomic Analysis
In this experiment, proteomic changes were performed by isobaric tags for relative and absolute quantitation (iTRAQ) [91]. The pistil samples of the different pollination treatments were extracted by SDT lysis [92], and then the protein content was quantified by the BCA method. The appropriate amount of proteins in each sample were trypsinized by the Filter Aid Proteome preparation (FASP) method [92]. The hydrolyzed peptides were desalted by C18 cartridge and lyophilized and reconstituted with 40 µL dissolution buffer. The peptides (100 µg) of each sample were labeled according to the AB SCIEX iTRAQ Labeling Kit instructions [91]. Each set of labeled peptides was mixed and fractionated using AKTA Purifier 100. Each fractionated sample was separated using a HPLC liquid phase system Easy nLC at 300 nL/min. The sample was chromatographed and subjected to mass spectrometry using a Q-Exactive mass spectrometer. Protein database was translated from the ISO-seq database. The library Mascot (Version 2.2) and Proteome Discoverer (Version 1.4) were used for identification and quantitative analysis with FDR < 0.01 using the RAW files of mass spectrometry. Principal component analysis (PCA) provides the evaluation of the reliability of experimental results, as well as operational stability. GOs' annotation was performed on the target protein collection using Blast2GO [93], and the KEGG pathway annotation was performed on the target protein collection using KAAS (KEGG Automatic Annotation Server) software [94]. The DEPs were defined according to the standard of expression fold changed more than 1.2 times and p-value < 0.05.

Metabolomic Analysis
Metabolomics methods based on HILIC and RPLC UHPLC-Q-TOF MS techniques were used [95]. The XCMS program (http://enigma.lbl.gov/xcms-online/) was used for peak alignment, retention time correction and peak area extraction. Metabolite structure identification was performed using a method of accurate mass-matching (<25 ppm) and secondary spectral matching to retrieve a self-built database [96,97]. For the data extracted by XCMS, ion peaks with missing values >50% in the group were removed. SIMCA-P 14.1 (Umetrics, Umea, Sweden) was used for pattern recognition, and the data was preprocessed by Pareto-scaling. Multidimensional statistical analysis, including unsupervised PCA analysis, supervised partial least squares discriminant analysis (PLS-DA) and orthogonal partial least squares discriminant analysis (OPLS-DA), were performed. DEMs were identified using VIP (variable importance for the projection) >1 and p-value < 0.1 as screening criteria.

Transcript, Protein and Metabolite Correlation Analysis
Through the statistics of the data, the metabolic pathways in which DETs, DEPs and DEMs were detected in self-and cross-pollinated pistils in C. oleifera were analyzed. To compare the concordance among transcriptome, proteome and metabolome changes, trend and correlation analysis were performed on DETs ( Figure S4), DEPs ( Figure S6) and DEMs ( Figure S7). First, DETs with the same expression pattern could be found by trend analysis with online software (https://www.omicshare.com/ tools/Home/Soft/trend). Nine trends were selected with p < 0.05, in which the transcripts expression trends were closely related to LSI in C. oleifera. The Cor program in R Project [98] was used to calculate the Pearson's correlation coefficient of DETs, DEPs and DEMs. With the Pearson's correlation coefficient greater than 0.8 at the threshold, DEPs and DEMs with the same expression pattern of DETs were screened out.

Quantitative RT-PCR
Total RNA was extracted using plant RNA kit (OMEGA, Norcross, GA, city, state abbrev., USA) according to manufacturer's instructions, while RNase-free DNase I (Fermatas, Toronto city, Canada) was used to eliminate the remaining potential genomic DNA. The total RNA was reverse-transcribed into the first single-stranded cDNA by the PrimeScript TM RT reagent kit with gDNA Eraser (TaKaRa, Japan). All quantitative RT-PCR (qRT-PCR) were performed with the CFX96 TM real-time PCR system (BIO-RAD, Hercules, CA, city, state abbrev., USA) by SYBR Premix ExTaqTM (2×) (TaKaRa, Japan). CoGAPDH was used as the reference gene for C. oleifera [99]. The reactions were carried out in triplets using independent biological samples. The primers used were listed in Table S10. The mRNA expression levels were analyzed by Software Bio-Rad CFX Manager with 2 −∆∆CT method [100].

Statistical Analysis
All experimental data were expressed as the mean of three or six independent biological repeats and the standard deviation (mean ± SD). Statistical analyses were performed by SPSS software using the Duncan's test.

Availability of Sequence Data
The PacBio SMRT reads and the Illumina SGS reads generated in this study have been deposited in the NCBI Sequence Read Archive (http://www.ncbi.nlm.nih.gov/sra). The BioProject are PRJNA606665 and PRJNA606862, respectively. The mass spectrometry proteomics data by iTRAQ have been deposited to the ProteomeXchange Datasets (http://www.proteomexchange.org) with the submission reference 1-2020015-117203.  Figure S5. The Venn diagram of the common metabolic pathways shared by DETs, DEPs and DEMs in self-and cross-pollinated pistils at 48 and 72 hours after pollination. Table S1. The sample names of transcriptome, proteome, and metabolome.

Conflicts of Interest:
The authors declare no conflicts of interest.