The seeds of P. polyphylla var. yunnanensis were harvested from plants grown in Yunnan Province, China in October 2012, and the anther and other tissue samples were collected in March 2017. Paris Polyphylla plants were grown at an altitude of 1995 m, with an annual average temperature 14.26˚C and an annual average rainfall of 957.12 mm, in the germplasm resource garden of the Medicine Research Institute, which affiliate to Yunnan Academy of Agricultural Sciences, located in County of Songming, Kunming City, China. Fifteen samples, including seven closed and four opened samples, were collected from three plants. Furthermore, four tissue samples, including leaf, petal, calyx, and stigma were collected to analyze the changes in anther (Table 1). All tissue samples were immediately frozen in liquid nitrogen and stored at −80˚C until RNA extraction.

For Illumina sequencing, the total RNA was extracted using the RNeasy^® plant kit (BioTeke, Beijing, China) and quantified using an Agilent 2100 Bioanalyzer (Agilent technologies, USA). The total RNA concentration of all the samples was equal to or greater than 400 ng∙μL⁻¹; the OD_260/280 ranged from 1.8 to 2.2; the RNA 28S:18S was higher than 1.0; and the RNA integrity number was higher than 7.0. The RNA samples were enriched with NEBNext Poly (A) mRNA

Magnetic Isolation Module (New England Biolabs, USA), and then a cDNA library was constructed using NEBNext mRNA Library Prep Master Mix for Illumina (New England Biolabs, USA) and NEBNext Multiplex Oligos for Illumina (New England Biolabs, USA). The length of each cDNA fragment was verified by 1% agarose gel electrophoresis and quantified by the real-time PCR using the KAPA Library Quantification Kit by Illumina GA Universal (KAPA, USA). The quantified library was used for cluster generation with an Illumina cBot (Illumina Inc., USA), and then sequenced on an Illumina HiSeq^TM 2500 (Biomarker Biotechnological Co., Beijing, China).

The original paired-end reads were filtered to obtain the “clean reads” for de novo assembly. This process included the removal of reads with adaptor contamination or unknown nucleotides, and low-quality reads with ambiguous sequence “N.” The “clean reads” from all samples of P. polyphylla were used for the de novo assembly using Trinity software [16]. This software assembles the reads into longer transcripts to form contigs, clusters the contigs according to paired-end information and similarity among contigs, and selects the main transcripts as unigenes.

Functional annotations were performed by sequence comparison against public databases, including the NCBI NR (http://www.ncbi.nlm.nih.gov/), Swiss-Prot (http://www.expasy.ch/sprot/), COG (http://www.ncbi.nlm.nih.gov/cog/), and KEGG databases (http://www.genome.jp/keg/) using BLAST software (E-value < 1e⁻⁵). The functional assignments were mapped to GO terms (http://www.geneontology.org/), and classification was performed using WEGO software (http://wego.genomics.org.cn/cgi-bin/wego/index.pl). The reads were mapped to the assembled unigenes using Bowtie [17], and their expression levels were estimated using the Expectation-Maximization (RSEM) software [18]. Gene expression level was determined by calculating the reads per kilobase per million (RPKM) for mapped reads [19].

We detected the differentially expressed genes either using DESeq (), when the samples fit the replication required for an analysis, or by using EBSeq [20]. Benjamini-Hochberg method [21], using the FDR, was applied to correct the P value of the original hypothesis. Here, The false discovery rate (FDR) was set to a threshold of <0.01 [22], and fold change was set to a threshold of ≥2 as the criterion [23]. The unweighted pair-group method with arithmetic mean (UPGMA) cluster analysis was used to compare the relationship between differential unigene clusters.

After filtering out low-quality reads and vector sequences, we obtained 72.75 GB of data with 13,981,216 - 16,999,694 paired-end reads from each sample. Of the clean reads, more than 93% had Phred-scaled mapping scores equal to the corresponding error probability of 0.01 (Q20) (Table 2). These results showed that the quality of these sequences was reliable. Trinity [16] was used to mix the samples and assemble a consolidated unigene library that included 12,083,102 contigs, 280,777 transcripts, and 79,815 unigenes. The N50 length of transcripts was 1215 bp and the mean length of the unigenes was 1098 bp (Table 3). Bowtie analysis [17] revealed that each read had a BLAST efficiency of >62.19%.

Dehisced anthers of P. polyphylla close every evening and reopen every morning; to study the mechanism of this closure phenomenon, the relevant samples were classified as follows: before opening in the morning (sample T1), semi opened (T2), opened (T3), closing initiation (T4), semi closed (T5), fully closed (T6), and never opened samples (T7). Furthermore, the samples treated by watering included the following: opened before watering (T8), closed after watering (T9), semi opened after watering (T10), and fully opened after watering (Table 1). To analyze the DEGs involved in the regulation of opening and closing of anthers by external stimulus, the opened and closed samples due to night, the T1, T2, and T3 samples, which represent opened samples; and the T4, T5, and T6 samples (Table 1), which represent closed samples, were compared. The results showed that 1528 DEGs between closing and opening (log2FC > 1.5) occurred, of which 1515 were up-regulated (mean RPKM value of 30.75) and identified as the night closing cluster. In comparison, up-regulated unigenes were expressed at low levels or not at all during dehiscence (mean RPKM value of 1.3), suggesting that DEGs in charge of anther closing are primarily the result of the up-regulation of these unigenes. (Figure 1(a)).

As the anthers of P. polyphylla can close when watering, a 10-minute watering treatment was carried out. After watering, the anthers moved from a fully opened stage to closed, and were divided into opening before watering (T8), water-closing (T9, T10), and opening after watering (T11). There were DEGs between opening before or after watering (T8 and T11) and closing after watering (T9, T10), then a total of 1220 DEGs were obtained and named as water-closing; of them 901 unigenes were up-regulated and 319 unigenes were down-regulated (FDR < 0.01), and most of the 901 unigenes (762, 84.4%) overlap with closing cluster owing to night. The clusters of water-closing related unigenes and closing in evening contain 762 of the same unigenes (Figure 1(b)).

Significantly enriched pathways (Fisher’s exact test, P < 0.05) revealed that during the transition from the open to closed stages (including night closing and watering-induced closing), the DEGs involved in the pathway were mostly up-regulated. Conversely, during the transition from closed to open stage, the genes involved in this pathway were mostly down-regulated. In addition, the common pathway for night closing and watering-induced closing is consistently involved in oxidative phosphorylation (energy generation) and ribosomes (protein translation), but the gene cluster for opening contained fewer unigenes involved in energy pathways, meanwhile, those unigenes involved in energy metabolism during opening were down-regulated (Table 4). These results suggest that closing of dehisced anthers is an energy dissipation process that requires the synthesis of related proteins. Conversely, the opening process does not require such high levels of energy and protein synthesis. That is, the closing of dehiscent anther is an abnormal stress response process and opening is a naturally restorative process, therefore, we inferred that closure is a response to withstand adverse stress.

In order to analyze the DEGs involved in the regulation of opening and closing of anthers by external stimulus including night and watering, open samples T3 (opening in the morning), T8 (opening before watering) and T11 (opening after watering), and closed samples T1 (closing before opening in the morning), T6 (closing in the evening), T7 (never opened sample), and T9 (closing after watering) (Table 1), were compared. The clustering analyses based on the RPKM values showed that opening samples (T3, T8, T11) were prominently separated from the closest cluster (samples T1, T9, T6, T7). Differential expression analysis

Note: Up, upregulated; Down, down regulated.

showed the DEGs between the two clusters is reliable. A total of 9819 DEGs were obtained from comparation of opened and closed anther samples; 4679 (47.7%) unigenes were up-regulated and 5140 (52.3%) unigenes were down-regulated. In total, 6231 (63.5%) unigenes were annotated to KEGG, Swiss-Prot, and NR databases; among them, 3770 (60.5%) unigenes were down-regulated with the mean RPKM value of 7.7, and 2462 (25.1%) unigenes were up-regulated with the mean RPKM value of 30.75.

Of the 6321 annotated DEGs, the unigenes involved in “response to external stimuli” were prominent in all biological processes (Figure 2(a)). Further analysis was carried out based on speculation that the external stimuli might transfer a signal to the nucleus, and finally induce “DNA binding” or “chromatin binding”. In total, 215 unigenes were annotated to chromatin binding, the “chromatin binding” unigenes showed prominent in number (215 in 279, 77%). Further analysis showed that among the 215 unigenes, 120 (56%) unigenes (Figure 2(b)) are involved in silencing by epigenetic modification, including histone acetylation, histone H3-K9 methylation, histone lysine methylation, DNA methylation,

chromatin silencing by small RNA, miRNA synthesis involved in gene silencing by miRNA, RNA splicing, histone phosphorylation, histone H3-S10 phosphorylation, methylation-dependent chromatin silencing, posttranscriptional gene silencing by RNA, covalent chromatin modification, ubiquitination, and posttranscriptional gene silencing (Figure 2(c)). Interestingly, among the unigenes involving epigenetics, 50 were annotated to seed dormancy- or germination-related function simultaneously (Figure 2(d)). Moreover, 202 unigenes were located in mitochondria; among them, nine unigenes were involved in seed dormancy, germination, and development. These results showed that epigenetic modification genes have important roles in impacting seed dormancy, germination and development.

Furthermore, the results of KEGG pathways Enrichment analysis of DEGs between opening vs. closing and between anthers vs. control (leaf, petal, calyx, and stigma), showed the “plant hormone signal transduction” pathway is the most significant in former, but not in latter (Figure 3), and the variation in the expression of DEGs involved in the pathway analysis showed that the key genes, such as GID1, are up-regulated in the signal transduction to germination while closing, indicating down regulation of germination. Meanwhile, key genes, such as PYR/PYL and PP2C, which were down-regulated, and SnRK2 which up-regulated in the signal transduction were induced seed dormancy while closing (Figure 4), indicating a closing impact to seed dormancy through the “plant hormone signal transduction” pathway.

To verify whether frequent abnormal closing of anthers affects biological processes that follow pollination and fertilization, such as seed dormancy and germination, the seed dormancy-related unigenes obtained from the mesodermal transcriptome of P. polyphylla var. yunnanensis [11] were compared with unigenes involved in anther closure (night closing and water-induced closing). The results revealed the significantly DEGs owning to anther closing including in the all type of seed dormancy-related unigenes found in embryo of seed, including phytohormone related, seed maturation related, cell wall growth related, circadian rhythms, flavonol biosynthesis related, cytochrome P450 and others. Among them, at leat four genes differential expression in anther closing showed similar expression pattern in embryo of seed (Table 5). interestingly, the four genes including three phytohormone related genes and one circadian rhythm gene. The three phytohormone related genes indicate similar expression pattern with showed in “plant hormone signal transduction” (Figure 4), where gibberellins up-regulated, abscisic acid down-regulated and brassinosteroid up-regulated, all which impact on seed dormancy or germination.

Furthermore, as a contrast, genes involved in the regulatory network related to the seed germination rate in Arabidopsis were used to retrieve in the “up-regulated” DEGs between closing and opening, finally, the retrieved unigenes

Notes: The expression pattern of seed dormancy-related unigenes in embryo of seed which compared with endosperm was referred to the paper of embryo transcriptome (Jianjun Qi, 2013); a means the number of unigenes; *up or down in the brackets mean the gene expression up-regulated or down-regulated. The overstriking words means the gene show similar expression pattern in embryo and closing anther.

could form a complete network, which induce from closing to seed germination (Figure 5). These network characterizing genes which might induce anther closing, including cell wall modifying genes, aquaporins, ion channels, and hormones. Cell wall structural proteins, including vegetative cell wall protein gp1, glycine-rich cell wall structural protein, and wall-associated receptor kinase 5, were up-regulated in this gene cluster. The aquaporins including NIP1-1 and

TIP2-1, voltage-dependent anion-selective channels, and chloride intracellular channel exc-4 were among the unigenes that were up-regulated during water-closing. Hormone-related genes up-regulated during anther closure were found to be related primarily to abscisic acid (ABA), jasmonic acid (JA), and gibberellin (GA), with concomitant strong up-regulation of transmembrane proteins, kinases, receptors, and molecules related to hormonal signal transduction. Indeed, these up-regulated genes are in regulatory pathways that can inhibit seed germination and prolong seed dormancy. For example, inhibition of germination via ABA and related signal transduction genes may involve strong up-regulation of superoxide dismutase and reduction of reactive oxygen species to inhibit germination. Up-regulation of cell wall-modifying enzymes and aquaporins can potentially inhibit seed germination. In addition, water-closing functions via increased expression of gibberellin 3-oxidase (GA3OX) and dihydroflavonol-4-reductase (DFR) can inhibit germination; moreover, water deprivation

can inhibit germination via aquaporins. Furthermore, these networks also comprised many chromatin remodeling proteins including histone-lysine N-methyltransferase (HLN), a small nuclear ribonucleoprotein (SNR), histone deacetylase (HD), and others. It is verified that up-regulation of these genes, especially chromatin remodeling genes, enable epigenetic modifications that can affect cellular processes following anther germination.