SMRT and Illumina RNA Sequencing and Characterization of a Key NAC Gene LoNAC29 during the Flower Senescence in Lilium oriental ‘Siberia’

Lily (Lilium spp.) is an important cut flower around the world. Flower senescence in lilies is characterized by the wilting and abscission of tepals, which results in a decrease in flower quality and huge economic loss. However, the mechanism underlying flower senescence in lilies is largely unknown. In this study, single-molecule, real-time (SMRT) and Illumina sequencing were carried out in L. oriental ‘Siberia’. Sequencing yielded 73,218 non-redundant transcripts, with an N50 of 3792 bp. These data were further integrated with three published transcriptomes through cogent analysis, which yielded 62,960 transcripts, with an increase in N50 of 3935 bp. Analysis of differentially expressed genes showed that 319 transcription factors were highly upregulated during flower senescence. The expression of twelve NAC genes and eleven senescence-associated genes (SAGs) showed that LoNAC29 and LoSAG39 were highly expressed in senescent flowers. Transient overexpression of LoNAC29 and LoSAG39 in tepals of lily notably accelerated flower senescence, and the promoter activity of LoSAG39 was strongly induced by LoNAC29. This work supported new evidence for the molecular mechanism of flower senescence and provided better sequence data for further study in lilies.


Introduction
Senescence is a natural process in which organs lose the capacity for cell division and gradually die. During vegetative growth in plants, under nutrient deficiency or moderate stress, old leaves initiate the process of senescence, allowing plants to use limited nutrients to sustain young leaves or to resist environmental stress [1]. During reproductive growth, plants use flowers to attract insects or birds for pollination. Once pollination is complete, ethylene is synthesized in the ovary, causing the flowers to senesce rapidly and deliver nutrients to seeds or other organs, as well as to maintain normal plant growth [2].
Plants can be divided into ethylene-sensitive and -insensitive types [3]. For ethylenesensitive flowers, ethylene plays an important role in flower senescence. In tulips (Tulipa gesneriana), flower senescence is accompanied by large amounts of ethylene synthesis. ACS is a key gene in ethylene synthesis, which encodes 1-aminocyclopropane-1-carboxylate (ACC) synthase. Silencing of ACS significantly delays tepal senescence in tulips [4]. Most of the lily cultivars, however, are less sensitive or insensitive to ethylene, so ethylene inhibitor silver thiosulfate (STS) or 1-methylcyclopropene (1-MCP) has little effect on prolonging the flower life of cut lily L. oriental 'Stargazer' [5].
In addition to ethylene, hormones such as abscisic acid (ABA) also play a role in promoting flower senescence. ABA promotes flower senescence in both ethylene-sensitive and -insensitive plants. The increase in ABA content in lily flowers is closely related to to 25 °C. The air humidity was kept at 60%-70%. The development stages of the lily flower were defined according to Shi et al. [17], with minor modification: stage 1, flower bud grows to the largest size, color of tepals does not turn white; stage 2, early opening, the upper one third of the tepals were opened with an angle of 30 degrees; stage 3, full opening, the tepals extend to the maximum extent; stage 4, flowers are senescent, the tepals are wilting (Figure 1). Flowers of different stages were sampled at the same time, with three biological replicates for each stage. The samples were quickly wrapped in foil, frozen in liquid nitrogen, transported to the lab, and stored in a −80 °C freezer. These samples were prepared for SMRT and Illumina sequencing. As for transient overexpression in lily tepals, cut flowers of lily (L. oriental 'Siberia') were bought from Nanhu flower market and transported to the laboratory within one hour. The flower stems were cut to a length of 30 cm and the leaves at the stem base were removed; the flowers were then placed in vases with distilled water.
Potted flower of lily (L. asiatic 'Orange Matrix') was bought from Nanhu flower market before the flowers were opened and further cultivated in the greenhouse. Growth conditions were the same as with lily (L. oriental 'Siberia'). The flower opening process was also divided into four stages as above ( Figure S1). Flowers of different stages were sampled at the same time, with three biological replicates for each stage. The samples were quickly wrapped in foil, frozen in liquid nitrogen, transported to the lab, and stored in a −80 °C freezer. These samples were prepared for gene expression analysis.

RNA Extraction, Illumina Sequencing, and SMRT Sequencing
RNA was extracted from lily samples with the hot borate method [21]. The concentration, purity, and integrity of RNA was checked with Agilent 2100 (Agilent, Palo Alto, CA, USA), Nanodrop 2000 (Thermo Fisher Scientific, Waltham, MA, USA), and agarose gel electrophoresis. After that, Illumina sequencing was carried out on Illumina HiSeq 2500 platform. Based on the quality of RNA, among the three biological repeats, the best one of each stage was chosen, and mixed to form one sample for SMRT sequencing with the PacBio RSII platform. Sequencing was finished by Novogene Bio Technology Co., Ltd. in Beijing, China. For raw data of SMRT sequencing, SMRTlink V6.0 (parameters: -minLength 50; -maxLength 15,000; -minPasses 2; -minPredictedAccuracy 0.8; -minZScore −9999; -maxDropFraction 0.8; -min_seq_len 200; -minReadScore 0.65) was used to remove adapters and low-quality reads, thus obtaining the subreads. After selfcorrection between the subreads, circular consensus sequences (CCSs) were obtained. The full-length non-chimeric (FLNC) and non-full-length (NFL) sequences were divided, according to whether the sequences contained 5′ and 3′ adapters and poly(A) tails. Cluster consensus sequences were obtained by clustering FLNC and sequences from the same transcript using the iterative clustering for error correction (ICE) algorithm [22]. Finally, As for transient overexpression in lily tepals, cut flowers of lily (L. oriental 'Siberia') were bought from Nanhu flower market and transported to the laboratory within one hour. The flower stems were cut to a length of 30 cm and the leaves at the stem base were removed; the flowers were then placed in vases with distilled water.
Potted flower of lily (L. asiatic 'Orange Matrix') was bought from Nanhu flower market before the flowers were opened and further cultivated in the greenhouse. Growth conditions were the same as with lily (L. oriental 'Siberia'). The flower opening process was also divided into four stages as above ( Figure S1). Flowers of different stages were sampled at the same time, with three biological replicates for each stage. The samples were quickly wrapped in foil, frozen in liquid nitrogen, transported to the lab, and stored in a −80 • C freezer. These samples were prepared for gene expression analysis.

RNA Extraction, Illumina Sequencing, and SMRT Sequencing
RNA was extracted from lily samples with the hot borate method [21]. The concentration, purity, and integrity of RNA was checked with Agilent 2100 (Agilent, Palo Alto, CA, USA), Nanodrop 2000 (Thermo Fisher Scientific, Waltham, MA, USA), and agarose gel electrophoresis. After that, Illumina sequencing was carried out on Illumina HiSeq 2500 platform. Based on the quality of RNA, among the three biological repeats, the best one of each stage was chosen, and mixed to form one sample for SMRT sequencing with the PacBio RSII platform. Sequencing was finished by Novogene Bio Technology Co., Ltd. in Beijing, China. For raw data of SMRT sequencing, SMRTlink V6.0 (parameters: -minLength 50; -maxLength 15,000; -minPasses 2; -minPredictedAccuracy 0.8; -minZScore −9999; -maxDropFraction 0.8; -min_seq_len 200; -minReadScore 0.65) was used to remove adapters and low-quality reads, thus obtaining the subreads. After self-correction between the subreads, circular consensus sequences (CCSs) were obtained. The full-length non-chimeric (FLNC) and non-full-length (NFL) sequences were divided, according to whether the sequences contained 5 and 3 adapters and poly(A) tails. Cluster consensus sequences were obtained by clustering FLNC and sequences from the same transcript using the iterative clustering for error correction (ICE) algorithm [22]. Finally, the NFL sequences were used to polish the consistent sequence with the Arrow algorithm (https://downloads. pacbcloud.com/public/software/installers/smrtlink_9.0.0.92188.zip, accessed on 10 May 2019) [23], and the high-quality polished consensus sequence was obtained for further analysis. Illumina sequencing data were used for further proofing the polished consensus sequence with LoRDEC software (parameters: -k 23, -s 3) [24], followed by redundancy removal with CD-HIT software with the following parameters: -c 0.99, -G 0, -aL 0.00, -aS 0.99, -AS 30, -M 0, -d 0, -p 1 [25]. Next, the non-redundant transcripts were generated and further annotated with seven public databases: NCBI non-redundant protein sequences (NR), NCBI non-redundant nucleotide sequences (NT), Protein family (Pfam), Clusters of Orthologous Groups of proteins (KOG/COG), a manually annotated and reviewed protein sequence database (Swiss-Prot), KEGG Ortholog database (KO), and Gene Ontology (GO).

Coding GENome Reconstruction Tool (Cogent) Analysis
Cogent analysis was carried out with default parameters, according to the method described in [27]. The k-mer profiles of non-redundant transcripts from our SMRT sequencing (group B) and three published transcriptomes based on Illumina sequencing in lilies [16][17][18] (group A) were created, pairwise distances of transcripts were calculated, and similar transcripts were clustered into families based on their k-mer values. For each family, similar transcripts were reconstructed into one new transcript by using a De Bruijn graph method. In the De Bruijn graph, Cogent analysis was carried out based on the coverage and identity of the overlap. For example, An from group A and Bm from group B were clustered into the same family. If the identity <95%, return Bm (category one) and, if the coverage >95%, the identity >95%, and the length of An < Bm, return Bm (category two). Category one and category two were merged and named Pacbio.un.blast. If the coverage >95%, the identity >95%, and the length of An > Bm, the sequence of An was assigned to Bm (category three, named Final); if the coverage <95% and the identity >95%, the sequence of An and Bm were reconstructed into one new transcript (category four, named Cogent) ( Figure 2).

TransDecoder Analysis
TransDecoder analysis (default parameters) was carried out according to the method described in [28]. Firstly, a minimum length of opening reading frame (ORF) (default 100 aa) was identified in the transcripts, to further maximize the sensitivity of the functional ORF. The ORF was compared with the UniProt and PFAM databases to identify the common protein domain. Finally, the final prediction was scored based on the alignment in the two databases. Based on the TransDecoder analysis, the transcripts were classified into four categories: complete, 5prime_partial, 3prime_partial, and internal. Complete means the sequence contains the complete opening reading frame (ORF), 5prime_partial means the 5 end of the sequence was missing, 3prime_partial means the 3 end was missing, and internal means both 5 and 3 parts were missing. For genes which were predicted to have multiple ORFs, the one with the highest score was reserved. For genes which were predicted as complete, 5prime_partial, or 3prime_partial, if the score was less than −20, they were merged with the internal category and renamed 'internal and others'. Genes 2021, 12, x FOR PEER REVIEW 5 of 19

Figure 2.
A scheme for cogent analysis of SMRT and Illumina sequencing data. Assembled Illumina sequencing data (group A) were compared with non-redundant transcripts coming from SMRT sequencing data (group B), and pairwise distances of transcripts were calculated. Based on the k-mer values, similar transcripts were clustered into the same family, and similar transcripts were reconstructed into one new transcript by using a De Bruijn graph method. In the De Bruijn graph, Cogent analysis was carried out based on the coverage and identity of the overlap. For example, An from group A and Bm from group B were clustered into the same family. If the identity <95%, return Bm (category one) and, if the coverage >95%, the identity >95%, and the length of An < Bm, return Bm (category two). Category one and category two were merged and named Pacbio.un.blast. If the coverage >95%, the identity >95%, and the length of An > Bm, the sequence of An was assigned to Bm (category three, named Final); if the coverage <95% and the identity >95%, the sequence of An and Bm were reconstructed into one new transcript (category four, named Cogent).

TransDecoder Analysis
TransDecoder analysis (default parameters) was carried out according to the method described in [28]. Firstly, a minimum length of opening reading frame (ORF) (default 100 aa) was identified in the transcripts, to further maximize the sensitivity of the functional ORF. The ORF was compared with the UniProt and PFAM databases to identify the common protein domain. Finally, the final prediction was scored based on the alignment in the two databases. Based on the TransDecoder analysis, the transcripts were classified into four categories: complete, 5prime_partial, 3prime_partial, and internal. Complete means the sequence contains the complete opening reading frame (ORF), 5prime_partial means the 5′ end of the sequence was missing, 3prime_partial means the 3′ end was missing, and internal means both 5′ and 3′ parts were missing. For genes which were predicted to have multiple ORFs, the one with the highest score was reserved. For genes which were predicted as complete, 5prime_partial, or 3prime_partial, if the score was less than −20, they were merged with the internal category and renamed 'internal and others'.

Quantitative Real-Time (qRT)-PCR Analysis
For qRT-PCR, the cDNA template was synthesized with 1 μg RNA by using cDNA Synthesis SuperMix (AE311-03, TransGen Biotech, Beijing, China) according to the manufacturer's instructions. The primers used for qRT-PCR were designed based on the SMRT sequencing data in lily (L. oriental 'Siberia'), as listed in Table S1. ACTIN in L. oriental 'Siberia' (LoACTIN) was used as the internal control gene, according to [29]. All reactions A scheme for cogent analysis of SMRT and Illumina sequencing data. Assembled Illumina sequencing data (group A) were compared with non-redundant transcripts coming from SMRT sequencing data (group B), and pairwise distances of transcripts were calculated. Based on the k-mer values, similar transcripts were clustered into the same family, and similar transcripts were reconstructed into one new transcript by using a De Bruijn graph method. In the De Bruijn graph, Cogent analysis was carried out based on the coverage and identity of the overlap. For example, An from group A and Bm from group B were clustered into the same family. If the identity <95%, return Bm (category one) and, if the coverage >95%, the identity >95%, and the length of An < Bm, return Bm (category two). Category one and category two were merged and named Pacbio.un.blast. If the coverage >95%, the identity >95%, and the length of An > Bm, the sequence of An was assigned to Bm (category three, named Final); if the coverage <95% and the identity >95%, the sequence of An and Bm were reconstructed into one new transcript (category four, named Cogent).

Quantitative Real-Time (qRT)-PCR Analysis
For qRT-PCR, the cDNA template was synthesized with 1 µg RNA by using cDNA Synthesis SuperMix (AE311-03, TransGen Biotech, Beijing, China) according to the manufacturer's instructions. The primers used for qRT-PCR were designed based on the SMRT sequencing data in lily (L. oriental 'Siberia'), as listed in Table S1. ACTIN in L. oriental 'Siberia' (LoACTIN) was used as the internal control gene, according to [29]. All reactions were performed with three biological replicates on LightCycler 96 (Roche, Basel, Switzerland). The data were calculated with 2 −∆∆CT method [30]. Two-sided Student's t-tests (* p < 0.05; ** p < 0.01) were used for statistical analysis.

Gene Cloning and Vector Construction
The sequences of LoNAC29 and LoSAG39 were extracted from the SMRT sequencing data, and the promoter of LoSAG39 was obtained by using fusion primer and nested integrated (FPNI)-PCR [31]. The sequences were corrected by using high-fidelity Phusion DNA Polymerase (Thermo Fisher Scientific, Waltham, MA, USA).
For overexpression (OE) of LoNAC29 and LoSAG39, the ORF carrying the proper restriction site was amplified by PCR, digested, and inserted into the pSuper1300 vector, which was derived from pCAMBIA1300 (laboratory of Dr. Zhizhong Gong, China Agricultural University). For promoter activity analysis, the promoter of LoSAG39 harboring the appropriate restriction site was digested and inserted into the PBI121 vector.

Transient Overexpression in Lily Tepals
Overexpression of LoNAC29 and LoSAG39 in lily tepals was performed as previously described [12] with minor modification. The pSuper-LoNAC29, pSuper-LoSAG39 or pSuper (empty vector) were transformed into Agrobacterium tumefaciens GV3101 separately. After culture overnight, the cells were harvested and resuspended in infiltration buffer (10 mM 2morpholinoethanesulfonic acid [MES], 10 mM MgCl 2 , 20 µM acetosyringone [As], pH 5.6). OD 600 nm was adjusted to 0.8 and incubated in the dark without shaking for two hours. Lily tepal discs with a diameter of 12 mm were obtained using a hole punch and infiltrated with A. tumefaciens GV3101 cells under a vacuum (0.6 atm). The discs were then washed in deionized water and cultured on a Petri dish with wet filter paper in the dark at 8 • C for three days. After that, the Petri dish containing the discs were taken out, and the time point was defined as 0 day. Thereafter, the discs were cultured in Petri dishes at 23 • C and 60% relative humidity under 5000 lux light.

Transient Transactivation in Tobacco (Nicotiana benthamiana)
The effector pSuper1300 empty vector, or pSuper::LoNAC29, and reporter PBI121 empty vector, or PBI121 harboring the promoter of LoSAG39, were transferred into A. tumefaciens GV3101. After overnight culture, cells were collected and resuspended in the infiltration solution as above, and the density was adjusted to OD 600 nm = 1.6. Agrobacterium carrying the effector or reporter were mixed with the ratio 1:1 (v/v) and incubated in dark at 23 • C without shaking for two hours. Agrobacterium were then injected into the back of N. benthamiana leaves using a 1 mL syringe (needle was removed before injection). After cultivation in the dark at 23 • C and a relative humidity of 40-60% for three days, the leaves of treated N. benthamiana were collected, immediately frozen in liquid nitrogen, and stored at −80 • C.

Determination of β-Glucuronidase (GUS) Activity
Determination of GUS activity was performed according to [32]. Briefly, treated leaves of N. benthamiana were ground in liquid nitrogen, and the soluble protein was extracted with extraction buffer. To measure the GUS activity, 4-Methylumbelliferyl β-D-glucuronide hydrate (4-MUG) was used as a substrate. The fluorescence of the 4-MU product was measured with a fluorescence spectrophotometer (F-4500; Hitachi, Tokyo, Japan) at 365 nm excitation and 455 nm emission. The GUS activity was defined as the amount of 4-MU produced per minute (unit: µM 4-MU mg −1 (protein) min −1 ).

SMRT Sequencing of Lily during Flower Opening
Through PacBio Sequel platform, 454,080 polymerase reads were obtained. The total data were 23.26 G nucleotides, and the average length was 51,217 bp. After removing the adapter and the original reads whose length is less than 50 bp, 13,782,679 subreads were obtained, with an average length and N50 of 1613 bp and 2281 bp, respectively. After checking the consistency of subreads going through the ZMW wells, 396,493 CCS (circular consensus sequence) sequences were obtained, with a mean length of 2397 bp and N50 of 3240 bp. Based on the 5 primer, 3 primer, and poly A, 310,912 FLNC sequences were found in CCS, with a proportion of 78.42%. After removing the redundancy, 73,218 nonredundant transcripts were obtained, with the total number of 203,392,716 nucleotides and N50 of 3792 bp (Table 1 and Table S2). The result showed that 60,676 unigenes have at least one annotation, and 16,774 unigenes have annotations in all seven databases ( Figure S2 and Table S3). Among the 56,644 unigenes annotated in NR database, 12,347 unigenes with top hits to Elaeis guineensis homolog genes, followed by Phoenix dactylifera (10,145 unigenes), Ananas comosus (3415 unigenes), Musa acuminata (3277 unigenes), Asparagus officinalis (2417 unigenes), and Anthurium amnicola (2350 unigenes) ( Figure S2).

Cogent and TransDecoder Analysis of SMRT Sequencing Data
Several Illumina sequencing data in lily were published recently [16][17][18], which supplied very important information for the study of lily, since its genome sequencing has not been reported yet. In order to make good use of the published transcriptome data to optimize the SMRT sequencing data, Cogent analysis was conducted. After Cogent analysis, 36,471 Pacbio.un.blast sequences, 16,167 Final sequences and 10,322 Cogent sequences were obtained. The total was 62,960, composed of 186,674,771 nucleotides, and the N50 was 3935 bp (Table 1 and Table S4). The quality of data was much higher than the original SMRT sequencing data and published second-generation transcriptome data.
The integrated sequencing data were further analyzed with TransDecoder. The result showed that in the original SMRT sequencing data, 43,998 sequences (60%) had the full length, 12,938 sequences (18%) were 5prime_partial, 699 sequences (1%) were 3prime_partial, and 15,583 sequences (21%) were internal and others which were missing both the 5 and 3 ends or the score of alignment in UniProt and PFAM databases were lower than −20. After Cogent analysis, 42,027 sequences (67%) were full length, 7774 (12%) and 548 (1%) were missing their 5 and 3 end, respectively, and 12,611 (20%) were internal and others (Table 2).

Analysis of DEG and Transcription Factor (TF) during Flower Opening
By using DESeq software, DEGs (fold change > 2, padj < 0.05) were determined between different opening stages in lily. Stage 1 versus stage 4 pairwise had the most DEG genes, 5071 genes were upregulated and 10,172 genes were downregulated. In contrast, stage 2 versus stage 3 had the least DEG genes ( Figure S3). The DEGs between different stages were gathered, and a total of 20,650 DEGs were obtained (Table S5). Clustering of 20,650 DEGs revealed that 665 genes were DEG in all four pairwises, including stage 1 versus stage 3, stage 1 versus stage 4, stage 2 versus stage 4, and stage 3 versus stage 4, and, compared with other stages, 5135 genes had a high expression level in stage 4, while 3554 genes had a low expression level in stage 4 ( Figure 3 and Table S5). Further, 20,650 DEGs were clustered into 12 subclusters. In subcluster 5, subcluster 7, and subcluster 8, the expression of DEGs was much higher in stage 4 than other stages. Meanwhile, in subcluster 1, subcluster 2, and subcluster 11, the DEGs had a much lower expression in stage 4 than in other stages ( Figure S4).

Analysis of Color-and Scent-Related Genes during Flower Opening
During stage 1 of the flower opening in L. oriental 'Siberia', the tepals were green, and with the opening of flowers, they gradually turned white ( Figure 1). The chlorophyll content was more than two times higher in stage 1 than in other stages, and there was no notable difference between stage 2, stage 3, and stage 4 ( Figure S5). Chlorophyll metabolism was tightly related to photosynthesis. The KEGG pathway analysis showed photosynthesis and the chlorophyll metabolism pathway were enriched in stage 1 versus stage 2, stage 1 versus stage 3, and stage 1 versus stage 4. Next, 29 genes related to chlorophyll metabolism and 51 genes related to photosynthesis were downregulated in stage 2, stage 3, and stage 4 compared with stage 1 (Tables S8 and S9). Most of the genes related to chlorophyll metabolism were classified into subcluster 1, subcluster 2, and subcluster 11. Additionally, their expression was highly correlated with 114 transcription factors ( Figure S4 and Table S10), whether these transcription factors were involved in regulation of chlorophyll metabolism will be checked in the further studies.
Flower opening is often accompanied with a release of flower scent. Terpenoids are major components in the flower scent of L. oriental 'Siberia', and the release of flower scent in lily mainly occurs in the full-flowering period (stage 3) [17]. KEGG pathway analysis indicated that the terpenoid backbone biosynthesis pathway was enriched in stage 1 versus stage 2 and stage 1 versus stage 3, and the expression of 54 genes related to terpenoid synthesis increased with flower opening, peaked at stage 3, and dropped at stage 4 (Tables S11 and S12). There were 30 genes related to terpenoid biosynthesis that were enriched in subcluster 9 and subcluster 10. Their expression pattern was in accordance with Genes 2021, 12, 869 9 of 18 28 transcription factors ( Figure S4 and Table S13), and some of them were being verified whether they participated in terpenoid synthesis.

Analysis of Senescence-Associated Genes during Flower Opening
Visible flower senescence occurred in stage 4, and DEG analysis indicated that, compared with the other three stages, 5135 genes were strongly induced in stage 4 (Table S5). Among them, eleven genes, annotated as senescence-specific cysteine protease, KDEL cysteine protease, or senescence-associated protein, were selected (Table S14). qRT-PCR analysis showed that the expression of the 11 SAG genes sharply increased in stage 4 ( Figure 4). Among these 11 SAG genes, the expression of SAG1 is more obvious.
major components in the flower scent of L. oriental 'Siberia', and the release of flower scent in lily mainly occurs in the full-flowering period (stage 3) [17]. KEGG pathway analysis indicated that the terpenoid backbone biosynthesis pathway was enriched in stage 1 versus stage 2 and stage 1 versus stage 3, and the expression of 54 genes related to terpenoid synthesis increased with flower opening, peaked at stage 3, and dropped at stage 4 (Tables S11 and S12). There were 30 genes related to terpenoid biosynthesis that were enriched in subcluster 9 and subcluster 10. Their expression pattern was in accordance with 28 transcription factors ( Figure S4 and Table S13), and some of them were being verified whether they participated in terpenoid synthesis.

Analysis of Senescence-Associated Genes during Flower Opening
Visible flower senescence occurred in stage 4, and DEG analysis indicated that, compared with the other three stages, 5135 genes were strongly induced in stage 4 (Table S5). Among them, eleven genes, annotated as senescence-specific cysteine protease, KDEL cysteine protease, or senescence-associated protein, were selected (Table S14). qRT-PCR analysis showed that the expression of the 11 SAG genes sharply increased in stage 4 ( Figure  4). Among these 11 SAG genes, the expression of SAG1 is more obvious.  The expression of the 11 SAG genes was also checked in different organs, including root, stem, mature leaf, flower, and senescent leaf. Relative to the expression in the root, the expression of SAG1, SAG5, and SAG8 was extremely high in the flower ( Figure S6). When compared with the mature leaf, the expression of SAG5, SAG6, and SAG8 was more than 30,000 times that of the senescent leaf ( Figure S7), which suggests that these genes might be also involved in leaf senescence. Since the expression of SAG1 was more obvious in flowers undergoing senescence, the full length of SAG1 was cloned, and phylogenetic analysis showed that the protein sequence of SAG1 had high homology with the SAG39 proteins in multiple species, so the gene was named LoSAG39 (Figure S8, Genbank accession no. MW520943).
Transient overexpression of LoSAG39 was carried out in lily tepal discs. The results showed that, after four days, in the super1300 vector control samples, about half of the tepal discs were senescent, while in LoSAG39 OE samples, nearly 100% of the tepal discs showed the senescent phenotype. qRT-PCR analysis indicated that the expression of LoSAG39 remarkably increased, compared with control samples. The ion leaking rate also showed that the damage of the cell membrane system is more serious in LoSAG39 OE samples ( Figure 5). Transient overexpression of LoSAG39 was carried out in lily tepal discs. The results showed that, after four days, in the super1300 vector control samples, about half of the tepal discs were senescent, while in LoSAG39 OE samples, nearly 100% of the tepal discs showed the senescent phenotype. qRT-PCR analysis indicated that the expression of LoSAG39 remarkably increased, compared with control samples. The ion leaking rate also showed that the damage of the cell membrane system is more serious in LoSAG39 OE samples ( Figure 5).

LoNAC29 Accelerated Flower Senescence Partly by Activating Several SAGs
Analysis of TF showed that 37.33% (28/75) NAC genes were upregulated during stage 4, when flowers were going through aging. We selected 12 NAC genes for further analysis ( Figure 3 and Table S6). The expression of the 12 genes was checked in stage 1 to stage 4 of L. oriental 'Siberia' flowers, the results indicated that most of them, including NAC2, NAC3, NAC4, NAC5, and NAC10, were sharply induced in stage 4 ( Figure 6). To confirm whether the high expression of NAC genes in stage 4 was related to flower senescence, the expression of NAC2 and NAC4 were also checked in different opening stages of lily (L. Asiatic 'Orange Matrix') flowers. The result showed that the expression of NAC2 rapidly increased more than 1500 times in stage 3 and 2000 times in stage 4, compared with stage 1. However, the change in NAC4 was not as strong as NAC2. Additionally, both NAC2 and NAC4 were induced by ABA treatment (Figure S9).
whether the high expression of NAC genes in stage 4 was related to flower senescence, the expression of NAC2 and NAC4 were also checked in different opening stages of lily (L. Asiatic 'Orange Matrix') flowers. The result showed that the expression of NAC2 rapidly increased more than 1500 times in stage 3 and 2000 times in stage 4, compared with stage 1. However, the change in NAC4 was not as strong as NAC2. Additionally, both NAC2 and NAC4 were induced by ABA treatment (Figure S9). The full length of NAC2 was cloned, and phylogenetic analysis showed that it had high homology with AtNAC29 in Arabidopsis thaliana, so the gene was named LoNAC29 ( Figure S10, Genbank accession no. MW548579). Overexpression of LoNAC29 significantly accelerated the senescence of lily flowers. In LoNAC29 OE samples, the expression of LoNAC29 was significantly upregulated compared with pSuper vector control, and the ion leakage rate was also much higher in LoNAC29 OE samples, which suggested that overexpression of LoNAC29 resulted in more serious flower senescence (Figure 7). The full length of NAC2 was cloned, and phylogenetic analysis showed that it had high homology with AtNAC29 in Arabidopsis thaliana, so the gene was named LoNAC29 ( Figure S10, Genbank accession no. MW548579). Overexpression of LoNAC29 significantly accelerated the senescence of lily flowers. In LoNAC29 OE samples, the expression of LoNAC29 was significantly upregulated compared with pSuper vector control, and the ion leakage rate was also much higher in LoNAC29 OE samples, which suggested that overexpression of LoNAC29 resulted in more serious flower senescence (Figure 7).
In LoNAC29 OE samples, the expression of LoSAG39, SAG2, and SAG7 was notably upregulated ( Figure 8A). The promoter of LoSAG39 was cloned by using FPNI-PCR, and the promoter fragment with a length of 734 bp was obtained. Transient transactivation assay was carried out in the leaves of N. benthamiana, and the result showed that the promoter activity of LoSAG39 was also significantly induced by LoNAC29 ( Figure 8B). In LoNAC29 OE samples, the expression of LoSAG39, SAG2, and SAG7 was notably upregulated ( Figure 8A). The promoter of LoSAG39 was cloned by using FPNI-PCR, and the promoter fragment with a length of 734 bp was obtained. Transient transactivation assay was carried out in the leaves of N. benthamiana, and the result showed that the promoter activity of LoSAG39 was also significantly induced by LoNAC29 ( Figure 8B).

Discussion
The flower is an important organ of angiosperm that evolved from leaves [38]. In cut flowers, the supply of water and nutrients are cut off after harvest, and how to maintain flower quality is a big problem. The cut lily flowers cannot tolerate a long period of storage and transportation, and the life of the flowers is relatively short. Delaying the senescence

Discussion
The flower is an important organ of angiosperm that evolved from leaves [38]. In cut flowers, the supply of water and nutrients are cut off after harvest, and how to maintain flower quality is a big problem. The cut lily flowers cannot tolerate a long period of storage and transportation, and the life of the flowers is relatively short. Delaying the senescence of the cut lily flowers is important for improving the ornamental value and prolonging the vase life of cut lily flowers.

Integration of SMRT and Illumina Sequencing Is an Efficient Approach for Study in Plants without a Reference Genome
Ploidy of lilies (L. spp.) is very complex, as most of the cultivars are diploid or tetraploid, and some are triploid. L. oriental 'Siberia' is diploid (2n = 2x = 24) [39]. The genome of lilies has not been reported yet, but it is estimated to be 36 Gb in size, since it contains a large number of repeats [40]. Previous studies on lilies have focused on the formation of floral color and fragrance, heat tolerance, and bulb dormancy. Due to the lack of a reference genome, many studies used Illumina sequencing to obtain gene sequences and study the function of candidate genes [18,19]. However, in Illumina sequencing, transcripts are typically cut into multiple pieces. Due to the short length of reads, the assembled transcripts were often not complete, thereby impeding the cloning of target genes and further study of gene function.
The third-generation sequencing technology, represented by Pacbio SMRT, developed rapidly and played an important role in biological research [20]. In this study, we used SMRT and Illumina sequencing to study the molecular mechanism of flower senescence in lily. A total of 73,218 transcripts were obtained, and the N50 was 3792 bp. In contrast, in the published transcriptome of lily based on Illumina sequencing, the N50 was 981 bp [16], 1038 bp [17], and 1443 bp [18] (Table S15). Further, we combined our SMRT sequencing data with the published Illumina sequencing data for Cogent analysis. We obtained 62,960 transcripts, and the N50 was further extended to 3935 bp, which was much better than the published transcriptome data of lily (L. spp.). Among the 62,960 transcripts, 42,027 (67% of the sequences) were predicted to have the full length by TransDecoder analysis, which was also better than our original SMRT sequencing data ( Table 2).
In terms of the regulatory mechanism underlying flower senescence, it was reported that an ethylene-induced HD ZIP transcription factor RhHB1 inhibits GA20ox1, a key gene for gibberellin synthesis, thus accelerating flower senescence [10]. On the other hand, RhERF113 promotes the synthesis of cytokinin and thus delays the senescence of rose flowers [11]. In this study, 2081 transcription factor genes were screened by SMRT and Illumina sequencing, and 319 of them were significantly upregulated at stage 4, compared with the other three stages. Among these upregulated transcription factors, many genes come from the NAC, WRKY, bHLH, and C2H2 Zinc finger families. Their expression displayed more significant changes ( Figure 3 and Table S6).
Since the discovery of a NAC gene AtNAP (ANAC029) that plays a key role in Arabidopsis leaf senescence [8], many studies have focused on the identification of key NAC genes in other species [45]. It was found that NAC genes promoted leaf senescence partly through inducing chlorophyll degradation [33], ABA synthesis [46], or ROS accumulation [47].
Some studies have also shown that the NAC gene regulates flower senescence in Japanese morning glory (Ipomoea nil cv. Violet). Knock-out of NAC gene EPHEMERAL 1 significantly delays senescence in Japanese morning glory (Ipomoea nil) [9]. However, the detailed regulatory mechanism of the NAC gene in flower senescence is not clear.
We searched the homologous genes in our SMRT sequencing data with published transcription factors which were reported to be involved in leaf or flower senescence and found 13 NAC unigenes, as well as 1 HB-HD-ZIP unigene and 1 ERF unigene were homologous to ANAC029 (AtNAP) [8], ANAC046 [33], ANAC032 [34], ANAC087 [35], ANAC022(NAC1) [36], ANAC002 (ATAF1) [37], Ipomoea nil EPHEMERAL 1 (NAC092) [9], RhHB1 [10], and RhERF113 [11] (Table S7), and their expression was significantly upregulated at stage 4, suggesting that they might be important candidate transcription factors that regulate flower senescence in lilies. In addition, we confirmed that LoNAC29 is a key regulator in the process of lily flower senescence, and OE of LoNAC29 significantly promotes the senescence of lily tepals. In the LoNAC29 OE samples, the expression of LoSAG39, SAG2, and SAG7 was significantly increased, and LoNAC29 strongly induced the promoter activity of LoSAG39 (Figure 8). LoSAG39, SAG2, and SAG7 are cysteine proteases, which may play a role in protein degradation. Overexpression of LoSAG39 also accelerated the senescence of lily tepals ( Figure 5).

Changes of Flower Color and Scent Are Accompanied by Flower Senescence in L. oriental 'Siberia'
The process of flower senescence in lily involves the degradation of proteins as well as chlorophyll degradation. Here, in L. oriental 'Siberia', the flower color is green at stage 1 and, with the opening of flowers, the tepals gradually turn white. White flowers are formed partly as a result of chlorophyll degradation [48] and partly as a result of genetic mutations or reduced expression levels of anthocyanin and carotenoid synthesis-related genes [49,50]. Similarly, in this study, we also found that chlorophyll content decreased with flower opening in L. oriental 'Siberia' and, accordingly, the expression level of chlorophyll synthesis-related genes and photosynthesis-related genes also decreased gradually (Table S9).
Some studies have also reported the mechanism of chlorophyll metabolism and regulation in green flowers. For example, in chrysanthemum (Chrysanthemum morifolium Ramat.), a NAC transcription factor was found to mediate chlorophyll synthesis partly through direct activation of HEMA1 and CRD1 genes [51]. In L. oriental 'Siberia', most of the genes involved in chlorophyll synthesis and photosynthesis were clustered into subcluster 1, subcluster 2, and subcluster 11. These clusters contained 114 transcription factors ( Figure S4, Table S10). In the process of flower opening, the decline of chlorophyll synthesis accompanies flower senescence, and whether these transcription factors were involved in chlorophyll synthesis or flower senescence needs further investigation. Metabolism of chlorophyll was tightly correlated with the accumulation of photosynthetic products, and carbohydrates, as energy substances, play an important role in delaying flower senescence [52]. It was speculated that preventing chlorophyll degradation, or promoting its synthesis, may be a new strategy for delaying flower senescence.
The flower opening of lilies (L. oriental 'Siberia') is also accompanied by the release of flower scent. The main components of floral fragrance are terpenoids, benzenoids/phenylpropanoid, and derivatives of fatty acids. Terpenes are synthesized by the MEP and MVA pathway. The main aroma components in L. oriental 'Siberia', cis-ocimene, linalool, and other terpenes, are released in large amounts at full bloom stage [18]. In this study, we also found that 68 MEP and MVA genes were expressed at full bloom stage (stage 3). At present, a few scant reports indicate that synthesis of terpene is regulated by MYC2 and MYB21 in freesia (Freesia hybrida) flowers [53] and ERF71 in the peel of sweet orange (Citrus sinensis Osbeck) [54]. Here, in cluster analysis, 28 transcription factors were found to be associated with the expression patterns of genes involved in terpene biosynthesis (Table S13). Among them, there were five MYB unigenes, including i0_HQ_lily14_c35718/f4p7/940, i1_HQ_lily14_c50407/f3p4/1037, i1_LQ_lily14_c39458/f1p4/1091, i0_LQ_lily14_c41028/f1p0/945, and i1_HQ_lily14_c99358 /f4p4/1023; the function of related genes needs to be further studied. It was reported that the flower longevity is different between two groups of fragrant and non-fragrant roses, in fragrant varieties, such as Rosa bourboniana, the flower life is much shorter, compared with most R. hybrida varieties, which are non-fragrant and widely used as cut flowers [55]. Synthesis of aroma components consumes a lot of energy and, in many species of the palm family, much energy is allocated to scent production in order to attract insects pollinators [56]. In future studies, controlling the synthesis of aroma components might be a new way to reduce the energy consumption and prolong the flower longevity.

Conclusions
Overall, through this study, we obtained a high-quality version of transcripts and made new insights into flower senescence in lily (L. oriental 'Siberia'), which provided a good foundation for the further study of gene cloning and related molecular mechanisms in lily (L. spp.). Additionally, we found that the key regulator LoNAC29 promotes lily flower senescence by inducing the expression of LoSAG39.