D unigenes were examined using three technical replicates and three biological replicates for each stage. The qRT-PCR was conducted using an ABI 7500 Real-Time PCR System (Applied Biosystems), SYBR Green, and a PrimeScriptTM RT Reagent qPCR Kit (Takara, Dalian, China). Relative transcript abundances were calculated according to the comparative cycle threshold method, with GAPDH as an internal standard [68]. Pearson correlation analyses were completed using the SPSS Statistics 17.0 software package (SPSS Inc., Chicago, IL, USA). Correlation analyses were conducted using the log ratios (i.e., log2 fold-change between stages) of unigene expression levels determined by qRT-PCR and RNA-seq.Chlorophyll content measurementsTo obtain high-quality reads, raw data were filtered to remove adaptor sequences and reads with unknown or low-quality bases. De novo assembly was performed using the Trinity program (release 20130225 PRIMA-1 [64]). We first combined the reads with default parameters to form fragments longer than 200 bp, which were defined as contigs. The reads were then mapped to the contigs to obtain longer sequences using the paired-end reads as templates. Finally, cap3/PriceTI was used to connect the contigs and obtain sequences that could not be extended on either end. Such sequences were defined as unigenes. Unigenes PubMed ID:https://www.ncbi.nlm.nih.gov/pubmed/3081428 were annotated using the BLASTx function of selected protein databases, including the NCBI-nr [13] and TAIR databases [14], considering an E-value threshold of 10-5. Pathway analyses were conducted using KAAS [15], while GO classifications were completed using WEGO [65] based on GO annotation terms provided by the Blast2GO program [66]. Gene expression PubMed ID:https://www.ncbi.nlm.nih.gov/pubmed/16474207 levels were calculated based on FPKM values using Cufflinks (version 1.0.3) [67]. The significance of the gene expression level differences between two stages was assessed via the Student's t-test using FPKM values obtained from three cDNA libraries for each stage. To identify the DEGs between two stages, a threshold FDR of < 0.05 was used to judge the significance of gene expression differences. Hierarchical cluster analyses were conducted with the normalized FPKM values using complete linkage groupings with the aid of Cluster 3.Unigene expression level analysisThe chlorophyll contents were measured in five biological replicates according to the method of Arnon [69]. Chlorophylls were extracted from 100-mg leaf samples using 80 acetone. The extracts were spectrophotometrically analyzed at 645 and 663 nm. The Student's t-test was used to compare the chlorophyll contents in the YG and W stages with those in the G stage.Quantification of leaf metabolites by UPLC qQ-MSTotal RNA extracted using the RNeasy Plus Mini Kit was treated with TURBO DNase to remove genomic DNA. First-strand cDNA was synthesized from 1 g DNA-free RNA in a reverse-transcription reaction using random hexamer primers and the MultiScribe reverse transcriptase from a High Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Foster City, CA,For each stage, 100-mg leaf samples were submerged in 500 L methanol:water (3:1, v/v) with ultrasonic mixing for 5 min. The mixtures were centrifuged at 12,000 rpm for 10 min at 4 , and the supernatants were analyzed using a UPLC qQ-MS instrument. Five replicates were analyzed for each sample. A 2-L aliquot of the supernatant was injected into a Zorbax Eclipse Plus C18 column (50 ?2.1 mm, 1.8-m particle size) (Agilent Technologies, Palo Alto, CA, USA) that was maintained.