Six samples were selected and used for metabolic analysis at two stages (three replicates), including 5DPA (days post flowering) and 15DPA. Generally, Cotton fibers initiate near the day of flower opening, and 5DPA is the fiber elongation stage while 15DPA is transition stage from primary cell wall to secondary cell wall thickening, suggesting that they are two important stages of fiber development. Finally, 321 metabolites were obtained (Fig. 1a), including 204 Pos metabolites and 161 Neg metabolites, among which 84 metabolites with protein were identified (Supplementary Table 1). Differentially expressed analysis was conducted based on all metabolites detected, and the results can be found in (Figs. 1b and 1c).

Partial least squares discrimination analysis (PLS-DA) was an effective method to compare significant differential metabolites, and OPLS-DA was an analysis tool that could be used to modify PLS-DA by filtering the noise unrelated to the classification information to improve the analytical ability and effectiveness of the model. The results suggest that this model was stable and reliable (Fig. 2, Supplementary Table 2).

This model could be used to evaluate the quantity of each metabolite and identify significantly differential metabolite molecules at biological level. Analysis of metabonomics was always performed based on the strict standard of OPLS-DA VIP > 1 and P value < 0.05, which were also the standard for selecting the significantly differential metabolites. In our study, 11627 and 8630 metabolites were obtained with pos-model and neg-model, respectively. We list several metabolites with significant differences (orange) based on the value of OPLS-DA VIP > 1 and P value < 0.05 in (Tables 1 and 2). Subsequent analyses mainly focused on these metabolites, including 13 metabolites with pos-model and 15 metabolites with neg-model.

In our transcriptome data, six different fiber samples, including three 5DPA and three 15DPA samples, were obtained using Illumina sequencing technology (San Diego, CA, USA), and the statistical results of the transcriptomes are shown in Table 3. The ratio of clean reads and the total mapped rate were 90.0% and 93.0%, respectively. A total of 36 G of data were obtained after quality control, and the Q30 base ratio was above 92.0%. In addition, the uniquely mapped rate was higher than 94.0%. All these indicators demonstrated that the transcriptomic data were highly accurate for subsequent analyses.

DESeq was used to analyse the differentially expressed genes (DEGs) based on the criteria of |log2FoldChange| > 1. Finally, 7175 DEGs were obtained, including 3351 that were upregulated and 3824 that were downregulated. The transcriptional results are shown in Fig. 3. The correlation coefficients among all samples were between 0.8–1, indicating that the gene expression level among samples reflects that the samples we selected are reasonable (Fig. 3a). The majority of genomic DEGs were located on the A05 and D05 chromosomes, suggesting that these two chromosomes are closely related to fiber development (Fig. 3b). These DEGs were mainly enriched in ATP binding, pyrophosphatase activity, nucleoside-triphosphatase activity and hydrolase activity (Fig. 3c). In addition, we found that four differentially expressed CesA genes were identified from the transcriptome data (Supplementary Table 3). One target gene belong to CESA3 member named Gh_A08G144300 (GO: 0005524 and KEGG: K08900) with high expression were obtained in the process of cotton fiber length development (Fig. 4b).

Based on the results of significantly differential metabolites and DEGs, conjoint analysis was conducted. A total of 84 metabolites with corresponding genes and enzymes were obtained, including 36 metabolites with pos-model and 48 metabolites with neg-model (Supplementary Tables 4 and 5). In addition, fold changes in expression were also investigated, and the results are shown in (Supplementary Tables 6 and 7). To further investigate the function of the target gene Gh_A08G144300 in more detail, the expression levels in different tissues and fiber development stages were detected at 5, 10, 15, 20 and 25DPA using a fluorescence quantification method. The results indicated that the expression level of this gene in the fiber was significantly higher than that in other tissues (Fig. 4). In addition, the expression levels of this gene at different fiber development stages were analysed. However, the Gh_A08G144300 gene reached a peak at 10 and 15DPA in fiber development and then decreased significantly. This gene was highly expressed throughout the entire fiber period. In addition, the metabolite contents and correlation network of the two genes were also analysed (Fig. 4c), indicating their important regulatory roles in fiber development.

To study the possible biological functions of CesA genes during the process of plant growth, we selected Gh_A08G144300 to construct the CesA plant interference vector (Wu et al., 2011) and quantified the results of the cotton phenotype (Fig. 5). Paraffin section method of the fiber cell was used to study the cause of reduced fibers. Significant differences in cell volume and number and fiber length were observed between silencing plants, in which the target gene Gh_A08G144300 was silenced, and normal plants. This method was used to study the cause of reduced fibers of the TRV-Ch _CesA knockout, and the film was observed under a 200X optical microscope (Figs. 5a–5d). Fiber growth was inhibited significantly in Gh_A08G144300-silenced plants compared with that of the control plants (CK) (Figs. 5e and 5f). The number and size of cotton bolls was reduced with the average length decreasing by 20% from 35 mm (wild type, wt) to 28 mm (Fig. 5h). The average length decreased by 13% from 25 mm to 18 mm (Fig. 5i). The variations in cotton fibers may be caused by changes in the cell volume and number. In addition, the relative gene expression of Gh_A08G144300 was also investigated, and the results indicated that the target gene Gh_A08G144300 was significantly downregulated, verifying the vital function of this gene. In addition, the cellulose content of silencing plants and normal plants was measured and the results showed the cellulose content in silencing plants (5.02 mg/g) was significantly lower than the normal (9.15 mg/g) according the method described by Updegraff (1969). All results suggested that this gene plays an important role in fiber growth.

We used Gossypium hirsutum (‘TM-1’) as the material, a standard system for genetics provided by the Institute of Cotton Research of Chinese Academy of Agricultural Sciences (Anyang, China). New roots, stems, young leaves, flowers (during the flowering period) and different stages of fibers during development were frozen in liquid nitrogen and stored at –80°C until RNA was extracted for tissue expression analysis. The flowering day of cotton was labeled as 0 day and the next was labeled as 1st day, and so on. Samples of cotton fibers were obtained at the same position of cotton plant depending on the flowering time. The cotton tissues (80 mg leaves or flower tissue) were quickly frozen in liquid nitrogen immediately after collection and ground into fine powder with a mortar and pestle. Then, 1000 μL methanol/acetonitrile/H₂O (2:2:1, v/v/v) was prepared and added to the homogenized solution for metabolite extraction before the mixture was centrifuged for 15 min (14000 g, 4°C). The supernatant was dried in a vacuum centrifuge. For LC–MS analysis, the samples were redissolved in 100 μL acetonitrile/water (1:1, v/v) solvent. Three replications were performed for the transcriptome and metabolome analysis.

Analyses were performed using an UHPLC (1290 Infinity LC, Agilent Technologies) coupled to a quadrupole time-of-flight (AB Sciex TripleTOF 6600) at Shanghai Applied Protein Technology Co., Ltd. For HILIC separation, samples were analysed using a 2.1 mm × 100 mm ACQUIY UPLC BEH 1.7 μm column (Waters, Ireland). In both ESI positive and negative modes, the mobile phase contained A = 25 mM ammonium acetate and 25 mM ammonium hydroxide in water and B = acetonitrile. The gradient was 85% B for 1 min and was linearly reduced to 65% in 11 min, reduced to 40% in 0.1 min and kept for 4 min, and then increased to 85% in 0.1 min, with a 5 min re-equilibration period employed. For RPLC separation, a 2.1 mm × 100 mm ACQUIY UPLC HSS T3 1.8 μm column (Waters, Ireland) was used. In ESI positive mode, the mobile phase contained A = water with 0.1% formic acid and B = acetonitrile with 0.1% formic acid; in ESI negative mode, the mobile phase contained A = 0.5 mM ammonium fluoride in water and B = acetonitrile. The gradient was 1% B for 1.5 min and was linearly increased to 99% in 11.5 min and kept for 3.5 min. It was then reduced to 1% in 0.1 min, and a 3.4 min re-equilibration period was employed. The gradients were at a flow rate of 0.3 mL/min, and the column temperatures were kept constant at 25°C. A 2 μL aliquot of each sample was injected.

The ESI source conditions were set as follows: Ion Source Gas1 (Gas1) as 60, Ion Source Gas2 (Gas2) as 60, curtain gas (CUR) as 30, source temperature: 600°C, and IonSpray Voltage Floating (ISVF) ± 5500 V. In MS only acquisition, the instrument was set to acquire over the m/z range 60–1000 Da, and the accumulation time for TOF MS scan was set at 0.20 s/spectra. In auto MS/MS acquisition, the instrument was set to acquire over the m/z range of 25–1000 Da, and the accumulation time for the product ion scan was set at 0.05 s/spectra. The product ion scan was acquired using information-dependent acquisition (IDA) with high sensitivity mode selected. The parameters were set as follows: the collision energy (CE) was fixed at 35 V with ± 15 eV; declustering potential (DP), 60 V (+) and −60 V (−); excluding isotopes within 4 Da, candidate ions to monitor per cycle: 10.

The raw MS data (wiff.scan files) were converted into MzXML files by using roteoWizard MSConvert before they were imported into freely available XCMS software (Tautenhahn et al., 2012). For metabolite peak picking, the following parameters were used: centWave m/z = 25 ppm, peak width = c (10, 60) and prefilter = c (10, 100). For peak grouping, bw = 5, mzwid = 0.025 and minfrac = 0.5 were used. CAMERA (Collection of Algorithms of MEtabolite pRofile Annotation) was used for annotation of isotopes and adducts. In the extracted ion features, only the variables having more than 50% of the nonzero measurement values in at least one group were kept. Compound identification of metabolites was performed by comparing the accuracy m/z value (<25 ppm) and MS/MS spectra with an in-house database established with available authentic standards.

After normalization to the total peak intensity, the processed data were analysed using the R package ropls and subjected to multivariate data analysis, including Pareto-scaled principal component analysis (PCA) and orthogonal partial least-squares discriminant analysis (OPLS-DA). Sevenfold cross-validation and response permutation testing were used to evaluate the robustness of the model. The variable importance in the projection (VIP) value of each variable in the OPLS-DA model was calculated to indicate its contribution to the classification. Metabolites with a VIP value > 1 were further applied to Student’s t-test at the univariate level to measure the significance of each metabolite, and P values less than 0.05 were considered statistically significant.

Sequencing data were retained by removing the reads with adaptations and low quality, which would cause great interference to the subsequent information analysis, and the filter criteria were as follows: a) remove the sequences with adapters at the 3’ end using the CutAdapt tool; b) remove the reads with an average mass score below Q20. Upgraded TopHat2 HISAT2 (http://ccb.jhu.edu/software/hisat2/index.shtml) software was used to blast the filtered reads onto the reference genome. HISAT2 uses an improved BWT algorithm (Siren et al., 2014) to achieve faster speeds and a lower resource footprint. For HisAT2 alignment, default parameters are used for nonstrand-specific libraries, which need to specify the library type (i.e., first use--RNA-Strandness RF, second use--RNA-Strandness FR). If the reference genome is selected properly and there is no pollution in the relevant experiments, the mapping ratio of the sequencing sequence is generally higher than 70%. The reasons for the low mapping ratio may be as follows: 1) The reference genome was poorly assembled, or the species tested had a distant relationship with the reference genome; 2) The special pretreatment of the sample or the variation of the sample itself was too large relative to the reference genome, resulting in a relatively low mapping rate. High-quality assembled TM-1 genome was used as the reference genome in the study (Zhang et al., 2015). We used FPKM value for the Nosrmalization of the expression quantity. We generally consider that genes with the FPKM value > 1 are expressed. This threshold is generally used and is a good indicator of gene expression levels.

Based on transcriptome sequencing and metabolomics, we first obtained quantitative results and extracted the differentially expressed metabolites and transcripts. We then downloaded the corresponding transcripts of corresponding enzymes from the KEGG database (https://www.kegg.jp/dbget-bin/www_bfind?compound). The metabolites and related transcripts were then mapped onto corresponding metabolic pathways.

To determine the relationship between different metabolites and related enzymes in the KEGG database (P < 0.05), we first obtained relevant information on metabolites and transcripts. For metabolites, we obtained information from the small molecule database of KEGG (https://www.kegg.jp/dbget-bin/www_bfind?compound). For each transcript, additional annotation information can be obtained by searching their homologous genes in the KEGG database (https://www.kegg.jp/kegg/ko.html).

The cotton plant was split into several parts, including the root, shoot, leaves, flowers and fibers. Fibers were also collected at different stages (5, 10, 15, 20 and 25DPA) after flowering. Each fiber sample was taken from bolls in the same position on the plant. The RNA was extracted separately. A quantitative real-time PCR experiment was conducted using TIANGEN RealUniversal Colour PreMix (SYBR Green) (QKD-201, Tiangen Biotech, Beijing, China) following the manufacturer’s instructions. GhHistone3 was used as the reference gene. A total of 50 μL reaction solution was used, which included 10 ng of cNDA, 0.4 μL of forward primer, 0.4 μL of reverse primer, 10 μL of qPCR supremix and ddH₂O. Primer sequences of Gh_A08G144300 (forward primer: GTGCATTTCCTGTCTGCCGC; reverse primer: ATCAGCATCACCATCCTCTTCTC) were designed using an online program (NCBI). The relative expression of target Gh_A08G144300 was obtained with the 2−ΔΔCT method (Eid et al., 2009).

Three-day-old cotton fibers were fixed in a fixing solution that contained 4% FAA (formaldehyde-glacial acetic acid-absolute ethanol) for 24 h, dried under vacuum and then incubated overnight at 4°C. The samples were dehydrated using a series of gradient concentrations of ethanol (50%, 70%, 85%, 95% and 100%); samples were soaked in each gradient for 30 min. The soaked tissues were embedded in liquid paraffin and then cooled at –20°C. The samples were cut into 4 μm-thick sections with a paraffin section base. The sections were suspended in a 40°C water bath to flatten them before their placement on a glass slide. They were completely dried overnight at 37°C. The sections were then stained with safranin and Fast Green and photographed with a digital camera under a microscope.

Full-length Gh_A08G144300 was amplified from G. hirsutum (‘TM-1’) cDNA. The EcoRI and KpnI sites of pTRV:RNA2 were used as cloning sites, and the target sequences were inserted. The recombinant vectors were transformed into Agrobacterium GV3101 competent cells following the manufacturer’s instructions. The Agrobacterium transformant was cultivated in liquid Luria-Bertani (LB) medium containing 25 μg/mL rifampicin to an OD600 from 1.8 to 2.2. The OD600 of the medium was adjusted to 1.5 using buffer that contained 0.5 mol/L MES, 200 mmol/L acetosyringone and 1.0 mol/L MgCl₂ for transfection. Liquid medium containing pTRV:RNA2-Gh_A08G144300 and pTRV:RNA2 was mixed with pTRV at an equimolar ratio. They were then injected into cotton cotyledons at the three-leaf stage. Each group had 10 replications. The cotton plants were exposed to negative pressure and subsequently grown in the dark for 48 h. The plants were then moved to a greenhouse and grown under a 16-hour photoperiod for 30 days. Finally, they were grown under an 8-hour photoperiod to trigger the differentiation of flower buds.