Carpel-specific down-regulation of GhCKXs in cotton significantly enhances seed and fiber yield

Carpel-specific enhancement of cytokinin significantly increases cotton ovule number and cotton yield without detrimental phenotypes.


Introduction
Cotton is a major source of natural fibers for the global textile industry. Cotton seeds contain about 23% protein and 21% oil, and therefore are also an important source of foodstuff, feed, and edible oil (Zhang et al., 2002;Sunilkumar et al., 2006;Chen et al., 2007;Ma et al., 2016). A cotton boll contains approximately 30 seeds, and approximately 25% of the ovular epidermal cells of each ovule are able to differentiate into the commercially important lint fibers (Kim and Triplett, 2001;Mansoor and Paterson, 2012). Hence, the seed size and seed number of cotton determine the yield of fibers and seeds. However, seed size is negatively correlated with seed number due to the limitation of space and nutrition, and enhancing fiber abundance usually results in smaller seeds (J. Zhang et al., 2005;M. Zhang et al., 2011).
The signal transduction pathway of cytokinins has been well investigated in Arabidopsis. The signaling pathway is initiated by binding of cytokinin to Arabidopsis histidine kinase receptors (AHKs), following phosphoryl group transport via Arabidopsis histidine proteins (AHPs) to Arabidopsis response regulators (ARRs). The B-type ARRs that contain conserved GARP DNA binding and activation domains can bind to a short 5ʹ-(A/G) GAT(T/C)-3ʹ core DNA sequence to activate the transcription of downstream genes (Sakai et al., 2000(Sakai et al., , 2001Mason et al., 2004Mason et al., , 2005Argyros et al., 2008;Werner and Schmülling, 2009;Argueso et al., 2010;Zubo and Schaller, 2020). Among ARR-activated genes, members of the AG subfamily, such as STK, AG1, and SHP, have been known to be involved in ovule initiation (Bartrina et al., 2011;Zu et al., 2022). However, details about cytokinin signaling pathway in carpel development need to be investigated.
In this study, we used a chimeric carpel-and stamen-specific promoter (Busch et al., 1999;Deyholos and Sieburth, 2000), proAGIP, to down-regulate a functional cotton CKX gene (proAGIP::GhCKX3b-RNAi) in cotton carpels. We showed that the carpel-specific down-regulation of GhCKXs could enhance CK levels at the position where the ovule initiates in the carpel, without causing abnormal growth phenotypes. Cytokinin-activated GhARR1, GhARR2a, and GhARR11 in turn promoted the expression of GhSTK2, GhAG1, and GhSHP by binding their promoter sequences, thus boosting carpel development and ovule formation. As a result, seed and fiber yield of proAGIP::GhCKX3b-RNAi cotton was significantly increased compared with that of the non-transgenic control. Our results indicate the biotechnological potential of manipulation of CKs in cotton carpels for the concurrent improvement of seed and fiber yield.

RNA extraction and real-time quantitative PCR
Total RNA was extracted using the EASY spin plant RNA extraction kit (Aidlab Biotech, China). Approximately 1 μg RNA was transcribed into first-strand cDNA using the NovoScript Plus All-in-one First Strand cDNA Synthesis SuperMix (gDNA Purge, Novoprotein, China). The real-time quantitative PCR (RT-qPCR) assays were performed on a CFX Connect Real-Time System (Bio-Rad Laboratories) with 2×NovoStart SYBR qPCR SuperMix plus (Novoprotein, China). GhHis3 (Zhang et al., 2011;Wan et al., 2016;Wu et al., 2018) and GhUbiquitin (Walford et al., 2011;Wu et al., 2018) served as internal references. Gene specific primers used for RT-qPCR are listed in Supplementary  Table S2. The expression data were calculated with the ΔΔC t method. For the RT-qPCR analysis, three individual biological replicates with two technical replicates for each gene were used. Mean values and standard errors were calculated using the data from the three replicates. The compliance of the RT-qPCR experiments with the Minimum Information for Publication of Quantitative Real-Time PCR Experiments (MIQE) is shown in a MIQE checklist (Supplementary Table S3).

Sequence retrieval, phylogenetic analysis, and sequence alignment
The amino acid sequences of AtCKXs and B-type AtARRs were obtained from the Arabidopsis genome databases (https://www.Arabidopsis.org/, accessed on 30 December 2019). The GhCKX and B-type GhARR homologs were identified using the BLASTP tool with default parameters in the CottonFGD database (https://cottonfgd.org/, accessed on 30 December 2019) (Zhu et al., 2017) using AtCKX and B-type AtARR sequences, respectively. The cutoff values were 0 for the sequence retrieval of GhCKXs and 1.0 −140 for the sequence retrieval of B-type GhARRs. Then, the selected GhCKX sequences were used for further identification of GrCKXs and GaCKXs by searching the databases https://phytozome. jgi.doe.gov/pz/portal.html and https://cottonfgd.org/, respectively. The phylogenetic trees of deduced CKX and B-type ARR amino acids were constructed by the neighbor-joining algorithm with default parameters, with 1000 bootstrap replicates as implemented in MEGA5.0 software. The GhCKX sequences were aligned using Megalign DNAstar software (Burland, 2000) and analysed with the Gendoc software (Nicholas, 1997).

Histochemical staining and quantification of β-glucuronidase activity
Histochemical staining of β-glucuronidase (GUS) was performed as previously described by Jefferson et al. (1987). Briefly, detached or handsectioned tissues were immediately immersed in the staining solution (Zeng et al., 2019) and then placed in the dark at 37 °C for 12 h. The stained samples were bleached and fixed in 75% ethanol before photographing. Images were captured on a stereo-microscope imaging system (SteREO Discovery V20, Zeiss, Germany). Fluorometric assays of GUS activity in vegetative and reproductive organs of proTCS::GUS transgenic cotton were performed as described by Hou et al. (2008). Each sample was ground in liquid nitrogen. Protein estimation was performed using the method of Bradford (1976). GUS activity was calculated as pmol 4-methylumbelliferone (4-MU) per minute per microgram protein and each test was represented by three biological replicates.

In situ hybridization
The linearized DNA template of the gene-specific GhCKX3b probe was amplified directly from a vector carrying the coding sequence of GhCKX3b. In situ hybridization of GhCKX3b mRNA was performed following the method described in Zhang et al. (2017). The sections incubated with the sense RNA probe served as the negative control. Images were captured on a microscope (CKX41, Olympus, Japan). Gene specific primers are listed in Supplementary Table S1.

Quantification of endogenous cytokinin
Buds of cotton were harvested at the pinhead square stage (approximately −21 d post-anthesis; DPA), and then carpels (100 mg fresh weight) were separated and ground in liquid nitrogen. Extraction and detection of endogenous cytokinin was performed following the method described in Yoshimoto et al. (2009) and Zeng et al. (2019). Analytical parameters of LC-MS/MS are listed in Supplementary Table S4.

Immunohistochemical localization of cytokinins
Immunolocalization of cytokinins was performed following the method described by Zhang et al. (2017) with some modifications. Briefly, sections (10 μm) of cotton bud at the pinhead square stage were incubated with antibody (against trans-zeatin riboside (tZR) and trans-zeatin (tZ), Agrisera, Sweden) and then the signal was detected by using DyLight 550-labeled secondary antibody (Abcam, UK), and visualized on a laserscanning confocal microscope (SP8, Leica, Germany). Sections incubated without the primary antibody served as the negative control.

Transient expression
Four-week-old leaves of Nicotiana benthamiana were used for transient expression. Agrobacterium tumefaciens strain GV3101 containing a plant expression vector was cultured overnight at 28 °C to OD 600 of 1.0. The pelleted cells were resuspended and diluted with infiltration buffer (Chen et al., 2021) to OD 600 of 0.01-0.05. The infiltrated leaves were used for analysis 3 d later.

Microscopic observations
GUS-stained samples were observed using a stereo-microscope imaging system (SteREO Discovery V20, Zeiss, Germany). Fiber initiation was observed on an S-3400N scanning electron microscope (Hitachi, Japan) as described previously by Zeng et al. (2019). The protein subcellular localization and cytokinin immunolocalization was observed on a laser-scanning confocal microscope (SP8, Leica, Germany). The fluorescence signal was detected by an HyD detector under a ×40 oil immersion objective lens. The imaging condition was set up manually: DyLight 550 (excitation: 552 nm, emission: 560-600 nm) and YFP (excitation: 514 nm, emission: 520-560 nm). The intensity of fluorescence signal was quantified using software Leica Application Suite X.

Southern blot
Southern blot was conducted as previously described by Zhang et al. (2011). Briefly, gDNA of proAGIP::GhCKX3b-RNAi transformants and the non-transgenic control cotton was digested overnight (10 µg) with the restriction enzyme EcoRI. An NPTII fragment amplified from the vector p5 served as the probe. DIG High Prime DNA Labeling and Detection Starter Kit II (Roche, Germany) were employed to prepare digoxigenin-labeled probes and to detect hybridization signals. Gene specific primers are listed in Supplementary Table S1.
Yeast one-hybrid assay Y1H assays were performed as previously described by Li et al. (2021). In brief, the pAbAi bait vector was transformed into the yeast strain Y1H Gold by the LiAc-mediated method. The pGADT7 prey vector was transformed into Y1H Gold cells harboring the Bait-pAbAi. Transformants were screened in the synthetic dextrose medium (SDM) containing 100 or 200 ng ml −1 aureobasidin A (AbA) and lacking Ura and Leu (SDM/+AbA/−Leu). The yeast growth was captured after 5 d inoculation at 30 °C.

Dual-luciferase reporter assay
A dual-luciferase reporter assay was performed as previously described by Hellens et al. (2005) and Yan et al. (2018). Briefly, the promoter fragments were amplified and ligated into the pGreenII 0800-LUC vector to produce firefly luciferase (LUC) reporters, and the coding regions of GhARRs were amplified and ligated into the pLGN vector to produce effectors. The reporters and effectors were co-infiltrated into 4-week-old leaves of N. benthamiana as described in 'Transient expression'. The effector expressing pro35S::YFP served as internal control. After 3 d of growth at 25 °C, 10 μM tZ was infiltrated into the leaves 5 h before sampling. The leaves were infiltrated with the same volume of dimethyl sulfoxide (DMSO) as the negative control (0 μM tZ). The Dual-Glo Luciferase Assay System (Promega, USA) was employed to measure the LUC activity, which was calculated based on the ratio of LUC/Renilla luciferase (REN). Gene specific primers are listed in Supplementary Table S1.

Statistical analysis
Statistical analysis was performed with Student's t-test or oneway ANOVA followed by Tukey multiple comparisons test (P<0.05). Each experiment comprised at least three replicates. The intensity of fluorescence signal was calculated using Leica Application Suite X software. Standard errors and standard deviations were calculated using Microsoft Excel (2016) and IBM SPSS Statistics (version 19).

GhCKX3b is preferentially expressed in carpels
By detecting GUS activity, we estimated cytokinin activities in vegetative and reproductive organs of cotton expressing the cytokinin signaling reporter proTCS::GUS (Müller and Sheen, 2008;Zeng et al., 2019). The maximum GUS activity was discernible in the flower bud at the pinhead square stage (approximately −21 DPA) when the ovule formation begins (Fig. 1A, B), suggesting that a high activity of cytokinins is required for ovule morphogenesis.  To identify cotton CKXs, which are the key negative regulators of cytokinins in plants (Schmülling et al., 2003), we screened the genomic database of Gossypium hirsutum (https:// cottonfgd.org/). Twenty-seven GhCKX homologous genes to Arabidopsis AtCKXs (https://www.arabidopsis.org/) were identified ( Supplementary Fig. S1). Among them, GhCKX3b, GhCKX3c, GhCKX5a, and GhCKX6b were expressed preferentially in the carpel and stamen ( Fig. 1C; Supplementary  Fig. S2). Our previous study demonstrated that GhCKX3b is a functional cytokinin oxidase (previously named GhCKX; Zeng et al., 2012;Zhao et al., 2015). In situ mRNA hybridization confirmed a strong signal of GhCKX3b transcript in the carpel at the place where ovules formed (Fig. 1D).
To avoid the side effects on plant growth from the global overproduction of cytokinins, while increasing the seed number through cytokinin manipulation, we decided to increase the cytokinin content spatiotemporally in carpels. To this end, we used proAGIP, a chimeric carpel-and stamenspecific promoter (Busch et al., 1999;Deyholos and Sieburth, 2000), to direct the transcription of a GhCKX3b-RNAi sequence ( Supplementary Fig. S3). The GUS pattern in proAGIP::GUS transgenic tobacco (Nicotiana tabacum) confirmed the carpel-specificity of proAGIP promoter (Supplementary Fig. S4; Deyholos and Sieburth, 2000;Wang et al., 2008).
No discernable phenotypic alteration in plant growth and development was observed in transgenic cotton holding proAGIP::GhCKX3b-RNAi (ACR) (Fig. 2C-E). RT-qPCR showed that the transcriptional level of GhCKX3b in the carpel of transgenic lines was obviously reduced. Two transgenic lines, ACR1 and ACR5, in which GhCKX3b was largely down-regulated, were selected for further study (Figs  2A, B, 3).
Then, we performed LC-MS/MS to detect the content of six primarily active cytokinins, namely tZ, tZR, N6-isopentenyladenine, N6-isopentenyladenosine, dihydrozeatin, and dihydrozeatin riboside, in carpels of cotton. Besides a nontransgenic negative control, a transgenic pro35S::GhCKX-RNAi line, 35SCR6, which showed a moderate increase of cytokinins with normal growth, and had the best yield performance among the pro35S::GhCKX-RNAi cottons (Zhao et al., 2015), was used as a positive control. In carpels of ACR1 and ACR5 transgenic cotton, the total cytokinins were 29.53 ± 1.46 ng g −1 and 27.46 ± 1.63 ng g −1 , respectively, significantly higher than in the 35SCR6 line (23.46 ± 1.81 ng g −1 ) and the non-transgenic control (15.05 ± 1.73 ng g −1 ) (Table 1), indicating a carpel-specific increase of cytokinins by proAGIP::GhCKX3b-RNAi. Immunolocalization assays supported that tZ and tZR signals, which made up the majority (85.0%) of the active CKs (Table 1), mainly appeared at the position where the ovule initiated, and the signals from ACR1 and ACR5 were visibly stronger than those in 35SCR6 and the control (Fig. 4).

Increase of cytokinin promotes the expression of ovule initiation-related genes
It has been reported that the expression of genes of the AG subfamily, such as those for Seedstick (STK), Shatterproof (SHP), and Agamous (AG), was able to promote the formation of carpels and ovules (Favaro et al., 2003;Pinyopich et al., 2003;Xu et al., 2004;Guo et al., 2007;Liu et al., 2009Liu et al., , 2010de Moura et al., 2017;Nardeli et al., 2018). In line with previous reports, AG subfamily genes, including GhSTK1, GhSTK2, GhSHP, and GhAG1 were up-regulated in the carpel of the ACR1 and ACR5 lines ( Fig.  5A; Supplementary Table S5). To confirm these results, we treated 0 DPA wild-type ovules with 50 μΜ tZ, an active cytokinin, for 12 h. The exposure resulted in a noticeable increase in the expression of AG subfamily genes (Fig. 5B), confirming the promotive effect of the cytokinin on the expression of these genes. Cytokinins regulate the expression of downstream signal transduction genes by activating B-type ARRs (Hwang and Sheen, 2001;Meng et al., 2017). To identify B-type ARR genes that are involved in cotton carpel development, we conducted a comparative phylogenetic analysis using data from Arabidopsis (https://www.arabidopsis.org/) and Gossypium hirsutum (https://cottonfgd.org/) (D'Agostino et al., 2000;Sakai et al., 2000;Mason et al., 2004Mason et al., , 2005Zubo and Schaller, 2020). Five carpel expressive GhARRs, i.e. GhARR1, GhARR2a, GhARR2b, GhARR11, and GhARR12, which predominantly localize to the nuclei of N. benthamiana pavement cells, were identified (Supplementary Fig. S5). Dual-luciferase reporter and Y1H assays indicated that GhARR1 could bind directly to the promoter sequences of the AG subfamily genes GhSTK2 and GhAG1, GhARR2a to the promoter of GhAG1, and GhARR11 to the promoter of GhSHP (Fig. 5E; Supplementary Fig. S6). This binding could activate the transcription of these AG subfamily genes (Fig. 5C-D), thereby promoting the expression of AG downstream genes.

Carpel-specific down-regulation of GhCKXs significantly increases seed and fiber yield
By counting, we found that the number of ovules per locule (0 DPA) of ACR1 and ACR5 was significantly higher than that of the non-transgenic control as well as the constitutively down-regulated 35SCR6 line ( Fig. 6; Supplementary  Fig. S7). The average number of ovules per locule of the ACR1 and ACR5 lines was 9.6 ± 0.5 and 9.3 ± 0.4, respectively, increasing to 21.5% and 17.7%, respectively, over that of the non-transgenic control (7.9 ± 0.6), and 11.6% and 8.1%, respectively, over that of the 35SCR6 line (8.6 ± 0.4). For ACR1 locules, the majority (59.2%) contained 10 or 11 ovules in each locule. In contrast, for the non-transgenic cotton, the majority (77.5%) had seven or eight ovules, and only a small part (1.1%) had 10 ovules. For the 35SCR6 cotton, 4.3% of locules had 10 ovules, and none had 11 ovules. The increase of ovules per locule of ACR cotton was also constant during the flower stage of cotton development ( Supplementary Fig. S8).
More ovules per locule usually means more seeds per boll. To assess the agronomic performance of the proAGIP::GhCKX3b-RNAi transgenic cotton, we conducted field trials at the experimental farm located in Chongqing, China (29°49ʹ11″N, 106°24ʹ43″E) in 2018 and 2019. The consecutive two-year field experiments showed a significant increase of seeds per boll in ACR lines ( Table 2). The seed number per boll of ACR1 and ACR5 was 31.1 ± 0.7 and 29.8 ± 0.6, respectively, significantly higher than that of 35SCR6 (26.1 ± 0.9) and the control (24.8 ± 0.2). Consequently, the seed yield of ACR1 and ACR5 increased 22.4% and 26.5%, respectively, as compared with that of the control. The increased seeds resulted in a significant increase of lint yield. The lint yield of ACR1 and ACR5 was increased by 32.3% and 25.8%, respectively, compared with the control. The seed yield of the ACR1 line was lower (but not statistically significant) than that of 35SCR6 cotton, due to a slight decrease in seed size (indicated by seed index, the weight in grams of 100 seeds). Nevertheless, the lint yield of the ACR1 line was still significantly higher than that of the 35SCR6 line (Table 2), with the advantage of more seeds per boll. The fiber quality, fiber length, and fiber strength of ACR1 were lower than those of the control; but the values of ACR5 were not significantly different from those of the control (Supplementary Table S6; Supplementary Fig. S9).

Discussion
The promotive effect of cytokinins on seed yield was first observed by exogenous application of cytokinins in various crops (Dyer et al., 1987;Atkins and Pigeaire, 1993;Sawan et al., 2000;Zuñiga-Mayo et al., 2018), followed by genetic expression of cytokinin biosynthetic genes, such as the gene for isopentenyltransferase, to endogenously increase the cytokinin levels in transgenic plants (Ma et al., 2008;Atkins et al., 2011;Peleg et al., 2011). CKX, which inactivates cytokinin irreversibly in plant cells, has been identified as a key negative regulator of cytokinin content in monocots and dicots (Jones and Schreiber, 1997;Schmülling et al., 2003;Kowalska et al., 2010;Zeng et al., 2012;Zhao et al., 2015;Ogonowska et al., 2019). Accumulating evidence has demonstrated that CKX is a key regulator for seed number, and down-regulation of CKXs is an effective strategy for yield improvement of seed crops (Zalewski et al., 2010;Bartrina et al., 2011;Li et al., 2013;Yeh et al., 2015;Schwarz et al., 2020). In our previous study, we generated cytokinin-enhanced transgenic cotton in which a CKX gene was constitutively down-regulated. We found that some transgenic cottons displayed cytokinin overproductionrelated aberrations, including dwarf growth, shortened root, and sterility. Only a moderate increase of cytokinin could show a positive effect on cotton yield (Zhao et al., 2015). In this study, we revealed that GhCKX3b, GhCKX3c, GhCKX5a, and GhCKX6b were expressed preferentially in the carpel and stamen ( Fig. 1C; Supplementary Fig. S2), suggesting a possible function of these genes in carpel development. We thus designed a GhCKX3b-RNAi sequence, which could interfere with the expression of GhCKX3b, GhCKX3c,GhCKX5a,and GhCKX6b (Figs 3,5A;Supplementary Fig. S3;Supplementary Table S5). The cytokinin content in the carpels of transgenic proAGIP::GhCKX3b-RNAi cotton lines was significantly higher not only over the wild-type control but also over the positive control pro35S::GhCKX3b-RNAi carpels (Table 1; Fig. 4). Compared with pro35S::GhCKX3b-RNAi cotton, of which some exhibited abnormal phenotypes, all proAGIP::GhCKX3b-RNAi cotton lines grew well without any levels of AG subfamily genes in tZ-treated ovules and the control. Total RNA was extracted from 0 DPA ovules treated with 50 μM tZ or the same volume of DMSO for the negative control, for 12 h. (C) Schematic representation of the constructs used in the dual-luciferase assay. (D) The promoters of AG subfamily genes involved in ovule initiation were activated under the expression of GhARRs and treatment with 10 μM tZ via a dual-luciferase assay. The effector expressing pro35S::YFP served as internal control. The leaves were infiltrated with 10 μM tZ or the same volume of DMSO for the negative control (0 μM tZ). The transactivation activity of AG subfamily gene promoters by GhARRs was calculated based on the ratio LUC/REN. Error bars indicate SD of three biological replicates. Asterisks represent significant difference (versus internal control, or 0 μM tZ) as determined by Student's t-test (*P<0.05; **P<0.01). (E) Yeast one-hybrid assay of GhARRs binding to the promoters of AG subfamily genes. Promoter fragments were inserted into the pAbAi vector to construct pAbAi bait vectors, and the full-length coding sequences of GhARRs were inserted into the pGADT7 vector to construct pGADT7 prey vectors. The pGADT7 prey vector was transformed into Y1H Gold cells harboring the Bait-pAbAi. The Bait-pAbAi and 53-pGADT7 (53-AD) vectors served as the positive control. Possible interactions were screened in synthetic dextrose medium (SDM) containing 100 or 200 ng ml −1 aureobasidin A (AbA) and lacking Ura and Leu (SDM/+AbA/−Leu).
cytokinin overproduction-related aberrations. More importantly, the fiber yield of the two transgenic proAGIP::GhCKX3b-RNAi cotton lines was significantly higher than that of both wild-type control and the pro35S::GhCKX3b-RNAi line (Table 2), indicating an advantage of the tissue-specific regulation strategy over the constitutive one. Cytokinins regulate growth and development of plants through the cytokinin-ARR signaling pathway (Hwang and Sheen, 2001;Meng et al., 2017). Among ARR-activated genes, AG subfamily genes STK2, AG1, and SHP are positive regulators of ovule identity and initiation (Favaro et al., 2003;Pinyopich et al., 2003;Xu et al., 2004;Guo et al., 2007;Liu et al., 2009Liu et al., , 2010de Moura et al., 2017;Nardeli et al., 2018). Our transcriptome and RT-qPCR data showed that the expression of AG subfamily genes was obviously increased in the ACR transgenic cotton carpels at the place where ovules formed ( Fig. 5A; Supplementary Table S5). Our Y1H and dual-luciferase assays further revealed that GhARR1, GhARR2a, and GhARR11 could directly bind to the upstream sequences of GhAG1 and GhSTK2, GhAG1, and GhSHP, respectively, and in turn activate the transcription of these AG subfamily genes ( Fig. 5C-E; Supplementary Fig. S6). The enhanced expression of these AG subfamily genes could in turn promote carpel development and ovule formation, thus increasing the yield of seeds. The mechanism for increasing cotton yield by the carpel-specific enhancement of cytokinins is summarized in Fig. 7.
Fiber length, strength, and fineness are the three major traits determining the quality and economic value of cotton (Han et al., 2013;Long et al., 2018). It has been reported that a high concentration of kinetin (>5 µM), a type of cytokinin, inhibits fiber elongation, whereas a low concentration (<0.5 µM) stimulates fiber elongation (Beasley and Ting, 1974;Yu et al., 2000b). In addition, fiber elongation was inhibited in transgenic cotton expressing cytokinin biosynthesis isopentenyltransferase gene, ipt, under the control of seed-specific promoter Ph/P (Yu et al., 2000a). Our previous study demonstrated that constitutive overexpression of GhCKX-RNAi had little negative effect on fiber quality, such as length, strength, and fineness (Zhao et al., 2015). In this current study, we found that the fiber length and fiber strength of the ACR1 transgenic line was significantly decreased, whereas  (Hwang and Sheen, 2001;Meng et al., 2017), which in turn promotes the transcription of AG subfamily genes (e.g. GhSTK2, GhAG1, and GhSHP) by binding the upstream region of the genes. The up-regulation of these AG subfamily genes boosts carpel differentiation and ovule formation (Favaro et al., 2003;Pinyopich et al., 2003;Xu et al., 2004;Guo et al., 2007;Liu et al., 2009;Liu et al., 2010;de Moura et al., 2017;Nardeli et al., 2018). This promotion of ovule formation results in production of more seeds in a boll, thus increasing the yield of both seeds and fibers of cotton. there was no significant alteration in ACR5. We noticed that the seed number per boll of ACR1 (31.1 ± 0.7) was higher than that of ACR5 (29.8 ± 0.6), while its seed size (indicated by the seed index) declined (Table 2). One explanation for the deterioration in quality of transgenic ACR1 fibers is that more seeds per boll means a lower nutritional share, which may reduce the quality of the fibers. The impact of manipulation of cytokinins on fiber quality awaits further investigation. Taken together, our data indicate that without any hindrance of plant development, carpel-specific up-regulation of endogenous cytokinins by down-regulation of CKX is a feasible and effective strategy for seed yield improvement, not only for cotton but perhaps also for other dicotyledons, such as canola and soybean.

Supplementary data
The following supplementary data are available at JXB online. Fig. S1. Phylogenetic analysis of CKX proteins. Fig. S2. Transcription levels of GhCKXs in different wildtype upland cotton tissues. Fig. S3. Sequence alignment of partial GhCKXs which were preferentially expressed in the carpel. Fig. S4. The activity pattern of proAGIP in Nicotiana tabacum. Fig. S5. Expression patterns and nuclear localization of GhARRs. Fig. S6. Diagram of ARR binding elements in the AG subfamily gene promoters. Fig. S7. Comparison of ovule number per 0 DPA locule or boll between T 0 proAGIP::GhCKX3b-RNAi transgenic cottons and wild type. Fig. S8. Comparison of ovule number per 0 DPA locule or boll between T 2 proAGIP::GhCKX3b-RNAi transgenic cotton and the non-transgenic segregated line at different growth temperatures. Fig. S9. Phenotypes of proAGIP::GhCKX3b-RNAi transgenic cotton bolls and seeds. Table S1. Primer and fragment information for plasmid construction. Table S2. Primers pairs used for RT-qPCR assay. Table S3. MIQE checklist. Table S4. Analysis parameters for CKs using LC-MS/MS. Table S5. Transcriptomic analysis of transcripts that were significantly changed between proAGIP::GhCKX3b-RNAi transgenic cotton and the non-transgenic segregated line. Table S6. Comparison of mature fiber qualities of proAGIP::GhCKX3b-RNAi transgenic cottons and the nontransgenic segregated line in field trial.

Author contributions
YP and JZ conceived the experiments and wrote the manuscript; JZ, XY, WB, YC, JZ, XD, RL, FW, HR, JZ, BD, HL, MZ, XL, LH, and YX performed the experiments and analysed the data; YP, JZ, YX, XY, and MZ contributed to the interpretation of the results. All authors reviewed and approved the manuscript.