Dynamic Quantitative Trait Loci Mapping for Plant Height in Recombinant Inbred Line Population of Upland Cotton

Plant height (PH) is a key plant architecture trait for improving the biological productivity of cotton. Ideal PH of cotton is conducive to lodging resistance and mechanized harvesting. To detect quantitative trait loci (QTL) and candidate genes of PH in cotton, a genetic map was constructed with a recombinant inbred line (RIL) population of upland cotton. PH phenotype data under nine environments and three best linear unbiased predictions (BLUPs) were used for QTL analyses. Based on restriction-site-associated DNA sequence (RAD-seq), the genetic map contained 5,850 single-nucleotide polymorphism (SNP) markers, covering 2,747.12 cM with an average genetic distance of 0.47 cM. Thirty-seven unconditional QTL explaining 1.03–12.50% of phenotypic variance, including four major QTL and seven stable QTL, were identified. Twenty-eight conditional QTL explaining 3.27–28.87% of phenotypic variance, including 1 major QTL, were identified. Importantly, five QTL, including 4 stable QTL, were both unconditional and conditional QTL. Among the 60 PH QTL (including 39 newly identified), none of them were involved in the whole period of PH growth, indicating that QTL related to cotton PH development have dynamic expression characteristics. Based on the functional annotation of Arabidopsis homologous genes and transcriptome data of upland cotton TM-1, 14 candidate genes were predicted within 10 QTL. Our research provides valuable information for understanding the genetic mechanism of PH development, which also increases the economic production of cotton.

Plant height (PH) is a key plant architecture trait for improving the biological productivity of cotton. Ideal PH of cotton is conducive to lodging resistance and mechanized harvesting. To detect quantitative trait loci (QTL) and candidate genes of PH in cotton, a genetic map was constructed with a recombinant inbred line (RIL) population of upland cotton. PH phenotype data under nine environments and three best linear unbiased predictions (BLUPs) were used for QTL analyses. Based on restrictionsite-associated DNA sequence (RAD-seq), the genetic map contained 5,850 singlenucleotide polymorphism (SNP) markers, covering 2,747.12 cM with an average genetic distance of 0.47 cM. Thirty-seven unconditional QTL explaining 1.03-12.50% of phenotypic variance, including four major QTL and seven stable QTL, were identified. Twenty-eight conditional QTL explaining 3.27-28.87% of phenotypic variance, including 1 major QTL, were identified. Importantly, five QTL, including 4 stable QTL, were both unconditional and conditional QTL. Among the 60 PH QTL (including 39 newly identified), none of them were involved in the whole period of PH growth, indicating that QTL related to cotton PH development have dynamic expression characteristics. Based on the functional annotation of Arabidopsis homologous genes and transcriptome data of upland cotton TM-1, 14 candidate genes were predicted within 10 QTL. Our research provides valuable information for understanding the genetic mechanism of PH development, which also increases the economic production of cotton.

INTRODUCTION
As the most important natural textile fiber, cotton (Gossypium spp.) supplies about 35% of global fiber consumption, which is also used to make feed and oilseeds, and upland cotton (Gossypium hirsutum L.) accounts for more than 90% of global cotton production (Chen et al., 2007;Wang et al., 2015;Zhang et al., 2015;Ma et al., 2018b). Cotton production is a part of the cash crop industry (Li, 2021). With the change of the cotton industry trade situation and the promotion of mechanization, it is very important to cultivate high-yield and good-quality cotton varieties. Plant architecture breeding can effectively increase cotton yield and improve the fiber quality, thereby affecting the economic production of cotton (Li et al., 1998Song and Zhang, 2009;Shang et al., 2015b;Fu et al., 2019). Plant architecture refers to the three-dimensional structure of the aerial part of the plant, which is a comprehensive agronomic trait, including many factors such as PH, fruit branch number, angle between stem and fruit branch, morphological characters and distribution of boll and leaf, and so forth (Yang et al., 1996;Wang and Li, 2006;Wu et al., 2021). It is worth noting that the Green Revolution that emerged in the 1960s changed the plant architecture by transferring dwarf genes into crops, resulting in a significant increase in yields (Khush, 2001). For example, reducing PH enhanced resistance and significantly increased yield of wheat and rice (Peng et al., 1999;Sasaki et al., 2002). Besides, using the compact and medium dwarf corn varieties increased planting density and significantly improved the production (Peiffer et al., 2014).
Plant height (PH) is a main factor affecting the plant architecture (Jiao et al., 2010;Zhang et al., 2019). In addition, the appropriate PH of cotton (80-90 cm) is also beneficial for lodging resistance and mechanical harvesting (Chen, 2013;Pei et al., 2021). Therefore, a study on PH of cotton will be helpful for cotton plant architecture breeding. However, PH is a quantitative trait controlled by multiple genes and affected by the environment easily (Cui et al., 2011;Wurschum et al., 2015;Liu et al., 2018). With the rapid development of molecular marker technology, the molecular genetic mechanism of quantitative trait can be studied by the linkage analysis and association analysis. QTL-seq and QTL mapping techniques were used to identify the main PH QTL on the A10 chromosome in two F 2 populations of Brassica napus (Dong et al., 2021). In maize, a major QTL controlling PH was mapped by three different methods, including genome wide association study (GWAS), and was located in a 600 kb region (Pan et al., 2017). In wheat, seven stable PH QTL were detected on six chromosomes, among which QPht.cau-3D1 explaining 28.36-38.08% of phenotypic variance was stable QTL (Chen et al., 2020). In rice, a PH QTL qPH7-2 explaining 36.77-56.82% of phenotypic variance was detected in an RIL population under three environments (Lei et al., 2018). The molecular genetic mechanism of cotton PH is mainly studied by the linkage analysis. Fourteen PH QTL were identified on 11 chromosomes of an upland cotton F 2 population, with 9.64% of the contribution rate of qPH-D12-1 (Jia et al., 2021). A stable QTL qPH-Dt1-1, as well as a gene GhPIN3 regulating cotton PH, was identified in an interspecific G. hirsutum × G. barbadense BIL population using the SLAF-seq technology . Twenty-seven PH QTL were mapped on 18 chromosomes using an RIL population of upland cotton, with 3.81-8.54% of phenotypic variance . In addition, naturally dwarfing cotton mutants are very important for identifying PH genes of cotton. A gene EXTR-DWARF related to dwarfing was screened from a natural dwarf mutant of cotton (Ji, 2018).
Cotton PH QTL previously reported were mainly mapped on phenotypic PH data of mature plants, which largely ignored the dynamic development process of PH. The net effect of QTL expression in the time period from t-1 to t may be evaluated based on the time-related genetic model of quantitative traits (Zhu, 1995). Forty-one unconditional QTL and 23 conditional QTL were detected using an intraspecific recombinant inbred line population of upland cotton, and some QTL could not be detected at the later stage of maturity (Shang et al., 2015b). Recently, thirty-two unconditional QTL and 24 conditional QTL for PH were detected using an intraspecific testcross population of upland cotton (Ma et al., 2018a). These results indicated that the expression of PH QTL during plant growth has temporal characteristics.
At present, there are relatively few studies on the dynamic QTL mapping of cotton PH. In this study, unconditional QTL and conditional QTL for PH in different environments and developmental stages were detected, and candidate genes affecting PH development in cotton were predicted. Our results shed substantial light on the genetic basis of cotton PH, will lay a foundation for further functional verification of candidate genes, and enable molecular selection toward ideal plant architecture.

Plant Materials
An F 2:7 RIL population, containing 201 lines coming from an intraspecific upland cotton cross of WCT-4 (female) × NK-4 (male), was constructed using a single seed descent method. The WCT-4 has short PH and super fiber quality, and the NK-4 is tall with average fiber quality. Both parents were developed by the Cotton Breeding & Cultivation team of Shandong Agricultural University. The WCT-4 is a strain from the cross of Shannong269//(Bersiruo/Xinhai2hao) F 4 . The NK-4 is a strain from the cross Shannong6/Shiyuan321//shannong6. The RIL population with two parents were grown in nine environments across two locations and 5 years in Shandong Province, China, including Heze from 2017 to 2021 (17HZ, 18HZ,19HZ,20HZ,21HZ), and Dezhou from 2018 to 2021 (18DZ, 19DZ, 20DZ, and 21DZ). The materials planting followed a randomized complete block design with two replications at each location under all environments. Each replicate had one row of 5 m. The row width and average plant space were 0.9 and 0.23 m in HZ in all 5 years and were 0.8 and 0.25 m in DZ in all 4 years. Planting dates for all environments were April 24-28. Local cultivation management methods were used in each location.

Phenotypic Measurements and Analysis
Plant height refers to the length from the cotyledonary node to the main stem growth point. Dynamic PH was investigated four times during cotton growth, with an interval of about 10 days, in all environments except for in 17HZ, 19DZ, 21HZ, and 21DZ, of which only three times array were carried out. The investigation time points were t1: June 25th, t2: July 5th, t3: July 15 in 19DZ, 21HZ, and 21DZ, and t2: July 5th, t3: July 15th, t4: July 25th in 17HZ, and t1: June 25, t2: July 5, t3: July 15, t4: July 25 in other environments. t1-2 referred to the net increase of PH between t1 and t2, t2-3 was the net increase of PH from t2 to t3, and t3-4 was the net increase of PH from t3 to t4. The descriptive data of RIL population PH traits were sorted out and calculated by IBM SPSS Statistics 25. Trait variance and generalized heritability (H 2 ) were calculated with the QTL IciMapping 4.2 (Meng et al., 2015;Ma et al., 2019). Generalized heritability above 40%, between 20 and 40%, and less than 20% were regarded as high, medium, and low heritability, respectively .

Genetic Map Construction
Sample DNA was extracted using a plant genomic DNA extraction kit. SNP markers were developed through RAD-seq at ROI-BIO Co., Ltd. The parental NK-4 and WCT-4 obtained 41.09 and 38.94 Gb clean reads, respectively. For each RIL line, 2.22-15.68 Gb clean reads were obtained. The genetic map was constructed with an RIL population (F 2:8 generation), and the SNP markers were filtered according to the sequencing results. First, the parental heterozygous markers, no polymorphic markers between the two parents, the bi-allelic markers, and the markers with a deletion rate of greater than 50% were eliminated. Then, the linkage group was constructed on the LOD value, and then the markers that were too far to be used to construct the genetic map were deleted. The genetic map was constructed with the Kosambi mapping function in JoinMap 4.0 (Ooijen, 2006) and created by MapChart 2.2 (Voorrips, 2002). The TM-1 genome sequence of Zhejiang University  was downloaded from the CottonFGD. 1 TBtools was used to extract the genome chromosome length information and draw the collinearity relationship diagram (Chen et al., 2018). IBM SPSS Statistics 25 was used to calculate the Spearman coefficient.

Quantitative Trait Loci Analysis
The PH data of the RIL population under nine environments and their BLUPs (named as BLUP3) across all the tests were used for QTL analysis. Two other BLUPs, namely, BLUP1 for HZ experimental site and BLUP2 for DZ site, were also estimated and used for QTL analysis to detect specific QTL of the experimental site. The lme4 package in the R software was used to estimate the BLUPs (de Los Campos et al., 2013;Huang et al., 2017). Unconditional QTL was the cumulative effect from the initial time to t (Liu, 2011), which was detected by t1, t2, t3, and t4, respectively. The net genetic effect of PH increase from t-1 to t was revealed by the conditional genetic analysis (Zhu, 1995). The growth of PH in the past adjacent time period ( t1-2, t2-3, and t3-4) were used to identify conditional QTL. Consistent QTL were determined by integrating unconditional QTL and conditional QTL. QTL were identified based on the PH value of nine environments and three BLUPs by the ICIM-ADD method in QTL IciMapping 4.2 (Meng et al., 2015), with 1 cM of the parameter step value, 0.05 value of P of type I error, and 0.01 value of the PIN; meanwhile, the threshold was calculated in 1,000 permutations for each trait. The QTL confidence interval (95%) was set as the mapping distance interval corresponding to reduce 1 LOD on either side of the peak (Yu et al., 2012;Liu et al., 2019). When the QTL confidence intervals partially or completely overlap, it was regarded as the same QTL. QTL detected in two or more tests or environments with LOD value greater than 2.5 and phenotypic variation explanation rate greater than 10% were 1 https://cottonfgd.org/ defined as major QTL. It was regarded as stable QTL, which was detected in at least three environments (Sun et al., 2012;Fan et al., 2015;Liu et al., 2018). The name of QTL starts with q, followed by the trait name, chromosome, and QTL number (McCouch et al., 1997), and the names of unconditional QTL and conditional QTL were respectively prefixed with U and C. If a conditional QTL shared confidence interval with an unconditional QTL, these two QTL were renamed as one QTL without prefixed both U and C in the name. The QTL were displayed on the genetic map by MapChart 2.2 (Voorrips, 2002). GraphPad Prism 9 was used to do boxplot of QTL with phenotypic data of BLUP3. The positive additive effect of QTL indicated that the positive allele was derived from the female parent WCT-4; otherwise, the positive allele was derived from the male parent NK-4.

Candidate Gene Analysis
To annotate the gene function, we used the COTTONOMICS database 2 to extract candidate genes from the confidence intervals of QTL with large effect. TBtools and omicstudio 3 were used to do the Gene Ontology (GO) enrichment analysis on genes, while novomagic 4 was utilized to perform the Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis on genes. To uncover the general expression patterns of candidate genes, the transcriptome data of TM-1 roots, stems, and leaves were used as a reference . TBtools was used to create the genes expression heatmap (Chen et al., 2018).

Plant Height Variation of Parents and the Recombinant Inbred Line Population
The PH of male parent NK-4 was higher than that of female parent WCT-4 at all the time points (i.e., t1, t2, t3, and t4) and most time periods (18/23) in the nine environments. The PH of the RIL population showed a two-way transgressive segregation in all time points under all environments except for the t1 in 17HZ with a one-way transgressive segregation ( Table 1). The values of skewness and kurtosis in each environment were close to 0, and the PH of the RIL population showed a normal distribution. Under the nine environments, the PH of the parents and RIL population increased rapidly in the t1-2 and t2-3, with an average growth of 19.48 and 13.90 cm, and slowed down in the t3-4, with an average increase of 5.97 cm ( Table 1). The results showed that the PH of parents and RIL population had dynamic developmental characteristics. The analysis of ANOVA on the RIL population showed that the environment, genotype, and genotype × environment had significant variations on PH at t1, t2, t3, and t4 (P < 0.01) (Supplementary Table 1). Furthermore, the generalized heritability estimates (H 2 ) were 78.19, 83.62, 85.57, and 84.78% at t1, t2, t3, and t4, respectively, suggesting that PH was highly heritable in this RIL population (Supplementary Table 1).    Uq and Cq refer to unconditional and conditional QTL, respectively; HZ refers to Heze, Shandong (2017, 2021; DZ refers to Dezhou, Shandong (2018, 2021; BLUP1 refers to the best linear unbiased predicted value calculated in the 5 Heze environments; BLUP2 refers to the best linear unbiased predicted value calculated in the 4 Dezhou environments; BLUP3 refers to the best linear unbiased predicted value calculated in the nine environments; and PVE and additive means the explanation of phenotypic variation and additive effect, respectively.

Genetic Map Construction
By comparing the RAD-seq sequencing results with the reference genome of TM-1, a total of 555,699 SNP markers were obtained. According to the genotyping criteria, SNP marker filtering was performed. First, 87,613 SNP markers remained after filtering out SNP markers that were heterozygous and non-polymorphic between parents, and then, 87,596 SNP markers remained after filtering out SNP markers of diallelic sites. Finally, after filtering out SNP markers with a deletion rate greater than 50%, 10,054 SNP markers were used for genetic map construction (Supplementary Table 2). A high-density genetic map of 5,850 SNP markers was consequently constructed. The genetic map included 52 linkage groups (LGs), and A13 contained the most SNP markers (1,067), followed by D01 (730) (Supplementary Table 3). The total length of the genetic map was 2,747.12 cM with an average genetic distance of 0.47 cM between adjacent markers (Figure 1 and Supplementary Table 3). The longest LG was D07-1, which spanned 162.45 cM and contained 290 SNP markers with an average of 0.56 cM. The shortest LG was D11-1, which spanned 6.21 cM and contained 26 SNP markers with an average of 0.24 cM. Among the 52 LGs, 24 LGs had Spearman coefficient greater than 0.80, and 26 LGs had Spearman coefficient greater than 0.70 Table 3). Spearman coefficient of each LG was found to have an average value of 0.80, indicating a high degree of collinearity between the genetic map and the physical map. It indicated that the order of most markers in the genetic map was consistent with the order in the physical map, according to the collinearity analysis based on the physical location on the TM-1 reference genome and the position in the linkage group of the genetic map (Figure 2 and Supplementary Table 3).

Conditional Quantitative Trait Loci Mapping
In 3 time periods (i.e., t1-2, t2-3, and t3-4) across the 12 sets of phenotypic data of PH, a total of 28 conditional QTL were identified on 17 chromosomes, which explained 3.27-28.87% of phenotypic variance with LOD value ranging from 3.18 to 28.78 (Supplementary Table 5). Among the 28 QTL, 13 and 15 QTL were detected on the A subgenome and the D subgenome, respectively. The alleles from WCT-4 for 15 QTL had positive additive effects on PH, while alleles from NK-4 for other 13 QTL had positive additive effects on PH. Furthermore, CqPH-A11-1 was identified in 2 tests as the major QTL explaining 4.91-5.99% of phenotypic variance with an LOD value ranging from 4.96 to 7.67 (Table 2 and Figure 3). Seven QTL and 23 QTL were detected in t1-2 and t2-3, respectively, and no QTL was detected in t3-4, indicating that conditional QTL were more positively expressed in t2-3.

Consistent Quantitative Trait Loci Mapping
By compared 37 unconditional QTL with 28 conditional QTL, 5 QTL (i.e., qPH-A08-1, qPH-A11-1, qPH-D07-1, qPH-D09-1, and qPH-D13-1) shared overlapping or common confidence intervals and were regarded as consistent QTL (Table 3 and Figure 3). In summary, a total of 60 QTL (31 on the A subgenome and 29 on the D subgenome) were identified across the 12 sets of phenotypic data of PH (Supplementary Table 6). In addition, no QTL could be detected at all stages in the same environment, which accorded with the characteristics of PH dynamic development.

Plant Height Candidate Genes Prediction
To predict candidate genes associated with cotton PH, homologous genes functional annotation of Arabidopsis were performed on these 948 genes within the physical interval of five stable unconditional QTL and five consistent QTL. Among the 778 genes with functional annotation information of homologous genes in Arabidopsis, to reduce candidate genes prediction errors, only genes whose homologous genes in Arabidopsis have been annotated to be involved in PH regulation, or genes that were previously proved to regulate PH, were included in the predicted candidate genes. Then, we finally predicted 14 candidate genes for cotton PH (Supplementary Table 10).
In this study, the gene GH_D11G2212 was annotated as AGAMOUS-like 29 in Arabidopsis, and the genes GH_D11G2122 and GH_D11G2351 were annotated as AGAMOUS-like 104 in Arabidopsis (Supplementary Table 10). The homologous gene Gh_D03G0922 of Arabidopsis was annotated as AGAMOUSlike 8, which had been proved to be related to cotton PH development (Su et al., 2018). The gene GH_D11G2128 is annotated as gibberellin-regulated family protein in Arabidopsis (Supplementary Table 10). Gibberellin can regulate plant stem elongation and other developmental processes (Braun et al., 2019;. The gene GH_D09G1177 was annotated as auxin response factor 4 in Arabidopsis, and the gene GH_D11G2121 was annotated as auxin-responsive GH3 family protein in Arabidopsis (Supplementary Table 10). The gene GH3 couples IAA to amino acids to maintain auxin homeostasis. It was indicated that the GH3 gene participated in PH regulation and its high expression resulted in shorter plant (Yu et al., 2002;Staswick et al., 2005). The gene GH_D11G2197 is annotated as squalene epoxidase 2 in Arabidopsis (Supplementary Table 10) and may be involved in sterol biosynthesis (Laranjeira et al., 2015). The genes GH_D09G1074, GH_D11G2118, GH_D11G2301, GH_D11G2302, GH_D11G2303, GH_D11G2399, and GH_D11G2415 were annotated as cytochrome P450 in Arabidopsis (Supplementary Table 10). Cytochrome P450related gene mutations may hinder the synthesis of brassinolide and cause plants to become shorter (Yang et al., 2014;Wu et al., 2016;Qi et al., 2017).
The gene expression heatmap indicated that the gene GH_D09G1177 was specifically expressed in the stem and root of TM-1, the genes GH_D11G2121, GH_D09G1074, and GH_D11G2118 were specifically expressed in the root of TM-1, and the gene GH_D11G2399 was specifically expressed in the stem of TM-1 (Hu et al., 2019; Figure 5). Since the PH is mainly determined by the growth of the stem, we can give priority to the gene GH_D11G2399, which is preferentially expressed in the stem, and its expression level is significantly higher than that in other important tissues (leaf and root) of TM-1.

Comparison With Previously Reported Plant Height Quantitative Trait Loci
Since quantitative traits are easily affected by the interaction between genes and the environments, the phenotypic data were usually collected from various environments to reduce the effect of environments. It can improve the accuracy of prediction by the phenotypic regression analysis on hundreds of thousands of mutations through a genome-wide regression model at the same time (de Los Campos et al., 2013). BLUP is a genome regression model method that is commonly used in animal and plant populations to improve prediction accuracy (de Los Campos et al., 2013;Huang et al., 2017). In this study, the BLUPs of 5 HZ environments, 4 DZ environments, and all environments were calculated for QTL mapping to improve the accuracy of QTL mapping. Among the 60 unconditional and conditional QTL detected in this study, 10 QTL were regarded as stable or consistent QTL. Furthermore, the annotation of homologous genes in Arabidopsis detected candidate genes with proved gene function involving PH regulation in these stable QTL confidence intervals. In addition, stable QTL detected at specific location was valuable for developing cultivars suitable to that location. The results of this study showed that 27 QTL were only detected in the HZ environment including a stable QTL (UqPH-D02-1) and a consistent QTL (qPH-D13-1), and 21 QTL were only detected in the DZ environment including a stable QTL (UqPH-A09-1).
Plant architecture is related to traits such as yield and fiber quality and has an important impact on high yield and good quality (Fu et al., 2019). Moreover, there are few QTL detected for PH in cotton, so it is necessary to use high-density genetic map to detect QTL for PH (Li et al., 2013;Liu et al., 2020). In this study, we used the constructed high-density genetic map and QTL mapping based on the PH phenotype data of the five generations of F 9 -F 13 , and a total of 60 PH QTL were mapped. To identify the newly discovered QTL in this study, the upland cotton PH QTL included in CottonGen 5 (Huang et al., 2017;Liu et al., 2020;Yu et al., 2021), and PH QTL identified in recent years (Said et al., 2013;Huang, 2018;Su et al., 2018;Ma et al., 2019) were compared with the QTL of this work. The physical location of QTL is determined according to markers of confidence interval. If the physical intervals between QTL shared partially or fully overlapped, they are considered to be the same QTL. Due to the difference of marker types used in previous QTL mapping research, only QTL mapped with SNP markers were included in QTL comparison. It was found that 21 QTL (35.0%) mapped in this study were previously reported (Supplementary Table 6), including 12 unconditional QTL, eight conditional QTL, and one consistent QTL. The rest 39 QTL detected here were preliminarily considered as new cotton PH QTL. In addition, seven QTL were detected by BLUP, suggesting that using BLUP to detected QTL was reliable.

Conditional Quantitative Trait Loci Mapping
unconditional QTL and conditional QTL after the comparison. The number of QTL varies at different stages of PH development, most unconditional QTL were detected in the early and middle stages, while most conditional QTL were detected in the middle stages, indicating that PH QTL have spatiotemporal expression characteristics (Shang et al., 2015a(Shang et al., , 2016Wang et al., 2019). The same QTL can also be expressed in different periods, for example, consistent and stable QTL qPH-A08-1 was detected in 12 periods of t1, t2, t3, t4, and t2-t3 of 5 environments, and qPH-D09-1 was detected in five periods of t2, t3, t4, and t1-t2 of 3 environments (Table 3). However, no QTL was found to be expressed in all stages of an environment, indicating that different QTL are expressed at different stages in the process of PH development.

CONCLUSION
Plant height is a key plant architecture factor affecting cotton fiber yield and other traits. A high-density genetic map of 5,850 SNP markers was constructed using an upland cotton intraspecific recombinant inbred line (RIL) population. Based on phenotypic PH data from nine environments and three BLUPs, 37 unconditional QTL and 28 conditional QTL sharing five common QTL were identified for PH, including 21 previously detected PH QTL. Fourteen candidate genes for PH were predicted within 10 stable QTL intervals. This study shed substantial light on the genetic basis of cotton PH, will lay a foundation for further functional verification of candidate genes, and enable molecular selection toward ideal plant architecture.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding authors.

AUTHOR CONTRIBUTIONS
JW completed the phenotypic data collection, analyzed and summarized all the data, and wrote the manuscript. LM and JT participated in the phenotypic data collection. XW completed the candidate genes enrichment and wrote the results section. HZ, MX, YS, YZ, GZ, ZZ, YW, MC, and LW managed and collected the phenotype data. XLS and XZS designed the research. XLS revised the manuscript. All authors read and approved the final manuscript.

ACKNOWLEDGMENTS
We are grateful to the laboratory members for help, advice, and discussion.