Boqi Li, Qin Tin, Xuwen Wng, Bei Hn, Li Liu, Xinhui Kong,Aijun Si,Jun Wng, Zhongxu Lin, Xinlong Zhng,Yu Yu,*, Xiyn Yng,*

aNational Key Laboratory of Crop Genetic Improvement,Huazhong Agricultural University,Wuhan 430070,Hubei,China

bCotton Institute,Xinjiang Academy of Agriculture and Reclamation Science,Shihezi 832000,Xinjiang,China

Keywords:Upland cotton Water-limited conditions Fiber yield Fiber quality QTL hotspot

A B S T R A C T Global warming is limiting availability of water resources in arid and semi-arid regions,and so understanding water use efficiency (WUE) is increasingly important for agricultural production in those areas. As China is the largest cotton producing area, the problem of balancing WUE and efficient cotton production is a major issue. In this study, we used a natural population of 517 Upland cotton accessions to conduct a water-controlled trial in south and north of Xinjiang over two years. A total of 18 traits including agronomic traits,fiber yield indices and fiber quality indices, were investigated for broad-sense heritability and coefficient of variation. Appropriate water limitation was found to promote the establishment of favorable agronomic traits in cotton, associated with an increased cotton yield of 8.46% in Xinjiang, at the expense of a certain degree of fiber quality, such as decreased fiber length and an over-higher micronaire value.We detected 33 QTL related to response to water limitation using a drought resistance coefficient (DRC), and 6 QTL were found using a comprehensive indicator of CIDT(comprehensive index of drought tolerance)at the genetic level by integrating resequencing data.Two novel QTL-hotspots including six differentially expressed genes (DEGs) were further identified related to the drought response of cotton. These findings not only suggested a new approach to irrigation of cotton fields in Xinjiang, but also provided abundant genetic evidence for genetic breeders to study drought improvement of crops.

1. Introduction

Cultivation of cotton(Gossypium hirsutum)has been carried out for thousands of years.As an important global economic crop,cotton fibers are indispensable raw material for processing by the textile industry to make comfortable clothing, and cottonseeds are also used for oil production [1]. China is the world’s largest cotton importer (2.10 Mt) but little at export with 0.05 Mt. and far less than the export volumes of the United States (3.21 Mt) and Brazil (1.31 Mt). Although China ranks as the first largest cotton production country (6.04 Mt),contributing approximately 23.31% to the world’s total, the yield only ranks third(1726 kg ha?1in 2018/2019)[2].Xinjiang,located in the arid and semi-arid northwest [3], with the suitable climatic conditions of plentiful sunshine, the large temperature differences between day and night,and little rain during the boll development stage, is the largest cotton planting and production area in China [4,5], and currently dominated more than 80% of the country’s high-quality cotton in 2018[6].

Due to the arid and semi-arid environments,more than 95%water resources in Xinjiang was used for agriculture.Water use efficiency (WUE) is increasingly important for agricultural production in arid and semi-arid regions,especially in Xinjiang with little rainfall.It is therefore essential to breed water-saving and drought-resistant crops to improve agricultural economics[7]. Submembrane drip irrigation has been developed in Xinjiang for cotton and other crops.This irrigation system was considered to be the most promising water-saving irrigation technology with the advantages of saving water, increasing production and reducing consumption [8]. However, the problems of water shortage and WUE could only be fundamentally solved by screen and cultivate drought-resistant varieties.Since the early 1980s, the two main strategies of drought breeding have been adjusting the amount of irrigation water to achieve maximum crop yield,and directly controlling water restrictions for screening high-yielding varieties [9,10]. There were two different ways to screen and identify the drought-resistant cotton germplasm.The results of limited irrigation in the field were reliable and true,but they are affected by the season and consume a lot of labor costs,while another method is through the water deficit stress in the greenhouse[11].In addition to pot culture experiments to control soil water content in greenhouse[12], chemical reagents including PEG and mannitol rapidly produce osmotic stress to simulate drought have also been applied to cotton research[13].Abdelraheem et al.[14]analyzed 661 abiotic and biotic stress resistance QTL and identified 98 QTL for drought tolerance under greenhouse and 150 QTL in field conditions.

Though cotton is a relatively drought-resistant economic crop, the worsening drought in global environment also threatens the yield and quality of cotton[13,15,16].The effects of water deficit on crops depend on the stage of development,the degree of drought and the duration of drought.For cotton,water deficit causes flower bud shedding, reduced fiber elongation, plant height, and fiber quality. In Xinjiang in particular, which is facing serious water resource challenges,water-saving irrigation and drought-resistance breeding has been an intractable problem [17–19]. Complex genetic mechanisms hinder the improvement of drought-tolerant cotton varieties, and plants are sensitive to diverse severe environmental changes[20],and there is an increase in arid and semiarid regions across the world [21]. With the continuous development of sequencing technology and the gradual improvement of crop genomes, the use of large natural populations rich in genetic variation to mine and locate drought candidate genes is increasingly conducive to longterm natural selection and artificial breeding.

In previous studies, genome-wide association studies(GWAS) [22] have been widely used on cotton with large natural resequencing population in recent years, and many QTL related to seed quality,morphological traits,fiber quality and productivity traits have been identified in different cotton populations [23–28]. However,most of the identified QTL had a relatively large interval that did not finely target a smaller genetic region or a single gene.Therefore,it was necessary to exploit high-resolution and high-density single nucleotide polymorphism (SNP) markers to precisely locate candidate genes. In the past decade, great progress has been made in dissecting the genetic basis of drought resistance through QTL mapping using parental or natural populations [29]. Zheng et al.[30]detected 35 QTL under water-limited conditions in the field using an F2:3population. Ulloa et al. [31] detected more than 150 QTL associated with productivity and fiber quality traits using two recombinant inbred lines (RIL) populations under regular-water and low-water irrigation field conditions.Islam et al. [32] identified the functional gene of GhRBB1_A07 related to superior fiber quality by using a multi-parent advanced generation inter-cross (MAGIC) population [33]based on GWAS. Hou et al. [34] identified 20 SNPs related to drought-tolerance traits using 319 upland cotton accessions by integrating GWAS and RNA-seq under PEG treatment in the greenhouse.

In the present study, appropriate water rather than too much water supply has been proven to contribute to the increase of cotton production although at the expense of fiber quality. Eighteen traits were obtained by conducting controldrought experiments with a panel of 517 Upland cotton accessions, and then the QTL hotspots and candidate genes related to drought tolerance were identified by GWAS in cotton. The objectives in our study were to examine the plasticity of cotton yield and its agronomic traits under wellwatered and water-limited conditions in the field, as well as identify QTL for corresponding drought stress in cotton. The findings in our research provide new insights into cotton drought study in the field and should be useful for molecular breeding of crop resistance.

2. Materials and methods

2.1. Plant materials and experiments design

A natural population of 517 Upland cottons(Table S1)were collected mainly from China (five cotton growing regions:NNIR, NSEMR, YRR, YtRR, SC) and two foreign countries(United States and the Soviet Union) as described previously[35]. These were planted in the north and south of Xinjiang(Shihezi, 44.27°N, 85.94°E; and Korla, 41.68°N, 86.06°E) during 2016 and 2017, under either well-watered or water-limited conditions. The field experiments were conducted in a completely randomized block arrangement with two replicates. Each accession was planted in a block with 2.2 m long and 0.9 m wide in double rows in each replicate. The control panels were normally irrigated 12 times in south Xinjiang(6768 m3ha?1) and 10 times in north Xinjiang (4950 m3ha?1).The water limitation treatment panels were supplied with 50% the water volume of the controls, as soon as seedlings emerged (Table S2). The plants in south Xinjiang received an extra irrigation (4500 m3ha?1) to reduce the saline and alkali content in the soil on March 15, and plants in north Xinjiang received an extra irrigation (240 m3ha?1) for seeding on April 20.

2.2. Phenotypic evaluation

2.2.1. Phenotype acquisition

Phenotypes were investigated at seedling, flowering and boll development stages. The calculation of growth period (GP)was from seedling stage (50% cotyledon spreading) to first fruit branch bolting(50%of all plants).Seven indices including plant height (PH), first fruit spur height (FFSH),first fruit spur branch number (FFSBN), fruit spur branch number (FSBN),empty fruit spur branch number (EFSBN), effective boll number (EBN) and peripheral boll number (PBN) were investigated with the selection of ten plants with identical growth in late September.Thirty bolls in the upper,middle and lower parts were harvested for indoor seed testing after boll opening and four indices of boll weight (BW), seed weight (SW), lint weight (LW) and lint percentage (LP) were acquired. Yield(with seeds) per hectare(YPH) was converted from the actual seed cotton weigh of each block.Fiber quality traits including fiber length (FL), fiber strength (FS), fiber uniformity (FU),micronaire value (MV) and fiber elongation (FE) were determined by 14 g lint in each replicate of each variety. Fiber quality testing was carried out in the Cotton Research Institute of Xinjiang Academy of Agricultural Reclamation Sciences using a high-volume instrument(HFT9000).Analysis was carried out at an ambient temperature of (20± 2) °C, and relative humidity of (65±2)%.

2.2.2. Statistical analyses

The above eighteen phenotypic traits were analyzed by integrating multi-environmental data using best linear unbiased predictions(BLUPs)based on a linear model according to a previous study [26]. Analysis of variance (ANOVA) was performed to evaluate the effects of genotype (G), environment (E) and the interactions between genotype by environment (G × E) [26]. Broad-sense heritability was estimated by using ANOVA based on the two year's phenotypic data: H2=Vg/(Vg+Vll/n+Vly/n+Vr/n2)[36];Vgrepresents the genotype variance, Vlland Vlystand for variance between lines and locations, lines and years, and Vrstand for residuals; n represents the number of replicates. The statistical analyses of mean and standard deviation were calculated using the overall BLUP results, and P-value was performed by pairedsamples t-test.

2.2.3. Evaluation of drought response

Coefficient of variation (CV) among 517 accessions was used to determine the discrete degree of observational data when comparing different attributes of traits, and was calculated by the formula:CV=standard deviation/arithmetic mean[37].The drought resistance coefficient(DRC;ratio between traitwater-limitedand traitwell-watered) of all traits was used to analyse the performance of the accessions under water-limited conditions.In order to comprehensively evaluate the drought response of cotton, an indicator named comprehensive index of drought tolerance (CIDT) was defined as [CIDT = DRC (LP + LW+ YPH +MV +FSBN + EFSB N) / (GP + SW + FL + FS + FE + PH + FFSH +FFSBN + EBN + PBN)], based on the significant (P-value < 0.001)positive and negative effects of traits on water-limited conditions,according to Dilnur et al. [38].

2.3. Genome-wide association studying and casual DR genes identifying

2.3.1. Genotyping, mapping and SNP calling

A subset of 314 cotton accessions were resequenced for this study, and data of another 203 resequenced accessions [27]was analyzed following download from NCBI. Young leaf tissues were sampled for DNA extraction and at least 5 μg was used to construct a sequencing library with an Illumina TruSeq DNA Sample Prep Kit, according to manufacturer's protocols. Paired-end sequencing (PE 150 bp) of each library was performed on an Illumina NovaSeq platform. Allotetraploid cotton (Gossypium hirsutum L.) genome acc. TM-1 was used as the reference genome [39]. Clean paired-end reads were mapped to the TM-1 genome by BWA software [40] using the default parameters.

2.3.2. Genome-wide association study and QTL discovery

Genetic variants were annotated using the software SnpEff(version 4.3) [41]. All SNPs were categorized as being in intergenic regions, upstream or downstream regions, in exons or introns by default parameters. All filtered SNPs used in association analysis were compressed into binary formats using plink software (version 1.90b6.8) [42]. GWAS was performed with the Fast-LMM (version 2.02) [43] as previously described. The traits of DRC and CIDT were used for GWAS in our study. Strict significant association threshold values were set at ?lg (0.05/n) (n, total filtered SNP number) for control panel and a relative strict threshold value were set at –lg (1/n) for DRC and CIDT. Data with DRC was conducted to factored spectrally transformed linear mixed models (FaST-LMM), α = 0.05 was set as the Bonferroni correction, and the strict significance P-value was set at 1.95 × 10?8. To identify causal drought-related (DR) genes, we integrated the GWAS results and a previous RNA-seq data for a drought stress profile of cotton in our previous study.Differentially expressed genes (P-adj < 0.05) in these QTL were considered to be DR genes. The plot of QTL on chromosomes was drawn using the R package ‘RIdeogram’[44]. The method to compare the QTL in this study with the reported QTL was described by Wen et al. [45]. The candidate genes were annotated by sequence blast from NCBI-nr database.

2.3.3. qRT-PCR analysis of candidate genes

Quantitative real-time PCR (qRT-PCR) was carried out using the ABI 7500 system (Applied Biosystems, Foster city, CA). Leaf samples of a cotton material of ZY7 were used as the template from control and drought samples. Gh_UBQ7 was used as an internal control for other genes by analysing the relative expression in qRT-PCR tests. The 2?ΔΔCTmethod with two biological and three technical replicates was used to calculate relative changes in gene expression levels [46]. All the genes sequences and primers are listed in Table S3.

3. Results

3.1. Analysis for phenotypic variation in eighteen traits

Eighteen traits including agronomic, yield and quality traits for a natural cotton population comprising 517 Upland accessions were acquired through two year-experiments with two treatments (well-watered and water-limited conditions)at two locations (Shihezi and Korla) in Xinjiang. The broadsense heritability (H2) of the 517 accessions under different treatment conditions (H2wwfor H2in well-watered; H2wlfor H2in water-limited) was estimated separately with BLUP based on multi-year multi-point experimental data (Table S4)(0.16 < H2ww< 0.78; 0.08 < H2wl< 0.76). The results indicate that high heritability (H2> 0.6) was found for eight traits: FL(H2ww= 0.78; H2wl= 0.76), LW (H2ww= 0.75; H2wl= 0.76), FFSH(H2ww= 0.75; H2wl= 0.76), SW (H2ww= 0.68; H2wl= 0.68), FS(H2ww= 0.72; H2wl= 0.75), PH (H2ww= 0.67; H2wl= 0.68), MV(H2ww= 0.64; H2wl= 0.66), and LP (H2ww= 0.63; H2wl< 0.70) under well-watered and water-limited conditions. Traits with high heritability were stable from year to year, while traits with low heritability varied greatly from year to year (Table S4). ANOVA was used to analyze the effects of G, E, and G × E for 18 traits in multi-environments under two different water irrigation conditions. The results indicated that all traits were significantly (P-value < 0.001) subjected to genotypes and environments under different water conditions, with less G × E effect on fiber quality traits, including three traits under both wellwatered and water-limited conditions (FL, FE, and FS), and two traits under water-limited conditions (FU and MV) (Table S5).

The CV of 18 traits were analyzed in both environments over two years. Two traits of BW (P-value = 0.74) and FU (Pvalue = 0.84) were not significantly affected by water limitation. While other 16 traits displayed a significantly (Pvalue < 0.001) increase or decrease under water deficit,including LP, FL, MV, GP, PH, etc. (Table S6). PH had a clear response that the CV changed from 7.90% under well-watered conditions to 10.17% under water-limited conditions. The CV of FFSBN showed a narrow variation (1.07%) under water stress, while it showed a relative higher CV at 5.11% under well-watered conditions. Intriguingly, FFSH had a higher variability between different varieties (13.17%), and it was still high under water-limited conditions (13.74%). This might be due to the fact that cotton FFSH was sensitive to water and would decrease in overall level under drought conditions.However, the CVs of some traits (FU, FE, FSBN, and GP) were very low.

3.2. Performance of cotton yield, fiber quality and related traits under different watering conditions

According to the data integrated by BLUP, some vegetative traits and traits related to fiber quality were found to be significantly affected by water limitation (Table S6). GP ranged between 130 and 141 days with a mean of 136 days under wellwatered conditions, while it ranged between 128 and 138 days with a mean of 133 days under water-limited conditions. PH ranged from 51.78 to 84.97 cm (mean 67.78 cm) under wellwatered and from 41.20 to 79.58 cm (mean 62.51 cm) under water-limited conditions. FFSH ranged from 13.14 to 32.25 cm(mean 21.91 cm) under well-watered and from 12.84 to 31.77 cm (mean 20.79 cm) under water-limited conditions.The mean values of GP, PH and FFSH were decreased by 1.88%,7.78% and 5.12%, respectively.

Traits related to fiber quality were also adversely affected by water deficit. The mean value of FL decreased from 28.86 mm to 28.44 mm, and FS decreased from 28.42 to 28.22 g tex?1. MV was also affected by drought (mean values of 4.46 under water-limited vs. 4.19 under well-watered). Conversely, LP increased by 2.36% with mean values of 39.56% and 40.49% under the well-watered and water-limited conditions.For the important trait of YPH, it ranged from 2958 to 5414 kg under well-watered and from 3592 to 5608 kg under waterlimited conditions (Fig. 1), and the mean value increased from 4142 to 4492 kg (8.46%). More details of trait variations were supplied in Table S6.

3.3. Analysis and integration of resequencing data and phenotypes

3.3.1. Genotyping and variant calling

To analyse the molecular basis of these traits related to drought resistance, a population of 517 natural Upland cotton accessions was selected and the resequencing data were integrated for variant calling. After mapping against the allotetraploid reference genome TM-1 and filtering out lowquality reads, a total of 14,706,082 single nucleotide polymorphisms (SNP) were obtained. About 70% of mutations occurred in the intergenic regions, while fewer mutations were found at upstream (12.18%) or downstream (11.23%) regions of genes(Table S7). However, start codons were extremely unlikely to undergo SNP mutations (0.01‰). The variant rate ranged from 1/174 bp on Chr D04 to 1/72 bp on Chr A08, with an average variant rate of 1/112 bp across the 26 chromosomes of the cotton genome. The highest variant rate was observed on Chr A08 which has the longest chromosome length, while the shortest chromosome Chr D03, has the lowest frequency of variants at 287,352 and with the lower variant rate of 1/162 bp(Table S8). By further filtering out minor alleles (MAF < 0.05)and scaffold loci on chromosomes, the remaining 2,564,238 SNPs were obtained for subsequent genome-wide association studies (GWAS).

3.3.2. GWAS and candidate water limitation stress-related loci identification

GWAS was performed across the cotton accessions for the DRC value of 18 traits. A total of 33 candidate QTL were colinked on 12 chromosomes (A01, A04, A06, A07, A08, A09, A10,A12, D03, D05, D06, D11) (Fig. 2) using DRC, of which DRC-QTL4 was co-localized by two traits of PH and LP, and DRC-QTL15 was co-localized by two traits, FFSBN and PH (Table S9).Moreover, six significant QTL intervals were localized on six chromosomes (A05, A06, A07, A12, D01, D07) using CIDT (Table S10). A Manhattan plot using CIDT is shown in Fig. 3a, and two 2 Mb intervals was determined by the significant association SNPs on Chr A06 and Chr A12, respectively (Fig. 3b, c). The lead SNPs beyond the threshold (?lg P = 6.41) on these two chromosomes were A06_94140421 and A12_58357256, which contributed to associating candidate genes.

3.3.3. Assessment of DR-related candidate genes by transcriptome analysis and qRT-PCR

To find candidate DR genes in the identified QTL, we integrated the above regions with reported transcriptome data from our previous research. A number of 168 differentially expressed genes (DEGs) were co-localized to these 33 DRC-QTL(Table S9) and 37 DEGs (Table S10) were co-localized to the CIDTQTL. Intriguingly, a region was co-localized by DRC-QTL15 and CIDT-QTL2 on Chr A06, and two DEGs, namely Gh_A06G1325(encoding predicted pyridoxal phosphate-dependent enzyme,YBL036C type) and Gh_A06G1327 (encoding Core-2/I-branching beta-1,6-N-acetylglucosaminyltransferase family protein) were identified within this region (Fig. 3d). Another larger interval was co-localized by DRC-QTL23 and CIDT-QTL4 on Chr A12, with four DEGs, i.e. Gh_A12G0874 (encoding alpha/beta-Hydrolases superfamily protein), Gh_A12G0882 (encoding RWD domain-containing protein), Gh_A12G0886 (DUF3598), and Gh_A12G0888 (cyclin p4;1)(Fig. 3e).

The above six candidate DEGs in two QTL-hotspots were chosen for further verification by qRT-PCR. Gh_A06G1325 encodes a protein with a predicted conserved PLPDE domain and a cDNA of 795 bp, and Gh_A06G1327 encodes a protein with a predicted conserved Branch domain with a cDNA of 1158 bp. Both of these two genes showed the clear upregulation under drought conditions in the leaf of ZY7, while Gh_A06G1325 is highly expressed and Gh_A06G1327 expressed in a relatively low level (Fig. 4). In the other interval,Gh_A12G0874 (1011 bp) and Gh_A12G0882 (759 bp) also showed upregulation under drought conditions. Gh_A12G0874 encodes an alpha/beta-Hydrolase protein and Gh_A12G0882 encodes a protein related to RWD domain-containing proteins. No reported conserved domains have been identified for Gh_A12G0876 (1272 bp), but its homologous Arabidopsis gene AT2G44760 has a domain of unknown function (DUF3598). A relatively short gene, Gh_A12G0878 (612 bp), encodes an important conserved CYCLIN domain and was downregulated under drought conditions. All of these six DEGs in the QTLhotspots have therefore been verified to be transcriptionally modulated by drought stress.

4. Discussion

4.1. Water limitation affects cotton development and morphogenesis

In Xinjiang, where annual rainfall is extremely low, water resources are scarce [3,47]. The yield of cotton depends on the genotype and environmental factors,as are fiber quality and several other agronomic traits [48–50]. Water deficit affects the entire reproductive process in cotton, including seedling, flowering and boll stages. In this study, plant height was significantly influenced by water limitation,and the coefficient of variation gets broader to 10.17%from 7.90%under sufficient water conditions. The important indicator FFSH, related to machine harvesting, still presented a high coefficient of variation (13.74%) under water-limited as opposed to well-watered conditions(13.17%).This indicated that the variation of the FFSH trait among populations was inherently large. After applying water stress, the overall level of variation of all varieties was affected to a similar extent. In addition, FU was not affected by drought. Two traits,EFSBN and PBN,showed high coefficients of variation in both drought and control conditions,which might be due to the relatively low absolute value of statistics used for investigating traits. Although the CV value of the growth period was not large (1.58% for well-watered; 1.52% for water-limited),this indicator was still significantly affected by drought(P-value<0.001),and it has great significance for cotton production.

Fig.3– Candidate genes in QTL-hotspot. (a)Manhattan plot for the trait of CIDT.The horizontal gray dashed lines show the genome-wide threshold (?lg P= 6.41).(b) A 2 Mb interval plot for QTL-hotspot on Chr A06.The arrow points to the significant SNP A06_94140421.(c)A 2 Mb interval plot for QTL-hotspot on Chr A12.The arrow points to the significant SNP A12_58357256.(d,e) Differentially expression genes(DEGs)in the LD decay interval.Gene regions are represented by rectangular graphs.

4.2.Controlled water limitation contributes to improvement of cotton production

Here,a network links yield traits(YPH,BW,LP,LW),fiber quality(FL,FU,MV,FS,FE)and agronomic traits(PH,FFSH,FSBN,EFSBN,GP) around prevailing field water constraints was depicted. We showed that fiber traits,such as FL,LW,SW,FS,MV,and LP,as well as two agronomic traits of FFSH and PH, exhibited high heritability under different conditions across multiple years.The traits GP,FL,PH,FFSH,PBN showed an adverse response trend to water-limited conditions, while LP, YPH, MV, EFSBN showed improvements. Controlled water limitation was found to promote the establishment of favorable agronomic traits resulting in an increased cotton yield of 8.46%.Fiber quality traits,including FL and MV,were significantly adversely affected by water deficit,though there was some increased production under these conditions. MV was a comprehensive index reflecting the fineness and quality of cotton fibers,and a value of between 3.7 and 4.2 might have economic benefit [51]. We found the mean value increased from 4.19 to 4.46 after water limitation in our study,indicating worse benefit.Some research has reported the effects of water deficits on vegetative growth and reproductive growth in crops and fruits.It was suggested that water limitation during the growing period influenced the plant's growth period by reduced vegetative growth and promoting reproductive growth, which might also apply to cotton. It was therefore important to consider the balance between water resources,yield and fiber quality linked to productivity of cotton fields. On the basis of our results and the actual cotton production needs in Xinjiang, the increased yield and high lint percentage would be good for farmer economics, but at the expense of reduced fiber quality. The lower plant height is also suitable for the rational close planting of cotton in Xinjiang, an important indicator related to mechanized harvesting,by sacrificing FFSH[52].

Fig.4– Relative expression of candidate genes.qRT-PCR verification of the transcript levels of candidate genes under control and drought conditions.Error bars represent standard deviations of the mean based on three biological replicates;statistical significance was determined by a two-sided t-test:*P <0.05,** P< 0.01.

When comparing the yield of cotton harvested under different water supply conditions in multiple environments, the results showed that the mean yield had significantly increased under water-limited conditions compared to well-watered in 2017 whether in the north or south, and was slightly lower in 2016(Fig. S1). The underlying cause might be different climatic conditions and a changing ecological environment during the two years.In addition,the annual cotton yield in south Xinjiang was significantly higher than that in the north,whether in 2016 or 2017. This might be due to either better sunshine in south Xinjiang, or the lack of water there, both of which could contribute to increased production. The YPH increased under water-limited conditions in our study was not in agreement with previous studies[53–55].However,the yield was estimated by boll weight and boll numbers or harvesting several rows in previous studies [53,55], the actual harvest yield was used in the current study.One reason might be the more rainfall in the early stage of cotton growth in 2016.The experiments were carried out in the field, the level of water-limitation depended on the irrigation amount with less rainfall;another reason was that the varieties with increased yield had the relative long GP under well-watered condition, while all bolls opened and were harvested under water-limited condition in late September.

In general,the potential yield and fiber quality were critical for cotton economic and both are negatively affected by insufficient irrigation in field [14,56,57]. During the development of fiber,water was needed to maintain cell expansion and carbohydrate assimilation.Therefore,the cells expansion in cotton leaves was affected, which led to the decrease of photosynthesis and carbohydrate supply during the development of cotton bolls,and then affected the development of fiber [58]. However,according to previous studies,fiber quality traits are less affected by water restriction stress[59,60],which were not consistent with our findings except FU in this study. It indicated that development of whole fiber might be affected to the same extent.Besides,the lint percentage was not reduced by water-limited irrigation in our study,which was the same as the study by Ulloa et al.[31].In order to select the elite germplasm with little effect on both yield and drought resistance, we chose germplasm resources with multiple excellent drought resistance traits. By integrating the DRC of YPH(>1)and CITD(>0.7),we found six varieties,including Shaanmian 13,Binchuan 373,Ganmian 13,Guannong 1,Guangye Daizimian, and Shaan 5245, presented high drought resistance with relative high yield potential, which would be the useful breeding materials for molecular and classical breeders.

4.3. Six important genes related to cotton drought-tolerant production

Compared with QTL mapping,GWAS more effectively correlated genotypes and phenotypes in natural populations by detecting underlying natural allelic variations[61–63].Due to its advantages of high resolution, high cost-effectiveness and no need to construct populations, GWAS has been widely used in many crops to dissect complex properties [64]. GWAS was also well used in the analysis of cotton genetics and breeding,especially in cotton fiber trait analysis[27,28,64].In a recent study,Dilnur et al.[38]used a comprehensive salt tolerance index to integrate seven traits for association with 215 accessions of Asiatic cotton, and 583 markers were associated. This reminded us that using integrated phenotypic indicators might contribute to a more efficient discovery of DR genes, since drought response was a complex quantitative trait. In our present study, GWAS was performed using 14,706,082 SNPs with 517 cotton germplasms of to detect DR genomic regions.We analyzed the genetic variation of 18 agronomic and fiber quality traits, as well as predicted candidate genes. We identified 33 QTL related to drought resistance and 6 QTL by a comprehensive indicator of CIDT.Importantly,two QTL-hotspots based on CIDT were detected and six DEGs (Gh_A06G1325, Gh_A06G1327, Gh_A12G0874,Gh_A12G0882, Gh_A12G0886, and Gh_A12G0888) were identified within these intervals by integrating a previous gene expression profile from samples under drought stress.Comparing these two identified QTL-hotspots with previous meta-analysis of QTL studies [14], CIDT-QTL2 and CIDT-QTL4 were two novel loci with drought stress. Five genes were further verified to be significantly upregulated,Gh_A06G1325 encodes pyridoxal phosphate (PLP)-dependent enzyme, Gh_A06G1327 encodes Core-2/IBranching enzyme, Gh_A12G0874 encodes palmitoyl protein thioesterase (PPT), Gh_A12G0882 encodes RWD domain, while no reported annotation with Gh_A12G0886. Another gene(Gh_A12G0888) with the annotation of regulating cyclindependent protein serine/threonine kinase (CDK) activity(GO:0000079), which was reported as core cell cycle regulators and played an important role in abiotic stress[65],was verified to be significantly downregulated under drought conditions compared with controls. These results represented a major step forward in identifying candidate genes involved in the drought response, and future research will be directed to understanding the regulatory mechanisms. The results would also provide valuable molecular markers for improved cotton breeding programs, and new strategies for cotton field irrigation in Xinjiang.

5. Conclusions

In summary, the northwest inland cotton region has its own unique geographical and climatic advantages,but also has the disadvantage of water shortage. According to water demand in Xinjiang, water-saving strategies and drought-resistant breeding in cotton were important ways to balance the development of agricultural production and sustainable use of water resources. To achieve higher yields, we need to optimize the use of water resources and combine the potential of mechanical harvesting. Furthermore, it is also necessary to explore better varieties of cotton adapted to drought or drought resistance from the perspective of molecular breeding. We should cultivate elite varieties to further improve yield production and fiber quality. More water limitation gradient tests should be applied to find the most suitable water supply for cotton growth.The cultivation of fine drought-resistant varieties and the saving of water resources are conducive to the health and sustainable development of cotton crops.

Supplementary data for this article can be found online at https://doi.org/10.1016/j.cj.2020.02.003.

Declaration of competing interest

The authors have declared that no conflict of interest exists.

Acknowledgments

This work was supported by the National Key Research and Development Program of China(2018YFD1000907).

Author contributions

X.Y. and Y.Y. conceived the project and designed the experiments; Q.T. and X.W. conducted the field experiments;B.L., B.H., L.L., X.K., A.S. and J.W. participated in the phenotyping;Z.L.and Y.Y. provided the seed materials;X.Z. offered writing advice; B.L. wrote the manuscript and managed the main data analysis, including statistical analysis, figure and table design;X.Y.revised the manuscript.