Genome-wide association study on root traits under different cultivation patterns in wheat

Background Roots are critical for water and nutrient acquisition, environmental adaptation and yield formation. Here, 196 accessions from Yellow and Huai Winter Wheat Region (YHW) of China were collected for investigating the performance of six root traits under three cultivation patterns—Indoor Hydroponic Culture (IHC), Outdoor Hydroponic Culture (OHC) and Outdoor Pot Culture (OPC) at different growth stages—for three consecutive years. In the same growth period, OPC root traits always varied greatly, followed by OHC and IHC. The correlation coecients between IHC and OPC at stooling stage (SS) were lower (0.016 ~ 0.278) than those between OHC and OPC (0.29 ~ 0.378). Root traits were negatively correlated grain yield (GY), the canonical correlation coecient between root traits and yield was the highest (0.232) at SS. Genome-wide association study (GWAS) was furtherly conducted by a wheat 660K SNP array. It was revealed that 1105 SNP loci were signicantly associated with root traits. A co-localized chromosomal segment regulating total root length (TRL) was detected on chromosome 4A, spanning from 737.85 to 742.00 Mb, under different cultivation patterns at stooling stage. Another co-localization region regulating total root area (TRA) was detected on chromosome 5A, an approximately 6.17 ~ 18.76 Mb region at SS, wintering stage (WT) and jointing stage (JS) under OPC. LD analysis and blast comparison revealed 27 and 31 genes related to root development were found from these two segments, respectively. Among them, TraesCS4A02G493900, TraesCS4A02G494200, TraesCS5A02G021700, TraesCS5A02G021800 and TraesCS5A02G011600 were predicted to be highly expressed in root. SS and WT was stronger than the correlations involving JS and MS. The results indicated that root traits showed stronger correlations between OHC and OPC and weaker correlations between IHC and OPC. However, the signicant correlations among root traits revealed strong coordination under the three cultivation patterns and four growth stages (Table S6). The coecients of correlation between TRL and TRA and between TRA and total root volume (TRV) were the highest among the traits (r > 0.85); the correlation coecients among TRL, TRA, TRV and number of root tips (NRT) were moderately high (r > 0.60), while the correlation between ARD and TRV was weak (r = 0.5). In contrast, ARD was not correlated with TRL, TRA, NRT or root dry weight (RDW), with correlation coecients close to zero or negative (except for MS). in eight pots, with two plants per pot, and two pots with four plants were measured in each stage. To facilitate management, the total 1568 pots, representing all accessions, were distributed among 20 quadrats (length, 150 cm; width, 160 cm), with each quadrat containing 80 pots. A border was planted to protect the plants around the quadrats from marginal effects. The root experiments were conducted at Zhengzhou (34.7°N, 113.6°E), Henan, China.

Due to the polyploidy of wheat and the lack of a reliable reference sequence, research on quantitative traits in hexaploid wheat has lagged behind that in other crops, especially research on root traits. Genome-wide association study (GWAS) is a powerful tool for identifying loci that are signi cantly associated with target traits based on linkage disequilibrium (LD) in natural populations. In recent years, with the increasing availability of genomic sequences and genotyping by sequence data, GWAS has become a rapid and costeffective way to detect and transform markers for marker-assisted breeding [18]. Many types of SNP arrays from functional genes have been successfully developed for the identi cation of QTLs for various traits in natural populations; such arrays include 9K [19], 55K [20], 90K [21,22] and 660K arrays [23]. Yield traits have been reported to be closely related to root traits, and some QTLs for root traits have been found to co-localize with yield traits [11,24].
The root system is underground and di cult to identify in situ. Previous research on wheat roots has mainly involved QTL mapping of seedlings from speci c populations (RIL, NIL and DH populations) by using a small number of markers under hydroponic indoor culture; thus, the molecular basis of many traits remain unclear. In view of this, in the present study, 196 wheat accessions from YHW of China were used as representatives of natural populations, and root traits were investigated under three cultivation patterns at four growth stages (SS, WT, JS and maturation stage (MS)). In addition to investigating yield in the eld, GWAS of the panel was conducted with the wheat 660K SNP array. The aim of this study was to elucidate the relationships between root and yield and explore associated SNPs or candidate genes for root traits. This work not only provides valuable genetic information for genetic improvements to the wheat root system but can also be used to screen favorable genotypes for breeding new lines with better root systems.

Phenotypic variation of root traits in wheat
The variations in root traits among the three cultivation patterns at the four growth stages of the 196 wheat accessions are provided in Table S3. The frequency distributions of the six studied root traits were continuous and distributed normally (Fig. S1). The joint variance analysis of root traits indicated that the effects of genotype (variety) and genotype × environment were highly signi cant (p < 0.01) ( Table 1). These results indicated that the root traits were quantitative traits and controlled by minor polygenes.

Correlations Among Root Traits
Canonical correlation analysis showed that among the three cultivation patterns at stooling stage in wheat, the strongest correlation of root traits was between OHC and OPC (0.515), followed by OHC and IHC (0.421) and then IHC and OPC (0.385) ( Table S5). The correlation coe cients of all root traits except average root diameter (ARD) were highest for IHC and OHC (r > 0.3). The correlations between IHC and OHC were moderately high, with correlation coe cients of approximately 0.2. However, the correlations between IHC and OPC were low, with correlation coe cients close to zero ( Table 2). The canonical correlation analysis showed that the correlation between IHC and OPC was low at every stage. The correlation between OHC and OPC was higher than that between IHC and OPC and highly signi cant at SS (0.515) and WT (0.409). Under OPC, the correlation between SS and WT was stronger than the correlations involving JS and MS. The results indicated that root traits showed stronger correlations between OHC and OPC and weaker correlations between IHC and OPC. However, the signi cant correlations among root traits revealed strong coordination under the three cultivation patterns and four growth stages (Table S6). The coe cients of correlation between TRL and TRA and between TRA and total root volume (TRV) were the highest among the traits (r > 0.85); the correlation coe cients among TRL, TRA, TRV and number of root tips (NRT) were moderately high (r > 0.60), while the correlation between ARD and TRV was weak (r = 0.5). In contrast, ARD was not correlated with TRL, TRA, NRT or root dry weight (RDW), with correlation coe cients close to zero or negative (except for MS).

Correlations Of Root Traits With Yield
Canonical correlation analysis indicated that root traits and yield had the strongest relationship during SS (0.232), followed by JS   Blast search against the Chinese Spring sequence (IWGSC V1.1) on the NCBI website revealed that 55 of the 58 genes were annotated. Functional prediction of orthologous genes in other species was performed; 46 and 18 genes were found to be homologous to those in rice and Arabidopsis, respectively (Table S11).
The gene expression data for the 58 genes in different wheat tissues were downloaded (Table S12) from the WheatExp website. Among them, 29 genes were expressed in all of the tissues root, stem, leaf, ear and grain, and eight genes were expressed in only some tissues; the remaining 21 genes were not expressed. In particular, TraesCS4A02G493900, TraesCS4A02G494200, TraesCS5A02G021700 and TraesCS5A02G021800 showed high expression levels in roots at three plant growth stages, and TraesCS5A02G011600 was speci cally expressed in the roots (Fig. 2). These ve genes may participate in regulating root development.

Discussion
Effects of different cultivation patterns on wheat root development Methods to acquire root system data for crops have always been a focus of root studies. Currently, approaches are mainly divided into two categories: eld detection and indoor detection [25]. Traditional trench digging, soil core sampling [26] and soil column testing [27] in the eld are labor-intensive and time-consuming methods for root phenotyping. Furthermore, traditional trench digging is not suitable for studies with large-population root QTLs or high-throughput analysis. Indoor detection involves growing plants in an incubator, an arti cial climate chamber or a greenhouse to identify root traits. This method is convenient for obtaining the complete root system, although the root system may not be re ective of the root system in a complex eld environment [28]. In this study, comparative analysis of root morphology under three cultivation patterns revealed that root development was least similar between IHC and OPC among the combinations of culture methods. Similarly, in rice, the ability of roots to penetrate a solid wax layer was found to be limited, and therefore, deep root lines were not consistently observed in the eld [29]. Bai et al. [30] found that the correlation between laboratory-based root length screening data and eld-based root depth data was weak and that there was no consistency in root depth across two years. However, in the present study, root development was comparable between OHC and OPC, and the root traits under OHC had a higher broad heritability. Under IHC, the root phenotypes showed little variation; however, the environment of root development in this type of system cannot re ect the complex root environment in the eld. Under OHC, root development is affected by not only genotype but also the environment and the interaction of environment and genotype. Furthermore, under OPC, the soil environment is complex, reducing the effect of genotype on the root system relative to that in hydroponic culture. Considering the integrity and feasibility of root trait collection, OHC can be used to perform multiyear and multisite GWASs in large populations, and it can incorporate genomics information to elucidate the genetic basis of root development. Hence, OHC is a rapid root identi cation method worthy of promotion.
Although the root system is very sensitive to the external environment, correlations between root traits were observed under different cultivation methods. As indicated by their u values, a few wheat varieties had more uniform root development; Luomai 23 had a poor root system, and Yumai 54 had a large root system under all three cultivation patterns. These varieties may contain the main genes responsible for root development and provide breeding materials for wheat root research.

Genetic Basis Of Wheat Root Traits
A total of 1105 SNP loci associated with root system traits were detected. Alignment against the reference wheat genome (Chinese Spring) revealed that seven SNP loci detected in this study were at the same or similar positions as loci related to root system traits detected in previous studies. For example, under OHC, TRL-associated AX-94990677 (2.49 Mb), detected on chromosome 2A, was in the same interval as Xbarc1138_2A detected by previous researchers [31]. Furthermore, RDW-associated XPsr1327_1A on chromosome 1A was close to the locus AX-111044604 (583.36 Mb) detected at SS under OPC in the present study [32]. The GWAS results also showed that some SNP loci that control multiple root traits were detected under all three cultivation patterns or at all stages except the MS. These results are consistent with previous studies on roots in the seedling stage [14,33]. A similar phenomenon for other wheat traits has been detected in previous studies. For example, disease resistance genes Lr26, Yr9 and Pm8 [34] and insect resistance genes are located on the 1B/1R chromosome translocation; these genes enhance yield and help maintain greenness [35].
In the present study, although the correlations of root traits among the four developmental stages were not high, some shared SNPs or

Relationships Between Root Traits And Yield
Previous studies have identi ed some yield QTLs on chromosomes 2B, 2D, 4A, 4B, 5A, 7A, 7B and 7D [19][20][21][22][23]. Furthermore, QTLs of yield traits have been found to overlap with QTLs of root number and TRL in tetraploid wheat [3]. A QTL for TRL and seminal root number was mapped to chromosome 7DS and found to be associated with GY [11,24,36]. In addition, studies of the relationship between the root traits and yield of indoor seedlings revealed that the seminal root number and TRL of wheat were positively correlated with GY [3,36] but negatively correlated or uncorrelated with thousand-kernel weight [25]. Atta et al. [37] found that wheat root traits had weak positive correlations with grain number per spike, spikelet number, water use e ciency and yield. In the present study, root traits were negatively correlated with GY, inconsistent with previous studies. It is possible that in previous studies, the focus was on the indoor seedling stage, where the light is lower than natural light and the cultivation period is short. Therefore, only the seminal roots but no nodal roots were produced. In wheat production, the root system comprises both seminal and nodal roots [3]. During the middle and late growth stages, nodal roots are the main component of the root system, accounting for more than 70% of the root system [21].
Large roots can extend the grain-lling period. Developed roots lead to strong competition for effective assimilation within the plant (limited sources), and the formation of spikes may be restricted [38]. Accordingly, root traits were negatively correlated with GY, and the correlations were signi cant at stooling stage. An excessive root mass has been shown to cause ine cient consumption and reductions in yield [39]. Ma et al. [40] showed that reducing the root mass can improve water use e ciency in the late owering stage of wheat and thus increase the harvest index. What's more, root traits were negatively correlated with GY and that yield was most strongly correlated with root traits at stooling stage. A robust root system in seedlings is conducive to the uptake of soil resources (such as N, P and others) during the early growth stage of plants [9], thereby largely relieving internal competition within the plant.
However, some wheat grows in arid regions that lack water resources, and deep root systems are conducive to the use of deep water. Therefore, the root system at stooling stage may play a key role in wheat yield.

Prediction Of Candidate Genes Related To Wheat Root Development
Through GWAS and haplotype analysis, ve out of 59 candidate genes regulating root development were found on chromosomes 4A and 5A. Functional annotation identi ed several similar genes that might affect root development. For example, two candidate genes on chromosome 5A, TraesCS5A02G022100 and TraesCS5A02G022300, encode zinc nger domain proteins, these genes potentially play important roles in plant growth regulation and development, signaling and responses to abiotic stress [17,41]. In wheat roots, the overexpression of TaZFP34, a gene encoding a zinc nger protein, can inhibit aboveground growth and increase the root-to-shoot ratio of plants [15]. Functional annotation showed that a root development-related candidate gene, TraesCS4A02G484800, is homologous to OSINV3 and AT1G12240. The glycoside hydrolase family 32 encoded by the AT1G12240 gene in Arabidopsis is involved in the decomposition and transformation of sucrose, which can promote hypocotyl elongation [42]. Similarly, OSINV3 in rice is a gene encoding glycoside hydrolase family proteins. Electron microscopy revealed that the panicles of OsINV3 mutant plants had small and few cells on the inner and outer membranes, indicating that OsINV3 plays an important role in cell expansion [43]. However, the function of TraesCS4A02G484800 in wheat has not been reported. This gene might regulate the elongation and formation of root cells through glycoside hydrolase, thereby affecting the development of the root system.
Expression data of ve genes, TraesCS4A02G493900, TraesCS4A02G494200, TraesCS5A02G011600 TraesCS5A02G021700 and TraesCS5A02G021800, revealed high expression levels in roots at three plant growth stages. Gene annotation revealed that the threonine dehydratase encoded by TraesCS5A02G021800 is the rst enzyme in the isoleucine biosynthetic pathway, and this gene's homologous gene, OMR1, similarly encodes a threonine dehydratase. Yu et al. [44] studied an Arabidopsis mutant with low isoleucine biosynthesis that has defects in cell proliferation and cell expansion during root development and found that the mutant root phenotype can be completely restored when externally supplementing isoleucine or introducing the OMR1 gene into the mutant. Thus, OMR1 might be involved in cell differentiation and formation during root development.

Conclusions
Within a given growth period, among the cultivation systems, OPC yielded the greatest RDW, OHC yielded rapid root elongation, and IHC yielded minimal variation in root morphological traits. The correlation coe cients of root traits were high between OHC and OPC, and root traits under OHC had higher broad heritability than those under the other systems; thus, OHC can be considered a rapid root identi cation method worthy of promotion. Root traits were negatively correlated with GY; the correlations were signi cant at SS. Because of the close relationships among root traits, the SNP loci detected by GWAS were found to be pleiotropic. Five candidate genes regulating root development were found on chromosomes 4A and 5A and might participate in regulating root development. Among these candidate genes, TraesCS4A02G493900, TraesCS4A02G494200, TraesCS5A02G021700 and TraesCS5A02G021800 showed high expression levels in roots at three plant growth stages, and TraesCS5A02G011600 was exclusively expressed in roots. The proteins of these genes were identi ed as mainly involved in carbon metabolism, nitrogen metabolism, signal transduction, stress responses and DNA synthesis. Among the 196 wheat accessions, Luomai 23 had a poor root system, and Yumai 54 had a large root system under all three cultivation patterns.

Plant materials
The 196 wheat accessions used in this study included newly bred varieties, elite varieties and historical varieties from YHW. The accessions comprised 158 accessions from Henan Province, nine from Shaanxi, eight from Jiangsu, 10 from Shandong, four from Hebei, four from Beijing, one from Anhui, one from Shanxi and one land variety from Sichuan (Chinese Spring) (Table S1). Seeds were provided by Collaborative Innovation Center of Henan Grain Crops, Henan Agricultural University.

Experimental design
Hydroponic culture Individuals were grown in an IHC system in a greenhouse, and the experiment was repeated three times. The greenhouse conditions included a 16 h/8 h light/dark photoperiod at 20 °C/16 °C and a light intensity of 1000 μmol•m -2 •s -1 (Fig. 3a). Plants grown in an OHC system were cultivated for three consecutive years from 20 th October to 20 th November in 2016, 2017 and 2018. The seeds were germinated in Petri dishes and grown on germination substrate for seven days with sterile water. Then, the germinated seeds with residual endosperm removed were transferred to plastic pots containing 20 L nutrient solution (Fig. 3b). The composition of the nutrient solution was as described by Ren et al. [14]. The nutrient solution was refreshed every three days, with the pH maintained at

Soil-lled pot experiments in the eld
Pots were placed in the eld in 2017~2018 and 2018~2019. Each pot (height, 10 cm; diameter, 12 cm) was lled with 2 kg of sieved tillage soil, and the top edge of each pot was on the same plane as the ground surface (Fig. 3c). The plants were regularly watered to 70~80% of eld capacity at 15-day intervals after sowing. Other eld management followed local agronomic practices. Uniform germinating seeds were placed in each pot. Ten days after sowing, each plot was thinned to two seedlings. Due to labor limitations, each wheat accession was planted in eight pots, with two plants per pot, and two pots with four plants were measured in each stage.
To facilitate management, the total 1568 pots, representing all accessions, were distributed among 20 quadrats (length, 150 cm; width, 160 cm), with each quadrat containing 80 pots. A border was planted to protect the plants around the quadrats from marginal effects.

Phenotyping of the 196 wheat accessions
For the seedlings in the hydroponic culture system, 30 days after sowing, the roots were rinsed with sterile water, and root traits were recorded. For the plants in the soil-lled pots, 30 days (SS), 60 days (WT), 150 days (JS) and 220 days (MS) after sowing, roots were removed from the soil and rinsed with fresh water before the root traits were measured.
The WinRHIZO system (Canada Regent Instruments, LA6400XL) was used to scan and analyze root morphology and measure ARD, TRL, TRV, TRA and NRT. After scanning, the roots were oven-dried at 80 °C to determine RDW using an analytical balance (Germany SARTORIUS, QUINTIX224-ICN) ( To facilitate a comprehensive evaluation, each root trait was converted into u values by the standardized normal distribution method using the equation where, for a given trait, is the average value for the different environments and ` and s are the arithmetic mean and standard deviation, respectively, of the 196 accessions. Then, the average u value was calculated for each accession at each stage. For each trait, a higher average u value suggested a larger root system.

GWAS and prediction of candidate genes
Genetic diversity, population structure and LD analyses of panel data were conducted according to Chen et al. [23]. Previous reports indicated that the 196 accessions could be divided into two subpopulations, which were largely consistent with the pedigrees and geographic origins. The optimal model for traits was taken into account when the actual -log 10 (p) value was found to be closest to the expected value of -log 10 (p). A mixed linear model correcting for both the Q-matrix and K-matrix (PCA + Q) was used to reduce population structure and relative kinship errors [46]. The best model was used to analyze associations among phenotypic data and BLUE values for each trait in Tassel v5.0. Here, the threshold for the p-value of -log 10 (p-value) was determined according to a uniform suggestive genome-wide signi cance threshold (-log 10 (p) ≥ 3.5) [47]. Manhattan and Q-Q plots were generated in R using the CMplot package. Haploview 4.2 software was used to analyze local LD decay and haplotype block structure. The identi cation of candidate genes for root system traits and the analysis of expression patterns were performed according to Zheng et al. [36].

Consent for publication
Not applicable.

Availability of data and materials
Not applicable. The datasets used and/or analyzed during the current study are available from the corresponding author on request.

Competing interests
The authors declare that they have no competing interest.

Funding
The work was supported National Key Research and Development Program "Science and Technology Innovation of High Grain Production E ciency" of China (2018YFD0300701). The supporters did not play any role in the design analysis, or interpretation of this study and the relevant data. Shulin Chen was supported with funds from China Scholarship Council (CSC).
Author's contributions FX, KZ and DH conceived the topic. SC and KZ provided gene chip, YH, LZ, MS and YH collected the phenotype regulation data. SZ, XY and XC performed the data analysis. FX wrote the manuscript with edits from other co-authors. All authors read and approved the