Association analysis of physiological traits in spring barley (Hordeum vulgare L.) under water‐deficit conditions

Abstract In the present study, 148 commercial barley cultivars were assessed by 14 AFLP primer combinations and 32 SSRs primer pairs. Population structure, linkage disequilibrium, and genomic regions associated with physiological traits under drought stress were investigated. The phenotypic results showed a high level of diversity between studied cultivars. The studied barley cultivars were divided into two subgroups. Linkage disequilibrium analysis revealed that r 2 values among all possible marker pairs have an average value of 0.0178. The mixed linear model procedure showed that totally, 207 loci had a significant association with investigated traits. 120 QTLs out of 207 were detected for traits under normal conditions, and 90 QTLs were detected for traits under drought stress conditions. Identified QTLs after validation and transferring to SCAR markers in the case of AFLPs can be used to develop MAS strategies for barley breeding programs. Some common markers were identified for a particular trait or some traits across normal and drought stress conditions. These markers show low interaction with environmental conditions (stable markers); therefore, selection by them for a trait under normal conditions will improve the trait value under stress conditions, too.

Genetic tolerance to drought stress is complex and understanding the genotype, environment and genotype by environment interaction effects are important and vital in designing drought resistant breeding programs (Bänziger & Araus, 2007;Yue et al., 2006).
Thanks to the development of molecular markers technology, it is now possible to decipher the genetics of complex traits and identify functional genes or markers closely linked to their controlling genes (Pinto et al., 2010). Genome-wide association study (GWAS) is a suitable method for identifying molecular markers associated with the genomic region involved in controlling complex traits such as drought tolerance (Kazemi et al., 2020;Yan et al., 2011). Association analysis is a method in the search for detecting relationships between phenotypic diversity and genetic polymorphisms in natural populations (Remington et al., 2001;Thornsberry et al., 2001;Wilson et al., 2004).
As association analysis is based on LD, then the investigation of LD extent in population group is important for successful association analysis (Abdurakhmonov et al., 2008;Sorkheh et al., 2008). In natural populations, LD can be occurred by different factors such as physical linkage, migration, individual relationships, and population structure. Except for physical linkage, the other factors produce LD between loci across different chromosomes, something that is not important in terms of plant breeding. Therefore, population structure and individual relationships are factors that affect the resolution of association analysis and should be considered to avoid false-positive associations (Bradbury et al., 2007;Zhu et al., 2008).
The presence of population structure can be detected by some statistical approaches such as model-based clustering (Pritchard et al., 2000) and principal component analysis. In the mixed linear model (MLM), Q-matrix from population structure analysis and the relatedness among individuals are included in the model as covariates to overcome spurious associations between markers and traits . Association studies in barley have been concentrated on flowering time (Stracke et al., 2009), yield (Gawenda et al., 2015;Kraakman et al., 2004), disease resistance (Massman et al., 2011;Roy et al., 2010), drought tolerance (Varshney et al., 2012;Wójcik-Jagła et al., 2018), salinity tolerance (Fan et al., 2016), and freezing tolerance (Rapacz et al., 2010;Visioni et al., 2013).
Association analysis can be also conducted based on the candidate genes approach. However, in most studies, a genome-wide association approach has been used. There are some studies about association mapping based on the analysis of candidate genes, including the Dwarf8  and phytoene synthase locus in maize (Palaisa et al., 2003), flowering time genes in barley (Stracke et al., 2009), the PsyI-AI locus in wheat (Singh et al., 2009), and rhg-1 gene in soybean (Li et al., 2009).
The objectives of the present work were to identify population structure in a collection of Iranian barley germplasm genotypes and investigate the association of AFLP and SSR markers with physiological traits in the plant under drought stress conditions. Physiological traits are important in selecting drought tolerance genotypes in plants including barley (Del Pozo et al., 2012), but most of them have quantitative behavior and are controlled by many genes (Mora-Poblete et al., 2015). Finding QTLs associated with quantitative traits can help breeders to develop a MAS strategy for their breeding programs.

| Experimental design
Genotypes were evaluated using alpha-lattice design with two replications in well-watered (irrigation at 90% FC) and drought stress (irrigation at 40% FC) conditions during two successive years. Each replicate includes 11 incomplete blocks with 14 plots, and each replication contains 148 barley cultivars and six local barley varieties (Local, 5-White cluster salinity, Nomar, Zahak, NP-90-113 and Nimroz). First irrigation was performed for germination, next irrigation was done after the soil moisture reached 90% of the field capacity for well-watered treatment and drought stress, and irrigation was done after reaching humidity to 40% of field capacity. Moisture measurement was done by Time-Domain Reflectometer (TDR) method. Physiological traits including canopy temperature (CT), relative water content (RWC), proline content (PRO), water-soluble carbohydrate concentration (WSC), relative chlorophyll content (RCC), the maximum potential quantum efficiency of PSII photochemistry (Fv/Fm) (PSII), chlorophyll a (Chl a), chlorophyll b (Chl b), the chlorophyll a/b ratio (Chl a/b), carotenoid (Car), catalase (CAT), guaiacol peroxidase (POD), and ascorbate peroxidase (APOX) were evaluated in both years in both irrigation conditions.

| Genotyping and population structure
Genotyping of individuals was performed using 14 AFLP primer combinations and 11 SSRs pairs primers (Kraakman et al., 2004(Kraakman et al., , 2006 following the method described by Kraakman et al. (2006). Also, we used 21 new SSR markers from the previously reported map (Aghnoum et al., 2010). Totally, 407 polymorphic markers were used in the present study.

| Data analysis
Variance components of phenotypic data were calculated using GenStat version 15 (Payne et al., 2011). Correlation among all studied traits was calculated by SPSS version 24 and heritabilities (Family mean basis) were estimated using SAS version 9.0 software.
Best Linear Unbiased Estimates (BLUEs) of phenotypic data based on G × E variances were used in association analysis (Haseneyer et al., 2010). Estimation of the population structure was performed with the Bayesian clustering model (Pritchard et al., 2000) using Structure 2.3. Burn-in period length was 100,000, and Markov Chain Monte Carlo (MCMC) replications was 100,000. ΔK index was determined for obtaining the optimal subpopulations number, Q-matrix was derived (Falush et al., 2003;Kraakman et al., 2004;Pasam et al., 2012). The neighbor-joining dendrogram (NJ) was performed based on the genetic distance matrix using Tassel 5. The linkage disequilibrium (LD) was estimated with Haploview 4.01 software (Barrett et al., 2005). Association analysis was performed using a mixed linear model (MLM) considering Q-and K-matrices  as covariates in the model in TASSEL software. For identifying significant marker-trait associations, the threshold pvalue of .03 was estimated and used for all traits according to Chan et al. (2010) and Pasam et al. (2012).

| Phenotypic variation
High levels of variation were observed among genotypes for studied traits according to ANOVA. The variance analysis showed that the effect of the environment was significant on some studied traits such as proline content (PRO), water-soluble carbohydrate concentration (WSC), catalase (CAT), guaiacol peroxidase (POD), and ascorbate peroxidase (APOX). The effect of genotype (G), genotype × year (G × Y), genotype × environment (G × E), and genotype × environment ×year (G × E×Y) were significant on all studied traits ( Table 1)  (chl a/b), chlorophyll b (chl b), relative chlorophyll content (RCC), and canopy temperature (CT), the heritability was medium. Whereas for relative water content (RWC), proline content (PRO), water-soluble carbohydrate concentration (WSC), PSII, and chlorophyll a + b low heritability was observed (Table 2). A positive correlation was observed among chl a, chl b, chl (a + b), and CAR in both water treatment conditions. Proline content was showed a high correlation with chl a, chl b, chl (a + b), and CAR. In well-watered conditions, chl a, chl b, chl (a + b), and CAR had a significant and positive correlation with APOX but in drought states, they had a significant and negative correlation with APOX. In drought state, a significant and negative correlation and significant and positive correlation was observed between POD and CT and between POD and CAT, respectively. In well-watered conditions, RWC was not significantly correlated with other traits except POD (Table 3).

| Population structure
The genetic fingerprint of 148 barley genotypes was investigated

| Linkage disequilibrium and association mapping
Linkage disequilibrium analysis was performed using molecular markers on the association panel. The r 2 values among all possible marker pairs showed an average value of 0.0178 ( Figure 3). A mixed-linear model (MLM) method using Q-and K-matrices as covariates were conducted for identifying molecular markers associated with genes controlling physiological traits under well-watered and drought stress conditions. Results showed the significant association of 207 AFLP and SSR markers with genomic region controlling the fourteen studied traits. In this study, 22 molecular markers were found to be significantly associated with CT from which 11 markers were associated with the trait in well-watered conditions and the other was associated with the trait in drought stress conditions. From marker associated with the trait in well-watered conditions, the location of two markers was on linkage group 2H (2016 and 2017) and three were on linkage groups 1H, 6H, and 4H. The location of others was unknown. In drought stress conditions, the location of all identified markers was unknown except to two markers located on linkage groups 2H and 1H. (Table 4).
Seventeen markers were found to be significantly associated with RWC from which 10 markers were associated with the trait in well-watered conditions and the rest 7 markers were associated with the trait in drought stress conditions. The location of 11 out of 17 identified markers was unknown. The location of three markers, identified for the trait under well-watered conditions, were on the 2H linkage group but in different parts, and the location of two markers was on linkage group 7H in the same region (140.172). The location of one marker identified for the trait under drought stress conditions was on linkage group 5H (Table 4).
Twenty-two markers were found to be significantly associated  (Table 5).
Fourteen markers were found to be significantly associated with RCC, from which 10 markers linked with gene controlling trait under normal conditions and four were associated with genes controlling traits under drought stress conditions. From identified markers for the trait under normal state, the location of two markers was on linkage group 2H but in different regions and the location of three markers was on linkage groups 3H, 4H, and 1H.
The location of five out of 10 markers was unknown. Concerning identified markers for the trait under drought stress conditions, the location of two markers was on linkage group 1H but in different positions and the location of two markers was on linkage groups 3H and 4H (Table 6).
Concerning to PSII trait, a total of 14 markers were identified from which four markers were associated with trait under normal conditions and the rest 10 markers associated with trait under drought stress conditions. From identified markers for the trait under normal conditions, the location of all markers was clear and known; one was on linkage group 2H and three were on linkage group 5H but in different positions. From identified markers for the trait under drought stress conditions, the location of all markers except one was unknown (Table 6).
Twelve markers showed significant associations with chl a, from which seven markers associated with trait under normal conditions and the rest five markers associated with trait under drought stress conditions. The location of five out of seven markers was known, TA B L E 2 Descriptive statistics and heritability (h 2 ) for studied physiological traits in spring barley (Hordeum vulgare L.) under wellwatered (W) and drought stress (D) conditions across two years  (Table 7).
Concerning chl b, eight markers were identified as significantly associated with genes controlling trait under normal conditions. The locations of five markers were on linkage groups 5H, 4H, 2H, 6H, and 3H. The locations of the three markers were unknown. Six markers associated with genes controlling trait under drought stress conditions. The location of one marker was on linkage group 7H, and the location of the other was unknown (Table 7).
Totally 11 DNA markers were identified for chl (a + b), from which seven markers associated with trait under normal conditions and the rest four markers associated with trait under drought stress conditions. The location of five markers identified for the trait under normal conditions was on linkage groups 3H, 5H, 4H, 6H, and 2H and the locations of two markers were unknown. The location of markers identified for the trait under drought stress conditions was unknown (Table 8).
For chl a/b, 12 DNA markers were found to be significantly  (Table 8).
Nineteen markers were found to be significantly associated with Carotene content (Car); among these 11 markers were associated with genes controlling trait under normal conditions and the rest eight markers associated with genes controlling traits under drought stress conditions. The location of six markers associated with trait under normal conditions was clear and know, on linkage groups 2H, 3H, 4H, and 5H. Indeed, two markers were located on linkage group 3H but in the same location (151.031 cm), two markers located on linkage group 4H and two markers located on linkage groups 5H and 2H. Concerning drought stress conditions, except for one QTL located on linkage group 4H, the location of others was unknown (Table 8).
For CAT, totally, 13 markers were found to be significantly associated with genes controlling trait. Seven out of 13 were found for the trait under normal conditions. The location of one marker is known; on linkage group 2H, the locations of six other markers were unknown. Six markers from 13 identified markers were associated with trait under drought stress conditions. The locations of three were known; on linkage groups, 3H and 5H and the locations of three others were unknown (Table 9).
For POD, 13 DNA markers were identified. The location of all of them was unknown except to two markers on linkage groups 7H and 6H, identified for the trait under normal conditions (well-watered treatment) and one on linkage group 5H, identified for the trait under drought stress conditions (Table 9).   (Table 9). According to Stansfield (1991), if the heritability of the trait is more than 0.5, the heritability is considered high, if it is between 0.2 to 0.5, the heritability is considered medium and in this case, if it is lower than 0.2, the heritability is considered low. Then, in this study catalase (CAT), guaiacol peroxidase (POD) and ascorbate peroxidase (APOX) had high heritability.

| D ISCUSS I ON
In population structure analysis, the studied association panel was subdivided into 2 subpopulations. Population structure affects the efficiency of association analysis (Sorkheh et al., 2008). If it exists and is not considered in the association model, probability some false-positive markers will be identified that are not important because of marker-assisted selection (Pritchard et al., 2000). Barley has a wide level of population structure because of its two-rowed versus six-rowed cultivars or spring versus winter barley cultivars (Pasam et al., 2012). Because of the complex population structure, there could be a higher challenge in GWAS than QTL mapping by producing false-positive markers (Myles et al., 2009;Pasam et al., 2012). The mixed linear model can overcome these false associations . LD is another critical factor that affects the resolution of association analysis (Remington et al., 2001). The In this study, a total of 207 DNA markers were identified for studied physiological traits under normal (well-watered) and drought stress conditions. A few studies presented reports about the identification of QTLs for water-soluble carbohydrates (Diab et al., 2004;Teulat et al., 2001) and proline content (Fan et al., 2015;Sayed et al., 2012) in barley. Teulat et al. (2001)    common DNA markers had also shown significant correlations at phenotypic levels. For instance, based on correlation analysis of phenotypic data, Chl a, Chl b, and Chl (a + b) had a significant correlation and GWAS analysis showed that E42M48-380 was common for these traits under normal conditions. The common markers between some of the traits can be due to linkage or pleiotropic effects. The common markers are useful because they lead to an increase in the efficiency of marker-assisted selection. Some common markers were identified for a particular trait or some traits across normal and drought stress conditions. For example,

CO N FLI C T O F I NTE R E S T
The authors declare that they have no competing interests.

DATA AVA I L A B I L I T Y S TAT E M E N T
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.