Identification of manganese efficiency candidate genes in winter barley (Hordeum vulgare) using genome wide association mapping

Background Manganese (Mn) has several essential functions in plants, including a role as cofactor in the oxygen evolving complex (OEC) of photosystem II (PSII). Manganese deficiency is a major plant nutritional disorder in winter cereals resulting in significant yield reductions and winter kill in more severe cases. Among the winter cereals, genotypes of winter barley are known to differ considerably in tolerance to Mn deficiency, but the genes controlling the Mn deficiency trait remains elusive. Results Experiments were conducted using 248 barley varieties, cultivated in six distinct environments prone to induce Mn deficiency. High-throughput phenotyping for Mn deficiency was performed by chlorophyll a (Chl a) fluorescence analysis to quantify the quantum yield efficiency of PSII. High-throughput phenotyping in combination with ICP-OES based multi-element analyses allowed detection of marker-trait associations by genome wide association (GWA) mapping. Several key candidate genes were identified, including PSII subunit proteins, germin like proteins and Mn superoxide dismutase. The putative roles of the encoded proteins in Mn dependent metabolic processes are discussed. Conclusions Fifty-four candidate genes were identified by Chl a fluorescence phenotyping and association genetics. Tolerance of plants to Mn deficiency, which is referred to as Mn efficiency, appeared to be a complex trait involving many genes. Moreover, the trait appeared to be highly dependent on the environmental conditions in field. This study provides the basis for an improved understanding of the parameters influencing Mn efficiency and is valuable in future plant breeding aiming at producing new varieties with improved tolerance to cultivation in soil prone to induce Mn deficiency. Electronic supplementary material The online version of this article (doi:10.1186/s12864-016-3129-9) contains supplementary material, which is available to authorized users.


Background
Deficiency of the essential micronutrient manganese (Mn) remains an unsolved problem that has a severe impact on crop production worldwide [1][2][3][4]. In addition to substantial yield losses, suboptimal use of nitrogen, phosphorus and water are marked side-effects of Mn deficiency. It is prevalent in areas with well aerated and high pH soils containing free carbonates and with high organic matter content [5]. Manganese deficiency often occurs as a latent disorder with no visual symptoms making it difficult to diagnose and follow up with timely Mn remediation [6]. Mn deficient plants have a decreased lignin content [7] and are therefore more prone to be infected by pathogens [8,9] and have marked decreased winter hardiness [6,10]. Application of soluble Mn-fertilizers to the soil is an ineffective way to correct Mn-deficiency, as the added Mn is instantaneously made unavailable by oxidation to MnO 2 , due to soil chemical conditions [5]. Foliar applications of soluble manganous sulphate are more effective, but this is time consuming, expensive and often impractical for farmers cultivating marginal lands [11]. A second way for farmers to fight Mn deficiency induced yield losses is by deploying plant varieties with an improved tolerance to growth in soils with low Mn availability, defined as Mn efficient varieties [3,12]. Improving nutrient efficiencies by exploiting genetic diversity in plants and strategies to implement such traits into crop breeding have previously been suggested to improve plant productivity [13][14][15][16].
Previous studies on barley varieties have identified various phenotypes in terms of tolerance to low Mn availability in soil [3,17], implying a genetic control of the trait. However, the genetic mechanisms involved in the ability of plants to cope with low amounts of plantavailable Mn have not yet been clarified. Several physiological mechanisms have been suggested to be involved in Mn efficiency in barley. For instance, it has been shown in barley that the Mn efficient variety Vanessa has a four-fold higher Mn uptake capacity compared with the Mn inefficient variety Antonia when exposed to sub-nanomolar Mn concentrations [18]. A follow-up study has suggested that this difference in uptake capacity is caused by different expression levels of the Mn transporter HvIRT1 [19]. However, Mn uptake and acropetal translocation involve many different transport pathways and their role in controlling Mn efficiency remains unknown [20]. It has also been proposed that chloroplasts are a main target for Mn deficiency and that Mn efficiency in barley is significantly influenced by processes linked to the stability and photochemical efficiency of the photosynthetic apparatus [21]. The photosynthetic apparatus seems more unstable when exposed to Mn limitations, with the amount of PsbA (D1) protein being reduced in the Mn inefficient variety Antonia compared to the Mn efficient variety Vanessa [22]. In addition, the ability to perform state transitions is only significantly decreased in the Mn inefficient varieties [21]. Furthermore, it has recently been suggested that exudation of enzymes to the rhizosphere (phytases dissolving organic chelated Mn) is involved in the superior Mn efficiency of certain ancient barley landraces [23].
Manganese activates more than 35 enzymes in plant metabolism, including processes such as the activation of the enzyme phenylalanine ammonia-lyase (PAL) in the shikimic pathway [7], Golgi localized glycosyl transferases [24,25], decarboxylases and dehydrogenases in the tricarboxylic acid cycle [5]. A few enzymes and processes have an irreplaceable requirement for Mn in plants: i) oxalate oxidase (OxO) resulting in hydrogen peroxide production involved in pathogen attack [26]; ii) Mn superoxide dismutase (Mn-SOD) as a key enzyme involved in scavenging of reactive oxygen species in mitochondria and peroxisomes [27,28]; iii) the oxidation of water (Hill-reaction) occurring in the oxygen-evolving complex (OEC) of photosystem II (PSII). Chl a fluorescence measurements to diagnose Mn deficiency have proven to be a powerful research tool for measuring the severity of Mn deficiency and the impact of different Mn fertilizers [6,10]. Chl a fluorescence is an in-vivo non-invasive and nondestructive technique enabling quantification of photosynthetic efficiency such as quantum yield efficiency of PSII (F V /F M ) [29]. This method can be extended to a high-throughput phenotyping method allowing for screening a large collection of cultivars.
In addition to phenotyping methods, the tremendous progresses made in genotyping major crop plants have enabled the search for QTLs and the identification of the underlying genes [30,31]. However, very little work has so far been undertaken on the genetic dissection of the quantitative traits controlling the adaptive response of crops to abiotic stress, including micronutrient deficiencies. QTL mapping and GWA have previously been shown to be promising methods to identify and characterize loci for nutrient efficiency [13,[32][33][34][35]. Initial genetic studies using yield improvements in response to Mn fertilization as a quantitative trait have suggested that Mn efficiency might be controlled by a single locus [36,37]. Several RFLP markers have been identified on barley chromosome 4HS linked to the Mel1 locus for Mn efficiency. The role of Mel1 has been confirmed in field trials and glasshouse experiments and included in breeding programs in South Australia [38].
Barley is the fourth most important cereal crop in the world [39] and it is a widely geographically adapted and stress tolerant crop. In addition, co-localization of barley genetic loci within other cereal species (synteny) is available [40,41], and the major progress achieved by sequencing the barley genome [42,43] and its diploid nature, makes barley a versatile crop for genetic studies. Together with high-throughput genotyping single nucleotide polymorphism (SNP) platforms combined with fast phenotyping methods, GWA is a suitable tool to perform genetic studies of many traits and identified new genes.
The objectives of the current study were: to induce Mn deficiency in a collection of 248 European winter barley varieties cultivated under field and greenhouse conditions, to determine the variation in Mn efficiency by Chl a fluorescence and leaf tissue Mn concentrations by ICP-OES, and subsequently to carry out a GWA in order to provide a set of SNPs associated with the trait, followed by the identification of candidate genes involved in Mn efficiency.

Experimental design Plant material
The population consisted of 248 winter barley accessions (Hordeum Vulgare) collected across Europe (Additional file 1). The countries of origin most represented by the population were Denmark, France, Germany, Italy and the United Kingdom. A first set of 111 varieties came directly from the ExBarDiv collection (Genomics-assisted analysis and exploitation of barley diversity, ERA-PG funded project), with commercial cultivars released over the last 60 years. The collection was supplemented by 18 double-haploid lines from Sejet Plant Breeding (Denmark) and 27 varieties from CREA (Italy). Finally 92 varieties from European plant breeding companies were added. The population contained 139 two-row type and 109 six-row type varieties.

Greenhouse experiment
Plants were cultivated under greenhouse conditions in the autumn 2013. In order to induce Mn deficient conditions, soil was collected from Sweden (55.58°N, 14.05°E, Skåne area) that is known to induce Mn deficiency in winter crops (unpublished data). Mn deficiency has been observed in uncompact and loose soil, therefore the soil was mixed with fine and coarse Leca® (Light Expanded Clay Aggregate) and perlite. In terms of volume, the proportions were 45/25/15/15 % respectively of soil, perlite, fine Leca® (2-6 mm) and coarse Leca® (6-10 mm). Natural light was used and temperatures were maintained between 7 and 13°C. The whole collection was sown in 2 L pots containing three plants of the same variety. A randomized complete block design was used with four replicates of each variety. Pots were watered every two weeks directly from the bottom of the pots by capillarity for a period of 10 min before the water was removed.

Field experiments
Experiments were carried out in 2011, 2012 and 2013 (  (Table 1). Chl a fluorescence is affected by chilling and freezing temperatures [44], thus screening was performed at temperature above 5°C. For each line in the field or pot in the greenhouse, three plants were measured using the youngest fully emerged leaves. After 25 min of dark-adaptation using Hansatech leaf clips, fluorescence measurements were recorded for 2 s after illumination with a saturating light pulse of 3000 μmol photons m −2 s −1 on the adaxial leaf surface. Fluorescence transients (Fig. 1) and JIP-test parameters [45] were extracted using PEA Plus Software (v1.10). The JIP-test parameters used for the study are summarized ( Table 2). The maximum quantum efficiency of PSII (F V /F M ) was used as the main stress indicator [46] as it has been demonstrated to be sensitive for Mn deficiency diagnosis [3,6,21]. However, Mn deficiency leads also to other marked changes in the transients [21]. Therefore, besides F V /F M , also the parameters V I , V J , V K and Area were analysed as the main parameters for association genetic analysis (Table 2). A full Mn-containing design as control was not realized. Indeed in regards to the PSII and the quantum yield efficiency parameter measured (F V /F M ), no diversity could be observed. Under those conditions, the quantum yield efficiency of PSII will reach its maximum with a value of F V /F M around 0.82. Instead of such Mn-containing design, control pots (or lines) were sprayed with a solution of manganese sulphate (MnSO 4 )

Mn concentration in leaf tissue
In LJ13, only the controls samples were harvested and decontaminated following the method described by NIST 1515 Apple leaf certified reference material (National Institute of Standards and Technology, Gaithersburg, Maryland, USA) was analysed together with the samples to evaluate the accuracy and precision of the analysis. Data were accepted when the limit of quantification (LOQ) was exceeded and the accuracy was within ±10 % of the certified level and the coefficient of variation (CV%) among replicates was below 5 %. Four technical replicates were made for all samples and seven for the reference material. The contents of 13 elements were determined and expressed in μg.g −1 of leaf dry weight.

Phenotypic analysis
All phenotypic analyses were carried out using R version 3.1.0 [47]. To estimate the trait value across environments and replicates, estimates of the variance components were calculated using the general linear mixed model. Best linear unbiased predictors (BLUP) for each variety were calculated from a two-dimensional spatial mixed-model with measurement error using model from ASReml-R package [48,49]. BLUPs are reported as being suitable for plant breeding trials [50,51], moreover twodimensional spatial analysis allows taking into account variation in Mn efficiency amongst field and greenhouse experiments. BLUPs were calculated within environments using the formula: where w() indicates within environment, y ij is the response of the i-th variety for the j-th replicated-measurement, z is the intercept, g w(i) is the i-th variety effect, r w(j) is the j-th replicated measurement effect, v w(ij) is the measurement error of the ij-th variety-replicated measurements. Finally, e w(ij) is the residual error which fits separable first order where σ g 2 is the genetic variance, σ r 2 is the variance of replicated measurements,σ 2 υ is the variance of measurement error and σ e 2 is the residual variance.

Genetic analyses Collection genotyping
The whole population of winter barley varieties was genotyped with SNP markers by TraitGenetics GmbH (Gatersleben, Germany). Samples were genotyped using the Illumina® iSelect 9 k barley Infinium chip [52]. Allele calling was performed at TraitGenetics using the company's own cluster file based on a diversity panel. Intra-chromosomal positions used for the GWA were based on the physical map of barley [42] in order to locate the associated SNP on the genome. After pre-processing with minor allele frequencies (MAF) < 0.01 and with a call rate > 0.75, a final set of 5,706 SNPs was retained in the study, including 4,761 SNPs with known positions on the reference map [42].

Population structure calculation
The software package STRUCTURE [53] was used to infer the population structure. This approach uses multi-locus genotypic data to assign individuals to groups (k) without prior knowledge of population genetic relationship under Hardy-Weinberg equilibrium assumption. The calculation was done using the ParallelStructure R package [54]. STRUCTURE runs were distributed over 22 cores on a server to speed up the calculations. To determine the number of subpopulation (k), a first run was done, with a burn-in period of 25000 and 250000 Monte Carlo Markov Chain (MCMC) iterations, for k ranged from 1 to 10 with 15 independent replicates for each k. The admixture model with correlated allele frequencies was applied. The optimum number of subpopulation estimation was based on the ad hoc statistic (ΔK) [55] implemented in Structure Harvester website [56], based the ad hoc statistic (ΔK). The number of optimum subpopulation k was set to k = 4, therefore a new run was used to assign the genotypes to each subpopulation cluster with a setup with burn-in period of 100000, 500000 MCMC and 40 replicates for the estimate k. Results of independent runs for the same value of k were summarized using the CLUMPP software [57]. Estimation of membership coefficient was calculated with the Greedy permutation algorithm option with 100000 random input order repeats and plotted using Distruct software [58].

GWA model
The GAPIT R package was used for GWA [59]. The final GWA was performed using a modified mixed linear model taking. It takes into account multiple levels of relatedness and population matrix as cofactor [60,61]. The pairwise relatedness matrix also called the genomic relationship or Kinship matrix was used to correct for relatedness in the GWA model and calculated using the Loiselle method [62]. Its dimension was (n x n), where n is the number of individuals. The developed model follows: where Y is the phenotypic response vector, X is the molecular marker matrix, β is the vector of fixed effect for the marker to be estimated, Q is the posterior probabilities matrix belonging to each population obtained for k = 4, υ is the vector of fixed effect for population structure, Z is the Kinship matrix, u is the vector of random effect for co-ancestry, and e is the vector of residuals. Only SNP markers with a P-value < 0.001 are presented in the results.

Bioinformatics on candidate genes
SNPs associated with the Chl a fluorescence based parameters were blasted in the BARLEYMAP database [63]. The data-base refers to the barley physical map [42] and specify gene annotations upstream and downstream of the SNP in question. The blast window was extended to half the extent of linkage disequilibrium (LD) ( Table 3) at both QTL region extremities, respectively and to each chromosome LD. In order to calculate LD extent, pairwise LD between loci was calculated in TASSEL version 5.0 [64] and the LD extent was implemented using the method described by Breseghello et al. [65].

Selection of environments
Six different environments were used in the study to screen the 248 varieties of the winter barley collection. Mn deficiency induction cannot be easily controlled in fields and soil pots. Environmental effects such as temperature, soil humidity and redox states, all have major impacts on Mn plant availability and consequently on Mn deficiency development [66]. Consequently genotype-by-environment interactions were expected. Therefore to study these interactions, genetic determination of the F V /F M related trait, also called heritability (H 2 ), and genetic correlation were calculated (Table 4). H 2 characterizes the variance of the trait captured by the genotypic effects relative to the total phenotypic variance. The two field sites ST12 and LJ12 showed very low H 2 , at 0.06 for both whereas H 2 was up to 0.31 for the greenhouse experiment GH13A. The low H 2 estimated show that only a minor fraction of the total variance is due to genetic effects. In addition, when looking at F V /F M genotypic estimates in the 6 environments (Fig. 2), the response range to low Mn availability ranged from 0.38 to 0.71 in LJ13 whereas LJ12 had F V /F M values from 0.72 to 0.76 and ST12 from 0.66 to 0.74. Indeed LJ12 and ST12 were phenotype in spring whereas the other environments were screened in winter. The winter 2012/13 was too wet and mild to induce Mn deficiency, while it is induced mostly under dry and cold conditions. Consequently, because of low H 2 , narrow variability and inadequate phenotyping time, it was decided to discard ST12 and LJ12 from the GWA analysis.
Calculation of genetic correlations across environments (Table 4) provided a better understanding of how genetic control of Mn efficiency is related across the different environments. The weak genetic correlations confirmed that the Mn efficiency trait had differential genetic determination depending on environments also named genotype × environment interactions. These contrasts across environments can be explained by several factors including climate (solar irradiation, soil humidity and temperature) and physiological age of the plants relative to the time of Chl a measurements. Indeed, the severity of Mn deficiency is subject to major fluctuations over time and consequently it was also expected that the genetic variability among the varieties would be affected. Moreover, the ranking of variety's F V /F M performances fluctuated over the environments and no grouping of variety in regards to their origins or row-type was observed.

Assessment of plant response to Mn deficiency
To validate the induction of Mn deficiency, plants in soil pots and in fields were sprayed with soluble Mn and measured two days after spraying. In all four environments, a clear impact of MnSO 4 foliar application was recorded on Chl a fluorescence curves and F V /F M values (Fig. 3). The fluorescence yield F V /F M of plants treated rose to a higher level (between 0.75 and 0.83) whereas the control plants remained at lower levels (between 0.47 and 0.66) (Fig. 3  b), confirming that the plants were indeed Mn deficient, but without any visual leaf symptoms (latent Mn deficiency). Therefore the ability to fully restore F V /F M to its maximum after foliar application of Mn sulfate confirmed that Mn is the only abiotic stress parameters influencing the F V /F M values under the conditions prevailing at the field sites and in the green house experiment. The J and I steps of the transients were flattened and less defined, and in severe cases a notable K-step was developed (Fig. 3 a). The most severe Mn deficient conditions were seen in the GH13B experiment in which Mn deficient plants showed flat J and I slopes. Furthermore, when Mn deficient plants in LJ13 were sprayed with MnSO 4 , the Mn concentration in leaf tissues increased significantly (Fig. 3 c). Therefore, spraying of Mn salts confirmed that Mn deficiency was induced and that plants can fully recover the maximum range of F V /F M . Moreover, the ICP-OES quantification of other nutrients indicated that their concentration in leaf tissues remained stable (Additional file 2). Consequently only Mn was responsible for the change of F V /F M .
Mn leaf concentration ranged from around 7 to 11.5 μgg −1 of DW and the H 2 of Mn concentration in leaf tissue was 0.33. The Pearson correlation and its associated Pvalue showed the highest correlation between F V /F M and Mn concentration at 0.59. The correlations were much lower with the other fluorescence parameters (Table 5).
Based on the differences in F V /F M , on the shape of the curves and on Mn leaf tissue concentration, it was evident that plants were exposed to Mn deficiency.

Grouping from the population structure
The structuration analysis allowed us to distinguish four groups (Fig. 4). Two groups of two-rows from mid-northern Europe were identified. One group of sixrows and one group with a mixture of six-rows and two-rows from a southern European origin were also detected. In blue, the group with two-rows contained most of the breeding lines of Scandinavian origin. It confirmed the significance of adding population structure into the GWA model.

GWA results
Taken together for the four environments considered, 54 SNP markers were found to be significantly associated with the F V /F M trait (P-value < 0.001) and were distributed over 6 chromosomes. Of these, 11 were unmapped SNPs with missing annotations. Marker-trait association displayed distinct peaks corresponding to QTLs (Fig. 5).
Other associations were detected for all JIP parameters and are discussed below. Despite strong genotype × environment interactions, a meta-analysis including all environments (data not shown) resulted in the identification of the same variants. However it does not reflect the reality as most of the associations are environment-specific. Different SNPs were detected in each environment (Fig. 5) implying that marker-trait associations are severely affected by the environment.
In contrast, only four SNPs were identified using the Mn leaf concentrations (Additional file 3). Three markers constitute one peak on chromosome 6H and one marker on chromosome 5H.

Putative candidate genes
Recent advance in barley genome sequencing [42] allows candidate genes to be sought for in the flanking region of the SNP associations. Furthermore, the iSelect array was designed with SNP markers located in the genic region of known genes [67]. Therefore, the significant SNPs for each of the peaks were blasted against databases [63] to identify candidate genes in the surrounding region of associations. Putative candidate genes controlling Mn efficiency and their positions on the genome were identified in the corresponding blasted genic region. Based on prior knowledge about physiological mechanisms involving Mn, numerous genes coding for proteins have been reported and classified for F V /F M , Area, V I , V J and V K fluorescence parameters ( Table 6, Additional file 4). However, no putative genes have been reported for Mn leaf concentration data obtained by ICP-OES. It is suggested that all the candidate genes play a role in Mn dependent pathways and these are presented below.

Photosystem II subunits
Different Psb subunits from PSII were found at the exact position or close to the SNP associated with the trait by GWA. Of these, a highly significant association (-log 10 (P val ) = 3.7) was found for F V /F M in KS11 with SNP 11_10254 located on 5H chromosome at 159.5 cM. Gene AK251925.1 was found at the same position coding for Psb28 PSII subunit. Other marker-associations pointed out genes coding for PSII subunits; AK249774.1 (54.3 cM on chromosome 4H), MLOC_77860.1 (124.6 cM on chromosome 7H) coding for PsbP PSII subunit were also identified with a close co-localization of significant SNPs (Table 6). Besides these results, PsbW and PsbN (F V /F M trait) and PsbQ (Area trait) PSII subunits coding genes were also found significant markerassociations, however they remained always more distant to the closest SNP.

Mn-SOD
The manganese superoxide dismutase (Mn-SOD) gene was identified in GH13A and LJ13 on chromosome 7H. Associations were found for V I with SNP SCRI_RS_158266 located at 12.8 cM (-log 10 (P val ) = 3.1) at the exact same position of the coding gene MLOC_69817.1 (Additional file 4).

Photosystem I subunits
Photosystem I (PSI) coding genes were found by GWA. Of these, PsaA/PsaB were found with the coding gene MLOC_40094.1 in GH13A for the F V /F M trait on chromosome 4H at 99.1 cM. SNP 12_30987 was detected as the most significant association (-log 10 (P val ) = 3.7) at the position 100.6 cM ( Table 6). In addition to these QTLs, PsaH (MLOC_53469.2) and PsaE (MLOC_73050.2) were detected for V I with SNPs 11_20908 (1H) and SCRI_RS_2824 respectively. They were 1 and 0 cM away from their respective genes. Finally, PsaN (AK368803) was identified with SNP 11_21057 (2H) at 0.1 cM from the marker position (Additional file 4).

Chlorophyll a/b binding protein
Genes coding for Chlorophyll a/b protein were identified for many associated SNPs on chromosomes 1H, 5H, 6H and 7H, in several copies and for all fluorescence parameters (Additional file 4). These proteins are also known as light-harvesting chlorophyll a/b binding (Lhc) proteins in higher plants and green algae.

Protein phosphatase 2C (PP2C) Mn/Mg aspartate binding
Many other SNPs were found to be significant and colocalized with PP2C genes. They were found in chromosomes 1H, 2H, 3H, 4H and 7H for all fluorescence parameters (Table 6, Additional file 4). From the GWA results, multiple copies of these genes were identified.

Quantification of Mn deficiency
Chl a fluorescence has previously been shown to be a powerful tool to quantify the severenes of Mn deficiency in plants [3,6,21]. However, little is known about differences in Mn uptake and Mn tissue concentrations in the plant tissue of varieties with contrasting Mn efficiency. Therefore, the quantification of Mn in leaves was used as an additional parameter.in the current study. Plants were clearly Mn deficient as indicated by the leaf tissue concentrations ranging from7-12 μg. −1 Mn g DW [21]. However, the correlation, 0.59, between Mn concentration and F V /F M does not suggest a strong link between the two parameters. GWA did not reveal any strong association with tissue Mn as indicated by the absence of a correlation between leaf Mn concentrations and the functional pool of Mn in PSII using Fv/ Fm values as a proxy.

Environment-specific QTLs
Mn efficiency appears to be a complex quantitative trait controlled by many genes, each contributing with minor effects. Genetic determination of the trait ranged between 0.18 and 0.31 in the four environments analysed (Table 4), which means that only 18 % to 31 % of the phenotypic variance was due to the genetic variance. This supports the variability of plant responses to Mn deficiency partly being determined by genetic background. However, in contrast to traits governed by a few loci with large effects, the variation in quantitative traits is caused by the segregation of multiple QTL with individually small effects that are sensitive to the environment [68]. Indeed, significantly associated SNPs were found with relatively small effects (0.011 to 0.039) for F V /F M (Table 6). Those markers are identified only within specific environments and the corresponding QTLs are consequently specifically related to a single environment. Collins et al. [13] introduced the concept of "constitutive" and "adaptive" QTLs controlling the response of crops to abiotic stress. A "constitutive" QTL is consistently detected across different environments whereas "adaptive" QTL is environment-specific. Stress sensitivity can therefore be due to the responsiveness of QTL regulation in specific environmental conditions or due to indirect causes contributing to the stress response. Staple or constitutive QTLs are naturally preferred in breeding.
In association mapping, detecting constitutive QTLs for abiotic stresses is challenging, since QTL effects are most likely small and controlled by one or several genes, and by epistasis and environments. Environment-specific QTL has been previously reported in cereals, in winter barley [69], in wheat [70,71], and in rice [72]. In the case of the Mn efficiency trait, the four environments included, had varying characteristics due to various climates in the three different years, differences in soil conditions and/or different growth stages at the screening date (Table 1).
Moreover, poor genetic correlations between environments also support the environment-specific feature of the trait, with correlations in a range from -0.11 to 0.61 for the four environments analysed (Table 4).
Nevertheless, with a genetic correlation of 0.61, the two greenhouse experiments imply that Mn efficiency in GH13A and GH13B was regulated by a set of shared genes under the controlled environmental conditions, with identical soil and comparable climates.
The results confirmed that environments have to be considered independently for the complex Mn efficiency trait. Consequently, different environmental conditions are needed to obtain a comprehensive picture of the trait and to identify the "adaptive" QTLs contributing to Mn efficiency.

Identification of candidate genes
Combining GWA, database screening and prior knowledge about the role of Mn in plants allowed several candidate genes to be identified, coding for PSI and PSII proteins, chlorophyll a/b binding proteins, GLP, Mn-SOD, and PP2C. Two different classes of encoding genes could be distinguished, the first related to photosynthesis and the others involved in various Mn dependent pathways.
In photosynthesis, reactions occur in the thylakoid membranes carried out by multi-protein complexes, namely ATP synthase, PSI, PSII, cytochrome b6f and light harvesting complexes. Proteins from these complexes were identified: chlorophyll a/b binding proteins, PsaA/PsaB, PsaE, PsaH and PsaN from PSI; Psb28, PsbN, PsbP, PsbQ and PsbW from PSII; and finally Lhc proteins in multiple copies at several genome positions.

Role of Mn in metabolic pathways
The light-harvesting chlorophyll a/b binding proteins belong to the Lhc family consisting of more than 20 different proteins and are connected with PSI and PSII [73]. Their primary function is related to the absorption of light through chlorophyll excitation and the transfer of the absorbed energy to photochemical reaction centres. Although Lhc are coded by well characterized and abundant CAB (chlorophyll a/b binding) genes [74,75], the link to Mn metabolism is not trivial. Nevertheless, previous studies showed a reduction in Chl a concentration in Mn deficient plants and changes in the phosphorylated forms of LhcII have been observed [21]. Standard error Allelic effect sign is estimated with respect to the minor allele. SNPs are ordered by environment and genome position Several protein subunits from PSI were also detected. Similarly to the chlorophyll a/b binding proteins there is no obvious and immediate links to Mn utilization. The subunits PsaA/PsaB are involved in electron transport and form the core centre of PSI [76]. PsaE is one of the three stromal subunits of PSI whereas PsaH and PsaN are specific plants subunits [77,78]. Although Mn is not a PSI cofactor, a decrease in the PSI activity in Mn-limited conditions has been reported previously. The composition of the PSI complex is altered and a decrease in PsaA subunit is pronounced [79].
PSII consists of at least 20 protein subunits and catalyses a light-driven process, splitting water into protons and molecular oxygen. [75,80]. The Psb28 protein is associated with the cytoplasmic side of the thylakoid membrane and known to interact with the core protein CP43 in the PSII monomer [81]. Even though no direct link with Mn has been demonstrated, recent studies have partly established its function. A Psb28 deletion mutant generated in Synechocystis showed slower growth rates and lower chlorophyll levels. [81,82]. Recent studies suggest a role of Psb28 in the assembly and repair of PSII under heat stress [83,84].
In barley, PsbP and PsbQ subunits are extrinsic proteins involved in the stabilization of the tetranuclear Mn cluster of OEC. [85,86]. The crucial role of PsbP in Mn binding to PSII has already been established [87]. It has been described as a Mn storage protein delivering Mn 2+ during PSII assembly. Its absence slows down the process of photoactivation. Whereas PsbP was found mandatory for OEC activity and showed a strong Mn dependence, PsbQ did not present a manganese-dependent activity [88]. However, recent studies established that PsbQ stabilizes PsbP binding and therefore contributes to the activity of the OEC in PSII complexes [89]. Finally, a study on the tobacco PsbQ mutant revealed the effect of PsbQ loss on Chl a fluorescence under low light conditions. For the PsbQ mutant, F V /F M dropped from 0.73 to 0.26 under low light conditions [90]. Hence, measuring Mn efficiency based on Chl a fluorescence appears to be linked to PsbQ activity. PsbN, a low molecular subunit encoded in the chloroplast, has exclusively been found in etioplasts, but its PSII function remains elusive [91]. Nevertheless, a recent study has demonstrated by reverse genetics in tobacco, that PsbN is involved in the biogenesis of PSI and PSII [92]. In addition, it has been demonstrated that PsbN is required for the repair and assembly of PSII [93].
Recent work with PsbW knock-out and antisense Arabidopsis mutants indicated an important role played by PsbW for the connection and stability of PSII-LHCII complexes [94]. Lower F V /F M values have been reported for mutants lacking PsbW, but despite the role of PsbW for the PSII complex, direct effects of Mn on PsbW remain unclear.
The identification of several candidate genes coding for PSII subunit proteins (PsbN, PsbP, PsbQ and PsbW) demonstrate that Mn is involved in the stability, in the assembly and repair of the PSII and in the loading of Mn into the photosystem machinery. Therefore, it suggests that some varieties are able to load Mn, assembly PSII and repair PSII better than other under Mn deficient condition.
Apart from photosystem related proteins, two other types of proteins were found to be involved in Mn efficiency: PP2C and GLP. PP2C is a well-known serine/ threonine specific protein phosphatase. With 76 candidate genes identified in Arabidopsis [95], the PP2C family is considered the largest protein phosphatase family in plants. They are identified on several barley chromosomes (1H, 2H, 3H, 4H and 7H). These phosphatases are involved in the regulation of several signalling pathways and stress responses in plants [96]. In the current context, an interesting feature of PP2C is the relatively high concentration of Mn or Mg required to maintain activity. In moss Physcomitrella patens, the activation of PP2C occurs at 0.1 mM of Mn whereas more than 5 mM of Mg was needed to activate the protein [97]. It therefore suggests that Mn has a special feature in activating PP2C, which appears to be a factor influencing Mn efficiency.
The remaining genes identified by GWA are coding for GLP and Mn-SOD. GLP are often described in crops as having two main enzymatic functions, OxO and SOD, in response to biotic and abiotic stress [98,99]. The two enzymes, Mn-SOD and OxO, generate hydrogen peroxide (H 2 O 2 ), a signalling molecule in plant defence. OxO enzymes are localized in the apoplast and Mn-SOD in mitochondria; they work in antioxidant defence by enzymatic mechanisms scavenging reactive oxygen species (ROS). A first study demonstrated the involvement of Mn as a cofactor in barley OxO activity [26]. Hence, it has been confirmed by spectroscopy and crystallography that Mn is a required cofactor for OxO catalysis [100,101]; moreover it revealed the Mn-SOD activity of GLP. GLP coding genes in barley are mapped on chromosomes 2H, 3H, 4H, 5H, 7H and classifies to five main subfamilies annotated HvGER-I to HvGER-V [102]. Finally, gene family studies on barley [103,104] confirmed the OxO and SOD activity of GLP, their function in defence resistance, the duplication of GLP genes, their developmental stage and their tissue-specific expression. The excellent characterization of the barley GLP genes and activity may contribute to a better understanding of the role of Mn.

Conclusions
Mn deficiency was the sole cause for the decrease in quantum yield efficiency (F V /F M ). The combination of Chl a fluorescence analysis and GWA approaches used in the current study has unravelled a series of main