Plant material and micromalting conditions

For this study, fifteen six-rowed malting barley genotypes were used. These genotypes represent the University of Minnesota core germplasm, and include: the parental lines (PL) Dickson and Bonanza; the elite lines developed between 1958 and 1967 (EL58-67) M1, Cree and Manker; the EL68-77 Morex, Robust and M44; the EL78-87 M46, Excel, M55, Stander and M66; and the EL88-98 M78 and Lacey, as described by [3]. Figure 1 shows the pedigree information of the 15 genotypes, which are a subset of the entire Minnesota breeding program whose pedigree can be found in [3]. Breeding lines were advanced by single-seed-decent to the F5 generation. Selection began in the F4 generation for plant height, maturity, lodging resistance, and straw strength. This was followed by 3-4 years of replicated field trials selecting for yield and malting quality.

The fifteen genotypes were grown in a randomized complete block experiment with three replicates in single row plots 2 m in length in St Paul, MN in the summer of 2005. The plots were harvested at maturity with a small plot combine (Wintesteiger, Ried, Germany) and the seed cleaned using a Sample Cleaner model SLN4 (A/S Rational Korn Service, Esbjerg, Denmark). The cleaned seed from the three replicates was bulked and mixed thoroughly for micromalting. Micromalting was performed at the USDA-ARS Barley and Malt Laboratory of the Cereal Crops Research Unit (CCRU) in Madison, Wisconsin, under standard CCRU micromalting conditions. Barley grains (175 grams) were steeped at 16°C with individual steep times adjusted from 24 to 32 hours to achieve a homogeneous steep-out moisture of 45%. This uniform grain moisture is necessary to equalize the starting point of the malting process and, hence, establish a comparable starting physiological state across the lines. Steeping involved alternating 4-hour periods of immersion in water and air-rest. The hydrated grains were transferred to germinators immediately after steeping and maintained at 17°C and 100% relative humidity in the dark. Samples were collected at the end of the steep period ("out of steep") and after three days in the germinator ("3d of germination"). Three samples for each time point and genotype were harvested and frozen in liquid nitrogen, and cleaned of emergent rootlets and acrospires prior to microarray analysis. Finished malts derived from this experiment were subjected to malting quality analysis. A total of six parameters commonly used to assess malt quality were measured following the standard ASBC (American Society of Brewing Chemists 2004) procedures: malt extract percentage (ME), grain protein content (GP), ratio of soluble protein to total malt protein (S/T), diastatic power (DP), α-amylase activity (AA), and wort β-glucan concentration (BG). ME is one of the most important parameters for maltsters and brewers and measures the amount of soluble sugars and nitrogenous compounds obtained upon mashing malt into wort [9], which is directly related to alcohol production during fermentation and hence brewery profitability. GP is a measure of the percentage of protein found in the grain at maturity (prior to malting) and influences many of the other quality traits. S/T reflects proteinase activity and the extent of protein mobilization. AA and DP are measures of the carbohydrate-degrading enzymatic activities in malt. DP includes the combined action of alpha-amylase, beta-amylase and the rest of the amylolytic enzymes [9]. Finally, BG reflects the amount of cell wall breakdown products present in wort after mashing, and provides a measure of the extent of malt modification.

To ensure that the malting profile of the samples used in this study was representative of the general performance of these lines, we examined malting quality data collected from field trials published in a previous study [13]. These prior malting quality data, based on the mean of five trials conducted in Minnesota in 2002 and 2003, are presented in Additional file 1.

RNA isolation, labeling and GeneChip hybridization

Total RNA was isolated from the ninety grain samples (fifteen genotypes, two time points, and three replicates) using TRIzol^® Reagent (Invitrogen, Carlsbad, CA), with a phenol/chloroform/isoamyl alcohol (49:49:2) pre-treatment extraction step and passed through RNeasy columns (Qiagen, Valencia, CA) for further clean up. Prior to labeling, RNA quality was examined by means of an Agilent Bioanalyzer 2100 (Agilent Technologies, Palo Alto, CA). cDNA synthesis was conducted with 15 μg of total RNA and T7-Oligo(dT) primer (Proligo, Boulder, CO) using the SuperScript Double-Stranded cDNA Synthesis Kit (Invitrogen, Carlsbad, CA). The cDNA was purified with the Affymetrix GeneChip Sample Cleanup Module (Affymetrix, Santa Clara, CA). To produce biotinylated cRNA, the cDNA was transcribed in vitro using the Enzo BioArray HighYield RNA Transcript Labeling Kit (Enzo Life Sciences, Farmingdale, NY) in the presence of biotinylated UTP and CTP. The biotin-labeled cRNA was purified with the Affymetrix GeneChip Sample Cleanup Module (Affymetrix, Santa Clara, CA). Labeled RNA (15 μg) was chemically fragmented using the Affymetrix GeneChip Sample Cleanup Module (Affymetrix), and used for hybridization. The chip hybridizations, washes, and data acquisition were conducted at the Biomedical Image Processing Facility at the University of Minnesota following standard Affymetrix procedures. Data files are available online at PLEXdb (experiment BB91, http://www.plexdb.org/modules/PD_browse/experiment_browser.php?experiment=BB91) and at NCBI-GEO (accession number GSE24534, http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE24534).

Data analysis

GeneChip data were analyzed using the software GeneSpring GX 9.0 (Silicon Genetics, Redwood City, CA, USA). Expression estimates were calculated using the gcRMA algorithm implemented in GeneSpring. Quality control on samples was performed using various criteria including internal control 3'/5' ratio, sample correlation matrix, principal component analysis (PCA), and hierarchical clustering. One replicate of genotype M44 and another replicate of genotype M55 were removed for subsequent analyses based on their correlation coefficients and position in the PCA and hierarchical clustering. Only probe sets with a signal intensity value greater than the 20^th percentile in all replicates for at least one condition were considered reliable and included for further analysis. Differentially expressed genes between the fifteen genotypes were identified using one-way ANOVA with Benjamini-Hochberg false discovery rate (FDR) correction [23]. Analysis was done using a FDR adjusted P value of 0.01 as the cutoff, followed by filtering for two-fold or greater changes. A SNK post-hoc test was applied to identify the genes differentially expressed by pairs of genotypes. For annotation purposes, BLASTX (E-score cutoff = e^-10) data was exported from HarvEST:Barley version 1.68 http://harvest.ucr.edu/. To associate gene expression with malting quality phenotypic data, Pearson's correlation coefficients were calculated for each of the six malting quality parameters using the normalized intensity values averaged over replicates. Analysis was carried out with the use of SAS software, version 9.1 (SAS Institute Inc., Cary, NC, USA).

The fifteen lines were genotyped with 1,524 SNPs from the pilot Oligonucleotide Pool Assay 1 (POPA1) [24] using the Illumina GoldenGate assay. Genotyping was conducted at the USDA-ARS small grains genotyping center at Fargo, North Dakota. Genetic similarities between the fifteen genotypes were calculated using the genetic similarity (GS) index proposed by Nei and Li [25] with POPA1 SNPs. Correlation coefficients between genetic similarities and the number of differentially expressed genes between genotypes were also calculated using Pearson's correlation implemented in SAS software (SAS Institute Inc., Cary, NC, USA).

Single-feature polymorphism (SFP) analysis

Oligonucleotide arrays such as the Barley1 GeneChip can be used to detect single-feature polymorphisms or SFPs [26]. To test if SFPs significantly affected gene expression estimates and, therefore, the detection of gene expression differences between genotypes, we compared the list of differentially expressed genes between Dickson and Bonanza at time point "out of steep" before and after removing the probes that contained SFPs. We chose these two genotypes as they showed the largest number of differentially expressed genes. SFPs were detected using the probe affinity shape power method described by Xu et al. [27]. After masking the SFP probes, differentially expressed genes between the two genotypes were identified by t-test with a FDR adjusted p-value of 0.01, followed by filtering for two-fold or greater changes. A venn diagram was used to compare the probe sets contained in both lists, showing that 96.3% of the genes were common. This revealed that SFPs do not significantly alter the perceived expression differences as had already been described in Arabidopsis by Kliebenstein et al. [15]. Therefore, probes with SFPs were not excluded from the analysis.

Genetic, phenotypic and functional characterization of a barley breeding program

Effect of advanced cycle breeding on gene expression diversity

To examine the relationship between genetic and the gene expression data we performed a linear regression analysis and calculated the correlation coefficient (r) for both time points (Figure 2). Significant correlations were detected between genetic similarities and the number of differentially expressed genes between lines at "out of steep" (r = -0.709; p < .0001) and at "3d of germination" (r = -0.711; p < .0001).

To study the effect that advanced cycle breeding exerts on gene expression diversity, we focused on the number of differentially expressed genes between each line and the last cultivar released by the breeding program (Lacey). The number of transcripts identified for each time point was added to the figure where the pedigrees are represented (Figure 1), together with the genetic similarity (GS) indexes relating to Lacey. Interestingly, in all cases, there were more transcripts differentially expressed at "out of steep" compared to "3d of germination".

In general, we observed that as the breeding program progressed, the genetic similarities became higher while the number of differentially expressed genes became smaller (Figure 1), indicating a reduction in both genetic and gene expression diversity through the breeding process. The only exceptions were: Robust, whose similarity coefficient with Lacey is lower than both of its parents; and M66, in which both genetic and gene expression data indicate that it is more different from Lacey than its parents. In the case of Robust, our similarity coefficient was not in accordance with the gene expression data. Interestingly, we found a higher number of differentially expressed genes between one of the parental lines (Dickson) and the last developed cultivar (Lacey) compared to the other parental line (Bonanza) and Lacey. The GS indexes in Bonanza and Dickson were in accordance with these gene expression values (Figure 1).

The expression of many genes has been fixed through the advanced cycle breeding as can be observed by the reduction in the number of differentially expressed genes in the most recently developed lines (Figure 1). However, there are still many genes whose expression has not been fixed yet, as demonstrated by the relatively high number of differentially expressed genes (116 genes at time point "out of steep" and 52 genes at "3d of germination") that were identified between the two most recently developed lines of the breeding program, M78 and Lacey.

Transcript accumulation correlated with malting quality traits

To identify candidate genes for the improvement of malting quality in the University of Minnesota breeding program, we used the following strategy: we selected the most recent lines of the breeding program based on their similarities to Lacey and we identified the differentially expressed genes between them that were associated with malting quality. The choice of these "most recent" lines was done considering genetic, gene expression and phenotypic data. Based on our similarity coefficients and number of differentially expressed genes related to Lacey (Figure 1), this group of genotypes includes M78, M55, Stander, Excel and M44 together with Lacey. We also focused on the Condón et al. [3] study since it included the lines used for this work. Using SSR data, these authors conducted a principle correspondence analysis in which Robust was located together with the previously mentioned lines. While the genetic similarity between Robust and Lacey was not very high (0.887), the numbers of differentially expressed genes were relatively small. Phenotypic data also supported the choice of Robust due to its importance as a malting cultivar, prevailing as a recommended variety and dominating US malting barley acreage for over 20 years ([5]; http://www.ambainc.org/media/AMBA_PDFs/Pubs/KYMBV_2010.pdf). Using these criteria, the group of most recent lines was expanded to include Robust, M44, Excel, Stander, M55, M78 and Lacey. The genes that are differentially expressed between these genotypes was determined by a one-way ANOVA (FDR adjusted P value ≤0.01) followed by a ≥2-fold filtering, and resulted in the identification of 86 and 39 genes at "out of steep" and "3d of germination", respectively.

Transcriptome variation among the breeding program is correlated with genetic distance

Transcriptome comparison of the 15 lines identified a total of 1,442 differentially expressed transcripts at time point "out of steep". A lower number of genes (851) was found to be differentially expressed between the lines at "3d of germination", indicating that, as the germination process progresses, gene expression differences between malting genotypes become smaller. A reduced amount of differentially expressed genes was also found after 3 days of germination when two malting genotypes were compared in a previous study [19]. Pair-wise comparisons revealed that the number of differentially expressed genes between any two genotypes across time points varied from 3 to 1,005 (Tables 1 and 2). Dickson and Bonanza were the lines with the largest number of differentially expressed genes, an expected result since they were two parent lines of this study that were introduced to the Minnesota breeding program from North Dakota State University and Agriculture & Agri-Food Canada [3]. The variation found in the transcriptome of these 15 lines is smaller than the transcript-level variation found within seven Arabidopsis thaliana accessions [15], but is within the range of transcript-level variation found between 5 maize inbreds [16] and within 4 malting barley varieties [17]. This is probably due to the more closely related genotypes used in the last two studies. However, further studies evaluating gene expression diversity in barley would be required to assess whether values presented herein are representative of the particular species or are due to the closely related genotypes used in the study.

The relationship of genetic to gene expression differences between genotypes have also been examined in Arabidopsis [15] and maize [16] to determine if DNA sequence polymorphism are associated with gene expression diversity. In this study, pair-wise genetic similarity indexes were calculated between the 15 lines after their genotyping with 1,524 SNP markers (Table 3). Their high values, ranging from 0.829 to 0.994, showed the narrow genetic base of the University of Minnesota breeding program. In accordance with the Kliebenstein et al. [15] and Stupar et al. [16] studies, we found a significant correlation between gene expression and genetic diversity (Figure 2).

Gene expression diversity is still present after forty years of advanced cycle breeding

Several studies have analyzed the decrease in genetic diversity due to breeding in North American six-rowed barley cultivars [3, 5, 7, 13, 14, 29]. However, the repercussions of consecutive cycles of breeding on gene expression diversity were not examined. Identification of the differentially expressed genes between each line and Lacey and their placement on the figure that represents the pedigree of the lines (Figure 1), allowed us to visualize how advanced cycle breeding has affected gene expression diversity. Gene expression diversity is reduced through the breeding process, as shown by the decreased numbers of differentially expressed genes between the last cultivar released by the breeding program and its closer relatives. Genetic diversity is also reduced as the breeding program progresses, as can be seen by the increase in the GS indexes (Figure 1). However, Robust had an unexpectedly lower GS index that was not correlated with its gene expression data. Two previous studies, based on information from SSR molecular markers [3] and coefficient of parentages [5], clustered Robust together with Lacey, which suggests that our gene expression data are probably a better measure of the relationship between Robust and Lacey. Although both genetic and gene expression values of M66 were also unexpected, they support the previous study of Condón et al. [3], which placed M66 out of the contemporary group of elite lines. Interestingly, our data showed that the level of gene expression diversity between one of the parental lines (Dickson) and Lacey is much higher than between the other parental line (Bonanza) and Lacey, suggesting that Bonanza made a higher contribution to Lacey than Dickson. Although variation in gene expression was reduced in this breeding program, the gene expression patterns have not been fixed between the lines released in the last decade. These results indicate that four decades of breeding have not eliminated the gene expression differences and imply the possibility of obtaining further improvements from crossing elite lines from the breeding program.

Identification of target genes for enhancing malting quality

Improvement of the complex phenotype "malting quality" is the main goal of the University of Minnesota breeding program and, therefore, most of the reduction in gene expression diversity will be predictably associated with this phenotype. The high number of quality-related genes (813) whose expression has been fixed during the breeding progress (Table 6; Additional file 3) supports this assumption. Malt extract (ME) is probably the single most important parameter for maltsters and brewers due to its influence on ethanol production and hence the quantity of beer produced. A previous study including the genotypes used in this work detected significant gains in this trait during the advanced cycle breeding [13]. Our phenotypic data, showing a ME increase of around 3% (Table 4), together with the finding of a high amount (67%) of ME-correlated genes with a fixed expression in the most recent lines of the program (Table 6), corroborates the selection pressure that has been exerted on this trait. In agreement with this previous work, an increase in the AA levels over time was also detected in our data obtained from finished malts derived from the microarray experiment (Table 4). This trait also has a similarly high number (70.8%) of associated genes with fixed transcript levels through the breeding process (Table 6). Unlike ME, selecting for higher AA was indirect and maybe due to the selection for related traits such as dormancy [13, 30]. Although many S/T-correlated genes exhibited fixed expression during the breeding process (Table 6), little improvement was observed for this trait (Table 4). Both high grain protein content (GP) and high wort β-glucan (BG) concentrations cause problems during brewing, and reductions in both parameters are desirable in malting barley varieties. Significant decreases in both traits have been reported during advanced cycle breeding [13]. An overall reduction in BG was also observed in our data (Table 4) and a high number of BG-correlated genes with a fixed expression were detected (Table 6; Additional file 3). However, no reduction over time was observed in our GP values (Table 4). The low correlation (0.30) between malting quality data derived from grain used in the the microarray experiment and previous field data for GP ([13]; Additional file 1) is likely due to the limited variation for this trait among the lines in the microarray experiment.

To identify the gene expression diversity still available for future enhancements on quality traits within the program, we focused on the genes differentially expressed between the most recent lines that were correlated to any of the six quality-assurance parameters. Considering both time points, 49 genes constituting potential targets for the improvement of malting quality were identified (Table 5). From the quality traits evaluated, the greatest potential for targeted improvements can be expected for AA and ME since they had the highest number of associated genes. Interestingly, these two traits (AA and ME) for which the greatest increases have been achieved in this advanced cycle breeding program, and for which the most genes have fixed their expression during the breeding program, are also the ones associated with the greatest number of specific target genes for future improvements. This is in agreement with a previous study suggesting that there are unknown genes for ME segregating in the breeding population, since the known chromosomal regions associated with ME had been fixed in these elite lines while the phenotypic variance had not been reduced over time [13]. These authors also suggested that new genetic diversity for AA must have been introduced to the program to explain its variable pattern of phenotypic variance over the four decades of breeding. In the case of GP, BG and S/T, gene expression diversity analysis has also suggested targets to be exploited for future gains, as shown by the relatively high number of correlated candidate genes (Table 5). However, the smallest improvements by intercrossing these elite lines can be expected for DP, given the low number (6) of candidate genes (Table 5). Moreover, a few DP-associated genes (59) fixed their expression during the breeding process (Table 6), indicating a low diversity for this trait within the breeding program. Although DP values exceed current AMBA criteria, a future improvement of this trait would require additional diversity introduced into the University of Minnesota gene pool.

Focusing on the 49 candidate transcripts, none of them corresponded to the well-characterized genes controlling malting quality in barley such as α- and β-amylases, α-glucosidase, carboxypeptidases and cysteine endoproteinases [31, 32]. This was an expected finding since the expression of those genes would have been fixed very quickly in the elite lines, as observed by the numerous genes coding for these proteins that were found to be fixed through the breeding process (Table 6; Additional file 3). Similarly, none of our candidate genes was coincident with the genes correlated with malting quality traits in recent studies [17, 22]. However, the barley aleurone transcript B12D identified at "out of steep" (Table 5) was also identified by [21] as an abundant transcript expressed during malting. This gene, which is known to be expressed during germination and to have GA response elements in its promoter [33], is negatively correlated to ME and positively to BG. Since desirable quality characteristics include high ME and low BG values [34], the expression of this gene in our lines has a negative influence on these two malting quality traits.

Historically, the greatest emphasis has been placed on the role of cysteine-class proteinases for overall endopeptidase activity [34] and their ability to degrade Hordein, the proline-rich storage protein, with more recent studies suggesting a role for metalloproteinases in protein solubilization [35]. However, some endopeptidases belonging to the serine family have been identified during germination and malting [19, 36]. Additional effects of the serine endopeptidases on malting quality have been suggested due to their ability to degrade malt beta-amylase, which would have an indirect effect in reducing DP [37, 38]. Interestingly, three of our candidate genes encoded serine-type endopeptidases (Table 5). One of them (subtilisin-like serine protease) was positively correlated to ME and negatively correlated to BG. This highlights the complexity of the roles of the various proteinases in affecting malting quality since Schmitt et al. [37] found overall negative correlations between bulk serine endoproteinase activity and ME along with a positive correlation between bulk serine proteinase activity and BG. The other two serine class proteinases (prolyl endopeptidases) showed negative correlations with GP. It is attractive to speculate that the negative link between the prolyl endopeptidases and GP is through the ability of the prolyl peptidases to degrade the proline-rich storage proteins (Hordeins) in barley. There has been significant interest recently in degradation of prolamins from a number of cereals, including barley, by prolyl endopeptidases both due to the involvement of the prolamins in haze formation in beer (see [39]), as well as involvement in dietary sensitivity (Celiac sprue) to proline-containing peptides [40–42]. While prolyl endopeptidases have not previously been shown to function in malting, the results here may suggest that they could have a role in malting quality by affecting grain protein levels during grain maturation.

Additionally, barley hordoindoline b was positively correlated with ME at "out of steep" (Table 5). These proteins have been found to be associated with grain hardness [43], which is significantly correlated with ME [44]. While many of the postulated roles of these genes in affecting malting quality are still speculative, in the absence of a demonstrated functional path from the gene product to the final malting quality phenotype, our results provide a number of novel hypotheses for selection targets for further improvements in malting quality.