Frost Resistance-1 (FR-1) and Frost Resistance-2 (FR-2) are the principal LT QTL on the group 5 chromosomes in the Triticeae (Francia et al., 2004; Skinner et al., 2005; Galiba et al., 2009). In barley, FR-H1 and FR-H2 are approximately 30 cM apart. The most likely candidate for FR-H1 is HvBM5A (Fu et al., 2005; von Zitzewitz et al., 2005), a MADS-box protein similar to the Apetala1 (AP1) gene in Arabidopsis thaliana (L.) Heynh. This gene was identified as the main source of allelic variation for the VRN response locus known as VRN-1 in the Triticeae (Danyluk et al., 2003; Trevaskis et al., 2003; Yan et al., 2003; von Zitzewitz et al., 2005). Recent evidence suggests that coincident VRN-H1 and FR-H1 QTL are the pleiotropic effects of HvBM5A (Dhillon et al., 2010). Further support comes from the observation that timing of maximum LTT is usually coincident with the timing of VRN saturation (Limin et al., 2007) and an unsatisfied VRN requirement maintains the vegetative state. On the other hand, there is evidence that LTT is not necessarily a function of VRN sensitivity. The barley variety ‘Dicktoo’, commonly used in LT research, is not VRN sensitive and LTT QTL map to FR-H1 in the Dicktoo × ‘Morex’ population (Pan et al., 1994). Dicktoo achieves a high degree of LTT under short day conditions and without VRN (Limin et al., 2007). The candidate genes for FR-H2 are one or more members of two physically linked clusters of at least 11 family members of C-repeat binding factor (CBF) genes, also known as DRE binding protein 1 (DREB1) (Francia et al., 2004; Skinner et al., 2005; Galiba et al., 2009).
Three loci (VRN-H1, VRN-H2, and VRN-H3) interact in an epistatic fashion to determine VRN sensitivity. All allelic configurations except Vrn-H2_ vrn-H1vrn-H1 vrn-H3vrn-H3 lead to a lack of significant VRN response (i.e., spring or facultative growth habits) (Takahashi and Yasuda, 1971). Deletions in intron I of HvBM5A are the functional polymorphisms of VRN-H1 accounting for the VRN insensitive (dominant) alleles (Fu et al., 2005; von Zitzewitz et al., 2005; Szűcs et al., 2007). A zinc finger–CCT (CONSTANS, CONSTANS-like, and TOC) domain transcription factor (ZCCT), encoding a flowering repressor downregulated by VRN, is considered the determinant of VRN-H2 (Yan et al., 2004). Allelic variation at this complex locus (there are actually three tightly linked ZCCT genes in barley: ZCCT-Ha, ZCCT-Hb, and ZCCT-Hc) is ascribed to loss-of-function mutations or complete deletion (Dubcovsky et al., 2005; Karsai et al., 2005). The candidate for VRN-H3 is HvFT1 (Yan et al., 2006; Faure et al., 2007). HvFT1 expression is induced by long days and may mediate the long-day flowering response (Turner et al., 2005). Allelic variation at HvFT1 is attributed to mutations in the first intron, with the relatively rare dominant alleles conferring very early flowering (Yan et al., 2006). According to Trevaskis et al. (2007) and Hemming et al. (2008), HvBM5A is a promoter of flowering that is activated by low temperatures. HvFT1 mediates the long-day flowering response and is induced by long days. This long-day induction of HvFT1 requires expression of HvBM5A, since in VRN responsive cereals HvFT1 is expressed only after plants have been vernalized. ZCCT-H is a repressor of flowering that is expressed in long days and delays flowering by suppressing long-day induction of HvFT1. HvBM5A also downregulates ZCCT-H and provides a mechanism to allow long-day induction of HvFT1 in vernalized plants.
The role of VRN sensitivity in LTT can be described in the context of the three growth habit types in barley: spring, winter, and facultative. Winter varieties are LT tolerant, highly responsive to VRN, and vary in PPD sensitivity. Spring varieties have little to no LTT and do not respond to VRN, and sensitivity or insensitivity to short day PPD is not relevant if they are grown under long-day (spring planted) conditions. The term facultative is generally used to describe genotypes that are LT tolerant, are not VRN sensitive, and may be PPD sensitive. Facultative varieties have winter allele haplotypes at the VRN-H1 locus and complete deletions of the ZCCT-H genes on 4H (Karsai et al., 2005; von Zitzewitz et al., 2005; Szűcs et al., 2007). Winter allele haplotypes have a full-length HvBM5A intron that includes a highly conserved 0.44 kbp “VRN critical” region. This critical region is the putative binding site, under long-day conditions, for the repressor encoded by VRN-H2 (Fu et al., 2005; von Zitzewitz et al., 2005). Deletions of various lengths are associated with variation in flower time (Szűcs et al., 2007), with large deletions (∼2.8 kbp) characteristic of spring growth habit types.
In addition to the VRN genes, the vegetative-to-reproductive transition and flowering time are controlled by PPD sensitivity genes. Allelic variation at HvFT3, the candidate for the PPD-H2 QTL on chromosome 1H, is due to deletions of (or within) the gene in accessions that are sensitive (e.g., remain vegetative) under short-day conditions (Faure et al., 2007; Kikuchi et al., 2009). In earlier QTL studies, PPD-H2 was shown to have significant effects on flowering under short photoperiods and in autumn-sown field experiments (Pan et al., 1994; Cuesta-Marcos et al., 2008b). Allelic variation at HvPRR7, the candidate for the PPD-H1 QTL on chromosome 2H, is particularly important in spring barley as the recessive allele confers insensitivity to long-day conditions, allowing for a prolonged growing period and consequently higher yield. Allelic variation at HvPRR7 is attributed to amino acid changes in the CCT domain leading to the recessive allele (Turner et al., 2005).
Mapping of winterhardiness QTL in the Triticeae has been achieved using biparental mapping populations of inbred (or doubled haploid) lines. Statistical methods for QTL detection in biparental mapping populations have been continuously improved (reviewed by Wang et al., 2005). These methods allow the estimation of QTL positions (within a confidence interval), effects, and interactions between QTL without the necessity of a high marker density (Piepho, 2000). There are, however, limitations to QTL detection in biparental populations that can compromise the subsequent success of marker-assisted selection. These include bias in the estimation of QTL effects due to reduced sample size (Vales et al., 2005), narrow genetic base and consequent limited scope of inference (Crepieux et al., 2005), and broad confidence intervals for QTL positions and effects (Darvasi et al., 1993; Hyne et al., 1995).
Genome-wide association mapping (GW-AM) provides a complementary or alternative approach to biparental mapping. Genome-wide association mapping can be performed using different types of germplasm and the individuals in the analysis do not need to trace back to a single cross. Therefore, a wider genetic base, including actual breeding lines representing a suitable spectrum of genetic diversity, can be sampled without having to develop populations (Flint-Garcia et al., 2003; Varshney et al., 2005). Quantitative trait loci detection by means of GW-AM is based on the existence of linkage disequilibrium (LD) between QTL and markers. In biparental mapping populations, associations between markers and QTL are only due to linkage, and the extension of LD depends on the number of recombination events that occurred during the development of the population. In GW-AM many factors, other than recombination, can be responsible for LD: mutation, admixture, different degrees of relatedness among individuals (kinship), genetic drift, and selection (Flint-Garcia et al., 2003). All these processes may create an underlying structure in the populations. Lack of consideration of population structure in the analysis may lead to false positives and false negatives (Pritchard et al., 2000). There are examples in which high-density markers and large population sizes have been used to accurately model for population structure and several methods have been implemented to increase computational speed in GW-AM (Kang et al., 2008, 2010; Zhang et al., 2010).
The development of genotyping platforms with sufficient marker density has made barley GW-AM possible. Genome-wide association mapping in barley is currently being implemented to identify and fine map traits directly in elite plant breeding material (Rostoks et al., 2006; Cockram et al., 2008; Gyawali et al., 2009; Beattie et al., 2010; Massman et al., 2010; Roy et al., 2010). We used the R software environment (R Development Core Team, 2009) with the efficient mixed model association (EMMA) package implementation (Kang et al., 2008) to empirically estimate the level of relatedness in our sample data and conduct GW-AM. Our objectives were to determine if mixed-model GW-AM using EMMA would be able to identify winterhardiness related QTL in our breeding program, and, if so, would it provide new perspectives on the relationship between LTT and VRN sensitivity.
Materials and Methods
The germplasm consists of breeding accessions and cultivars, most of which originated from the Oregon State University (Corvallis, OR) breeding program. There are two germplasm sets developed as part of the Coordinated Agricultural Project (available at http://barleycap.cfans.umn.edu [verified 8 Feb. 2011]): Oregon Barley coordinated agricultural project (CAP) I and CAP II. Based on the phenotypic criteria described in Table 1, CAP I consists of 16 winter, 35 facultative, and 27 spring habit accessions. Coordinated agricultural project II consists of 34 winter, 32 facultative, and four spring habit accessions. Of the total number of accessions (148), 39 accessions came from other breeding programs in the United States. The remainder trace to crosses among six parents: ‘Strider’ (winter, 6-row, and feed type), ‘Kold’ (winter, 6-row, and feed type), 88Ab536 (facultative, 6-row, and malt quality), ‘Orca’ (spring, 2-row, and malt quality), ‘Legacy’ (spring, 6-row, and malt quality), and ‘Excel’ (spring, 6-row, and malt quality). The two spring 6-rows were not included in the association analysis. The Hordeum Toolbox (THT) (available at http://hordeumtoolbox.org [verified 8 Feb. 2011]) contains genotype and phenotype data for all CAP accessions; data on the germplasm used for this study can be found in the THT database by searching for the Oregon State University breeding program and years 2006 (CAP I) and 2007 (CAP II). The identification number used for describing accessions in Fig. 1 matches the “line synonym” number in the THT database.
|LTT SPMN (% survival)||HD COR (S) (days to flower)|
|Number of lines||31||67||50||31||67||50|
Two winterhardiness traits were evaluated: LTT and VRN sensitivity. Low temperature tolerance was measured as the percentage of plants surviving LT stress in field and growth chamber tests. Vernalization sensitivity was measured under spring-sown field conditions and without VRN under greenhouse conditions. Flowering time per se was measured under fall-sown field conditions and under greenhouse conditions, with VRN. The various treatments, germplasm sets, years, and environments during which the experiments were conducted are summarized in Table 2.
|Germplasm set||Environment and trait measured|
|Field LT tolerance||Controlled LTT||Field VRN sensitivity (S)||Greenhouse VRN sensitivity (V–)||Field flowering time (F)||Greenhouse lowering time (V+)|
|CAP I||POR 2006||FCCO 2006||MRI 2006||COR 2008||COR 2008||COR 2006||COR 2008||COR 2008|
|CAP II||COR 2009||COR 2009||COR 2007||COR 2009||COR 2009|
|CAP I and II||SPMN 2009||COR 2009|
Controlled freeze tests were conducted at the Agricultural Research Institute of the Hungarian Academy of Sciences, Martonvásár, Hungary (MRI), as described by Skinner et al. (Skinner et al., 2006). Field assessments of winter survival were conducted at Fort Collins, CO (FCCO), at Pendleton, OR (POR), and at St. Paul, MN (SPMN). The trials were planted in the fall of 2006 (FCCO and POR) and 2009 (SPMN). At each location, plot sizes and experimental designs were in accordance with local practice. The percentage survival at all field sites was based on visual assessment when plots resumed growth after the winter.
For the greenhouse assessments, the VRN treatment consisted of maintaining seeds in moist soil in a growth chamber at a constant 4°C with no light for 6 wk. Seedlings were then moved to a greenhouse maintained at 18/16°C day/night. A 16/8 h light/dark light photoperiod was maintained using supplemental lights. Unvernalized plants were grown from seed planted 1 wk before the removal of the vernalized treatments in the same greenhouse. The vernalized and unvernalized plants were grown in a two replicate randomized complete block design. On both vernalized and unvernalized plants, heading date (HD) and final leaf number (FLN) were recorded on the first stem to flower. The experiments were terminated 150 d after planting the unvernalized treatment. Plants that had not flowered by this time were assigned an HD value of 150. Fall- and spring-planted experiments were conducted under field conditions at Corvallis, OR, (COR) using 1-m, 1-row plots. Each entry was replicated twice using a randomized complete block design. The fall planted experiments were sown in 2006 (CAP I) and 2007 (CAP II). The spring planted experiments were sown in 2009. The experiment was terminated 250 d after 1 January. Plants that did not flower were assigned an HD value of 250. Joint analyses of CAP I and CAP II for traits that were measured in different years (HD and FLN under controlled conditions and field flowering time under fall-sown conditions) were done using least squares adjusted means calculated with a set of common checks replicated in the different years.
Genomic DNA from each of the 148 CAP accessions was purified using Plant DNeasy (Qiagen, Valencia, CA) kits starting with 100 to 300 mg of seedling leaves. Under the auspices of the Barley CAP project all accessions were genotyped for 3072 single nucleotide polymorphisms (SNPs) using two Illumina GoldenGate oligonucleotide pool assays (OPA). Details on the development of the OPAs are described elsewhere (Close et al., 2009; Szűcs et al., 2009). Briefly, SNPs detected in expressed sequence tags (ESTs) and sequenced amplicons were used to design three Illumina 1536-plex pilot oligonucleotide pool assays (POPAs) (POPA1, POPA2, and POPA3). Single nucleotide polymorphisms were selected from three POPAs to generate two production barley oligonucleotide pool assays (BOPA1 and BOPA2). The BOPA assays were conducted at the USDA-ARS Small Grains Genotyping Center in Fargo, ND. From the 3072 SNPs, 2126 were informative in the combined CAP I and CAP II accessions. In addition, the VRN-H1, VRN-H2, VRN-H3, PPD-H1, and PPD-H2 loci were genotyped using allele-specific assays (Supplemental Table S1; http://barleyworld.org/winterdocs/winter_markers_12_10.pdf [verified 21 Feb. 2011]). The estimated positions of the SNPs are based on the consensus map developed by Close et al. (2009) and are available by downloading the 1.77 version of the barley HarvEST database (available at http://harvest.ucr.edu/HBarley178.exe [verified 21 Feb. 2011]).
Linear Mixed Model
The linear mixed model approach used in the association mapping analysis, including the estimation of multiple levels of relatedness between accessions, was previously described by Yu et al. (2006). We additionally used the changes in algorithms and kinship estimation introduced by Kang et al. (2008). The vector of phenotypes, y, is modeled as:in which X contains the marker data, β is a vector of marker allele effects to be estimated, Q contains the population assignments by STRUCTURE (Pritchard et al., 2000), ν is a vector of subpopulation effects, Z is an identity matrix, u is the random variance due to genome-wide relatedness, and e is the random variance due to error. The phenotypic covariance matrix is assumed to have the following form:in which K the matrix of kinship coefficients, is the genetic variance from the genome-wide effects, and is the residual variance.
The population assignment matrixes (Q matrices) for each of the CAP populations and for the combined set were generated with STRUCTURE following methods by Pritchard et al. (2000) and by using the linkage model described by Falush et al. (2003). A core set of 1527 SNP markers was selected after removing markers with minor allele frequencies ≤10% and ≥10% missing data. The kinship matrix was generated with the EMMA package implementation for the R software environment (Kang et al., 2008). We determined whether the Q matrix would improve the fit to our vector of phenotypes significantly or if the kinship estimation by EMMA would suffice as follows. Random SNPs are expected to be unlinked to the polymorphisms controlling the traits under study (H0: no SNP effect). An approach that appropriately controls for type I errors is expected to show a uniform distribution of p values (Yu et al., 2006). We chose the model that best explained our phenotypic data by plotting the cumulative distribution of the observed p values (generated with EMMA) for each model and population phenotypic data set against the expected, in which the diagonal line in these cumulative plots represents the ideal distribution.
The phenotypic means per environment and for the combined environments and SNPs were subjected to an analysis with the EMMA package implementation for the R software environment (Kang et al., 2008) using the publicly available package implementation (available at http://mouse.cs.ucla.edu/emma [verified 8 Feb. 2011]). The association analysis was performed by performing a linear mixed model association via t test with restricted maximum likelihood estimates. After obtaining the p values for each individual marker per phenotype, the threshold for the statistical significance was established by using the q package implementation for the R software environment value, which measures the significance in terms of the false discovery rate (FDR) associated with each tested SNP (Storey, 2002; Storey et al., 2004). The q value for a particular SNP is the expected proportion of false positives incurred when calling that feature significant. In all our experiments we used a FDR α level equal to 0.05.
Phenotypic Variation Explained
The R2 statistic is used in biparental QTL mapping to estimate the proportion of phenotypic variation explained by markers in the model. Linear mixed models have no well-established R2 calculation procedure. Sun et al. (2010) tested the performance of several R2-like statistics for linear mixed models and identified the previously described likelihood-ratio-based R2 () (Magee, 1990):in which logLM is the maximum log-likelihood of the model of interest, logL0 is the maximum log-likelihood of the intercept-only model, and n is the number of observations. reduces to the regular R2 and also provides a general measure for the QTL effects in linear mixed-model association mapping. The was used to calculate the variation explained by each individual significant marker.
As expected, all markers that are in high LD with each other will explain a similar proportion of the phenotypic variation. Therefore an LD heat map (the r2 statistic of linkage disequilibrium [Hill and Robertson, 1968]) was constructed with all markers with a p value above the 0.05 FDR α level. The LD plot was created with the snp.plotter package implementation for the R software environment (Luna and Nicodemus, 2007). One representative from each group of markers that were in complete LD (r2 = 1) with each other was retained for further analysis. We further created a multi-marker model to test the number of QTL present within the remaining markers. For this, marker selection was performed following a forward selection and backward elimination method, an approach regularly implemented in QTL detection (Basten et al., 1996; Cuesta-Marcos et al., 2008a). We used linear mixed models via t test with restricted maximum likelihood estimates implemented in the lme4 package implementation for the R software environment (Bates, 2005), using the markers as sources of variation. At each step, the marker with the lowest p value of its t statistic was added to the model. Markers with the lowest p value of the t statistic were then sequentially added to the model until no marker had a p value below the 0.05 threshold. We then checked whether all markers included in the model were still significant. For the remaining markers we applied a backward elimination by sequentially removing markers with p values above the 0.05 level until all markers left were significant and doing so we obtained the final model. With that final model, we conducted the likelihood ratio test (LRT) with maximum likelihood estimates to find significant interactions between markers. Then we calculated the of the model that included the significant markers and their significant interactions.
Results and Discussion
A comprehensive set of phenotyping experiments was conducted to measure LTT and VRN sensitivity (Table 2). Flowering time per se was also measured; although this trait is not a major focus of this report, the data serve as a useful baseline for assessing the VRN sensitivity assays. The Oregon Barley CAP I and CAP II germplasm sets were measured together for percent winter survival at SPMN and for field VRN sensitivity at COR. Greenhouse assays were conducted separately for CAP I and CAP II and the data were combined by calculating the least square means based on the use of common checks. Throughout this report, we focus on these larger data sets, as population size is essential for greater detection power in association mapping studies (Yu et al., 2006; Zhao et al., 2007; Myles et al., 2009). However, as supporting evidence there are three independent measures of LTT for CAP I—two field experiments, FCCO and POR, respectively, and one controlled environment at MRI.
There is abundant phenotypic variation for LTT, VRN sensitivity, and flowering time in the Oregon Barley CAP I and II germplasm arrays (Fig. 2 and 3; Supplemental Fig. S1 and S2). Starting with LTT and the full population data set from SPMN, the range of values are representative of those reported in the literature for field survival in stress environments (Pan et al., 1994). The facultative check Dicktoo was among the accessions with the highest survival (80%) whereas the survival for Orca, the spring growth habit check, was 0%. This differential winter injury was caused by minimum temperatures of −26°C (with 23 cm of snow cover) and −10°C (without snow cover). The percent survival for ‘Maja’, the facultative check used for the heading date studies, was 75%. Based on agronomic considerations, our definition of facultative growth habit includes high survival rates in target environments (Table 1). The minimum percent survival for facultative accessions was 45%. The results from the Oregon Barley CAP I germplasm tests at FCCO, POR, and MRI (Supplemental Fig. S1) corroborate the SPMN test. For these comparisons, the three individual CAP I tests are compared with the separate CAP I and CAP II data from SPMN. At FCCO the minimum temperature was −21°C (with snow coverage), −15°C (without snow coverage), and at POR the minimum temperature was −13°C (with snow cover) and −15°C (without snow cover).
In each of these single germplasm array datasets, Dicktoo (facultative) is among the most cold tolerant and Strider, the winter check, has somewhat lower survival than Dicktoo. Within the individual environments Maja showed 90% survival at POR, 68% survival at FCCO, and 78% survival at MRI. The MRI LTT freeze tests support previous reports for Dicktoo and Morex (spring check) (Hayes et al., 1993; Pan et al., 1994; Skinner et al., 2006). Using the same protocol, Skinner et al. (2006) report survival values for Dicktoo and Morex of 85 and 0% at −12.5°C, respectively. Overall, these data confirm that facultative accessions are as (or more) cold tolerant than winter accessions. Karsai et al. (2001) reported similar findings in a survey of winter and facultative germplasm evaluated under controlled freeze tests at the MRI. As shown in Table 1, the maximum and minimum percent survival values for facultative and winter germplasm were 85 and 45%, and 80 and 25%, respectively.
Our definition of facultative growth habit is based on a lack of VRN sensitivity. We used three measures of vernalization sensitivity: (i) HD under spring planted field conditions at COR, (ii) HD under greenhouse conditions, without VRN, and (iii) FLN under greenhouse conditions, without VRN. Heading date and FLN under greenhouse conditions were measured on the same plants. Under field conditions, Strider and 38 other winter growth habit types did not flower (Fig. 3). Maja (facultative) flowered 179 d after 1 January. Average HDs for the spring checks Orca, ‘Baronesse’, ‘Harrington’, ‘Robust’, ‘Tradition’, and ‘Lacey’ were 174, 178, 176, 174, 174, and 175 d, respectively. There was a clear separation of winter vs. facultative and/or spring HDs under field conditions. A limited number of accessions had intermediate HDs. These accessions were either too late maturing for production agriculture (e.g., 2 wk later than Maja) or they produced only a few tillers that flowered while the rest of the plant remained in a vegetative condition. Integrating the SPMN field survival data with the COR VRN sensitivity data we identified criteria for defining facultative growth habit in this germplasm array as follows: there are 11 facultative accessions with winter survival values between 45 and 55% and 56 facultative accessions above 60%. Within the winter growth habit accessions there were 11 that flowered, under spring-sown conditions at COR between 200 and 229 d; the remaining 39 accessions never flowered.
Under greenhouse conditions, without VRN, a continuum of phenotypes is observed from insensitive to sensitive and there are lines out of the range of the checks. Maja and Strider are consistently observed at opposite ends of the distributions for both HD and FLN. Maja and the spring habit checks had similar HD and FLN values, under greenhouse conditions and without VRN (Fig. 3; Supplemental Fig. S2). For example, average HD values for Maja, Baronesse, Harrington, Robust, Tradition, and Lacy were 48, 48, 46, 46, 49, and 46, respectively. Final leaf number values were 10, 10, 9, 9, 10, and 10, respectively. Over two-fold differences were observed between winter types (e.g., Strider) vs. the facultative and spring habit types (e.g., Maja) for HD and FLN. The greenhouse data underscore the advantage facultative germplasm could have over winter germplasm: rapid cycling of generations as with spring types. The highest correlation coefficients were observed between the three measures of VRN (Table 3; Supplemental Table S2). Heading date under field conditions is most agronomically relevant and is necessary for determining if facultative germplasm will be appropriate for spring sowing in any given environment. Under greenhouse conditions, HD is simpler to measure than FLN, but the latter is favored as an estimator of the transition from the vegetative to the reproductive phase of development (Limin et al., 2007; Cuesta-Marcos et al., 2008a, b). Similar distributions for greenhouse and field measures of vernalization sensitivity were observed for the separate Oregon Barley CAP I and CAP II germplasm arrays (Supplemental Fig. S2).
|Trait and environment||HD COR (S)||HD GH (V–)||FLN GH (V–)||HD COR (F)||HD GH (V+)||FLN GH (V+)|
|HD COR (S)||0.86||0.81||0.42||0.42||0.40|
|HD GH (V–)||0.93||0.33||0.48||0.49|
|FLN GH (V–)||0.31||0.46||0.53|
|HD COR (F)||0.17||0.16|
|HD GH (V+)||0.85|
Although flowering per se is not a primary focus of this research, it is worth noting that HD under fall-sown field conditions shows a very low correlation with HD and FLN in the greenhouse conditions with VRN (Table 3; Supplemental Table S2). This is likely due to the cumulative effects of changes in temperature and PPD duration under fall-sown conditions, whereas under greenhouse conditions a single long (16/8 h light/dark) PPD was supplied, together with a consistent temperature profile of 20/10°C day/night. The distributions for fall-sown HDs for Oregon Barley CAP I compared with CAP II are very different, due to differing numbers of winter vs. facultative accessions. Heading date per se in barley is a complex topic in its own right (Karsai et al., 1997b; Hay and Ellis, 1998; Sameri and Komatsuda, 2004; Cuesta-Marcos et al., 2008a, 2009).
Correlations between LTT and VRN sensitivity measures (spring-sown and greenhouse without VRN) are low but significant and positive. This can be explained by the fact that most (but not all) winter types have high LTT (Table 1). Likewise, fall-planted HD (per se) in the field is significantly and positively correlated with LTT. These patterns fit reports in the literature (Karsai et al., 2001). Positive correlations could be due to linkage or pleiotropy. To explore the genetic basis of these correlations we integrated the Oregon Barley CAP phenotype data with genotype data and conducted GW-AM.
Structure and Association Mapping
The two Oregon Barley CAP germplasm sets are somewhat different in their composition. In CAP I, 66 lines trace to various crosses involving one or more of six parents (Strider, Kold, 88Ab536, Orca, Legacy, and Excel). Strider and Kold are winter growth habit types. 88Ab536 is a facultative. Orca, Legacy, and Excel are spring types. All of these parental lines, except for Legacy and Excel, are also included in CAP I. The remaining eight accessions in CAP I do not figure in the recent pedigrees of other germplasm in the set. In CAP II, 31 lines trace to three parents (Strider, 88Ab536, and Kold). The remaining 39 accessions trace to other breeding programs and do not figure in the recent pedigrees of other Oregon CAP lines. Considering Oregon Barley CAP I and II, 97 lines share common parents and 51 (in terms of the short evolutionary history of a breeding program) are unrelated. The germplasm was selected according to the Barley CAP objectives, which were directed toward characterizing barley germplasm relevant to breeding programs. The Oregon winter malting barley program is focused on the simultaneous improvement of winterhardiness and malting quality, as described by Muñoz-Amatriain et al. (2010), which leads to narrow pedigrees. Unrelated accessions from other breeding programs (e.g., USDA-ARS, Aberdeen, ID, and Utah State University, Logan, UT) were included as possible sources of novel alleles for winterhardiness traits.
To reduce the number and likelihood of false positives in GW-AM of winterhardiness traits we accounted for structure by determining the Q and the K matrices (see Materials and Methods). For all traits, we performed a linear mixed model association with restricted maximum likelihood estimates using the K matrix (estimated with EMMA), with the Q matrix (Q + K model) and without the Q matrix (K model). Per Supplemental Fig. S3 we found that the expected and observed cumulative distributions of p values obtained using the K and Q + K models were comparable. There were no consistent or notable differences in the number and identity of markers showing significant associations with the two models. Our results are in accordance with those of Zhao et al. (2007) who estimated the K matrix by defining kinship coefficients simply as the proportion of shared haplotypes for each pair of individuals. These authors showed that correcting for population structure with this simpler estimate of kinship was as effective in reducing the false-positive rate as using the Q + K model of Yu et al. (2006). We estimated the K matrix using the Kang et al. (2008) method, which is similar to the method of Zhao et al. (2007), estimating a simple identical-by-state allele-sharing matrix. The K matrix estimated with EMMA also led to a biologically meaningful classification of the germplasm (Fig. 1) as is discussed in greater depth later in this report. Therefore, we report the results of association mapping using only the K matrix to account for structure in the Oregon Barley CAP I and II.
We performed GW-AM for all the data sets shown in Table 2. For LTT we emphasize the SPMN results in this report due to the larger population size and simultaneous application of the LT stress to all the germplasm. The CAP II germplasm was assessed in field tests at POR and FCCO the year following the assessment of the CAP I, but due to milder winter conditions no differential winter injury was observed. Likewise, for the VRN sensitivity phenotype in this report we emphasize the Oregon Barley CAP I and II simultaneously spring planted at COR and the greenhouse (GH) without VRN joint data sets. Although all markers were tested for marker-trait associations, no significant marker had a minor allele frequency ≤10%. For LTT at SPMN, all of the associations above the α = 0.05 FDR were on chromosome 5H. A few markers on 3H, 4H, and 6H approached the threshold (Fig. 4A). In the ∼30 cM interval on 5H, 30 markers were significant (Fig. 4B) and these markers occur in two linkage disequilibrium blocks corresponding to the reported positions of FR-H1 and FR-H2 (Francia et al., 2004).
The candidate genes determining FR-H2 are one or more members of a physically linked cluster of CBF family members (Skinner et al., 2005, 2006). A subset of the possible CBF gene family members at FR-H2 are represented on the BOPA. Marker 12_30854 is a SNP in HvCBF9. For the purposes of GW-AM these genes serve as markers and no inferences can be made as to which gene (or genes) are most likely determinants of the LTT phenotype. The FR-H1 candidate is HvBM5A (von Zitzewitz et al., 2005; Dhillon et al., 2010). A specific amplicon in the first intron of HvBM5A (marker VRN-H1a; Supplemental Table S1) is the most significant marker for the LTT association and additional significant markers are based on SNPs and indels in HvBM5A (Supplemental Tables S1 and S3). In the analysis of the SPMN data, and in the separate analyses of the CAP I data (see next section), VRN-H1 markers based on functional polymorphisms in HvBM5A always showed more significant associations than SNPs elsewhere in the gene. In the FR-H2 region, 12_31236 is more significant than 12_30854, the HvCBF9 marker. Interestingly, the EST in which 12_31236 is located could be involved in LTT, as it is annotated as a heat shock transcription factor (HSF). Heat shock transcription factor genes are transcriptional activators of heat shock proteins (HSPs). Heat shock transcription factors and HSPs are numerous and are an interaction point between multiple stress response pathways, including heat, cold, salt, and osmotic stress (Swindell et al., 2007). The highest homology with Arabidopsis of the protein encoded by the barley EST in which 12_31236 is located is AtHSFB2b. In Arabidopsis this gene is downregulated more than twofold after 12 h of exposure to 4°C (see supporting files from Swindell et al., 2007), and small HSPs have been shown to be induced by cold (Sabehat et al., 1998). Little is known about the low temperature responses of HSFs. The third most significant association in the 5H region was marker 11_11456. The EST in which this marker is located encodes for a protein that shows the highest similarity to Arabidopsis Glu-tRNA (Gln) amidotransferase subunit C. This gene has no obvious relationship with LTT. Other significant markers in the region could be of interest as candidate genes based on annotations in the HarvEST database with their respective unigene ID (Supplemental Table S3). However, in GW-AM markers in high linkage disequilibrium with the functional and/or causal polymorphisms may show higher significance than the functional and/or causal polymorphism itself (Weigel and Nordborg, 2005). The resolution of our GW-AM of LTT cannot, therefore, establish causal relationships between the phenotype and candidate genes nor can we confirm or refute that HvBM5A is the determinant of FR-H1 or that one or more HvCBF members are candidates for FR-H2. What is of particular note is that in this GW-AM scan we identified the one chromosome region—and the two QTL therein—reported to be associated with LTT in every biparental mapping study in the Triticeae reported over nearly two decades.
The GW-AM results for the experiments involving Oregon Barley CAP I are consistent in identifying the same region on chromosome 5H (Supplemental Fig. S4). In all cases, the most significant markers involve functional polymorphisms and SNPs in HvBM5A (Supplemental Table S3). Relying on the results from CAP I only carries the risk of false positive associations due to small population size. However, it is interesting that some significant associations are coincident with chromosome regions where genes and/or QTL are located that relate to the timing of the vegetative-to-reproductive transition. For example, the marker on the long arm of chromosome 4H showing the most significant association with LTT at POR is 12_30824, a SNP in HvBmy1, which is ∼4cM from VRN-H2. In CAP II SPMN, on chromosome 1H, 11_10686 approaches the FDR threshold and is ∼20 cM from PPD-H2, which cannot be assumed to be in LD with this SNP. The EST in which marker 11_10686 is located has the highest similarity to Arabidopsis ERF4 (Ethylene Responsive Element Binding Factor 4). Other significant marker associations are coincident or in the proximity of genes with possible roles in abiotic stress tolerance, for example, HvCBF8 on 2H. This gene has been reported as a pseudogene (Skinner et al., 2006), nevertheless other accessions could have functional alleles at the same locus. Annotations for other genes in which SNPs show significant or nearly significant associations in the CAP I and CAP I and II datasets do not have immediately apparent roles in abiotic stress resistance or growth and development. These associations may be due to false positives or to genes with effects too small to be detected as significant with the current germplasm, phenotype data, and marker density. Additional experiments involving GW-AM in larger populations with balanced phenotype data and higher marker density will be necessary to validate and exploit these associations.
The GW-AM scans for the three measures of VRN sensitivity (Fig. 5) identify a specific amplicon of the ZCCT-Hb gene at the VRN-H2 locus on chromosome 4H (Supplemental Tables S1 and S4) and SNPs within and/or near ZCCT gene family members. This validates the biparental QTL reports (Pan et al., 1994; Francia et al., 2004) and the functional models for the epistatic interaction of VRN genes in which, under long day conditions, the VRN-H2 locus represses flowering (Yan et al., 2004; Trevaskis et al., 2007; Hemming et al., 2008). This finding provides a genetic explanation for the very high phenotypic correlation observed between the three measures of the VRN sensitivity phenotype (Table 3; Supplemental Table S2). It further provides evidence that the complete deletion of ZCCT-H genes is sufficient to eliminate vernalization sensitivity. Neither VRN-H1 nor VRN-H3 individually were identified as significant determinants of vernalization sensitivity. There are a few markers approaching significance on 5H in the vicinity of VRN-H1, but these are not above the 0.05 FDR α level. The lack of a significant VRN-H1 effect is probably due to a very small numbers of lines (n = 13) that have spring alleles at VRN-H1 and the epistatic inheritance of the trait. The VRN-H3 situation is not as clear. It is likely that most, if not all, accessions in CAP I and II have recessive (winter) VRN-H3 alleles. However, the assignment of dominant and recessive allele types at this locus is not possible at this point: the proposed functional polymorphism in HvFT1 intron 1 (Yan et al., 2006) is not valid in broader arrays of barley germplasm (Cuesta-Marcos et al., 2010). A significant effect is observed in the greenhouse data on chromosome 2H. This is due to a specific amplicon of the HvPRR7 gene (PPD-H1) (Supplemental Tables S1 and S4), where some genotypes show later flowering under long days. The greenhouse was maintained at a 16/8 h light/dark photoperiod for the duration of the experiment. HvPRR7 nearly reaches the FDR threshold for FLN. The only individual data set in which PPD-H1 shows a clear and significant effect is with the Oregon Barley CAP II GH HD data, without VRN (Supplemental Fig. S5B). In this scan, marker 11_20074 (chromosome 7H) is also significant. Marker 11_20074 is a SNP within a cinnamoyl CoA Reductase 1 gene similar to the wheat Ta-CCR1, which is associated with stem elongation (Ma, 2007) and hence could be affected by daylength. This marker is approximately 17 cM from HvFT1 (VRN-H3). The basis of the low phenotypic correlations between fall-sown HD and the vernalized GH treatments is apparent in the scans (Fig. 5; Supplemental Fig. S5A and S5B). PPD-H2 drives HD in the field due to short days over the winter, delaying the vegetative-to-reproductive transition in competent accessions. In contrast, PPD-H1 drives flowering in the greenhouse due to the continuous long day photoperiods.
Based on the results of the association mapping of VRN sensitivity and LTT, the strategy for developing facultative germplasm is apparent: fix winter alleles at FR-H1 and FR-H2 on chromosome 5H and the ZCCCT-H deletion on chromosome 4H. Photoperiod responses can be targeted to specific environments by selecting for appropriate alleles at PPD-H1 and PPD-H2. Short-day sensitivity is likely to be appropriate for all environments, as it will delay the vegetative to reproductive transition. Long-day sensitivity could maximize yield by delaying maturity. However, the advantages may not be apparent in all environments. In some production zones, early winter barley avoids summer heat and water stress and in others earliness allows for relay cropping. All evidence, in this sample of germplasm, indicates that VRN sensitivity is not a prerequisite for maximum LTT. Therefore facultative germplasm seems to be the growth habit of choice: it will allow for accelerated cycles in the greenhouse for specific breeding purposes and production of the same variety under fall- or spring-sown conditions.
If breeding for facultative germplasm is the principal objective, the question that remains is whether to target FR-H1 or FR-H2 or both QTL. Interestingly, there are two distinct LD groups on chromosome 5H. To explore this issue in greater depth, we identified the phenotypic values of the four possible haplotypes at FR-H2 and FR-H1 and the proportion of phenotypic variance explained by the two-locus haplotype (Fig. 1; Table 4). This required an alternative approach to estimate the proportion of phenotypic variation explained by markers. The R2 calculation is standard in biparental QTL mapping but it has not been well established for linear mixed models. Sun et al. (2010) tested the performance of several R2-like statistics for linear mixed models and reported that reduces to the expected R2 and provides a general measure for QTL effects in linear mixed-model association mapping. We therefore used this approach and found an overall R2 for the FR-H2 and FR-H1 haplotype on winter survival (using the SPMN data) of 0.25 ( marker model − base model) (Fig. 1 inset; Table 4). The proportion of phenotypic variance accounted for by FR-H2 alone is 9% and by only FR-H1 is 15%. Francia et al. (2004), using a biparental mapping population, reported R2 values of 0.22 and 0.37 for FR-H2 and FR-H1, respectively, and an adjusted model, accounting for both loci, of 0.63. The differences in our estimates are likely due to the germplasm and perhaps the QTL mapping methodology. Francia et al. (2004) used a biparental population derived from a cross of two parents with extreme differences in LTT.
|Model||Model||Model form||–2 log likelihood||LRT||p value|
|1||Intercept||y = μ + e||1351.6|
|2||Q||y = μ + Qυ + e||1309.0||42.6||<0.001||0.25|
|3||K||y = μ + Zu + e||1276.6||75.0||<0.001||0.40|
|4||Q + K||y = μ + Qυ + Zu + e||1273.2||3.5||0.327||0.41|
|5||12_31236 + K||y = μ + m1 + Zu + e||1253.2||23.5||<0.001||0.49|
|6||VRN-H1a + K||y = μ + m2 + Zu + e||1234.9||41.8||<0.001||0.55|
|7||12_31236 + VRN-H1a + K||y = μ + m1 + m2 + Zu + e||1205.9||29.0||<0.001||0.63|
|8||12_31236 × VRN-H1a + K||y = μ + m1 + m2 + m1 × m2 + Zu + e||1195.3||10.6||0.014||0.65|
We further used the method proposed by Sun et al. (2010) with the Oregon Barley CAP I and II VRN sensitivity data (Table 5; Supplemental Table S5). After applying the forward selection–backward elimination method and determining their we found that the marker or markers that best explained the linear mixed model were for COR spring planted, VRN-H2b with an = 0.29; for HD GH without vernalization, PPD-H1 and VRN-H2b with an = 0.30; for FLN GH without vernalization, PPD-H1 and VRN-H2b and their interaction with an = 0.25; for HD COR fall planted, PPD-H2 with an = 0.05; for HD GH with vernalization, PPD-H1 with an = 0.30; and for FLN GH with vernalization, PPD-H1 with an = 0.19. Cuesta-Marcos et al. (2008a), using a biparental mapping approach, reported very similar R2 values for markers within the same genes.
|Model||Model||Model form||–2 log likelihood||LRT||p value|
|1||Intercept||y = μ + e||1436.7|
|2||Q||y = μ + Qυ + e||1392.2||44.5||<0.001||0.26|
|3||K||y = μ + Zu + e||1364.9||71.8||<0.001||0.38|
|4||Q + K||y = μ + Qυ + Zu + e||1364.4||0.5||0.327||0.39|
|5||VRN-H2b + K||y = μ + m1 + Zu + e||1272.2||92.7||<0.001||0.67|
In our analysis, the winter allele haplotype for FR-H2 and FR-H1 (coded as “BB”) is preponderant in the Oregon Barley CAP I and II germplasm (118 out of 148 accessions equals ∼80%). Accessions with this haplotype had an average winter survival at SPMN of 64%. The “BB” haplotype is found throughout the CAP I and II germplasm but only in winter (n = 52) and facultative (n = 66) accessions (Fig. 1). In contrast, the few accessions with only one favorable allele (“AB” and “BA” haplotypes) had average survival values of 30 and 14%, respectively. All of these accessions have spring germplasm in their pedigrees and persisted in the Oregon breeding program due to the fact that in the test environments routinely used by the program, no differential winter injury was observed during cycles of assessment and selection of this germplasm. Only one accession in the CAP I and II germplasm is a spring variety based on passport data (Orca). This variety and 10 other accessions had <10% survival at SPMN. All of these had “AA” haplotypes at FR-H2 and FR-H1 (Fig. 1 inset; Fig. 6). Of the “AB” and “BA” haplotypes, all were classified as spring types based on agronomic criteria except for two, which were classified as winter types based on VRN sensitivity (Table 1). These classifications underscore the challenges of classifying germplasm by growth habit. In terms of phenotype, spring growth habit types can be classified based on their lack of VRN sensitivity and low probability of winter survival in target environments. Winter growth habit types can be classified as vernalization sensitive. Cold tolerance is implicit but not defined. Facultative types lie somewhere between. Von Zitzewitz et al. (2005) proposed that the term facultative be used to describe accessions with the VRN-H2 deletion and a winter allele at VRN-H1. This two-locus haplotype corresponds to the “AB” plus “BB” accessions identified in Fig. 1 and the “AB(A)” plus “BB(A)” in Fig. 6. For most fall-sown environments in higher latitudes, we would amend this definition to specify the “BB” FR-H2 and FR-H1 haplotype, coupled with the “A” VRN-H2 deletion haplotype (“BBA”; Fig. 6). By these criteria, 66 accessions in the Oregon CAP I and II germplasm would be described as facultative. All facultative accessions have ZCCT-H deletions. Based on a threshold cutoff of 45% survival at SPMN, 67 accessions would meet the definition of facultative based on phenotype. Thus, in this sample of germplasm, the three-locus haplotype is 99% effective in predicting the phenotype.
This first use of GW-AM for LTT and VRN sensitivity in barley confirms that GW-AM is as effective as biparental QTL mapping for these traits. Genome-wide association mapping results were achieved using a sample of breeding lines, all of which, except for the spring variety Orca, had some agronomic potential (based on phenotype) in the Pacific Northwest of the United States. Field survival at SPMN was effective in identifying a subset of accessions with the capacity to survive the low temperature stresses encountered in this environment. In contrast, all biparental QTL populations used to date are based on winter and/or facultative × spring crosses in which at least 50% of the progeny are of no direct utility to the breeding program.
These results confirm that, in this sample of germplasm, maximum LTT can be achieved with facultative growth habit. This presents opportunities for rapid cycling of germplasm and the option to use the same varieties under fall- and spring-planted conditions. Facultative growth habit can be predicted with very high accuracy based on a three-locus haplotype. Two of the loci define the FR-H2 and FR-H1 haplotype. Unless QTL for other traits are found in the interval between the two LD groups defining the FR-H2 and FR-H1 haplotype, selection could target the complete ∼30 cM region on 5H. The third locus—VRN-H2—is not necessary or sufficient for LTT. These GW-AM results were found using a population of modest size (Oregon Barley CAP I + CAP II = 148) and many of the same results were obtained using only CAP I and or CAP II. In terms of methodology, we found that the K matrix alone was sufficient to account for structure in the sample of germplasm and that it generated a biologically meaningful classification of genetic relationships. The method of Sun et al. (2010) was effective in providing estimates of the proportion of phenotypic variance accounted for by significant associations. The modest estimate of (0.25) offers hope that additional gains from selection for LTT are possible in this germplasm. There are numerous regions in the genome where near-significant associations were observed. Genome-wide association mapping in larger sets of facultative germplasm in a balanced set of field and controlled environment trials should be effective in determining which of these possible associations are real and which are false positives. At the same time, genomic selection in this germplasm should be effective in fixing and validating the effects of alleles with small effects. Finally, this research was conducted using a small sample of related germplasm: abundant genetic resources in germplasm collections and other breeding programs have yet to be mined for alternative alleles at the 5H loci and at as yet unknown loci.
Supplemental Information Available
Supplemental material is available free of charge at http://www.crops.org/publications/tpg.