Molecular Markers for Sweet Sorghum Based on Microarray Expression Data

Abstract

Using an Affymetrix sugarcane genechip, we previously identified 154 genes differentially expressed between grain and sweet sorghum. Although many of these genes have functions related to sugar and cell wall metabolism, dissection of the trait requires genetic analysis. Therefore, it would be advantageous to use microarray data for generation of genetic markers, shown in other species as single-feature polymorphisms (SFPs). As a test case, we used the GeSNP software to screen for SFPs between grain and sweet sorghum. Based on this screen, out of 58 candidate genes, 30 had single-nucleotide polymorphisms (SNPs) from which 19 had validated SFPs. The degree of nucleotide polymorphism found between grain and sweet sorghum was in the order of one SNP per 248 base pairs, with chromosome 8 being highly polymorphic. Indeed, molecular markers could be developed for a third of the candidate genes, giving us a high rate of return by this method.

Introduction

The development of molecular markers is essential for marker-assisted selection in plant breeding as well as to understand crop domestication and plant evolution (Varshney et al. 2005). Single-nucleotide polymorphisms (SNPs) have become the marker of choice because of their abundance and uniform distribution throughout the genome (Gupta et al. 2008; Varshney et al. 2005; Zhu and Salmeron 2007). Around 90% of the genetic variation in any organism is attributed to SNPs (Varshney et al. 2005; Zhu and Salmeron 2007). They are discovered from genomic or expressed sequence tag sequences available in databases or through sequencing of candidate genes, PCR products, or even whole genomes (Varshney et al. 2005; Zhu and Salmeron 2007).

Recent studies have described the use of transcript abundance data from RNA hybridizations to Affymetrix microarrays to discover genetic polymorphisms that can be utilized as markers for genotyping in mapping populations (Borevitz and Chory 2004; Gupta et al. 2008; Hazen and Kay 2003; Shiu and Borevitz 2008; Zhu and Salmeron 2007). In an Affymetrix chip, each gene is represented by 11 different 25-bp oligonucleotides that cover features of the transcribed region of that gene (exons and 3′ untranslated regions). Each of these features is described as a perfect match (PM) and mismatch (MM) oligonucleotide. The PM exactly matches the sequence of a standard genotype, whereas the MM differs from the PM by a single base substitution at the central, 13th position (Borevitz and Chory 2004; Hazen and Kay 2003; Zhu and Salmeron 2007).

A new aspect of this approach is to discover sequence polymorphisms in cultivars or variants of species, where one of them has been sequenced but where no sequence information is yet available from the other ones. Here, the hybridization data from microarrays not only measure differential gene expression but also can yield information on sequence variation between two inbred lines. If two genotypes differ only in the amount of mRNA in a particular tissue, this should result in a relatively constant difference in hybridization throughout the 11 features. On the other hand, if the two genotypes contain a genetic polymorphism within a gene that coincides with one of the particular features, this will produce differential hybridization for that single feature. Such differences have been described as single-feature polymorphisms (SFPs) (Borevitz and Chory 2004; Borevitz et al. 2003; Hazen and Kay 2003; Zhu and Salmeron 2007). Thus, expression microarrays hybridized with RNA are able to provide us not only with phenotypic (variation in gene expression) but also with genotypic (marker) data (Zhu and Salmeron 2007). If two genotypes differ in the expression level of a particular gene, we can consider it as an expression level polymorphism or (ELP). Both ELPs and SFPs are dominant markers and can be mapped as alleles in segregating populations (genetical genomics), and ELPs can be considered as traits to determine expression quantitative trait loci (eQTLs) (Coram et al. 2008; Jansen and Nap 2001).

In Arabidopsis, SFPs have been used for several purposes such as mapping clock mutations through bulked segregant analysis (Hazen et al. 2005), the identification of genes for flowering QTLs (Werner et al. 2005), high-density haplotyping of recombinant inbred lines (RILs) (West et al. 2006), and natural variation in genome-wide DNA polymorphism (Borevitz et al. 2007). In plant species of agronomic importance, SFPs have been utilized to identify genome-wide molecular markers in barley and rice (Kumar et al. 2007; Potokina et al. 2008; Rostoks et al. 2005) as well as markers linked to Yr5 stripe rust resistance in wheat (Coram et al. 2008). However, an impediment to SFP discovery in crop plants based on DNA hybridization to Affymetrix expression arrays could be the size of gene families (Borevitz et al. 2003; Varshney et al. 2005; Zhu and Salmeron 2007). Because the coding regions of many gene clusters that arose by tandem gene amplification are quite conserved, hybridization-based approaches would not be sufficient to distinguish between allelic and paralogous copies (Xu and Messing 2008). Therefore, one would have to limit this analysis to low-copy genes. On the other hand, this approach does not aim at identifying candidate genes directly but rather linked genetic markers.

An area where gene discovery has become of general interest is the utilization of biomass for the production of alternative fuels. Because desirable traits for biofuel crops are very complex and involve many genes from different pathways, it becomes necessary to take genetic approaches to identify key genes so that molecular breeding can be employed to make performance improvements. The most successful biofuel crop today is sugarcane. However, it cannot be grown in moderate climate. Maize, which is a major biofuel crop in the USA, has a much lower yield of bioethanol per acreage than sugarcane, requires high input costs, and is a major food and feed source. A crop that bridges between the two is the close relative, sorghum. Sorghum tolerates harsher environmental conditions than sugarcane and maize, has a higher disease resistance than maize, and has a high stem sugar variant, sweet sorghum, which has potential yields of bioethanol like sugarcane. Moreover, sweet sorghum can be crossed with grain sorghum so that genetic analysis could uncover key regulatory factors that would increase sugar and decrease lignocellulose in the biomass. Therefore, sorghum could be used to identify both SFPs and ELPs linked to high sugar content.

We have recently reported the hybridization of RNAs derived from the stems of grain and sweet sorghum onto the sugarcane Affymetrix genechip (Calviño et al. 2008). A previous study demonstrated that cross-species hybridization did not affect the reproducibility of the microarray experiment (Cáceres et al. 2003). Moreover, an Affymetrix soybean genome array has been used to identify SFPs in the closely related species cowpea (Das et al. 2008).

Here, we have asked the question whether we could use the sugarcane chip analysis to extend the cross-species concept in SFP discovery in the grasses. We report the identification of SFPs in 58 sorghum genes by using the recently developed software GeSNP (Greenhall et al. 2007). These genes were described in our previous study to be differentially expressed between grain and sweet sorghum (Calviño et al. 2008). The utility of GeSNP has been successfully tested for SFP discovery in mice, humans, and chimpanzees (Greenhall et al. 2007), but there is no report on plants yet. In order to experimentally validate the SFPs identified in sorghum, we sequenced fragments from 58 genes and found SNPs in 30 of them, out of which 19 genes had a validated SFP. Furthermore, we develop molecular markers based on the SNPs found. The high experimental validation rate of SNPs of 50% of the candidate genes shows the potential of this method for the development of molecular markers and, in principle, the applicability to any trait of interest.

Results

SFP discovery and validation from differentially expressed genes in sorghum

Previously, we reported the use of an Affymetrix genechip from sugarcane to identify differentially expressed genes in the stem of grain and sweet sorghum (Calviño et al. 2008). Such a cross-species hybridization (CSH) approach allowed us to identify 154 genes harboring expression level polymorphisms between grain and sweet sorghum. In order to discover single-feature polymorphisms within these genes as well, we uploaded the sugarcane Affymetrix CEL files previously obtained into the GeSNP software. Indeed, we found that, from 154 genes, 57 harbored a SFP with a t value ≥7 (Fig. 1 and Table 1). Based on existing data (Greenhall et al. 2007), we adopted a t value of 7 or higher as a threshold. Chromosomes 1, 2, and 3 had the highest number of genes displaying both ELPs and SFPs, whereas chromosomes 5 and 6 had the lowest number of ELPs and SFPs, respectively (Fig. 1).

Fig. 1

Histogram showing the proportion of ELPs and SFPs between BTx623 and Rio for each sorghum chromosome. The number of genes with ELPs previously reported by Calviño et al. 2008 were plotted for each chromosome along with the number of SFPs found in this study. Only SFPs with t values ≥7 were taken into consideration.

Table 1 Sorghum Genes with SFPs Predicted by the GeSNP Software

In order to validate the SFPs discovered and calculate the SFP discovery rate (SDR) of the GeSNP software, we cloned and sequenced the fragments from 57 genes harboring both ELPs and SFPs in addition to one gene harboring only SFPs (see below) from sweet sorghum Rio and aligned the sequences against the BTx623 reference genome. The software predicted a total of 125 SFPs (on average ∼2 per gene), and we could experimentally validate 32 of them (Table 1). We calculated the SDR as 25.6% . As expected, the SDR was dependent on the t value, with the lowest SDR (less than 10%) at t values between 7 and 10 and the highest SDR (80%) with t values from 22 to 25, respectively (Fig. 2a).

Fig. 2

The SFP discovery rate of GeSNP is dependent on the t value. The percentage of SFPs in sorghum genes that were validated through sequencing (and thus represented true SNPs between BTx623 and Rio) was plotted against their respective t values (a). For the validated SFPs, we calculated the frequency distribution of their respective t values (b).

Besides SFPs identified in genes that are differentially expressed, the GeSNP software also detected SFPs in genes that did not show differential expression under our experimental conditions (data not shown). Considering the high success rate of SNPs discovered in genes having both SFPs and ELPs, we extended our screen to genes that have predicted SFPs with t values of 22 to 25 but no ELP. This analysis allowed us to identify 35 sugarcane probe pairs that matched the sorghum genome sequence and have a high probability of representing SNPs in genes that have no ELPs between BTx623 and Rio but were expressed in the stem (see Table 2). For example, one of the sugarcane probe pairs (Sof.3814.1.S1_at) matched a sorghum gene coding for fructose bisphospate aldolase. Since the protein product of this gene has a role in the sucrose and starch metabolic pathway (our trait of interest), we cloned and sequenced the fragment containing the SFPs. As it is shown in Fig. 3, we found six SNPs, two of which were recognized by three sugarcane probe pairs. This result indicates that our approach is able to efficiently detect SNPs. From the 58 genes that were sequenced, 19 genes (∼33%) had a validated SFP, and 11 genes (19%) harbored SNPs outside the probe pairs at different location than the one predicted by GeSNP. Therefore, the total SNP detection rate was ∼52%. A list of genes with validated SFPs as well as the nature of the nucleotide change/s is provided in Table 3.

Table 2 Sugarcane Probe Pairs with t Values of 22–25 That Identify Sorghum Transcripts with SFPs but not ELPs

Fig. 3

SFP validation for fructose bisphosphate aldolase. A fragment from the gene fructose bisphosphate aldolase was cloned and sequenced from both BTx623 and Rio and SNPs predicted by the probe pairs #8, 9, and 11 were validated. The blue lines represent the sugarcane probe pairs that are identical to either the Rio sequence (probe pairs #8 and #9) or identical to the BTx623 sequence (probe pair #11).

Table 3 Nucleotide Change Conservation for Validated SFPs Between BTx623, Rio, and Sugarcane

Most of the validated SFPs had probe pairs with t values from 15 to 18 and greater than 25 (Fig. 2b). Since the SFP validation depends on the SNP position along the probe pair (Rostoks et al. 2005), we analyzed the SNP position from the edge of the sugarcane probe pair for those genes with validated SFPs (Fig. 4). We found that, from a total of 22 probe pairs (probes that recognized the same SNP were not counted), 19 of them recognized a SNP between the sixth and the 13th positions.

Fig. 4

The position of the SNP along the 25mer in the probe pair influences the SFP validation. The position of the SNP from the edge of the sugarcane probe pair was scored for each validated SFP. Most of the SNPs locate within positions 6 and 13 along the 25mer. If two or more SNPs were located on a single probe pair, their positions along the 25mer were not counted and thus not included in the graph.

With regard to genes involved in our traits of interest, that is, sugar accumulation and cell wall metabolism, we validated SFPs for five of them (Figs. 5 and 3). The SFPs in the cellulose synthase 1 and dolichyl-diphospho-oligosaccharide genes was based on a SNP, whereas the SFP in the LysM gene was due to a 13-bp indel (Fig. 5a, b). This indel allowed us to develop an allele-specific PCR marker (Fig. 5d). In the case of the 4-coumarate coenzyme A ligase gene, the SFP was based on a mis-spliced intron in Rio (Fig. 5c).

Fig. 5

GeSNP prediction of SFPs in sorghum genes related to biofuel traits. The hybridization intensity between the perfect match and the mismatch oligonucleotides was averaged and scaled (GeSNP software output) and plotted against each sugarcane probe pair. Graphs are shown for four genes related to biofuel traits that have SFPs with t values of ≥7 and that were previously reported to be differentially expressed between grain sorghum BTx623 and sweet sorghum Rio (a). The SFP present in lysM identified a 13-bp indel, whereas the SFPs present in cellulose synthase 1 and dolichyl-disphospho-oligosaccharide identified an A/G and G/A SNP between BTx623 and Rio, respectively (b). In Rio, the third intron of the gene 4-coumarate coenzyme A ligase is mis-spliced and detected in the sugarcane probe pair #2 (c). Molecular markers for the genes lysM, cellulose synthase 1, and dolichyl-diphospho-oligosaccharide were generated based on allele-specific PCR (d). In the case of lysM, a primer spanning the 13-bp deletion in BTx623 was used to selectively amplify the allele from Rio. In the case of cellulose synthase 1 and dolichyl-diphospho-oligosaccharide, primer pairs specific for the SNP in question were generated by the WebSNAPER software and tested empirically.

To calculate the number of SNPs per total sequence length, we determined the genome size of the Rio line by flow cytometry. The Rio line appeared to have the same genome size than the sequenced BTx623 (data not shown). Based on 87 SNPs in 21,612 bp of sequence from both parental lines, we concluded that there is an average of one SNP every 248 base pairs of sequence between BTx623 and Rio. Taking in consideration that the genome size is in the order of 730 Mbp (Paterson et al. 2009), we suggest that 2,938,800 SNPs could exist between grain sorghum BTx623 and sweet sorghum Rio and that at least 0.4% of the genome could be polymorphic between the two lines. We also looked at the SNP density per sorghum chromosome in order to see if there is any difference among them. Surprisingly, we found that the level of polymorphism is higher for chromosomes 8 and 9 and lower for chromosome 3 compared to the average SNP density per Kb of sequence (4 SNPs/Kbp) (Fig. 6a). However, if we consider the frequency of probe pairs with t values between 22 and 25 for each sorghum chromosome as it is shown in Fig. 6b, chromosome 3 had the highest number of probes. On the other hand, chromosome 8 had the second highest number of probes with t values between 22 and 25 together with a high SNP density (Fig. 6a, b). This might suggest an unusual level of polymorphism for this chromosome between BTx623 and Rio. However, we have not sufficient data (genes sequenced) to test whether the SNP density differences among the chromosomes are statistically significant.

Fig. 6

SNP density per sorghum chromosomes. The number of SNPs per kb of sequence was calculated based on the number of genes sequenced belonging to a given chromosome. Only those chromosomes with five or more genes sequenced are represented (a). Frequency distribution along sorghum chromosomes of sugarcane probe pairs with t values between 22 and 25 (b).

Sorghum genes harboring validated SFPs allowed us to investigate if such nucleotide substitutions were conserved or not within grain sorghum BTx623, sweet sorghum Rio, and sugarcane. Indeed, we found that from 22 SNPs discovered through 29 validated SFPs (one sugarcane probe pair can recognize more than one SNP), 15 of them were conserved between BTx623 and sugarcane, whereas only eight SNPs were conserved between Rio and sugarcane (Table 3).

Development of molecular markers based on validated SFPs

The identification of SNPs between BTx623 and Rio provided a direct way to develop molecular markers that can be used in mapping populations. From 58 candidate genes, we were able to develop allele-specific PCR markers for 18 (Table 4). We utilized the Single Nucleotide Amplified Polymorphism (SNAP) technique to develop markers based on SNPs (Drenkard et al. 2000), as it is shown for the gene alanine aminotransferase (Fig. 7). These markers were tested also in other grain and sweet sorghum lines to see whether the SNPs were conserved or not (Table 4). In fact, we found a marker within the gene Sb09g029170 that distinguished the grain sorghums from the sweet sorghums cultivars used in this study. The protein product encoded by this gene is a putative ketol-acid reductoisomerase enzyme that is involved in the biosynthesis of valine, leucine, and isoleucine amino acids (www.phytozome.net/cgi-bin/gbrowse/sorghum/). SNAP markers were also developed for the cellulose synthase 1 and dolichyl-diphospho-oligosaccharide genes (Fig. 5d).

Table 4 Primer Sequences of SNAP Markers within Sorghum Genes

Fig. 7

Development of a molecular marker for alanine aminotransferase based on SFP discovery and the SNAP technique. The SFP detected by the probe pair #5 in the sugarcane probe set Sof.1326.1.S1_a_at was validated through sequencing (a). Specific primers for either A or G nucleotides were designed with WebSNAPER (b) and tested through PCR in ten sorghum lines (c).

It has been suggested that Dale and Della sweet sorghums share a common genetic background (Ritter et al. 2007). In agreement with this, we found that from ten SNAP markers that gave a PCR product in both lines, they always represented the same allele (Table 4). In addition, the sweet sorghum lines Top 76-6 and Simon have been identified as attractive contrasting pairs for mapping purposes based on their difference not only in genetic distance (D) but also in sugar content (measured as Brix degree) (Ali et al. 2008). In our work, we identified six SNAP markers within the genes Sb01g044810, Sb03g027710, Sb04g0037170, Sb08g008320, Sb09g006050, and Sb10g002230, respectively, which were polymorphic between Top 76-6 and Simon. These markers will be useful for mapping purposes when these lines are used as parents.

Discussion

A significant proportion of the phenotypic variation in any organism can be attributed to polymorphisms at the DNA level. Thus, these DNA polymorphisms can be used for genotyping, molecular mapping, and marker-assisted selection applications. The association of a particular trait of interest with a DNA polymorphism is essential for breeding purposes. Microarrays have been used to identify abundant DNA polymorphisms throughout the genome (Gupta et al. 2008; Hazen and Kay 2003). In particular, ELPs and SFPs can be identified from RNA hybridization studies. SFPs are detected by oligonucleotide arrays and represent DNA polymorphisms between genotypes within an individual oligonucleotide probe pair that is detected by the difference in hybridization affinity (Borevitz et al. 2003). In addition, SFPs present in a transcribed gene may be the underlying cause of the difference in a phenotype of interest. In most of the cases, SNPs are the cause of SFPs as have been demonstrated by sequence analysis (Borevitz et al. 2003; Rostoks et al. 2005).

Here, the goal was to identify SFPs from an Affymetrix sugarcane genechip dataset of closely related species (Calviño et al. 2008). The Affymetrix sugarcane genechip was used to survey the SFPs with the GeSNP software between two sorghum cultivars that differ in the accumulation of fermentable sugars in their stems, with the objective to develop genetic markers for mapping purposes. This is the first report to our knowledge of the use of GeSNP to identify SFPs within closely related grass species and the development of molecular markers based on validated SFPs.

We cloned and sequenced gene fragments harboring SFPs with t values equal or higher than 7 from 58 sweet sorghum genes comprising 125 SFPs in total. In this study, we found a SFP discovery rate of 25.6%, which is sufficient for most applications. Still, there are several possibilities to increase the SDR. First, the number of biological replicates suggested for using the GeSNP software is 4 or more. In contrast, we had only three replicates for both grain and sweet sorghum. Second, the cross-species hybridization of sorghum RNAs to probe sets of the sugarcane array is not as sensitive as intra species hybridization. Third, false positives could be due to the cross-hybridization of paralogous gene targets to individual probes, which may affect the specificity of the SFP calling. This problem would also arise from using next-generation sequencing for SNP detection. Nevertheless, we could show that the use of expression analysis in conjunction with GeSNP is an efficient and inexpensive way to develop new molecular markers.

The sugarcane probe pairs with t values between 22 and 25 had the highest SDR (80%) found in our study. One of these probe pair sets matched a sorghum gene coding for fructose bisphosphate aldolase (cytoplasmic isozyme) and the identified SFP was confirmed through DNA sequence analysis (Fig. 3). This gene codes for a glycolytic enzyme that catalyzes the cleavage of fructose 1,6 bisphosphate to glyceraldehyde 3-phosphate and dihydroxyacetone phosphate (Tsutsumi et al. 1994).

One third (33%) of the 58 genes that we have sequenced have a validated SFP. In addition, we could detect SNPs in 19% of all sequenced genes at a different position than indicated by GeSNP. This is attributable to the fact that the probe pair set does only cover a part of the gene, which implies that any SNP outside this region is not reported by GeSNP. We estimated the average SNP density between BTx623 and Rio to one SNP every 248 bp. This is probably an underestimation because the sugarcane probe sets were designed from genic regions and are, therefore, more conserved than other regions in the genome.

Although the sorghum chromosomes 1, 2, and 3 had the highest numbers for both ELPs and SFPs, chromosomes 8 and 9 were the most polymorphic ones, measured as the number of SNPs per Kb sequence (Figs. 1 and 6). Our data are in agreement with a previous report by Ritter et al. (2007) in which amplified fragment-length polymorphism markers on chromosome 8 could unambiguously distinguish grain from sweet sorghum lines (Ritter et al. 2007). Furthermore, sugar content QTLs have been located in this chromosome with a RIL derived from a dwarf derivative of Rio as one of the parents. In addition, we found that a marker within the gene Sb09g029170 coding for a putative ketol-acid reductoisomerase could discriminate the grain sorghums from the sweet sorghum lines used in this study (Table 4). This enzyme is the second in the biosynthesis of branched amino acids valine, leucine, and isoleucine (Leung and Guddat 2009). When the SNPs found through validated SFPs were compared between BTx623, Rio, and sugarcane, we found that SNPs between BTx623 and sugarcane are twice as high as between Rio and sugarcane.

Allelic genetic diversity among sweet sorghum cultivars has previously been investigated based on simple sequence repeat markers (Ali et al. 2008). This study described the correlations between allelic diversity and the degree of stem sugar. Indeed, one could envision a simpler approach, using the microarray described here by hybridizing stem-derived RNAs from these lines to the sugarcane genechip, and identify both ELPs and SFPs for subsequent mapping of sugar content QTLs. Furthermore, the SNPs identified in our study provided us with the opportunity to develop molecular markers within genes. So far, there is no report of SNP-based molecular markers in transcribed genes in sorghum. The SFPs generated from transcriptome studies are also useful for the development of markers in those species that lack sequence resources such as Miscanthus and switchgrass, further extending the use of microarrays of one species for related ones.

Materials and methods

Plant material

The grain sorghum lines Heilong (accession number PI 563518), IS 9738C (PI 595715), and SC 1063C (PI 595741) were obtained from the National Plant Germplasm System (NPGS), USDA. The other lines used in this study were previously described (Calviño et al. 2008). Two-week-old seedlings were harvested for the extraction of genomic DNA.

SFP discovery and validation from Affymetrix transcript data

The microarray analysis for differentially expressed transcripts in stems of grain and sweet sorghum with a sugarcane genechip was previously described (Calviño et al. 2008). The CEL files from the microarray work were uploaded into the publicly available GeSNP software at http://porifera.ucsd.edu/∼cabney/cgi-bin/geSNP.cgi, and an excel file was obtained with all the probe sets in the array harboring an SFP together with their respective t values. The excel file also contained the average hybridization intensity between the PM and MM probe pairs (average scaled PM–MM) as well as their variance values that were converted to standard deviations. These values were used to generate the graphs displaying differences in hybridization intensity between BTx623 and Rio along the 11 sugarcane probe pairs for a given probe set.

From the transcripts previously described as being differentially expressed between grain sorghum BTx623 and sweet sorghum Rio, we selected those harboring SFPs with t values ≥7 for further validation through sequencing. In total, we sequenced gene fragments corresponding to 58 different genes.

Total RNA from Rio stem tissue was extracted at the time of flowering from three independent plants. RNA extraction was performed with the RNeasy Plant Mini Kit from QIAGEN. cDNA synthesis was performed for each of the three samples from 1 μg of total RNA with the SuperScript III First-Strand Synthesis kit from Invitrogen. cDNAs from Rio were pooled respectively and used for the amplification of genes with SFPs.

The reverse transcription polymerase chain reaction products were checked by agarose gel electrophoresis in order to verify that a single band amplification product from each gene was present. The PCR products were purified with the QIAquick PCR Purification kit from Qiagen and cloned into the pGEM-T easy vector from Promega. Twelve clones per gene were sequenced in order to identify any sequencing or reverse transcriptase errors. The consensus sequence for each gene was then used to find SNPs between BTx623 and Rio.

Development of molecular markers using WebSNAPER software

Once a SNP was identified between BTx623 and Rio for a particular gene of interest, the sequence harboring the SNP in question was uploaded into the publicly available WebSNAPER software (http://pga.mgh.harvard.edu/cgi-bin/snap3/websnaper3.cgi). The SNAP procedure has been previously described (Drenkard et al. 2000). Several primer pairs per SNP were tested, and the ones that successfully distinguished the SNP in one line or the other were selected. The primer sequences used to distinguish SNPs are provided in Table 4.

Genomic DNA from 2-week-old seedlings was extracted with the PrepEase Genomic DNA Isolation kit from USB. Several concentrations of genomic DNA were tested, and 50 ng was used for testing the SNAP primer pairs through PCR. The conditions used for PCR reaction were as follows: 94°C for 2 min, then 30× [94°C 30 s, 64°C 30 s, 72°C 30 min] and a final extension at 72°C for 2 min.