By applying higher density SFP analysis than previously, a total of 1208 SFP probes were detected in the present analysis (Figure 1, Additional file 1). Plots of the log intensities, affinity differences and individual outlying scores for a representative probe set (Os.33510.1.S2_at) are shown in Figure 2. The intensity differentiation between Pokkali and FL478 is highest at probes 4 and 3, indicating polymorphism at these probe positions. A representative alignment of the amplicon sequences with the target sequence of Os.33510.1.S2_at probe set is shown in Figure 3. Several SNPs were detected, but only probe positions 3 and 4 span a SNP. Probe 4 was selected as a SFP by the RPP method based on a higher outlying score than that of probe 3 (Figure 2).

We explored the source of the Saltol region in FL478 because several reports demonstrated the importance of this region for salt tolerance, and because our prior report 3 suggested that the Saltol region of FL478 may have originated from the salt sensitive parent. Bonilla et al. (2002) 8 initially delimited Saltol as a QTL controlling three traits (low Na⁺ absorption, high K⁺ absorption and low Na⁺/K⁺ ratio) within a 15 cM segment of the rice genetic map with peak LOD score > 6.7 (Figure 4). A major QTL for high shoot K⁺ concentration under salt stress also was identified in the same region 11. More recently, Ren et al. (2005) identified the SKC1 gene encoding a sodium transporter and demonstrated that it is a determinant of salt tolerance in the Saltol region 12.

In prior work we reported that all of the SFPs detected in the Saltol region of FL478 were consistent with an IR29 origination (salt sensitive parent) 3, indicating either that FL478 received its salt tolerance from other QTL or that we did not have sufficient SFP marker density in this region to detect a small region of the genome from the salt tolerant parent. Subsequent to the Walia et al. (2005) work 3, we extended the list of SFPs to examine the Saltol region in more detail. This was accomplished by: 1) considering all probe sets including those with "_s", "_x" or "_a" in the probe set name in order to give higher SFP density and 2) updating the gene model annotations available from http://www.tigr.org/tdb/e2k1/osa1. An explanation of these suffixes is in the Affymetrix GeneChip design manual, which is available from the Affymetrix website. The suffix "_at" at the end of every probe set means antisense transcript. A lack of another suffix means that all probes in the probe set are unique to the particular sequence used for the array design. The "x" indicates that at least one probe is a perfect match to another sequence. The "a" indicates that all probes are a perfect match to another sequence in the same gene family and the "s" indicates that all probes are a perfect match to a sequence in another gene family.

These actions revealed additional SFPs in the Saltol region, increasing the total to 21 SFPs among which one corresponding to gene model LOC_Os01g20120 was identical to the Pokkali allele (Table 1, Figure 4), not IR29. This gene model is adjacent to the SKC1 gene (LOC_Os01g20160) which as stated above is known to be a salt tolerance gene 12.

In order to confirm the SFPs detected in the Saltol region, we examined the SFP locations by amplicon sequencing. Alignments of the amplicon sequences are shown in Figure 5. For probe set Os.55011.1.S1_x_at, which corresponds to gene model LOC_Os01g20120, one SNP was found in the amplicon sequence at the SFP probe position and the FL478 allele was the same as in the Pokkali genotype. These results confirmed the SFP detection data, which suggested that FL478 contains a Pokkali-derived gene near SKC1 (LOC_Os01g20160). To further examine this region we checked additional genes in the vicinity of LOC_Os01g20120. We found that three additional genes (LOC_Os01g19220, LOC_Os01g19400, and LOC_Os01g20160 [SKC1]) within a < 1 Mb segment of FL478 also are of a non-IR29 origination (Figure 6). Taken together, it appears that FL478 contains a small non-IR29 haplotype block including SKC1 (Figure 4C), which we did not detect previously. We could not detect a SFP in the SKC1 gene in either the previous work or the present study because the expression level from the probe set (Os.30563.1.S1_at) for SKC1 was not "present" in all expression datasets, which is a requirement of our statistical filtering method. The SKC1 sequences are shown in Figure 6C. Surprisingly, in an apparently highly variable region, FL478 contains a haplotype that is not identical to either of the presumed parents. We confirmed this by sequencing amplicons from independent reactions from each genotype, making use of high fidelity Taq polymerase (Platinum pfx DNA polymerase, Invitrogen, USA). The existence in FL478 of an allele that matches neither IR29 nor the genotype which we know as Pokkali could be explained by either parent being genetically not uniform when the crosses to make RILs including FL478 were made. This notion is consistent with records now showing that there are actually at least eight distinct accessions named Pokkali in the germplasm collection at International Rice Research Institute http://www.iris.irri.org/.

We examined a total of 64 putative SFPs by amplicon sequencing (Additional file 2). Among them, 62 were found to cover polymorphisms (~97% validation). Among these 62 confirmed SFPs, 51 (82.2%) were positioned over a single SNP, seven (11.3%) were positioned over an INDEL, two (3.2%) spanned one SNP and one INDEL, one (1.6%) spanned > 1 SNP and no INDEL, and one spanned > 1 SNP and > 1 INDEL. From this we assert that at the threshold of top 20 percentile outlying scores, our detection method is correct about 97% of the time (2 false positive in 64) in a priori identification of SFPs from the Affymetrix rice genome array data using RNA-based datasets. Winzeler et al. (1998) identified more than 3,000 polymorphisms between two yeast strains at a 5% error rate using DNA hybridization 4. Also, about 1,000 SFPs were identified at 3~7% error rates in yeast using mRNA hybridization 5. In Arabidopsis, among 3,806 predicted SFPs, 97% of known polymorphisms were detected, which established a false negative rate of 3% 1. Rostoks et al. (2005) used a probe level analysis of transcriptome data in barley to identify 10,504 putative SFPs, which included ~40% false positives 13. More recently, rice genomic DNA was hybridized to an oligonucleotide microarray to detect SFPs 6 with an up to 20% false discovery rate. The 97% validation rate (3% false positives) from our method of RNA-based SFP detection by RPP compares favourably to these other performance metrics.

In the single nucleotide polymorphism database (dbSNP) of the National Center for Biotechnology Information (NCBI), more than 5 million polymorphisms including SNPs, small INDELs and microsatellite repeat variations have been catalogued. Also, the International Rice Research Institute has initiated a project to identify a large fraction of the SNPs in germplasm pertinent to cultivated rice through whole-genome comparisons 14. This will provide additional millions of rice SNPs. Our work has shown that the existing Affymetrix rice genome array can be used to provide some thousands of SFP markers from a pairwise rice genotype comparison. Because a number of researchers have been using Affymetrix microarrays for transcriptome analyses in a range of rice RILs, NILs and germplasm accessions, existing data files provide abundant opportunities for the identification of additional SFP markers and resolution of trait determinants without additional expenditure on materials or data acquisition. Therefore, application of the RPP method to existing data could augment, or sometimes obviate the need for, other markers to meet objectives such as map-based cloning and sub-Mb resolution of the position of trait determinants. Examples of such applications would be to define introgressed regions in NILs or to generate moderate density linkage maps from RIL populations. Also, SFPs can provide a reliable discovery component in the development of markers for other detection systems including SNPs, CAPS, DArT, and SSRs.

Seeds of rice (Oryza sativa) genotypes Pokkali, IR29 and FL478 were obtained from G. B. Gregorio at the International Rice Research Institute in the Philippines and then propagated at the USDA/ARS George E. Brown, Jr., US Salinity Laboratory in Riverside, CA. Seedlings of the three genotypes were grown and stored at -80°C until DNA extraction.

Genomic DNA was extracted from seedlings of the three genotypes using a DNeasy Plant Mini Kit (Qiagen, USA) according to the manufacturer's protocol. For each genotype, more than seven seedlings were ground and about 0.1 g of pulverized tissue was processed. Purified genomic DNA was quantified at 260 nm using a spectrophotometer.

We produced RNA expression data using the Affymetrix rice GeneChip hybridized with cRNA synthesized from shoot tissue RNA of young seedling of three rice genotypes with and without salt stress, essentially as described previously 3. The dataset was from seven chips with Pokkali RNA, five chips with IR29 and six chips with FL478. The Affymetrix rice GeneChip consists of probe sets designed for 48,564 japonica and 1,260 indica sequences http://www.affymetrix.com/. For SFP detection, we applied the RPP method to each probe set that had a "present" call in all chip samples from each pair of genotypes under comparison: (1) Pokkali versus IR29, (2) Pokkali versus FL478, (3) IR29 versus FL478. Using the top 20 percentile of all overall outlying scores as a cutoff, SFP probes were compiled. FL478 alleles presumed to be inherited from IR29 were then obtained as the SFPs detected in comparisons (1) and (2) but not (3). Similarly FL478 alleles presumed to be from Pokkali were obtained as the SFPs detected in (1) and (3) but not (2). As described in Cui et al. (2005) 2, the RPP method first measures the overall outlyingness of each probe set. Probe sets with significantly high outlying scores are then analyzed at the probe level and the probes that make a sufficiently large contribution to overall outlyingness of the probe set are identified as SFP probes.

We obtained the target sequence of each probe set from the sequence information file (SIF) for the Affymetrix rice genome array http://www.affymetrix.com/. The target sequence corresponds to the 5' end of the 5'-most probe to the 3' end of the 3'-most probe. To obtain the corresponding indica rice genomic sequences, each target sequence was searched using BLASTN against the indica rice whole genome shotgun sequences in the NCBI database http://www.ncbi.nlm.nih.gov/BLAST/Genome/PlantBlast.shtml?10. The indica sequences (cv. 93-11) were aligned with the target sequence using AlignX in Vector NTI Advance 10 (Invitrogen, USA). HarvEST:RiceChip 15 was used to check the position of SFP probes in each target sequence. Primers were designed using Primer3 http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi/16. The primers are listed in Additional file 3.

PCR was performed in 20 μl containing 25~50 ng of genomic DNA, 0.1 μM of specific primers, 0.2 mM dNTPs, and 1 unit of Taq (GenScript Corp., USA) DNA polymerase. The reaction included a 5 min denaturation at 95°C followed by 35 cycles of PCR (94°C, 30 sec; 55~65°C, 70 sec; 72°C, 60 sec), and a final 5 min at 72°C. Aliquots (4 μl) of the PCR products were separated on a 1.2% agarose gel to check the band size and quantity. PCR products were purified using QIAquick PCR purification Kit (Qiagen, USA) to prepare for sequencing.

DNA sequencing was performed by the dideoxynucleotide chain termination method 17. The amplified PCR products (amplicons) were sequenced with an ABI-PRISM 3730×l Autosequencer (ABI, USA). These sequences were then compared with the target sequence of each probe set using AlignX (Invitrogen, USA). Comparisons of nucleotide sequence similarity were displayed using GeneDoc 18. Rice genomic amplicon sequences have been deposited in the GenBank Data Library under accession numbers [GenBank:EF589163–EF589342 and EU099042–EU099056].