The ebi-1 mutant was identified in a screen for mutants with altered temporal expression of CAB2 from an EMS-mutagenized population. The M₂ population was generated from the Ws-2 accession of Arabidopsis carrying the CAB2:LUC+ reporter construct (transgenic line 6A, Nottingham Arabidopsis Stock Centre (NASC) ID N9352). The screen involved growing plants in 12-h light/12-h dark cycles before screening LUC activity over 36 h in constant darkness 18 . The ebi-1 mutant was isolated as a plant with a 1.5- to 2-h early peak phase of CAB2 expression in constant dark (Figure 1a).

To clarify whether the early phase was the result of altered circadian clock function in the ebi-1 mutant, we analyzed CAB2 expression under constant red light. Under these conditions CAB2 expression in the ebi-1 mutant oscillated with short period (wild type (WT), 23.3 h, standard error (SE) 0.06, n = 53; ebi-1, 22.4 h, SE 0.05, n = 79; Figure 1b), consistent with the early phase of CAB2 expression in the dark. To further investigate the phenotype, we assayed circadian rhythms of leaf movement under constant white light (Figure 1c). Similarly, the leaves in the ebi-1 mutant oscillated with a shorter period than the WT (WT, 24.6 h, SE 0.11, n = 12; ebi-1, 23.5 h, SE 0.05, n = 11). Although the phenotype is subtle, it is comparable to the 1-h period difference observed for the cca1-11 and lhy-21 mutants 19 . Our data are supportive of the ebi-1 mutant perturbing multiple clock outputs. Furthermore, the ebi-1 mutation appears to affect equally the clock output in darkness (as manifested by an early phase) and light, suggesting it has a light-independent effect, and its primary defect may therefore not be in the light signaling pathway. Collectively, these results suggest that ebi-1 plays a role in the central circadian system of Arabidopsis.

To positional clone ebi-1, we took a standard approach, out crossing ebi-1 with Col-0, then re-isolating ebi-1 mutants in the F₂ mapping population. This process was very difficult for two reasons: firstly, because of the subtle phenotype of the mutant and the stochastic variation in clock timing from one individual to another, the mutant and WT clock phenotypes overlapped (Figure 1b, inset); secondly, there is more plasticity in clock function in Col-0 compared to the mutated background Ws-2 (Additional file 1). Therefore, in parallel to the mapping, we sequenced the genomes of Ws-2 and ebi-1 in an attempt to identify candidate polymorphisms.

The ebi-1 mutant was backcrossed four times with the original parent line (Ws-2 CAB2:LUC+ 6A, used to generate the EMS population) to remove EMS-induced SNPs not associated with the phenotype. Whole genomic DNA was isolated from the original parent Ws-2 CAB2:LUC+ 6A and the backcrossed ebi-1 mutant.

In total, 8 Gbp (ebi-1) and 8.5 Gbp (Ws-2, N9352) of raw color-space sequence data were generated for this study using the ABI SOLiD (version 2) sequencing machine. The number of uniquely mapping tags available for SNP calling after mapping to the Col-0 reference genome is summarized in Additional file 2 and varied between 26.7 and 39.5% of the total depending on genome and schema used. Also depending on the schema used, an average of 12.9% of the genome failed to have any tags mapping to it, which likely resulted from a combination of coverage, insertions, deletions and hyper-variable regions between Ws-2 and Col-0. In this project we focused exclusively on SNPs because insertion and deletion are not associated with EMS mutagenesis.

SNP counts before and after filtering are summarized in Additional file 3. Filtering criteria were determined empirically; working on the assumption that all loci for both mutant and WT should be homozygous, any SNP reported as heterozygous was considered, a priori, to be low confidence (an assumption confirmed by the fact that the majority occurred within obvious repeat-rich regions of the reference genome). The assumption was based on the fact that we knew that the SNP responsible for the phenotype would be homozygous. On this basis, selection criteria were identified that minimize the numbers of heterozygous SNPs, whilst maximizing the number of homozygous, and thus potentially high-confidence, SNPs. Output from the corona_lite SNP-discovery pipeline (Life Technologies, Foster city, CA, USA) provided several parameters for assessing the quality of SNP calls. We found that two parameters in particular, coverage and SNP score, when applied simultaneously to both genomes, were most effective at eliminating false positive SNPs.

By ignoring loci below a threshold coverage depth on either of the genomes being compared, we could eliminate many low-confidence SNPs. It was important to consider loci with sufficiently high coverage for two reasons: to adequately distinguish real SNPs from the ubiquitous low background of false positives generated through systematic error; and to ensure loci on both genomes were sufficiently covered to allow for SNP calling (a SNP shared by ebi-1 and Ws-2 could be mistaken for a SNP unique to one or other of these genomes if coverage in one or the other was too low).

Secondly, we found that the SOLiD SNP score provided a robust means of filtering out low-confidence SNPs. The higher the score the greater the confidence in the SNP, the score being weighted to take into account the location of the SNP within the read. Thus, SNP calls relying on more error-prone bases towards the distal end of reads were scored lower than those supported by base calls at the proximal end. The method is schematically illustrated in Figure 2.

To this end, based on an analysis of the data, only those SNPs reported where coverage exceeded 5× in both ebi-1 and Ws-2 and with a SOLiD score of 0.7 or greater were considered. We found that these cutoff values applied equally to all five of the matching schemas used.

Nevertheless, even after application of this filtering regime, examination of the remaining SNPs revealed that an unacceptably high number of low-confidence SNP calls were being reported regardless of matching schema employed (Additional file 3); interestingly, these were not the same low-confidence SNPs for each of the different schemas. Investigation revealed that the reason for this was that the different schema varied in their sensitivity to the various filtering strategies used. Thus, applying our filtering regime to schemas allowing the fewest mismatches (for example, 35_2) resulted in SNPs predominately being discarded due to too low coverage. Conversely, the same regime applied to higher mismatch schemas (for example, 35_4) led to more SNPs being eliminated due to a poor score.

The reason for this observation is clear: allowing for fewer mismatches resulted in fewer reads successfully mapping to the reference, leading to lower coverage overall, hence more loci being discarded because coverage was too low for one or other of the genomes. Conversely, accommodating more mismatches led to a higher depth of coverage, but also an increased number of SNPs called from the more error-prone proximal end and thus with poorer SNP scores.

We took advantage of this difference in filtering sensitivity to increase our filtering stringency: thus, cross-referencing results from all schemas, we identified SNPs that had high enough coverage in both genomes to be identified by low-mismatch schema, whilst at the same time having sufficiently high SNP scores to enable identification by the higher mismatch schema. The resulting SNPs are summarized in Tables 1 and 2. As a very conservative approach, we decided to cross-reference the results of all five of the schemas used (25_2, 25_3, 35_3, 35_4, 35_5). Whilst undoubtedly a highly conservative approach, with schema 25_2 in particular providing very strict matching criteria, we found that excluding the 25-mer schemas did not greatly increase the number of true SNPs whilst allowing more low-confidence SNPs. The limitation of this conservative strategy was that 11.5% of the genome had reads but failed to meet the filtering criteria and was therefore not interrogated for SNPs.

The accuracy of the SNP calling was validated using 454 sequencing. A single run of a 454-FLX sequencer (Roche) was carried out using Titanium™ chemistry on a whole genome shotgun library of the Ws-2 strain. This generated roughly 3× coverage of the genome (data not shown). SNPs were called using the Newbler read mapping software against the chromosome 5 sequence and the results compared to the SOLiD SNP calls. The software only called SNPs where there were data in the forward and reverse directions and where there were at least three reads. We only compared SNPs where the 454 phred score was ≥40 and the SNP was not adjacent to a homo-polymer. The 454 data called 15,751 SNPs at this threshold on chromosome 5; this low number reflects the reduced coverage using 454 and the scoring threshold used. Of these, 15,597 were also called using SOLiD, indicating that our SNP calls were correctly identifying at least 99% of the SNPs present between the two varieties.

To further validate our scoring and ability to accurately predict SNPs, we tested 17 SNPs between ebi-1 and Ws-2 on chromosome 5 and 4 SNPs on chromosome 1 using cleaved amplified polymorphic (CAPS) and derived cleaved amplified polymorphic (dCAPS) markers 20 . All 21 SNPs were validated. In addition, we considered five borderline SNPs, which had been filtered out because of low coverage either because they were below threshold scoring or they were not identified in all schemas. Of these borderline SNPs, four failed to be confirmed and one was heterozygous (Additional file 4). Both the 454 and the validation using CAPS/dCAPS markers together supported the accuracy of our SNP detection and our scoring and threshold setting.

Using our SOLiD data we identified 144,797 SNPs shared by Ws-2 and ebi-1 between Col-0. We also observed far fewer mutations leading to protein truncation (expected 5% under neutral selection, observed 0.4%) or amino acid substitutions (expected 65% under neutral selection, observed 44%) than predicted by chance, supporting natural selection against these types of mutations (Table 1). As the aim of this re-sequencing project was to identify EMS-induced SNPs between Ws-2 and ebi-1, we made no attempt to identify deletions or to de novo assemble sequences that failed to align with the reference. The number of SNPs we identified was far lower than that reported between Burren, Eire (Bur-0) and Col-0 (549,064) and between Tsu, Japan (Tsu-1) and Col-0 (483,352) 21 . This is likely due to the relatively close geographical proximity of Col-0 (Germany) and Ws-2 (Ukraine) on the same land mass.

To identify the EMS-induced SNPs in ebi-1, we compared the sequence generated for both lines. While 144, 797 SNPs between Col-0 and Ws-2 were shared between Ws-2 and ebi-1, 109 were unique to ebi-1 (Table 2). Based on an 8.5-Mb region of chromosome 5, we would estimate a mutation rate of approximately 1 mutation per 112 kb. This is still likely to be an underestimate as we have not considered repetitive DNA within this region. The figure closely matches previous estimates from a large-scale TILLING project using a comparable EMS dose and calculated as being 1 mutation per 170 kb 22 . We found that approximately 29.3% of mutations in genes were synonymous and 70.7% non-synonymous/nonsense, which reflects the rate expected under neutral selection. This is consistent with the fact that little selection had been placed on the plants other than their ability to set viable seed.

The EMS-induced SNPs were not spread evenly over the genome but were grouped on the north arm of chromosome 5 (76) and to a lesser extent on chromosome 1 (27) (Figure 3). The groupings, rather than a random distribution, were the result of backcrossing ebi-1 with the original parent. Rough mapping had placed the mutation on the north arm of chromosome 5 and the grouping of EMS mutations on chromosome 5 was the result of mutations 'hitchhiking' with the ebi-1 mutation during the backcrossing processes. All mutations were consistent with those expected from EMS G/C to A/T transitions 22 . However, what we had expected was that mutation types would be random, that is, equal numbers of G to A and C to T, and this was not the case. In the clustered group of EMS mutations on chromosome 5, 96% of the mutations were C to T transitions (Additional file 5), whereas 100% of the mutations on chromosome 1 were G to A transitions (Additional file 6). This is probably because the plant had arisen from germ-line cells that inherited only a single alkylated strand of DNA for each chromosome: a daughter cell of an original mutated cell line. Thus, mutations will have occurred in only one direction. In plants, previous studies have looked at bias in populations of EMS mutant plants rather than in single plants. This is also an excellent indication of the accuracy with which we are identifying SNPs and that the thresholds we have set are unlikely to have identified false positive SNPs.

Rough mapping had already confirmed that ebi-1 was located in the north arm of chromosome 5. Furthermore, using the EMS mutations on chromosome 1, backcrossed lines were identified that failed to have the EMS mutated region on chromosome 1. These lines still displayed an ebi-1 phenotype (Additional file 7); therefore, we focused on the chromosome 5 SNPs, where 32 of the 76 SNPs were non-synonymous. Based on the assumption that most clock components are themselves rhythmically expressed, we investigated the circadian expression pattern of the 32 non-synonymous SNP-containing genes using Diurnal 23 24 . We considered two transcriptomic experiments where seedlings had been entrained in 12-h light/12-h dark cycles and their gene expression then assayed in constant light 25 26 and a third where seedlings had been entrained in constant light with temperature cycles with their gene expression assayed upon transfer to constant dark 27 . We screened the temporal expression pattern of 32 SNP-containing genes, scoring an expression profile as rhythmic if it had a correlation (>0.85) with an expression pattern model consistent with circadian regulation (Additional file 8). Only one SNP-containing gene was robustly rhythmic in all our tested conditions, PSEUDO RESPONSE REGULATOR 7 (PRR7, At5g02810; 0.95 correlation with a circadian time (ct) 7-h spike and 0.93 correlation with a ct 6-h spike in the constant light data sets, and a 0.87 correlation with a ct 6-h spike in the constant dark data set. A second gene, AtNFXL-2 (At5g05660), a zinc finger transcription factor, was not rhythmic in constant light but had a 0.91 correlation with a sine wave in constant dark and was therefore a strong potential candidate. Two other genes, At5g19850, a predicted hydrolase, and At5g12470, an organelle protein of unknown function, had good correlation with a cosine wave but only in one set of the constant light data. All other genes failed to show rhythmic patterns of expression.

The obvious strong candidate was the non-synonymous SNP in PRR7. Sanger sequencing and a dCAPS marker were used to validate the SNP. The gene PRR7 has already been shown to play a key role in the circadian clock, with the T-DNA insertion mutant prr7-3 causing a lengthening of the circadian period 28 , opposite to the affect of ebi-1. The point mutation in PRR7 in ebi-1 caused an R to be substituted with an H. However, the amino acid did not lie in a functional domain and was not conserved across species; in fact, in Brassica napus, the endogenous PRR7 has an H at this position (Additional file 9).

The other strong candidate SNP, based on the circadian regulation and molecular function, was in AtNFXL-2. The mutation caused a C to T transition, which was confirmed by Sanger sequencing and a dCAPS marker. The AtNFXL-2 protein shares homology with the mammalian zinc finger transcription factor NF-X1 29 . Arabidopsis has two NF-X1-like genes, AtNFXL-1 (At1g10170) and AtNFXL-2 (At5g05660) 30 . No previous study has suggested a role for the AtNFXL genes in the circadian clock. The SNP resulted in an amino acid substitution (V to I) in the gene At5g05660. The valine is relatively conserved across species and is either valine or methionine and lies within a zinc finger motif (Figure 4). However, in the Arabidopsis homolog, AtNFXL-1, the residue is a leucine.

From our functional genomics analysis two clear candidate SNPs remained. Based on the location of the SNP in a conserved domain, AtNFXL-2 was a strong candidate. We used SNP markers for AtNFXL-2 and PRR7, identified by our re-sequencing of ebi-1, to screen a backcrossed ebi-1 F₂ population to identify recombinant individuals. To exclude the mutation in PRR7, we identified two lines (ebi-1-clean-1 and ebi-1-clean-2) that contained the AtNFXL-2 SNP but were WT for the PRR7 gene. We then identified a further two lines (prr7-clean-1 and prr7-clean-2) that were WT for AtNFXL-2 but retained the PRR7 SNP. We analyzed CAB2 expression under constant red light in all the lines. Both ebi-1-clean-1 and ebi-1-clean-2 had phenotypes identical to the original ebi-1 mutant while prr7-clean-1 and prr7-clean-2 had almost WT phenotypes, thus demonstrating that the mutation in PRR7 does not contribute significantly to the ebi-1 phenotype (Figure 5a). Furthermore, by combining new mapping data with SNP information, we were able to further narrow down the candidate SNPs to the AtNFXL-2 SNP, which lies between molecular markers nga158 and CIW18, thus excluding PRR7.

Finally, a T-DNA insertion line was ordered, SALK_128255.54.50.n, which contains a T-DNA inserted in the promoter region of the EBI gene (ebi-2). The insertion does not stop EBI expression but it significantly reduces the expression level (Figure 5d). A homozygous T-DNA line was transformed with the CAB2:LUC+ reporter gene and the circadian phenotype of transformed lines analyzed. Like ebi-1, ebi-2 had a short period in constant light (WT, (Col-0) 26.74 h, SE 0.17, n = 27; T-DNA line, 25.67 h, SE 0.44, n = 28; Figure 5b) and peaked early in constant dark (Figure 5c).

Experiments were carried out with ebi-1 that had been backcrossed four times to the parental transgenic line 6A carrying the CAB2:LUC+ reporter construct (NASC ID N9352).

The T-DNA line SALK_128255.54.50.n was obtained from NASC and plants homozygous for the T-DNA were confirmed by PCR using primers 5'-ttgccgcagtaacaaaggtac-3', 5'-agtttatccggaagcaaatgg-3' (WT band in Col-0, no band in homozygous SALK line). The left border sequence was amplified with 5'-agtttatccggaagcaaatgg-3' and LBb primer. CAB2:LUC+ was introduced using Agrobacterium-mediated transformation and dipping protocol 36 .

The mutagenesis and screening have been described in 18 . Briefly, Arabidopsis Ws-2 transgenic seeds carrying the CAB2:LUC+ transgene (described above) were mutagenized by soaking in 100 mM EMS for 3 h. The resulting M₁ population was sown and self-fertilized, and the M₂ population was screened for seedlings with altered timing of CAB2:LUC+ expression in constant darkness.

Seedlings were then sown on Murashige and Skoog medium containing 3% sucrose and 1.5% agar. They were entrained in a growth chamber in light/dark cycles at 22°C for 7 days before transfer to constant light and temperature. Two methods where used to measure CAB2:LUC+ activity. For the initial screen and preliminary characterization of the mutant in constant dark an automated luminometer was used (Topcount, Perkinelmer, Cambridge, UK)as described 37 . The second method for the characterization of the mutant in constant light and subsequent characterization of backcrossed lines and T-DNA mutants was a low-light video imaging system as described in 37 . The method for measuring rhythms in leaf movement used older 12-day-old seedlings and a method identical to that described in 38 .

DNA was isolated using a plant DNeasy kit (Qiagen, Crawley, West Sussex, UK) Two read tag libraries were prepared, one for ebi-1 and one for Ws. Emulsion PCR using the standard SOLiD protocol was performed on each library. The libraries were deposited onto separate slides and sequenced in a single run using the SOLiD analyzer version 2 (Life Technologies).

For the 454 genome sequencing, 5 μg of Ws-2 DNA was fragmented by nebulization. Fragmented DNA was analyzed using a Bioanalyzer (Agilent Technologies, Wokingham, Berkshire, UK)to ensure that the majority of the fragments were between 350 and 1,000 bp. The purified fragmented DNA was processed according to the 454 FLX Titanium Library construction kit and protocol (Roche Applied Science, Burgess Hill, East Sussex, UK). Library fragments were added to emulsion PCR beads at a ratio of 1:1 to emPCR at the optimal of 1.5 DNA molecules per bead and amplified according to the manufacturer's instructions (Roche Applied Science) and a full pico-titre plate was sequenced.

The resulting 35-character color-space tags from both sequencing runs were then mapped to the 119.7 Mbp Col-0 reference sequence 39 using the matching pipeline of the off-machine SOLiD data analysis package Corona Lite 40 employing a range of matching schemas, based on the full-length 35-character color-space tags as well as schemas based on tags trimmed to 25 characters to remove the most error-prone positions. Putative SNPs relative to Col-0 were then called for each genome using Corona Lite's SNP detection pipeline.

The resulting SNP list for ebi-1 was then cross-referenced with that of Ws-2 to identify SNPs shared by both genomes, as well as SNPs occurring only in ebi-1 or only in Ws-2. At this stage low-confidence SNPs were filtered out by excluding all SNP loci where coverage was 5 or less, SOLiD SNP scores were less than 0.7, or the SNP was heterozygous, in either genome. To ensure only high-confidence SNPs were considered, a further screening round was undertaken in which only those reported by all matching schemas employed were considered for subsequent analysis.

Using current (TAIR 8) annotations 39 as a guide, high-confidence SNPs were classified and enumerated. The sequence data for Ws-2 are archived at TAIR and available as a track on the Arabidopsis genome hosted at TAIR [SpeciesVariant:393] 41 .

To validate the SNPs between ebi-1 and Ws-2, we used a simple PCR-based approach of CAPS and dCAPS analysis. PCR primers for CAPS/dCAPS analysis were designed using dCAPS finder 2.0 42 . A standard PCR protocol was used to amplify products from ebi-1 and Ws-2, and the PCR products were digested and run on a 4% agarose gel and scored. The primers, restriction sites and product sizes are summarized in Additional file 4. The SNPs in PRR7 and EBI were further validated by standard sequencing methods.

Seedlings were grown under 12-h light/12-h dark cycles for 6 days. Seedlings were harvested directly into liquid nitrogen at 1 h after dawn and 1 h after dusk using a green safety light. The RNA was subsequently extracted using an RNeasy Plant Mini Kit (Qiagen, Hilden, Germany). cDNA was synthesized from 1 μg of total RNA using the iScript™ cDNA synthesis kit (Bio-Rad Laboratories, Inc., Hercules, CA, USA). Real-time PCR was performed with a MyIQ™, ICycler or CFX96 Real-Time PCR Detection System (Bio-Rad Laboratories, Hempstead, Hertfordshire, UK), using iQ SYBR^® Green Supermix (Bio-Rad Laboratories). The efficiency of amplification was assessed relative to β-TUBULIN (βTUB) expression. The measurements were repeated at least two times with independent biological material. Expression levels were calculated relative to the reference gene using a comparative threshold cycle method 43 . The results show the mean of four biological replications, each with three technical repeats, and expressed relative to the mean of the wild-type series after standardization to βTUB. Primers for βTUB have been published previously 44 . The EBI-specific primers were as follows: EBI-F, 5'-TGC GAG AAT ATG CTT AAT TGC-3'; EBI-R, 5'-CCA CAA CAT CAC AAG ACA AG-3'.

An F₂ mapping population was made between ebi-1 and Col-0. A set of approximately 20 individuals from this population, which had their ebi-1 phenotype confirmed in the F₃, had recombination events in chromosome 5 and placed the ebi-1 mutation on the north arm of chromosome 5. This mapping population was increased and with two individuals we were further able to limit the mapping interval to between CIW18 and nga158.