The characteristics of microsatellite occurrences were surveyed among the different genomic regions in the Arabidopsis genome. It was obvious that microsatellites were found to be highly abundant in the regulatory regions, and the over-representation of CT/GA and CTT/GAA repeats contributed most to the increase of microsatellites in these regions (Figure 1A, 1B). This preference of CT/GA and CTT/GAA repeat occurrences indicated that they might have the role in regulating genes.

Regulatory sequence elements within promoter DNA are often short, orientation independent and contain frequent gaps of variable size. Thus, we determined the conserved noncoding microsatellite sequences as candidate regulatory elements based upon the following criteria: that there were at least 6-bp overlapping regions of the corresponding microsatellites between the aligned sequences. According to the criteria, we identified 247 Arabidopsis-Brassica orthologous CNMSs and 122 Arabidopsis paralogous CNMSs [see Additional file 1], involving 491 CT/GA and CTT/GAA repeats respectively (Table 1), which accounted for 10.6% of these types located in the 500-bp regions upstream of coding sequences in the Arabidopsis genome. These CNMSs do not randomly occur in different noncoding regions and they tend to be found more frequently near the initiation codon (Figure 2A, 2B).

In order to validate the above study and to ensure that the observation of CNMSs was not simply due to its over-representation in plant genomes, a similar analysis was carried out on three different random datasets, i.e. the 1000 homologous pairs of 5' noncoding sequences in Arabidopsis as dataset 1, the 1000 randomly shuffled pairs of 5' noncoding sequences as dataset 2 and the 1000 random pairs of genomic DNA sequences as dataset 3, as well as the three corresponding datasets of Arabidopsis and Brassica sequence pairs with the same data size. Figure 3 showed the frequencies of CNMS (CT/GA)_n and (CTT/GAA)_n in dataset 1, dataset 2 and dataset 3, respectively. Obviously, there was very little probability that CNMSs were found in the 500-bp genomic DNA sequence pair by chance. In contrast with the random pairs of noncoding sequences, the homologous noncoding sequences showed significant high in the frequency of CNMS occurrences. Taken together, these tests indicated that some microsatellites in regulatory regions were conserved from common ancestors during plant evolution.

To gain insight into the evolutionary relationship of Arabidopsis-Brassica and Arabidopsis-Arabidopsis CNMSs, the synonymous substitution rate (Ks) was calculated for the corresponding gene pairs. For Arabidopsis-Brassica orthologous CNMS gene pairs, the frequency distribution showed a clear peak for Ks values of 0.4 to 0.5 (Figure 4A), suggesting that these CNMSs were conserved from a common ancestor over a 15 million years (Myr) period, which was consistent with the divergence time frame estimated at 14.5 to 20.4 Myr based on mitochondrial DNA data 23. On the other hand, we noticed two peaks in the Ks distribution of Arabidopsis paralogous CNMS gene pairs, and the Ks values were 0.8 to 0.9 and 1.2 to 1.3, respectively (Figure 4B). The former group contained most of the paralogous CNMSs which were originated from large scale gene duplication over 28 Myr ago, which was consistent with the recent polyploidization event during evolution of the Arabidopsis genome 24. The latter group were duplicated from the common ancestor over 42 Myr ago, which probably occurred at the time of the divergence of brassicaceae family 25.

The results from the evolutionary relationships of Arabidopsis-Brassica and Arabidopsis-Arabidopsis CNMSs suggested that most paralogous CNMSs pre-dated the divergence of the two species; hence, many paralogous CNMSs in Arabidopsis were likely to find their counterparts in Brassica. Further comparisons of paralogous and orthologous genes from Arabidopsis and Brassica were made for common CNMSs (Figure 5A, 5B). With the same criteria, we identified 18 CNMSs found in Arabidopsis paralogous pairs that also were coincident with CNMSs from at least one orthologs in Brassica (Table 2). We called these conserved elements, shared among paralogous and orthologous genes, Ultra-CNMSs. An example of such Ultra-CNMSs was shown in Figure 5C, and the three homologous CT repeats were highly conserved from a common ancestor over 48 Myr.

As expected, analysis of regulatory regions of related gene families revealed that many Ultra-CNMSs were conserved across a number of more distantly homologous genes in Brassicaceae species and other plants. Figure 6A showed that CNMSs (CT)_n were conserved among orthologous genes from Arabidopsis, Brassica, Medicago and rice, as well as among more distantly paralogous genes in Arabidopsis. These genes are representatives of a larger family of transmembrane receptor kinases and related non-transmembrane kinases in plant genomes. Many of them arised from a common ancestor of dicots and monocots. Another striking CNMS was found in the regulatory regions of GATA transcription factor genes from Brassica, Arabidopsis and rice. Of 14 members in subfamily I in the Arabidopsis genome 26, five of them have the same CNMSs found in their regulatory regions (Figure 6B). These CNMS associated transcription factor genes that fell into two subgroups indicated they diverged before the dicotyledonous and monocotyledonous plants 26. It was obvious that these Ultra-CNMSs had been passed down from a common ancestor of dicots and monocots under extreme purifying selection for more than 170 Myr 27.

We tested whether CNMSs in genes were influenced by the function of the proteins they encode. There were 206 Arabidopsis-Brassica and 194 Arabidopsis-Arabidopsis CNMS associated genes with known function in the Arabidopsis genome. We looked for categories of biological process and molecular function defined in the Gene Ontology (GO) database that were significantly enriched or depleted in these genes 2829. These CNMS associated genes showed significant functional enrichment for transcription factor activity (P < 1.8 × 10^-7 for orthologous CNMS genes, and P < 7.7 × 10^-7 for paralogous CNMS genes, against all GO annotated Arabidopsis genes) and transcription (P < 4.5 × 10^-4 for orthologs and P < 6.6 × 10^-5 for paralogs) (Figure 7), and genes that performed the functions of the two types accounted for about 23% of all known genes. However, they were obviously depleted for DNA metabolism (P < 2.2 × 10^-4 for orthologous CNMS genes, and P < 1.8 × 10^-2 for paralogous CNMS genes). These findings suggested that CNMSs might be specifically associated with regulation of transcription at the DNA level, but not involved in DNA metabolism.

To further investigate the regulatory nature of these CNMSs, we employed a computational method to discover cis-elements that were similar to function assigned elements based on the PlantCARE 3031 and PLACE databases 3233. The identification of cis-elements showed that some binding sites were clustered in the consevered microsatellite regions, and these regulatory elements were involved in plant-specific functions in response to some environmental stimuli (Table 3). The CNMS (CT)_n include the TCTCtCT sequences similar to the TCCC motif known as part of conserved DNA module array AtpCD-CMA involved in light responsiveness 3435. Another function of CNMS (CT)_n may be as an enhancer due to the same motif (TCTCTCTCT) found in a 60-nt region downstream of the transcription start site of the CaMV 35S RNA, which can enhance gene translation in plant protoplasts 36. As complementary sequences to (CT)_n, (GA)_n serve as regulatory element having similar functions, which contain a series of overlapped GAG motifs (AGAGAGa) involved in light regulation 3537. In soybean, it is clear that the 18-bp GAGA element sequence within the Gsal promoter can be recognized by GBP encoded by a light-regulated gene 12. The CNMS (CTT)_n contain sequences similar to the TCA-element (TCATCTTCTT) which is a binding site for salicylic acid-inducible proteins 38. Similarly, the CNMS (GAA)_n contain AGAA sequences having the characteristic of the core recognition sequence (tcAGAAgagg) for salicylic acid-responsive genes 39.

Although CNMSs (CT)_n/(GA)_n and (CTT)_n/(GAA)_n are similar to known regulatory elements, most of them have no experimental verification for their functions. Therefore, all CNMS (CTT)_n/(GAA)_n associated genes were selected to investigate their changes in expression levels after the treatment of salicylic acid. The abundance of gene transcripts evaluated by the MPSS showed these CNMS (CTT)_n/(GAA)_n associated genes had distinct expression characters with salicylic acid treatment (Figure 8A, 8B). About 70–80% of CNMS (CTT)_n/(GAA)_n associated genes in Arabidopsis leaves were regulated by salicylic acid, while others were undetectable with and without salicylic acid treatment. Among these salicylic acid-responsive genes, only about 15–23% of them were up-regulated by salicylic acid, and most of them were inhibited after the treatment. Seven CNMS (CTT)_n/(GAA)_n associated genes were additionally analyzed for expression patterns after salicylic acid treatment. The RT-PCR showed that these investigated genes, excepted At2g05920 and At5g67360, were obviously down-regulated by salicylic acid (Figure 9), which were consistent with the patterns of gene expression from the Arabidopsis MPSS database. According to the expression patterns by RT-PCR, we found the preliminary correlation between repeat number of CTT/GAA motif and gene in response to salicylic acid. The (CTT)₄/(GAA)₄ sequences were associated with gene down-regulation with salicylic acid stimulus, but the (CTT)₅ and (CTT)₇ associated genes were not obviously regulated by salicylic acid. These findings implied that regulation of CNMS associated gene expression by salicylic acid might be dependent on the number of CTT/GAA repeats.

The Arabidopsis plants were grown in soil in a growth chamber at 20°C with 8 hours of light for 40 days. Plants were sprayed to run-off with 1 mM salicylic acid in 0.5% dimethyl sulfoxide (DMSO) for different time scales. One, four, twelve and forty-eight hours post-treatment, leaves were cut and harvested respectively, quick-frozen in liquid nitrogen, then stored at -80°C. Total RNA was later extracted using Plant RNA Mini Kit (Watson Biotechnologies INC., China).

The annotated sequences of the five chromosomes of Arabidopsis (accession numbers: NC_003070, NC_003071, NC_003074, NC_003075, and NC_003076, updated 25-JAN-2005) were downloaded from the Genomes Division of GenBank 4546. Intergenic regions were defined as being a part of DNA from the end of the last exon of one gene to the beginning of the first exon of the following gene. A set of 16223 full-length cDNA sequences containing both 5' and 3'UTRs for Arabidopsis was extracted from the TAIR database 47. The preliminary sequences of Brassica genome were obtained from The Institute for Genomic Research website 48.

To identify putative Arabidopsis-Brassica orthologous gene sets, each preliminary sequence from Brassica was searched against 1-kb sequences (fragments from the position -500 to +500 relative to the translation initiation) of all genes from Arabidopsis using BLASTN 49 and then the fragments from Brassica were clustered according to the best match gene of the Arabidopsis genome. Conversely, each 1-kb gene sequence from Arabidopsis was searched against the contigs from Brassica. Two sequences were defined as orthologs if each of them was the best hit of the other in the aligned regions and if the expect value (E) was <le-10. A list of the identified Arabidopsis-Brassica orthologs in the study is provided as supplementary data [see Additional file 2].

For identifying the paralogous gene pairs from a recently common ancestor in the Arabidopsis genome, each annotated coding sequence was searched against all other coding sequences using BLASTN. The best pair was considered significant if each of them was the best hit of the other and the expect value was <le-10. A file of the list of the paralogous gene pairs is included as supplementary data [see Additional file 3]. To avoid the negative conservation of microsatellites caused by the effects of insufficient randomizing mutations, the tandemly repeated gene pairs separated by less than 25 intermediate genes were ignored in further analysis.

Microsatellites were found in sequences using the modified Sputnik repeat-finder 50. Di-and trinucleotide repeats were identified when a total size of at least 12-bp, allowing up to about 10% deviation from a perfect repeat. Repeat motifs consisting of different frames (e.g. GAA, AGA and AAG) were regarded as the same type of repeat.

Because gene fragments of Brassica were derived from preliminary contigs with no annotated open reading frames, each pair of Arabidopsis-Brassica sequences were aligned using DiAlign2 with translation option to identify the 5' noncoding sequences and coding regions in the Brassica orthologs 51. The 5' noncoding sequence pairs were aligned using DiAlign2 for finding conserved microsatellites. To exclude nonspecific alignments, a stringent threshold parameter of 3 was used. The CNMSs were identified when the corresponding loci had at least 6-bp overlapping sequences between the aligned microsatellite sequences.

To ensure that CNMSs were not to occur by chance, we used two different datasets of random pairs as negative controls to validate the results. One control dataset contained 1000 random pairs of 500-bp upstream noncoding sequences in the Arabidopsis genome, and another control dataset of 1000 pairs was randomly generated from the 500-bp sequences of Arabidopsis genomic DNA fragments. The reference dataset of equal data size was randomly selected from the 500-bp paralogous noncoding sequence pairs in Arabidopsis. The 1000 paralogous pairs, the 1000 shuffled pairs of noncoding sequences and the 1000 random pairs of genomic sequences were respectively referred as dataset 1, dataset 2 and dataset 3 in further analysis. Similarly, three corresponding datasets of Arabidopsis and Brassica sequence pairs were generated with the same data size. The dataset 1 consisted of 1000 Arabidopsis-Brassica orthologous pairs of 5' noncoding sequences, and the dataset 2 contained 1000 random pairs of Arabidopsis and Brassica upstream noncoding sequences, and the dataset 3 of 1000 pairs was randomly generated from the Arabidopsis and Brassica genomic DNA sequences. The same criteria of CNMS detection was applied in the test. Occurrences of CNMSs were analyzed in analogous manner for 10 different random sets with equal data size.

We used the level of synonymous substitution of CNMS associated coding sequences to estimate the Ks of CNMSs. For each pair of CNMS associated genes, the two protein sequences were aligned by ClustalW, and the resulting alignment was then used as a guide to align the nucleotide sequences 52. After removing gaps, the level of synonymous substitution was estimated using the yn00 program in PAML 23. The time of divergence (T), between two sequences was calculated from this as T = Ks/2λ, where Ks is the fraction of synonymous substitutions per synonymous site and λ is the mean rate of synonymous substitution. The estimate value for λ in dicots is 1.5 synonymous substitutions per 10⁸ years 53.

Gene expression level was estimated using the data from the Massively Parallel Signature Sequencing (MPSS) database of Arabidopsis 5455. The MPSS data of three different libraries was generated from untreated leaves and treated leaves 4 and 52 hours after salicylic acid treatment, respectively. For the three libraries, a total of 9,081,200 17-bp signatures were obtained in multiple sequencing runs and in two sequencing frames. The abundance for each signature was normalized to transcripts per million (TPM) to facilitate comparisons across libraries.

RT-PCR of Arabidopsis CNMS associated genes was conducted using the one-step RNA PCR kit (TaKaRa) with gene specific primers [see Additional file 4]. The 0.5 μg total RNA was used as the template to be amplified with the following program: an initial 50°C for 30 min and 94°C for 2 min, followed by 25 cycles of 94°C for 30s, 54°C for 30s and 72°C for 1 min. The house-keeping gene actin2 (At3gl8780) was used as an internal control in RT-PCR reaction.