The Myb gene family has broadly expanded in plants during evolution. The amplification of the Myb gene family occurred before the divergence of monocots and dicots 12. In our study, 130 Myb genes were found in the Arabidopsis genome and 85 in Oryza sativa L. ssp. indica. The large size of this gene family was also confirmed in Zea mays and sorghum 12. Although most plant Myb genes contain only two repeats, there have been three-repeat Mybs reported in Arabidopsis 1, maize 13 and other plants 12. To date, only three-repeat Myb genes have been detected in animals, and it has been proposed that two-repeat Myb genes died out in the animal lineage 12. The broad presence of three-repeat Myb genes in diverse species indicates the antiquity of these genes. Using homology search in the GenBank non-redundant database, two three-repeat Myb proteins (accession numbers NP_913483 and BAC79618.) were identified in Oryza sativa L. ssp. japonica. However, no three-repeat Mybs were detected in rice (indica) in our study. This could be due to the incompleteness of the Oryza indica dataset.

On the basis of sequence similarity and the topology of the phylogeny, we clustered the Myb genes into 42 subgroups, ranging in size from two to 14 Myb genes (Figure 1). The phylogenetic topology and subgroup structures are consistent with previous reports 110. The detailed comparison is described in Additional data file 1 (also available with all other supplementary material at 14). However, because of the large number of taxa, the bootstrap values are low (data not shown). Therefore, we sought other evidence to support the reliability of the subgroup designations.

Interestingly, AtMyb33, 65, 101, 104 and At3g60460 were complementary, with few mismatches, to Arabidopsis Myb microRNA (noncoding RNA) miR159 15. The sequence is 21 nucleotides long and located in the 3' untranslated region (3' UTR) of all five genes. MicroRNAs are proposed to act as regulators of gene expression through interactions with complementary mRNA sequences. Importantly, these five Arabidopsis Mybs are located in subgroup G12 (Figure 1). This clustering provides additional evidence for the reliability of the subgroup designations in our analysis.

The phylogenetic topology and subgroup structures are based on sequence comparisons of the complete predicted Myb genes. To test the reliability of the subgroup designations using independent criteria, we investigated the exon-intron structure of Myb genes subgroup by subgroup. A majority of Arabidopsis (59%) and rice (53%) Myb genes have a conserved splicing pattern of three exons and two introns in R2R3 domains (represented by subgroup G13; Figure 2a). Either or both of the two introns are absent in 19% of Arabidopsis Myb genes and 12% of rice Myb genes. Variable splicing patterns different from G13 were detected in 22% of Arabidopsis and 35% of rice Myb genes, respectively (data not shown). Strikingly, the exon-intron structure is conserved within each subgroup, but varies between subgroups (Figures 2b,c). This supports the subgroup designations from the independent criterion of splicing pattern.

Interestingly, the Myb gene splicing patterns constitute four major blocks in the Myb gene phylogeny (Figure 1). Block A lacks both introns 1 and 2. There are three splicing patterns in block B: subgroup G15 lacks both introns; subgroup G17 lacks only intron 2; and the remaining genes have altered splicing sites when compared to subgroup G13. Myb genes in block C have the major splicing pattern (81.2%) typified by G13, with some individual genes lacking intron 1 (9.4%), intron 2 (4.7%) or both introns (1.9%), or having minor splicing patterns (2.8%). In contrast, 58.2% of Myb genes in block D retain the typical splicing sites, and the rest lack only intron 1 (G02, G05 and half of the genes in G06).

In addition to splice-site locations, we also examined the position of splicing with respect to the open reading frame (ORF) - the intron phase. The splicing of each intron is designated as occurring in one of three phases: in phase 0, splicing occurs after the third nucleotide of the first codon; in phase 1, splicing occurs after the first nucleotide of the single codon; and in phase 2, splicing occurs after the second nucleotide. Figure 2a shows not only the conserved locations in the Myb-domain protein sequences but also the conserved phases of introns within the same subgroup. Moreover, there is a significant excess of phase 1 and 2 introns as well as an excess of nonsymmetric exons in Myb genes. Symmetric exons are exons that are flanked by introns of the same phase.

According to the intron-early theory 16, an excess of phase 0 introns and symmetric exons may facilitate exon shuffling by avoiding interruptions of the ORF, and thus could accelerate the rate of recombinational fusion and exchange of protein domains. Our results suggest that ancient Myb genes had a compact size without introns. During evolution, under some unknown mechanisms, introns were inserted into Myb domains and resulted in the observed splicing patterns. One splicing pattern remained unchanged in the subsequent gene amplification, resulting in the major splicing pattern typified by G13. Consistent with this, transposition of introns occurs very infrequently during evolution 17. This intron-gain model is consistent with previous results showing that numerous introns have been inserted into plants and retained in the genome 18. A similar approach to gene classification using intron/exon structure has been applied in the kinesin family 19 and the bHLH family 20, and the results support a similar evolutionary pattern.

Although the splicing sites are conserved, the sizes of both introns vary greatly for different Myb genes. Approximately 85% of introns 1 and 2 of Myb genes is shorter than 300 bp in Arabidopsis and rice. Detailed information about the distribution of intron sizes of Myb genes is available in Additional data file 2. It is worth noting that the size of intron 2 of maize p1 and p2 orthologs is very large, around 5 kb. This intron-size information may be helpful for aligning expressed sequence tags (ESTs) with genomic sequences.

Strikingly, a 743-base fragment was found in intron 2 of maize P1-rr and P1-wr alleles, but not in P1-rw and p2 alleles. A 10-base direct repeat (5'-TGATTTTGAC-3') flanks this fragment. Interestingly, no Ac elements were found inserted in its adjacent 3.2-kb intronic region, but frequent Ac insertion occurred in other regions. This could be due to a particular chromatin structure refractory to Ac insertion in this region 21. BLAST search detected this fragment (94% identity over 723 base-pairs (bp)) at one other locus in the maize genome, but with a new flanking direct repeat (5'-GGATATCCA-3'). The GenBank accession number is AF466202 (located 84795..85689, 12 March 2002 version). These results are consistent with a previous proposal that some transposable elements could insert into the genome as intronic sequences, a mechanism that has been proposed for the insertion of nuclear introns 22.

Most plant Myb genes are thought to encode transcription factors that activate or repress target gene expression either independently or together with cofactors. The topology of Myb phylogeny (Figure 1) indicates that some Myb genes in the same subgroup have the same function and that some Myb genes with similar functions are located in the same subgroup. For example, two Myb orthologs, snapdragon PHAN gene and maize rs2 gene, are located in subgroup G18, and both are involved in organ development: PHAN has been shown to regulate the development of the proximo-distal axis and dorso-ventral asymmetry of lateral organs such as leaves, bracts and petal lobes 23, while the rs2 gene controls the development of maize lateral organ primordia by repressing expression of knox (knotted1-like homeobox) genes that are required for the normal initiation and development of lateral organs 24. In another example, the Arabidopsis genes GL1 and WER located in subgroup G07 are both involved in epidermal cell development: GL1 activates the GLABRA2 homeobox gene for trichome (hair cell) development in some parts of the leaf and in the stem 2526, while WER controls the formation of the root epidermis by regulating expression of the GLABRA2 gene 27.

Similar results are observed for Myb genes involved in the phenylpropanoid biosynthetic pathway (Figure 1, subgroups N08, N09, N14): C1 28, Pl, TT2 29, AN2 30, p1 31, p2 32, FaMyb1 33, PAP1 and PAP2 34. These genes all encode a transcription factor that activates enzymes for phenylpropanoid synthesis, except that the FaMyb1 transcription factor suppresses anthocyanin and flavonol accumulation 33. In addition, the functional conservation among some Myb genes during evolution could be observed in the cell-cycle protein CDC5 (Figure 1, G19). The CDC5 protein performs an essential function in cell-cycle control at G2/M, and also participates in pre-mRNA splicing 35.

The extent of the Myb R1, R2 and R3 repeats is based on similarity to the previously-published consensus Myb repeat sequences 36. We used computational methods to identify additional conserved sequences downstream of the Myb repeats. A total of 18 motifs were identified in the carboxy-terminal regions, with each motif ranging in size from 9 to 32 amino acids (Figure 1). An exceptionally large domain (91 residues) was found in subgroup G19, gene CDC5, which is a conserved Myb paralog that originated prior to Myb-family amplification 12. In addition, Myb genes maize C1, Pl and AtMyb123 (TT2) in subgroup N08 have a nine-amino-acid motif previously reported 137. This motif has a high e-value (4.5e-008), so it was excluded from our analysis. Three other motifs identified by Stracke et al. 1 were excluded from our analysis because of their high e-values (see Additional data file 3).

Interestingly, three short fragments directly following the Myb R3 repeat are highly conserved in some subgroups (Figure 1, black boxes). We designated these extension motifs, E1, E2 and E3. Subgroups with an extension motif contain few or zero motifs in their carboxy-terminal coding regions when compared to those subgroups without extension motifs. One exception is subgroup G03, which contains motif E1 and two other carboxy-terminal motifs - 1 and 2 (Figure 1). In subgroup G08, a short conserved segment following E1 is termed E2. The three extension motifs are relatively small, ranging from 8 to 13 residues, but they are much more conserved than other motifs (Table 1). In the group of three extension motifs, 28 (of 33) sites are occupied by a single residue in more than 50% of the Myb proteins, and this value is greater than twice the relative frequency of the second most frequent residue.

To test the reliability of the motif predictions, the similarity scores were calculated over the motif plus its flanking regions. The similarity plots produced much higher scores in the motif region than in the flanking regions (Figure 3a), thus supporting the identified motifs. Similar results were observed in the nonsynonymous (dN) substitution analysis, which is a typical way of examining the degree of functional constraints on proteins using evolutionary comparisons 37. The results indicated that motif regions were less frequently subject to substitution than flanking regions. The distribution of dN values showed that most dN values are equal to or less than 0.6 in motif regions, and greater than 1 in flanking regions (Figure 3, Table 2). Interestingly, there are seven other motifs identified by Stracke et al. 1 that had a high dN value and did not pass this test (see Additional data file 3). The presence of carboxy-terminal motifs could reflect either the long-term conservation of critical sequences from antiquity or more recent gene duplications. The low dN values in the motif regions compared with the flanking regions suggest that the motifs are ancient sequences which have been conserved over long periods of time, rather than being the result of more recent duplications.

We wished to determine whether the detected motifs are specifically present in Myb genes, but are absent in non-Myb genes. Therefore, we used protein motif sequences as query sequences and performed Blastp searches in the Swiss-Prot database. For the motifs with size equal to or less than 15 amino acids, the homologous hits with 85% of the query motif length are all Myb domain containing proteins. For the motifs with size greater than 15 amino acids, the corresponding homologous hits with 70% of the query motif length contain Myb domains. The search result is described in Additional data file 4.

We obtained similar results in EST searches. When translated into proteins, all the 14 ESTs detected from an extension motif search also contain a Myb domain. Interestingly, we detected more ESTs from E1 than from the other extension motifs. Most probably this is due to the presence of E1 in more Myb genes than other motifs (Figure 1). The search result is described in Additional data file 5.

After comparing these two search results, we found that not all carboxy-terminal motifs detected homologous ESTs. This could be due to the low levels of expression of some Myb genes so that their EST sequences are not yet available. In some cases, these ESTs did not contain Myb domains; however, because the carboxy-terminal motifs are located downstream some distance from the Myb domains, the returned ESTs are probably too short to reach the Myb domains. However, alignment of ESTs with known Myb genes showed high identity not only in the motif sequence but also for considerable lengths in the flanking regions. This suggests that such ESTs are very likely from Myb genes.

Interestingly, we checked each carboxy-terminal motif in the 258 Myb proteins and found they are subgroup specific. For example, motif 1 from subgroup G03 was not detected in other subgroups.

In addition to the carboxy-terminal motifs detected in the predicted Myb proteins, we wanted to test whether any conserved DNA sequence motifs could be identified among the Myb gene subgroups. We applied motif-searching tools to detect conserved regulatory elements in the promoter region plus 5' UTR of the Myb genes and in intron regions. In contrast to the carboxy-terminal coding regions, no conserved DNA sequence motifs were identified in the Myb gene noncoding regions. This could be due to the fact that the Myb genes clustered in each subgroup are probably not orthologs or paralogs. In contrast, within the subgroup N14 (Figure 1) containing the maize p1 and p2 genes, and orthologs/paralogs from sorghum and rice, a highly conserved scheme of TATA-box, transcription start site sequences, and 5' UTR CA-box were found (data not shown). Otherwise, no significant regulatory elements were detected in noncoding regions of other Myb genes. However, it should be noted that segments of intron sequence closer to flanking exons are significantly more conserved than interior intron sequence. It has been reported that this level of intron sequence conservation may have a functional role in gene regulation 38. Our results suggest that it will be difficult to directly identify regulatory motifs in noncoding regions using only existing computational techniques. The chance of identification of regulatory elements will be increased in orthologs/paralogs. Possibly, the identification of co-regulated genes using microarray analysis will assist in the identification of common regulatory elements.

At the initiation of our project, the international rice genome sequencing project was not finished. Two finished rice genomes - one by Monsanto, the other by Syngenta (Oryza sativa L. ssp. japonica) - were not available to the public. Only the rice (indica) genome sequenced by Beijing Genomics Institute was publicly available. The sequence-quality assessments through sequence-tagged site (STS) markers, UniGene clusters and nonredundant cDNAs showed that 92% of the functional sequences that encode genes, and their immediate regulatory elements were present in the assembled sequences 9. Therefore, we chose subspecies indica rather than japonica for this study.

Rice genome sequences (scaffold dataset) were obtained from Beijing Genomics Institute 39. FGeneSH has been used successfully to predict genes in rice 9, and GenScan was used together with it to predict genes by taking rice genomic sequences as input. The two prediction results were combined as the complement of rice proteins. We performed Blastp and HMMER 40 searches to identify Myb genes from this rice protein dataset. For Blastp, we used a set of Myb R2R3 domains as query sequences. For HMMER, we used the Myb profile from Pfam. We parsed and combined the results of both searches, and obtained the final complement of rice Myb proteins with manual inspection of each sequence to confirm the identification of bona fide Myb genes. In the end, 85 typical Myb genes with complete R2R3 domains (one R0R1R2R3 and 84 R2R3) and 28 partial Myb genes were detected in the rice genome. Partial Myb genes contain a segment similar to one or a partial Myb repeat. The sequences of rice Myb genes are listed in Additional data file 6.

The complement of Arabidopsis proteins from GenBank were used to identify Myb proteins with complete R2R3 domains. The same methods as above were applied. We obtained 130 typical Myb proteins containing complete R2R3 domains (one R0R1R2R3 Myb, five R1R2R3 proteins, 124 R2R3 protein) and 11 partial Myb proteins. The results are consistent with previous findings 1.

To collect reference information on Myb gene functions, we used Blastp search against the nonredundant dataset in GenBank. The search yielded 43 plant Myb proteins with complete R2R3 domains; for most of these, some experimental information regarding functions or expression patterns was deposited by individual researchers.

For phylogenetic analysis, the above 258 Myb proteins (130 Arabidopsis, 85 rice and 43 from various other plants) with complete R2R3 domains were included. The sequences were aligned by ClustalX (version 1.81). The phylogenetic tree was constructed by the neighbor-joining method using MEGA version 2.0 41, with the setting of pairwise gap deletion and Poisson distance. Bootstrapping (1,000 replicates) was performed to evaluate the degree of support for a particular grouping in the neighbor-joining analysis. To enable the identification of motifs in the carboxy-terminal regions within each subgroup, we did not employ complete gap deletion as this may tend to exclude the contribution of carboxy-terminal residues because of their high divergence. The p-distance represents the simplest sort of genetic distance calculation and can be highly biased, so it was not used. In addition, attempts to use only the carboxy-terminal regions in construction of phylogeny were negative as a result of the high divergence. Therefore, we used the complete Myb proteins in clustering.

Three trees were constructed with the above settings, then taxa were classified into subgroups based on the topology of the phylogeny. Tree I used the 43 landmark Myb proteins with 130 Arabidopsis Myb proteins. The clustering result is consistent with the previous report 1; that is, the taxa that were clustered as subgroups in Stracke et al.'s findings 1 are located within a subgroup in tree I. Tree II replaced the 130 Arabidopsis Myb proteins in tree I with 85 rice Myb proteins. Tree III used the total 258 Myb proteins. We found that the clustering result of Myb proteins from Arabidopsis, rice and the landmark Myb proteins was consistent among all three trees. Therefore, we used tree III as the representative in this study.

Within each subgroup, motifs were detected using MEME 42 with the following parameter settings: the distribution of motifs: zero or one per sequence; maximum number of motifs to find: 16; minimum width of motif: 6; maximum width of motif: 117, in order to identify long R2R3 domains; minimum number of sites for each motif: the number of sequences, i.e., the motif must be present in all members within the same subgroup. Other options used the default values. Only motifs with e-value <= 1e-10 were kept for further analysis.

To confirm the reliability of the 38 motif candidates identified by MEME, we used PlotSimilarity from GCG package from Genetics Computer Group, Inc. to calculate the similarity score of each motif plus its 10-residue flanking fragments (protein sequences). There were 33 motifs with values above the average score in the motif region and below the average score in the flanking regions, and these were tested further using the dN values. The program YN00 from PAML package 43 was applied to analyze the conservation of each motif plus its flanking regions (coding DNA sequences). The frequency of synonymous substitution is too high to detect the conservation. Therefore, nonsynonymous substitution value was calculated. Low dN values indicate conservation whereas high dN values indicate divergence. We detected 18 motifs with dN <0.5 in the motif region and >1 in flanking regions; these 18 motifs included 3 extension and 15 carboxy-terminal motifs.

Originally, MEME identified 38 motif candidates with e-value <= 1e-10. Then five motifs were removed in the similarity score test. Later, 15 motifs were discarded in the nonsynonymous substitution test. This result suggests that the similarity score test is not sufficiently powerful to determine the reliability of motif candidates, and may be safely ignored in the future.

To test whether the carboxy-terminal motifs are specific to Myb genes, motif sequences were used to perform homology search in Swiss-Prot database and EST data set from GenBank. The latter can also provide information on the expression pattern of Myb genes. Low complexity was turned off for optimal short sequence search in both homology searches. In addition, in EST search, for motifs less than 15 residues 10 downstream residues were appended, and this elongated sequence was used as query sequence to perform EST search. The corresponding Myb R2 repeats were used in a tblastn EST search as an internal positive control.