To identify new eukaryotic intein genes, we used the sequences of a wide variety of previously identified inteins (of eukaryotic, prokaryotic and viral origins) to perform BLAST searches of the publicly available eukaryotic sequence databases (including GenBank, and genome sequencing centre databases containing data not yet released to GenBank; see Methods section). High-quality matches identified in the BLAST searches, putatively representing new inteins, were then examined in detail for features characteristic of inteins 15. For example, most inteins begin with a Cys or a Ser residue, end with the dipeptide His-Asn, and are followed by a Cys, Ser or Thr residue as the first amino acid of the C-extein. Inteins contain a number of conserved motifs associated with splicing, and most inteins contain a homing LAGLIDADG endonuclease domain. Inteins also appear as specific inserts within other proteins, and often appear at highly conserved sites within highly conserved proteins. Another important indicator of the presence of an intein encoding sequence is that the sequence is present in some homologues of the host gene, but is absent in most. Table 1 summarises the novel inteins described in this report.

P. (Stagonospora) nodorum is a filamentous ascomycete belonging to the class Dothideomycetes. It is a major pathogen of wheat crops. Strain SN15 has been sequenced by the whole genome shotgun method to >10-fold coverage at the Broad Institute 32. The assembled sequence has been made publicly available in GenBank. Using the sequence of the C. eugametos chloroplast Ceu ClpP intein as a query in a TBLASTN search of the whole genome shotgun (WGS) sequence division of the GenBank database we detected a high quality match in P. nodorum sequence AAGI01000064 (E = 3 × 10^-6; bases 49105–50471). This sequence can also be found on supercontig 1.4 (from base pair 1094221 to 1095587) at the Broad Institute website 32. The matching region appears as an insert within a gene encoding the second largest subunit of RNA polymerase (Figure 1). Sequence comparisons (Figure 1) and phylogenetic analyses (Figure 2, additional file 1) indicate that the gene encodes a subunit of RNA polymerase I. The insert has numerous features indicating that it encodes an intein. It appears as an insertion within the RNA polymerase gene and consists of an uninterrupted open-reading frame (ORF) encoding 455 amino acids, in phase with the RNA polymerase ORF. The insert begins with a Cys residue and is followed by a Cys residue. The N- and C-terminal regions of the insert contain sequences corresponding to the conserved splicing domains of inteins (alignments of conserved intein domains can be viewed at the InBase website 6). The central region contains the characteristic motifs of an LAGLIDADG homing endonuclease domain, although this appears to be degenerate and is unlikely to be still active. An unusual feature of this new intein is that it ends with the dipeptide Gly-Asn rather than the more common His-Asn. Gly-Asn termini have, however, been identified previously in inteins, for instance, in Ceu ClpP and in the ThrRS inteins of C. tropicalis and C. parapsilosis. This new intein has been named Pno RPA2, following intein naming conventions. It is the first intein to be identified in a gene encoding the second largest subunit of an RNA polymerase. The only previously identified RNA polymerase inteins (the archaeal inteins Mja rPol A' and Mja rPol A" from M. janaschii and Nph rPol A" from Natronomonas pharaonis) all appear in archaeal homologues of the gene encoding the largest subunit of eukaryotic RNA polymerase.

Note that the sequence AAGI01000064 has a frameshift in the region corresponding to the intein. Comparisons (not shown) with the P. nodorum sequences in the GenBank trace archives, however, suggest that this is a sequencing error resulting from the insertion of a G residue at position 50225 and a C at 50260. These were removed to generate the full sequence of the RNA polymerase gene with an uninterrupted ORF.

C. reinhardtii is a unicellular green alga. An intein in this species was first detected in several C. reinhardtii expressed sequence tag (EST) sequences using a TBLASTN search of the GenBank EST databases with the Ctr ThrRS intein sequence as a query. A full-length sequence of the intein was then retrieved from version 2 of the C. reinhardtii genome sequence assembly, available from the Joint Genome Institute 33. The intein lies on scaffold 5, contig 26 (bases 289701–290993 on the minus strand). The intein, Cre RPB2, appears as an uninterrupted ORF encoding 431 amino acids inserted within the coding region of the C. reinhardtii gene for the second largest subunit of RNA polymerase II (Figures 1, 2). Like many other inteins, it begins with a Cys residue, is followed by a Cys residue in the C-extein, and contains the conserved splicing domains and an LAGLIDADG homing endonuclease domain (see InBase for alignments 6). Like the Pno RPA intein, Cre RPB2 ends with a Gly-Asn dipeptide rather than the more common His-Asn. This is the first intein encoded in a nuclear genome to be found outside of the fungi.

Three further inteins have been found in genes encoding the second-largest subunits of RNA polymerase II. The sequences of these genes were generated as part of the Assembling the Fungal Tree of Life (AFTOL) project 34, which is using, among other things, RNA polymerase II sequences to assist in determining the relationships among a wide variety of fungi. Inteins appear in RPB2 sequences from Spiromyces aspiralis (DQ302790), a zygomycete fungus, and Coelomomyces stegomyiae (DQ302766) and Batrachochytrium dendrobatidis (DQ302769), both members of the Chytridiomycota. These inteins again have the conserved splicing and endonuclease domains characteristic of inteins (see InBase 6). They also appear as inserts within the RNA polymerase sequences. The C. stegomyiae intein, Cst RPB2, and the S. aspiralis intein, Sas RPB2, are inserted at homologous sites and are therefore allelic inteins. The B. dendrobatidis intein is inserted at a different site. Both of these sites are distinct from the insertion site of the C. reinhardtii RNA polymerase II intein, Cre RPB2. To distinguish the three intein insertion sites in RNA polymerase II genes they have been denoted "a", "b" and "c", according to the order in which they were identified: Cre RPB2 is in the "a" site, Cst RPB2 and Sas RPB2 are in the "b" site, and Bde RPB2 is in the "c" site. The allelic RPB2-b inteins are present in two very distantly related species, a zygomycete and a chytrid.

D. discoideum is a slime mould classified within the Mycetozoa. The whole genome sequence has been determined and described 35. An intein in D. discoideum was detected as an insertion of 464 amino acid residues within the second largest subunit of RNA polymerase III (Figures 1, 2; GenBank protein ID no. EAL63250). The intein, Ddi RPC2, appears as a specific insert within the RNA polymerase subunit relative to homologues from other species, and it is inserted at a different site from the P. nodorum RNA polymerase I intein and from any of the RNA polymerase II inteins. The Dictyostelium intein begins with a Cys residue, ends with a standard His-Asn dipeptide and is followed by a Cys residue. The N- and C-terminal parts contain the conserved splicing domains characteristic of inteins 6, while the central region contains a possibly degenerate LAGLIDADG homing endonuclease. Ddi RPC2 contains several low-complexity regions or short runs of the same amino acid. For instance, it contains a region of 13 amino acid residues, of which 11 are Asn residues. It also contains a region with seven consecutive Asn residues and two regions with seven consecutive Gln residues. Such low-complexity regions appear to be common features in D. discoideum proteins 35. In the D. discoideum RNA polymerase III protein, they are restricted to the segment corresponding to the endonuclease domain of the intein, i.e. they are not found in the intein splicing domains or in the RNA polymerase sequence.

P. ramorum is a member of the oomycetes, belonging to the kingdom Stramenopiles, which also includes diatoms, golden-brown algae and brown algae 36. The genome sequence has been determined by the Joint Genome Institute 3337. Using the Ceu ClpP intein as a query in a TBLASTN search, we detected a high quality match (E = 3.0 × 10^-15) in the P. ramorum genome (scaffold 19, bases 14734–15744 on the minus strand). This sequence has numerous features suggesting that it is an intein. For instance, it begins with a Cys residue and contains sequences similar to the splicing domains of other inteins (not shown). These are separated by a region containing an LAGLIDADG homing endonuclease domain similar to that of previously identified inteins. Immediately upstream of this putative intein is a long ORF encoding a putative protein homologous to the second largest subunit of RNA polymerase. The site at which the putative intein interrupts the RNA polymerase ORF is highly conserved, although it is distinct from the insertion sites of the previously identified RNA polymerase inteins (Figure 1).

In addition to having these similarities to other inteins, this putative intein has unusual features. Firstly, instead of being an uninterrupted ORF, the region encoding the intein-like sequence contains two frameshift mutations, which result in the appearance of stop codons within the coding reading frame. Secondly, although it contains most of the conserved motifs associated with intein splicing, it lacks the conserved residues (usually a His-Asn dipeptide) corresponding to the extreme C-terminal ends of inteins; instead, the corresponding sequence consists of a stop codon and an Arg codon (see additional file 2). These features suggest that the sequence no longer represents a functional intein (comparisons with sequences in the trace archives suggest that most of these are genuine mutations, although one of the frameshifts within the intein is likely to be a sequencing error; data not shown). Likewise, the RNA polymerase gene, in which the putative intein gene is inserted, has some unusual features. Firstly, it also appears to be non-functional; about 780 bp upstream of the intein insertion site, the RNA polymerase coding sequence contains a frameshift mutation and there is a nonsense mutation six codons upstream of the putative intein. Secondly, the section of the RNA polymerase gene expected to lie downstream of the intein gene (i.e. the coding sequence for the C-extein) is missing (additional file 2). Comparisons with the trace archives suggest that these are all genuine mutations. Phylogenetic analyses indicate that this degenerate RNA polymerase gene is not closely related to eukaryotic RNA polymerase I, II or III genes (Figure 2). Instead, it is most closely related (100% bootstrap support) to an RNA polymerase from African swine fever virus (ASFV), a large double-stranded DNA virus that is a member of the nuclear-cytoplasmic large dsDNA virus (NCLDV) group 38. In addition, three intact genes encoding the second largest subunits of RNA polymerases I, II and III can be found in the P. ramorum genome (Figure 2). Close relatives (not shown) of these three genes also appear in the genome sequence of the related species Phytophthora sojae (also sequenced by the JGI), but no close relative of the degenerate ASFV-like RNA polymerase gene is present in the P. sojae genome.

Further analyses of the sequences surrounding this RNA polymerase gene reveal a likely explanation for its unusual features; when the predicted products of the ORFs in the regions close to the RNA polymerase gene are used in BLASTP searches against the protein sequences in GenBank, the strongest hits are (as with the RNA polymerase itself) often proteins encoded by ASFV (additional files 3 and 4). Most of these proteins do not have close relatives in the P. sojae genome. The ORFs further away from the degenerate RNA polymerase gene, however, do have close matches in P. sojae, and are not closely related to genes found in ASFV. It is therefore likely that a previously unidentified virus related to ASFV has integrated into the P. ramorum genome. This integration would have occurred after the divergence of the lineages leading to P. ramorum and P. sojae, as no trace of the putative viral relic appears in P. sojae. After its integration into the P. ramorum genome, the viral sequence has started to degenerate.

A putative intein was also identified in a sequence from an unclassified species (IBEA_CTG_SVAEH23TF) found in the environmental samples division of GenBank (accession no. AACY01369547). The sequence was generated as part of the shotgun sequencing of samples from the Sargasso Sea 39. The complementary strand of this sequence encodes the N-terminal part of an intein, which includes the conserved splicing motifs and the first motif of a homing endonuclease domain, and is preceded by part of an RNA polymerase. The intein in this sequence is inserted at the same site as that in the putative viral relic in P. ramorum, i.e. they are allelic inteins. Similarity searches at InBase indicate that the most closely similar annotated intein to this Sargasso Sea sequence is Ceu ClpP, the intein from the chloroplast of C. eugametos (E = 2 × 10^-13). BLAST2 comparisons suggest a closer sequence similarity between the Ceu ClpP intein and the Sargasso Sea sequence (E = 1 × 10^-17) than between the Sargasso Sea sequence and the P. ramorum virus intein fragment (E = 4 × 10^-5).

Phylogenetic analyses (not shown) indicate that the RNA polymerase from which this sequence is derived is most closely related to eukaryotic RNA polymerase II, although it is not highly similar to any sequence of known origin. E-values derived from TBLASTN searches at NCBI 40 indicate that the 59-residue fragment of this RNA polymerase is most similar (52–56% amino acid identity) to RNA polymerases (RPB2) from fungi (Table 2). It is less likely that the sequence is from a marine virus such as one of the large double-stranded DNA viruses from the NCLDV group that infect eukaryotes (40–46% amino acid identity). These large viruses, some of which infect marine organisms, encode RNA polymerase II-like proteins.

Previous analyses have suggested that inteins usually appear at highly conserved sites within their host proteins 5. One possible reason for this preference is that inteins inserted at such sites are less likely to be removed. Highly conserved sites in proteins usually have important and sequence-specific functions, and, therefore, any deletion that removes the intein sequence would have to be very precise or it would result in a non-functional host gene. In contrast, inteins inserted at poorly conserved sites might be successfully removed by a wide range of imprecise deletions. A second possibility is that inteins inserted at highly conserved sites may be more likely than inteins inserted at poorly conserved sites to spread successfully throughout the gene pool of a species or to undergo a successful horizontal transmission to a new species, as the homing endonuclease recognition site is more likely to be conserved. With six distinct insertion sites within homologous genes, these new RNA polymerase inteins provide a good opportunity to examine this phenomenon in detail in a eukaryote system. We therefore created an alignment of eukaryotic RNA polymerases, plotted the level of conservation in 10 amino-acid windows across the alignment, and mapped the intein insertion sites onto the plot (Figure 3). As can be seen, the intein insertion sites each correspond to one of the peaks in the sequence conservation plot, indicating that each is inserted into one of the most highly conserved sites within these genes. Not all highly conserved sites can act as intein refuges because they must contain appropriate flanking residues (for example the C-extein Cys or Ser).

We also mapped onto the plot the positions of all the spliceosomal introns from the intein-containing RNA polymerase genes. There are two introns in the genes from P. nodorum, B. dendrobatidis and D. discoideum, and 20 in the C. reinhardtii gene. None was found in the S. aspiralis or C. stegomyiae genes or in the putative proviral gene from P. ramorum. As can be seen in Figure 3, some introns are inserted at highly conserved positions, but others are inserted at sites that are only moderately or are poorly conserved, showing that, in contrast to inteins, RNA polymerase introns do not preferentially appear at highly conserved sites.

Having mapped the intein insertion sites onto the eukaryotic RNA polymerase sequence conservation plot (Figure 3) and determined that these were in regions of high sequence conservation, we wished to discover where these sites occurred in the three-dimensional protein, including which structural domains correspond to the intein insertion sites. It is of interest to plot these sites because inteins at some positions might be more easily processed during protein folding. Because the RNA polymerase subunits RPA2, RPB2 and RPC2 are similar in structure, we used as our common template the structure of RPB2 from S. cerevisiae 4243. It is possible to map onto the protein structure the position of the RPA2, RPB2, RPC2 and viral RPO II intein insertion sites, as if they were all inserted into the homologous RPB2 protein. Using data from the Protein Data Bank 44 (entry 1I3Q) and the MacPyMOL molecular visualisation system 45, we highlighted six residues immediately adjacent to each intein insertion site. Figure 4 illustrates the assembly of RPB2 and RPB1 into the core heterodimer. Using the terminology of Cramer et al 42, both the Ddi RPC2 and Bde RPB2 insertion sites are in the 'fork' domain, the Pno RPA2, Cst RPB2, Sas RPB2 and the viral RPO intein found in P. ramorum are all inserted within the 'hybrid binding' domain, and the Cre RPB2 intein insertion site is in the anchor domain. All of these sites are close to the active site of RPB2. None of the insertion sites are found on the external surface of the protein; all are on the surface of the 'cleft' formed by the RPB1/RPB2 heterodimer or are on the interface between these subunits (Figure 4). It is highly probable, therefore, that inteins inserted into these sites in any of the homologues will need to be accurately spliced out before the protein subunits can assume their correct folds and the active RNA polymerase complex can be assembled. In contrast, if inteins were present on the surface of the heterodimer, they could undergo inactivation and progressive deletion without necessarily impairing the assembly and function of the RNA polymerase.

Previous work with eukaryotic inteins has shown that allelic inteins are usually each other's closest relatives. For instance, the wide variety of PRP8 inteins identified in ascomycete and basidiomycete fungi form a monophyletic group, relative to all other known inteins. Similarly, the yeast VMA1 inteins also appear as a monophyletic group. There is some evidence to suggest that many of the previously identified eukaryotic nuclear inteins (i.e., VMA1, PRP8, GLT1 and CHS2) may be more closely related to each other than they are to most inteins encoded by non-nuclear genes 20. There is no evidence suggesting that the nuclear-encoded inteins are closely related to eukaryotic inteins encoded by chloroplast genes, or inteins encoded by eukaryotic viruses. Indeed, some of these latter inteins are alleles of, and closely related to, inteins found in prokaryotes. For instance, the DNA polymerase B inteins of the A. polyphaga mimivirus and HaV01 are most closely related to allelic DNA polymerase inteins from various archaea 17.

To study the relationships among the new RNA polymerase inteins and previously identified inteins, we constructed phylogenies based on alignments of conserved intein splicing domains. Homing endonuclease domains were not included in this analysis because they are absent from mini-inteins and because, even in full-length inteins, they are often degenerate and therefore might produce misleading results. The sequences to be aligned were edited so as to remove all of the intein residues between the end of the N-terminal splicing domain and the beginning of the C-terminal splicing domain. These domains were determined by comparison with other intein sequences available at InBase 6. The resulting alignment has 102 positions and is available as supplementary data (additional data file 5). An example of a tree containing all the new RNA polymerase inteins and a wide variety of previously identified inteins, including most of the known eukaryotic inteins and representatives of many of the allele groups of prokaryotic inteins, is shown in Figure 5. Most of the relationships observed on this tree are consistent with results obtained previously. For instance, the PRP8 inteins all group together, as do the VMA1 inteins.

The new eukaryotic RNA polymerase inteins do not generally appear to be closely related to each other, despite being present in homologous (in some cases paralogous) genes. This is not perhaps unexpected, however, as most are inserted at different sites in these genes and therefore are not allelic inteins. The C. reinhardtii RNA polymerase II intein appears to be most closely related to the threonyl transfer RNA synthetase inteins from C. tropicalis and C. parapsilosis. This grouping receives a high level of bootstrap support (100%). This is unusual as these inteins are not alleles and are found in different kingdoms. The P. nodorum RNA polymerase I intein is not closely related to any other known intein, although it does fall within a moderately supported (68%) group that also includes the C. eugametos chloroplast Ceu ClpP intein, Sas RPB2, Cst RPB2, the putative viral intein embedded within the P. ramorum genome, and a variety of prokaryotic inteins. All these inteins (including Cre RPB2, Cpa ThrRS and Ctr ThrRS, Pno RPA2, Sas RPB2 and Cst RPB2, and Ceu ClpP), together with a set of prokaryotic inteins, form a well-supported (99%) cluster distinct from all other inteins. The B. dendrobatidis RNA polymerase II intein (Bde RPB2) and the D. discoideum RNA polymerase III intein (Ddi RPC2) lie outside of this cluster. Although on this tree, they appear as each other's closest known relative, this grouping does not receive high levels of support (60%) and the two inteins are not particularly similar in sequence (~20% identity), so the significance of the grouping is uncertain.

The topology of the distance trees generated from the alignment data is generally very similar if different tree-building algorithms such as quartet puzzling or parsimony analyses are used. The bootstrap values generated follow a similar pattern also, with one exception; the node that joins the PRP8 allelic inteins with the VMA allelic inteins can attract values ranging from 56% to 96%. The bootstrap value of the node that groups many of the newly described RNA polymerase inteins into a cluster distinct from all other inteins ranges from 70% (fast heuristic search, no branch-swapping) to 99% (heuristic search with branch-swapping).

The sequence databases used were:

• The Joint Genome Institute 33.

• The Wellcome Trust Sanger Institute 46.

• The Broad Institute 32.

• Washington University Genome Sequencing Center 47.

• National Center for Biotechnology Information 48.

General sequence analyses were carried out using the programs of the GCG package 49. Sequence similarity searches were carried out using the BLAST servers at GenBank 40, InBase 6 or the various genome-sequencing centres mentioned above. Multiple sequence alignments were constructed using CLUSTAL_X 50 and refined using SEAVIEW 51. Phylogenetic analyses were performed using PAUP* 52 using the default settings unless otherwise noted.

Intein protein sequences were retrieved from InBase 6 under the standard intein names. Protein sequences for the second largest subunits of RNA polymerase sequences were retrieved from GenBank 48 using the following protein ID numbers.

• Schizosaccharomyces pombe Pol. I, CAB66435; Pol. II, Q02061; Pol III, CAA93558.

• Aspergillus fumigatus Pol. I, EAL88681; Pol. II, EAL84702; Pol III, EAL87958.

• Saccharomyces cerevisiae Pol. I, AAA34993; Pol. II, CAA99357; Pol III, CAA99422.

• Dictyostelium discoideum Pol. I, EAL60592; Pol. II, EAL63310; Pol III, EAL63250.

• Homo sapiens Pol. I, AAX81999; Pol. II, AAH23503; Pol III, AAH46238.

• Drosophila melanogaster Pol. I, AAF51503; Pol. II, AAF55024; Pol III, AAF58590.

• Arabidopsis thaliana Pol. I, AAG52049; Pol. II, CAB36815; Pol III, BAB11387.

• Cryptosporidium parvum Pol. I, EAK88354; Pol. II, EAK90367; Pol III, EAK87469.

• Encephalitozoon cuniculi Pol. I, CAD26190; Pol. II, CAD25744; Pol III, CAD25947.

• The P. nodorum RNA polymerase sequences were predicted from the genes on the following sequences: RNA Pol. I, AAGI01000064; RNA Pol. II, AAGI01000234; RNA Pol. III, AAGI01000034.

• The P. ramorum RNA polymerase sequences were predicted from the genes on the following sequences: RNA Pol. I, scaffold 163; RNA Pol. II, scaffold 60; RNA Pol. III, scaffold 33.

• The Chlamydomonas reinhardtii RNA polymerase II gene sequence was assembled from sequences from version 2 of the genome assembly (Scaffold 5. contigs 26, 27 and 28) combined with sequences from the trace archive (589516860, 651002588, 591226556, 650233847, 587272333). Introns were identified by comparison to other RNA polymerases.

• African swine fever virus AAA65283

• Emiliana huxleyi virus 86 CAI65861

• Acanthamoeba polyphaga mimivirus AAQ09583

• Chilo iridescent virus AAK82288

• Grouper iridovirus AAV91067

• Frog virus 3 AAT09722

• Lymphocystis disease virus AAU10873

• Rock bream virus AAT71848

• Swinepox virus AAL69852

• Orf virus AAR98326

• Melanoplus sanguinipes entomopoxvirus T28316

• Amsacta moorei entomopoxvirus AAG02772

• Vaccinia virus AAB96526

• Pyrococcus furiosus AAL81688

• Ferroplasma acidiarmanus EAM93828.

• Staphylococcus aureus AAW37698

• Crocosphaeria watsonii EAM50876