A set of 7846 potential and known coding sequences from Streptomyces coelicolor was used as the starting point. Low complexity regions were masked using 'seg' 13. A comparison of each protein against itself was carried out using Prospero 14. Prospero returns the highest scoring self-self matches with an E-value score measuring the significance of each alignment.

Highest scoring matches were retained for each sequence and a series of filters were applied to remove matches that are unlikely to be novel domains. Firstly, all matches that had an E-value greater than 0.001 were discarded. Given the size of the Streptomyces coelicolor genome we would expect very few false alignments to be detected at this threshold. Secondly, alignments with a length of less than 30 residues were removed. Thirdly, alignments where the start points of each subsequence were separated by less than 45 residues ('shift') were discarded. Such short duplications are unlikely to be genuine domains. These are more likely to be structural repeats that are not stable in isolation. From this set any that overlapped a Pfam-A family were also discarded unless both subsequences occurred within the boundaries of single Pfam-A family. Such an occurrence indicates that the family contains more than one domain or repeat and needs refining. An overlap is defined as there being residues that occur in both the test alignment and the Pfam-A family alignment.

The alignments generated by Prospero were used as an initial alignment to make profile-HMMs using the HMMER 2.2 software 15. If the pair of sequences in the Prospero alignment overlapped then these overlap regions were removed from the alignment. Profile HMMs were built in local (fs) and global (ls) mode. The resulting profile-HMMs were scanned against the SWISS-PROT and TrEMBL databases 16. An inclusion threshold of 0.01 was chosen and an alignment of all homologues detected was constructed using the hmmalign program from the HMMER package. This alignment was then compared again to the Pfam-A database to see if the profile-HMM searches had detected any similarities to known families. This step removed distant homologues of previously described families. In some cases the missing members were subsequently added to the Pfam SEED alignments.

The previous three steps help to narrow down the number of potential domains to analyze. The final step is a careful manual inspection of the family to extend its membership as well as improve the multiple sequence alignment and hopefully to determine the domains function. This analysis uses a wide variety of tools and methods (see below).

A set of 597 short proteins (≤ 100 residues) was assembled. An all-against-all BLAST was carried out and the proteins clustered using single-linkage clustering with a cut-off threshold of 50 bits, which we determined was sufficiently high to prevent clustering of unrelated proteins.

All clusters that corresponded to Pfam-A families and single proteins that did not cluster were then removed from the set. This step also provides a useful check on the stringency of the clustering cut-off score. The clustered sequences were then aligned using T-Coffee 17.

The aligned clusters were then used as seeds for an iterative search process using HMMER 2.2, similar to above. The families were iterated until convergence. They were then realigned with T-Coffee and a single round of searching carried out. If any new family members were identified then the iterative search process was repeated.

Manual analysis as carried out in Step 4 of Method 1 (also see below).

See Figure 2 for example alignment. The domain is typically seventy residues in length and is predicted to have an α-helix-only fold. It appears to mostly only be found in the streptomycetes, though an HA-containing helicase is found in Chlamydia muridarum, and a protein consisting of three copies of the domain (Swiss:Q98RX4) in the lower eukaryote Guillardia theta. The gene in C. muridarum is likely to be a result of a lateral transfer event 27. Examination of the position of the HA domain-containing proteins, using Artemis, on the Streptomyces coelicolor genome revealed a surprising result. From each end of the linear S. coelicolor chromosome the second and third ORFs contain HA domains. The second gene from each end is identical to the other (SCO0002 and SCO7845) as are the HA-containing genes third from each end (SCO003 and SCO7844). SCO0002 and SCO7845 have an N-terminal DEAH/D helicase domain and 4 C-terminal HA repeats; SCO003 and SCO7844 have 6 C-terminal HA repeats and N-terminal region of unknown function, though it may contain a helix-turn-helix DNA-binding motif (score = 3.12, ~50% probability as predicted at http://npsa-pbil.ibcp.fr/cgi-bin/primanal_hth.pl). One more gene encoding a single HA domain is found more centrally on the chromosome (SCO0034).

Specific complexes are required for maintaining the ends of the linear streptomycete chromosomes, and the appearance of the genes encoding these domains specifically at the ends suggests that the proteins may be involved in forming these complexes. This is further evidenced by the observation 28 that similar helicases appeared at the end of several of the steptomycete chromosomes investigated as well as the linear plasmids. A knockout mutation experiment they carried out was inconclusive; chromosome linearity was maintained, but the region of protein substituted lay between the helicase domain and the HA domains, so it is possible that the helicases still retained functionality. The identification of an HA-containing helicase (SCP1.136) in the SCP1 plasmid, which is also linear and has the same type of telomere, further confirms this hypothesis.

There are no clear conserved catalytic residues in the alignment, suggesting that these domains have a binding function. The secondary structure prediction of the HA domain as a three-helical bundle is also suggestive of the Myb-like domain – a general DNA-binding domain. Aligning the sequence of the DNA-binding domain of Htrf1 (human telomeric protein) against the Pfam SEED alignment with T-Coffee showed interesting similarities between them. Two of the three key tryptophan residues in Myb-like DNA binding domain align to tryptophan residues in HA; in the place of a third is a leucine, which is a structurally conservative replacement. The first helix appears to align well, however the second is longer in HA whereas the third is shorter. As to whether there is a true evolutionary or functional relationship between the HA domain and the Myb-like domain, the evidence is not conclusive but the number of similarities is at least striking. Eukaryotic and Streptomycete telomeres are significantly different in structure, but the Myb-like domain may provide a plausible structure model for determining if and how the HA domains interact with DNA.

The following family was an interesting case, and has been previously mentioned as an uncharacterized domain 29. Although a repeat was detected with an E-value of 4.73 × 10^-4 using Prospero on the masked sequence, the validity of the repeat could not be verified by other means. However the amino terminal region was related to a number of other bacterial proteins and was investigated further; see Figure 3 for alignment. The BTAD domain is found in small set of bacterial regulatory proteins that occur in the streptomycetes and the closely related Mycobacteria, though one is also found in Rhizobium loti (MLR2443/Q98IE9). One of the proteins it is found in – AfsR – is a global secondary metabolite regulator of S. coelicolor 30. This protein has two basic functions – binding DNA and recruiting RNA polymerase. The first of these is carried out by the OmpR-like DNA-binding domain (PF00486), whereas the second is carried out by the region C-terminal to the BTAD domain. This region includes the ATP-binding NB-ARC domain (PF00931) and three TPR repeats (PF00515). AfsR's DNA-binding activity is modulated by serine/threonine phosphorylation 31; however there are no conserved serines or threonines in the BTAD domain so the phosphorylatyed residues are likely to occur in the DNA-binding domain. A mutation analysis by 32 on DnrI suggests that the BTAD domain is essential to its function. A possible explanation is that it mediates oligomerisation with other transcription complex proteins, or even mediates interactions between DnrI monomers that are binding tandem repeats in a promoter region. There are eleven pathway-specific regulatory proteins in S. coelicolor that contain this domain, including DnrI and RedD, five of which are found in antibiotic synthesis clusters. It is possible that the BTAD domain mediates interactions between the global regulator AfsR and the downstream pathway-specific regulators.

This family occurs as two sets of four forty-five residue tandem repeats in three S. coelicolor proteins. The repeats have a predicted secondary structure of three α-helices (See Figures 4 &5). The unusual architecture of these proteins is of note. To the C-terminus of each set of repeats is a low-complexity or coiled-coil region. For all three proteins InterProScan http://www.ebi.ac.uk/interpro/scan.html finds a chemotaxis sensory transducer region (IPR:004089; PS50111) between the two ALF-repeat regions. However searching these regions with HMMER 2.2 against SWISS-PROT and TrEMBL found no significant homology to other chemotaxis proteins; similarly using PSI-BLAST at the NCBI found several false-positives (data not shown), but no chemotaxis signal transduction proteins. The sequence in this stretch is very alanine rich, and so could lead to high-scoring matches on the basis of the apparent conservation of the alanines despite a lack of conservation in other positions. So it seems likely that the apparent homology is incorrect. One of the proteins, SCP1.201 (Swiss: Q9ACV2), also contained an intein (N-terminus: SM00306, IPR003587; C-terminus: PS50818, IPR002203) at its C-terminus, which is the first identified in S. coelicolor.

Two of the proteins, SCO6198 (Swiss: Q9Z5A4) and SCO6593 (Swiss: O87848), are located on the chromosome adjacent or close to secreted esterases (SCO6199 and SCO6590) and several other probable secreted proteins of unknown function (SCO6197; SCO6592, SCO6591, SCO6594). SCP1.201 is located on the SCP1 plasmid. Again this gene is located near a secreted esterase (SCP1.199) and a secreted protein of unknown function (SCP1.200). Homology searches showed that SCO6197, SCO6591 and SCP1.200 are all homologues, though no other homologues were found. No relationships were found for SCO6592, while SCO6594 was found to be homologous to the C-terminal portion of SCO0545. SCO0545 does not have a known function but there are several catabolic enzymes in the same region.

Given the conservation of the associated genes it seems likely that they represent a conserved pathway and that the ALF regions act as a substrate- or product-recognition domain that passes a signal to or from the secreted esterases. The intein does not contain the homing endonuclease, and so is probably no longer an active mobile genetic element; this concurs with the apparent lack of other inteins in the S. coelicolor genome. This implies that the plasmid has passaged through another species that has mobile intein elements.

This domain typically occurs in pairs, is approximately 90 residues in length and has two conserved tryptophans and a proline (See Figure 6). It is only found in a region of the S. coelicolor that is believed to be an integrated genetic element, e.g. a plasmid or transposon 1. The region appears to consist of two sections: a 'core' mobile element region with the essential replication genes and a flanking region containing a polyketide synthase and arsenic resistance genes. So this element may be important in mobilising these loci between strains. All of the SPDY domains occur in the core region, indicating that they are important in the replication of the element – though it is not possible to assign them a precise role. The lack of occurrences of this domain in any other known proteins indicates that this region of the genome represents a previously undescribed type of mobile genetic element.

The PASTA domain is discussed in greater detail in 33. It is a small (~70 residues) globular domain that binds cell wall peptidoglycan. With regards to S. coelicolor's genome it shows an unusual distribution. Typically organisms that have PASTA domains have one PASTA-containing serine/threonine protein kinase (pPSTK), which is, putatively, the master regulator of cell wall peptidoglycan cross-linking and essential to growth and development, and one PASTA-containing penicillin-binding protein (pPBP), which is the primary cross-linking enzyme. For a type example see Streptococcus pneumoniae. However, uniquely amongst the sequenced microbial genomes, S. coelicolor has three pPSTKs and no pPBP. The PASTA domains show very little identity to each other in each PSTK. The simplest explanation is that each pPSTK regulates different stages of growth and division, each of which uses different peptidoglycans. This also fits there being no pPBP as it would be specific to a single peptidoglycan structure; so we propose it uses an alternative localisation system, perhaps similar to that used by Deinococcus radiodurans or Gram-ve bacteria. Intriguingly S. coelicolor has three principle cell morphologies and it may be that each pPSTK regulates the development of each type.

This domain normally occurs as tandem repeats, is approximately 70 residues in length, and is predicted to be composed of 2 α-helices (See Figure 7). It is mostly found in prokaryotes, though four Arabidopsis proteins were identified with multiple HHE repeats and a Schizosaccharomyces pombe protein. Typically an HHE-containing protein consists of two HHE domains only, though there are exceptions like the Arabidopsis proteins (e.g. Q9LJQ1). There are two conserved histidines, both in the middle of predicted helices, and a conserved glutamate. It shows a slightly disparate phylogenetic distribution, but is found in eubacteria, archaea, fungi and plants. In several cases it appears to be involved in NO response – for instance DnrN from Pseudomonas stutzeri 34. Deletion of dnrN leads to slower response to nitrite of the nirSTB operon, so it may be involved in regulation or signal recognition. However, in Ralstonia eutropha deletion of the HHE-containing genes norA1 and norA2, despite being co-transcribed with the NO reductases-encoding norB1 and norB2, does not appear to affect growth or ability to cope with NO stress 35. It is also found in the ScdA protein of Staphylococcus aureus, which has been implicated in growth, development, and peptidoglycan cross-linking 36. The two conserved histidines and the glutamate are suggestive of a cation-binding site, such as the binding of Zn²⁺ in Carboxypeptidase A. This hypothesis is supported by its occurrence in the putative cation-transporting ATPase SCO0164 (Swiss:Q9RJ01) where it might sequester cations for transport.

These domains are typically ninety residues in length and found at the C-termini of secreted peptidases (See Figure 8). Surprisingly these domains are found in at least four different classes of peptidases. The PPC domain is found in some members of metallopeptidase families M4, M9 and M28 as well as the serine peptidase family S8 37. The PPC domains are cleaved off subsequent to secretion, but prior to activation of the peptidase. The actual function of them is not clear but they may aid secretion/localisation or inhibit the peptidase until needed. Visual inspection of the alignment, as well as predicted similarities in the secondary structure, suggests that it may be related to the PKD domain (PF00801), but no significant homology was detected using computational methods. They are often found in the same protein as the PKD domain and in very similar contexts, and it is tempting to suggest that they are functionally interchangeable (see Figure 9 for example domain architectures). PKD domains are thought to be involved in protein-protein interactions. Unlike the PKD domain the PPC domain is only found in bacteria and archaea, and not in eukaryotes.

This domain represents a sixty residue region that includes an FMN-binding site (indicated in alignment, Figure 10), as determined in the NqrC proteins of Vibrio cholerae 38 and Vibrio alginolyticus 39. Interestingly the NqrB proteins, which also bind FMN through a threonine residue and are part of the same complex, do not show any homology. The region is found in several electron transport chain proteins; for example the RnfG electron transport protein, part of a chain that supplies electrons to both nitrogen fixation and DNP reduction in Rhodobacter capsulatus 40. Other examples include the NosR/NirI nitrous oxide reduction regulatory proteins. FMN_bind-containing proteins appear to split into two groups, which relate length to function. The shorter proteins, typically 200–350 residues, are components of electron transport chains whereas the longer proteins, typically 680–800 residues, have a regulatory function. The regulatory proteins typically have five transmembrane helices in the C-terminal half of the protein. Members of both groups often have 4Fe-4S domains present, suggesting that the regulatory mechanisms also involve charge movement.

This domain is named after the MbtH protein from Mycobacterium tuberculosis (Swiss: O05821). The domain is typically 70 residues in length and covers the full length of the protein, though NikP1 from Streptomyces tendae (Swiss:Q9F2E7) also contains two domains common to antibiotic synthesis proteins: an AMP-binding domain (PF00501) and a Phosphopantetheine attachment site domain (PF00550). It is found in the Actinomycetes, the Proteobacteria gamma subdivision and Rhizobium leguminosarum. Several of these proteins have been implicated in antibiotic biosynthesis in several streptomycetes (for instance nikkomycins: 41; simocyclinone: 42; coumermycin A1: 43, and the formation of siderophores such as E. coli's enterobactin or M. tuberculosis's mycobactin (reviewed in 44). In the biosynthesis of siderophores they do not seem to have a direct role, as a complete synthetic pathway can be built up of mycobactin without assigning to a role to MbtH (and similarly with enterobactin and the Mbth-like YbdZ); so it is likely that it is involved in either regulation of expression or transport of the siderophores out of the cell, with a similar role in antibiotic synthesis. There are several conserved residues, including three tryptophans that may have functional importance (See alignment in Figure 11).

This domain family has previously only been reported in eukaryotes, but in fact it contains a diverged sub-group that occurs in eubacteria as well. An alignment of the eukaryotic and prokaryotic versions show that the principle difference is the absence in bacteria of the conserved cysteine residues, which form disulphide bridges, whereas the proposed active site 45 (see Figure 12) is mostly conserved. In order to try and determine its function in bacteria a review of the information available for the eukaryotic domains was carried out.

So far all SCP-containing proteins appear to be secreted, and this is backed up by the consistent prediction of signal peptides at the N-terminus. There is very little direct evidence of their general function currently, however many examples have been found to be involved in signaling. For instance they are involved in several mammalian developmental processes, most notably sperm maturation 46 and sperm-egg fusion 47, and are up-regulated in several tumors (4849). Clear evidence has been found, in Xenopus, of sperm following the concentration of 'Allurin' – an SCP-containing protein 50. They are also commonly used by insects and reptiles as mammalian toxins.

However these proteins are very big for direct signaling molecules, typically being 200 or 400 residues (1 or 2 SCP domains). It has been suggested that there is an active site, based on analysis of the 3D NMR image of plant PR14a and comparison with human GliPR 45, and three of the four residues predicted to make up the site are conserved between the eukaryotic and prokaryotic subfamilies (See Figure 12). This would imply that the domain generates a smaller signaling molecule. However no evidence has been found of such a molecule and several pieces of evidence conflict with this hypothesis. Firstly the nematode SCP-containing Neutrophil Inhibitory Factor (NIH) binds directly to integrins CD11b/CD18 on the neutrophil cell surface 51. Secondly pseudochetoxin (from King brown snake) appears to bind the extracellular portion of cyclic-nucleotide gated ion channels (CNG channels) blocking their function 52. In the second case there does appear to be time-lag between association or disassociation and blocking or release of the gate. This does seem to suggest that its mode of action is not simply as a steric block.

SCP-containing proteins are involved in a tremendously wide range of processes, and found to be essential in plants (PR1-like proteins), mammals, lizards, insects (venom allergens) and nematodes. It appears likely that they are similarly important and similarly multi-function in bacteria, and hence are an important target for further analysis.

Several S. coelicolor proteins were identified that were found to be related to FG-GAP repeats. The Pfam family from version 7.4 contained only 5 bacterial members; the updated family in Pfam 7.5 is found in thirty nine bacterial proteins – including fourteen in S. coelicolor (see Figure 13 for distribution of FG-GAP repeats in bacteria). An extra thirty-four eukaryotic family members are also identified, as well as an archaeal protein (Swiss:O28333). The FG-GAP repeats have been predicted to assume a β-propeller conformation. The occurrence of this repeat as sets of four or five tandem copies casts doubt on this (e.g. Swiss:ITA2_DROME), as they are normally six or seven bladed 3. However the hemopexin repeat (PF00045) has been seen as a four-bladed propeller e.g. as in mammalian blood serum haemopexin glycosylated-native protein (PDB:1qjs), so perhaps FG-GAP repeats might be more structurally similar to these repeats.