Control of directionality in integrase-mediated recombination: examination of recombination directionality factors (RDFs) including Xis and Cox proteins

Abstract

Similarity between the DNA substrates and products of integrase-mediated site-specific recombination reactions results in a single recombinase enzyme being able to catalyze both the integration and excision reactions. The control of directionality in these reactions is achieved through a class of small accessory factors that favor one reaction while interfering with the other. These proteins, which we will refer to collectively as recombination directionality factors (RDFs), play architectural roles in reactions catalyzed by their cognate recombinases and have been identified in conjunction with both tyrosine and serine integrases. Previously identified RDFs are typically small, basic and have diverse amino acid sequences. A subset of RDFs, the cox genes, also function as transcriptional regulators. We present here a compilation of all the known RDF proteins as well as those identified through database mining that we predict to be involved in conferring recombination directionality. Analysis of this group of proteins shows that they can be grouped into distinct sub-groups based on their sequence similarities and that they are likely to have arisen from several independent evolutionary lineages. This compilation will prove useful in recognizing new proteins that confer directionality upon site-specific recombination reactions encoded by plasmids, transposons, phages and prophages.

INTRODUCTION

Recombination directionality factors (RDFs) are a diverse group of proteins involved in controlling the directionality of integrase-mediated site-specific recombination reactions. Typically, RDFs are small DNA-binding proteins acting as accessory factors to influence the choice of substrates that are recombined by their cognate recombinase. While the majority of the RDFs that have been described are components of phage-encoded site-specific recombination systems [e.g. Lambda (1)], RDF proteins are also associated with a variety of other recombination systems including those encoded by plasmids [e.g. pSAM2 (2)] and transposons [e.g. Tn916 (3)].

The best studied RDF is that encoded by phage Lambda, Xis. The phage-encoded integrase, a member of the tyrosine family of site-specific recombinases, catalyzes both integration and excision reactions. The integration reaction utilizes the phage attP and bacterial attB DNA sites as substrates and generates recombinant junctions, attL and attR, as products. The excision reaction involves recombination between attL and attR to generate attP and attB as products. Both of these reactions require the integration host factor (IHF) in addition to integrase (4,5). Lambda Xis is required for the excision reaction, but inhibits integrative recombination (6).

Lambda Xis determines the directionality of recombination by influencing the formation of specific protein–DNA architectures. The Lambda integrase is composed of three domains, two of which confer different DNA binding specificities; both DNA binding valences can be occupied simultaneously leading to both intramolecular and intermolecular integrase-mediated bridges. The formation of these bridges is facilitated by IHF, which binds to specific sites in the DNA substrates and introduces DNA bends (7). Lambda Xis also binds to specific recognition sequences and introduces sharp DNA bends, which in the attR substrate promotes the formation of protein–DNA structures that can undergo excisive recombination, but which in attP prevents the formation of the architectures needed for integration (8,9). There is also some additional evidence that the C-terminal end of Lambda Xis interacts with the integrase protein. However, this interaction is not absolutely required and seems primarily to be involved in stabilizing binding of integrase to the DNA (10).

Although more than 100 phage-encoded integrases have been described (11), not all of these belong to the tyrosine family of recombinases (12,13). More than two dozen large serine recombinases have been described of which nearly half function in phage integration and excision. These recombinase proteins contain a 140 residue N-terminal domain with strong similarity to the catalytic domain of transposon resolvases and DNA invertases which utilize a serine near the N-terminus as the catalytic residue, but they also have a C-terminal domain that is much larger than that of the typical resolvase or invertase. We will thus refer to this class of integrase as the serine-integrases (Int-S) to distinguish them from the tyrosine-integrases (Int-Y). However, little is known about how the directionality of these systems is regulated. In the best-studied Int-S example, encoded by Streptomyces phage φc31, the recombination sites are small, there is no evidence of a host integration factor and no RDF has been identified (14). In phage TP901 (15) and in the heterocyst development system in Anabaena (16), RDF proteins have been identified although their mechanism of action is not understood.

In the case of the Lambda recombination system, the only known role of the Xis protein is in controlling the directionality of recombination. However, this is not true for the RDFs encoded by phages HP1 and P2 where they also act as transcriptional regulators (17,18). These are referred to as Cox proteins, and they can both negatively and positively regulate transcription initiation. Initial characterization of the HP1 excision reaction showed that the Cox protein binds specifically to attP DNA and forms several specific DNA–protein complexes in the presence of attP and Cox (17). In spite of the fact that the Cox proteins are also transcriptional regulators it seems likely that they regulate recombination directionality in a similar way as described for the Lambda integration system.

While a number of RDFs have been identified through experimental approaches, others have proven difficult to find using comparative sequence methods, mainly because the proteins are small, typically containing <100 amino acids, and contain few, if any, highly conserved residues. Moreover, while many RDFs are basically charged, there are several examples of others that are acidic. Because of the great diversity of these RDF proteins, there previously has been no systematic effort to catalog, classify or explore the possible evolutionary relationships among them.

In this study, we present our attempt to identify all of the likely RDFs for which sequence information has been determined, including those that have not been identified previously. A variety of database search techniques were used to identify sequences that have similarity to known RDFs, generating a list of 63 known or putative RDFs which can be further sub-classified into at least seven smaller groups on the basis of sequence similarity. This compilation of proteins has been analyzed for homologous groups and different chemical characteristics. Many of the identified RDFs fall into specific groups for which some have discernible differences in amino acid composition.

Since the group of proteins influencing the directionality of site-specific recombination reactions are highly diverse, are sometimes involved in additional processes, and are likely to have originated more than once during the course of evolution, we have chosen to describe them as RDFs. While xis and cox are adequate as gene names, the term RDF avoids the use of the potentially misleading terms ‘excise’ and ‘excisionase’.

RESULTS AND DISCUSSION

The identification of RDFs has previously proven difficult due to their small size and sequence diversity. In order to compile as complete a list of RDFs as possible we used a two-step approach. First, we used text-based searches to identify all of the previously annotated RDFs, noting which of these were accompanied by experimental support for their function. Secondly, we conducted a broad database search of both protein and nucleotide records using each of the previously identified RDFs as query sequences.

Determination of annotated excisionases

To collate all of the currently annotated RDFs, we first identified all existing GenBank records containing the text terms excisionase, excisase or xis. From this search a total of 204 records were found, of which 55 were duplicate records with the same accession number, and 62 of the remaining 149 records were duplicate entries of the same sequence with different accession numbers. After these were removed, the list contained records representing 87 unique sequences. Closer examination of these revealed that 39 were included only because the records contained cross-references to RDFs (24 to cognate integrases and 15 to other proteins) and these were also removed to leave a list of 48 putative RDFs.

There are compelling reasons to think that several of the protein entries appearing in this list do not in fact function as RDFs. For example, in mycobacteriophages L5 and D29, gene 34 was initially annotated as a putative excisionase since it encodes a small protein and is closely linked to the integrase gene (19,20). However, another reading frame, 34.1, was subsequently considered as a more likely candidate for encoding Xis since it is more highly conserved between these two phages (21). It is now known that neither 34 nor 34.1 of these phages encodes the RDF and there is good experimental evidence that this activity is provided by gene 36 (22). However, the incorrect annotations resulted in the errant assignment of orf53 from pREAT701 (23) which was annotated as a putative excisionase based on its similarities to D29 34.1. Each of these entries was removed from the list. In three other cases, in plasmid pME2200 and phages phiAR29 and A2, a gene was assigned as a putative excisionase on the basis that it is adjacent to an integrase gene, but without any additional supporting evidence. Since none of these genes has sequence features shared by other RDFs (see below) they were also removed from this list, although we obviously cannot exclude the possibility that they do provide RDF activity. We also removed gene SCE39.01c since it is incomplete and represents part of the longer gene SCE29.20c. Of the 38 remaining RDFs, all have either been shown experimentally to be required for excisive recombination, or are reasonably close relatives of those that have.

It is possible that other RDFs were missed in the text search simply due to variations in nomenclature. For example, the P2 cox gene—which acts as a transcriptional regulator as well as controlling directionality in recombination—was only included because of a linked protein record which identified it as an excisionase. There are three other annotated cox genes (encoded by phages HP1, K139 and WPhi) that were missed in the initial list, but were identified in subsequent sequence searches (see below). Finally, there are two serine integrases, encoded by phage TP901 (15) and Anabaena (XisF) (16) that require an RDF for excisive recombination. There is good experimental evidence for the RDF encoded by phage TP901 as well as the involvement of two proteins, XisH and XisI, in XisF-mediated recombination. The addition of these three proteins raises the number of unique RDFs that we have identified through text searches of existing database entries to 41.

Searches for unidentified excisionases

The compilation of RDFs was expanded by using sequence similarity based methods to search GenBank’s non-redundant (nr) database of proteins. A variety of search algorithms were used including PROBE (24), BLAST (25) and PSI-BLAST (26) (all three algorithms available from NCBI) and each of the 41 annotated RDFs were used as query sequences. At the beginning of the search we recognized that the small size and sequence diversity of this group of proteins may present serious difficulties in identifying RDFs based on sequence characteristics alone. This was particularly evident with the PROBE program, which utilizes iterative rounds of BLAST searches in conjunction with model building to search for distantly related proteins. While this is an effective method for identifying protein families (24), it was not helpful in identifying other RDFs since we frequently found that the resulting list of similar proteins contained none of the original sequences (presumably due to the small size of the query sequences). We therefore focused on using manual iterations of BLAST, and PSI-BLAST, in which proteins that were obviously not RDFs, typically larger proteins containing small segments of similarity, could be excluded.

Since these sequence-based searches generated lists containing many sequences, of which we suspected that only a subset are likely to function as RDFs, we used a secondary criterion to screen the results. Each of the proteins found with a moderate BLAST score (E < 10) was examined for the presence of a nearby integrase gene since RDF and recombinase genes are frequently closely linked. However, for putative RDFs within a bacterial genome (most likely as part of a prophage) we looked as far as 50 kb in either direction for an associated integrase, since this is a reasonable distance that an RDF and recombinase gene could be separated by in an integrated phage genome if these genes flank the attachment site. Using these two criteria, we identified a total of 16 previously unidentified RDFs.

An additional three putative RDFs were identified that did not appear to be accompanied by a recombinase gene. These are encoded by phages DR1455, phi-R67 and TM1. However, these all appeared with good BLAST scores (E < 10^–4) and these three were thus included in the RDF compilation. Two other closely related proteins (B2168_C1_172 and SCE68.26c) also did not appear to have a cognate recombinase although their BLAST scores are also relatively poor (E = 10^–3). Both of these protein sequences came from incomplete genome projects (of Mycobacteria leprae and Streptomyces coelicolor respectively) so that a thorough examination of the flanking sequences was not possible. However, we noted that in the case of B2168_C1_172, there was a good match (E = 10^–22, identified through a TBLASTN search) to a segment of an integrase-like pseudogene located ∼41 kb away. Both of these putative RDFs were added to the list.

Since RDF genes are small, and some even overlap adjacent integrase genes, it is easy for them to escape annotation during analysis of the nucleotide sequence. We therefore collected the GenBank nucleotide records containing ‘bacteriophage’ or ‘integrase’ as key words using ENTREZ. These sequences were translated in six frames (using translate.pl written for this work; J.A.Lewis, unpublished) and formatted as a separate database (using FORMATDB from NCBI) that could be more readily searched using the BLAST program. By using each of the identified RDFs as query sequences, we found two new putative RDFs, one of which was previously described as a pseudogene in prophage DLP12 (27). Since only part of the gene could be identified this was not included in the list. The second candidate (Pspu, encoded by Pseudomonas putida, Table 1) overlaps an adjacent integrase gene and has strong similarity to the Lambda family of RDFs. This was added to the list to generate a final compilation of 63 known or putative RDF proteins. The complete list of RDFs is shown in Table 1.

Table 1. The 63 RDFs analyzed in this study.

Familya	Nameb	Locationc	Host	Evidd	Refs	RDF gi #	Rec typee	Rec gi #
L5	B2168_C1_172	Prophage	Mycobacteria leprae	P	(34)	467073	ND	ND
L5	DR1455	Prophage	Deinococcus radiodurans	P	(35)	6459215	ND	ND
L5	L5	Phage	mycobacteria	E	(22)	15892	Tyr	465416
L5	Rv2657c	Prophage	Mycobacteria tuberculosis	P	(32,36)	1550698	Tyr	1550700
L5	SCE68.26c	Prophage	Streptomyces coelicolor	P	(37)	5123673	ND	ND
L5	D29	Phage	mycobacteria	P	(21,38)	2358239	Tyr	3172283
P22	APSE-1	Phage	Acyrthosiphon pisum	P	(39)	6118035	Tyr	6118033
P22	P22	Phage	Escherichia coli	E	(40)	75990	Tyr	76009
P22	SfV	Phage	Shigella flexneri	P	(41)	2465478	Tyr	2465477
P22	SfX	Phage	S.flexneri	P	(42)	4099029	Tyr	4099030
pSAM2	pNL1	Plasmid	Sphingomonas aromaticivorans	P	(43)	3378297	Tyr	3378303
pSAM2	pSAM2	Plasmid	Streptomyces ambofaciens	E	(44,45)	3043524	Tyr	3043525
pSAM2	pSE101	Plasmid	Saccharopolyspora erythraea	P	(46)	1076058	Tyr	541467
pSAM2	pSE211	Plasmid	S.erythraea	E	(47)	152673	Tyr	152674
pSAM2	Rv2310	Prophage	M.tuberculosis	P	(36)	3261643	Tyr	1524291
pSAM2	Rv3750c	Prophage	M.tuberculosis	P	(36)	2960174	Tyr	2960175
pSAM2	SCE29.20c	Prophage	S.coelicolor	P	(37)	4490998	Tyr	4490997
pSAM2	TM1	Prophage	Arthrobacter sp. TM1	P	(48)	8517283	ND	ND
SLP1	P4	Phage	E.coli	P	(49)	140147	Tyr	15176
SLP1	phi-R73	Phage	E.coli	P	(50)	93825	Tyr	93827
SLP1	SLP1	Plasmid	S.coelicolor	E	(51)	312936	Tyr	312937
SLP1	yp43	Prophage	Yersinia pestis	P	(52)	4106643	Tyr	4106629
HP1	HP1	Phage	Haemophilus influenzae	E	(53)	459180	Tyr	459175
HP1	K139	Phage	Vibrio cholerae	P	(54)	4530499	Tyr	4530503
HP1	phi-R67	Prophage	E.coli	P	(55,56)	141342	ND	ND
HP1	S2	Phage	H.influenzae	P	(57)	1679810	Tyr	1679808
P2	P2	Phage	E.coli	E	(58,59)	76820	Tyr	6136261
P2	WPhi	Phage	E.coli	P	(60)	5824357	Tyr	5824355
L54a	L54a	Phage	staphylococcus	E	(28)	76013	Tyr	76011
L54a	pXO1	Plasmid	Bacillus anthracis	P	(61)	4894317	Tyr	4894320
L54a	T12	Phage	Streptococcus pyogenes	P	(62)	1877428	Tyr	1877429
Tn916	Tn1545	Transposon	Streptococcus pneumoniae	E	(63)	75987	Tyr	76007
Tn916	Tn1549	Transposon	Enterococcus faecalis	P	(64)	8100683	Tyr	8100684
Tn916	Tn5382	Transposon	Enterococcus faecium	P	(65)	3243184	Tyr	3243185
Tn916	Tn916	Transposon	E.faecalis	E	(66,67)	532534	Tyr	532535
Lambda	434	Phage	E.coli	P	(68)	801887	Tyr	215353
Lambda	e14	Prophage	E.coli	P	(69)	7466710	Tyr	3024035
Lambda	H19J	Phage	E.coli	P	(70)	4490351	Tyr	4490352
Lambda	HK022	Phage	E.coli	E	(71)	15761	Tyr	15760
Lambda	HK97	Phage	E.coli	P	(72)	6901614	Tyr	6901614
Lambda	Lambda	Phage	E.coli	E	(1,73)	215134	Tyr	215133
Lambda	Pspu	Prophage	Pseudomonas putida	P	(74)	4520377f	Tyr	4520380
Lambda	P21	Phage	E.coli	E	(75,76)	215449	Tyr	215450
Tn5276	ICESt1	Transposon	Streptococcus thermophilus	P	(77)	6782410	Tyr	6782411
Tn5276	Tn5252	Transposon	S.pneumoniae	P	(78)	1361379	Tyr	1361380
Tn5276	Tn5276	Transposon	L.lactis	E	(3)	1075727	Tyr	1075733
misc	11	Phage	staphylococcus	E	(79)	455128	Tyr	166159
misc	16–3	Phage	Rhizobium meliloti	E	(80)	5824336	Tyr	5824335
misc	186	Phage	E.coli	E	(81)	3337276	Tyr	3337277
misc	D3	Phage	Pseudomonas aeruginosa	P	(82)	9635627	Tyr	9635596
misc	Gifsy-1	Phage	Salmonella typhimurium	P	(83)	3294479	Tyr	3294478
misc	mv4	Phage	Lactobacillus plantarum	P	(84)	684926	Tyr	684925
misc	Mx8	Phage	Myxococcus xanthus	P	(29)	2105132	Tyr	2149006
misc	phi-80	Phage	E.coli	E	(40)	75989	Tyr	75992
misc	phig1e	Phage	lactobacillus	P	(85)	1926325	Tyr	1926326
misc	SgiI	Prophage	Salmonella enterica DT104	P	(86)	9944850	Tyr	9944851
misc	Tn4555	Transposon	Bacteroides fragilis	P	(87)	5453491	Tyr	5453489
misc	VT2-Sa (933W)	Phage	E.coli	P	(88,89)	5881594	Tyr	5881593
misc	ydaQ	Prophage	E.coli	P	(69)	1787608	Tyr	1787607
misc	Rv1584c	Prophage	M.tuberculosis	P	(36)	7476829	Ser	7476830
misc	TP901-1	Phage	L.lactis	E	(15)	2924238	Ser	6808404
misc	xisH	Chromosomal	anabaena	E	(16)	1613875	Ser	1075645
misc	xisI	Chromosomal	anabaena	E	(16)	1613876	Ser	1075645

^aThe family groupings based on sequence similarity.

^bThe name used for the protein in this paper. For phages, plasmid and transposons locations, the name of the element is used to name the protein. For prophage and chromosomal locations, the gene name is used directly.

^cLocation of the gene as found in database searches.

^dType of evidence indicating functionality as an RDF: E, experimental; P, putative.

^eThe type of recombinase as classified by its catalytic residue (Tyr or Ser).

^fThe xis gene has not been annotated as an open reading frame. It is in this nucleotide record (4773–4531).

The main concern in establishing this list of putative RDFs was that we include as many candidate RDFs as possible while excluding all of those that do not function as RDFs. While we believe that in general this was achieved, we also identified proteins that matched some but not all of the criteria, and for which excision function cannot be ruled out. These borderline protein sequences can be viewed at the web site http://www.pitt.edu/~gfh/rdf.html.

Recombination directionality factor classification

Throughout the course of the extensive database searching described above it became evident that the RDFs and putative RDFs do not belong to a single group of closely related proteins. Attempts to align them all using multiple sequence alignment programs were problematic and we could not clearly identify any residues that were highly conserved in all or most members of the group. We therefore attempted to place them into family groups that were more closely related to each other than they were to other putative RDFs. An indication of appropriate groupings was first obtained by examining phenogram representations using guide trees generated by ClustalX (available from BioWeb at http://www.web.tiscalinet.it/biologia) analysis (Fig. 1). While many different trees can be generated, depending on the input order of the sequences and other features of the heuristics (only one example is represented in Fig. 1), those sequences lying close together in the phenogram typically remain together, suggesting that the list might be further sub-divided into smaller groups. This is supported by further PSI-BLAST analyses, where searches identify predominantly those proteins that are near one another in the phenograms.

Figure 1.

Phenogram of RDFs. A tree based on degrees of similarities between RDFs was calculated with CLUSTALX (using the default parameters from http://web.tiscalinet.it/biologia/) and rendered with the DrawGram program (from the PHYLIP package at http://evolution.genetics.washington.edu). The vertical bars indicate groups of RDFs that stay together during multiple cycles of tree generation. The groups are named (as shown on the right) according to a member for which there is experimental evidence of RDF activity.

From these analyses, we propose that 46 of the protein sequences can be assembled into 10 distinct groups or families. Since each of these families contains at least one RDF for which there is experimental support for its function, we named the families after such a member (i.e. P22, L5, pSAM2, SLP1, Tn916, L54a, P2, HP1, Lambda and Tn5276 families). Seventeen sequences did not assemble into families and are listed as miscellaneous. For most of the 10 families the membership is fairly obvious and family members are identified in early rounds of reiterative PSI-BLAST searches; in at least four of the families (Tn5275, Lambda, Tn916 and L54a) the PSI-BLAST searches converged without inclusion of any other putative RDFs. In contrast, members of the L5, pSAM2 and SLP1 subgroups repeatedly identified matches to each other, and these were thus joined into a larger family that we will refer to as the L5–SAM–SLP1 family. Members of the P22 family also appeared in these PSI-BLAST searches, but because of several insertions that are common to the P22 family and absent from the L5–SAM–SLP1 family, we will consider the P22 group as a separate family. The P2 and HP1 groups also appeared to be sufficiently similar to warrant inclusion into a single family (P2–HP1 family). This grouping is also supported by the observation that at least one member of both groups has been shown experimentally to function as cox genes in transcriptional regulation as well as recombination.

Amino acid sequence alignment illustrates the relationships among the RDF family members (Fig. 2). The largest group (with 17 members) is the L5–SAM–SLP1 family. The members of this group vary considerably in distance from the initiating amino acid to the closely related core segment of ∼50 residues which they all share. While this may indicate poor conservation of these parts of the proteins, it could also result from errant assignment of the translation initiation codons. There are no amino acids that are absolutely conserved among all the members, but there are many positions where the chemical character of the amino acids is shared (Fig. 2). The P2–HP1 family (which has six members) has a similar character with several well conserved residues but not that are common to all members.

Figure 2.

Sequence alignments of RDFs. RDFs within individual families are shown using alignments derived from CLUSTALX analysis (using the default parameters as in Fig. 1). Amino acid residues that are identical in 65% of the sequences are highlighted in red and residues that are similar among at least 75% of sequences are shown in blue. Similarity groupings were based on positive scoring substitutions as determined by the BLOSUM 85 substitution matrix (33). The location of a putative helix–turn–helix DNA binding motif is shown above the L5–SAM–SLP1 and P22 families. In cases where RDFs from different sources are of identical sequence, only one was used in the alignment and the view of each alignment is limited to a 150 residue segment containing the related sequences.

The remaining five groups each have fewer members, and it is perhaps not surprising that there are a number of absolutely conserved residues. In the P22 family, there is a segment of ∼110 residues, of which 21 are present in all four members. The Tn916 family is also a tight group (although with only three members; the Tn1545 Xis was not included since it is identical to that of Tn916, and their cognate integrases are also extremely similar), with a common segment of ∼70 residues of which 42 are present in all members. The Lambda, Tn5276 and L54a groups are rather more diverse than these. The Lambda group contains eight family members although the Xis encoded by phages HK97, HK022, 434 and H19J were excluded from the alignment because they are identical to the Lambda Xis. The L54a family is clearly the most diverse of these groups although the pXO1, L54a and T12 RDFs do appear to be more similar to each other than other RDFs, and they share properties regarding protein charge that are unique to this family (see below).

Protein charge

It has been reported previously that phage RDFs are typically basic, with pIs in the range of 9–10 (22,28). However, analysis of the pI values of the RDFs shown in Table 1 shows that while about half of the proteins fall into this category, there is actually a much broader range of pI values represented (Fig. 3). Interestingly, 10 RDFs have a pI <7, and these are listed in Table 2. This includes all three of the L54a family members as well as two (Mv4 and phig1e) that are in the miscellaneous group, but which appear to function with integrases that, along with those associated with the L54a RDF family, are in the LC3 group of recombinases (11) (Table 3). It therefore seems plausible that at least these five may have a shared ancestry.

Figure 3.

Distribution of isoelectric focusing point (pI) values among RDFs. The predicted pI was calculated for each RDF using Compute pI/MW (http://www.expasy.ch). The number of RDFs with pIs within 0.5 pH intervals was determined and plotted. The majority of the proteins are basic, with only 10 of the 63 RDFs having a pI < 7. Five of the seven proteins (Lambda, HK97, HK022, 434, H19J) in the 11–11.5 range have identical sequences.

Table 2. Acidic RDFs.

Name	pI	RDF family	Integrase familya
Mv4	6.51	misc	LC3
phig1e	4.57	misc	LC3
L54a	4.53	L54a	LC3
T12	5.15	L54a	LC3
pXO1	6.87	L54a	misc
11	4.13	misc	phi11
WPhi	6.58	P2	P2
orfA	6.93	pSAM2	N/A
xisH	4.78	misc	serine
xisI	5.88	misc	serine

^aIntegrase family was based on Esposito’s classifications (11).

Table 3. RDF versus integrase classifications.

Namea	RDF family	Integrase familyb
L5	L5	FRAT1
Rv2657c	L5	FRAT1
D29	L5	FRAT1
APSE-1	P22	P22
P22	P22	P22
SfV	P22	P22
SfX	P22	P22
pNL1	pSAM2	misc
pSAM2	pSAM2	misc
pSE101	pSAM2	pSE
pSE211	pSAM2	pSE
SCE29.20c	pSAM2	pSE
P4	SLP1	P4
phi-R73	SLP1	P4
SLP1	SLP1	P4
yp43	SLP1	P4
Tn1545	Tn916	Tn916
Tn1549	Tn916	Tn916
Tn5382	Tn916	Tn916
Tn916	Tn916	Tn916
ICESt1	Tn5276	LC3
Tn5252	Tn5276	LC3
Tn5276	Tn5276	LC3
L54a	L54a	LC3
pXO1	L54a	misc
T12	L54a	LC3
HP1	HP1	P2
K139	HP1	P2
S2	HP1	P2
P2	P2	P2
Wphi	P2	P2
434	Lambda	Lambda
e14	Lambda	Lambda
H19J	Lambda	Lambda
HK022	Lambda	Lambda
HK97	Lambda	Lambda
Lambda	Lambda	Lambda
Pspu	Lambda	Lambda
P21	Lambda	Lambda
11	misc	phi11
16-3	misc	misc
186	misc	P2
D3	misc	P22
Gifsy-1	misc	Phi-80
mv4	misc	LC3
Mx8	misc	misc
phi-80	misc	Phi-80
phig1e	misc	LC3
SgiI	misc	misc
Tn4555	misc	misc
VT2-Sa (933W)	misc	phiCTX
ydaQ	misc	p4

^aThe name of the excisionase gene as described in Table 1. Only proteins for which an Int-Y was found are included.

^bIntegrase family was based on Esposito’s classifications (11).

Other RDF properties

Previous studies suggested that some RDFs have a putative helix–turn–helix DNA binding motif that is responsible for DNA binding (22,29). We thus examined all of the RDFs for the presence of this motif. Using the HTHpred program [written for this work using the method of Dodd and Egan (30)], 23 of the putative RDFs have a probability of >25% of containing this motif, and three others (SCE29.20c, SCE68.26c and TM1) had scores of 2.2 or greater, only just missing the 2.5 cut-off value representing a probability of 25% (30); these are therefore also reasonable candidates for having this motif (Fig. 2). The majority of the predicted DNA binding domains (19 of the 26), are in proteins that are in the L5, P22, pSAM2 and SLP1 families. Thus, of the 22 total members of these families, only three (pSE101, pSE211, SLP1) were not predicted to contain a helix–turn–helix motif by this analysis. None of the Lambda group of RDFs was predicted to contain a helix–turn–helix motif (31).

Evolutionary considerations

The compilation and grouping of these known and putative RDFs reveals at least three important insights into their evolution. First, while the members of some groups (e.g. the Lambda group) are encoded by phages that infect related hosts, this is not true for all families. For example, in the L5 sub-family, there are members encoded by mycobacteriophages as well as an RDF encoded by a putative prophage of Deinococcus radiodurans. In the pSAM2 sub family there are RDFs encoded by both plasmids and prophages within hosts as diverse as Sphingomonas aromaticivorans, Arthrobacter sp., Streptomyces ambofaciens, Saccharopolyspora erythraea, Mycobacterium tuberculosis and S.coelicolor. Assuming that the members of each family or sub-family do indeed arise from common ancestry, this suggests that the RDFs have disseminated broadly throughout the phage and plasmid population by extensive lateral exchange, which appears to be a common theme in phage evolution (32).

The second observation is that the RDFs appear to co-evolve with their cognate integrases. A comparison of independently generated families for the RDFs (this work) and phage integrases (11) reveals considerable congruence (Table 3). For 52 of the RDFs in Table 1, cognate full-length tyrosine integrase sequences have been identified, 40 of which were previously classified (11) and another 12 which have only recently been identified (D.Esposito, Invitrogen Corporation, personal communication). Most of the RDFs that can be grouped into families form groups which correspond to the family groupings of their cognate recombinases. For example, the four members of the Tn916 family of RDFs form the same group as those of their cognate integrases (Table 3). The simple interpretation of this pattern is that although the RDF genes may indulge in widespread lateral movement throughout the phage population that they tend to do so in partnership with their cognate integrase.

A significant departure of this general pattern is seen with the L54a and Tn5276 families of RDFs. In this case, the cognate integrases (ICESt1, Tn5252 and Tn5276 in the Tn5276 family, and L54a, pXO1 and T12 in the L54a family) all form a single family of LC3 integrases (the pXO1 integrase has been classified as a miscellaneous member). The lack of congruence cannot be simply explained as an artifact of the sequence analysis of the RDFs, since the L54a and Tn5276 RDF families are not only evidently different in their primary sequences (Fig. 2) but are substantially different in their overall protein charge (Fig. 3). It therefore seems likely that, at least in this case, the integrase and RDFs have followed distinct evolutionary paths.

The third consideration is whether all of these RDFs are derived from a single common ancestor or whether it is plausible that they have multiple origins. We suggest that there are likely to have been at least four separate origins for the following reasons. First, the L54a group not only shares little or no sequence similarity with the other RDF families but they are quite acidic with respect to their charge, in marked contrast with the other RDF families. It thus seems likely that these arose independently from other RDFs. Secondly, the P2–HP1 cox genes differ from other RDFs in that they are the only ones that also act as transcriptional regulators. We therefore suggest that these may also have arisen independently. Thirdly, the L5, SAM, SLP1 and P22 families all are predicted to contain helix–turn–helix DNA binding motifs, which are not observed in the Tn916, Tn5276 or Lambda families. Taken together with the extent of sequence divergence, it seems plausible that these could also have arisen from an independent origin. Finally, while the Tn916, Tn5276 and Lambda families are distinctly different from each other with regard to their primary structures, they are all small, basic proteins and their evolutionary relationships to each other are less clear. Finally, we recognize that speculating on the origins of these proteins must be cautious, since their small sizes and roles as architectural rather than catalytic molecules may enable more frequent acquisition of function through convergent evolution than would be expected for larger enzymes.

Further considerations

In light of the great diversity of RDFs described here and the likelihood of multiple origins, what proportion of RDFs has been identified? We note that more than 120 Int-Y have been described (11), most of which are likely to utilize an accessory protein to regulate directionality. Since the total number of RDFs described here is only ∼50% of this number, it seems that there are many RDFs for which sequence information already exists, whose functions have not yet been identified. Since we have not found these by the exhaustive database analyses described above, their identification will have to await experimental dissection of other recombination systems.

The architectural role of Xis in Lambda recombination is well established and it seems probable that other RDFs associated with tyrosine recombinases will act similarly. However, the mechanism of directionality regulation has not been clearly established in any of the integration systems that use a large serine recombinase. The only such system that has been investigated biochemically is that encoded by Streptomyces phage φc31, although no φc31 RDF has been identified. However, the φc31 attP site is substantially smaller than those of the Int-Y that have been studied, suggesting a quite different recombinational process that does not require the formation of complex higher order protein–DNA structures (13,14). Nevertheless, there are at least four RDFs that are associated with an Int-S, all of which are classified as miscellaneous. One putative RDF (Rv1584c) associated with an Int-S (Rv1586c) has weak sequence similarity to the L5 family of RDFs, of which Rv2657c is a member. Rv2657c, which is associated with an Int-Y (Rv2659c), and Rv1584c are both found in prophage-like elements (φRv2 and φRv1 in M.tuberculosis) that have a colinear arrangement of these genes, adding further support to the identification of Rv1584c as a putative RDF. There is experimental evidence supporting the function of the TP901 RDF as well as Anabaena XisH and XisI (15,16). However, the role that the RDF proteins play in control of directionality in the Int-S systems remains to be elucidated.

Finally, the compilation and classification of RDF and putative RDFs should be helpful in the future annotation of plasmid, phage and prophage sequences. As new RDFs are identified and added to this list we expect to see the formation of additional families of RDF proteins and to gain further insights into their evolution and function.

Web site

Further details on the RDFs, putative RDFs, and borderline protein sequences are available on the World Wide Web at http://www.pitt.edu/~gfh/rdf.html. Additional information on the cognate integrases is available at the tyrosine recombinase web site at http://members.home.net/domespo/trhome.html.