GENERAL INTRODUCTION

Top Previous Next References

Host-Controlled ModificationA Historical Perspective

Awareness of the phenomenon of restriction and modification and the consequent revolution in molecular biology grew from the observations of microbiologists in the early 1950s that the host range of a bacterial virus (phage) could be influenced by the bacterial strain in which the phage was last propagated. While phages produced in one strain of a bacterial species would readily infect a culture of the same strain, they might only rarely achieve successful infection of cells from a different strain of the same species. This finding implied that the phages carried an "imprint" that identified their immediate provenance. The occasional successful infection of a different strain resulted in the production of phages that had lost their previous imprint and acquired a new one, i.e., they acquired a new host range, hence the term host-controlled modification (see reference 5 for an early review).

In the 1960s, elegant molecular experiments showed the imprint to be a DNA modification that was lost when the phage DNA replicated within a different bacterial strain; phages that conserved one of their original DNA strands retained the modification, whereas phages containing two strands of newly synthesized DNA did not (6). The modification was shown to protect the DNA against an endonuclease, the barrier that prevented, or "restricted," the successful propagation of an incoming phage genome. Later it was proven that the modification and restriction enzymes both recognize the same target, a specific nucleotide sequence. The modification enzyme is a DNA methyltransferase (161) that methylates specific bases within the target sequence, and in the absence of the specific methylation the target sequence renders the DNA sensitive to the restriction endonuclease. When DNA lacking the appropriate modification imprint enters a restriction-proficient cell, it is therefore recognized as foreign and degraded by the endonuclease. For each unmodified target sequence, there is only a low probability that it will become modified and escape attack by the endonuclease. Since the host-controlled barrier to successful infection by phages that lack the correct modification was referred to as restriction, the relevant endonucleases have acquired the colloquial name of restriction enzymes. Similarly, the methyltransferases are more commonly termed modification enzymes. Classically, a restriction enzyme is accompanied by its cognate modification enzyme, and the two comprise a restriction and modification (R-M) system. Most restriction systems conform to this classical pattern. There are, however, some restriction endonucleases that attack DNA only when their target sequence is modified; such modification-dependent restriction enzymes do not, therefore, coexist with a cognate modification enzyme.

The classical R-M systems and the modification-dependent restriction enzymes share the potential to attack DNA derived from different strains and thereby restrict DNA transfer. They differ in that in one case an associated modification enzyme is required to protect DNA from attack by the cognate restriction enzyme, while in the other a modification enzyme specified by one strain imparts a signal that provokes the degradative activity of a restriction endonuclease in another.

Early examples of host-controlled modification, though they were not always recognized as such, were reviewed by Luria (107). Two papers in particular stimulated interest in the phenomenon. In one, Bertani and Weigle (17), using temperate phages ( and P2), identified the classical R-M systems of Escherichia coli K-12 and E. coli B. In the other, Luria and Human (108) identified a restriction system of the second, nonclassical kind.

T-even phages were used by Luria and Human as test phages, and after their growth in a mutant E. coli host, these phages were restricted by wild-type E. coli K-12 but not by Shigella dysenteriae. An understanding of this restriction phenomenon requires knowledge of the special nature of the DNA of T-even phages. When the DNA of T-even phages is replicated, the unusual base 5-hydroxymethylcytosine (HMC) substitutes for cytosine, and the hydroxymethyl group is then glucosylated in a phage-specific pattern. In the mutant strain of E. coli used by Luria and Human, glucosylation fails. The vulnerable nucleotide sequences of T4, normally protected by glucosylation, are recognized by an endonuclease in E. coli K-12 because they include the modified base HMC rather than cytosine residues. S. dysenteriae does not restrict the nonglucosylated phage because it lacks the relevant endonuclease. Later it was discovered that methylated cytosine residues in the context of the correct nucleotide sequence also evoke restriction by these modification-dependent endonucleases (145).

The biological tests for a classical R-M system are illustrated by the pioneering experiments of Bertani and Weigle (17). Phage  grown on E. coli strain C (.C or .0), where E. coli C is a strain that apparently lacks an R-M system, forms plaques with poor efficiency (efficiency of plating [EOP] of 2 × 10⁴) on E. coli K-12 because the phage DNA is attacked by a restriction endonuclease. Phage  grown on E. coli K-12 (.K) forms plaques with equal efficiency on E. coli K-12 and E. coli C, since it has the modification required to protect against the restriction system of E. coli K-12 and E. coli C has no restriction system. In contrast, .K will form plaques with very low efficiency on a third strain, E. coli B, since E. coli B has an R-M system with different sequence specificity from that of E. coli K-12.

The classical restriction endonucleases of E. coli K-12 and B were not only the first to be detected but the first to be purified (104, 117). The demonstration that the restriction endonuclease from E. coli K-12 produced a digest of phage  DNA comprising large DNA fragments (117) was exciting, for it implied a highly specific target for the endonuclease, a view supported by genetic studies which showed that target sequences in phage genomes could be mutated and mapped (54, 93, 124). The rational expectation was that type I restriction enzymes would cut DNA close to their target sequences, but Horiuchi and Zinder (68) showed otherwise. The enzyme from E. coli B cut the DNA of phage f1 nonspecifically at considerable distances from the unmodified target sequences. Type I restriction enzymes therefore failed to provide the anticipated analytical reagents, but they raised the alert so that Ham Smith immediately appreciated the significance of his observation that Haemophilus influenzae strain Rd degraded P22 phage DNA, and as a consequence, he purified HindII, the first type II restriction enzyme (160). The phenomenon of restriction, identified for type I and methylation-dependent systems, laid the foundations for modern molecular biology, yet the molecular complexity and biological importance of these systems remain to be fully understood.

Classification. R-M systems are classified on the basis of their composition and cofactor requirements, the nature of their target sequence, and the position of the site of DNA cleavage with respect to the target sequence. Currently three distinct, well-characterized types of classical R-M systems are known (I, II, and III), although a few do not fit well into any of these. The first R-M systems to be identified, those characteristic of E. coli K-12 and E. coli B, were designated type I, but the type II systems became better known, since they, unlike type I restriction enzymes, cut DNA into discrete fragments. A summary of the properties of the principal types of classical R-M systems is given in Fig. 1, and the reviews selected (5, 10, 18, 65, 83, 92, 118, 144, 149, 170, 192) include some of particular historical significance.

View larger version (18K): [in this window] [in a new window]   FIG. 1.   Distinguishing characteristics and organization of the genetic determinants and subunits of the different types of R-M systems. ENase, restriction endonuclease; Mtase, methyltransferase. Modified with permission from a figure in reference 83.

INTRODUCTION TO TYPE I R-M SYSTEMS

Top Previous Next References

General Characteristics

The finding that type I R-M systems exist as closely related members of a family has been of fundamental value to their analysis (10). Evidence for related systems was first indicated by the demonstration that mutants with defects in the allelic genes encoding EcoKI and EcoBI could complement each other. On the basis of such tests, it was inferred that each enzyme comprised three subunits, that the subunits of EcoKI and EcoBI were interchangeable, and that the subunit encoded by one gene, hsdS, confers target sequence specificity on the multimeric complex (22, 58, 69). EcoKI and EcoBI became the founder members of a family of type I systems, type IA. The essential difference between two members of one family resides in the regions of the HsdS subunit that confer sequence specificity.

Hybridization screens of bacterial DNAs and serological screens of bacterial extracts first suggested that allelic genes might also encode sufficiently dissimilar type I R-M systems to warrant their separation into different families (122). As expected, the nucleotide sequences of the hsd genes for EcoKI and EcoBI would hybridize to each other and antibodies raised against EcoKI reacted with EcoBI, but in contrast, DNA probes comprising the EcoKI genes failed to hybridize with those of E. coli 15T, which encoded EcoAI; similarly, antibodies against EcoKI did not cross-react with EcoAI. The hsd genes in these two strains behave as alleles in genetic tests (7) but have very different nucleotide sequences (34, 80). EcoAI defines a second family of type I systems, type IB.

A third family, type IC, headed by EcoR124I, includes plasmid-encoded members (141) and a chromosomally encoded relative (EcoprrI) identified in a natural isolate of E. coli. The first representative of a fourth family, ID (180), is the R-M system of Salmonella enterica serovar blegdam, identified initially on the basis of biological tests (26). The genes encoding the ID systems, but not those for EcoprrI (183), map to the same region of the E. coli chromosome as those for type IA and type IB. Currently, each type I R-M system identified in E. coli, or in a close relative, has been allocated to one of the four families (the members are listed later in Table 2).

ENZYMES

Top Previous Next References

Introduction

The restriction endonucleases from three families of type I systems (IA, IB, and IC) have been purified and characterized. Each is a large oligomeric complex of relative molecular weight ~400,000 to 500,000, which in the presence of Mg²⁺, ATP, and AdoMet functions as an endonuclease on a DNA substrate that includes unmethylated target sequences but catalyzes the transfer of methyl groups from AdoMet to DNA substrates that include hemimethylated target sequences (187). Early evidence indicated that EcoKI comprises subunits of three sizes, with approximate molecular weights of 135,000, 62,000, and 52,000. Estimates of the relative amounts of each subunit in EcoKI indicated two of each of the larger polypeptides and one of the smallest (118). An analysis of the polypeptides specified by the cloned hsd genes of E. coli K-12 permitted the correlation of polypeptides with genes (151) and the consequent suggestion that the stoichiometry of the subunits within the complex is R₂M₂S₁.

Convincing evidence that endonuclease activity is associated only with the R₂M₂S₁ complex was obtained much later (45). For a time it appeared that EcoR124I and -II, members of the IC family, have one rather than two R subunits and that they retain endonuclease activity in the absence of AdoMet (75). However, it has now been shown that the EcoR124I complex has a tendency to lose one HsdR polypeptide (76) and that AdoMet copurifies with EcoR124I and other type IC enzymes (41, 133; P. Janscak and T. A. Bickle, personal communication). Type IB complexes readily lose both HsdR subunits (167), but it seems probable that the predominant active complex for any type I R-M system includes two HsdR subunits.

A modification enzyme without endonuclease activity was demonstrated in extracts from E. coli B, in addition to the larger complex with both activities (53, 100). Strains of bacteria encoding EcoKI (169), EcoAI (167), and EcoR124I (174) are now known to possess a complex with only methyltransferase activity, as well as the large R-M complex. In all cases, the stoichiometry of the enzyme lacking endonuclease activity is M₂S₁.

In the following sections, evidence for the functional domains within each subunit is summarized before the activities of the enzyme complexes are considered.

Variable sequences. When the sequences of specificity genes (hsdS) of members of the type IA family were searched for differences that would correlate with the recognition of different target sequences, each pairwise comparison revealed two long regions (~450 to 500 bp) of apparently unrelated sequence (referred to as variable regions), in addition to some minor differences in the intervening conserved sequences. Together, the two variable regions comprise the major portion of the gene, but given the bipartite nature of the target sequences, it was inescapable that each variable region might encode polypeptide sequences responsible for the recognition of one component of the target sequence (62). Subsequently, comparisons of the nucleotide sequences of hsdS genes within the IB (80), IC (64), and ID (A. J. B. Titheradge and N. E. Murray, unpublished data) families also revealed two long variable regions flanking a conserved region, as shown for a IC member in Fig. 2a.

View larger version (43K): [in this window] [in a new window]   FIG. 2.   Specificity polypeptide of EcoR124I. (a) Organization indicating the two variable regions (TRDs) and the regions conserved in sequence for all members of one family. Repeats of similar sequence (below the arrows) are identified by their conserved marking. The number of the amino acid residue at the beginning and end of each region is given below the diagram. (b) Model of Kneale (86), in which the repeated sequences form linkers joining the TRDs in a rotationally symmetrical configuration. The nucleotide sequences identified by each TRD are shown. (c) Consequent model of the methyltransferase, in which the two HsdM subunits bind to the linker region to generate an enzyme with pseudodyad symmetry.

Conserved sequences. Two structural roles are anticipated for the conserved regions of HsdS polypeptides. These are the maintenance of the relative positions of the two TRDs and the specific associations of HsdS with other components of the R-M complex. Evidence that the conserved peptide sequence that links the two TRDs serves to position the TRDs on the target sequence came from the chance occurrence of a derivative of EcoR124I with a new specificity. The sequence of the hsdS gene of this R-M system, EcoR124II (originally referred to as EcoR124/3), is consistent with a recombination event in which unequal crossing-over between a misaligned 12-bp duplication in the central conserved sequence of the hsdS gene, or slipped mispairing during DNA replication, has led to a triplication of that sequence. As a result, the two components of the target sequence of EcoR124II are separated by 7 bp rather than the 6 bp in the target sequence of EcoR124I (63, 142). The region in which the four additional amino acids occur is predicted to be strongly helical, and the increase in length is likely to be sufficient to accommodate a change in spacing of 1 bp (142).

Imposition of symmetry. In addition to the regions conserved in HsdS subunits of the same family, there are regions of similarity within each HsdS subunit (Fig. 2a). For EcoKI (type IA), this was noted by Argos (8) for regions in the central and carboxy-terminal conserved sequences. In the IB family, a segment of amino acids is common to both the amino-terminal and central conserved sequences (Fig. 3c), in addition to regions similar to those in EcoKI (80). The analysis of sequence similarities in the specificity subunit of a type IC member indicated that a sequence in the central conserved region was incompletely repeated at the carboxy terminus, with the remainder of the repeat located at the amino terminus (Fig. 2a). This "split repeat" led Kneale (86) to propose a model in which the amino and carboxy termini of HsdS are in close proximity so that they associate to form a linker domain similar to that provided by the central conserved sequence (Fig. 2b). These two domains of similar sequence would then dictate the symmetrical association of the two HsdM subunits (Fig. 2c). The repeated sequences identified within the HsdS subunits of the type I systems of the gram-positive bacterium Lactococcus lactis (153) are remarkably reminiscent of those identified in the type IC family, and a close analysis of members of the type IA family finds some evidence for a short sequence at the amino terminus of HsdS that has similarity with part of the repeat in the central conserved region (86).

View larger version (34K): [in this window] [in a new window]   FIG. 3.   Experimental support for the symmetrical arrangement of the TRDs (large open circles or arrows) within specificity subunits. (a) Predicted arrangement of two truncated subunits of EcoDXXI and the consequent recognition of a hyphenated symmetrical target sequence. (b) The HsdS subunits of StySKI (type IB) and EcoR124I (type IC). The IB family of enzymes have a long N-terminal conserved sequence, while the IC family has a long C-terminal conserved tail. The amino-terminal TRD of StySKI and the carboxy-terminal TRD of EcoR124I (indicated by arrows) have 36% amino acid identity. The DNA targets recognized are indicated. The complement of the EcoR124I 3' target sequence (CGAY) is a degenerate version of the StySKI 5' target sequence (CGAT), consistent with a symmetrical organization of two similar TRDs. (c) Circularly permuted variants of the specificity gene of EcoAI (type IB) can retain activity. The normal organization of regions within an HsdS polypeptide is given in the first diagram, followed by an active but circularly permuted variant. The black segments identify the well-conserved repeats.

Association of subunits. Interaction with other subunits was an anticipated role for the conserved sequences of HsdS polypeptides (62). Convincing evidence in support of this prediction derives from the truncated derivatives of HsdS polypeptides of EcoDXXI (Fig. 3a) and EcoR124I. Two truncated polypeptides can substitute for one normal HsdS subunit if the subunits retain their ability to interact with HsdM. Active complexes, i.e., those in which HsdS retains the ability to bind HsdM, are recognized by their new specificities (1, 109, 113). Analyses of truncated polypeptides implicate the three conserved regions of HsdS (Fig. 2a) in binding HsdM (109). As predicted, a deletion within the central conserved region of an HsdS subunit prevents binding to HsdM (189).

Molecular model. Sturrock and Dryden (166) supplemented sequence data from known type I R-M systems with data for putative systems identified in genomic sequences to derive a molecular model for the recognition region of a TRD. These authors used amino acid sequences combined with secondary-structure prediction to align 51 TRDs. The inclusion of secondary-structure prediction enhances the strength of the amino acid alignments, making distant similarities more apparent. This is particularly helpful because the amino acid identities between TRDs that recognize different target sequences are usually less than 25%. The alignments suggest a common tertiary structure, and secondary-structure predictions with strong similarity to the known structure of the TRD of the type II HhaI methyltransferase. Of the 51 sequences compared, that of the amino-terminal TRD of HsdS from EcoKI shows the closest similarity to the TRD of M · HhaI, sufficiently so to suggest that EcoKI, like M · HhaI, might interact with DNA via two short polypeptide loops flanking a -strand.

View larger version (47K): [in this window] [in a new window]   FIG. 4.   Model of the amino acid segment comprising residues 43 to 157 of the amino terminus of EcoKI interacting with DNA as proposed by Sturrock and Dryden (166). (a) Side view. (b) Bird's eye view. Residues in red are substitutions that cause the loss of restriction and modification; those in yellow have no detectable effect on activity. Residue 141, in gray, was previously thought to lead to a loss of both activities but is now known to impair rather than inactivate. The DNA structure is from the complex of M · HhaI with its DNA target sequence (85) and therefore shows an extrahelical cytosine rather than adenine. The figure is modified with permission from that published in O'Neill et al. (131).

HsdM needed for DNA binding. The HsdS subunits of EcoKI (193) and EcoR124I (132) are insoluble in the absence of HsdM, but the HsdS subunit of EcoR124I has been produced in soluble form as a glutathione-S-transferase (GST) fusion product (97). The resulting fusion protein is unable to make specific complexes with DNA. The GST moiety at the N terminus interferes with the recruitment of the second HsdM subunit, preventing assembly of the active methylase; its removal allows HsdS to assemble with HsdM units to form an active modification enzyme (115). These findings are consistent with the model (Fig. 2c) in which the conserved sequence at the N terminus forms part of the region that contacts HsdM. The HsdS subunit of EcoAI, a type IB enzyme, is soluble in the absence of HsdM, but it, too, requires HsdM to make sequence-specific complexes (74). For EcoKI, even the association with one HsdM subunit is sufficient to promote a sequence-specific interaction, though with a higher K_d than the full M₂S₁ complex (136). It would appear that HsdM is important for the positioning of the domains of HsdS, at least in part by maintaining its rotationally symmetrical configuration (74, 86).

Like HsdS, HsdM is essential for restriction as well as modification. AdoMet, the methyl donor for modification, is an essential cofactor for restriction (117), and early experiments showed that AdoMet binds to the HsdM subunit of EcoKI (24). Consistent with this, the amino acid sequence predicted from the nucleotide sequence of the hsdM gene of E. coli K-12 (106) includes a version of a motif, N/DPPF/Y/W, that is characteristic of both N-6 adenine and C-4 cytosine methyltransferases and is now generally referred to as motif IV (for reviews of methyltransferases, see references 29, 42, and 129).

Active site. The HsdM subunit is well conserved (~90%) within a family (156, 183), but comparisons between members of different families generally indicate only 25 to 30% identity (156). Site-directed mutagenesis of the hsdM gene of E. coli K-12 demonstrated the relevance of two conserved sequences to methyltransferase activity (191). A change in the sequence predicted to be motif I (D/E/SXFXGXG) abolished the binding of AdoMet, while changes in motif IV (N/DPPF/Y/W) prevented catalysis but did not affect binding of the cofactor. Nevertheless, a tryptophan residue substituted for phenylalanine in motif IV of EcoKI is sufficiently close to the AdoMet-binding site to enhance cross-linking with the methyl donor (191). Subsequently, molecular modeling of type I HsdM subunits, based on sequence alignment and predicted secondary structures, suggested a domain in HsdM subunits which resembles that of the  class of type II N-6 adenine methyltransferases (44) and includes the six motifs found in the catalytic domain of M · TaqI (99). The model of the catalytic domain for the HsdM subunit of EcoKI is consistent with the location of proteolysis-sensitive sites (31) and the mutational analysis of Willcock et al. (191). In this model, the glycine at amino acid residue 177 (motif I) is located close to the cofactor, and the phenylalanine at residue 269 (motif IV) is positioned at the edge of the active site, where it can interact with the target adenine should this be flipped out of the DNA helix, as has been shown for the cytosine residue during methylation by M · HhaI (85). Base flipping allows access to the target base and is therefore predicted to be common to all DNA methyltransferases as well as some DNA repair enzymes (150).

A complex with only methyltransferase activity has been purified from representatives of the type IA, IB, and IC families. The stoichiometry of this active methyltransferase is M₂S₁ (167, 174). EcoKI and EcoBI (type IA) dissociate to M₁S₁ and M₁ (43, 45, 100, 136), but for EcoKI the active form has been shown to be M₂S₁ (43). The type I modification enzymes catalyze the transfer of the methyl group from AdoMet to the N-6 position of specific adenine residues in their respective target sequences, probably utilizing the cationic- interactions proposed for those type II enzymes that methylate adenine at the N-6 position or cytosine at the N-4 position (152).

Discrimination of methylation state. The three type I families differ significantly in their relative responses to hemimethylated versus unmethylated target sequences. EcoAI, the type IB representative, like most known methyltransferases, shows little or no preference for hemimethylated DNA (167). In contrast, in vitro experiments indicate that members of the IA and IC families have a very strong preference for hemimethylated DNA (approximately 100-fold compared to unmethylated DNA); the data for the type IA system have been obtained from experiments with both plasmid DNA (169) and oligonucleotide duplexes (43) as substrates, and those for IC were obtained from oligonucleotide duplexes (175). Consistent with the in vitro evidence, unmodified phage DNA is a poor substrate in vivo for methylation by a type IA modification system (82, 105), but in apparent contradiction to the in vitro evidence, unmodified phage DNA was found to be a good substrate in vivo for methylation by an rm⁺ type IC system (S. Makovets and N. E. Murray, unpublished data).

+/

Base flipping. The footprinting of M · EcoR124I provides the following circumstantial evidence that the adenine residues flip out to provide access for methylation (114). Two sites that are hypersensitive to hydroxyl radical cleavage have been identified within the target sequence, one per strand, each associated with the adenine that is the substrate for methylation. The authors argue that it is unlikely to be fortuitous that the hypersensitive site on each strand coincides with the adenine that is the target for methylation. A plausible explanation is that binding of the enzyme induces a marked conformational change in the structure of the sugar-phosphate backbone of the DNA in the region around those bases that are the targets for methylation. Mernagh et al. (116) also showed that M · EcoR124I binds more strongly when either uracil or an abasic site is substituted for one of the target adenines. Again, this would be consistent with the idea that the adenine residues, like the cytosine residues in the HhaI target sequence (85, 150), are flipped out during the methylation reaction. Base analog experiments with the HhaI methyltransferase have shown that this enzyme binds more strongly to substrates in which the target base is mismatched (84).

The polypeptide sequences of HsdR subunits representative of the four families of type I systems have only 20 to 30% identity (180). All these sequences, however, include motifs characteristic of ATP-binding proteins, consistent with the ATP dependence of restriction. In addition, there are other conserved sequences indicating the presence of motifs previously identified in ATP-dependent helicases and putative helicases (59, 123, 180). There are several superfamilies of ATP-dependent helicases. Members of superfamilies 1 and 2 include the motif DEAD or a variant of this motif and are often referred to as DEAD-box proteins (103). The motifs in HsdR subunits are indicative of superfamily 2 (59). The structures of DEAD-box helicases indicate that the motifs form a nucleoside triphosphate (NTP)-binding pocket and a portion of a nucleic acid-binding site (88, 186; see reference 64a for a recent review). It is suggested that the conserved motifs define an "engine" that powers translocation on single-stranded DNA and unwinding of duplex DNA (64a). Velankar et al. (186) present an elegant model for the coupling of the energy to DNA translocation in which the enzyme "inchworms" along a single DNA strand using unpaired bases. Nearly all reported mutations affecting DEAD-box motifs impair the hydrolysis of NTP or the coupling of NTP hydrolysis to nucleotide unwinding (64a). Mutagenesis of the hsdR gene of EcoKI showed that each of the seven DEAD-box motifs is essential for restriction in vivo (35, 188).

The number of bacterial genomes for which sequences are available has increased significantly since the alignment of HsdR sequences reported by Titheradge et al. (180). The additional sequences have improved the reliability of alignments and the prediction of the secondary structure. Comparisons of HsdR sequences with Rep and PcrA, DNA helicases of known structure (88, 186), suggest that the HsdR subunits have the same secondary structure as the helicases in the region that includes the DEAD-box motifs (36). In Rep and PcrA, the motifs reside in two domains that couple ATP hydrolysis to DNA helicase activity (20). The fragmentation patterns produced by limited proteolysis of HsdR are consistent with the location of the DEAD-box motifs in two domains similar to those first observed in Rep and PcrA, in which the DEAD-box motifs cluster around a cleft between two domains (36).

The earlier alignments of HsdR sequences (180) detected an additional conserved sequence in the N-terminal part of the polypeptide which proteolysis experiments indicate to be in a separate domain from those including the DEAD-box motifs (36). This additional conserved sequence (Fig. 5) has similarities with motifs associated with DNA nicking in both type II restriction enzymes and the RecB family of nucleases. Site-directed mutagenesis proved the relevance of this motif to the endonuclease activities of EcoAI (78) and EcoKI (36, 37).

View larger version (12K): [in this window] [in a new window]   FIG. 5.   Domains and motifs of HsdR of EcoKI. The N- and C-terminal regions are omitted, since their roles are not known. The two domains that include the DEAD-box motifs correlate with IA and 2A, as determined for the structures of DNA helicases (36). Substitutions for the underlined amino acids confer a restriction-deficient phenotype.

The earliest biochemical interest in restriction enzymes was as proteins that made specific interactions with DNA by a recognition process assumed to be intolerant of errors. The question of specificity for type I R-M systems extends to include the mechanism by which each enzyme not only recognizes the methylation state of its target sequence, but then reacts as a methyltransferase if the target sequence is already hemimethylated or as an endonuclease if it is unmethylated. Finally, of special interest is the mechanism by which type I R-M enzymes translocate DNA for considerable distances before breaking phosphodiester bonds in both strands of the DNA duplex. Associated with this mechanism is the problem of what halts translocation and triggers DNA cleavage. Complete answers to all of these complex molecular problems are not yet available (see reference 170 for a recent review).

DNA binding. The restriction pathway is presumed to be initiated by the ATP-dependent conformational change originally reported for EcoKI (19) but analyzed only recently by footprinting (137). EcoKI remains the best-studied R-M complex in terms of sequence-specific binding, the effect of the cofactor AdoMet on representatives of the type IB and IC enzymes being less easy to assess because it is difficult to separate AdoMet from these enzymes (41, 133; P. Janscak and T. A. Bickle, personal communication). EcoKI binds to DNA in the absence of ATP. A strong footprint of 42 to 46 bp is detected only if the DNA includes a target sequence. On the addition of ATP in the presence of AdoMet, unmethylated or hemimethylated target sequences remain protected, and the footprint shrinks to 30 bp; both fully modified target sequences and nonspecific DNA lose all protection. ATP and AdoMet are both needed for the conformational change in response to target sequences, though S-adenosyl homocysteine may be substituted for AdoMet and a nonhydrolyzable analog may be substituted for ATP (137).

Recognition of methylation state. In the presence of the functional cofactors, EcoKI methylates hemimethylated DNA but initiates ATP-dependent translocation if the target sequence is unmethylated. AdoMet has long been implicated in determination of the methylation state of the target sequence. Burkhardt et al. (27) suggested that the HsdM subunits would use the methyl group of the AdoMet as a probe for the presence of the methylated base in the major groove. For modified DNA, steric hindrance from the methyl groups on the adenine residues could prevent a conformational change. Only if both adenines were unmethylated would the HsdM subunits enter the major groove to give the "closed conformation," in which the HsdR subunits are appropriately positioned to initiate DNA translocation. Recent data indicate that the TRDs of HsdS enter the major groove (28), and by analogy with type II modification enzymes, the bases that are to be modified are anticipated to flip out of the helix (116). Nevertheless, AdoMet could still serve as a probe for the methylation state of the target DNA (134). Steric hindrance by AdoMet could block the positioning of a methylated base within the active site after the adenine residues become exposed. This refinement of the model maintains a critical role for AdoMet in restriction as well as modification, and a simple prediction is that loss of AdoMet binding to HsdM should cause an rm phenotype. Consistent with this, a mutant of EcoKI in which AdoMet binding to the HsdM subunit is blocked as the result of a single-amino-acid substitution in motif I (D/E/SXFXGXG) (191) is deficient in restriction as well as modification (V. Doronina and N. E. Murray, unpublished data).

Communication. The means by which the enzyme communicates the methylation state of its target sequence has been probed by looking for mutations in hsdM or hsdS that prevent access to the restriction pathway. Analysis of an hsdM or hsdS mutant with an rm⁺ phenotype could identify an enzyme locked into the modification mode, or it could indicate that assembly of the R-M complex is prevented. The rm⁺ mutants of the EcoKI system resulted from substitutions in the amino-terminal third of HsdM and also showed enhanced modification, i.e., they were m* mutants, more suggestive of defects in communication than assembly (82). Indeed, one of these mutants has been shown to make an R-M complex that is defective in restriction (D. T. F. Dryden and N. E. Murray, unpublished data).

Subunit interactions. Mutations conferring an rm⁺ phenotype have been identified in the hsdS genes of type IA and IC systems. In EcoR124I (189) these mutations are at the junction between a conserved region and a TRD, while in EcoKI they can be close to the junction (197) or at various positions within the TRD (131). A cautious interpretation of r phenotypes for EcoKI, however, is required by the recent discovery that mutations in hsdM that result in inadequate modification induce host-mediated alleviation of restriction (111) (see section on modulation of restriction activity). A mutation in hsdS may prevent binding of HsdS to HsdM or, as suggested by Weiserova and Firman (189), the TRD may influence the precise positioning of HsdR, perhaps in response to the methylation state of the target sequence. HsdM interacts and communicates with HsdR, but for HsdS the evidence is still unclear.

Functional analyses of motifs. The hsdR gene for EcoKI is now well charted, with mutations in regions that encode sequences common to all type I R-M systems (Fig. 5). These mutations indicate that each conserved region is important for restriction. Amino acid substitutions in the DEAD-box motifs do not prevent the conformational change associated with tight binding to the target sequences in the presence of AdoMet and ATP. These proteins, however, are all deficient in ATPase activity and DNA translocation, consistent with a role for the DEAD-box motifs in the coupling of ATP hydrolysis to DNA translocation (35, 37). The failure to translocate DNA was demonstrated by an in vivo assay in which wild-type EcoKI translocates the T7 genome from the phage particle into the bacterial cell (57).

DNA translocation. The type I R-M enzymes cleave DNA at variable positions remote from their recognition sequences (68). Electron microscopy has been used to identify possible intermediates in the reaction leading to cleavage of both linear and covalently closed plasmid DNA (52, 195). The ATP-dependent formation of DNA loops, both twisted and untwisted, has been detected. A model was proposed in which the enzyme binds to its recognition sequence, makes a second nonspecific contact with DNA, and subsequently moves the DNA past the bound complex, generating loops in an ATP-dependent process (195). It was also suggested that if type I R-M enzymes were topoisomerases, they would overcome the conformational problems encountered should the enzyme, while anchored to its target sites, maintain contact with the major groove as the DNA is pulled towards the complex. Elegant experiments using catenated plasmid DNA (171) provided direct confirmation that communication between recognition and cleavage sites stems from a process in which the enzyme follows the contour of the DNA substrate. Meanwhile, using linear DNA, Studier and Bandyopadhyay (165) provided evidence for a model in which DNA is pulled past the bound protein on both sides of the recognition sequences, and endonuclease activity results when two translocating complexes meet (Fig. 6). This model was derived from the analysis of products obtained from the digestion of T7 DNA with EcoKI. When the restriction reaction was synchronized by the addition of ATP to protein-DNA complexes in the presence of AdoMet, diffuse bands of DNA were detected on gels, consistent with the production of relatively discrete DNA fragments. The ends of the fragments, according to the model, would focus around the midpoints between the adjacent recognition sequences. Earlier, though less discriminating, in vivo experiments suggested the cutting of DNA between recognition sites (23). According to the Studier model, a common stimulus for cleavage of the DNA could be the hindrance of translocation (Fig. 6). Recent experiments demonstrate the stimulation of DNA cutting when two translocating enzymes from different families collide and even when a translocating enzyme encounters a Holliday junction that is unable to migrate (77).

View larger version (24K): [in this window] [in a new window]   FIG. 6.   ATP-dependent DNA translocation. (a) Model of Studier and Bandyopadhyay (165). EcoKI bound to adjacent target sequences translocates DNA towards itself. Collision blocks translocation and stimulates the nicking of each DNA strand. Two domains of HsdR flanking the DNA are indicated. (b) In this variant, the EcoKI complexes have dimerized prior to translocation (51). (c) When translocation is impeded by some other protein or structure, endonuclease activity is stimulated (77).

The assembly of EcoKI has been analyzed in vitro (45). An assembly pathway relevant to the bacterial cell was proposed on the basis of experiments quantifying the interactions between intermediate complexes and subunits. The methyltransferase (M₂S₁) is formed from the reversible association of M with M₁S₁. HsdR binds very tightly to both M₁S₁ and M₂S₁, but the only complex with endonuclease activity is R₂M₂S₁.

The relative strengths of protein-protein interactions determined in vitro can be used, at least in part, to explain regulation of the R-M activities in vivo. It is obvious that regulation is essential when hsd genes are transferred to a modification-deficient recipient. Experiments fail to find evidence for transcriptional regulation of hsdR (106, 139). The assembly pathway could, in the following way, form the basis of posttranslational control after expression of the hsd genes. Initially, HsdM and HsdS will exist mainly as HsdM and M₁S₁ with very little M₂S₁ until the concentrations of the subunits reach a critical level. HsdR will bind M₁S₁ and consequently will be unavailable for binding to M₂S₁, thereby imposing a further delay in the production of R₂M₂S₁. Recent experiments for EcoKI indicate that there is a lag of many generations (~11) before cells become fully modification proficient, as assessed by their ability to modify infecting .0 (S. Makovets and N. E. Murray, unpublished data) and longer, ~15 generations, before restriction proficiency is established (138). It was proposed that any of the intermediate complexes (M₁S₁ and R₁M₁S₁ or HsdR) could be targets for cellular proteases (45), thereby delaying the appearance of functional restriction enzyme. However, recent studies on the host-controlled modulation of the restriction activities (see next section) question a major role for assembly pathways in controlling the restriction activities of either EcoKI or EcoAI.

All the known type I R-M enzymes are likely to assemble by a similar pathway, although the affinities of analogous protein-protein interactions may differ. For members of the type IB and IC families, the M₁S₁ intermediate appears to be less prevalent than for the type IA family, but for EcoAI (type IB), R₂M₂S₁ readily dissociates to yield the methylase and free HsdR, while for type IC enzymes, the relative affinities of HsdR for M₂S₁ and R₁M₂S₁ are quite different. The preferential stability of the R₁M₂S₁ intermediate for EcoR124I (76) is currently the only explanation for the easy establishment of type IC hsd genes in a new host.

BIOLOGY OF TYPE I R-M SYSTEMS

Top Previous Next References

Host-Controlled Modulation of Restriction Activity

Self-protection. Bacteria are assumed to tolerate the presence of a classical restriction endonuclease because the cognate modification enzyme maintains the methylation of target sequences within chromosomal DNA. A maintenance methylase does not, however, explain the long-established fact that genes encoding R-M systems are readily transferred to recipient cells in which the chromosomal DNA is unmodified. Under these circumstances, a delay in the appearance of restriction activity in the recipient cell is necessary to allow time for methylation of the unmodified chromosome (138). Other experiments demonstrate that the maintenance methyltransferase activity of an r⁺m⁺ cell is sometimes unable to cope with the protection of unmethylated targets that arise in response to DNA damage (111). In both these cases, it is now known that there is an additional level of control over the endonuclease activity of some type I R-M systems which enables the bacteria to survive in the absence of complete modification of chromosomal target sequences. This section of the review will trace the development of our current understanding of the mechanisms by which restriction activity is modulated.

Proteolytic control. The heterooligomeric nature of type I R-M systems provides opportunity for the regulation of the R-M activities purely on the basis of the affinities with which different subunits bind to intermediates in the assembly pathway (45). Nevertheless, efficient transmission of the genes encoding EcoKI was shown to depend on a host function (139). A number of energy-dependent proteases are now known to play important regulatory roles in bacteria, and these were obvious candidates for this host function. The energy-dependent proteases identified in bacteria are often large oligomeric assemblies within which substrate recognition is imposed by one component and protease activity is imposed by another. The unfolded protein substrate is translocated to a chamber within the oligomeric complex and is then degraded processively (for reviews, see references 60 and 61). The protease specified by clpX and clpP was implicated in the regulation of restriction activity; in the absence of either ClpX or ClpP, acquisition of the hsd genes for EcoKI or for EcoAI led to the death of m recipients (110). While ClpXP is a protease, ClpX itself can function as a substrate-specific chaperone. Loss of ClpX imposed a bigger barrier to gene transfer than loss of ClpP, suggesting a dual role for ClpX. This could imply a requirement for ClpX as a chaperone in addition to its role as a component of the ClpXP protease (110).

View larger version (30K): [in this window] [in a new window]   FIG. 7.   Generation of unmodified target sequences following UV irradiation. Methylated strands of DNA are shown as thick lines, and unmethylated strands are shown as thin lines. Homologous recombination, involved in the repair of double-strand breaks or postreplicative repair, can generate regions of unmethylated, double-stranded DNA via annealing of two unmethylated strands (regions within boxes). In addition, the SOS mutagenesis pathway leads to new (unmodified) target sequences as the result of base changes.

View larger version (38K): [in this window] [in a new window]   FIG. 8.   Model for the mechanism of ClpXP-dependent RA (111). When a type IA or IB R-M complex binds to an unmodified target sequence, translocation is initiated and ClpXP recognizes the HsdR subunit of the complex. ClpXP degrades HsdR, preventing further translocation and endonuclease activity.

Significance. Modification has been assumed to be essential to protect the resident chromosome from restriction. However, a mutation in hsdM (F269G) which results in a modification-deficient, restriction-proficient EcoKI complex is not lethal. Residue 269 of HsdM is in motif IV (NPPF) of the catalytic domain of the methyltransferase (see section on the HsdM subunit). In the absence of methyltransferase activity, the chromosome of E. coli K-12 will contain ~600 unmodified targets, yet the cell survives even when repair of DNA breaks is blocked by a mutation in recA. Survival is dependent on ClpXP, consistent with the ClpXP-dependent degradation of HsdR (111). RA apparently protects the bacterial chromosome from restriction in the complete absence of modification, but when analyzed by infection with unmodified , the alleviation is not complete; the EOP of .0 is 10¹. These facts raise two new but possibly related questions. First, why does phage DNA entering the cell show some susceptibility to restriction, whereas the resident chromosome does not? Second, why do unmodified targets on the chromosome but not those of infecting  DNA induce the alleviation of restriction? The classical explanation in which modification provides the basis for EcoKI to distinguish host DNA from foreign DNA is no longer adequate, but it can be argued that DNA modification is necessary to allow high levels of restriction activity. Important questions concerning the distinction between unmodified phage and unmodified host DNA remain to be answered. It may be relevant that the two substrates differ in their location, in their association with other proteins, or in their topology. Any of these differences could be critical for the regulation of restriction activity.

Many phages and transmissible plasmids have some means of moderating their susceptibility to R-M systems. These include unusual modification of DNA, a low frequency of target sequences, and the production of a protein that interferes with one or more of the activities of an R-M system (for reviews, see references 18 and 92). DNA that enters a bacterium in a single-stranded form, as is the case for transmissible plasmids and some phages, does not avoid restriction but becomes susceptible after the synthesis of the second strand. This is true for bacterial DNA transferred by conjugation (5). Unmodified targets in newly synthesized DNA are susceptible to restriction. It is not known whether recombination is a prerequisite for unmodified targets to acquire the status of resident DNA and induce RA.

Glucosylation of the HMC residues of the DNA of T-even phages provides effective protection against most restriction enzymes; importantly, it also identifies phage DNA so that phage-encoded enzymes can selectively degrade the host chromosome (98, 146). The Mom function of phage Mu converts adenine residues within a specific nucleotide sequence to N₆-(1-acetamido)adenine, a modification that is known to interfere with restriction by EcoKI and EcoBI (92).

The nucleotide sequences of some plasmids and the genomes of many phages have been analyzed for the frequencies of the target sequences recognized by those restriction enzymes that the transmissible agents might encounter (155, 190; see reference 18 for a review). Most of the statistically significant examples of underrepresentation are for the targets of type II rather than type I restriction enzymes. The apparent lack of effect on targets for type I systems may in itself may be significant, since E. coli, the species for which most data are available, maintains a great variety of type I R-M systems encoded by allelic genes that confer different specificities (see section on Detection, Distribution, and Diversity). The promiscuous plasmid RP4 was found to be deficient in target sequences for potentially relevant type II endonucleases but not for the known type I R-M systems of E. coli (190). It was noted, however, that there is little evidence for the presence of promiscuous plasmids in E. coli strains collected prior to 1954.

Most commonly studied coliphages (18), many plasmids (14, 15, 16, 30, 148), and at least one conjugative transposon (154) specify proteins that interfere with restriction, some of which provide general protection against all known type I systems (15). The efficacy of antirestriction functions poses the intriguing question of how a protein specified by the unmodified DNA of a transmissible agent is able to act before the genome that encodes it is attacked by the restriction enzyme. Some interesting answers emerge.

The simplest solution is encountered for bacteriophage P1. This phage defends its DNA from type I restriction enzymes by proteins, designated Dar for defense against restriction, which are coinjected with any encapsidated DNA (71). This mechanism allows efficient generalized transduction between strains with different type I R-M systems as well as unrestricted infection by P1 phages. How Dar protects DNA is unclear. It does not function in trans and therefore may remain associated with the incoming DNA. It was suggested (71) that Dar protein bound to DNA might hinder the translocation of DNA, but this does not readily explain the requirement for different Dar proteins to combat or antagonize different families of type I enzymes.

The genomes of phages T3 and T7 enter E. coli according to a lengthy program in which the leading genes are transcribed prior to internalization of the remainder of the genome. The antirestriction gene of these phages (0.3) is the first gene to be transcribed, and its product protects the remainder of the T7 genome from attack by type I restriction enzymes. Experimental evidence, derived from phages manipulated to include two targets for EcoKI in the leading region of the T7 genome, indicates that during the first few minutes of infection, the leading region is accessible to RNA polymerase but not EcoKI, and it was suggested that early transcription occurs in an EcoKI-free compartment (121), but this compartment has not been defined.

Transmissible plasmids that encode an antirestriction function, called Ard for "alleviation of restriction DNA," resemble T3 and T7 in that the ard gene is located in the leading end of the DNA, but for plasmids it is single-stranded DNA that is transferred (5' to 3') by conjugation. Convincing though indirect evidence supports an elegant model for the action of ArdA in which the genes in the leading end of the DNA are transcribed from special promoters recognized within secondary structures of single-stranded DNA (4, 12). This allows transcription of ardA and the accumulation of antirestriction protein before the transferred strand is converted into duplex DNA, the substrate for the restriction enzyme.

The entire  genome enters the bacterial cell before any phage gene is expressed, and the DNA of phage  is an immediate target for restriction (46). This is also true for the T7 genome when it enters the bacterium by courtesy of a  capsid (57). Ral, the "restriction alleviation" function of , serves only to enhance the modification of those phages that escape restriction (105, 196). Enhanced modification is a significant advantage for a phage if the modification activity of the system, like that of EcoKI, is relatively ineffective on unmodified DNA. Ral affects members of the IA family but not type IB enzymes, for which modification of unmethylated DNA is efficient (105). The in vivo evidence implicates Ral as an antagonist of restriction by enhancing the modification activity of type IA systems on unmethylated DNA, but no molecular basis for this mechanism has been established. Rac, a cryptic prophage of E. coli K-12, specifies an analog of Ral, termed Lar; the presence of the Rac prophage enhances the EOP of .0 approximately 10-fold (105).

The 0.3 gene products of T3 and T7 are the only antirestriction functions available in significant quantities for detailed molecular analyses. These proteins, sometimes referred to as Ocr, for "overcoming restriction," bind to type I R-M enzymes and inactivate both the endonuclease and methyltransferase activities (9). Purified 0.3 protein binds tightly to EcoKI or EcoBI, preventing the enzyme from binding to DNA (9). These early experiments indicate that 1 or 2 dimers of 0.3 protein per EcoKI complex are sufficient to saturate the binding sites. More recent experiments indicate a value of 1 for the methyltransferase and 2 for the endonuclease (C. Atanasiu and D. T. F. Dryden, personal communication). The 0.3 protein of T3 but not that of T7 is endowed with an additional means of antagonizing type I restriction systems. This second activity degrades AdoMet, the cofactor for restriction and the methyl donor for modification (164).

Belogurov and Delver (13) noted that the 0.3 protein of T7 includes acidic sequences similar to a motif in various Ard proteins. This acidic sequence is also similar to those sequences that are common to the HsdS subunits of type I R-M systems, and these authors proposed that the antirestriction proteins compete with HsdS for assembly with the HsdM and HsdR subunits of the enzyme, but there is no evidence that the HsdS subunit of the methyltransferase is displaced by 0.3 protein in vitro (C. Atanasiu and D. T. F. Dryden, personal communication).

The glucosylated T4 genome is insensitive to type I R-M systems, but nevertheless T4 encodes a minute polypeptide, Stp, that inhibits the chromosomally encoded type IC system EcoprrI. This R-M system is notable because the purified complex includes an additional protein, an anticodon nuclease specified by a gene that separates hsdR from hsdS. The anticodon nuclease remains latent until Stp inactivates EcoprrI (see reference 80a for a detailed review).

Type I R-M systems have been detected by the barrier they impose on gene transfer, as exemplified by the original identification of the R-M systems of E. coli K-12 and E. coli B (17). Phages are a useful and sensitive test for the presence of R-M systems in laboratory strains of bacteria, but they do not provide a general screen for detecting new restriction systems. Many bacterial strains, even within the same species, and particularly when isolated from natural habitats, are unable to support the propagation of the available test phages (10).

The technical importance of type II restriction enzymes promoted intensive screens of extracts from bacterial cells for the presence of endonuclease activities that cleaved DNAs into discrete fragments. Screens for endonucleases have not been used for type I systems and are unlikely to be effective, given that the products of digestion are not discrete DNA fragments and that ATP will promote the activity of exonuclease V.

The identification of new R-M genes via sequence similarities initially relied on experimental approaches. Screening by DNA hybridization is applicable to any bacterial species, but its usefulness is limited by the availability of suitable probes. Genes encoding members of known families of type I systems may be detected by hybridization to DNA probes derived from representative hsd genes (10). More generally, screening databases of predicted polypeptide sequences identifies putative R-M systems in genomic sequences even when the genes would escape detection using probes derived from known families.

Biological evidence currently documents functional type I R-M systems in Bacillus subtilis (194), Citrobacter freundii (34), Klebsiella pneumoniae (101, 184), L. lactis (153), Mycoplasma pulmonis (47), Pasteurella haemolytica (66), Salmonella enterica (26), and Staphylococcus aureus (158), in addition to those originally identified in E. coli. All these systems have been shown to provide a barrier to phage infection, either in their original host or after cloning of their genes in E. coli. The three genes diagnostic of a type I system were commonly identified by coding sequences, but for the R-M system of S. aureus (SauS2I), the biological activity has not been associated with gene sequences. Screening of the present database of complete genomic sequences indicates that about half the bacterial genomes include closely linked genes that are predicted to encode the three polypeptides comprising a type I R-M enzyme (Table 1), and others include remnants, i.e., one or two hsd genes, commonly hsdS. Some strains of one species include a type I R-M system, and others do not, e.g., B. subtilis (182). It is clear that complex restriction enzymes are not predominantly associated with any group of bacteria: previous bias toward their presence in members of the Enterobacteriaceae reflects the bacteria commonly used for genetic analysis. A more surprising observation is that one strain of Helicobacter pylori uses more than 1% of its small genome to encode R-M systems, three of which are type I (181).

For E. coli, the genomes of wild-type isolates have been screened with probes specific for each of the three families of chromosomally encoded type I R-M systems (IA, IB, and ID) identified in the Enterobacteriaceae. Thirty-seven strains from the E. coli reference library (ECOR strains) were tested, and 17 included sequences identified by one of the probes. No genome was found to hybridize with more than one probe, as expected if the hsd genes are alleles of one locus and consistent with the subdivision of the systems into discrete families using DNA hybridization as a test for sequence similarity (11). The serB-linked hsd genes identified in eight serotypes of S. enterica (26) include representatives of the three families, but the hsd genes in three of the eight strains were not detected with any probe (10, 180). A summary of the classification of the type I R-M systems in the Enterobacteriaceae is given in Table 2. The allocation to families has not always been confirmed by complementation tests, but interspecific tests show that defects in EcoKI (IA) can be complemented by StyLTIII (185) and those in StySBLI (ID) can be complemented by subunits of both EcoR9I (A. J. B. Titheradge and N. E. Murray, unpublished data) and KpnAI (J. Ryu, personal communication).

The two strains of H. pylori for which complete genomic sequences are available, strains 26695 (181) and J99 (3), each include coding sequences for three putative type I R-M systems. The predicted polypeptide sequences for systems within a strain, either 26695 or J99, show only modest levels of amino acid identity (20 to 40%). The three systems in strain 26695 have homologs in J99, based on amino acid identities (80 to 90%) of HsdR and HsdM. Some of the HsdS sequences are sufficiently short to suggest that they are not functional. The HsdS sequences of one pair of homologs (HP0462 and jhp0414) are of normal length for known type I systems. Their sequences have high identity in the amino-terminal TRD but low identity in the carboxy-terminal TRD, consistent with allelic genes that confer a different specificity.

A modified concept of families is emerging from studies of L. lactis, in which it appears that HsdS subunits can be exchanged between two families of type I R-M systems. The HsdR and HsdM subunits specified by two plasmids have the high levels of amino acid identity (~90%) expected for members of one family (153). The sequences of these plasmid-encoded subunits, however, show only 40% identity with those specified by chromosomally encoded genes. These low levels of identity between the plasmid- and chromosomally encoded systems of L. lactis are consistent with the concept of two families of enzymes. In contrast, all the sequence data for HsdS subunits, whether plasmid or chromosomal in origin, indicate a single family of enzymes with differences in HsdS sequences found principally within the TRDs. The conserved regions of HsdS subunits include those sequences that interact with HsdM, subunits that are well conserved within a family (62, 86, 109, 189). Schouler et al. (153) note that, irrespective of the family designation, the C-terminal parts of the HsdM subunits of the lactococcal systems have a common sequence that could identify a region involved in the association of HsdM with HsdS. Complementation tests indicate that the HsdS subunits of plasmid systems can form functional complexes with the chromosomally encoded HsdM and HsdR subunits. The present data suggest that a reservoir of independent hsdS genes on plasmids provides an effective "combinational" system for varying the target specificity of the catalytic subunits specified by the host.

The type I R-M systems of M. pulmonis illustrate the most sophisticated system for the expression of different specificities; in this case, variation is found among the descendants of one bacterium (48). The bacterial chromosome includes two hsd loci, each of which contains two hsdS genes flanking hsdM and hsdR. Site-specific DNA inversion can occur using any of three sites within the hsdS genes. There is only one promoter for each hsd locus, so that only one orientation permits transcription of a functional hsd operon, but the alternative sites of inversion create a family of functional HsdS subunits by gene rearrangements. This switching of specificities is reminiscent of the phase variation of virulence determinants.

Of all known R-M systems, those classified as type I appear to be uniquely suited for diversification of sequence specificity. In part this is because a single subunit is responsible for the sequence specificity of both enzyme activities. Importantly, therefore, the specificities of the endonuclease and methyltransferase change concomitantly. In addition, some, perhaps all, type I R-M systems have coevolved with their hosts to ensure that neither the acquisition of such a system nor a change in the specificity of an established system leads to endonucleolytic attack of the host chromosome (110, 130). While both the shared specificity subunit and the control of restriction activity facilitate change of sequence specificity, the scope for diversification is enhanced by the presence of two different TRDs within one specificity subunit. The two TRDs of a homodimeric type II restriction endonuclease dictate a rotationally symmetrical target sequence. In contrast, for known type I systems, the two TRDs recognize different sequences, and it was shown that specificity can be changed not only by reassortment of TRDs, but also by altering the spacing between them (Fig. 9). Chance examples of these two routes to diversification were encountered in the laboratory. The new R-M system found as the by-product of P1-mediated transduction (25) arose from a genetic exchange between different hsdS genes; a cross-over within the conserved DNA sequences in the interval separating the variable regions that specify TRDs generated a new combination of TRDs (55, 127). The second serendipitous change in DNA specificity arose in a type IC system (70). In this family, a short duplication of 12 bp within the conserved sequence of an hsdS gene provides the opportunity for mispairing and unequal crossing-over. The target sequences of the new specificity are separated by an extra base pair as the consequence of a triplication of the 12-bp repeat (142).

View larger version (55K): [in this window] [in a new window]   FIG. 9.   Evolution of type I R-M systems with new specificities. (a) Recombination between hsdS genes produces hybrid genes and chimeric S polypeptides. StySPI and StyLTIII are naturally occurring type I R-M systems (see Table 2). StySQ and StySJ have hybrid hsdS genes (55, 56). The regions originating from StySPI are hatched, and those originating from StyLTIII are stippled. Reassortment of the target recognition domains (TRDs) accordingly gave rise to recombinant recognition sequences (56, 127). Site-directed mutagenesis of the central conserved region of the StySQ hsdS gene produced StySQ*, comprising only the amino-terminal variable region from StySPI and the remainder from StyLTIII. The StySQ* target sequence confirms that the amino-terminal variable region is in fact a TRD responsible for recognition of the trinucleotide component of the sequence (33). (b) Sequence specificity may also be altered by changing the length of the nonspecific spacer of the target sequence. The S polypeptides of EcoR124I and EcoR124II differ only in the number of times a short amino acid motif (X = TAEL) is repeated within their central conserved regions (142), resulting in extension of the spacer in the target sequence from six nucleotides (N6) in EcoR124I in N7 in EcoR124II. The recognition sequence of EcoDXXI also contains a nonspecific spacer of seven nucleotides, corresponding to three TAEL repeats in its S polypeptide (64). Chimeric S polypeptides recognize the predicted target sequences (64). Modified from a figure in Barcus and Murray (10) with permission.

New specificities as the result of the reassortment of TRDs depend upon a natural reservoir of related hsdS genes available for genetic recombination. The target sequences of five members of the IA family are known, and only two have a common component (Table 2). However, the generation of a TRD with a different specificity has never been witnessed, despite experiments designed to select for a relaxation in the specificity of EcoKI [5'AAC(N₆)GTGC] to StySPI [5'AAC(N₆)GTRC, where R is a purine] (N. E. Murray, unpublished data). It seems probable that the evolution of a new specificity will require an appropriate combination of amino acid changes.

The allelic hsdS genes of a family of type I systems retain a record of some of the evolutionary changes associated with the diversification of sequence specificity, while the HsdM and HsdR polypeptides are conserved to associate with variant HsdS subunits. Nevertheless, detailed comparisons of nucleotide sequences of the hsdM genes of E. coli K-12 and B revealed that they are far more divergent in terms of both synonymous and nonsynonymous changes than any of 29 other pairs of genes available for comparison (156). It is readily argued that the diversity in specificity reflects the action of natural selection, and it is tempting to speculate that the high level of divergence between the hsdM alleles also reflects selection. Levin (102) has pointed out that phage-mediated selection of bacterial R-M systems is likely to be frequency dependent, thereby favoring the retention of rare genotypes and the promotion of diversity. Phage-mediated selection is expected to act directly on hsdS. Some new hsdS alleles will be acquired by recombination, and therefore the high divergence at hsdM may result from its tight linkage with hsdS. Allelic diversity at the hsd locus of E. coli parallels the polymorphisms noted for cell surface antigens where frequency-dependent selection has been invoked for favoring rare antigens (21).

The high level of sequence divergence within a family (123, 156), the abnormal G+C content within hsdS genes (39), and the alternative families of hsd genes encoded by allelic genes (11) support the concept of lateral transfer of hsd genes. Lateral transfer of genes of very dissimilar nucleotide sequence does not explain their common chromosomal location or, in the case of the ID family, a different gene order (180). Would one expect the maintenance of sufficient DNA homology in flanking regions to direct insertion of different families of hsd genes at the same site? At one locus in B. subtilis, strain-specific variation identifies the alternative to genes for a type I R-M system as those specifying type II systems (182). In this case the transfer of alternative sets of genes to B. subtilis 168 has been shown to rely on flanking homology (T. A. Trautner, personal communication), but why would closely related bacteria have such different gene sequences at a common location?

The hyphenated, symmetrical target sequences detected for mutant type I systems in which the hsdS genes are truncated (Fig. 3a) imply that two "half" specificity subunits (HsdS*) within one TRD can substitute for a single wild-type HsdS with two TRDs (1, 113). This is consistent with a complex of R₂M₂S*₂ rather than the normal R₂M₂S₁. Sequence analyses provide evidence for the origin of the present hsdS gene by gene duplication. The evidence for the type IA family is relatively weak (8), probably because the first part of the gene has suffered a deletion, but the evidence remains strong for members of the IB family (80). In this case a sequence of 192 nucleotides at the beginning of hsdS is repeated in the center, with 55 of the 65 encoded amino acids being identical. The recently published sequences of the hsdS genes in the gram-positive bacterium L. lactis (153) reinforce the evidence for gene duplication, identifying a split repeat within which 47 of 59 encoded amino acids are identical (Fig. 10).

View larger version (11K): [in this window] [in a new window]   FIG. 10.   Repeated amino acid sequences in an HsdS polypeptide of L. lactis. The sequence (153) has been aligned to emphasize the repeats flanking the predicted target recognition domains.

Ancestral enzymes, in which HsdS has only a single TRD, probably had the composition R₂M₂S₂. Some unusual R-M systems are now known to comprise an endonuclease that is like a type II enzyme but is accompanied by a modification enzyme more akin to a type I methyltransferase. The AhdI methyltransferase, for example, is made up of subunits resembling HsdM and a half HsdS. The enzyme recognizes a hyphenated, symmetrical target sequence, and the subunits have more sequence similarity with type I systems than has been found for classical type II methyltransferases (G. G. Wilson, personal communication).

At the level of DNA sequence, little evidence exists for homology between either the hsd alleles of different families of enzymes or even the variable regions of hsdS genes within a family unless the variable regions specify the same target sequence. Nevertheless, evidence for homology remains in the nucleotide sequence of variable regions of hsdS genes from different families when the TRDs specify the same target sequence (32).

It seems significant that the amino acid sequences and predicted secondary structures of type I enzymes have led to model structures similar to those deduced experimentally for type II enzymes, in particular, the active site of type I methyltransferases (44) and the regions of the TRDs located close to the DNA target (131, 166). Irrespective of family, or even type, the evidence from the protein domains is consistent with the concept that R-M systems have common building blocks.

R-M systems are both widespread and diverse in nature, yet there is no indication that a strain of E. coli deprived of its restriction system is disadvantaged under normal laboratory conditions. It is evident from surveys of E. coli that allelic variation provides diversity (11), and this diversity is therefore expected to have relevance within the species. The classical explanation for the presence of R-M systems is that restriction is a barrier to phage infection. There is much evidence for the impact of R-M systems on phages (see section on Mechanisms by Which Plasmids and Phages Avoid Restriction), but there is no proof that bacteria maintain R-M enzymes to protect themselves against phages. Experiments indicate that bacteria sharing a habitat with phages rapidly acquire resistance to infection, and restriction-proficient bacteria encoding a system with a novel specificity are likely to be at an advantage only when colonizing a new habitat (89, 90, 102). This advantage may be sufficient to impose frequency-dependent selection for diversity.

As early as 1973, it was realized that the DNA breaks induced by restriction endonucleases could be recombinogenic (S. Lederberg, cited in reference 143), and it has been pointed out that because the breaks made by type I enzymes are not within the target sequences, any gene has a chance to be close to a break (52a, 140).

For a time there appeared to be a conflict between a recombinogenic role for RecBCD, alias exonuclease V, in salvaging the products of restriction by recombination and its proven importance in the degradation of the linear DNA fragments produced following the restriction of phage (157). This apparent contradiction was resolved by an appreciation of the "recombinator" sequence Chi, an 8-base sequence missing in phage  and well represented in the bacterial chromosome (140, 163). RecBCD travels along the DNA from an end, and on encountering an appropriately oriented Chi sequence, DNA degradation is attenuated and recombination activity is stimulated (40, 162). Following conjugation, most recombination is believed to result in the incorporation of long "chunks" of donor DNA (159), with the first exchange occurring within the DNA close to the origin of transfer. This recombination is mediated by the RecBCD pathway. Fragmentation of DNA by restriction should increase the opportunity for the incorporation of short, rather than long, stretches of DNA. Sequence comparisons of chromosomal DNA from several E. coli reference strains show that members of these natural isolates differ by short lengths of nucleotide sequence (119). Analyses of gene transfer between natural isolates of E. coli and E. coli K-12 by transduction and conjugation document the influence of R-M systems on the pattern of DNA fragments incorporated into the recipient chromosome, creating mosaic genomes of short sequences originating from different strains (112, 120). R-M systems therefore will influence the flux of genetic material between bacterial populations, enhancing the opportunity for the acquisition of advantageous coding sequences in the absence of deleterious ones (120).

The groups of Kobayashi (87, 128) and Ehrlich (94) have provided experimental evidence that type II R-M genes behave as "selfish" entities; the R-M genes are not readily lost from the host cell but maintain the presence of the plasmids that encode them. These results imply that the loss of type II R-M genes leads to descendants that die because they retain some endonuclease activity at a time when there is insufficient modification enzyme to protect all the target sequence within the chromosome. An explanation for diversity of sequence specificity in the case of selfish R-M genes stems from the finding that one system excludes another if both recognize the same target sequence (96). In other words, systems with the same target sequence behave like plasmids in the same compatibility group, and selection could arise from competition for recognition sequences rather than being driven by phages.

The type I systems studied, EcoKI, EcoR124I, and EcoAI, do not behave like type II systems in many relevant respects. Loss of the genes specifying a type I system has no detectable effect on bacterial viability (95, 110, 130). For EcoKI, a mutation that blocks modification activity but leaves hsdR functional, e.g., deletion of hsdM and hsdS or a missense mutation in hsdS, can be detrimental to viability but only in the absence of both ClpXP and some other unknown function, and even then only in stationary-phase cultures (S. Makovets and N. E. Murray, unpublished data). Apparently, in the absence of ClpXP, a mutation in hsdS can lead to progeny in which dilution of the modification enzyme, following cell growth and division, results in the inadequate modification of chromosomal DNA, thus generating targets susceptible to residual restriction endonuclease. Normally, ClpXP-dependent RA successfully protects the bacterial chromosome from damage by the endonuclease activity of EcoKI. On the current model (Fig. 8), any EcoKI molecule caught in the act of translocating chromosomal DNA is prevented from breaking the DNA. For EcoKI and EcoAI, a refined target-specific system exercises tight control of endonuclease activity. This is probably particularly necessary for a system like EcoKI, in which modification of unmethylated DNA is very inefficient.

It has been suggested that the proteolytic control of endonuclease activity permits efficient variation of the specificity of type I R-M systems (111). The organization of type I enzymes, the bipartite asymmetrical nature of the target sequence, and the host-mediated alleviation of restriction all emphasize the importance of diversity. Selection is presumed to contribute to the maintenance of allelic diversity. R-M systems in general may have evolved under a variety of pressures, but it is tantalizingly difficult to prove the biological relevance of such varied and sophisticated molecular systems.