Chapter 52

DNA Restriction and Modification Systems

NICOLE REDASCHI and THOMAS A. BICKLE

INTRODUCTION

The phenomenon of restriction and modification (R-M) was first observed in the course of studies on bacterial viruses in the early 1950s. Several authors reported that certain strains of bacteria inhibited ("restricted") the growth of bacterial viruses previously propagated on a different strain (10, 83). The molecular explanation for this effect was discovered in the early 1960s: the restriction of viral growth was due to endonucleolytic cleavage of the viral DNA by site-specific endonucleases (2, 4, ). Some of these restriction endonucleases were found to possess very useful properties, and their subsequent exploitation in the 1970s was the key to the development of genetic engineering technology (90, 111).

Restriction systems occur exclusively in unicellular organisms, mainly bacteria, and in some of the viruses of these organisms (64, 112). Their main purpose seems to be the protection of a cell from invading foreign DNA, the most obvious sources of which are bacteriophage genomes and conjugative plasmids (15, 72). A prerequisite to this biological function is the ability to distinguish between foreign DNA and the cell’s own genome. This is done in one of two ways. Some systems protect the cellular DNA from restriction by modification methylation of adenosyl or cytosyl residues within the specific sequences recognized by the restriction enzymes (the classic R-M systems). Other systems recognize and cut DNA bearing foreign modifications. In this second case, the host DNA is protected from restriction because it is not modified, and consequently, these systems do not include a modification methylase.

RESTRICTION SYSTEMS SPECIFIC FOR MODIFIED DNA

The first restriction systems described in Escherichia coli were the Rgl (restricts glucoseless DNA) systems (83, 109, 110). The RglA and RglB systems are active against T-even phage mutants that contain nonglycosylated 5-hydroxymethylcytosine (hm5C) in their DNA. Wild-type T-even phages have glycosylated hm5C residues in their DNA instead of cytosine, and this modification renders them immune to restriction by the Rgl systems as well as to restriction by most of the classic R-M systems. Since hm5C is not a common modification, the Rgl systems were long considered to be specific for T-even phages (72, 109). Not until the late 1980s was it realized that the Rgl systems are in fact responsible for the difficulties encountered with cloning methylcytosine-containing DNA from various sources, including higher plants and animals, in E. coli (16, 91, 104, 106, 136, 144, 145). It turned out that the Rgl systems restrict not only DNA containing hm5C but also DNA harboring 5-methylcytosine (m5C) and N ⁴-methylcytosine (m4C). The systems were renamed Mcr (methyl-cytosine restricting) to more appropriately describe their specificities (103). There are two Mcr systems in E. coli; they are called McrBC and McrA.

McrBC (RglB) is encoded by two genes (mcrB and mcrC) that form an operon (reviewed in reference 102). It restricts DNA containing hm5C, m5C, and m4C in specific sequence contexts that are not yet fully defined. Restriction seems to require pairs of R-methylcytosine (where R is either purine) separated by 40 to 80 bp of DNA, with cleavage occurring between the modified cytosyl residues at multiple positions on both strands (126). A feature distinguishing McrBC from all other known DNases is its absolute requirement for GTP to cleave an appropriately modified DNA in vitro, whereas ATP is a competitive inhibitor of the reaction (126).

McrA (RglA) is encoded by a prophagelike element, e14, present in many strains of E. coli K-12 (52, 105). So far, the only known substrates of McrA are DNAs methylated by the HpaII and SssI methylases (both cytosine methylases) (52, 63, 97, 106). Both McrA and McrBC restrict hemimethylated DNA as well as DNA methylated on both strands (126). Another system in E. coli, Mrr (modified adenine recognition and restriction), restricts some DNA sequences that include N ⁶-methyladenine (m6A) or m5C residues (51, 63, 71, 135); however, the exact sequence specificity is unknown. Interestingly, the genes encoding the McrBC and Mrr systems in E. coli K-12 lie adjacent to the hsd genes that encode the classic type I R-M system EcoKI. The three closely linked restriction systems gave this part of the genome the nickname "immigration control region" (63, 105).

CLASSIC R-M SYSTEMS

The classic R-M systems comprise pairs of enzymes with matching DNA sequence specificities but different enzyme activities: the restriction enzyme is an endodeoxyribonuclease that cleaves DNA only if its recognition sequences are not modified. Restriction enzymes produce DNA ends with 5' phosphates and 3' hydroxyl groups. Depending on the particular enzyme, the ends may be fully double stranded or have short 3' or 5' single-stranded protrusions. The modification enzyme is a DNA methyltransferase that specifically methylates either adenosyl or cytosyl residues within the recognition sequence, thereby rendering the DNA resistant to the activity of the cognate restriction endonuclease. The primary function of the modification methylase is to protect the cell’s own genome from restriction.

R-M systems have been classified according to cofactor requirements, subunit composition of the enzymes, structure of the recognition sequences, and position of DNA cleavage relative to the recognition sequences (recently reviewed in references 15 and 141). Presently, three distinct, well-characterized types of R-M systems are known. However, several of the recently identified R-M systems do not fit well into any of the three types (I, II, or III), thus demanding an expansion of the current classification scheme.

Type I R-M Systems

Type I R-M systems appear to be confined to the family Enterobacteriaceae. Fewer than a dozen natural type I R-M systems have been discovered so far, in E. coli, and in Citrobacter, Salmonella, and Klebsiella spp. They may well be more common than this figure implies, but their complex characteristics have hindered a systematic search so far. In fact, the type I R-M systems are the most complex so far discovered. Three different subunits form a multifunctional enzyme complex that catalyzes both restriction and modification of DNA. In addition, the type I enzymes are restriction-dependent ATPases and show DNA topoisomerase activity (reviewed in references 11, 12, 13, 147, and 149).

A type I enzyme alone has little affinity for DNA. Interaction with the cofactor S-adenosylmethionine (AdoMet) is necessary to transform the protein into a form capable of binding DNA. Apart from acting as an allosteric effector, AdoMet is also the methyl donor and sole cofactor for the methylation reaction. The restriction endonuclease activity, on the other hand, absolutely requires Mg²⁺ and ATP in addition to AdoMet. Nonhydrolyzable analogs of ATP cannot substitute for ATP in the DNA cleavage reaction. Following, or possibly concomitantly with, DNA cleavage, the enzyme is irreversibly transformed into a potent ATPase. The endonucleolytic activity is lost after one DNA cleavage event, and cleavage is thus not catalytic in the sense that the enzyme does not turn over. By contrast, the ATPase activity is extremely long-lasting in vitro, but whether the extended ATP hydrolysis occurs in vivo as well and what the biological function would be are not known (14, 48, 146).

A type I monofunctional modification methylase devoid of restriction activity can also be isolated (43, 77, 124, 125, 131). It consists of only two of the three subunits of the multifunctional enzyme. Its physiological role is unclear, since the three-subunit enzyme is also an efficient modification methylase (15, 125).

The DNA sequences recognized by type I enzymes are asymmetric and split into two recognition components, one of 3 bp and the other of 4 to 5 bp, separated by a nonspecific spacer of 6 to 8 bp. Some of the recognition sequences are also degenerate at one or two positions. All known type I enzymes methylate adenosyl residues, one on each strand of the recognition sequence. Cleavage occurs randomly at considerable distances from the recognition sites, rarely less than 400 bp and often as far as 7,000 bp away. Three different models have been proposed to explain how these enzymes can act so far from their binding site (114, 122, 148).

The three subunits of the type I enzyme complex are coded for by three contiguous genes, hsdR, hsdM, and hsdS (host specificity for DNA). The hsdM and hsdS genes form an operon, whereas hsdR is expressed from its own promoter (36, 81, 100, 115, 123). The two subunits encoded by hsdM (M) and hsdS (S) are both necessary and sufficient for methylation, as shown by the fact that the hsdR gene product (R) is absent in the monofunctional modification methylase. The R subunit is required only for restriction, while the M subunit is necessary for modification and forms part of the multifunctional restriction enzyme. The S subunit is responsible for specific DNA recognition in both modification and restriction reactions. On the basis of genetic complementation, sequence homology, and antigenic cross-reactivity, the type I R-M systems were further divided into three families, A, B, and C (12). Within each family, the hsdR and hsdM genes are largely homologous and thus exchangeable. The hsdS genes share conserved regions that may be responsible for protein-protein interactions with the M and R subunits, but they also contain two extensive regions of nonhomology that are thought to code for sequence-specific DNA recognition domains (41, 60). Between families, complementation is not observed, and there is essentially no sequence homology apart from a few short amino acid sequence motifs common to DNA adenine methylases and to ATP-binding proteins (13, 15, 117, 141). The type IA and IB systems have the same genetic organization, with the hsdR gene preceding the hsdM-hsdS operon. Since the genes of the two families share the same chromosomal location, they could be considered functional alleles (28, 36, 60). Most type IC systems are plasmid encoded, and the order of the two transcriptional units is reversed relative to that in families IA and IB (80, 100, 119, 134).

Type I R-M systems can change their sequence specificities by purely natural processes, and this may confer a selective advantage on type I systems compared to other restriction systems. The first spontaneous alteration of sequence specificity observed in the laboratory was the result of a transduction experiment. By homologous recombination within the central conserved region of the hsdS genes of StySPI and StySBI, a new specificity was generated, called StySQI (20, 35, 37). The StySQI recognition site is a hybrid of those of the two parents: the 5' component of the split recognition site is identical to the 5' part of the StySPI site, and the 3' component is the same as the 3' half of the StySBI site (88, 89). Thus, the S polypeptides contain two independent protein domains, each one recognizing one half of the split recognition site. By reassortment of domains from different S proteins, new specificities can be created. This idea was tested by in vitro recombination of different hsdS genes within the central conserved region, and the constructed recombinants all showed the predicted specificity, a hybrid recognition site composed of different parental half-sites (38, 44). There is evidence that these kinds of reassortments also occur outside the laboratory: if the amino acid sequences of the variable regions of different hsdS genes are compared, there is essentially no homology between them, with the notable exception of those specificity domains that recognize identical half-sites. The type IA systems, EcoKI and StySPI, both recognize AAC as the 5' part of their recognition sequences, and their N-terminal variable regions show 90% amino acid sequence identity (37). Likewise, the type IB systems, EcoAI and EcoEI, recognize the same 5' half-site, GAG, and their N-terminal variable regions share 80% amino acid sequence identity. StySBI also recognizes GAG as its 5' half-site, and although it belongs to a different family, its amino variable region still shows considerable homology to that of EcoAI (about 50% identity) (27).

Recombination is also the most probable mechanism for a spontaneous switch back and forth between two alternative specificities, StyR124I and StyR124/3I, that can be observed at a low frequency in vivo (40). The recognition sites of the two systems differ only in the length of the nonspecific spacer (101). The hsdS genes are identical apart from a 12-bp in-frame sequence in the central conserved region that is directly repeated twice in the StyR124I gene and three times in the StyR124/3I gene (100). Unequal crossing over between these repeats could change their number and thus switch the specificity of the system. This example of specificity alteration strongly suggests that at least part of the central conserved region forms a spacer that positions the two DNA-binding domains at the right distance and angle, allowing specific interactions with the two components of the recognition site. The flexibility of this interdomain linker represents another avenue for the generation of new specificities (43, 44).

Another natural mechanism by which type I enzyme specificity can be altered has been discovered recently. Truncation of the hsdS genes of type IC systems by deletion or transposon insertion does not destroy enzyme activity but instead alters the specificity of the systems. The truncated hsdS genes still encode both the N-terminal DNA-binding domain and the conserved region that defines the length of the nonspecific spacer of the recognition site, but the C-terminal DNA-binding domain is missing. The mutant enzymes specifically interact with interrupted palindromic recognition sequences in which the 5' half-site of the wild-type recognition site is repeated in inverse orientation. It was proposed that the mutant enzymes use two molecules of the truncated hsdS subunit to recognize their binding sites (1, 86).

The modular composition of type I R-M systems seems to be a key element that allows them to rapidly evolve new DNA sequence specificities. Changes in just one module, the S subunit, will concomitantly alter the specificity of both modification and restriction activities. The S subunit itself contains several modular protein domains that contribute to its specificity. These domains are partially interchangeable and thus allow the development of new specificities by reassortment of protein domains.

Type II R-M Systems

Type II R-M systems comprise separate restriction and modification enzymes that act independently of each other and have simple cofactor requirements: restriction depends on the presence of Mg²⁺, and modification requires AdoMet. Type II restriction endonucleases are relatively easily to detect in crude cell extracts, and they cleave substrate DNA at fixed positions within (or, for a few enzymes, close to) their recognition sequences, producing DNA fragments of defined sizes. These useful properties have made them indispensable tools for molecular biology and have stimulated a systematic screen of many taxonomically diverse bacteria for restriction enzymes with new DNA sequence specificities. As a result, several thousand type II enzymes have been identified, including more than 150 from E. coli and Salmonella strains (data taken from ftp://vent.neb.com/pub/rebase/ type2.* [113]); however, apart from their DNA recognition properties, most have not been further characterized.

Currently, the DNA sequences of the structural genes of over 50 type II R-M systems are available (139). Surprisingly, no distinct amino acid sequence homologies between cognate pairs of restriction and modification enzymes can be detected, even though the enzymes recognize the same DNA sequence (21, 76). The lack of homology suggests that these enzymes must have evolved independently. The same is probably true for the different endonucleases, since, with the exception of a few pairs of isoschizomers, they show no sequence homologies and thus cannot easily be traced back to one or a few common ancestors (9, 61, 121, 141). In contrast, the methyltransferases show marked amino acid sequence similarities, mainly related to the enzymatic reaction mechanism. They can be grouped into three major classes that differ in the nature of the modification introduced: m5C, m4C, or m6A (21, 68, 76, 140). The m5C methyltransferases form a relatively homogeneous class with a common building plan. The m6A methyltransferases also share some consensus motifs, but in general, the homologies are less extensive than for the m5C methylases. The m4C methyltransferases are more closely related to m6A methylases than to m5C methylases. This may reflect the fact that both m4C and m6A methyltransferases methylate an exocyclic amino group, while m5C methylases interact with a ring carbon.

The m5C methyltransferases share 10 conserved sequence motifs that occur in invariant order, alternating with variable sequence elements (for a recent review, see reference 74). The largest variable region is thought to encode the DNA sequence specificity. Evidence for this comes from the analysis of the multispecific Bacillus phage m5C methyltransferases (reviewed in reference 92). Each of these enzymes methylates several unrelated sequences, and the existence of mutations that affect the methylation of only one sequence helped define distinct specificity domains (75, 137). Hybrid specificities could be created by interchanging specificity domains among the different methylases (8, 133). The picture that emerged for the multispecific methylases is that each DNA sequence is recognized by a different domain in a set of sequentially arranged domains that form the largest variable region. Hybrids of monospecific m5C methyltransferases indicate that specificity determinants for both the DNA sequence to be recognized and the specific residue to be methylated are located within the largest variable region (67, 87).

Most type II recognition sequences are essentially symmetric, comprising four to eight specific base pairs that may be continuous or interrupted. Cleavage as well as methylation occurs symmetrically within the recognition sites. The endonucleases generally act as homodimers, with each subunit interacting with one half of the palindromic recognition sequence to facilitate the coordinate cleavage of both strands of the duplex. The crystal structures of several type II restriction enzymes, including EcoRI and EcoRV from E. coli, have been determined (66, 142). Type II methyltransferases are normally monomeric enzymes that methylate unmodified DNA one strand at a time. This fits in well with the fact that their customary substrate is hemimethylated DNA (141).

A subgroup of type II R-M systems recognizes asymmetric DNA sequences 4 to 7 bp in length. The endonucleases of this group cleave the DNA at a precise distance (1 to 20 bp) outside their recognition sequences, thus giving the subclass the name type IIS (shifted cleavage [128]). Their grouping as type II enzymes is based on their cofactor requirements and the fact that the endonuclease and methyltransferase act independently. In other features, however, they represent a different type of system altogether.

The type IIS endonucleases appear to act as monomers. The spatial separation of the recognition and cleavage sites suggests a protein structure in which there are separate domains for specific binding and cleavage. This modular domain model is backed up by studies on proteolytic fragments and deletion mutants of the FokI endonuclease, which defined an N-terminal DNA-binding domain and a C-terminal domain with nonspecific DNA cleavage activity (78, 79). Recently, a chimeric restriction endonuclease was constructed by linking the Drosophila ultrabithorax (Ubx) homeodomain to the cleavage domain of the FokI endonuclease (65). The hybrid enzyme shows the same DNA sequence binding preference as Ubx and cleaves the DNA away from the recognition site.

Several systems originally assigned to the type IIS group are very irregular and might be better classified as new types. The Eco57I system comprises a joint endonuclease-methyltransferase and a separate methyltransferase, and DNA cleavage is stimulated by AdoMet. It was proposed that Eco57I be classified as a type IV R-M system (58). Preliminary results indicate that GsuI is another member of this new type of enzymes (59, 93). Two additional enzymes which may also constitute a new class are the BcgI and Bsp24I endonucleases. They are unusual in that cleavage occurs outside the recognition sequence on both sides, excising a 34-bp and a 30-bp DNA fragment, respectively, containing the recognition site (33, 69, 70).

Type III R-M Systems

As with type I R-M systems, no systematic search for type III enzymes has been undertaken, and only four members of this class are known to date: EcoP1, EcoP15I, StyLTI, and HinfIII. EcoP1 and EcoP15I are encoded, respectively, by the P1 prophage and the closely related p15B plasmid found in E. coli 15T^– (4, 5). StyLTI is encoded on the chromosomes of most Salmonella strains (19, 23). HinfIII is produced by Haemophilus influenzae Rf (with an isoschizomer in strain Re) (96). Recently, an open reading frame (ORF) that showed amino acid sequence homology to part of the EcoP1 and StyLTI systems was detected on the Bacillus cereus chromosome; however, it is not known whether a functional protein is expressed (50).

With type III restriction enzymes, a single hetero-oligomeric protein catalyzes both the restriction and the modification reactions. Modification requires the cofactor AdoMet and is stimulated by Mg²⁺ and ATP. Restriction absolutely depends on Mg²⁺ and ATP and is stimulated by AdoMet. In the presence of all three cofactors, DNA cleavage and methylation are competing reactions (45, 62, 108). The recognition sequences of type III enzymes are asymmetric, uninterrupted, and 5 to 6 bp in length. Cleavage occurs on the 3' side of the recognition sequence at a distance of approximately 25 to 27 bp.

These bifunctional enzymes are composed of two subunits, the products of the mod and res genes (32, 56). The Res subunit has no enzymatic activity when not complexed with Mod, whereas a monofunctional modification methylase composed of the Mod protein only can be isolated (30, 46, 53). Thus, the Res subunit is required only for restriction, while Mod confers DNA specificity for both modification and restriction as well as catalyzing modification. The Mod protein can therefore be considered a functional analog of the complex formed by the S and M subunits of type I R-M systems.

The two systems EcoP1 and EcoP15I are closely related, with complementation as well as recombination between the structural genes being possible (5). Electron microscopic analysis of DNA heteroduplexes between the resP1 and resP15I genes revealed that they are largely homologous, with only a short region of partial homology toward the beginnings of the genes (56). The modP1 and modP15I genes have been sequenced, and a comparison of the deduced amino acid sequences reveals a structure composed of conserved and nonconserved regions (55). The N- and C-terminal thirds of the genes are highly homologous, while the central region shows few similarities. Replacing a conserved region from one gene with the equivalent region from the other does not change the specificity of the system, suggesting that sequence-specific DNA recognition is encoded by the central variable regions (H. E. Eberle, N. Redaschi, and T. A. Bickle, unpublished data). Point mutations localized in this variable region indicate that it is also at least partially responsible for the binding of the methyl donor AdoMet (107). Two conserved sequence motifs are common to all m6A and m4C methylases and are therefore presumed to be important for the catalytic activities of these enzymes (68, 140). In the EcoP1 and EcoP15I genes, one of these two motifs is contained within the N-terminal conserved region, and the other is found at the transition between the variable and the C-terminal conserved regions. Thus, the conserved N and C termini of the two proteins are likely to be involved in AdoMet binding and the transmethylation reaction. In addition, they probably mediate subunit interactions, analogously to the function proposed for the conserved regions of the type I hsdS gene products (26, 41, 43, 86).

Recently, the StyLTI structural genes have been sequenced (29), and the deduced amino acid sequences were compared with those of the EcoP1 system. The StyLTI and EcoP1 Mod proteins are 36% identical, and the two Res proteins show 32% overall identity. Thus, the two systems are clearly related. For the Mod proteins, the homologies are somewhat stronger in the N- and C-terminal regions than in the central portions of the proteins. With the two Res proteins, a region of about 50 amino acids in the middle of the proteins is almost totally conserved, suggesting an important function in either subunit interactions or the catalysis of DNA cleavage.

In all three systems, EcoP1, EcoP15, and StyLTI, the mod and res genes are contiguous, with the mod gene preceding the res gene, and both genes are transcribed in the same direction (29, 55). However, the expression of the genes must be controlled by different mechanisms, as the EcoP1 and EcoP15 genes are freely transferable from cell to cell by phage infection, conjugation, and transformation, whereas horizontal transfer of the StyLTI system results in death of the recipient cell caused by extensive DNA breakdown (3, 4, 6, 31).

Type III modification enzymes belong to the class of m6A methyltransferases. Surprisingly, only one strand of the asymmetric type III recognition sites is methylated (7, 30, 47, 85, 94, 95). Most R-M systems methylate residues in both strands of their recognition site, so that after DNA replication, each daughter molecule inherits one of the parental methyl groups per site. Such hemimethylated recognition sequences are resistant to restriction, and they are substrates for the modification enzyme which subsequently methylates the unmodified strand. Replication of an asymmetrically modified site will yield a completely unmodified recognition sequence in one of the two daughter molecules. This unmodified site ought to be subject to restriction, which would be lethal for the cell (7, 47). A solution to this problem followed the observation that phage T3 is restricted by EcoP15I, while its close relative T7 is not. A correlation was made to the constellation of EcoP15I sites in the two phage genomes. The T7 chromosome contains 36 sites for EcoP15I, but strikingly, they all have the same orientation along the DNA, whereas the T3 DNA has sites arranged in both orientations (73, 116). This led to the hypothesis that type III endonucleases require at least two recognition sites in inverse orientation. This model was confirmed by in vitro experiments using artificial M13 constructs with different combinations of EcoP15I sites in both direct and inverse orientations. Any site could be modified, but restriction occurred only when the sites were in inverted orientation (84). One can thus consider type III recognition sequences to be interrupted symmetrical sites with a nonspecific spacer of variable length at the center of symmetry. DNA replication will always leave the nonmodified sites in the daughter chromosomes in direct orientation.

REGULATION OF EXPRESSION OF R-M SYSTEMS

It is generally assumed that the expression of modification and restriction activities must be regulated to avoid autorestriction and cell death. Restriction activity should be expressed only when the cellular genome is modified. The need for this type of regulation is most obvious during establishment of an R-M system in a new host cell whose chromosome is completely unmodified when the R-M genes enter the cell. Modulation of restriction activity can also be important during changes in the physiology of the cell that may lead to a transient undermodification of the genome (e.g., starvation associated with a deficiency of methyl donors). Many R-M systems can be transferred to new host cells by conjugation, transfection, or transformation and become established without apparent difficulties. However, the regulatory mechanisms allowing establishment in naive cells are unclear. Either the recipient cells must be able to repair DNA damage caused by uncontrolled expression of restriction activity, or the expression of endonuclease activity must be delayed until complete protection of the host DNA is achieved by the methylase.

Regulation of Type I R-M Systems

Both the chromosomally coded systems, type IA and IB, and the plasmid-borne systems, type IC, can be readily transferred from one strain to another by conjugation, transfection, and transformation (17, 20, 22, 24, 25, 36, 115, 119, 123, 143). The first evidence that the expression of restriction activity is regulated upon gene transfer came from studies of the type IB systems, EcoAI and EcoEI. An F' plasmid carrying the EcoAI genes could be transferred to naive recipient cells without difficulties, whereas conjugation into a recipient host expressing the hsdR gene was lethal, presumably because of restriction of the host genome (123). Likewise, a λ phage encoding the hsdS gene of either EcoAI or EcoEI could not lysogenize bacteria that carried a plasmid expressing the hsdR and hsdM genes of EcoAI. The few surviving phage were found to have mutated to hsdS (36). These results imply that on transfer of an entire hsd region, the modification methylase is functional before the endonuclease is. The genetic organization of the hsd genes potentially allows differential expression of the two opposing activities. The hsdR gene is transcribed from its own promoter, Pres, whereas hsdM and hsdS form an operon that is controlled by a separate promoter, Pmod. If transcription from the Pmod promoter occurs first, a monofunctional modification enzyme devoid of endonuclease activity can assemble and modify the recipient cell’s genome, thereby conferring protection against subsequent expression of the hsdR gene and the establishment of restriction activity (36, 81, 115, 123). Although this is a feasible model, there has so far been no direct demonstration of control at the transcriptional level.

Recently, the establishment of the type IA system EcoKI in naive cells was investigated (98, 99). An F' plasmid carrying the EcoKI genes was transferred into non-K-modified recipient cells, and the phenotypic expression of modification and restriction activities was monitored. Modification activity could be detected immediately following the conjugal transfer, whereas restriction activity was observed only after about 15 generations (99). This sequential expression explains the efficient establishment of the EcoKI system in a nonmodified host and suggests regulation of restriction activity after gene transfer. To test whether the sequential expression of EcoKI modification and restriction activities was achieved by transcriptional control of the hsdR gene, conjugation studies were performed with hsd-lacZ operon fusions on F' plasmids. Those experiments indicate that Pmod and Pres are expressed simultaneously following conjugative transfer (98). Thus, the EcoKI hsd genes do not appear to be regulated at the transcriptional level. Insight into the regulatory mechanisms associated with the establishment of the EcoKI system in naive cells may be gained from the characterization of a serendipiditously isolated spontaneous mutant of E. coli C that is killed upon conjugative transfer of the EcoKI genes. It was proposed that the affected function is normally involved in delaying expression of the restriction activity after gene transfer (98).

Regulation of Type II R-M Systems

Type II R-M systems have been identified in a wide variety of taxonomically diverse bacteria. The lack of familiarity with the majority of these organisms renders the analysis of many type II systems difficult. Genetic characterization depends to a large degree on transfer of genes from the bacteria in which they occur to E. coli. In this heterologous host, the expression of restriction and modification activities may be disturbed, because regulation need not be the same as in the original strain. Many type II systems become established in E. coli without difficulty (82, 138). However, for two systems, DdeI and BamHI, the genes had to be introduced sequentially: the M gene was introduced first, to allow modification of the new host chromosome, and then the R gene was introduced (18, 54, 127). Some systems have resisted all attempts to clone them, and others have yielded only the methylase gene. To date, approximately 100 complete type II systems have been cloned (139). Among those analyzed, all four possible arrangements of M and R genes have been found. When M and R genes have opposite orientations, each gene is transcribed by its own promoter, providing the opportunity for independent transcriptional regulation. However, the genes for most type II systems have the same orientation, usually adjoining so closely that they are unlikely to have separate promoters.

In theory, type II systems could be regulated by passive mechanisms. For example, most endonucleases are homodimers, so that subunits must accumulate before active enzyme is formed, while most methylases are effective as monomers. This may result in a lag between the appearance of modification and restriction activities (42). Also, there is evidence that E. coli can repair a certain amount of endonucleolytic DNA damage. For the systems PaeR7I, TaqI, AvaII, HaeII, HinfI, PstI, and XbaI, the endonuclease gene could be cloned without the accompanying methylase gene (39, 82, 120, 132, 138). However, unmodified cells containing one of these R genes differ strikingly from modified cells. The bacterial colonies have a translucent, flat appearance, and with a high frequency, they lose the plasmid with the R gene or mutate to R^–. Also, they usually synthesize active endonuclease at a reduced level compared to that in modified cells. These observations suggest that E. coli may cope transiently with a low amount of DNA cleavage but that normal expression of endonuclease activity requires protection of the chromosome by modification.

In some type II systems, the regulation of modification and restriction activities seems to be mediated by the product of an additional ORF tightly linked to the R-M gene cluster (130). Of the type II R-M systems that have been sequenced, nine have an additional ORF associated with their R and M genes (139). Five of these ORFs show considerable sequence homology with each other (130). Mutations in one C gene can be complemented in trans by a wild-type copy of any of the homologous C genes (J. E. Brooks and C. I. Ives, FASEB Summer Research Conference, abstr., p. 15, 1993). Since there is no homology between the corresponding endonucleases and only limited homology among the methyltransferases, it seems likely that the C genes form a family that has evolved independently of the R and M genes (130). The deduced amino acid sequences of the C proteins show homology to DNA-binding proteins of the helix-turn-helix class (49), indicating that they may function as transcriptional repressors or activators (57, 129, 130). It was suggested that the C genes could mediate temporal regulation of gene expression to facilitate the establishment of the R-M system in a new host. According to the proposed model, transcription of the M gene takes place immediately after the system has entered the cell, whereas the expression of the R gene will be delayed until enough C protein has accumulated to stimulate R gene transcription. There is preliminary evidence that additional control mechanisms may be operating posttranscriptionally (57, 130).

Regulation of Type III R-M Systems

The type III system, EcoP1, is carried on the temperate bacteriophage P1 and is thus frequently transferred to new host cells. It was demonstrated in 1962 that establishment of the EcoP1 system is achieved by a temporal delay in the expression of restriction activity (4). Modification activity was detected a few minutes after infection of E. coli with phage P1, whereas restriction activity attained its normal level only some hours later (3, 6). The regulatory mechanism underlying this sequential expression of modification activity followed by restriction activity has remained unclear.

The EcoP1 mod and res genes abut and are transcribed in the same direction (55). This genetic organization suggests that the two genes form an operon, with mod being the promoter-proximal gene. Transposon insertions into the mod gene are completely polar for the expression of res, supporting the operon model (56). However, there is some evidence that the res gene can be transcribed independently of the mod gene. The results of an R-loop analysis of the EcoP1 genes indicate that each of the genes is transcribed in vitro from its own promoter and, in addition, that res transcripts may be produced by readthrough from the mod promoter (56). Further transcriptional analysis of the modP1 and resP1 genes led to the conclusion that two separate promoters control the transcription of the two genes in vivo (118). On the basis of comparative expression studies, a simple control mechanism for expressing the mod gene before the res gene that was based solely on differential promoter strength was considered very unlikely. Still, two transcriptional units would allow independent regulation by trans-acting factors.

Surprisingly, the chromosomally encoded StyLTI system cannot be transferred horizontally by conjugation, transfection, or transformation to a recipient cell lacking the corresponding modification. Such transfer proved to be lethal because of recipient DNA degradation, indicating uncontrolled expression of restriction activity (31). It was necessary to clone the StyLTI genes in two steps. First, the methylase gene was subcloned, and then the complete system could be introduced into methylase-producing cells on a compatible vector (32). There is evidence that the StyLTI mod and res genes can be transcribed as distinct units. Tn5 transposition into the mod gene showed no polar effects on expression of the downstream res gene (29).

The StyLTI Mod and Res proteins show a good deal of homology with the EcoP1 and EcoP15 proteins (29). The gene organization is also alike in all three systems (29, 55). Despite all these similarities, regulation of the expression of the enzymatic activities differs profoundly. In the EcoP1 and EcoP15 systems, the expression of restriction activity is tightly regulated, allowing efficient establishment of the systems in naive cells, whereas transfer of the StyLTI system is lethal for the recipient cell because of uncontrolled expression of restriction activity (3, 4, 6, 31).

ACKNOWLEDGMENTS

We thank Maria MacWilliams for helpful suggestions during the preparation of the text. Work from this laboratory has been supported by the Swiss National Science Foundation.