ABSTRACT
The S subunits of type I DNA restriction/modification enzymes are responsible for recognising the DNA target sequence for the enzyme. They contain two domains of approximately 150 amino acids, each of which is responsible for recognising one half of the bipartite asymmetric target. In the absence of any known tertiary structure for type I enzymes or recognisable DNA recognition motifs in the highly variable amino acid sequences of the S subunits, it has previously not been possible to predict which amino acids are responsible for sequence recognition. Using a combination of sequence alignment and secondary structure prediction methods to analyse the sequences of S subunits, we predict that all of the 51 known target recognition domains (TRDs) have the same tertiary structure. Furthermore, this structure is similar to the structure of the TRD of the C5-cytosine methyltransferase, HhaI, which recognises its DNA target via interactions with two short polypeptide loops and a [beta] strand. Our results predict the location of these sequence recognition structures within the TRDs of all type I S subunits.
A major aim of many studies of sequence specific protein-DNA interactions has been to determine how certain sequences of amino acids can recognise, with great fidelity, a DNA target sequence. Structural analysis of protein-DNA complexes has shown how [alpha] helices, [beta] strands and loops can be used to give sequence specificity (1 -3 ).
DNA methyltransferases (mtases) of restriction/modification (R/M) systems use target recognition domains (TRD) of 50-150 amino acids to recognise their DNA target sequence (4 ). The TRD is the major determinant in DNA target specificity with separate catalytic domains being required for enzymatic activity. The crystal structures of two monomeric type II C5-cytosine mtases, HhaI and HaeIII, bound to their DNA targets show that the TRD uses a conserved structure comprising two loops and one [beta] strand to accomplish sequence recognition (5 ,6 ). The amino acid sequences of TRDs of many different C5-cytosine DNA mtases have been compared. The level of sequence identity in these comparisons is very low and confined to several very short amino acid sequences corresponding to the recognition region in the two crystal structures (7 ,8 ), however, experimental support has been obtained for the involvement of this region in DNA recognition by C5-cytosine mtases other than HhaI and HaeIII (8 ). The N6-adenine and C4-cytosine mtases also contain a TRD and a catalytic domain (9 ), however, no cocrystal structure of one of these enzymes with DNA has been solved. A model of DNA recognition by the TaqI N6-adenine mtase, whose structure is known in the absence of DNA, has been constructed (10 ,11 ).
All characterised type I R/M systems recognise N6-adenine methylation of a bipartite target sequence (12 -14 ). They are large, oligomeric, multifunctional enzymes encoded by the hsdR, M and S genes, combining both restriction endonuclease (R) and modification mtase (M) subunits with a DNA specificity (S) subunit. Type I R/M systems of enteric bacteria have been grouped into four families based on subunit complementation, DNA hybridisation and antibody cross-reactivity experiments (14 ,15 ). The amino acid sequence identity is very high within a family for the R, M and parts of the S subunit outwith the TRDs (16 -21 ). This is believed to reflect conservation of residues in the subunit interfaces and the nuclease and mtase catalytic sites.
The S subunits of type I R/M systems contain two TRDs of 150-180 amino acids (12 -14 ). Each TRD is responsible for recognising one of the two parts of the bipartite DNA target. The amount of amino acid sequence conservation between TRDs is either below ~25% for TRDs recognising different targets, or 40-90% when a target is shared dependent on whether the S subunits are in a different or the same family. The remainder of the ~50 kDa S subunit contains amino acid sequences which show a high degree of conservation between type I systems. These regions are responsible for defining the length of the non-specific DNA spacer in between the two TRD target sequences (22 ) and for binding the M and R subunits (23 -26 ). It is believed that each TRD fits into the major groove to recognise the DNA, and the M subunits are arranged on either side of the S subunit allowing them to encircle the DNA and gain access to the methylation targets (27 ,28 ). Methylation is predicted to occur via a base flipping mechanism in which the adenine base is displaced out of the DNA helix into the catalytic pocket of the M subunits. This implies that the M subunit is on the other side of the DNA helix from the S subunit.
Table 1
The TRDs of type I S subunits recognise a wide variety of 3, 4 or 5 bp targets and it would be of interest to define which amino acids within the large and highly variable sequence of the TRDs are responsible for sequence specificity. In this paper we use amino acid sequence alignment combined with secondary structure prediction methods. The use of secondary structure predictions enhances the strength of the amino acid alignment making distant similarities more apparent. These alignments of the TRDs suggest that all have the same tertiary structure and that they are the products of divergent evolution. A comparison of the secondary structure predictions with the known structure of the TRD of the HhaI mtase shows a strong similarity which has allowed us to define potential DNA recognition loops for all of the type I TRDs and to model the tertiary structure of these domains in a manner amenable to experimental verification.
Most nucleotide or amino acid sequences of the S subunits were obtained from published references or GenBank and a database constructed which separated the sequences into TRDs and conserved spacer regions. The amino acid sequences for the S subunits of BsuCI and KpnAI were generously provided by Prof. T. Trautner and Dr G. Xu (Berlin) and Dr J. Ryu (Loma Linda). The locations of the TRDs in the S subunit sequence and, if known, their DNA target and type I family are given in Table 1 . The amino terminal TRD and carboxyl terminal TRD are indicated by the suffix -1 or -2 appended to the type I system's name in Figure 1 .
A database of TRDs was made by separating conserved and unconserved regions of the S subunits and discarding the conserved regions. This database was inverted (every member was compared with every other one) using sss_align (47 , S. Sturrock and A. Coulson, manuscript in preparation), a new implementation of the Smith and Waterman algorithm (48 ) using the Dayhoff PAM scoring scheme (49 ). The Smith and Waterman algorithm is a mathematically rigorous and exhaustive method to optimally align a pair of sequences. Each TRD sequence, along with its closest homologues, was then sent to the PHD program (50 ,51 ) and a secondary structure prediction acquired. PHD is a secondary structure prediction method which uses a neural network trained with a set of known tertiary structures combined with multiple sequence alignment. On average secondary structure is predicted with >70% confidence. Each prediction was then placed in a new database and again inverted using sss_align but this time including the secondary structure prediction as well as amino acid sequence. sss_align takes into account the reliability index assigned by PHD on a residue by residue basis. This database inversion was performed with the QVAL parameter set at 80, PAMS 150, GAPOPEN 20 and GAPEXTEND 4. The QVAL parameter sets the amount of weighting given to the secondary structure when performing the sequence alignment. A value of 0 means the alignment uses only secondary structure information while a value of 100, as used in the first database inversion above, indicates that only sequence information is used in the alignment. A value of 80 was found to give the optimal statistical significance to the alignments of type I TRDs. In this instance, sss_align performs better than normal sequence alignment methods in aligning two distantly related sequences because the addition of secondary structure information, whether derived from a real structure or a prediction as in this case, is used to help the alignment pass through regions of very low sequence identity. The output of sss_align was again used to cluster sequences of high similarity and overlap these clusters with others until nearly all the TRD sequences were successfully aligned. Some TRD sequences could not be inserted into this alignment by the program due to a lack of obvious homology and these were aligned manually. These sequences are indicated in Figure 1 by an asterisk after the TRD name. In addition, sss_align also aligned the known tertiary structure of HhaI mtase (5 ) with the EcoKI-1 TRD to provide a key for the prediction of the location of loops and strands involved in DNA recognition. In this case, settings of PAMS 250, GAPOPEN 8, GAPEXTEND 1 and QVAL 60 were used to obtain the optimal alignment. These values were used successfully in the CASP2 competition (see below). To obtain a measure of the statistical significance of our alignment of type I TRDs with HhaI, we enlarged our database of TRDs by merging it with the approximately 2000 unique sequences with known tertiary structures in the protein databank. This enlarged database was then searched using sss_align and the TRD of HhaI to find homologous structures.
sss_align can be accessed at http://www.icmb.ed.ac.uk/sss_align/. Using secondary structure information from known structures, sss_align has been shown to successfully align sequences with only 15% amino acid identity (personal communication A. Coulson; CASP2, Second meeting on the critical assessment of techniques for protein structure prediction on World Wide Web URL: http://iris4.carb.nist.gov/casp2/). sss_align also adjusts for the variation in the reliability of the secondary structure predictions by using the residue by residue reliability of PHD predictions. This allows the program to align sequences even if parts have incorrectly predicted secondary structure.
Normal sequence alignment methods have been applied to complete S subunit sequences in the past (16 ,52 ). These studies were hampered not only by the limited number of sequences available but also by the high degree of sequence similarity in the conserved regions of the subunits. These restricted areas of high homology almost totally obscured any sequence similarity between TRDs except when the TRDs recognised identical DNA targets whereupon the similarity was so high as to preclude any prediction of amino acids involved in sequence recognition.
Combining the methods of multiple sequence alignment and secondary structure prediction within the sss_align program has facilitated the alignment of all 51 known or putative type I TRDs, overcoming the difficulties imposed by the large size of the TRDs and their very limited sequence conservation. The alignment bears a significant similarity to a short section responsible for DNA sequence specificity in the HhaI mtase. We suggest that this implies that all TRDs of type I S subunits are the products of divergent evolution with a conserved tertiary structure and that a small part of this structure, by analogy with HhaI mtase, is involved in DNA sequence recognition.
A variety of experiments such as UV-induced crosslinking to DNA (57 ), chemical modification of lysines (58 ) and random mutagenesis of TRDs (personal communication, M. O'Neill and N. E. Murray) have been applied to the best characterised type I R/M systems, EcoKI and EcoR124I, to identify amino acids involved in sequence recognition. These experiments provide preliminary support for our identification of a DNA binding region.
Chemical modification of EcoR124I showed that several lysines in the second TRD were susceptible to modification especially in the absence of bound DNA (58 ). Lysines 261, 297 and 327 within the TRD were particularly strongly modified. Lys297 is the most strongly modified residue and lies within the second proposed recognition loop. These three lysine residues are also conserved in the first TRD of StySKI which recognises the same DNA target as the second TRD of EcoR124I therefore supporting a role for them in sequence recognition (39 ). The other less strongly modified lysines in the second TRD may be required for non-specific DNA binding as they are not conserved in StySKI and lie outside of our predicted recognition region.
Random mutagenesis of the first TRD of EcoKI has so far changed 62 out of 150 amino acids (personal communication, M. O'Neill and N. E. Murray). Most of the mutations are silent, but five of seven mutations that impair restriction and modification are within the two putative recognition loops. The other two mutations occur shortly after the position of [beta] 3 in Figure 1 .
UV-crosslinking demonstrated that Tyr27 in the first TRD of EcoKI was in contact in the major groove with the 3' thymine base in the sequence complementary to the 5' AAC part of the EcoKI target (57 ). This residue is outside of our predicted recognition loops, however, it has been found that changing it to other amino acids has a minor effect on DNA specificity suggesting that it may be involved in a non-sequence specific interaction with the DNA (personal communication, M. O'Neill and N. E. Murray).
Genes similar to the hsd genes of enteric bacteria have now been found in non-enteric bacteria and archaebacteria (see references in Table 1 ) indicating that type I R/M systems are widespread in nature. It has been suggested that diversity within genes such as those forming type I R/M systems would be advantageous to a bacterial population (59 ,60 ). Furthermore, the diversity in hsd gene sequences observed in enteric bacteria provides support for horizontal gene transfer and a very ancient origin for the hsd genes (19 ,21 ). The presence of type I R/M systems on conjugative plasmids would assist the spread of hsd genes by horizontal transfer (61 ). The existence of a common tertiary structure for TRDs, as implied by Figure 1 , would support this model for the distribution of type I systems in nature. Gene duplication of TRDs and transfer of TRDs by recombination is evident, not only from genetic and sequencing experiments (16 ,18 ,44 ,52 ,62 ), but also from biochemical results on the domain structure of the S subunit (23 -26 ). Recombination was responsible for the generation of two new type I target specificities, StySQI and EcoR124/3I (22 ,63 ,64 ), and evidence for recombination of a short stretch of the hsdS gene between E.coli B and S.enterica serovar Potsdam has been found (19 ). It is possible that other recombination events could encompass the short region within the TRD which we have predicted to be involved in DNA recognition, thereby allowing the generation of new specificities. These experiments suggest that the type I S subunit is a fusion of two half S subunits each containing one TRD to give a 2-fold rotationally symmetric arrangement of the TRDs and a bipartite DNA target (24 ,25 ,27 ,28 ,65 ). Horizontal gene transfer has also been proposed for the type II R/M systems (66 ). The range of organisms in which type I systems have been found or postulated, and their diversity within species such as E.coli and S.enterica, could also suggest that a large pool of TRDs existed before the evolution of different bacterial species. Therefore, it may be possible to have similar TRDs in different species even without invoking horizontal transfer, if they both carried with them the same range of TRDs when the species diverged (67 ).
If our alignments are realistic, then the similarity between TRDs of type I N6-adenine mtases and the TRDs of C5-cytosine mtases may extend further to many, if not all, TRDs of type II N6-adenine mtases, type III mtases and other mtases which do not fit current classifications. This would support the proposal (68 ) that all mtases have evolved from a common ancestor consisting of a small monomeric TRD, such as that still found in AquI mtase (69 ) and EcoHK31I mtase (70 ), associated with a separate catalytic subunit. It has been proposed that the mtase catalytic subunit may have developed from early DNA repair enzymes which use the same base flipping method to gain access to their target base as the mtases (71 ). The normal rate of mutation and gene duplication events coupled with the selection pressure within a bacterial population to expand the range of DNA target sequences, has virtually obscured this common origin. A conserved tertiary structure within TRDs implies that it may eventually be feasible to derive the amino acid recognition code used by TRDs to recognise DNA sequences as is currently being revealed for zinc finger-DNA recognition (3 ).
Pasteurella haemolytica also appears to contain a type I system belonging to the ID family. See S. K. Highlander and O. Garza (1996), Gene, 178, 89-96 for the gene sequences and alignment of the S subunit with that of HI1286. The TRDs of this system fit into the alignment scheme shown in Figure 1 .
We wish to thank Professor Noreen Murray, Dr Andrew Coulson and our colleagues in their laboratories, particularly Dr Mary O'Neill, for provision of unpublished data and many useful discussions. We also thank Professor Thomas Trautner and Dr Guoliang Xu (Berlin), and Dr Junichi Ryu (Loma Linda) for the provision of unpublished sequences and other information. This work would not have been possible without the support of The Royal Society and The Darwin Trust. David Dryden thanks the Royal Society for a University Research Fellowship and Shane Sturrock thanks the Biochemical and Biological Sciences Research Council for a studentship.
*To whom correspondence should be addressed. Tel: +44 131 650 7053; Fax: +44 131 650 8650; Email: david.dryden@ed.ac.uk
Family
Namea
DNA target if known
S subunit
length
1st TRD
approximate
location2nd TRD
approximate
locationReferenceb
IA
EcoKI
AAC N6 GTGC
464
11-157
214-368
29
IA
EcoBI
TGA N8 TGCT
474
11-158
215-379
30,31,32
IA
EcoDI
TTA N7 GTCY
444
11-128
185-348
33
IA
StyLTIII
GAG N6 RTAYG
469
11-153
209-375
34
IA
StySPI
AAC N6 GTRC
463
11-157
214-367
34
IA
EcoR5I
>140
1-140
35,36
IA
EcoR10I
>131
1-131
35,36
IA
EcoR13I
>152
1-152
35,36
IB
EcoAI
GAG N7 GTCA
589
110-247
403-540
37,38
IB
EcoEI
GAG N7 ATGC
594
109-247
403-545
17
IB
CfrAI
GCA N8 GTGG
578
108-236
385-529
18
IB
StySKI
CGAT N7 GTTA
587
108-249
396-538
39
IB
StySTI
>146
1-146
36
IB
EcoR17I
ATR....
>138
1-138
35,36
IC
EcoR124I
GAA N6 RTCG
409
25-142
215-350
40
IC
EcoDXXI
TCA N7 RTTC
406
23-139
211-341
20
IC
EcoprrI
CCA N7 RTGC
405
22-159
232-360
41
ID
StySBLI
CGA N6 TACC
434
1-153
229-405
15
ID
EcoR9I
464
1-188
264-435
35, N. Murray pers. comm.
ID
KpnAI
439
1-155
231-410
42, J. Ryu pers. comm.
IC?
BsuCI
GAY N7 TGGA
405
23-162
219-355
43, G. Xu and T. Trautner, pers. comm.
IC?
MpuAI
401
1-139
221-359
44
IC?
MpuBI
336
1-139
188-324
44
ID?
HI0216
385
20-138
198-333
45
ID?
HI1286
459
1-176
268-445
15,45
?
mj0130
425
23-161
231-371
46
?
mj1218
425
28-155
226-368
46
?
mj1531
425
28-170
241-371
46
REFERENCES