| Nucleic Acids Research | Pages |
Sticky Egyptians: a technique for assembling genes encoding constrained peptides of variable length
Introduction
Materials And Methods
Results
Discussion
Acknowledgements
References
Sticky Egyptians: a technique for assembling genes encoding constrained peptides of variable length
ABSTRACT Naturally occurring peptides, such as those produced by the poisonous marine snails of the genus Conus, have the ability to form tight, highly specific molecular interactions. The rigidity of the peptide framework which promotes these interactions is usually maintained by disulphide bonds, and it seems that the overall main chain conformation (or fold) of the peptide is determined by its length and the sequence distribution of the pairs of cysteine residues participating in these bonds. The fold of the peptide in turn is largely responsible for its shape. Since highly effective molecular interactions occur between species complementary in shape, we reasoned that peptides with the greatest potential in therapy or diagnosis would be found in a library of shapes, those peptides with a shape complementary to a given target being identified, for example, by selection. As a first step towards constructing such a peptide shape library, we have developed a method for assembling DNA fragments which encode an even number of cysteine residues and which are of variable length. We describe this method here.
INTRODUCTION
We are interested in exploiting the natural chemical diversity of peptides for medical applications. With this interest in mind, one important consideration is how strongly the peptide interacts with its target. Tight binding requires the peptide to be complementary to the target and to have a rigid architecture. Both factors influence the free energy change accompanying complex formation: complementarity encourages favourable enthalpic contributions (e.g. from van der Waals interactions); rigidity minimizes adverse contributions from conformational entropy losses. Therefore for clinical situations in which tight binding is critical, it would be advisable to begin with a peptide which is complementary to the desired target, and one which has a fixed shape.
Theshape of a peptide is broadly dependent on its main chain conformation. This may be constrained and, therefore the shape of the peptide fixed, by one or more disulphide bonds. Structural work on the conotoxins (1), neurotoxic peptides from cone snail venom, and inspection of their amino acid sequences (2), suggest that the particular conformation, or fold, adopted depends upon the length of the sequence and the distribution of the disulphide bonded cysteine residues within that sequence. Given this dependence, which is not unlike that seen in the antigen-binding loops of antibodies (3), we thought that it might be possible to generate a large variety of peptide shapes by altering the position of a small, even number of cysteine residues within short amino acid sequences of varying length. This `shape library' might then be used to select for peptides complementary to a given target. Once selected, a peptide could be subjected to mutagenesis to improve the strength of its interaction with the target. Such a strategy, again invoking antibodies, would be very similar to the affinity maturation process of the immune system (4).
This paper describes a scheme for assembling DNA fragments encoding a peptide library of the type outlined above (for conciseness we refer to these fragments as genes), the intention being to include these genes in a format suitable for peptide selection, e.g. phage display. The scheme (Fig. 1) involves the ligase-catalysed polymerization of sticky-ended double-stranded oligonucleotide building blocks and hinges upon the fact that there are two codons for cysteine, 5[prime]-UGU-3[prime] and 5[prime]-UGC-3[prime].
Figure 1. Scheme for constructing variable length genes encoding an even number of cysteine residues. Dark rectangle, restriction site for vector incorporation; clear rectangle, spacer; encircled C, cysteine residue. The assembled gene shown encodes four cysteine residues, but genes encoding two, six, eight... cysteine residues may also be formed. Box, supplementary propagating blocks providing for the cysteine arrangements CC and CXC, where X is a single amino acid and N represents 3 bp. Most cysteine arrangements may be generated with these blocks plus four others: equivalent TGC[CC] and TGC[CXC] blocks, a TGT[CC] block in which the internal TGT/ACA section and the spacer have exchanged positions and a TGT[CXC] block in which N and the spacer have exchanged positions. Spacers encode two, three, four... amino acids. There are four main types of building block in the scheme, two which propagate polymerization (TGT and TGC, the Egyptians of the title) and two which terminate polymerization (N-CAP and C-CAP). The blocks bear 3 nt overhangs representing one of the two cysteine codons. (Propagating blocks are named according to the sequence of their 5[prime]-overhang, e.g. a TGC block has a 5[prime]-TGC-3[prime] overhang; see also legend to Fig. 2A for block nomenclature.) These overhangs are arranged such that formation of genes encoding an even number of cysteines is favoured, thereby permitting an integral number of disulphide bonds in the encoded peptide. Each block also possesses a spacer (Fig. 1). The spacers allow the length of the encoded peptide to vary and also most arrangements of cysteines. Spacers for the propagating blocks encode two or more amino acids. [Note that a propagating block with a spacer encoding one amino acid is likely to be unstable (5) and a propagating block `encoding' 0 amino acids would be non-existent. We did, however, explore the possibility of using a spacer encoding one amino acid in this work.] Spacers for the terminating blocks encode one or more amino acids, or may be absent altogether. Arrangements involving CC and CXC (where X is a single amino acid) are catered for by supplementary propagating blocks (Fig. 1, Box) which insert between two standard blocks while still preserving an even number of cysteines. However, there are some arrangements which the scheme does not allow, specifically those involving only two consecutive cysteines (e.g. CC...CC...CC) or two cysteines separated by a single amino acid (e.g. CXC...CXC...CXC) or both (e.g. CC...CXC...CC). The scheme also forbids three or more consecutive cysteines (as may be found in one conotoxin; 2). Figure 2. (A) Blocks and primers used for this work. Propagating blocks are designated according to the sequence of their 5[prime]-overhang and the number of amino acids encoded by the double-stranded spacer. The supplementary propagating blocks encoding two consecutive cysteines at their `N-terminus' have an additional [CC] preceding the numeral; similarly, supplementary blocks encoding two cysteines separated by a single amino acid at the `N-terminus' would be denoted by an additional [CXC] preceding the numeral. In supplementary blocks encoding CC or CXC at their `C-terminus', the numeral would precede the square brackets. Three other TGC2 blocks were used in which the double-stranded spacer of the original block was replaced by recognition sites for XhoI, BsrBI and EarI (5[prime]-CTCGAG-3[prime], 5[prime]-GAGCGG-3[prime] and 5[prime]-CTCTTC-3[prime] respectively). The TGC1 block comprised 5[prime]-TGCTTG-3[prime] (top strand) and 3[prime] -AACACA-5[prime] (bottom strand). (B) Assembled gene and primers for its amplification. The last 3 nt of both PCRL(A) and PCRL(B) are complementary to the 3 nt encoding the terminal cysteines of the peptide. (C) Primers for amplification and sequencing of the gene ligated into pCR[trade]3. Below we present experimental evidence supporting the validity of the proposed scheme.
MATERIALS AND METHODS
Standard recombinant DNA techniques (6) were used throughout unless otherwise stated. Restriction enzymes, T4 polynucleotide kinase and T4 DNA ligase were purchased from New England Biolabs. Taq DNA polymerase, used for all PCRs, was obtained from Bioline. Oligonucleotides were purchased from Perkin-Elmer Applied Biosystems. DNA sequencing was performed on an ABI PRISM 377 DNA Sequencer.
Double-stranded oligonucleotide building blocks for the ligation reactions were constructed by annealing the appropriate oligonucleotides. All oligonucleotides were phosphorylated at the 5[prime]-terminus, apart from the upper strand of N-CAP and the lower strand of C-CAP. Oligonucleotides for the TGC4/TGT4 blocks (see Fig. 2A and its legend for a description of the blocks used in this work) employed in the ligation leading to the PCR screen shown in Figure 3B were purchased pre-phoshorylated. The remaining oligonucleotides were phosphorylated with T4 polynucleotide kinase according to the supplied instructions, except that, in preparation for ligation, T4 DNA ligase buffer was used. Annealing was performed by mixing an equal number of moles of each of the oligonucleotides constituting a particular block, making the solution up to 25 µl with T4 DNA ligase buffer and water and then lyophilizing. When mixing the oligonucleotides, care was taken to keep the volume of residual glycerol (from the T4 polynucleotide kinase storage solution) below 0.05 µl. Residual volumes greater than this prevented proper lyophilization. Each annealed block was dissolved in 25 µl water at 4°C.
Figure 3. Experimental verification of the proposed scheme. (A) PCR products of various ligation mixtures. Lane 1, TGC4 and TGT4. Each band corresponds to a gene encoding an even number of cysteine residues. The three smallest genes are indicated schematically on the left side of the figure (encircled N, N-CAP; encircled C, C-CAP; open circle, TGC4; filled circle, TGT4). Lane 2, TGC4, TGT4, TGC[CC]5 and TGT[CC]5. Lane 3, TGC2 and TGT4. Lane 4, TGC4 and TGT2. The unmarked gradations to the right side of the photograph correspond to 50 bp increments. (B) PCR screen of 17 transformants arising from a restriction digested TGC4/TGT4 ligation mixture cloned into the vector pCR[trade]3. Brackets I-IV identify bands corresponding to genes encoding two, four, six and eight cysteines respectively (constructs I-IV in the text). (C) PCR screen of 18 transformants arising from a restriction digested TGC2-6/TGT2-6/TGC[CC]5/TGT[CC]5 ligation mixture cloned into pCR[trade]3. Mr, 50 bp DNA ladder (Gibco). All PCR products were run on 3% agarose gels. Most ligations were performed with a ratio of N-CAP:TG-C:TGT:C-CAP of essentially 1:10:6:1. For the ligations leading to the PCR products shown in lanes 1, 3 and 4 of Figure 3A, concentrations were (N-CAP:TGC:TGT:C-CAP) 1:10:6:1 pmol. For the ligation leading to the PCR products shown in lane 2 of Figure 3A, concentrations were (N-CAP:TGC4:TGT4:TG-C[CC]5:TGT[CC]5:C-CAP) 1:6:2:2:2:1 pmol, maintaining the concentrations of propagating block overhangs implicit in the original 1:10:6:1 ratio. For the ligation leading to the PCR screen of Figure 3B, the concentrations of (N-CAP:TGC:TGT:C-CAP) were increased to 5.5:55:33:5.5 pmol, so that the mixture could be visualized on a gel. For the initial multi-block ligations involving TGC2-6, TGT2-6, TGC[CC]5 and TGT[CC]5, the concentration of each standard TGC block was 10 pmol and of each TGT, TGC[CC]5 and TGT[CC]5 block 5 pmol, with N-CAP and C-CAP 6 pmol. In subsequent multi-block ligations involving the TGC2 blocks containing a recognition site for a restriction enzyme, the concentration of TGC2 was 30 pmol, TGC3-6 each 6 pmol, TGT2 18 pmol, TGT3-6, TGC[CC]5 and TGT[CC]5 each 3.6 pmol and N-CAP and C-CAP both 6 pmol. To perform the ligations, the propagating blocks were mixed, T4 DNA ligase added (4 U/µl reaction volume) and the mixture left for 1 h at 0 (TGC1/TGT4 ligation) or 4°C (all other ligations). (Ligations were performed at low temperature to enhance the stability of the blocks.) Subsequently N-CAP and C-CAP were added, together with more T4 DNA ligase (200 U), and the new mixture left for a further 2 h at 4°C. PCR amplification of ligation mixtures (0.5 µl) was achieved with primers PCRL(A) and PCRL(B) (Fig. 2B) using the programme: [94°C, 1 min; 65°C, 1 min; 72°C, 1 min] × 25; 72°C, 5 min. Including the two 3 bp terminal joints, amplification with these primers appended 54 bp to the insert between N-CAP and C-CAP [e.g. for a TGC4:TGT4:TGC4 insert, the size of the PCR product would be 54 + (3 × 12) + (2 × 3) = 96 bp]. After alcohol precipitation, the relevant ligations were restriction digested with HindIII and XbaI and purified by agarose gel electrophoresis, excising DNA in the range 50-150 bp (TGC4/TGT4 ligation) and 60-150 bp (TGC2-6/TGT2-6/TGC[CC]-5/TGT[CC]5 ligations). Following digestion, the mixture was ligated into HindIII- and XbaI-cut pCR[trade]3 (Invitrogen) and transformed into Escherichia coli. PCR screening of ampicillin-resistant colonies was performed with the primers T7 and Sp6 (T7, 5[prime]-TAATACGACTCACTATAGGG-3[prime]; Sp6, 5[prime]-GATTTAGGTGACACTATAG-3[prime]) (Fig. 2C), using the programme: 94°C, 5 min; [94°C, 1 min; 55°C, 1 min; 72°C, 1 min] × 30; 72°C, 5 min. Together with the two terminal 3 bp joints, these primers appended 103 bp to the insert. Restriction digestions of inserts amplified in PCR screens were carried out after they were alcohol precipitated.Sequencing was performed with the primer pCR3seq, 5[prime]-AGGTCTATATAAGCAGAGCT-3[prime] (Fig. 2C).
RESULTS
To test the principle of our scheme, we performed a ligation with TGT and TGC blocks comprising 12 putative base pairs (TGC4 and TGT4 respectively, where 4 represents the number of amino acids encoded by the spacer; see legend to Fig. 2A). Amplification of a small part of the ligation mixture with primers specific for N-CAP and C-CAP gave a ladder of bands (Fig. 3A, lane 1). Each band corresponded in size to a gene encoding an even number of cysteine residues (Fig. 3A, schematic). A similar TGC4/TGT4 ligation mixture was restriction digested, purified on an agarose gel, isolating DNA expected to include genes encoding two, four, six and eight cysteines only (designated constructs I-IV respectively), and inserted into a vector. The vector population was transformed into E.coli and transformants screened for inserts (Fig. 3B). Of 35 screened transformants, 33 gave bands corresponding to genes encoding an even number of cysteine residues, with a bias towards construct I (I, 14; II, 6; III, 8; IV, 5). Two bands were larger than expected and probably corresponded to two constructs ligated together. We did not encounter this problem again. Twelve (three of each construct) inserts were sequenced; each sequence confirmed the fidelity of the assembly process.
Having shown successful assembly for two particular propagating blocks, we then went on to investigate the stability, under the ligation conditions used, of two of the smallest double-stranded propagating blocks (TGC2 and TGT2) needed for the scheme. We did two ligations, one with TGC2 and TGT4 and one with TGT2 and TGC4. On amplification, each ligation mixture gave a ladder diagnostic of correct gene formation (Fig. 3A, lanes 3 and 4), although the extent of large construct formation was greater for TGC2/TGT4 (lane 3). Correct ladder formation was not observed when a TGC1/TGT4 ligation mixture was PCR amplified (data not shown).
Subsequently we investigated the ability of the supplementary blocks to insert between standard blocks by performing a TGC4/TGT4 ligation in the presence of TGC[CC]5 and TGT[CC]5. Amplification of this mixture gave a ladder consistent with correct insertion events (Fig. 3A, lane 2), with the lowest band unaffected and the higher bands smeared. (The second lowest band probably corresponds to constructs containing a standard propagating block ligated to a supplementary one.)
Finally, we performed a ligation involving TGC2-6, TGT2-6, TGC[CC]5 and TGT[CC]5. Following restriction digestion of the ligation mixture, DNA expected to include genes encoding between four and twelve cysteine residues was isolated. We deliberately tried to avoid genes encoding two cysteines in the light of the bias towards construct I seen in the TGC4/TGT4 experiment described above. (These genes could in any case be obtained in a separate ligation involving the TGC2-6 blocks alone.) Despite this precaution, about a quarter of the bands from PCR screens were suggestive of genes encoding two cysteines. The remaining bands were within the expected upper size limit; the bands were also encouragingly varied in size (Fig. 3C).
Fifty nine inserts from eight independent ligations were sequenced. The majority of these sequences (fifty four) showed that polymerization had occurred as desired; the remaining five sequences indicated undesirable assembly events (e.g. TGC6 ligated to TGC5). Forty three of the fifty four correctly assembled genes were unique. Concatamers of the supplementary blocks were observed in only two cases, both involving TGT[CC]5 and comprising two and three blocks (we wanted to avoid extensive concatenation as it would give genes encoding dense arrays of cysteines, leading to peptides unlikely to fold productively). Inserts which had assembled correctly contained one, two, three, four, five, six and seven propagating blocks (five, ten, twenty, six, seven, five and one inserts respectively) and encoded two, four, six, eight and ten cysteines (five, twenty nine, thirteen, six and one insert respectively). Theoretically we might also have encountered inserts with 8-10 propagating blocks.
The sequences also revealed certain trends for the individual blocks. Thus TGT5 was rarely incorporated, while TGC5, although it occurred regularly, was not often found ligated to N-CAP. TGC[CC]5, on the other hand, was only found ligated to N-CAP, unlike its sister block TGT[CC]5, which did insert between the standard propagating blocks and was also found ligated to C-CAP. TGC[CC]5 was able to self-polymerize in the absence of other blocks, however (data not shown).
Far more worrying was the complete absence of TGC2 in inserts from the initial multi-block ligations and the infrequent appearance of TGT2 (about one in twelve inserts). In an attempt to rectify this situation, the concentration of both blocks was raised to five times that of their larger counterparts. This increased the frequency with which TGT2 occurred to a more acceptable level (approximately one in seven inserts), but TGC2 was still undetectable. To enable large numbers of inserts to be screened for the presence of TGC2, a new TGC2 block was designed, TGC2[Xho] (see legend to Fig. 2A), in which the putative double-stranded region comprised a XhoI recognition site (TGT2 already had an EcoRI site for this purpose). Like TGT2, this new block had a G:C base pair at either end which we hoped would make the block less susceptible to possible fraying. It also had two internal G:C base pairs, lending it extra stability. Digestion of amplified inserts from multi-block ligations involving TGC2[Xho] did indeed suggest that it had been included in some cases. On sequencing these inserts, however, it was apparent that the oligonucleotides comprising TGC2[Xho] and TGT2 had self-annealed (TGC2[Xho] had also annealed as required) and that these aberrant blocks had facilitated each other's incorporation into the assembly product, severely disrupting the ligation scheme. This mutually assisted misincorporation had not been possible with the previous TGC2/TGT2 pairing since only the TGT2 oligonucleotides could self-anneal. Self-annealing of the TGT2 oligonucleotides would lower the concentration of the TGT2 block proper, possibly explaining its infrequent inclusion in the multi-block ligations and the attenuated ladder of bands from the PCR-amplified TGC4/TGT2 ligation mixture (Fig. 3A, lane 4).
To resolve the self-annealing problem in the TGC2 block (we were less concerned about TGT2 since we had already observed its correct incorporation) two new blocks were devised in which the palindromic XhoI site was replaced with the asymmetrical recognition sequences of BsrBI and EarI (see legend to Fig. 2A). Like TGC2[Xho], both blocks have terminal and internal G:C pairs which ought to make them relatively robust. In restriction digestion screens of amplified inserts from multi-block ligations involving either TGC2[Bsr] or TGC2[Ear], both sites were detected at satisfactory frequencies (about one in five inserts and one in three inserts respectively). Sequencing of inserts containing these sites showed that both blocks had incorporated as intended.
DISCUSSION
We have conceived and tested a scheme for assembling genes encoding an even number of cysteine residues and which vary in length. The scheme involves the ligation of double-stranded blocks of DNA with sticky ends. It works reasonably well. Most assembled genes do encode an even number of cysteines, although some (<10%) exhibit an inappropriate assembly event resulting in these genes encoding an odd number of cysteines. The genes encode quite diverse arrangements of cysteines and their length, which may be controlled at the stage when the assembled DNA fragments are isolated by gel electrophoresis, also shows good variation (Fig. 3C).
Nearly all the blocks with which we attempted the scheme participated in gene assembly. These blocks had spacers which encoded between two and six amino acids. (A block whose spacer encoded one amino acid, TGC1, did not participate in assembly, as expected.) The behaviour of two of the smallest blocks, TGC2 and TGT2, and a supplementary block, TGC[CC]5, was of some concern, however. In order to detect TGC2 and TGT2 at frequencies comparable with the other blocks, the concentrations of both had to be raised relative to these blocks. In addition, only those TGC2 blocks with a G:C base pair at either end of the double-stranded region were observed (the TGT2 block fortuitously already had G:C base pairs at these locations). If one wanted to include some sequence variation in the smallest blocks, this apparent G:C requirement would restrict the first amino acid encoded by the blocks to just ten (L, P, H, Q, R, V, A, D, E and G), assuming these blocks were made by annealing complementary oligonucleotides, as they were here (see Materials and Methods), although it would not limit the identity of the second amino acid. However, the sequence diversity encoded by the genes is more critically affected by the format in which their peptide products are selected (see final paragraph).
The other problematical block, TGC[CC]5, was only found ligated to N-CAP and could not be detected inserted between two standard propagating blocks. Again, although irksome, this preference would not be fatal to the scheme, which may be realized with three alternative sets of supplementary blocks, (TGT-[CC][X1-6], TGT[CXC][X1-6], TGT[X1-6][CC], TGT[X1-6]-[CXC], TGC[CC][X1-6] and TGC[CXC][X1-6) or (TGC-[CC][X1-6], TGC[CXC][X1-6], TGC[X1-6][CC], TGC[X1-6]-[CXC], TGT[X1-6][CC] and TGT[X1-6][CXC]) or a combination of all eight types of supplementary block (see legend to Fig. 2A for supplementary block nomenclature). Using the first set of supplementary blocks, the scheme requires that the TGC supplementary blocks ligate to N-CAP alone or as self-polymerized units (and TGC[CC]5 does self-polymerize) and that the TGT supplementary blocks ligate to C-CAP as well as insert between the standard propagating blocks, which TGT[CC]5 can do.
Overall, therefore, it seems that the scheme could be used to construct genes encoding most combinations of cysteines. If the peptides in a potential library were restricted in length to [le]30 residues, with 0-6 residues between consecutive cysteines, roughly in keeping with the conotoxins (2), the maximum number of folds attainable with the scheme is ~107. [The number of folds = number of possible cysteine arrangements × number of ways of pairing cysteines, by disulphide bonds, in a particular arrangement. The number of possible cysteine arrangements in a sequence of length L containing N cysteines is the coefficient of XL - N in the escargot polynomial (1 + X + X2 +...+ X6)N - 1 once it has been expanded. So for a peptide containing 30 amino acids, eight of which are cysteines, the relevant term is 59 710X22 and thus the number of possible cysteine arrangements in this peptide is 59 710. The polynomial arises by considering sequences as products, e.g. CXXCXCXC = C4X4. In formulating it, it was assumed that amino acids beyond the two terminal cysteines contribute little to the peptide fold; L is therefore defined by these two termini. The polynomial may also be used to compute the number of sequences in the library. The number of ways of pairing N cysteines is given by the expression (N - 1) × (N - 3) ×...× 1.] Some of these potential folds will be excluded on stereochemical grounds, while others will be disfavoured energetically. Therefore the actual number of folds will almost certainly be <107.
Sequence variation could be introduced into the library with blocks encoding random amino acids. Theoretically these blocks could give rise to ~1033 different sequences in the 30 amino acid size range, even taking into account sequence restrictions imposed on the TGC2 and TGT2 blocks. This greatly exceeds the library size manageable with a standard selection format such as phage display (the size of conventional phage display libraries is [le]108; 7). Thus the number of amino acid sequences in a library created through this scheme would be only a tiny fraction of those possible. However, the scheme seeks primarily to generate a collection of shapes into which sequence diversity may be subsequently introduced following selection of a shape complementary to a given target. There are clearly limitations to this approach (for example, as a consequence of the poor sequence diversity some shapes may be electrostatically incompatible with the target due to charge repulsions and, therefore may not be selected), but despite these limitations we feel that the approach is one worth pursuing.
ACKNOWLEDGEMENTS
This work was suggested by a talk given by Bruce Livett at the University of Bath in 1995 and profited from very helpful conversations with Mark Searle, Dek Woolfson and Alan Cox. Financial support from The Royal Society, The Wellcome Trust, the BBSRC and the University of Bath is gratefully acknowledged, as is the use of laboratory space in the Department of Biology and Biochemistry at Bath. Further details of the mathematics used in this paper may be found on the Internet at http://www.bath.ac.uk/~masgcs/escargot.html
REFERENCES
This article has been cited by other articles:
This page is run by Oxford University Press, Great Clarendon Street, Oxford OX2 6DP, as part of the OUP Journals
Comments and feedback: www-admin{at}oup.co.uk
Last modification: 19 May 1998
Copyright©Oxford University Press, 1998.
CiteULike 
Connotea 
Del.icio.us What's this?
S. G. Conticello, Y. Gilad, N. Avidan, E. Ben-Asher, Z. Levy, and M. Fainzilber
Mechanisms for Evolving Hypervariability: The Case of Conopeptides
Mol. Biol. Evol.,
February 1, 2001;
18(2):
120 - 131.
[Abstract]
[Full Text]
This Article 
Abstract
Print PDF (92K)
Alert me when this article is cited
Alert me if a correction is posted
Services
Email this article to a friend
Similar articles in this journal
Similar articles in ISI Web of Science
Similar articles in PubMed
Alert me to new issues of the journal
Add to My Personal Archive
Download to citation manager
Search for citing articles in:
ISI Web of Science (3)
Request Permissions
Commercial Re-use Guidelines
for Open Access NAR Content
Google Scholar
Articles by Palmer, S. J.
Articles by Cox, J. P.
Search for Related Content
PubMed
PubMed Citation
Articles by Palmer, S. J.
Articles by Cox, J. P.
Social Bookmarking
What's this?