Vascular endothelial growth factors encoded by Orf virus show surprising sequence variation but have a conserved, functionally relevant structure

Orf virus, the type species of the Parapoxvirus genus, causes a contagious, proliferative skin infection in sheep, goats and humans (Mercer & Haig, 1999 ). Several reports have described prominent vascular changes associated with lesions caused by Orf virus. The discovery that Orf virus encodes a homologue of mammalian vascular endothelial growth factor (VEGF-A) suggested a possible explanation for these observations (Lyttle et al., 1994 ). VEGF-A is a mitogen specific for vascular endothelial cells, a potent inducer of vascular permeability and plays a critical role in the formation of new blood vessels during both vasculogenesis and angiogenesis (Ferrara, 1999 ; Robinson & Stringer, 2001 ; Stacker & Achen, 1999 ). Disruption of the VEGF-A gene results in embryonic lethality because of failed development of the vasculature. VEGF-A also plays an important role in the angiogenesis associated with tumour formation and a number of other pathological conditions. In addition to a series of isoforms of VEGF-A, the VEGF family of growth factors includes VEGF-B, VEGF-C, VEGF-D and placental growth factor (PlGF). All are secreted homodimeric glycoproteins that share 30 to 45% amino acid sequence identity. All contain a VEGF homology domain (VHD) that includes eight cysteine residues involved in intra-or inter-chain disulphide bonds and are hence members of the cystine-knot family of proteins. The VEGF family are ligands for a set of tyrosine kinase receptors. VEGF-A binds and activates VEGF receptor 1 (VEGFR-1) (Flt1) and VEGFR-2 (Flk1/KDR) but not VEGFR-3 (Flt4). PlGF and VEGF-B bind only VEGFR-1 while VEGF-C and VEGF-D bind both VEGFR-2 and VEGFR-3 but not VEGFR-1. The role of each receptor-ligand interaction remains to be precisely defined but in general terms it appears that VEGFR-1 plays a role in vascular endothelial differentiation and migration, VEGFR-2 is the main signal-transducing VEGF receptor for mitogenesis of endothelial cells and VEGFR-3 is involved in angiogenesis of the lymphatic vasculature.

We and others have recently demonstrated that VEGF encoded by Orf virus is a cystine-knot homodimer that has some of the functional features of mammalian VEGF (Meyer et al., 1999 ; Ogawa et al., 1998 ; Wise et al., 1999 ). The purified factor derived from Orf virus strain NZ2 was able to stimulate proliferation of vascular endothelial cells, promote vascular permeability and was shown to bind VEGFR-2 but not VEGFR-1 or VEGFR-3. As such the viral VEGF forms a new member of the VEGF family of molecules with a unique profile of receptor recognition. We have also shown that infection of sheep with a recombinant Orf virus in which the viral VEGF gene was partially deleted resulted in lesions that lacked the extensive dermal vascularization seen in wild-type infections (Savory et al., 2000 ). No other virus has been reported to encode a form of VEGF.

The original report of a possible viral VEGF described two independent New Zealand isolates of Orf virus (Lyttle et al., 1994 ). The isolates (NZ2 and NZ7) encode predicted peptides of similar size with similar levels of sequence relatedness to mammalian VEGFs and both peptides have been shown to have similar VEGF-like activities (Ogawa et al., 1998 ; Wise et al., 1999 ). However, the two viral genes are unexpectedly different from each other sharing only 41·1% predicted amino acid sequence identity. This raised the possibility that further variants of VEGF are encoded by other isolates of Orf virus. In order to investigate the sequence variation and distribution of VEGF genes of Orf virus we examined a selection of isolates from diverse sources.

Viruses and cells.
Orf virus strains NZ2, NZ9, NZ10, NZ12, NZ52, NZ66 and NZ180 are derived from naturally infected sheep in New Zealand (Robinson et al., 1987 ). NZ7 was provided by Coopers Animal Health New Zealand Ltd. With the exception of NZ2, NZ7 and NZ10, the samples analysed were derived directly from clinical cases of contagious ecthyma in lambs. Strains NZ2, NZ7 and NZ10 were plaque-purified in bovine testis cells and then inoculated onto sheep. Scabs that formed were recovered and used as the source of viral DNA (Mercer et al., 1987 ). Strain NZh1 was recovered in Dunedin, New Zealand from a naturally infected adult woman. Strain Acsl is a contagious ecthyma vaccine originating from Commonwealth Serum Laboratories, Australia. MRI1 and MRI3 are samples derived from naturally infected sheep in South Glamorgan, Wales and Inverness, Scotland, respectively, and have been passaged once on sheep (Gilray et al., 1998 ). MRIsc is a Scottish reference strain maintained on sheep and which has not been adapted to grow in cell culture (elsewhere referred to as MRI scab), while MRI11 is a Scottish reference strain maintained by passage in primary bovine testis or foetal lamb muscle and elsewhere referred to as MRI orf11 (McInnes et al., 2001 ). Strains ITtor, ITC2, IT7, IT90, IT19 and IT20 are from Italy. IT7, IT19 and IT20 are scab samples from naturally infected sheep. The other three have been adapted to grow in ovine testis cells. IT90 and ITtor are derived from samples taken from chamois while ITC2 is from sheep. Strains D15, D23, D47 and D1701 were recovered from naturally infected sheep or goat (D47) in Germany and have been propagated in cultured bovine foetal lung cells or BK-KL3A cells. D1701 is a highly attenuated strain derived after multiple cell culture passages (Cottone et al., 1998 ).

DNA manipulations.
Virus was purified and DNA recovered as previously described (Mercer et al., 1987 ). The primer pairs used in PCRs were derived from the first and last six codons of D1701 VEGF (Meyer et al., 1999 ), NZ2 VEGF or NZ7 VEGF (Lyttle et al., 1994 ). PCRs were performed using 30 cycles of 50 °C 20 s, 72 °C 20 s and 95 °C 20 s. The DNA sequences of the PCR products were either determined directly or the amplified products were cloned into plasmid vectors and the sequence determined from at least five different clones. Double-stranded DNA templates were prepared and sequenced by procedures recommended by Applied Biosystems Inc. (ABI). The products of sequencing reactions were analysed with an ABI model 373A sequencing system. Assembly and analysis of nucleotide sequence was conducted using programs of the DNASTAR package. Phylogenetic analysis of aligned sequences was conducted using programs of the PHYLIP package version 3·6 (Felsenstein, 1989 ); genetic distances were calculated using PROTDIST (JonesTaylorThornton matrix) and phylogenetic trees constructed by the neighbour-joining method (NEIGHBOR).

Prediction of the tertiary structure of the VEGF-variants of Orf virus.
The structure of the VEGF-variants of Orf virus were modelled using SWISSMODEL and the SWISSPDBVIEWER protein modelling program (version 3.5; Guex & Peitsch, 1997 ), which are both available from the ExPASy website (http://www.expasy.ch/spdbv). The sequences of the variants were aligned using SWISSMODEL against protein subunits A and B of human VEGF-A (Muller et al., 1997a ) (PDB identifier, 2VPF). The Iterative Magic Fit function was used for energy minimization and the alignments were manually optimized. Ramachandran plots for the viral VEGF models, VEGF-A and PlGF, were compared to determine if the viral models contained residues that did not conform to acceptable φ and/or ψ angles.

Sequence comparisons of viral VEGFs
The presence of a VEGF-like gene in 21 isolates of Orf virus was examined by PCR or by nucleotide sequence determination of cloned genomic fragments (isolates NZ2, NZ7, D15, D1701, MRIsc and MRI11). PCRs were performed using scab samples, infected cell lysates or DNA recovered from purified virus and used primer pairs which incorporated the first or last 18 nt of the coding regions of NZ2, NZ7 or D1701-derived VEGF genes. All isolates tested generated a PCR product of appropriate size with one of the primer pairs (not shown). The nucleotide sequence of each PCR product was determined and compared with published sequences. Three isolates were found to encode sequences very closely related to the VEGF of NZ7. The nucleotide sequences derived from NZ52 and IT7 were identical to that of the VEGF of NZ7 while the NZ180-derived sequence had a single nucleotide difference at position 187 (A to G) resulting in the predicted amino acid changing from threonine to alanine (not shown). These data may slightly under-represent the extent of variation between VEGF genes of Orf virus since the PCR products were generated using primers which included a total of 36 nt from the termini of the VEGF open reading frame. Attempts to use primers derived from sequences external to the VEGF coding region were not successful.

The majority of isolates tested (18) encoded VEGF-like genes with sequences related to that of NZ2 VEGF. An alignment of the predicted amino acid sequences of these NZ2-like VEGFs is shown in Fig. 1 along with the established sequences for NZ2, D1701 and NZ7. The nucleotide and amino acid relationships between the NZ2-like VEGFs are listed in Table 1. In only one case (NZ9) was the VEGF nucleotide sequence identical to that of NZ2. The NZ2-like VEGF genes were on average 93·5% identical to NZ2 VEGF in nucleotide sequence. In all 190 combinations of the 20 NZ2-like VEGFs, there were only eight combinations that were identical at the amino acid level and the average amino acid identity between pairs of sequences was 86·1% with a standard deviation of 7·6.

(36K): Fig. 1. Comparison of NZ2-like VEGFs from Orf virus. The deduced amino acid sequences of the NZ2-like VEGFs were aligned with each other using the MEGALIGN program of the DNASTAR sequence analysis package using the Jotun Hein method with PAM250 residue weight table. The complete sequence of VEGF from Orf virus NZ2 is shown in the top line. Residues identical to those of the NZ2-encoded VEGF are indicated by dots, while variant residues are listed. Also included are the amino acid sequences of NZ7-encoded VEGF and the VEGF homology domain (VHD) of human VEGF121 (accession no. P15692). The eight cysteine residues of the cystine-knot motif are indicated above the NZ2 sequence. Potential O-linked glycosylation sites as predicted by NETOGLYC 2.0 (http://genome.cbs.dtu.dk/services/NetOGlyc) are indicated by # symbols and a potential N-linked glycosylation recognition site is indicated by * symbols.

Table 1. NZ2-like VEGF sequence comparisons

The sequence variations we report here are not the result of alterations that have occurred during propagation in the laboratory. Poxviruses including Orf virus are known to undergo genome alterations such as duplication, transposition and deletion when they are propagated in cultured cells (Fleming et al., 1995 ). However, the majority of our data is derived from samples taken directly from naturally infected animals (MRI1, MRI3, IT7, IT19, IT20, NZ9, NZ12, NZ52, NZ66 and NZ180). Strain MRI11 has been adapted to grow in cultured cells and this is the most-variant of the NZ2-like VEGFs, with average sequence identities to the other clones of 76·99±0·77% (nucleotide) and 71·84±1·53% (amino acid). However, the VEGFs derived from other strains known to have been repeatedly passaged in cultured cells (D1701 and ITC2) are no more variant than scab-derived samples. Furthermore, a passaged derivative of strain NZ2 that is known to have undergone substantial genomic alterations (Fleming et al., 1995 ) carries a VEGF gene identical in nucleotide sequence to that of the parent strain (unpublished). Alignment of DNA sequences revealed only 31·9% sequence identity for MRI11 and NZ7 but 73·1% identity for MRI11 and NZ2, confirming that, despite its extreme variation, MRI11 is correctly included in the NZ2-like group. Conditioned medium from MRI11-infected cells showed activity in a VEGFR-2 bioassay (Wise et al., 1999 ), suggesting that the MRI11-encoded VEGF is biologically active (unpublished).

The inclusion of MRI11 in the NZ2-like group is also supported by the phylogenetic analysis shown in Fig. 2. This phylogenetic tree illustrates the relationships between the viral VEGFs and demonstrates the clear separation of NZ2-like and NZ7-like VEGFs. It also reveals some evidence of geographical clustering with peptides of identical (ITtor, ITC2 and IT90, NZ10, Acsl and NZh1) or very similar sequences (IT19 and IT20, and NZ2, NZ66 and NZ9) likely to have come from the same country. However, the differences between MRI1 and MRI3, between NZ2 and NZ12 and between IT90 and IT20 clearly indicate that more than one clone is present in each country. The amino acid sequence identity between the human derived clone (NZh1) and NZ10 suggests that there is not a specific clone infecting humans rather than sheep and is consistent with the evidence that human infection by Orf virus is sourced from sheep. Similarly, the sequence identity between IT90, ITtor and ITC2 suggests that the same clones of Orf virus infect both sheep and chamois.

(7K): Fig. 2. Phylogenetic tree of VEGFs encoded by Orf virus. In addition to NZ2- and NZ7-like viral VEGFs, the tree includes the VEGF homology domain (VHD) of human VEGF121 (accession no. P15692). The tree was produced using the neighbour-joining method. Bootstrap values of greater than 75% (of 100 replicates) are shown in support of each node. Vertical distances are arbitrary; horizontal branch lengths are proportional to genetic distance.

The extent of sequence variation seen in VEGF coding regions from independent isolates of Orf virus is greater than that usually reported for poxvirus genes. A comparison can be made with the orthologue of mammalian epidermal growth factor which several poxviruses have been shown to encode (Chang et al., 1987 ; Shchelkunov et al., 1998 ; Twardzik et al., 1985 ). This factor is not encoded by Orf virus but, like the VEGF of Orf virus, it is a secreted cytokine that is a ligand for membrane-bound receptors and stimulates the replication of target cells. Amongst the epidermal growth factors from four strains of vaccinia virus (Copenhagen, WR, Ankara and Tian Tan; accession nos P20494, P01136, AAB96482 and AAF33857 respectively) the extent of amino acid sequence identity ranges from 91·4 to 98·9% with an average of 95·5%. Examination of the data for three isolates of variola virus (Bangladesh 1975, India 1967 and Garcia 1966; accession nos AAA60749, P33804, and C72150 respectively) showed amino acid sequence identity ranging from 97·1 to 99·3%. Comparison of the vaccinia virus factors with the variola virus factors revealed that the amino acid sequence identity between species ranged from 82·9 to 88·6%. A similar range was observed between these sequences and the predicted protein of a third species of orthopoxvirus, cowpox virus. These comparisons illustrate that the extent of sequence variation seen in VEGFs from Orf virus is unusually high and is more similar to the variation seen in homologous proteins from different species of the same genus of poxviruses rather than between independent isolates of the same species.

The sequence variation seen in VEGF genes of Orf virus is not a general feature of genes of this virus. For example, Orf virus also encodes a homologue of the cytokine interleukin-10 (IL-10) (Fleming et al., 2000 ). The predicted amino acid sequences for the IL-10s encoded by Orf virus strains NZ2 and NZ7 differ by only 2·7%. Two further examples are provided by an interferon resistance element and an inhibitor of GM-CSF encoded by Orf virus. Inter-strain amino acid sequence variation for both these elements is less than 3% (Deane et al., 2000 ; McInnes et al., 1998 ).

Nor is the sequence variation seen in VEGF genes of Orf virus likely to reflect accelerated genetic drift of silent genes. In all isolates key functional motifs such as the cystine-knot motif are conserved. Furthermore, in addition to the characterized activity of the VEGFs encoded by strains NZ2, D1701 and NZ7 (Meyer et al., 1999 ; Ogawa et al., 1998 ; Wise et al., 1999 ), we detected VEGF-like activity in conditioned medium of cells infected with each of strains D15, D23, D47 or MRI11 (not shown).

Receptor-binding motifs
Members of the VEGF family share a VEGF homology domain (VHD) containing eight cysteine residues that form the cystine-knot motif and link the subunits of the anti-parallel homodimer. Despite the sequence variation detailed above, all eight cysteines are conserved in all of the VEGF-variants of Orf virus (Fig. 1). Also conserved in each of the VEGFs of Orf virus are a potential signal sequence and potential N- and O-linked glycosylation sites (Fig. 1). A threonine/proline-rich motif seen in the C terminus of VEGFs encoded by both NZ2 and NZ7 but not seen in mammalian VEGFs is retained in all viral VEGFs (Fig. 1).

The conservation of structure within the VEGF family of growth factors suggests that residues important in mediating the binding of different members to their receptors are likely to be conserved. We aligned representatives of the VEGF family with the viral VEGFs and looked for any correlation with the ability to bind VEGFR-1 or VEGFR-2 (Fig. 3). We focused initially on residues that have been implicated in mediating binding to these receptors.

(45K): Fig. 3. Comparison of viral VEGFs with other members of the VEGF family. The VEGFs encoded by strains NZ2 and NZ7 were used as representatives of the diversity within the VEGFs encoded by Orf virus. The VHD of the viral VEGFs were aligned with the VHD of human PlGF (GenBank accession no. CAA38698), human VEGF-B (AAC50721), human VEGF-C (P49767), human VEGF-D (O43915) and human VEGF-A (AAF19659). Residues are shaded and coloured to indicate that mutational analysis has implicated them in determining the binding of VEGF-A to VEGFR-1 or VEGFR-2, or that they contribute to a groove structure hypothesized to bind the domain 23 linker region of VEGFR-1. The colour coding scheme is as follows with the residue numbers indicating the relevant VEGF-A residues. Yellow residues (Q22, L66, E67) bind to VEGFR-1 only. Red residues (Q79,I83, P85) bind to VEGFR-2 only. Blue residues (D34, F36, S50) are a groove component only. Cyan residues (F17, Y21) bind to both VEGFR-1 and VEGFR-2. The green residue (D63) binds VEGFR-1 and is a groove component. Purple residues (I43, I46) bind to VEGFR-2 and are a groove component. The white residue (E64) binds to both VEGFR-1 and VEGFR-2 and is a groove component. Secondary structural elements of VEGF-A are indicated below the alignments by lower case letters with a representing alpha-helices and b representing beta-strands.

Mutational analyses have implicated seven residues in the binding of VEGF-A to VEGFR-1 (Phe-17, Tyr-21, Gln-22, Asp-63, Glu-64, Leu-66 and Glu-67; Keyt et al., 1996 ; Li et al., 2000 ). Ala substitutions have implicated residues Phe-17, Tyr-21, Ile-43, Ile-46, Glu-64, Gln-79, Ile-83 and Pro-85 as playing important roles in the binding of VEGF-A to VEGFR-2 (Li et al., 2000 ; Muller et al., 1997a ). Comparisons of these positions in VEGF family members, including the viral VEGFs, revealed little evidence of correlations with the ability or inability to recognize VEGFR-1 or VEGFR-2 (Fig. 3). Ile-43 of VEGF-A is conserved in VEGF-As of 12 mammalian species and is matched by a Val at this position in four species of VEGF-B and three species of PlGF-1 (Fig. 3). These factors all recognize VEGFR-1, whereas in VEGF-C (four species) and in VEGF-D (three species), which do not recognize VEGFR-1, a Thr is conserved at this position. Furthermore, all 24 isolates of Orf virus (both NZ2 and NZ7-like) also carry a Thr at this position. Despite these correlations, an Ala substitution of Ile-43 of VEGF-A has been shown to reduce binding to VEGFR-1 only modestly (Li et al., 2000 ) and has a greater effect on the recognition of VEGFR-2 (Li et al., 2000 ; Muller et al., 1997b ).

Correlations with the ability to recognize specific VEGF receptors were detected in other residues. Ser-50 of VEGF-A is retained in all factors that recognize VEGFR-1 but is not present in any of the factors that do not recognize this receptor (Fig. 3). The NZ2-like viral VEGFs along with VEGF-C and VEGF-D generally have a Pro at this position. Mutational analysis of this residue has not been reported.

A strong correlation is apparent between binding of VEGFR-2 and the presence of Asn at a position equivalent to Asn-62 of VEGF-A. This residue is conserved in 11 of 12 species of VEGF-A, four species of VEGF-C, three species of VEGF-D and all 24 viral isolates. In contrast this residue is not found in PlGF-1 (three species have Gly) or VEGF-B (4 species have Pro) which do not recognize VEGFR-2. Crystal structure determination of VEGF-A has shown that Asn-62 is an accessible residue on the receptor-binding face of the ligand (Muller et al., 1997a ). However substitution of this residue by Ala reduced the relative affinity of VEGF-A for VEGFR-2 only modestly (Muller et al., 1997b ).

Structural modelling
Predicted structures of the VEGF-variants of Orf virus were determined by comparison to the solved crystal structure of subunits A and B of VEGF-A. Despite only moderate sequence identity, the structures predicted for the viral VEGF monomers are very similar both to each other and to the structure determined by X-ray crystallography for VEGF-A (Fig. 4). They conserve the central anti-parallel, four-stranded β sheet and two α-helical segments. With the exception of the VEGFs encoded by Orf virus strains NZ7 (ORFV_NZ7VEGF) and MRI11 (ORFV_MRI11VEGF), the modelling program we used did not detect interruptions in the fourth and fifth β sheets so as to generate β4/β5 and β6/β7 as determined for published structures of VEGF (Muller et al., 1997a ). However, for greater clarity we have retained the published nomenclature of seven β sheets. The orientations of the cysteines forming the intra-chain disulphide bonds responsible for the cystine-knot motif are maintained in the viral VEGFs (Muller et al., 1997a ). Also maintained is the three-stranded anti-parallel β sheet (β2, β5 and β6) at the opposite end of the monomer to the cystine-knot and shown to be part of the receptor-binding face of VEGF-A (Muller et al., 1997a , b ). The viral VEGF monomers also contain the three variable loop regions, connecting strands β1 to β3, β3 to β4 and β5 to β6. The third loop region is extended by three and five amino acids in ORFV_MRI11VEGF and ORFV_NZ7VEGF, respectively.

(25K): Fig. 4. Ribbon representations of the predicted structures of monomers of selected members of the VEGF family. The sequences used in the model constructions were the VHDs listed in Fig. 3. Secondary structure elements of VEGF-A (Muller et al.,1997a ) and the N and C termini are labelled. Models were derived as described in Methods.

As a test of validity of the predicted models of the variant VEGFs, Ramachandran plots were made for each structure and compared to that of the solved crystal structures of VEGF-A and PlGF. For the NZ2-like VEGF structures, 9899·5% of residues fell within allowed conformational areas. For ORFV_NZ7VEGF and ORFV_MRI11VEGF, 7 and 4% of residues, respectively, fell outside the accepted conformational areas. The majority of these residues corresponded to the extended loop-3 regions of each structure that could not be modelled with any degree of certainty.

VEGF-A dimerizes in an anti-parallel, side-by-side fashion, with the monomers being covalently linked by two symmetrical disulphide bonds between Cys-51 and Cys-60. These cysteine residues are also conserved in the viral VEGFs, some of which have been shown to act as dimers (Meyer et al., 1999 ; Ogawa et al., 1998 ; Wise et al., 1999 ). The viral VEGFs were therefore predicted to dimerize in a similar fashion to VEGF-A and modelling revealed dimeric structures very similar to that of VEGF-A (Fig. 5).

(44K): Fig. 5. Ribbon representations of the predicted structures of dimers of VEGF-A, ORFVNZ2VEGF (NZ2) and ORFVNZ7VEGF (NZ7). The sequences used in the model constructions were the VHDs listed in Fig. 3. Models were derived as described in Methods. One monomer of each dimer is shown in a darker shade.

Formation of the VEGF-A dimer creates receptor-binding sites at opposite poles of the dimer and it is believed that each dimer binds two receptor molecules (Keyt et al., 1996 ; Muller et al., 1997a ; Wiesmann et al., 1997 ). The binding sites for VEGFR-1 and VEGFR-2 appear to overlap and each extends across the subunit interface. Fig. 6 shows surface rendering representations of VEGF-A, ORFV_NZ2VEGF and ORFV_NZ7VEGF in two orientations. Residues which make up the receptor-binding face are shaded in a dark grey. This face is built from strands β2, β5 and β6 from one monomer and helix 1, the β3-β4 loop and the β7 strand from the other monomer (Muller et al., 1997a ; Wiesmann et al., 1997 ). In addition, some residues of VEGF-A are coloured to indicate that they have been implicated in mediating the binding of VEGF-A to VEGFR-1 (yellow), VEGFR-2 (blue) or both receptors (green) (Keyt et al., 1996 ; Li et al., 2000 ; Muller et al., 1997b ). Where these residues are conserved in ORFV_NZ2VEGF or ORFV_NZ7VEGF they are coloured in the same manner. The models shown in Fig. 6 confirm the structural similarities between the viral VEGFs and VEGF-A, including the location of the receptor-binding face. Furthermore, where residues implicated in VEGFR-binding are conserved in the viral factors, they occupy similar three-dimensional locations. However, the rather small number of these residues conserved in the viral VEGFs is clearly illustrated (Fig. 6).

(93K): Fig. 6. Surface rendering of VHD domains of dimeric VEGF-A, ORFVNZ2VEGF (NZ2) and ORFVNZ7VEGF (NZ7) showing the predicted locations of amino acids implicated in receptor binding. Residues of VEGF-A are coloured to indicate that they have been implicated in mediating the binding of VEGF-A to VEGFR-1 (yellow, Q22, D63, L66 and E67), VEGFR-2 (blue, I43, I46, Q79, I83 and P85) or both receptors (green, F17, Y21 and E64). Where these residues are conserved in ORFVNZ2VEGF or ORFVNZ7VEGF (see Fig. 3) they are coloured in the same manner. Residues which make up the receptor-binding face are shaded in a dark grey. The dimers are shown in two orientations, related by a rotation of 90° about the horizontal axis. The top row shows the membrane-facing side of the dimers and the bottom row shows end-on views. The sequences used in the model constructions were the VHDs listed in Fig. 3. Models were derived as described in Methods.

Receptor-binding groove
The extracellular portion of VEGFR-1 consists of seven immunoglobulin (Ig)-like domains. It has been determined that the second and third Ig-like domains bind VEGF-A and that the fourth domain promotes receptor dimerization (Wiesmann et al., 1997 ). It has been proposed that domain 2 of VEGFR-1 binds to the receptor-binding end of VEGF-A while the region linking domains 2 and 3 of the receptor occupies a 6·5 wide groove between the VEGF-A monomers and thereby positions domain 3 in contact with the bottom face of VEGF-A (Wiesmann et al., 1997 ). The grooves at each end of the VEGF-A dimer connect the ends of the dimer to the flat membrane-facing side (bottom) of VEGF-A. The walls of the groove in VEGF-A are formed by Asp-63 and Glu-64 on one side and Ile-43, Ile-46 and Phe-36 on the other side, while the floor is formed by Asp-34 and Ser-50 (Iyer et al., 2001 ; Wiesmann et al., 1997 ). Fig. 7 shows a surface rendering representation of VEGF-A in which the residues forming the surfaces of the groove have been shaded. The expanded view shows the groove more clearly with one side of the groove represented by Glu-64, the other side by Ile-46 and the floor by Ser-50. Matched representations of ORFV_NZ2VEGF and ORFV_NZ7VEGF are also shown. In the viral VEGFs, residues occupying the same predicted locations (see Fig. 3) as those forming the walls of the groove in VEGF-A are also shaded and marked. It can be seen that the strong interaction between residues Glu-74 and Arg-56 of ORFV_NZ2VEGF appears to block the opening to the groove. This effect is less apparent between the equivalent residues (Asp-81 and Gln-63) of ORFV_NZ7VEGF but the presence of Arg-67 in place of Ser-50 of VEGF-A has the effect of raising the floor and filling the groove. This effect can also be seen in the end-on view of ORFV_NZ2VEGF shown in Fig. 6. It is intriguing that in the correlations discussed above the presence of Ser-50 was unusual in that it showed a strong correlation with the ability to recognize VEGFR-1 (Fig. 3). These models raise the possibility that the inability of the viral VEGFs to bind VEGFR-1 is related to their lack of a groove able to bind effectively the domain 23 linker region of the receptor. Structural analysis of PlGF, which recognizes only VEGFR-1, has shown that the groove seen in VEGF-A is conserved in PlGF (Iyer et al., 2001 ).

(106K): Fig. 7. Surface rendering of VHD domains of dimeric VEGF-A, ORFVNZ2VEGF (NZ2) and ORFVNZ7VEGF (NZ7) showing the predicted locations of residues forming a possible receptor-binding groove. The walls of the groove in VEGF-A are formed by Asp-63 and Glu-64 on one side and Ile-43, Ile-46 and Phe-36 on the other side, while the floor is formed by Asp-34 and Ser-50. These residues and the corresponding residues of ORFVNZ2VEGF and ORFVNZ7VEGF (see alignment in Fig. 3) are shaded dark grey. In the expanded view of one end of the VEGF-A dimer, three residues of the groove are labelled and the possible location of the VEGR-1 domain 23 linker is shown by an arrow. The expanded views of viral VEGFs suggest that in these dimers the groove is blocked by interaction between residues Glu-74 (E) and Arg-56 (R) of ORFVNZ2VEGF or by the presence of Arg-67 in the floor of the groove of ORFVNZ7VEGF.

Concluding remarks
The evolutionary significance of the variations seen in the sequence of the NZ2-like VEGFs is unknown. The lower G+C content of the viral genes, compared with the overall G+C content of the genome of Orf virus and particularly of sequences flanking the VEGF gene, has been suggested to indicate a relatively recent acquisition by the virus of this gene from a mammalian host (Lyttle et al., 1994 ). The sequence variations reported here might reflect genetic drift of the recently acquired gene although the rate of drift seems greater than generally seen in poxvirus genes. This explanation does not account for the very substantial difference between the NZ2-like and the NZ7-like VEGFs. It is possible that the NZ7-like VEGF was acquired by Orf virus independently of the NZ2 acquisition event and from a different source. An as yet unidentified member of the mammalian VEGF family is one possible source. An alternative source is another parapoxvirus. We have preliminary evidence of VEGF-like activity expressed by cells infected with other species of parapoxvirus (unpublished).

Recombination during co-infection by Orf virus and another species of parapoxvirus might have resulted in the transfer of the NZ7-like VEGF into Orf virus. The minimal variation seen in NZ7-like sequences might suggest that such an event happened recently. Despite the surprising extent of sequence variation among the viral VEGFs, key motifs of structural and functional importance are conserved. These include the eight cysteines of the cystine-knot motif and potential glycosylation sites. A sequence motif rich in threonine and proline seen in the C terminus of the VEGFs encoded by both NZ2 and NZ7 is unique to the viral VEGFs (Lyttle et al., 1994 ) and is retained in all isolates. Determining the functional significance of the sequence variations reported here will require further examination.

This collection of natural variants of VEGF provides opportunities to examine the importance of specific sequence motifs in the various activities of VEGF. The precise nature of the interaction between the viral VEGFs and VEGFR-2 is clearly of interest. It is intriguing that, despite the lack of retention of residues implicated as being critical in the binding of VEGF-A to VEGFR-2 and the lack of conservation of residues between NZ2- and NZ7-like viruses, both NZ2- and NZ7-like VEGFs have been shown to bind and activate VEGFR-2 (Meyer et al., 1999 ; Ogawa et al., 1998 ; Wise et al., 1999 ). This indicates that the viral VEGFs binding site(s) may differ, or that considerable variability can be tolerated while still maintaining the ability to bind VEGFR-2. We hypothesize that the failure of viral VEGFs to bind VEGFR-1 may relate to the absence in these factors of a groove able to accept the VEGFR-1 domain 23 linker. Site-directed mutagenesis of the relevant residues will be required to test this hypothesis.

We gratefully acknowledge the skilled technical assistance of Ellena Whelan, the advice of Dugald Hall with the phylogenetic analysis and the guidance of Nathan Hall in conducting computer predictions of the viral VEGF structures. This work was supported in part by funding from the Health Research Council of New Zealand.