Specificity determinants and diversification of the Brassica self-incompatibility pollen ligand

Abstract

Self-incompatibility in crucifers is effected by allele-specific interactions between the highly polymorphic stigmatic S locus receptor kinase (SRK) and its pollen ligand, the S locus cysteine-rich protein (SCR). Here we show that specificity in SCR function is determined by four contiguous amino acids in one variant, indicating that the minimum sequence requirement for gaining a new specificity can be low. We also provide evidence for an extraordinarily high degree of evolutionary flexibility in SCR, whereby SCR can tolerate extensive amino acid changes within the limits of maintaining the same predicted overall structure. This remarkable adaptability suggests a hypothesis for generation of new self-incompatibility specificities by gradual modification of SRK-SCR affinities and, more generally, for functional specialization within families of homologous ligands and receptors.

In the self-incompatibility (SI) response of crucifers, the recognition of self-related pollen by the stigma epidermis is effected through the activity of two tightly linked and highly polymorphic genes encoded by the S locus (1). The S locus receptor kinase gene SRK encodes a single-pass transmembrane serine/threonine kinase (2), which is an integral component of the plasma membrane of the stigma epidermis (3–5) and is displayed with its glycosylated N-terminal S domain external to the cell (6). The S locus cysteine-rich protein gene SCR (7) [also designated SP-11 (8)] encodes a small (<8 kDa) hydrophilic and positively charged peptide that is localized to the pollen coat. Both SRK and SCR exhibit an extraordinarily high degree of DNA and protein sequence variability. Furthermore, the number of S locus variants, or S haplotypes, and consequently of SRK and SCR alleles, is typically large, with as many as 100 S haplotypes occurring in one species (9), and these polymorphisms are estimated to be at least 20–40 million years old (10). These features are consistent with the expectation that new SI variants have a strong reproductive advantage and persist in populations for long periods.

Recent studies have shown that SCR is a ligand for SRK (11, 12) and that specificity in the SI response is determined by allele-specific interactions between the SRK receptor and the SCR ligand (11). In a self-pollination, the SCR protein delivered to the stigma binds to the ectodomain of its cognate “self” SRK and activates the SRK kinase. In turn, this activation presumably triggers a signal transduction pathway that results in the inhibition of pollen hydration, germination, and pollen tube penetration into the stigma epidermal cell wall. In contrast, in a cross-pollination, pollen tube growth proceeds unimpeded because the SCR variant that is delivered to the stigma by the pollen coat cannot bind nor activate “non-self” SRK. An implication of this model is that the SRK and SCR genes must coevolve to maintain their interaction. Thus, in this two-gene system, the generation of a novel SI specificity requires the occurrence of compensatory mutations in receptor and ligand encoded by the same S haplotype such that SRK–SCR binding is maintained. How this occurs is not known, however.

In a previous study of the S₆ and S₁₃ haplotypes of Brassica oleracea, we developed methods for expressing substantial amounts of SCR and of the SRK ectodomain (eSRK), for assaying their binding affinities and for testing the biological activity of SCR variants in a pollination bioassay (11). We demonstrated that SCR interacts directly with eSRK in vitro and that this interaction does not require additional molecules from the stigma or pollen. In this article, we describe our use of ligands and receptors derived from the S₆ and S₁₃ haplotypes for structure–function studies of SCR. SCR₆ sequence variants were generated by the swapping of specific domains between SCR₆ and SCR₁₃ variants and by in vitro mutagenesis. The primary goal of these studies was to map sequences that determine recognition specificity in SCR, and the secondary goal was to assess if residues that are largely conserved in SCRs are important for SCR₆ function. The implications of our results for hypotheses relating to the evolution of new SI specificities are discussed.

Materials and Methods

Modeling of SCR Structures. The procedure used to generate the 3D models of the SCR molecules involved two main steps. First, an appropriate 3D structure was identified from the Protein Data Bank to be used as a template, and the best possible alignment of the SCR sequence to that of the template was determined. Second, a 3D model for the SCR sequence was constructed and based on the template structure.

The first step was carried out by using the threading program (13). The 3d-pssm server utilizes a library of known protein structures onto each of which the query sequence is “threaded.” The information provided by 3d-pssm is subsequently analyzed, and, if the scores are significant, the selected template structures are retrieved from the Protein Data Bank and the automatic alignment provided by 3d-pssm is carefully analyzed. For each template, both the template structure and alignment are used as input for the program modeller (14), which generates the 3D model for our query sequence through the satisfaction of spatial restraints. modeller is a program commonly used in comparative protein modeling, which minimizes the violations of distance and dihedral-angle restraints derived from the template structure. The 3D models generated through modeller are then analyzed visually to determine their quality. Models with atomic overlaps or unreasonable loop topologies are discarded. Visual inspection is also used to improve the original alignment between the query and template sequences. In the SCR protein, in particular, the original alignments provided by 3d-pssm were usually altered to allow for the formation of the four S—S bridges known to form in SCRs (12).

Models for SCR₆ and SCR₁₃ were generated by using the structures of two plant defensins Rs-AFP1 (PDB ID code 1AYJ) and AH-AMP1 (PDB ID code 1BK8), respectively, as templates. Both templates are structurally similar and superimpose with a rms deviation of 1.7 Å for 47 (of 50) Cα atoms. After optimization, our final models deviate little from their respective templates with <2-Å rms deviation for 45 (of 50) Cα atoms used in the alignments. The NMR structure of SP11–8 (SCR₈) has been recently reported (15) and is similar to Rs-AFP1, with the Cα atoms of residues in the regions of secondary structure elements superimposing with an rms deviation of 1.86 Å.

Mutagenesis and Domain Swapping. SCRs and the ectodomains of SRKs were produced and purified as described (11). Recombinant SCR₆ and SCR₁₃ (minus the signal peptide) carrying a C-terminal myc-His₆ tag were targeted to the periplasm in Escherichia coli by using the pBAD/gIIIB vector (Invitrogen) and purified on Ni-NTA affinity agarose (Qiagen, Valencia, CA). The SRK₆ and SRK₁₃ ectodomains carrying a C-terminal FLAG epitope were expressed in Nicotiana benthamiana leaves by using the potato virus X expression system and purified on anti-FLAG M2 affinity gel (Sigma). Mutated residues were introduced with the QuikChange site-directed mutagenesis kit (Stratagene) according to the manufacturer's directions and with oligonucleotides synthesized by Integrated DNA Technologies (Coralville, IA). Only a few mutants, namely R23A, D24A, and R43A, were poorly expressed and were not used in the tests.

SCR chimeras were produced by swapping appropriate domains between SCR₆ and SCR₁₃. In all, the following chimeras were assayed: SCR₆ (3–4 and 5–6) [i.e., SCR₆ backbone with the C3–C4 and C5–C6 regions (see Fig. 1A) derived from SCR₁₃], SCR₆ (5–6), SCR₆(TDTQ), SCR₁₃ (2–3), SCR₁₃ (3–4), SCR₁₃ (5–6), SCR₁₃ (3–4 and 5–6), SCR₁₃ (2–3 and 3–4), SCR₁₃ (2–3 and 5–6), and SCR₁₃ (2–3, 3–4, and 5–6). Chimeras were produced by using the PCR, or by replacing the sequences by QuikChange, as in SCR₁₃ (TDTQ) and the SCR₁₃ chimeras containing the C2–C3 domain of SCR₆, which was introduced into SCR₁₃ by converting the “RNI” sequence to “GDS” (see Fig. 1A). The SCR₁₃ (5–6) chimera retained the Cys-5TDTQM sequence to maintain the length of the loop in this region, although its activity on S₆S₆ stigmas was similar to that of an SCR₁₃ (5–6) chimera lacking this additional sequence.

Fig. 1.

Sequence divergence and predicted structural conservation of SCR variants. (A) Sequence alignment of predicted mature SCR variants derived from S haplotypes of B. oleracea (Bo) and Brassica campestris (Bc). Residues that are conserved in most SCR variants are indicated, and the domains that were mutagenized or exchanged between SCR₆ and SCR₁₃ are shown below the alignment. (B) Threaded structures of SCRs from the B. oleracea S₆ and S₁₃ haplotypes. The conserved cysteines (pink), N/C termini (blue), and disulfide bonds (stick representation with sulfur atoms colored yellow) are shown. The structures in the top row show theα-helix (red), the β-sheet (blue), the C3–C4 region (green), and the C5–C6 region (purple). In the second row, the SCR₆ structure shows Y26 (red) and residues essential for SCR₆ function (green). To the right are two views of the SCR₁₃ model showing the SCR₁₃-specificity-determining TDTQ residues (red).

Analysis of SRK–SCR Interactions. SRK–SCR interactions were detected by using ELISA and pull-down assays as described (11), with at least two different preparations of each SCR variant and three replicates of each. For ELISAs, microtiter plate wells were coated with 0.5 μg of eSRK₆ and incubated with increasing amounts of SCR₆-myc-His₆ or SCR₁₃-myc-His₆. After incubation with anti-myc antibodies and alkaline phosphatase-conjugated secondary antibody (Sigma), the reaction was developed with 100 μl of Sigma Fast pNPP substrate, and absorbance was measured at 405 nm in an ELISA plate reader. Averages of the binding maximum (B_max) values obtained for an SCR variant were calculated and used to derive B_max mutant/wild-type ratios.

For pull-down assays and immunoblot analysis of SRK–SCR interactions, eSRK-FLAG was immobilized on anti-FLAG M2 affinity agarose and incubated with increasing amounts of SCR-myc-His₆. The bead–protein complexes were resolved on 15% (wt/vol) SDS/PAGE and detected with anti-myc antibodies (Invitrogen). Protein bands were visualized with horseradish peroxidase-conjugated secondary antibodies (Sigma) and the ECL+ system (Amersham Pharmacia Biotech).

SRK-containing microsomal fractions were prepared from stigmas as follows. Stigmas were ground in a buffer consisting of 30 mM Tris·HCl (pH 7.5)/75 mM NaCl/10 mM EDTA/10% (vol/vol) glycerol, and containing a protease inhibitor mixture (Sigma). The extract was centrifuged at 4,000 × g for 3 min, and the supernatant was then spun in a TLA100.4 centrifuge (Beckman Coulter) at 136,000 × g for 1 h. The microsomal pellet was resuspended in 50 mM Tris·HCl (pH 8.0)/150 mM NaCl/1% (vol/vol) Nonidet P-40/0.1% (wt/vol) SDS containing a protease inhibitor mixture. Changes in affinities obtained by ELISA were generally consistent with pull-down assays of SRK from stigma microsomal fractions.

Pollination Assays. The pollination bioassay was performed as described (11). In brief, recombinant SCR-myc-His₆ was mixed with pollen-coat protein extracted with cyclohexane from S₂ pollen and then applied on S₆S₆ or S₁₃S₁₃ stigmas. Typically, 300 ng of SCR were applied per stigma, because this amount had been shown to be effective for wild-type SCR to inhibit pollen germination on the stigma (11). The treated stigmas were allowed to dry for ≈1 h and then were cross-pollinated with non-self-pollen. Each assay for activity of SCR mutants included the following controls: wild-type SCR₆ or SCR₁₃, pollen-coat protein alone, no SCR controls, and stigmas expressing an S haplotype other than S₆ or S₁₃ treated with the preparation of SCR used in the experimental samples to verify that the SCR protein was not toxic to pollen. Pollinated stigmas were processed for microscopy and pollen tube growth was observed by UV-fluorescence microscopy (11). In previous applications of this bioassay (11), we had shown that pretreatment of S₆S₆ stigmas with wild-type SCR₆-myc-His₆ caused these stigmas to inhibit the germination of normally compatible (i.e., non-self) pollen, whereas pretreatment of stigmas expressing other S haplotypes with SCR₆-myc-His₆ did not.

Results

Predicted Conservation of Overall Structure Among Highly Diverged SCR Variants. Variants of highly polymorphic molecules typically exhibit conserved regions that reflect common function and highly diverged regions that are often good predictors of the variant specificity determinants. Consequently, sequence comparisons are often a useful guide for structure–function studies of such molecules. In SCR, however, sequence comparisons are of little value for predicting the variant specificity determinants because all regions of the protein are highly diverged. Indeed, SCR variants, with a few exceptions (16), differ in the type and number of amino acids over their entire length (7, 9, 17, 18). As shown in Fig. 1A, only a few residues are conserved among most variants of SCR, namely eight cysteines (hereafter designated Cys-1 through Cys-8), the glycine in the GlyxCys-2 motif (Gly-12 in SCR₆), and an aromatic residue (Tyr-26 in SCR₆) which is part of the Cys-3xxxTyr/Phe motif. Even for these residues, however, conservation is not absolute, because some SCRs lack Cys-8 and others lack the aromatic residue in the Cys-3xxxTyr/Phe motif (e.g., SCR₁₃ in Fig. 1A). Which residues in SCR determine allelic specificity, what features allow this protein to undergo extensive diversification while maintaining its function as a ligand, and how SCR proteins acquire new specificities are intriguing but unresolved questions.

To determine whether SCRs might be constrained to adopt a similar 3D structure, we generated crude structural models of several SCR variants by threading (see Materials and Methods). We found that, despite their extreme sequence divergence, the SCRs we tested can be threaded to fit a similar fold recognition composed of an α-helix and a triple-stranded antiparallel β-sheet, which are forced into a compact tertiary configuration by four intramolecular disulfide bonds (Fig. 1B). The disulfide bridge that links Cys-1 to Cys-8 tethers the N and C termini, whereas the bridges that link Cys-2 to Cys-5, Cys-3 to Cys-6, and Cys-4 to Cys-7 appear to form the hydrophobic inner core of the molecule. This structure, which has features characteristic of the cysteine-stabilized αβ (CSαβ) motif found in defensins (19), is in agreement with the structure recently determined for the SCR allele derived from the S₈ haplotype (ref. 15; Materials and Methods), suggesting that the SCR-threaded models we generated, although requiring experimental verification, are robust representations of actual SCR structures.

Identification of SCR Specificity Determinants by Domain Swapping. The SCR₆ and SCR₁₃ structural models (Fig. 1B) and previous surface probability predictions (17) point to the regions between Cys-3 and Cys-4 (C3–C4; Fig. 1A) and between Cys-5 and Cys-6 (C5–C6; Fig. 1A) as candidate specificity determinants, because they are predicted to be surface-exposed and available for interaction with SRK. To test this prediction, we generated SCR₆-SCR₁₃ chimeras in which various domains were exchanged either singly or in combination between the two SCR variants. Chimeras are designated by the SCR backbone followed by the exchanged region in parentheses [e.g., an SCR₆ protein in which the C3–C4 region is derived from SCR₁₃ is designated SCR₆ (3–4)]. The chimeras were expressed and targeted to the E. coli periplasm as efficiently as wild-type SCR₆ and SCR₁₃. One exception was the SCR₆ (3–4) chimera, only a small proportion of which was targeted to the periplasmic fraction, possibly reflecting poor protein folding; this chimera was not analyzed further. The chimeras were purified and tested (11) for binding to recombinant SRK₆ and SRK₁₃ ectodomains (designated eSRKs) and for their ability to elicit SI on S₆S₆ and S₁₃S₁₃ stigmas. Relative binding affinities for eSRKs were estimated from the ratio between maximal binding of chimeric and wild-type SCR (Fig. 2A), because most SCR chimeras differed from one another and from wild type in binding maximum (B_max) (see Materials and Methods).

Fig. 2.

Analysis of SCR chimeras. (A and B) Activity of SCR₆ (A) and SCR₁₃ (B) chimeras. Binding to 0.5 μg of eSRK₆-FLAG or eSRK₁₃-FLAG immobilized in microtiter wells was measured by ELISA (11). A B_max ratio of 1 between chimera and wild type represents an affinity identical with that of the wild-type protein. Asterisks indicate the binding of wild-type SCR₆ to eSRK₆ and of wild-type SCR₁₃ to eSRK₁₃ where values equal 1 and are therefore not visible on the graph. SCR activity in pollination assays is shown below the graph: +, SI is activated; –, SI is not activated. (C) Protein blot analysis of the interaction between the SCR₆ and SCR₁₃ chimeras and eSRK₆ and eSRK₁₃. Seven hundred fifty nanograms of eSRK₆-FLAG (lanes 2–5) or eSRK₁₃-FLAG (lanes 6–9) were immobilized on FLAG M2 affinity agarose and incubated with increasing amounts of wild-type or chimeric SCR (lanes 3 and 7, 50 ng; lanes 4 and 8, 100 ng; and lanes 5 and 9, 500 ng). Immunoblots of bead–protein complexes were treated with anti-myc antibodies to detect bound SCR. The positive controls in lanes 1 consisted of 1μg of wild-type SCR₆-myc-His₆ (upper three gels) or SCR₁₃-myc-His₆ proteins (lower four gels) and did not bind anti-FLAG agarose (lanes 2).

Among SCR₆ chimeras, only the SCR₆ (5–6) chimera exhibited modified specificity; it was no longer active on S₆S₆ stigmas but instead activated SI on S₁₃S₁₃ stigmas (Fig. 2A). Thus, replacement of the C5–C6 VPTGR sequence in SCR₆ with the C5–C6 TDTQMGTYS sequence from SCR₁₃ (Fig. 1A) is sufficient to confer S₁₃ specificity on SCR₆. Even more dramatically, replacement of the VPTGR sequence in SCR₆ with just the first four residues (TDTQ) from the SCR₁₃ C5–C6 region was sufficient to convert SCR₆ into a functional SCR₁₃ protein that bound eSRK₁₃ efficiently (Fig. 2 A and C) and activated SI on S₁₃S₁₃ stigmas (Fig. 2A). The SCR₆ (5–6) and SCR₆(TDTQ) chimeras were able to bind both eSRK₆ and eSRK₁₃, even though they activated the SI response only on S₁₃S₁₃ stigmas (Fig. 2A). Thus, the binding of SCR to eSRK can be uncoupled from its ability to activate SI.

In contrast to SCR₁₃, the specificity determinants of SCR₆ remain elusive. We were unable to incorporate SCR₆ specificity into SCR₁₃ by replacing, either individually or in pairwise combinations, the C3–C4 and C5–C6 regions (Fig. 2B) or the C2–C3 region (data not shown) in SCR₁₃ with the corresponding regions from SCR₆. Although these chimeras exhibited reduced affinity for eSRK₁₃ relative to wild-type SCR₁₃ and improved affinity for eSRK₆ relative to wild-type SCR₁₃ (Fig. 2 B and C), none of them gained the ability to activate the SI response on S₆ stigmas. Only a swap of the C2–C3, C3–C4, and C5–C6 regions together, which largely reconstituted SCR₆, produced a chimera that exhibited S₆ specificity (data not shown). Replacement of either the C3–C4 region alone or the C5–C6 region alone with the corresponding regions from SCR₆ had no effect on the ability of SCR₁₃ to activate the SI response on S₁₃ stigmas (Fig. 2 B and C).

Alanine-Scanning Mutagenesis of SCR. The fact that the four specificity-determining residues of SCR₁₃ can function in the context of the very different SCR₆ backbone suggests that mutations in much of the SCR molecule might be tolerated. Undoubtedly, however, constraints must exist for the SCR protein to maintain its overall shape and function as an activator of the SI response. The existence of these constraints is evident from the conservation, albeit not absolute, of some residues in SCRs (Fig. 1A). To assess how flexible the SCR protein is, we generated variants of SCR₆ by alanine-scanning mutagenesis of the C3–C4 and C5–C6 regions and by substituting the residues that are conserved in most SCRs with alanine or valine.

Several mutations in SCR₆ either modestly improved or had little effect on its affinity for eSRK₆, with no disruption of its ability to elicit SI on S₆S₆ stigmas (Fig. 3A). Only mutations of Tyr-32, Met-33, and Ser-36 in C3–C4, and Thr-42 and Gly-43 in C5–C6 were disruptive (Fig. 3A). These residues, which are located in a surface-exposed loop in the SCR₆-threaded model (Fig. 1C), might contribute to the eSRK₆-binding site. Similarly, and as expected for mutations that likely cause global structural perturbations by disrupting disulfide bonds, mutations of the conserved cysteines abolished activity of SCR₆, even when they did not affect or slightly improved affinity for eSRK₆ (Fig. 3B). Surprisingly, however, replacement of the conserved Gly-12 residue with valine was tolerated (Fig. 3B). Thus, this glycine residue is not required for SCR₆ function, and its conservation might reflect a role in synthesis or stability of SCR in planta. In contrast, the Y26A mutant failed to activate SI on S₆S₆ stigmas, even though its binding affinity for eSRK₆ was three times higher than wild-type SCR₆ (Fig. 3B), and its ability to pull down native SRK₆ from microsomal fractions of S₆S₆ stigmas was similar to wild-type SCR₆ (data not shown). This behavior is yet another demonstration of the uncoupling of SCR binding to eSRK from its ability to activate the SI response. In any case, the Tyr-26 residue is clearly required for SCR₆ function, and high-resolution structural analysis might explain why this aromatic residue is dispensable in SCR₁₃ and at least two other SCR variants.

Fig. 3.

Mutagenesis of SCR₆. (A) Alanine-scanning mutants of the C3–C4 and C5–C6 regions of SCR₆. The R23A, D24A, and R44A mutants were expressed poorly and were not analyzed. Results show B_max ratios for binding of the SCR mutants to eSRK₆ relative to wild-type SCR₆. (B) Contribution of the conserved cysteine residues, Gly-12, and Tyr-26 to SCR₆ function. Evaluation of the mutants for binding to eSRK in vitro and for activity in vivo was as described in Fig. 2 A and B.

Discussion

Despite the high sequence divergence between SCR₆ and SCR₁₃, we were able to engineer a variant of SCR₆ that binds tightly to eSRK₁₃ and activates SI on S₁₃S₁₃ stigmas. This change in specificity was achieved by tailoring the C5–C6 region in SCR₆ to be like SCR₁₃ and, in fact, required alteration of only four residues within this region. The TDTQ residues that impart SCR₁₃ specificity on SCR₆ are located in a surface-exposed loop in the SCR₁₃ structural model (Fig. 1C) and are thus likely to make contact with SRK₁₃, or they might cause structural changes that indirectly allow a productive interaction with eSRK₁₃. In any case, the finding that only a few contiguous residues are sufficient to determine SCR₁₃ specificity is consistent with results obtained for other recognition molecules such as plant disease resistance genes (20), the stylar S-RNase determinant of SI in solanaceous plants (21), and the mating-type genes of Basidiomycetes (22). However, the arrangement of specificity determinants is likely to vary significantly between SCR variants. Indeed, our inability to incorporate SCR₆ specificity into SCR₁₃, except in a chimera that largely reconstituted the SCR₆ protein, suggests that SCR₆ specificity might be determined by noncontiguous residues or segments. Alternatively, residues other than the specificity-determining residues might influence binding to SRK in this variant.

Although SCR specificity can be determined by a limited number of residues, many other residues are required for the interaction of SCR with eSRK and for activation of the SI response, as determined by the disruptive effect of mutating these residues. In our assays, binding of SCR to eSRK in vitro did not always translate into a capacity to activate SRK in vivo. For example, the Y26A mutant showed a significant increase in binding affinity for eSRK₆, but it did not activate the SI response in stigmas. Similarly, the SCR₁₃ (3–4 and 5–6) chimera bound SRK₆ with an affinity approaching that of wild-type SCR₆, yet it did not acquire the capacity to activate SRK₆. This lack of correlation between in vitro binding and in vivo activation cannot be dismissed as an artifact of our in vitro binding assays, because, when tested, the results of these binding assays usually correlated with binding of SCR to SRK from stigma microsomes. Instead, it appears that the capacity for SRK activation is more sensitive to changes in overall SCR structure than binding. It is possible that small stretches of residues contribute to overall binding affinity, whereas the extensive conformational changes usually associated with receptor activation might require multiple contact points between SRK and SCR, elimination of any one of which would result in a nonproductive interaction. A similar lack of correlation between binding and signal generated was also observed in studies of other receptor–ligand systems (23). In these studies, the initial docking of the ligand and the stable binding that actually results in receptor activation were found to be distinct events, with receptor activation dependent on the tightness and duration of binding.

Flexibility of the SCR Molecule. As in any polymorphic system, extant SCR variants must represent an evolutionary compromise to accommodate the drive for evolutionary novelty and the counterbalancing pressure to maintain function. An unusual feature of SCR lies in the apparently low sequence requirements for this protein to achieve a functional interaction with SRK. Our analysis of SCR mutants and chimeras demonstrated that significant changes can be made to the SCR protein with minimal effect. First, up to ≈30% of the SCR₁₃ protein could be replaced, as in the SCR₁₃ (3–4) chimera, without compromising its ability to activate the SI response on S₁₃ stigmas. Second, although several mutations disrupted SCR function, other relatively drastic modifications of SCR were tolerated with only minor effects on binding affinity and without adverse effect on signaling capacity. Finally, the recruitment of receptor-binding and activation properties from highly diverged SCRs by modification of only a few residues indicates that these residues can effect their recognition function within the context of highly diverged SCR backbones.

This remarkable evolutionary adaptability or “evolvability” (24) of SCR undoubtedly derives from the inherent versatility of the CSαβ motif, a scaffold apparently shared by SCR variants and found in a variety of proteins with little or no primary sequence similarity (19) and diverse activities, such as defense against fungi and bacteria, inhibition of α-amylase, disruption of membranes, and induction of signaling pathways (19). In SCR, this motif would allow for the adoption of a wide range of specificities within the framework of a similar protein fold.

Implications for Evolution of New SI Specificities. The significant flexibility of the SCR molecule revealed by our study has broad implications for the evolution of SI specificities in natural populations. Because little constraint is apparent on the SCR primary amino acid sequence within the limits of maintaining the same overall shape, the susceptibility of the SRK–SCR interaction to disruption by random mutation is likely to be low. These features would allow the persistence of random mutations as long as they retain capacity to bind and activate the appropriate allele of SRK and thus would lead to rapid sequence diversification of SCRs within each functional allelic class. We infer that in nature, mutations in SCR at many residues produced small changes in affinity for SRK rather than loss of the interaction, as observed in some laboratory-induced SCR mutants and chimeras described here. It is also likely that, among naturally occurring mutations, the generation of SCR variants with novel specificities required only a few critical nonsynonymous substitutions.

Attempts to explain how new specificities arise in a two-component SI system such as the SRK–SCR system of crucifers are faced with an intriguing and difficult problem. Two alternative schemes have been proposed. In one scheme, this diversification process occurs by means of self-compatible intermediates, with a change occurring in one gene that disrupts SI, followed by a compensatory mutation in the second gene that restores SI (25, 26). Such a scenario is problematic, however, because nonfunctional S haplotypes are unlikely to be maintained along with functional haplotypes within a population. They might be eliminated as a consequence of inbreeding depression and selection against self-fertilization, or they might persist in a population and drive to extinction the functional haplotypes from which they arise (27). The alternative scheme for the generation of a novel SI specificity by formation of dual-specificity intermediates (21) is equally problematic (25, 26).

On the basis of the apparent evolutionary adaptability of SCR and the inferred flexibility of the SRK–SCR physical interaction, we propose the hypothesis that new SI specificities are generated through self-incompatible intermediates by a process that changes both pollen (SCR) and stigma (SRK) components of an S locus haplotype but preserves allelic recognition. This hypothesis appears to be fully compatible with the data, it accounts for the observed variability within functional S haplotypic classes (28), and it avoids the problematic assumptions of self-compatible or dual-specificity intermediates that underlie other models. We propose that S locus diversification involved minor readjustments in the SRK/SCR interface without adverse effects on the signaling outcome, similar to the subtle changes in affinity that are thought to underlie functional diversification of isoforms within families of receptors and ligands in animal systems (29). Newly arising mutations in SCR may be favored if they either increase the activation of the appropriate SRK allele, or if they reduce the spurious activation of any of the mismatching SRK alleles. Specificity is maintained by this combination of positive and negative determinants of binding.

To understand the process whereby new functional allelic classes can arise without producing a self-compatible intermediate, consider the variability within each functional allelic class of SRK/SCR as a cluster (large circles in Fig. 4A) of slightly varying sequences or suballeles (dots and dashes within the circles in Fig. 4). These circles do not overlap, consistent with the observation that each SRK/SCR variant does not overlap in its recognition with any other variant. Variability within SCR and SRK can result in variability in binding affinity, and occasionally subsets of allelic sequences may arise within a functional haplotype that show some mutual partitioning of affinity (Fig. 4B). As soon as the 2* haplotype begins to show greater affinity between SCR_2* and SRK_2* than cross affinities with the other class 2 alleles, then variants that optimize the 2 vs. 2 and 2* vs. 2* recognition are favored (Fig. 4C). As the 2* × 2* affinities improve, loss of interaction can occur between 2 and 2* proteins with no loss of self-incompatibility at any stage, and this process ends up producing a novel fourth allele (Fig. 4D). In addition to this passive loss of interaction between SRK₂ and SCR₂, substitutions that are favored at this stage may occur in either SCR or SRK, and they may either reinforce the activation of SRK₂ by SCR₂ or they may be negative in their effects, reducing activation of SRK₂ by SCR₂.

Fig. 4.

Model for the generation of new SI specificities through self-incompatible intermediates. (A) Within each functional haplotype (circles), variability in both SCR and SRK (represented by dots and dashes within the circles) is tolerated, and mutual recognition exists (solid lines connecting the circles). (B–D) If by chance, in a subset of alleles, designated 2*, the SRK and SCR proteins exhibit stronger affinity with each other than with the corresponding proteins from other class 2 alleles, then natural selection will drive the strengthening of the SRK₂–SCR₂ and the SRK_2*–SCR_2* interactions, while at the same time tolerating the weakening of the SRK₂–SCR_2* and SRK_2*–SCR₂ interactions (dashed lines). This process can lead to the origination of a novel allele (designated 4) without producing any self-compatible intermediates.

This process of splitting of the cloud of allelic sequences within a functional haplotype into two may be accelerated in a subdivided population or under conditions when selection is relaxed, such as when the population size is restricted (30), but computer simulations have shown that it proceeds within a single panmictic population over a wide range of population sizes (data not shown). The previously uninvestigated feature of this model is that it allows the variability to be organized into cliques of suballeles whose specificity can subsequently be optimized. Analysis of the structure–function relationships of protein products of different SCR alleles has revealed not only properties that are important for specificity of receptor–ligand interactions, but those same results suggested an evolutionary model that solves a long-standing puzzle by being able to generate novel allelic specificities without a low-fitness, self-compatible intermediate.