Introduction

Gametophytic self-incompatibility (GSI) plays a key role in the genetic diversity within plants while changing patterns of lineage diversification in clades that employ these mechanisms1,2,3,4. GSI involves the participation of both factors from the pollen and pistil5, which play an important role in pollen recognition and rejection6. One of the important features of the GSI is the existence of an extracellular ribonuclease in the pistil, the SRNase (the female determinant of SI), and its involvement in the rejection of incompatible pollen cells7. The Rosaceae SRNase contains a single hypervariable region (RHV) that is the most hydrophilic part of the SRNase, suggesting that it could be the prime determinant of the SRNase specificity8. The high level of allelic polymorphism of the SLF/SFB gene, along with its pollen-specific expression and close physical distance to the SRNase gene in the S-locus supports the hypothesis that SLF/SFB is the male determinant of GSI9,10. Despite the identification of F-box genes as pollen-S candidates the mechanism by which SLF/SFB and SRNase interact to trigger the self-incompatibility reaction remains uncertain. Several studies have suggested divergence of pollen-S function among different taxa with some having a self-recognition system with a single factor while others represent a non-self recognition system with multiple factors11,12,13,14,15,16,17,18,19. The increase in the number of SLF genes implies that a larger repertoire of SRNases can be targeted, thus allowing more potential coupling partners, while the expansion in the number of SRNases will have the opposite effect. Accordingly, for a plant species it would be advantageous to have multiple SLF genes as this helps prevent inbreeding and provide diversity in plants.

The growing economic importance of fruit crops and the problems associated with their commercial production has placed the molecular basis of self-incompatibility under intense attention. This has led to the rise of many hypotheses about how the pistil S-ribonuclease (SRNase) and pollen-expressed S-haplotype-specific F-box (SFB) interact. Kubo and colleagues (2010) proposed a Collaborative Non-Self Recognition model for the interaction of SLF variants (brothers) with different SRNases. This model assumes that the pollen S-locus encodes several SLF proteins and different SLF variants can recognize a subset of SRNases20. It is therefore of great interest to evaluate and study this model in various species, as it provides clues about the mechanism of GSI employed and further assists in formulating future studies, especially those focused on apple due to its economic significance in worldwide agriculture. It is notable that successful breeding programs rely on the availability of diverse germplasm, which poses the need to develop and introduce new cultivars. Additionally, the GSI mechanism represents an obstacle for the breeding of apple by restricting crosses between apple cultivars. As such the knowledge of GSI as to how the SLF/SFB and SRNase proteins interact, can provide useful information to guide the implementation of breeding strategies.

To address this issue a model of interaction between S-locus genes is introduced by assessing the Collaborative Non-Self Recognition model in apple. This model aims at describing the specificities of SLF/SFB and their brothers (SFBBs) in recognizing a large repertoire of SRNases, using molecular modeling and docking techniques.

Results and Discussion

With the aim of inspecting the Collaborative Non-Self Recognition model in apple, Malus × domestica (Borkh.), the tertiary structure of known SRNases, along with SLFB3 and SLFB9, as well as their relevant brothers namely SFBB3α, SFBB3β and SFBB9α and SFBB9β, were predicted using the I-TASSER suite21 (Table S1-S7 and Fig. S1-S8). The folding topologies of the modeled SRNases structures were examined using the Dali server, which were found to be very similar to the topologies of Momordica charantia ribonuclease MC and MC1 (Table S4). Hence, it is concluded that these predicted structures have good homology with the experimentally solved structures of the RNase T2 family enzymes. These models can therefore be used confidently in a docking study to further assess the Collaborative Non-Self Recognition model in apple. With respect to SLF/SFBs and their relative brothers (SFBBs), the folding topologies of the modeled structures were found to be commonly very similar to the topologies of F-BOX/WD-repeat proteins (Table S3). Therefore, the result of Dali server clearly shows that the template and modeled structures could be part of the same protein family. Previous protein binding studies have shown that SLF interacts with the hypervariable (HV) region of S-RNase22, accordingly in this analysis only the residues within the predicted hypervariable region (Fig. 1) as identified previously by Ashkani and Rees23 were used as the interacting interface.

Figure 1
figure 1

Graphical representations of the location of hypervariable region (HV) on the alignment and secondary structures of Malus (A) SLF/SFBs and (B) SRNases. Highlighted regions on the primary and secondary structure of SLF/SFBs and SRNases are shown the location of HV region while ‘S’ referrers to SRNase. Ancestor sequences of SLF/SFBs and SRNases (i.e. SLFanc and SRNanc) were reconstructed using the Bayesian method as described by Hall35 as described by Ashkani and Rees23. HV regions containing 27 amino acids (from Lys-247 to Cys-273, on loop 19 and a section of β16-sheet) and 24 amino acids (from Ser-40 to Pro-63, on loop 2 and a section of α2-helix) were identified for SLF/SFBs and SRNases based on the ancestor amino acid sequences, respectively.

Full size image

ZDOCK24,25,26 was used to predict SLF and SRNase interactions in S-locus while the docking results were narrowed down to those predicted complexes that have shown a significant interaction based on the Wilcoxon rank-sum test (p-value < 0.0001) comparing binding energies (ZRANK scores). The null hypothesis for this analysis was based on the finding of Hua and Kao (2006) who suggested that the non-self interactions are much stronger than the self-interactions22. The accuracy of the predicted complexes by docking was further assessed using the root mean square deviation (RMSD) between the atoms of the predicted complex and the native or near-native complex, the model with the best binding energy (Figs S9 and S10).

The results of the Wilcoxon rank-sum test (Table 1) shows that SFBB3β from the S3-haplotype interacts more strongly with S10- and S25-RNase compared to the self S3-RNase while SLFB3 and SFBB3α are only strongly interacting with S10-RNase. With regards to SLFB9 and its brothers (SFBB9α and SFBB9β), they all commonly interact with S3-, S8-, S10-, S25- and S30-RNase. In addition, SLFB9 and SFBB9β commonly interact with S7- and S26-RNase while SFBB9α and SFBB9β commonly interact with S4-RNase (Fig. 2).

Table 1 The predicted binding energies for the interacted SLF/SFBs and SRNases.
Full size table
Figure 2
figure 2

Graphical representation of the introduced model of interaction between Malus S-locus proteins.

The figure was generated based on the results of Wilcoxon rank-sum test33. SLFB3 and SLFB9, and their brothers (SFBB3α, SFBB3β, SFBB9α, SFBB9β) are all predicted to interact with S10-RNase. In addition SFBB3β interact with S25 specifically. In terms of SLFB9 and its brothers (SFBB9α, SFBB9β), they all interact with S3-, S8-, S25- and S30-RNase with different binding affinities. However, SLFB9 and SFBB9β interact with S7- and S26-RNase, and SFBB9α and SFBB9β interact with S4-RNase. ‘S’ referrers to SRNase.

Full size image

Therefore, it is suggested that the self/non-self recognition system in apple is controlled by multiple factors where SLF/SFBs interact with non-self SRNases with different binding specificities and thus proteins originally identified as SLF-like (SLF/SFB brothers), may indeed be true SLF/SFB proteins in carrying out the process of self/non-self recognition. Consequently, the presence of extra subsets of SLF/SFB proteins in the S-locus, which can interact with certain subset of SRNases is expected while additional SLF proteins in S-locus have yet to be characterised in Malus. In line with this hypothesis, Minamikawa and colleagues18, and Okada and co-workers27 reported additional SFBB-like genes/alleles in apple, and it is likely that more SFBB genes remain to be identified. However, recently multiple male factors in pear (Pyrus pyrifolia) have also been identified28. In addition, common binding of a specific subset of SRNases to another subset of SLF/SFBs, are consistent with the scenario that a large repertoire of non-self SRNases are targeted and detoxified by multiple SLF/SFBs each of which recognizes a sub-fraction of SRNases28,29. These findings are in line with the Collaborative Non-Self Recognition model that the loss-of-function of one of multiple SLF/SFBs leads to a limited effect on the GSI phenotype.

Though our findings pertaining to the computational prediction of SLF/SFB and SRNase mechanism of interaction hold great value for future breeding studies, its further validation through direct protein-protein interaction studies is needed.

Conclusion

It is known that the specificity of the self-incompatibility response results from S-haplotype-specific molecular interactions of SLF/SFBs and SRNases. As such understanding the recognition mechanism of GSI requires detailed structure-function analysis of the S-locus proteins. Such structure-function studies require an experimental system that allows efficient in vivo functional analysis of large numbers of SLF and SRNase variants generated in vitro by site-directed mutagenesis or domain swapping between proteins that determine different GSI specificities. However, due to the experimental limitations associated with such studies, the computational approach undertaken here represents a powerful alternative. Hence, these findings provide the basis for studying the appearance of multiple subsets of SLF/SFBs and the biochemical basis that allows a particular subset of SLF/SFBs to recognize certain subset of SRNases but not others.

Methods

The I-TASSER suite21 was used to predict the tertiary structures of SLFB3, SFBB3α, SFBB3β, SLFB9, SFBB9α, SFBB9β and SRNase haplotypes (i.e. S1-S4, S7-S10, S16, S20, S24-S26, S28, S30, S31) (Table S1). The quality of the final refined models was subjected to a series of tests for their internal consistency and reliability. Backbone conformation and overall model quality were evaluated by PROCHECK30 in VADAR v1.831 and ProSa-web32, respectively. Furthermore, the folding topologies of the modeled structures were examined using the Dali server (http://ekhidna.biocenter.helsinki.fi/dali_server). The ZDOCK program was used to evaluate the interactions between currently known Malus x domestica (Borkh.) SLF/SFBs and SRNases. The PDB file of the modeled structures of SLF/SFBs and SRNases were used as inputs to ZDOCK as receptors and ligands respectively. For docking purpose, the number of top poses for SLF/SFB-SRNase complex was set to 2 000 with the root mean square division (RMSD) cut off the value of 10 while only the hypervariable regions identified by Ashkani and Rees23 was provided to ZDOCK as an interface. Furthermore, ZRANK was used to re-rank the ZDOCK scores while the binding energies from ZRANK were analysed using Wilcoxon rank-sum test33, as implemented in R34. The RMSD values were calculated with ProFit v3.1 (Martin, unpublished, http://www.bioinf.org.uk/software/profit). Assuming that a pose with the lowest binding energy is the nearest native, the structure with the top ZRANK from the previous analysis was provided to ProFit as the reference structure in PDB format. All the other structures (from 2 000 poses) were provided as mobile structures. The reference structure remains fixed while the mobile structures are fitted on to it. Finally, to determine the accuracy of the scoring function for docking models that were shown to be significant using Wilcoxon rank-sum test, the ZRANK scores were plotted against ligand RMSDs, L-RMSDs. Plotting the ZRANK scores against LRMSDs shows the distribution of the poses in the form of an energy funnel, which is suggestive of the accuracy of the scoring function, whereby most near-native complexes have lower binding energies and low L-RMSDs.

Additional Information

How to cite this article: Ashkani, J. and Rees, D. J. G. A simple, high-throughput modeling approach reveals insights into the mechanism of gametophytic self-incompatibility. Sci. Rep. 6, 34732; doi: 10.1038/srep34732 (2016).