The sporophytic self-incompatibility mating system is conserved in Olea europaea subsp. cuspidata and O. e. europaea

Fruit setting after self-pollination, crosses and free-pollination appears to be erratic in the cultivated olive tree [Olea europaea subsp europaea L. (O. e. europaea L.)] because of a lack of a suitable model to enable prediction of rates. The same lack of prediction also applies to the wild taxon Olea europaea subsp cuspidata (O. e. cuspidata). Because of their close phylogenetic relationships, we hypothesize that O. e. cuspidata and cultivated olive share the same self-incompatibility system. We used data recently published in a wide study involving four O. e. cuspidata accessions and four olive cultivars. Because the olive varieties have been deciphered for their S-allele pair, that infer determinants present in the stigma and pistil, and that coat the pollen, we deciphered the S-alleles carried by three of the O. e. cuspidata accessions. Data are too scarce and the number of accessions too small to speculate on the O. e. cuspidata genetic population structure. The working hypothesis is confirmed. This study and data from the Italian team will enable us to embark on a large-scale hybridization program between the two subsp. to obtain a wide range of progenies for screening for responses to cold, diseases and pests.


Introduction
Recently Cáceres et al. (2015) published new data on fruit setting after hybridization between two subspecies of Olea europaea, namely subsp europaea L. (O. e. europaea L.) and subsp cuspidata (O. e. cuspidata Wall ex G. Don.). Controlled crosses between four O. e. cuspidata accessions on the one hand and four olive cultivars-'Dolce Agogia', 'Favolosa' also called 'Fs17', 'Frantoio' and 'Leccino'-on the other hand, led to fifteen hybrid plants saved in a nursery. Fruit setting between pairs of the subspecific taxon O. europaea appeared erratic because there was no way of predicting success or failure. However, we have already attributed S-allele pairs to three of the olive varieties within the framework of the sporophytic model. We thus questioned whether O. e. europaea and O. e. cuspidata have the same self-incompatibility system.
As reported by Besnard et al. (2001) seventeen hybrid trees, which are now 31 years old, were obtained by Villemur by 1984. Hannachi et al. (2009) studied oil yield and some phenotypic traits such as fruit color and size in these trees. These trees are maintained in dense conditions in the orchard collection at INRA-Montpellier; they have lost their juvenile traits and blossom and produce fruit regularly each year with each tree yielding several kilograms of fruit Table 1.
Natural hybrids between subspecies O. e. cuspidata and O. e. europaea occur in areas surrounding olive orchards in South Africa (Costa 1998; H. Sommerlatte pers comm). There are no published reports of natural hybridization in countries other than South Africa, but there are speculations that they may occur (Breton et al. 2008;Cáceres et al. 2015 for review).
Recently Gibbs (2015) published a complete review on the different aspects of self-incompatibility including architecture self-incompatibility (ASI) type and late-acting self incompatibility. We can omit aspects that are well documented in the review. However, most researchers ignore the fact that SI distribution in the evolutionary tree of the Plantae is not in concordance with the species phylogeny. It appears that SI has appeared several times either as architecture SI, gametophytic self-incompatibility (GSI) and sporophytic self-incompatibility (SSI) (Fig. 1). Moreover, for close species in different genera SI may have been lost and become selfcompatible as in Prunus (P. persica), whereas other Prunus species (cherry, apricot and almond) harbor a GSI mechanism. Arabidopsis thaliana has lost SI whereas A. lyrata harbors a deep SSI system (Goubet et al. 2012).
Oleaceae are known to harbor different floral biology systems. In Jasminum heterostyly, either distyly and tristyly, is known in 24 families (Ganders 1979;Dommée et al. 1992). In Fraxinus different modes exist; F. excelsior and F. angustifolia (Mediterranean) are polygamous (Bochenek 2011) as each tree carries at least some hermaphroditic flowers and may be functionally dioecious. F. ornus displays androdioecy (Dommée et al. 1999). F. lanuginosa displays androdioecious behavior in northern Japan (Ishida and Hiura 2002); and Phillyrea displays androdioecy (Saumitou-Laprade et al. 2010). In Olea little is known apart from O. e. europaea; however, Hannachi and Marzouck (2012) recently discovered androdioecy in O. e. subsp. sylvestris (wild olive or oleaster). Although the GSI model has been considered as functioning in the cultivated olive (Zhu et al. 2013) it is now established that the SSI model works (Breton and Bervillé 2012;Breton et al. 2014;Farinelli et al. 2015;Koubouris et al. 2014). SSI and GSI differ in other features by the stage of deposit of S-determinants onto the pollen coat. With SSI the S-determinants are deposited early before the tetrad stage, and thus all pollen grains are coated by the same S-determinants . Without dominance between the S-alleles, both determinants coat all pollen grains produced by the individual. However, when dominance is expressed, only the S-determinant of the dominant S-allele is present on the pollen coat. Gibbs (2015) reported the history of emergence of the SSI model, and stressed the difficulties it may cause in analysing cross data. The singularities are due to dominance relationships ([) whereby (i) each pollen grain may not be coated by the S-determinant of the Sallele it carries; that is, the determinant of the recessive S-allele is lacking, and (ii) in reciprocal crosses with R2 [ R3 and R1 = R3 the cross $R1R3 9 R2R3# (determinant R2) is compatible, whereas the reciprocal cross $R2R3 9 R1R3# (determinant R1R3) is incompatible.
Differences in fruit setting after crossing, selfpollination or free-pollination between O. e. cuspidata and O. e. europaea resemble most olive cross data widely published in several countries and quoted by Cáceres et al. (2015). For three of these varieties Farinelli et al. (2015) have, within the framework of the sporophytic self incompatibility (SSI) system, already deciphered the S-allele pair in each variety. We aimed to verify if O. e. cuspidata harbors SSI as is the case in cultivated olive, and therefore we attempted to decipher the S-alleles in the four accessions from O. e. cuspidata. Based on the attributed S-allele pair of each olive variety we deduced the S-alleles in O. e. cuspidata. Cross fruit set is shown to match inferences from SSI for successes and failures. When reciprocal crosses were made the power of attributing a pair is higher than when crosses were made in only one direction. Data enabled us to decipher S-alleles in 16 crosses. It was fortunate that Cáceres et al. (2015) used several olive varieties that carry different S-allele pairs (including all S-alleles). Theoretical self-compatibility and self-incompatibility as open-pollination rates are discussed in relation to experimental data.

Materials and methods
The origin of the O. e. cuspidata studied in Montpellier was Stellenboch, South Africa. Elia Psillakis brought it from Chania (Greece) in the 1970s as fresh scions that were grafted onto an olive stock in a greenhouse in Montpellier. O. e. cuspidata was grafted onto the cultivated olive 'Olivière', and local varieties from the Hérault area were grown in the greenhouse. The grafted O. e. cuspidata tree was exposed to  The genetics of SI depend on one locus with several S-alleles; thus two S-alleles are expected for each variety. Until now six S-alleles have been hypothesized for olive, and named R1 to R6, thus the two by two pairwise combinations lead to fifteen host pairs (Breton and Bervillé 2012). Dominance relationships were revealed when one reciprocal pairwise combination leads in one direction to fruit set, and in the other direction to failure of the cross (Moutier 2002;Farinelli et al. 2006Farinelli et al. , 2015. Dominance relationships established by Breton and Bervillé (2012) that is, four levels of dominance were found. Among olive varieties for any S-allele pair, the higher the difference in dominance level the higher the self-compatibility level of the tree (Breton et al. 2016). However, on the male side the most important feature is that each pollen grain is coated by the same S-determinants-all pollen grains emitted by one tree carry the same S-determinant. S-proteins remain undetermined in olive at present. Consequently, to avoid any ambiguity in the coding of S-alleles and S-determinants for varieties the S-allele pair is italicized (SxSy), whereas the pollen determinants are presented with regular characters and underlined (Sx, when Sx [ Sy). Obviously, for one variety the S-allele pair in the stigma and the pollen determinant(s) is conserved, whether or not the other variety is male or female.
From all data published on olive crosses one can deduce that data from a cross that succeeded are more reliable than data from a cross that failed. In all publications, repeats of one cross in several bags shows that successes occur for about half of the expected compatible bags (Farinelli et al. 2015).
Routinely, when a variety is mated as male or female with a series of other varieties all with known S-allele pairs, then the S-alleles enabling success are identified. We proceeded with other cross data in a stepwise manner. It is only necessary to eliminate Salleles to reduce the number to two potential S-alleles. Thus, choice of the reference varieties for crossing is vital to complete deciphering. In the Cáceres et al. (2015) study all host varieties were emasculated, thus there was no risk of interference in fruit set by selfpollination, and therefore we can be confident with the data. However, emasculation may have damaged the stigma and pistil because the average fruit set appeared low for self-pollination and free pollination in the varieties. The C1 to C20 codes refer to crosses made and examined by Cáceres et al. (2015).

Results
The open-pollinated grafted O. e. cuspidata tree grown in the greenhouse at Montpellier produced many fruits. After sowing viable seedlings were grown in a greenhouse. After 31 years 17 viable adult trees remain in the INRA collection at Montpellier. Cáceres et al. (2015) published a set of data after crossing four cultivated olive varieties (Dolce-Agogia, 'Favolosa', Frantoio and Leccino) with four O. e. cuspidata individuals from South Africa (SA), China (CH), and India (IN). Farinelli et al. (2015) attributed the S-allele pairs R2R3-R2 (female genotype-pollen determinant), unknown; R4R5-R5, and R1R5-R1R5 to these varieties, respectively.
Because O. e. cuspidata SA-A as female was successfully mated by 'Dolce Agogia' (C1)-R2R3 as host and R2 as male (C16)-in both directions of crossing, SA-A cannot carry R2 or R3 but could carry R1, R4, R5, or R6. $SA-A was successfully mated by Frantoio (R5) (C3) and Leccino (R1R5) (C4), and consequently R1 and R5 can be eliminated as the possible S-alleles for SA-A leaving R4R6 as the genotype for SA-A. The SA-A cross by 'Favolosa' failed (C2, C19) as both SA-A and 'Favolosa' carry the common S-allele R6. Consequently SA-A and 'Favolosa' carried R4R6.

Discussion
Comparison of progeny size from O. e. cuspidata in Montpellier (present results) and in Italy (Cáceres et al. 2015) shows that embryo rescue is useless. The determination of self-incompatibility in olive is expensive and requires considerable investment.
S-alleles were identified for the cultivated olive variety 'Favolosa' as R4R6 using three O. e. cuspidata accessions, namely, the two from South Africa (SA-A R4R6 and SA-B R3R4) and one from China (R5R6). The fourth O. e. cuspidata accession from India remains partially deciphered as R1R5, R1R4 or R4R5. R4 is the lowest ranking recessive S-allele and thus did not influence the outcome of crosses unless the parent was homozygous for R4, but a homozygous R4R4 olive individual has not been documented. Data on self-pollination shows that all these pairs are partially self-compatible-thus R1R4 could be eliminated based on the Breton et al. (2016) model, and openpollination data showed that all host varieties received some compatible pollen. The Chinese accession is the first occurrence of R5R6 pair reported for an Olea individual.
The pedigree of 'Favolosa', a selection from openpollinated progenies from Frantoio R4R5, infers that 'Favolosa' should carry either R4 or R5 and an unknown S-allele from the pollination source. 'Favolosa' received R6 from free pollination. It is logical to suppose that a pollen grain coated by R5 had not pollinated Frantoio. However, R6 was not detected previously in Central Italy and is infrequent in Central Italy, but it is present in 'Nostrale Rigali' in Sicily (Farinelli et al. 2015). The frequency of R6 among about eighty varieties was 9/155 (0.06), whereas the frequency of R3 was 35/155 (23%) (unpublished statistics). The number of O. e. cuspidata individuals here is too low to speculate on the frequency of R6 in its populations.
The self-compatibility rate is expected to be higher for the R6R4 pair because of three levels of dominance difference between R6 and R4 (Breton et al. 2016). Mehlenbacher (1997) found nine levels of dominance among 39 S-alleles in Hazelnut, and self-fertility was present in cutleaf hazelnut (Mehlenbacher and Smith 2006). Self-fertility occurs in genotypes combinating S28 and S1, S2 or S26 S-alleles-all three being recessive to S28. However, the possibility that S28 alone leads to self-fertility by mutation in the male part cannot be eliminated.
There is concern because OeSA-B was mated by the other three accessions, but not by Frantoio. This is not important because 'Leccino' and 'Frantoio' carry the R5 S-allele. However, it is frequent that a mate predicted to succeed, leads to no fruit set (Farinelli et al. 2015). This does not contradict the SSI model. Cáceres et al. (2015) used % for O-PR and selfpollination of 100 emasculated flowers. However the results are ambiguous, because they did not say whether they had or had not counted male-sterile flowers, the proportion of which is variety dependent. The exact description should be 'fruit set for 100 hermaphrodite flowers' that standardizes data to enable comparison of fruit sets between varieties.
Self-pollination for the R3R4 pair is expected to be relatively low. Thus 0.28 (0.28 fruit for 100 emasculated flowers calculated for 1 fruit / 362 emasculated flowers) is probably overestimated, and the range in variation could be important on a larger sample.
Unambiguously O. e. cuspidata from China carries, for the first time, the R5R6 pair, a combination not found found in cultivated varieties Farinelli et al. 2015). Self-compatibility is expected to be relatively high (with Frantoio for reference). Openpollination is low probably because varieties carrying R5 are frequent in the area (Farinelli et al. 2015). All varieties that do not harbor R6 release compatible pollen for the host O. e. cuspidata in the Perugia collection. This illustrates the paradoxical inference from SSI as in any location free pollination of one host variety does not depend on the host variety itself, but on the surrounding varieties. Thus trees carrying R6 even at low frequency are pollinators for 99% of host varieties, and consequently it is not surprising that 'Favolosa' harbors R6. This is one of the consequences of the SSI model, and does not happen with the GSI model.

Conclusion
The olive subspecies O. e. europaea and O. e. cuspidata share many molecular markers and a few are specific of each species (Cáceres et al. 2015). Moreover they share the same self-incompatibility system. This study and that of Cáceres et al. (2015) will now enable us to start on a largescale hybridization program between the two subspecies to obtain large progeny populations for screening for agronomic traits such as resistance to cold, diseases and pests. This study shows that crosses can be chosen based on compatibility.