Introduction

Mating patterns are important because they determine which alleles unite in the same individuals to be tested by selection. When individuals with particular combinations of genotypes are more or less likely to mate with each other than expected by chance, this can bring together combinations of beneficial alleles in the offspring.  It can also bring recessive deleterious alleles together and cause inbreeding depression. There may also be variation between individuals in how many mates, and hence offspring, they have, causing heritable changes in the population. We would like to understand who mates with whom, and characterise variation in reproductive success.
For many plants mating is mediated by animal pollinators. Two aspects of pollinator behaviour are important. Firstly, pollinators make decisions about which plants to visit based on the phenotypes of the plants. They may preferentially visit flowers that are especially attractive or rewarding, causing differences in reproductive success between plants. There is an extensive literature exploring differences in the female component of reproductive success measured through seed set. However, although the majority of flowering plants are hermaphrodites [ref?], the male component of fitness, reflected in successful pollen export to other plants, has received much less attention, because this requires accurate assignment of paternity in seeds. Since every offspring receives exact half its genetic material from the mother and half from the father, by neglecting variation in male fitness we also potentially miss substantial variation in overall reproductive success.
Secondly, it is more energy efficient for pollinators to transition between plants that are close to one another than to move at random between plants. This means that plants are more likely to mate with close neighbours than more distant plants. The distribution of pollen dispersal distances is described by the pollen dispersal kernel. In plants, it is common to find that the dispersal kernels that show strong leptokurtosis, or overdispersion; most dispersal is between close neighbours, but there is a ’fat tail’ of long-distance dispersal events that cause the dispersal kernel to decay more slowly than exponentially. Although these events are relatively rare, they contribute disproportionally to the spread of migrants, allowing populations to expand rapidly in space, and to homogenise distant populations. This is especially important in a hybrid zone, where divergent populations meet and exchange migrants. Here, the extent to which alleles at individual loci are able to introgress across the hybrid zone is determined by the balance between the homogenising effect of dispersal, and selection against migrant alleles keeping populations spatially distinct. Given estimates of dispersal distance and distribution of alleles in space we can infer the strength of selection acting on those alleles. Accurate estimates of the shape of the dispersal kernel are this invaluable in understanding the evolutionary processes acting in natural populations.
Pedigree reconstruction has emerged as a powerful tool to investigate both both differences in mating success and dispersal \cite{Pemberton2008}. The idea is to reconstruct mating events by identifying the parents of a set of offspring asssuming Mendelian inheritance based on shared alleles at loci for which offspring and candidate parents have been genotyped. This is most successful when as much informative data as possible can be included in the analysis \cite{Wang2007}. One way to achieve this is to jointly infer sibling relationships and the paternity of those sibships . Likewise, inference of parentage jointly with other biological parameters that influence mating, such as those related to dispersal or mating success increases the accuracy to infer both the pedigree and biological parameters \cite{Hadfield2006,Neff2001}. It follows from this that an even more powerful approach would be to unify these approaches and jointly infer parentage, sibships and biological parameters. However, we currently lack software which is able to do this efficiently for modern SNP data.
Here, we examine a hybrid zone between two subspecies of the snapdragon, Antirrhinum majus that differ for flower colour. Flower colour in A. majus is controlled primarily by the genes Rosea , controlling the distribution of magenta anthocyanin pigments, and Sulfurea , controlling the distribution of yellow aurone pigments. In the Vall de Ribès in the Spanish Pyrenees, the yellow-flowered A. m. striatum and the magenta-flowered A. m. pseudomajus meet and form a narrow hybrid zone where loci at Rosea and Sulfurea recombine to give rise to six flower-colour morphs. Alleles at these loci nevertheless form sharp clines consistent with selection against recombinant genotypes over many generations. Furthermore, parentage analysis indicates genotype-by-environment interactions for fitness, with the A. m. striatum - and A. m. pseudomajus -like genotypes having an advantage in their home territory. As there are no obvious difference in habitat or other phenotypes across the hybrid zone, and given that pollinators are known to discriminate between flowers based on colour, non-random foraging behaviour by pollinators is an appealing explanation for these differences in fitness. Statement about female fitness (in the core only?) from \citet{Tastard2012}, but we still don’t know much about male fitness. Parentage estimates reflect the whole life cycle, which includes variation in mating success, seedling establish and survival, and we are not able to distinguish pollen and seed dispersal. As such, this is not ideal to investigate mating patterns directly to properly characterise male fitness and pollen dispersal.
In this study we reconstruct mating events between wild snapdragons in the hybrid zone. We use a panel of open-pollinated seedlings of known maternity collected in a natural hybrid zone and dense sampling of possible sires. This differs from the pedigree because there is no seed dispersal or establishment, so parentage reflects mating patterns through pollen movement only. We build on previous methods to jointly infer paternity, sibship relationships and the pollen dispersal kernel. Use these results to examine the extent to which pollinators cause assortative mating and differences in male fitness beyond what would be expected due to local population structure. Finally we use simulations to verify that these results are unlikely to be due to false positive paternity assignment due to insufficient statistical power or missing fathers. These results imply that pollinators are an important force mediating selection on flower colour.

Materials and Methods

Study population