The barriers that lead to divergence of species during the course of evolution were classified by Dobzhansky in two categories: prezygotic and postzygotic reproductive barriers 1 . Postzygotic reproductive barriers concern all the events that occur after fertilization, such as reduced hybrid viability and fertility, while prezygotic reproductive barriers concern isolation of sexual partners via ecological, temporal or behavioral isolation. Pheromones play a key role in pre-mating recognition of sexual partners 2 . These compounds are defined as substances released by an animal that are able to induce specific behavioral and/or endocrinological reactions in a sexual partner of the same species 3 . Through these reactions, they could be involved in mate choice and sexual selection.

Odorant binding proteins (OBP) are small soluble proteins that are present in the olfactory apparatus as well as in biological fluids such as saliva, urine or vaginal discharge, and are able to bind pheromones (for review see 4 ). OBP are assumed to be directly involved in chemical communication and in the pre-mating recognition process. Three hypotheses are proposed concerning their mechanism of action. The first is that olfactory receptors can recognize the OBP/pheromone complex, not just the pheromone alone. The second hypothesis is that the pheromone can be transferred to olfactory receptors only if assisted by the OBP. The third hypothesis is that the ligand can spontaneously dissociate from the complex with OBP and bind to the receptor as a "free pheromone" 5 .

The role of saliva in chemical communication between males and females is well established in pig 6 , like the role of urine in mouse 7 . In pig, saliva contains the pheromonal steroids 5α-androst-16-en-3-one and 5α-androst-16-en-3α-ol, as well as abundant quantities of salivary lipocalin (SAL1), the most abundant OBP isolated from submaxillary glands of mature males. When extracted from its source, this protein is associated with both pheromonal steroids 8 , and appears to play a key role in the standing reflex in the sow 6 and also in the boar's libido 9 . SAL1 is also expressed in the nasal and vomeronasal area, but devoid of ligand 10 11 . SAL1 exhibits a classical structure of lipocalins characterized by a fully conserved N-terminal -G-X-W- motif and the typical folding pattern of a nine-stranded antiparallel β-barrel forming an internal ligand binding site for small hydrophobic molecules 12 , despite relatively low sequence similarity 13 . SAL1 also possesses a glycosylation site on Asn53. Two natural variants have been identified in which in three residues differ (Val61, Ile64 and Ala89 of isoform A are respectively Ala, Val and Val in isoform B). Two residues (Val61 and Ala89) are located inside the β-barrel while the third residue (Ile64) is located next to the β-barrel, suggesting that these minor structural differences lead to ligand binding specificities 14 .

Olfactory receptors are located on the olfactory sensory neurons of the main olfactory system in mammals and on the vomeronasal organ in rodents and other non-primate species 15 . Several authors examined the evolution of olfactory receptors, but few studies of lipocalins and OBP have been performed. Ganfornina et al. 16 undertook phylogenetic analysis of prokaryotic and eukaryotic lipocalins and showed that this family appeared early and is composed of 13 monophyletic clades. These authors also showed that ancestral lipocalin clades in the phylogenetic tree are able to bind large ligands while more recent lipocalin clades, such as clades composed of OBP and MUP (Mouse Urinary Protein), bind smaller ligands. They also found that later clades had higher rates of amino acid substitution, more flexible protein structures and greater ligand-binding efficiency than more ancestral lipocalins.

Logan et al. 17 undertook an extensive study of the Mup cluster in the mouse genome. They identified 21 Mup genes and 21 Mup pseudogenes on chromosome 4. They also identified Mup gene expansion in rat (9 genes and 13 pseudogenes), in horse (3 genes) and in mouse lemur (2 genes and 1 pseudogene) in the same syntenic region. Orangutan, chimpanzee, dog, pig (with SAL1), bushbaby and rhesus monkey have only one Mup gene in the syntenic region. The inferred phylogeny, the accumulation of synonymous substitutions, and the genomic organization of the Mup loci suggest that gene expansion occurred independently in several species 17 .

In the light of previous analyses, the aim of the present work was to study the evolution of SAL1 which is involved in pre-mating recognition in pig. We wanted to determine if selective pressures act on these proteins and to check if positive selection may play a role in binding specificity toward ligands or olfactory receptors.

Phylogenomic analyses showed that the SAL1 family originated in eutherian mammals and that genes belonging to this family were duplicated after speciation events in five mammalian species. In certain living species, such as mouse lemur, bushbaby, orangutan and human, the gene has been lost, as it has in the Neanderthal genome. We can date the loss of the gene in hominid before the Neandertal-modern human split, 400,000 to 350,000 years ago 28 . The number of duplication events varies greatly among species. In mouse and rat, massive cis-duplication events have occurred, with respectively 42 and 22 genes in the cluster, followed by gene loss with respectively 21 and 11 pseudogenes.

Gene duplication represents a source of new genetic material, and can lead to evolutionary novelties. The fate of duplicated genes can follow different models of evolution, with different selective pressures acting on the genes 29 . We checked for a change in selective pressure in all paralogs identified in the SAL1 family and found that only a few paralogs underwent positive selection: one gene in cow, guinea pig, horse, mouse and rat. Moreover, few sites of these genes were identified. A large proportion (66%) of each gene evolved under neutrality and only a small proportion (2 to 5%) under positive selection. However, among the single genes identified as positively selected in guinea pig and horse, a larger proportion of sites evolved under positive selection (18 and 72%, respectively) and more sites were identified as being positively selected (15 sites in guinea pig and 6 sites in horse).

Because sequences of some paralogs share high similarity, we searched for gene conversion in our paralogous gene datasets and found extensive interlocus gene conversion events in mouse and guinea pig, and to a lesser extent in horse and rat. Karn and Laukaitis 30 compared the mouse Mup cluster with a gene tree published by Mudge et al. 31 and suggested that concerted evolution masked the common origin of the gene and neighboring pseudogenes 30 . Our results confirmed this hypothesis, indicating extensive gene conversion in the mouse Mup cluster. This extensive gene conversion phenomenon led to sequence homogenization and is the cause of the concerted evolution of these genes. Such extensive concerted evolution suggests that, at least in mouse and guinea pig, both maintenance of sequence homogeneity and increased gene dosage are important for these species. The evolution of SAL1 paralogs resembles the evolution of the β-globin gene family. In this family, paralogous copies evolved under a process of functional divergence and there is evidence for two gene conversion events in mouse and goat clusters composed of β-globin duplicated genes There is also evidence for variable selective pressure among sites for β and γ-globin genes with 4 to 9% of sites evolving under positive selection 32 .

By combining phylogenetic, gene conversion and selective pressure results on paralogs evolution, we can try to describe the fate of duplicated genes, in which duplication can be seen as an advantageous phenomenon for the species concerned, by combining two scenarios from Innan and Kondrashov 29 . In the first scenario, one could consider the massive duplication in rat and mouse as a gene amplification where the increase in dosage of these genes is beneficial. This scenario of evolution corresponds to category IIa described by Innan and Kondrashov 29 . In this model, if selection for the duplicated copy is weak, pseudogenization can occur if a null mutation is fixed, which is the case in both mouse and rat. The occurrence of gene conversions that maintain sequence similarity and promote conservation of gene copies could be consistent with that hypothesis, but the high frequency of gene conversion events is not restricted to mouse and rat. In fact, guinea pigs, which do not harbor large gene amplifications, have the highest frequency of conversion events per gene copy among the species tested. The beneficial increase in dosage has already been shown to apply to genes that mediate the interaction between the organism and the environment 33 , as is true of genes of the SAL1 family. However, we also showed that among the many duplicates in rat, mouse and guinea pig, one gene per species is under positive selection so increased gene dosage and gene conversion events are not the only driving force of the evolution of these genes in these species. For these positively selected duplicates, it is the scenario of the category III 29 which fits, where a new copy can be fixed and preserved by positive selection, leading to the possible emergence of a new function for the positively selected gene.

To study selective pressure in the SAL1 family in more detail, we tested the amino acids changing occurring in the 12 branches supporting species as positively selected by PAML. Our results showed that marmoset, dog, guinea pig, horse and mouse branches underwent positive selection just after divergence. This evolutionary scenario likely reflects the ability of the SAL1 family to diverge and to adapt to new behaviour between sexual partners. A previous study on mouse and rat genes identified 32 sites as positively selected on rodent co-orthologs of SAL1 34 . In that study, mouse and rat genes were considered together, whereas in our study, mouse and rat genes were analyzed separately, this explains the difference between the two results. Indeed, we only identified one site under positive selection in mouse and no positive selection in rat. The difference between the two results is also due to a difference in the probability threshold chosen to determine whether a site is subject to positive selection or not. In Emes et al. 34 , a site was said to be positively selected if the probability for one model is > 0.90 and > 0.50 for at least one other model. In our study, we chose to consider only sites whose probability was > 0.95 in order to minimize false positives.

Finally, we compared tests of variable selective pressures for the family using several PAML codon models. We found evidence for positive selection in a small proportion of sites. Because positive selection is known to play a role in the diversification of protein functions, we mapped all positively selected sites on the 3D structure, in order to assess their biological significance for the gene family as a whole and for each species independently. Apart from the three amino acids that were under positive selection and involved in ligand binding, the other amino acids identified by site models of PAML analyses projected out of the binding pocket. Moreover, the majority of these sites were exposed to solvent. If these sites were involved in the interaction with pheromones, they would be found preferentially in the hydrophobic core and would be buried. We thus propose that positive selection plays a role not only in the binding specificity but also in the interaction between the protein and its environment. We were not able to draw any conclusions concerning selective pressures on each site involved in ligand binding, because gaps in the multiple sequence alignment made these calculations impossible. Nevertheless, for the 16 amino acids involved in pheromone binding, we identified three sites that probably evolved under purifying constraints (87Y, 91N and 93F) and four sites that probably evolved under relaxed constraints (60F, 85V, 121E and 123Y). The three sites that evolved under purifying constraints may be essential for protein function, because they were well conserved during the evolution of the family. In rodent populations, Emes et al. 34 found that MUPs, which are co-orthologs of SAL1, exhibited amino acids under positive selection, and that these positively selected sites were located at the interface between MUPs and their receptors, probably V2R receptors on the vomeronasal organ. They also found evidence that olfactory receptors, such as V2Rs, underwent positive selection. The hypothesis they proposed is that this adaptation phenomenon is due to conspecific competition, resulting in well adapted pheromones, pheromone binding proteins such as MUP, and olfactory receptors 34 . Our results allow us to extend this hypothesis because positive selection also drives the evolution of pheromone binding proteins in other eutherian mammals. So for all the family, and not just for rodents, there is an adaptive evolution of these proteins to their ligands and maybe their receptors, too. It would be interesting to test if V2R receptors are subject to positive selection, not only in rodents but also in other mammals. Several authors reported evidence for positive selection on other OR genes in mammals 35 36 37 38 39 , with possible involvement of positively selected sites in the binding property of proteins. Moreno-Estrada 36 suggested that positive selection could be at the origin of a new ligand binding capability or the modification of odorant perception and could improve the overall degenerated OR gene repertoire, at least in human. In insects, co-evolution of the two enzymes involved in the pheromone biosynthetic pathway and in the pheromone receptor has been suggested to play a role in the speciation process 40 . It would be interesting to test co-evolution of enzyme/receptor, pheromone/receptor and OBP/receptor in mammals.

In mice, MUPs are important for the delivery, via urine, of chemical signals conveying information about the sex and hormonal status of the animal who release the scent mark 41 . In pig, SAL1 may be involved in pre-mating recognition by binding pig specific sex pheromones in saliva 8 . In both species, these proteins are involved in conspecific recognition in the context of reproduction. When the genomes of marine mammals are completed, it will be interesting to search for SAL1 orthologs. Indeed, in such a different environment, chemical communication between sexual partners is probably not mediated by the same olfactory cues as in terrestrial mammals. If a SAL1 ortholog is found in marine mammal genomes, it will be interesting to discover if it evolved under relaxed constraints or positive selection.

It is well established that reproduction is a very competitive process, and that selective pressures on genes involved in the process are not rare (for a review, see 42 ). Positive Darwinian selection is not atypical, especially for genes involved in sensory perception and mate choice 43 . Our results demonstrated that (i) positively selected sites differ between genes and (ii) positively selected sites are involved in ligand binding and are putatively involved in receptor binding. Such a selective pressure on these proteins could be at the origin of a divergence process between species and thus contribute to the speciation phenomenon by reinforcing prezygotic barriers. To test this hypothesis, we performed in vitro mutagenesis experiments on SAL1, but the poor folding of the resulting proteins prevented further experimentation.

The SAL1 gene family originated in eutherian mammals and duplicated after speciation in cow, horse, guinea pig and rodents. Some duplicated genes underwent concerted evolution with extensive gene conversion. Others were subject to positive selection at different sites, and our knowledge of the 3D structure of this protein suggests that the selected sites are involved in pheromone binding and possibly in olfactory receptor binding. This result suggests a functional divergence between species because positively selected sites differ between species. All these data suggest that the evolution of the SAL1 family allows a species-specific strategy to transduce pheromonal signals in mammals, reinforcing species divergence through species-specific sexual behaviour.

CM performed the main data collection and analyses. GP provided advice on bioinformatic analyses. IC performed protein structural analyses. CM, FB and PNLM performed mutagenesis experiments and protein production. PM designed the study and helped guide the general analyses. All authors read and approved the final manuscript.