Population genomics relies on the principle that loci across the genome are influenced by genome-wide evolutionary force, whereas selection is locus-specific and imprints a particular pattern of variability only on linked loci 12. Increasing attention has been put in understanding what proportion of a genome or which genes are being shaped by selection. Directional (Darwinian) selection can leave a set of signatures in the genes under its influence, such as the rapid divergence of functional sites among species and the depression of polymorphism within species. On the basis of these signatures, it is possible to identify genes or chromosomal regions which are likely targets of positive selection 34.

This is achieved by comparing all the loci across the genome that respond similarly to demography and neutral history of populations with outlier loci that show patterns of variation that deviate from the rest of the genome. Identification of outliers is important for two main reasons: i. such loci are potentially under selection and could be a sign of adaptive variation; ii. they could also bias estimates of population genetic parameters such as gene flow, population size and structure, and therefore should be excluded from these analyses. However, selected loci could be used in studies to better understand adaptation or to plan conservation-management strategies 5.

The identification of genes that have undergone positive selection is an important step in understanding how populations have adapted to environmental changes. Such studies are increasingly widespread 678910 and their application to livestock species can also reveal insights on their selection history.

Aberrant behaviour of a locus can range from having exceptionally high or low FST between populations, to having an excess or deficit of low frequency alleles in a population 5.

Several statistical methods, in which loci candidate for selection are identified in the extreme tails of empirical distributions, have become a widely used strategy in genome-wide scans for selection 11121314151617. However, several studies pointed out important methodological conditions to ensure the correct application of these methods 18 or stressed their lack of power 192021.

FST statistic can be used to assess if the variation of SNP allele frequencies among populations leads to signatures of selection 222. If FST is determined only by genetic drift, all loci across the genome are affected in a similar way. In the presence of locus-specific selection pressure, deviation in FST values is observed in selected loci and in linked genetic markers. See Table 3 for SNPs showing simulated FST.

In this study we want to compare outlier loci assessed using a FST based method 22 with those identified by a recently described method 8 based on spatial analysis (SAM). In the original paper Joost et al. tested the efficacy of SAM against FST based methods using AFLP and microsatellite data from two different species, showing a strong correspondence between the two approaches 8. SAM is an implementation of logistic regression resembling the method applied by Jump et al. to plant populations with AFLP markers 23; but it is used in an explorative way rather than for the confirmation of working hypotheses 824 and therefore the processing of all possible univariate association models between all environmental parameters and the presence or absence of all single alleles is carried out.

Here we evaluate how FST and SAM perform with a different kind of marker, single nucleotide polymorphisms (SNPs), which are increasingly employed due to their high number, low cost and easy of scoring. Besides, SNP can be found inside gene candidates for artificial or natural selection and therefore they might be more informative for this purpose rather than neutral or random markers, like microsatellites and AFLPs.

We analysed goat breeds of southern areas of the Mediterranean basin (Italy, Greece and Albania) sampled in the context of the European ECONOGENE Project http://www.econogene.eu using a set of SNPs previously described 25. See Table 2 for Breeds analysed, country of origin and their sample sizes (N).

Six alleles at five loci (Lipase, Casein Alpha S1, Interleukin 2, Integrin Beta-1 and Growth Hormone Receptor) out of 27 were found to lie outside the 95% confidence region of the conditional joint distribution of FST and mean heterozygosity (Figure. 1) by FDIST2 analysis. They are potentially under selection (P < 0.05) (Tab. 3).

With a significance threshold (ST) set to 95% (corresponding to P < 5.04E-06), SAM detected 16 loci associated with at least one environmental parameter, among which 3 are also identified by FDIST2 (LIPE, CSN1S1-5 and IL2_ln2). Conversely, Integrin beta-1 and Growth Hormone Receptor, identified by FDIST2, are not detected by SAM at any confidence level, and are not associated with any environmental parameter. With a ST of 1.01E-13 (confidence level of 99.99999999%, Bonferroni correction included,), SAM identified 3 alleles at 2 loci (CSN1S1 and LIPE) to be significantly associated with at least one environmental variable. Of the two SNPs analysed in CSN1S1 gene, one (CSN1S1_ex9) is involved in 38 significant models and is associated with 7 among 10 "families" of environmental parameters (duration of sunshine between may and September; relative humidity in January, May to September, and December; number of days with > 0.1 mm rain per month from may to September, and the yearly mean; temperature from April to September; precipitation from June to September; diurnal temperature range from march to December, and the yearly mean; number of days with ground frost per month from March to November, and the yearly mean). The second SNP in CSN1S1 (CSN1S1-5) is associated with 3 environmental parameters which are relative humidity in May, the number of days with ground frost per month in April, and diurnal temperature range in October. Finally, one LIPE allele is associated with relative humidity from May to September, the coefficient of variation of monthly precipitation in March, diurnal temperature range from March to December, as well as with the yearly mean, and with wind speed from April to May (see Figure 2).

One CSN1S1_ex9 allele is the unique involved in the 5 most significant models (1.01E-16 Bonferroni correction included). At this level of significance, it is associated with relative humidity in July and August, with the number of days with > 0.1 mm rain per month in August, with precipitation in August, and with the number of days with ground frost in April.

Rapid adaptation to environments is expected to have shaped at least some of the genetic diversity in marginal European goat breeds 8. We thus expected that evidence for divergent selection should be detected in goat breeds, as a response to human and environment mediated selection and to migration history 26.

Outlier approaches, in which genes potentially under selection are identified in the extreme tails of empirical distributions, have become a widely used strategy in genome-wide scans for ascertaining selection signatures 11141617.

Most studies have been conducted using neutral or random markers, like microsatellites or AFLP, however the increasing use of SNPs as a genetic marker due to their rapid discovery rate provides the opportunity to identify genome-wide signatures of selection, as shown in humans 2728. There is potentially little power to detect outlier loci when examining populations in pairwise comparisons with biallelic markers 29. On the contrary, in multi-population analysis the variability in FST among loci is reduced facilitating the detection of outlier loci 22. Lao et al. 30 have shown that the frequency distributions of SNPs with ability to detect geographical population structure are most likely shaped by local positive selection rather than by genetic drift. The set of SNPs used for this study have been proven very efficient in detecting geographical patterning (Pariset et al, Population analysis of goat breeds from Italy, Albania and Greece reveals precise geographical patterning. Manuscript in preparation).

By calculating FST and mean heterozygosis as a measure of genetic differentiation for each locus, we have identified six loci at five genes in goat populations, which have been potentially target of selection according to the distribution of their genetic variation. One of them is related to body size and can be easily selected for by man, two are related to nutritional property of milk (CSN1S1 and LIPE) and they can be both selected by man or by adaptive advantage of the offspring. Interleukin-2 is a gene showing evidence for positive selection in mammals 31 as involved in immune response. The last gene, ITGB1, affects several traits, many of which potentially under selection.

Moreover, by using univariate logistic regression analysis with geographical variables, we could identify 3 of the 5 genes ascertained using FDIST2. The combined approach reveals that these three genes are under a selection pressure driven by the environment. One allele at locus CSN1S1 is very significantly associated with relative humidity during warm months (July and August), with the frequency of humidity and of precipitation during a warm summer month as well (number of days with > 0.1 mm rain per month and precipitation in August), and with the number of days with ground frost in April. These conditions (warm, wet) are favourable for parasite number to increase over time.

The last association with ground frost is more difficult to interpret, but April could correspond – grosso modo – to the moment when the animals come out of the goatshed and this could explain why the association doesn't exist during the other months in winter. Integrin beta-1 and Growth Hormone Receptor, identified by FDIST2, are not detected by SAM at any confidence level, and are not associated with any environmental parameter. Indeed, such markers can only be detected by their low variance in frequency among populations, and this action requires a dedicated model. Thus, because of their nature they cannot be detected using a regression approach like SAM does. Outlier loci, possibly under selection, can also be driven by other kinds of selection than natural selection. SAM proposes an environment-centered analysis, but there is room left for non-selected environmental effects, and FDIST or BayeScan are able to identify these outliers. This is a reason why Sam can not be used independently, without comparing results with specialized software packages based on theoretical approaches in population genetics, and focused on the analysis of genetic data

While directional selection tends to reduce within-population genetic diversity and to increase among-population differentiation, the reverse is expected under balancing selection, where gene frequencies tend to some equilibrium that maintains polymorphism. In general, Kelley et al. 20 found that the outlier approaches determine genes that have been targets of positive selection. Hoffman et al. 32 and Minder and Widmer 33 show that it is possible to find candidates for balancing selection using the Beaumont & Nichols 22 approach, but it is difficult because the lower 95% confidence limit is typically close to zero 34.

Among the 6 outlier SNPs, 3 of them had FST values lower than the 99% confidence limit in the FDIST2 test. We hypothesize that balancing or stabilizing selection may be responsible for these outliers, two of which (IL2 and ITGB2) have defence-immunity function. Indeed, genes involved in immunity are affected by balancing selection 35 and proteins that perform a defence/immunity function present low FST values 11.

In contrast, 3 loci had FST values exceeding the 99% confidence limits of FDIST2 and two of them were identified also by SAM. These genes might be important in adaptation to different environments and were probably subject to human selection, as well. SAM is able to identify both loci under directional and in stabilizing selection, but requires the comparison with the results produced by FDIST2 to differentiate the two types of selection, as is the case for IL2. Moreover, SAM can independently validate the FST outcomes and find correlations wit geographical variables, pinpointing the probable cause of the selection.

The population divergence methods implemented in FDIST2 have been widely used in several papers 7151732 and have been shown to be quite robust by simulations among various demographic scenarios 34. Setting the same significance threshold, SAM detects more markers than FDIST2 does, but it was shown that when gradually lowering the confidence level in FDIST2, the method was also able to identify those additional loci detected by SAM 24. SAM's sensitivity is not mastered yet, and further studies will be necessary to establish precise relationships between population genomics approaches and these statistical measures of association.

Data appear congruent by using the two methods for FST values exceeding the 99% confidence limits, more markedly than shown in the Joost et al. 8 paper where microsatellites were used as markers in sheep populations.

It may surprise that a high percentage of loci appear under selection, but it should be considered that the loci were purposely chosen for influencing potentially selected traits. Moreover, an active trade of livestock was historically present in the studied area 36 and adaptation to novel environment should have been crucial for the survival of imported populations.

The methods of FST and SAM can independently detect signatures of selection and therefore can reduce the probability of finding false positives if employed together.

The outlier loci identified in this study can be important because they could indicate adaptive variation in the analysed species, which is characterized by a large range of climatic conditions in the rearing areas and by a history of intense trade, that implies plasticity in adapting to new environments.

The authors declare that they have no competing interests.

LP: designed the study, performed the FST analysis, and drafted the manuscript. SJ: developed the SAM approach, carried out the relative computations, and formulated parts of the paper. PAM: leader of he Econogene project, provided DNA samples and guidance for the project. AV: participated in developing ideas, in supervision and revision of the manuscript.