Introduction

Local adaptation is defined as the evolution, through natural selection, of traits that have high fitness in specific habitat conditions. Because functional genes are the foundation of these traits, their genetic diversity is believed to dictate the adaptive capacity of individuals and the persistence of a population through environmental changes (Ford 2002). As such, the identification of the processes that shape the diversity of these genes is a major concern for the prediction of evolutionary trajectories of populations and represents an important issue in conservation biology.

In this regard, genes that are directly associated with individual survival have received significant attention (Suprunova et al. 2004; Ingvarsson et al. 2006). Among gene candidates in this category, those of the major histocompatibility complex (hereafter MHC) have seemingly received the most attention. These genes, which provide the foundation of the immune system of vertebrates, encode several receptors that bind protein fragments of both viral and bacterial antigens and, therefore, reduce the occurrence of infection and disease (Bernatchez and Landry 2003; Fraser and Neff 2009). Despite numerous studies performed in experimental, semi-experimental, or natural environments (reviewed in Bernatchez and Landry 2003), generalizations about the mechanisms driving MHC evolution remains mostly unclear. While many have suggested long-term effects of selection (Sauermann et al. 2001; Miller et al. 2001; Froeschke and Sommer 2005; van Oosterhout et al. 2006; Heath et al. 2006; Wegner 2008; among others), a consensus has not been reached regarding the type of selection involved. The short-term mechanisms driving variation in allele frequencies within and among populations have been as much, if not more, ambiguous. Indeed, a growing number of studies suggest large variations in the relative importance of selection (Slade 1992; Landry and Bernatchez 2001; Miller et al. 2001; Campos et al. 2006; Aguilar and Garza 2007) and stochastic processes (Hayashi et al. 2005; Langefors 2005; Peters and Turner 2008) that shape MHC diversity in natural populations. Although these studies were restricted to MHC genes, the ambiguity among conclusions from different studies (sometimes performed on the same taxa) demonstrates the difficulties of interpreting natural evolutionary mechanisms when using few DNA segments from either a restricted number of populations or a limited spatial scale. The genetic diversity of populations is the result of neutral and/or non-neutral processes acting at different evolutionary time scales. As a result, the effects of natural selection are entangled with both the short-term effects of drift and gene flow and the long-term effects of mutation rate and allopatric fragmentation (Excoffier 2001; Otto 2000), both of which are often locally specific. This is especially relevant for populations living over post-glaciated areas, which are believed to have suffered from drastic reduction of genetic diversity during and since the colonization of these regions (Bernatchez and Wilson 1998; Hewitt 2000).

A joint analysis of sequence, frequency, and genotype diversities from numerous genetic markers observed within several populations located on a wide geographic area would be the most effective way that may help identify the evolutionary processes acting on genes. The objective of this study was to increase understanding of the mechanisms acting on functional genes over a territory once covered by Pleistocene glaciations; to do so, a common Cyprinidae species of North America was adopted to evaluate the differential effects of natural selection and neutral evolutionary processes on targeted coding regions. Level of diversity was then quantified on (1) a single exon from three genes: the MHC class IIβ, the growth hormone (GH), and the trypsin (TRY) digestion enzyme and (2) seven neutral markers (microsatellites) within a set of 27 natural populations sampled in an area over which the postglacial colonization has previously been investigated by Girard and Angers (2006b).

Materials and Methods

Model Species and Sampling

The longnose dace (Rhinichthys cataractae) is a fish species typical of turbulent rivers found in almost all the temperate habitats of North America, from Mexico to the Yukon (Scott and Crossman 1973; Fig. 1a). These fish have little interaction with humans; thus, they are expected to be almost undisturbed by stocking activities or translocations (Scott and Crossman 1973). A total of 542 individuals were sampled by electrofishing (LR-24, Smith-Root) from 27 rivers of the Quebec province (Canada) (Fig. 1b, c, d; Table 1). These rivers are located within four major drainage basins (James Bay, Saint Lawrence River, Outaouais River, and Saguenay River). Sample sizes ranged between 10 and 24 individuals. For each individual, a piece of the caudal fin was removed and stored in 95% ethanol for molecular analyses. In addition, individuals belonging to a closely related species, the blacknose dace (R. atratulus), were analyzed as an outgroup.

Fig. 1
figure 1

Geographic position of populations and their MHC (a), GH (b), and TRY (c) compositions. The current distribution of the longnose dace in North America is presented in gray. In maps, black and gray identifiers represent populations for which Fu or Li test was significant or not, respectively, while white identifiers indicate populations for which the test was not performed (see text). Finally, the minimum spanning network between the longnose dace and blacknose dace (Bd) alleles is shown for each gene. The colors in the trees correspond to those presented in pie charts. The black circles correspond to alleles not found in this study

Full size image
Table 1 Neutrality statistic tests computed on microsatellites and genes for each longnose dace population
Full size table

Molecular Analyses

Functional Genes

We focused our analyses on the second exon of the MHC class IIβ because of its key role in peptide binding (Ono et al. 1993) and its known polymorphism (Garrigan and Edwards 1999; Landry et al. 2001). Primers were designed on the 5′ end of this exon and the 3′ end of the third exon according to conserved sequences observed among several MHC DAB gene sequence data of Danio rerio available on Genbank. By polymerase chain reaction (PCR), the primers 5′-CCAGTGACTACAGTGATATGG-3′ and 5′-TGGAGGTCACATCTGAGG-3′ successfully amplified a segment of approximately 600 bp (which covered the 200 last bp of the second exon, the first 150 bp of the third exon, and the entire intron between them). The following reaction conditions were used: a 12.5-μl reaction aliquot containing 1.5 mmol l−1 of MgCl2, 2.5 nmol l−1 of each dNTP, 0.2 unit of Taq polymerase, 1.25 μl of 10× Taq polymerase buffer (Invitrogen), and approximately 20 ng of DNA. Reaction conditions included an initial denaturation of 10 s at 92°C; followed by 45 cycles combining 15 s at 92°C, 10 s at 54°C, and 30 s at 68°C; and a final extension of 120 s at 68°C. A reverse primer specific to the longnose dace and in closer proximity to the second exon was then designed (5′-AGTGGACACACATGGTGAG-3′), which led to the amplification of a segment of 295 bp (which covered 229 bp on the exon and 66 bp on the intron). The PCR reaction and conditions that optimized this amplification were the same as outlined above, except using an annealing temperature of 50°C. The polymorphism of this locus was then screened using the single strand conformation polymorphism (SSCP) (Orita et al. 1989). The locus was electrophoresed on a 6% nondenaturing gel for 10 h at 20 W in 0.5X TBE. To ensure the reliability of similar migration patterns among populations, the different patterns from each population were sequenced for at least one individual.

No clear information is available on the function and polymorphism of each exon of GH and TRY in fish. Consequently, determining which GH and TRY regions to analyze was not as straightforward as in the case of MHC. For each locus, we used the following procedure to identify an appropriate DNA segment. The longer exons were targeted among the five that constituted each gene. This choice was based on a strict probabilistic assessment that longer exons are more susceptible to accumulating mutations. Primers were thus designed (according to homologous sequences available for other fish species in Genbank) for the fourth (~160 bp) and the fifth (~201 bp) exons of GH, and the second (~160 bp) and third (~260 bp) exons of TRY. GH primers were designed according to conservation sequences found among very divergent species (Cyprinus carpio #X51969 and Hypophthalmichthys molitrix #M94348), and TRY primers were designed according to the genomic sequence #BX539313 of D. rerio. Successful amplifications were performed on longnose dace for all loci except the fourth exon of GH, which was discarded. A SSCP screening was then performed on a total of 12 R. cataractae individuals from four distant populations (populations 1, 8, 19, and 29) to evaluate the polymorphism potential of the three remaining loci. The fifth exon of GH and the third exon of TRY showed polymorphism. No variation was observed for the second exon of TRY and, as such, it was discarded. The PRC conditions of the two loci selected are presented below.

For GH, the primers 5′-CCATTTGTATCTGCACAG-3′ and 5′-TACAGGCATTGACTAAC-3′ successfully amplified a segment of 255 bp that covered the entire fifth exon and 36 bp of the following intron. For TRY, primer pairs 5′-CGTCTGGGTGAGCACAAC-3′ and 5′-TCCCCATCCAGAGATCAGAC-3′ amplified a segment of 222 bp of the third exon. The PCR reactions and conditions that optimized amplification were the same for both sets of primers. The following reaction was used: a 12-5 μl reaction aliquot containing 1.5 mmol l−1 of MgCl2, 2.5 nmol l−1 of each dNTP, 0.2 unit of Taq polymerase, 1.25 μl of 10× Taq polymerase buffer (Invitrogen), and approximately 20 ng of DNA. Reaction conditions included an initial denaturation of 10 s at 92°C: followed by 45 cycles combining 15 s at 92°C, 10 s at 50°C, and 30 s at 68°C; and a final extension of 120 s at 68°C. The polymorphisms of both genes were evaluated on all longnose dace individuals following the same procedure as that used for MHC and as described above. Sequencing was performed for each individual, which showed variation in the migration pattern as explained above in the case of MHC.

The expression of the three coding regions delineated above was confirmed through cDNA amplification on two individuals. For each individual, the total RNA was isolated and purified using the RNeasy Mini Kit (QIAGEN) from 30 mg of muscle tissue previously stabilized in RNA stabilization reagent (RNAlater). Total RNA was then reverse-transcripted into single-strand cDNA using the ProtoScript® First Strand cDNA Synthesis Kit. The resulting cDNA products were used directly in several PCR to confirm the expression of the three genes analyzed. As we were using primers designed in different exons of a given gene, we expected the amplification product that covered an intron in the genomic DNA to be shorter when applied on cDNA. PCRs were performed on cDNA, genomic DNA, and negative templates.

Microsatellite Polymorphism

Populations were screened with seven microsatellite loci (Rhca15b, Rhca16, Rhca20, Rhca23, Rhca31, Rhca34, and Rhca52) according to conditions described by Girard and Angers (2006a, 2008). The polymorphism of the microsatellite markers was evaluated with 6% denaturing urea-polyacrylamide gel electrophoreses.

Statistical Analyses

Sequence Analyses

Long-term effects of natural selection are expected to leave a signal on gene sequences that would be unexpected under neutral evolution. To detect these potential effects, we first performed the McDonald and Kreitman (MK) test (McDonald and Kreitman 1991), available in DNASP 4.10 (Rozas et al. 2003), which evaluates the level of synonymous and non-synonymous substitutions within and between species. Under neutrality, the ratio between among species divergence and within species polymorphism should be equal for synonymous and non-synonymous mutations. Therefore, sequences of each exon were aligned using CLUSTAL W (Thompson et al. 1994). MK test was then performed using three (Genbank #AJ297822, BX539313, and BC092664) and five (Genbank #BC124461.1, AY103492.1, L04805.1, L04808.1, and L04815.1) homologous sequences of D. rerio for TRY and MHC, respectively, and five sequences from C. carpio (Genbank #AJ640136.1, AY553378.1, AY553379.1, AY822624.1, and DQ350436.1) for GH. These outgroup species were chosen because no fixed differences were observed between longnose dace and blacknose dace sequences, which is a condition required to compute the MK test.

Sequences were also analyzed using the HKA tests (Hudson et al. 1987), which explore potential balancing selection effects. The test predicts that, under neutrality, regions of the genome that evolve at high rates will present high levels of polymorphism within species. The test was performed between each pair of genes. For these analyses, the Genbank sequences of D. rerio #NW_001513985.1, NW_001513092.1, and NW_001511399.1 were used for MHC, GH, and TRY homologous sequences, respectively.

Gene Diversity Within Populations

Genetic diversity within populations was calculated using both allele frequencies and sequence information. The Nei’s gene diversity (H E) and the allelic richness (R) according to the rarefaction index of El Mousadik and Petit (1996) were computed using the program FSTAT v.2.9.3 (Goudet 2001). Molecular diversity indices within each sample were evaluated from the observed number of segregating sites (θS) and from the mean number of pairwise differences (θπ) using ARLEQUIN v3.1 (Excoffier 2006).

Short-term effects of natural selection are believed to modify allelic and genotypic diversities in unexpected ways: as such, when natural selection occurs, it is no longer possible to predict the genetic diversity from the demographic history and structure among populations. We evaluated these impacts using several strategies. First, R and H E computed on genes in each populations were correlated to those computed on non-coding microsatellite loci. Under neutrality, a significant positive correlation should be observed. The significance of this correlation was evaluated using a Pearson statistic tested with 999 random permutations of the data. Second, the effects of selection on genes were assessed by (1) examining their deviation from mutation/drift expectations and (2) using the Hardy–Weinberg equilibrium (HWE). These effects were first evaluated on microsatellites in each population. Analyses were performed using the BOTTLENECK program (Cornuet and Luikart 1996): an infinite allele mutation (IAM) model, and a two-phase mutation (TPM) model (the latter used proportions of single-step mutations of 0.25, 0.50, or 0.75 and a multistep change variance of 30 repeats) were employed. The temporal stability of the effective size of each population was further evaluated using the Wilcoxon sign-rank test (Luikart and Cornuet 1998). Deviations from HWE were tested using the exact test of Guo and Thompson (1992) implemented in the software GENEPOP v3.4 (Raymond and Rousset 1995). Populations showing significant departures from mutation/drift and/or HW equilibriums on microsatellites were removed from further analyses because of the impossibility to distinguish natural selection effects from those of local demographic processes in such populations.

In the remaining populations, the accordance with the mutation/drift expectations was evaluated on the three genes with the Ewen–Watterson neutrality test (Watterson 1978), the D of Tajima (1989), and the F of Fu and Li (1993). These analyses were performed overall and within each sample using the program ARLEQUIN v3.1. Subsequently, deviation from HWE was evaluated using the same method that was used for the microsatellites.

Gene Diversity Among Populations

Impacts of natural selection could also be potentially observed on the among populations differentiation, which can be significantly higher (directional selection) or lower (balancing selection) for genes under selection than for neutral markers. Using FSTAT, overall F ST was computed with MHC, GH, TRY, and microsatellites. The neutral expectations of the overall population differentiation for MHC, GH, TRY, and microsatellites were evaluated by the method proposed by Beaumont and Nichols (1996) and implemented in the software FDIST2. The average overall F ST across markers (microsatellites and genes) was used in the simulations. This value was targeted after an exploratory analysis based on the indications of Beaumont that can be found in the FDIST2 package. Accordingly, a robust starting F ST value should yield an output where the observations will be distributed almost equally below and above the average line. Simulations were performed with an infinite allele model. A total of 50,000 simulations were performed to obtain an acceptable compromise between precision and computation time. Each simulation was composed of 100 demes, from which a number of populations were sub-sampled to match the number of populations in the longnose dace dataset. Sample size was defined as the average number of gene copies among the samples. The observed F ST value for each marker was then compared to the simulation intervals according to their H E.

Effects of natural selection among populations were also explored using the Bayesian method implemented in BAYESCAN 2.01 (Foll and Gaggiotti 2008). The approach introduced the effect of selection by decomposing each F ST coefficient into a population component (beta) which is shared by all loci, and a locus-specific component (alpha) which is shared by all populations. Given the data, posterior probabilities were computed over two models for each locus: one that included selection (alpha + beta) and one that exclude it (beta). In this study, posterior probabilities were defined using 50,000 iterations, which were sampled every 10 iterations; each chain was preceded by 50 pilot runs of 5000 iterations and an additional burning period of 10,000 iterations. For each run, the stationarity of the posterior distribution of all parameters through the sampled iterations was assessed using the Geweke statistic (Geweke 1992) available in the boa R package (Smith 2007). Finally, departure from neutrality was assessed using the ratio of the posterior probabilities (posterior odds) of each model. The prior odds of the neutral model were set to 1 (minimum value) and 10 (default value). As explained in the BAYESCAN user manual, prior odds influence both the power of the method and the probability of finding false positives. Increasing the prior odds value will tend to eliminate these errors but does so at the cost of reducing the power of the method.

Results

Sequence Analyses

The analyses performed on total RNA confirmed the transcription of the MHC, GH, and TRY genes. In each case, the PCR products resulting from cDNA amplifications were shorter than those obtained from genomic DNA. As expected, the size of the cDNA amplification products were equal to that of their respective spliced genomic DNA, which strongly suggests that introns were removed and exons joined within the cDNA of all the three genes analyzed.

Only four distinct alleles were observed on the MHC exon among the longnose dace populations. Sequencing replicates (see Method) confirmed that SSCP and sequencing were reliable. These alleles (MHC-I to MHC-IV; Genbank #EU848571–EU848574) were characterized by a total of 11 polymorphic sites. Among them, five were parsimony-informative (125, 126, 135, 139, and 144). Alleles differed by an average of 6.3 substitutions. Ten substitutions resulted in amino acid replacements, and only one substitution was synonymous. The non-synonymous nucleotide diversity (0.032) was twice that of the synonymous nucleotide diversity (0.014). In spite of an over-representation of non-synonymous substitutions in polymorphic versus divergent sites (which provided an NI (neutrality index) of 3.75), the MK test was not significant (G = 1.46; P = 0.23).

Both GH and TRY showed low polymorphisms, which are mostly represented by synonymous substitutions. Again, sequencing replicates confirmed SSCP reliability. The GH exon was close to fixation among the longnose dace populations. Only two alleles (GH-I and GH-II; Genbank #EU836876–EU836877) that differed by a single non-synonymous mutation were observed, which resulted in a nucleotide diversity of 0.004. The MK test was not significant (G = 0.618; P = 0.43). Three alleles were observed on the TRY exon (TRY-I to TRY-III; Genbank #EU836882–EU836884). These alleles diverged by a single synonymous (117) and two non-synonymous (16 and 109) substitutions. The nucleotide diversity of the non-synonymous sites (0.008) was slightly lower than that of synonymous sites (0.013), and the MK test was not significant (G = 0.387, P = 0.533). In addition, none of the HKA tests provided significant results, suggesting that polymorphisms of the MHC, GH, and TRY genes were not under selection.

The minimum spanning network of MHC alleles revealed that MHC-I and MHC-II are closer to blacknose dace alleles (average of 3.5 substitutions) than to MHC-III and MHC-IV (average of 8 substitutions; Fig. 1a). Interestingly, the same pattern was observed for GH and TRY, whereby one allele of each gene showed the identical sequence in both species (Fig. 1b, c).

Microsatellite Diversity Within Populations

Seven populations (9, 12, 13, 18, 26, 27, and 31) were not in accordance with HWE (Table 1). However, only three populations (12, 13, and 18) showed a highly significant deviation (P < 0.001) for more than one locus. Consequently, only these populations were considered as deviating from panmixia and were removed from the further analyses of genetic diversity within populations.

According to the Wilcoxon rank-tests, three populations (3, 6, and 9) were in discordance with mutation–drift equilibrium under three out of four sets of mutation assumptions with P < 0.05. All of the other populations appeared to be in accordance with mutation–drift equilibrium under all mutation assumptions, except populations 16 (IAM) and 26 (TPM with a proportion of single-step mutations of 0.75); for both of these populations, the analysis was significant at the 0.05 threshold. This discordance with demographic equilibrium was, however, not considered as severe as that of the three previous populations. These two populations were thus retained for further analyses.

Gene Diversity Within Populations

Genetic diversity within populations varied significantly among the three functional genes (Table 1). For instance, when R was computed on MHC, GH, and TRY, the resulting average was 2.4, 1.2, and 1.6 alleles by population, respectively. This difference is highly significant (F = 21.19; P < 0.0001). Significant differences were also observed for H E (F = 14.34; P < 0.001), θS (F = 41.27; P < 0.0001), and θπ (F = 30.93; P < 0.0001). GH and TRY were either fixed or in accordance with both HW and mutation–drift equilibriums (Table 1). In contrast, the MHC exon showed an excess of homozygotes (27 and 29), an excess of diversity (8), or both (2 and 14). That most of the populations were at HWE suggests the absence of null alleles for these genes. Results obtained with the D of Tajima were almost never significant, with the exception of two populations for MHC (7 and 24) and one for TRY (26; Table 1). Furthermore, the overall D for MHC (1.34; P = 0.08), GH (−0.40; P = 0.31), and TRY (0.81; P = 0.19) was not significant.

The F of Fu and Li showed high discordances between MHC and the two other genes. Nine populations (hereafter referred to as the “F SIGN subset”) showed a higher molecular diversity between alleles than expected under neutrality. In contrast, 10 populations (the F NS subset) followed the neutral expectation (Table 1). The populations from the F SIGN subset also showed a higher number of non-synonymous substitution differences between each pair of alleles. Indeed, between 1.24 (pop. 15) and 1.75 (pop. 1) non-synonymous substitutions were observed within the F SIGN subset as compared to between 0.27 (pop. 16) and 1.29 (pop. 30) in the F NS one (Fig. 2). The difference between the averages was highly significant (t = 7.591; P < 0.0001). This was not observed within populations (Table 1) or within over all populations for either GH (F = −0.42; P = 0.21) or TRY (F = 2.76; P = 0.14).

Fig. 2
figure 2

Average number of nonsynonymous substitutions between each pair of MHC alleles within populations of F SIGN (gray) and F NS (white) subsets. Average numbers of nonsynonymous substitutions between each pair of MHC alleles within subsets (dashed lines) and overall averages (solid lines) are also presented

Full size image

The MHC richness was significantly correlated to the richness of microsatellites (r = 0.42; P < 0.05; Fig. 3a). While the correlation remained significant when only the F NS populations were taken into account (r = 0.53; P < 0.05; Fig. 3a), a complete absence of correlation was observed when the F SIGN subset (r = −0.002; P = 0.55) was considered. Similar trends were observed for H E (Fig. 3b). Owing to its low level of diversity, the results of GH were not very informative (Fig. 3c, d); yet it is worth noting that TRY indices of diversity displayed strong correlations with those of microsatellites. In this context, the results of TRY were similar to those observed on MHC on the F NS subset (Fig. 3e, f).

Fig. 3
figure 3

Relationships between genes and microsatellites R and H E within each population. In the MHC figures, white squares and black circles represent populations within F SIGN and F NS subsets, respectively. Correlation values were calculated using all of the populations

Full size image

Diversity Among Populations

The neutral expectations of the overall population differentiation assessed using the simulation procedures of FDIST2 showed that all microsatellites were within the F ST 0.99 confidence interval, regardless of whether F SIGN and F NS populations were considered together (Fig. 4a) or separately (Fig. 4b, c). When all populations are taken into account, MHC, GH, and TRY all fell within the neutral F ST interval (Fig. 4a). The F ST values computed over the F NS subset for MHC and TRY were also in accordance with neutral expectations (Fig. 4c). Computation could not be performed for GH because it was monomorphic in this population subset. However, while GH and TRY showed an F ST value in the neutral interval in the F SIGN subset, the F ST computed on MHC was clearly below the lower limit of the neutral confidence interval (Fig. 4b).

Fig. 4
figure 4

Average (bold lines), 95% (solid lines), and 99% (dashed lines) confidence intervals for the relationships between F ST and H E according to FDIST2 simulations performed with a sample of 20 (a), 9 (b), and 11 (c) populations. Observed values of the seven microsatellites (empty circles) and the three genes (filled circles) computed overall F SIGN and F NS (a), only F SIGN (b), and only F NS (c) are superimposed on these simulation results. GH result is not presented in c because it was monomorphic within this subsample

Full size image

The BAYESCAN approach provided the same results when prior odds for the neutral model were set to 1. The approach was not powerful enough to detect selection on any loci when prior odds were set to 10 (results not shown). While F ST of MHC was not considered an outlier when either all populations (Fig. 5a) or only F NS subset (Fig. 5c) were taken into account, its value was above the neutrality maximum threshold when only the F SIGN populations were considered (Fig. 5b). The two other genes remained in accordance with the neutral expectations (Fig. 5a, b, c). The effect of selection on MHC across population subsets is also visible through the posterior probability density of its alpha parameter. Centered at 0 when all populations (Fig. 5d) and F NS subset (Fig. 5f) are considered, the distribution shifts toward an average of −1 when considering only the F SIGN subset (Fig. 5e). According to Foll and Gaggiotti (2008), this negative value is associated with balancing selection. Selection was also diagnosed on two microsatellites (Rhca34 and Rhca52). Interestingly, the F ST values of these two microsatellites were also positioned close to the upper bound of the 0.99 confidence interval defined by FDIST2; this is an additional indication that the two methods converged to the same results.

Fig. 5
figure 5

Posterior odds in favor of the selection model (PO) for each microsatellite (empty circles) and gene (filled circles) according to the BAYESCAN approach (whereby the locus-specific F ST is on the y-axis) computed overall (a), only F SIGN (b), and only F NS (c) subsets. Bold lines correspond to the PO thresholds leading to a false discovery rate of no more than 5%. These thresholds were computed using the R function provided with the BAYESCAN package. GH was removed from the F NS subset analysis because it was monomorphic within this subsample. Panels (d) to (f) correspond to the posterior distributions of the MHC alpha parameter according to BAYESCAN when performed overall (d), only on F SIGN (e), and only F NS (f) subsets

Full size image

Discussion

In this study, we presented an extensive comparison between neutral markers and coding genes among several populations to determine the effect of the short- and long-term evolutionary processes that shape the diversity of specific functional genes in longnose dace over a territory once covered by Pleistocene glaciations. For all three exons, the allelic richness was extremely low. While these results were not as surprising for GH and TRY, which are probably under the action of directional and/or purifying selection, the few distinct alleles (four alleles) found on the MHC exon were astonishing. Because of the potential importance of co-evolution in host–pathogen systems such as the one involving MHC, allelic diversity is believed to be of prime importance for the viability of populations (Bernatchez and Landry 2003). The MHC allelic richness observed in this study for the longnose dace, while being comparable with a few other studies (see Peters and Turner 2008; and references therein), remains surprisingly low compared with the majority of studies performed on MHC class ΙΙβ (Froeschke and Sommer 2005: 58 individuals, 20 alleles; Aguilar and Garza 2006: 450 individuals, 88 alleles; and Ottovà et al. 2007: 30 individuals, 48 alleles; and others), especially considering the high number of individuals analyzed (542). We acknowledge the fact that some rare MHC alleles may have been missed, as shown by the four populations with slight excesses of homozygotes. Nevertheless, we are confident that the most frequent MHC lineages were detected because (1) amplifications in this study resulted in at least one and no more than two bands in every individual analyzed, and (2) most populations were under HWE. Furthermore, the absence of STOP codons and the exon-coding expression revealed by the total RNA examination confirmed that the segment selected was not part of a pseudogene.

Long-term Processes

The sequence analyses performed in this study revealed few clues about the long-term impacts of natural selection on any of the genes surveyed. While the high ratio of non-synonymous/synonymous substitutions that was observed on MHC could be in accordance with a historical effect of diversifying selection (Hughes and Nei 1988; Froeschke and Sommer 2005; Schad et al. 2005; Aguilar and Garza 2006), the MK test remains non-significant, and the effect of stochastic mutations cannot be rejected. The HKA tests also failed to detect other forms of selection in any of the genes studied. Biological interpretations of these results remain difficult to perform since the restricted number of alleles observed within our population set has likely considerably decreased the power of these analyses.

The incomplete lineage sorting observed between longnose and blacknose daces for all the three genes must also be interpreted with caution. The maintenance of alleles (or allele lineages) through speciation events is often associated with the action of natural selection (Mayer et al. 1992; O’Brien and Yuhki 1999; Bryja et al. 2006). However, conservations of primer sites and ranges of overlap of allelic size between both species were also observed on several microsatellite loci expected to be neutral (Girard and Angers 2006a). Such similarity throughout the genome is not in accordance with selection and thus suggests that other evolutionary processes may have been operating. Potential hybridization between both Rhinichthys species (while never reported to our knowledge), large effective population size, or recent speciation appears more likely to explain incomplete lineage sorting between both the species (Nei 1987; Pamilo and Nei 1988; Takahata 1989).

Short-term Processes

Despite not having detected many alleles, analyses performed to identify short-term selection processes were successful, at least with regard to the MHC exon. One of the most striking results of this study is the separation of populations into two different subsets, which was suggested by the Fu and Li statistic. The genetic and molecular diversities of MHC appeared to be strongly related to those observed on neutral markers in almost half of the sampled populations (F NS populations). We cannot reject the hypothesis that the evolution of the MHC exon in these populations resulted mainly from random effects associated with demographic processes following the postglacial colonization. Campos et al. (2006) found similar results in brown trout (Salmo trutta) populations, where genetic drift had appeared to become the predominant evolutionary force shaping genetic variation and had eroded the effect of selection.

On the other hand, the MHC gene and molecular diversities of nine populations (F SIGN subset) appear to depart from neutral expectations. Significant F tests, lower population differentiation than expected, and the absence of a correlation between the diversity of MHC and microsatellites appear to reflect the effects of a balancing selection process. The fact that these results were not observed on the other genes (especially TRY) and microsatellites increases the plausibility of the hypothesis that specific selective evolutionary processes acted on MHC at least in these populations.

Interestingly, these populations have the highly divergent allele MHC-III in common. The combination of this allele with either MHC-I or MHC-II is clearly the reason why the Fu and Li test was significant in these specific populations. Despite their similar MHC composition, these populations are spread across the entire sampling area (Fig. 1a); this is not expected under neutral evolution considering the strong effects of post-glacial colonization on the geographic organization of the genetic diversity of these populations (Girard and Angers 2006b). The fact that the presence of the MHC-III allele has few relationships with the allopatric fragmentation pattern of the species may indicate the effects of non-neutral mechanisms, which seek to maintain high genetic diversity on this exon by keeping this highly divergent allele.

Nevertheless, the mechanisms behind the maintenance of the MHC-III allele have yet to be identified. In these populations, if selection was apparent on allelic frequencies, then all genotype frequencies were in accordance with neutral expectations. Consequently, if mechanisms such as overdominance, diversifying selection through heterozygote advantage, or disassortative mating did occur, they were not detected. While both small sampling sizes and low number of alleles could have reduced the power to detect these biological mechanisms, such discordance between selectively allelic and neutral–genotypic frequencies has already been observed on MHC (Hayashi et al. 2005). This discordance was interpreted as the result of natural selection on the long-term evolution of MHC, and drift was understood as the main process impacting short-term evolution. However, there is a clear necessity for additional investigations to ensure that current failure to detect either selection or the biological mechanisms guiding selection are not solely due to low gene diversity or sampling sizes.

In conclusion, there is no doubt that the lack of diversity observed considerably reduced our capacity to interpret the shaping mechanisms behind functional genes of longnose dace over the surveyed area. However, the depauperate allelic pool being observed on all genes has likely been influenced by the important diversity reduction during post-glacial colonization, which may have overcome the effects of natural selection. The fact that few alleles were observed (and seemed to evolve neutrally in several populations) may temper the importance of functional gene diversity (especially on MHC) on population viability, as suggested in previous study (Campos et al. 2006; Babik et al. 2009). However, while our study underlines the major impact of post-glacial colonization on the diversity of functional genes, it is noticeable that even with such a low number of alleles, locally dependant balancing selection was detected on MHC; this supports the idea that these specific genes evolve under strong selective evolutionary mechanisms. Nonetheless, the lack of diversity observed so far in this species on both neutral and non-neutral markers over the wide area surveyed is evident and may affect the persistence of these populations in the face of future environmental changes. Considering the significant imprint of the Pleistocene glaciations throughout the northern hemisphere, further studies on functional gene diversity appears appropriate to evaluate and generalize their impacts on the persistence of the numerous species affected by these major disturbances of the past.