Multiple parasites mediate balancing selection at two MHC class II genes in the fossorial water vole: insights from multivariate analyses and population genetics


Nathalie Charbonnel, CBGP – Campus International de Baillarguet, CS 30016, 34 988 Montferrier s/Lez, France. Tel.: +33 4 99 62 33 02; fax: +33 4 99 62 33 45; e-mail:


We investigated the factors mediating selection acting on two MHC class II genes (DQA and DRB) in water vole (Arvicola scherman) natural populations in the French Jura Mountains. Population genetics showed significant homogeneity in allelic frequencies at the DQA1 locus as opposed to neutral markers (nine microsatellites), indicating balancing selection acting on this gene. Moreover, almost exhaustive screening for parasites, including gastrointestinal helminths, brain coccidia and antibodies against viruses responsible for zoonoses, was carried out. We applied a co-inertia approach to the genetic and parasitological data sets to avoid statistical problems related to multiple testing. Two alleles, Arte-DRB-11 and Arte-DRB-15, displayed antagonistic associations with the nematode Trichuris arvicolae, revealing the potential parasite-mediated selection acting on DRB locus. Selection mechanisms acting on the two MHC class II genes thus appeared different. Moreover, overdominance as balancing selection mechanism was showed highly unlikely in this system.


Major histocompatibility complex (MHC) genes are ideal candidates for the study of selection in natural populations (Bernatchez & Landry, 2003). The way in which their extreme polymorphism evolved remains a matter of debate (review in Piertney & Oliver, 2006). Balancing selection has frequently been detected on these genes at the macroevolutionary timescale, as shown by studies of substitution patterns or trans-species polymorphism (e.g. Musolf et al., 2004; Bryja et al., 2006). Evidence has recently been found for selection acting over the microevolutionary timescale, based on comparisons of population genetic structures for MHC and neutral genes (e.g. Landry & Bernatchez, 2001; Charbonnel & Pemberton, 2005; but see Seddon & Ellegren, 2004). Contemporary selection can also be detected by investigating associations between MHC alleles and infection status. Parasite-mediated balancing selection (PMBS) has long been suspected because of the key immune function of these genes (foreign peptide recognition triggering specific immune response, Hughes & Yeager, 1998; Delves et al., 2006) and the ubiquitous nature of this immune function in different taxa (Klein & O’Huigin, 1994). Associations have thus frequently been reported in the context of human infectious diseases (Hill et al., 1991; Thursz et al., 1995; Carrington et al., 1999) or for a single parasite species in wild host populations (Briles et al., 1977; Langefors et al., 2001; Bonneaud et al., 2006). This approach may not meet the requirements of evolutionary studies, due to the difficulties involved in selecting a relevant, single, parasite species. Moreover, evidence suggests that multiple parasites are required to drive MHC polymorphism (McClelland et al., 2003; Wegner et al., 2003). Given this constraint, associations between MHC genes and infection have been explored focusing on parasite richness or burden (Prugnolle et al., 2005; Schad et al., 2005). This approach is more satisfactory but is also subject to certain limitation, as such indices of parasite pressure do not reflect the particular selection mediated by each parasite species (especially as concerns antagonistic effects between species, e.g. Meyer et al., 1994). The associations between MHC alleles/genotypes and the specific composition of the parasite community have, to our knowledge, never been studied.

In this study, we aimed to determine whether balancing selection was mediated by multiple parasites and to identify the parasite species acting on MHC polymorphism. The fossorial water vole (Arvicola scherman, Arvicolinae, Cricetidae) is a relevant model for exploring this question, because it suffers from only moderate numbers of macroparasite infections (Cerqueira et al., 2007), making it possible to carry out almost exhaustive screening. We considered gastrointestinal helminths, brain coccidias and viral agents of zoonoses. MHC class II genes were studied, as they encode proteins that present exogenous antigens to T helper cells, and are therefore involved in the recognition of extracellular parasites (Hughes & Yeager, 1998). Experiments on MHC class II-deficient mice have shown that MHC class II genes are also involved in immunity against viruses (Apanius et al., 1997). DQA and DRB are the most studied MHC class II genes in small mammals (Charbonnel et al., 2006). Both genes were included in this study, and the second exon of these genes was studied, as it encodes the peptide-binding site (PBS, in humans: Brown et al., 1993; and in voles: Bryja et al., 2006). Associations between host genetics and parasite load have already been reported for the DRB gene in several wild populations (Froeschke & Sommer, 2005; Schad et al., 2005; Schwensow et al., 2007). In Arvicolinae (voles and lemmings), evidence has been obtained for both historical and contemporary balancing selection acting on DQA and DRB, using phylogenetic and population genetic methods (Bryja et al., 2006, 2007; Oliver & Piertney, 2006).

The specific objectives of this study were to explore associations between genetics (MHC class II DQA and DRB genes, in terms of alleles and heterozygosity, and global genomic heterozygosity) and parasitism (parasite species and parasite richness) in natural populations of water vole. Multivariate analyses made it possible to consider each parasite species as a potential agent mediating selection while minimizing type I statistical errors due to multiple testing. Detecting associations would reveal the potential action of parasites as selective pressure. It would also help in discriminating between two possible mechanism of PMBS: (i) negative frequency dependence (or rare allele advantage, Doherty & Zingernagel, 1975), which requires associations between parasites and specific alleles; and (ii) overdominance (or heterozygote advantage, Takahata & Nei, 1990), which assumes associations with MHC heterozygosity or genomic heterozygosity (in the latter case, a mechanism of sexual selection would also be involved, Penn, 2002). We also used intrapopulation genetic analyses to detect balancing selection acting in the history of populations, and to evidence heterozygote excess, an argument favouring the hypothesis of overdominance.

Materials and methods

Species, sampling design and parasite screening

The fossorial water vole (A. scherman) is an arvicoline species that typically inhabits grasslands and experiences multiannual density cycles in various regions of France and Switzerland (Saucy, 1994). We sampled water vole populations at six sites in eastern France (Nozeroy, Jura, Franche-Comté, France, Fig. 1), during three successive years (2003, 2004 and 2005). A group of animals sampled from a single site in a particular year is referred to as a population. Only three populations are similar to those studied by Bryja et al. (2007): A = 03, B = 05, C = 06, in 2003. Sampling was carried out in October, as the prevalence levels of most viruses or helminth parasites of mammals globally increase from spring to autumn (e.g. Haukisalmi et al., 1988). This has been confirmed in these water vole populations considering viruses (Cowpox virus, Deter et al., unpublished data) and nematodes (Trichuris arvicolae and Syphacia nigeriana, Deter et al., 2007a). Autumn sampling is thus expected to reflect the highest potential pressure exerted by parasites and pathogens on water vole populations. The areas sampled each covered about 2 ha and were separated by 4 to 13 km. We used 80 Sherman live traps per site, but only one live trap per colony (i.e. nest), to prevent the sampling of related individuals. This led to the capture of about 20 animals per site per day of sampling. If more than 24 voles were trapped, we simply took a tail sample from the additional animals for genetic analyses and then released them. Voles were brought to the laboratory and housed individually for 1 week in standardized conditions for the need of physiological experiments (Charbonnel et al., 2008). Voles were then killed by cervical dislocation, as recommended by Mills et al. (1995), weighed and dissected. Only sexually mature individuals were retained for the analysis. Pregnant females and females that gave birth in the laboratory were excluded from the analyses. Toe tissue was kept in 95° ethanol for the extraction of DNA. Blood samples were taken from the heart and centrifuged. Serum samples were screened for IgG antibodies against Puumala virus (PUUV, Hantavirus), Cowpox virus (CPXV, Orthopox virus) and Lymphocytic choriomeningitis virus (LCMV, Arenavirus), by means of immunofluorescence assays (IFA), as described by Kallio-Kokko et al. (2005). Brains were crushed between slides and checked under a microscope for the presence of Frenkelia cysts (Coccidia, Apicomplexa: Sarcocystidae). Helminths from the liver and the body cavity were identified during dissection of the vole. Finally, the digestive tracts were stored in 95° ethanol before dissection. Helminths were identified under the microscope. The diseases caused by helminths are generally more dependent on infection intensity than on infection status (May & Anderson, 1979). We therefore counted the individuals detected for each helminth species when possible. Parasite richness is expressed as the number of parasite species detected per vole. We considered both helminth richness (Rh) and total richness (Rp), which included viruses and coccidia.

Figure 1.

 Location of the sampling sites. On the left, situation of the Jura Mountains (Franche-Comté) in France. On the right, map of the Nozeroy canton with the six sampling sites indicated.

MHC class II gene and microsatellite genotyping

Genomic DNA was extracted from a single vole toe, using the DNeasy® Tissue Kit (Qiagen, Hilden, Germany), according to manufacturer’s instructions, using 400 μL of elution buffer.

We amplified the complete exon 2 of DQA and part of the exon 2 of DRB for genotyping, as described in Bryja et al. (2005, 2007). Primers were fluorescently labelled with different dyes (FAM, HEX or NED). Capillary electrophoresis-single strain conformation polymorphism (CE-SSCP) analysis was performed on a MegaBACE 1000® DNA analysis system (Amersham Biosciences, Buckinghamshire, UK) (see details on the method in Bryja et al., 2005). The DQA gene is duplicated in water voles and the second copy (DQA2) is not present in all individuals (see Bryja et al., 2006). We identified two alleles at this locus by cloning and sequencing, but these alleles could not be distinguished by CE-SSCP. We therefore decided to consider only the polymorphism of DQA duplication, namely the presence/absence of DQA2.

All individuals were genotyped at nine polymorphic autosomal microsatellite loci specifically developed in A. scherman, as described by Berthier et al. (2004). This made it possible to infer the effects of neutral evolutionary processes on MHC gene polymorphism.

Detecting genetic/parasitological associations using co-inertia analyses

We used multivariate analyses to investigate relationships between: (i) an MHC class II genetic matrix including, for each individual screened, the presence/absence for each allele of the DQA1 and DRB genes, DQA1 and DRB heterozygosity (0 = homozygote and 1 = heterozygote), the presence/absence of DQA2, microsatellite heterozygosity (0 = low level of heterozygosity and 1 = high level of heterozygosity); and (ii) a parasitological matrix including, for each individual screened, the abundance of each parasite species and parasite richness indices (Rh and Rp).

We analysed jointly all vole individuals providing from 15 samplings. Indeed, genetic differentiation among them is expected to be very low in this system as the sites sampled are geographically very close (Fig. 1), and as sampling was carried out during the ‘increase’ and ‘outbreak’ phases of the cycle (years 2003–2004–2005, see Charbonnel et al., 2008), when the effective size of water vole populations was very high (Berthier et al., 2006; Bryja et al., 2007). However, as spatial structure may increase the probabilities of missing or identifying false parasitological/genetic associations, we checked for genetic structure among years and localities, computing global FST estimates over all microsatellites, and testing global genetic differentiation using Genepop 3.3 software (Raymond & Rousset, 1995).

Multivariate methods have been widely used in ecology, particularly for the study of species/environment relationships (Dolédec & Chessel, 1994), but have not previously been used in studies of genetic associations. Considering the large number of genetics and parasitological variables included here, more conventional univariate statistics would rely on hundreds of tests and would thus require corrections for multiple testing. The acceptable type I error in all univariate statistical tests would thus be very low (< 10−4) and such a method would be very conservative. In this way, multivariate approach revealed particularly relevant, avoiding the increase of type I errors. We carried out co-inertia analysis (ACO), searching for axes that maximize the covariance between the row coordinates of two matrices (Dray et al., 2003). We chose this method as it can be used with all types of variables and is robust to correlation between variables from a given matrix (Dray et al., 2003). This analysis involves two steps. Each matrix is first analysed independently – the MHC genetic matrix by correspondence analysis (CA) and the parasitological matrix by principal component analysis (PCA, correlation matrix), using the row weights of the CA. These analyses can be displayed on factorial maps, to evidence particular distributions of alleles or of parasites potentially affecting the interpretation of associations between genetics and parasites. The global relationship between the two matrices was then estimated from the co-inertia (vectorial correlation) of the matrices. Relationships between the two matrices were assessed by comparing the co-inertia estimated from the real data set with the co-inertia distribution estimated after 1000 permutations of the parasitological matrix lines. Particular associations were detected by eye from the distribution of genetic and parasitological variables on the ACO factor map. Variables located in a given direction with respect to the origin were considered to be positively associated, whereas those located in the opposite direction were considered to be antagonistic. Variables close to the origin were considered to be unrelated to the variables of the other matrix. These associations were then tested statistically, using a cross-validation approach (Draper & Smith, 1981; see, e.g. Hallgren et al., 1999). We randomly split the data into two independent data sets and carried out two independent ACO. Associations between genetic and parasitological data were considered significant if they were observed in both data sets. This approach also prevents from corroborating associations, which would only be artefacts of population structure. The relative risk (RR) associated with these genetic variables of interest was estimated as described by Haldane (1956). All multivariate analyses were carried out with ADE-4 software (Thioulouse et al., 1997).

According to the frequency-dependent selection hypothesis, we expected to find positive or negative associations between the presence of specific MHC alleles and parasite species presence/abundance. Negative associations may be interpreted as indicating alleles conferring resistance to the parasite species, whereas positive associations indicate susceptibility to the parasite species. Alternatively, in situations of overdominance, negative associations should be observed between estimation of genetic diversity (heterozygosity at MHC genes, at microsatellites or MHC locus duplication) and specific infection or parasite richness.

Within-population genetic analyses

As demography affects all the genome, whereas selection acts on a single locus, comparisons of patterns for MHC genes and microsatellites may provide evidence for balancing selection (Cavalli-Sforza, 1966; Luikart et al., 2003). We first tested for linkage disequilibrium for each pair of loci, including nine neutral microsatellite loci, DQA1 and DRB, in each population. We used exact tests and Markov chains, as implemented in Genepop (Raymond & Rousset, 1995). Corrections for multiple tests were performed using the false discovery rate (FDR) approach, as described by Benjamini & Hochberg (1995). We then described genetic diversity for the DQA1 and DRB genes, and for all the microsatellites considered together, in each vole population. We calculated expected and observed heterozygosity (HE and HO), using Genetix software (Belkhir et al., 1996–2004) and the unbiased estimator inline image of Wright’s inbreeding coefficient FIS, as described by Weir & Cockerham (1984), using Fstat 2.9.3 (Goudet, 1995).

Homogeneity in allelic frequencies is expected not only for loci under balancing selection (Watterson, 1978), but also for all the genome after a recent demographic bottleneck (Watterson, 1986). Such demographic events are likely to occur in water vole populations, which periodically decline to low densities (Saucy, 1994; Charbonnel et al., 2008). Under balancing selection, a significant deviation from mutation–drift equilibrium in MHC genes is expected, whereas microsatellite loci would be at equilibrium. For each locus, we tested for deviation of gene diversity from equilibrium using Bottleneck 1.2.02 program (Piry et al., 1999), simulating 10 000 data sets. Contrary to Ewens–Watterson test (Watterson, 1978), this test allows a variety of mutation model (Piry et al., 1999). We used an infinite allele mutation (IAM) model for MHC genes and a generalized stepwise mutation (GSM) model for microsatellites (with the variance of the geometric distribution fixed at 0.36, Estoup et al., 2001). Sequential Bonferroni corrections (Rice, 1989) were used to correct for multiple testing.

If overdominance occurs, selection should result in an excess of heterozygotes at MHC genes (Doherty & Zingernagel, 1975; Garrigan & Hedrick, 2003), with the other loci at equilibrium. This heterozygote excess might be detected through the demonstration of Hardy–Weinberg disequilibrium at MHC class II genes, with equilibrium at microsatellites (disequilibrium at all loci would reflect outbreeding, which may be MHC independent). An exact test of heterozygote excess was performed, using the Markov chain methods implemented in Genepop 3.3 to test this hypothesis.


Data collection

We sampled 302 water voles for parasite screening and 367 for population genetic analyses. In 2005, the three sites located in the north of the study area had such low population densities that it was not possible to trap any animals over a period of several days.

We tested for antibodies against PUUV, CPXV and LCMV. Seroprevalence results are shown in Table 1. Two coccidian species were observed in the brain (Frenkelia microti and Frenkelia glareoli), four nematodes (T. arvicolae, S. nigeriana, Aonchotheca sp. and Eucoleus bacillus) and four adult cestodes (Anoplocephaloides dentata, Paranoplocephala gracilis, Paranoplocephala omphalodes and Arostrilepis cf horrida) were detected in the digestive tract. Two larval cestodes were observed in the liver (Echinococcus multilocularis and Taenia taeniaeformis) and one in the body cavity (Taenia crassiceps). Detailed prevalence values are shown in Table 1. Individual helminth richness ranged between 0 and 5 and total parasite richness between 0 and 7. As the helminths E. bacillus, T. crassiceps and A. dentata were each found in only one individual (Table 1), we could not include these species in the association analyses.

Table 1.   List of viruses and macroparasites identified in the 302 water voles screened.
  1. Nem-, Cest and Cocc- refer to nematodes, cestodes and coccidia respectively. Np is the number of infected (macroparasites, protozoans) or seropositive (viruses) animals identified in the whole data set. Prevalence estimates are presented as percentages. An asterisk indicates parasites not included in co-inertia analyses because their prevalence was too low.

 Orthopox virusCowpox virusCPXV6019.60
 ArenavirusLymphocytic choriomeningitis virusLCMV123.92
 SarcocystidaeFrenkelia glareoliCocc-Fg4414.38
Frenkelia microtiCocc-Fm72.28
 NematodaSyphacia nigerianaNem-Sn8527.77
Trichuris arvicolaeNem-Ta13243.14
Aonchotheca sp.Nem-Asp278.82
Eucoleus bacillusNem-Eb*10.33
 Cestoda larvaeEcchinococcus multilocularisCest-Em268.49
Taenia taeniaeformisCest-Tt5618.3
Taenia crassicepsCest-Tc*10.33
 Cestoda adultsAnoplocephaloides dentataCest-Ad*10.33
Paranoplocephala gracilisCest-Pg41.31
Paranoplocephala omphalodesCest-Po7323.86
Arostrilepis cf horridaCest-Ah278.82

We identified five alleles at DQA1 and 12 alleles at DRB in the 367 water voles screened for DNA analyses. All these alleles have previously been detected, cloned and sequenced in water voles (Bryja et al., 2006, 2007). Two DRB alleles (Arte-DRB-12 and Arte-DRB-14) were found at very low frequencies (< 0.01) and so could not be included in association analyses.

Co-inertia analyses

Global genotypic differentiation over years and localities was low (FST = 0.017, < 0.001), especially compared with other A. sherman studies in closed localities (FST estimates ranged between 0.015 and 0.037; Berthier et al., 2006; Bryja et al., 2007).

Co-inertia analysis was first carried out for the 302 individuals screened for both the 19 genetic (five DQA1 alleles, 10 DRB alleles, DQA1, DRB and microsatellite heterozygosity, presence of DQA2) and 15 parasitological variables (three viruses, two coccidias, eight helminth species, helminth and total parasite richness). The binary variable reflecting individual microsatellite heterozygosity was obtained using the proportion of microsatellite locus being heterozygote. This variable thus makes out as ‘high level of heterozygosity’ the individuals having at least eight heterozygous microsatellites out of nine (21% of the individuals). As the choice of the threshold may influence the results, we also performed this analysis considering as ‘highly heterozygote’ the individuals presenting at least six heterozygous loci (76.8% of the individuals).

The CA based on the genetic table (Fig. 2) revealed that the two first axes accounted for 26.1% of the total variance (14.0% F1 and 12.1% F2). The first axis (F1) is structured by the alleles Arte-DQA-06 and Arte-DRB-09 opposed to Arte-DQA-07, Arte-DRB-16 and Arte-DRB-10. The second axis (F2) mainly opposes Arte-DRB-11 to Arte-DRB-13 (and Arte-DRB-15 to a lesser extent). Neither heterozygosity nor duplication makes a major contribution to the structure of this data set. The first two dimensions of the PCA performed on the parasitological table (Fig. 3) accounted for 29.4% of the total variance (18.4% F1 and 11.0% F2). The first axis is structured by the two parasite richness indices: Rp and Rh. The second axis opposes Orthopox virus (CPXV) and Arenavirus (LCMV) and F. glareoli (Cocc-Fg) to the two nematodes located in the caecum, T. arvicolae (Nem-Ta) and S. nigeriana (Nem-Sn).

Figure 2.

 Correspondence analysis (CA) of the genetic table including 19 variables and 302 voles. The variables DQA1 and DRB alleles coded by name, heterozygosity of DQA1 (Het-DQA1), DRB (Het-DRB) and microsatellites (Het-MS) and DQA duplication (Dupl) are projected on the F1 × F2 map. Variables reflecting genetic diversity are shown in bold and those structuring the data set are underlined using solid (F1) or dotted (F2) lines.

Figure 3.

 Principal component analysis (PCA) of the parasitological table including 15 variables and 302 individuals. The variables are projected on the F1 × F2 correlation circle. Nem-, Cest- and Cocc- correspond to nematodes, cestodes and coccidia respectively. The next two letters refer to the code used to identify species (see Table 1). Viruses are coded using CPXV for Orthopox virus, LCMV for Lymphocytic choriomeningitis virus and PUUV for Hantavirus (undetermined species). Rh and Rp correspond to helminth and total parasite richness respectively. Variables structuring the data are underlined: with solid lines for variables structuring F1 and dotted lines for those structuring F2.

We found no overall relationship between the genetic and parasitological matrices (global co-inertia = 0.112, = 0.84), although this does not exclude the existence of particular associations. Similar results were obtained when this ACO was carried out with each MHC gene independently (DQA1: = 0.170 and DRB: = 0.328).

The two first axes of the ACO accounted for 46.3% of the variance shared between the two matrices (31.2% F1 and 15.1% F2). The genetic variables were plotted on the F1 × F2 ACO factor map (Fig. 4a), and four alleles were identified as principally responsible for co-inertia. All were alleles of the DRB gene (F1: Arte-DRB-11, Arte-DRB-15 and F2: Arte-DRB-07, Arte-DRB-11, Arte-DRB-16). The heterozygosity of DQA1, DRB and microsatellites, and the presence of DQA2, were located very close to the origin on the co-inertia factor maps (Fig. 4a), indicating the absence of linkage between these four variables and parasitism. When considering as ‘high level of heterozygosity’ the individuals having at least six heterozygote loci, similar results were observed, confirming that microsatellite heterozygosity is not associated with parasitism. Richness indices, the nematode T. arvicolae and, to a lesser extent, the two viruses CPXV and LCMV, were all involved in co-inertia (Fig. 4b), and were previously found to structure parasitological data in the PCA. The cestodes P. omphalodes and A. horrida, which were not identified as important in the PCA, strongly structured the co-inertia (first and second axes respectively).

Figure 4.

 Co-inertia analysis (ACO) of the genetic and parasitological matrices. Projection on common ACO axes of (a) the genetic variables and (b) the parasitological variables, encoded as in Figs 2 and 3 respectively. Variables located in a given direction relative to the origin may be considered to be positively associated (susceptibility), whereas those located in the opposite direction may be considered to be negatively associated (resistance). Associations between genetic and parasite variables visually detected on the F1 axis are underlined with solid lines, whereas those detected on the F2 axis are underlined with dotted lines. Associations shown to be significant by cross-validation are surrounded. Variables reflecting genetic diversity are indicated in bold.

Superimposing the genetic and parasitological variables on this ACO factor map revealed positive (negative) associations on the first axis between Arte-DRB-11 (Arte-DRB-15) and richness indices, P. omphalodes and T. arvicolae. Another positive association detected on both axes concerned Arte-DRB-11 and A. horrida. The associations detected on the second axis were less important, due to the smaller proportion of the variance accounted for by this axis. Associations were observed between the two viruses CPXV and LCMV and the Arte-DRB-16 and Arte-DRB-10 (positive associations) and Arte-DRB-07 and Arte-DRB-11 (negative) alleles.

We randomly split the data set into two independent data sets, to highlight nonrandom genetic/parasitological associations (Supplementary material). We only confirmed the positive (negative) associations between Arte-DRB-11 and richness indices (Rh and to a lesser extent Rp) and T. arvicolae and the negative association between Arte-DRB-15 and these factors. The ACO analyses performed with the DRB locus only also revealed this strong opposition between Arte-DRB-11 and Arte-DRB-15 (Fig. 5). Individuals carrying Arte-DRB-11 are 1.5 and 6.2 times more likely than other individuals to be infected with at least three parasites or with high loads of T. arvicolae respectively (Table 2). Alternatively, individuals with the Arte-DRB-15 allele have a lower risk of infection with at least three helminths or parasites (by factors of 0.3 and 0.5 respectively). None of the individuals carrying Arte-DRB-15 was infected with high loads of T. arvicolae (Table 2). Note that T. arvicolae intensity was significantly correlated with parasite richness (Spearman, S = 3.1 × 106, < 10−8, ρ = 0.326) and helminth richness (S = 2.5 × 106, < 10−8, ρ = 0.456). When removing T. arvicolae from the richness estimates (new variables Rp-T, and Rh-T), these relationships disappeared (Rp-T: S = 5.0 × 106, = 0.14, and Rh-T: S = 4.5 × 106, = 0.81). We also performed ACO using Rp-T and Rh-T instead of Rp and Rh. These two variables and T. arvicolae abundance still appeared associated with the alleles Arte-DRB-11 and Arte-DRB-15 (data not shown). However, only the associations with the nematode could be confirmed through cross-validation.

Figure 5.

 Associations between DRB and parasitological variables projected on the F1 axis of the co-inertia analysis (ACO) based on DRB only. This first axis accounts for 33.1% of the total variance. Grey circles indicate positive associations and white squares, negative associations. The size of the symbol indicates the strength of the association. Variables involved in previously confirmed associations are surrounded.

Table 2.   Correspondence table between alleles and parasitological variables selected by co-inertia analyses.
  1. The number in each cell indicates sample size. Parasite richness and T. arvicolae distributions are given both for individuals with and without each genetic factor identified (Arte-DRB-11 and Arte-DRB-15). RR corresponds to the relative risk associated with each allele, estimated as described by Haldane (1956).

Helminth richness Rh
  ≥ 3836143
  1 or 22816812184
 RR (Rh ≥ 3)1.498 0.282 
Parasite richness Rp
  ≥ 31464375
  1 or 22217013179
 RR (Rp ≥ 3)1.475 0.485 
Trichuris arvicolae intensity, Nem-Ta
  ≥ 63306
 RR (Nem-Ta ≥ 6)6.25 0 

The other associations found with the whole data set were not confirmed by cross-validation method. Some associations were not observed, regardless of the subset of data considered. This was the case for the association between Arte-DRB-11 (positive) or Arte-DRB-15 (negative) and P. omphalodes, or Arte-DRB-10 (positive) and LCMV or CPXV. Other associations were observed in only one of the two data sets. This was true for the positive associations between Arte-DRB-07 or Arte-DRB-11 and A. horrida (data set 2), and for the positive associations between Arte-DRB-16 and LCMV or CPXV (data set 1).

Population genetics

We performed 815 tests to analyse the linkage disequilibrium between genes (55 pairs of loci × 15 populations minus 10 pairs of loci as one microsatellite was monomorphic in one population). After FDR correction, four tests (0.48%) remained significant. If these remaining significant comparisons were stochastic (random) events, 0.48% of the 55 pairs of loci would correspond to 0.26 cases. Nevertheless, three of the four significant comparisons included DQA1 and DRB, suggesting weak linkage between these two genes.

Significant deviation from mutation–drift equilibrium (Table 3) was detected for DQA1 locus in five populations after Bonferroni correction (all 15 populations had < 0.05). For DRB locus, only one population stayed significant after Bonferroni correction (four had < 0.05). Considering microsatellite loci, the mean number of population exhibiting deviation from mutation–drift equilibrium (< 0.05) was 3.4, ranging from zero (AT3) to eight (AT25). However, P-values were closed to 0.05 (Table 3) and no microsatellite locus differed significantly from mutation–drift equilibrium after Bonferroni correction, whatever the population considered. In this way, DQA1 locus, but not DRB locus, strongly differed from neutral microsatellites in allelic frequency.

Table 3.   Detection of overdominance and balancing selection in water vole populations: six sampling sites followed for 3 years.
YearSiteNHeterozygote excess (FIS)Mutation–drift neutrality tests
DQA1DRB9 MSDQA1DRBSignificant microsatellite loci
  1. N is the number of voles screened for genetic analyses. Estimates of FIS, are given separately for each type of marker. Probabilities of conformity to Hardy–Weinberg equilibrium (H1 heterozygote excess), obtained from exact tests, are indicated in brackets. Results of mutation–drift neutrality tests (Wilcoxon sign test probabilities) are provided, for each MHC gene (IAM), for each microsatellite only if it is significant (GSM). P-values are considered significant after Bonferroni correction are indicated in bold.

2003A30−0.077 (0.150) −0.110 (0.140) 0.065 (0.871)0.0090.231AT9 (= 0.018)
AT25 (= 0.030)
AT19 (= 0.048)
B30−0.095 (0.249) −0.004 (0.497) 0.018 (0.930)0.0200.002AT13 (= 0.017)
AT22 (= 0.015)
AT9 (= 0.028)
AT25 (= 0.028)
C260.224 (0.980)0.022 (0.476)−0.002 (0.509)0.0050.145AT2 (= 0.026)
AT9 (= 0.034)
D270.132 (0.878)0.120 (0.868)0.004 (0.463)0.0000.034AT13 (= 0.047)
AT25 (= 0.029)
E250.166 (0.810)0.006 (0.435)−0.030 (0.187)0.0240.100AT13 (= 0.024)
AT19 (= 0.032)
F28−0.095 (0.316)−0.107 (0.143)0.043 (0.958)0.0020.152AT24 (= 0.029)
2004A200.215 (0.875)0.164 (0.937)0.019 (0.951)0.0470.114AT9 (= 0.005)
AT25 (= 0.005)
B220.103 (0.774)0.079 (0.930)0.064 (0.975)0.0410.183AT13 (= 0.018)
AT19 (= 0.010)
C220.103 (0.844)−0.077 (0.187)0.057 (0.659)0.0010.134AT9 (= 0.021)
AT19 (= 0.025)
D230.215 (0.980)0.025 (0.511)0.028 (0.786)0.0170.006 
E210.341 (0.997)−0.128 (0.090) 0.038 (0.801)0.0090.144AT23 (= 0.016)
AT25 (= 0.022)
AT19 (= 0.040)
F220.091 (0.714)−0.101 (0.094)−0.025 (0.398)0.0030.004AT23 (= 0.048)
2005D230.174 (0.866)−0.093 (0.243)−0.018 (0.518)0.0080.447AT13 (= 0.043)
AT25 (= 0.028)
E240.157 (0.959)0.044 (0.640)−0.080 (0.015) 0.0060.097AT23 (= 0.033)
AT13 (= 0.029)
AT25 (= 0.014)
F240.343 (0.998)−0.129 (0.085)0.016 (0.779)0.0020.163AT25 (= 0.008)
AT19 (= 0.037)

Whatever the gene considered, no population deviated from Hardy–Weinberg equilibrium, revealing an absence of significant heterozygosity excess (Table 3).


This is the first study, to our knowledge, to have explored the potential roles of both parasite species and parasite burden as selective pressures acting on immune genes. An approach combining multivariate statistics and population genetics was applied to a genetic data set, including polymorphism of two MHC class II genes and nine microsatellites, and an almost exhaustive parasitological data set.

Evidence of selection

Evidence of balancing selection acting on both MHC class II genes first came from the extensive polymorphism found within the small study area (five DQA1 alleles and 12 DRB alleles in a 100 km2 area). Moreover, we observed contrasted patterns of genetic diversity at MHC genes and neutral microsatellite markers. After correction for multiple testing, no population deviated from demographic neutrality for microsatellites. The weak impact of demographic bottlenecks on neutral genetic diversity is consistent with the results of a previous study of A. scherman populations (Berthier et al., 2006). On the contrary, homogeneous allelic frequencies (Watterson, 1978) were found in five of 15 populations for the DQA1 gene and in only one population for the DRB gene. In this way, DQA1 locus patterns strongly deviated from mutation–drift equilibrium neutral results, whereas it was not really the case for DRB. This locus-specific (DQA1) homogeneity in allelic frequencies can therefore be attributed to balancing selection (Cavalli-Sforza, 1966; Luikart et al., 2003).

We found no significant global co-inertia between the genetic and parasitological matrices, but this is not surprising given the difficulties related in the literature to find MHC-parasite associations (Hedrick & Kim, 2000). This result is not inconsistent with PMBS if: (i) some important parasites, such as bacteria or viruses, were not analysed; (ii) other genes influence parasite burden (see Charbonnel et al., 2006 for examples of genes likely to be associated with macroparasites); (iii) various strains of parasites are different enough that different alleles might confer resistance, and the genetic diversity of parasites should be included in such studies (e.g. Webster & Woolhouse, 1998; Moore et al., 2002); or if (iv) only a few main associations drive MHC polymorphism.

Associations between two specific MHC alleles, parasite load and parasite species were found. One allele, Arte-DRB-15, was negatively associated with parasite richness (mainly helminth richness Rh) and with the presence of the nematode T. arvicolae. An antagonistic allele, Arte-DRB-11, was positively associated with the same parasitological variables. These associations were weak, but the large sample size used in this study made it possible to confirm the observed associations using cross-validation. Nevertheless, correlation between parasites involved in associations could lead to spurious associations. Here, parasite and helminth richness were correlated with T. arvicolae intensity as a consequence of their definition (sum of each parasite presence/absence, an effect which may be intensified by the fact that T. arvicolae is the most prevalent parasite). When removing T. arvicolae from parasite richness definition, associations with these new variables appeared in the whole data set but were not confirmed through cross-validation. In this way, we confirmed the association of two alleles (Arte-DRB-11 and Arte-DRB-15) with the nematode T. arvicolae (confirmed though cross-validation). The association of the same alleles with parasite richness could be direct, or alternatively, could be driven by T. arvicolae associations. These findings highlight the relevance of considering each parasite species separately to reveal the true variable with which MHC alleles are associated.

Associations between parasitism and MHC polymorphism reveal PMBS only if the parasite species can affect host fitness (Little, 2002). Most of the parasite species studied here were known to affect host fecundity or mortality (e.g. Feore et al., 1997 for Cowpox virus, Fichet-Calvet et al., 2004 for Frenkelia, Vervaeke et al., 2006 for Echinococcus). In particular, a recent laboratory study showed that T. arvicolae affects fecundity in another vole, Microtus arvalis (Deter et al., 2007b). These associations between MHC and parasitism therefore provide only indirect evidence of selection acting on DRB or on another closely related gene. Associations between DRB genotypes and fitness would provide direct evidence of selection, but such data are difficult to obtain (but see De Eyto et al., 2007 for impact on survival, Jäger et al., 2007 on reproductive parameters and Sauermann et al., 2001 on reproductive success).

Insights into selection mechanisms

Whatever the individual estimator considered (DQA1, DRB or microsatellites heterozygosity or DQA duplication), allelic diversity was not involved in the co-inertia between genetic and parasitological matrices. Thus, being heterozygous or having a supplementary copy of one gene may not affect the susceptibility of these water voles to parasites. Several other studies reported no effect on parasite burden of MHC heterozygosity (e.g. Paterson et al., 1998; Schad et al., 2005; Schwensow et al., 2007; but see Froeschke & Sommer, 2005), nor of whole genome heterozygosity (e.g. Ortego et al., 2007; but see Coltman et al., 1999). Furthermore, none of the water vole populations studied deviated from Hardy–Weinberg equilibrium, whatever the locus considered. Thus, none of the approaches used provided any evidence of overdominance in these populations. These results support the idea emerging from empirical (see references above) and theoretical (Borghans et al., 2004; De Boer et al., 2004) studies that particular alleles may be more important for disease resistance than heterozygosity (Bernatchez & Landry, 2003; Piertney & Oliver, 2006).

Associations between single alleles and parasite load would be expected under the negative frequency dependence hypothesis. Here, we report weak associations between two particular alleles of the DRB gene and the nematode species T. arvicolae. The low water vole genetic differentiation observed over years and localities and the cross-validation approach confirmed that these associations were not artefacts of spatial structure. These two DRB alleles seem to play antagonistic roles in the susceptibility of voles to parasites. Similar results were previously obtained concerning the resistance of sheep to gastrointestinal nematodes (Paterson et al., 1998). We notice that these two DRB alleles encode proteins whose amino-acid sequences differ by 10 amino acids, nine of which seem to be involved in peptide recognition (Bryja et al., 2007).

Confirming frequency dependence mechanism in this system would require a temporal monitoring of allelic frequencies, revealing fluctuations (Little, 2002; e.g. Dybdahl & Lively, 1998). However, as associations are weak and parasite effects on fitness are low, selection is expected to be soft, and this monitoring would need to be very long to allow observation of parasite-driven genetic change. On the other hand, the vole populations of the Jura Mountains are particularly likely to undergo spatio-temporal variations in selective pressures (Hedrick, 2002) as they go through marked density cycles (Saucy, 1994; Charbonnel et al., 2008). Bryja et al. (2007) investigated this hypothesis and detected variations in selection patterns, both in space (local adaptation detected in one population during the low-density phase of the cycle) and in time (the population structure observed during the low-density phase disappeared at high density, when balancing selection homogenized allelic frequencies). Spatio-temporal variations in selective pressures are thus another, nonexclusive, possible mechanism of balancing selection likely to mediate selection in these populations.

Contrasted patterns of selection for two MHC class II genes

Within rodents, the precise chromosomal position of MHC class II genes is known only for the laboratory mouse (Blake et al., 2003) and rat (Gunther & Walter, 2001), in which the DQA and DRB genes are tightly linked. We found a weak signal indicating linkage disequilibrium between DQA1 and DRB genes and the action of different selection pressures on these two genes. The DQA gene seemed to play a lesser role in genetic/parasitological associations than DRB. Parasite-mediated selection would mostly act on DRB in the present generation (Garrigan & Hedrick, 2003). By contrast, homogeneity in allelic frequencies was observed mostly for the DQA1 gene. This pattern reflects balancing selection acting on this gene over the history of the population (Garrigan & Hedrick, 2003). Bryja et al. (2007) obtained similar results for neighbouring populations of A. scherman: a strong signature of balancing selection on DQA1 and limited evidence for such selection on DRB. These results challenge the common idea that the patterns observed at one locus can be used as an indicator of all MHC patterns (Bryja et al., 2007; and references therein), and encourage the simultaneous study of several MHC genes, although such studies remain scarce (but see Aguilar et al., 2004). This would provide new insight into the potential effects of different selective pressures acting on MHC genes, now that recombination within the MHC is increasingly widely recognized to be common (e.g. Richman et al., 2003; Schaschl et al., 2005).


This study proposes a new approach to detect associations between immunogenetic factors and parasitism. This question has frequently been tackled in the field of evolutionary biology over the last 10 years, but most studies have been subject to major flaws. First, they consider the relationship between MHC and either one species or global parasite load independently. Second, they do not take into account the multiple tests used to investigate all possible associations, strongly increasing the risk of type I error. Simultaneous evaluation of the influences of each parasite species on each MHC variable was made possible by applying multivariate co-inertia approach. This revealed particularly relevant as genetic associations with global parasite load may ultimately be due to only one parasite species. This method permits to analyse data sets containing numerous variables (several genes and numerous parasite species as in this study) and as it has no size limitation, this approach could also be used to study intraspecific parasite data. Nevertheless, rare alleles and parasite would challenge the utilization of cross-validation for confirming the significance of such rare associations observed over a whole data set. Possible solutions would be: (1) to pool alleles in supertypes (Schwensow et al., 2007) and parasites in functional families; or (2) to enlarge sample size, but spatio-temporal variations of parasite pressure may prevent confirmation of previous results (e.g. Bonneaud et al., 2006).

However, correlative field studies of MHC/parasite associations will always show main limitations as many confounding factors are likely to influence parasite burden, e.g. (1) host genetic background because parasite susceptibility is thought to be influenced by many genes (Woolhouse et al., 2002); (2) probability of exposure, as a result of differences in dispersion rate (Deter et al., 2007a) for example; and (3) condition-based effects (Little, 2002), in particular immune status may be highly variable between and within individuals. Experimentation remains the only way to get free from these confounding effects, and thus to enhance the probability to detect MHC/parasites associations and to confirm the importance of immunogenetics on susceptibility to parasites. Nevertheless, this type of correlative field study would still be an important prerequisite to extract genetic and parasitological variables that will have to be tested through such experimental infestations.


We thank Matthieu Faure, Esteve Martinez Garcia, Hervé Lépissa and Sylvain Piry for assistance with the field work. We thank Jean-Dominique Lebreton for advices in statistical analyses and Carine Brouat for comments on the manuscript. This work received financial support from the Institut National de la Recherche Agronomique, the Region of Franche-Comté, the DIREN Franche-Comté, the French Ministry of the Environment, and was partially funded by the EU grant GOCE-2003-010284 EDEN. This paper is catalogued by the EDEN Steering Committee as EDEN0070 (