Genomic analysis of MHC-based mate choice in the monogamous California mouse

Genes of the Major Histocompatibility Complex (MHC) have been implicated in mate-choice decisions in vertebrates. Theory predicts that individuals should prefer MHC-dissimilar mates. We show that, in monogamous California mice, reproductive partners are not more dissimilar at MHC genes than randomly selected male–female pairs. Instead, our data suggest that pathogen exposure, not disassortative mating, underlies variability at MHC Class I and II genes in California mice.

Abstract

Variation at Major Histocompatibility Complex (MHC) genes is thought to be an important mechanism underlying mate choice in vertebrates, with individuals typically predicted to prefer MHC-dissimilar reproductive partners. However, analyses based on individual MHC loci have generated contradictory results regarding the role of these genes in mate-choice decisions. To provide a more comprehensive assessment of relationships between MHC variation and mating behavior, we used an exome capture strategy to characterize variability at 13 MHC loci, 312 innate immune system genes, and 1044 nonimmune genes in 25 obligate monogamous pairs of California mice (Peromyscus californicus) from 2 free-living populations of this species in Monterey County, California. Pairwise genotypic comparisons and analyses of SNP-based allelic differences failed to detect disassortative mating based on MHC variability; reproductive partners were not more dissimilar than randomly generated male–female pairs at MHC, innate or nonimmune loci. Within populations, individuals tended to be more closely related at MHC genes than at innate or nonimmune genes. Consistent with the functional role of immunogenes, the 2 study populations were highly differentiated at MHC and innate genes but not at nonimmune loci. Collectively, our results suggest that MHC genetic variation in California mice reflects local differences in pathogen exposure rather than disassortative mating based on variability at MHC Class I and II genes.

INTRODUCTION

The genes of the Major Histocompatibility Complex (MHC) form an integral part of the vertebrate immune system (Klein 1986). Due to their role in detecting antigens originating from parasites and pathogens, Class I and Class II MHC genes are expected to be subject to strong balancing selection (Hedrick 1999; Bernatchez and Landry 2003; Eizaguirre et al. 2012), leading to often-extreme polymorphism at these loci (Garrigan and Hedrick 2003). This variability, in turn, has been argued to be important in mate-choice decisions in multiple species (Tregenza and Wedell 2000). Specifically, females are predicted to prefer mates whose genotypes maximize MHC variability among young, either by increasing the number of MHC alleles or the diversity of the allelic sequences present in offspring (Penn and Potts 1999; Neff and Pitcher 2005). These potential adaptive benefits are often referred to collectively as choice of MHC-dissimilar partners (Penn and Potts 1999; Tregenza and Wedell 2000).

Initially, analyses of MHC-based patterns of mate choice emphasized laboratory studies of highly inbred strains of mice, the genotypes of which differed primarily at MHC loci (Tregenza and Wedell 2000; Yamazaki and Beauchamp 2007). Although these studies tended to confirm the importance of disassortative mating with regard to MHC genotypes, subsequent analyses of free-living populations of vertebrates have generated more variable outcomes (Kamiya et al. 2014). One of the factors that may contribute to variability in relationships between mate choice and MHC loci is the number of immunogenes examined per study (Kamiya et al. 2014). Studies of natural populations have tended to rely on data from one to at most a few genes, with the MHC Class II DRβ locus being a particularly common target for analysis (e.g. Wedekind et al. 1995; Paterson et al. 1997; Sommer et al. 2002; Sommer 2005; Radwan et al. 2008; Roberts et al. 2008; Schwensow et al. 2008a; Schwensow et al. 2008b; Huchard et al. 2010; Setchell et al. 2010; Schad et al. 2011; Cutrera et al. 2012; Knafler et al. 2012; Huchard et al. 2013; Ferrandiz-Rovira et al. 2015; Wa Sin et al. 2015; Galaverni et al. 2016; Ferrandiz-Rovira et al. 2016; Rymešová et al. 2017; Santos et al. 2017). Patterns of variability and evidence for selection can vary markedly among MHC genes (Kuduk et al. 2012; Huchard et al. 2013) as well as among different species genotyped at the same MHC locus (reviewed in Kamiya et al. 2014). As a result, conclusions based on a limited number of MHC loci may not be indicative of genome-wide relationships between mate preferences and immunogenetic variability (i.e. variability at genes with known immune function) and, accordingly, may not provide reliable information regarding the adaptive benefits of this mechanism for selecting reproductive partners (Chaix et al. 2008; Kamiya et al. 2014).

To generate a more comprehensive understanding of the role of MHC genetic variability in mate choice, we characterized genome-level genetic variation in California mice (Peromyscus californicus) from 2 populations located in coastal central California. The California mouse is unusual among mammals in that it is both socially and genetically monogamous (Ribble 1991), with reproductive partnerships typically persisting until the death of one member of a monogamous pair (Ribble and Salvioni 1990). Accordingly, because females typically select a single reproductive partner during their lifetime, mate-choice decisions by members of this species may have particularly profound impacts on individual fitness. If variability at MHC genes contributes to the choice of reproductive partners in this species, then dissimilarity at MHC genes is predicted to be greater between reproductive partners than between randomly selected pairs of adult males and females; in contrast, reproductive partners are not expected to be more genetically differentiated than random pairs at innate and nonimmune loci. To test these predictions, we generated more than 1300 single nucleotide polymorphic (SNP) markers from the exome of P. californicus, including multiple MHC Class I and Class II genes. To provide a broader perspective on variability at immune-related genes and mate choice, we also examined variability at SNPs in genes associated with the innate immune response. Finally, to place these analyses in the context of overall patterns of genetic variability in our study populations, we assessed variability at SNPs contained in genes with no known immunological function. In addition to providing one of the first analyses of mate preferences in relation to multiple functional categories of immunologically important genes, our analyses generate important new insights into patterns of immunological versus neutral genetic variation in free-living, nonhuman populations of mammals.

METHODS

Field sites

Field sampling of P. californicus was conducted from July 2013 to June 2014 at the Hastings Natural History Reservation (HNHR) and the Landels-Hill Big Creek Reserve (BCR), Monterey County, CA (Figure 1). Both localities are characterized by a Mediterranean climate, with almost all precipitation occurring during winter months. Although these sites are separated by only ~32 km, they occur on opposite sides of the Santa Lucia Mountains and thus the habitats at these locations differ markedly. HNHR is situated on the drier, eastern side of the Santa Lucia range and receives a mean of ~50 cm of rain per year. The habitats at this locality consist primarily of open grassland and coastal live oak Savannah. In contrast, BCR is located on the substantially wetter western side of the Santa Lucia range and receives a mean of ~200 cm of precipitation per year; habitats at this locality are dominated by redwood forests. Despite these environmental differences, P. californicus is abundant at both localities.

Figure 1.

Location of the study populations in central coastal California. The geographic distribution of P. californicus is shown in dark gray. Study sites at the Hastings Natural History Reservation (HNHR) and the Big Creek Reserve (BCR) are indicated, with photos depicting typical habitat at each site. Also included is a photograph of a California mouse (Peromyscus californicus) from the HNHR.

Trapping and tissue sample collection

Mice at both study sites were captured using Sherman live traps baited with rolled oats and containing cotton balls that served as nesting material. A total of 200 traps were laid out in pairs, with approximately 6 m allowed between successive pairs of traps. Traps were opened at 1800 h and closed at 0300 h. Trapping was conducted for 20 consecutive nights per study site per summer field season. All individuals captured were identified to species based on size and pelage characteristics, after which they were weighed and their sex and reproductive status (e.g. pregnant, lactating, scrotal) determined. At the time of first capture, each animal was permanently marked by attaching a uniquely numbered metal tag (Monel 1005-1, National Band and Tag Company, Inc.) to its right ear pinna, after which a nondestructive tissue sample was collected by using sterile surgical scissors to remove a small (~ 2 × 5 mm) sliver of skin from the margin of the pinna. Tissue samples were stored in 95% ethanol at ambient temperature until analysis (see DNA extraction and sonication). Upon completion of these procedures, each animal was released at the location at which it had been caught.

Identification of reproductive partners

To characterize potential genetic dissimilarity among reproductive partners, it was necessary to determine which individuals were members of the same monogamous pair. A male and female were suspected to be reproductive partners if they were captured in adjacent (paired) traps on more than 3 occasions during the same 20-night trapping effort. These assignments were confirmed using fluorescent powder tracking (Ribble and Salvioni 1990; Kalcounis-Rueppell et al. 2001). Specifically, the female in each putative pair was covered from neck to tail in one of 6 colors of nontoxic Eco Pigments (DayGlo) fluorescent powder just prior to release at the point of capture. Previous studies of this species have demonstrated that, when a dusted female returns to her nest, powder is transferred from the female to the adult male with which she is partnered (Ribble and Salvioni 1990; Ribble 1991). By recapturing the female and her partner the following night, this transfer can be detected visually (either directly or with a hand-held black light), thereby confirming physical contact between the adults in question. In the few cases in which the male was not caught the following night, additional powder was applied to the female and the process was repeated. A male displaying significant powder transfer was determined to be a female’s reproductive partner; significant transfer was defined as that visible without the assistance of a UV light. Although transfer of powder could be detected on any part of a male’s body, it was most common on the pinnae, muzzle, and around the eyes as well as on the feet, tail, and genitalia. As a final check on our assignments of individuals to reproductive pairs, the fluorescent powder tracking process was repeated for each putative pair using a different color of powder.

All field work involving live mice was approved by the Animal Care and Use Committee at the University of California, Berkeley, and was consistent with the Guidelines for the Use of Wild Mammals in Research published by the American Society of Mammalogists (Sikes et al. 2016).

DNA extraction and sonication

Genomic DNA was extracted from tissue samples obtained from 50 P. californicus representing 25 male–female reproductive pairs (HNHR: N = 12 pairs; BCR: N = 13 pairs). A standard salt extraction protocol was used, after which DNA quality and the distribution of size fragments in our extracts were assessed via electrophoresis on 1% agarose gels followed by staining with ethidium bromide and visualization under UV light. Samples were then sheared to 200 bp fragments using a BioRuptor UCD-200 (Diagenode). Sonication time was dependent on the quality of the individual extract; although extracts containing high molecular weight DNA were sonicated for 15 min, more degraded samples were sonicated for 5–10 min. After sonication, DNA fragment sizes were confirmed via visualization on 1% agarose gels (see above); additional sonication was performed if necessary to complete shearing of an extracted sample.

Library preparation

DNA libraries were prepared following the protocol of Meyer and Kircher (2010), as modified for half reactions. Additional modifications to this protocol included 1) use of a 1.3-fold volume of SPRI beads for cleaning reactions, 2) insertion of an additional SPRI bead clean-up immediately after sonication, and 3) completion of 2 indexing PCR reactions (9 cycles each) per sample to reduce PCR duplicates. Prior to library construction, DNA concentrations were quantified using a Qubit fluorometer (Thermo Fisher Scientific). On average, each indexing PCR reaction yielded 600 ng of amplified DNA. Final indexed, cleaned PCR products were eluted in ddH₂O, after which amplification success was assessed via electrophoresis of PCR products on a 2% agarose gel followed by visualization with ethidium bromide. To increase the concentration of reaction products and the proportion of unique reads, 2 indexing reactions per individual were combined; the concentrations of these combined reactions were determined using a NanoDrop spectrophotometer (Thermo Fisher Scientific). A 900-ng aliquot of this combined indexing reaction was then pooled with comparable samples from other individuals. Qubit analysis revealed the final concentration of this pooled sample to be 21 μg, of which 20 μg were used for exome capture (see below); the remaining 1 μg was retained to allow comparisons of pre and postcapture concentrations and evaluation of the success of the capture process.

De novo transcriptome annotation and probe design

Tissue sample (liver) preparation and transcriptome sequencing required to produce a reference transcriptome for P. californicus were performed by Crawford JC (unpublished data). Raw reads were cleaned using a custom Perl script, after which transcriptome assembly was conducted using Trinity (Grabherr et al. 2011). We annotated the assembled transcripts based on NCBI reference protein sequences from Mus, Rattus, Microtus, Peromyscus, and humans using the blastx (Altschul et al. 1997), exonerate (http://www.genome.iastate.edu/bioinfo/resources/manuals/exonerate/exonerate.man.html, accessed 4 July 2018), and frameDP (Gouzy et al. 2009) software programs; the pipelines used for de novo transcriptome annotation are available at https://github.com/CGRL-QB3-UCBerkeley/DenovoTranscriptome (accessed 4 July 2018). We categorized these annotated transcripts as “adaptive” if they coded for MHC Class I or Class II proteins and “innate” if they coded for MHC Class III proteins, cytokines (i.e. chemokines, interleukins, interferons, transforming growth factors), or a series of other known innate immune system functions (i.e. tumor necrosis factors, solute carrier proteins, toll-like receptors). Transcripts not categorized as either “adaptive” or “innate” and lacking a known immune function were categorized as “non-immune.” Genes that a priori had been identified as being of interest (e.g. MHC Class II or I loci) but that were not present among our annotated transcripts were added to the final list of loci targeted for analysis by extracting the relevant mRNA sequences from the Peromyscus maniculatus reference transcriptome.

We combined all candidate immune loci and >1000 randomly selected nonimmune genes to compile a “targets file.” We then filtered these target sequences for duplicates by self-blasting for similarity. For sequences with multiple hits, a single unique copy was included in the design of our bait capture chip. The final suite of genes included on the chip consisted of 16 adaptive, 347 innate, and 1100 nonimmune genes, for a total of 4.3 Mbp of targeted sequence (Supplementary Table 1). The 16 MHC Class II genes on the chip included 2 frequently analyzed loci, the MHC Class II DRβ gene (murine ortholog H2-IEβ) and the MHC Class II DQβ gene (rat ortholog RT1-Bβ). The chip did not include a third commonly analyzed locus, the MHC Class II DQα gene; this locus was removed from our target gene set through self-blasting for similarity due to its resemblance to other, potentially nonfunctional loci, including DQα pseudogenes.

We used an Agilent SureSelect custom 1M-feature microarray to capture the target sequences. Probes were designed as per the recommendations in Hodges et al. (2009). To increase capture efficiency, 60 bp probes were superimposed at 4 bp intervals across individual exon targets; because we did not have sequences for introns, we decreased the super-imposition frequency to every 1 bp at exon ends to improve coverage in these regions.

Sequence capture procedure

Capture of target genes in our pooled, indexed libraries was performed following the protocol of Hodges et al. (2009), with the substitution of P5 and P7 xGEN (Integrated DNA Technologies) as blocking oligonucleotides. Libraries were hybridized to the bait capture chip at 65 °C for 65 h, with constant rotation at 12 rpm. The final library amplifications followed the Indexing PCR temperature profile suggested by Meyer and Kircher (2010). Capture efficiency was assessed primarily by comparison of pre and postcapture qPCR results for four positive control genes (Rag1, Adh1, Ghr, and Rasgrf1), with secondary assessment based on the observed doubling of postcapture concentrations for the target sequences on a Bioanalyzer 2100 DNA 1000 chip (Agilent Technologies). Postcapture concentrations were determined using both the NanoDrop and the Qubit high-sensitivity kit (Themo Scientific). Postcapture fragment sizes were assessed on a Bioanalyzer 2100 DNA 1000 chip; postcapture libraries had an average fragment size of 337 bp. Postcapture libraries were sequenced at Berkeley’s QB3 Vincent J. Coates Genomics Sequencing Laboratory on 2 lanes of Illumina HiSeq2000 with 100 bp paired-end reads.

Analyses of captured sequences

Data generated by the exon capture procedure were analyzed using a custom Perl pipeline (Bi et al. 2012) developed in the Evolutionary Genetics Laboratory of the Museum of Vertebrate Zoology (https://github.com/CGRL-QB3-UCBerkeley/denovoTargetCapturePopGen, accessed 4 July 2018). This pipeline functioned to clean raw de-multiplexed Illumina reads, assemble contigs, and identify target loci (those contained on the bait capture chip) for organisms without a reference genome. In brief, the raw data were first cleaned by removing adaptor sequences (cutadapt: Martin 2011) and trimmed based on quality (PHRED < 20) (Trimmomatic: Bolger et al. 2014). Overlapping reads were then merged (COPE: Liu et al. 2012 and FLASH: Magoč and Salzberg 2011) and duplicate reads were removed (Bowtie2: Langmead and Salzberg 2012). The pipeline also identified and removed reads originating from bacterial contamination by aligning sequences to the Escherichia coli genome (Bowtie2: Langmead and Salzberg 2012). Once these steps were completed, the remaining sequence reads were assembled using a multiple-kmer approach (ABySS: Simpson et al. 2009 and Bi et al. 2012), after which redundant sequences were removed from the final assembly (cap3: Huang and Madan 1999; blat: Kent 2002; and cd-hit-est: Li and Godzik 2006; Fu et al. 2012).

A pseudo-reference was constructed by identifying the assembled contigs that were derived from the target sequences included in the bait capture chip (BLASTn: Kent 2002 and cd-hit-est: Li and Godzik 2006; Fu et al. 2012). Alignment of the cleaned reads to the pseudo-reference was completed using Novoalign (http://www.novocraft.com, accessed 4 July 2018) and only uniquely mapped reads were retained. We used Picard (http://broadinstitute.github.io/picard/, accessed 4 July 2018) and GATK (McKenna et al. 2010) to perform realignment around insertions and deletions. The output alignments were analyzed using SAMtools and its associated “bcftools” as a quality control step, after which these data were filtered using SNPcleaner (https://github.com/tplinderoth/ngsQC/tree/master/snpCleaner, accessed 4 July 2018). Throughout this process, we employed data quality filtering steps recommended by Bi et al. (2013). To account for uncertainties when assigning genotypes based on our data (coverage ~20×), we called single nucleotide polymorphisms (SNPs) and estimated allele frequencies using an empirical Bayesian framework implemented in the program ANGSD (http://www.popgen.dk/angsd/index.php/ANGSD, accessed 4 July 2018). We only assigned genotypes to those sites that had a posterior probability ≥0.95 and for which the probability that the site was variable (likelihood ratio test) was ≤10⁻⁶.

Genetic diversity and departures from neutrality

We estimated average (π) and total (θ_w) genetic diversity for each population and each category of genes using ANGSD (Watterson 1975; Nei and Li 1979; Nei and Tajima 1981; Korneliussen et al. 2013). Under equilibrium, average nucleotide differences per-site (π) are not expected to differ from the number of segregating sites (θ_w), indicating that π = 4N_eμ = θ_w. To test for possible deviations from this expectation, we used the Tajima’s D test (Tajima 1989), as implemented in ANGSD. The statistical significance of these analyses was determined using DnaSP v5 (Librado and Rozas 2009). To assess potential departures from neutrality at the population level, we also conducted Hardy–Weinberg tests using the “HardyWeinberg” package in R (Graffelman and Morales-Camarena 2008; Graffelman 2015).

Genetic differentiation among populations

To examine potential differences in genetic structure between the study populations, for each SNP, we estimated F_ST between individuals from BCR and HNHR; mean values of F_ST were then calculated for each gene category (adaptive, innate, nonimmune). Estimates were generated using the “hierfstat” package in R (Goudet 2005) to calculate the Nei (1987) pairwise estimator of F_ST. In addition, we estimated F_ST using the “pegas” package in R (Paradis 2010); this procedure employs the Weir and Cockerham (1984) estimator, which is thought to be less biased for estimates of F_ST based on smaller population sizes (Willing et al. 2012).

Evidence for disassortative mating

Separate analyses of disassortative mating were conducted for each population and, within populations, for each gene category. We evaluated disassortative mating using 3 distinct measures of genetic differences between individuals: 1) coefficients of identity (fraction of identical alleles), 2) allelic dissimilarity (mean allelic dissimilarity), and 3) absolute genetic distance (average number of pairwise differences between alleles). All of these tests estimate genetic dissimilarity between pairs of individuals but the information captured by each analysis is distinct and offers a potentially different perspective on genetic differences among reproductive partners. For example, coefficients of identity are a measure of genetic similarity based on allele sharing between individuals and are therefore a binary measure of genetic variation. In contrast, allelic dissimilarity captures information about base-pair differences between individuals, which encompasses a greater range of potential variation between animals. Absolute genetic distance, in turn, incorporates information about heterozygosity at each SNP, thereby avoiding biases that may occur in measures (e.g. coefficients of identity, allelic dissimilarity) based on the estimates of relative genetic differences among individuals. Collectively, these analyses provide a more robust perspective on potential disassortative mating than would analyses based on any one of these individual metrics.

First, coefficients of identity (Q_ab) were calculated for all male–female pairs (both reproductive partners and randomly generated pairs) as the fraction of identical alleles shared between 2 individuals (Derti et al. 2010), with pairs of animals assigned values of 0 (2 alleles in common), 1 (1 allele in common), or 2 (no alleles in common) for each SNP locus. To evaluate the mean difference in Q_ab between reproductive and random pairs, we standardized Q_ab values to account for population level variation using R_ab= (Q_ab − Q_mean)/(1 − Q_mean), where Q_mean is the mean Q value for all male–female pairs (Chaix et al. 2008; Derti et al. 2010) and R_ab is the coefficient of relatedness calculated following Rousset (2002). We then generated a distribution of values of R_ab for randomly selected male–female pairs (1000 permutations with resampling; true reproductive partners excluded), after which we compared the mean value of R_ab for reproductive partners against this distribution (Hoover and Nevitt 2016). We estimated P values for each population as the proportion of the random distribution that was less than or equal to the mean value for reproductive partners.

Second, values for allelic dissimilarity (AD_ab) were calculated for all male–female pairs as the mean allelic variation between 2 individuals, with allelic variation at each SNP locus assigned as described above (0, 1, or 2 for each SNP locus). To evaluate the difference in AD_ab between reproductive partners and random male–female pairs, we first used z-scores to standardize AD_ab values, thereby accounting for potential differences in population level genetic variation. For this analysis, Z_ab= (AD_ab − AD_mean)/(sd(AD_ab)), where AD_mean is the mean and sd(AD_ab) is the standard deviation of AD_ab values for all male–female pairs. We then generated a distribution of Z_ab values for randomly selected pairs (1000 permutations with resampling; true reproductive partners excluded), after which we compared mean value of Z_ab for reproductive partners against this distribution; P values for these comparisons were estimated as above.

Third, we estimated pairwise genetic distances between reproductive partners and randomly generated male–female pairs by calculating a global Dxy_i (average number of differences per-site) for every male–female pair using a custom “perl” script (PopStats: https://github.com/CGRL-QB3-UCBerkeley/denovoTargetCapturePhylogenomics, accessed 4 July 2018). As in Crawford et al. (2015; Supplementary Information), Dxy_i = (h + 2H)/(2L), where h is the number of sites at which one or both members of a pair are heterozygous, H is the number of sites at which members of a pair are homozygous for different alleles, and L is the number of sites for which genotypic data are available for both members of the pair. We then generated a distribution of Dxy_i values for randomly selected pairs (1000 permutations with resampling; true reproductive partners excluded), after which we compared the mean value of Dxy_i for reproductive partners against this distribution; P values for these comparisons were estimated as above.

As an additional means of assessing potential disassortative mating, we calculated the degree of kinship between all possible male–female pairs within each study population. Pairwise estimates of coefficients of relatedness (or coefficients of co-ancestry, f) were generated using the “SNPRelate” package in R (Zheng et al. 2012), which employs a maximum likelihood algorithm to assess identity by descent (IBD). By definition, values of f in diploid organisms are equal to twice the Pearson’s coefficient of relatedness (r) and thus we used r = 2f (Falconer and Mackay 1996) to generate the coefficient of relatedness between members of each male–female pair. Values of r for reproductive partners were compared to those for randomly generated pairs using Wilcoxon’s signed rank tests.

Finally, to compare genetic dissimilarity at adaptive genes to overall genetic dissimilarity (innate and nonimmune genes), we calculated Pearson correlation coefficients between adaptive and the 2 other gene categories examined. Separate correlations were conducted for reproductive partners versus randomly generated pairs; in both cases, correlation analyses were conducted for estimates of Q_ab as well as estimates of AD_ab.

Locus-specific analyses of disassortative mating

To test hypotheses regarding the specific roles of the MHC Class II DRβ and DQβ loci in mate choice, we used the methods outlined above to generate locus-specific estimates of Q_ab, AD_ab, and Dxy_i for each of these genes. We then compared mean values of these parameters for reproductive partners with the distributions of values for randomly generated male–female pairs using the procedures described above.

Evidence for selection at the MHC Class II DRβ locus

We tested for evidence of selection on exon 2 of the MHC Class II DRβ locus using comparative data from 3 species of Peromyscus: P. californicus, P. eremicus, and P. maniculatus. Exon 2 of this gene codes for the peptide-binding region (PBR) of the DRβ protein; in addition to displaying high levels of nucleotide variation and nonsynonymous substitutions (Hughes and Nei 1989; Brown et al. 1993), this exon has been found to be under selection in multiple populations of wild vertebrates (reviewed in Radwan et al. 2010). DRβ sequence data for P. eremicus and P. maniculatus were downloaded from Genbank (AY219820.1, AF516929.1) and aligned to our sequences for P. californicus using the BLAST aligner (Altschul et al. 1990). Although we were able to identify the exon 2 region for the MHC Class II DRβ gene, we could not confidently identify this region for the DQβ locus and thus tests for selection were performed only for the DRβ gene.

Evidence of selection on exon 2 of the DRβ locus was assessed using the branch-site model of positive selection contained in the codeml package in PAML (Yang 1997; Yang 2007). Specifically, we contrasted a null model (no selection) for sequence variation against a model based on positive selection. Both models employed a phylogenetic tree containing the 3 focal species, with P. californicus and P. eremicus as sister species; for both models, we designated P. californicus as the “foreground” branch in the tree, with the other 2 species designated as “background” branches. In the model of positive selection, the proportion of sites under selection in the foreground branch was allowed to vary relative to the background branches (MA; model = 2, fix_omega = 0, NSsites = 2). In contrast, for the null model, sites were not allowed to vary in any branches of the tree (MA_null; model = 2, fix_omega = 1, NSsites = 2). The 2 models were compared using a log-likelihood ratio test in which the value of twice the difference between the log-likelihood of each model (2ΔlnL = 2(lnL1 − lnL0); l₁ = log-likelihood of MA and l₀ = log-likelihood of MA_null) was compared to a X² distribution with 1 degree of freedom. Individual sites under apparent positive selection were detected using the Bayes Empirical Bayes (BEB) approach as implemented in codeml (Yang et al. 2005).

RESULTS

Tissue samples were obtained from a total of 208 P. californicus. Seventy-eight (37.5%) of these animals were captured at HNHR; the remaining 130 animals were captured at BCR. Live-trapping and fluorescent powder marking identified 16 male–female pairs at HNHR and 27 male–female pairs at BCR; of these, 12 pairs from HNHR and 13 pairs from BCR were used to evaluate evidence of disassortative mating in the study populations. We chose to genotype this subset of randomly selected pairs (rather than all pairs identified) to optimize the depth of sequence coverage obtained given the number of loci targeted for analysis. Typical coverage was 20× per individual. Samples for all individuals met our quality thresholds; on average, each individual generated 1448 Mb of raw sequence data and 459 Mb of cleaned and edited sequence. A total of 1369 (93.5%) of our 1463 target loci were captured by our analyses. Of the 1369 loci included in the final data set, 13 were identified as part of the adaptive immune response (Table 2), 312 were identified as part of the innate immune response, and 1044 had no known immune function. From these loci, we identified a total of 60 adaptive, 1880 innate, and 3694 nonimmune SNPs.

Table 2.

Tests for departures from neutrality at the 13 MHC genes examined

			π %		θw %		Tajima’s D
MHC Class	Protein ID	Gene name	BCR	HNHR	BCR	HNHR	BCR	HNHR
II	XP_006997447.1	H2-IEα	0.002	0.002	0.01	0.01	−0.13	−0.15
	XP_006997450.1	RT1-Bβ (DQβ)	0.23	0.20	0.29	0.23	−0.69	−0.35
	XP_006997454.1	H2-EKα	0.22	0.23	0.15	0.15	0.97	1.20
	XP_006985191.1	HLA-DRγ	0.15	0.16	0.16	0.22	−0.36	−1.13
	XP_006997106.1	H2-Mβ1	0.21	n/a	0.30	n/a	−0.87	n/a
	XP_006997110.1	HLA-DOα	0.23	0.34	0.36	0.23	−1.08	1.23
	XP_006997452.1	H2-IEβ (DRβ)	0.76	0.97	0.52	0.80	1.29	0.69
	XP_006997101.1	HLA-DOβ	0.46	0.43	0.37	0.29	0.78	1.54
I	XP_006998053.1	H2-DKα	1.13	1.12	1.10	0.98	0.11	0.52
	XP_006998409.1	Saoe-2CAα	0.04	0.09	0.25	0.49	−1.93	−2.36 *
	XP_006985711.1	Mr1	0.15	0.22	0.31	0.23	−1.34	−0.08
	XP_006997342.1	Non-RT1-Aα	0.06	0.05	0.11	0.15	−1.08	−1.76
	XP_006997374.1	H2-D37α	0.21	0.23	0.32	0.26	−0.97	−0.37

For each locus, the NCBI protein code and gene name are given, as are values of per-site Pi (π%) and Theta (θ_w%), and Tajima’s D. Data for each study population are presented separately; sample sizes were 26 for BCR and 24 for HNHR. The only significant departure (Fisher’s 2-tailed test; P < 0.05) from neutral expectations detected was for estimates of Tajima’s D at the Saoe-2CA gene, as indicated by an asterisk.

Nucleotide variation

Mean (±SD) nucleotide diversity (π) in the BCR population was 0.0011 (±0.001), with values ranging from 0.00000002 to 0.01 (Table 1). Mean (±SD) π in the HNHR population was also 0.0011 (±0.001), with values ranging from 0.00000007 to 0.01; not surprisingly, mean values of π did not differ between the study populations (Wilcoxon’s signed rank test, 2-tailed P = 0.99). In both populations, the locus with the greatest nucleotide diversity (π ca.1%) was an adaptive immune system gene, specifically the MHC Class I H2-DKα locus (Table 2). No significant differences between study populations were detected when separate mean values of π were calculated for adaptive, innate, or nonimmune genes (Kolmogorov–Smirnov Tests, all D < 0.19, all P > 0.05). Within each population, mean π was significantly greater for adaptive immune genes than for innate immune genes or nonimmune genes (Mann–Whitney U tests, all one-tailed P < 0.01); mean π was greater for innate immune genes than for nonimmune genes (Mann–Whitney U tests, all 1-tailed P < 0.01). When comparing the distributions of π values across gene categories, the range of π values for adaptive genes was greater than those for either innate immune or nonimmune genes and the range of π values for innate immune genes was greater than that for nonimmune genes (Kolmogorov–Smirnov Tests, all tests D > 0.14, all P < 0.001). Thus, in general, nucleotide diversity was greatest for adaptive genes, followed by innate genes, with nonimmune genes displaying the lowest levels of such diversity.

Table 1.

Measures of genetic diversity in P. californicus

	Mean ± SD π %		Mean ± SD θw %		D ± SD
Gene categories	BCR	HNHR	BCR	HNHR	BCR	HNHR
Adaptive (N = 13)	0.30 ± 0.2	0.34 ± 0.4	0.33 ± 0.1	0.34 ± 0.3	−0.40 ± 0.97	−0.08 ± 1.22
Innate (N = 312)	0.13 ± 0.1	0.12 ± 0.1	0.15 ± 0.1	0.14 ± 0.1	−0.55 ± 0.92	−0.51 ± 0.88
Nonimmune (N = 1044)	0.11 ± 0.1	0.11 ± 0.1	0.13 ± 0.1	0.13 ± 0.1	−0.56 ± 0.92	−0.49 ± 0.96
All categories (N = 1369)	0.11 ± 0.1	0.11 ± 0.1	0.13 ± 0.1	0.13 ± 0.1	−0.56 ± 0.92	−0.49 ± 0.94

Data are from 26 individuals (13 male–female pairs) at BCR and 24 individuals (12 male–female pairs) at HNHR. Mean (±SD) values are shown for per-site nucleotide diversity (π%), number of segregating sites (θ_w%), and Tajima’s D; data were analyzed for all loci as well as for loci in each functional gene category examined. The number of genes in each functional category is indicated in parentheses. No significant differences (Wilcoxon’s signed rank tests, all P > 0.05) were detected between the study populations for any of these parameters.

Number of segregating sites

Mean number of segregating sites (θ_w) in the BCR population was 0.0013 (±0.001), with values ranging from 0.0000001 to 0.01 (Table 1). Mean θ_w at HNHR was also 0.0013 (±0.001), with values ranging from 0.0000004 to 0.01. Mean θ_w did not differ between the study populations (Wilcoxon’s signed rank test, 2-tailed P = 0. 25).

Departures from neutral expectations

Values for Tajima’s D were generally negative, with no difference between the frequency distributions of locus-specific estimates of D at HNHR and BCR (Kolmogorov–Smirnov Test, D = 0.0166, P = 0.63), suggesting an overall excess of low-frequency polymorphisms in both study populations. When these data were parsed by gene category, mean values of D for each category remained negative in both populations. At BCR, frequency distributions for D did not differ between innate, adaptive, and nonimmune genes (Kolmogorov–Smirnov Tests, all tests D < 0.06, all P > 0.05; Figure 2). In contrast, at HNHR, although distributions of D did not differ between innate and nonimmune genes (Kolmogorov–Smirnov Test, D = 0.04, P = 0.46; Figure 2), the distribution of D values for adaptive genes was more skewed toward positive values than the distributions for either innate (Kolmogorov–Smirnov Test, D = 0.35, P = 0.05) of nonimmune genes (Kolmogorov–Smirnov Test, D = 0.34, P = 0.06). Locus-specific values of D revealed only 1 adaptive gene—the MHC Class I gene Saoe-2CA gene—that displayed notable deviations from neutrality in both populations. Closer inspection of these data revealed that although Tajima’s D for this locus was strongly negative in the BCR population, D did not differ significantly from neutral expectations (D = −1.93, Fisher’s 2-tailed test, P > 0.05). In contrast, at HNHR, the value of D for this locus was significantly negative, providing potential evidence of purifying selection on this gene (D = −2.36, P < 0.05).

Figure 2.

Frequency density distributions for values of Tajima’s D from (A) BCR and (B) HNHR. Distributions are based on analyses of 13 adaptive, 312 innate, and 1044 nonimmune genes in each study population. The x axis depicts values of Tajima’s D; the y axis indicates the frequency density of genes. Distributions of D as a function of gene category did not differ at BCR; at HNHR, the distribution of values for adaptive immune genes was significantly different from those for innate or nonimmune genes (Kolmogorov–Smirnov tests, all P < 0.05).

Potential departures from neutral expectations were also examined using Hardy–Weinberg (HW) tests. At BCR, 118 (8.6%) of the 1368 genes analyzed revealed significant departures from Hardy–Weinberg expectations; at HNHR, 129 (8.6%) of these loci departed from HW expectations. When divided according to functional category, 1 (8.6%) of 13 adaptive genes, 41 (13.1%) of 312 innate genes, and 76 (7.2%) of 1044 nonimmune genes at BCR displayed significant departures from HW expectations. At HNHR, these figures were 0 (0%), 32 (10.3%), and 96 (0.1%) for adaptive, innate, and nonimmune genes, respectively. A total of 34 (2.5%) genes showed significant deviations from neutrality in both populations; this included genes categorized as innate (6.4%) and nonimmune (2.4%). No genes from the adaptive immune category displayed significant deviations in both populations.

Genetic differentiation between populations

Per-locus estimates of F_ST revealed that the extent of genetic differentiation between the study populations varied with the category of genes considered. For estimates generated according to Nei (1987), values for adaptive (F_ST = 0.69) and innate (F_ST = 0.66) genes indicated strong differentiation among populations; in contrast, the value for nonimmune loci (F_ST = 0.01) indicated little differentiation for this category of genes. Estimates generated using the method of Weir and Cockerham (1984) were similar (adaptive: F_ST = 0.57; innate: F_ST = 0.56, nonimmune: F_ST = 0.007; Figure 3). Distributions of locus-specific estimates of F_ST (Weir and Cockerham 1984) for adaptive and innate genes did not differ (Kolmogorov–Smirnov test, D = 0.08, P = 0.47), but both distributions were significantly different from that for nonimmune genes (Kolmogorov–Smirnov tests: innate versus nonimmune, D = 0.79, P < 2.2^–16; adaptive versus nonimmune, D = 0.88, P < 2.2^–16). Collectively, these findings indicate that while genes associated with immune function were strongly differentiated between the study populations, much less differentiation was evident for nonimmune genes.

Figure 3.

Per-locus estimates of F_ST between the study populations at BCR and HNHR. Data for adaptive (N = 60 SNPs), innate (N = 1880 SNPs), and nonimmune (N = 3694 SNPs) genes are shown separately. Estimates of F_st were generated following the method of Weir and Cockerham (1984). The dashed line denotes F_st = 0.25 (great genetic differentiation); the solid line denotes the F_st = 0.15 (moderate genetic differentiation; Hartl and Clark 1997).

Disassortative mating—identity coefficients and allelic dissimilarity

We identified a total of 60 adaptive, 1880 innate, and 3694 nonimmune SNPs (variants) from among the genes analyzed. For both populations and all gene categories, comparisons of mean values for R_ab and Z_ab for reproductive partners to the associated distributions of these variables for randomly selected male–female pairs were nonsignificant (all P > 0.05; 1000 permutations) (Table 3; Table 4; Figure 4). Values of Q_ab and AD_ab for reproductive partners did not differ between gene categories in either population (Dunn’s Tests with Bonferroni correction; both P > 0.05).

Table 3.

Pairwise coefficients of identity (Q_ab) for reproductive partners (N = 13 for BC; N = 12 for HNHR) and randomly generated male–female pairs of P. californicus (1000 permutations)

	Gene categories	Q ab				R ab		P value
		Mean ± SD		Median		Mean ± SD
		Rep. partners	Nonpartners	Rep. partners	Nonpartners	Rep. partners	Nonpartners
BCR	Adaptive	0.60 ± 0.11	0.61 ± 0.10	0.57	0.60	−0.04 ± 0.29	−0.008 ± 0.25	0.42
	Innate	0.62 ± 0.07	0.62 ± 0.06	0.64	0.64	0.01 ± 0.19	0.01 ± 0.15	0.47
	Nonimmune	0.64 ± 0.06	0.64 ± 0.05	0.66	0.66	0.01 ± 0.17	0.01 ± 0.15	0.42
	MHC-DRβ	0.36 ± 0.35	0.43 ± 0.39	0.33	0.33	−0.13 ± 0.61	0.002 ± 0.68	0.54
	MHC-DQβ	0.61 ± 0.24	0.65 ± 0.18	0.66	0.67	0.06 ± 0.59	0.13 ± 0.44	0.27
HNHR	Adaptive	0.59 ± 0.12	0.60 ± 0.11	0.61	0.62	−0.01 ± 0.30	0.01 ± 0.27	0.40
	Innate	0.64 ± 0.09	0.64 ± 0.09	0.67	0.66	0.02 ± 0.26	0.003 ± 0.26	0.41
	Nonimmune	0.66 ± 0.09	0.65 ± 0.09	0.68	0.68	0.02 ± 0.25	0.005 ± 0.25	0.39
	MHC-DRβ	0.50 ± 0.48	0.46 ± 0.36	0.50	0.33	0.08 ± 0.89	−0.001 ± 0.66	0.59
	MHC-DQβ	0.80 ± 0.19	0.78 ± 0.17	0.83	0.83	0.06 ± 0.95	−0.016 ± 0.82	0.30

For each study population and functional gene category, we present mean and median values of Q_ab as well as mean standardized values of R_ab. Significant differences in standardized values were assessed by comparing values for reproductive partners to the distribution of values for randomly generated pairs; no significant contrasts were detected.

Table 4.

Pairwise measures of allelic dissimilarity (AD_ab) for reproductive partners (N = 13 for BC; N = 12 for HNHR) and randomly generated male–female pairs of P. californicus (1000 permutations)

Gene categories		AD ab				Z ab
		Mean ± SD		Median		Mean ± SD
		Rep. partners	Nonpartners	Rep. partners	Nonpartners	Rep. partners	Nonpartners	P value
BCR	Adaptive	0.21 ± 0.08	0.20 ± 0.07	0.22	0.19	0.19 ± 1.09	−0.01 ± 1.00	0.60
	Innate	0.24 ± 0.02	0.24 ± 0.01	0.23	0.24	−0.18 ± 1.46	0.02 ± 1.00	0.43
	Nonimmune	0.23 ± 0.01	0.23 ± 0.01	0.23	0.23	−0.09 ± 1.19	0.02 ± 1.00	0.45
	MHC-DRβ	0.13 ± 0.25	0.10 ± 0.22	0	0	0.10 ± 1.11	−0.001 ± 1.00	0.81
	MHC-DQβ	0.21 ± 0.31	0.15 ± 0.23	0	0	0.07 ± 1.38	−0.14 ± 0.99	0.79
HNHR	Adaptive	0.24 ± 0.07	0.22 ± 0.07	0.23	0.22	0.25 ± 1.00	0.01 ± 1.00	0.58
	Innate	0.23 ± 0.01	0.23 ± 0.01	0.23	0.23	0.04 ± 0.49	0.01 ± 1.00	0.45
	Nonimmune	0.22 ± 0.01	0.22 ± 0.01	0.22	0.22	−0.28 ± 1.16	0.004 ± 1.00	0.34
	MHC-DRβ	0.26 ± 0.43	0.24 ± 0.35	0	0	0.05 ± 1.23	−0.0001 ± 1.00	0.62
	MHC-DQβ	0.08 ± 0.16	0.09 ± 0.13	0	0	−0.10 ± 1.24	−0.04 ± 1.00	0.63

For each population and functional gene category, we present mean and median values of AD_ab as well as mean standardized values of Z_ab. Significant differences in standardized values were assessed by comparing values for reproductive partners with the distribution of values for randomly generated pairs; no significant contrasts were detected.

Figure 4.

Pairwise estimates of coefficients of identity (Q_ab) for (A) BCR and (B) HNHR. For each population, separate coefficients of identity were calculated for each functional gene category. The x axis indicates the estimated coefficient of identity; the y axis depicts the frequency density of pairs displaying a given degree of identity. The distributions of values for reproductive partners (N = 13 for BC; N = 12 for HNHR) are indicated with dashed lines; the distributions for randomly generated male–female pairs (1000 permutations) are indicated with solid shapes. Inverted triangles denote coefficients of identity for each reproductive pair sampled. Vertical lines indicate the means for partners (dashed) and randomly generated pairs (solid).

Disassortative mating—genetic distance

For both populations and for all gene categories, values of Dxy_i did not differ between reproductive partners and randomly selected male-female pairs (all P > 0.05; 1000 permutations). Further, values of Dxy_i did not differ between gene categories in either population (Dunn’s Tests with Bonferroni correction; both P > 0.05).

Disassortative mating at MHC genes

Analyses of the MHC Class II DRβ and DQβ genes revealed no significant differences between reproductive partners and randomly selected male–female pairs for any of the metrics considered (coefficients of identity, allelic dissimilarity, genetic distance; all P > 0.05; 1000 permutations per metric; Tables 3, 4, and 5).

Table 5.

Pairwise nucleotide differences (genetic distance) between individuals

Gene categories		Mean ± SD		Median		P value
		Reproductive partners	Nonpartners	Reproductive partners	Nonpartners
BCR	Adaptive	1.62–04 ± 5.06−05	1.40–04 ± 5.28−05	2.0–04	1.0–04	0.55
	Innate	2.00–04 ± 0	1.94–04 ± 2.47–05	2.0–04	2.0–04	1
	Nonimmune	1.61–04 ± 5.06−05	1.35–04 ± 4.78–05	2.0–04	1.0–04	0.60
	MHC Class II DRβ	9.23−05 ± 2.40–04	9.2−06 ± 6.00−05	0	0	0.97
	MHC Class II DQβ	1.54–04 ± 1.98–04	6.93−05 ± 1.44–04	0	0	0.76
HNHR	Adaptive	1.75–04 ± 6.22−05	1.93–04 ± 7.42−05	2.0–04	2.0–04	0.28
	Innate	1.92–04 ± 2.89–05	1.90–04 ± 3.35–05	2.0–04	2.0–04	0.17
	Nonimmune	1.58–04 ± 5.15−05	1.57–04 ± 4.94–05	2.0–04	2.0–04	0.45
	MHC Class II DRβ	2.33–04 ± 4.00-04	2.64–04 ± 3.66–04	0	0	0.61
	MHC Class II DQβ	9.17−05 ± 1.57–04	1.32–04 ± 1.45–04	0	2.0–04	0.45

Data for each population and functional gene category are shown separately. Estimates of mean genetic distance (Dxy_i) are shown for reproductive partners (N = 13 for BC; N = 12 for HNHR) and randomly generated male–female pairs (1000 permutations). Significant differences were assessed by comparing values for reproductive partners with the distribution of values for randomly generated pairs; no significant contrasts were detected.

Selection on the MHC-DRβ gene

We found no within-population polymorphisms or fixed differences between populations within exon 2 (207 bp) of the MHC Class II DRβ locus. Tests for selection on this exon revealed no significant deviations from neutral expectations in P. californicus. Comparison of the models examined using a likelihood ratio test (LTR; 2 × −380.59694 − (−381.749144)) = 2.30) failed to detect evidence of significant positive selection on the P. californicus lineage (X² = 2.30, P value > 0.10). No individual sites were found to be under positive selection (w > 1, Bayesian posterior probability < 0.95).

Kinship among male–female pairs

When all SNP markers were considered, the mean pairwise coefficient of relatedness (r) among randomly generated male–female pairs tended to be low for both study populations (Hastings: r = 0.004 ± 0.03; Big Creek: r = 0.008 ± 0.04), suggesting that, overall, males and females in our study populations were not closely related. In both populations, the largest r-values detected (r ~ 0.40) occurred between members of 4 randomly generated pairs of individuals (N = 3 in BCR; N = 1 in HNHR). When analyses were restricted to reproductive partners, only 2 (0.8%) of the 25 pairs analyzed (both populations pooled) had r values > 0, indicating that in general reproductive partners were not closely related to each other; the 2 exceptions detected, both from BCR, had r values of ~0.06, suggesting that members of these pairs were also not closely related to each other.

When separate coefficients of relatedness were calculated for each gene category, values for innate and nonimmune loci revealed generally low levels of relatedness among all individuals in both populations (Table 6). Mean r values among individuals for adaptive loci (0.12 ± 0.20 in both populations) were significantly greater than those for either of the other gene categories (Mann–Whitney U tests, all one-tailed P < 0.001; Table 6; Figure 5). There were no statistically significant differences in the degree of relatedness between reproductive partners and randomly generated pairs for any of the gene categories examined (Wilcoxon’s signed rank tests, all one-tailed p > 0.05).

Table 6.

Mean pairwise coefficients of relatedness (r) for P. californicus

		Mean ± SD
Gene categories		Reproductive partners	Nonpartners	ALL
BCR	Adaptive	0.08 ± 0.20	0.12 ± 0.19	0.12 ± 0.20
	Innate	0.014 ± 0.034	0.004 ± 0.05	0.012 ± 0.05
	Nonimmune	0.008 ± 0.030	0.0014 ± 0.05	0.008 ± 0.05
	All categories	0.02 ± 0.12	0.04 ± 0.12	0.008 ± 0.04
HNHR	Adaptive	0.08 ± 0.15	0.12 ± 0.19	0.12 ± 0.20
	Innate	0	0.008 ± 0.032	0.006 ± 0.03
	Nonimmune	0.001 ± 0.003	0.0016 ± 0.034	0.004 ± 0.03
	All categories	0.02 ± 0.08	0.04 ± 0.12	0.004 ± 0.03

For each population and functional gene category, mean (±SD) values of r are shown for reproductive partners (N = 13 for BC; N = 12 for HNHR), randomly generated male–female pairs (N = 156 for BC; N = 132 for HNHR), and all individuals sampled (N = 169 for BC; N = 144 for HNHR). No significant differences (Wilcoxon’s signed rank test, all P > 0.05) were detected between reproductive partners and randomly generated pairs.

Figure 5.

Pairwise estimates of coefficients of relatedness (r) for all possible male-female pairs at (A) BCR and (B) HNHR. Data for adaptive (N = 60 SNPs), innate (N = 1880 SNPs), and nonimmune (N = 3694 SNPs) genes are indicated with differently dashed lines. The x axis indicates values of r; the y axis depicts the frequency density of pairs displaying a given degree of relatedness. No significant differences among gene categories were detected for either population.

Correlations between gene categories

Because our study populations displayed similar patterns of genetic variation, to increase statistical power, data from BCR and HNHR were pooled for analyses of correlations between measures of disassortative mating for different gene categories. For both reproductive partners and randomly generated pairs, coefficients of identity were significantly positively correlated for all gene categories (all r > 0.70; all P < 0.001). Similarly, for both reproductive partners and randomly generated pairs, values of allelic dissimilarity were significantly positively correlated between innate and nonimmune genes (partners: r = 0.53; P = 0.006; random pairs: r = 0.26; P < 0.001). Among reproductive pairs, values for allelic dissimilarity at adaptive genes, however, were not significantly related to either innate genes (r = −0.30; P = 0.14) or nonimmune genes (r = −0.27; P = 0.18). Among randomly selected pairs, the correlation between adaptive and innate genes was not significant (r = −0.08; P = 0.13) whereas that between adaptive and nonimmune genes was marginally significant (r = 0.48; P = 0.04). The latter outcome suggests that allelic dissimilarity between reproductive partners at adaptive genes is independent of overall genetic dissimilarity between these animals. More generally, it indicates that coefficients of identity and allelic dissimilarity provide distinct measures of genetic differences among individuals.

DISCUSSION

Our results do not support the hypothesis that the choice of reproductive partners in California mice is based on disassortative mating at MHC Class I or II genes. Analyses of nucleotide diversity and allelic dissimilarity at 13 MHC loci failed to provide evidence that reproductive partners were more dissimilar than randomly generated male–female pairs. Similarly, coefficients of relatedness revealed that the degree of kinship did not differ between reproductive partners and randomly generated pairs for any of the gene categories considered. We found no evidence of positive selection on exon 2 of the MHC Class II DRβ locus, one of the most commonly analyzed markers in studies of natural populations of vertebrates. Although overall levels of genetic variability did not differ between our study populations, these populations were more genetically differentiated at adaptive immune genes than at innate immune or nonimmune genes, consistent with differences in selection on adaptive genes imposed by local communities of pathogens. Although the number of reproductive pairs examined may have limited the power of our analyses to detect genetic differences among pairs of individuals, none of the analytical approaches employed revealed evidence disassortative mating based on MHC genes for either study population. Thus, although analyses of larger numbers of reproductive pairs are warranted, our data suggest that local pathogen exposure, not disassortative mating, underlies variability at MHC Class I and II genes in California mice.

Comparisons between study populations

Our analyses revealed similar patterns of genetic variation in both populations of P. californicus examined. In particular, data from BCR and HNHR were consistent in indicating that reproductive partners were not more genetically distinct than randomly selected male–female pairs for any of the categories of genes considered. Furthermore, reproductive partners were not more genetically distinct than randomly generated pairs at the MHC Class II DRβ and DQβ loci. Thus, the absence of disassortative mate choice based on MHC or other adaptive immune genes appeared to be consistent across study populations.

In both populations, although relatively greater levels of diversity were detected at adaptive genes, the highest percentages of loci that deviated from Hardy–Weinberg expectations occurred among innate genes. We did not observe deviations from Hardy–Weinberg expectations at MHC Class I and II loci. Furthermore, estimates of Tajima’s D were generally negative in both populations for all categories of genes considered, providing no evidence for balancing selection at MHC Class I and II loci. Locus-specific analyses revealed only 1 MHC Class I gene that was under potential purifying selection in both study populations, with no evidence of enhanced selection on the widely studied exon 2 of the MHC Class II DRβ locus. These results, though contrary to expectations that MHC Class I and II genes are subject to balancing selection (Hedrick 1999; Jeffery and Bangham 2000; Bernatchez and Landry 2003; Spurgin and Richardson 2010; Eizaguirre et al. 2012), are not unprecedented. Reduced polymorphism and selection at MHC Class I and Class II genes may reflect decreased exposure to pathogens (Slade 1992, Klein et al. 1993; Klein 1998), which is in line with expectations for monogamous systems (McLeod and Day 2014). Although BCR and HNHR are located in relatively close geographic proximity, these sites are separated by the Santa Lucia Mountains and thus differ ecologically in several potentially important ways, including rainfall regimes and dominant habitats (California observed precipitation maps: www.cnrfc.noaa.gov; 1981–2010 Climate normals: www.cnrfc.noaa.gov; Thornthwaite 1931). Although values of F_ST between these localities tended to be low for nonimmune genes, values of F_ST were considerably higher for adaptive and innate gene categories, suggesting greater response to local environmental conditions at immunologically relevant loci. Thus, though overall patterns of genomic variability did not differ between the study populations, potentially important differences in genetic variation at immune related loci were found between study sites.

Genomic perspectives on mate choice

Previous studies of immunogenetic diversity and mate choice have typically focused on a limited set of markers, often one to a few Class II MHC genes. Although some of these analyses have also included data from neutral or nonimmune markers (e.g. Huchard et al. 2010; Ferrandiz-Rovira et al. 2016; Galaverni et al. 2016; Table 1), these additional data have typically been derived from microsatellite markers, the suitability of which for assessing genome-wide patterns of neutral variability has been questioned (Vali et al. 2008; Miller et al. 2014). More generally, the results of studies based on one to a few MHC loci often differ markedly, with relationships between mate choice and these genes varying across loci and, in some cases, across taxa analyzed using the same loci (e.g. Schwensow et al. 2008b; Huchard et al. 2013; Kamiya et al. 2014). Together, these aspects of locus-specific studies have made it challenging to draw general conclusions regarding the role of MHC or other immunologically important genes in mate-choice decisions.

In part, the limited number of genes included in previous studies reflects technical constraints imposed by traditional Sanger sequencing strategies; these limitations do not affect genomic sequencing procedures, which allow the generation of much larger quantities of genetic information (Mardis 2008). Although the resulting data allow evaluation of variability at large numbers of loci, they may not be as effective for analyses of specific genes. For example, although our data set encompassed more than a dozen loci associated with the adaptive immune system, the final list of genes examined did not include the MHC Class II DQα gene—a commonly used locus in previously published studies of MHC variation and behavior (e.g. Sommer et al 2002; Cutrera and Lacey 2006; Chaix et al. 2008; Galaverni et al. 2016). This gene failed to meet the quality control standards implemented during our analyses, probably due to the established tendency for this locus to be characterized by gene duplications and the presence of multiple pseudogenes (Nei et al. 1997; Edwards and Hedrick 1998). As a result, we were unable to compare directly the results of our analyses with those for previously published studies based on the DQα locus, although our data set did allow direct assessment of another well-studied immunogene, the MHC Class II DRβ locus. Because relationships between mate choice and individual immunologically active loci are expected to vary, we suggest that the considerably larger number of adaptive immune genes contained in genomic data sets provide a more robust assessment of the role of MHC Class I and II variability in mate-choice decisions.

Importance of mating system

One factor that may substantially affect the role of MHC diversity in mate choice is the larger context of the mating system in which reproductive decisions are made (Sugg et al. 1996). Although Classic sexual selection theory predicts that careful choice of reproductive partners should always be favored among females (Trivers 1972), the relative importance of such choice probably varies among mating systems, as does the extent to which females benefit by choosing MHC dissimilar partners. Such choices may be particularly critical for females in monogamous populations, in which a female’s fitness may depend largely, if not entirely, upon her choice of a reproductive partner (Orians 1969). Although P. californicus is socially and genetically monogamous (Ribble and Salvioni 1990; Ribble 1991)—suggesting that the choice of reproductive partners has substantial fitness consequences for both females and males in this species—we found no evidence of disassortative mating at adaptive immune genes in our study populations. The only other study to have considered MHC variation and mate choice in a genetically monogamous mammal, the Malagasy giant rat (Hypogeomys antimena), reported that male–female pairs tended to be more similar at the MHC Class II DRβ locus than randomly selected pairs of adults (Sommer 2005). Although Kamiya et al. (2014) reported a tendency toward greater MHC dissimilarity regardless of the mating system type (i.e. social or extra-pair mating), their meta-analyses focused on the degree of extra-pair parentage of young rather than social monogamy versus other behavioral mating systems, thereby limiting the utility of their data for assessing the effects of monogamy on MHC-based patterns of mate choice. Thus, although mating system is expected to be an important determinant of the nature and extent of mate-choice decisions by females, these relationships are complex and are unlikely by themselves to provide clear predictors of patterns of immunogenetic variability among reproductive partners.

Role of environment

Within the context of a given mating system, the specific traits used by females to select mates may vary according to the benefits associated with different male phenotypes. Those benefits, in turn, may be substantially influenced by the environment in which a female lives. The primary benefit expected from disassortative mating at MHC Class I and Class II genes is increased immunogenetic diversity among offspring (Penn and Potts 1999; Neff and Pitcher 2005), the relative importance of which probably varies across both species and environments. For example, the diversity of sexually transmitted pathogens detected in P. californicus at BCR was significantly less than that in a sympatric population of the highly promiscuous P. maniculatus (MacManes 2011a; MacManes 2011b), suggesting that the benefits of disassortative mating based on MHC genetic variation should be greater in the latter species; consistent with this, selection on the MHC-DQα gene appeared to be enhanced in the latter species (MacManes and Lacey 2012). Although differences in selection on adaptive loci were not detected between our study populations, differentiation between BCR and HNHR was greater for these genes than for either innate or nonimmune genes, a finding that may reflect response to differences in local pathogen exposure. Accordingly, future efforts to examine MHC mediated mate choice should take into consideration the infection status of animals as well as the behavioral, life history, and environmental variables that affect that status.

SUPPLEMENTARY MATERIAL

Supplementary data are available at Behavioral Ecology online.

FUNDING

This work was supported by the Museum of Vertebrate Zoology (Carl B. Koford Memorial Fund); the American Society of Mammalogists (Grants-in-Aid); the Animal Behavior Society (Student Research Grant); the Sigma Xi society (Grants in Aid of Research); and the University of California Natural Reserve System (Mildred E. Mathias Graduate Student Research Grant). This work used the Vincent J. Coates Genomics Sequencing Laboratory at UC Berkeley. This work used the Vincent J. Coates Genomics Sequencing Laboratory at UC Berkeley, supported by NIH S10 Instrumentation Grants S10RR029668 and S10RR027303.

The Museum of Vertebrate Zoology, the American Society of Mammalogists, the Animal Behavior Society, the Sigma Xi society, and the University of California Natural Reserve System.

Data accessibility: Analyses reported in this article can be reproduced using the data provided by Meléndez-Rosa et al. (2018).