Introduction

The spread of infectious diseases among humans, wildlife, and domestic animals continues to impact human heath, the global economy, and wildlife conservation (Daszak et al. 2000; Cunningham et al. 2017). Approximately 75% of emerging infectious diseases of humans are zoonotic (Jones et al. 2008), including multiple coronavirus epidemics, which transferred from bats to intermediate hosts to humans (Frutos et al. 2021). As a result of devastating wildlife epidemics (e.g. chytridiomycosis, a fungal infection that caused declines and extinctions of numerous amphibian species (Fisher and Garner 2020)), infectious diseases are increasingly recognized as important for wildlife management and conservation (Blanchong et al. 2016). Across multiple study systems and at both the population and individual scales, heterozygosity has been associated with reduced infection prevalence and abundance. Yet, not all studies have found such associations (reviewed in (Blanchong et al. 2016)) and the relationship between heterozygosity and infection can be context-dependent (Gompper et al. 2011; Ruiz-López et al. 2012). Thus, whether, when, and how genome-wide genetic diversity buffers individuals and populations against infectious disease risk is a critical and open question for understanding and managing infectious diseases in wildlife, and also has implications for public health (Blanchong et al. 2016; DeCandia et al. 2018).

Correlations between genome-wide patterns of heterozygosity and fitness-related traits are the product of at least three non-exclusive hypotheses: 1) the direct fitness benefits of the loci studied, 2) the fitness benefits of linked loci, or 3) the general fitness costs of homozygosity throughout the genome (Hansson and Westerberg 2002). Indeed, both candidate gene and neutral genetic diversity correlated with parasite resistance in Mediterranean mouflon (Portanier et al. 2019). Adding infection into the heterozygosity-fitness paradigm seems simple at first; parasite infection can mediate inbreeding depression because less heterozygous individuals may have weaker resistance to infection (Acevedo-Whitehouse et al. 2005; Reid et al. 2007), incur higher infection rates (Luikart et al. 2008; Voegeli et al. 2012), or experience higher parasite-induced morbidity (e.g. reduced condition)(Acevedo-Whitehouse et al. 2006; Murray et al. 2013). Tolerance, an organism’s ability to limit the fitness costs of infection, also is known to have a genetic basis (Råberg et al. 2007). However, these mechanisms are physiologically linked within hosts, with bi-directional interactions well established among immunity, infection, condition, and reproduction (French et al. 2009). Indeed, a study of wild Alpine ibex used a path analysis approach to infer that the positive relationship between heterozygosity and horn growth, a fitness-related trait, was mediated by helminth abundance and body mass (Brambilla et al. 2015). Similarly, more inbred Soay sheep experienced higher overwinter mortality only during periods of high population density, but deworming ameliorated this effect (Coltman et al. 1999). Considering these within within-host interactions might help elucidate relationships among environmental conditions, infection, and heterozygosity.

Since multiple factors are known to contribute to infection in wildlife, we examined the relationship between genome-wide diversity, as well as contributions of specific associated loci, in deer mice (Peromyscus maniculatus) and infection with a gastro-intestinal, fecal–oral transmitted coccidian parasite. Coccidia infections are a prevalent, endemic infection in mice and have been shown to affect hosts’ susceptibility to other infections (Pedersen and Antonovics 2013). Moreover, resistance to coccidia is known to have a genetic basis in other systems (e.g., Kim et al. 2006; Beraldi et al. 2007a). Deer mice carry multiple zoonotic infections, seven of which have been deemed “high priority” by the National Institute of Allergy and Infectious Disease (Barbour 2017, NIAID 2018). Although they have been proposed as a model organism for studying natural variation and adaption (Bedford and Hoekstra 2015) and have numerous genetic and genomic resources available (Kenney-Hunt et al. 2014; Harris et al. 2015; BCM-HGSC 2016; Long et al. 2019), there are surprisingly only two study of heterozygosity and infection in Peromyscus. Specifically, population-level allo-enzyme heterozygosity was negatively correlated with the prevalence of a gastrointestinal nematode (Meagher 1999), and a study of a single MHC-II gene found that rare alleles, but not heterozygosity, was associated with gastrointestinal helminth infection (André et al. 2017). We hypothesize that host genetic variation contributes to likelihood of coccidia infection, infection severity, and infection intensity, and will affect the likelihood of nematode coinfection.

We utilize full genome sequencing to examine per-locus measures of heterozygosity to assess the contribution of neutral genetic diversity to infection susceptibility and severity, as well as association tests to identify potential candidate genes of interest. Additionally, per-individual measures of genome-wide heterozygosity were used to test for relationships among host condition, heterozygosity, and infection. This genome-wide approach is more comprehensive than earlier neutral microsatellite methods, which may not be a good proxy for genome-wide variation and have yielded mixed findings about heterozygosity-infection relationships in wildlife (reviewed in: Blanchong et al. 2016; DeCandia et al. 2018). Genome-wide methods are also increasingly preferred over targeted gene approaches (e.g. MHC or TLR receptors) because they are not focused on a single aspect of defense (Blanchong et al. 2016; DeCandia et al. 2018). For example, genome-wide association tests, such as we use here, can uncover potential functional genes contributing to coccidia infection that may be missed from a targeted gene approach. Our combination of genome-wide per-locus and per-individual heterozygosity approaches, along with genome-wide association tests, provides a way to begin understanding the contributions of genetic factors and potential environmental-dependencies of parasite-mediated inbreeding depression.

Materials and methods

Field sampling methods

Deer mice (Peromyscus maniculatus) were live-trapped at the Bernard Field Station in Claremont, CA from June 27 to August 9, 2019. Eight 40 × 40 m grids with traps spaced 10 m apart (25 traps/grid) were established in paired locations throughout the sage-scrub habitat. At each grid, Sherman traps were baited with seeds for three nights in a row. Three-night trap** sessions were repeated at each grid every two weeks. Upon their first capture each session, mice were weighed, measured, and a retro-orbital blood sample was collected. Each mouse was also given an individually-numbered ear tag upon its first capture. Females were considered reproductive if they were pregnant or lactating, and males were classified as reproductive based on testes size. Fecal samples from each individual were collected from the traps, weighed, and stored in neutral-buffered formalin. Blood samples were kept on ice, then spun at 2000 g for 10 min to separate plasma. Plasma and pellet fractions were frozen separately at -80C. Host trait and parasite infection data were used from each individual’s initial capture, while the blood samples sequenced sometimes came from subsequent recaptures. An average of 3.9 ± 0.6 (mean ± se) mice were included from each grid (range: 2–7 mice).

Parasite and condition measurements

To determine body condition, we regressed weight with body length then used the residuals as a measure of condition. Mass-length residuals are the most common condition metric used in condition-infection studies (Sánchez et al. 2018), and this metric has been shown to be highly correlated with body fat in deer mice (Schulte-Hostedde et al. 2005). The number of coccidia oocysts and helminth eggs per g of feces were counted using a modified McMaster egg fecal counting protocol (MAFF 1980). There were 9 uninfected and 22 coccidia-infected individuals. Following the methods of Ezenwa et al. (2021), we used a model-based approach to identify statistically significant clusters in the oocyst intensity data (S3) using the mclust package (Fraley et al. 2021). We then classified infected individuals into two groups based on these clusters [Low (< 100 oocysts per gram feces (opg)): n = 11; High (> 100 opg): n = 10]. This also corresponded to a natural break in the data since no samples had between 80–135 opg. For two individuals, feces weight data were not collected, so only infection status, not abundance, could be determined. One of these two individuals only had a single coccidia oocyst detected, so it was included in the Low infection severity category.

Lastly, we classified individuals as whether or not they were coinfected with coccidia (Eimeria sp.) and gastrointestinal nematodes (Aspicularis americana (a pinworm) and 3 stryongylid (hookworm) species) (Uninfected: n = 9; Low coccidia (< 100 opg): n = 9; High coccidia (> 100 opg): n = 7; Coinfected: n = 5; Table S2). Due to their similar biology and the general suite of host immune responses they trigger, grou** all gastrointestinal nematodes together for analysis is common for genetic studies of host resistance (Oppelt et al. 2010; Brambilla et al. 2015; Portanier et al. 2019; Ezenwa et al. 2021; Ahbara et al. 2021) as well as coccidia-nematode coinfection studies (Knowles et al. 2013; Pedersen and Antonovics 2013; Gorsich et al. 2014; Yan et al. 2021).

Sequencing and bioinformatics workflow

DNA was extracted from heparin-treated blood pellet samples with the Qiagen DNEasy Blood and Tissue Extraction Kit. DNA was quantified on a Qubit™ 2.0 fluorometer using a dsDNA High Sensitivity Assay Kit (Invitrogen). Sequencing libraries were prepared by Admera Health with a KAPA Hyper Prep Kit. Individually barcoded libraries were sequenced on an Illumina NovaSeq S4 (paired end, 2 × 150 bp), at a target depth of ~ 200 million PE reads per sample.

We conducted an initial assessment of read quality using FASTQC v0.11.8 with default values (Andrews 2010). TRIMMOMATIC v0.39 was then used to trim adapter sequences, the leading and trailing three bases, and low quality sequences ( leading:3 trailing:3 slidingwindow:4:15 minlen:36; Bolger et al. 2014). Filtered sequences were mapped to the P. maniculatus genome (GCF_000500345.1_Pman_1.0_genomic.fna) using BWA mem v0.7.17 with default values (Li). After sorting (“sort”) and indexing (“index”) BAM files with SAMTOOLS v1.14 (Li et al. 2009), PICARD v 2.26.1 was used to remove duplicates (“MarkDuplicates”) and add read groups ("AddOrReplaceReadGroups"; validation_stringency = lenient; Broad Institute 2019).

We called variants in the parallel mode of FREEBAYES v1.3.5 with 16 threads, resulting in an initial total of 153,221,207 variant sites (Garrison and Marth 2011; Danecek et al. 2021). After filtering, 573,838 high quality SNPs were retained for downstream analyses.

Genetic diversity analyses

We predicted that individuals with higher levels of genetic diversity would be less likely to be infected and have less severe infections. To this end, we compared genetic per-locus diversity metrics across individuals varying in (a) coccidia infection status (Uninfected/infected); (b) coccidia severity (Uninfected/Low/High); and (c) coinfection status (Uninfected/Low/High/coinfected). Per-locus measures of diversity average heterozygosity metrics across each SNP for individuals within each infection-status group for each site in the genome (n = 573,838 SNPs per group). We used the basicstats function of hierstat v0.5–7 to calculate per-locus observed heterozygosity (HO), expected heterozygosity (HE), and the inbreeding coefficient (FIS)(Goudet and Jombart 2020). In addition to the measures of heterozygosity described here, we wanted to estimate the number of segregating alleles per group or “allelic richness”. Because allelic richnness can be sensitive to small sample size (Pruett and Winker 2008), we used a rarefaction approach in ADZE v1.0 to calculate per-locus allelic richness (AR) and private allelic richness (PAR) (Szpiech et al. 2008). For AR, we standardized sample sizes from two to the number of individuals in the largest group (infection status: 22; severity: 11; coinfection: 9). For PAR, we set the standardized sample size to the size of the smallest group (9 for infection status and severity; 5 for coinfection). All other parameters were default values. Because the number of coinfected individuals is lower than the other categories, we only considered analyses that accounted for differences in sample sizes (i.e., PAR and AR).

Per-individual genetic diversity metrics were calculated in vcftools v0.1.17 and included HO, HE, and FIS. Per-individual measures of diversity average across all loci in the genome for each individual (n = 31 individuals). We assessed statistical differences of genetic diversity means among groups using Welch’s t-test (infection status: Uninfected vs. Infected; sex: male vs. female) or ANOVA (severity: Uninfected vs. Low vs. High; coinfection: Uninfected vs. Low vs. High vs coinfected) in R version 4.0.4 (R Core Team 2021). Tukey’s HSD was applied for post-hoc comparisons where appropriate. For rarefaction metrics (AR, PAR), we calculated 95% confidence intervals in ADZE v1.0. VCFs were manipulated using the vcfR v1.12.0 (Knaus and Grünwald 2017) package and figures were made with the ggplot2 v3.3.5 (Wickham 2016) and cowplot v1.1.1 (Wilke 2020) packages in R.

We used a linear modeling approach to determine how per-individual genetic diversity, as well as other factors, influenced the response variables of infection status, abundance and intensity. For infection status, individuals were coded as Uninfected and Infected (as above). Abundance was measured as the log oocysts per gram (opg) coccidia of all individuals and included Uninfected individuals. Intensity was measured as the log oocysts per gram coccidia in Infected individuals only. Covariates included a metric of individual-level genetic diversity (either HO or FIS), sex (male/female), age (adult/sub-adult), grid (trap** location: HM, NK, RC, FH), and residual body condition. Genetic diversity metrics for each individual were standardized by subtracting the population mean and dividing by the standard deviation. To make sure that potential relatedness of mice within grids was not affecting our results, we used ANOVAs to confirm that grid was not associated with per-individual HO, HE, and FIS (HO: F = 0.41, p = 0.90; HE: F = 0.81, p = 0.60; FIS: F = 0.43, p = 0.89). Additionally, all association tests (below) controlled for pairwise genetic relatedness.

For testing the relationship between per-individual heterozygosity and infection status, we built general linearized models fitted to a binomial distribution with a logit link function to determine the significance of fixed effects. For abundance and intensity, we built general linearized models fitted to gaussian distributions. Backward stepwise model selection was performed by comparing AIC values in R with the stepAIC function from the MASS package (Venables and Ripley 2002). Linear models for host condition were parameterized similarly, with covariates including: a metric of individual-level genetic diversity (either standardized HO or FIS), sex (male/female), age (adult/sub-adult), grid (trap** location: HM, NK, RC, FH), and log-transformed coccidia abundance. We also included an interaction between the metric of genetic diversity and coccidia abundance as a measure of tolerance. Backward stepwise model selection was performed by comparing AIC values in R with the stepAIC function. Post-hoc power analyses for regression and binomial analyses were conducted using the pwr and pwr2ppl packages, respectively (Aberson 2019; Champely et al. 2020).

Association tests

We used GEMMA v0.98.4 to perform univariate single-marker association tests to identify loci associated with infection status (Zhou and Stephens 2012, 2014). Sex was included as a covariate and a centered pairwise genetic relatedness matrix was calculated to account for population structure (gk -1). Infection status was defined as Uninfected = 1 and Infected = 2. The lrt P-values were adjusted using a false discovery rate (Benjamini and Hochberg 1995). We used the relevant.gff file to determine whether significant loci were in genes. When necessary, location IDs were converted to gene names in NCBI gene (https://www.ncbi.nlm.nih.gov/gene). For the set of genes that contained a significantly associated locus, we performed a gene ontology analysis using the gost function from the gprofiler2 package v0.20.0 (Kolberg et al. 2020). Gene functions were ascertained in the Online Mendelian Inheritance in Man (https://www.omim.org/), NCBI gene (https://www.ncbi.nlm.nih.gov/gene), and/or Uniprot (https://www.uniprot.org/) databases. Because the genomic and immunological basis of resistance to coccidia infection has been previously investigated (e.g. Chapman et al. 2013; Boulton et al. 2018), we cross-referenced the functions of genes with associated loci with previously identified potential genes and pathways involved in coccidia resistance.

Results

Sampling and sequencing summary

We performed whole genome sequencing and disease phenoty** on 31 P. maniculatus (Table S1). Mice were classified according to infection status (Infected vs. Uninfected), severity (Uninfected, Low, High), and coinfection with nematodes. After trimming adapters and low quality regions, individuals had an average of 350 million reads retained yielding an average coverage of 20.77. Initial variant calling produced 153,221,207 variable sites. After filtering, 573,838 high quality SNPs were retained for downstream analyses.

Infection status

When considering infection status, Uninfected mice had higher levels of per-locus genetic diversity than Infected mice across most metrics (Fig. 1). Per-locus, Uninfected mice had higher levels of HO (two-sided t-test; Inf = 0.486 ± 0.0004, Uninf = 0.492 ± 0.0004, t1138387 = 11.944, p < 0.001), HE (two-sided t-test; Inf = 0.357 ± 0.0002, Uninf = 0.358 ± 0.0002, t1136475 = 3.179, p = 0.001479), and FIS (two-sided t-test; Inf = -0.286 ± 0.0004, Uninf = -0.311 ± 0.0.0005, t1075762 = -38.226, p < 0.001) than Infected mice (Fig. 1a,b,c). Although HO, though not HE, can be sensitive to smaller sample sizes, we note that any bias would predict groups with the lowest number of samples to have lower diversity values (Pruett and Winker 2008). Instead, we find that the smaller group (Uninfected mice, n = 9), has significantly higher levels of all three heterozygosity metrics than the larger group (Infected mice, n = 22). While AR was not different across all sample sizes tested (overlap** error bars), PAR was lower in Infected (0.0851; 95% CIs: 0.0847–0.0855) than Uninfected mice (0.091; 95% CI: 0.0906 – 0.0914) at the sample size equivalent to the Uninfected group size (n = 9; Fig. 1d,e). We also note that our SNP filtering may have affected our ability to detect differences in allelic richness, as we excluded any SNPs with more than 2 alleles and removed indels.

Fig. 1
figure 1

Genome-wide per-locus diversity metrics for mice categorized according to infection status (Uninfected: n = 9; Infected n = 22). Per-locus genomic diversity metrics are significantly lower in Uninfected vs Infected mice for a) observed heterozygosity (HO; p < 0.001), b) expected heterozygosity (HE; p = 0.001479), c) the inbreeding coefficient (FIS; p < 0.001) and, d) private allelic richness (PAR; non-overlap** 95% CIs). Across all sample sizes, e) allelic richness (AR) does not differ among groups (overlap** confidence intervals). Points are means and error bars are ± one standard error, with the exception of FIS which is shown with ± two standard errors for visualization purposes. For AR (e), the curve is the smoothed conditional mean for Uninfected (dashed) and Infected (solid) mice, and the shading represents 95% confidence intervals

Full size image

Infection severity

Mice with higher levels of per-locus genetic diversity had less severe infections across several metrics of genetic diversity (Fig. 3). HO was highest in Uninfected mice (0.492 ± 0.0004), at intermediate levels in Low-coccidia mice (0.491 ± 0.0004), and lowest in High-coccidia mice (0.482 ± 0.0004; ANOVA: F1715013 = 209.9, p < 0.001; Tukey’s HSD p < 0.001 in all pairwise comparisons; Fig. 2a). High-coccidia mice had significantly lower levels of HE (0.357 ± 0.0002) than Low-coccidia (0.358 ± 0.0002) and Uninfected (0.358 ± 0.0002) mice, though Low-coccidia and Uninfected mice were not significantly different from each other (ANOVA: F1715013 = 20.29, p < 0.001; Tukey’s HSD: High vs Low p < 0.001, High vs Uninfected p < 0.001, Low vs Uninfected: p = 0.710; Fig. 2b). Inbreeding coefficients (FIS) were associated with severity in the predicted direction (High = -0.286 ± 0.0005, Low = -0.298 + 0.0005, Uninfected -0.311 ± 0.0005; ANOVA: F1608741 = 612.7, p < 0.001; Tukey’s HSD P < 0.001 in all pairwise comparisons; Fig. 2c). Again, our results are in the opposite direction of any bias imposed by small sample sizes, as we find the smallest group (Uninfected mice) to have the largest HO. However, HE, which is less sensitive to sample size issues, only detected differences in High vs. the other two categories (Pruett and Winker 2008); this suggests that highly infected mice have lower neutral diversity, but differences in low vs uninfected mice should be interpreted with caution.

Fig. 2
figure 2

Genome-wide per locus diversity metrics for mice categorized according to infection severity (Uninfected: n = 9; Low: n = 11; High: n = 10). Per-locus genomic diversity metrics include a) observed heterozygosity (HO; all comparisons p < 0.001), b) expected heterozygsity (HE; High vs Low/Uninfected p < 0.001; Low vs Uninfected ns), c) the inbreeding coefficient (FIS; all comparisons p < 0.001), d) private allelic richness (PAR; Uninfected vs Low/High non-overlap** CIs; Low vs High ns), and e) allelic richness (AR; overlap** CIs). Points are means and error bars are ± two standard errors. For AR (e), the curve is the smoothed conditional mean for Uninfected (white, dotted), Low (medium gray, long dash), and High mice (dark grey, solid), and the shading represents 95% confidence intervals

Full size image

Allelic diversity showed mixed patterns. PAR was higher in Uninfected mice (0.0343; 95% CI: 0.0341–0.0346) than other groups, though Low (0.0313; 95% CI: 0.0311–0.0316) and High mice (0.0317; 95% CI: 0.0314–0.0319) had similar levels of PAR (Fig. 3d). AR did not differ across groups at all sample sizes tested (overlap** error bars; Fig. 2e), and previous caveats about SNP filtering apply here as well.

Fig. 3
figure 3

Coinfected individuals show (a) significantly lower per-locus levels of private allelic richness (PAR; non-overlap** CIs) and (b) a non-significant trend towards lower levels of per-locus allelic richness (AR). Points are means and error bars are ± two standard errors. For AR (e), the curve is the smoothed conditional mean for Uninfected (dashed, lightest gray), Low (long dash, medium gray), High (dotted, dark gray), and coinfected (solid, black) groups, and the shading represents 95% confidence intervals. Group colors are the same in (a) and (b). Samples sizes are as follows: Uninfected (n = 9), Low (n = 9), High (n = 7) and coinfected (n = 5)

Full size image

Coinfection

Coinfected individuals had lower levels of per-locus allelic diversity than other groups. At a sample size of 5 individuals, PAR was significantly lower in coinfected mice (0.0231; 95% CI: 0.0229–0.0232) than all other groups (Fig. 3a; Uninfected – 0.0315, 95% CI: 0.0313–0.0317; Low – 0.0287, 95% CI: 0.0285–0.0289; High – 0.0301, 95% CI: 0.0299–0.0303). Indeed, all groups had significantly different levels of PAR, with Uninfected mice having the highest values, followed by High, Low, and coinfected mice. Coinfected mice also showed a nonsignificant trend towards lower levels of AR across all samples sizes tested (Fig. 3b; low overlap of confidence intervals). While rarefaction analyses such as AR are recommended for small samples sizes, we note that our results for coinfection are in the predicted direction of any bias. Here, we find the group with the smallest sample size (coinfected, n = 5) also shows the lowest values of per-locus allelic diversity. Further work with larger sample sizes should be done to confirm this finding.

Individual-level analyses

In addition to per-locus metrics, we used a linear modeling approached to examine how per-individual levels of heterozygosity (standardized HO and standardized FIS), condition, and other factors influenced infection. Neither standardized HO or FIS affected the likelihood that an individual was infected with coccidia, though both terms were retained in reduced models (Table 1). In those models, body condition was negatively associated with infection, indicating that individuals with higher body condition were less likely to be infected. Both sex and age were retained in the final models and had marginal p-values, with females and adults perhaps being slightly more likely to be infected (Table 1).

Table 1 Results of reduced linear models for coccidia infection status (Infected vs. Uninfected), intensity (log opg), and abundance (log opg), versus residual body condition. The results of the analyses for condition as a function of HO and FIS are identical (and thus reported only once) because the heterozygosity measures drop from both models. Significant p-values are in bold, marginal values are in italics
Full size table

Looking at the quantitative measures of infection, HO and FIS were the only terms retained in the final linear models for coccidia intensity (Table 1). Their p-values were marginal with a negative trend observed for intensity and HO (p = 0.078), and a positive trend for FIS (p = 0.069). HO was not associated with coccidia abundance, but a marginal positive trend was again observed for FIS (Fig. 4a, Table 1, p = 0.083,). However, these analyses were slightly underpowered (power ≈ 0.71 for a large effect size). Condition was negatively associated with coccidia abundance (Fig. 4b, Table 1, p = 0.007). Similar to the infection status analyses, the abundance models also retained sex and age; they had marginal p-values indicating females and adults might have slightly higher infection abundances than males and subadults, respectively (Table 1, p = 0.054 (age), p = 0.074 (sex)).

Fig. 4
figure 4

Partial residual plots from linear regression models show a) the marginally positive correlation between per-individual FIS and coccidia abundance (p = 0.083), b) the negative correlation between residual body condition and coccidia abundance (p = 0.007), and c) the lack of relationship between per-individual FIS and residual body condition (p = 0.76)

Full size image

Both HO and FIS were dropped from linear models of condition (Table 1), and visually it is clear that there is no relationship between FIS and condition (Fig. 5c, p = 0.76). The interaction between genetic diversity and parasite load, a proxy measure of tolerance, also dropped from the models (Table 1). Coccidia abundance was negatively associated with condition, and subadults had lower condition indices than adults (Table 1, p = 0.030 (condition), p = 0.001 (age)). We also examined whether individual-level heterozygosity metrics varied between the sexes and found no significant differences (HO: p = 0.115, HE: p = 0.335, FIS: p = 0.119).

Fig. 5
figure 5

Effects of genomic variation on host fitness can be mediated by parasite infection via effects of heterozygosity on immune function and infection. The gray box depicts an individual host and the bi-directional interactions among immunity, infection, condition, and reproduction. The lower body condition observed in infected animals could be a) a consequence of infection, b) a cause of infection, or c) both. The observed relationships between per-locus host genomic diversity and infection could be a direct effect or mediated via host immunity, and could interact with the infection-condition association to affect host fitness and future genomic diversity (gray dashed arrow)

Full size image

Association tests

We found 73 SNPs associated with coccidia infection status that passed filtering after applying a false discovery rate (Table S3). Sixteen of the associated loci fell within annotated genes (Table 2). For 35 loci, the associated SNP was associated with a reduction in disease prevalence (6 within annotated genes; i.e., negative beta coefficient), while 38 associate loci were associated with an increase in prevalence (10 within annotated genes; i.e., positive beta coefficient). The gene ontology analysis revealed significant enrichment for one molecular function (protein serine/phosphatase activity; GO:0,004,722; p = 0.024), one cellular component (RNA polymerase II, holoenzyme; GO:0,016,591; p = 0.036), and no biological processes.

Table 2 Single nucleotide polymorphism (SNP) loci that are significantly associated with disease prevalence. The complete list of associated SNPs is detailed in Table S3. Note: Scaffold number; SNP position; minor allele (A1); major allele (A2), frequency of the minor allele in the full population (f(A1)), frequency of the minor allele in uninfected individuals (f(U)); frequency of the minor allele in infected individuals (f(I)); beta coefficient; standard error of beta; P calculated from a likelihood ratio test; gene name; gene annotation
Full size table

Discussion

Identifying the factors that determine host susceptibility to infection can inform the management and evolutionary consequences of wildlife disease. We explored the genetic factors contributing to coccidia infection in Peromyscus maniculatus. Our results suggest that both neutral diversity, as well as the presence of certain genetic variants, influence coccidia infection. We find strong evidence that per-locus genetic diversity is negatively associated with infection status and severity, and weaker evidence that lower genetic diversity increases likelihood of coinfection. We also placed per-individual heterozygosity in context with several other factors that may contribute to infection state, including condition. Condition, along with a trend for individual-level heterozygosity, were negatively correlated with infection abundance and intensity, though the effects of condition and heterozygosity, since they were uncorrelated, are independent. Our results underscore the importance of considering genomic variation, in conjunction with other measures of fitness, to understand the ecology and evolution of host-parasite interactions.

Genomic diversity and infection

Loss of genomic diversity and inbreeding are hypothesized to increase host susceptibility to pathogens through decreasing the adaptive potential of hosts and unmasking deleterious recessive alleles (Spielman et al. 2004; Blanchong et al. 2016). Here, we find that genome-wide neutral variation is an important factor contributing to parasite infection in P. maniculatus. Specifically, lower levels of per-locus genomic variation result in a higher likelihood of infection and increased infection severity. These results are consistent whether considering patterns of heterozygosity (HO), measures of nucleotide diversity (HE), relative inbreeding (FIS), or allelic variation (PAR). The main effect of genomic diversity on infection is mediated through differences in patterns of heterozygosity, rather than allelic richness, with the exception that private alleles also play a role. In per-individual analyses factoring in host traits, we observed marginal relationships between standardized heterozygosity metrics and infection intensity and abundance.

Our per-locus analyses show genomic diversity affects host resistance to coccidia, suggesting more heterozygous hosts reduce their parasite load once infected (e.g. lower intensity) and perhaps clear some infections entirely (e.g. lower likelihood of infection). However, our individual-scale analyses, which had lower power due to sample size, found no relationship between heterozygosity and infection status. Previous studies from diverse systems often (Acevedo-Whitehouse et al. 2005, 2006; Townsend et al. 2018; Banks et al. 2020; DeCandia et al. 2021), but not always (Acevedo-Whitehouse et al. 2009; Townsend et al. 2018; Kubacka et al. 2020; Martin et al. 2021), find negative associations between genome-wide variation and infection status (reviewed in (Blanchong et al. 2016)). While there is well-described functional genetic variation for host susceptibility to coccidia infection (Klesius and Hinds 1979; Kim et al. 2006; Boulton et al. 2018; present study) reviewed in (Chapman et al. 2013)), here we also document a role for genome-wide neutral diversity. In P. maniculatus, the population-level prevalence of a different fecal–oral parasite (Capillaria hepatica) negatively correlates with genetic diversity, after controlling for density (Meagher 1999). Other systems that explore fecal–oral parasites have found similar relationships with genetic diversity and infection status (e.g., Coltman et al. 1999; Acevedo-Whitehouse et al. 2006), though this is not always the case (Schwensow et al. 2007; Townsend et al. 2018).

In addition to associations with infection status, we documented negative correlations between per-locus genomic diversity and parasite load. Recent work on sarcoptic mange in wolves that also used genome-wide SNP data found a similar negative relationship between per-locus measures of genomic variation and disease severity (DeCandia et al. 2021). In other systems, multi-locus heterozygosity negatively correlates with overall fecal–oral parasite loads (Brambilla et al. 2015; Mitchell et al. 2017), though the effect disappeared when examining coccidia load separately in at least one taxa (Mitchell et al. 2017). A reduction in coccidia load has also been linked to heterozygosity at specific immune loci (Rödel et al. 2020), as well as the presence of specific alleles contributing toward resistance (Kim et al. 2006; Worley et al. 2010; Boulton et al. 2018). The per-locus severity relationship (and trends at the individual level) could be due to higher parasite replication within the host. Because Eimeria undergo an asexual reinfective cycle prior to the sexual reproduction that produces the oocysts (Bowman 2009), reduced immunity in relatively inbred individuals could result in higher coccidia oocyst shedding.

Nematode-coccidia coinfected individuals had lower levels of private allelic richness than other groups and overall allelic richness followed a similar trend. Specifically, individuals with less allelic variation were more likely to be coinfected, and coinfected individuals had the fewest number of unique alleles. Both having a smaller number of alleles (AR), as well as the presence of particular unique alleles (PAR), could increase an individual’s risk of coinfection. For example, in African buffalo, particular alleles were associated with the likelihood and severity of nematode infection, and a worm-resistant phenotype increased the progression and severity of bovine tuberculosis coinfection (Ezenwa et al. 2021). Notably, our analyses were limited by low sample size due to the low prevalence of nematodes in the study population. Additionally, since we found no individuals infected with only nematodes, we cannot distinguish the possibility that the lower PAR is a consequence of nematode-infection, rather than coinfection. Therefore, further investigations with increased sampling are required to know whether our coinfection results are robust.

A positive relationship between heterozygosity and sequencing depth, as well as mis-map**, can confound estimates of genome-wide heterozygosity in studies such as the present one. At least five lines of reasoning mitigate these concerns here: i) we applied very stringent filtering to variant sites, including retaining only sites with high read depth, map** quality, and genotype quality (< 0.5% of initial calls were kept for final analyses), ii) only sites that were in Hardy–Weinberg Equilibrium in Uninfected individuals were kept, iii) such sequencing or map** errors do not predict systematic biases that vary according to infection state, iv) patterns are consistent across all metrics of genomic variation, and v) a preliminary analysis using a different bioinformatic workflow recovered similar relationships between heterozygosity and coccidia infection (PLATYPUS)(WGS500 Consortium et al. 2014). As further evidence that the infection-based patterns are real, we evaluated whether heterozygosity varied according to sex (i.e., another way to categorize individuals) and found no marginal or significant differences.

Condition, heterozygosity and infection

Host body condition was negatively correlated with coccidia infection and abundance, but was not associated with either measure of individual-level heterozygosity (HO and FIS). Thus, we found no evidence that general costs of homozygosity throughout the genome were manifest in lower body condition. Two non-exclusive mechanisms could be driving the negative correlation between infection and host condition: 1) since immune responses are energetically costly, individuals in lower condition might be less resistant to infection, and/or 2) lower condition could be a consequence of parasite infection (Beldomenico and Begon 2010; Sánchez et al. 2018). If condition is affecting infection via altered immunity (Fig. 5a), then condition and heterozygosity are having synergistic effects on infection status and intensity. However, we cannot rule out the possibility that the effects of heterozygosity on infection are also mediated via changes in host immunity and resistance. We found no evidence that heterozygosity affected host tolerance. Future analyses focused on whether immune genes experience elevated diversity relative to non-immune genes could elucidate the mechanism underlying the per-locus infection-heterozygosity correlations as well as examine if the individual-level trends reach significance with larger sample sizes and multiple timepoint infection measures.

Alternatively, the lower condition observed in infected individuals could be a consequence of coccidia infection, and thus an indirect effect of host heterozygosity (Fig. 5b). Individuals with higher parasite loads may not necessarily have decreased infection-imposed fitness. For example, while more inbred sea lions had a higher prevalence of hookworm, anemia was only associated with homozgyosity at a single locus (Acevedo-Whitehouse et al. 2006). However, in a similar longitudinal field study of deer mice, the frequency of infection with the coccidian parasite Eimeria arizonensis was associated with lower body weight and decreased overwinter survival, but these effects were dependent on sex and reproductive status (Fuller 1996). Specifically, effects of coccidia on mouse weight were found only in lactating females, and effects on survival were found only in male mice (Fuller 1996). In our study population, longitudinal infection and condition data could help determine if condition is a cause or consequence, or both a cause and consequence, of infection (Fig. 5c).

Associated loci point towards a role for the inflammatory response and echo previous work

Because coccidia resistance is known to be heritable (Kim et al. 2006; Beraldi et al. 2007b) and specific immune loci have been linked to coccidia resistance (e.g., Kim et al. 2006; Worley et al. 2010; Boulton et al. 2018; Rödel et al. 2020), we performed a genome-wide association test to identify functional candidates associated with coccidia infection in Permoyscus. Previous work has shown that the genomic basis of coccidia resistance is polygenic with both many loci of small effect and a few loci of large effect contributing to this complex trait (e.g., Kim et al. 2006, 2010; Hamzić et al. 2015; Boulton et al. 2018). Our study identified numerous SNPs across at least 16 distinct genes that are significantly associated with the coccidia infection status (Table 2). The discovery of specific associated loci complements genome-wide patterns of diversity, suggesting that both the broad benefits of heterozygosity as well as certain genetic variants contribute to disease resistance. This finding mirrors other studies of coccidia resistance in wild populations of terrestrial mammals; both diversity at candidate immune genes and neutral genetic diversity were correlated with gastrointestinal nematode abundance in European mouflon, though surprisingly this study found no relationship between heterozygosity and coccidia abundance (Portanier et al. 2019).

Many of the associated loci found here echo previous work from poultry and lab mice, suggesting that the genetic pathways controlling coccidia infection may be shared across diverse domesticated and wild species (Table 2). A host’s ability to control coccidia requires a strong, rapid inflammatory response and contributing genetic factors link to cell-mediated, humoral, and innate components of immunity (Chapman et al. 2013; Boulton et al. 2018). We find numerous associated loci that are important in the inflammatory response. At least five associated SNPs are in genes that interact with the transcription factor NF- κB (e.g., PSME4, KLF3, CD3000LD3, BIRC3, HAVCR1), a pivotal mediator of inflammatory responses that is at the center of pathways conferring coccidia resistance in other species (Alcala-Canto and Ibarra-Velarde 2008; Psifidi et al. 2016; Boulton et al. 2018; ** et al. 2020; Knights et al. 2020). One exciting functional candidate is CD300LD3 (also known as CLM-3 or CMRF35-like molecule 3), which can activate cytokine production in mast cells and lead to increased production of proinflammatory cytokines (Wu et al. 2011). CMRF35-like molecules also stimulate toll-like receptor 9 pathways and subsequently NF- κB, both of which are implicated in coccidia resistance in chickens (Boulton et al. 2018). Additionally, several candidate genes are known to show enhanced expression in neutrophils (SIRPB1, CD3000LD3, MAP7; https://www.proteinatlas.org/) or to modulate inflammatory signaling (e.g., BIRC3). A GWAS performed in chickens found that leukotriene B4 receptor 2 (LTB4R2), a cell surface receptor of the inflammatory leukotriene B4, was significantly associated with resistance to coccidia parasitism (Psifidi et al. 2016). Our study also implicates the leukotriene B4 pathway, as we find a significantly associated SNP in CYP4F14 (leukotriene-B4 omega-hydroxylase 3), a protein involved in leukotriene B4 degradation. Our results also offer novel candidates for contributing to coccidia resistance. For example, a significantly associated SNP is located in the helper-T cell-surface receptor HAVCR1 (Hepatitis A virus cellular receptor 1 homolog), which regulates cytokine production (McIntire et al. 2003). After mice were infected with hepatitis A, certain variants of this gene conferred protective immunity against non-viral diseases (McIntire et al. 2003). Future work can focus on molecular experiments with these functional candidates to investigate and validate their role in coccidia infection. Taken together, the resonance of specific associated loci found here with previously documented associations underscores our ability to detect host genetic factors contributing to the coccidia infection, despite small sample sizes and a single snapshot of a sublethal pathogen.

Conclusions and future directions

We find that increased neutral genomic diversity, as well as the presence of specific genetic variants, contribute to decreased coccidia infection in wild populations of deer mice. Our genome-wide association tests identify well-described, as well as novel, functional candidates that can be the focus of future molecular work exploring the effect of particular loci on coccidia resistance. But what is the mechanistic basis of associations between neutral genomic diversity and infection? Our study does not distinguish among whether fitness benefits of heterozygosity are conferred through direct effects of loci, functional effects of linked (likely immune) loci, or general benefits of heterozygosity (Hansson and Westerberg 2002). While it is clear that the immune system plays a role, it is not yet clear how host resistance intersects with genome-wide variation. It is likely that increased diversity at immune genes contributes in some way to our pattern, but the fact that we examined > 550,000 loci with low linkage disequilibrium suggests a genome-wide advantage of heterozygosity that is not mediated through improved condition (e.g., differences in costly anti-parasite behaviors (Luong et al. 2007)). Future analyses on candidate genes, as well as transcriptomics or experimental challenges, can help to disentangle the role of genome-wide heterozygosity vs. single-locus effects in coccidia infections. Further, our work underscores the importance of preserving genetic diversity in wild populations.

Coccidia infections can be dynamic, yet in this study, the parasite infection data came from a single timepoint, so it is surprising that the per-locus patterns were detectable, and perhaps why, in addition to sample size, the individual-level ones were marginal. The classification of individuals into uninfected, low, and high oocyst shedding groups, as was done for the per-locus analyses, might also better reflect individuals’ general infection propensities than the single oocyst count used in the individual-level intensity and abundance analyses. Given that the lifecycle of coccidia includes both intracellular and extracellular stages, host immunological defenses to coccidia are similarly multi-faceted and include both multiple innate and adaptive components of immunity (Yun et al. 2000). Thus, the relationships between per-locus heterozygosity and resistance to coccidia and the candidate genes associated with some of these SNPs may be informative for other intracellular and extracellular parasites that involve similar immune pathways. In P. maniculatus, heterozygosity-coccidia interactions are far easier to study and can provide insight into how host genetic diversity may affect rodent-borne zoonotic diseases like Sin Nombre virus (Barbour 2017). While bats carry some of the most deadly zoonotic viral infections to humans (Brook and Dobson 2015), by far more zoonoses have been identified as coming from rodents (Han et al. 2016). Moreover, it is likely that rodents will also be reservoirs of future zoonoses we have yet to encounter (Han et al. 2015), so it is critical to better understand how both neutral and functional genetic diversity contribute to disease resistance.