Immunogenetic novelty confers a selective advantage in host–pathogen coevolution

Among guppies that survived to the end of the experiment (n = 209), fish carrying only novel alleles or supertypes (designated as N/N genotypes and N/N supergenotypes, respectively; Materials and Methods) experienced G. turnbulli infections that were significantly less severe than fish carrying only local alleles or supertypes (L/L) (Fig. 2). Infection severity was measure in “worm days” (the area under a graph of number of worms against time), and analyses used AIC_C-based multimodel inference (Materials and Methods). The N/N genotypes (n = 44) and N/N supergenotypes (n = 14) respectively experienced 35.1% and 34.7% fewer worm days compared with the L/L genotype and supergenotype fish (n = 59 and 88, respectively; P = 0.003 and 0.012; SI Appendix, Tables S2.1 a and b and S2.2 a and b). L/N genotypes (n = 106; i.e., fish carrying both novel and local alleles) experienced infections of comparable intensity to L/L genotypes (P = 0.65; SI Appendix, Table S2.1b), whereas L/N supergenotypes (n = 107) experienced intermediate infection intensities that were only marginally nonsignificant relative to L/L (P = 0.055; SI Appendix, Table S2.2b). Direct comparison of the best allele-based and supertype-based models of worm days, using a constant set of covariates, indicated that allele-based groupings produced the better fit (ΔAIC_C = 4.22; SI Appendix, Tables S2.1a and S2.2a); however, fish carrying at least one novel supertype (n = 121) experienced 27.1% fewer worm days than fish with no novel supertypes but at least one novel allele (n = 29; top-ranked model; P = 0.02; SI Appendix, Tables S2.3 a and b and Fig. S2.1). We did not detect a significant interaction between replicate population and MHC genotype/supergenotype class (ΔAIC_C = +7.37/+5.71, P ≥ 0.13/0.24). Neither the number of alleles nor the number of supertypes carried by a host—measures used as analogs of heterozygosity – were significantly associated with the number of worm days experienced by hosts (ΔAIC_C = +1.82/+1.28, P = 0.52/0.32; SI Appendix, Tables S2.1 a and b and S2.2 a and b). No genetic variables were significant predictors of host mortality (SI Appendix, Tables S2.5 a–c and S2.6 a–c), despite worm load itself being a significant predictor of mortality from infection day 3 onwards (SI Appendix, Table S2.4).

Our results support the novel MHC variant hypothesis: N/N hosts experienced parasite infections that were significantly less severe than those of L/L hosts. This was the case whether novelty was defined by amino acid sequences (alleles) or by physicochemical functional groups (supertypes). Differences in parasite burden between the genotype classes did not translate into a detectable effect on host survival. However, this may reflect the relatively benign and stable conditions of the experiment: In the wild, fish weakened by infection may be more susceptible to predation (28) and secondary infections (29), and to environmental stressors such as river spates (30)—even one additional worm can reduce a wild guppy’s survival probability (30). Furthermore, besides reducing survival, parasites may reduce host fitness by affecting reproductive potential, as has previously been demonstrated in guppies (31).

The novel variant advantage that we observed could, in theory, result in either balanced polymorphism or fixation of the novel variant—i.e., it is consistent with both balancing and positive selection. When a novel variant is introduced into a natural population (by point mutation, microrecombination, or introgression), both processes are cooccurring and indistinguishable, likely resulting in an increase of the novel allele’s frequency. We explored this potential using computer simulations parameterized from our current data on the effects of novel alleles/supertypes on gyrodactylid load, and from the effect of gyrodactylid load on the survival of guppies in the field from a mark-recapture study (30). These simulations show that upwards of 11% of novel amino acid variants should successfully establish in a population, and upwards of 54% if the variant is a novel supertype (compared with <0.1% in neutral simulations; SI Appendix, Appendix S10).

In the long term, the same process that leads to novel variant advantage—adaptation of parasites to local MHC genotypes—should diminish the advantage of the variant, leading to balanced polymorphism via negative frequency-dependent selection (12, 16, 17, 32). Such dynamics, whereby a novel (or rare) allele increases in frequency, loses its advantage, and decreases in frequency again, have yet to be demonstrated. Alternatively, in the absence of balancing selection, novel variants could spread to fixation in a population, but such a scenario is inconsistent with the high MHC polymorphism observed in most study systems being coupled with strong signatures of positive selection.

Novel variant advantage may also explain the striking transspecies polymorphism observed at the MHC (33): If such polymorphisms are derived from hybridization (as opposed to being ancestral; ref. 34), novel variant advantage may accelerate introgression and promote interspecies sharing of polymorphism. Furthermore, novel allele advantage might also affect the evolution of MHC-based mating preferences, an important factor in shaping MHC diversity (35–37). Our results suggest that preferences for partners with MHC alleles that are novel in a population, rather than those that just differ from self MHCs, should be strongly favored.

Negative frequency-dependent dynamics [favoring both novel and (sufficiently) rare alleles] and heterozygote advantage are two important types of balancing selection maintaining MHC diversity (12). Although frequency dependence and heterozygote advantage are not mutually exclusive, their relative influences are notoriously hard to test independently (12) and may be impossible to separate by observational studies alone (38). Other processes that can contribute to MHC diversity further complicate the separation (2, 12, 39). However, the crosses in our experiment produced genotypes/supergenotypes that would not normally be found in natural populations at the time at which novelty enters/arises. In particular, we generated hosts that were “homozygous” with respect to MHC novelty (N/N) while controlling for genetic background. Hence, we were able to separate the effects of novelty from the effects of simply carrying more MHC variants, and this showed that the number of alleles or supertypes carried by an individual was not significantly associated with infection intensity (SI Appendix, Tables S2.1 a and b and S2.2 a and b). Furthermore, we did not detect an overdominance-type advantage for L/N fish (SI Appendix, Tables S2.1b and S2.2b). The simplicity of the guppy-Gyrodactylus system may explain the lack of heterozygote advantage, which previous work has shown to be particularly important in multipathogen systems (40). Importantly, however, our results show that, against a natural host–pathogen genetic background, novel MHC variants can be selectively advantageous by virtue of properties arising from their novelty/extreme rarity (17, 32), rather than by simply being present in heterozygotes.

Direct comparison of the best allele-based and supertype-based models of infection intensity indicated that allele-based groupings produced the better fit (SI Appendix, Tables S2.1a and S2.2a). However, fish with at least one novel supertype experienced infections that were significantly less severe than fish with no novel supertypes but at least one novel allele (i.e., a novel amino acid sequence variant within a shared supertype). This suggests that functional novelty may be more important than simple allelic novelty, whereby the unique binding properties of the novel supertypes are more likely to fill an immune response void (41, 42). The disparity between the statistical model and empirical observations on parasite loads of guppies with and without novel supertypes highlights that the relative fitness contributions of novel supertypes and novel alleles within supertypes remain to be determined. Despite this uncertainly, our experiment demonstrates a general advantage of novel MHC variants.

While the breeding design of our study controls for population-level linkage between MHC class II genes and other genes that may affect immune responses [28 chromosome pairs (43), plus recombination when F₁s reproduce], we cannot exclude possible effects from genes that may be in close physical linkage with the MHC without knockdown experiments or isogenic guppy lineages. Unlike for tetrapods, linkage with MHC class I can be ruled out for teleost fish (44)—a pertinent point because, although MHC class I usually targets intracellular pathogens, these genes have been coopted into roles more typical of class II in some teleosts (45). Concerning the exact mechanism by which the MHC may influence responses to skin ectoparasites, one possibility is through antigen-presenting skin cells (dendritic cells) mediating production of proinflammatory cytokines. This has been demonstrated in vitro with zebrafish skin tissue (46) and implied by gene expression studies on salmonids infected with sea lice (47–49). That supertypes are associated with lower infection intensities also suggests a functional rather than linkage-based influence of MHC.

Our results suggest that novel variants rather than locally adapted variants are associated with lower levels of parasite infection, and the lack of a significant interaction with replicate population suggests this finding may be generalizable. This contrasts with several studies on stickleback (Gasterosteus aculeatus) MHC–parasite interactions. When comparing or crossing lake and river stickleback populations, these studies have variously shown local adaptation (24), immigrant advantage (50), and a mixture of signals (51). This variation likely reflects the highly divergent ecologies and parasite faunas of lake and river sticklebacks, which may exert a complex suite of selection pressures (24, 50–54). Our study builds on the important insights from these studies by overcoming some of their limitations, such as the use of only a single population pair (24, 50); no experimental control of population-level linkage (ref. 50; although this study’s use of statistical control means it is uniquely able to demonstrate effects of such linkage); and the stock caveats of snapshot observational studies (ref. 51; e.g., uncertainty regarding where an allele may be in a frequency-dependent dynamic, and the difficulty of separating effects of rare alleles from heterozygosity; ref. 12). Also, these previous studies did not test for effects of supertypes. The most pertinent stickleback experiment to our result presents the simplest finding: MHC variants that confer resistance, irrespective of being rare or common, tend to increase in frequency (55). Many traits and circumstances may make a variant resistant; our study shows that novelty is likely to be one of them.

Based on our results, we predict that novel variants, including those coming from immigrants, have a reasonable probability of establishing and spreading. However, inferring the consequences of novel variant advantage from MHC-based population genetic structure is more challenging, especially in populations connected by gene flow (42). Furthermore, genuine novelty of a variant can only be ascertained by exhaustively sampling a population’s MHC diversity over an extended period of time. Introgression of novel alleles may be easier to observe if it occurs among well-diversified populations coming into secondary contact, including between species (e.g., ref. 56) or between allopatric populations (such as those we investigated). We suspect that the small number of alleles shared between Trinidad and Tobago (SI Appendix, Appendix S5) might come from fish introduced to Tobago by humans, as most of these alleles were found in a population close to human settlements and communication hubs. If so, our simulations suggest that these alleles are likely to spread across Tobago in the nearby future, subject to migration rate to other Tobagonian populations.

Our data constitute empirical demonstration of a long-posited, fundamental model of the evolution of MHC variability: Novel immune variants confer a selective advantage, consistent with Red Queen scenarios in which parasites adapt to local host immune genotypes (8, 15, 16). We show this effect using wild-sourced host and pathogen genetic variation, indicating that demonstrations of novel MHC advantage in congenic laboratory systems (18, 19) may be applicable in the context of natural MHC diversity. Furthermore, because we show this with replicate population crosses while controlling for population genetic background, our finding should be pertinent regardless of the source of novelty (e.g., mutation, recombination, introgression) or the rarity with which novelty enters a system. Looking beyond the MHC, although Red Queen dynamics have been shown in several nonvertebrate host–pathogen systems (57–59), examples of experimentally tested molecular mechanisms are rare, even for simple systems such as bacteria–phage interactions (60). In contrast, we explicitly link phenotype (infection intensity) to genotype (novel/local MHC) with an a priori hypothesis, using a gene family with a well-characterized immunological function. Overall, our results show that the advantage of immunogenetic novelty is a key selective agent underpinning Red Queen coevolution, and that, despite skepticism (1, 61), Red Queen coevolution may be an important force in shaping the immune genes of complex organisms.

All methods are described in more detail in SI Appendix, Appendix S1. At our field laboratory in Tobago, we conducted controlled gyrodactylid infection experiments on five replicate guppy populations. Each population was the F₂ descendant of a cross between a wild Trinidad guppy population and a wild Tobago guppy population (Fig. 1). Sampling locations and numbers of founding pairs per replicate are given in SI Appendix, Appendix S3. We reared the founding females for each replicate from wild-caught juveniles to ensure they were (i) virgins and (ii) free from gyrodactylids before being used for breeding (husbandry/screening details in SI Appendix, Appendix S1.1). The males were freshly caught wild adults. To ensure a 1:1 sex ratio at foundation, and to minimize the risk of parasite transfer from males to females, we used artificial insemination (62) to make the crosses. Between-island crosses allow a more powerful proof-of-principle test of the novel variant hypothesis than within-island crosses because they minimize the risk of crossed guppy populations having shared/exchanged MHC alleles or parasites in the recent past (between-island population structure is stronger than within-island structure; refs. 42 and 63; SI Appendix, Appendix S5). Between-island crosses also removed a hierarchical factor (“island”) that we would have struggled to achieve adequate replication to control. All males and females were fin-clipped for DNA (caudal fin, 2–4 mm²; preserved in 0.3 mL of 97% ethanol).

After insemination, we released the females to 800-L mesocosms (one per crossed population) and provided supplementary feeding (SI Appendix, Appendix S1.1). One month after observing the first F₁s in each mesocosm, we removed all surviving founding females and allowed the F₁s to mature and mate among themselves. One month after observing the first F₂s in each mesocosm, we removed and fin-clipped all F₁s. We did not attempt to control the mating of F₁s because of logistical constraints and because population-level replication coupled with the testing of a very general allelic property (novelty) should minimize potential family effects. However, as a precaution, we tested whether F_IS among F₂s differed significantly from zero (see below). At both removal stages, we verified that the fish were free from ectoparasites (64). We found no endoparasites in a subsample of four to eight fish per mesocosm that were dissected fresh after the experiment.

Three months after observing the first F₂s, we began controlled gyrodactylid (“gyros”) infection experiments on these parasite-naïve fish. For each experimental population, we caught fresh wild guppies from one founder population (SI Appendix, Table S3.2), screened these for gyros, and used fish with suitably large infections as parasite donors. Heavier donor infections are both easier to work with (gyros jump to new hosts more readily when the donor has a higher gyro load) and likely to be important sources of new infections in the wild. To minimize the handling of infected fish and the time for which gyros were held in captivity before making experimental infections, we did not attempt to identify gyros to species level before the experiments, despite the likelihood of multiple species being present (see molecular identification below). We did not captive-breed individual lineages because we wanted wild parasite genetic variation, free from artificial selection (65). Under weak anesthesia (MS-222), each recipient F₂ fish was infected with two gyros. Full details of the infection protocol are given in SI Appendix, Appendix S1.2. After being revived, each recipient fish was transferred to a separate 400-mL isolation container. After infection on day 0, we briefly anesthetized all infected fish and counted the number of gyros on day 1, and then every other day thereafter for 17 d (22, 23). Isolation containers were kept at ambient outside temperature in shade (husbandry details in SI Appendix, Appendix S1.2). Any fish found dead (checked every 3–12 h, depending on infection intensity) was promptly preserved whole in 1 mL of 97% ethanol, with the ethanol replaced after 6 h and again after 24 h. Fish that survived until day 17 were fin-clipped.

The work conducted in these experiments was approved by Cardiff University’s animal ethics committee and covered by UK Home Office License PPL 302876. All Tobago-sourced wild fish were collected with permission from the Tobago House of Assembly (permit 004/2014). No specific permits are required on Trinidad, but we collected only from areas where guppies were reasonably abundant.

From genomic DNA extracted from fin clips, we genotyped all P generation and F₂ fish at a 217-bp fragment of the MHC class II that codes for the highly polymorphic β-chain of the MHC molecule’s antigen binding groove (66). All genotypes used in downstream analyses had ≥300 allelic reads (median = 1,042 reads). Full details of primers, PCR conditions, and genotyping bioinformatics (67, 68) parameters are given in SI Appendix, Appendix S1.3.

We used custom Python scripts to assign MHC alleles in each replicate as coming from the maternal or paternal founding populations. Assignment was unambiguous—there was only one linkage block (one to two allelic variants per haplotype), all variants within population crosses differed by at least one nonsynonymous mutation, and we observed no allele sharing between any crossed population pair (SI Appendix, Appendix S5). We then designated alleles as either “L” (local) when they belonged to the same host population as the worms used for that replicate, or “N” (novel) when detected only in the “other” founding population, and allocated F₂s to three genotype groups based on these designations: N/N (two novel haplotypes); L/L (two local haplotypes); and L/N (mixed/“heterozygous”). Although a nominally novel allele could be present in the local population at a frequency too low for our sample to detect, population genetic analyses on a larger dataset suggested that this is likely to be extremely rare (e.g., only 8/214 alleles in our dataset were present on both islands; SI Appendix, Appendix S5). Because of the breeding design, all three genotype groups should have the same average genetic background with respect to population hybridization and heterozygosity [23 chromosome pairs (43), plus recombination when F₁s reproduce].

Amino acid (AA) substitutions vary in their functional consequences for an MHC molecule’s antigen binding profile, such that alleles with different AA sequences may be functionally similar. These MHC supertypes are predicted to bind similar antigenic “supermotifs” and may better characterize the breadth of host defense than alleles (25–27). Using 15 guppy MHC codons previously identified as being under positive selection (69), five physicochemical descriptors of each AA (70), and discriminant analysis of principal components (71, 72), we reduced the list of allele sequences to 14 supertype clusters (full description in SI Appendix, Appendix S1.3). Supertype designations were used to assign fish into L/L, N/N, and L/N “supergenotypes” analogous to allele-based groupings, except that supertypes shared between a pair of crossed populations (all pairs shared at least one supertype; SI Appendix, Appendix S5) were treated as local. Supergenotypes thus do not describe the haplotype makeup of individuals—multiallelic novel haplotypes may contain one or more shared supertypes. Only nine individuals, over three replicate populations, had a number of supertypes lower than their number of alleles (i.e., they carried 2+ alleles of the same supertype). Of these, only one changed novel/local categorization between allelic and supertype-based analyses (from L/N to L/L). Novel and local alleles did not differ systematically in the supertype physicochemical parameter space, and we thus concluded that the only “special” property of novel variants was their novelty (details in SI Appendix, Appendix S6).

We used mitochondrial barcoding to identify two to four gyros per replicate to species level (full details in SI Appendix, Appendix S1.3). All sequences showed their strongest matches (98–100% identity) against published G. turnbulli (Gt) sequences, except for those representing 35 fish from the AV/SS population, all infected on the same day, which matched G. bullatarudis (Gb; 98–100%). These fish also showed markedly different pathology, consistent with Gb (SI Appendix, Appendix S1.3). Given the taxonomical and pathological differences between Gt and Gb and the absence of indications of Gb elsewhere in the experiment, we excluded these fish from our analyses.

The analyses described in the next two paragraphs were performed separately for alleles and supertypes. We first tested if MHC group (L/L, L/N, N/N) predicted whether individual fish survived the experiment. We used corrected Akaike information criterion (AIC_C) ranking (73) of logistic regressions to explore all combinations of the following main effects: MHC group; number of MHC variants (alleles/supertypes; continuous, 1–4); standard length (continuous, z-transformed); age/sex (one variable: “male” if in possession of a fully shaped gonopodium, i.e., “hook and hood” visible; “female” if length >13.0 mm and no gonopodium evident; “juvenile” for all others); temperature (mean daily maximum over monitored period, z-transformed); and experimental population. We also included interactions between experimental population and both MHC group and number of MHC variants. We then examined models comprising the top two units of ranked AIC_C. For clarity of presentation in the main text, we only report the ΔAIC_C of the highest ranked model to include MHC group and the P values (two-tailed) for certain contrasts, but in SI Appendix, Appendix S2, we give a fully nuanced account of the analysis.

We used a similar process to test whether MHC group predicted infection burden among fish that survived the experiment. We used worm days as the response variable, calculated as the total area under a fish’s 17-d infection trajectory line. Worm days are tractable to analyze (no zero-inflation or random effects) and provide an ecologically relevant summary metric of an infection trajectory: Fish that experience more worm days can reasonably be considered to have endured a greater parasite burden, and, in the wild, be more vulnerable to decreased condition and associated consequences (28–31). For this analysis, we used generalized linear models with negative binomial errors (log link function), and the same AIC_C approach and set of predictors used for analyzing probability of death. As a follow-up test of the relative importance of AA sequence novelty versus putative functional novelty, we used model ranking to test whether worm days differed significantly between fish carrying at least one novel supertype and fish carrying no novel supertypes but at least one novel AA variant (the latter are novel alleles that belong to shared supertypes; SI Appendix, Tables S2.3 a and b and Fig. S2.1).

Twenty-eight fish from replicate population DR/SC were missing gyro counts for either day 7 or day 9 because of a 36-h power cut that rendered microscope work impossible. We dealt with this in the main analysis by dropping these same days from the worm-day calculation for all fish, rather than excluding the affected fish. However, N/N fish still experienced significantly fewer worm days than L/L fish when we used the full area data restricted to only complete cases (SI Appendix, Appendix S7).

F_IS was significantly different from zero (0.216, SE = 0.060, P = 0.001) in HC/Guan, suggesting family effects may be distorting haplotype frequencies in this replicate population. Repeating the main analyses without this replicate did not result in a different interpretation (SI Appendix, Appendix S8).

To visualize average infection trajectories for each MHC genotype/supergenotype, we analyzed the number of gyros per fish for each infection day separately, using AIC_C to find the highest ranked model to include MHC group as a predictor. From these models, we generated predicted daily worm loads for each MHC group that we then used (i) to plot Fig. 2; (ii) to estimate the percentage reduction in number of worm days of N/N fish relative to L/L fish that was minimally affected by the incomplete cases; and (iii) as a post hoc exploration of which MHC groups had the lowest/highest loads on which days, and which fish were the most likely to clear their infections (full details in SI Appendix, Appendix S9).

We thank Tobago House of Assembly for permission to conduct this project; staff of the Environmental Research Institute Charlotteville, Tobago, for support in the field; P. Turpin for renting us the field station; A. Pilastro for training in guppy artificial insemination; members of the Cardiff University parasite research group who helped with parasite work; J. Lighten for sharing an as-then unpublished manuscript; M. Migalska for Fig. 1; and M. Konczal, M. Migalska, D. S. Richardson, J. Kaufmann, L. G. Spurgin, W. Babik, C. A. Morrison, J. Lighten, W. Smallbone, A. R. Ellison, M. Hammers, M. J. G. Gage, and three anonymous reviewers for comments on earlier manuscript versions. This project was funded by Polish National Science Centre Harmony Grant UMO-2013/10/M/NZ8/00253 (to J. Radwan, C.v.O., and J.C.).