Evolutionary gain and loss of a pathological immune response to parasitism

Abstract

Parasites impose fitness costs on their hosts. Biologists often assume that natural selection favors infection-resistant hosts. Yet, when the immune response itself is costly, theory suggests selection may sometimes favor loss of resistance, which may result in alternative stable states where some populations are resistant and others are tolerant. Intraspecific variation in immune costs are rarely surveyed in a manner that tests evolutionary patterns, and there are few examples of adaptive loss of resistance. Here, we show that when marine threespine stickleback colonized freshwater lakes they gained resistance to the freshwater-associated cestode, Schistocephalus solidus. Extensive peritoneal fibrosis and inflammation is a commonly observed phenotype that contributes to suppression of cestode growth and viability, but also impose a substantial cost of reduced fecundity. Combining genetic mapping and population genomics, we find that opposing selection generates immune system differences between tolerant and resistant populations, consistent with divergent optimization.

One Sentence Summary:

Fish evolve fibrosis to suppress parasite growth, but costly decreases in fecundity then select for infection tolerance via fibrosis suppression.

Parasites impose strong selection on their hosts, driving rapid immune evolution (1, 2). However, selection may not always favor the evolution of ever-greater resistance. Resistance-based immunity can have costly side-effects that reduce host fitness by consuming limited resources or inducing auto-immune pathology, both of which influence survival and reproduction (3, 4). For example, adult Soay sheep with higher self-reactive antibody titers had lower reproductive rates (5), though immune costs have not been found in other systems, perhaps because costs are condition dependent (6). Evolutionary optimization theory suggests that the marginal benefit of evolving increased parasite resistance must be balanced against the marginal costs imposed by the immune response (7). As a result, populations may evolve towards an intermediate optimum with partial resistance and some autoimmune costs. Or, populations may evolve tolerance (8), allowing infection to persist while mitigating mortality and fecundity costs of that infection (including direct effects of the parasite, or immune costs). Tolerance itself may be costly, either requiring energy for damage repair, or by permitting infection. Either resistance or tolerance should fix in a population when these strategies are controlled by separate mechanisms with linear and additive costs; mixed strategies can be maintained under different cost functions (9). Here, we show that a particular immune response against parasitic cestodes entails benefits and costs and has been repeatedly gained and lost in a small fish. These results provide a stark example of conflicting selection pressures acting on immune variation in the wild, leading to among-population divergence in immune or tolerance strategies.

Marine populations of threespine stickleback (Gasterosteus aculeatus) rarely encounter the freshwater cestode Schistocephalus solidus (Fig. 1A), and so have not evolved resistance (10). Following post-Pleistocene glacial retreat, marine stickleback established populations in freshwater lakes where they repeatedly evolved greater resistance (10). However, infection prevalence varies among lakes (Fig. 1A). Ecological factors (i.e. fish size, diet, and lake elevation and area) partly explain infection variation (Fig. S1–S2), but even similar lakes differ in S.solidus prevalence. For instance, stickleback from Roberts and Gosling Lakes (R and G hereafter) on Vancouver Island exhibit infection rates of 0% and >50% respectively, despite close geographic proximity, similar consumption of cyclopoid copepods (the first host) (11), and nesting pairs of loons in both lakes (the final host). We hypothesized these differences in infection prevalence reflected evolution of resistance by R stickleback, or tolerance of G fish.

Fig. 1. Population differences in infection success and immune phenotypes.

A) S.solidus prevalence varies between stickleback in different aquatic habitats on Vancouver Island, British Columbia (binomial GLM deviance=148.8 df=2, P<0.0001, 2009 sample). Prevalence also varies significantly among lakes (𝞆²=4629, df=49, P<0.0001; estimate and ± SEM). Three focal populations are labeled: high-infection Gosling Lake (G), infection-free Roberts Lake (R) and a low-infection susceptible marine population from Sayward Estuary (S) representing the ancestral state. B. Cestode mass is greatest in lab-infected G stickleback, least in R, and intermediate in F1, F2, and reciprocal backcross hybrids. In a general linear model there was a significant effect of R ancestry proportion t=−11.6, P<0.0001; controlling for covariates including sex t=−2.5, P=0.013, log fish mass t=3.9, P=0.001, and coinfection intensity t=−1.0, P=0.308. C. Reactive oxygen production by granulocytes (median fluorescent intensity, MFI, of gated granulocytes) differs between lab-raised G and R stickleback, and their F1 and F2 hybrids are intermediate (sample sizes listed for each cross time; MFI depends on cross type (F=43.14, P<0.0001), and log mass (F=27.06, P<0.0001) but not fish sex (F=0.815, P=0.4428) nor cestode presence (F=2.06, P=0.1518). A weak cross*Infection interaction (F=2.47, P=0.0308) exists because RBC backcross fish decrease ROS significantly when infected, whereas all other crosses exhibit no response. D. Image of infected fish with dissected cestodes. E. Peritoneal fibrosis is more common in wild-caught R than G fish, and more common in lab-reared cestode-exposed RBC and F2 hybrids than GBC hybrids (Cross ²=155, P<0.0001, sex 𝞆²=2.53 P=0.281, log mass 𝞆²=23.83 P<0.0001, cestode infection 𝞆²=98.48 P<0.0001, cross*infection 𝞆²=18.45 P=0.002; points are means ± SEM). F. Masson’s Trichrome stain section of stickleback visceral organs from Alum injected and PBS control injected fish. Arrow indicates fibrosis (blue-stained collagen) connecting the intestinal wall to the spleen. Black scale bar in the top left corner equals 500μm.

We experimentally exposed lab-raised R×G recombinant hybrids, including intercross (F2) and backcross hybrid stickleback to S.solidus (Fig. S3;(12)). Infection rates were lower in hybrids with greater R ancestry (Fig. S4), consistent with prior trends (10, 11). R ancestry also reduced the growth rate of cestodes that established infections (Fig. 1B). Because all parasites grew for the same period of time (12), growth rate is proportional to ln(cestode mass). When the cestode reaches reproductive size (50mg), it manipulates fish to facilitate predation by birds (terminal hosts) (13). Cestode growth suppression in R-backcross (RBC) and many F2 stickleback prevented the parasite from reaching this size threshold, protecting the fish from increased predation risk.

Infection increased head kidney granulocyte abundance in both fish genotypes (Fig. S5). However, R ancestry increased the inflammatory Reactive Oxygen Species (ROS) generated per granulocytes (Fig. 1C). R genotypes also exhibited a visually striking response to infection involving severe peritoneal fibrosis (14) (Fig. 1D). In vertebrates, tissue damage from parasites can cause inflammation, fibroblast proliferation, and collagen deposition (fibrosis) forming adhesions between the viscera. In healthy stickleback the abdominal organs move freely, but cestode exposure induces extensive adhesion in some stickleback. Similar adhesions can be induced by intraperitoneal injection of aluminum phosphate (alum) or cestode protein (15), and have been observed in a wide variety of other fish species (16). Trichrome stain confirms the adhesions entail excessive collagen deposition (Fig. 1E, Fig. S6). Following cestode exposure of F2 hybrid fish, fibrosis was positively associated with R ancestry (Figs. 1D), especially when a cestode was present (Fig. S7). Fibrosis is rare in G fish (in the wild and in the lab), and in G-backcross (GBC), regardless of infection. Marine stickleback (a proxy for the ancestral state) also lack fibrosis in the wild and after lab infections. Fibrosis is common in independently colonized watersheds in both Alaska and Vancouver Island, confirming that fibrosis evolved repeatedly during freshwater colonization (Fig. S8).

Fibrosis confers two benefits to lab fish. First, fibrotic F2 hybrids had 87.7% smaller cestodes than non-fibrotic F2 (Fig. 2A, mean mass 0.316g and 2.558g respectively). Considering all infected hybrids, cestode mass depended on fibrosis (P=0.0072) and cross direction (P<0.0001), but does not depend on a fibrosis*cross interaction. ROS production was also negatively correlated with cestode growth within F2 hybrids (Fig. 2B). Second, fibrosis increased the likelihood that small cestodes would be encased in cysts (Fig. S9; RBC odds ratio(OR)=6.6, P<0.0001; F2 fish OR=13.2, P<0.0001; GBC fish OR=52, P=0.0359), as collagen is a key component in cyst formation. These cysts often contained dead or visibly degraded parasites, or amorphous material with cestode RNA. Cysts were more common in R ancestry fish (Fig. 2C). Thus, fibrosis is associated with both growth suppression and elimination of cestodes. Importantly, fibrosis is also a persistent lesion that remains long after the immune challenge (15), which is why we observe fibrosis in cestode-exposed but uninfected fish. The negative effect of R ancestry on cestode growth arises via multiple direct and indirect paths explaining 34.3% of cestode size variation (Fig. 2D): R ancestry significantly promotes fibrosis and ROS,both these responses independently contribute to cestode growth suppression, and R ancestry explains additional variation in cestode size.

Fig. 2. Benefits and costs of fibrosis and ROS in lab (A-D,G) and wild (E,F,H) stickleback.

A. Cestode mass is lower in fish with greater R ancestry (t=−11.6 P<0.0001; sex t=−2.505 P=0.0127; log mass t=3.86 P=0.0014; number of coinfecting cestodes t=−1.02 P=0.3082), and when fibrosis is present (t=−2.909, P=0.0039), with no cross*fibrosis interaction (F_2,255=1.25, P=0.2873). Cross types are color coded green (R backcross), turquoise (F2 intercross) and blue (G backcross). B. Stickleback with greater ROS production have smaller cestode mass (F_1,236= 5.25, P=0.0228) controlling for fibrosis (F_1,236=13.18, P<0.001), cross (F_2,236=37.9, P<0.001), with no ROS*cross interaction (P>0.05). C. Stickleback with fibrosis are more likely to have cysts encasing tapeworms (pictured in Fig. S9), particularly in F2 and RBC fish (R ancestry effect χ2=83.9, P<0.0001). D. Path analysis to explain variation in cestode mass: dashed lines denote negative correlations, solid lines are positive. Thick lines are significant, thin lines are non-significant (* P<0.05, ** P<0.01, *** P<0.001). E. Wild-caught individuals with fibrosis tend to have lower cestode mass than fibrosis-free individuals. Each trendline is a population (sampled in 2016), solid and dashed lines represent significant and non-significant within-lake trends. The main effect of fibrosis is significant in a model with random population effects (χ2= 20.9, df=1, P<0.0001, controlling for host mass χ2=0.075, df=1, P=0.7833). F. Confirming lab results, wild stickleback populations with more fibrosis tend to have smaller cestodes (t=−1.91, one-tailed P=0.048, controlling for host mass and infection intensity, 2013 sample). Circle size represents infection prevalence (t=−1.91, controlling for crowding infection intensity). G. Fibrosis, but not cestode infection, was associated with reduced ovary development in lab-raised fish (20.4% versus 5.4% gravid in non-fibrotic versus fibrotic females respectively, binomial GLM deviance=13.53, P=0.0002). H. Fibrosis was associated with low female fecundity in wild fish from Roselle Lake in 2018: 70% of non-fibrotic females were gravid compared to 6.6% with fibrosis (z=−2.7, P=0.0057), with no effect of cestode infection. Error bars are ± 1 SEM.

These laboratory results are corroborated by data from wild stickleback populations. In a 2016 sample of 13 lakes in British Columbia, fibrosis prevalence varies significantly among lakes (χ²=90.8, P<0.0001). Fibrosis is more likely in fish with cestode infections (Fig. S10), but unrelated to the only other common helminth in the peritoneum (Eustrongylides sp., χ²=0.98, P=0.197). Fibrosis also drove a 56.6% reduction in cestode size (Fig. 2E), controlling for host mass and lake. A 2013 survey of 16 other lakes found population fibrosis frequency was positively related to cestode prevalence (r=0.580) and negatively related to mean cestode mass (Fig. 2F).

Both fibrosis and ROS are costly responses to infection. In our experimentally exposed hybrid lab fish, fibrotic females were 73.4% less likely to be gravid at euthanaisia (Fig. 2G; binomial GLM P=0.0002). Reproductive readiness also was negatively related to ROS (Fig. S11), increased with fish mass, and was unrelated to cross or infection. These results indicate that cestode infection undermines female fitness indirectly via fibrosis, rather than directly. Field samples corroborate this inference: in 2018, Roselle Lake infected fish are more likely to exhibit fibrosis. Female reproductive maturity is reduced 90.5% by fibrosis (Fig. 2H), but was not significantly affected by infection itself. Similarly, male stickleback with fibrosis are less likely to nest (Odds ratio=0.41 in Roselle Lake, OR=0.56 in Boot Lake, both P<0.01), controlling for infection status and body size (17).

Given these benefits and costs of fibrosis, we hypothesized that natural selection may be optimizing fibrosis, favoring gain (or loss) depending on a population’s infection risk or ecological factors affecting costs. Optimization should be reflected in the genetic architecture of the relevant traits: rather than systematic directional selection for ever-stronger fibrosis, stabilizing selection would act on a combination of pro- and anti-fibrotic genes (positive and negative effects). To test these optimization expectations, we used a triangulation approach (merging QTL mapping, population genomics, and transcriptomics). We mapped the genetic basis of variation in fibrosis (and related traits), determined QTL effect directions, and determined which populations experienced selection for gain- or loss-of-function

To map infection-related traits, we genotyped 647 of the experimentally exposed F2, GBC, and RBC fish, for 234 informative markers. Mapping quantitative trait loci (QTLs) for all traits of interest (Fig. S3) revealed 7 significant QTLs explaining variation in cestode mass, fibrosis, granulomas, and ROS, but not cestode presence/absence (Fig. 3A; Table 1, Table S1, Figs. S12–13). Fibrosis maps to a QTL on chromosome 2, where R alleles increase fibrosis frequency (Fig. 3B). Each QTL explains a small to moderate amount of phenotypic variance in resistance traits. For example, a QTL on Chr12 explains 9.3% of cestode mass variation (Table 1). Several traits map to overlapping QTL (e.g., ROS and cestode mass map to Chr15, Fig. 3A, Table 1), representing potential pleiotropy. An optimization process is expected to yield a mixture of positive and negative effect QTLs (18) within a given population. We observe such a mixture of effect directions for some traits (ROS and cestode mass). Although R granulocytes generally produce more ROS, QTLs are not uniformly in this direction: within the ROS QTL on Chr15 R alleles confer higher ROS, but at the QTL on Chr11 R alleles reduce ROS (Fig. S14).

Fig. 3. Genetic mapping and population genomic analyses of immune differences between stickleback populations.

A. Quantitative Trait Locus (QTL) mapping identified chromosomal regions associated with R versus G differences in cestode growth, fibrosis, granuloma (without fibrosis) and ROS; here we present LOD scores for focal QTL. Tick marks on the x axis represent population-informative SNPs for mapping. B. Effect plot showing association between Chr2 marker X52 genotype and the frequency of fibrosis, for the three hybrid cross types (blue RBC, bluegreen F2 intercross, green GBC), with means and standard errors. Jitter is added to distinguish overlapping points. C. Within the Chr2 QTL for fibrosis, PoolSeq data indicates the strongest target of divergent selection between R and G is in and adjacent to the 3’ end of PU1 gene (top subpanel is F_ST between R and G, bottom subpanel is the population branch statistic for G and R showing accelerated evolution is in Gosling Lake, including the fixation of a deletion within the intron (nucleotides 8843941–8844018) containing a regulatory CTCF binding motif. D. The gene SPI1 (ENSGACT00000020522.1) produces PU1 and is more highly expressed in fibrotic than non-fibrotic fish controlling for infection and cross (lfc=0.199; padj=0.0324) and is more strongly expressed in infected fish (lfc=0.254, padj=0.0036). E. Effect plot of Chr12 marker X308 genotype on log live cestode mass. F. Within the Chr12 QTL for cestode mass, the strongest genomic targets of selection (i.e., highest population branch statistic) are tightly clustered around the genes STAT6 and Cyp3a48. G. STAT6 contains nearly fixed deletions in Gosling Lake just 3’ to the start of the gene, and within exon 5. H. Cyp3a48 contains a 3kb deletion within exon 2 that is fixed in G fish, as indicated by the large window of reduced coverage in G but not R or Sayward poolseq data. In Panels B,D, and E error bars are ± 1 SEM.

Table 1.

The locations of quantitative trait loci (QTL) contributing to variation in cestode mass and stickleback immune traits.

Chromosome	Marker ID	Locus	Genomic Position	Trait
4	X86	2.3	1406842	Cestode Mass
12	X308	17.8	14316110	Cestode Mass
15	X359	25.4	1930615	Cestode Mass
2	X52	7.8	6190955	Fibrosis
3	X78	34	11999766	Granuloma
11	X275	46.3	14553738	ROS
15	X362	47.7	4462742	ROS

QTL mapping identifies large chromosomal regions associated with a phenotype, containing large numbers of genes. Genomic signatures of natural selection are typically much narrower, allowing a more precise genetic map. We obtained whole-genome sequencing of pooled population samples (PoolSeq) for R, G, and a marine population (Sayward, or S), and calculated population branch statistics (PBS) to detect targets of selection. We focus on genes that (1) lie within a trait QTL, (2) exhibit strong evidence of selection, (3) have known effect on the QTL trait, and (4) show some feature of molecular evolution, such as changes in coding or known regulatory regions, or or differentially expressed between R and G genotypes (19, 20). Only a few loci met three or four of these criteria. PBS analysis, using marine fish as the outgroup, allowed us to determine which population experienced selection on a given gene. Several of the loci we identified experienced strong positive selection in the susceptible/tolerant G population, others evolved in the resistant R population.

Within the fibrosis QTL (Chr2, Fig. 3B, S15), the strongest target of selection is a narrow area containing the gene SPI1b (Figs. 3C, S16, F_ST~1.0). This gene produces a transcription factor, PU.1, previously shown to regulate fibroblast polarization and tissue fibrosis (21, 22). PoolSeq data reveal a 78-base pair (bp) deletion in the second intron that is nearly fixed in G but absent in R or S, (Fig. 3C). The Ensembl human genome browser identifies a CTCF binding site in this region, but sequence identity is poor. SPI1b is expressed more in fibrotic compared to non-fibrotic F2 hybrid fish (Fig. 3D), and in infected fish. Additional genes within the QTL are differentially expressed between genotypes, or as a function of fibrosis (Fig. S17). But, SPI1b is the only gene in the QTL to meet all four criteria, providing a strong candidate gene to explain the fibrotic lesions of R fish. Remarkably, this gene exhibits a large allele frequency change in Gosling rather than Roberts Lake (Fig. 3C, larger population branch statistic), implying that selection favored evolutionary loss of fibrosis in G fish. This result is corroborated by PoolSeq data from 12 high- and 13 low-fibrosis populations showing that SPI1b evolution accelerated in about half the low-fibrosis populations (Fig. S18).

The strongest QTL for cestode mass is on Chr12 (Fig. 3E) and contains two strong targets of selection. One locus shows rapid evolution in G, and contains STAT6, which regulates inflammation and activation of alternatively activated macrophages (23), and helminth resistance in lab mice (24). G is nearly fixed for a small frame-shift deletion in exon 5 of STAT6 (Fig. 3G) that should render mRNA isoforms with this exon non-functional. G fish also evolved a 100bp deletion 80bp upstream of the STAT6 start codon. The second selected locus within the QTL is driven by evolution in R fish (Figs. 3F, S19) and contains a cytochrome P450 subfamily 3a gene–Cyp3a48. Proteins in this subfamily metabolize a large variety of endogenous and exogenous products, with inflammatory ROS being major enzymatic byproducts of these reactions (25). G fish are nearly fixed for a 3kb coding-sequence deletion in Cyp3a48 (Fig. 3H); deletion of this gene is consistent with G fish’s weaker fibrosis response. Intriguingly, Cyp3a48 is also among the strongest targets of selection between each of two benthic and limnetic stickleback species pairs (26), which also differ in S.solidus prevalence. A third site within the QTL is a weaker target of selection (Fig. S19E) but is noteworthy for containing hnf4α, a fibrosis suppressor (27). Ingenuity Pathway Analysis of transcriptome data identified hnf4α as an upstream regulatory switch responsible for differential expression of a set of genes (20) that are induced by infection (z-score=0.481, padj=0.032) in G but not R fish (z-score=−0.8, padj<0.0001). Infection-induced expression of this fibrosis suppressor in G fish may contribute to their lack of fibrosis. But it is R which experienced selection on this locus (high PBS), leading us to conclude that R fish evolved fibrosis in part by reducing this down-regulatory pathway. Selected regions within other QTL are presented in Figure S20.

In evolutionary ecology it is commonly assumed that immunity is preferable to susceptibility (e.g., (28)), and hence population differences in parasite resistance reflect evolutionary gain-of-function in the more resistant populations (but see (29)). Experiments with birds have confirmed that greater resistance can evolve despite short-term immune costs (30). But, cost-benefit trade-offs mean selection may favor either loss or gain of resistance, as we illustrate here. Although marine stickleback have the ancient genetic pathway for peritoneal fibrosis (31), this response is not activated following cestode exposure, leading to large, virulent parasites. Replicated evolutionary gain of a strong fibrosis response to S.solidus (Fig. S8) across numerous freshwater stickleback populations indicates that fibrosis is an adaptive trait. The adaptive benefits of fibrosis include suppressed cestode growth, and increased cyst formation and cestode death. These benefits are noteworthy because fibrosis in humans is seen mostly as a pathology (32). In stickleback, fibrosis has pathological aspects as well, reducing reproductive success in both sexes. Given this cost-benefit balance, optimization theory leads us to expect stabilizing selection on immune traits leading to both heritable gains and losses of function. Our genetic analyses confirm these expectations. QTL mapping revealed pleiotropic and polygenic evolution of immune traits, with a mix of positive- and negative-effect QTLs (e.g., ROS; Fig. S14), consistent with theoretical expectations for stabilizing selection (18) rather than relentless directional selection. Using poolSeq to polarize the direction of evolutionary change within these QTL, we found strong selection in G favoring rapid fixation of deletions within three pro-fibrotic genes: SPI1, STAT6, and Cyp3a48. We did not detect these deletions in either R or marine fish. Meanwhile, R fish exhibit selection around pro-fibrotic TMEM39A, and anti-fibrotic hnf4α whose expression is unresponsive to infection in R fish (but responsive in G fish). Thus, although Roberts Lake exhibits a phenotypic gain-of-function relative to both G and ancestral marine fish, selection actually acted in both lakes in opposing directions (summarized in Figure S21).

Phylogenetic ancestral trait reconstruction also suggests that fibrosis is gained during the repeated colonization of freshwater, and then repeatedly lost in a subset of lakes (Fig. S8). As a result, cestode resistance and fibrosis severity differ among populations, which then contributes to the dramatic variation in parasite prevalence among populations. These differences are consistent with theoretical models (8,9), which suggest that host-parasite coevolution can lead to alternative stable states in which some populations evolve resistance (despite costs of immunity) whereas other populations evolve tolerance (despite costs of infection) (8). Such alternative states can allow parasites to persist in host metapopulations, which may facilitate cases of zoonotic disease spillover to humans (32). But, infection prevalence (and any zoonosis risk) will be unevenly distributed among populations depending on where they evolve along the resistance/tolerance axis. Thus, to understand parasite epidemiology we must consider how the mix of immune costs and benefits may lead to evolution towards, or away from, resistance to infection.

Data and materials availability:

Data and analytical code are publicly archived on Dryad/Zenodo (https://doi.org/10.5061/dryad.d51c5b060), sequence data are available at the Short Read Archive (submission SUB11518774).