Contemporary evolution of the innate immune receptor gene TLR3 in an isolated vertebrate population

Understanding where genetic variation exists, and how it influences fitness within populations is important from an evolutionary and conservation perspective. Signatures of past selection suggest that pathogen‐mediated balancing selection is a key driver of immunogenetic variation, but studies tracking contemporary evolution are needed to help resolve the evolutionary forces and mechanism at play. Previous work in a bottlenecked population of Seychelles warblers (Acrocephalus sechellensis) show that functional variation has been maintained at the viral‐sensing Toll‐like receptor 3 (TLR3) gene, including one nonsynonymous SNP, resulting in two alleles. Here, we characterise evolution at this TLR3 locus over a 25‐year period within the original remnant population of the Seychelles warbler, and in four other derived, populations. Results show a significant and consistent temporal decline in the frequency of the TLR3C allele in the original population, and that similar declines in the TLR3C allele frequency occurred in all the derived populations. Individuals (of both sexes) with the TLR3CC genotype had lower survival, and males ‐ but not females ‐ that carry the TLR3C allele had significantly lower lifetime reproductive success than those with only the TLR3A allele. These results indicate that positive selection on the TLR3A allele, caused by an as yet unknown agent, is driving TLR3 evolution in the Seychelles warbler. No evidence of heterozygote advantage was detected. However, whether the positive selection observed is part of a longer‐term pattern of balancing selection (through fluctuating selection or rare‐allele advantage) cannot be resolved without tracking the TLR3C allele over an extended time period.

populations (Haldane, 1992). Consequently, immunogenetic locii.e., those involved in the detection and combating of pathogens -are excellent candidates in which to investigate the evolutionary forces underlying the maintenance of genetic variation (Croze et al., 2016;Sommer, 2005). Indeed, pathogen-mediated selection is thought to be a key driver of balancing selection (Spurgin & Richardson, 2010). Three non-mutually exclusive mechanisms driving pathogen-mediated selection have been proposed: heterozygote advantage (Doherty & Zinkernagel, 1975), rare allele advantage (Slade & McCallum, 1992), and fluctuating selection (Hill et al., 1991). These three mechanisms -along with other forces such as sexual selection -can act independently, in concert, or in trade-off with one other (Apanius et al., 1997;Ejsmond et al., 2014;Spurgin & Richardson, 2010).
Immunogenetic research on wild populations has focused mainly on receptor genes of the acquired immune system: in particular on the exceptionally polymorphic major histocompatibility complex (MHC) (reviewed in Piertney & Oliver, 2005). However, high levels of diversity (Hedrick, 1994), gene duplication (Bollmer et al., 2010), conversion, recombination (Miller and Lambert, 2004a), and epistasis (van Oosterhout, 2009) makes it hard to tease apart the evolutionary forces driving MHC variation (Spurgin & Richardson, 2010). In contrast, the genes involved in the innate immune response, while still often polymorphic, exhibit relatively lower complexity. Furthermore, the innate immune system is the host's first line of response to pathogens, enabling a broad defence against an assortment of organisms (Aderem & Ulevitch, 2000). Consequently, innate immune genes can be more tractable candidates with which to study the evolutionary forces shaping immunogenetic variation in wild populations (Acevedo-Whitehouse & Cunningham, 2006).
Toll-Like Receptor (TLR) genes encode receptor molecules which bind to pathogen-associated molecular patterns -evolutionary conserved molecular motifs that are integral to the pathogen's survival (Medzhitov, 2001). Once bound, the TLR molecule triggers a cascade of processes associated with the innate and adaptive immune responses (Akira et al., 2006). Vertebrate TLRs can be divided into six families, depending on the pathogen-associated molecular patterns they detect (Roach et al., 2005). For example, TLR3 binds to viral dsRNA (Barton, 2007), while TLR5 binds to bacterial flagellin (Brownlie & Allan, 2011). While the majority of the TLR structure is structurally conserved (Roach et al., 2005), there is variation in the leucine-rich repeat domain of TLR genes, resulting in functional variation at the binding site. Such TLR polymorphisms have been associated with resistance (Antonides et al., 2019), or susceptibility to specific pathogens (Kloch et al., 2018), or associated with increased survival (Bateson et al., 2016;Grueber et al., 2013). TLRs can evolve rapidly as a result of pathogen-mediated selection (Downing et al., 2010) and evidence of balancing selection at TLR genes has been reported for various taxa (e.g., Areal et al., 2011;Velová et al., 2018). Nevertheless, most of these studies only inferred past selection from sequence variation and could not determine if selection was still acting, or determine the specific mechanisms involved. Moreover, in various bottlenecked populations, genetic drift may override selection as the dominant evolutionary force shaping TLR variation (Gonzalez-Quevedo et al., 2015;Grueber et al., 2013).
Here, we investigate the contemporary evolution of TLR variation in a natural population of Seychelles warblers (Acrocephalus sechellensis). The last remaining population of this species on Cousin island underwent a bottleneck in the 1900s resulting in decreased genome-wide genetic variation . Extensive longitudinal monitoring and a lack of dispersal (Komdeur et al., 2004) means that virtually all individual warblers on Cousin Island are sampled, marked and tracked throughout their entire lives (Hammers et al., 2015;Komdeur, 1992). This allows for accurate measures of survival and reproductive success (Hammers et al., 2019). As part of a conservation programme, individuals have been translocated from Cousin to establish populations on four additional islands (Komdeur, 1994;Richardson et al., 2006;, allowing spatial TLR variation to be investigated. A previous study found that five of seven TLR loci examined in the contemporary population (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008) of Seychelles warbler on Cousin Island were polymorphic and detected a signature of past positive selection at two loci, one of these being TLR3 -a viral sensing TLR (Gilroy et al., 2017). One of the three SNPs at this TLR3 loci was singled out for investigation because it is nonsynonymous, found within the functionally important leucine-rich repeat domain region, and had a relatively high minor allele frequency (32%, n = 28). However, if and how balancing selection maintains variation at this locus has yet to be investigated.
We first assessed how the frequency of this TLR3 SNP has changed over 25-years in the Seychelles warbler on Cousin Island. We then tested the role of selection in shaping TLR3 variation in this population; specifically, if survival and reproductive success are associated with individual TLR3 genotypes. Lastly, we compared patterns of TLR3 evolution over time in, and between, the Cousin population and the newly established (translocated) populations. These analyses allow us to better understand which evolutionary forces shape immunogenetic variation in small populations of conservation concern.
2 | MATERIAL S AND ME THODS

| Study species and system
The Seychelles warbler is a small (ca 15 g) insectivorous passerine endemic to the Seychelles. The species was distributed across the archipelago prior to human colonisation , but underwent a severe population reduction in the 1900s due to anthropogenic effects, with just ca 29 individuals remaining on Cousin Island (4°20′S, 55°40′E; 0.29 km 2 ) by the 1960s (Crook, 1960). After intensive conservation, the population recovered to carrying capacity on Cousin (ca 320 adults present in ca 110 territories) by the 1980s (Brouwer et al., 2009;Komdeur, 1992). Additional populations were established by translocations to four nearby islands (Table S1) (Komdeur, 1994;Richardson et al., 2006;Wright, Spurgin, et al., 2014). Founder individuals (all from Cousin) were selected based on sex, age, body condition, and breeding experience but without reference to genetic characteristics . Translocations to Aride and Cousine were undertaken before blood sampling became routine, whereas sampling of all the founders of the Denis and Frégate populations was undertaken . Of the translocated populations, two are now at carrying capacity (Aride: ca 1850 individuals; Cousine: ca 210 individuals ), while the populations on the other islands are still increasing (Denis: ca 424 birds in 2015 (Doblas & McClelland, 2015); Frégate: ca 141 birds in 2016 (Johnson et al., 2018)).
The Seychelles warbler population on Cousin Island has been monitored since 1986 (Hammers et al., 2019;Komdeur, 1992). A comprehensive population census has taken place every year during the major breeding season (June-September), and since 1997 also during the minor breeding season (November-March), except in -2002(Brouwer et al., 2010. Individuals were recorded as present if caught, or observed, during the field season.
The other populations have not been censused regularly, and only sporadic census data are available.
The rate of annual resighting of individuals on Cousin is high (0.98, Brouwer et al., 2010) and there is virtually no inter-island dispersal (0.1%, Komdeur et al., 2004), thus enabling accurate survival estimates (Brouwer et al., 2006). Individuals can be confidently presumed dead if not seen for two consecutive breeding seasons; the date of death is assigned as the end of the last season in which a bird was observed (Hammers et al., 2013). Ages were rounded to the nearest 0.5 years. Adult annual survival is high (84%), with mortality being greatest in first-year birds (Brouwer et al., 2006). Median lifespan is 5.5 years post-fledging, and maximum lifespan is 19 years (Hammers & Brouwer, 2017).
Females typically lay single-egg clutches (Richardson et al. 2001) and only occasionally two or three eggs (Komdeur, 1991). They are facultatively cooperative breeders, with a socially monogamous dominant breeder pair defending strict territories year-round (Komdeur, 1992). Some adult birds delay independent breeding and become subordinates , and may help raise offspring (Hammers et al., 2019;Komdeur, 1992). Although 44% of female subordinates gain reproductive success by cobreeding, male subordinates rarely gain paternity (Raj Pant et al., 2019;Richardson et al., 2002). Extra-pair paternity is frequent in this species (Richardson et al., 2001), with 41% of offspring fathered outside the natal territory (Raj Pant et al., 2019).
Individuals are caught either by mist-net, or as nestlings, and are aged based on hatch date, behaviour, and eye colour at first catch (for details see Komdeur, 1992;Wright, 2014). Each bird is given a metal British Trust for Ornithology (BTO) ring and a unique combination of three colour rings (Richardson et al., 2001). Routine blood sampling began in 1993. Since 1997, >96% of the Cousin population has been ringed and blood sampled (Raj Pant et al., 2019). Samples

| Molecular methods
Genomic DNA was extracted from blood using either a salt extraction technique (Richardson et al., 2001) or, since 2013, the DNeasy blood and tissue kit (Qiagen). Sex was determined via PCR (Griffiths et al.,1998). Individuals were genotyped at 30 polymorphic microsatellite loci (Richardson et al., 2001). Parentage assignment was carried out using MasterBayes 2.52 (Hadfield et al., 2006); for full details see Sparks et al. (2021). Parentage assignment was conducted for 1966 offspring that hatched between 1993-2018, with 89% of fathers and 86% of mothers assigned at ≥80% accuracy. Standardised individual and maternal microsatellite heterozygosity (H s ) was calculated using the R package genhet 3.1 (Coulon, 2010). Two of the microsatellite loci were excluded from this heterozygosity analysis due to pooled alleles (see Sparks et al., 2021). Variation at exon 3 of the MHC class I loci (MHC-I) had previously been screened in individuals from Cousin (1148 individuals hatched between 1992-2009) (Richardson & Westerdahl, 2003;Wright, Spurgin, et al., 2014).
Variation within the leucine-rich repeat domain of TLR3 exon 4 had previously been characterised; of the three SNPs found, only one SNP was nonsynonymous and had a minor allele frequency of >0.05 (Gilroy et al., 2017). This focal SNP is found at 198 bp in the Seychelles warbler TLR3 reference sequence (NCBI accession number: KM657704.2), where the presence of an A or C nucleotide caused a change of amino acid from lysine (+ charge), to asparagine (polar). Variation at KM657704.2:g.198A>C (hereafter referred to as TLR3 SNP) was genotyped in 1647 individuals using the KASP genotyping technology by LGC Genomics, Hertfordshire.

| Analyses
Unless otherwise stated, all analyses were conducted in R 3.6.1.

| Temporal patterns of TLR3 variation on Cousin
In total, 1190 birds hatched on Cousin from four cohorts 1992-1994, 1997-1999, 2005-2010, and 2016-2018, were sequenced at the TLR3 SNP. The earliest and latest of the sampled cohorts were used to assess temporal changes. In addition, the years 1997-1999 and 2005-2010 were selected; (a) to avoid hatch years in which translocations happened (2004,2011), as the subsequent reduction in population density may have a positive effect on juvenile (<1 year) survival in that year (Brouwer et al., 2006), and, (b) to focus on individuals with the most complete MHC and life-history data. Temporal allelic variation was analysed using a linear model (LM) and significance was assessed using the F-statistic. Frequency of the TLR3 allele in the sampled adult or juvenile population was the response variable, while year was the fixed factor.

| Survival
A mixed-effects Cox proportional hazards model in the package coxme 2.2-14 (Therneau, 2019) was used to determine whether TLR3 genotypes differed in survival. Model diagnostics using Schoenfeld's residuals confirmed that proportional hazards assumptions were met (Grambsch & Therneau, 1994). Age at death was standardised to bi-annual levels corresponding to the major and minor seasons.
Fieldwork was not conducted for four minor breeding seasons (2000)(2001)(2002)2006), so accurate bi-annual survival estimates could not be calculated for 77 individuals. Instead, the minimum date of death was assigned (i.e., the last season an individual was observed).
Excluding these individuals did not qualitatively alter the results, so they were retained in the model. Birds first caught as an adult (>1 year, n = 21) were excluded to prevent any survivorship bias by including individuals that have already survived the first year of life, and because Seychelles warblers cannot be reliably aged past one year of age (Wright, 2014). Individuals that were translocated to other islands (n = 39), and those still alive after the major 2018 breeding season (n = 42) were right-censored. Previous work has found that in low-quality seasons maternal heterozygosity affected offspring survival (Brouwer et al., 2007), and MHC diversity positively affected survival in juveniles, while individuals with the MHC-I allele (Ase-ua4) have a greater life expectancy (Brouwer et al., 2010).
Due to these fitness component differences, and the fact that MHC-I has similar properties to TLR3 in that it primarily binds intracellular peptides, we also include MHC-I characteristics in subsequent analysis. TLR3 genotype (TLR3 AA /TLR3 AC /TLR3 CC ), MHC diversity  between 1997-1999 or 2005-2010, for which these data were available, were included (n = 517).
Cox proportional hazards models in the package survival 2.44-1.1 (Therneau & Lumley, 2015), without the random effects, were used to plot Kaplan-Meier survival curves.

| Reproductive success
A zero-inflated generalised linear mixed model (GLMM) with a Poisson error structure was run using the package glmmtmb 0.2.3 (Brooks et al., 2017) to test whether lifetime reproductive success (LRS) was associated with TLR3 variation. LRS was measured as the number of offspring that survived to independence (3 months) produced throughout an individual's lifespan. Both social and extra-pair offspring were included. Individuals that were translocated, or still alive after the minor 2018 season, were excluded due to incomplete data. Individuals first caught over one year of age, for which we did not have accurate age and longevity data, were also excluded. All other birds hatched on Cousin between 1997Cousin between -1999Cousin between and 2005Cousin between -2010 were included (n = 487). TLR3 genotype, MHC diversity, presence of the Ase-ua4 allele, and individual H s were fixed factors in the model, with year of hatch as a random factor to control for cohort effects.
The sexes were modelled separately as it is likely that different factors and constraints act upon males and females.
As LRS is strongly correlated with longevity (GLMM, p < .001, Table 2), and survival was strongly correlated with TLR3 genotype (COXME, p = .026, Figure 2, Table 1), we tested if lifetime reproductive rate (defined as reproduction controlling for longevity) was associated with TLR3 genotype. The model and data set used was the same as used for LRS, except for two key differences: (a) Individuals which died before reaching adulthood (i.e., one year of age) were excluded from this analysis (resulting in n = 323), (b) Age at death (i.e., longevity and longevity 2 ) were included as fixed factors. The inclusion of longevity, and the exclusion of non-adult individuals, allows reproductive success to be isolated from survival; thus gaining a measure of the rate of reproduction during the individual's adult life.
For both LRS and rate of reproduction models all continuous factors were standardised (scaled and centred) using the package arm 1.10-1 (Gelman et al., 2018). Collinearity between fixed effects was tested using variance inflation factors. We used the package dharma 0.2.4 (Hartig, 2017) to confirm that there was no over or under dispersion, residual spatial or temporal autocorrelation in the GLMM models. We used model averaging using the dredge function in the mumin package 1.43.6 ( Barton & Barton, 2019) to select plausible models. All models within seven AICc of the top model were included in the averaged model to get the final conditional model.

| Selection coefficient
Mean values of LRS were calculated for each genotype from the raw data, relative fitness per TLR3 genotype was calculated by dividing the mean for all three genotypes by the mean from the genotype with the greatest fitness. The data set used was the same as that used for LRS -except that mean LRS was measured as the total number of offspring produced by an individual that survived to recruitment (>1 year) as this is a more accurate measure of genotype contribution to the next generation.

| Hardy-Weinberg Equilibrium in young birds on Cousin
Deviation from Hardy-Weinberg Equilibrium (HWE) was tested using exact tests (Guo & Thompson, 1992) based on allelic frequencies in genepop 4.2 (Rousset, 2008). p-values were estimated with Markov chain algorithms (1000 dememorisations, 100 batches, 1000 iterations), and F IS values are presented using Robertson and Hill estimates (Robertson & Hill, 1984). First, all birds from Cousin first caught before three months of age (before independence) were tested (n = 591).
Second, to determine if early-life mortality changed HWE proportions, this test was repeated including only individuals that survived until adulthood (n = 361). To determine if any deviation from HWE was caused by a temporal Wahlund-like effect (as in Pusack et al., 2014) we also re-ran the analysis separately for each hatch year.

| Spatial and temporal TLR3 variation across islands
The earliest available samples from the source population Cousin (120 birds caught in 1993 and 1994), were used to provide a proxy estimate of the initial TLR3 diversity on Aride and Cousine (which were established in 1988 and 1990 before sampling took place).
Samples from 56 of the 58 birds translocated to Denis, and all 59 birds translocated to Frégate were used to determine initial TLR3 diversity on these islands. The most recent population samples were  (Table S1). genepop 4.2 (Rousset, 2008) was used to test if the different island populations conformed to HWE (as above). We tested for temporal and spatial divergence in TLR3 frequencies among populations using genic differentiation tests (Raymond & Rousset, 1995) in genepop 4.2 (Rousset, 2008). Fisher's exact test and the Markov chain algorithm parameters were as above. First, we tested for differentiation between the initial (translocated or 1993-1994 samples) and most recent samples from each population. Second, we tested for differentiation among populations using the most recent samples.

| Ethics statement
Fieldwork was carried out in accordance with local ethical regulations and agreements. The Seychelles Department of Environment and the Seychelles Bureau of Standards approved the fieldwork.

| Temporal patterns of TLR3 variation on Cousin
In the adult population on Cousin, the frequency of the minor TLR3 C allele decreased significantly over time from 0.40 in 1993 to 0.29 in 2018, with a corresponding increase in the TLR3 A allele (LM: Figure 1). Likewise, the minor TLR3 C allele also significantly decreased over time in the juvenile population from 0.44 in 1993 to 0.23 in 2018 (LM: R 2 = 0.68, F 1,12 = 28.7, p < .001, Figure 1).

| Testing for contemporary selection on TLR3 variation on Cousin
There were significant differences in lifetime survival probabilities between TLR3 genotypes. Individuals (first caught as juveniles) with the TLR3 CC genotype had a 37% increased mortality risk compared to those with the TLR3 AC or TLR3 AA genotypes, with a median age of death of 1, 2, and 2.5 years respectively (COXME, p = .024, Figure 2, Table 1). Thus, individuals with at least one copy of the TLR3 A allele had an increased probability of survival than those without (p = 0.025, Table S2). Independently -and as found previously in a smaller data set (Brouwer et al., 2010) -individuals with the Ase-ua4 MHC-I allele had a 25% reduced risk of mortality than those without, corresponding to a median age of death at 3.5 years (compared to 2 years for those individuals without) (COXME, p = .028, Table 1).

| Hardy-Weinberg equilibrium in fledglings sampled on Cousin
There was a significant deviation from HWE among fledglings (individuals < 3 months of age) on Cousin, with a deficiency of heterozygotes (n = 591, F IS = 0.12, p = .002, Table S4, Figure S1a). However, there was no deviation from HWE in those individuals that survived until adulthood (individuals >1 year, n = 380, F IS = 0.08, p = .13 Figure   S1b). Individuals caught < 3 months of age were then separated into hatch year, and HWE was assessed for each year. The heterozygote deficiency was consistent across most years (indicated by a positive  Table S4).

| Spatial and temporal TLR3 variation across islands
No significant deviation from HWE was observed in any of the different island populations, either pre-or post-translocation (Table   S5). All populations showed the same overall trend, with TLR3 C alleles decreasing in frequency over time (Figure 4), but the rate of change differed between islands (Table 3, Figure 4). As shown above for adults and juveniles, TLR3 C allele frequencies on Cousin were significantly lower for individuals caught in 2018 compared to 1993-1994 (p < .001, Table 3, Figure 4). Of the translocated populations, only Denis showed a significant decline in TLR3 C allele frequency between the initial and most recent sample (15 years difference; p = .002; Figure 4; Table 3). TLR3 allele frequency temporal differences for Frégate (7 years difference), and between the oldest samples from the source population (Cousin) and the contemporary samples from Aride and Cousine (20 or 28-year difference respectively) were not significant ( Figure 4; Table 3).

F I G U R E 3
Effects of TLR3 genotype on reproductive success in the Cousin population of the Seychelles warbler: (a) Lifetime reproductive success (offspring surviving >3 months) for all birds; n = 487), (b) Rate of reproduction (i.e. offspring surviving to >3 months/ longevity for focal birds that survived to adulthood; n = 323). Data are raw means and standard errors, with female data shown in light grey and males in black separated by genotype, with associated sample sizes at the bottom. ***p < .001, **p < .01, *p < .05 , Ase-ua4 = Present.

TA B L E 2 Reproductive success in male and female Seychelles warblers on Cousin Island in relation to
Significant terms are in bold and underlined.

| DISCUSS ION
We detected spatial and temporal changes in TLR3 variation in the Seychelles warbler. On Cousin, we found a consistent decline in the minor allele frequency of the nonsynonymous TLR3 C allele in the adult population, from 40% in 1993, to 29% in 2018 ( Figure 1). Importantly, differential survival was associated with TLR3 genotype; individuals with the TLR3 CC genotype had 37% increased mortality risk compared to those with TLR3 AC or TLR3 AA genotypes. Furthermore, males -but not females -with TLR3 CC or  TLR3 AC genotypes had lower LRS than those with the TLR3 AA genotype ( Figure 3a). Even when controlling for longevity, males with the TLR3 AC genotype had reduced reproduction compared to those with the TLR3 AA genotype (Figure 3b). Notably, the TLR3 genotypes of nestlings/fledglings deviated from Hardy-Weinberg expectations.
Lastly, although we found differences in the TLR3 minor allele frequency among the island populations (Figure 4), they all showed the same pattern of a decrease in the minor allele frequency.
The temporal pattern in our data -with the TLR3 C allele declining in the population on Cousin over a 25-year period -could be driven by a number of evolutionary forces. However, the lack of migration in or out of Cousin (Komdeur et al., 2004), means it cannot be caused by gene flow. Importantly, our results show that individuals of either sex that were homozygous for TLR3 C had lower survival and that as it is with the differential fitness of the TLR3 C allele, suggest that selection is currently the prevailing force acting upon this locus in this population. Furthermore, a previous study showed that neither neutral microsatellite diversity, nor functional MHC allelic richness, changed over an 18-year time period in the Cousin population, while the mean MHC diversity per individual increased over that time . This lack of a change at these other loci may suggest that the effect of genetic drift is limited in this already genetically depauperate (Hansson & Richardson, 2005;Richardson & Westerdahl, 2003) population over the timeframe observed here.
While various studies have linked TLR variation with pathogen infection (Quemere et al., 2015;Tschirren et al., 2013), few have found direct links between TLR variation and fitness in wild populations. In the pale-headed brushfinch (Atlapetes pallidiceps), decreased survival was associated with high overall TLR diversity (Hartmann et al., 2014), whilst in song sparrows (Melospiza melodia) there was no relationship between survival and TLR heterozygosity (Nelson-Flower et al., 2018), although in both cases the effect of specific alleles was not tested. In the Stewart Island robin (Petroica australis rakiura), early life mortality was reduced in individuals with the TLR4 BE genotype, compared to other TLR4 genotypes, despite it being a synonymous substitution (Grueber et al., 2013). Finally, in Attwater's prairie-chicken (Tympanuchus cupido attwateri) the presence of a specific TLR1B allele was associated with reduced survival (Bateson et al., 2016). Like the latter two studies, we found the presence of a specific allele to confer differential survival; the TLR3 A allele conferred a selective advantage via increased survival, predominantly in early life. Given the importance of TLR3 as an innate immune receptor (Barton, 2007), and that the SNP investigated causes a functional difference in the binding region, it is likely that the survival differences seen here are due to differential pathogen recognition.
In this study, we also found some evidence of TLR3 genotypes conferring differential reproductive success in male, but not female, warblers. To our knowledge, this is the first-time variation at a TLR gene has been associated with reproductive success in a wild population. In vertebrates, longevity is generally strongly positively correlated with LRS (Clutton-Brock, 1988), indeed we found longevity to be the greatest predictor of reproductive success in the Seychelles warbler. However, even after controlling for fitness effects associated with offspring genotype, ability to breed, and longevity we found an effect of male TLR3 genotype. Combined with differential survival, this resulted in TLR3 AA males having considerably greater overall LRS than other genotypes. This observed difference in the reproductive output of males, but not females, could be driven by male-male competition -with males in better condition (through differential immune response due to the TLR3 variation) better able to outcompete others and gain more social or extra-group offspring.
For example, specific alleles at both immune and non-immune loci have been associated with increased competitive ability and increased reproductive success in male vertebrates (Johnston et al., 2013;Sepil et al., 2013).
If female choice is occurring based on the TLR3 variant in the Seychelles warbler this could explain how only male, and not female, individuals had differential reproduction associated with different TLR3 genotypes. Previous studies, on both the Seychelles warbler Wright et al., 2016) and other vertebrate taxa, have focused on MHC-based female mate choice (reviewed in Kamiya et al., 2014;Milinski, 2006). As we found a TLR3 heterozygote deficiency in offspring, it is possible that assortative mating could be taking place, whereby individuals mate with individuals similar to themselves more frequently than expected by chance (Sin et al., 2015). Likewise, as TLR3 heterozygous individuals do not have greater fitness than TLR3 homozygous individuals, mate choice is unlikely to be based on TLR3 heterozygosity. Further investigation should focus on 'good genes' or assortative mating as potential candidate mechanisms in driving the differential reproduction observed in this study.
A third possibility that could explain the pattern of reproductive success linked to TLR3 variation is that the heterozygote deficit in offspring is due to selection on those offspring. For example, males with TLR3 AA genotypes are unable to produce TLR3 CC offspring (whoever they breed with), so those males will never suffer from reduced reproductive success caused by the higher mortality of TLR3 CC offspring, and thus will have higher LRS. Nonetheless, if this were the sole determinant of the differential reproductive success found in this study, one would expect an equivalent outcome for females. However, there was no effect of TLR3 genotype on female overall LRS or rate of repro-  (Hansson & Richardson, 2005;Richardson & Westerdahl, 2003), may indicate that balancing selection is acting on this locus over the long-term. Given the role this locus plays in the innate immune response, this is likely to be pathogen-mediated. Of the three main mechanisms by which balancing selection is thought to maintain immune variation (reviewed in Spurgin & Richardson, 2010), our study shows that this is not caused by heterozygote advantage (Doherty & Zinkernagel, 1975); TLR3 AC individuals did not gain greater LRS or have increased survival than the homozygote genotypes. The variation observed could potentially be driven by rare allele advantage (Slade & McCallum, 1992), or fluctuating selection (Hill et al., 1991), or both. However, differentiating the relative importance of these two mechanisms in driving genetic variation, and separating them from other evolutionary mechanisms is complicated and beyond the scope of the present study (reviewed in Spurgin & Richardson, 2010).
In the present study, we identified a decrease in the TLR3 C allele frequency over time across all five island populations (Figure 4) though they did differ in rate of change. These temporal patterns of TLR3 C loss suggest that whatever selective agent is acting on Cousin is present on the other islands. Given their very close proximity, and similarity to Cousin -compared to the more isolated islands of Denis  (Hutchings, 2009). Influenza A has been reported in Procellariformes (petrels and shearwaters) in the Seychelles (Lebarbenchon et al., 2015), but whether this could be passed to the warblers is unknown. It is possible that we just do not see visible signs of a pathogen that is circulating in the warblers because of mild virulence or evolved host tolerance (Råberg, 2014. Furthermore, individuals may only show visible symptoms during the acute phase of infection when they are also least active, consequently they may be unlikely to be observed before recovery or death (LaPointe et al., 2009).
Even if there are no virulent pathogens currently in the populations, maintaining immunogenetic variation could have important consequences for the future success of this species. If selection continues, the SNP investigated here will go to fixation, and potentially important immunogenetic variation will be lost in the system. This is particularly important given the reduced diversity already present at this, and other innate immune genes, in the Seychelles warbler (Gilroy et al., , 2017. The innate immune response is often the organism's first line of defence against pathogens and plays an important role in the evolution to novel disease outbreaks (Bonneaud et al., 2012). Thus, knowing the underlying variation present, and understanding the mechanisms driving evolutionary change at these key functional sites could be important for future species conservation. This is important in small populations and/or those of conservation concern which often have reduced genetic variation. Managing genetic variation in such populations could be important for their adaptive potential, while monitoring pathogen presence may be important to identify and control disease outbreaks -both of which may be crucial for the population's long term survival.

| CON CLUS ION
We found strong evidence that selection, acting through both survival and to a lesser degree reproduction, was associated with TLR3 locus variation in the contemporary Cousin population. This suggests that an unknown pathogen is present in the Seychelles warbler population, driving evolution at this TLR3 locus. It is possible that this current positive selection may be part of a much longer-term pattern of balancing selection, but only further monitoring will be able to determine this.

ACK N OWLED G EM ENTS
We thank the Seychelles Bureau of Standards and the Department

DATA AVA I L A B I L I T Y S TAT E M E N T
All metadata, along with R scripts used to run analyses, are available in the Dryad Digital Repository, https://doi.org/10.5061/dryad. m905q fv06.