Convergent balancing selection on an antimicrobial peptide in Drosophila

Summary

Genes of the immune system often evolve rapidly and adaptively, presumably driven by antagonistic interactions with pathogens [1–4]. Those genes encoding secreted antimicrobial peptides (AMPs), however, have failed to exhibit conventional signatures of strong adaptive evolution, especially in arthropods (e.g., [5, 6]) and often segregate for null alleles and gene deletions [3, 4, 7, 8]. Furthermore, quantitative genetic studies have failed to associate naturally occurring polymorphism in AMP genes with variation in resistance to infection [9–11]. Both the lack of signatures of positive selection in AMPs and lack of association between genotype and immune phenotypes have yielded an interpretation that AMP genes evolve under relaxed evolutionary constraint, with enough functional redundancy that variation in, or even loss of, any particular peptide would have little effect on overall resistance [12, 13]. In stark contrast to the current paradigm, we identified a naturally occurring amino acid polymorphism in the antimicrobial peptide, Diptericin, that is highly predictive of resistance to bacterial infection in Drosophila melanogaster [13]. The identical amino acid polymorphism arose in parallel in the sister species D. simulans, by independent mutation with equivalent phenotypic effect. Convergent substitutions to arginine at the same amino acid residue have evolved at least five times across the Drosophila genus. We hypothesize that the alternative alleles are maintained by balancing selection through context-dependent or fluctuating selection. This pattern of evolution appears to be common in antimicrobial peptides, but is invisible to conventional screens for adaptive evolution that are predicated on elevated rates of amino acid divergence.

Results

An amino acid variant in Diptericin predicts immune defense in D. melanogaster and D. simulans

Diptericin is an antimicrobial peptide produced by dipteran flies. We previously discovered a naturally occurring polymorphism at residue 69 of the Diptericin mature peptide that was strongly predictive of resistance to infection by Providencia rettgeri, a Gram-negative natural pathogen of Drosophila [14]. The ancestral serine residue is phosphorylated and hence negatively charged (99.5% confidence, http://www.cbs.dtu.dk/services/NetPhos/, [15]). The derived arginine allele, carried by 15% of the lines in the mapping panel, is positively charged and is associated with a strong susceptibility to infection (Figure 1A, Figure S1A, [14]).

Figure 1. Convergent arginine and null mutations in Diptericin decrease D. melanogaster and D. simulans resistance to Providencia rettgeri.

A) Three Dpt genotypes of D. melanogaster: the ancestral serine residue, the derived arginine residue, and a presumed null genotype (see also Table S1, Figure S2). B) Bacterial load (CFU) is higher in D. melanogaster lines carrying null (white) and arginine (blue) alleles than serine alleles (red) at 24 hours post-infection. C) D. melanogaster lines homozygous for serine (red) survive P. rettgeri infection better than lines homozygous for arginine (blue) or null alleles (black). The dashed grey line represents sterile-wound controls. D) Four Dpt genotypes of D. simulans: the ancestral serine residue, the derived arginine residue, and two putative null genotypes. E) D. simulans lines bearing arginine alleles (blue) and presumed null haplotypes (light blue/red) have higher pathogen loads (CFU) at 24 hours post-infection than lines bearing serine (red). F) D. simulans lines homozygous for serine (red) survive infection better than lines homozygous for arginine (blue) or presumptive nulls (light blue/red). The dashed gray line indicates sterile-wound controls.

We confirmed that lines homozygous for the arginine allele are more susceptible to P. rettgeri infection than line homozygous for serine. The average pathogen load for an arginine homozygote was 20 times higher than that of a serine homozygote (Figure 1B, p < 0.001), and arginine homozygotes were almost 4 times more likely to die from infection (Figure 1C, p < 0.001). Heterozygous flies had statistically intermediate bacterial loads, indicating incomplete dominance of the serine allele (Figure S1B).

We examined the Diptericin locus in D. simulans, a sister species 1–5 million years diverged from D. melanogaster [16]. We found that an arginine polymorphism has convergently arisen at the same residue through independent mutation of the codon (D. melanogaster: AGC -> AGA; D. simulans: AGC -> AGG). In both species, the serine/arginine polymorphism is segregating in populations throughout the world, although arginine is rare in D. melanogaster and common in D. simulans (Figure S2, Table S1). We infected D. simulans with P. rettgeri and found that lines homozygous for the arginine allele carried 3 times higher bacterial loads than lines homozygous for serine (Figure 1C, p = 0.008) and virtually never survived infection (Figure 1D, p < 0.001). The derived arginine alleles of D. melanogaster and D. simulans are convergent in phenotype as well as genotype.

Diptericin null alleles in D. melanogaster and D. simulans are associated with extremely poor immune defense

We previously [14] identified two D. melanogaster lines that carry a premature stop codon in the mature peptide, and four more that carry a 12 bp deletion removing 4 residues from the mature peptide (Figure S2A). The two lines bearing the premature stop codon sustained the absolute highest pathogen loads in the initial study and the three lines carrying the deletion were in the top 7%. In the present study, the two lines homozygous for the premature stop sustained higher pathogen loads than any other lines evaluated (Figure 1B) and no flies from either line survived for more than 48 hours after infection (Figure 1C). Tissue specific RNAi knockdown of Dpt (see Supplemental Methods) also resulted in 0% survival (0/40 flies alive), compared to 70% survival (28/40 flies alive after 48 hours) in control flies (Fisher Exact Test p <0.001). These data demonstrate that Diptericin plays a vital role in D. melanogaster defense against P. rettgeri and suggest that the premature stop codon renders the gene nonfunctional.

In an additional convergence, we found a polymorphic loss-of-function allele in D. simulans. This mutation is a 6 bp deletion that begins in the 5′ UTR and removes the start codon (Figure 1D). D. simulans lines carrying this deletion sustained significantly higher P. rettgeri loads (p = 0.028; Figure 1E) and mortality after infection (p = 0.011; Figure 1F). Thus, both D. melanogaster and D. simulans are additionally polymorphic for parallel mutations that eliminate Diptericin function and reduce resistance to infection.

Allele-specific protection by Diptericin is pathogen-dependent

Previous studies using other bacterial pathogens failed to find an association between alleles of Diptericin and resistance to infection in D. melanogaster [9–11]. To test the specificity of the serine/arginine polymorphism, we measured bacterial load after infection with four other pathogens: Providencia alcalifaciens, Providencia sneebia, Serratia marcescens and Enterococcus faecalis (all Gram-negative except the Gram-positive E. faecalis). The serine allele provided greater protection against P. alcalifaciens (p < 0.001, Figure 2), but had no effect on resistance to any of the other bacteria, even though expression of the Dpt gene is strongly induced by all four Gram-negative bacteria (Figures S1C and S1D). The two lines carrying the premature stop codon had some of the highest pathogen burdens after infection with P. alcalifaciens and P. rettgeri, but were not exceptional after infection with the other three pathogens (Figure 2, Figure S1D), indicating that the phenotypic effects of Diptericin alleles are pathogen-specific.

Figure 2. Effects of Diptericin alleles are pathogen-specific.

D. melanogaster lines were infected with five different bacteria and bacterial load (CFU) was measured 24 hours after infection in genotypes homozygous for arginine (blue), serine (red), null (white). There was no effect of the Dpt allele on resistance to P. sneebia, S. marcescens, or E. faecalis infections. See also Figure S1.

There are at least five independent mutations to the arginine residue across the Drosophila phylogeny

Having observed a convergent serine/arginine polymorphism in D. melanogaster and D. simulans, we asked how many additional times such variants may have arisen in the subgenus Sophophora (genus Drosophila). D. mauritiana and D. sechellia, which are close sister species to D. simulans, are both fixed for serine (D. mauritiana: n=107, Nolte et al. 2012; D. sechellia: n=18, Daniel Matute unpublished data). In comparing the reference genome sequences of Drosophila species, we found five independent substitutions to arginine at this codon (Figure 3) as well as two independent mutations to glutamine and one substitution to asparagine. This substitution rate is highly elevated relative to expectations under a model of strictly diversifying selection (ratio of posterior odds to prior odds or maximum Bayes factor = 680.94; a value greater than 20 is considered significant).

Figure 3. Convergence across the Drosophila phylogeny.

The derived arginine state has arisen at least five times independently, glutamine twice, and asparagine once, in the Sophophora subgenus of Drosophila [36]. Codon 69 sequence is given to the right of each species name. See also Figure S3.

Recombination at the Diptericin locus obscures the signatures of balancing selection in D. melanogaster

Given that arginine is the derived state but provides weaker immune defense, we hypothesized that the polymorphism is segregating in D. melanogaster and D. simulans due to condition-dependent balancing selection with the arginine allele presumed to be beneficial under some conditions. For example, arginine might provide protection against an unknown pathogen, alter gut microbial composition, or reduce autoimmune damage.

Balancing selection can be classically inferred from sequence data, where deep divergence times between balanced alleles may result in elevated nucleotide diversity and an excess of polymorphisms at intermediate frequency [17–19]. But despite the strong phenotypic evidence that the serine/arginine polymorphism is balanced in Drosophila, we do not observe the classically predicted molecular evolutionary signatures. This may be because the convergent alleles are recently derived, and thus have short coalescent histories. Atlernatively, the large effective population sizes and high recombination rates in Drosophila [20–22] may obscure the signature of balancing selection. There is virtually no linkage disequilibrium in the Diptericin gene. In an independent set of almost 200 D. melanogaster alleles directly sampled from a natural African population [23], we observe unambiguous recombination within 33 bp upstream and 97 bp downstream of the focal polymorphism (Figure S3A, other populations not shown exhibit similar patterns). The meiotic recombination rate in the Dpt chromosomal region is in the top 20% genome-wide [24]. Thus, selection may be able to act on the serine/arginine site without leaving a measurable population genetic footprint at flanking positions (Figures S3A and S3B) and eliminating the prospect of testing for balancing selection [19] with population genetic data.

A tandem duplication of Diptericin segregates in D. simulans

In D. simulans, we found a tandem duplication of Diptericin segregating in 23 out of 37 African inbred lines (Figure 4A). These duplicates are annotated in the D. simulans genome release 1.4 as GD11417 (the derived duplicate, hereafter Dpt A2) and GD11418 (the ancestral paralog, hereafter Dpt A1, Figure 4A). The phenotypic results described above were obtained from D. simulans lines that carry a single copy of Diptericin, but we were intrigued by the possibility that duplication may have additional phenotypic consequence. We infected a group of D. simulans lines that carried the duplication and compared them to a group that only contained a single copy. Genotype at the Dpt A1 allele is the strongest predictor of resistance to P. rettgeri (Figure 4B and 4C), although lines that carry the duplication have higher survival (p<0.001) and lower pathogen burden (p=0.039) after P. rettgeri infection than those that do not. Recent duplicates often experience considerable rates of gene conversion between paralogs. We see conversion tracts in the primary sequence data where both paralogs share the same nucleotide polymorphisms, including the serine/arginine polymorphism (Figure 4D). Figure S4 shows the phylogenetic relationship of all paralogs from a set of inbred African D. simulans lines, and the two paralogs are not reciprocally monophyletic even at well-resolved nodes (Figure S4). The non-independence among substitutions arising from paralogous gene conversion violates the assumptions of population genetic tests for balancing selection and other forms of adaptive evolution [25], so these cannot be applied to the D. simulans Diptericin gene.

Figure 4. A gene duplication of Diptericin in D. simulans influences resistance to P. rettgeri and has experienced recurrent gene conversion between paralogs.

A) Schematic of the tandem duplication that generated Diptericn A2. B) Bacterial load (CFU) 24 hours after P. rettgeri infection of each D. simulans haplotype based on Ser/Arg genotype at DptA1 and DptA2 and the deletion in DptA1. C) Survival of each haplotype defined by Ser/Arg genotype at DptA1 and DptA2 and the deletion in DptA1. D) Shared polymorphism between DptA1 and DptA2 reveals recurrent gene conversion (shaded sequence blocks and asterisks). Sites are numbered relative to translational start along the top, with allele counts shown on the left. The serine/arginine polymorphism is boxed. Sites identical to the most common DptA1 allele are denoted with a period. See also Figure S4.

Discussion

A single, convergently arisen amino acid polymorphism in the Diptericin antimicrobial peptide has a large effect on resistance to bacterial infection in D. melanogaster and D. simulans. Both species also segregate for null alleles that result in high susceptibility to infection. Arginine alleles have arisen independently at least five times across the Sophophora subgenus of Drosophila. The presence of this evolutionarily convergent, large-effect amino acid variant does not support the prevailing hypothesis that insect AMPs are functionally redundant or that variation in individual AMPs has little effect on defense. Instead, it suggests that individual sites within AMPs can be targets of natural selection and may be sites of co-evolution between host and pathogen. These new data are more consistent with observations from human defensins, which have been associated with variation in disease susceptibility [26].

Previous studies have not found evidence of recurrent positive selection or balancing selection at the AMP genes of Drosophila and other insects [6], although long-term maintenance of allelic variation has been suggested for some vertebrate AMPs [27–31]. Yet selective maintenance of the serine/arginine polymorphism in Diptericin is the most plausible explanation for the repeated substitutions to the susceptible arginine across Drosophila. The polymorphism might be maintained by either direct fitness tradeoffs or alternating adaptiveness to fluctuating environments, but other potential explanations can be convincingly rejected. The sequence context around the recurrently substituted codon does not appear hypermutable, and distinct nucleotide mutations of the codon repeatedly give rise to parallel amino acid variants. The frequency of the susceptible arginine allele in D. melanogaster and D. simulans is far too high to be consistent with mutation-selection balance or genetic drift, and arginine can be assumed to be similarly common (if not fixed) in D. orena, D. ficusphila, and D. willistoni.

Our results highlight a general issue in population genetics: although we find phenotypic and molecular evolutionary evidence of a balanced polymorphism, we see no real population genetic signature of balancing selection because the high rate of recombination at the selected locus. This is likely to be a pervasive problem in Drosophila and other organisms that have large effective population sizes [32]. Frustratingly, the small footprint of selection will be winnowed even further for ancient alleles that have had more evolutionary time over which to recombine, and more recently derived alleles may have had insufficient time in which to accumulate the flanking neutral mutations that generate the signature of selection [19]. For these reasons, the general role that balancing selection plays in maintaining genetic variation may be severely underestimated by genome scans and population genetic surveys, especially in organisms with large population sizes.

The frequent incidence of natural loss-of-function alleles of antimicrobial peptides suggests that AMP function in immune defense is balanced by deleterious effects of AMPs in the absence of infection. Serial pseudogenization and duplication may explain previously observed gene family dynamics. In the Diptericin gene family of D. simulans, both loss-of-function alleles and tandem duplicates are segregating. One possibility is that during epidemics, null alleles are quickly lost from the population, but are regenerated and are possibly even beneficial when pathogen pressure is low. The evidence for adaptive divergence after duplication in arthropods is mixed [5, 33, 34], but in the case of Diptericin in D. simulans, gene conversion between the recently-duplicated paralogs currently hinders adaptive divergence.

Lazzaro and Clark [7] found evidence of paralogous gene conversion between Attacin A and Attacin B and a signature of a rapid rise in frequency of the converted region. Like the current observation at Diptericin in D. simulans, gene conversion and selection reduced sequence divergence between the Attacin paralogs. Attacin A was also found to be segregating for a loss-of-function allele, as well as a 9 bp insertion/deletion polymorphism, with the insertion present in D. simulans and D. sechellia but absent from D. mauritiana [7]. Using data from Nolte et al. [35] we have found that the 9 bp insertion is polymorphic in D. mauritiana, suggesting another incidence either of convergent mutation or long-term maintenance of the polymorphism. The previously unappreciated similarities in the evolutionary patterns of Diptericin and Attacin genes of Drosophila, combined with similar observations in organisms such as mussels [38] and vertebrates [32–36] may reveal general rules of AMP evolution.