Samples used in the population genetic analysis

For D. melanogaster, the sequenced genomes of 197 haploid embryo Siavonga lines from the Drosophila Genome Nexus Project 3 were used [25]. These lines represent a single ancestral population of D. melanogaster from Zambia. For D. simulans, genome sequences from 20 isofemale lines, 10 collected in Madagascar and 10 collected in Kenya, were used [24]. In addition, a reference sequence of D. yakuba was used as an outgroup [26].

Curating genes of interest and control genes from the literature

A list of genes known to be involved in phagocytosis and autophagy was established by reviewing the primary literature (S1 and S2 Tables). In addition to the known phagocytosis and autophagy genes in Drosophila, Drosophila homologs of phagocytosis or autophagy genes characterized in other organisms were included. Each gene was assigned to a functional category (e.g. Autophagy induction) (S1 and S2 Tables; Fig 1). To test whether phagocytosis and autophagy pathways show different evolutionary patterns than canonical humoral pathways, known humoral immune signaling and recognition genes were used as a comparison. To control for effects of gene structure and chromosomal position, control genes were chosen to be similar to the focal immune genes in gene length (0.5-2x length of the coding region of the focal gene) and gene location (within 60kb ± of the start site of the focal gene). Three or four control genes that matched the criteria and had no annotated immune function were chosen for each focal gene. If a focal gene had fewer than three control genes that matched the criteria, it was removed from downstream analysis. Pairing focal and control genes controlled for potential effects of gene length and recombination rate. To ensure that focal and control genes indeed are matched for local recombination rate, the predicted recombination rates for all genes were examined using the Drosophila melanogaster Recombination Rate Calculator (RRC, version 2.3, [27]). This tool provides estimates of recombination rates based on the Marey map approach [27] and direct measurement [28]. Both RRC and Comeron estimates showed that the correlation between recombination rates of focal and control genes in D. melanogaster is strong, confirming that focal and control genes truly share a similar recombination environment (R² = 0.994 based on RRC midpoint and R² = 0.890 based on Comeron midpoint, S1 Fig). The presence of inversions can create strong haplotype structure and influence patterns of polymorphisms [29]. To minimize the effect of inversions, no focal gene or control gene used in the study was near the boundaries of known inversions. Populations collected in Africa sometimes contain genomic segments inferred to have recent cosmopolitan (non-African) ancestry [30]. To specifically analyze African genetic variation and eliminate the effect of relatedness among individual lines, genomic regions that were thought to have come from cosmopolitan (non-African) ancestry and that showed evidence of identity-by-descent in D. melanogaster were masked using Perl scripts obtained from http://www.johnpool.net/genomes.html.

Processing sequence data prior to population genetics analyses

Using custom scripts and bedtools [31], the coding sequence of each gene was extracted based on coordinates from the General Feature Format (GFF) file of the reference sequence for D. melanogaster (FlyBase release 5.25) and from the GFF file provided by the Thornton Lab GitHub (https://github.com/ThorntonLab) for D. simulans. When multiple isoforms were available for a gene, the longest sequence was chosen for downstream analysis. Custom filters were applied to exclude sequences with sites containing more than 10% missing data (noted as Ns) and then alignment was performed using PRANK [32]. Genes with poor sequencing or alignment quality, with large regions of gap, or with no annotation in any of the species were excluded from downstream analysis. To standardize the number of lines surveyed for each gene, 149 D. melanogaster lines and 14 D. simulans lines were randomly subsampled.

Surveying the level of polymorphisms and divergence

To evaluate the patterns of polymorphisms in each species, the following population genetic statistics were calculated: Watterson’s θ (θ_w), Tajima’s D (TajD), normalized Fay and Wu’s H (nFWH), Ewens-Watterson test statistic (EW), and the compound statistics that combined Tajima’s D, normalized Fay and Wu’s H with Ewens-Watterson test statistic (DHEW). θ_w indicates the level of DNA sequence variation, and a reduction in sequence diversity can be due to a recent selective sweep [33]. Tajima’s D test statistic compares the number of pairwise differences between individuals to the total number of segregating sites and detects the level of mutations of intermediate frequency relative to mutations that segregate at low frequencies [34]. The mutations that initially arise after variation is purged in a selective sweep will necessarily be at low frequency, skewing the site frequency spectrum and causing Tajima’s D to become negative [23]. Negative Tajima’s D values can also be caused by population expansion. Fay and Wu’s H test statistic detects excess of high frequency alleles in the derived state, which is a signature of selective sweeps that is less likely under population expansion [35]. The normalized version of the H test was used to increase statistical power. The EW test statistic measures haplotype homozygosity by comparing the observed homozygosity to the expected homozygosity [36, 37]. A P-value greater than 0.95 suggests that allele frequencies are more unevenly distributed than the neutral expectation, which suggests directional selection. The DHEW statistic is a composite statistic that provides more power to detect cases of positive selection [38, 39]. By combining Tajima’s D and normalized Fay and Wu’s H with Ewens-Wattersons expected homozygosity (EW), which is generally not sensitive to recombination, the DHEW overcomes the effect of recombination rates on the site frequency spectrum and makes the inference more conservative. The DHEW test first calculates required statistics separately and then combines the respective P-values into a vector to determine whether this vector deviates from the expectation under selective neutrality [38, 39]. A significant P-value points to positive selection. To calculate aforementioned statistics, the program DH was used [38–40]. Each test statistic was compared to 1) null distributions created using coalescent simulations with no recombination to obtain a P-value [39], and 2) the test statistics obtained from genomic control genes.

To evaluate the patterns of amino acid divergence, the ratio of non-synonymous to synonymous polymorphic sites in each species (Pn/Ps) and the ratio of non-synonymous to synonymous differences between the species (Dn/Ds) were calculated and the McDonald-Kreitman test (MK test) was performed [41] using a custom script. If the observed value of Pn/Ps is much different from Dn/Ds at a locus as determined by Fishers exact test, that locus is rapidly diverging between the two species at the level of amino acid sequence, which is consistent with adaptive evolution. In addition, the Direction of Selection (DoS) coefficient was calculated [42]. DoS >0 indicates the action of positive selection and DoS <0 indicates the action of purifying selection. To correct for multiple testing, the p.adjust function in R based on the Benjamini and Hochberg method was implemented [43].

Processing statistics and testing for significance

To assess whether phagocytosis and autophagy pathways as a whole show a departure from their control genes for various test statistics, the difference between the value of a given statistic in the focal gene and the median value of the statistic for the matched control genes, hereafter called a ‘comparison score’, was calculated for each focal-control pairing. We then evaluated whether the mean of all comparison scores was significantly different from 0 using both a t-test (parametric) and the Wilcox rank sum test (non-parametric) (Table 1). If the 95% confidence interval contains 0, the test provides no evidence for any statistically significant difference between the two groups as a whole. To identify individual genes that bear patterns of non-neutral sequence evolution in each species, the P-values of, D, H, EW, and the compound test statistic DHEW were used. To ensure that the pattern seen on focal genes is indeed due to a local selection, the comparison score was calculated for Tajima’s D, Fay and Wu’s H, and DoS for each focal-control pair and the scores were ranked from largest to smallest (listed as ‘rank’ column in Tables 2 and 3). A total of 68 and 65 comparison values make up the distributions for D. melanogaster and D. simulans, respectively. Randomly assigning a control gene to be "focal" and re-calculating the ranks of comparison scores for these statistics created a distribution of ranks. Then comparison scores from true coupling of focal and control genes were compared to this permutated distribution (listed as ‘rank against null’ column in Tables 2 and 3). A total of 245 and 232 values were used to build null distributions for D. melanogaster and D. simulans, respectively. The more extreme this rank is, the stronger the confidence is that the focal gene differs from the control genes for a given statistic. We set a threshold for concluding evidence that a gene had experienced a selective sweep as requiring 1) a significant P-value for DHEW, 2) significant corrected P-values for two of any of the test statistics, and 3) comparison scores for D, H, or DoS ranking in the top 10% for the ‘rank’ test and the top 5% for the ‘rank against null’ test. Individual genes that have potentially undergone adaptive divergence on each lineage were identified as having a statistically significant (P <0.05) MK test result and a Direction of Selection (DoS) coefficient >0.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Evaluation of polymorphism and divergence at the pathway level.

https://doi.org/10.1371/journal.pone.0205024.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Genes that show significant and nominal evidence of recent selection in D. melanogaster and in D. simulans.

https://doi.org/10.1371/journal.pone.0205024.t002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Genes that show significant and nominal evidence of amino acid divergence on the D. melanogaster lineage and on the D. simulans lineage.

https://doi.org/10.1371/journal.pone.0205024.t003

Data

The sequences and the scripts to process them are available in https://github.com/imjoohyu.

Survey of autophagy and phagocytosis genes in D. melanogaster

Survey of autophagy and phagocytosis genes in D. simulans

Positive selection on autophagy genes

We looked for signatures of natural selection in autophagy genes, which have both defensive and housekeeping roles. Besides removal of pathogens, autophagy degrades damaged host proteins and organelles to recycle nutrients during stressful conditions, such as starvation [5]. The core autophagy machinery is conserved from yeast to higher eukaryotes [4]. The role of autophagy in maintaining organismal homeostasis could act as a constraint on adaptation in response to pathogen pressure, which might explain why we did not see pervasive signatures of positive selection in autophagy genes of either D. melanogaster or D. simulans. However, we did identify signatures of recent positive selection in Atg14 and Ykt6, which remained significant after multiple testing correction in D. melanogaster. Atg14 is conserved in both Drosophila and mammals and is involved in the nucleation of phagosphore membrane [3]. Upon infecting mouse and human cell lines, the intracellular bacterium Brucella abortus forms a Brucella-containing vacuole using the host proteins Atg14 and Beclin-1, the mammalian homolog of Atg6, in order to be trafficked to the ER where the bacterium proliferates [54]. Similar interactions with pathogens may have led to Atg14 adapting to play a more specific role in immune defense and to our observed evolutionary pattern in D. melanogaster. While the test statistics of Atg16 did not meet the multiple testing correction threshold, its nominally significant test statistics indicate that it may have experienced a strong selective pressure because it is responsible for removing both intra- and extracellular pathogens [54].

Positive selection on glutamate transporters

We also identified a glutamate transporter gene, gb, to have elevated amino acid sequence divergence on the D. simulans lineage (Table 3). gb encodes a transporter that controls the extracellular glutamate levels in the nervous system [51]. When gb is mutated, glutamate level in the hemolymph is reduced, synthesis of glutathione (a major antioxidant) is disrupted, and the intracellular ROS is increased [55], leading to a failure in producing mature phagosomes and in phagocytosis of Staphylococcus aureus in Drosophila [51]. Another putative glutamate transporter gene polyph that also plays a role in regulating glutamate level showed a significant MK result prior to multiple testing correction [51]. Although it is unknown whether these proteins physically interact with each other, both genes are expressed in Drosophila blood cells and share a function [51]. Due to this shared function and evolutionary pattern, it is tempting to speculate that positive selection may be acting on these proteins together. A host protein that evolves to escape pathogen interference may also evolve away from its native function within the host. Thus, compensatory mutation in interacting proteins that restore full function could become adaptive [56].

Positive selection on recognition genes in phagocytosis

Opsonins and phagocytic receptors have been hypothesized to evolve under host-pathogen co-evolution because they directly bind to molecules from pathogens in order to promote phagocytosis [57–61]. Previous research reported that emp, pes, and scb, which encode phagocytic receptors that bind to pathogens bore evidence of recent and adaptive evolution in Drosophila [58, 59, 62]. TepII, TepIV (opsonin), Sr-CI (scavenger receptor) and nimrod genes (NimC1, NimB4) also show an excess of non-synonymous fixations between D. melanogaster and D. simulans in [57–59]. While emp and pes in D. melanogaster and pes and scb in D. simulans in this study showed some indication of selection, the test statistics were not significant after multiple testing correction. These distinctions between previous studies and the present one may be due to differences in methods since our method employed a composite method in addition to typical test statistics used in the field to determine significance.

No parallel adaptation between D. melanogaster and D. simulans

One objective of this study was to evaluate whether the same genes in D. melanogaster and D. simulans show similar patterns of non-neutral evolution as a test for parallel adaptation. However, we did not find evidence for parallel adaptation at the gene level. While D. melanogaster and D. simulans are closely related species, the locations in which they were collected, their ecology, and their demographic histories differ [63]. While D. melanogaster and D. simulans are known to share some viruses, such as the Galbut Virus [64], there is no information about whether they share any bacterial pathogens in the wild and it is reasonable to suppose they may generally be exposed to different pathogens because of difference in micro- or macro-ecology. If these distinct pathogens interact differently with the two host species, then parallel adaptation may be unlikely. It is important to note that detection of parallel evolution relies on the detection of statistically significant patterns in both species. Thus, they may be under-detected if the analysis has low statistical power. In our particular study, the number of lines sampled from each species differed, so the power to detect selection was not equivalent across the species. Nevertheless, our finding is consistent with previous reports suggesting that parallel adaption is not common. For example, a previous study noted that while selective sweeps affect antiviral pathways in many insect species, the affected genes varied considerably across species [65]. Another study reported that a humoral signaling gene, Relish, known to have undergone adaptive divergence in D. simulans, did not exhibit as strong of evidence of natural selection in three other sister species [66].

Soft sweeps and balancing selection

Our study mainly focused on detecting strong signatures of natural selection, as would be expected from hard selective sweeps favoring novel mutations [67], and on recurrent adaptive amino acid substitutions in a gene [41, 68]. These are the signatures expected from classic arms race model, which was our core biological hypothesis. We have less power to detect other forms of adaptation that might act on immune system genes, including soft sweeps of adaptation from standing genetic variation and balancing selection. Soft sweeps refer to a mode of adaptation where multiple distinguishable adaptive alleles are present in the population at the same time [69]. Soft sweeps may be common in D. melanogaster [70, 71] but are not easily detected by frequency-based statistics such as Tajima’s D or Fay and Wu’s H because genetic diversity is not as severely reduced when the selected site is on multiple haplotype backgrounds or is not driven to complete fixation [72]. A previous study that looked for genes evolving via soft sweeps in D. melanogaster [71] showed evidence for soft sweeps in chromosomal regions that include Vamp7 and Sr-CII, which did not show statistically significant results in our study, although the precise targets of these sweeps remain unidentified.

In the context of host-pathogen interactions, balancing selection can theoretically be generated if polymorphisms arise at co-evolving loci of both hosts and pathogens and two or more alleles are maintained at static intermediate or oscillating frequencies, or if alleles are costly in the absence of infection so cannot be driven to fixation [11, 73–75]. The genomic signatures of balancing selection can be detected in organisms whose breeding structures or population sizes result in linkage disequilibrium that extends over long physical stretches of chromosomes, such as Arabidopsis [73, 76] and humans [77, 78]. However, it is much more difficult to detect balancing selection in organisms like D. melanogaster that have large population sizes and high rates of recombination [79, 80]. Therefore, we cannot rule out the possibility that some components of autophagy and phagocytosis systems in Drosophila may have evolved under undetected balancing selection.