Footprints of Worldwide Adaptation in Structured Populations of Drosophila melanogaster Through the Expanded DEST 2.0 Genomic Resource

Abstract

Large-scale genomic resources can place genetic variation into an ecologically informed context. To advance our understanding of the population genetics of the fruit fly Drosophila melanogaster, we present an expanded release of the community-generated population genomics resource Drosophila Evolution over Space and Time (DEST 2.0; https://dest.bio/). This release includes 530 high-quality pooled libraries from flies collected across six continents over more than a decade (2009 to 2021), most at multiple time points per year; 211 of these libraries are sequenced and shared here for the first time. We used this enhanced resource to elucidate several aspects of the species' demographic history and identify novel signs of adaptation across spatial and temporal dimensions. For example, we showed that the spatial genetic structure of populations is stable over time, but that drift due to seasonal contractions of population size causes populations to diverge over time. We identified signals of adaptation that vary between continents in genomic regions associated with xenobiotic resistance, consistent with independent adaptation to common pesticides. Moreover, by analyzing samples collected during spring and fall across Europe, we provide new evidence for seasonal adaptation related to loci associated with pathogen response. Furthermore, we have also released an updated version of the DEST genome browser. This is a useful tool for studying spatiotemporal patterns of genetic variation in this classic model system.

Keywords: Drosophila melanogaster; dataset; ecological genomics; local adaptation; population structure; seasonal selection.

Fig. 1.

Spatial and temporal scales of DEST. a) World map showing 530 high-quality samples part of DEST 1.0 (Kapun et al. 2020), DEST 2.0 (this study), and the DGN (Lack et al. 2016). b) Sampling density across years of sampling contained in the DEST dataset. The colors are consistent with (a).

Fig. 2.

Patterns of filtering, genetic variation, and recombination in DEST 2.0. a) Visualization of filtering information of samples using PCA. Each dot is a sample's QC metric, and the color indicates the filtering decision (legend: Pass: samples that pass the filter and are used in downstream analyses; Collapse: biological and/or technical replicates collapsed into a single representative sample; otherwise, samples were excluded due to abnormal p_N/p_S levels of high levels of missing data or contamination). b) Autosomal nucleotide diversity (π) calculated across continents (see the “Estimation of Nucleotide Diversity” section in Materials and Methods for details). c) Recombination landscape of chromosome 2L in samples representative of the 74 D. melanogaster populations analyzed (one gray line per sample). Light blue area highlights the region spanning the ln(2L)t inversion. Average (black line) and overall distribution envelope (orange shaded ribbon; delineated by the average values ±1.96 SD) are shown.

Fig. 3.

PCA and projections. a) PCA projections showing PCs 1 and 2. Analyses were done for each chromosome arm and all arms combined. The proportion of variance explained (VE) is shown at the corners of each axis. b) Projections of PCs 1 and 2 relative to latitude for Europe and North American pools. c) Same as (b) but for longitude. Notice that, in this analysis, Asia refers primarily to samples from Turkey (which is located in Western Asia).

Fig. 4.

Spatial population structure and admixture in worldwide Drosophila. a) Clustering map, based on PCA projections 1 to 3 built using k = 4 (as reported in DEST 1.0). b) Same as (a) but with k = 8 (the optimal number of clusters as defined by a heuristic Gap statistic search, as shown in the inset). c) Zoom view of k = 8 into Europe to show the hypothetical overlap zone. d) Zoom view of k = 8 into North America showing the hypothetical “Latin America” cluster (green) and Southeast cluster (yellow).

Fig. 5.

Patterns of admixture across the Americas and Australia. a) Correlation between admixture estimates obtained using moments as well as with the linear modeling method. b) Coefficients of linear admixture for Australia and North America, inferred using moments. c) Map projection of levels of African ancestry in North American samples (moments estimate).

Fig. 6.

Genetic differentiation analyses. a) Values of the F_ST estimates over all DEST samples and their 95% CI (corresponding to ±1.96 SE estimated using block−jackknife with blocks of 50,000 consecutive SNPs). Note that the hF_ST, F_GT, and F_SG statistics were estimated using the hierarchical F_ST model, over all DEST samples grouped according to the k = 4 clustering analysis and their 95% CI. Colors indicate autosomes and X chromosomes. b) Pairwise comparisons between cluster-continents (under k = 4) results in a heat map. In this plot, 1-Africa” refers to Sub-Saharan African populations, and “3-Africa” refers to North Africa. The clusters “Australia-3” and “Australia-4” represent samples with low and high levels of African admixture, respectively. c) F_ST estimates within clusters from the k = 4 analysis. For (b) and (c), the population names represent the continent of provenance as well as the cluster number identified in Fig. 4a.

Fig. 7.

Geographically informative markers. a) Bi-plot of d_hav from the DEST 1.0 GIMs. City model (y-axis) and Region model (x-axis). b) Mean and 95% CIs of d_hav for the DEST 1.0 and 2.0 GIM model (to improve readability, the x-axis has been log₁₀ transformed, and CIs < 0 were set to 1; as 0 is logarithmically undefined). The mean distance to the true value is shown by dashed vertical lines (red for DEST 1.0, blue for DEST 2.0, models). c) Quality of predictions for the GIM DEST 2.0 model. The color indicates the average distance between the real d_hav of a sample and its predicted d_hav. Yellow are good predictions (accuracy = 0 to 10 m), white are “adequate” predictions (10 to 100 m), and red are poor predictions (1,000 to 10,000 m).

Fig. 8.

Temporal genetic differentiation due to overwintering. a) F_ST values across DEST 2.0 samples as a function of latitude. Broken-stick regression and breakpoint are shown, for samples below latitude 50.3, the regression is shown with and without Turkey. The color indicates the average temperature in Celsius between the samples for which the F_ST was calculated. b) Distribution of year-to-year F_ST values across DEST 2.0 samples as a function of latitude, for comparisons spanning one winter only. Outliers (i.e. data above the 75th percentile) are shown in red. c) Distribution of temporal F_ST values as a function of the mean temperature in Turkey (Yesiloz) samples collected between 2015 and 2020 (logit transformed; correlation between F_ST and mean temperature; r = 0.135; P = 4.60 × 10⁻⁷). d) Same as (c) but for comparisons of 2020 and 2021, a historical heatwave year in Turkey and in Southern Europe (correlation between F_ST and mean temperature; r = −0.100; P = 7.74 × 10⁻¹³). e) Mean year-to-year F_ST overlaid over a world map of northern seasonal habitats.

Fig. 9.

Local and seasonal adaptation in Drosophila. a) Schematic of sampling for the local and seasonal analysis. In total, we used 138 samples collected in 26 European localities across an 8-year period. We selected localities where there was more than one sample per year and designated the first sample as “spring” and the last sample as “fall.” There is no overlap between the samples used here and the samples used in seasonal analysis in Machado et al. (2021), Bergland et al. (2014), and Nunez et al. (2024). b) Results of the GLMM analysis. The permutations are shown in gray (95% CIs) and the real data in red. There are more SNPs with low seasonal P-values than expected by permutations. c) We performed the contrast analysis using BayPass 2.4. The contrast score (C₂ statistic) is the test statistic for the seasonal term and follows a χ² distribution with 1 degree of freedom. The x-axis is the −log₁₀(P-value) from the GLMM. The red horizontal line represents the 99.9% significance threshold from the pseudo-observed data (POD) for ∼10 M simulated sites. The red vertical line represents the 99.9% significance threshold from the permutations of the GLM analysis. d) Bayesian local adaptation scan. The plot shows the log₁₀ transformed wZa P-value of the local adaptation (XtX*) BayPass analysis. For d, e, and f, regions of interest are highlighted in yellow. Inversions are demarcated along the top of the figure. e) Bayesian seasonal adaptation scan. The plot shows the log₁₀ transformed wZa P-value of the contrast (C₂) adaptation BayPass analysis. f) GLMM seasonal adaptation scan. The plot shows the log₁₀ transformed wZa P-value of the LRT of base and seasonal models.