Schematic overview of experimental evolution, genome-wide DNA sequencing and analysis, and conservation-based cross-species interpretation of candidate loci.

Divergence in starvation tolerance and triglyceride content, and genome-wide allele-frequency differentiation in starvation-selected versus control populations.

Starvation-selected (SS) populations exhibit a pronounced extension of survival (A) under acute starvation relative to control (C) populations in females (Nested ANOVA, P < 1 × 10⁻9), indicating strong and repeatable phenotypic adaptation to starvation stress. (B) Relative triacylglycerol level (fold change) in SS versus C populations. Bars represent mean ± SD. (C) Principal component analysis of genome-wide allele-frequency data reveals clear and consistent separation between C and SS populations along PC1, which explains 53.7% of the total genetic variance. (D) Pairwise Pearson product–moment correlations of reference allele frequencies across all six populations. Darker shading denotes higher correlation coefficients. Control populations cluster tightly, whereas starvation-selected populations show reduced correlation with controls but high correlations among themselves.

Heterozygosity and Tajima’s D plotted against selective sweep signatures along major chromosome arms of D. melanogaster.

(A) Control (C) and (B) starvation-selected (SS) genome-wide profiles of heterozygosity (blue dots) and Tajima’s D (purple triangles) along chromosome arm 3L. Horizontal colour blocks indicate the genomic positions of inferred selective sweep regions, classified as sweep regions shared across all SS populations (blue), line-specific sweep regions detected in individual SS populations (red), and no sweep (gray). (C) Distribution of selective sweep lengths (mean ± S.E.M.) across chromosome arms (X, 2L, 2R, 3L, and 3R) in C and SS populations. (D) Lengths of shared selective sweep regions for each chromosome arm in C and SS populations

Genome-wide allele-frequency shifts after long-term starvation selection.

Genome-wide minor allele-frequency changes (|ΔMAF|) are shown for each starvation-selected population (SS1–4) after replicate-specific drift filtering. Each point represents a single SNP, plotted by genomic position and the magnitude of allele-frequency change relative to the ancestral population. SNPs exceeding the 99.9% drift envelope are highlighted in red, whereas those within the drift envelope are shown in black. Drift-exceeding SNPs are nonrandomly distributed and cluster across multiple chromosome arms, including X, 2L, 2R, 3L, and 3R. Their recurrence in corresponding genomic intervals across independent SS replicates indicates parallel allele-frequency change under starvation selection.

Convergent mito–nuclear and mitochondrial signatures of starvation selection.

(A) Overlap among genes in sweep–low-heterozygosity regions, drift-filtered candidates, and the MitoCarta gene set. (B) Circos plot of MitoCarta genes overlapping sweep–low-heterozygosity regions (pink) or drift-filtered candidates (green); nonoverlapping genes are shown in black. Inner ribbons denote associated biological processes. (C) Mitochondrial genes harboring differentiated SNPs are highlighted in dark gray; the strongest differentiation occurs near mtORI (ΔAF up to ∼0.4). (D) mtDNA copy number is higher in starvation-selected populations than in controls (mean ± SEM; Student’s t test, P = 0.03).

Increased longevity in starvation-selected populations.

Mean adult lifespan (days) of control (C) and starvation-selected (SS) flies. Bars show means across replicate populations; error bars indicate ±SD. Starvation-selected flies lived significantly longer than controls (Nested ANOVA, P < 0.001).

Workflow summarizing SNP filtering, ortholog mapping, and cross-species comparison with MitoCarta and human PBS candidates.

Allele-frequency differentiation in human orthologs of fly starvation-selected genes.

Genome-wide PBS distributions are shown for each focal population. (A) Bengali in Bangladesh (BEB) vs Gujrati in Houston (GIH) vs Finnish in Finland (FIN), (B) Luhya in Webuye, Kenya (LWK) vs Yoruba in Ibadan, Nigeria (YRI) vs Utah residents with northern and western European ancestry (CEU). (C) Sri Lankan Tamil in the UK (STU) vs Indian Telugu in the UK (ITU) vs FIN and (D) YRI vs Esan in Nigeria (ESN) vs CEU, with vertical dashed lines indicating the 95th (top 5%) and 99th (top 1%) percentile thresholds. Human orthologs of Drosophila starvationselected genes are highlighted, and many falls within the extreme upper tails of the PBS distributions. In the top 5% PBS tail, 30, 62, 28, and 66 genes overlapped with sweep–low-heterozygosity regions, and 125, 202, 91, and 208 genes overlapped with drift-filtered genes in BEB, LWK, STU, and YRI, respectively (binomial test; P = 5.54 × 10⁻⁵³, 1.81 × 10⁻⁶³, <2.5 × 10⁻¹¹⁴, and 0.0157 for BEB, LWK, STU, and YRI, respectively), blue indicates genes in sweep–low-heterozygosity regions; red indicates drift-filtered genes. Notably, genes involved in nutrient sensing, mitochondrial function, and lipid or amino-acid metabolism, including PTPN2, DAB2IP, PRKAG2, SDHA, FARS2, PNPLA6, DBT, and HSD17B4, exceed the 95th and 99th percentile PBS thresholds, indicating unusually strong lineagespecific differentiation at loci functionally linked to metabolic regulation. (E) Overlap among genes falling within the top 5% PBS tail across populations for overlapping sweep-low-heterozygosity regions. (F) Overlap among genes falling within the top 1% PBS tail across populations for driftfiltered genes.

Genome-wide reductions in heterozygosity in starvation-selected populations.

(a–d) Genome-wide heterozygosity profiles estimated using sliding windows (100-kb windows with a 2-kb step size) for each control (C) and starvation-selected population (SS) pair (C1–SS1 through C4–SS4). SS1-4 populations (orange) exhibit extensive, spatially coherent reductions in heterozygosity across multiple chromosomal regions, consistent with strong directional selection. In contrast, C1-4 populations (blue) maintain relatively uniform heterozygosity across the genome. The x-axis denotes genomic position along chromosome arms X, 2L, 2R, 3L, and 3R.

Starvation selection increases heterozygosity variance and generates large, recurrent selective sweep regions.

(a) Starvation-selected (SS) populations show a pronounced increase in heterozygosity variance relative to controls (σ²_SS = 0.008–0.015 vs. σ²_Control = 0.002–0.003; Levene’s test, P < 1 × 10⁻⁶). (b) Number of contiguous low-heterozygosity blocks (H < 0.05) identified per population. Starvation-selected populations harbour dramatically more such blocks (734–3,027 per line) than controls populations. (c) Chromosomal distribution of low-heterozygosity blocks across SS replicates, showing multiple regions shared among independent populations. Shared blocks are indicated in the top row. (d) Size distribution of contiguous low-heterozygosity blocks in SS populations. Sweep sizes range from 0.6 to 1.1 Mb ± SEM and are significantly larger than those observed in control populations.

Genome-wide reduction in nucleotide diversity in starvation-selected populations.

(a–d) Genome-wide nucleotide diversity (π) estimated for each matched control–starvation-selected (SS) population pair (C1–SS1 through C4–SS4). In all four comparisons, SS populations exhibit significantly lower nucleotide diversity than their corresponding C populations, consistent with widespread loss of genetic variation under sustained directional selection. Boxplots show median, interquartile range, and whiskers extending to 1.5× the interquartile range; points represent individual windows. Statistical significance was assessed using the Wilcoxon rank-sum test; ***P < 0.0001.

Heterozygosity and Tajima’s D profiles across the major chromosome arms aligned with selective-sweep signatures.

Genome-wide estimates of heterozygosity (blue points) and Tajima’s D (purple triangles) are shown for control and starvation-selected (SS) populations. Colored blocks mark inferred sweep intervals: sweeps shared across all SS populations (blue), line-specific sweeps detected in individual SS populations (red), and regions with no sweep signal (gray).

Functional enrichment of genes under starvation-driven selection.

(a) GO enrichment analysis of genes within sweep–low-heterozygosity regions reveals distinct functional signatures, including pathways involved in RNA processing, chromatin modification, organellar maintenance, intracellular transport, and mitochondrial mRNA processing. (b) Gene Ontology (GO) enrichment analysis of drift-filtered genes identifies significant overrepresentation of biological process (BP), molecular function (MF), and cellular component (CC) categories related to amino acid transport and activation, lipid metabolic processes, mitochondrial function, and membrane-associated compartments. Bars indicate P values for significantly enriched GO terms.

Frequency-dependent estimates of effective population size and neutral drift dynamics.

(a) Relationship between starting allele frequency and the effective population size (Nₑ) estimated from control populations based on variance in allele frequencies across replicates. Autosomal sites (black) and X-linked sites (red) show similar frequency-dependent patterns, with Ne peaking at low initial allele frequencies (∼0.05) and declining at higher frequencies, consistent with reduced sampling variance at intermediate allele frequencies. (b) Posterior distributions of allele-frequency change under neutrality generated by Monte Carlo simulations for discrete starting-frequency bins. Darker curves correspond to higher initial allele frequencies. The pronounced skew toward low post-drift frequencies for rare alleles illustrates the asymmetric nature of drift-driven dynamics in pooled population sequencing.

Starvation-associated variants alter predicted RNA secondary structure of mtORI.

Predicted mtORI RNA structures for control (C; left) and starvation-selected (SS; right) populations were generated using identical folding parameters. Variants alter folding topology and local stem–loop architecture, with MFE values shown (C: −26.56; SS: −27.12 kcal/mol). Base pairing is color-coded by structural context (red–orange, highly paired; yellow–green, intermediate).

Starvation-associated variants alter predicted RNA secondary structure of S6k.

Predicted RNA secondary structures for S6k in control (C; left) and starvation-selected (SS; right) populations. Structures were generated from reference (control) and starvation-associated variant DNA sequences using identical folding parameters. Starvation-associated variants produce alterations in RNA folding topology and local stem–loop architecture, suggesting potential effects on RNA stability, processing, or translation. Base pairing is color-coded by structural context, with red–orange indicating highly stable, strongly paired regions and yellow–green indicating regions of intermediate stability.