Climate-induced range shifts drive adaptive response via spatio-temporal sieving of alleles

The study investigates how species responded to Quaternary climate fluctuations by integrating both neutral demographic processes and adaptive evolution. Prior work emphasized range shifts inferred from neutral genetic patterns and species distribution models, often treating adaptation separately. A key hypothesis, proposed conceptually by Davis and Shaw (2001), posits that as species shift ranges, local environments act as sieves that filter genotypes, driving adaptive change during colonization and establishment. The authors aim to directly test this interplay by reconstructing both neutral and adaptive histories of Dianthus sylvestris across the Alpine, Apennine, and Balkan regions. Leveraging whole-genome resequencing and gene–environment modelling, they propose and quantify a “glacial genomic offset” to measure adaptive genomic shifts from Last Glacial Maximum (LGM) refugia to present, testing whether adaptive responses arose concomitantly with post-glacial range expansion and how spatial heterogeneity shaped present-day adaptive variation.

The paper reviews how Quaternary climate shifts shaped species’ distributions and genetic structures, noting that most studies inferred range shifts via fossil records, occurrence data, and neutral genetic variation. The conceptual framework of genotype “sieving” during range shifts (Davis and Shaw, 2001) suggests adaptation can accompany migration as local environments filter standing genetic variation. Recent gene–environment association studies model genotype frequencies along environmental gradients and can project genotypes across time under the space-for-time assumption, but typically neglect demography (migration, drift). Advances in population genomics now allow integration of demography with adaptive inference, motivating an approach that jointly reconstructs neutral demographic history and adaptive genomic landscapes.

- Sampling and sequencing: 1261 individuals from 115 populations (5–20 per population) of Dianthus sylvestris across its range were sequenced at low coverage (~2×) using Illumina NovaSeq 6000 (150 bp PE). Reads were trimmed (Trimmomatic), mapped to a D. sylvestris reference (~440 Mb) with BWA-MEM, duplicates removed (Picard), indel realignment (GATK), base recalibration (ATLAS), overlapping reads clipped (bamUtil). Analyses used genotype likelihoods (ANGSD, ATLAS). - Population structure and barriers: PCA (PCAngsd) performed on full and balanced datasets; admixture proportions inferred with PCAngsd (K=2–6). Geographic visualization via interpolation/kriging. Genetic barriers identified using Monmonier’s algorithm on pairwise FST; effective migration surfaces (EEMS) to infer regions of elevated/depressed differentiation. - Admixture vs bottlenecks: Chromosome painting (MixPainter) and badMIXTURE residuals used to differentiate admixture from recent bottlenecks in contact zones. - Expansion signal (Alpine lineage): Directionality index computed from unfolded two-population SFS (polarized using reconstructed ancestral states). TDOA algorithm localized expansion origin; founder effect strength estimated. - Demographic inference: Phylogenetic placement of three D. sylvestris lineages within Dianthus using distance-based trees (ngsDists, FastME, ASTRAL-III). Moments used to fit unfolded 3D joint SFS (one representative population per lineage; 20 individuals each) under 18 demographic models with varying size changes and migration across epochs. Model selection via log-likelihood, AIC, and GIM-based likelihood ratio tests; parameter uncertainty from 100 SFS bootstraps. Generation time assumed 3 years. - Species and lineage distribution modelling: Occurrence data compiled from multiple repositories. Environmental predictors included 19 CHELSA bioclim variables, topo metrics (elevation, slope, aspect), soil type and pH (SoilGrids). Variable selection based on importance (GLM, GAM, MaxEnt, RF), VIF (<10) and correlation (|r|<0.7), yielding 10 predictors. Ensemble SDMs (GLM, GAM, MaxEnt, RF) with 5-fold CV were projected to present and LGM climate (PMIP3 ensemble of 4 GCMs). Distinct LGM refugia identified via DBSCAN clustering of predicted occurrences. Expansion routes from LGM to present reconstructed at 100-year steps with dispersal constraint (12 km per century) and competitive exclusion between lineages. - Gene–environment modelling and adaptive projections: Gradient Forest (GF) modelled allele frequency turnover of 390,262 exon SNPs across 43 Alpine populations (14 individuals each), weighting SNPs by environmental association R². Covariates included latitude/longitude or alternatively a Moran’s Eigenvector Map (MEM) reflecting expansion history. GF turnover functions applied to present-day and LGM climates to generate maps of adaptive genomic composition (visualized via PCA of transformed environmental space). - Glacial genomic offset: Defined as multivariate Euclidean distance between each present-day population’s GF-transformed genomic composition and that of its geographically closest predicted LGM refugial source, explicitly retaining spatial covariates (or MEM) to incorporate isolation-by-distance and expansion (drift) alongside environmental differences. This integrates isolation by environment, IBD, and expansion effects. - Validation using population genetics: Correlated glacial genomic offset with site-frequency-spectrum-based statistics (Zeng’s E, Fay & Wu’s H, Tajima’s D, Fu & Li’s F) computed genome-wide and weighted around environmentally associated loci (weights = GF R²). Compared nucleotide diversity (π) between geographically proximate high- vs low-elevation population pairs and across all Alpine populations binned by elevation, both genome-wide and at GF-weighted loci.

- Population structure and history: Three geographically structured lineages (Alpine, Apennine, Balkan) with strong genetic separation and barriers at the Adriatic Sea and Po Plain; contact zones in the French Prealps/Maritime Alps (Apennine–Alpine) and Brenner–Puster–Gail region (Alpine–Balkan). Demographic inference dates lineage divergences to the Penultimate Glacial–Interglacial Period: initial Balkan split 178–217 kya (95% CI), subsequent Apennine–Alpine split 114–132 kya, with minimal migration over the last ~115 kya. - Alpine postglacial expansion: Evidence for an east-to-west expansion within the Alps originating near Monte Baldo and the western Dolomites (inferred by TDOA), with a serial founder pattern and clinal diversity. Estimated founder effect strength ~1% per 139 km. - Refugia and environmental changes: SDMs identified three discrete LGM refugia corresponding to the Alps, Apennines, and Balkans. Postglacial environmental space shifted heterogeneously, with warm habitats expanding at the expense of alpine habitats in the Alps. - Adaptive genomic projections: GF projections indicate LGM refugial adaptive genotypes resemble present-day alpine genotypes. Present-day low-elevation valley genotypes are distinct from alpine genotypes, implying contemporary populations evolved from an alpine-like ancestral state; low-elevation populations harbor a subset of adaptive variation found in high-elevation/refugial populations. - Glacial genomic offset pattern: Low offsets near alpine areas close to refugia; high offsets in valleys and regions far from refugia, reflecting combined environmental and geographic distances. - Population genetic validation: Glacial genomic offset correlates with signatures of selective sweeps—positive correlation with Zeng’s E and negative with Fay & Wu’s H (p < 1×10⁻⁵). Correlations were stronger around environmentally associated loci (H_GF: r = −0.57, R² = 0.31; E_GF: r = 0.63, R² = 0.38) than genome-wide (H_GW: r = −0.48, R² = 0.21; E_GW: r = 0.53, R² = 0.26). - Diversity patterns: Nucleotide diversity is higher around environmentally associated loci than genome-wide (π_GF > π_GW; Mann–Whitney U p < 1×10⁻¹⁵). Low-elevation populations have significantly lower diversity than high-elevation populations (both π_GW and π_GF; p < 0.001), consistent with founder effects and/or polygenic selection. The low–high elevation diversity difference (Δπ_GF) increases from east to west along the expansion axis, indicating the lowest adaptive diversity at the expansion front and environmental margin.

The results provide empirical support that adaptation and range shifts acted jointly during Quaternary climate change. As D. sylvestris expanded from alpine-like LGM refugia into newly available, warmer low-elevation habitats, local environmental conditions selectively increased frequencies of warm-associated alleles while populations persisting in alpine habitats retained ancestral-like genotypes. This spatio-temporal sieving produced structured adaptive genomic patterns that align with theoretical expectations. The glacial genomic offset integrates effects of isolation by environment, distance, and expansion and is validated by SFS-based sweep signals and diversity patterns, reinforcing the inference of adaptive shifts. These findings highlight that present-day adaptive potential is shaped by historical demographic routes and environmental distances from ancestral habitats. In the Alps, low-elevation expansion-front populations may have constrained adaptive diversity, potentially limiting responses to multifaceted future climatic changes, though prior warm adaptation may confer some tolerance to further warming.

This study reconstructs both neutral demographic history and adaptive genomic change in Dianthus sylvestris, demonstrating that climate-driven range shifts and adaptation were tightly coupled via spatio-temporal sieving of alleles. The authors introduce and validate the glacial genomic offset as an integrative metric capturing evolutionary change from refugia to the present. Findings reveal east-to-west Alpine expansion from LGM refugia, distinct lineage histories, and adaptive shifts leading to reduced adaptive diversity at warm, low-elevation expansion fronts. The work underscores the enduring influence of past climate on contemporary adaptive landscapes and provides a framework that complements forward-looking predictions by adding historical evolutionary context. Future research could apply this integrative approach to other taxa and regions, refine environmental reconstructions through time, and further disentangle demographic and selection components of adaptive change.

- Gene–environment projections assume that spatial gene–environment associations hold across time (space-for-time substitution). - Glacial genomic offset relies on SDM-inferred refugia, which assumes niche conservatism; although lineage-specific niches and genetic data were used to mitigate this. - Soil pH was assumed constant through time due to lack of paleo-soil models—a stated strong assumption. - Climate between LGM and present was interpolated linearly at 100-year steps; actual climate change was non-linear. CHELSA-TraCE21k was not used due to inconsistencies with selected PMIP3 models. - Incorporation of distance effects using MEM and inverse path distance weighting can be coarse and prone to artefacts when structure is not clinal. - Limited number of geographically proximate high–low elevation population pairs (n=4) for paired diversity comparisons. - Low-coverage sequencing entails uncertainty, addressed by genotype-likelihood-based methods but still a potential source of noise.