Environmental selection, rather than neutral processes, best explain regional patterns of diversity in a tropical rainforest fish

The study investigates how environmental selection versus neutral processes structure genetic and morphological diversity in a tropical rainforest fish, Melanotaenia splendida splendida, across the Wet Tropics of Queensland. Tropical rainforests are biodiversity hotspots with complex evolutionary histories and high vulnerability to environmental change. Prior work suggests both vicariant isolation and ecological gradients can drive divergence, but disentangling adaptive from neutral processes remains challenging, especially in under-studied tropical freshwater systems. The authors hypothesize that hydroclimate (temperature and precipitation-related factors) strongly drives intraspecific diversity in this tropical ecotype, as seen in related rainbowfishes. They address three questions: (1) To what extent are genetic and morphological diversity correlated with hydroclimate beyond expectations from neutral genetic structure? (2) If ecological associations exist, can genotype–phenotype–environment links support a heritable (genetic) component to morphology? (3) How much does catchment (drainage) structure contribute to patterns of divergence? The work integrates landscape genomics and geometric morphometrics to inform eco-evolutionary processes and conservation under climate change.

The paper reviews debates on mechanisms generating tropical rainforest diversity, contrasting vicariant models (refugia, landscape breaks) with ecological divergence across gradients and ecotones. It highlights the paucity of integrated genomic and geospatial studies in tropical systems relative to temperate regions. Landscape genomics, especially genotype–environment association methods, can detect selection but require accounting for demography. Phenotype–environment associations and geometric morphometrics help reveal adaptive morphology, though plasticity can confound inference. Integrative genotype–phenotype–environment approaches can strengthen adaptive inference. In rainbowfishes, prior studies have shown heritable, habitat-associated morphological variation linked to hydrology, selection on thermal response plasticity, and hydroclimate-associated genomic divergence. However, genome-wide analyses were lacking for a tropical representative like M. s. splendida.

Sampling: In March 2017, 267 wild M. s. splendida were collected from nine rainforest creek sites across five drainages in the Queensland Wet Tropics. Of these, 208 were photographed for morphometrics; 210 high-quality fin-clip DNA samples were retained; 180 individuals had both high-quality genomic and morphological data. DNA and sequencing: DNA extracted via salting-out. ddRAD libraries prepared (Peterson et al. protocol with modifications) and sequenced on Illumina HiSeq platforms. Bioinformatics: Reads demultiplexed/trimmed (TRIMMOMATIC/DDOCENT), mapped to a Melanotaenia duboulayi reference genome (Bowtie2, GATK pipeline). Variants called with BCFTOOLS; filtered with VCFTOOLS to retain high-quality unlinked SNPs. Individuals with <700k reads removed. Final set: 14,540 SNPs. Neutral vs outlier loci: Population structure (K) inferred with fastSTRUCTURE; BAYESCAN used to flag outliers (FDR<0.05). Putatively neutral dataset: 14,478 loci (210 individuals). Genetic diversity and structure: Calculated He, π, PP (ARLEQUIN); global, pairwise, and site-specific F-statistics (HIERFSTAT). Built NJ tree (PAUP) and a scaled covariance matrix of site allele frequencies (Ω) using BAYPASS core model. Clustering via fastSTRUCTURE and DAPC (adegenet). Recent migration rates estimated with BA3-SNPS (BayesAss for SNPs). Environmental variables: From National Environmental Stream Attributes v1.1.3 (250 m resolution), six uncorrelated variables selected: ASPECT, RDI (river disturbance index), RUNSUMMERMEAN (summer mean runoff), STRANNRAIN (annual mean rainfall), STRANNTEMP (annual mean temperature), STRDENSITY (total upstream segment length/stream density). Genotype–environment associations (GEA): Tested associations between the full SNP set and the six environmental variables, controlling for neutral structure using multiple approaches: BAYPASS auxiliary covariate model (accounting for Ω) and redundancy analysis (RDA) with partial RDAs controlling for (1) PCs of Ω, (2) PCs of pairwise FST, or (3) PCs of waterway distance. Candidate loci identified from significant loadings/p-values. Geometric morphometrics: Eighteen homologous landmarks digitized on fish images (TPSDIG2). Procrustes superimposition performed (MorphoJ), PCA on covariance matrix to summarize shape variation. Allometric regression of shape on log centroid size conducted; residuals used in canonical variate analyses (CVA) to test shape differences among sites and catchments with permutation tests. Phenotype–environment associations (PEA): RDA of significant shape PCs (Broken-Stick) against environmental variables, with partial RDAs controlling for body size (log centroid) and neutral structure proxies (Ω, pairwise FST, and waterway distance). Genotype–phenotype–environment (GxPxE): Global and partial RDAs testing whether putatively adaptive SNPs (from GEA, 864 loci) are explained by body shape PCs (controlling for body size), to infer heritable components of environmentally associated morphology and identify candidate loci. Functional annotation: Extracted 300 bp flanking sequences around candidate SNPs from the reference genome; BLASTX against UniProtKB/Swiss-Prot (e≤1e-3); annotated hits with GO terms and functional descriptions.

• Data and diversity: After filtering, 14,540 SNPs retained; 14,478 considered neutral for population analyses. Neutral genomic diversity was moderately high (He 0.278–0.321; mean 0.293). Mean proportion of polymorphic loci PP=0.329 (range 0.252–0.391). Global FST=0.165; global FIS=0.205; site-specific FIS not significant. Pairwise FST within drainages: 0.017–0.029 (mean 0.024); among drainages: 0.071–0.208 (mean 0.120), indicating strong drainage-driven structure.
• Population structure: Clustering (fastSTRUCTURE, DAPC) supported K=5, grouping by drainage; NJ tree showed reciprocal monophyly by drainage. Contemporary inter-drainage migration was very low (m=0.0033–0.0224), versus high within-population assignment (0.9521–0.9865).
• Genotype–environment associations: Global RDA (no covariates) linked six environmental variables to 23% of genetic variation (p<0.001). Partial RDAs remained highly significant (p<0.001): • Controlling for Ω: environment explained 16.6% of SNP variation; 864 candidate loci (p≤0.0027). STRANNRAIN and STRANNTEMP most influential. • Controlling for pairwise FST: 12.1% variation; STRANNRAIN and STRANNTEMP most influential. • Controlling for waterway distance: 15.7% variation; STRANNTEMP and ASPECT most influential. • BAYPASS auxiliary covariate model identified 176 candidates: 88 with STRANNRAIN, 56 STRANNTEMP, 12 ASPECT, 10 RDI, 9 STRDENSITY, 1 RUNSUMMERMEAN. Twenty percent (36) overlapped with pRDA candidates.
• Morphology and environment: Four significant shape PCs captured variation in body depth, curvature, fin position/length, and head/mouth orientation. CVAs showed significant shape differences among most sites and all drainages. Global RDA linked environment to ~24% of body shape variation (p<0.001). After controlling for size and Ω, 14% of shape variation remained associated with environment (p<0.001), with STRANNTEMP and STRDENSITY most influential; alternative covariables yielded 11.9% and 11.6%. Shape PCs most associated with environment were PC2 (dorsal flattening, ventral curvature, upturned head) and PC4 (fin positions/widths, body depth, caudal peduncle length).
• GxPxE: Putatively adaptive genetic variation was significantly associated with morphology: 6.8% (global) and 6.5% (controlling for size) of adaptive SNP divergence attributed to shape PCs (p<0.001). Sixty-one loci were candidates for climate-adaptive morphological variation (p<0.0455).
• Functional annotation: Of 864 adaptive candidates, 128 matched Swiss-Prot proteins. Five candidates were also implicated by GxPxE, including cdh2 and SPTAN1 with known phenotypic roles in zebrafish development and fin/neuromuscular traits. Overall: Environmental selection, especially hydroclimate (temperature and precipitation), had comparable or greater explanatory power than neutral structure for genome-wide patterns and greater power for morphology, supporting adaptive divergence across the riverscape.

Findings indicate that while drainage boundaries limit gene flow and structure neutral genetic variation, contemporary environmental selection—predominantly hydroclimatic factors—better explains regional patterns of both genetic and morphological divergence. Strong GEAs for rainfall and temperature, robust to different neutral covariates and methods, suggest climate-driven selection. PEAs showed that morphology is tightly linked to environment beyond neutral expectations, with specific shape components (PC2, PC4) reflecting ecologically meaningful traits (e.g., fin positioning related to hydrodynamics; head shape associated with surface feeding or oxygen dynamics). Significant GxPxE associations and annotated candidate genes support a heritable basis for environmentally associated morphological variation, aligning with prior evidence in congeneric rainbowfishes. Drainage structure still contributes to demographic divergence, reinforcing the combined roles of environmental selection and riverscape connectivity. The results underscore potential vulnerability of rainforest freshwater taxa to climate change, as shifts in temperature and precipitation regimes could challenge locally adapted populations, particularly in dendritic systems where connectivity is constrained.

The interplay of hydroclimatic variation and drainage connectivity has shaped genetic and morphological diversity in Melanotaenia splendida splendida across the Wet Tropics. Environmental selection, especially linked to temperature and precipitation, exerts strong influence on genome-wide variation and body shape, with three-way genotype–phenotype–environment associations indicating heritable, climate-adaptive morphology. This appears to be the first freshwater tropical rainforest example using high-resolution genomics to link climate with phenotype and genotype in this way. Future research should further integrate environmental, genomic, and phenotypic datasets, expand spatial and temporal sampling, and experimentally validate candidate adaptive traits and loci to inform conservation under climate change.

• Environmental associations are putative; unmeasured or collinear environmental factors may confound detected signals.
• Morphological divergence could include plastic components; although GxPxE suggests heritability, common-garden or experimental validation was not performed here.
• Sampling spanned nine sites and a single time period; temporal variation and broader spatial coverage may reveal additional patterns.
• ddRAD captures a subset of the genome; causal variants may be missed or only linked.
• Reference mapping used a closely related species’ genome, which may affect alignment and annotation.
• Choice of six environmental variables, while justified, may omit other influential ecological factors.
• Different GEA methods yielded differing candidate sets, reflecting methodological sensitivity to demography and selection strength.