Adaptive Variation Amid High Gene Flow Across a Broad Latitudinal Gradient in the Reef‐Building Honeycomb Worm, Sabellaria alveolata Authors: Edward J. Wilson, Antony M. Knights, Louise B. Firth, Charlotte Corporeau, Amelia Curd, Stanislas F. Dubois, Fernando P. Lima, Rui Seabra, Flavia L. D. Nunes Published: 2026 ABSTRACT ABSTRACTClimate change is already altering, and will continue to reshape, the persistence and spatial structure of marine populations. Understanding metapopulation connectivity and local adaptation is therefore critical for identifying both vulnerable populations and those with adaptive resilience. We applied a seascape genomics approach to investigate connectivity patterns and adaptive responses in the honeycomb worm, Sabellaria alveolata —an intertidal ecosystem engineer that supports high biodiversity and provides essential ecological functions. A total of 286 individuals from 17 sites spanning the entire latitudinal range were genotyped to assess population structure and local adaptation. Outlier analyses (BayeScan and OutFlank) identified six adaptive and 2117 neutral SNPs. Neutral SNP analysis revealed low to moderate genetic differentiation, indicating well‐connected populations. Admixture with possible ancestral gene pools in Southern Edge and Biscay populations suggests the persistence of unique genetic diversity. While connectivity is likely driven by natural larval dispersal, human‐mediated transport resulting from support vessels, aquaculture, marine debris and expanding offshore and coastal infrastructure may also facilitate gene flow, shaping observed genetic patterns. Despite this high connectivity, we detected genetic signatures of adaptation to temperature, salinity and net primary productivity, with adaptive SNPs linked to genes involved in molecular and physiological responses to these environmental factors. Our findings suggest that S. alveolata is likely resilient to climate change and other anthropogenic pressures, making it a valuable model for investigating the genetic and physiological underpinnings of marine population resilience. Introduction Population connectivity and gene flow are fundamental to a species' fitness and long‐term persistence, as they help maintain genetic diversity (Hanski and Ovaskainen 2003). Connectivity and gene flow promote the exchange of genotypes that can enhance resilience against various pressures, including environmental change, overexploitation and disease (Morecroft et al. 2012; Rowley et al. 2014; Xuereb et al. 2021). Traditionally, marine broadcast spawners with long pelagic larval durations were assumed to exhibit high levels of genetic admixture (Cowen and Sponaugle 2009; Ovenden 2013). However, recent research challenges this assumption, revealing fine‐scale genetic structuring and adaptive divergence to be more common than previously thought, even in species with high dispersal potential (Allendorf 2017; De Wit et al. 2020; Jupe et al. 2020). Consequently, research is shifting from solely evaluating neutral genetic diversity to also identifying adaptive loci under selection (Benestan et al. 2016). This combined approach greatly improves our understanding of a population's capacity for environment‐specific adaptation (Sanford and Kelly 2011; Funk et al. 2012; Grummer et al. 2019). In Europe and North America, population bottlenecks followed by subsequent recolonization due to Pleistocene glaciations have led to reduced genetic diversity in northern range edges and high genetic diversity in southern range edges in many terrestrial and marine species (Hewitt 2000; Hewitt 1999; Maggs et al. 2008). A large proportion of studies have focused on commercial species or those having conservation interest, such as fish (Matala et al. 2014; Johansen et al. 2020), marine mammals (Pratt et al. 2022) and molluscs (Sandoval‐Castillo et al. 2018). However, these studies have often been constrained to limited spatial scales, with few examining populations across a species' full geographic range. This has left important range‐edge populations underexplored, particularly in the southern range edges, which, for some species, encompass less accessible or studied areas (Vendrami et al. 2019; Provan and Maggs 2012; Assis et al. 2013). The lack of information on range‐edge populations creates a significant gap in our understanding of genetic diversity and adaptive capacity across the full distribution of a species. Range edges, shaped by environmental instability, serve as sentinels for the evolution of adaptive traits that enhance resilience to extreme conditions (Pinsky et al. 2020). Populations at these edges commonly exhibit greater genetic variability due to fluctuating selection pressures, making them invaluable reservoirs of biogenetic resources (Provan and Maggs 2012; Assis et al. 2013). They are also possibly mal‐adapted to range edge abiotic conditions and may have reduced biological fitness. Such diversity is crucial for conservation efforts, particularly those aimed at rehabilitating and restoring degraded ecosystems (Pinsky et al. 2020). By neglecting to study species across their entire geographic range, we hinder our ability to fully assess the adaptive potential of natural populations. Bridging this gap is essential for developing effective conservation strategies to address future environmental challenges. This need is especially urgent for ecosystem engineer species that play a role in creating three‐dimensional habitats that support high species diversity (Goldberg 2013; Jones et al. 2018; Muller et al. 2021). In the intertidal marine realm, ecosystem engineers buffer extremes in environmental conditions, such as temperature, desiccation and anoxia, thereby contributing to the stability and resilience of the ecosystem (Teagle et al. 2017; Firth, Curd, et al. 2021; Curd et al. 2023). While many intertidal species are already suspected to live close to their physiological limits (Helmuth et al. 2006; Sunday et al. 2012), maladaptation to rapid environmental shifts may drive the local extinction of ecosystem engineers (Firth et al. 2011; Filbee‐Dexter et al. 2020). Such losses would disrupt ecosystem structure and function, leading to cascading effects on biodiversity and overall ecological stability (Morecroft et al. 2012; Curd et al. 2023). Therefore, clear identification and monitoring of adaptive responses in ecosystem engineers to environmental stresses are imperative for developing robust, long‐term conservation strategies that safeguard biodiversity and bolster ecosystem resilience amidst ongoing environmental change. The honeycomb worm, Sabellaria alveolata, is a gregarious reef‐building intertidal polychaete with a wide‐ranging distribution spanning northern Morocco to southwest Scotland (Curd et al. 2023; Curd et al. 2025). Throughout its range, it acts as an important ecosystem engineer by consolidating coarse sand and bioclastic fragments into hardened tubular structures, thereby providing complex biogenic habitat for myriad taxa (Dubois et al. 2002; Jones et al. 2020; Muller et al. 2021). In situ larval surveys have estimated pelagic larval duration to be between 4 and 10 weeks (Dubois et al. 2007), leading to marked differences in expected dispersal distances (David et al. 2022; James et al. 2023). Consequently, definitive characterisation of larval dispersal remains to be determined and is likely to differ between populations experiencing contrasting environmental conditions (Dubois et al. 2007; James et al. 2019). As a result of inshore larval aggregation behaviours and preference for conspecific settlement, S. alveolata can create both small (10m2) and large (10,000m2) inter‐connected reefs, which can support high densities of individuals and tubes (up to 60,000m2, Dubois et al. 2002). Large reef formations have been shown to enhance local biodiversity (Dubois et al. 2002; Jones et al. 2018), convey associational defence (Firth, Harris, et al. 2021) and influence coastal processes, including wave attenuation and sediment transport (Desroy et al. 2011). By providing a range of ecological functions, S. alveolata reefs act as significant biogenic coastal filters, contributing to ecosystem health and stability (Dubois et al. 2009; Jones et al. 2020; Ventura et al. 2024). Due to their essential role in supporting biodiversity and delivering ecological services, reefs are protected under the European Union Habitats Directive (92/43/EEC). Although S. alveolata has been extensively studied from molecular (Buffet et al. 2018; Muir et al. 2020; Nunes et al. 2021) and physiological perspectives (Muir et al. 2016; Curd et al. 2019), research has predominantly focused on northern and central populations within the Bay of Biscay, France. These studies reveal significant genetic structuring and population discontinuities across this region (see figure 1; Muir et al. 2020; Nunes et al. 2021; David et al. 2022). In contrast, the southern range edge remains notably understudied, creating critical gaps in understanding how genetic and physiological traits vary across environmental gradients. This knowledge is especially relevant for assessing regional adaptation and resilience to environmental change (e.g., Buffet et al. 2018; Muir et al. 2020). To address this limitation, incorporating data from multiple peripheral populations, particularly range edges, is essential for improving our understanding of its population structure and adaptive potential. Current genetic studies have yet to provide definitive evidence of local adaptation in S. alveolata (Muir et al. 2020); however, physiological variations suggest otherwise. For example, differences in membrane lipid composition between northern and southern populations indicate temperature‐driven adaptations (Muir et al. 2016). Similarly, variations in reproductive ecophysiological traits further support the presence of local adaptation (Curd et al. 2021). While much of the existing research has centred on responses to rising temperatures (Muir et al. 2016, 2020), other selective pressures—such as low temperatures, food availability and seston availability—have received comparatively little attention (Dubois et al. 2009). This oversight is significant, as prolonged exposure to low temperatures and limited food and seston availability have been proposed as major population structuring drivers across the range (Dubois et al. 2002, 2007; Firth et al. 2015; Firth, Harris, et al. 2021; Nunes et al. 2021). Investigating these additional selective pressures could provide critical insights into the mechanisms shaping population dynamics and resilience in S. alveolata . In this study, a 2b‐RAD genome scan was applied to genotype individuals of S. alveolata from 17 populations spanning the latitudinal range from northern England to Morocco, including multiple populations in the less‐studied southern range edge encompassing the Iberian Peninsula and northern Africa. Using single nucleotide polymorphisms (SNPs), we aimed to uncover novel evidence of: (1) population structuring, (2) local adaptation and (3) the identification of abiotic and biotic factors that drive selection through genotype‐environment associations. This study offers valuable insights into the range‐wide genetic structure, local adaptation and ecological resilience of a key ecosystem engineer, providing a foundation for understanding how S. alveolata may respond to changing environmental conditions. Materials and Methods Field Sampling, DNA Extraction and Library Preparation Between March 2018 and September 2019, 286 adults of S. alveolata were sampled from 17 locations spanning the known latitudinal distribution from Morocco to Northern UK (Figure 1). To minimise DNA contamination from the digestive tract, the thorax of individuals was dissected in situ. Tissues were preserved in guanidine thiocyanate buffer and stored at room temperature until processing. DNA was extracted using equal volumes of lysed tissue and phenol:chloroform:isoamyl alcohol (25:24:1) followed by ethanol precipitation (99% for initial precipitation and 75% for 2 subsequent washes). Extracted DNA was then quantified by QuBit and normalised to 200 ng/μl per sample before digestion with the Alf‐I enzyme (Thermo‐Fisher) for 12 h. Digestion with Alf‐I produced 36 bp fragments, which were then ligated to Illumina Nextera sequencing adapters with T4 ligase (New England Biolabs) for 12 h. Ligation products were then amplified with the Q5 high fidelity polymerase (New England Biolabs) for 25 PCR cycles, using the standard Q5 thermocycler profile. The PCR products were migrated on a 2% agarose gel and the fragment corresponding to the 176 bp construct (36 bp fragment plus the Illumina adapters) was excised from the gel then purified with the Macherey‐Nagel Gel and PCR Clean‐up kit. The purified products were then quantified by spectrometry using the AccuBlue NextGen dsDNA quantitation kit (VWR), standardised to the same concentration and pooled in equivalent volumes per individual for each sampling site. Final concentrations of each library were measured with the Qubit High Sensitivity kit (ThermoFisher) and library quality was assessed by Bioanalyzer (Agilent Technologies). Each population was composed of 12–20 individuals which were genotyped in 5 separate runs on an Illumina NextSeq550 at the Centre Hospitalier Regional et Universitaire Morvan, Brest. Due to differences in read output, 43 individuals were re‐sequenced to increase coverage. Sampling locations and genomic summary statistics of 17 sampled populations of the honeycomb worm, S. alveolata : Number of samples per site genotyped (n; 12–20), number of samples per site after quality control (Fn; 7–16), inbreeding coefficient (F IS), and observed and expected heterozygosity (Ho, He). Geographic information is also provided with the respective sampling date. Dotted line breaks represent genetic breaks observed in previous studies of S. alveolata : A. Firth, Curd, et al. (2021), B. David et al. (2022), C‐C1. Muir et al. (2020). *La Fontaine aux Bretons, *Moledo do Minho. Extraction of Environmental Variables Environmental variables for the 17 populations were derived from modelled data covering the years 2016–2019 (see Supporting Information). Daily physical and biological seawater parameters were obtained from the three‐dimensional Copernicus products “IBI_MULTIYEAR_PHY_005_002” (10.48670/moi‐00028) and “IBI_MULTIYEAR_BGC_005_003” (10.48670/moi‐00028) with a 1/12° horizontal resolution. Suspended inorganic particulate matter (SPIM, mg m−3) and chlorophyll‐a concentrations (Chl‐a, μg m−3) were sourced from IFREMER's OC5 algorithm, which integrates outputs from MERIS, MODIS and SeaWiFS sensors (Saulquin et al. 2010). OC5 data, validated against in situ data loggers in the Bay of Biscay and the English Channel, were obtained at a 1.1 km2 resolution (Gohin 2011). Additional daily atmospheric variables were extracted from Météo France's ARPEGE product (Déqué et al. 1994). Tidal data was sourced from the OTIS Regional Tidal Solutions database (Egbert and Erofeeva 2002), providing high‐resolution models for regional tidal dynamics. Bioinformatic Processing Pipeline Bioinformatic analysis was conducted on Ifremer's computing cluster, DATARMOR (https://datarmor.ifremer.fr/). Raw reads (available at ENA: PRJEB88855) were demultiplexed, and multiple fastq files produced for the same individual were concatenated. Reads where the restriction sites were absent were removed from the dataset using a custom python script, as the “Alf‐I” restriction enzyme employed in 2b‐RAD is not currently available in the commonly used process_radtags function in Stacks v2.64 (Rochette et al. 2019). Fastp v0.20.1 (Chen et al. 2018) was used to trim reads to a length of 36 bps. Any reads with a mean quality score below 30 were excluded from downstream analysis. A final quality control was assessed using FastQC v0.11.9 (Andrews 2010) and MultiQC v1.13 (Ewels et al. 2016) before downstream analysis commenced. The S. alveolata genome (raw reads available at ENA: PRJEB86308; assembled and annotated genome available at ENA: GCA_965245925) was used as a reference against which 2b‐RAD reads could be mapped. Reads were aligned to the reference genome using the BWA‐MEM algorithm (bwa v0.7.17; Li et al. 2009). The resulting SAM files were then processed using samtools v1.16.1 (Li 2013) for conversion to BAM format. SNPs were identified by reading aligned reads using Stacks v2.64 (Rochette et al. 2019). To address missing genomic data, a two‐step “bad apple” procedure described by Cerca et al. (2021) was applied. Loci and associated SNPs were phased by applying gstacks across all individuals within a population. Populations were then processed using populations with the parameters ‐R 0.8 ‐p 1 to retain loci shared by at least 80% of individuals within a single population. Any individuals that exhibited missingness exceeding 15% were identified using the vcfTools v0.1.16 call –missing‐indv (Danecek et al. 2011) and excluded from further analysis. Sequencing depth and missingness were compared across populations to check for possible differences in these two metrics. Remaining individuals that passed the initial threshold and quality control procedures were combined into a pooled dataset. Two SNP matrices were generated with populations: one for analysis related to neutral genetic variation (n.maf) and another for adaptive variation (a.maf). Both n.maf and a.maf retained SNPs based on the following criteria: (1) shared across 70% of all individuals, (2) present in 16 out of the 17 populations and (3) one random SNP retained per locus in order to minimise using linked SNPs from the same RAD locus (O'Leary et al. 2018) (−R 0.7, −p 16, −write‐random‐snp). They differ only in the –min‐maf option, with a value of 0.05 being used in n.maf and 0.01 in a.maf. Different values of “min‐maf” were applied according to previous studies highlighting biases associated with using minor‐allele‐frequency (maf) values lower than 0.05 when assessing demographic processes (Roesti et al. 2012; Benestan et al. 2015). All downstream statistical analysis and environmental data extrapolation and processing was conducted in R (R core Team 2021). Individuals that passed the initial quality control procedures were analysed to identify potential clonal genotypes. Multilocus genotypes (MLGs) were calculated using the adegenet package (Jombart 2008) in R. Individuals sharing identical allelic profiles across all loci were assigned the same MLG. A frequency table of MLGs was generated with the poppr package (Kamvar et al. 2014), allowing for the identification of genotypes shared by multiple individuals. Potential clones were those MLGs present in more than one individual, indicating possible clonal reproduction or recent common ancestry. To identify putative loci under selection, two differential genetic outlier analyses were applied: the R package OutFlank v0.2 (Default parameters; Whitlock and Lotterhos 2015) and BayeScan v2.1 (−n 10,000 ‐thin 10 ‐nbp 20 ‐pilot 5000 ‐burn 50,000 ‐pr_odds 100; Foll and Gaggiotti 2008). Both neutral and adaptive SNPs filtered by populations (a.maf & n.maf) were converted into GESTE formatting using pgdspider v2.1.1.5 (Lischer and Excoffier 2012) for use in BayeScan v2.1 (Foll and Gaggiotti 2008). In both OutFlank and BayeScan analyses, a Q‐value of ≤ 0.05 was used to indicate statistical significance of adaptive outliers (loci under diversifying selection). The outcome of both methods was intersected to identify a panel of eventual consensus outliers. The number of loci identified as under balancing selection was recorded, then removed from the dataset for downstream analyses. Loci under directional selection that were shared between methods were subset into two lists of putatively “adaptive” and “neutral” SNPs. To create informative datasets, both SNP lists (containing loci positions; POS) were filtered against the associated “n.maf” and “a.maf” using VCFtools (–vcf –positions –recode –snps; Danecek et al. 2011) to create two separately filtered *.vcf files of putatively “neutral” and “adaptive” SNPs. The two outputs hereafter referred to as “neutral.vcf” and “adaptive.vcf”, were the final iterations used for subsequent downstream analysis. Genetic Diversity and Population Structure All demographic statistics were computed using the neutral.vcf dataset converted into genlight format using vcfR v1.14.0 (Knaus and Grünwald 2017). The package hierfstat v0.5–11 (Goudet 2005) was used to calculate the observed and expected heterozygosity and inbreeding coefficient (F IS). The global fixation index, F ST, was estimated using varcomp.glob() in hierfstat. Significance was assessed using 999 permutations, with the p‐values being calculated as the proportion of permuted F ST values equal to or exceeding the observed value. To calculate pairwise F ST between populations (Weir and Cockerham 1984; Pembleton et al. 2013), the stamppFst function was used (nboots = 10,000, percent = 95) with the StAMPP v1.6.3 package. Heatmaps were then used to visualise genetic differentiation (pairwise FST). Values smaller than 0.01 were represented as 0. Furthermore, principal component analysis (PCA) and discriminant analysis of principal components (DAPC) were conducted in adegenet v2.1.10 (Jombart 2008) to further visualise genetic differentiation among populations. The PCA was performed using the glPCA function (nf = 10), while the DAPC was computed using the dapc function retaining 40 principal components (n.pca = 40) and two discriminant functions (n.da = 2). The optimum number of principal components was calculated using xvalDAPC. A secondary DAPC that excluded the most divergent populations found in the first DAPC was conducted to investigate divergence across a reduced geographic scale. To test for isolation by distance (IBD), a Mantel test was used to correlate the genetic distance matrix (DFST) with the geographical distance matrix (Dgeo) using the mantel function (method = “pearson”) from the package vegan v2.6–4 (Oksanen et al. 2013). The Dgeo matrix was generated by calculating the least cost distances between each pair of populations using the package marmap v1.0.10 (Pante and Simon‐Bouhet 2013). To prepare the “neutral.vcf” dataset for population structure analysis, pgdspider v2.1.1.5 (Lischer and Excoffier 2012) was used to convert the file into STRUCTURE formatting. Genetic clustering was then investigated using STRUCTURE v2.3 (Pritchard et al. 2000), performing 10 iterations for each value of K clusters ranging from 1 to 17. We applied an admixture model with a Markov Chain Monte Carlo (MCMC) consisting of 500,000 iterations and a burn‐in period of 300,000. The most likely number of genetic clusters (K) was determined with StructureSelector (Li and Liu 2018), integrating multiple estimators including ΔK (Evanno et al. 2005) and Puechmaille's metrics (Puechmaille 2016). STRUCTURE output files were processed using the R package pophelper (Francis 2017) to align replicate runs and to generate visualisations of individual ancestry coefficients for the selected K values (K = 2, K= 3, K = 4). Redundancy Analysis, Genotype‐Environment Associations We performed a partial redundancy analysis (pRDA) using the rda function in the vegan v2.6‐4 package (Oksanen et al. 2013) to examine the relationship between population‐level allele frequencies of putatively adaptive loci, identified as being under diversifying selection (adaptive.vcf) and various environmental parameters. Missing genotypes were imputed with the modal values per population prior to calculating allele frequencies, ensuring minimal bias while preserving the overall genetic structure within each population. To account for spatial autocorrelation, we computed distance‐based Moran's eigenvector maps (dbMEMs) using the geographic coordinates of each sampling site. The dbMEMs capture spatial patterns at multiple scales, allowing us to model spatial structure that could confound the relationship between genetic variation and abiotic and biotic parameters. A forward selection procedure was used to identify significant dbMEMs that explain a substantial amount of spatial variation in the data. Only significant dbMEMs (MEM1 and MEM2) were retained and included as conditioning variables in the pRDA to control for spatial structure. Prior to conducting the pRDA, a stepwise procedure was applied using the ordistep function (Oksanen et al. 2013) to select the abiotic and biotic parameters that best explained the variance in allele frequencies. To adjust for residual collinearity, environmental variables were additionally omitted from model implementation if they attained a variance inflation value (VIF) greater than 10 (Dormann et al. 2013). The final pRDA model only included environmental predictors that were selected through the stepwise procedure. All abiotic and biotic variables were scaled and standardised to ensure uniformity and comparability across variables. To assess the significance of the overall partial redundancy analysis (pRDA) model and the contributions of individual explanatory variables, permutation‐based ANOVA was performed using the anova.cca function in the vegan package (Oksanen et al. 2013) with 1000 permutations. Identification of Gene Function To identify putative gene functions, the adaptive.vcf file was imported into the genome browser JBrowse and mapped to the S. alveolata reference genome, enabling precise visualisation of adaptive SNP loci. Sequences flanking each SNP were translated into amino acids for three possible reading frames. The reading frame with the fewest stop codons was selected, and the longest nucleotide sequence was used as a query in NCBI's ORFfinder to identify potential protein‐coding regions. Each identified ORF was translated into its corresponding protein sequence, then subsequently queried against the UniProtKB/Swiss‐Prot database using BLASTp to find homologous proteins and infer potential gene functions. The most likely gene function was assigned based on the highest sequence identity and alignment score. Protein structure homology was also used to provide functional annotations to the genes associated with adaptive SNP loci. Homology of 3D protein structure can be a reliable method to generate functional traits of the protein by structure homology with UniProtKB/Swiss‐Prot database‐annotated protein. The SWISS‐MODEL repository was used to predict and model the 3D protein structure of ORFs based on sequence homology, utilising its automated mode (Bienert et al. 2017). Results In total, 554,902,287 raw reads were obtained from 286 individuals sequenced in 3 Illumina runs, of which 331,833,597 reads were retained after quality filtering, corresponding to 60% of the original reads. Following read processing and alignment, 292,625,858 BAM records were retained, of which 54,457,491 represented primary alignments (Figure S1). Nearly all populations (14 out of 17) were represented by individuals sequenced on multiple runs, reducing possible batch effects associated with a single run (Figure S2). In total, 483,868 genotyped loci were identified prior to filtering (Figure S1). Application of the “bad apples” procedure retained 189 individuals for further analysis, representing ~66% of the 286 individuals genotyped, with 7–16 individuals retained per population (See Figure 1; Fn). Following population‐level filtering, 22,804 adaptive MAF‐filtered loci and 6254 neutral MAF‐filtered loci were retained for separate analysis, respectively (Figure S1). The median sequencing depth ranged from 2 to 8 reads per SNP, and the median missingness ranged from 10% to 31% for most populations (Figure S3). Missingness was substantially higher in CASA, at 45.5% (Figure S3). Results from this population should therefore be interpreted with caution. Because sequencing depth and missingness were broadly comparable across the remaining populations, any biases associated with sequencing effort or batch effects are likely to be minimal. Prior to further analysis we tested for the presence of clones; however, none were found in the dataset. A total of 2117 (33.85%) putatively neutral SNPs were identified for demographic inferences (Figure S1). SNPs under balancing selection accounted for 61% of the a.maf dataset, and 66% of the n.maf dataset. Outflank identified 47 SNPs (0.7%) as potentially divergent, while BayeScan highlighted only six SNPs (0.03%) with potential divergence (Figure S1). For subsequent analysis, the six SNPs common to both Outflank and BayeScan were included in the partial redundancy analysis. Genetic Diversity and Population Structure The inbreeding coefficients (FIS) for all populations of the honeycomb worm were marginally negative (−0.056 ± 0.045) indicating absence of inbreeding and slight heterozygote surplus (Ho>He; Ho = 0.156 ± 0.028, He = 0.143 ± 0.019) relative to the Hardy–Weinberg Equilibrium (Figure 1). In general, populations appear to be outcrossed. The global F ST estimated across all populations was 0.038 (p = 0.001, calculated as the proportion of permuted F ST values equal to or exceeding the observed value). Pairwise F ST identified low to moderate levels of genetic differentiation across the range. Among the 17 populations sampled, pairwise F ST values ranged from 0 to 0.11 (Figure 2). Genetic differentiation was most pronounced among the Southern Edge 0.077 ± 0.01 (Mean ± SD) and Biscay 0.06 ± 0.03 (Mean ± SD) regions, relative to all other pairs of regions (Figure 2). Similarly, genetic differentiation was also observed to be higher between Douarnenez and Locquirec in relation to populations in Biscay and the south of the range, remaining relatively well‐connected to all northern sites aside from Maryport. Relatively higher values (F ST > 0.1) observed in these populations suggest that gene flow is weaker between these locations. Inter‐regional F ST values were observed to be low on average within northern (0.037 ± 0.024) and southern populations (0.02 ± 0.01). With low values of F ST 0.036 ± 0.019 (Mean ± SD) observed between all populations in the north and south, indicating genetic similarity to exist between these two regions. Isolation by distance was weak but significant (R = 0.268, p = 0.028) when using geographic distances, but was not statistically significant when calculated with least‐cost distances (R = −0.034, p = 0.59; Figure S4). Pairwise F ST values for 17 populations of the honeycomb worm, S. alveolata. Top triangle: F ST values denoted by an asterisk (*) indicate significant differentiation after correction for multiple testing (FDR/Bonferroni, p < 0.05). Bottom triangle: seaway (least‐cost) distances (Dgeo, km). Population genetic structure visualised by PCA and DAPC showed patterns consistent with those observed in the pairwise F ST analysis (Figure 3A–C). The PCA revealed a clear separation of the two Southern Edge populations (ESS and TAG, southern Morocco) which were positioned at one extreme of PC1, while Biscay populations clustered at the opposite end (Figure 3A). The DAPC showed an even clearer demarcation, with ESS and TAG being genetically distinct from all other populations (Figure 3B). After excluding these southern populations, finer scale sub‐structure became apparent, with a distinct separation along PC2 between La Fontaine aux Bretons (LFB) and Oléron (OLE) (central Bay of Biscay) and the remaining sites (Figure 3C). Principal Component Analysis (PCA), Discriminant Analysis of Principal Components (DAPC) and STRUCTURE analysis illustrating genetic differentiation among 189 individuals of the honeycomb worm, based on 2117 putatively neutral single nucleotide polymorphism (SNP) markers. Each point represents an individual assigned to sampling location (codes according to Figure 1) and genetic cluster (C1/C2 = blue; C3 = red, C4 = orange). (A) PCA showing the individuals from the Southern Edge populations positioned on one extreme of PC1, while individuals from the Biscay populations clustered at the opposite end. (B) DAPC based on 17 populations, showing genetic distinction of populations from southern Morocco (ESS and TAG); (C) DAPC after excluding the highly differentiated sites (ESS and TAG). The Biscay populations (LFB and OLE) show genetic differentiation with respect to remaining sites (D) In the STRUCTURE analysis, sites are arranged latitudinally from the northern range edge (left) to the southern range edge (right). The most likely number of clusters was identified as K = 4, based on StructureSelector metrics (see Figure S5 for details). Evidence of fine‐scale structuring was further evaluated in STRUCTURE. Admixture analysis using 2117 neutral SNPs confirmed low levels of population structuring to be present among populations. StructureSelector indicated a LnP(K) that increased up to K = 7, followed by slowly increasing values up to K = 10. The Evanno method showed a first peak at ΔK = 4, followed by subsequent decrease in ΔK values, and two additional peaks at K = 10 and K = 12. In contrast, Puechmaille's estimators (MedMedK and MedMeanK) supported K = 1, indicating an absence of stable clustering, while MaxMedK and MaxMeanK suggested at most K = 2. These results point to limited genetic structure in a context of isolation‐by‐distance (Pritchard et al. 2000; Evanno et al. 2005). STRUCTURE scenarios K = 2 to K = 4 were visualised (Figure S6), and compared with the PCA and DAPC results; K = 4 is shown in Figure 3D. All populations were found to share a large proportion (~80%) of their genotypes as evidenced by cluster 1 (C1 in Figure 3D). Notably, the proportion attributed to cluster 2 was observed to be drastically lower or absent across divergent populations, suggestive of northern origin (C2 in Figure 3D). Evidence of population fine‐scale structuring was primarily observed within the populations of Essaouira, Taghazout, La Fontaine aux Bretons and Oléron. Notably, individuals from La Fontaine aux Bretons and Oléron displayed around 20% of their genetic composition associated with cluster 3 whilst individuals from Essaouira and Taghazout exhibited approximately ~25% of their genetic makeup to be attributed to cluster 4, suggesting a likely southern edge origin (C3 and C4 in Figure 3D, respectively). Both results align with those obtained within pairwise FST, PCA and DAPC analyses (Figure 2 and Figure 3A–C). Associations With Environmental Parameters After applying stepwise selection with a significance threshold of p < 0.05 and removing variables exhibiting high collinearity, six out of the initial 38 environmental variables were retained in the pRDA model. The final pRDA model was globally significant (permutation‐based ANOVA; F 6,7 = 2.8637, p < 0.001), explaining 55.49% of the total variance, including the effects of the conditioning variables (MEM1, MEM3 and MEM4). After adjusting for the number of predictors and excluding variance explained by the conditioning variables, the adjusted R 2 was 44.4% (Figure 4), indicating that these environmental variables substantially influence the adaptive genetic differentiation observed among populations at these loci. Partial Redundancy Analysis (pRDA) illustrating genotype–environment associations for six putatively adaptive single nucleotide polymorphisms (SNPs) of S. alveolata . Four environmental variables with significant associations are displayed, with significance levels indicated as follows: * p < 0.05, ***p < 0.001. Abbreviations are as follows: Sea Surface Temperature (SST; °C), concentration of Chlorophyll‐a in seawater (Chl‐a; μg m−3), Practical Salinity Units (PSU), Tidal Amplitude (Tideamp), Net Primary Productivity of carbon (NPPV). Individual SNPs are represented by black points and labelled according to their Contig IDs: 1 = Contig_17228, 2 = Contig_18407, 3 = Contig_30,212, 4 = Contig_35888, 5 = Contig_3724, 6 = Contig_38,214. Location codes are listed in Figure 1. Four of the six explanatory variables used in the RDA were statistically significant (p < 0.05). Mean seawater temperature emerged as the most influential variable, explaining 21.4% of the total variance (F 1,7 = 5.1617, p < 0.001). Mean tidal amplitude accounted for 13.7% of the variance (F 1,7 = 3.0712, p < 0.05), and mean Net Primary Productivity and seawater salinity explained 13.73% (F 1,7 = 3.1539, p < 0.05) and 5.91% (F 1,7 = 3.0701, p < 0.05) of the total variance, respectively. Neither current velocity (F 1,7 = 1.4035, p = 0.246) nor mean analysed SPIM (F 1,7 = 1.3219, p = 0.260) were statistically significant. Putative Gene Functions of Loci Under Selection A total of six Open Reading Frames (ORFs) were identified across the contigs containing the six adaptive SNPs analysed, with ORF lengths ranging from 37 to 110 amino acids. BLASTp and Swiss model structural homology revealed several putative protein matches, providing insights into protein function (Figure S7). Contig_17728 (SNP 1) showed sequence homology to “transcription initiation factor TFIID subunit 2”, a transcription factor responsible for the initiation of transcription of RNA polymerase II (Orphanides et al. 1996). SNP1 showed structural homology to the transcriptional factor “myocyte‐specific enhancer factor 2A” that activates many muscle‐specific, growth factor‐induced genes (Black and Olson 1998; Dai et al. 2024). Contig_18407 (SNP 2) showed sequence homology to “Signalling mucin HKR1” a putative osmosensor (Tatebayashi et al. 2007; De Nadal et al. 2007), and structural homology to “Calcium‐binding signalling molecule” (Feng et al. 2010). No functional ORFs were found for Contig_30,212 (SNP 3). Contig_35888 (SNP 4) showed sequence homology to “uridine 5’‐monophosphate synthase”, an enzyme that catalyses the formation of uridine monophosphate (UMP), an energy‐carrying molecule in many important biosynthetic pathways. SNP4 showed structural homology to a protein‐like polyhedral crystal structure, involved in long‐term stabilisation of proteins in insect viruses (Ji et al. 2015). Contig_3724 (SNP 5) showed sequence similarity to “S‐adenosylmethionine: tRNA ribosyltransferase‐isomerase” a multifunctional enzyme involved in tRNA modification, a process that stabilises tRNA structure and protects tRNAs from degradation under stressful conditions (Fontecave et al. 2004; Yared et al. 2024), and is also known to contribute to cellular thermotolerance (Ohira and Suzuki 2024). SNP5 showed structural homology to “ubiquitin‐like‐specific protease 2 (USP2)”, an enzyme that removes ubiquitin molecules from proteins, thereby stabilising them and preventing their degradation by the proteasome (Cruz et al. 2021). Contig_38,214 (SNP 6) showed sequence homology to “venom metalloproteinase BumaMPs1”. Metalloproteases are enzymes that use a metal ion to catalyse the cleavage of peptide bonds in proteins, and are involved in numerous functions including extracellular membrane remodelling and tissue repair (Stamenkovic 2003), regulation of signalling molecules such as desintegrin (Giebeler and Zigrino 2016) and regulating inflammation (Fingleton 2017). SNP6 showed structural homology with “Type IV Secretion System (T4SS) related protein”, a secretion system used for transporting DNA and proteins across cells, including the transfer/evacuation of toxic effector proteins (Wallden et al. 2010). Allele frequencies of these six mutations were correlated with environmental factors in the RDA (Figure 4). Contig_17728 (SNP 1) was most strongly related to temperature, Contig_18407 (SNP 2) and Contig_30,212 (SNP 3) were inversely correlated with temperature and salinity, Contig_35888 (SNP 4) and Contig_38,214 (SNP 6) were correlated with net primary productivity and Contig_3724 (SNP 5) did not show a strong relationship with any of the environmental variables analysed, having a central position in the RDA. Discussion Across heterogenous seascapes, ecosystem engineers such as S. alveolata are vital for the continued health of marine ecosystems resultant of the numerous ecosystem functions and cascading benefits that they provide (Dubois et al. 2002, 2007; Jones et al. 2020; Ventura et al. 2024). Based on the RAD‐associated genome scan of 2117 neutral and six adaptive SNPs, this study provides novel insights into the genetic structure of the honeycomb worm across its entire latitudinal range and provides the first genetic evidence of adaptation in this common and widespread ecosystem engineer. Genetic Structure and Gene Flow The population genomic analysis of the honeycomb worm across its broad distribution indicates that the species is predominantly outcrossed, with moderately low genetic differentiation (average global pairwise F ST = 0.05 ± 0.03). These results suggest that gene flow is generally well‐maintained across populations, supporting high spatial connectivity between localities. The observed levels of genetic differentiation found here are consistent with general expectations for marine broadcast‐spawning invertebrates with similar pelagic larval durations (PLD), particularly those analysed using RADseq approaches. For instance, Holothuria scabra (PLD ~4 weeks) exhibits an average global pairwise F ST of 0.03 (Brown et al. 2022), while the bat star Patiria miniata (PLD 6–10 weeks) has an F ST of 0.06 (Sunday et al. 2014). These values align closely with those observed for S. alveolata (F ST = 0.05 ± 0.03), reinforcing the established relationship between larval duration and genetic connectivity in marine species (Cowen and Sponaugle 2009; Ovenden 2013). While these findings support the expectation that marine broadcast spawners generally maintain high levels of connectivity, they diverge from previous estimates for the honeycomb worm based on a smaller panel of SNPs, which reported stronger genetic structuring across similar spatial scales (F ST = 0.28 ± 0.10; Muir et al. 2020). This discrepancy may reflect methodological differences, particularly the larger number of sampling sites employed in this study, which included additional “stepping‐stone” sites along the Atlantic coasts of Spain, Portugal and Morocco. Additionally, the increased number of SNPs retained in our analysis provides greater statistical power and spatial resolution, leading to a more refined estimate of population differentiation (Kimura and Weiss 1964; Benestan et al. 2016). Overall, high genetic similarity was apparent throughout much of the honeycomb worm's distribution (Figure 3D), suggesting extensive connectivity across distances exceeding 3000 km. One possible explanation for this pattern is post‐glacial recolonisation, which may have re‐established connectivity among refugial populations following the last glacial maximum (Neiva et al. 2016; Nunes et al. 2021). Alternatively, the pelagic larval duration (PLD) for the honeycomb worm may be considerably longer than the previously estimated four weeks, with local larval surveys suggesting a PLD of up to ten weeks (Dubois et al. 2007). For instance, laboratory experiments have shown that larvae of S. alveolata can remain viable in culture for several months (Wilson 1968), and field observations indicate a PLD extending up to nine months (Wilson 1971). If even a modest fraction survives in the plankton beyond the typical four‐week period, extended viability could significantly enhance connectivity among populations. Although post‐glacial admixture and extended larval duration may explain greater genetic homogeneity, direct gene flow across such a broad latitudinal gradient appears unlikely in light of prevailing oceanographic currents and hydrodynamic connectivity modelled for this species (Ayata et al. 2010; Muir et al. 2020; David et al. 2022). Notably, populations from northern England to northern Morocco (spanning ~25° of latitude) showed surprisingly low genetic differentiation (average F ST = 0.036 ± 0.019), suggesting a potential role of other mechanisms beyond natural dispersal. Sporadic, long‐distance “skipping‐stone” dispersal events, facilitated by human‐mediated transport, may periodically bridge non‐adjacent populations, effectively circumventing natural barriers. Such anthropogenic dispersal has been inferred in other marine invertebrates and is known to enhance genetic connectivity in species such as Stramonita haemastoma (El Ayari et al. 2017), Ciona robusta (Bouchemousse et al. 2016), Magallana gigas (Lallias et al. 2015) and Diopatra biscayensis (Galaska et al. 2021). Notably, when M. gigas was introduced for aquaculture in the early to mid‐20th century, it demonstrated the capacity for rapid range expansion through ecological dispersal networks that stabilised within a few generations following its establishment (Clubley et al. 2024). A similar process may be influencing modern‐day gene flow and connectivity in S. alveolata , allowing populations to expand quickly and maintain connectivity across regions where natural dispersal alone would be insufficient. For S. alveolata , transport via ship ballast water appears to be a likely vector (Bailey et al. 2020), particularly as the species has multiple spawning periods, spanning from spring to autumn (Gruet and Lassus 1983). Although the transport of adults may be less frequent, it cannot be ruled out entirely—particularly as the honeycomb worm is known to colonise artificial structures and marine litter (Firth et al. 2015), some of which may occasionally be relocated. Furthermore, shellfish farming operations are thought to contribute to the transport and spread of numerous associated species, both native or introduced (Massé et al. 2023). Given that honeycomb worm is frequently observed on the shells of cultivated species such as oysters (Dubois et al. 2006), aquaculture practices may also inadvertently contribute to its dispersal. Consequently, human‐mediated dispersal may help account for the unexpectedly low F ST values observed between regions in this study. Nonetheless, direct evidence (e.g., confirming S. alveolata larvae in ballast water) is currently lacking, and the relative frequency of these anthropogenic “skipping‐stone” events compared to natural stepping‐stone dispersal remains unclear. Even sporadic occurrences, however, could have a substantial impact on large‐scale connectivity patterns. Although sporadic “skipping‐stone” events may promote broad connectivity, they may not act uniformly across the species' range. Consequently, the results from PCA, DAPC and STRUCTURE analyses reveal notable genetic divergence in two key areas: the southern edge of the species' range (Essaouira and Taghazout, Morocco) and parts of the Bay of Biscay (La Fontaine aux Bretons and Oléron), in the centre of the range (Figures 3, 4). The southernmost populations show both the highest differentiation relative to all other sites and the presence of unique genetic diversity (cluster C4). They lie on either side of Cape Ghir, where seasonal upwelling likely disrupts larval transport (Sangrà 2015; Santana‐Falcón et al. 2020). Reduced connectivity in these edge populations may result from offshore advection or local eddy formation, as well as seasonal fluctuations in nutrient availability that could drive population instability and genetic drift (Broquet et al. 2013). Within the Bay of Biscay, our findings concur with previous studies, which also identified limited larval exchange between the Bay of Biscay and the English Channel, potentially due to the Ushant tidal front (Ayata et al. 2010; Muir et al. 2020; Nunes et al. 2021; David et al. 2022). The genetic analysis shown here indicates that present‐day gene flow may be stronger than that predicted by dispersal modelling alone. However, the STRUCTURE analysis does indicate the presence of two genetic clusters in this area (C2 and C3—possibly the ancestral gene pool in this region), with admixture being inferred in most individuals from these two populations. Collectively, these findings confirm that S. alveolata displays high levels of connectivity throughout its range, likely maintained by a combination of stepping‐stone dispersal, post‐glacial admixture, and human‐mediated transport. Nevertheless, unique genetic diversity is observed in both the Southern Edge and Biscay populations, reflecting the influence of localised oceanographic processes and demographic factors. By incorporating dense sampling and a comprehensive suite of SNP markers, this study refines our understanding of S. alveolata population structure and underscores the complexity of marine connectivity over large spatial scales. Local Adaptation and Environmental Drivers of Selection This study provides the first genetic evidence of adaptation for the honeycomb worm, S. alveolata . Seawater temperature emerged as the primary driver of adaptive variation, in agreement with previous findings that showed local adaptation in membrane lipid composition to maintain membrane fluidity under temperature stress (Muir et al. 2016). In addition to temperature (21.4%), tidal amplitude (13.74%), net primary productivity (13.73%), and seawater salinity (5.91%) were also associated with adaptive loci in the pRDA (Figure 4), each explaining a lower but significant proportion of the variance. These results suggest that while temperature is likely the dominant selective force, other environmental factors, particularly those linked to hydrodynamics and resource availability, also contribute to shaping adaptive variation in the honeycomb worm. Six loci were identified as being under diversifying selection, and sequence and structural homologies associated with these loci indicate possible molecular processes involved in physiological adaptation to the local environment. SNP1 shared sequence homology with “transcription initiation factor TFIID” and was strongly associated with temperature, further supporting the role of thermal selection in shaping adaptive variation in S. alveolata . General transcription factors that make part of the transcriptional machinery “PIC” are mobilised during heat shock, particularly TFIID and SAGA (Zanton and Pugh 2006). Furthermore, TFIID is known to bind to heat shock factors (Sandaltzopoulos and Becker 1998) and heat shock proteins (Emanuel and Gilmour 1993). Sequence variation in SNP1 is therefore likely related to thermal adaptation that may require changes in the general regulation of transcriptional activity of mRNA by RNA polymerase II. SNP2 shared sequence homology with “signalling mucin HKR1”, an osmosensor (De Nadal et al. 2007; Tatebayashi et al. 2007) and was inversely correlated to temperature and salinity. SNP2 further shared structural homology to calcium‐binding signalling molecules, which are sensing molecules involved in the activation of stress‐responsive intracellular pathways, controlling a broad range of physiological activities, such as cell growth, proliferation, migration and biomineralisation (Vogel et al. 2002). Sequence variation in SNP2 is possibly an adaptation for detecting environmental, possibly osmotic stress. Two SNPs were correlated with high net primary productivity: SNP4 (“uridine 5’‐monophosphate synthase”) and SNP6 (“venom metalloproteinase BumaMPs1”). UMP‐synthase is an enzyme involved in numerous biosynthesis pathways, particularly in the synthesis of nucleotides containing pyrimidine, while metalloproteases are enzymes that degrade or remodel proteins, and many are involved in the degradation of extracellular matrix (Cui et al. 2017). Correlation of these two SNPs with net primary productivity may indicate that they may provide advantages in breaking down organic matter and/or more efficiently rendering it available for biosynthesis of building blocks needed for cell metabolism, growth and maintenance. SNP 5 did not have any strong correlation with environmental variables studied here. However, it had sequence homology to “S‐adenosylmethionine: tRNA ribosyltransferase‐isomerase”, a tRNA modification enzyme. tRNA modification enhances the structural stability, decoding accuracy and functionality of tRNA required for cell translation processes under temperature, oxidative or nutrient stress (Huang and Hopper 2016; Edwards et al. 2020; Pollo‐Oliveira et al. 2022; Ohira and Suzuki 2024). Therefore, SNP5 may be advantageous for tRNA modifications possibly associated with various kinds of environmental stress, as an adaptation to maintain protein synthesis. Putative drivers of diversifying selection that examine physiological performance under environmental stress (i.e., metabolism or thermal performance curves) are needed to confirm if and how these sequence variants affect fitness in the honeycomb worm. Finally, approximately 60% of SNPs were identified as being under balancing selection. While this high proportion may partly reflect the sensitivity of outlier detection methods in populations with high gene flow, it nonetheless is an indication that these loci may represent an additional reservoir of evolutionary novelty, as suggested for natural populations of European hake (Pita et al., 2022). Balancing selection maintains multiple functional alleles for a given locus, that might otherwise be lost to directional selection or genetic drift. For an organism inhabiting a highly dynamic intertidal zone, this standing genetic variation may act as a vital ‘genomic buffer’. By preserving a wide array of adaptive pathways, these loci likely contribute to the species' resilience, providing the raw material necessary to persist through fluctuating environmental stressors and rapid habitat shifts. Future research utilising whole‐genome sequencing and functional annotation will be essential to validate these candidates and decipher the specific metabolic or physiological pathways that this balanced polymorphism maintains. Conclusion Genome‐wide SNP analysis indicates that present‐day populations of the honeycomb worm maintain strong connectivity across its range along the Atlantic coast of Europe. Gene flow occurs across large distances, possibly facilitated by human‐mediated dispersal. No population demonstrated reduced genetic diversity or signs of isolation. Despite the potential for disruptive gene flow, local adaptation was observed, a pattern also observed in other marine organisms such as Haliotis rufescens (red abalone, De Wit and Palumbi 2013) and Kelletia kelleti (Kellet's whelk, Lee et al. 2025). Whether gene flow hinders or contributes to diversification (Tigano and Friesen 2016) requires further investigation for the honeycomb worm. Six SNPs showed signs of selection and were primarily correlated with temperature, salinity and net primary productivity. Gene functions associated with these SNPs corroborate possible adaptations to these environmental drivers of selection. The results presented here indicate that the honeycomb worm is capable of local adaptation to environmental factors and is likely to be resilient to environmental change, highlighting its potential as a model species for studying putative climate change “winners”. Author Contributions Edward J. Wilson – formal analysis (lead); visualisation (lead); writing – original draft (lead); writing – review and editing (lead). Antony M. Knights – conceptualization (supporting); resources – sampling (equal); supervision (equal); writing – review and editing (equal). Louise B. Firth – conceptualization (supporting); funding acquisition (supporting); resources – sampling acquisition (equal); supervision (equal); writing – review and editing (equal). Charlotte Corporeau – formal analysis (supporting); writing – original draft (supporting); writing – review and editing (equal). Amelia Curd – resources – sampling (equal); writing – review and editing (equal). Stanislas F. Dubois – funding acquisition (lead); resources – sampling (equal); writing – review and editing (equal). Fernando P. Lima – resources – sampling (equal); writing – review and editing (equal). Rui Seabra – resources – sampling (equal); writing – review and editing (equal). Flavia L.D. Nunes – conceptualization (lead); data curation (lead); formal analysis (supporting); resources – sampling (equal); supervision (equal); writing – original draft (supporting); writing – review and editing (equal). Funding This work was supported by the Total Foundation (1512 215 588/F, 2015). EW received a mobility grant from the ISblue project, Interdisciplinary graduate school for the blue planet (ANR‐17‐EURE‐0015) and was co‐funded by a grant from the French government under the program “Investissements d'Avenir” embedded in France 2030. Ethics Statement This study was conducted as part of the master's thesis of Edward J. Wilson. Other benefits from this research include the sharing of our data and results on public databases. Conflicts of Interest The authors declare no conflicts of interest. Supporting information Figure S1: SNP statistics for the honeycomb worm 2bRAD dataset. Figure S2: Number of retained individuals assigned to each Illumina sequencing run, shown by population. Figure S3: (A) Depth of sequencing and (B) Missingness (%) per population. Figure S4: Isolation by distance. Figure S5: Metrics used for the selection of genetic clusters for STRUCTURE. Figure S6: Comparison of STRUCTURE plots for varying number of clusters (K = 2, K = 3 and K = 4). Figure S7: Predicted gene functions and peptide structure homology for SNPs under selection.