Genomic Footprints of Hybridisation in North Atlantic Eels (Anguilla anguilla and A. rostrata)

ABSTRACT

Understanding interspecific introgressive hybridisation and the biological significance of introgressed variation remains an important goal in population genomics. European ( Anguilla anguilla ) and American eel ( A. rostrata ) represent a remarkable case of hybridisation. Both are panmictic and spawn in partial sympatry in the Sargasso Sea, occasionally producing viable, fertile hybrids, primarily found in Iceland. We studied introgressive hybridisation from American into European eel using whole‐genome sequences of 78 individuals, including European, American and 21 putative hybrid eels. Previous studies using few genetic markers could not resolve whether hybridisation involved simple unidirectional backcrossing or a more complex hybrid swarm scenario. However, local ancestry inference along individual chromosomes revealed that Icelandic hybrids were primarily F1 or first‐generation backcrosses towards European eel, with some showing more complex backcrossing. All European eels outside Iceland contained short chromosomal blocks from American eel, indicating a porous genome. We found no evidence for previously hypothesised geographical gradients of introgression in European eel outside Iceland. Several chromosomal regions showed high interspecific divergence, but haplotype blocks introgressed from American eel were identified both within and outside these regions. There was little correspondence between regions of high relative (F _ST) and absolute divergence (d _XY ), with the former reflecting selective sweeps within species or reduced recombination rather than barrier loci. A single genomic region showed evidence of repeated introgression from American into European eel under positive selection in both species. The study illustrates that species can maintain genetic integrity despite porous genomes and that standing variation in one species can potentially be available for future adaptive responses in the other species.

1. Introduction

Speciation and hybridisation represent intermingled phenomena, with hybridisation occurring when reproductive barriers between species are not yet complete. Incomplete barriers leading to hybridisation and introgression sometimes are found even between otherwise well‐established species (Abbott et al. 2013; Edelman and Mallet 2021; Mallet 2005), underpinning the complex and often prolonged processes of speciation.

The genome of an admixed individual represents a mixture of alleles inherited from the parental species. In diploid organisms, first‐generation (F1) hybrids inherit exactly one chromosome of each chromosome pair from either parent. In subsequent generations, recombination causes the exchange of genetic material between chromosomes, which over time leads to an increasingly mosaic‐like structure of chromosomes consisting of multiple blocks of DNA originating from each parental species.

The novel combinations of parental genotypes exhibited by admixed individuals are subject to natural selection, and depending on their effect on fitness some alleles and chromosomal segments introgress more easily than others (Abbott et al. 2013; Edelman and Mallet 2021; Harrison and Larson 2014; Moran et al. 2021; Ravinet et al. 2017; Wu 2001). Particular allelic variants at different loci originating from each species may be fully or partially incompatible and cause decreased fitness or even inviability of the carrier, that is Bateson–Dobzhansky–Muller hybrid incompatibility (Dobzhansky 1937; Orr 1995, 1996). Also, alleles may be selected against in the environment experienced by the species that alleles are introgressed into (Moran et al. 2021; Nosil, Vines, and Funk 2005). Conversely, introgression may fuel adaptative processes. Hybridisation may in itself make new niches available to organisms and lead to speciation and adaptive radiation (Lamichhaney et al. 2018; Meier et al. 2017; Payseur and Rieseberg 2016; Rieseberg et al. 2003). Moreover, adaptively important variation may introgress even against a backdrop of genome‐wide purifying selection (Anderson et al. 2009; Jones et al. 2018; Liu et al. 2015). Hence, with ongoing hybridisation and introgression genomes are expected to be porous, with some regions resisting gene flow and other regions being more permeable (Coyne and Orr 2004; Parchman et al. 2013; Stolting et al. 2013; Wu 2001).

During the past few decades, the use of genome‐wide data for unravelling footprints of speciation has attracted considerable interest (Cruickshank and Hahn 2014; Feder, Egan, and Nosil 2012; Nosil, Funk, and Ortiz‐Barrientos 2009; Ravinet et al. 2017; Seehausen et al. 2014). Accordingly, genomic islands of divergence have often been interpreted in a speciation‐with‐gene‐flow framework (Feder, Egan, and Nosil 2012), although it has later become apparent that such genomic islands may reflect a variety of factors and processes other than genomic regions resistant to gene flow, such as selective sweeps, purifying selection and variation of recombination across the genome (Cruickshank and Hahn 2014; Ravinet et al. 2017). It would therefore be of considerable interest to compare genomic islands of divergence with the distribution of introgressed chromosomal segments in admixed individuals, as this would allow for testing if these islands coincide with regions exhibiting reduced introgression.

Anguillid eels represent remarkable cases of hybridisation and incomplete reproductive isolation between species pairs (Albert, Jónsson, and Bernatchez 2006; Avise et al. 1990; Barth et al. 2020; Nikolic et al. 2020). The two sister species of North Atlantic eels, European eel ( Anguilla anguilla ) and American eel ( A. rostrata ), represent the most studied example of this. They are catadromous, semelparous fishes that spawn in partial sympatry in the Sargasso Sea, a region of the Northern Atlantic Ocean bounded by ocean currents (Schmidt 1923; Tesch 2003). Eel larvae are transported by the Gulf Stream and other currents towards the European and American coasts respectively. Here they grow up in freshwater and coastal areas before returning to the Sargasso Sea to spawn. Both species are distributed across a wide environmental and climatic range, with European eel occurring from northern Norway to Morocco and American eel from southern Greenland to Venezuela and Colombia. Nevertheless, despite the heterogeneity of habitats both species are suggested to be panmictic (Als et al. 2011; Côté et al. 2013; Dannewitz et al. 2005; Enbody et al. 2021; Palm et al. 2009; Pujolar et al. 2014b; Ulmo‐Diaz et al. 2024).

European and American eels are able to hybridise, but hybrids have so far almost exclusively been found in Iceland, with findings of only a few later‐generation hybrids in continental Europe and North America (Albert, Jónsson, and Bernatchez 2006; Avise et al. 1990; Enbody et al. 2021; Gagnaire et al. 2009; Jacobsen et al. 2017; Pujolar et al. 2014a). Analysis of speciation history sheds further light on the presence of hybridisation. Initial analyses of divergence time based on mitochondrial DNA (mtDNA) yielded estimates ranging from 1.5 to 10 million years ago (mya) (Aoyama and Tsukamoto 1997; Avise et al. 1990; Jacobsen, Pujolar, Gilbert, et al. 2014; Minegishi et al. 2005). More recently, Nikolic et al. (2020) analysed genomic data (RAD sequencing and whole‐genome sequencing), and the results pointed to a complex speciation scenario. The species were found to have diverged between ca. 1.3 and 2.4 mya, presumably in allopatry. This was followed by a period of secondary contact, lasting from between 0.8 and 1.4 mya until the present. Analysis of demographic history of the two species using Stairway Plots (Liu and Fu 2020) further showed that these events were associated with abrupt changes in population size (Feng, Liu, and Hansen 2022). The results by Nikolic et al. (2020) also suggested that ca. 64% of the genome experiences restricted gene flow between species, which conversely suggests that a sizeable fraction of the genome is porous.

Whereas knowledge about hybridisation between the two species has advanced considerably, many questions remain unresolved. First, studies using allozyme and mtDNA (Avise et al. 1990) and amplified fragment length polymorphism (AFLP) markers (Albert, Jónsson, and Bernatchez 2006; Gagnaire et al. 2009) provided evidence of post‐F1 hybrids, raising the possibility of introgression in the long term. Pujolar et al. (2014a) analysed species‐diagnostic SNPs, defined as SNPs showing F _ST of > 0.95 between the two species. These markers allowed a clearer distinction between F1 and later‐generation hybrids, but also suggested surprising patterns of backcrosses including the presence of individuals with a hybridisation history of backcrossing to both species. This raises the question of whether (1) hybridisation between these species leads primarily to backcrossing in each of the species, or whether (2) hybrid offspring have an increased tendency to mate with either species across generations, the latter scenario resulting in a hybrid swarm (a population formed by extensive interbreeding between two species or genetically differentiated populations, resulting in a wide range of intermediate genotypes and phenotypes).

Second, the presence of hybrids almost exclusively in Iceland is striking, especially since spawning does not take place in Iceland but instead thousands of kilometres away in the Sargasso Sea. How does this relate to genetic differences between the species? Intrinsic barriers against gene flow between the two species undoubtedly exist, illustrated by recent evidence for demethylation and presumably reactivation of transposons in hybrids (Liu et al. 2022). However, there is also considerable evidence for the importance of ecological selection. The two species experience a similar range of environmental conditions and climate gradients on each side of the North Atlantic, but a major difference between them consists in the migration distance between the Sargasso Sea and the continents; > 1500 km for American eel and > 5000 km for European eel (Tesch 2003). This also translates into a shorter larval stage for American than for European eel (Lecomte‐Finiger 1992; Tesch 1998, 2003). Genome scans involving the two species show candidate regions under selection enriched with genes involved in energetics and development (Jacobsen, Pujolar, Bernatchez, et al. 2014; Pujolar, Jacobsen, and Bertolini 2022), and species‐specific co‐adaptation between the mitochondrial ATP6 gene and the nuclear ATP5c1 gene has also been suggested (Gagnaire, Normandeau, and Bernatchez 2012; Jacobsen et al. 2017). At the transcriptomic level, it has also been found that up or downregulation of genes occurs later in European eel as compared to American eel larvae, presumably reflecting differences in larval phase duration (Bernatchez et al. 2011). This provides further evidence that migration distance and larval phase duration represent major biological differences between these species, possibly under diversifying selection. These differences could also explain why hybrids, expected to exhibit intermediate larval phase duration relative to the parental species, end up in Iceland between the North American and European continents along the route represented by the Gulf Stream (Albert, Jónsson, and Bernatchez 2006; Avise et al. 1990; Jacobsen, Pujolar, Bernatchez, et al. 2014; Pujolar et al. 2014a). As a logical extension of this scenario, it raises the possibility that gradients of introgression also occur in each species outside Iceland, with for instance Northern European eels showing more introgression than eels in Southern Europe and North Africa (Albert, Jónsson, and Bernatchez 2006; Wielgoss et al. 2014). In other words, this would represent a scenario of environmental sorting of admixed genotypes, with Iceland being the site with most F1 hybrids and recent backcrosses, whereas the degree of admixture of individuals would gradually decrease on either side of Iceland along the Gulf stream. It would, however, require a high resolution of genome‐wide introgression to firmly test this hypothesis.

Third, the fate of introgressed chromosomal segments remains poorly known. Gagnaire et al. (2009) presented evidence for anonymous AFLP markers under positive selection in Icelandic eels following introgression from American eel. Local (i.e., spatially varying) single‐generation selection, including polygenic selection, has been suggested in both American and European eel (Babin et al. 2017; Gagnaire, Normandeau, Côté, et al. 2012; Pavey et al. 2015; Pujolar et al. 2014b; Williams, Koehn, and Mitton 1973), but the possible contribution of introgressed variation is not known and would be complicated to assess. At the level of entire species, however, identification of chromosomal segments resulting from introgression that are simultaneously under positive selection would provide evidence for ongoing adaptive introgression. Enbody et al. (2021) reported evidence for a ca. 0.2 Mb segment in Chromosome 15, possibly an inversion, introgressed from American into European eel and possibly under selection. However, specific evidence for positive selection remains unclear and it is not known if a single or multiple events of introgression have occurred.

We analysed whole‐genome sequences of 78 individuals encompassing European and American eels and hybrids between them. First, we made use of the power provided by whole‐genome sequencing data to clarify the hybrid status of Icelandic eels; is hybridisation exclusively a matter of initial hybridisation followed by generations of backcrossing with European eel? Or do Icelandic eels represent a more complex case of a hybrid swarm with backcrosses in both directions, as suggested by some results (Albert, Jónsson, and Bernatchez 2006)? Second, we aimed to test the hypothesis that genomes of Atlantic eels can be considered porous (Barth et al. 2020; Nikolic et al. 2020), in which case introgressed segments of American eel genome should be commonly observed in European eel. Third, we tested the hypothesis that, along the European and North African coasts, gradients of admixture occur coinciding with the transport distance of juvenile eels from the Sargasso Sea (Albert, Jónsson, and Bernatchez 2006; Wielgoss et al. 2014). Fourth, we investigated if adaptive introgression from American eel into European eel is ongoing, by searching for chromosomal segments with signatures of both introgression and positive selection. We also looked further into the dynamics of such adaptive introgression and investigated whether this was the case of a single or multiple hybridisation events. Conversely, we assessed if genomic islands of high divergence between the two species show a lower prevalence of introgressed haplotype blocks across individuals owing to purifying selection.

2. Materials and Methods

2.1. Sampling

We analysed a total of 50 European eels (of which one later turned out to be a hybrid), seven American eels and 21 individuals of admixed ancestry. We note that the admixed individuals do not represent a random selection of hybrids; 10 individuals had beforehand been identified as F1 hybrids and 11 as different types of backcrosses: bAA (first‐generation backcross to European eel, N = 3), bAAxAA (second‐generation backcross to European eel, N = 3) and bAR × AA (second‐generation backcross, with backcrossing first to American eel and subsequently to European eel, N = 5). This initial hybrid identification was based on 68 species‐diagnostic SNPs and analyses using Structure 2.3.4 (Falush, Stephens, and Pritchard 2003; Pritchard, Stephens, and Donnelly 2000) and NewHybrids 1.1 (Anderson and Thompson 2002) as detailed in Pujolar et al. (2014a). As demonstrated in this paper using both simulated and empirical data, there is high power for distinguishing ‘pure’ American and European eel from each other and from recently admixed individuals.

The analysed individuals were collected between 2001 and 2017 using electrofishing or net fishing and encompassed both juvenile glass eels and adult eels within the continental ranges of American and European eels (see Figure 1). Two eel larvae collected by ring net in the Sargasso Sea were also included. For individual sampling details, see Table S1. A total of 22 individuals, including 17 hybrids and five American eels, were sequenced for the present study. The remaining 56 individuals, encompassing primarily European eels, had previously been analysed using whole‐genome sequencing for the purpose of analysing speciation between European and American eel and genome‐wide methylation in European eel (Liu et al. 2022; Nikolic et al. 2020).

FIGURE 1.

Location of sampling sites. The geographic range (nonbreeding) of American eel (ASSG 2022) and European eel (AESG 2020) is shown in blue and orange respectively. The light grey area shows the approximate spatial extent of the Sargasso Sea, and the red arrows demonstrate how eel larvae are transported by the Gulf Stream.

2.2. Mapping and Variant Calling

Genomic DNA was extracted using a standard phenol‐chloroform extraction or E.Z.N.A. purification columns (Omega Bio‐Tek, Norcross, Georgia, USA). Whole‐genome sequencing was outsourced to BGI (Beijing Genomics Institute, Hongkong, China) (nine individuals) and NOVOGENE (Hong Kong, China) (all other individuals). Sequencing libraries were constructed using the NEBNext DNA Library Prep Kit (New England Biolabs, MA, USA). Genomic DNA was randomly fragmented to a size of 350 bp by shearing, and PCR amplification was conducted. Paired‐end Illumina sequencing was conducted using the Illumina HiSeq 2500 platform with a read length of 150 bp and aimed for a coverage of ~20×.

Using BWA MEM v.0.7.17 (Li 2013; Li and Durbin 2009) with default parameter settings, the reads were mapped to a recent chromosome level European eel genome assembly (Rhie et al. 2021) (GenBank accession: GCA_013347855.1). Mapped reads were converted to BAM files, sorted and indexed using SAMtools v.1.9 (Li et al. 2009). A VCF file of SNPs encompassing all 78 individuals was generated from the BAM files with BCFtools v.1.9 (Li 2011; Li et al. 2009) using a minimum mapping quality threshold of 20. Initial filtering of the SNPs was performed using VCFutils.pl. (Li et al. 2009) and VCFtools v.0.1.16 (Danecek et al. 2011). Only biallelic SNPs with minimum variant quality of 20, minor allele count ≥ 3, no missing data and pooled read depth between 900 and 1500 across all individuals were kept. The thresholds were determined based on inspection of the SNP depth distribution (Figure S1). SNPs on nonanchored scaffolds were discarded.

We furthermore produced an ‘all sites’ data set by using BCFtools call without the ‐‐variants‐only input option. Similarly to the ‘variant sites’ data set, we filtered the VCF to discard indels and retain only sites with minimum mapping quality of 20 and pooled read depth between 900 and 1500 across all individuals. No filtering for missing data and minor allele frequency was performed.

2.3. Population Structure and Genetic Diversity

Genetic relationships among the individuals were explored through a PCA using the ‘prcomp’ function in R v.3.5.1 (R Core Team 2020). The analysis was based on the ‘variant sites’ data set thinned using VCFtools, so that no two SNPs are within > 1000 bp from one another—a distance at which linkage disequilibrium (LD) is virtually absent in the species (Jacobsen, Pujolar, Bernatchez, et al. 2014).

Based on the ‘all sites’ data set, two measures of genetic diversity were calculated. Genome‐wide observed heterozygosity (HO) was calculated as the number of variant sites divided by the total number of genotyped sites for each individual. Nucleotide diversity (π) was calculated in windows of 10 kb for each species and the hybrids using PIXY 1.0 (Korunes and Samuk 2021), after which the raw counts were averaged across all windows to obtain genome‐wide estimates.

2.4. Hybrid Identification Using Species‐Diagnostic SNPs

Initial hybrid identification was conducted using species‐diagnostic SNPs developed by Pujolar et al. (2014a). These SNPs were identified using RAD sequencing data (Baird et al. 2008) aligned to the first draft reference genome assembly of European eel with N50 of 77.6 kb (Henkel et al. 2012), with a maximum of one SNP chosen per scaffold. As an uneven distribution of species‐diagnostic SNPs within the genome could induce biases in hybrid identification, we determined the position of these markers in the recent chromosome‐level assembly. We identified 100 bp on each side of each SNP in the scaffolds of the assembly by Henkel et al. (2012), and determined the position of the sequences in the assembly by Rhie et al. (2021) (GenBank accession: GCA_013347855.1) using the blastn tool at https://blast.ncbi.nlm.nih.gov.

We subsequently identified a new set of species‐diagnostic SNPs based on the chromosome‐level assembly. This was done by searching for SNPs showing fixation between species (F _ST = 1 in VCFtools) and choosing one SNP for each 10 Mb along the chromosomes. Based on this new set of markers, the reclassification of the 22 previously identified hybrids was conducted using NewHybrids 1.1 (Anderson and Thompson 2002) with 100,000 iterations for burn‐in followed by 500,000 Markov Chain Monte Carlo iterations. The genotype frequency class file was expanded to include third‐generation hybrid classes, as detailed in Pujolar et al. (2014a).

2.5. Statistical Phasing

For the purpose of inferring local ancestry along chromosomes and scanning for selective sweeps, we created a phased data set for the reference populations. The previously produced VCF file was split between hybrids and nonhybrids. The former file was not phased, as statistical phasing introduced a large number of switch errors (results not shown, but see e.g., Smeds et al. (2021)). Statistical phasing was performed using SHAPEIT v.2 (r900) (Delaneau, Zagury, and Marchini 2013). To increase accuracy, we increased the number of conditioning states from 100 states per SNP (default) to 200 states. Each chromosome was phased individually, and no recombination map was specified. The American and European eels were phased simultaneously and the phased VCF file was subsequently subset by species.

2.6. Admixture Analyses and Local Ancestry Inference

We used PCAdmix v.1.0 (Brisbin et al. 2012) to infer local ancestry along chromosomes for the recently admixed individuals. PCAdmix assigns ancestry to windows of consecutive, nonoverlapping SNPs by PCA by projecting them onto the PC loadings of individuals from two or more ancestral reference populations. The unphased VCF file containing all recently admixed individuals (n = 22) was input as the admixed file and the phased VCF files containing all American eels (n = 7) and all European eels (n = 49), respectively, were input as ancestral files. We specified a fixed window size of 0.01 Mb (‐wMb 0.01), a MAF filter of 0.05 for the ancestral reference populations, and that no LD pruning should take place (‐prune 0). Each chromosome was analysed independently. Based on the PCAdmix output, we calculated the number of switches occurring between ancestry blocks of inferred European and American ancestry (henceforth denoted ‘ancestry switches’) and the genome‐wide proportion of windows assigned to each reference population for each admixed individual.

Using the same parameter settings as described above, we also inferred local ancestry for the putatively pure European eels to detect any chromosomal elements introgressed from American eel. This was done by sequentially removing all individuals from one sampling location from the VCF file containing pure European eels to use them as the admixed population in PCAdmix, while using all the American eels and the remaining European eels as the ancestral populations. As the performance of PCAdmix is highly sensitive to the choice of window size, we discarded any inferences of American ancestry for windows spanning less than 20 SNPs following recommendations in the PCAdmix manual.

2.7. Islands of Divergence

Differentiation among loci between closely related species or populations has frequently been interpreted as genomic islands resistant to gene flow. However, relative measures of divergence are biased towards regions of low recombination. Hence, regions of reduced diversity may be mistaken for regions of reduced gene flow when inferences are based solely on F _ST (Cruickshank and Hahn 2014). On that account, we calculated both F _ST (Weir and Cockerham 1984) and d _XY (Nei 1987), a measure of absolute divergence, in sliding windows of 10,000 bp containing at least (‐m) 20 SNPs and moving at steps (‐s) of 1000 bp using popgenWindows.py (https://github.com/simonhmartin/genomics_general). The calculations were based on the ‘all sites’ data set. Windows with values exceeding defined thresholds (0.1 for F _ST and 0.02 for D _XY ) were identified, and the top 5% of the genome‐wide distribution of these extreme values was retained. Windows identified by both methods were considered candidate islands of divergence resistant to gene flow between populations and were examined for genes.

2.8. Selection Scans

We searched for selective sweeps using XP‐EHH (cross‐population extended haplotype homozygosity) (Sabeti et al. 2007), XP‐nSL (cross‐population number of segregating sites by length) (Szpiech et al. 2021) and XP‐CLR (cross‐population composite likelihood ratio) (Chen, Patterson, and Reich 2010), in addition to the previously described F _ST scan. These tests capture different genetic patterns suggestive of positive selection. Specifically, XP‐EHH and XP‐nSL detect alleles that have increased in frequency faster than expected from random genetic drift, manifesting as unusually long‐range LD, while XP‐CLR utilises measures of multi‐locus genetic differentiation between populations, assuming that differentiation around a variant under selection will be larger than from genetic drift. For an in‐depth introduction to these methods, as well as those introduced below, see Vatsiou, Bazin, and Gaggiotti (2016). As cross‐population methods do not allow for the detection of alleles putatively under selection in both populations in a pair, we additionally utilised single‐population selection scans iHS (integrated haplotype score) (Voight et al. 2006) and nSL (number of segregating sites by length) (Ferrer‐Admetlla et al. 2014); the counterparts to XP‐EHH and XP‐nSL respectively. However, as these methods that compare between alleles in the same population have low predictive power with small sample sizes (Pickrell et al. 2009), these analyses were only performed for the European eels (N = 49) and not for American eels (N = 7).

XP‐EHH, XP‐nSL, iHS and nSL analyses were performed using selscan v.2.0 (Szpiech 2021; Szpiech and Hernandez 2014) with default parameter settings. For XP‐EHH and iHS, calculations were based on physical distances instead of genetic distances (‐‐pmap). The statistics were normalised across the genome using norm v.1.3.0 implemented in selscan. All statistics were summarised in sliding windows of 10,000 bp containing at least 10 SNPs and moving at steps of 1000 bp. For XP‐EHH and XP‐nSL, the summarisation was performed separately for positive scores, suggesting adaptation in the focal population, and negative scores, suggesting adaptation in the reference population. XP‐CLR (https://github.com/hardingnj/xpclr) was calculated using the default recombination rate of 1·10⁻⁸ per bp, scanning the genome in sliding windows of (‐‐size) 25,000 bp containing at least 10 SNPs and moving at steps of (‐‐step) 10,000 bp. Since selection is only detected in the object population with XP‐CLR, the analysis was performed twice, alternating which population constituted the objective and the reference population respectively. XP‐CLR scores were normalised by default. All selection scans were based on the previously produced phased data set.

Genome scans were visualised using the ‘manhattanplot’ function in the R package ‘rehh’ (Gautier, Klassmann, and Vitalis 2017). For select sites, extended haplotype homozygosity (EHH) and underlying haplotype bifurcation patterns—constituting the basis of iHS and XP‐EHH statistics—were visualised using functions ‘plot.ehh’ and ‘plot.furcation’.

Windows with values exceeding a defined threshold of 2 were identified, and the top 1% of the genome‐wide distribution of these extreme values was retained for each statistic. Calculation and identification of regions of high F _ST proceeded as described in the preceding section on islands of divergence. To maintain consistency despite having performed more selection scans for European eels than American eels, candidate regions under selection were identified separately for cross‐population and single‐population statistics. For the cross‐population statistics (XP‐EHH, XP‐nSL, XP‐CLR and F _ST) computed for both species, only windows that were identified by at least two out of four methods were kept as candidate regions under selection. For the single‐population statistics (iHS and nSL) computed only for European eels, only windows that were identified by both methods were likewise retained as candidate regions. Overlapping regions were merged to produce a single list of candidate regions under selection for each species to be interrogated for genes.

2.9. GO Enrichment Analysis

Any genes overlapping, partially or wholly, with candidate regions were considered candidate genes under selection. To identify the biological significance of the set of genes putatively under selection in each population, we performed a gene ontology (GO) enrichment analysis for the candidate genes under selection in each population using the R package ‘topGO’ (Alexa and Rahnenfuhrer 2016). We used the ‘fisher’ statistics with the ‘weight01’ algorithm for statistical tests and reported unadjusted p‐values for significantly enriched GO terms. Subsequently, we used GO‐Figure! (Reijnders and Waterhouse 2021) to simplify and reduce the redundancy of significant GO terms by grouping terms based on their semantic similarity. The ‘topGO’ output files were used as input (‐‐topgo), and analyses were run with default parameters.

3. Results

3.1. Whole‐Genome Sequencing

Whole‐genome sequences showed sequencing depths ranging between 8.80× and 29.10× (see Table S2 for summary statistics). A total of 25,849,491 SNPs distributed on 19 chromosomes remained after filtering, and the thinned ‘variant sites’ data set used for the PCA contained 835,600 SNPs. The ‘all sites’ data set contained 685,428,200 sites (70% of the genome) with individuals exhibiting between 0.2% and 0.9% missing genotypes.

3.2. Population Structure and Genetic Diversity

The PCA (Figure 2) clearly separated the pure European and American individuals into two clusters and confirmed the admixed status of the 21 known hybrids, all of which appeared between the two species clusters. One additional individual (Bur_05) which had not previously been analysed for hybridisation, also appeared as admixed and was therefore treated as such in the following analyses, increasing the number of hybrids from 21 to 22 individuals.

FIGURE 2.

Principal component analysis of genetic differentiation among 78 individuals. American eels and European eels are represented by blue circles and orange squares, respectively, and hybrids are shown as grey triangles. One individual appearing as admixed between species, but which was not previously tested for hybridisation, is marked by an asterisk.

Observed nucleotide diversity ($\pi$) was highest for the admixed individuals (0.0120), and higher for American eels (0.0112) than for European eels (0.0105).

3.3. Hybrid Identification Using Species‐Diagnostic SNPs

Upon identifying the positions of the 68 SNPs developed by Pujolar et al. (2014a) in the recent chromosome‐level genome assembly, it turned out that their distribution was highly uneven, with 47 found in Chromosome 7 and 9 in Chromosome 2 (see Table S4). Whereas the markers should still be able to identify pure individuals, F1 hybrids (heterozygous for all loci) and backcrosses (homozygous for species‐specific alleles at some loci), this is likely to bias identification of the specific type of backcross.

NewHybrids analysis using the new set of species‐diagnostic SNPs derived from the chromosome‐level assembly yielded results that, in several cases, contrasted with hybrid classification based on the Pujolar et al. (2014a) SNPs (Table S3). Among the 22 hybrids, 10 were clearly identified as F1 hybrids, 9 as bAA and a single individual (Ell_09) was suggested to be bARxAA. The remaining two individuals (Vog_04 and Vog_05) had the highest probability of being bAA, but also had a high probability of being bARxAA.

3.4. Whole‐Genome Admixture Analyses and Local Ancestry Inference

Local ancestry of individuals was inferred for the 22 admixed individuals using PCAdmix. Initially, we observed between 1131 and 1829 ancestry switches across the genome of admixed individuals. The number of ancestry switches generally increases with the number of generations since admixture due to recombination. However, the majority of the observed switches produced very short ancestry blocks dispersed among otherwise very long blocks (see Figure S2 for an example of this). This pattern is biologically implausible, as it suggests an unusually high number of recombination events occurring in close proximity. Such short ancestry blocks are more likely attributable to ‘switch errors’, in this case not due to faulty phasing but because the genotypes of the admixed individuals were not phased. Alternatively, it is possible that some of them could represent remains of introgression occurring many generations ago (see below). We manually corrected for putative switch errors by reversing the inferred pattern in any case where an assigned ancestry block spanned ≤ 250 windows of size 0.01 Mb, thereby reducing the number of switches to between 0 and 21 per individual. Conforming with the nature of the inheritance process, the 10 individuals previously identified as F1 hybrids (Ell_06, Ell_08, Ell_10, Ste_01, Sto_03, Vog_01, Vog_07, Vog_09, Vog_10 and Vog_11) exhibited one chromosome originating from either parental species in every chromosome pair (Figure 3; Figure S3) and a genome‐wide European ancestry proportion of 0.5 following the correction. For the remaining individuals that exhibited more mosaic‐like haplotypes, the genome‐wide European ancestry proportion ranged from 0.619 to 0.796 and the number of ancestry switches from 10 to 21 (Table S3). The distribution of ancestry switches and individual ancestry proportion values suggests that most of these hybrid individuals are likely first‐generation backcrosses to European eel (bAA) for which the expectation is a genome‐wide European ancestry proportion of 0.75 and ~17 ancestry switches (assuming a genome length of 1709 cM as inferred for the related Japanese eel ( A. japonica ) (Nomura et al. 2018)). While recombination is a stochastic process and some variation is expected among hybrid individuals of the same type, three individuals were particularly deviant: The individual Vog_02 exhibited a much higher European ancestry proportion and the highest number of ancestry switches out of all, which raises the possibility that it could be a second‐generation backcross to European eel. On the other hand, the individual Ell_09 exhibited the lowest ancestry proportion of all at 0.619 and Vog_04 exhibited a lower European ancestry proportion as well as much fewer ancestry switches than expected for a bAA, suggesting that the origin of these individuals might be more complicated than simple backcrossing to one species.

FIGURE 3.

Local ancestry inferred with PCAdmix along Chromosome 1 (88.06 Mb) for 12 recently admixed eels and one first‐generation (F1) hybrid, Ste_01, plotted for reference. Each line represents a diploid individual. The x‐axis shows position in Mb. The colour scheme indicates ancestry assignment of chromosomal blocks to either the European reference population (orange) or the American reference population (blue).

3.5. Local Ancestry Inference in Putatively Pure European Eel

Local ancestry inference showed that every putatively pure European individual, regardless of sampling location, carried chromosomal blocks originating from American eel (Figure S4). Genome‐wide American ancestry proportion ranged between 0.000227 and 0.00103 for the 49 eels. An analysis of variance showed no significant difference in mean ancestry proportion between the seven European and North African sampling sites ($F\left(6,41\right)=0.273,p=0.946$). Hence, there was no geographical gradient of introgression; on average, eels from different sampling locations in Europe and North Africa exhibited similar levels of introgressed American chromosomal segments in their genome.

A total of 3.3% of all analysed windows included at least one haplotype of American ancestry. A single region exhibiting a much‐elevated rate of introgression was identified on chromosome 15; the region stretches from 11.95 to 12.13 Mb, in which 11 haplotypes out of 98 were of inferred American ancestry (Figure 4). This genomic region overlaps with the putative inversion reported by Enbody et al. (2021).

FIGURE 4.

Local ancestry inferred with PCAdmix along Chromosome 15 (37.22 Mb) for 49 European eels. Each line represents a diploid individual. The x‐axis shows position in Mb. The colour scheme indicates ancestry assignment of chromosomal blocks to either the European reference population (orange) or the American reference population (blue).

3.6. Islands of Divergence

The genome‐wide F _ST scan (Figure S5) revealed 878 regions of high relative genetic differentiation between the species with a total length of 27.3 Mb. Concordant with the results of Pujolar, Jacobsen, and Bertolini (2022), we observed large regions or islands of divergence on several chromosomes. The most pronounced regions were located on chromosome 6 (3.4 Mb, primarily in region 29.6–37.0 Mb), chromosome 7 (4.6 Mb, primarily in region 27.8–36.2 Mb) and chromosome 10 (4.4 Mb, primarily in region 22.9–31.3 Mb) (Figure 5). A total of 3.5% of windows analysed with PCAdmix were located in these highly divergent regions. Within these windows, the American ancestry proportion for the European eels ranged between 0 and 0.0074. We found no significant difference between the mean American ancestry proportion within the divergent regions and the mean genome‐wide American ancestry proportion ($t\left(48\right)=2.008,p=0.976$). Thus, the regions of high relative genetic divergence appeared not to be resistant to gene flow.

FIGURE 5.

Examples of the correlation between (a) F _ST, (b) d _XY and (c) the cross‐population selection scan XP‐EHH for American and European eel. The three chromosomes with the largest, most pronounced regions of high F _ST are plotted. The x‐axes show chromosomes, and, for XP‐EHH, the y‐axis indicates the direction of selection.

We observed that absolute genetic divergence, measured as d _XY (Figure S5), was not elevated in regions of high relative divergence. Rather, absolute divergence was generally slightly lower in the regions of high relative divergence than in the surrounding regions (Figure 5). Comparing identified regions of high relative and high absolute divergence, we found only 16 common regions, totalling 0.18 Mb (Table S6). Of those 16 regions, likely representing regions of reduced gene flow, a total of eight overlapped with at least one gene (Table S7).

3.7. Signatures of Selection and Introgression

Based on an array of selection scans (Figures S5 and S6), we identified 297 candidate regions under selection, spanning a total of 5.3 Mb and overlapping with 339 genes, in European eel (Tables S8 and S9) and 183 candidate regions, spanning 3.1 Mb and overlapping with 239 genes, in American eel (Tables S10 and S11).

The GO enrichment analysis showed 99 and 67 significantly enriched GO terms for European eel (Table S12) and American eel (Table S13) respectively. Grouping semantically similar GO terms revealed that most enriched GO terms in American eel described a diverse array of biological processes including development and cell differentiation, transcription, metabolism and ion homeostasis, and in particular regulation thereof (Figure 6a, Figure S7). Similarly, a number of enriched GO terms in European eel detailed regulation of, for example, transcription and translation, and metabolism. By far the most enriched GO terms in European eel, however, were related to embryonic development (Figure 6b, Figure S8). A single GO term was significantly enriched within both species; calcitonin receptor activity (GO:0004948), which influences bone formation and metabolic processes in fish (Nag et al. 2007).

FIGURE 6.

Semantic similarity scatterplot of enriched gene ontology (GO) terms related to biological processes for (a) American eel and (b) European eel. Circles represent clusters of semantically similar GO terms. The 50 most significant clusters are plotted, and the 20 most significant clusters are labelled by their most significant GO term. Circle colour shows the significance (log(p)) obtained from the enrichment analyses, and size represents the number of GO terms in a cluster.

Among the candidate regions showing the strongest signatures of selection in American eel, two were located within the previously identified region (chromosome 15:11.95–12.13 Mb) where introgression from American eel into European eel occurred at a high rate. Upon closer inspection, two of the cross‐population selection scans, XP‐EHH and XP‐nSL, exhibited large peaks of positive value within the region (Figure 7a). This signifies that the haplotypes in American eel were longer than haplotypes in the same region for European eel and is a sign of recent positive selection in American eel. Within the same region, the two single‐population selection scan statistics calculated for European eel, iHS and nSL, were also elevated (Figure 7b). This suggests recent positive selection occurring in European eel as well and indicates that, for a given biallelic site, the haplotypes carrying one allele are longer than haplotypes carrying the other allele. The decay of extended haplotype homozygosity (EHH) for the two positions with the most extreme values within the region is visualised in Figure 7c,d. This statistic is the basis of iHS and XP‐EHH selection scans and is defined as the probability of two randomly chosen chromosomes carrying an allele of interest being homozygous at a given position (Sabeti et al. 2002). At both extreme positions, American eel was fixed for one allele (denoted the American allele), while two alleles (the American allele and a second allele denoted the European allele) were segregating in European eel. EHH decayed rapidly on chromosomes carrying the European allele and, in both cases, reached 0 at a distance of less than 1 kb on each side of the focal marker. On the other hand, on chromosomes carrying the American allele, EHH only decayed below 0.05 upon reaching a distance of up to 17 kb on each side of the two markers (Figure 7c,d). In summary, these results demonstrated that this chromosomal segment, introgressed from American eel to European eel, is under positive selection in both species. Furthermore, the haplotype bifurcation patterns suggest that introgression into European eel has occurred on multiple occasion. Providing more detail than the EHH values, reflecting how many haplotypes are identical at increasing distances from a focal marker, the bifurcation plots (Figure 7e; Figure S9) directly capture the breakdown of haplotype homozygosity with branching points or bifurcations occurring when haplotypes diverge from each other. The European eel chromosomes did not constitute a single branch of homozygous haplotypes for either of the focal markers investigated, as would otherwise be expected if they had originated from the same introgression event. Rather, some European eel chromosomes were highly similar to certain American eel chromosomes, while other European eel chromosomes were more similar to other American eel chromosomes, suggesting multiple introgression events in this region. The introgressed region overlapped with three genes (Table S14), all of which were also identified as candidate genes under selection.

FIGURE 7.

Signatures of recent positive selection on Chromosome 15. (a) Cross‐population selection scans, XP‐EHH and XP‐nSL, indicating positive selection acting on American eel and (b) single‐population selection scans, iHS and nSL, indicating positive selection acting on European eel. The shaded areas show the boundaries of a genomic region introgressed from American eel into European eel as suggested by PCAdmix analyses. Decay of extended haplotype homozygosity around the focal positions (c) Chr_15:11987480 and (d) Chr_15:12097370, associated with the most extreme values in the introgressed region. (e) Haplotype furcation around the marker Chr_15:11987480 in the introgressed region. The thickness of the line corresponds to the number of chromosomes sharing an extended (identical) haplotype, which translates directly into the EHH values shown in panel (c), and the branching points (bifurcations) occur when haplotypes diverge at greater distances from the focal marker. American eels are labelled in light blue and European eels in orange. Only the haplotypes carrying the American allele, for which the American population is fixed, are shown.

4. Discussion

Our results based on whole‐genome sequencing provided significant new knowledge about patterns and dynamics of hybridisation between European and American eel. First, and in contrast to previous studies based on relatively few markers (Albert, Jónsson, and Bernatchez 2006; Gagnaire et al. 2009; Pujolar et al. 2014a), we found that the hybrids analysed from Iceland were primarily either F1 hybrids or first‐generation backcrosses in the direction of European eel. Second, we found that all European eels outside Iceland contained short chromosomal blocks from American eel, thus providing direct evidence for a porous genome. However, we found no evidence for geographical gradients of introgression in European eel outside Iceland as otherwise hypothesised previously (Albert, Jónsson, and Bernatchez 2006; Wielgoss et al. 2014). Third, we identified several chromosomal regions of high divergence between the two species, but haplotype blocks introgressed from American eel were identified both within and outside these regions. Finally, we identified a single genomic region with evidence for being introgressed from American into European eel and at the same time being under positive selection at the species level. This corresponds to the same segment in Chromosome 15 as reported by Enbody et al. (2021). We found that this was not due to a single incidence of introgression, but that multiple independent introgression events had occurred. We discuss these findings in more detail in the following.

4.1. Classification of Hybrids From Iceland

Important discrepancies were observed between hybrid classification based on a previously developed set of diagnostic markers (Pujolar et al. 2014a) on the one hand, and genome‐wide ancestry proportions and a new set of diagnostic markers based on a chromosome level genome assembly on the other hand. For instance, Ell_07, Sto_07 and Vog_08 had been categorised as resulting from F1 hybrids that backcrossed first to American eel and in the next generation to European eel (bARxAA). However, the expected admixture proportion of European eel for this type of cross (0.625) was considerably lower than the inferred genome‐wide ancestry proportions (0.730, 0.766 and 0.748 respectively) using PCAdmix, which rather indicate first‐generation backcrosses (bAA). This re‐classification was further supported by results from the new set of markers, which also identified these individuals as first‐generation backcrosses.

We discovered that the diagnostic markers by Pujolar et al. (2014a) showed an uneven genomic representation with the majority of SNPs being situated on Chromosomes 2 and 7 in regions showing high divergence between the species, likely to bias identification of specific classes of backcrosses. Hence, depending on specific recombination in these genomic regions, a disproportionally high number of SNPs could be either heterozygous or homozygous for European alleles. In the former case, this could cause a first‐generation backcross to European eel (bAA; expected American eel admixture proportion of 25%) to show a much higher American eel admixture proportion leading to erroneous classification as a backcross to first American and subsequently European eel (bARxAA; expected American eel admixture proportion of 37.5%). In the latter case, this could lead to erroneous classification of bAA individuals as second‐generation backcrosses (bAAxAA). Indeed, this appeared to be the case for several individuals in this study, of course under the assumption that genome‐wide ancestry proportions and the new set of diagnostic markers resulted in correct hybrid classifications. These results raise caveats when selecting diagnostic markers based on highly fragmented genome assemblies; even when care is taken to select markers from different scaffolds, they might represent only a few high‐divergence regions.

We note that the individuals analysed do not represent a random selection of hybrids from Iceland. The analysed post‐F1 individuals were, however, selected a priori to encompass different types of backcrosses. Nevertheless, they turned out to represent primarily first‐generation backcrosses: Both the results obtained by PCAdmix and the results from NewHybrids using the new diagnostic markers suggested two main groups of hybrids, corresponding to F1 hybrids and bAA backcrosses. One individual, however, showed a high probability of being bAR×AA (using NewHybrids) and exhibited a low European eel genome‐wide ancestry proportion, while for two additional individuals, bAR×AA status could not be entirely ruled out. Despite the uncertainty regarding the latter individuals, the resolution provided by whole‐genome data suggests a simpler scenario than previous studies reporting complex backcrosses to both species and even F2 hybrids (crosses between two F1 individuals) (Albert, Jónsson, and Bernatchez 2006; Gagnaire et al. 2009; Pujolar et al. 2014a). Most post‐F1 hybrids are simple backcrosses in the direction of European eel, and although a few cases of more complex backcrossing between both species exist this can hardly be considered a hybrid swarm.

4.2. Geographical Gradients of Introgression

Analysis of the putatively nonhybrid European eels (i.e., not being of recent hybrid ancestry) showed that all individuals contained short haplotype blocks introgressed from American eel, in accordance with assumptions of porous genomes. The short lengths of these blocks owe to several generations of recombination and in principle, some could represent ancient hybridisation, similar to, for example, reminiscences of Neanderthals and Denisovans in modern human genomes (Racimo et al. 2015). Nikolic et al. (2020) analysed site frequency spectra (SFS) of the species and estimated that ca. 64% of the genome experiences restricted gene flow and conversely 36% is permeable. Our finding, based on nonhybrid eels, of ca. 3.3% of the European eel genome containing at least one American haplotype block is considerably lower than this, but would be expected to increase if more individuals were analysed. In this context, it should also be noted that PCAdmix (Brisbin et al. 2012) has the highest power to detect introgression in regions where species are highly divergent. If introgression occurs frequently, regions are expected to become less divergent, meaning that introgression is less likely to be detected.

Whereas European × American eel hybrids are common in Iceland, only a few later‐generation backcrosses have previously been identified in continental regions of both Europe and North America (Jacobsen, Pujolar, Bernatchez, et al. 2014; Jacobsen et al. 2017; Pujolar et al. 2014b), including a first‐generation backcross from Ireland in the present study. Furthermore, when considering the nonhybrid eels, there was no evidence for geographical clines in admixture, hence not providing support for the hypothesis by Wielgoss et al. (2014). Even nonhybrid eels from Iceland showed similar low levels of introgression as individuals from other geographical regions.

How then should we view the system of hybridisation between European and American eels? First, it is important to consider that the life history of Atlantic eels is opposite that of other well‐known systems such as salmonid fishes (Avise 2011), with spawning in panmixia in the Sargasso Sea and subsequent dispersal to continental regions. Iceland should therefore not be considered a hybrid zone between the species, but rather a locality where recently admixed individuals showing intermediate phenotypes have a higher tendency of occurring. If a geographically defined hybrid zone exists between the species, it should be present in the spawning regions in the Sargasso Sea. Interestingly, however, analysis of eel larvae sampled in both 2007 and 2014 revealed only a few admixed individuals showing no obvious geographical patterns in their occurrence (Als et al. 2011; Jacobsen et al. 2017), even though the 2014 samples encompassed the entire known spawning region of European eel (Miller et al. 2019). It therefore remains unknown where or under which circumstances hybridisation occurs. However, when considered at the species level, which is the unit to consider in these panmictic species, a clear picture emerges with most hybrids being F1 or recent backcrosses. This adds to the general pattern observed in several other species, where strong purifying selection occurs during the first generations followed by more relaxed selection in later generations (Moran et al. 2021), the latter reflected in our study by the findings of short introgressed chromosomal blocks in all nonhybrid eels.

4.3. Introgression and Differential Selection Between Species

It is well established that the rate of introgression varies between loci, with some loci subject to negative directional or purifying selection and other being favoured by positive directional selection (adaptive introgression) (Harrison and Larson 2014). F _ST scans in the present study revealed numerous highly divergent genomic regions between the species, but there was no significant tendency for these regions to show reduced introgression of chromosomal segments from American eel into European eel genomes. This could reflect relaxed or slow purifying selection in later generations after initial hybridisation, as noted above. However, as there were only few cases of concomitant high relative (F _ST) and absolute (d _XY ) divergence, it is also possible that these regions primarily reflect reduced recombination and/or selective sweeps within species rather than resistance to gene flow (Cruickshank and Hahn 2014).

Our combined results from different types of selection scans revealed hundreds of genomic regions under possible selection in the two species, but presumably reflecting selective sweeps within species rather than differential introgression. Similar to previous GO‐term analyses of genes under putative selection between the two species (Jacobsen, Pujolar, Gilbert, et al. 2014; Pujolar, Jacobsen, and Bertolini 2022), we found overrepresentation of terms related to broad categories of energy conversion, metabolism and development. However, although these functional categories align well with major biological differences between the species (Jacobsen, Pujolar, Bernatchez, et al. 2014), we ascribe selection to have primarily occurred within species rather than representing direct selection in these genomic regions in admixed individuals.

4.4. Adaptive Introgression

Introgressed genomic segments of inferred American ancestry into European eel occurred at a very high frequency in a ~ 180 kb region on chromosome 15. This coincided with a putative recent selective sweep detected in both American and European eels suggesting that the introgressed chromosomal segment confers a selective advantage in both populations. Our findings support those of Enbody et al. (2021), who reported the existence of a putative inversion on chromosome 15, at 11.95–12.15 Mb, which initially underwent a selective sweep in American eel and since introgressed into European eel. The introgressed segment detected in the present study is shorter and does not overlap with the gene LOC118213596 (G patch domain‐containing protein 8‐like) which Enbody et al. (2021) hypothesised was associated with a fitness effect due to a large number of amino acid changes. Rather, it overlaps with the two genes LOC118214452 (FERM) and LOC118214186 (dedicator of cytokinesis protein 9‐like), of which only the latter appeared to be under selection in the two species in the present study. It will require more specific identification of the sites under selection, including possible co‐adaptation, and understanding of the function of the associated genes to assess the actual traits under selection and the associated mechanisms of selection.

The bifurcation tree analysis provided additional insights into the introgression process. Hence, it was evident that introgression from American into European eel has occurred in several independent instances. This illustrates the continuous process of gene flow between the species and shows that they are not completely evolutionary independent; at least in the longer term, they may share a common source of standing variation. Secondary contact and introgression are known to have supplied both adaptive and deleterious variation in a number of instances, most famously documented in hybridisation between ancient hominids having consequences for adaptation and health in contemporary populations (Huerta‐Sánchez et al. 2014; Reilly et al. 2022). Other examples include adaptive introgression from domestic dogs into wolves ( Canis lupus ) (Anderson et al. 2009) and from domestic goats ( Capra aegagrus hircus) into Alpine ibex ( Capra ibex ibex) (Grossen et al. 2014). In the case of Atlantic eels, panmixia combined with historically very high effective population sizes has led to high levels of variation (Feng, Liu, and Hansen 2022; Nikolic et al. 2020). However, both species are under strong anthropogenic pressure and endangered (Jacoby et al. 2015), and the possibility of drawing on a combined pool of standing variation could be important for future adaptive responses.

Given the evidence for hybridisation and introgression in Atlantic eels, it may in fact seem surprising that not more instances of adaptive introgression were uncovered. Detection of the region in chromosome 15 was facilitated by the fact that it presumably represents an inversion (Enbody et al. 2021), its genomic imprint thus being highly visible. It is well established that inversions can harbour complex co‐adapted variation protected from recombination (Wellenreuther and Bernatchez 2018), and for instance, in Atlantic cod ( Gadus morhua ) inversions show a strong association with migratory ecotypes (Sodeland et al. 2016). Other instances of adaptive introgression not representing inversions could have occurred but remained undetected. In this regard, it should also be considered that panmixia in both species rules out genetically based local adaptation (Als et al. 2011; Côté et al. 2013; Enbody et al. 2021; Pujolar et al. 2014b; Ulmo‐Diaz et al. 2024). This does not mean, however, that selection does not occur at all, although it is assumed to occur within single generations and represents complex temporally and spatially fluctuating dynamics (Babin et al. 2017; Gagnaire, Normandeau, Côté, et al. 2012; Pujolar et al. 2014b). It is likely that introgressed variation could be involved in such selection, and it would be interesting albeit challenging to track the possible introgressed origin of the involved alleles.

5. Conclusions

We made use of a genome‐level approach based on inferred local ancestry to corroborate recent hybrid categories of Atlantic eels. This allowed us to clarify and de‐mystify some aspects of hybridisation between the two eel species. Hence, we demonstrated that the occurrence of hybrids in Iceland is primarily a result of simple patterns of backcrossing towards European eel rather than representing a hybrid swarm scenario, although a few cases of repeated backcrossing between both species were suggested. We also found no support for a previously stated hypothesis of a gradient of introgression across the distributional range of European eel (Albert, Jónsson, and Bernatchez 2006; Wielgoss et al. 2014). Instead, Iceland stands out as a hotspot for the occurrence of hybrids, with few individuals of recent hybrid origin found outside this region.

We also made the noteworthy finding that all supposedly ‘pure’ European eels contained short genomic segments introgressed from American eel, demonstrating that their genomes are indeed porous. We did not find any obvious correspondence between genomic regions showing high divergence and restricted introgression, possibly reflecting that purifying selection is slow after the initial generations of backcrossing. Whereas selection was evident in both species, we ascribe this primarily to processes within the species and only to a minor extent to genomic regions showing restricted interspecific gene flow. Finally, similar to Enbody et al. (2021), we found evidence for adaptive introgression of a putative inversion in chromosome 15 from American eel into the European eel gene pool. We further demonstrated that introgression had occurred multiple times independently.

Anguillid eels represent unusual patterns of species boundaries and hybridisation. Hence, even though speciation occurred millions of years ago, gene flow is ongoing and with introgression found throughout the genome, as evidenced by the present study and Barth et al. (2020). Significant recent interest has revolved around gene flow and introgression occurring among otherwise well‐established species (Edelman and Mallet 2021; Mallet 2005), as illustrated by, for example, studies of systems such as Heliconius butterflies (Dasmahapatra et al. 2012; Rougemont et al. 2023; Van Belleghem et al. 2021). For anguillid eels the question is rather how the species can maintain their genomic and phenotypic integrity despite ongoing hybridisation and apparently only weak purifying selection. Perhaps, the huge effective population sizes and consequently low drift in eels (Feng, Liu, and Hansen 2022; Nikolic et al. 2020) render even low purifying selection efficient in the longer term. As a whole, the present results demonstrate that the gene pools of the two species are not isolated; hybridisation is an ongoing process, all individuals of European eel show evidence of introgression, and adaptive introgression occurs at least from American to European eel and presumably also in the opposite direction.

Author Contributions

Aja Noersgaard Buur Tengstedt, Michael M. Hansen and Shenglin Liu conceived and designed the study. Aja Noersgaard Buur Tengstedt conducted bioinformatics and statistical analyses with contributions from Shenglin Liu and Michael M. Hansen. Bjarni Jónsson provided samples and information, and Aja Noersgaard Buur Tengstedt and Michael M. Hansen wrote the manuscript with input from all other authors. All authors read and approved the final version of the manuscript.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1

Table S1

Data Availability Statement

Raw whole‐genome sequencing reads are available in the NCBI Sequence Read Archive under project number PRJNA1127634. VCF files used for the analyses; encompassing a file with raw SNP calls, a file with quality‐filtered SNPs, a file with statistically phased SNPs and an LD‐thinned file, in addition to an ‘all sites’ file containing both variant and invariant sites; have been uploaded to DRYAD (https://www.doi.org/10.5061/dryad.x0k6djhvh) (Tengstedt et al. 2025). Workflows and scripts are available on GitHub (https://github.com/atengstedt/Eel_hybrids).