DNA methylation patterns linked to salinity and geography in the American eel (Anguilla rostrata)

Abstract

The panmictic American eel (Anguilla rostrata) displays a wide range of intraspecific phenotypic variation as well as geographical sex bias and differential recruitment. By definition, panmictic species lack genetic structure, thus local adaptation through genetic variation cannot explain the presence of intraspecific variation. As a result, the contrasting phenotypes observed in the American eel could be attributed to either spatially varying selection, phenotypic plasticity (often mediated through epigenetic changes), or the interaction of both processes. Here we explore, for the first time, the role of DNA methylation in acclimatization in a panmictic species, the American eel, as well as its association with salinity and geography in Northeastern Canada. Using whole genome bisulfite sequencing in 72 individuals, we found that DNA methylation patterns were associated with geography and to a lesser degree with salinity. We identified a genomic region with differential methylation associated with salinity that falls inside the SOCS2 gene, which has been previously linked to salinity differences in other fish species, as well as to metabolism and somatic growth regulation. This study advances our understanding of how panmictic species or populations with high gene flow acclimatize to variable environments in the absence of heritable genetic local adaptation.

Introduction

In panmictic species, the lack of population structure in a single mating pool prevents heritable local adaptation through genetic variation. Any observed acclimatization to heterogeneous environments is usually the result of within-generation spatially varying selection (SVS) [1–6] or phenotypic plasticity. SVS is a type of temporary local selection within a generation, before the gene pool is reshuffled in the unique reproductive pool. The presence of SVS has previously been described in Atlantic killifish (in a panmictic population) [7, 8] and North Atlantic eels (European eel Anguilla anguilla [9]; American eel A. rostrata [1, 4, 10]; both species [3, 11, 12, reviewed in 13]. Through phenotypic plasticity, different local phenotypes can be triggered, in some cases as an adaptive response, by different local environmental conditions. Phenotypic plasticity mediating acclimatization in panmictic species has been reported in rodents [14].

Epigenetic mechanisms such as cytosine methylation are often involved in the regulation of genes that ultimately produce alternative phenotypes [15]. A proliferation of studies has recently underlined the role of DNA methylation in shaping phenotypic plasticity, along with the ecological, adaptive, and evolutionary effects in natural and domestic plant and animal species [16–24]. Different genome-wide methylation patterns often characterize developmental stages, life history traits, phenotypes, and response to seasonal and environmental changes. Methylation can leave transient or permanent signals; thus, timing is paramount for detecting differential patterns. For example, Viitaniemi et al. [25] found that temporal samples of female great tits (Parus major) presented small changes in methylation during the breeding season. Methylation variability in plants has been associated to local and wide-range environmental differences, as well as to adaptive responses to such environmental conditions [26]. In fishes, methylation patterns have been associated with rearing environment salinity [27] or context (e.g. hatchery vs. wild) [28, 29]. In the European eel, different methylation patterns have been linked to developmental stages and heavy metal exposure levels [30–34].

The American eel comprises a single panmictic population over its broad geographical distribution range from Greenland to Colombia [e.g. 35, 36]. Larvae hatch in the proximity of the Sargasso Sea and promptly disperse as glass eels/elvers to reach their feeding and growing areas, where immature yellow eels live until they reach sexual maturity, become silver eels and migrate back to the Sargasso Sea to reproduce and die (for more details, see Fig. 1b in 36). Although heritable genetic local adaptation to environmentally dissimilar habitats cannot arise in this species, spatial phenotypic variation is well documented. Different rearing environments are characterized by different eel phenotypes and gene expression patterns [4, 37–40]. Typically, American eels found upstream in freshwater are prone to become females, and females overall reach larger sizes than males. In contrast, brackish waters, like estuaries or bays, have a higher number of males that reach maturity earlier, but at smaller sizes [41]. Potentially due to early developmental plasticity windows, phenotypic differences linked to rearing grounds are partly maintained even when eels are reared in a common environment [42, 43] or translocated early in life, for example from the Maritimes provinces to Lake Ontario and the Upper Saint Lawrence River and tributaries [38, 44–47] and in the Chesapeake Bay watershed [48]. Recruitment, which is linked to life history, is also contrasted among areas. The Upper Saint Lawrence River and the Great Lakes are marked by a decline in new eel arrivals, while the number of elvers remains relatively stable in the Gulf of Saint Lawrence and the Maritimes [49]. The mechanisms by which such patterns of spatial phenotypic variation are attained have yet to be fully explained. SVS has been proposed as a mechanism for temporary adaptation, based on allele frequency differences associated with latitudinal variations (allozymes [50, 51]; single nucleotide polymorphisms (SNPs) [1, 10]), salinity (SNPs [4]), or contamination (SNPs [11]). However, the relative importance of phenotypic plasticity and its relationship with SVS are not known.

Figure 1.

(a) Map showing sampling sites. (b) and (c) Overall methylation patterns with all CpG loci. (b) PCA results where each point represents an individual, with point colour indicating sampling sites, and point shape indicating habitat. The 0.95 confidence ellipses group sampling sites into geographic areas, with colour indicating the broader geographic area. (c) Dendrogram with colours indicating geographic area, and individual sampling site indicated at the end of each branch (as per Panel b). For similar dendrogram with bootstrapped nodes estimated using pvclust, see Supplementary Figure S2.

Given the phenotypic and life history variation of this panmictic species, we hypothesized that epigenetic differences would contribute to acclimatization to different environments. We tested if individuals from six sampling sites with contrasting salinities showed contrasting DNA methylation patterns and found methylation differences associated with salinity and geography at both individual loci and longer genomic regions. One particular region of the SOCS2 gene appears to be linked with salinity. Our results showcase how methylation can contribute to acclimatization and plasticity in panmictic species or species with high gene flow.

Materials and methods

We used fin tissue from 74 wild caught yellow eels that were captured near six sites from three geographic areas in Eastern Canada (River, Gulf, Ocean; 12–13 individuals per site; see Fig. 1a and Supplementary Table 1). These eels were sampled in 2008 and 2011 for a study analysing genetic differentiation between eels from freshwater and brackish/saltwater [4]. In the River area, eels were captured in the fluvial section of the St. Lawrence near Beauharnois (B) and in the Petite-Trinité (PT) river, which reaches the St. Lawrence Estuary close to its eastern limit. In the Gulf area, eels were captured in the southwest portion of the Gulf of St. Lawrence near Iles-de-la-Madeleine (IM), in the Cascumpeque Bay (CB) and in Lake of Shining (LS). IM and CB are brackish/saltwater sites. LS is a freshwater pond separated from the Gulf by only 200 metres. In the Ocean area, eels were captured near Riviere Bourgeoise (RB) in brackish water/saltwater following the classification by [4]. The age of the eels was determined for PT samples using otoliths (Supplementary Table 1), showing that these individuals belong to different age cohorts.

DNA was extracted following the salt-based extraction protocol developed by Aljanabi and Martinez [52]. Whole genome bisulfite sequencing libraries were performed at the Génome Québec Innovation Centre using NEBNext Ultra II DNA library prep kits and sequenced on the Illumina HiSeqX platform (150 bp paired-end reads; target coverage 18.5×). After quality control with MultiQC [53], raw sequences were prepared for analysis using the bwa-meth pipeline (https://github.com/enormandeau/bwa-meth_pipeline.git). Briefly, sequences were trimmed and filtered for quality with fastp [54] and aligned to a reference genome (GCA_018555375.2) with bwa-meth v0.2.2 [55]. Duplicates were removed with Picard v1.119 function MarkDuplicates (Broad Institute, 2019) and methylation biases were detected with Methyl-Dackel v0.4.0 (https://github.com/dpryan79/MethylDackel). This package was also used to extract merged CpG dinucleotide methylation data (using function mergeContext) in bedgraph format and nonmerged cytosines in methylKit [56] format after trimming bases with biased methylation calls.

Because we cannot distinguish between a T nucleotide from a SNP vs. a T nucleotide from an unmethylated cytosine, C/T SNPs (and A/G SNPs at the G position of the CpG site) lead to inaccuracies in methylation calling. To mask C/T polymorphisms, we used the genomic data from 96 individuals from the same localities (low-coverage whole genome sequencing with approximately 4× average coverage; 7 636 736 loci when using -maxdepth 9× and Canadian localities only) [36]. Following Wellband et al. [57], both C/T and A/G biallelic SNPs were used to mask against the bisulfite sequences using the intersect function in bedtools v2.26.0 [58]. After masking, and in order to equalize sample size, we eliminated the samples with the lowest average coverage from each site. All downstream analyses were carried out with 12 samples per site (72 samples in total, we eliminated one from RB and one from B). Global average methylation was estimated in each site to detect samples with methylation degradation due to long-term storage (Supplementary Fig. S1) but no obvious outliers were found and no samples were discarded following this quality control step. Lastly, we only retained CpG loci identified within chromosomes, filtering out those in unplaced scaffolds and the mitochondria.

Overall methylation patterns

We analysed overall methylation patterns using the methylKit library [56] in R v4.1 (R Core Team, 2022) after retaining sites with coverages between 10× to 100× (filterbyCoverage parameters lo.count = 10, lo.perc = NULL, hi.count = 100, hi.perc = 99.9). We normalized the data by coverage using the median, and kept loci present in 91.6% of samples (66 samples) while merging both strands. For overall analysis, we first filtered by strand and then across individuals while merging because we wanted to retain a more stringent set of loci to decrease noise. We performed a principal component analysis (PCA) using the function PCAsamples with default arguments but without discarding low variation loci (sd.filter = FALSE) and visualized it with ggplot2 [59]. We also performed hierarchical clustering (Euclidean distance and Ward.D2 method) with the clusterSamples function (without discarding low variation loci) and visualized the results with the dendextend package. We used pvclust [60] with nboot = 1000 to obtain a dendrogram with estimated approximate unbiased P-values and bootstrap values for the nodes. As suggested by Lea et al. [17], overall data analysis compensates for small sample sizes and allows for successfully detecting small divergence typically associated with many environmental variables.

To associate principal components with the covariates ‘sampling site’, ‘geography’, and ‘salinity’, a subset of the dataset with no missing data (excluding missing data points) was obtained and the function assocComp (methylKit) was used. To estimate how much of the variance could be attributed to salinity or geography, we used the function varpart from vegan package [61] using percentage of methylation as input (function percMethylation in methylKit). The same estimation was done with the package variancePartition [62] developed for high-throughput genomics experiments.

Differential methylation between groups

We identified differentially methylated CpG loci (DML) and differentially methylated regions (DMRs), as well as their location in the genome, using DSS [63] in R v4.1 (R Core Team, 2022). Starting with already merged CpG loci, we again retained only those with coverages between 10× and 100×, and present in at least 66 of 72 total individuals (91.6%), using the custom filter code publicly available (https://github.com/cvenney/dss_pipeline). We tested for differential DNA methylation in association with salinity (freshwater vs. saltwater habitat) and geographic areas (River vs. Gulf, River vs. Ocean, Gulf vs. Ocean). In both cases, we used the Wald test for all pairwise comparisons, and a minimum delta of 0.2 (20%) was required for both DML and DMRs Methylation data were smoothed over 500 bp, which is the default for DSS. Differentially methylated loci were obtained with the function callDML with parameter p.threshold = 0.001. We used DSS function callDMR considering a genomic region as a DMR when it contained at least six CpGs in a minimum of 100 bp (again with p.threshold = 0.001). DMRs were merged when within 50 bp from one another. Additionally, we used the function DMLfit.multiFactor to fit a generalized linear model (GLM) including salinity and geography for hypothesis testing (salinity with geography as covariate). We kept the same parameters (with the exception of delta, which is not used by the function) as before when using the function callDMR in the multifactor design.

DMRs were visualized using the heatmap scripts found in https://github.com/kylewellband/methylUtil.git. Briefly, we obtained the percent methylation estimates for each DMR by averaging the number of methylated reads (M) by the total number of reads (coverage) for each CpG site inside each DMR, using the function ‘DMR_heatmap’ and plotting with the package ComplexHeatmap [64]. We plotted all loci q-values (Benjamini and Hochberg method) and highlighted those loci that were the start of DMRs in Manhattan plots obtained with the CMplot library [65].

Finally, overlaps between DMRs lists provided by the Wald tests and the GLM were obtained with the function findOverlapPairs from the package IRanges [66]. To annotate the genome, we used GAWN (https://github.com/enormandeau/gawn) with default options and using the European eel transcriptome. This produced a simplified genome annotation table (Supplementary Table 2) with the putative locations of annotated genes, which was linked to the locations of DMRs associated with salinity by both analysis methods.

Results

Overall methylation patterns

After filtering and uniting the strands in methylKit [56], we analysed 67 020 CpG sites (reduced to 9678 CpG loci when no missing data are allowed) with approximately 23× to 47× average coverage per sample. Overall, methylation PCA analysis shows that samples from the Gulf (LS, CB, and IM) and River (B and PT) sampling sites cluster relatively close together; they are partially overlapping, but widespread along PC1 (Fig. 1b). Samples from the River and the Gulf spread from a cluster with most individuals from PT to samples from B, IM, CB and finally LS. However, high dispersion within the Gulf clusters and overlapping with mostly B samples creates a soft shift between sites. All samples from the Ocean (RB) cluster together, revealing a distinct overall methylation pattern. They are segregated from all other sites on PC2 but are closer to Gulf sites on PC1.

The hierarchical cluster dendrogram (Fig. 1c; for dendrogram with bootstrap support obtained with pvclust [60], see Supplementary Fig. S2) shows samples grouping by geographic areas. One of the two main branches (River + Gulf) includes all River samples and eight Gulf samples. Most samples from PT and B are located on different inner branches (all samples from PT and B are located on different inner branches in the dendrogram obtained with pvclust [60], Supplementary Fig. S2), and the Gulf samples are intermingled with B samples. The second major branch (Ocean + Gulf) includes samples from the Ocean located on a distinct inner branch, while most Gulf samples form another inner branch. Interestingly, all but one sample from LS (all samples from LS in the dendrogram obtained with pvclust [60], Supplementary Fig. S2), the only freshwater site in the Gulf, are grouping with the other samples from this area on the Gulf branch. Overall, we see clear epigenetic differentiation between Ocean and River samples, with Gulf samples split between the two groups.

The association analysis confirms that both PC1 and PC2 are correlated with sampling sites and geographic areas (P < 0.001), but no association is found with salinity (P > 0.1). Partition analysis [61, 62] also indicates that a larger portion of the variation is linked to geographic areas (vegan: 6.3%; variancePartition: 5.5–8.1%; vegan’s varpar or variancePartition’s median-mean across all loci estimation, respectively). A small proportion of the variance is linked to salinity (vegan: 0.9%; variancePartition: 1.5–2.3%; per the same estimators).

Differential methylation between groups

After filtering the merged bedgraphs, we analysed 4 270 475 CpG sites with approximately 13.7× to 24.0× coverage per sample (average sample methylation 18×).

Methylation was markedly different between geographic areas (Fig. 2a–c), with numerous DMRs widespread across the genome (Fig. 2d and f). Pairwise Wald tests showed that the Ocean area was most different from both River and Gulf, with more than 80 000 DML and 3000 DMRs (Fig. 2b and c). In contrast, we observed fewer DML and DMRs in the Gulf vs. River comparison (4501 DML and 147 DMRs, Fig. 2a and d). This matches with the observations in the PCA, with the River and the Gulf samples overlapping while the Ocean samples cluster independently.

Figure 2.

Pairwise differential methylation between geographic areas. Heatmaps show DMR methylation percentage (rows) per individual (columns) for pairwise comparisons with the Wald method between (a) River vs. Gulf, (b) River vs. Ocean, and (c) Gulf vs. Ocean. Corresponding Manhattan plots show q-values for each locus, with the start position of each DMR highlighted in red for pairwise comparisons between (d) River vs. Gulf, (e) River vs. Ocean, and (f) Gulf vs. Ocean.

Differential methylation patterns between freshwater and saltwater were more apparent with the GLM method, which included geographic area as a covariate, than with the Wald test, which did not control for geographic area (Fig. 3a and b). Both methods identified only a few DMRs clustered in the genome (Fig. 3c and d). With the Wald test, the Ocean and all three Gulf sampling sites shared similar methylation patterns despite LS being a freshwater site (Fig. 3a). The 36 DMRs (and 731 DML) between salinity groups were located mostly on chromosomes 8, 11, and 17 (Fig. 3c, Supplementary Table 3). With the GLM, all freshwater sites are visibly grouping together, including the freshwater site located in the Gulf (LS) and the two River sites. This pattern is supported by only five DMRs, one of which (top DMR in Fig. 3b) strongly differentiates between salinities. This DMR (chromosome 11 from 23 932 771 to 23 933 153) partially overlaps with one of the 36 DMRs obtained with the Wald test (chromosome 11 from 23 932 758 to 23 933 040). The remaining DMRs do not overlap with those identified by the Wald test and are present on chromosomes 8 and 15.

Figure 3.

Pairwise differential methylation between freshwater and brackish/saltwater. Heatmaps show DMR methylation percentage (rows) per individual (columns) for pairwise comparisons with (a) the Wald method and (b) the GLM method. Corresponding Manhattan plots present each locus q-value, with the start position of each DMR highlighted in red for pairwise comparisons using (c) the Wald method and (d) the GLM method.

All but one DMR identified with the Wald test between salinities fully or partially overlaps with DMRs in pairwise comparisons involving the River area. The nonoverlapping DMR is the one also present in the GLM comparison between salinities. None of the DMRs obtained through this method overlapped with any DMR obtained with geography comparisons using the Wald test. Curiously, there is no overlap between the DMRs observed in the freshwater vs. saltwater comparison using the Wald test and the DMRs observed in the Gulf vs. Ocean. There are 15 overlaps between River vs. Gulf and Gulf vs. Ocean and 2174 overlaps between River vs. Ocean and Gulf vs. Ocean, while 92 DMRs fully or partially overlap in the River vs. Gulf and River vs. Ocean comparisons. A summary of overlapping DMRs can be found in Supplementary Table 4.

When comparing salinities, we obtained only one partially overlapping DMR statistically supported by both detection methods. The genomic region corresponding to the partially overlapping DMR falls inside the annotation for the predicted gene suppressor of cytokine signalling 2 (SOCS2) and is located on chromosome 11 (from 23 930 390 to 23 936 139, see Supplementary Table 2).

Discussion

This study reveals that American eels from distinct geographic areas display genome-wide epigenetic differentiation in the absence of genetic differences, both in overall patterns of methylation and at specific loci and genomic regions. This supports the idea that epigenetic mechanisms could contribute to phenotypic plasticity in panmictic species such as eels, facilitating acclimatization in different habitats when local genetic adaptation is impossible. Epigenetic differences were widely distributed across the genome, as may be expected given the heterogeneity of the environment across the study area. However, one DMR associated with salinity exposure was located, exemplifying how epigenetic marks can be associated with specific and important habitat characteristics.

Our results suggest the Gulf is an intermediate area of horizontal movements between salinities. Eels are known to be facultative catadromous [67–70], which means that some eels take definitive residence (up until the reproductive migration) either in freshwater or salt/brackish water, while others shift between habitats and different salinities. These inter-habitat movements can occur once or several times in an individual life [69, 70] and according to otolith chemistry, most brackish water individuals have ‘experienced freshwater at least once’ [68]. This could explain why overall methylation patterns of a few individuals from the Gulf are similar to those seen in the River, as evidenced by River and Gulf samples clustering on a main branch in the dendrogram and overlapping in the PCA (Fig. 1).

In both the PCA and the hierarchical clustering, samples from the Ocean (Rivière Bourgeoise, RB) all group together but are more similar to the Gulf than to River samples. The comparisons with the Ocean are also the ones with the highest number of DMRs. This could be an indication that RB individuals are strictly saltwater residents while the Gulf ones move between salinities or endure different environmental conditions while still sharing more with them than with River sampling sites. These differences compounded would be reflected in the methylation and create a particular ‘ocean methylation pattern’. Due to the convenience nature of our samples acquisition, the sampling design is unbalanced with only one locality to represent the ocean region, which could ultimately result in bias. Further research including other Ocean locations would be necessary to confirm this pattern. Alternatively, individuals could indeed span a gradient from freshwater residents to saltwater residents on PC1, but the segregation of Ocean eels on PC2 (and in the dendrogram) could be due to different life stages (yellow vs. silver). This is to say, eels caught in RB may have been individuals initiating the silvering metamorphosis, but inadvertently labelled as yellow eels, as the methods usually used to differentiate life stages, like the skin color or other morphological characteristics, are unreliable [71, 72]. Indeed, silver eels’ reproductive migration in Nova Scotia happens around the fall months [73], and RB eels were sampled in October. Then, we would expect to see major differences in DNA methylation, as documented by Trautner et al. [33] in gills and brain tissue of European eel. This explanation seems unlikely, as life stage differences in DNA methylation would likely be more marked than environmental or life history differences (hence drive PC1, not PC2), as observed by Liu et al. (Fig. 2b in [34]) between juveniles and adult European eels.

Methylation profiles of eels from the two River sites (B and PT) were similar in all analyses. This finding is quite surprising given that the sampling sites are distant by more than 670 km away, and under different environmental conditions. Beauharnois (B) is far upstream in the fluvial section of the Saint Lawrence River, while Petite-Trinité (PT) is a tributary river very close to the beginning of the Gulf. Fluvial and estuarine sections of the Saint Lawrence present contrasting conditions in terms of water chemistry, turbidity, currents and tides, productivity, and food chain composition [e.g. 74, 75]. This may explain why most samples from PT cluster together as a distinct group in the dendrogram and the PCA.

Comparison of methylation patterns considering only the sampling site salinity suggests that differences are primarily driven by geographic motifs rather than the mere difference between salinities, or at least that both are confounded. Considering the life cycle of the eel, it could be that these epigenetic patterns are the product of transient plasticity associated with recruitment in the studied areas. Alternatively, epigenetic marks may already be present when elvers arrive in an area (maybe as a product of developmental plasticity), and result in those individuals staying in a given environment rather than being the result of such environment. As such, epigenetic marks could principally reflect how far yellow eels have migrated towards growing grounds. Indeed, freshwater sites in the River area are much further away from the ocean proper, leaving plenty of time for a distinct methylation pattern to develop. However, when geography is used as a covariate to compare sampling sites with different salinities, methylation patterns proved similar in all freshwater sites. Indeed, Lake of Shining (LS), the sole freshwater site in the Gulf, then displayed a methylation pattern similar to that of River sites. This site is so close to brackish water, however, that movements between different salinity environments may also explain why eels from LS display a methylation pattern less typical of other freshwater sites, and why LS is more similar to other Gulf sampling sites than to River sites in all other comparisons (dendrogram and PCA analysis).

One partially overlapped DMR was highlighted by both statistical methods for detecting DMRs in freshwater vs. saltwater, providing statistical support for its association with salinity. This genomic region overlaps with the annotation for the suppressor of cytokine signalling 2 (SOCS2) gene. Differential expression of the SOCS2 gene has been associated with salinity changes or exposure in several fish species [76–78], including the Japanese eel (A. japonica) [79]. This gene is also associated with growth and metabolism regulation in vertebrates [80–82]. An in vitro study in rainbow trout (Oncorhynchus mykiss) liver suggests that it modulates the stress-growth axis, and cortisol-induced upregulation of SOCS2 (and SOCS1) seems to inhibit insulin-like growth factor-1 (IGF-1) mRNA levels mediated by growth hormone (GH) levels [83, 84]. Additionally, SOCS2 knockout mice have higher growth rates and display physiologic and morphologic phenotypes characteristic of dysregulated GH/IGF-1 pathway [85]. The DMR associated to this gene body shows hypermethylation in saline sampling sites, which is sometimes associated to upregulation of the transcription [86–88]. Interestingly, in freshwater eels, brackish environments and aquaculture settings are linked to elevated cortisol (due to higher densities) and usually higher growth rates (but smaller final size and age) often leading to more males [reviewed in 89]. Differences in the expression of SOCS2 may thus have pleiotropic effects with multiple links to the eel complex life cycle.

Additionally, while single nucleotide polymorphisms in SOCS2 have also been found to be associated with differential salinity in other species [90, 91], the gene was not among those found to discriminate between salinities in the American eel by Pavey et al. [4], suggesting its acclimatization role in this species is mostly performed through epigenetic mechanisms. However, differential expression of this gene was not observed in previous studies looking for differential transcription associated with salinity in the species [2]. Certainly, further research on this gene [similar to 39, 92], or qualitative transcriptional analysis like alternative splicing, is warranted, especially from wild-caught eels.

A shortcoming of our study is that geography may be confounded with sex ratios, which vary across the study zone. Indeed, there are mostly females in upstream areas of the Saint Lawrence system, whereas males are more often found in estuaries and downstream areas [40, 43, 93]. For example, natural migrants to Lake Ontario are usually females [43], while both sexes can be found in different ratios in Nova Scotia rivers like the Mira River (Dr Martha Jones, personal communication in Côté et al. [2]). Thus, differences observed in this study could partly be due to the different sex ratios naturally occurring in the study zone. If so, the variance observed along PC1 and the inner branches of the dendrogram (Gulf + Ocean vs. Gulf + River) could partially reflect sex differences in methylation patterns, with females clustering in the River + Gulf branch and at the left of the PCA. Differential methylation linked to sex has been observed in other species (e.g. Planococcus citri [94]). Additionally, we cannot disregard the possibility of confounding age/life stage introducing variance in our data (as discussed previously for RB in Ocean). Future developments in epigenetic aging molecular clocks in the species, as well as further research on life stage and sex characteristic methylation patterns, could shed more light into it.

Our exploration of methylation patterns in eels inhabiting environmentally different sampling sites offers a glance at how potentially adaptive phenotypic plasticity can result from epigenetic mechanisms These findings add to those supporting spatially varying selection in the same localities [4] and other eel localities [1] in the northern part of the species range. Epigenetic regulation mediated through methylation could work together with SVS by upregulating or downregulating the expression of genes under SVS. Genetic variants selected by SVS could also work in a different layer of epigenetic regulation, with genetic variants influencing DNA methylation patterns [e.g. 95, 96]. Differentially methylated CpG loci or genomic regions that are related to salinity should be further studied, ideally in a common garden experiment controlling for sex and age variables and in conjunction with other methods and transcriptional analysis. Also, tissues more relevant to salinity, like gill or liver tissue, should be included. It could be interesting to research if arriving glass eels share the patterns observed here, or if it is possible to assign them to different categories based on such patterns. Further research should also tackle the link between genetic variation and methylation, since several DMRs are found in areas where structural variants have been observed [36].

Overall, our results suggest that differences in DNA methylation allow the panmictic American eel to acclimate to different environments across Northeastern Canada despite the lack of genetic structure across their range. It is possible that plasticity early in life plays a role in said acclimation, perhaps triggered by environmental and social clues, like salinity and density. From a conservation perspective, our results suggest that active conservation and management measures like translocation of glass eels from different geographical areas may be misguided, in agreement with previous translocation experiments showing that translocated elvers from donor sites in the Maritimes developed most phenotypic characteristics typical of their origin localities while growing in the Upper Saint Lawrence [38, 44–47]. This highlights the importance of passive conservation strategies like the creation of eel ladders or seasonal pauses of hydroelectric dam turbines though any management recommendations would require further research for confirmation and should be context-dependent. Our results also introduce the notion that methylation patterns could potentially be used to identify rearing site, life history variations, sex, or developmental stage in panmictic species like the American eel. Additionally, our results highlight the importance of phenotypic plasticity and epigenetic mechanisms in acclimatization of panmictic species with high dispersal that encounter diverse environments. In a context where the American eel faces major threats such as dams, habitat pollution, and climate change, with increased demand for glass eels driving prices higher, advancing our comprehension of the mechanisms helping this species to cope with different environments will be paramount for developing effective management and conservation strategies.

Author contributions

Gabriela Ulmo-Diaz (Conceptualization, data curation, formal analysis, funding acquisition, investigation, software, visualization, writing original draft, writing—review & editing), Celine Audet (Conceptualization, funding acquisition, writing—review & editing), Eric Normandeau (Data curation, formal analysis, software, writing—review & editing), Clare Venney (Data curation, software, writing—review & editing), Kyle Wellband (Data curation, software, writing—review & editing), Julie Turgeon (Conceptualization, supervision, writing—review & editing), and Louis Bernatchez (Conceptualization, data curation, funding acquisition, project administration, resources, supervision)

Conflict of interest

The authors are not aware of any potential conflict of interest.

Funding

This work was supported by the NSERC/CRSNG Vanier scholarship and the project GÉCAS: Génomique et Épigénétique pour la conservation de l’anguille du Saint-Laurent. This project was funded by the Réseau Québec Maritime (Intersectorial network supported by the Ministère de l’Économie, de l’Innovation et de l’Énergie du Québec and Fonds de recherche du Québec) grant to Louis Bernatchez (U. Laval), Céline Audet (UQAR), Lyne Létourneau (U. Laval), Louis-Étienne Pigeon (U. Laval), and Alexandre Montpetit (McGill U.) as well as Ressources Aquatiques Québec (Regroupement stratégique, Fonds de recherche du Québec—Nature et Technologies).

Data availability

The data here analysed can be found in NCBI SRA under accession number PRJNA1167745.