Two Decades of DNA Methylation Shifts Reveal Epigenomic Responses to Environmental Change in Wild Salmo salar From Its Southernmost Populations

ABSTRACT

Atlantic salmon ( Salmo salar ) populations have been declining across the North Atlantic, with southern populations, including those in rivers draining into the southwestern Bay of Biscay, showing some of the strongest impacts. These rivers mark the species' southernmost distribution limit and are highly vulnerable to global change stressors such as rising marine and freshwater temperatures, ocean acidification, altered food webs, and pollution. Epigenetic mechanisms, including DNA methylation, can mediate environmentally induced phenotypic plasticity and may contribute to rapid adaptation. We used Reduced Representation Bisulphite Sequencing (RRBS) to examine genome‐wide, single‐base DNA methylation patterns in fin tissue from Atlantic salmon sampled at three time points over two decades (early 2000s, early 2010s and early 2020s) from two river systems (Nalón–Narcea and Sella, northern Spain). Analysis of 715,240 common CpGs in 120 adults identified differentially methylated cytosines (DMCs) and regions (DMRs), including the promoters of two Hox genes essential for embryonic development. Gene Ontology and KEGG enrichment analyses revealed significant overrepresentation of developmental pathways, notably the ECM–receptor interaction pathway involved in morphogenesis and vertebral formation. Several enriched pathways were also related to environmental information processing and cell signalling, suggesting a potential epigenetic response to environmental stressors. These findings provide evidence of an epigenetic footprint of environmental change over time. DNA methylation may influence phenotypic variation in declining southern salmon populations, which might facilitate adaptation. Future research should incorporate controlled experiments to link methylation changes to specific stressors and assess their causal role in resilience and adaptation.

1. Introduction

Global change is resulting in rapid environmental changes leading to modifications in the geographical distribution of species (Sirami et al. 2017), population collapses (Brook et al. 2008), or changes in food availability (Moloney et al. 2011), among other impacts. One of the main drivers of these anthropogenic changes is atmospheric CO₂ enrichment, which leads to ocean acidification and climate change (Sage 2020). In response to these variations, species can adapt either slowly via natural selection on standing genetic variation (DuBois et al. 2022; Hoffmann and Sgrò 2011) or fast by phenotypic plasticity, which is mainly driven by epigenetic mechanisms (Rey et al. 2016; Zetzsche and Fallet 2024). Environmental epigenetics represents the process by which the environment affects biology (Skinner 2023). However, to what extent epigenetics are contributing to global change effects adaptation is under debate (Harney et al. 2022; McGuigan et al. 2021).

Epigenetics could be defined as the study of the “genome‐associated mechanisms of non‐DNA sequenced‐based inheritance” (Deans and Maggert 2015), which result in stable gene expression states that are transmissible through cell divisions (Richards 2006). DNA methylation (DNAm) involves direct chemical modification of the DNA through the transfer of a methyl group from S‐adenyl methionine to the fifth carbon of a cytosine to form a 5‐methylcytosine (5‐mC) catalysed by a DNA methyltransferase. In vertebrates, this mainly occurs in cytosine‐guanine dinucleotides (CpG) and regulates gene expression by recruiting proteins implicated in gene repression or by impeding the binding of transcription factors (Héberlé and Bardet 2019; Moore et al. 2013). Moreover, it plays a role in alternative splicing leading to distinct transcripts in response to environmental factors (Maor et al. 2015). Alternative splicing is relevant for adaptive evolution, especially at the short timescales (Singh and Ahi 2022). The study of ancient RNA is possible when samples are preserved by cold, desiccation or chemical treatment (Friedländer and Gilbert 2024) and relevant discoveries have been recently reported. For instance, Mármol‐Sánchez et al. (2023) analysed historical RNA from skin and muscle tissue of a ∼130‐year‐old Tasmanian tiger ( Thylacinus cynocephalus ), representing the first successful attempt to obtain transcriptional profiles from an extinct animal species. However, in the absence of ancient RNA, temporal gene expression changes or transcript variants through alternative splicing are usually inferred from DNAm patterns. The study of ancient samples is growing steadily (Hernando‐Herraez et al. 2015; Orlando et al. 2021), and important findings have been made, particularly regarding humans and ancient hominids. For example, Gokhman et al. (2014) reconstructed the full methylation maps of Neanderthals and Denisovans and compared these ancient maps with those of present‐day humans, thereby shedding light on anatomical and disease‐related differences.

Environmental epigenetic changes associated with global change have been documented in various species through laboratory‐based experiments. In European seabass ( Dicentrarchus labrax ) larvae, DNAm changes in ecologically relevant genes were detected following rearing at elevated temperature conditions (Anastasiadi et al. 2017). Moreover, in this same species, the DNAm changes induced by a simulated marine heatwave during early development persisted after 3 years of exposure to the increased temperature (Anastasiadi, Shao, et al. 2021). More recently, paternal multi‐generational temperature‐induced DNAm marks were identified, with approximately 5% of all DMRs being both temperature‐induced and inherited from sires to offspring (Sánchez‐Baizán et al. 2025). In reef‐building corals ( Stylophora pistillata ), long‐term DNAm variation due to pH stress was found. These changes were related to phenotypic plasticity and suggest an epigenetic component in phenotypic acclimatization (Liew et al. 2018). Later, inter‐generational epigenetic inheritance was detected in that species (Liew et al. 2020). As a final example, embryonic DNAm reprogramming changes were reported in marine stickleback ( Gasterosteus aculeatus ) under ocean warming scenarios (Fellous et al. 2022), and inter‐generational heritability of DNAm marks was proven analysing fin tissue of F1 and F2 generations of natural stickleback populations from different ecotypes (Hu et al. 2021).

In Atlantic salmon ( Salmo salar ), environmental stressors such as cold‐shock during embryogenesis, absence of tank enrichment during larval‐stage or long‐term CO₂ exposure have been reported to affect the transcriptome and epigenome (Mota et al. 2020; Uren Webster et al. 2018). Among these environmental stressors, temperature, particularly during gametogenesis and spawning (Alix et al. 2020; Dahlke et al. 2020), can lead to phenotypic plasticity influencing life‐history traits through epigenetic mechanisms (Jonsson and Jonsson 2019). For example, increases in embryonic water temperature resulted in a 2‐week delay in the return of salmon from feeding areas (Jonsson and Jonsson 2018). Moreover, male size dimorphism in this species has been hypothesized to be an environmentally induced, epigenetically mediated trait (Morán and Pérez‐Figueroa 2011). Environmentally induced changes in DNA methylation may alter gene regulation, ultimately influencing the phenology and fitness of different populations along the species distribution (Poloczanska et al. 2016; Saino et al. 2017; Venney et al. 2023).

The evolutionary importance of DNAm relies on the extent of epigenetic reprogramming, which is itself sensitive to environmental variations (Fellous et al. 2022). Though the two rounds of epigenetic reprogramming in mammals, during primordial germ cell determination and after fertilization, have been widely studied (Zeng and Chen 2019), species‐specific processes occur in fish. Contrasting methylation programs have been described between medaka ( Oryzias latipes ) (Wang and Bhandari 2019), zebrafish (Danio rerio) (Jiang et al. 2013; Potok et al. 2013), marine stickleback (Fellous et al. 2022), and mangrove rivulus (Kryptolebias marmoratus) (Fellous et al. 2018). Germline epigenome reprogramming is not extensive in teleost fishes, easing epigenetic inheritance (Anastasiadi, Venney, et al. 2021). In Atlantic salmon, DNAm inheritance has been previously reported in response to captive breeding (Rodriguez Barreto et al. 2019; Wellband et al. 2021) and suggested after micronutrient supplementation (Saito et al. 2022).

Atlantic salmon is an ecologically and culturally important anadromous species, native to the rivers of North America and Europe that flow into the North Atlantic Ocean, as well as the Barents and Baltic Seas. It exhibits different life strategies, the main implying a juvenile freshwater phase followed by an oceanic period in feeding grounds before migrating back to rivers for spawning (Klemetsen et al. 2003). The species exhibits a strong homing behaviour, which has led to structured populations per river basin (Gilbey et al. 2018; Griffiths et al. 2010) showing unique local adaptations (Garcia de Leaniz et al. 2007), although some gene flow occurs between rivers (Fontaine et al. 2025). All Atlantic salmon populations have declined substantially in the last decades (Dadswell et al. 2022), with reduced marine survival identified as a major driver of this decline (Gillson et al. 2022). Overfishing (Parrish et al. 1998), both in rivers and in the ocean, pollution (Gillson et al. 2022), migration barriers (Smialek et al. 2021), and climate change (Jonsson and Jonsson 2009; Thorstad et al. 2021), are pointed as the main reasons of the decline. The South Bay of Biscay represents the southernmost natural distribution of the species, where individuals are exposed to the highest temperatures and face more intense effects of climate change (Almodóvar et al. 2019; Nicola et al. 2018). Climate change impacts are particularly severe in the region. For instance, between 2004 and 2024, snow cover extent has been reduced by up to 16% per decade in some basins of the Cantabrian Mountains (Melón‐Nava and Gómez‐Villar 2025). From 1998 to 2019, sea surface temperature (SST) has increased, and marine heatwaves have become more frequent, longer, and more intense (Izquierdo et al. 2022). Additionally, in the last four decades, there has been a general reduction in river streamflow, especially in spring and summer (Martínez‐Fernández et al. 2013). This reduction is mainly due to decreased winter precipitation, particularly since the 1980s (Halifa‐Marín et al. 2022). Moreover, differences in size and age at maturity (Jonsson et al., 2016) and reduced survival and growth (Almodóvar et al. 2019) have been correlated with environmental and food availability changes happening in the North Atlantic Ocean feeding grounds.

These environmental and anthropogenic changes make these salmon populations valuable models for studying epigenetic microevolutionary processes and their potential role in mediating responses to global change. In rapidly changing environments, epigenetic effects may arise over only a few generations (McGuigan et al. 2021). Historical samples may shed light on this, as comparisons of DNA methylation patterns between contemporary and archival samples can reveal epigenetic population variation (Gokhman et al. 2016; Rubi et al. 2020). Although variation in DNAm among generations does not itself demonstrate inheritance, it may provide insight into adaptive processes occurring over short evolutionary timescales. The field of historical epigenomics is advancing rapidly (Holleley and Hahn 2025); however, well‐preserved genomic DNA suitable for epigenetic analyses remains scarce. To our knowledge, this study represents one of the few investigations assessing temporal variation in DNAm across decades in a wild fish species, providing novel insights into how global environmental change may influence epigenetic variation in populations located at the limits of their distribution.

In this study, we aim to identify differentially methylated cytosines and regions (DMCs & DMRs) by comparing the methylomes of wild salmon from two populations at the southern limit of the species´ distribution across two decades characterized by intense environmental change. These patterns may reflect either signs of environmentally induced epigenetic inheritance and/or intra‐generational environmentally induced epigenetic changes. However, these two mechanisms cannot be distinguished with the dataset used in the present work. Although the temporal scale considered (20 years) is relatively short, the environmental changes occurring during this period have been substantial and highly relevant at the biosphere level (Sage 2020). To minimise confounding effects related to age, we selected returning adult salmon of similar body size (4–6 kg), indicative of similar age‐at‐sea and, thus similar lifetime environmental exposure. Genome‐wide DNAm was assessed at base‐pair resolution using Reduced Representation Bisulphite Sequencing (RRBS), with samples collected from the two main river systems in the Cantabrian region: the Nalón–Narcea and the Sella. Through the analysis of temporal methylation patterns, we aim to assess the potential role of epigenetics and plasticity as mechanisms enabling rapid responses to environmental change in this emblematic species.

2. Methodology

2.1. Atlantic Salmon in the Study Region

Atlantic salmon spend their early life in rivers, typically between one and 4 years as juveniles (parr) (Hansen and Quinn 1998; Klemetsen et al. 2003). After smoltification, a physiological transformation that prepares them for marine life, they migrate to the ocean, where they feed and grow from one to several years (Thorstad et al. 2012). In Asturias (southwestern Bay of Biscay, northern Spain), the typical life cycle includes one to 2 years in freshwater and one to 3 years at sea, resulting in a spawning population composed of multiple adult cohorts alongside male parr that mature entirely in the river (Juanes et al. 2007; Valiente et al. 2005).

In Asturias, the fishing season currently spans 3 months, from mid‐April to mid‐July, although previous fishing regulations allowed a longer season. Every capture must be reported at a Salmon Tagging Center (Centro de Precintaje), an official facility where fish are registered and fin, muscle and scale samples are collected. This mandatory procedure supports catch monitoring, traceability, and conservation of wild salmon populations. The rivers analysed in this study (Figure 1) flow into the Cantabrian Sea and are among the longer and most important in the region, holding some of the largest salmon populations along the Cantabric Coast (Utanda Moreno 2003). Fishing catches have declined dramatically over recent decades. For example, during the 1980s, an average of approximately 3500 individuals were captured per fishing season. In contrast, during the last decade (2016–2025), the average annual catch decreased to 437 individuals, with only 127 salmon captured in the 2025 season (Hoz Regules et al. 2025). Current population numbers have placed these populations at high risk of extinction (Nicola et al. 2018).

FIGURE 1.

Nalón–Narcea and Sella rivers in Asturias, Spain. In the present study, Atlantic Salmon samples for two rivers and three time points (the early 2000s, the 2010s and the 2020s) were analysed to look for temporal DNA methylation changes.

The inhabiting biota of the rivers analysed in the present study is generally suffering from the effects of anthropogenic global change as recently published in Mytilus galloprovincialis (Gutierrez‐Rodriguez et al. 2025), Anguilla anguilla (Gutierrez‐Rodriguez et al. 2026) and Atlantic salmon (Joseph et al. 2026). Geological differences, legacy pollution, and habitat fragmentation vary between rivers, with the Sella being less disturbed than the Nalón–Narcea (Garcia‐Ordiales et al. 2018; Gutierrez‐Rodriguez et al. 2025; Rivas‐Iglesias et al. 2024). The Nalón–Narcea basin has a history of intensive mining and industrial activities that have impacted water quality (Garcia‐Ordiales et al. 2018) and is fragmented by multiple barriers, which negatively affect biodiversity (Fernández et al. 2018). Atlantic salmon distribution in the Nalón–Narcea basin is limited by several unpassable dams while the Sella river is completely accessible (Mortera Piorno and Hoz Regules 2020). Atlantic salmon phenotypic variation is these rivers has been reported. For instance, sea run timing is now delayed compared to 50 years before (Valiente et al. 2011) and the number of one‐sea‐winter (OSW) breeders have decreased (Hoz Regules et al. 2025). However, epigenetic or gene expression temporal analysis are lacking.

2.2. Sampling

Historical and contemporary samples are part of the laboratory collection and come from wild adult Atlantic salmons fished by recreational anglers. Fin clip samples were preserved in absolute ethanol and stored at −20°C until DNA extraction. Samples from adults of a similar size (mean length of 76.14 cm with a SD of 4.45 in the Nalón–Narcea and 75.72 cm with a SD of 5.68 in the Sella, corresponding predominantly to individuals spending 2 years at sea (MSW) in these populations (Hoz Regules et al. 2025)) were selected. Of these samples, only those exhibiting high DNA integrity, as indicated by the presence of high‐molecular‐weight genomic DNA, were retained for subsequent analyses (see DNA extraction). This resulted in reduced groups for the 2000s datasets as good historic gDNA is scarce. The final experimental design comprised 120 samples: six from Nalón–Narcea early 2000s (2002–2003), 12 from Sella early 2000s (2003), 26 from Nalón–Narcea early 2010s (2011), 22 from Sella early 2010s (2012), 27 from Nalón–Narcea early 2020s (2023) and 27 from Sella early 200 s (2023).

2.3. Tissue Choice

Adipose fin tissue was selected for this study for several reasons. First, it can be sampled easily and non‐lethally, facilitating straightforward implementation in population studies. Moreover, this tissue comprises multiple cell types, including epithelial, blood, and lymphatic cells (Hou et al. 2020), and exhibits active expression of neuron‐ and glial cell‐marker genes, suggesting a mechanoreceptive function (Koll et al. 2019). Consequently, this tissue possesses relevant functional activity. Furthermore, samples were obtained from wild fish captured by recreational anglers, making the collection of other tissues, such as gonads, liver, or brain, logistically unfeasible. Nevertheless, we acknowledge that findings derived from adipose fin tissue should be interpreted with caution when extrapolating to biological processes occurring in other tissues or organs.

2.4. DNA Extraction

Genomic DNA was extracted from frozen salmon adipose fin tissue using a phenol‐chloroform methodology, with tissue amounts ranging from 12 to 15 mg. Extracted DNA was diluted in 40 μL nuclease‐free water and preserved at −20°C. DNA integrity was assessed by 1% (w/v) agarose gel electrophoresis. To remove the low molecular weight DNA, samples were purified using the GeneMATRIX Agarose‐Out DNA Purification Kit (E3540, EurX). Then, DNA concentration was quantified using the Qubit dsDNA BR Kit (Q32850, ThermoFisher), and aliquots were diluted to a final concentration of 25 ng/μl. Only samples showing a high quality gDNA band of around 10 kb and more than 100 ng of gDNA were selected following the recommendations of the Premium RRBS Kit V2 (Hologic Diagenode).

2.5. Library Preparation and Sequencing

Libraries were prepared using the Premium RRBS Kit V2 (C02030036 (24) + C02030037 (96), Hologic Diagenode) according to the manufacturer's protocol, using 100 ng of genomic DNA per sample. RRBS provides a cost‐effective DNA methylation analysis with a focus on CpG loci. This method detects site‐specific methylation levels via bisulphite treatment, which converts unmethylated cytosine residues into uracil while methylated cytosines remain unchanged. MspI restriction enzyme (5′‐C^CGG‐3′) was employed for DNA digestion. After library preparation, samples were pooled in four random groups, one of 24 and three of 32 samples. For quality assurance the final libraries concentrations were determined by Qubit dsDNA HS Kit (Q33230, ThermoFisher) and a 2100 Bioanalyzer (2100 Expert Software version B.02.10.SI764, Agilent) run was performed. Sequencing, performed by Macrogen (Korea), was done in four NovaSeq X lanes (150 bp paired‐end mode, 350Gb per lane) with a 20% of PhiX.

2.6. Bioinformatics and Statistics

Initial quality control of raw sequencing reads was performed using FastQC (Andrews et al. 2010). Demultiplexing was conducted using fumi_tools demultiplex (Fehlmann 2019), followed by individual quality assessments with FastQC. Summarized quality reports were generated using MultiQC (Ewels et al. 2016). Quality and adapter trimming was performed with Trim Galore! (Krueger 2015), low quality base pairs and the end of reads were removed (Phred score > 20), and reads shorter than 15 bp were discarded. The −rrbs, ‐‐paired, and −non_directional options were selected. After trimming, quality was again checked individually with FastQC. Alignment to the reference genome of Salmo salar (GCF_905237065.1; NCBI) was performed using Bismark aligner (Krueger and Andrews 2011). A minimum alignment score (‐‐score_min) needed for an alignment to be considered was set to L, 0, −0.6 with the −pbat option. Though the Bismark alignment creates an alignment report per sample, additional alignment control was performed with Samtools flagstats (Li et al. 2009). Bisulphite conversion efficiency was assessed using methylated and unmethylated spike‐in controls included in the RRBS kit.

For the CpG methylation analysis, the R package methylKit (Akalin et al. 2012) was employed. For DMRs, calculation data was filtered by a minimum coverage of three reads, bases that have more than 99.9th percentile of coverage were also discarded. Later, coverage across samples was normalized to reduce bias. After filtering, samples with insufficient data were removed (17) to prevent inflating the group sizes with samples of low contribution. Evaluated windows were defined as 1000 bp, with 1000 bp step size, that have at least 10 cytosines, each with a minimum coverage of three reads. Region size was of that length as CpG islands are of 1 kb on average (Deaton and Bird 2011). After that, a pool per group was created with a minimum requirement of three samples supporting each region. The aim is to obtain the DNA methylation information that is characteristic of a population at a specific moment (Barouch et al. 2024), while dealing with the specific limitations of working with salmonids RRBS data and big sample sizes (El Kamouh et al. 2024). Then, DMRs with |∆meth > 20%| and q‐value of 0.01 between the early 2000s and the early 2020s samples were calculated per river using the Fisher's exact test. To generate a more stringent list of regions, only DMRs common between both rivers, showing the same tendency were retained for further analysis.

For DMCs, the workflow was similar but lacking the tilling window analysis. Genomic annotation was done using ChIPseeker (Wang et al. 2022; Yu et al. 2015). Gene ontology (GO) and the Kyoto Encyclopaedia of Genes and Genomes (KEEG) enrichment analysis were performed using clusterProfiler (Yu et al. 2012). CpG context annotation was done using the EMBOSS newcpgreport tool. The CpG island annotation was used to define CpG shores (2 kb flanking regions up‐ and downstream of CpG islands), shelves (2 kb flanking regions adjacent to shores), and open sea regions (regions outside the others). For those identified DMRs, the Jonckheere‐Terpstra (JT) non‐parametric test (with FDR correction) was used to determine the significance of the tendency including samples from 2011 to 2012 from both rivers. For this, the mean methylation of each sample contributing to that region was used.

3. Results

The number of raw sequences per sequencing pool was 808.1 M, 1022.8 M, 1582.6 M and 1263.4 M. After quality filtering and demultiplexing, a mean of 28.04 M (SD = 2.42 M) high quality paired end RRBS reads were retained per sample (Table S1). The mean mapping efficiency to the reference genome was 52.56%, like other RRBS studies working with this species (Katirtzoglou et al. 2024; Uren Webster et al. 2018). The analysis of the spike methylation controls yielded conversion efficiencies of 98.44%, 98.41%, 98.06%, and 98.41% for the four respective libraries, indicating a high and consistent level of bisulphite conversion efficiency across all samples. Out of the initial 120 samples, 17 were excluded from further analysis due to low number of reads, low mapping efficiency or low mean global methylation (Table S1). The final group sizes were: six samples from Nalón–Narcea early 2000s (Nar03), 22 from Nalón–Narcea early 2010s (Nar11), 23 from Nalón–Narcea early 2020s (Nar23), 12 from Sella early 2000s (Sel03), 15 from Sella early 2010s (Sel12), and 25 from Sella early 2020s (Sel23). A total of 14.1% of the samples were excluded from the pooling process as they contributed little to their respective pools despite increasing overall sample sizes.

After pooling, the number of CpG sites identified was as follows: Nar03, 960,760 CpGs; Sel03, 1,516,143 CpGs; Nar11, 2,497,153 CpGs; Sel12, 1,980,500 CpGs; Nar23, 3,066,826 CpGs; and Sel23, 3,120,187 CpGs. Among these, 715,240 CpG sites were common across all six pools. Of the shared CpGs, 43.75% were mapped to distal intergenic regions, 33.62% to introns, 14.73% to promoters, and 7.28% to exons. Regarding their genomic context, 43.19% were located within CpG islands, 24.66% in shores, 7.38% in shelves, and 24.78% in open sea regions. CpG islands are CpG‐rich regions, predominantly nonmethylated, and frequently coincide with transcription starting sites. These regions influence chromatin structure and gene expression (Deaton and Bird 2011). The genome‐wide average methylation was 81.79%, similar to previously reported means (Katirtzoglou et al. 2024; Uren Webster et al. 2018). No significant differences in mean methylation were observed across sampling years (Kruskal–Wallis, 𝝌² = 9.1127, df = 5, p‐value = 0.1047), indicating that DNA deamination was not evident in older samples (Figure S1).

3.1. Shared Epigenetic Variation

There were 13,017 DMCs (|∆meth > 20%|) common for both rivers showing the same trend when comparing samples from the early 2000s with those of the early 2020s; 6968 were hypermethylated and 6049 were hypomethylated. Following genomic annotation, 43% of DMCs mapped to distal intergenic regions, 33.4% to introns, 14.6% to promoters, and 8.1% to exons. Regarding CpG context, 37.6% of sites were located in open sea regions, 28.04% in CpG islands, 23.52% in shores, and 10.85% in shelves. GO enrichment analysis (Tables S2–S5) identified a total of 75 significantly enriched GO terms across all categories. When sorted by category, 40 terms corresponded to Biological Process (BP) (Figure 2a), 13 to Cellular Component (CC) (Figure 2b), and 15 to Molecular Function (MF) (Figure 2c). The Top 10 enriched GO terms included beta‐catenin binding (GO:0008013), catenin complex (GO:0016342), multicellular organism development (GO:0007275), adherens junction (GO:0005912), cadherin binding (GO:0045296), axon guidance (GO:0007411), cell migration (GO:0016477), cell–cell junction assembly (GO:00070439), calcium‐dependent cell–cell adhesion via plasma membrane cell adhesion molecules (GO:0016339), and adherens junction organization (GO:0034332). KEGG pathway enrichment analysis revealed nine significant pathways (Figure 2d), with extracellular matrix (ECM)‐receptor interaction (Environmental Information Processing; signalling molecules and interaction), TGF‐beta signalling pathway (Environmental Information Processing; Signal transduction), and cell adhesion molecules (Environmental Information Processing; signalling molecules and interaction) among the most significant (Table S6).

FIGURE 2.

Gene ontology enrichment analysis by GO category: (a) biological processes (BP), (b) cellular component (CC), (c) molecular function (MF) and (d) KEGG pathway enrichment.

Regarding differentially methylated regions (DMRs), the comparison between historical (early 2000s) and contemporary (early 2020s) samples revealed 36 DMRs common for both rivers and exhibiting the same directional trend. These regions were primarily located in introns (36.1%), distal intergenic regions (30.6%), promoters (19.4%), and exons (11.1%). In terms of CpG context, 47.22% were mapped to CpG islands, 25% to open sea regions, 13.89% to shelves, and 13.89% to shores. Of the 36 DMRs, 17 showed a significant tendency (Jonckheere–Terpstra; p‐adj. < 0.05) when including the early 2010s (2011/12) data (Figure 3; Table S7). The promoter‐associated regions (Figure 4) included NC_059455.1:86,940,001‐86,941,000, located 1001 bp downstream of the transcription start site (TSS) of hoxa2aa, which encodes homeobox protein HoxA2aa (Mean ∆meth = −28.48%, JT adj. p‐value = 0.014); NC_059453.1: 62,084,001‐62,085,000, encompassing the TSS of hoxc8ba, which encodes the homeobox protein HoxC8ba (Mean ∆meth = 28.97%, JT adj. p‐value = 0.018); NC_059454.1:75,562,001‐75,563,000, located 558 bp upstream of the TSS of LOC106568006, predicted to encode a G protein‐coupled receptor kinase 6‐like (Mean ∆meth = 28.06%, JT adj. p‐value = 0.016); and NC_059443.1: 94,442,001–94,443,000, located 964 bp of the TSS of LOC106592199, predicted to encode the sodium/glucose cotransporter 5, protein FAM83G‐like (Mean ∆meth = 34.80%, JT adj. p‐value = 0.047).

FIGURE 3.

DMRs identified between the early 2000s and the early 2020s pools that show a significant tendency common for both rivers (Jonckheere‐Terpstra; p‐adj. < 0.05). Positive and negative DMRs tendencies by rivers can be found in Figures S2 and S3.

FIGURE 4.

Dot plots showing DNA methylation percentages across Differentially Methylated Regions (DMRs) and samples. The analysed DMRs correspond to promoter regions. Intra‐group variability may arise because CpG sites within each region are not necessarily consistent across samples. Only regions with at least 10 CpGs meeting coverage criteria were selected, and methylation levels were calculated accordingly. In the plots, blue triangles represent samples from the Sella River, while red circles represent samples from the Nalón–Narcea River.

Seven DMRs were mapped within intronic regions (Table S7; Figure S4). The most relevant included: NC_059459.1:60,949,001‐60,950,000, located in intron 1 of 7 of the gene XM_014155831.2/LOC106577616, which encodes the Kv channel‐interacting protein 4 (Mean ∆meth = 22.39%, JT adj. p‐value = 0.014); NC_059463.1:42,499,001‐42,500,000, located in intron 3 of 28 of the gene XM_014167623.2/cacna1da, encoding the calcium channel, voltage‐dependent, L type, alpha 1D subunit, variant a (Mean ∆meth = 21.03%, JT adj. p‐value = 0.014); NC_059447.1: 19,429,001‐19,430,000, corresponding to intron 3 of 16 of the gene XM_045719885.1/LOC106606554, which encodes the protein mono‐ADP‐ribosyltransferase PARP14 (Mean ∆meth = 28.41%, JT adj. p‐value = 0.031); NC_059442.1: 124,787,001‐124,788,000, located in intron 2 of 26 of the gene XM_014125432.2/LOC106561454, coding the guanine nucleotide exchange factor subunit RIC1 (Mean ∆meth = −25.55%, JT adj. p‐value = 0.040); NC_059459.1: 19,257,001‐19,258,000, located in intron 5 of 35 of the gene LOC106577001, coding the protocadherin‐15‐like protein (Mean ∆meth = −25.87%, JT adj. p‐value = 0.0412); and NC_059463.1: 54,205,001‐54,206,000, within intron 26 of 38 of the gene LOC106582526, predicted to encode a fibronectin‐like protein (Mean ∆meth = −26.17%, JT adj. p‐value = 0.0146).

The remaining DMRs were mapped to distal intergenic regions or were uncharacterized (Figure 3). For the GO enrichment analysis, the 36 DMRs were used. Only one term, GO:0003950 to NAD+ poly‐ADP‐ribosyltransferase activity, was significantly enriched, although with limited statistical support (Fold enrichment 27.2, p‐adj = 0.048).

3.2. Between Rivers Epigenetic Variation

Temporal DNA methylation differences between rivers showing a opposite trend were also analysed. However, these results should be interpreted with caution as genetic differences between both populations do exist. The comparison between historical (early 2000s) and contemporary (early 2020s) samples revealed 18 DMRs common for both rivers and exhibiting the opposite directional trend. Of those, six had significant tendencies when including data from the early 2010s at least for one of the rivers (Table S8; Figure S5). This DMRs corresponded to four promoters: NC_059465.1:15,106,001‐15,107,000, located −2740 bp upstream of the TSS of the gene the cdk7, coding the Cyclin dependent kinase 7 (∆meth Nar 2000s–2020s = 29.32%; ∆meth Sel 2000s–2020s = −26.87%; JT_padj Nar 2000s–2020s = > 0.05; JT_padj Sel 2000s–2020s = 0.005); NC_059452.1:58,548,001‐58,549,000, located 167 bp downstream of the TSS of the LOC106562559 gene, coding the Double C2‐like domain‐containing protein beta (∆meth Nar 2000s–2020s = −28.49%; ∆meth Sel 2000s–2020s = 26.31%; JT_padj Nar 2000s–2020s = > 0.05; JT_padj Sel 2000s–2020s = 0.007); NC_059442.1:21,764,001‐21,765,000, located within the promoter of the gene LOC106575475 coding the COUP transcription factor 2‐like (∆meth Nar 2000s‐20s = 21.25%; ∆meth Sel 2000s–2020s = −26.13; JT_padj Nar 2000s–2020s = 0.032; JT_padj Sel 2000s–2020s = > 0.05) and the NC_059449.1:4,303,001‐4,304,000, located −561 bp upstream of the TSS of the gene LOC106609964 (unknown coding) (∆meth Nar 2000s–2020s = −26.09%; ∆meth Sel 2000s–2020s = 20.50%; JT_padj Nar 2000s–2020s = 0.040; JT_padj Sel 2000s–2020s = 0.040). Additionally, one DMR was exonic: NC_059457.1:63,599,001‐63,600,000, located in the exon 9 of 9 of the gene LOC106574379 coding for a Parathyroid hormone 2 receptor‐like (∆meth Nar 2000s–2020s = 23.83%; ∆meth Sel 2000s–2020s = −28.43%; JT_padj Nar 2000s–2020s = > 0.05; JT_padj Sel 2000s–2020s = 0.026). Finally, one DMR was in the upstream UTR region: NC_059464.1:9,498,001‐9,499,000, located −42,158 bp from the TSS of the gene LOC106584031 which codes an inactive tyrosine‐protein kinase transmembrane receptor ROR1.

4. Discussion

In the present study, temporal genome‐wide methylation patterns in Salmo salar from two rivers, the Nalón–Narcea and Sella, were analysed using Reduced Representation Bisulphite Sequencing (RRBS) across three time periods (early 2000s, 2010s, and 2020s). These rivers harbour unique salmon populations located at the southern limit of the species' distribution. Consequently, they are particularly exposed to the effects of global change and have experienced marked population declines (Jonsson and Jonsson 2009; Nicola et al. 2018; Thorstad et al. 2021), with salmon abundance in the region decreasing by approximately 70% over recent decades (Hoz Regules et al. 2025). The focus of this study is on DNAm epialleles (DMCs and DMRs) exhibiting consistent temporal changes in both rivers, independently of the genetic background of the populations (pure epialleles) (Richards 2006). The current lack of knowledge regarding epigenetic reprogramming in this species, together with the design of our study, precludes distinguishing between inherited epigenetic variation and convergent intra‐generational changes induced by environmental conditions. Nevertheless, temporal epigenetic shifts may shed light into microevolutionary processes under natural conditions and provide a deeper understanding of the role of natural epigenetic variation in adaptive evolution (Chapelle and Silvestre 2022). These potentially heritable epigenetic modifications can act as an additional layer of heritable variation upon which natural selection might act. Recent reviews emphasize that epigenetic signals triggered by the environment can persist long enough to influence adaptive phenotypes, especially in rapidly changing or fluctuating environments (Korolenko and Skinner 2024; Sabarís et al. 2023). Here, we identified differentially methylated cytosines (DMCs) and regions (DMRs), including in the promoters of two Hox genes essential for embryonic development. Gene ontology and KEGG pathway enrichment analyses showed a significant overrepresentation of developmental pathways, including the ECM–receptor interaction pathway, which is essential for morphogenesis and vertebral formation. In addition, several enriched pathways were associated with environmental information processing and cell signalling, suggesting a potential epigenetic response to environmental stressors.

Global environmental change is exerting profound impacts across ecosystems (Walker and Steffen 1997), with temperature rise and ocean acidification among the most critical stressors for aquatic organisms. Industrial CO₂ emissions increase both atmospheric and dissolved CO₂ levels, leading to ocean acidification, reduced oxygen availability, and elevated water temperatures (Doney et al. 2012). These stressors, individually and in combination, can disrupt key developmental and physiological processes in fish, thereby influencing growth, survival, and reproduction. In this context, environmentally responsive epigenetic mechanisms such as DNAm may play an important role in mediating phenotypic plasticity and facilitating short‐term acclimation to changing environmental conditions.

Epigenetic mechanisms as DNAm produce plastic phenotypes leading to local adaptation to the environment, though epigenetic changes usually depend on the genotype‐environment interaction (Berbel‐Filho et al. 2019). Types of epialleles were first defined by Richards (2006), who stated that the significance of meiotic transmission of epigenetic modifications relies on the relationship between the epigenetic state and the genotypic context; defining three types of epigenetic variation: obligatory epigenetic variation (where the epigenotype is a direct consequence of the genotype), facilitated epigenetic variation (where the epigenotype depends partially on both the genotype and the environment), and pure epigenetic variation, which only depends on the environment. However, this classification is population specific, and the pure variation identified in this study may be facilitated if genotypic variation between samples is expanded. These epialleles may contribute to population microevolution by facilitating rapid responses to changing and stressful environmental conditions, and may arise through either environmental induction or stochastic mechanisms (Chapelle and Silvestre 2022). The transcriptomic results of epigenetic variation are finally gene expression changes or alternative splicing events leading to different transcripts or different abundance of those (Maor et al. 2015; Moore et al. 2013; Singh and Ahi 2022). The classification and quantification of epigenetic variation is of concern for population epigenetic, as pure DNAm variation is the only type of epigenetic variation that can lead to an autonomous, genotype‐independent contribution to evolutionary processes (Mueller et al. 2025).

Atlantic salmon has been widely used in epigenetic studies and increasing evidence of the effects of anthropogenic environmental conditions on the phenotype are being highlighted (Best et al. 2018). Factors such as hypoxia, acidification, temperature, salinity, pollution, or nutrition have all been shown to influence the epigenome (Best et al. 2018). In the present study, we identified DNA methylation changes of at least 20% in 1 kb regions, many of which show consistent trends even when including samples from the early 2010s (2011–2012) (Figures 3 and 4). Particularly noteworthy are the Hox genes, whose promoter regions exhibit differential methylation. Hox genes are essential transcription factor coding genes that play a central role in body plan development and have been crucial to vertebrate evolution (Mallo 2018). This gene family is also essential for fin development in teleost fishes (Cumplido et al. 2024). As a consequence of two rounds of genome duplication (Lien et al. 2016), Atlantic salmon has 118 Hox genes organized in 13 Hox clusters (Mungpakdee et al. 2008). The hoxa2aa is a gene encoding the homeobox protein HoxA2aa. This gene is relevant for the hindbrain and the pharyngeal arches development (Scemama et al. 2006; Seifert et al. 2015). However, its DNA methylation state has also been recently associated with breast carcinogenesis in humans (De Palma et al. 2025) and is considered one of the top 20 genes involved in colorectal cancer development (Li et al. 2019). The hoxc8ba encodes the homeobox protein HoxC8ba. The hoxC‐related clusters include regions that regulate the development of the dorsal and anal fins in fish along the anterior–posterior axis (Adachi et al. 2024). This gene has been reported to be downregulated in Atlantic salmon post‐smolts following long‐term exposure to CO₂, suggesting its responsiveness to the environmental stressor (Mota et al. 2020). The methylation state of these genes may have important biological implications, particularly during development and morphogenesis (Cumplido et al. 2024). Although the functional roles of these genes are most prominent during early life stages, epigenetic marks established during development can persist throughout life. Indeed, persistent epialleles have been used to infer past environmental conditions across multiple tissues (Anastasiadi, Shao, et al. 2021). In addition to that, the promoter of the gene encoding G protein‐coupled receptor kinase 6 (GRK6) was found to be significantly hypermethylated in this study. This kinase is a key modulator of G protein‐coupled receptors (GPCRs) activity through phosphorylating activated receptors and is involved in diverse physiological processes, including immune cell migration, inflammation resolution, and dopamine receptor signalling (Stegen and Frey 2022). Finally, the promoter region of the FAM83G gene also exhibited hypermethylation along the two decades period in both rivers. This gene has been associated with solid tumours development and progression (Jiang et al. 2023; Zhao et al. 2024), suggesting potential deleterious effects in salmon.

Generally, hypermethylation in promoter regions is associated with gene expression repression, whereas hypomethylation is associated with gene expression activation. Nonetheless, the relationship between promoter methylation and gene expression is complex and context‐dependent, requiring validation through transcriptomic analyses (de Mendoza et al. 2022). Potential developmental alterations in response to changing conditions, such as temperature increase or ocean acidification, are plausible as the methylation state of two hox genes was significantly altered along the studied temporal samples for both rivers. Elevated temperatures have been shown to induce developmental defects due to changes in cell proliferation, differentiation, and gene expression in Atlantic salmon (Burgerhout et al. 2017). Moreover, skeletal deformities have been linked to hyperthermia and are related to extracellular matrix genes (ECM) (Ytteborg et al. 2010).

Some DMRs identified in the present study were located in intronic regions. DNAm in gene bodies, both in introns and exons, is involved in promoter usage, alternative splicing (Maor et al. 2015; Shayevitch et al. 2018) and transcription elongation (Jones 2012). These processes contribute to phenotypic plasticity enabling the emergence of different phenotypes from a single genotype. DNA methylation in the first intron, which is rich in enhancers, and the first exon, have shown an inverse correlation with gene expression across tissues and species (Anastasiadi et al. 2018; Brenet et al. 2011); however, the specific effects of DNA methylation in gene bodies needs to be further investigated. In this study, the first intron of the gene encoding for the Kv channel‐interacting protein 4 showed increased hypermethylation over the two‐decade period, suggesting the potential downregulation of this gene. Interestingly, the same gene was reported to be downregulated in response to elevated temperature in whole‐body larval samples of the estuarine longfin smelt ( Spirinchus thaleichthys ) (Jeffries et al. 2016). Moreover, cacna1da, encoding the calcium channel, voltage‐dependent, L type, alpha 1D subunit was hypermethylated in the intron 3 of 28, which was also downregulated in response to temperature in the longfin smelt (Jeffries et al. 2016). Calcium channels are relevant in neural excitability and neurotransmitter release. This gene has been reported to be downregulated in olfactory bulbs from coho salmon exposed to high CO₂ affecting sensory systems (Williams et al. 2019). These systems are essential for salmon as they rely on them during their homing migrations back to the natal streams and support the hypothesis of potential hyperthermia effects. PARP14 is involved in ADP‐ribosylation and has functions in transcriptional regulation, immune responses and metabolism (Đukić et al. 2023) and has shown antiviral immune function in Atlantic salmon (Clark et al. 2023). These genes have shown different responses to environmental stimuli and stress. Additionally, the gene encoding the protocadherin‐15‐like protein was significantly hypomethylated in the intron 5 of 35 over the temporal period analysed. In relation to this, cadherin‐4 DNA methylation was also affected by hypoxia and acidification in Large Yellow Croaker ( Larimichthys crocea ) (Yue et al. 2023). Finally, the fibronectin‐like protein gene was hypomethylated in the intron 26 of 38. Fibronectin is essential for normal vertebra development, it plays adhesive roles in the extracellular matrix and is key in morphogenesis (Guillon et al. 2020). The fact that some genes that show responsiveness to CO₂ or temperature in the literature are differentially methylated in Atlantic salmon may suggest potential environmental influence in the methylome of this species in Southern Europe. Nevertheless, the functional and adaptive significance of the observed temporal DNA methylation changes remains to be determined, and future studies integrating epigenomic, transcriptomic, and phenotypic data will be needed to further assess their biological relevance.

With regards to the GO enrichment analyses performed with the DMCs, it is worth noting that terms as multicellular organism development, axon guidance, cell–cell junction assembly, cell morphogenesis, calcium‐dependent cell–cell adhesion molecules, catenin complex, glutamatergic synapse were identified. These GO terms highlight biological processes as cell communication, synapsis and organism development, supporting the hypothesis that developmental alterations may result from environmental stressors potentially associated with the species being at the edge of its distribution range. Moreover, nine KEGG pathways have been also significantly overrepresented when analysing the DMCs, belonging five of them to the category environmental information processing. ECM‐receptor interaction was the pathway more enriched. It consists of structural and functional macromolecules playing important roles in tissue and organ morphogenesis and the maintenance of cell and tissue structure and function (Melo‐Braga et al. 2014). This pathway has being identified as relevant for adaptive response to hypoxia in marine animals (Huo et al. 2024) and was enriched in response to cortisol mediated stress in rainbow trout ( Oncorhynchus mykiss ) (Aravena‐Canales et al. 2024). Moreover, as previously mentioned, hyperthermia has shown to induce developmental defects in Atlantic salmon leading to skeletal deformities related to ECM (Ytteborg et al. 2010). In relation with this, cell adhesion molecules pathway was also enriched in the present study and in response to hypoxia and acidification stress in large yellow croaker (Yue et al. 2023). This pathway is essential for homeostasis, the immune response, inflammation, embryogenesis, and development of neuronal tissue (Aplin et al. 1999). Finally, TGF‐beta signalling pathway was enriched in domesticated Nile tilapia (Podgorniak et al. 2022). It is known that domestication effects occur early even within the first generation (Milla et al. 2021). This can lead to introgression in wild populations when captive reared salmon is released in the wild and mates wild individuals (Rodriguez Barreto et al. 2019). In the rivers studied here, supportive breeding is generally applied (Hoz Regules et al. 2025) although it was found to account for < 10% of the census (Juanes et al. 2011). Moreover, in some cases genetic diversity losses occur due to an inadequate strategy of crosses (Horreo et al. 2008; Machado‐Schiaffino et al. 2007).

In the present study, epigenetic differences between rivers were also identified over the two‐decade period, as DMRs with opposite significant trends were found. However, although both rivers may show different abiotic and biotic conditions due to their differences in historical use and environmental conditions (Dopico et al. 2009; Garcia‐Ordiales et al. 2018), these differences may account, at least partially, for the genetic differences existing between both rivers as they constitute two separate breeding units (Griffiths et al. 2010) and the possible random temporal variations in allele frequencies that may likely be happening due to population decline (Dadswell et al. 2022).

These findings suggest that environmentally responsive DNA methylation may play a role in the phenotypic plasticity and potential adaptive responses of Atlantic salmon facing climatic and ecological stressors at the southern edge of their range (Best et al. 2018; Sabarís et al. 2023). However, it is worth noting that the observed methylation patterns may reflect environmentally induced changes rather than strictly heritable epialleles (Richards 2006). As these populations occupy the southernmost part of the species' distribution, they are likely exposed to chronic environmental stress as rising temperatures, which may lead to stress‐induced epigenetic modifications (Thorstad et al. 2021). Such changes do not necessarily confer an adaptive benefit and may instead reflect physiological strain or maladaptive responses to ongoing environmental deterioration (Anastasiadi, Shao, et al. 2021). Moreover, decreasing demographic changes in North Iberian salmon may additionally lead to epigenetic changes due to changes in allele frequency by genetic drift. Among the potential limitations of this study, it is important to acknowledge the absence of a temporal genetic variation analysis, which would help determine whether some of the observed epigenetic patterns could be partially driven by changes in allele frequencies over time. Additionally, the sample sizes of the historical groups were smaller than desired due to difficulties associated with historical tissue preservation, which may limit the robustness of some conclusions. Finally, because all results were obtained from fin‐clip tissue, tissue‐specific methylation patterns cannot be entirely ruled out. Nevertheless, we detected significant temporal methylation changes that are consistent with the environmental variation expected over the studied period.

While our results suggest that epigenetic variation may be linked to environmental changes occurring, at the southern edge of the Atlantic salmon distribution, where populations are undergoing marked declines, we consider that formally testing correlations with single environmental variables would not provide meaningful evidence in this case. Given that our data span only three time points, any quantitative correlation with long‐term trends (e.g., temperature increase) would be statistically underpowered and potentially misleading. Therefore, we interpret our findings in a qualitative ecological context, acknowledging that the observed epigenetic patterns are likely to reflect complex interactions between multiple environmental stressors and demographic processes rather than a single explanatory variable. Further functional validation, particularly through transcriptomic analyses and multi‐generational studies, will be essential to distinguish between plastic, transient responses and heritable epigenetic modifications that could contribute to population‐level adaptation under natural conditions (Chapelle and Silvestre 2022; Korolenko and Skinner 2024).

5. Conclusions

In this study we investigate whether environmental, among others, changes have left a detectable epigenetic signature in Atlantic salmon ( Salmo salar ) populations at the southern edge of the species' distribution over time. By comparing adult fin clip samples collected over a 20‐year period (2003–2004, 2011–2012, and 2023), we identified consistent, temporally structured changes in DNA methylation, particularly in genes involved in development and stress response. These differentially methylated cytosines and regions were detected across independent river systems, suggesting that the observed epigenetic variation is not purely stochastic or population‐specific, but potentially linked to shared environmental stressors such as rising temperatures and acidification, among others. These DNA methylation changes showed a significant constant tendency over the two‐decade period analysed.

While causality and heritability remain unresolved, our findings support the hypothesis that environmentally induced epigenetic modifications may contribute to phenotypic plasticity and could, under certain conditions, participate in rapid adaptive responses. Given the endangered status of southern Atlantic salmon populations and the intensifying impacts of global change, future research integrating epigenomic, transcriptomic, and fitness data across generations will be essential to clarify the functional relevance and evolutionary potential of these epigenetic signals. Such integrative approaches may ultimately inform conservation strategies that account for both genetic and epigenetic dimensions of resilience.

Author Contributions

A.G.‐R.: conceptualization, methodology, data curation, formal analysis, writing – original draft. C.B.‐F.: investigation, methodology, review and editing. E.G.‐V.: methodology, writing – review and editing, Funding acquisition. G.M.‐S.: conceptualization, methodology, writing – review and editing, supervision, funding acquisition.

Funding

This study received funds from the Spanish Ministry of Science and Innovation, Grant PID2022‐138523OBI00, and the Principality of Asturias (Spain) with the Grants GRU‐GIC‐24‐051 & MRR‐24‐BIODIVERSIDAD‐BIO12. Alvaro Gutierrez‐Rodriguez is funded by the University of Oviedo (Grant PAPI‐24‐TESIS‐18).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Mean global methylation of every sample by group. There are not significant differences between river nor years (Kruskal–Wallis p‐value > 0.05).

Figure S2: DMRs identified between the early 2000s and the early 2020s pools that show a significant negative tendency, common for both rivers, by river (Jonckheere‐Terpstra; p‐adj. < 0.05).

Figure S3: DMRs identified between the early 2000s and the early 2020s pools that show a significant positive tendency, common for both rivers, by river (Jonckheere‐Terpstra; p‐adj. < 0.05).

Figure S4: Dot plots depicting DNA methylation percentages across Differentially Methylated Regions (DMRs) located within intronic regions and samples. Intra‐group variability may reflect differences in CpG site composition across samples within each region. Only regions containing at least 10 CpGs with adequate coverage were analysed. Triangles indicate samples from the Sella River, while circles indicate samples from the Nalón–Narcea River.

Figure S5: Dot plots depicting DNA methylation percentages across Differentially Methylated Regions (DMRs) showing opposite and significative trend between rivers (Jonckheere–Terpstra test with FDR correction). Intra‐group variability may reflect differences in CpG site composition across samples within each region. Only regions containing at least 10 CpGs with adequate coverage were analysed. Triangles indicate samples from the Sella River, while circles indicate samples from the Nalón–Narcea River.

Table S1: Sample statistics. Only samples labelled with the tag “Pass” were used for the analysis.

Table S2: Enriched GO terms without considering GO categories for the DMCs (|∆meth > 20%|) common for both rivers and showing the same trend between the early 2000s and the early 2020s samples.

Table S3: Enriched GO terms for the biological process category of the DMCs (|∆meth > 20%|) common for both rivers and showing the same trend between early 2000 and early 2020s samples.

Table S4: Enriched GO terms for the cellular component category of the DMCs (|∆meth > 20%|) common for both rivers and showing the same tendency between early 2000s and early 2020s samples.

Table S5: Enriched GO terms for the molecular function category of the DMCs (|∆meth > 20%|) common for both rivers and showing the same tendency between early 2000s and early 2020s samples.

Table S6: KEGG enrichment analysis of the DMCs (|∆meth > 20%|) obtained when comparing early 2000s and early 2020s samples. This DMCs are filtered so that they are common for both rivers and show the same trend.

Table S7: DMRs shared by both rivers with the same and significant trend (Jonckheere–Terpstra test with FDR correction) and |∆methylation (2000s–2020s)| > 20%.

Table S8: DMRs shared by both rivers with opposite and significant trends (Jonckheere–Terpstra test with FDR correction) and |∆methylation (2000s–2020s)| > 20%.

Data Availability Statement

The raw Reduced Representation Bisulphite Sequencing data has been uploaded to the NCBI Short Read Archive (SRA) as BioProject PRJNA1345345.